]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blame - mm/memory.c
mtd: jedec_probe: fix NEC uPD29F064115 detection
[mirror_ubuntu-bionic-kernel.git] / mm / memory.c
CommitLineData
1da177e4
LT
1/*
2 * linux/mm/memory.c
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 */
6
7/*
8 * demand-loading started 01.12.91 - seems it is high on the list of
9 * things wanted, and it should be easy to implement. - Linus
10 */
11
12/*
13 * Ok, demand-loading was easy, shared pages a little bit tricker. Shared
14 * pages started 02.12.91, seems to work. - Linus.
15 *
16 * Tested sharing by executing about 30 /bin/sh: under the old kernel it
17 * would have taken more than the 6M I have free, but it worked well as
18 * far as I could see.
19 *
20 * Also corrected some "invalidate()"s - I wasn't doing enough of them.
21 */
22
23/*
24 * Real VM (paging to/from disk) started 18.12.91. Much more work and
25 * thought has to go into this. Oh, well..
26 * 19.12.91 - works, somewhat. Sometimes I get faults, don't know why.
27 * Found it. Everything seems to work now.
28 * 20.12.91 - Ok, making the swap-device changeable like the root.
29 */
30
31/*
32 * 05.04.94 - Multi-page memory management added for v1.1.
33 * Idea by Alex Bligh (alex@cconcepts.co.uk)
34 *
35 * 16.07.99 - Support of BIGMEM added by Gerhard Wichert, Siemens AG
36 * (Gerhard.Wichert@pdb.siemens.de)
37 *
38 * Aug/Sep 2004 Changed to four level page tables (Andi Kleen)
39 */
40
41#include <linux/kernel_stat.h>
42#include <linux/mm.h>
43#include <linux/hugetlb.h>
44#include <linux/mman.h>
45#include <linux/swap.h>
46#include <linux/highmem.h>
47#include <linux/pagemap.h>
48#include <linux/rmap.h>
49#include <linux/module.h>
0ff92245 50#include <linux/delayacct.h>
1da177e4 51#include <linux/init.h>
edc79b2a 52#include <linux/writeback.h>
8a9f3ccd 53#include <linux/memcontrol.h>
cddb8a5c 54#include <linux/mmu_notifier.h>
3dc14741
HD
55#include <linux/kallsyms.h>
56#include <linux/swapops.h>
57#include <linux/elf.h>
1da177e4
LT
58
59#include <asm/pgalloc.h>
60#include <asm/uaccess.h>
61#include <asm/tlb.h>
62#include <asm/tlbflush.h>
63#include <asm/pgtable.h>
64
42b77728
JB
65#include "internal.h"
66
d41dee36 67#ifndef CONFIG_NEED_MULTIPLE_NODES
1da177e4
LT
68/* use the per-pgdat data instead for discontigmem - mbligh */
69unsigned long max_mapnr;
70struct page *mem_map;
71
72EXPORT_SYMBOL(max_mapnr);
73EXPORT_SYMBOL(mem_map);
74#endif
75
76unsigned long num_physpages;
77/*
78 * A number of key systems in x86 including ioremap() rely on the assumption
79 * that high_memory defines the upper bound on direct map memory, then end
80 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
81 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
82 * and ZONE_HIGHMEM.
83 */
84void * high_memory;
1da177e4
LT
85
86EXPORT_SYMBOL(num_physpages);
87EXPORT_SYMBOL(high_memory);
1da177e4 88
32a93233
IM
89/*
90 * Randomize the address space (stacks, mmaps, brk, etc.).
91 *
92 * ( When CONFIG_COMPAT_BRK=y we exclude brk from randomization,
93 * as ancient (libc5 based) binaries can segfault. )
94 */
95int randomize_va_space __read_mostly =
96#ifdef CONFIG_COMPAT_BRK
97 1;
98#else
99 2;
100#endif
a62eaf15
AK
101
102static int __init disable_randmaps(char *s)
103{
104 randomize_va_space = 0;
9b41046c 105 return 1;
a62eaf15
AK
106}
107__setup("norandmaps", disable_randmaps);
108
109
1da177e4
LT
110/*
111 * If a p?d_bad entry is found while walking page tables, report
112 * the error, before resetting entry to p?d_none. Usually (but
113 * very seldom) called out from the p?d_none_or_clear_bad macros.
114 */
115
116void pgd_clear_bad(pgd_t *pgd)
117{
118 pgd_ERROR(*pgd);
119 pgd_clear(pgd);
120}
121
122void pud_clear_bad(pud_t *pud)
123{
124 pud_ERROR(*pud);
125 pud_clear(pud);
126}
127
128void pmd_clear_bad(pmd_t *pmd)
129{
130 pmd_ERROR(*pmd);
131 pmd_clear(pmd);
132}
133
134/*
135 * Note: this doesn't free the actual pages themselves. That
136 * has been handled earlier when unmapping all the memory regions.
137 */
9e1b32ca
BH
138static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd,
139 unsigned long addr)
1da177e4 140{
2f569afd 141 pgtable_t token = pmd_pgtable(*pmd);
e0da382c 142 pmd_clear(pmd);
9e1b32ca 143 pte_free_tlb(tlb, token, addr);
e0da382c 144 tlb->mm->nr_ptes--;
1da177e4
LT
145}
146
e0da382c
HD
147static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
148 unsigned long addr, unsigned long end,
149 unsigned long floor, unsigned long ceiling)
1da177e4
LT
150{
151 pmd_t *pmd;
152 unsigned long next;
e0da382c 153 unsigned long start;
1da177e4 154
e0da382c 155 start = addr;
1da177e4 156 pmd = pmd_offset(pud, addr);
1da177e4
LT
157 do {
158 next = pmd_addr_end(addr, end);
159 if (pmd_none_or_clear_bad(pmd))
160 continue;
9e1b32ca 161 free_pte_range(tlb, pmd, addr);
1da177e4
LT
162 } while (pmd++, addr = next, addr != end);
163
e0da382c
HD
164 start &= PUD_MASK;
165 if (start < floor)
166 return;
167 if (ceiling) {
168 ceiling &= PUD_MASK;
169 if (!ceiling)
170 return;
1da177e4 171 }
e0da382c
HD
172 if (end - 1 > ceiling - 1)
173 return;
174
175 pmd = pmd_offset(pud, start);
176 pud_clear(pud);
9e1b32ca 177 pmd_free_tlb(tlb, pmd, start);
1da177e4
LT
178}
179
e0da382c
HD
180static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
181 unsigned long addr, unsigned long end,
182 unsigned long floor, unsigned long ceiling)
1da177e4
LT
183{
184 pud_t *pud;
185 unsigned long next;
e0da382c 186 unsigned long start;
1da177e4 187
e0da382c 188 start = addr;
1da177e4 189 pud = pud_offset(pgd, addr);
1da177e4
LT
190 do {
191 next = pud_addr_end(addr, end);
192 if (pud_none_or_clear_bad(pud))
193 continue;
e0da382c 194 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
1da177e4
LT
195 } while (pud++, addr = next, addr != end);
196
e0da382c
HD
197 start &= PGDIR_MASK;
198 if (start < floor)
199 return;
200 if (ceiling) {
201 ceiling &= PGDIR_MASK;
202 if (!ceiling)
203 return;
1da177e4 204 }
e0da382c
HD
205 if (end - 1 > ceiling - 1)
206 return;
207
208 pud = pud_offset(pgd, start);
209 pgd_clear(pgd);
9e1b32ca 210 pud_free_tlb(tlb, pud, start);
1da177e4
LT
211}
212
213/*
e0da382c
HD
214 * This function frees user-level page tables of a process.
215 *
1da177e4
LT
216 * Must be called with pagetable lock held.
217 */
42b77728 218void free_pgd_range(struct mmu_gather *tlb,
e0da382c
HD
219 unsigned long addr, unsigned long end,
220 unsigned long floor, unsigned long ceiling)
1da177e4
LT
221{
222 pgd_t *pgd;
223 unsigned long next;
e0da382c
HD
224 unsigned long start;
225
226 /*
227 * The next few lines have given us lots of grief...
228 *
229 * Why are we testing PMD* at this top level? Because often
230 * there will be no work to do at all, and we'd prefer not to
231 * go all the way down to the bottom just to discover that.
232 *
233 * Why all these "- 1"s? Because 0 represents both the bottom
234 * of the address space and the top of it (using -1 for the
235 * top wouldn't help much: the masks would do the wrong thing).
236 * The rule is that addr 0 and floor 0 refer to the bottom of
237 * the address space, but end 0 and ceiling 0 refer to the top
238 * Comparisons need to use "end - 1" and "ceiling - 1" (though
239 * that end 0 case should be mythical).
240 *
241 * Wherever addr is brought up or ceiling brought down, we must
242 * be careful to reject "the opposite 0" before it confuses the
243 * subsequent tests. But what about where end is brought down
244 * by PMD_SIZE below? no, end can't go down to 0 there.
245 *
246 * Whereas we round start (addr) and ceiling down, by different
247 * masks at different levels, in order to test whether a table
248 * now has no other vmas using it, so can be freed, we don't
249 * bother to round floor or end up - the tests don't need that.
250 */
1da177e4 251
e0da382c
HD
252 addr &= PMD_MASK;
253 if (addr < floor) {
254 addr += PMD_SIZE;
255 if (!addr)
256 return;
257 }
258 if (ceiling) {
259 ceiling &= PMD_MASK;
260 if (!ceiling)
261 return;
262 }
263 if (end - 1 > ceiling - 1)
264 end -= PMD_SIZE;
265 if (addr > end - 1)
266 return;
267
268 start = addr;
42b77728 269 pgd = pgd_offset(tlb->mm, addr);
1da177e4
LT
270 do {
271 next = pgd_addr_end(addr, end);
272 if (pgd_none_or_clear_bad(pgd))
273 continue;
42b77728 274 free_pud_range(tlb, pgd, addr, next, floor, ceiling);
1da177e4 275 } while (pgd++, addr = next, addr != end);
e0da382c
HD
276}
277
42b77728 278void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,
3bf5ee95 279 unsigned long floor, unsigned long ceiling)
e0da382c
HD
280{
281 while (vma) {
282 struct vm_area_struct *next = vma->vm_next;
283 unsigned long addr = vma->vm_start;
284
8f4f8c16
HD
285 /*
286 * Hide vma from rmap and vmtruncate before freeing pgtables
287 */
288 anon_vma_unlink(vma);
289 unlink_file_vma(vma);
290
9da61aef 291 if (is_vm_hugetlb_page(vma)) {
3bf5ee95 292 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
e0da382c 293 floor, next? next->vm_start: ceiling);
3bf5ee95
HD
294 } else {
295 /*
296 * Optimization: gather nearby vmas into one call down
297 */
298 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
4866920b 299 && !is_vm_hugetlb_page(next)) {
3bf5ee95
HD
300 vma = next;
301 next = vma->vm_next;
8f4f8c16
HD
302 anon_vma_unlink(vma);
303 unlink_file_vma(vma);
3bf5ee95
HD
304 }
305 free_pgd_range(tlb, addr, vma->vm_end,
306 floor, next? next->vm_start: ceiling);
307 }
e0da382c
HD
308 vma = next;
309 }
1da177e4
LT
310}
311
1bb3630e 312int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
1da177e4 313{
2f569afd 314 pgtable_t new = pte_alloc_one(mm, address);
1bb3630e
HD
315 if (!new)
316 return -ENOMEM;
317
362a61ad
NP
318 /*
319 * Ensure all pte setup (eg. pte page lock and page clearing) are
320 * visible before the pte is made visible to other CPUs by being
321 * put into page tables.
322 *
323 * The other side of the story is the pointer chasing in the page
324 * table walking code (when walking the page table without locking;
325 * ie. most of the time). Fortunately, these data accesses consist
326 * of a chain of data-dependent loads, meaning most CPUs (alpha
327 * being the notable exception) will already guarantee loads are
328 * seen in-order. See the alpha page table accessors for the
329 * smp_read_barrier_depends() barriers in page table walking code.
330 */
331 smp_wmb(); /* Could be smp_wmb__xxx(before|after)_spin_lock */
332
c74df32c 333 spin_lock(&mm->page_table_lock);
2f569afd 334 if (!pmd_present(*pmd)) { /* Has another populated it ? */
1da177e4 335 mm->nr_ptes++;
1da177e4 336 pmd_populate(mm, pmd, new);
2f569afd 337 new = NULL;
1da177e4 338 }
c74df32c 339 spin_unlock(&mm->page_table_lock);
2f569afd
MS
340 if (new)
341 pte_free(mm, new);
1bb3630e 342 return 0;
1da177e4
LT
343}
344
1bb3630e 345int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
1da177e4 346{
1bb3630e
HD
347 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
348 if (!new)
349 return -ENOMEM;
350
362a61ad
NP
351 smp_wmb(); /* See comment in __pte_alloc */
352
1bb3630e 353 spin_lock(&init_mm.page_table_lock);
2f569afd 354 if (!pmd_present(*pmd)) { /* Has another populated it ? */
1bb3630e 355 pmd_populate_kernel(&init_mm, pmd, new);
2f569afd
MS
356 new = NULL;
357 }
1bb3630e 358 spin_unlock(&init_mm.page_table_lock);
2f569afd
MS
359 if (new)
360 pte_free_kernel(&init_mm, new);
1bb3630e 361 return 0;
1da177e4
LT
362}
363
ae859762
HD
364static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
365{
366 if (file_rss)
367 add_mm_counter(mm, file_rss, file_rss);
368 if (anon_rss)
369 add_mm_counter(mm, anon_rss, anon_rss);
370}
371
b5810039 372/*
6aab341e
LT
373 * This function is called to print an error when a bad pte
374 * is found. For example, we might have a PFN-mapped pte in
375 * a region that doesn't allow it.
b5810039
NP
376 *
377 * The calling function must still handle the error.
378 */
3dc14741
HD
379static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
380 pte_t pte, struct page *page)
b5810039 381{
3dc14741
HD
382 pgd_t *pgd = pgd_offset(vma->vm_mm, addr);
383 pud_t *pud = pud_offset(pgd, addr);
384 pmd_t *pmd = pmd_offset(pud, addr);
385 struct address_space *mapping;
386 pgoff_t index;
d936cf9b
HD
387 static unsigned long resume;
388 static unsigned long nr_shown;
389 static unsigned long nr_unshown;
390
391 /*
392 * Allow a burst of 60 reports, then keep quiet for that minute;
393 * or allow a steady drip of one report per second.
394 */
395 if (nr_shown == 60) {
396 if (time_before(jiffies, resume)) {
397 nr_unshown++;
398 return;
399 }
400 if (nr_unshown) {
1e9e6365
HD
401 printk(KERN_ALERT
402 "BUG: Bad page map: %lu messages suppressed\n",
d936cf9b
HD
403 nr_unshown);
404 nr_unshown = 0;
405 }
406 nr_shown = 0;
407 }
408 if (nr_shown++ == 0)
409 resume = jiffies + 60 * HZ;
3dc14741
HD
410
411 mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL;
412 index = linear_page_index(vma, addr);
413
1e9e6365
HD
414 printk(KERN_ALERT
415 "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n",
3dc14741
HD
416 current->comm,
417 (long long)pte_val(pte), (long long)pmd_val(*pmd));
418 if (page) {
1e9e6365 419 printk(KERN_ALERT
3dc14741
HD
420 "page:%p flags:%p count:%d mapcount:%d mapping:%p index:%lx\n",
421 page, (void *)page->flags, page_count(page),
422 page_mapcount(page), page->mapping, page->index);
423 }
1e9e6365 424 printk(KERN_ALERT
3dc14741
HD
425 "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
426 (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
427 /*
428 * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y
429 */
430 if (vma->vm_ops)
1e9e6365 431 print_symbol(KERN_ALERT "vma->vm_ops->fault: %s\n",
3dc14741
HD
432 (unsigned long)vma->vm_ops->fault);
433 if (vma->vm_file && vma->vm_file->f_op)
1e9e6365 434 print_symbol(KERN_ALERT "vma->vm_file->f_op->mmap: %s\n",
3dc14741 435 (unsigned long)vma->vm_file->f_op->mmap);
b5810039 436 dump_stack();
3dc14741 437 add_taint(TAINT_BAD_PAGE);
b5810039
NP
438}
439
67121172
LT
440static inline int is_cow_mapping(unsigned int flags)
441{
442 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
443}
444
ee498ed7 445/*
7e675137 446 * vm_normal_page -- This function gets the "struct page" associated with a pte.
6aab341e 447 *
7e675137
NP
448 * "Special" mappings do not wish to be associated with a "struct page" (either
449 * it doesn't exist, or it exists but they don't want to touch it). In this
450 * case, NULL is returned here. "Normal" mappings do have a struct page.
b379d790 451 *
7e675137
NP
452 * There are 2 broad cases. Firstly, an architecture may define a pte_special()
453 * pte bit, in which case this function is trivial. Secondly, an architecture
454 * may not have a spare pte bit, which requires a more complicated scheme,
455 * described below.
456 *
457 * A raw VM_PFNMAP mapping (ie. one that is not COWed) is always considered a
458 * special mapping (even if there are underlying and valid "struct pages").
459 * COWed pages of a VM_PFNMAP are always normal.
6aab341e 460 *
b379d790
JH
461 * The way we recognize COWed pages within VM_PFNMAP mappings is through the
462 * rules set up by "remap_pfn_range()": the vma will have the VM_PFNMAP bit
7e675137
NP
463 * set, and the vm_pgoff will point to the first PFN mapped: thus every special
464 * mapping will always honor the rule
6aab341e
LT
465 *
466 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
467 *
7e675137
NP
468 * And for normal mappings this is false.
469 *
470 * This restricts such mappings to be a linear translation from virtual address
471 * to pfn. To get around this restriction, we allow arbitrary mappings so long
472 * as the vma is not a COW mapping; in that case, we know that all ptes are
473 * special (because none can have been COWed).
b379d790 474 *
b379d790 475 *
7e675137 476 * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP.
b379d790
JH
477 *
478 * VM_MIXEDMAP mappings can likewise contain memory with or without "struct
479 * page" backing, however the difference is that _all_ pages with a struct
480 * page (that is, those where pfn_valid is true) are refcounted and considered
481 * normal pages by the VM. The disadvantage is that pages are refcounted
482 * (which can be slower and simply not an option for some PFNMAP users). The
483 * advantage is that we don't have to follow the strict linearity rule of
484 * PFNMAP mappings in order to support COWable mappings.
485 *
ee498ed7 486 */
7e675137
NP
487#ifdef __HAVE_ARCH_PTE_SPECIAL
488# define HAVE_PTE_SPECIAL 1
489#else
490# define HAVE_PTE_SPECIAL 0
491#endif
492struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr,
493 pte_t pte)
ee498ed7 494{
22b31eec 495 unsigned long pfn = pte_pfn(pte);
7e675137
NP
496
497 if (HAVE_PTE_SPECIAL) {
22b31eec
HD
498 if (likely(!pte_special(pte)))
499 goto check_pfn;
500 if (!(vma->vm_flags & (VM_PFNMAP | VM_MIXEDMAP)))
501 print_bad_pte(vma, addr, pte, NULL);
7e675137
NP
502 return NULL;
503 }
504
505 /* !HAVE_PTE_SPECIAL case follows: */
506
b379d790
JH
507 if (unlikely(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))) {
508 if (vma->vm_flags & VM_MIXEDMAP) {
509 if (!pfn_valid(pfn))
510 return NULL;
511 goto out;
512 } else {
7e675137
NP
513 unsigned long off;
514 off = (addr - vma->vm_start) >> PAGE_SHIFT;
b379d790
JH
515 if (pfn == vma->vm_pgoff + off)
516 return NULL;
517 if (!is_cow_mapping(vma->vm_flags))
518 return NULL;
519 }
6aab341e
LT
520 }
521
22b31eec
HD
522check_pfn:
523 if (unlikely(pfn > highest_memmap_pfn)) {
524 print_bad_pte(vma, addr, pte, NULL);
525 return NULL;
526 }
6aab341e
LT
527
528 /*
7e675137 529 * NOTE! We still have PageReserved() pages in the page tables.
7e675137 530 * eg. VDSO mappings can cause them to exist.
6aab341e 531 */
b379d790 532out:
6aab341e 533 return pfn_to_page(pfn);
ee498ed7
HD
534}
535
1da177e4
LT
536/*
537 * copy one vm_area from one task to the other. Assumes the page tables
538 * already present in the new task to be cleared in the whole range
539 * covered by this vma.
1da177e4
LT
540 */
541
8c103762 542static inline void
1da177e4 543copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
b5810039 544 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
8c103762 545 unsigned long addr, int *rss)
1da177e4 546{
b5810039 547 unsigned long vm_flags = vma->vm_flags;
1da177e4
LT
548 pte_t pte = *src_pte;
549 struct page *page;
1da177e4
LT
550
551 /* pte contains position in swap or file, so copy. */
552 if (unlikely(!pte_present(pte))) {
553 if (!pte_file(pte)) {
0697212a
CL
554 swp_entry_t entry = pte_to_swp_entry(pte);
555
556 swap_duplicate(entry);
1da177e4
LT
557 /* make sure dst_mm is on swapoff's mmlist. */
558 if (unlikely(list_empty(&dst_mm->mmlist))) {
559 spin_lock(&mmlist_lock);
f412ac08
HD
560 if (list_empty(&dst_mm->mmlist))
561 list_add(&dst_mm->mmlist,
562 &src_mm->mmlist);
1da177e4
LT
563 spin_unlock(&mmlist_lock);
564 }
0697212a
CL
565 if (is_write_migration_entry(entry) &&
566 is_cow_mapping(vm_flags)) {
567 /*
568 * COW mappings require pages in both parent
569 * and child to be set to read.
570 */
571 make_migration_entry_read(&entry);
572 pte = swp_entry_to_pte(entry);
573 set_pte_at(src_mm, addr, src_pte, pte);
574 }
1da177e4 575 }
ae859762 576 goto out_set_pte;
1da177e4
LT
577 }
578
1da177e4
LT
579 /*
580 * If it's a COW mapping, write protect it both
581 * in the parent and the child
582 */
67121172 583 if (is_cow_mapping(vm_flags)) {
1da177e4 584 ptep_set_wrprotect(src_mm, addr, src_pte);
3dc90795 585 pte = pte_wrprotect(pte);
1da177e4
LT
586 }
587
588 /*
589 * If it's a shared mapping, mark it clean in
590 * the child
591 */
592 if (vm_flags & VM_SHARED)
593 pte = pte_mkclean(pte);
594 pte = pte_mkold(pte);
6aab341e
LT
595
596 page = vm_normal_page(vma, addr, pte);
597 if (page) {
598 get_page(page);
c97a9e10 599 page_dup_rmap(page, vma, addr);
6aab341e
LT
600 rss[!!PageAnon(page)]++;
601 }
ae859762
HD
602
603out_set_pte:
604 set_pte_at(dst_mm, addr, dst_pte, pte);
1da177e4
LT
605}
606
607static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
608 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
609 unsigned long addr, unsigned long end)
610{
611 pte_t *src_pte, *dst_pte;
c74df32c 612 spinlock_t *src_ptl, *dst_ptl;
e040f218 613 int progress = 0;
8c103762 614 int rss[2];
1da177e4
LT
615
616again:
ae859762 617 rss[1] = rss[0] = 0;
c74df32c 618 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
1da177e4
LT
619 if (!dst_pte)
620 return -ENOMEM;
621 src_pte = pte_offset_map_nested(src_pmd, addr);
4c21e2f2 622 src_ptl = pte_lockptr(src_mm, src_pmd);
f20dc5f7 623 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
6606c3e0 624 arch_enter_lazy_mmu_mode();
1da177e4 625
1da177e4
LT
626 do {
627 /*
628 * We are holding two locks at this point - either of them
629 * could generate latencies in another task on another CPU.
630 */
e040f218
HD
631 if (progress >= 32) {
632 progress = 0;
633 if (need_resched() ||
95c354fe 634 spin_needbreak(src_ptl) || spin_needbreak(dst_ptl))
e040f218
HD
635 break;
636 }
1da177e4
LT
637 if (pte_none(*src_pte)) {
638 progress++;
639 continue;
640 }
8c103762 641 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
1da177e4
LT
642 progress += 8;
643 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
1da177e4 644
6606c3e0 645 arch_leave_lazy_mmu_mode();
c74df32c 646 spin_unlock(src_ptl);
1da177e4 647 pte_unmap_nested(src_pte - 1);
ae859762 648 add_mm_rss(dst_mm, rss[0], rss[1]);
c74df32c
HD
649 pte_unmap_unlock(dst_pte - 1, dst_ptl);
650 cond_resched();
1da177e4
LT
651 if (addr != end)
652 goto again;
653 return 0;
654}
655
656static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
657 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
658 unsigned long addr, unsigned long end)
659{
660 pmd_t *src_pmd, *dst_pmd;
661 unsigned long next;
662
663 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
664 if (!dst_pmd)
665 return -ENOMEM;
666 src_pmd = pmd_offset(src_pud, addr);
667 do {
668 next = pmd_addr_end(addr, end);
669 if (pmd_none_or_clear_bad(src_pmd))
670 continue;
671 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
672 vma, addr, next))
673 return -ENOMEM;
674 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
675 return 0;
676}
677
678static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
679 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
680 unsigned long addr, unsigned long end)
681{
682 pud_t *src_pud, *dst_pud;
683 unsigned long next;
684
685 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
686 if (!dst_pud)
687 return -ENOMEM;
688 src_pud = pud_offset(src_pgd, addr);
689 do {
690 next = pud_addr_end(addr, end);
691 if (pud_none_or_clear_bad(src_pud))
692 continue;
693 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
694 vma, addr, next))
695 return -ENOMEM;
696 } while (dst_pud++, src_pud++, addr = next, addr != end);
697 return 0;
698}
699
700int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
701 struct vm_area_struct *vma)
702{
703 pgd_t *src_pgd, *dst_pgd;
704 unsigned long next;
705 unsigned long addr = vma->vm_start;
706 unsigned long end = vma->vm_end;
cddb8a5c 707 int ret;
1da177e4 708
d992895b
NP
709 /*
710 * Don't copy ptes where a page fault will fill them correctly.
711 * Fork becomes much lighter when there are big shared or private
712 * readonly mappings. The tradeoff is that copy_page_range is more
713 * efficient than faulting.
714 */
4d7672b4 715 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
d992895b
NP
716 if (!vma->anon_vma)
717 return 0;
718 }
719
1da177e4
LT
720 if (is_vm_hugetlb_page(vma))
721 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
722
34801ba9 723 if (unlikely(is_pfn_mapping(vma))) {
2ab64037 724 /*
725 * We do not free on error cases below as remove_vma
726 * gets called on error from higher level routine
727 */
728 ret = track_pfn_vma_copy(vma);
729 if (ret)
730 return ret;
731 }
732
cddb8a5c
AA
733 /*
734 * We need to invalidate the secondary MMU mappings only when
735 * there could be a permission downgrade on the ptes of the
736 * parent mm. And a permission downgrade will only happen if
737 * is_cow_mapping() returns true.
738 */
739 if (is_cow_mapping(vma->vm_flags))
740 mmu_notifier_invalidate_range_start(src_mm, addr, end);
741
742 ret = 0;
1da177e4
LT
743 dst_pgd = pgd_offset(dst_mm, addr);
744 src_pgd = pgd_offset(src_mm, addr);
745 do {
746 next = pgd_addr_end(addr, end);
747 if (pgd_none_or_clear_bad(src_pgd))
748 continue;
cddb8a5c
AA
749 if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
750 vma, addr, next))) {
751 ret = -ENOMEM;
752 break;
753 }
1da177e4 754 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
cddb8a5c
AA
755
756 if (is_cow_mapping(vma->vm_flags))
757 mmu_notifier_invalidate_range_end(src_mm,
758 vma->vm_start, end);
759 return ret;
1da177e4
LT
760}
761
51c6f666 762static unsigned long zap_pte_range(struct mmu_gather *tlb,
b5810039 763 struct vm_area_struct *vma, pmd_t *pmd,
1da177e4 764 unsigned long addr, unsigned long end,
51c6f666 765 long *zap_work, struct zap_details *details)
1da177e4 766{
b5810039 767 struct mm_struct *mm = tlb->mm;
1da177e4 768 pte_t *pte;
508034a3 769 spinlock_t *ptl;
ae859762
HD
770 int file_rss = 0;
771 int anon_rss = 0;
1da177e4 772
508034a3 773 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
6606c3e0 774 arch_enter_lazy_mmu_mode();
1da177e4
LT
775 do {
776 pte_t ptent = *pte;
51c6f666
RH
777 if (pte_none(ptent)) {
778 (*zap_work)--;
1da177e4 779 continue;
51c6f666 780 }
6f5e6b9e
HD
781
782 (*zap_work) -= PAGE_SIZE;
783
1da177e4 784 if (pte_present(ptent)) {
ee498ed7 785 struct page *page;
51c6f666 786
6aab341e 787 page = vm_normal_page(vma, addr, ptent);
1da177e4
LT
788 if (unlikely(details) && page) {
789 /*
790 * unmap_shared_mapping_pages() wants to
791 * invalidate cache without truncating:
792 * unmap shared but keep private pages.
793 */
794 if (details->check_mapping &&
795 details->check_mapping != page->mapping)
796 continue;
797 /*
798 * Each page->index must be checked when
799 * invalidating or truncating nonlinear.
800 */
801 if (details->nonlinear_vma &&
802 (page->index < details->first_index ||
803 page->index > details->last_index))
804 continue;
805 }
b5810039 806 ptent = ptep_get_and_clear_full(mm, addr, pte,
a600388d 807 tlb->fullmm);
1da177e4
LT
808 tlb_remove_tlb_entry(tlb, pte, addr);
809 if (unlikely(!page))
810 continue;
811 if (unlikely(details) && details->nonlinear_vma
812 && linear_page_index(details->nonlinear_vma,
813 addr) != page->index)
b5810039 814 set_pte_at(mm, addr, pte,
1da177e4 815 pgoff_to_pte(page->index));
1da177e4 816 if (PageAnon(page))
86d912f4 817 anon_rss--;
6237bcd9
HD
818 else {
819 if (pte_dirty(ptent))
820 set_page_dirty(page);
4917e5d0
JW
821 if (pte_young(ptent) &&
822 likely(!VM_SequentialReadHint(vma)))
bf3f3bc5 823 mark_page_accessed(page);
86d912f4 824 file_rss--;
6237bcd9 825 }
edc315fd 826 page_remove_rmap(page);
3dc14741
HD
827 if (unlikely(page_mapcount(page) < 0))
828 print_bad_pte(vma, addr, ptent, page);
1da177e4
LT
829 tlb_remove_page(tlb, page);
830 continue;
831 }
832 /*
833 * If details->check_mapping, we leave swap entries;
834 * if details->nonlinear_vma, we leave file entries.
835 */
836 if (unlikely(details))
837 continue;
2509ef26
HD
838 if (pte_file(ptent)) {
839 if (unlikely(!(vma->vm_flags & VM_NONLINEAR)))
840 print_bad_pte(vma, addr, ptent, NULL);
841 } else if
842 (unlikely(!free_swap_and_cache(pte_to_swp_entry(ptent))))
843 print_bad_pte(vma, addr, ptent, NULL);
9888a1ca 844 pte_clear_not_present_full(mm, addr, pte, tlb->fullmm);
51c6f666 845 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
ae859762 846
86d912f4 847 add_mm_rss(mm, file_rss, anon_rss);
6606c3e0 848 arch_leave_lazy_mmu_mode();
508034a3 849 pte_unmap_unlock(pte - 1, ptl);
51c6f666
RH
850
851 return addr;
1da177e4
LT
852}
853
51c6f666 854static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
b5810039 855 struct vm_area_struct *vma, pud_t *pud,
1da177e4 856 unsigned long addr, unsigned long end,
51c6f666 857 long *zap_work, struct zap_details *details)
1da177e4
LT
858{
859 pmd_t *pmd;
860 unsigned long next;
861
862 pmd = pmd_offset(pud, addr);
863 do {
864 next = pmd_addr_end(addr, end);
51c6f666
RH
865 if (pmd_none_or_clear_bad(pmd)) {
866 (*zap_work)--;
1da177e4 867 continue;
51c6f666
RH
868 }
869 next = zap_pte_range(tlb, vma, pmd, addr, next,
870 zap_work, details);
871 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
872
873 return addr;
1da177e4
LT
874}
875
51c6f666 876static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
b5810039 877 struct vm_area_struct *vma, pgd_t *pgd,
1da177e4 878 unsigned long addr, unsigned long end,
51c6f666 879 long *zap_work, struct zap_details *details)
1da177e4
LT
880{
881 pud_t *pud;
882 unsigned long next;
883
884 pud = pud_offset(pgd, addr);
885 do {
886 next = pud_addr_end(addr, end);
51c6f666
RH
887 if (pud_none_or_clear_bad(pud)) {
888 (*zap_work)--;
1da177e4 889 continue;
51c6f666
RH
890 }
891 next = zap_pmd_range(tlb, vma, pud, addr, next,
892 zap_work, details);
893 } while (pud++, addr = next, (addr != end && *zap_work > 0));
894
895 return addr;
1da177e4
LT
896}
897
51c6f666
RH
898static unsigned long unmap_page_range(struct mmu_gather *tlb,
899 struct vm_area_struct *vma,
1da177e4 900 unsigned long addr, unsigned long end,
51c6f666 901 long *zap_work, struct zap_details *details)
1da177e4
LT
902{
903 pgd_t *pgd;
904 unsigned long next;
905
906 if (details && !details->check_mapping && !details->nonlinear_vma)
907 details = NULL;
908
909 BUG_ON(addr >= end);
910 tlb_start_vma(tlb, vma);
911 pgd = pgd_offset(vma->vm_mm, addr);
912 do {
913 next = pgd_addr_end(addr, end);
51c6f666
RH
914 if (pgd_none_or_clear_bad(pgd)) {
915 (*zap_work)--;
1da177e4 916 continue;
51c6f666
RH
917 }
918 next = zap_pud_range(tlb, vma, pgd, addr, next,
919 zap_work, details);
920 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
1da177e4 921 tlb_end_vma(tlb, vma);
51c6f666
RH
922
923 return addr;
1da177e4
LT
924}
925
926#ifdef CONFIG_PREEMPT
927# define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
928#else
929/* No preempt: go for improved straight-line efficiency */
930# define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
931#endif
932
933/**
934 * unmap_vmas - unmap a range of memory covered by a list of vma's
935 * @tlbp: address of the caller's struct mmu_gather
1da177e4
LT
936 * @vma: the starting vma
937 * @start_addr: virtual address at which to start unmapping
938 * @end_addr: virtual address at which to end unmapping
939 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
940 * @details: details of nonlinear truncation or shared cache invalidation
941 *
ee39b37b 942 * Returns the end address of the unmapping (restart addr if interrupted).
1da177e4 943 *
508034a3 944 * Unmap all pages in the vma list.
1da177e4 945 *
508034a3
HD
946 * We aim to not hold locks for too long (for scheduling latency reasons).
947 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
1da177e4
LT
948 * return the ending mmu_gather to the caller.
949 *
950 * Only addresses between `start' and `end' will be unmapped.
951 *
952 * The VMA list must be sorted in ascending virtual address order.
953 *
954 * unmap_vmas() assumes that the caller will flush the whole unmapped address
955 * range after unmap_vmas() returns. So the only responsibility here is to
956 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
957 * drops the lock and schedules.
958 */
508034a3 959unsigned long unmap_vmas(struct mmu_gather **tlbp,
1da177e4
LT
960 struct vm_area_struct *vma, unsigned long start_addr,
961 unsigned long end_addr, unsigned long *nr_accounted,
962 struct zap_details *details)
963{
51c6f666 964 long zap_work = ZAP_BLOCK_SIZE;
1da177e4
LT
965 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
966 int tlb_start_valid = 0;
ee39b37b 967 unsigned long start = start_addr;
1da177e4 968 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
4d6ddfa9 969 int fullmm = (*tlbp)->fullmm;
cddb8a5c 970 struct mm_struct *mm = vma->vm_mm;
1da177e4 971
cddb8a5c 972 mmu_notifier_invalidate_range_start(mm, start_addr, end_addr);
1da177e4 973 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
1da177e4
LT
974 unsigned long end;
975
976 start = max(vma->vm_start, start_addr);
977 if (start >= vma->vm_end)
978 continue;
979 end = min(vma->vm_end, end_addr);
980 if (end <= vma->vm_start)
981 continue;
982
983 if (vma->vm_flags & VM_ACCOUNT)
984 *nr_accounted += (end - start) >> PAGE_SHIFT;
985
34801ba9 986 if (unlikely(is_pfn_mapping(vma)))
2ab64037 987 untrack_pfn_vma(vma, 0, 0);
988
1da177e4 989 while (start != end) {
1da177e4
LT
990 if (!tlb_start_valid) {
991 tlb_start = start;
992 tlb_start_valid = 1;
993 }
994
51c6f666 995 if (unlikely(is_vm_hugetlb_page(vma))) {
a137e1cc
AK
996 /*
997 * It is undesirable to test vma->vm_file as it
998 * should be non-null for valid hugetlb area.
999 * However, vm_file will be NULL in the error
1000 * cleanup path of do_mmap_pgoff. When
1001 * hugetlbfs ->mmap method fails,
1002 * do_mmap_pgoff() nullifies vma->vm_file
1003 * before calling this function to clean up.
1004 * Since no pte has actually been setup, it is
1005 * safe to do nothing in this case.
1006 */
1007 if (vma->vm_file) {
1008 unmap_hugepage_range(vma, start, end, NULL);
1009 zap_work -= (end - start) /
a5516438 1010 pages_per_huge_page(hstate_vma(vma));
a137e1cc
AK
1011 }
1012
51c6f666
RH
1013 start = end;
1014 } else
1015 start = unmap_page_range(*tlbp, vma,
1016 start, end, &zap_work, details);
1017
1018 if (zap_work > 0) {
1019 BUG_ON(start != end);
1020 break;
1da177e4
LT
1021 }
1022
1da177e4
LT
1023 tlb_finish_mmu(*tlbp, tlb_start, start);
1024
1025 if (need_resched() ||
95c354fe 1026 (i_mmap_lock && spin_needbreak(i_mmap_lock))) {
1da177e4 1027 if (i_mmap_lock) {
508034a3 1028 *tlbp = NULL;
1da177e4
LT
1029 goto out;
1030 }
1da177e4 1031 cond_resched();
1da177e4
LT
1032 }
1033
508034a3 1034 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
1da177e4 1035 tlb_start_valid = 0;
51c6f666 1036 zap_work = ZAP_BLOCK_SIZE;
1da177e4
LT
1037 }
1038 }
1039out:
cddb8a5c 1040 mmu_notifier_invalidate_range_end(mm, start_addr, end_addr);
ee39b37b 1041 return start; /* which is now the end (or restart) address */
1da177e4
LT
1042}
1043
1044/**
1045 * zap_page_range - remove user pages in a given range
1046 * @vma: vm_area_struct holding the applicable pages
1047 * @address: starting address of pages to zap
1048 * @size: number of bytes to zap
1049 * @details: details of nonlinear truncation or shared cache invalidation
1050 */
ee39b37b 1051unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
1da177e4
LT
1052 unsigned long size, struct zap_details *details)
1053{
1054 struct mm_struct *mm = vma->vm_mm;
1055 struct mmu_gather *tlb;
1056 unsigned long end = address + size;
1057 unsigned long nr_accounted = 0;
1058
1da177e4 1059 lru_add_drain();
1da177e4 1060 tlb = tlb_gather_mmu(mm, 0);
365e9c87 1061 update_hiwater_rss(mm);
508034a3
HD
1062 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
1063 if (tlb)
1064 tlb_finish_mmu(tlb, address, end);
ee39b37b 1065 return end;
1da177e4
LT
1066}
1067
c627f9cc
JS
1068/**
1069 * zap_vma_ptes - remove ptes mapping the vma
1070 * @vma: vm_area_struct holding ptes to be zapped
1071 * @address: starting address of pages to zap
1072 * @size: number of bytes to zap
1073 *
1074 * This function only unmaps ptes assigned to VM_PFNMAP vmas.
1075 *
1076 * The entire address range must be fully contained within the vma.
1077 *
1078 * Returns 0 if successful.
1079 */
1080int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address,
1081 unsigned long size)
1082{
1083 if (address < vma->vm_start || address + size > vma->vm_end ||
1084 !(vma->vm_flags & VM_PFNMAP))
1085 return -1;
1086 zap_page_range(vma, address, size, NULL);
1087 return 0;
1088}
1089EXPORT_SYMBOL_GPL(zap_vma_ptes);
1090
1da177e4
LT
1091/*
1092 * Do a quick page-table lookup for a single page.
1da177e4 1093 */
6aab341e 1094struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
deceb6cd 1095 unsigned int flags)
1da177e4
LT
1096{
1097 pgd_t *pgd;
1098 pud_t *pud;
1099 pmd_t *pmd;
1100 pte_t *ptep, pte;
deceb6cd 1101 spinlock_t *ptl;
1da177e4 1102 struct page *page;
6aab341e 1103 struct mm_struct *mm = vma->vm_mm;
1da177e4 1104
deceb6cd
HD
1105 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
1106 if (!IS_ERR(page)) {
1107 BUG_ON(flags & FOLL_GET);
1108 goto out;
1109 }
1da177e4 1110
deceb6cd 1111 page = NULL;
1da177e4
LT
1112 pgd = pgd_offset(mm, address);
1113 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
deceb6cd 1114 goto no_page_table;
1da177e4
LT
1115
1116 pud = pud_offset(pgd, address);
ceb86879 1117 if (pud_none(*pud))
deceb6cd 1118 goto no_page_table;
ceb86879
AK
1119 if (pud_huge(*pud)) {
1120 BUG_ON(flags & FOLL_GET);
1121 page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE);
1122 goto out;
1123 }
1124 if (unlikely(pud_bad(*pud)))
1125 goto no_page_table;
1126
1da177e4 1127 pmd = pmd_offset(pud, address);
aeed5fce 1128 if (pmd_none(*pmd))
deceb6cd 1129 goto no_page_table;
deceb6cd
HD
1130 if (pmd_huge(*pmd)) {
1131 BUG_ON(flags & FOLL_GET);
1132 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
1da177e4 1133 goto out;
deceb6cd 1134 }
aeed5fce
HD
1135 if (unlikely(pmd_bad(*pmd)))
1136 goto no_page_table;
1137
deceb6cd 1138 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
1da177e4
LT
1139
1140 pte = *ptep;
deceb6cd 1141 if (!pte_present(pte))
89f5b7da 1142 goto no_page;
deceb6cd
HD
1143 if ((flags & FOLL_WRITE) && !pte_write(pte))
1144 goto unlock;
6aab341e
LT
1145 page = vm_normal_page(vma, address, pte);
1146 if (unlikely(!page))
89f5b7da 1147 goto bad_page;
1da177e4 1148
deceb6cd
HD
1149 if (flags & FOLL_GET)
1150 get_page(page);
1151 if (flags & FOLL_TOUCH) {
1152 if ((flags & FOLL_WRITE) &&
1153 !pte_dirty(pte) && !PageDirty(page))
1154 set_page_dirty(page);
bd775c42
KM
1155 /*
1156 * pte_mkyoung() would be more correct here, but atomic care
1157 * is needed to avoid losing the dirty bit: it is easier to use
1158 * mark_page_accessed().
1159 */
deceb6cd
HD
1160 mark_page_accessed(page);
1161 }
1162unlock:
1163 pte_unmap_unlock(ptep, ptl);
1da177e4 1164out:
deceb6cd 1165 return page;
1da177e4 1166
89f5b7da
LT
1167bad_page:
1168 pte_unmap_unlock(ptep, ptl);
1169 return ERR_PTR(-EFAULT);
1170
1171no_page:
1172 pte_unmap_unlock(ptep, ptl);
1173 if (!pte_none(pte))
1174 return page;
1175 /* Fall through to ZERO_PAGE handling */
deceb6cd
HD
1176no_page_table:
1177 /*
1178 * When core dumping an enormous anonymous area that nobody
1179 * has touched so far, we don't want to allocate page tables.
1180 */
1181 if (flags & FOLL_ANON) {
557ed1fa 1182 page = ZERO_PAGE(0);
deceb6cd
HD
1183 if (flags & FOLL_GET)
1184 get_page(page);
1185 BUG_ON(flags & FOLL_WRITE);
1186 }
1187 return page;
1da177e4
LT
1188}
1189
672ca28e
LT
1190/* Can we do the FOLL_ANON optimization? */
1191static inline int use_zero_page(struct vm_area_struct *vma)
1192{
1193 /*
1194 * We don't want to optimize FOLL_ANON for make_pages_present()
1195 * when it tries to page in a VM_LOCKED region. As to VM_SHARED,
1196 * we want to get the page from the page tables to make sure
1197 * that we serialize and update with any other user of that
1198 * mapping.
1199 */
1200 if (vma->vm_flags & (VM_LOCKED | VM_SHARED))
1201 return 0;
1202 /*
0d71d10a 1203 * And if we have a fault routine, it's not an anonymous region.
672ca28e 1204 */
0d71d10a 1205 return !vma->vm_ops || !vma->vm_ops->fault;
672ca28e
LT
1206}
1207
b291f000
NP
1208
1209
1210int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
9d73777e
PZ
1211 unsigned long start, int nr_pages, int flags,
1212 struct page **pages, struct vm_area_struct **vmas)
1da177e4
LT
1213{
1214 int i;
b291f000
NP
1215 unsigned int vm_flags = 0;
1216 int write = !!(flags & GUP_FLAGS_WRITE);
1217 int force = !!(flags & GUP_FLAGS_FORCE);
1218 int ignore = !!(flags & GUP_FLAGS_IGNORE_VMA_PERMISSIONS);
4779280d 1219 int ignore_sigkill = !!(flags & GUP_FLAGS_IGNORE_SIGKILL);
1da177e4 1220
9d73777e 1221 if (nr_pages <= 0)
900cf086 1222 return 0;
1da177e4
LT
1223 /*
1224 * Require read or write permissions.
1225 * If 'force' is set, we only require the "MAY" flags.
1226 */
deceb6cd
HD
1227 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
1228 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
1da177e4
LT
1229 i = 0;
1230
1231 do {
deceb6cd
HD
1232 struct vm_area_struct *vma;
1233 unsigned int foll_flags;
1da177e4
LT
1234
1235 vma = find_extend_vma(mm, start);
1236 if (!vma && in_gate_area(tsk, start)) {
1237 unsigned long pg = start & PAGE_MASK;
1238 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
1239 pgd_t *pgd;
1240 pud_t *pud;
1241 pmd_t *pmd;
1242 pte_t *pte;
b291f000
NP
1243
1244 /* user gate pages are read-only */
1245 if (!ignore && write)
1da177e4
LT
1246 return i ? : -EFAULT;
1247 if (pg > TASK_SIZE)
1248 pgd = pgd_offset_k(pg);
1249 else
1250 pgd = pgd_offset_gate(mm, pg);
1251 BUG_ON(pgd_none(*pgd));
1252 pud = pud_offset(pgd, pg);
1253 BUG_ON(pud_none(*pud));
1254 pmd = pmd_offset(pud, pg);
690dbe1c
HD
1255 if (pmd_none(*pmd))
1256 return i ? : -EFAULT;
1da177e4 1257 pte = pte_offset_map(pmd, pg);
690dbe1c
HD
1258 if (pte_none(*pte)) {
1259 pte_unmap(pte);
1260 return i ? : -EFAULT;
1261 }
1da177e4 1262 if (pages) {
fa2a455b 1263 struct page *page = vm_normal_page(gate_vma, start, *pte);
6aab341e
LT
1264 pages[i] = page;
1265 if (page)
1266 get_page(page);
1da177e4
LT
1267 }
1268 pte_unmap(pte);
1269 if (vmas)
1270 vmas[i] = gate_vma;
1271 i++;
1272 start += PAGE_SIZE;
9d73777e 1273 nr_pages--;
1da177e4
LT
1274 continue;
1275 }
1276
b291f000
NP
1277 if (!vma ||
1278 (vma->vm_flags & (VM_IO | VM_PFNMAP)) ||
1279 (!ignore && !(vm_flags & vma->vm_flags)))
1da177e4
LT
1280 return i ? : -EFAULT;
1281
1282 if (is_vm_hugetlb_page(vma)) {
1283 i = follow_hugetlb_page(mm, vma, pages, vmas,
9d73777e 1284 &start, &nr_pages, i, write);
1da177e4
LT
1285 continue;
1286 }
deceb6cd
HD
1287
1288 foll_flags = FOLL_TOUCH;
1289 if (pages)
1290 foll_flags |= FOLL_GET;
672ca28e 1291 if (!write && use_zero_page(vma))
deceb6cd
HD
1292 foll_flags |= FOLL_ANON;
1293
1da177e4 1294 do {
08ef4729 1295 struct page *page;
1da177e4 1296
462e00cc 1297 /*
4779280d
YH
1298 * If we have a pending SIGKILL, don't keep faulting
1299 * pages and potentially allocating memory, unless
1300 * current is handling munlock--e.g., on exit. In
1301 * that case, we are not allocating memory. Rather,
1302 * we're only unlocking already resident/mapped pages.
462e00cc 1303 */
4779280d
YH
1304 if (unlikely(!ignore_sigkill &&
1305 fatal_signal_pending(current)))
1306 return i ? i : -ERESTARTSYS;
462e00cc 1307
deceb6cd
HD
1308 if (write)
1309 foll_flags |= FOLL_WRITE;
a68d2ebc 1310
deceb6cd 1311 cond_resched();
6aab341e 1312 while (!(page = follow_page(vma, start, foll_flags))) {
deceb6cd 1313 int ret;
d06063cc 1314
d26ed650
HD
1315 ret = handle_mm_fault(mm, vma, start,
1316 (foll_flags & FOLL_WRITE) ?
1317 FAULT_FLAG_WRITE : 0);
1318
83c54070
NP
1319 if (ret & VM_FAULT_ERROR) {
1320 if (ret & VM_FAULT_OOM)
1321 return i ? i : -ENOMEM;
1322 else if (ret & VM_FAULT_SIGBUS)
1323 return i ? i : -EFAULT;
1324 BUG();
1325 }
1326 if (ret & VM_FAULT_MAJOR)
1327 tsk->maj_flt++;
1328 else
1329 tsk->min_flt++;
1330
a68d2ebc 1331 /*
83c54070
NP
1332 * The VM_FAULT_WRITE bit tells us that
1333 * do_wp_page has broken COW when necessary,
1334 * even if maybe_mkwrite decided not to set
1335 * pte_write. We can thus safely do subsequent
878b63ac
HD
1336 * page lookups as if they were reads. But only
1337 * do so when looping for pte_write is futile:
1338 * in some cases userspace may also be wanting
1339 * to write to the gotten user page, which a
1340 * read fault here might prevent (a readonly
1341 * page might get reCOWed by userspace write).
a68d2ebc 1342 */
878b63ac
HD
1343 if ((ret & VM_FAULT_WRITE) &&
1344 !(vma->vm_flags & VM_WRITE))
deceb6cd 1345 foll_flags &= ~FOLL_WRITE;
83c54070 1346
7f7bbbe5 1347 cond_resched();
1da177e4 1348 }
89f5b7da
LT
1349 if (IS_ERR(page))
1350 return i ? i : PTR_ERR(page);
1da177e4 1351 if (pages) {
08ef4729 1352 pages[i] = page;
03beb076 1353
a6f36be3 1354 flush_anon_page(vma, page, start);
08ef4729 1355 flush_dcache_page(page);
1da177e4
LT
1356 }
1357 if (vmas)
1358 vmas[i] = vma;
1359 i++;
1360 start += PAGE_SIZE;
9d73777e
PZ
1361 nr_pages--;
1362 } while (nr_pages && start < vma->vm_end);
1363 } while (nr_pages);
1da177e4
LT
1364 return i;
1365}
b291f000 1366
d2bf6be8
NP
1367/**
1368 * get_user_pages() - pin user pages in memory
1369 * @tsk: task_struct of target task
1370 * @mm: mm_struct of target mm
1371 * @start: starting user address
9d73777e 1372 * @nr_pages: number of pages from start to pin
d2bf6be8
NP
1373 * @write: whether pages will be written to by the caller
1374 * @force: whether to force write access even if user mapping is
1375 * readonly. This will result in the page being COWed even
1376 * in MAP_SHARED mappings. You do not want this.
1377 * @pages: array that receives pointers to the pages pinned.
1378 * Should be at least nr_pages long. Or NULL, if caller
1379 * only intends to ensure the pages are faulted in.
1380 * @vmas: array of pointers to vmas corresponding to each page.
1381 * Or NULL if the caller does not require them.
1382 *
1383 * Returns number of pages pinned. This may be fewer than the number
9d73777e 1384 * requested. If nr_pages is 0 or negative, returns 0. If no pages
d2bf6be8
NP
1385 * were pinned, returns -errno. Each page returned must be released
1386 * with a put_page() call when it is finished with. vmas will only
1387 * remain valid while mmap_sem is held.
1388 *
1389 * Must be called with mmap_sem held for read or write.
1390 *
1391 * get_user_pages walks a process's page tables and takes a reference to
1392 * each struct page that each user address corresponds to at a given
1393 * instant. That is, it takes the page that would be accessed if a user
1394 * thread accesses the given user virtual address at that instant.
1395 *
1396 * This does not guarantee that the page exists in the user mappings when
1397 * get_user_pages returns, and there may even be a completely different
1398 * page there in some cases (eg. if mmapped pagecache has been invalidated
1399 * and subsequently re faulted). However it does guarantee that the page
1400 * won't be freed completely. And mostly callers simply care that the page
1401 * contains data that was valid *at some point in time*. Typically, an IO
1402 * or similar operation cannot guarantee anything stronger anyway because
1403 * locks can't be held over the syscall boundary.
1404 *
1405 * If write=0, the page must not be written to. If the page is written to,
1406 * set_page_dirty (or set_page_dirty_lock, as appropriate) must be called
1407 * after the page is finished with, and before put_page is called.
1408 *
1409 * get_user_pages is typically used for fewer-copy IO operations, to get a
1410 * handle on the memory by some means other than accesses via the user virtual
1411 * addresses. The pages may be submitted for DMA to devices or accessed via
1412 * their kernel linear mapping (via the kmap APIs). Care should be taken to
1413 * use the correct cache flushing APIs.
1414 *
1415 * See also get_user_pages_fast, for performance critical applications.
1416 */
b291f000 1417int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
9d73777e 1418 unsigned long start, int nr_pages, int write, int force,
b291f000
NP
1419 struct page **pages, struct vm_area_struct **vmas)
1420{
1421 int flags = 0;
1422
1423 if (write)
1424 flags |= GUP_FLAGS_WRITE;
1425 if (force)
1426 flags |= GUP_FLAGS_FORCE;
1427
9d73777e 1428 return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas);
b291f000
NP
1429}
1430
1da177e4
LT
1431EXPORT_SYMBOL(get_user_pages);
1432
920c7a5d
HH
1433pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr,
1434 spinlock_t **ptl)
c9cfcddf
LT
1435{
1436 pgd_t * pgd = pgd_offset(mm, addr);
1437 pud_t * pud = pud_alloc(mm, pgd, addr);
1438 if (pud) {
49c91fb0 1439 pmd_t * pmd = pmd_alloc(mm, pud, addr);
c9cfcddf
LT
1440 if (pmd)
1441 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1442 }
1443 return NULL;
1444}
1445
238f58d8
LT
1446/*
1447 * This is the old fallback for page remapping.
1448 *
1449 * For historical reasons, it only allows reserved pages. Only
1450 * old drivers should use this, and they needed to mark their
1451 * pages reserved for the old functions anyway.
1452 */
423bad60
NP
1453static int insert_page(struct vm_area_struct *vma, unsigned long addr,
1454 struct page *page, pgprot_t prot)
238f58d8 1455{
423bad60 1456 struct mm_struct *mm = vma->vm_mm;
238f58d8 1457 int retval;
c9cfcddf 1458 pte_t *pte;
8a9f3ccd
BS
1459 spinlock_t *ptl;
1460
238f58d8 1461 retval = -EINVAL;
a145dd41 1462 if (PageAnon(page))
5b4e655e 1463 goto out;
238f58d8
LT
1464 retval = -ENOMEM;
1465 flush_dcache_page(page);
c9cfcddf 1466 pte = get_locked_pte(mm, addr, &ptl);
238f58d8 1467 if (!pte)
5b4e655e 1468 goto out;
238f58d8
LT
1469 retval = -EBUSY;
1470 if (!pte_none(*pte))
1471 goto out_unlock;
1472
1473 /* Ok, finally just insert the thing.. */
1474 get_page(page);
1475 inc_mm_counter(mm, file_rss);
1476 page_add_file_rmap(page);
1477 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1478
1479 retval = 0;
8a9f3ccd
BS
1480 pte_unmap_unlock(pte, ptl);
1481 return retval;
238f58d8
LT
1482out_unlock:
1483 pte_unmap_unlock(pte, ptl);
1484out:
1485 return retval;
1486}
1487
bfa5bf6d
REB
1488/**
1489 * vm_insert_page - insert single page into user vma
1490 * @vma: user vma to map to
1491 * @addr: target user address of this page
1492 * @page: source kernel page
1493 *
a145dd41
LT
1494 * This allows drivers to insert individual pages they've allocated
1495 * into a user vma.
1496 *
1497 * The page has to be a nice clean _individual_ kernel allocation.
1498 * If you allocate a compound page, you need to have marked it as
1499 * such (__GFP_COMP), or manually just split the page up yourself
8dfcc9ba 1500 * (see split_page()).
a145dd41
LT
1501 *
1502 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1503 * took an arbitrary page protection parameter. This doesn't allow
1504 * that. Your vma protection will have to be set up correctly, which
1505 * means that if you want a shared writable mapping, you'd better
1506 * ask for a shared writable mapping!
1507 *
1508 * The page does not need to be reserved.
1509 */
423bad60
NP
1510int vm_insert_page(struct vm_area_struct *vma, unsigned long addr,
1511 struct page *page)
a145dd41
LT
1512{
1513 if (addr < vma->vm_start || addr >= vma->vm_end)
1514 return -EFAULT;
1515 if (!page_count(page))
1516 return -EINVAL;
4d7672b4 1517 vma->vm_flags |= VM_INSERTPAGE;
423bad60 1518 return insert_page(vma, addr, page, vma->vm_page_prot);
a145dd41 1519}
e3c3374f 1520EXPORT_SYMBOL(vm_insert_page);
a145dd41 1521
423bad60
NP
1522static int insert_pfn(struct vm_area_struct *vma, unsigned long addr,
1523 unsigned long pfn, pgprot_t prot)
1524{
1525 struct mm_struct *mm = vma->vm_mm;
1526 int retval;
1527 pte_t *pte, entry;
1528 spinlock_t *ptl;
1529
1530 retval = -ENOMEM;
1531 pte = get_locked_pte(mm, addr, &ptl);
1532 if (!pte)
1533 goto out;
1534 retval = -EBUSY;
1535 if (!pte_none(*pte))
1536 goto out_unlock;
1537
1538 /* Ok, finally just insert the thing.. */
1539 entry = pte_mkspecial(pfn_pte(pfn, prot));
1540 set_pte_at(mm, addr, pte, entry);
1541 update_mmu_cache(vma, addr, entry); /* XXX: why not for insert_page? */
1542
1543 retval = 0;
1544out_unlock:
1545 pte_unmap_unlock(pte, ptl);
1546out:
1547 return retval;
1548}
1549
e0dc0d8f
NP
1550/**
1551 * vm_insert_pfn - insert single pfn into user vma
1552 * @vma: user vma to map to
1553 * @addr: target user address of this page
1554 * @pfn: source kernel pfn
1555 *
1556 * Similar to vm_inert_page, this allows drivers to insert individual pages
1557 * they've allocated into a user vma. Same comments apply.
1558 *
1559 * This function should only be called from a vm_ops->fault handler, and
1560 * in that case the handler should return NULL.
0d71d10a
NP
1561 *
1562 * vma cannot be a COW mapping.
1563 *
1564 * As this is called only for pages that do not currently exist, we
1565 * do not need to flush old virtual caches or the TLB.
e0dc0d8f
NP
1566 */
1567int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
423bad60 1568 unsigned long pfn)
e0dc0d8f 1569{
2ab64037 1570 int ret;
e4b866ed 1571 pgprot_t pgprot = vma->vm_page_prot;
7e675137
NP
1572 /*
1573 * Technically, architectures with pte_special can avoid all these
1574 * restrictions (same for remap_pfn_range). However we would like
1575 * consistency in testing and feature parity among all, so we should
1576 * try to keep these invariants in place for everybody.
1577 */
b379d790
JH
1578 BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)));
1579 BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) ==
1580 (VM_PFNMAP|VM_MIXEDMAP));
1581 BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags));
1582 BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn));
e0dc0d8f 1583
423bad60
NP
1584 if (addr < vma->vm_start || addr >= vma->vm_end)
1585 return -EFAULT;
e4b866ed 1586 if (track_pfn_vma_new(vma, &pgprot, pfn, PAGE_SIZE))
2ab64037 1587 return -EINVAL;
1588
e4b866ed 1589 ret = insert_pfn(vma, addr, pfn, pgprot);
2ab64037 1590
1591 if (ret)
1592 untrack_pfn_vma(vma, pfn, PAGE_SIZE);
1593
1594 return ret;
423bad60
NP
1595}
1596EXPORT_SYMBOL(vm_insert_pfn);
e0dc0d8f 1597
423bad60
NP
1598int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
1599 unsigned long pfn)
1600{
1601 BUG_ON(!(vma->vm_flags & VM_MIXEDMAP));
e0dc0d8f 1602
423bad60
NP
1603 if (addr < vma->vm_start || addr >= vma->vm_end)
1604 return -EFAULT;
e0dc0d8f 1605
423bad60
NP
1606 /*
1607 * If we don't have pte special, then we have to use the pfn_valid()
1608 * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must*
1609 * refcount the page if pfn_valid is true (hence insert_page rather
1610 * than insert_pfn).
1611 */
1612 if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) {
1613 struct page *page;
1614
1615 page = pfn_to_page(pfn);
1616 return insert_page(vma, addr, page, vma->vm_page_prot);
1617 }
1618 return insert_pfn(vma, addr, pfn, vma->vm_page_prot);
e0dc0d8f 1619}
423bad60 1620EXPORT_SYMBOL(vm_insert_mixed);
e0dc0d8f 1621
1da177e4
LT
1622/*
1623 * maps a range of physical memory into the requested pages. the old
1624 * mappings are removed. any references to nonexistent pages results
1625 * in null mappings (currently treated as "copy-on-access")
1626 */
1627static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1628 unsigned long addr, unsigned long end,
1629 unsigned long pfn, pgprot_t prot)
1630{
1631 pte_t *pte;
c74df32c 1632 spinlock_t *ptl;
1da177e4 1633
c74df32c 1634 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
1da177e4
LT
1635 if (!pte)
1636 return -ENOMEM;
6606c3e0 1637 arch_enter_lazy_mmu_mode();
1da177e4
LT
1638 do {
1639 BUG_ON(!pte_none(*pte));
7e675137 1640 set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot)));
1da177e4
LT
1641 pfn++;
1642 } while (pte++, addr += PAGE_SIZE, addr != end);
6606c3e0 1643 arch_leave_lazy_mmu_mode();
c74df32c 1644 pte_unmap_unlock(pte - 1, ptl);
1da177e4
LT
1645 return 0;
1646}
1647
1648static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1649 unsigned long addr, unsigned long end,
1650 unsigned long pfn, pgprot_t prot)
1651{
1652 pmd_t *pmd;
1653 unsigned long next;
1654
1655 pfn -= addr >> PAGE_SHIFT;
1656 pmd = pmd_alloc(mm, pud, addr);
1657 if (!pmd)
1658 return -ENOMEM;
1659 do {
1660 next = pmd_addr_end(addr, end);
1661 if (remap_pte_range(mm, pmd, addr, next,
1662 pfn + (addr >> PAGE_SHIFT), prot))
1663 return -ENOMEM;
1664 } while (pmd++, addr = next, addr != end);
1665 return 0;
1666}
1667
1668static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1669 unsigned long addr, unsigned long end,
1670 unsigned long pfn, pgprot_t prot)
1671{
1672 pud_t *pud;
1673 unsigned long next;
1674
1675 pfn -= addr >> PAGE_SHIFT;
1676 pud = pud_alloc(mm, pgd, addr);
1677 if (!pud)
1678 return -ENOMEM;
1679 do {
1680 next = pud_addr_end(addr, end);
1681 if (remap_pmd_range(mm, pud, addr, next,
1682 pfn + (addr >> PAGE_SHIFT), prot))
1683 return -ENOMEM;
1684 } while (pud++, addr = next, addr != end);
1685 return 0;
1686}
1687
bfa5bf6d
REB
1688/**
1689 * remap_pfn_range - remap kernel memory to userspace
1690 * @vma: user vma to map to
1691 * @addr: target user address to start at
1692 * @pfn: physical address of kernel memory
1693 * @size: size of map area
1694 * @prot: page protection flags for this mapping
1695 *
1696 * Note: this is only safe if the mm semaphore is held when called.
1697 */
1da177e4
LT
1698int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1699 unsigned long pfn, unsigned long size, pgprot_t prot)
1700{
1701 pgd_t *pgd;
1702 unsigned long next;
2d15cab8 1703 unsigned long end = addr + PAGE_ALIGN(size);
1da177e4
LT
1704 struct mm_struct *mm = vma->vm_mm;
1705 int err;
1706
1707 /*
1708 * Physically remapped pages are special. Tell the
1709 * rest of the world about it:
1710 * VM_IO tells people not to look at these pages
1711 * (accesses can have side effects).
0b14c179
HD
1712 * VM_RESERVED is specified all over the place, because
1713 * in 2.4 it kept swapout's vma scan off this vma; but
1714 * in 2.6 the LRU scan won't even find its pages, so this
1715 * flag means no more than count its pages in reserved_vm,
1716 * and omit it from core dump, even when VM_IO turned off.
6aab341e
LT
1717 * VM_PFNMAP tells the core MM that the base pages are just
1718 * raw PFN mappings, and do not have a "struct page" associated
1719 * with them.
fb155c16
LT
1720 *
1721 * There's a horrible special case to handle copy-on-write
1722 * behaviour that some programs depend on. We mark the "original"
1723 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
1da177e4 1724 */
4bb9c5c0 1725 if (addr == vma->vm_start && end == vma->vm_end) {
fb155c16 1726 vma->vm_pgoff = pfn;
895791da 1727 vma->vm_flags |= VM_PFN_AT_MMAP;
4bb9c5c0 1728 } else if (is_cow_mapping(vma->vm_flags))
3c8bb73a 1729 return -EINVAL;
fb155c16 1730
6aab341e 1731 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
1da177e4 1732
e4b866ed 1733 err = track_pfn_vma_new(vma, &prot, pfn, PAGE_ALIGN(size));
a3670613 1734 if (err) {
1735 /*
1736 * To indicate that track_pfn related cleanup is not
1737 * needed from higher level routine calling unmap_vmas
1738 */
1739 vma->vm_flags &= ~(VM_IO | VM_RESERVED | VM_PFNMAP);
895791da 1740 vma->vm_flags &= ~VM_PFN_AT_MMAP;
2ab64037 1741 return -EINVAL;
a3670613 1742 }
2ab64037 1743
1da177e4
LT
1744 BUG_ON(addr >= end);
1745 pfn -= addr >> PAGE_SHIFT;
1746 pgd = pgd_offset(mm, addr);
1747 flush_cache_range(vma, addr, end);
1da177e4
LT
1748 do {
1749 next = pgd_addr_end(addr, end);
1750 err = remap_pud_range(mm, pgd, addr, next,
1751 pfn + (addr >> PAGE_SHIFT), prot);
1752 if (err)
1753 break;
1754 } while (pgd++, addr = next, addr != end);
2ab64037 1755
1756 if (err)
1757 untrack_pfn_vma(vma, pfn, PAGE_ALIGN(size));
1758
1da177e4
LT
1759 return err;
1760}
1761EXPORT_SYMBOL(remap_pfn_range);
1762
aee16b3c
JF
1763static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1764 unsigned long addr, unsigned long end,
1765 pte_fn_t fn, void *data)
1766{
1767 pte_t *pte;
1768 int err;
2f569afd 1769 pgtable_t token;
94909914 1770 spinlock_t *uninitialized_var(ptl);
aee16b3c
JF
1771
1772 pte = (mm == &init_mm) ?
1773 pte_alloc_kernel(pmd, addr) :
1774 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1775 if (!pte)
1776 return -ENOMEM;
1777
1778 BUG_ON(pmd_huge(*pmd));
1779
38e0edb1
JF
1780 arch_enter_lazy_mmu_mode();
1781
2f569afd 1782 token = pmd_pgtable(*pmd);
aee16b3c
JF
1783
1784 do {
2f569afd 1785 err = fn(pte, token, addr, data);
aee16b3c
JF
1786 if (err)
1787 break;
1788 } while (pte++, addr += PAGE_SIZE, addr != end);
1789
38e0edb1
JF
1790 arch_leave_lazy_mmu_mode();
1791
aee16b3c
JF
1792 if (mm != &init_mm)
1793 pte_unmap_unlock(pte-1, ptl);
1794 return err;
1795}
1796
1797static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1798 unsigned long addr, unsigned long end,
1799 pte_fn_t fn, void *data)
1800{
1801 pmd_t *pmd;
1802 unsigned long next;
1803 int err;
1804
ceb86879
AK
1805 BUG_ON(pud_huge(*pud));
1806
aee16b3c
JF
1807 pmd = pmd_alloc(mm, pud, addr);
1808 if (!pmd)
1809 return -ENOMEM;
1810 do {
1811 next = pmd_addr_end(addr, end);
1812 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1813 if (err)
1814 break;
1815 } while (pmd++, addr = next, addr != end);
1816 return err;
1817}
1818
1819static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1820 unsigned long addr, unsigned long end,
1821 pte_fn_t fn, void *data)
1822{
1823 pud_t *pud;
1824 unsigned long next;
1825 int err;
1826
1827 pud = pud_alloc(mm, pgd, addr);
1828 if (!pud)
1829 return -ENOMEM;
1830 do {
1831 next = pud_addr_end(addr, end);
1832 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1833 if (err)
1834 break;
1835 } while (pud++, addr = next, addr != end);
1836 return err;
1837}
1838
1839/*
1840 * Scan a region of virtual memory, filling in page tables as necessary
1841 * and calling a provided function on each leaf page table.
1842 */
1843int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1844 unsigned long size, pte_fn_t fn, void *data)
1845{
1846 pgd_t *pgd;
1847 unsigned long next;
cddb8a5c 1848 unsigned long start = addr, end = addr + size;
aee16b3c
JF
1849 int err;
1850
1851 BUG_ON(addr >= end);
cddb8a5c 1852 mmu_notifier_invalidate_range_start(mm, start, end);
aee16b3c
JF
1853 pgd = pgd_offset(mm, addr);
1854 do {
1855 next = pgd_addr_end(addr, end);
1856 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1857 if (err)
1858 break;
1859 } while (pgd++, addr = next, addr != end);
cddb8a5c 1860 mmu_notifier_invalidate_range_end(mm, start, end);
aee16b3c
JF
1861 return err;
1862}
1863EXPORT_SYMBOL_GPL(apply_to_page_range);
1864
8f4e2101
HD
1865/*
1866 * handle_pte_fault chooses page fault handler according to an entry
1867 * which was read non-atomically. Before making any commitment, on
1868 * those architectures or configurations (e.g. i386 with PAE) which
1869 * might give a mix of unmatched parts, do_swap_page and do_file_page
1870 * must check under lock before unmapping the pte and proceeding
1871 * (but do_wp_page is only called after already making such a check;
1872 * and do_anonymous_page and do_no_page can safely check later on).
1873 */
4c21e2f2 1874static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
8f4e2101
HD
1875 pte_t *page_table, pte_t orig_pte)
1876{
1877 int same = 1;
1878#if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1879 if (sizeof(pte_t) > sizeof(unsigned long)) {
4c21e2f2
HD
1880 spinlock_t *ptl = pte_lockptr(mm, pmd);
1881 spin_lock(ptl);
8f4e2101 1882 same = pte_same(*page_table, orig_pte);
4c21e2f2 1883 spin_unlock(ptl);
8f4e2101
HD
1884 }
1885#endif
1886 pte_unmap(page_table);
1887 return same;
1888}
1889
1da177e4
LT
1890/*
1891 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1892 * servicing faults for write access. In the normal case, do always want
1893 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1894 * that do not have writing enabled, when used by access_process_vm.
1895 */
1896static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1897{
1898 if (likely(vma->vm_flags & VM_WRITE))
1899 pte = pte_mkwrite(pte);
1900 return pte;
1901}
1902
9de455b2 1903static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma)
6aab341e
LT
1904{
1905 /*
1906 * If the source page was a PFN mapping, we don't have
1907 * a "struct page" for it. We do a best-effort copy by
1908 * just copying from the original user address. If that
1909 * fails, we just zero-fill it. Live with it.
1910 */
1911 if (unlikely(!src)) {
1912 void *kaddr = kmap_atomic(dst, KM_USER0);
5d2a2dbb
LT
1913 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1914
1915 /*
1916 * This really shouldn't fail, because the page is there
1917 * in the page tables. But it might just be unreadable,
1918 * in which case we just give up and fill the result with
1919 * zeroes.
1920 */
1921 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
6aab341e
LT
1922 memset(kaddr, 0, PAGE_SIZE);
1923 kunmap_atomic(kaddr, KM_USER0);
c4ec7b0d 1924 flush_dcache_page(dst);
0ed361de
NP
1925 } else
1926 copy_user_highpage(dst, src, va, vma);
6aab341e
LT
1927}
1928
1da177e4
LT
1929/*
1930 * This routine handles present pages, when users try to write
1931 * to a shared page. It is done by copying the page to a new address
1932 * and decrementing the shared-page counter for the old page.
1933 *
1da177e4
LT
1934 * Note that this routine assumes that the protection checks have been
1935 * done by the caller (the low-level page fault routine in most cases).
1936 * Thus we can safely just mark it writable once we've done any necessary
1937 * COW.
1938 *
1939 * We also mark the page dirty at this point even though the page will
1940 * change only once the write actually happens. This avoids a few races,
1941 * and potentially makes it more efficient.
1942 *
8f4e2101
HD
1943 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1944 * but allow concurrent faults), with pte both mapped and locked.
1945 * We return with mmap_sem still held, but pte unmapped and unlocked.
1da177e4 1946 */
65500d23
HD
1947static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1948 unsigned long address, pte_t *page_table, pmd_t *pmd,
8f4e2101 1949 spinlock_t *ptl, pte_t orig_pte)
1da177e4 1950{
e5bbe4df 1951 struct page *old_page, *new_page;
1da177e4 1952 pte_t entry;
83c54070 1953 int reuse = 0, ret = 0;
a200ee18 1954 int page_mkwrite = 0;
d08b3851 1955 struct page *dirty_page = NULL;
1da177e4 1956
6aab341e 1957 old_page = vm_normal_page(vma, address, orig_pte);
251b97f5
PZ
1958 if (!old_page) {
1959 /*
1960 * VM_MIXEDMAP !pfn_valid() case
1961 *
1962 * We should not cow pages in a shared writeable mapping.
1963 * Just mark the pages writable as we can't do any dirty
1964 * accounting on raw pfn maps.
1965 */
1966 if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
1967 (VM_WRITE|VM_SHARED))
1968 goto reuse;
6aab341e 1969 goto gotten;
251b97f5 1970 }
1da177e4 1971
d08b3851 1972 /*
ee6a6457
PZ
1973 * Take out anonymous pages first, anonymous shared vmas are
1974 * not dirty accountable.
d08b3851 1975 */
ee6a6457 1976 if (PageAnon(old_page)) {
ab967d86
HD
1977 if (!trylock_page(old_page)) {
1978 page_cache_get(old_page);
1979 pte_unmap_unlock(page_table, ptl);
1980 lock_page(old_page);
1981 page_table = pte_offset_map_lock(mm, pmd, address,
1982 &ptl);
1983 if (!pte_same(*page_table, orig_pte)) {
1984 unlock_page(old_page);
1985 page_cache_release(old_page);
1986 goto unlock;
1987 }
1988 page_cache_release(old_page);
ee6a6457 1989 }
7b1fe597 1990 reuse = reuse_swap_page(old_page);
ab967d86 1991 unlock_page(old_page);
ee6a6457 1992 } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) ==
d08b3851 1993 (VM_WRITE|VM_SHARED))) {
ee6a6457
PZ
1994 /*
1995 * Only catch write-faults on shared writable pages,
1996 * read-only shared pages can get COWed by
1997 * get_user_pages(.write=1, .force=1).
1998 */
9637a5ef 1999 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
c2ec175c
NP
2000 struct vm_fault vmf;
2001 int tmp;
2002
2003 vmf.virtual_address = (void __user *)(address &
2004 PAGE_MASK);
2005 vmf.pgoff = old_page->index;
2006 vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
2007 vmf.page = old_page;
2008
9637a5ef
DH
2009 /*
2010 * Notify the address space that the page is about to
2011 * become writable so that it can prohibit this or wait
2012 * for the page to get into an appropriate state.
2013 *
2014 * We do this without the lock held, so that it can
2015 * sleep if it needs to.
2016 */
2017 page_cache_get(old_page);
2018 pte_unmap_unlock(page_table, ptl);
2019
c2ec175c
NP
2020 tmp = vma->vm_ops->page_mkwrite(vma, &vmf);
2021 if (unlikely(tmp &
2022 (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
2023 ret = tmp;
9637a5ef 2024 goto unwritable_page;
c2ec175c 2025 }
b827e496
NP
2026 if (unlikely(!(tmp & VM_FAULT_LOCKED))) {
2027 lock_page(old_page);
2028 if (!old_page->mapping) {
2029 ret = 0; /* retry the fault */
2030 unlock_page(old_page);
2031 goto unwritable_page;
2032 }
2033 } else
2034 VM_BUG_ON(!PageLocked(old_page));
9637a5ef 2035
9637a5ef
DH
2036 /*
2037 * Since we dropped the lock we need to revalidate
2038 * the PTE as someone else may have changed it. If
2039 * they did, we just return, as we can count on the
2040 * MMU to tell us if they didn't also make it writable.
2041 */
2042 page_table = pte_offset_map_lock(mm, pmd, address,
2043 &ptl);
b827e496
NP
2044 if (!pte_same(*page_table, orig_pte)) {
2045 unlock_page(old_page);
2046 page_cache_release(old_page);
9637a5ef 2047 goto unlock;
b827e496 2048 }
a200ee18
PZ
2049
2050 page_mkwrite = 1;
1da177e4 2051 }
d08b3851
PZ
2052 dirty_page = old_page;
2053 get_page(dirty_page);
9637a5ef 2054 reuse = 1;
9637a5ef
DH
2055 }
2056
2057 if (reuse) {
251b97f5 2058reuse:
9637a5ef
DH
2059 flush_cache_page(vma, address, pte_pfn(orig_pte));
2060 entry = pte_mkyoung(orig_pte);
2061 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
954ffcb3 2062 if (ptep_set_access_flags(vma, address, page_table, entry,1))
8dab5241 2063 update_mmu_cache(vma, address, entry);
9637a5ef
DH
2064 ret |= VM_FAULT_WRITE;
2065 goto unlock;
1da177e4 2066 }
1da177e4
LT
2067
2068 /*
2069 * Ok, we need to copy. Oh, well..
2070 */
b5810039 2071 page_cache_get(old_page);
920fc356 2072gotten:
8f4e2101 2073 pte_unmap_unlock(page_table, ptl);
1da177e4
LT
2074
2075 if (unlikely(anon_vma_prepare(vma)))
65500d23 2076 goto oom;
557ed1fa
NP
2077 VM_BUG_ON(old_page == ZERO_PAGE(0));
2078 new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address);
2079 if (!new_page)
2080 goto oom;
b291f000
NP
2081 /*
2082 * Don't let another task, with possibly unlocked vma,
2083 * keep the mlocked page.
2084 */
ab92661d 2085 if ((vma->vm_flags & VM_LOCKED) && old_page) {
b291f000
NP
2086 lock_page(old_page); /* for LRU manipulation */
2087 clear_page_mlock(old_page);
2088 unlock_page(old_page);
2089 }
557ed1fa 2090 cow_user_page(new_page, old_page, address, vma);
0ed361de 2091 __SetPageUptodate(new_page);
65500d23 2092
2c26fdd7 2093 if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL))
8a9f3ccd
BS
2094 goto oom_free_new;
2095
1da177e4
LT
2096 /*
2097 * Re-check the pte - we dropped the lock
2098 */
8f4e2101 2099 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
65500d23 2100 if (likely(pte_same(*page_table, orig_pte))) {
920fc356 2101 if (old_page) {
920fc356
HD
2102 if (!PageAnon(old_page)) {
2103 dec_mm_counter(mm, file_rss);
2104 inc_mm_counter(mm, anon_rss);
2105 }
2106 } else
4294621f 2107 inc_mm_counter(mm, anon_rss);
eca35133 2108 flush_cache_page(vma, address, pte_pfn(orig_pte));
65500d23
HD
2109 entry = mk_pte(new_page, vma->vm_page_prot);
2110 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
4ce072f1
SS
2111 /*
2112 * Clear the pte entry and flush it first, before updating the
2113 * pte with the new entry. This will avoid a race condition
2114 * seen in the presence of one thread doing SMC and another
2115 * thread doing COW.
2116 */
cddb8a5c 2117 ptep_clear_flush_notify(vma, address, page_table);
9617d95e 2118 page_add_new_anon_rmap(new_page, vma, address);
64d6519d
LS
2119 set_pte_at(mm, address, page_table, entry);
2120 update_mmu_cache(vma, address, entry);
945754a1
NP
2121 if (old_page) {
2122 /*
2123 * Only after switching the pte to the new page may
2124 * we remove the mapcount here. Otherwise another
2125 * process may come and find the rmap count decremented
2126 * before the pte is switched to the new page, and
2127 * "reuse" the old page writing into it while our pte
2128 * here still points into it and can be read by other
2129 * threads.
2130 *
2131 * The critical issue is to order this
2132 * page_remove_rmap with the ptp_clear_flush above.
2133 * Those stores are ordered by (if nothing else,)
2134 * the barrier present in the atomic_add_negative
2135 * in page_remove_rmap.
2136 *
2137 * Then the TLB flush in ptep_clear_flush ensures that
2138 * no process can access the old page before the
2139 * decremented mapcount is visible. And the old page
2140 * cannot be reused until after the decremented
2141 * mapcount is visible. So transitively, TLBs to
2142 * old page will be flushed before it can be reused.
2143 */
edc315fd 2144 page_remove_rmap(old_page);
945754a1
NP
2145 }
2146
1da177e4
LT
2147 /* Free the old page.. */
2148 new_page = old_page;
f33ea7f4 2149 ret |= VM_FAULT_WRITE;
8a9f3ccd
BS
2150 } else
2151 mem_cgroup_uncharge_page(new_page);
2152
920fc356
HD
2153 if (new_page)
2154 page_cache_release(new_page);
2155 if (old_page)
2156 page_cache_release(old_page);
65500d23 2157unlock:
8f4e2101 2158 pte_unmap_unlock(page_table, ptl);
d08b3851 2159 if (dirty_page) {
79352894
NP
2160 /*
2161 * Yes, Virginia, this is actually required to prevent a race
2162 * with clear_page_dirty_for_io() from clearing the page dirty
2163 * bit after it clear all dirty ptes, but before a racing
2164 * do_wp_page installs a dirty pte.
2165 *
2166 * do_no_page is protected similarly.
2167 */
b827e496
NP
2168 if (!page_mkwrite) {
2169 wait_on_page_locked(dirty_page);
2170 set_page_dirty_balance(dirty_page, page_mkwrite);
2171 }
d08b3851 2172 put_page(dirty_page);
b827e496
NP
2173 if (page_mkwrite) {
2174 struct address_space *mapping = dirty_page->mapping;
2175
2176 set_page_dirty(dirty_page);
2177 unlock_page(dirty_page);
2178 page_cache_release(dirty_page);
2179 if (mapping) {
2180 /*
2181 * Some device drivers do not set page.mapping
2182 * but still dirty their pages
2183 */
2184 balance_dirty_pages_ratelimited(mapping);
2185 }
2186 }
2187
2188 /* file_update_time outside page_lock */
2189 if (vma->vm_file)
2190 file_update_time(vma->vm_file);
d08b3851 2191 }
f33ea7f4 2192 return ret;
8a9f3ccd 2193oom_free_new:
6dbf6d3b 2194 page_cache_release(new_page);
65500d23 2195oom:
b827e496
NP
2196 if (old_page) {
2197 if (page_mkwrite) {
2198 unlock_page(old_page);
2199 page_cache_release(old_page);
2200 }
920fc356 2201 page_cache_release(old_page);
b827e496 2202 }
1da177e4 2203 return VM_FAULT_OOM;
9637a5ef
DH
2204
2205unwritable_page:
2206 page_cache_release(old_page);
c2ec175c 2207 return ret;
1da177e4
LT
2208}
2209
2210/*
2211 * Helper functions for unmap_mapping_range().
2212 *
2213 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
2214 *
2215 * We have to restart searching the prio_tree whenever we drop the lock,
2216 * since the iterator is only valid while the lock is held, and anyway
2217 * a later vma might be split and reinserted earlier while lock dropped.
2218 *
2219 * The list of nonlinear vmas could be handled more efficiently, using
2220 * a placeholder, but handle it in the same way until a need is shown.
2221 * It is important to search the prio_tree before nonlinear list: a vma
2222 * may become nonlinear and be shifted from prio_tree to nonlinear list
2223 * while the lock is dropped; but never shifted from list to prio_tree.
2224 *
2225 * In order to make forward progress despite restarting the search,
2226 * vm_truncate_count is used to mark a vma as now dealt with, so we can
2227 * quickly skip it next time around. Since the prio_tree search only
2228 * shows us those vmas affected by unmapping the range in question, we
2229 * can't efficiently keep all vmas in step with mapping->truncate_count:
2230 * so instead reset them all whenever it wraps back to 0 (then go to 1).
2231 * mapping->truncate_count and vma->vm_truncate_count are protected by
2232 * i_mmap_lock.
2233 *
2234 * In order to make forward progress despite repeatedly restarting some
ee39b37b 2235 * large vma, note the restart_addr from unmap_vmas when it breaks out:
1da177e4
LT
2236 * and restart from that address when we reach that vma again. It might
2237 * have been split or merged, shrunk or extended, but never shifted: so
2238 * restart_addr remains valid so long as it remains in the vma's range.
2239 * unmap_mapping_range forces truncate_count to leap over page-aligned
2240 * values so we can save vma's restart_addr in its truncate_count field.
2241 */
2242#define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
2243
2244static void reset_vma_truncate_counts(struct address_space *mapping)
2245{
2246 struct vm_area_struct *vma;
2247 struct prio_tree_iter iter;
2248
2249 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
2250 vma->vm_truncate_count = 0;
2251 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
2252 vma->vm_truncate_count = 0;
2253}
2254
2255static int unmap_mapping_range_vma(struct vm_area_struct *vma,
2256 unsigned long start_addr, unsigned long end_addr,
2257 struct zap_details *details)
2258{
2259 unsigned long restart_addr;
2260 int need_break;
2261
d00806b1
NP
2262 /*
2263 * files that support invalidating or truncating portions of the
d0217ac0 2264 * file from under mmaped areas must have their ->fault function
83c54070
NP
2265 * return a locked page (and set VM_FAULT_LOCKED in the return).
2266 * This provides synchronisation against concurrent unmapping here.
d00806b1 2267 */
d00806b1 2268
1da177e4
LT
2269again:
2270 restart_addr = vma->vm_truncate_count;
2271 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
2272 start_addr = restart_addr;
2273 if (start_addr >= end_addr) {
2274 /* Top of vma has been split off since last time */
2275 vma->vm_truncate_count = details->truncate_count;
2276 return 0;
2277 }
2278 }
2279
ee39b37b
HD
2280 restart_addr = zap_page_range(vma, start_addr,
2281 end_addr - start_addr, details);
95c354fe 2282 need_break = need_resched() || spin_needbreak(details->i_mmap_lock);
1da177e4 2283
ee39b37b 2284 if (restart_addr >= end_addr) {
1da177e4
LT
2285 /* We have now completed this vma: mark it so */
2286 vma->vm_truncate_count = details->truncate_count;
2287 if (!need_break)
2288 return 0;
2289 } else {
2290 /* Note restart_addr in vma's truncate_count field */
ee39b37b 2291 vma->vm_truncate_count = restart_addr;
1da177e4
LT
2292 if (!need_break)
2293 goto again;
2294 }
2295
2296 spin_unlock(details->i_mmap_lock);
2297 cond_resched();
2298 spin_lock(details->i_mmap_lock);
2299 return -EINTR;
2300}
2301
2302static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
2303 struct zap_details *details)
2304{
2305 struct vm_area_struct *vma;
2306 struct prio_tree_iter iter;
2307 pgoff_t vba, vea, zba, zea;
2308
2309restart:
2310 vma_prio_tree_foreach(vma, &iter, root,
2311 details->first_index, details->last_index) {
2312 /* Skip quickly over those we have already dealt with */
2313 if (vma->vm_truncate_count == details->truncate_count)
2314 continue;
2315
2316 vba = vma->vm_pgoff;
2317 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
2318 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
2319 zba = details->first_index;
2320 if (zba < vba)
2321 zba = vba;
2322 zea = details->last_index;
2323 if (zea > vea)
2324 zea = vea;
2325
2326 if (unmap_mapping_range_vma(vma,
2327 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
2328 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
2329 details) < 0)
2330 goto restart;
2331 }
2332}
2333
2334static inline void unmap_mapping_range_list(struct list_head *head,
2335 struct zap_details *details)
2336{
2337 struct vm_area_struct *vma;
2338
2339 /*
2340 * In nonlinear VMAs there is no correspondence between virtual address
2341 * offset and file offset. So we must perform an exhaustive search
2342 * across *all* the pages in each nonlinear VMA, not just the pages
2343 * whose virtual address lies outside the file truncation point.
2344 */
2345restart:
2346 list_for_each_entry(vma, head, shared.vm_set.list) {
2347 /* Skip quickly over those we have already dealt with */
2348 if (vma->vm_truncate_count == details->truncate_count)
2349 continue;
2350 details->nonlinear_vma = vma;
2351 if (unmap_mapping_range_vma(vma, vma->vm_start,
2352 vma->vm_end, details) < 0)
2353 goto restart;
2354 }
2355}
2356
2357/**
72fd4a35 2358 * unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file.
3d41088f 2359 * @mapping: the address space containing mmaps to be unmapped.
1da177e4
LT
2360 * @holebegin: byte in first page to unmap, relative to the start of
2361 * the underlying file. This will be rounded down to a PAGE_SIZE
2362 * boundary. Note that this is different from vmtruncate(), which
2363 * must keep the partial page. In contrast, we must get rid of
2364 * partial pages.
2365 * @holelen: size of prospective hole in bytes. This will be rounded
2366 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
2367 * end of the file.
2368 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
2369 * but 0 when invalidating pagecache, don't throw away private data.
2370 */
2371void unmap_mapping_range(struct address_space *mapping,
2372 loff_t const holebegin, loff_t const holelen, int even_cows)
2373{
2374 struct zap_details details;
2375 pgoff_t hba = holebegin >> PAGE_SHIFT;
2376 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2377
2378 /* Check for overflow. */
2379 if (sizeof(holelen) > sizeof(hlen)) {
2380 long long holeend =
2381 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
2382 if (holeend & ~(long long)ULONG_MAX)
2383 hlen = ULONG_MAX - hba + 1;
2384 }
2385
2386 details.check_mapping = even_cows? NULL: mapping;
2387 details.nonlinear_vma = NULL;
2388 details.first_index = hba;
2389 details.last_index = hba + hlen - 1;
2390 if (details.last_index < details.first_index)
2391 details.last_index = ULONG_MAX;
2392 details.i_mmap_lock = &mapping->i_mmap_lock;
2393
2394 spin_lock(&mapping->i_mmap_lock);
2395
d00806b1 2396 /* Protect against endless unmapping loops */
1da177e4 2397 mapping->truncate_count++;
1da177e4
LT
2398 if (unlikely(is_restart_addr(mapping->truncate_count))) {
2399 if (mapping->truncate_count == 0)
2400 reset_vma_truncate_counts(mapping);
2401 mapping->truncate_count++;
2402 }
2403 details.truncate_count = mapping->truncate_count;
2404
2405 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
2406 unmap_mapping_range_tree(&mapping->i_mmap, &details);
2407 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
2408 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
2409 spin_unlock(&mapping->i_mmap_lock);
2410}
2411EXPORT_SYMBOL(unmap_mapping_range);
2412
bfa5bf6d
REB
2413/**
2414 * vmtruncate - unmap mappings "freed" by truncate() syscall
2415 * @inode: inode of the file used
2416 * @offset: file offset to start truncating
1da177e4
LT
2417 *
2418 * NOTE! We have to be ready to update the memory sharing
2419 * between the file and the memory map for a potential last
2420 * incomplete page. Ugly, but necessary.
2421 */
2422int vmtruncate(struct inode * inode, loff_t offset)
2423{
61d5048f
CH
2424 if (inode->i_size < offset) {
2425 unsigned long limit;
2426
2427 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
2428 if (limit != RLIM_INFINITY && offset > limit)
2429 goto out_sig;
2430 if (offset > inode->i_sb->s_maxbytes)
2431 goto out_big;
2432 i_size_write(inode, offset);
2433 } else {
2434 struct address_space *mapping = inode->i_mapping;
1da177e4 2435
61d5048f
CH
2436 /*
2437 * truncation of in-use swapfiles is disallowed - it would
2438 * cause subsequent swapout to scribble on the now-freed
2439 * blocks.
2440 */
2441 if (IS_SWAPFILE(inode))
2442 return -ETXTBSY;
2443 i_size_write(inode, offset);
2444
2445 /*
2446 * unmap_mapping_range is called twice, first simply for
2447 * efficiency so that truncate_inode_pages does fewer
2448 * single-page unmaps. However after this first call, and
2449 * before truncate_inode_pages finishes, it is possible for
2450 * private pages to be COWed, which remain after
2451 * truncate_inode_pages finishes, hence the second
2452 * unmap_mapping_range call must be made for correctness.
2453 */
2454 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2455 truncate_inode_pages(mapping, offset);
2456 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
2457 }
d00806b1 2458
acfa4380 2459 if (inode->i_op->truncate)
1da177e4
LT
2460 inode->i_op->truncate(inode);
2461 return 0;
61d5048f 2462
1da177e4
LT
2463out_sig:
2464 send_sig(SIGXFSZ, current, 0);
2465out_big:
2466 return -EFBIG;
1da177e4 2467}
1da177e4
LT
2468EXPORT_SYMBOL(vmtruncate);
2469
f6b3ec23
BP
2470int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
2471{
2472 struct address_space *mapping = inode->i_mapping;
2473
2474 /*
2475 * If the underlying filesystem is not going to provide
2476 * a way to truncate a range of blocks (punch a hole) -
2477 * we should return failure right now.
2478 */
acfa4380 2479 if (!inode->i_op->truncate_range)
f6b3ec23
BP
2480 return -ENOSYS;
2481
1b1dcc1b 2482 mutex_lock(&inode->i_mutex);
f6b3ec23
BP
2483 down_write(&inode->i_alloc_sem);
2484 unmap_mapping_range(mapping, offset, (end - offset), 1);
2485 truncate_inode_pages_range(mapping, offset, end);
d00806b1 2486 unmap_mapping_range(mapping, offset, (end - offset), 1);
f6b3ec23
BP
2487 inode->i_op->truncate_range(inode, offset, end);
2488 up_write(&inode->i_alloc_sem);
1b1dcc1b 2489 mutex_unlock(&inode->i_mutex);
f6b3ec23
BP
2490
2491 return 0;
2492}
f6b3ec23 2493
1da177e4 2494/*
8f4e2101
HD
2495 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2496 * but allow concurrent faults), and pte mapped but not yet locked.
2497 * We return with mmap_sem still held, but pte unmapped and unlocked.
1da177e4 2498 */
65500d23
HD
2499static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2500 unsigned long address, pte_t *page_table, pmd_t *pmd,
30c9f3a9 2501 unsigned int flags, pte_t orig_pte)
1da177e4 2502{
8f4e2101 2503 spinlock_t *ptl;
1da177e4 2504 struct page *page;
65500d23 2505 swp_entry_t entry;
1da177e4 2506 pte_t pte;
7a81b88c 2507 struct mem_cgroup *ptr = NULL;
83c54070 2508 int ret = 0;
1da177e4 2509
4c21e2f2 2510 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
8f4e2101 2511 goto out;
65500d23
HD
2512
2513 entry = pte_to_swp_entry(orig_pte);
0697212a
CL
2514 if (is_migration_entry(entry)) {
2515 migration_entry_wait(mm, pmd, address);
2516 goto out;
2517 }
0ff92245 2518 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
1da177e4
LT
2519 page = lookup_swap_cache(entry);
2520 if (!page) {
a5c9b696 2521 grab_swap_token(mm); /* Contend for token _before_ read-in */
02098fea
HD
2522 page = swapin_readahead(entry,
2523 GFP_HIGHUSER_MOVABLE, vma, address);
1da177e4
LT
2524 if (!page) {
2525 /*
8f4e2101
HD
2526 * Back out if somebody else faulted in this pte
2527 * while we released the pte lock.
1da177e4 2528 */
8f4e2101 2529 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
1da177e4
LT
2530 if (likely(pte_same(*page_table, orig_pte)))
2531 ret = VM_FAULT_OOM;
0ff92245 2532 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
65500d23 2533 goto unlock;
1da177e4
LT
2534 }
2535
2536 /* Had to read the page from swap area: Major fault */
2537 ret = VM_FAULT_MAJOR;
f8891e5e 2538 count_vm_event(PGMAJFAULT);
1da177e4
LT
2539 }
2540
073e587e
KH
2541 lock_page(page);
2542 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2543
2c26fdd7 2544 if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) {
8a9f3ccd 2545 ret = VM_FAULT_OOM;
bc43f75c 2546 goto out_page;
8a9f3ccd
BS
2547 }
2548
1da177e4 2549 /*
8f4e2101 2550 * Back out if somebody else already faulted in this pte.
1da177e4 2551 */
8f4e2101 2552 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
9e9bef07 2553 if (unlikely(!pte_same(*page_table, orig_pte)))
b8107480 2554 goto out_nomap;
b8107480
KK
2555
2556 if (unlikely(!PageUptodate(page))) {
2557 ret = VM_FAULT_SIGBUS;
2558 goto out_nomap;
1da177e4
LT
2559 }
2560
8c7c6e34
KH
2561 /*
2562 * The page isn't present yet, go ahead with the fault.
2563 *
2564 * Be careful about the sequence of operations here.
2565 * To get its accounting right, reuse_swap_page() must be called
2566 * while the page is counted on swap but not yet in mapcount i.e.
2567 * before page_add_anon_rmap() and swap_free(); try_to_free_swap()
2568 * must be called after the swap_free(), or it will never succeed.
03f3c433
KH
2569 * Because delete_from_swap_page() may be called by reuse_swap_page(),
2570 * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry
2571 * in page->private. In this case, a record in swap_cgroup is silently
2572 * discarded at swap_free().
8c7c6e34 2573 */
1da177e4 2574
4294621f 2575 inc_mm_counter(mm, anon_rss);
1da177e4 2576 pte = mk_pte(page, vma->vm_page_prot);
30c9f3a9 2577 if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) {
1da177e4 2578 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
30c9f3a9 2579 flags &= ~FAULT_FLAG_WRITE;
1da177e4 2580 }
1da177e4
LT
2581 flush_icache_page(vma, page);
2582 set_pte_at(mm, address, page_table, pte);
2583 page_add_anon_rmap(page, vma, address);
03f3c433
KH
2584 /* It's better to call commit-charge after rmap is established */
2585 mem_cgroup_commit_charge_swapin(page, ptr);
1da177e4 2586
c475a8ab 2587 swap_free(entry);
b291f000 2588 if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page))
a2c43eed 2589 try_to_free_swap(page);
c475a8ab
HD
2590 unlock_page(page);
2591
30c9f3a9 2592 if (flags & FAULT_FLAG_WRITE) {
61469f1d
HD
2593 ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte);
2594 if (ret & VM_FAULT_ERROR)
2595 ret &= VM_FAULT_ERROR;
1da177e4
LT
2596 goto out;
2597 }
2598
2599 /* No need to invalidate - it was non-present before */
2600 update_mmu_cache(vma, address, pte);
65500d23 2601unlock:
8f4e2101 2602 pte_unmap_unlock(page_table, ptl);
1da177e4
LT
2603out:
2604 return ret;
b8107480 2605out_nomap:
7a81b88c 2606 mem_cgroup_cancel_charge_swapin(ptr);
8f4e2101 2607 pte_unmap_unlock(page_table, ptl);
bc43f75c 2608out_page:
b8107480
KK
2609 unlock_page(page);
2610 page_cache_release(page);
65500d23 2611 return ret;
1da177e4
LT
2612}
2613
2614/*
8f4e2101
HD
2615 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2616 * but allow concurrent faults), and pte mapped but not yet locked.
2617 * We return with mmap_sem still held, but pte unmapped and unlocked.
1da177e4 2618 */
65500d23
HD
2619static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2620 unsigned long address, pte_t *page_table, pmd_t *pmd,
30c9f3a9 2621 unsigned int flags)
1da177e4 2622{
8f4e2101
HD
2623 struct page *page;
2624 spinlock_t *ptl;
1da177e4 2625 pte_t entry;
1da177e4 2626
557ed1fa
NP
2627 /* Allocate our own private page. */
2628 pte_unmap(page_table);
8f4e2101 2629
557ed1fa
NP
2630 if (unlikely(anon_vma_prepare(vma)))
2631 goto oom;
2632 page = alloc_zeroed_user_highpage_movable(vma, address);
2633 if (!page)
2634 goto oom;
0ed361de 2635 __SetPageUptodate(page);
8f4e2101 2636
2c26fdd7 2637 if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL))
8a9f3ccd
BS
2638 goto oom_free_page;
2639
557ed1fa
NP
2640 entry = mk_pte(page, vma->vm_page_prot);
2641 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1da177e4 2642
557ed1fa
NP
2643 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2644 if (!pte_none(*page_table))
2645 goto release;
2646 inc_mm_counter(mm, anon_rss);
557ed1fa 2647 page_add_new_anon_rmap(page, vma, address);
65500d23 2648 set_pte_at(mm, address, page_table, entry);
1da177e4
LT
2649
2650 /* No need to invalidate - it was non-present before */
65500d23 2651 update_mmu_cache(vma, address, entry);
65500d23 2652unlock:
8f4e2101 2653 pte_unmap_unlock(page_table, ptl);
83c54070 2654 return 0;
8f4e2101 2655release:
8a9f3ccd 2656 mem_cgroup_uncharge_page(page);
8f4e2101
HD
2657 page_cache_release(page);
2658 goto unlock;
8a9f3ccd 2659oom_free_page:
6dbf6d3b 2660 page_cache_release(page);
65500d23 2661oom:
1da177e4
LT
2662 return VM_FAULT_OOM;
2663}
2664
2665/*
54cb8821 2666 * __do_fault() tries to create a new page mapping. It aggressively
1da177e4 2667 * tries to share with existing pages, but makes a separate copy if
54cb8821
NP
2668 * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid
2669 * the next page fault.
1da177e4
LT
2670 *
2671 * As this is called only for pages that do not currently exist, we
2672 * do not need to flush old virtual caches or the TLB.
2673 *
8f4e2101 2674 * We enter with non-exclusive mmap_sem (to exclude vma changes,
16abfa08 2675 * but allow concurrent faults), and pte neither mapped nor locked.
8f4e2101 2676 * We return with mmap_sem still held, but pte unmapped and unlocked.
1da177e4 2677 */
54cb8821 2678static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
16abfa08 2679 unsigned long address, pmd_t *pmd,
54cb8821 2680 pgoff_t pgoff, unsigned int flags, pte_t orig_pte)
1da177e4 2681{
16abfa08 2682 pte_t *page_table;
8f4e2101 2683 spinlock_t *ptl;
d0217ac0 2684 struct page *page;
1da177e4 2685 pte_t entry;
1da177e4 2686 int anon = 0;
5b4e655e 2687 int charged = 0;
d08b3851 2688 struct page *dirty_page = NULL;
d0217ac0
NP
2689 struct vm_fault vmf;
2690 int ret;
a200ee18 2691 int page_mkwrite = 0;
54cb8821 2692
d0217ac0
NP
2693 vmf.virtual_address = (void __user *)(address & PAGE_MASK);
2694 vmf.pgoff = pgoff;
2695 vmf.flags = flags;
2696 vmf.page = NULL;
1da177e4 2697
3c18ddd1
NP
2698 ret = vma->vm_ops->fault(vma, &vmf);
2699 if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
2700 return ret;
1da177e4 2701
d00806b1 2702 /*
d0217ac0 2703 * For consistency in subsequent calls, make the faulted page always
d00806b1
NP
2704 * locked.
2705 */
83c54070 2706 if (unlikely(!(ret & VM_FAULT_LOCKED)))
d0217ac0 2707 lock_page(vmf.page);
54cb8821 2708 else
d0217ac0 2709 VM_BUG_ON(!PageLocked(vmf.page));
d00806b1 2710
1da177e4
LT
2711 /*
2712 * Should we do an early C-O-W break?
2713 */
d0217ac0 2714 page = vmf.page;
54cb8821 2715 if (flags & FAULT_FLAG_WRITE) {
9637a5ef 2716 if (!(vma->vm_flags & VM_SHARED)) {
54cb8821 2717 anon = 1;
d00806b1 2718 if (unlikely(anon_vma_prepare(vma))) {
d0217ac0 2719 ret = VM_FAULT_OOM;
54cb8821 2720 goto out;
d00806b1 2721 }
83c54070
NP
2722 page = alloc_page_vma(GFP_HIGHUSER_MOVABLE,
2723 vma, address);
d00806b1 2724 if (!page) {
d0217ac0 2725 ret = VM_FAULT_OOM;
54cb8821 2726 goto out;
d00806b1 2727 }
2c26fdd7 2728 if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL)) {
5b4e655e
KH
2729 ret = VM_FAULT_OOM;
2730 page_cache_release(page);
2731 goto out;
2732 }
2733 charged = 1;
b291f000
NP
2734 /*
2735 * Don't let another task, with possibly unlocked vma,
2736 * keep the mlocked page.
2737 */
2738 if (vma->vm_flags & VM_LOCKED)
2739 clear_page_mlock(vmf.page);
d0217ac0 2740 copy_user_highpage(page, vmf.page, address, vma);
0ed361de 2741 __SetPageUptodate(page);
9637a5ef 2742 } else {
54cb8821
NP
2743 /*
2744 * If the page will be shareable, see if the backing
9637a5ef 2745 * address space wants to know that the page is about
54cb8821
NP
2746 * to become writable
2747 */
69676147 2748 if (vma->vm_ops->page_mkwrite) {
c2ec175c
NP
2749 int tmp;
2750
69676147 2751 unlock_page(page);
b827e496 2752 vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE;
c2ec175c
NP
2753 tmp = vma->vm_ops->page_mkwrite(vma, &vmf);
2754 if (unlikely(tmp &
2755 (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) {
2756 ret = tmp;
b827e496 2757 goto unwritable_page;
d0217ac0 2758 }
b827e496
NP
2759 if (unlikely(!(tmp & VM_FAULT_LOCKED))) {
2760 lock_page(page);
2761 if (!page->mapping) {
2762 ret = 0; /* retry the fault */
2763 unlock_page(page);
2764 goto unwritable_page;
2765 }
2766 } else
2767 VM_BUG_ON(!PageLocked(page));
a200ee18 2768 page_mkwrite = 1;
9637a5ef
DH
2769 }
2770 }
54cb8821 2771
1da177e4
LT
2772 }
2773
8f4e2101 2774 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
1da177e4
LT
2775
2776 /*
2777 * This silly early PAGE_DIRTY setting removes a race
2778 * due to the bad i386 page protection. But it's valid
2779 * for other architectures too.
2780 *
30c9f3a9 2781 * Note that if FAULT_FLAG_WRITE is set, we either now have
1da177e4
LT
2782 * an exclusive copy of the page, or this is a shared mapping,
2783 * so we can make it writable and dirty to avoid having to
2784 * handle that later.
2785 */
2786 /* Only go through if we didn't race with anybody else... */
54cb8821 2787 if (likely(pte_same(*page_table, orig_pte))) {
d00806b1
NP
2788 flush_icache_page(vma, page);
2789 entry = mk_pte(page, vma->vm_page_prot);
54cb8821 2790 if (flags & FAULT_FLAG_WRITE)
1da177e4 2791 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1da177e4 2792 if (anon) {
64d6519d 2793 inc_mm_counter(mm, anon_rss);
64d6519d 2794 page_add_new_anon_rmap(page, vma, address);
f57e88a8 2795 } else {
4294621f 2796 inc_mm_counter(mm, file_rss);
d00806b1 2797 page_add_file_rmap(page);
54cb8821 2798 if (flags & FAULT_FLAG_WRITE) {
d00806b1 2799 dirty_page = page;
d08b3851
PZ
2800 get_page(dirty_page);
2801 }
4294621f 2802 }
64d6519d 2803 set_pte_at(mm, address, page_table, entry);
d00806b1
NP
2804
2805 /* no need to invalidate: a not-present page won't be cached */
2806 update_mmu_cache(vma, address, entry);
1da177e4 2807 } else {
5b4e655e
KH
2808 if (charged)
2809 mem_cgroup_uncharge_page(page);
d00806b1
NP
2810 if (anon)
2811 page_cache_release(page);
2812 else
54cb8821 2813 anon = 1; /* no anon but release faulted_page */
1da177e4
LT
2814 }
2815
8f4e2101 2816 pte_unmap_unlock(page_table, ptl);
d00806b1
NP
2817
2818out:
b827e496
NP
2819 if (dirty_page) {
2820 struct address_space *mapping = page->mapping;
8f7b3d15 2821
b827e496
NP
2822 if (set_page_dirty(dirty_page))
2823 page_mkwrite = 1;
2824 unlock_page(dirty_page);
d08b3851 2825 put_page(dirty_page);
b827e496
NP
2826 if (page_mkwrite && mapping) {
2827 /*
2828 * Some device drivers do not set page.mapping but still
2829 * dirty their pages
2830 */
2831 balance_dirty_pages_ratelimited(mapping);
2832 }
2833
2834 /* file_update_time outside page_lock */
2835 if (vma->vm_file)
2836 file_update_time(vma->vm_file);
2837 } else {
2838 unlock_page(vmf.page);
2839 if (anon)
2840 page_cache_release(vmf.page);
d08b3851 2841 }
d00806b1 2842
83c54070 2843 return ret;
b827e496
NP
2844
2845unwritable_page:
2846 page_cache_release(page);
2847 return ret;
54cb8821 2848}
d00806b1 2849
54cb8821
NP
2850static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2851 unsigned long address, pte_t *page_table, pmd_t *pmd,
30c9f3a9 2852 unsigned int flags, pte_t orig_pte)
54cb8821
NP
2853{
2854 pgoff_t pgoff = (((address & PAGE_MASK)
0da7e01f 2855 - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
54cb8821 2856
16abfa08
HD
2857 pte_unmap(page_table);
2858 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
54cb8821
NP
2859}
2860
1da177e4
LT
2861/*
2862 * Fault of a previously existing named mapping. Repopulate the pte
2863 * from the encoded file_pte if possible. This enables swappable
2864 * nonlinear vmas.
8f4e2101
HD
2865 *
2866 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2867 * but allow concurrent faults), and pte mapped but not yet locked.
2868 * We return with mmap_sem still held, but pte unmapped and unlocked.
1da177e4 2869 */
d0217ac0 2870static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma,
65500d23 2871 unsigned long address, pte_t *page_table, pmd_t *pmd,
30c9f3a9 2872 unsigned int flags, pte_t orig_pte)
1da177e4 2873{
65500d23 2874 pgoff_t pgoff;
1da177e4 2875
30c9f3a9
LT
2876 flags |= FAULT_FLAG_NONLINEAR;
2877
4c21e2f2 2878 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
83c54070 2879 return 0;
1da177e4 2880
2509ef26 2881 if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
65500d23
HD
2882 /*
2883 * Page table corrupted: show pte and kill process.
2884 */
3dc14741 2885 print_bad_pte(vma, address, orig_pte, NULL);
65500d23
HD
2886 return VM_FAULT_OOM;
2887 }
65500d23
HD
2888
2889 pgoff = pte_to_pgoff(orig_pte);
16abfa08 2890 return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte);
1da177e4
LT
2891}
2892
2893/*
2894 * These routines also need to handle stuff like marking pages dirty
2895 * and/or accessed for architectures that don't do it in hardware (most
2896 * RISC architectures). The early dirtying is also good on the i386.
2897 *
2898 * There is also a hook called "update_mmu_cache()" that architectures
2899 * with external mmu caches can use to update those (ie the Sparc or
2900 * PowerPC hashed page tables that act as extended TLBs).
2901 *
c74df32c
HD
2902 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2903 * but allow concurrent faults), and pte mapped but not yet locked.
2904 * We return with mmap_sem still held, but pte unmapped and unlocked.
1da177e4
LT
2905 */
2906static inline int handle_pte_fault(struct mm_struct *mm,
65500d23 2907 struct vm_area_struct *vma, unsigned long address,
30c9f3a9 2908 pte_t *pte, pmd_t *pmd, unsigned int flags)
1da177e4
LT
2909{
2910 pte_t entry;
8f4e2101 2911 spinlock_t *ptl;
1da177e4 2912
8dab5241 2913 entry = *pte;
1da177e4 2914 if (!pte_present(entry)) {
65500d23 2915 if (pte_none(entry)) {
f4b81804 2916 if (vma->vm_ops) {
3c18ddd1 2917 if (likely(vma->vm_ops->fault))
54cb8821 2918 return do_linear_fault(mm, vma, address,
30c9f3a9 2919 pte, pmd, flags, entry);
f4b81804
JS
2920 }
2921 return do_anonymous_page(mm, vma, address,
30c9f3a9 2922 pte, pmd, flags);
65500d23 2923 }
1da177e4 2924 if (pte_file(entry))
d0217ac0 2925 return do_nonlinear_fault(mm, vma, address,
30c9f3a9 2926 pte, pmd, flags, entry);
65500d23 2927 return do_swap_page(mm, vma, address,
30c9f3a9 2928 pte, pmd, flags, entry);
1da177e4
LT
2929 }
2930
4c21e2f2 2931 ptl = pte_lockptr(mm, pmd);
8f4e2101
HD
2932 spin_lock(ptl);
2933 if (unlikely(!pte_same(*pte, entry)))
2934 goto unlock;
30c9f3a9 2935 if (flags & FAULT_FLAG_WRITE) {
1da177e4 2936 if (!pte_write(entry))
8f4e2101
HD
2937 return do_wp_page(mm, vma, address,
2938 pte, pmd, ptl, entry);
1da177e4
LT
2939 entry = pte_mkdirty(entry);
2940 }
2941 entry = pte_mkyoung(entry);
30c9f3a9 2942 if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) {
1a44e149 2943 update_mmu_cache(vma, address, entry);
1a44e149
AA
2944 } else {
2945 /*
2946 * This is needed only for protection faults but the arch code
2947 * is not yet telling us if this is a protection fault or not.
2948 * This still avoids useless tlb flushes for .text page faults
2949 * with threads.
2950 */
30c9f3a9 2951 if (flags & FAULT_FLAG_WRITE)
1a44e149
AA
2952 flush_tlb_page(vma, address);
2953 }
8f4e2101
HD
2954unlock:
2955 pte_unmap_unlock(pte, ptl);
83c54070 2956 return 0;
1da177e4
LT
2957}
2958
2959/*
2960 * By the time we get here, we already hold the mm semaphore
2961 */
83c54070 2962int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
d06063cc 2963 unsigned long address, unsigned int flags)
1da177e4
LT
2964{
2965 pgd_t *pgd;
2966 pud_t *pud;
2967 pmd_t *pmd;
2968 pte_t *pte;
2969
2970 __set_current_state(TASK_RUNNING);
2971
f8891e5e 2972 count_vm_event(PGFAULT);
1da177e4 2973
ac9b9c66 2974 if (unlikely(is_vm_hugetlb_page(vma)))
30c9f3a9 2975 return hugetlb_fault(mm, vma, address, flags);
1da177e4 2976
1da177e4 2977 pgd = pgd_offset(mm, address);
1da177e4
LT
2978 pud = pud_alloc(mm, pgd, address);
2979 if (!pud)
c74df32c 2980 return VM_FAULT_OOM;
1da177e4
LT
2981 pmd = pmd_alloc(mm, pud, address);
2982 if (!pmd)
c74df32c 2983 return VM_FAULT_OOM;
1da177e4
LT
2984 pte = pte_alloc_map(mm, pmd, address);
2985 if (!pte)
c74df32c 2986 return VM_FAULT_OOM;
1da177e4 2987
30c9f3a9 2988 return handle_pte_fault(mm, vma, address, pte, pmd, flags);
1da177e4
LT
2989}
2990
2991#ifndef __PAGETABLE_PUD_FOLDED
2992/*
2993 * Allocate page upper directory.
872fec16 2994 * We've already handled the fast-path in-line.
1da177e4 2995 */
1bb3630e 2996int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
1da177e4 2997{
c74df32c
HD
2998 pud_t *new = pud_alloc_one(mm, address);
2999 if (!new)
1bb3630e 3000 return -ENOMEM;
1da177e4 3001
362a61ad
NP
3002 smp_wmb(); /* See comment in __pte_alloc */
3003
872fec16 3004 spin_lock(&mm->page_table_lock);
1bb3630e 3005 if (pgd_present(*pgd)) /* Another has populated it */
5e541973 3006 pud_free(mm, new);
1bb3630e
HD
3007 else
3008 pgd_populate(mm, pgd, new);
c74df32c 3009 spin_unlock(&mm->page_table_lock);
1bb3630e 3010 return 0;
1da177e4
LT
3011}
3012#endif /* __PAGETABLE_PUD_FOLDED */
3013
3014#ifndef __PAGETABLE_PMD_FOLDED
3015/*
3016 * Allocate page middle directory.
872fec16 3017 * We've already handled the fast-path in-line.
1da177e4 3018 */
1bb3630e 3019int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
1da177e4 3020{
c74df32c
HD
3021 pmd_t *new = pmd_alloc_one(mm, address);
3022 if (!new)
1bb3630e 3023 return -ENOMEM;
1da177e4 3024
362a61ad
NP
3025 smp_wmb(); /* See comment in __pte_alloc */
3026
872fec16 3027 spin_lock(&mm->page_table_lock);
1da177e4 3028#ifndef __ARCH_HAS_4LEVEL_HACK
1bb3630e 3029 if (pud_present(*pud)) /* Another has populated it */
5e541973 3030 pmd_free(mm, new);
1bb3630e
HD
3031 else
3032 pud_populate(mm, pud, new);
1da177e4 3033#else
1bb3630e 3034 if (pgd_present(*pud)) /* Another has populated it */
5e541973 3035 pmd_free(mm, new);
1bb3630e
HD
3036 else
3037 pgd_populate(mm, pud, new);
1da177e4 3038#endif /* __ARCH_HAS_4LEVEL_HACK */
c74df32c 3039 spin_unlock(&mm->page_table_lock);
1bb3630e 3040 return 0;
e0f39591 3041}
1da177e4
LT
3042#endif /* __PAGETABLE_PMD_FOLDED */
3043
3044int make_pages_present(unsigned long addr, unsigned long end)
3045{
3046 int ret, len, write;
3047 struct vm_area_struct * vma;
3048
3049 vma = find_vma(current->mm, addr);
3050 if (!vma)
a477097d 3051 return -ENOMEM;
1da177e4 3052 write = (vma->vm_flags & VM_WRITE) != 0;
5bcb28b1
ES
3053 BUG_ON(addr >= end);
3054 BUG_ON(end > vma->vm_end);
68e116a3 3055 len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE;
1da177e4
LT
3056 ret = get_user_pages(current, current->mm, addr,
3057 len, write, 0, NULL, NULL);
c11d69d8 3058 if (ret < 0)
1da177e4 3059 return ret;
9978ad58 3060 return ret == len ? 0 : -EFAULT;
1da177e4
LT
3061}
3062
1da177e4
LT
3063#if !defined(__HAVE_ARCH_GATE_AREA)
3064
3065#if defined(AT_SYSINFO_EHDR)
5ce7852c 3066static struct vm_area_struct gate_vma;
1da177e4
LT
3067
3068static int __init gate_vma_init(void)
3069{
3070 gate_vma.vm_mm = NULL;
3071 gate_vma.vm_start = FIXADDR_USER_START;
3072 gate_vma.vm_end = FIXADDR_USER_END;
b6558c4a
RM
3073 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
3074 gate_vma.vm_page_prot = __P101;
f47aef55
RM
3075 /*
3076 * Make sure the vDSO gets into every core dump.
3077 * Dumping its contents makes post-mortem fully interpretable later
3078 * without matching up the same kernel and hardware config to see
3079 * what PC values meant.
3080 */
3081 gate_vma.vm_flags |= VM_ALWAYSDUMP;
1da177e4
LT
3082 return 0;
3083}
3084__initcall(gate_vma_init);
3085#endif
3086
3087struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
3088{
3089#ifdef AT_SYSINFO_EHDR
3090 return &gate_vma;
3091#else
3092 return NULL;
3093#endif
3094}
3095
3096int in_gate_area_no_task(unsigned long addr)
3097{
3098#ifdef AT_SYSINFO_EHDR
3099 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
3100 return 1;
3101#endif
3102 return 0;
3103}
3104
3105#endif /* __HAVE_ARCH_GATE_AREA */
0ec76a11 3106
f8ad0f49
JW
3107static int follow_pte(struct mm_struct *mm, unsigned long address,
3108 pte_t **ptepp, spinlock_t **ptlp)
3109{
3110 pgd_t *pgd;
3111 pud_t *pud;
3112 pmd_t *pmd;
3113 pte_t *ptep;
3114
3115 pgd = pgd_offset(mm, address);
3116 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
3117 goto out;
3118
3119 pud = pud_offset(pgd, address);
3120 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
3121 goto out;
3122
3123 pmd = pmd_offset(pud, address);
3124 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
3125 goto out;
3126
3127 /* We cannot handle huge page PFN maps. Luckily they don't exist. */
3128 if (pmd_huge(*pmd))
3129 goto out;
3130
3131 ptep = pte_offset_map_lock(mm, pmd, address, ptlp);
3132 if (!ptep)
3133 goto out;
3134 if (!pte_present(*ptep))
3135 goto unlock;
3136 *ptepp = ptep;
3137 return 0;
3138unlock:
3139 pte_unmap_unlock(ptep, *ptlp);
3140out:
3141 return -EINVAL;
3142}
3143
3b6748e2
JW
3144/**
3145 * follow_pfn - look up PFN at a user virtual address
3146 * @vma: memory mapping
3147 * @address: user virtual address
3148 * @pfn: location to store found PFN
3149 *
3150 * Only IO mappings and raw PFN mappings are allowed.
3151 *
3152 * Returns zero and the pfn at @pfn on success, -ve otherwise.
3153 */
3154int follow_pfn(struct vm_area_struct *vma, unsigned long address,
3155 unsigned long *pfn)
3156{
3157 int ret = -EINVAL;
3158 spinlock_t *ptl;
3159 pte_t *ptep;
3160
3161 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3162 return ret;
3163
3164 ret = follow_pte(vma->vm_mm, address, &ptep, &ptl);
3165 if (ret)
3166 return ret;
3167 *pfn = pte_pfn(*ptep);
3168 pte_unmap_unlock(ptep, ptl);
3169 return 0;
3170}
3171EXPORT_SYMBOL(follow_pfn);
3172
28b2ee20 3173#ifdef CONFIG_HAVE_IOREMAP_PROT
d87fe660 3174int follow_phys(struct vm_area_struct *vma,
3175 unsigned long address, unsigned int flags,
3176 unsigned long *prot, resource_size_t *phys)
28b2ee20 3177{
03668a4d 3178 int ret = -EINVAL;
28b2ee20
RR
3179 pte_t *ptep, pte;
3180 spinlock_t *ptl;
28b2ee20 3181
d87fe660 3182 if (!(vma->vm_flags & (VM_IO | VM_PFNMAP)))
3183 goto out;
28b2ee20 3184
03668a4d 3185 if (follow_pte(vma->vm_mm, address, &ptep, &ptl))
d87fe660 3186 goto out;
28b2ee20 3187 pte = *ptep;
03668a4d 3188
28b2ee20
RR
3189 if ((flags & FOLL_WRITE) && !pte_write(pte))
3190 goto unlock;
28b2ee20
RR
3191
3192 *prot = pgprot_val(pte_pgprot(pte));
03668a4d 3193 *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT;
28b2ee20 3194
03668a4d 3195 ret = 0;
28b2ee20
RR
3196unlock:
3197 pte_unmap_unlock(ptep, ptl);
3198out:
d87fe660 3199 return ret;
28b2ee20
RR
3200}
3201
3202int generic_access_phys(struct vm_area_struct *vma, unsigned long addr,
3203 void *buf, int len, int write)
3204{
3205 resource_size_t phys_addr;
3206 unsigned long prot = 0;
2bc7273b 3207 void __iomem *maddr;
28b2ee20
RR
3208 int offset = addr & (PAGE_SIZE-1);
3209
d87fe660 3210 if (follow_phys(vma, addr, write, &prot, &phys_addr))
28b2ee20
RR
3211 return -EINVAL;
3212
3213 maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot);
3214 if (write)
3215 memcpy_toio(maddr + offset, buf, len);
3216 else
3217 memcpy_fromio(buf, maddr + offset, len);
3218 iounmap(maddr);
3219
3220 return len;
3221}
3222#endif
3223
0ec76a11
DH
3224/*
3225 * Access another process' address space.
3226 * Source/target buffer must be kernel space,
3227 * Do not walk the page table directly, use get_user_pages
3228 */
3229int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write)
3230{
3231 struct mm_struct *mm;
3232 struct vm_area_struct *vma;
0ec76a11
DH
3233 void *old_buf = buf;
3234
3235 mm = get_task_mm(tsk);
3236 if (!mm)
3237 return 0;
3238
3239 down_read(&mm->mmap_sem);
183ff22b 3240 /* ignore errors, just check how much was successfully transferred */
0ec76a11
DH
3241 while (len) {
3242 int bytes, ret, offset;
3243 void *maddr;
28b2ee20 3244 struct page *page = NULL;
0ec76a11
DH
3245
3246 ret = get_user_pages(tsk, mm, addr, 1,
3247 write, 1, &page, &vma);
28b2ee20
RR
3248 if (ret <= 0) {
3249 /*
3250 * Check if this is a VM_IO | VM_PFNMAP VMA, which
3251 * we can access using slightly different code.
3252 */
3253#ifdef CONFIG_HAVE_IOREMAP_PROT
3254 vma = find_vma(mm, addr);
3255 if (!vma)
3256 break;
3257 if (vma->vm_ops && vma->vm_ops->access)
3258 ret = vma->vm_ops->access(vma, addr, buf,
3259 len, write);
3260 if (ret <= 0)
3261#endif
3262 break;
3263 bytes = ret;
0ec76a11 3264 } else {
28b2ee20
RR
3265 bytes = len;
3266 offset = addr & (PAGE_SIZE-1);
3267 if (bytes > PAGE_SIZE-offset)
3268 bytes = PAGE_SIZE-offset;
3269
3270 maddr = kmap(page);
3271 if (write) {
3272 copy_to_user_page(vma, page, addr,
3273 maddr + offset, buf, bytes);
3274 set_page_dirty_lock(page);
3275 } else {
3276 copy_from_user_page(vma, page, addr,
3277 buf, maddr + offset, bytes);
3278 }
3279 kunmap(page);
3280 page_cache_release(page);
0ec76a11 3281 }
0ec76a11
DH
3282 len -= bytes;
3283 buf += bytes;
3284 addr += bytes;
3285 }
3286 up_read(&mm->mmap_sem);
3287 mmput(mm);
3288
3289 return buf - old_buf;
3290}
03252919
AK
3291
3292/*
3293 * Print the name of a VMA.
3294 */
3295void print_vma_addr(char *prefix, unsigned long ip)
3296{
3297 struct mm_struct *mm = current->mm;
3298 struct vm_area_struct *vma;
3299
e8bff74a
IM
3300 /*
3301 * Do not print if we are in atomic
3302 * contexts (in exception stacks, etc.):
3303 */
3304 if (preempt_count())
3305 return;
3306
03252919
AK
3307 down_read(&mm->mmap_sem);
3308 vma = find_vma(mm, ip);
3309 if (vma && vma->vm_file) {
3310 struct file *f = vma->vm_file;
3311 char *buf = (char *)__get_free_page(GFP_KERNEL);
3312 if (buf) {
3313 char *p, *s;
3314
cf28b486 3315 p = d_path(&f->f_path, buf, PAGE_SIZE);
03252919
AK
3316 if (IS_ERR(p))
3317 p = "?";
3318 s = strrchr(p, '/');
3319 if (s)
3320 p = s+1;
3321 printk("%s%s[%lx+%lx]", prefix, p,
3322 vma->vm_start,
3323 vma->vm_end - vma->vm_start);
3324 free_page((unsigned long)buf);
3325 }
3326 }
3327 up_read(&current->mm->mmap_sem);
3328}
3ee1afa3
NP
3329
3330#ifdef CONFIG_PROVE_LOCKING
3331void might_fault(void)
3332{
95156f00
PZ
3333 /*
3334 * Some code (nfs/sunrpc) uses socket ops on kernel memory while
3335 * holding the mmap_sem, this is safe because kernel memory doesn't
3336 * get paged out, therefore we'll never actually fault, and the
3337 * below annotations will generate false positives.
3338 */
3339 if (segment_eq(get_fs(), KERNEL_DS))
3340 return;
3341
3ee1afa3
NP
3342 might_sleep();
3343 /*
3344 * it would be nicer only to annotate paths which are not under
3345 * pagefault_disable, however that requires a larger audit and
3346 * providing helpers like get_user_atomic.
3347 */
3348 if (!in_atomic() && current->mm)
3349 might_lock_read(&current->mm->mmap_sem);
3350}
3351EXPORT_SYMBOL(might_fault);
3352#endif