ia64/xen-unstable

view linux-2.6-xen-sparse/mm/memory.c @ 13631:db3d03dfe92f

add libelf: an ELF binary parser library.

This patch adds a library with a small collection of helper functions
to parse and load elf binaries. The library handles endianess and
elfsize at runtime.

The patch also shuffles around the include files a bit. Now there is
*one* include file holding all the elf structures
(xen/include/public/elfstructs.h) which is included by everyone who
needs them.

It's dead code with this patch only, putting the code into use happens
in followup patches.

Signed-off-by: Gerd Hoffmann <kraxel@suse.de>
---
tools/libxc/xc_elf.h | 525 -----------------------------------
xen/arch/x86/boot/mkelf32.c | 2
xen/common/Makefile | 2
xen/common/libelf/Makefile | 4
xen/common/libelf/README | 1
xen/common/libelf/libelf-dominfo.c | 420 ++++++++++++++++++++++++++++
xen/common/libelf/libelf-loader.c | 156 ++++++++++
xen/common/libelf/libelf-private.h | 51 +++
xen/common/libelf/libelf-relocate.c | 345 +++++++++++++++++++++++
xen/common/libelf/libelf-tools.c | 225 +++++++++++++++
xen/include/public/elfstructs.h | 527 ++++++++++++++++++++++++++++++++++++
xen/include/public/libelf.h | 238 ++++++++++++++++
xen/include/xen/elf.h | 490 ---------------------------------
13 files changed, 1972 insertions(+), 1014 deletions(-)
author Emmanuel Ackaouy <ack@xensource.com>
date Thu Jan 25 22:16:52 2007 +0000 (2007-01-25)
parents 3adf00179a63
children 7180d2e61f92
line source
1 /*
2 * linux/mm/memory.c
3 *
4 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
5 */
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 */
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 */
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 */
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 */
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>
50 #include <linux/delayacct.h>
51 #include <linux/init.h>
53 #include <asm/pgalloc.h>
54 #include <asm/uaccess.h>
55 #include <asm/tlb.h>
56 #include <asm/tlbflush.h>
57 #include <asm/pgtable.h>
59 #include <linux/swapops.h>
60 #include <linux/elf.h>
62 #ifndef CONFIG_NEED_MULTIPLE_NODES
63 /* use the per-pgdat data instead for discontigmem - mbligh */
64 unsigned long max_mapnr;
65 struct page *mem_map;
67 EXPORT_SYMBOL(max_mapnr);
68 EXPORT_SYMBOL(mem_map);
69 #endif
71 unsigned long num_physpages;
72 /*
73 * A number of key systems in x86 including ioremap() rely on the assumption
74 * that high_memory defines the upper bound on direct map memory, then end
75 * of ZONE_NORMAL. Under CONFIG_DISCONTIG this means that max_low_pfn and
76 * highstart_pfn must be the same; there must be no gap between ZONE_NORMAL
77 * and ZONE_HIGHMEM.
78 */
79 void * high_memory;
80 unsigned long vmalloc_earlyreserve;
82 EXPORT_SYMBOL(num_physpages);
83 EXPORT_SYMBOL(high_memory);
84 EXPORT_SYMBOL(vmalloc_earlyreserve);
86 int randomize_va_space __read_mostly = 1;
88 static int __init disable_randmaps(char *s)
89 {
90 randomize_va_space = 0;
91 return 1;
92 }
93 __setup("norandmaps", disable_randmaps);
96 /*
97 * If a p?d_bad entry is found while walking page tables, report
98 * the error, before resetting entry to p?d_none. Usually (but
99 * very seldom) called out from the p?d_none_or_clear_bad macros.
100 */
102 void pgd_clear_bad(pgd_t *pgd)
103 {
104 pgd_ERROR(*pgd);
105 pgd_clear(pgd);
106 }
108 void pud_clear_bad(pud_t *pud)
109 {
110 pud_ERROR(*pud);
111 pud_clear(pud);
112 }
114 void pmd_clear_bad(pmd_t *pmd)
115 {
116 pmd_ERROR(*pmd);
117 pmd_clear(pmd);
118 }
120 /*
121 * Note: this doesn't free the actual pages themselves. That
122 * has been handled earlier when unmapping all the memory regions.
123 */
124 static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd)
125 {
126 struct page *page = pmd_page(*pmd);
127 pmd_clear(pmd);
128 pte_lock_deinit(page);
129 pte_free_tlb(tlb, page);
130 dec_zone_page_state(page, NR_PAGETABLE);
131 tlb->mm->nr_ptes--;
132 }
134 static inline void free_pmd_range(struct mmu_gather *tlb, pud_t *pud,
135 unsigned long addr, unsigned long end,
136 unsigned long floor, unsigned long ceiling)
137 {
138 pmd_t *pmd;
139 unsigned long next;
140 unsigned long start;
142 start = addr;
143 pmd = pmd_offset(pud, addr);
144 do {
145 next = pmd_addr_end(addr, end);
146 if (pmd_none_or_clear_bad(pmd))
147 continue;
148 free_pte_range(tlb, pmd);
149 } while (pmd++, addr = next, addr != end);
151 start &= PUD_MASK;
152 if (start < floor)
153 return;
154 if (ceiling) {
155 ceiling &= PUD_MASK;
156 if (!ceiling)
157 return;
158 }
159 if (end - 1 > ceiling - 1)
160 return;
162 pmd = pmd_offset(pud, start);
163 pud_clear(pud);
164 pmd_free_tlb(tlb, pmd);
165 }
167 static inline void free_pud_range(struct mmu_gather *tlb, pgd_t *pgd,
168 unsigned long addr, unsigned long end,
169 unsigned long floor, unsigned long ceiling)
170 {
171 pud_t *pud;
172 unsigned long next;
173 unsigned long start;
175 start = addr;
176 pud = pud_offset(pgd, addr);
177 do {
178 next = pud_addr_end(addr, end);
179 if (pud_none_or_clear_bad(pud))
180 continue;
181 free_pmd_range(tlb, pud, addr, next, floor, ceiling);
182 } while (pud++, addr = next, addr != end);
184 start &= PGDIR_MASK;
185 if (start < floor)
186 return;
187 if (ceiling) {
188 ceiling &= PGDIR_MASK;
189 if (!ceiling)
190 return;
191 }
192 if (end - 1 > ceiling - 1)
193 return;
195 pud = pud_offset(pgd, start);
196 pgd_clear(pgd);
197 pud_free_tlb(tlb, pud);
198 }
200 /*
201 * This function frees user-level page tables of a process.
202 *
203 * Must be called with pagetable lock held.
204 */
205 void free_pgd_range(struct mmu_gather **tlb,
206 unsigned long addr, unsigned long end,
207 unsigned long floor, unsigned long ceiling)
208 {
209 pgd_t *pgd;
210 unsigned long next;
211 unsigned long start;
213 /*
214 * The next few lines have given us lots of grief...
215 *
216 * Why are we testing PMD* at this top level? Because often
217 * there will be no work to do at all, and we'd prefer not to
218 * go all the way down to the bottom just to discover that.
219 *
220 * Why all these "- 1"s? Because 0 represents both the bottom
221 * of the address space and the top of it (using -1 for the
222 * top wouldn't help much: the masks would do the wrong thing).
223 * The rule is that addr 0 and floor 0 refer to the bottom of
224 * the address space, but end 0 and ceiling 0 refer to the top
225 * Comparisons need to use "end - 1" and "ceiling - 1" (though
226 * that end 0 case should be mythical).
227 *
228 * Wherever addr is brought up or ceiling brought down, we must
229 * be careful to reject "the opposite 0" before it confuses the
230 * subsequent tests. But what about where end is brought down
231 * by PMD_SIZE below? no, end can't go down to 0 there.
232 *
233 * Whereas we round start (addr) and ceiling down, by different
234 * masks at different levels, in order to test whether a table
235 * now has no other vmas using it, so can be freed, we don't
236 * bother to round floor or end up - the tests don't need that.
237 */
239 addr &= PMD_MASK;
240 if (addr < floor) {
241 addr += PMD_SIZE;
242 if (!addr)
243 return;
244 }
245 if (ceiling) {
246 ceiling &= PMD_MASK;
247 if (!ceiling)
248 return;
249 }
250 if (end - 1 > ceiling - 1)
251 end -= PMD_SIZE;
252 if (addr > end - 1)
253 return;
255 start = addr;
256 pgd = pgd_offset((*tlb)->mm, addr);
257 do {
258 next = pgd_addr_end(addr, end);
259 if (pgd_none_or_clear_bad(pgd))
260 continue;
261 free_pud_range(*tlb, pgd, addr, next, floor, ceiling);
262 } while (pgd++, addr = next, addr != end);
264 if (!(*tlb)->fullmm)
265 flush_tlb_pgtables((*tlb)->mm, start, end);
266 }
268 void free_pgtables(struct mmu_gather **tlb, struct vm_area_struct *vma,
269 unsigned long floor, unsigned long ceiling)
270 {
271 while (vma) {
272 struct vm_area_struct *next = vma->vm_next;
273 unsigned long addr = vma->vm_start;
275 /*
276 * Hide vma from rmap and vmtruncate before freeing pgtables
277 */
278 anon_vma_unlink(vma);
279 unlink_file_vma(vma);
281 if (is_vm_hugetlb_page(vma)) {
282 hugetlb_free_pgd_range(tlb, addr, vma->vm_end,
283 floor, next? next->vm_start: ceiling);
284 } else {
285 /*
286 * Optimization: gather nearby vmas into one call down
287 */
288 while (next && next->vm_start <= vma->vm_end + PMD_SIZE
289 && !is_vm_hugetlb_page(next)) {
290 vma = next;
291 next = vma->vm_next;
292 anon_vma_unlink(vma);
293 unlink_file_vma(vma);
294 }
295 free_pgd_range(tlb, addr, vma->vm_end,
296 floor, next? next->vm_start: ceiling);
297 }
298 vma = next;
299 }
300 }
302 int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
303 {
304 struct page *new = pte_alloc_one(mm, address);
305 if (!new)
306 return -ENOMEM;
308 pte_lock_init(new);
309 spin_lock(&mm->page_table_lock);
310 if (pmd_present(*pmd)) { /* Another has populated it */
311 pte_lock_deinit(new);
312 pte_free(new);
313 } else {
314 mm->nr_ptes++;
315 inc_zone_page_state(new, NR_PAGETABLE);
316 pmd_populate(mm, pmd, new);
317 }
318 spin_unlock(&mm->page_table_lock);
319 return 0;
320 }
322 int __pte_alloc_kernel(pmd_t *pmd, unsigned long address)
323 {
324 pte_t *new = pte_alloc_one_kernel(&init_mm, address);
325 if (!new)
326 return -ENOMEM;
328 spin_lock(&init_mm.page_table_lock);
329 if (pmd_present(*pmd)) /* Another has populated it */
330 pte_free_kernel(new);
331 else
332 pmd_populate_kernel(&init_mm, pmd, new);
333 spin_unlock(&init_mm.page_table_lock);
334 return 0;
335 }
337 static inline void add_mm_rss(struct mm_struct *mm, int file_rss, int anon_rss)
338 {
339 if (file_rss)
340 add_mm_counter(mm, file_rss, file_rss);
341 if (anon_rss)
342 add_mm_counter(mm, anon_rss, anon_rss);
343 }
345 /*
346 * This function is called to print an error when a bad pte
347 * is found. For example, we might have a PFN-mapped pte in
348 * a region that doesn't allow it.
349 *
350 * The calling function must still handle the error.
351 */
352 void print_bad_pte(struct vm_area_struct *vma, pte_t pte, unsigned long vaddr)
353 {
354 printk(KERN_ERR "Bad pte = %08llx, process = %s, "
355 "vm_flags = %lx, vaddr = %lx\n",
356 (long long)pte_val(pte),
357 (vma->vm_mm == current->mm ? current->comm : "???"),
358 vma->vm_flags, vaddr);
359 dump_stack();
360 }
362 static inline int is_cow_mapping(unsigned int flags)
363 {
364 return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE;
365 }
367 /*
368 * This function gets the "struct page" associated with a pte.
369 *
370 * NOTE! Some mappings do not have "struct pages". A raw PFN mapping
371 * will have each page table entry just pointing to a raw page frame
372 * number, and as far as the VM layer is concerned, those do not have
373 * pages associated with them - even if the PFN might point to memory
374 * that otherwise is perfectly fine and has a "struct page".
375 *
376 * The way we recognize those mappings is through the rules set up
377 * by "remap_pfn_range()": the vma will have the VM_PFNMAP bit set,
378 * and the vm_pgoff will point to the first PFN mapped: thus every
379 * page that is a raw mapping will always honor the rule
380 *
381 * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT)
382 *
383 * and if that isn't true, the page has been COW'ed (in which case it
384 * _does_ have a "struct page" associated with it even if it is in a
385 * VM_PFNMAP range).
386 */
387 struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, pte_t pte)
388 {
389 unsigned long pfn = pte_pfn(pte);
391 if (unlikely(vma->vm_flags & VM_PFNMAP)) {
392 unsigned long off = (addr - vma->vm_start) >> PAGE_SHIFT;
393 if (pfn == vma->vm_pgoff + off)
394 return NULL;
395 if (!is_cow_mapping(vma->vm_flags))
396 return NULL;
397 }
399 /*
400 * Add some anal sanity checks for now. Eventually,
401 * we should just do "return pfn_to_page(pfn)", but
402 * in the meantime we check that we get a valid pfn,
403 * and that the resulting page looks ok.
404 */
405 if (unlikely(!pfn_valid(pfn))) {
406 if (!(vma->vm_flags & VM_RESERVED))
407 print_bad_pte(vma, pte, addr);
408 return NULL;
409 }
411 /*
412 * NOTE! We still have PageReserved() pages in the page
413 * tables.
414 *
415 * The PAGE_ZERO() pages and various VDSO mappings can
416 * cause them to exist.
417 */
418 return pfn_to_page(pfn);
419 }
421 /*
422 * copy one vm_area from one task to the other. Assumes the page tables
423 * already present in the new task to be cleared in the whole range
424 * covered by this vma.
425 */
427 static inline void
428 copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm,
429 pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma,
430 unsigned long addr, int *rss)
431 {
432 unsigned long vm_flags = vma->vm_flags;
433 pte_t pte = *src_pte;
434 struct page *page;
436 /* pte contains position in swap or file, so copy. */
437 if (unlikely(!pte_present(pte))) {
438 if (!pte_file(pte)) {
439 swp_entry_t entry = pte_to_swp_entry(pte);
441 swap_duplicate(entry);
442 /* make sure dst_mm is on swapoff's mmlist. */
443 if (unlikely(list_empty(&dst_mm->mmlist))) {
444 spin_lock(&mmlist_lock);
445 if (list_empty(&dst_mm->mmlist))
446 list_add(&dst_mm->mmlist,
447 &src_mm->mmlist);
448 spin_unlock(&mmlist_lock);
449 }
450 if (is_write_migration_entry(entry) &&
451 is_cow_mapping(vm_flags)) {
452 /*
453 * COW mappings require pages in both parent
454 * and child to be set to read.
455 */
456 make_migration_entry_read(&entry);
457 pte = swp_entry_to_pte(entry);
458 set_pte_at(src_mm, addr, src_pte, pte);
459 }
460 }
461 goto out_set_pte;
462 }
464 /*
465 * If it's a COW mapping, write protect it both
466 * in the parent and the child
467 */
468 if (is_cow_mapping(vm_flags)) {
469 ptep_set_wrprotect(src_mm, addr, src_pte);
470 pte = *src_pte;
471 }
473 /*
474 * If it's a shared mapping, mark it clean in
475 * the child
476 */
477 if (vm_flags & VM_SHARED)
478 pte = pte_mkclean(pte);
479 pte = pte_mkold(pte);
481 page = vm_normal_page(vma, addr, pte);
482 if (page) {
483 get_page(page);
484 page_dup_rmap(page);
485 rss[!!PageAnon(page)]++;
486 }
488 out_set_pte:
489 set_pte_at(dst_mm, addr, dst_pte, pte);
490 }
492 static int copy_pte_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
493 pmd_t *dst_pmd, pmd_t *src_pmd, struct vm_area_struct *vma,
494 unsigned long addr, unsigned long end)
495 {
496 pte_t *src_pte, *dst_pte;
497 spinlock_t *src_ptl, *dst_ptl;
498 int progress = 0;
499 int rss[2];
501 again:
502 rss[1] = rss[0] = 0;
503 dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl);
504 if (!dst_pte)
505 return -ENOMEM;
506 src_pte = pte_offset_map_nested(src_pmd, addr);
507 src_ptl = pte_lockptr(src_mm, src_pmd);
508 spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING);
510 do {
511 /*
512 * We are holding two locks at this point - either of them
513 * could generate latencies in another task on another CPU.
514 */
515 if (progress >= 32) {
516 progress = 0;
517 if (need_resched() ||
518 need_lockbreak(src_ptl) ||
519 need_lockbreak(dst_ptl))
520 break;
521 }
522 if (pte_none(*src_pte)) {
523 progress++;
524 continue;
525 }
526 copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss);
527 progress += 8;
528 } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end);
530 spin_unlock(src_ptl);
531 pte_unmap_nested(src_pte - 1);
532 add_mm_rss(dst_mm, rss[0], rss[1]);
533 pte_unmap_unlock(dst_pte - 1, dst_ptl);
534 cond_resched();
535 if (addr != end)
536 goto again;
537 return 0;
538 }
540 static inline int copy_pmd_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
541 pud_t *dst_pud, pud_t *src_pud, struct vm_area_struct *vma,
542 unsigned long addr, unsigned long end)
543 {
544 pmd_t *src_pmd, *dst_pmd;
545 unsigned long next;
547 dst_pmd = pmd_alloc(dst_mm, dst_pud, addr);
548 if (!dst_pmd)
549 return -ENOMEM;
550 src_pmd = pmd_offset(src_pud, addr);
551 do {
552 next = pmd_addr_end(addr, end);
553 if (pmd_none_or_clear_bad(src_pmd))
554 continue;
555 if (copy_pte_range(dst_mm, src_mm, dst_pmd, src_pmd,
556 vma, addr, next))
557 return -ENOMEM;
558 } while (dst_pmd++, src_pmd++, addr = next, addr != end);
559 return 0;
560 }
562 static inline int copy_pud_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
563 pgd_t *dst_pgd, pgd_t *src_pgd, struct vm_area_struct *vma,
564 unsigned long addr, unsigned long end)
565 {
566 pud_t *src_pud, *dst_pud;
567 unsigned long next;
569 dst_pud = pud_alloc(dst_mm, dst_pgd, addr);
570 if (!dst_pud)
571 return -ENOMEM;
572 src_pud = pud_offset(src_pgd, addr);
573 do {
574 next = pud_addr_end(addr, end);
575 if (pud_none_or_clear_bad(src_pud))
576 continue;
577 if (copy_pmd_range(dst_mm, src_mm, dst_pud, src_pud,
578 vma, addr, next))
579 return -ENOMEM;
580 } while (dst_pud++, src_pud++, addr = next, addr != end);
581 return 0;
582 }
584 int copy_page_range(struct mm_struct *dst_mm, struct mm_struct *src_mm,
585 struct vm_area_struct *vma)
586 {
587 pgd_t *src_pgd, *dst_pgd;
588 unsigned long next;
589 unsigned long addr = vma->vm_start;
590 unsigned long end = vma->vm_end;
592 /*
593 * Don't copy ptes where a page fault will fill them correctly.
594 * Fork becomes much lighter when there are big shared or private
595 * readonly mappings. The tradeoff is that copy_page_range is more
596 * efficient than faulting.
597 */
598 if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) {
599 if (!vma->anon_vma)
600 return 0;
601 }
603 if (is_vm_hugetlb_page(vma))
604 return copy_hugetlb_page_range(dst_mm, src_mm, vma);
606 dst_pgd = pgd_offset(dst_mm, addr);
607 src_pgd = pgd_offset(src_mm, addr);
608 do {
609 next = pgd_addr_end(addr, end);
610 if (pgd_none_or_clear_bad(src_pgd))
611 continue;
612 if (copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd,
613 vma, addr, next))
614 return -ENOMEM;
615 } while (dst_pgd++, src_pgd++, addr = next, addr != end);
616 return 0;
617 }
619 static unsigned long zap_pte_range(struct mmu_gather *tlb,
620 struct vm_area_struct *vma, pmd_t *pmd,
621 unsigned long addr, unsigned long end,
622 long *zap_work, struct zap_details *details)
623 {
624 struct mm_struct *mm = tlb->mm;
625 pte_t *pte;
626 spinlock_t *ptl;
627 int file_rss = 0;
628 int anon_rss = 0;
630 pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
631 do {
632 pte_t ptent = *pte;
633 if (pte_none(ptent)) {
634 (*zap_work)--;
635 continue;
636 }
638 (*zap_work) -= PAGE_SIZE;
640 if (pte_present(ptent)) {
641 struct page *page;
643 page = vm_normal_page(vma, addr, ptent);
644 if (unlikely(details) && page) {
645 /*
646 * unmap_shared_mapping_pages() wants to
647 * invalidate cache without truncating:
648 * unmap shared but keep private pages.
649 */
650 if (details->check_mapping &&
651 details->check_mapping != page->mapping)
652 continue;
653 /*
654 * Each page->index must be checked when
655 * invalidating or truncating nonlinear.
656 */
657 if (details->nonlinear_vma &&
658 (page->index < details->first_index ||
659 page->index > details->last_index))
660 continue;
661 }
662 ptent = ptep_get_and_clear_full(mm, addr, pte,
663 tlb->fullmm);
664 tlb_remove_tlb_entry(tlb, pte, addr);
665 if (unlikely(!page))
666 continue;
667 if (unlikely(details) && details->nonlinear_vma
668 && linear_page_index(details->nonlinear_vma,
669 addr) != page->index)
670 set_pte_at(mm, addr, pte,
671 pgoff_to_pte(page->index));
672 if (PageAnon(page))
673 anon_rss--;
674 else {
675 if (pte_dirty(ptent))
676 set_page_dirty(page);
677 if (pte_young(ptent))
678 mark_page_accessed(page);
679 file_rss--;
680 }
681 page_remove_rmap(page);
682 tlb_remove_page(tlb, page);
683 continue;
684 }
685 /*
686 * If details->check_mapping, we leave swap entries;
687 * if details->nonlinear_vma, we leave file entries.
688 */
689 if (unlikely(details))
690 continue;
691 if (!pte_file(ptent))
692 free_swap_and_cache(pte_to_swp_entry(ptent));
693 pte_clear_full(mm, addr, pte, tlb->fullmm);
694 } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0));
696 add_mm_rss(mm, file_rss, anon_rss);
697 pte_unmap_unlock(pte - 1, ptl);
699 return addr;
700 }
702 static inline unsigned long zap_pmd_range(struct mmu_gather *tlb,
703 struct vm_area_struct *vma, pud_t *pud,
704 unsigned long addr, unsigned long end,
705 long *zap_work, struct zap_details *details)
706 {
707 pmd_t *pmd;
708 unsigned long next;
710 pmd = pmd_offset(pud, addr);
711 do {
712 next = pmd_addr_end(addr, end);
713 if (pmd_none_or_clear_bad(pmd)) {
714 (*zap_work)--;
715 continue;
716 }
717 next = zap_pte_range(tlb, vma, pmd, addr, next,
718 zap_work, details);
719 } while (pmd++, addr = next, (addr != end && *zap_work > 0));
721 return addr;
722 }
724 static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
725 struct vm_area_struct *vma, pgd_t *pgd,
726 unsigned long addr, unsigned long end,
727 long *zap_work, struct zap_details *details)
728 {
729 pud_t *pud;
730 unsigned long next;
732 pud = pud_offset(pgd, addr);
733 do {
734 next = pud_addr_end(addr, end);
735 if (pud_none_or_clear_bad(pud)) {
736 (*zap_work)--;
737 continue;
738 }
739 next = zap_pmd_range(tlb, vma, pud, addr, next,
740 zap_work, details);
741 } while (pud++, addr = next, (addr != end && *zap_work > 0));
743 return addr;
744 }
746 static unsigned long unmap_page_range(struct mmu_gather *tlb,
747 struct vm_area_struct *vma,
748 unsigned long addr, unsigned long end,
749 long *zap_work, struct zap_details *details)
750 {
751 pgd_t *pgd;
752 unsigned long next;
754 if (details && !details->check_mapping && !details->nonlinear_vma)
755 details = NULL;
757 BUG_ON(addr >= end);
758 tlb_start_vma(tlb, vma);
759 pgd = pgd_offset(vma->vm_mm, addr);
760 do {
761 next = pgd_addr_end(addr, end);
762 if (pgd_none_or_clear_bad(pgd)) {
763 (*zap_work)--;
764 continue;
765 }
766 next = zap_pud_range(tlb, vma, pgd, addr, next,
767 zap_work, details);
768 } while (pgd++, addr = next, (addr != end && *zap_work > 0));
769 tlb_end_vma(tlb, vma);
771 return addr;
772 }
774 #ifdef CONFIG_PREEMPT
775 # define ZAP_BLOCK_SIZE (8 * PAGE_SIZE)
776 #else
777 /* No preempt: go for improved straight-line efficiency */
778 # define ZAP_BLOCK_SIZE (1024 * PAGE_SIZE)
779 #endif
781 /**
782 * unmap_vmas - unmap a range of memory covered by a list of vma's
783 * @tlbp: address of the caller's struct mmu_gather
784 * @vma: the starting vma
785 * @start_addr: virtual address at which to start unmapping
786 * @end_addr: virtual address at which to end unmapping
787 * @nr_accounted: Place number of unmapped pages in vm-accountable vma's here
788 * @details: details of nonlinear truncation or shared cache invalidation
789 *
790 * Returns the end address of the unmapping (restart addr if interrupted).
791 *
792 * Unmap all pages in the vma list.
793 *
794 * We aim to not hold locks for too long (for scheduling latency reasons).
795 * So zap pages in ZAP_BLOCK_SIZE bytecounts. This means we need to
796 * return the ending mmu_gather to the caller.
797 *
798 * Only addresses between `start' and `end' will be unmapped.
799 *
800 * The VMA list must be sorted in ascending virtual address order.
801 *
802 * unmap_vmas() assumes that the caller will flush the whole unmapped address
803 * range after unmap_vmas() returns. So the only responsibility here is to
804 * ensure that any thus-far unmapped pages are flushed before unmap_vmas()
805 * drops the lock and schedules.
806 */
807 unsigned long unmap_vmas(struct mmu_gather **tlbp,
808 struct vm_area_struct *vma, unsigned long start_addr,
809 unsigned long end_addr, unsigned long *nr_accounted,
810 struct zap_details *details)
811 {
812 long zap_work = ZAP_BLOCK_SIZE;
813 unsigned long tlb_start = 0; /* For tlb_finish_mmu */
814 int tlb_start_valid = 0;
815 unsigned long start = start_addr;
816 spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL;
817 int fullmm = (*tlbp)->fullmm;
819 for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) {
820 unsigned long end;
822 start = max(vma->vm_start, start_addr);
823 if (start >= vma->vm_end)
824 continue;
825 end = min(vma->vm_end, end_addr);
826 if (end <= vma->vm_start)
827 continue;
829 if (vma->vm_flags & VM_ACCOUNT)
830 *nr_accounted += (end - start) >> PAGE_SHIFT;
832 while (start != end) {
833 if (!tlb_start_valid) {
834 tlb_start = start;
835 tlb_start_valid = 1;
836 }
838 if (unlikely(is_vm_hugetlb_page(vma))) {
839 unmap_hugepage_range(vma, start, end);
840 zap_work -= (end - start) /
841 (HPAGE_SIZE / PAGE_SIZE);
842 start = end;
843 } else
844 start = unmap_page_range(*tlbp, vma,
845 start, end, &zap_work, details);
847 if (zap_work > 0) {
848 BUG_ON(start != end);
849 break;
850 }
852 tlb_finish_mmu(*tlbp, tlb_start, start);
854 if (need_resched() ||
855 (i_mmap_lock && need_lockbreak(i_mmap_lock))) {
856 if (i_mmap_lock) {
857 *tlbp = NULL;
858 goto out;
859 }
860 cond_resched();
861 }
863 *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm);
864 tlb_start_valid = 0;
865 zap_work = ZAP_BLOCK_SIZE;
866 }
867 }
868 out:
869 return start; /* which is now the end (or restart) address */
870 }
872 /**
873 * zap_page_range - remove user pages in a given range
874 * @vma: vm_area_struct holding the applicable pages
875 * @address: starting address of pages to zap
876 * @size: number of bytes to zap
877 * @details: details of nonlinear truncation or shared cache invalidation
878 */
879 unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address,
880 unsigned long size, struct zap_details *details)
881 {
882 struct mm_struct *mm = vma->vm_mm;
883 struct mmu_gather *tlb;
884 unsigned long end = address + size;
885 unsigned long nr_accounted = 0;
887 lru_add_drain();
888 tlb = tlb_gather_mmu(mm, 0);
889 update_hiwater_rss(mm);
890 end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details);
891 if (tlb)
892 tlb_finish_mmu(tlb, address, end);
893 return end;
894 }
895 EXPORT_SYMBOL(zap_page_range);
897 /*
898 * Do a quick page-table lookup for a single page.
899 */
900 struct page *follow_page(struct vm_area_struct *vma, unsigned long address,
901 unsigned int flags)
902 {
903 pgd_t *pgd;
904 pud_t *pud;
905 pmd_t *pmd;
906 pte_t *ptep, pte;
907 spinlock_t *ptl;
908 struct page *page;
909 struct mm_struct *mm = vma->vm_mm;
911 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
912 if (!IS_ERR(page)) {
913 BUG_ON(flags & FOLL_GET);
914 goto out;
915 }
917 page = NULL;
918 pgd = pgd_offset(mm, address);
919 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
920 goto no_page_table;
922 pud = pud_offset(pgd, address);
923 if (pud_none(*pud) || unlikely(pud_bad(*pud)))
924 goto no_page_table;
926 pmd = pmd_offset(pud, address);
927 if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd)))
928 goto no_page_table;
930 if (pmd_huge(*pmd)) {
931 BUG_ON(flags & FOLL_GET);
932 page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE);
933 goto out;
934 }
936 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
937 if (!ptep)
938 goto out;
940 pte = *ptep;
941 if (!pte_present(pte))
942 goto unlock;
943 if ((flags & FOLL_WRITE) && !pte_write(pte))
944 goto unlock;
945 page = vm_normal_page(vma, address, pte);
946 if (unlikely(!page))
947 goto unlock;
949 if (flags & FOLL_GET)
950 get_page(page);
951 if (flags & FOLL_TOUCH) {
952 if ((flags & FOLL_WRITE) &&
953 !pte_dirty(pte) && !PageDirty(page))
954 set_page_dirty(page);
955 mark_page_accessed(page);
956 }
957 unlock:
958 pte_unmap_unlock(ptep, ptl);
959 out:
960 return page;
962 no_page_table:
963 /*
964 * When core dumping an enormous anonymous area that nobody
965 * has touched so far, we don't want to allocate page tables.
966 */
967 if (flags & FOLL_ANON) {
968 page = ZERO_PAGE(address);
969 if (flags & FOLL_GET)
970 get_page(page);
971 BUG_ON(flags & FOLL_WRITE);
972 }
973 return page;
974 }
976 int get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
977 unsigned long start, int len, int write, int force,
978 struct page **pages, struct vm_area_struct **vmas)
979 {
980 int i;
981 unsigned int vm_flags;
983 /*
984 * Require read or write permissions.
985 * If 'force' is set, we only require the "MAY" flags.
986 */
987 vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD);
988 vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE);
989 i = 0;
991 do {
992 struct vm_area_struct *vma;
993 unsigned int foll_flags;
995 vma = find_extend_vma(mm, start);
996 if (!vma && in_gate_area(tsk, start)) {
997 unsigned long pg = start & PAGE_MASK;
998 struct vm_area_struct *gate_vma = get_gate_vma(tsk);
999 pgd_t *pgd;
1000 pud_t *pud;
1001 pmd_t *pmd;
1002 pte_t *pte;
1003 if (write) /* user gate pages are read-only */
1004 return i ? : -EFAULT;
1005 if (pg > TASK_SIZE)
1006 pgd = pgd_offset_k(pg);
1007 else
1008 pgd = pgd_offset_gate(mm, pg);
1009 BUG_ON(pgd_none(*pgd));
1010 pud = pud_offset(pgd, pg);
1011 BUG_ON(pud_none(*pud));
1012 pmd = pmd_offset(pud, pg);
1013 if (pmd_none(*pmd))
1014 return i ? : -EFAULT;
1015 pte = pte_offset_map(pmd, pg);
1016 if (pte_none(*pte)) {
1017 pte_unmap(pte);
1018 return i ? : -EFAULT;
1020 if (pages) {
1021 struct page *page = vm_normal_page(gate_vma, start, *pte);
1022 pages[i] = page;
1023 if (page)
1024 get_page(page);
1026 pte_unmap(pte);
1027 if (vmas)
1028 vmas[i] = gate_vma;
1029 i++;
1030 start += PAGE_SIZE;
1031 len--;
1032 continue;
1035 #ifdef CONFIG_XEN
1036 if (vma && (vma->vm_flags & VM_FOREIGN)) {
1037 struct page **map = vma->vm_private_data;
1038 int offset = (start - vma->vm_start) >> PAGE_SHIFT;
1039 if (map[offset] != NULL) {
1040 if (pages) {
1041 struct page *page = map[offset];
1043 pages[i] = page;
1044 get_page(page);
1046 if (vmas)
1047 vmas[i] = vma;
1048 i++;
1049 start += PAGE_SIZE;
1050 len--;
1051 continue;
1054 #endif
1055 if (!vma || (vma->vm_flags & (VM_IO | VM_PFNMAP))
1056 || !(vm_flags & vma->vm_flags))
1057 return i ? : -EFAULT;
1059 if (is_vm_hugetlb_page(vma)) {
1060 i = follow_hugetlb_page(mm, vma, pages, vmas,
1061 &start, &len, i);
1062 continue;
1065 foll_flags = FOLL_TOUCH;
1066 if (pages)
1067 foll_flags |= FOLL_GET;
1068 if (!write && !(vma->vm_flags & VM_LOCKED) &&
1069 (!vma->vm_ops || !vma->vm_ops->nopage))
1070 foll_flags |= FOLL_ANON;
1072 do {
1073 struct page *page;
1075 if (write)
1076 foll_flags |= FOLL_WRITE;
1078 cond_resched();
1079 while (!(page = follow_page(vma, start, foll_flags))) {
1080 int ret;
1081 ret = __handle_mm_fault(mm, vma, start,
1082 foll_flags & FOLL_WRITE);
1083 /*
1084 * The VM_FAULT_WRITE bit tells us that do_wp_page has
1085 * broken COW when necessary, even if maybe_mkwrite
1086 * decided not to set pte_write. We can thus safely do
1087 * subsequent page lookups as if they were reads.
1088 */
1089 if (ret & VM_FAULT_WRITE)
1090 foll_flags &= ~FOLL_WRITE;
1092 switch (ret & ~VM_FAULT_WRITE) {
1093 case VM_FAULT_MINOR:
1094 tsk->min_flt++;
1095 break;
1096 case VM_FAULT_MAJOR:
1097 tsk->maj_flt++;
1098 break;
1099 case VM_FAULT_SIGBUS:
1100 return i ? i : -EFAULT;
1101 case VM_FAULT_OOM:
1102 return i ? i : -ENOMEM;
1103 default:
1104 BUG();
1107 if (pages) {
1108 pages[i] = page;
1110 flush_anon_page(page, start);
1111 flush_dcache_page(page);
1113 if (vmas)
1114 vmas[i] = vma;
1115 i++;
1116 start += PAGE_SIZE;
1117 len--;
1118 } while (len && start < vma->vm_end);
1119 } while (len);
1120 return i;
1122 EXPORT_SYMBOL(get_user_pages);
1124 static int zeromap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1125 unsigned long addr, unsigned long end, pgprot_t prot)
1127 pte_t *pte;
1128 spinlock_t *ptl;
1130 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
1131 if (!pte)
1132 return -ENOMEM;
1133 do {
1134 struct page *page = ZERO_PAGE(addr);
1135 pte_t zero_pte = pte_wrprotect(mk_pte(page, prot));
1136 page_cache_get(page);
1137 page_add_file_rmap(page);
1138 inc_mm_counter(mm, file_rss);
1139 BUG_ON(!pte_none(*pte));
1140 set_pte_at(mm, addr, pte, zero_pte);
1141 } while (pte++, addr += PAGE_SIZE, addr != end);
1142 pte_unmap_unlock(pte - 1, ptl);
1143 return 0;
1146 static inline int zeromap_pmd_range(struct mm_struct *mm, pud_t *pud,
1147 unsigned long addr, unsigned long end, pgprot_t prot)
1149 pmd_t *pmd;
1150 unsigned long next;
1152 pmd = pmd_alloc(mm, pud, addr);
1153 if (!pmd)
1154 return -ENOMEM;
1155 do {
1156 next = pmd_addr_end(addr, end);
1157 if (zeromap_pte_range(mm, pmd, addr, next, prot))
1158 return -ENOMEM;
1159 } while (pmd++, addr = next, addr != end);
1160 return 0;
1163 static inline int zeromap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1164 unsigned long addr, unsigned long end, pgprot_t prot)
1166 pud_t *pud;
1167 unsigned long next;
1169 pud = pud_alloc(mm, pgd, addr);
1170 if (!pud)
1171 return -ENOMEM;
1172 do {
1173 next = pud_addr_end(addr, end);
1174 if (zeromap_pmd_range(mm, pud, addr, next, prot))
1175 return -ENOMEM;
1176 } while (pud++, addr = next, addr != end);
1177 return 0;
1180 int zeromap_page_range(struct vm_area_struct *vma,
1181 unsigned long addr, unsigned long size, pgprot_t prot)
1183 pgd_t *pgd;
1184 unsigned long next;
1185 unsigned long end = addr + size;
1186 struct mm_struct *mm = vma->vm_mm;
1187 int err;
1189 BUG_ON(addr >= end);
1190 pgd = pgd_offset(mm, addr);
1191 flush_cache_range(vma, addr, end);
1192 do {
1193 next = pgd_addr_end(addr, end);
1194 err = zeromap_pud_range(mm, pgd, addr, next, prot);
1195 if (err)
1196 break;
1197 } while (pgd++, addr = next, addr != end);
1198 return err;
1201 pte_t * fastcall get_locked_pte(struct mm_struct *mm, unsigned long addr, spinlock_t **ptl)
1203 pgd_t * pgd = pgd_offset(mm, addr);
1204 pud_t * pud = pud_alloc(mm, pgd, addr);
1205 if (pud) {
1206 pmd_t * pmd = pmd_alloc(mm, pud, addr);
1207 if (pmd)
1208 return pte_alloc_map_lock(mm, pmd, addr, ptl);
1210 return NULL;
1213 /*
1214 * This is the old fallback for page remapping.
1216 * For historical reasons, it only allows reserved pages. Only
1217 * old drivers should use this, and they needed to mark their
1218 * pages reserved for the old functions anyway.
1219 */
1220 static int insert_page(struct mm_struct *mm, unsigned long addr, struct page *page, pgprot_t prot)
1222 int retval;
1223 pte_t *pte;
1224 spinlock_t *ptl;
1226 retval = -EINVAL;
1227 if (PageAnon(page))
1228 goto out;
1229 retval = -ENOMEM;
1230 flush_dcache_page(page);
1231 pte = get_locked_pte(mm, addr, &ptl);
1232 if (!pte)
1233 goto out;
1234 retval = -EBUSY;
1235 if (!pte_none(*pte))
1236 goto out_unlock;
1238 /* Ok, finally just insert the thing.. */
1239 get_page(page);
1240 inc_mm_counter(mm, file_rss);
1241 page_add_file_rmap(page);
1242 set_pte_at(mm, addr, pte, mk_pte(page, prot));
1244 retval = 0;
1245 out_unlock:
1246 pte_unmap_unlock(pte, ptl);
1247 out:
1248 return retval;
1251 /*
1252 * This allows drivers to insert individual pages they've allocated
1253 * into a user vma.
1255 * The page has to be a nice clean _individual_ kernel allocation.
1256 * If you allocate a compound page, you need to have marked it as
1257 * such (__GFP_COMP), or manually just split the page up yourself
1258 * (see split_page()).
1260 * NOTE! Traditionally this was done with "remap_pfn_range()" which
1261 * took an arbitrary page protection parameter. This doesn't allow
1262 * that. Your vma protection will have to be set up correctly, which
1263 * means that if you want a shared writable mapping, you'd better
1264 * ask for a shared writable mapping!
1266 * The page does not need to be reserved.
1267 */
1268 int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, struct page *page)
1270 if (addr < vma->vm_start || addr >= vma->vm_end)
1271 return -EFAULT;
1272 if (!page_count(page))
1273 return -EINVAL;
1274 vma->vm_flags |= VM_INSERTPAGE;
1275 return insert_page(vma->vm_mm, addr, page, vma->vm_page_prot);
1277 EXPORT_SYMBOL(vm_insert_page);
1279 /*
1280 * maps a range of physical memory into the requested pages. the old
1281 * mappings are removed. any references to nonexistent pages results
1282 * in null mappings (currently treated as "copy-on-access")
1283 */
1284 static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd,
1285 unsigned long addr, unsigned long end,
1286 unsigned long pfn, pgprot_t prot)
1288 pte_t *pte;
1289 spinlock_t *ptl;
1291 pte = pte_alloc_map_lock(mm, pmd, addr, &ptl);
1292 if (!pte)
1293 return -ENOMEM;
1294 do {
1295 BUG_ON(!pte_none(*pte));
1296 set_pte_at(mm, addr, pte, pfn_pte(pfn, prot));
1297 pfn++;
1298 } while (pte++, addr += PAGE_SIZE, addr != end);
1299 pte_unmap_unlock(pte - 1, ptl);
1300 return 0;
1303 static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud,
1304 unsigned long addr, unsigned long end,
1305 unsigned long pfn, pgprot_t prot)
1307 pmd_t *pmd;
1308 unsigned long next;
1310 pfn -= addr >> PAGE_SHIFT;
1311 pmd = pmd_alloc(mm, pud, addr);
1312 if (!pmd)
1313 return -ENOMEM;
1314 do {
1315 next = pmd_addr_end(addr, end);
1316 if (remap_pte_range(mm, pmd, addr, next,
1317 pfn + (addr >> PAGE_SHIFT), prot))
1318 return -ENOMEM;
1319 } while (pmd++, addr = next, addr != end);
1320 return 0;
1323 static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd,
1324 unsigned long addr, unsigned long end,
1325 unsigned long pfn, pgprot_t prot)
1327 pud_t *pud;
1328 unsigned long next;
1330 pfn -= addr >> PAGE_SHIFT;
1331 pud = pud_alloc(mm, pgd, addr);
1332 if (!pud)
1333 return -ENOMEM;
1334 do {
1335 next = pud_addr_end(addr, end);
1336 if (remap_pmd_range(mm, pud, addr, next,
1337 pfn + (addr >> PAGE_SHIFT), prot))
1338 return -ENOMEM;
1339 } while (pud++, addr = next, addr != end);
1340 return 0;
1343 /* Note: this is only safe if the mm semaphore is held when called. */
1344 int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr,
1345 unsigned long pfn, unsigned long size, pgprot_t prot)
1347 pgd_t *pgd;
1348 unsigned long next;
1349 unsigned long end = addr + PAGE_ALIGN(size);
1350 struct mm_struct *mm = vma->vm_mm;
1351 int err;
1353 /*
1354 * Physically remapped pages are special. Tell the
1355 * rest of the world about it:
1356 * VM_IO tells people not to look at these pages
1357 * (accesses can have side effects).
1358 * VM_RESERVED is specified all over the place, because
1359 * in 2.4 it kept swapout's vma scan off this vma; but
1360 * in 2.6 the LRU scan won't even find its pages, so this
1361 * flag means no more than count its pages in reserved_vm,
1362 * and omit it from core dump, even when VM_IO turned off.
1363 * VM_PFNMAP tells the core MM that the base pages are just
1364 * raw PFN mappings, and do not have a "struct page" associated
1365 * with them.
1367 * There's a horrible special case to handle copy-on-write
1368 * behaviour that some programs depend on. We mark the "original"
1369 * un-COW'ed pages by matching them up with "vma->vm_pgoff".
1370 */
1371 if (is_cow_mapping(vma->vm_flags)) {
1372 if (addr != vma->vm_start || end != vma->vm_end)
1373 return -EINVAL;
1374 vma->vm_pgoff = pfn;
1377 vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP;
1379 BUG_ON(addr >= end);
1380 pfn -= addr >> PAGE_SHIFT;
1381 pgd = pgd_offset(mm, addr);
1382 flush_cache_range(vma, addr, end);
1383 do {
1384 next = pgd_addr_end(addr, end);
1385 err = remap_pud_range(mm, pgd, addr, next,
1386 pfn + (addr >> PAGE_SHIFT), prot);
1387 if (err)
1388 break;
1389 } while (pgd++, addr = next, addr != end);
1390 return err;
1392 EXPORT_SYMBOL(remap_pfn_range);
1394 #ifdef CONFIG_XEN
1395 static inline int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd,
1396 unsigned long addr, unsigned long end,
1397 pte_fn_t fn, void *data)
1399 pte_t *pte;
1400 int err;
1401 struct page *pmd_page;
1402 spinlock_t *ptl;
1404 pte = (mm == &init_mm) ?
1405 pte_alloc_kernel(pmd, addr) :
1406 pte_alloc_map_lock(mm, pmd, addr, &ptl);
1407 if (!pte)
1408 return -ENOMEM;
1410 BUG_ON(pmd_huge(*pmd));
1412 pmd_page = pmd_page(*pmd);
1414 do {
1415 err = fn(pte, pmd_page, addr, data);
1416 if (err)
1417 break;
1418 } while (pte++, addr += PAGE_SIZE, addr != end);
1420 if (mm != &init_mm)
1421 pte_unmap_unlock(pte-1, ptl);
1422 return err;
1425 static inline int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud,
1426 unsigned long addr, unsigned long end,
1427 pte_fn_t fn, void *data)
1429 pmd_t *pmd;
1430 unsigned long next;
1431 int err;
1433 pmd = pmd_alloc(mm, pud, addr);
1434 if (!pmd)
1435 return -ENOMEM;
1436 do {
1437 next = pmd_addr_end(addr, end);
1438 err = apply_to_pte_range(mm, pmd, addr, next, fn, data);
1439 if (err)
1440 break;
1441 } while (pmd++, addr = next, addr != end);
1442 return err;
1445 static inline int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd,
1446 unsigned long addr, unsigned long end,
1447 pte_fn_t fn, void *data)
1449 pud_t *pud;
1450 unsigned long next;
1451 int err;
1453 pud = pud_alloc(mm, pgd, addr);
1454 if (!pud)
1455 return -ENOMEM;
1456 do {
1457 next = pud_addr_end(addr, end);
1458 err = apply_to_pmd_range(mm, pud, addr, next, fn, data);
1459 if (err)
1460 break;
1461 } while (pud++, addr = next, addr != end);
1462 return err;
1465 /*
1466 * Scan a region of virtual memory, filling in page tables as necessary
1467 * and calling a provided function on each leaf page table.
1468 */
1469 int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
1470 unsigned long size, pte_fn_t fn, void *data)
1472 pgd_t *pgd;
1473 unsigned long next;
1474 unsigned long end = addr + size;
1475 int err;
1477 BUG_ON(addr >= end);
1478 pgd = pgd_offset(mm, addr);
1479 do {
1480 next = pgd_addr_end(addr, end);
1481 err = apply_to_pud_range(mm, pgd, addr, next, fn, data);
1482 if (err)
1483 break;
1484 } while (pgd++, addr = next, addr != end);
1485 return err;
1487 EXPORT_SYMBOL_GPL(apply_to_page_range);
1488 #endif
1490 /*
1491 * handle_pte_fault chooses page fault handler according to an entry
1492 * which was read non-atomically. Before making any commitment, on
1493 * those architectures or configurations (e.g. i386 with PAE) which
1494 * might give a mix of unmatched parts, do_swap_page and do_file_page
1495 * must check under lock before unmapping the pte and proceeding
1496 * (but do_wp_page is only called after already making such a check;
1497 * and do_anonymous_page and do_no_page can safely check later on).
1498 */
1499 static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd,
1500 pte_t *page_table, pte_t orig_pte)
1502 int same = 1;
1503 #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT)
1504 if (sizeof(pte_t) > sizeof(unsigned long)) {
1505 spinlock_t *ptl = pte_lockptr(mm, pmd);
1506 spin_lock(ptl);
1507 same = pte_same(*page_table, orig_pte);
1508 spin_unlock(ptl);
1510 #endif
1511 pte_unmap(page_table);
1512 return same;
1515 /*
1516 * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when
1517 * servicing faults for write access. In the normal case, do always want
1518 * pte_mkwrite. But get_user_pages can cause write faults for mappings
1519 * that do not have writing enabled, when used by access_process_vm.
1520 */
1521 static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma)
1523 if (likely(vma->vm_flags & VM_WRITE))
1524 pte = pte_mkwrite(pte);
1525 return pte;
1528 static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va)
1530 /*
1531 * If the source page was a PFN mapping, we don't have
1532 * a "struct page" for it. We do a best-effort copy by
1533 * just copying from the original user address. If that
1534 * fails, we just zero-fill it. Live with it.
1535 */
1536 if (unlikely(!src)) {
1537 void *kaddr = kmap_atomic(dst, KM_USER0);
1538 void __user *uaddr = (void __user *)(va & PAGE_MASK);
1540 /*
1541 * This really shouldn't fail, because the page is there
1542 * in the page tables. But it might just be unreadable,
1543 * in which case we just give up and fill the result with
1544 * zeroes.
1545 */
1546 if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE))
1547 memset(kaddr, 0, PAGE_SIZE);
1548 kunmap_atomic(kaddr, KM_USER0);
1549 return;
1552 copy_user_highpage(dst, src, va);
1555 /*
1556 * This routine handles present pages, when users try to write
1557 * to a shared page. It is done by copying the page to a new address
1558 * and decrementing the shared-page counter for the old page.
1560 * Note that this routine assumes that the protection checks have been
1561 * done by the caller (the low-level page fault routine in most cases).
1562 * Thus we can safely just mark it writable once we've done any necessary
1563 * COW.
1565 * We also mark the page dirty at this point even though the page will
1566 * change only once the write actually happens. This avoids a few races,
1567 * and potentially makes it more efficient.
1569 * We enter with non-exclusive mmap_sem (to exclude vma changes,
1570 * but allow concurrent faults), with pte both mapped and locked.
1571 * We return with mmap_sem still held, but pte unmapped and unlocked.
1572 */
1573 static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma,
1574 unsigned long address, pte_t *page_table, pmd_t *pmd,
1575 spinlock_t *ptl, pte_t orig_pte)
1577 struct page *old_page, *new_page;
1578 pte_t entry;
1579 int reuse, ret = VM_FAULT_MINOR;
1581 old_page = vm_normal_page(vma, address, orig_pte);
1582 if (!old_page)
1583 goto gotten;
1585 if (unlikely((vma->vm_flags & (VM_SHARED|VM_WRITE)) ==
1586 (VM_SHARED|VM_WRITE))) {
1587 if (vma->vm_ops && vma->vm_ops->page_mkwrite) {
1588 /*
1589 * Notify the address space that the page is about to
1590 * become writable so that it can prohibit this or wait
1591 * for the page to get into an appropriate state.
1593 * We do this without the lock held, so that it can
1594 * sleep if it needs to.
1595 */
1596 page_cache_get(old_page);
1597 pte_unmap_unlock(page_table, ptl);
1599 if (vma->vm_ops->page_mkwrite(vma, old_page) < 0)
1600 goto unwritable_page;
1602 page_cache_release(old_page);
1604 /*
1605 * Since we dropped the lock we need to revalidate
1606 * the PTE as someone else may have changed it. If
1607 * they did, we just return, as we can count on the
1608 * MMU to tell us if they didn't also make it writable.
1609 */
1610 page_table = pte_offset_map_lock(mm, pmd, address,
1611 &ptl);
1612 if (!pte_same(*page_table, orig_pte))
1613 goto unlock;
1616 reuse = 1;
1617 } else if (PageAnon(old_page) && !TestSetPageLocked(old_page)) {
1618 reuse = can_share_swap_page(old_page);
1619 unlock_page(old_page);
1620 } else {
1621 reuse = 0;
1624 if (reuse) {
1625 flush_cache_page(vma, address, pte_pfn(orig_pte));
1626 entry = pte_mkyoung(orig_pte);
1627 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1628 ptep_set_access_flags(vma, address, page_table, entry, 1);
1629 update_mmu_cache(vma, address, entry);
1630 lazy_mmu_prot_update(entry);
1631 ret |= VM_FAULT_WRITE;
1632 goto unlock;
1635 /*
1636 * Ok, we need to copy. Oh, well..
1637 */
1638 page_cache_get(old_page);
1639 gotten:
1640 pte_unmap_unlock(page_table, ptl);
1642 if (unlikely(anon_vma_prepare(vma)))
1643 goto oom;
1644 if (old_page == ZERO_PAGE(address)) {
1645 new_page = alloc_zeroed_user_highpage(vma, address);
1646 if (!new_page)
1647 goto oom;
1648 } else {
1649 new_page = alloc_page_vma(GFP_HIGHUSER, vma, address);
1650 if (!new_page)
1651 goto oom;
1652 cow_user_page(new_page, old_page, address);
1655 /*
1656 * Re-check the pte - we dropped the lock
1657 */
1658 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
1659 if (likely(pte_same(*page_table, orig_pte))) {
1660 if (old_page) {
1661 page_remove_rmap(old_page);
1662 if (!PageAnon(old_page)) {
1663 dec_mm_counter(mm, file_rss);
1664 inc_mm_counter(mm, anon_rss);
1666 } else
1667 inc_mm_counter(mm, anon_rss);
1668 flush_cache_page(vma, address, pte_pfn(orig_pte));
1669 entry = mk_pte(new_page, vma->vm_page_prot);
1670 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
1671 lazy_mmu_prot_update(entry);
1672 ptep_establish(vma, address, page_table, entry);
1673 update_mmu_cache(vma, address, entry);
1674 lru_cache_add_active(new_page);
1675 page_add_new_anon_rmap(new_page, vma, address);
1677 /* Free the old page.. */
1678 new_page = old_page;
1679 ret |= VM_FAULT_WRITE;
1681 if (new_page)
1682 page_cache_release(new_page);
1683 if (old_page)
1684 page_cache_release(old_page);
1685 unlock:
1686 pte_unmap_unlock(page_table, ptl);
1687 return ret;
1688 oom:
1689 if (old_page)
1690 page_cache_release(old_page);
1691 return VM_FAULT_OOM;
1693 unwritable_page:
1694 page_cache_release(old_page);
1695 return VM_FAULT_SIGBUS;
1698 /*
1699 * Helper functions for unmap_mapping_range().
1701 * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __
1703 * We have to restart searching the prio_tree whenever we drop the lock,
1704 * since the iterator is only valid while the lock is held, and anyway
1705 * a later vma might be split and reinserted earlier while lock dropped.
1707 * The list of nonlinear vmas could be handled more efficiently, using
1708 * a placeholder, but handle it in the same way until a need is shown.
1709 * It is important to search the prio_tree before nonlinear list: a vma
1710 * may become nonlinear and be shifted from prio_tree to nonlinear list
1711 * while the lock is dropped; but never shifted from list to prio_tree.
1713 * In order to make forward progress despite restarting the search,
1714 * vm_truncate_count is used to mark a vma as now dealt with, so we can
1715 * quickly skip it next time around. Since the prio_tree search only
1716 * shows us those vmas affected by unmapping the range in question, we
1717 * can't efficiently keep all vmas in step with mapping->truncate_count:
1718 * so instead reset them all whenever it wraps back to 0 (then go to 1).
1719 * mapping->truncate_count and vma->vm_truncate_count are protected by
1720 * i_mmap_lock.
1722 * In order to make forward progress despite repeatedly restarting some
1723 * large vma, note the restart_addr from unmap_vmas when it breaks out:
1724 * and restart from that address when we reach that vma again. It might
1725 * have been split or merged, shrunk or extended, but never shifted: so
1726 * restart_addr remains valid so long as it remains in the vma's range.
1727 * unmap_mapping_range forces truncate_count to leap over page-aligned
1728 * values so we can save vma's restart_addr in its truncate_count field.
1729 */
1730 #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK))
1732 static void reset_vma_truncate_counts(struct address_space *mapping)
1734 struct vm_area_struct *vma;
1735 struct prio_tree_iter iter;
1737 vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX)
1738 vma->vm_truncate_count = 0;
1739 list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list)
1740 vma->vm_truncate_count = 0;
1743 static int unmap_mapping_range_vma(struct vm_area_struct *vma,
1744 unsigned long start_addr, unsigned long end_addr,
1745 struct zap_details *details)
1747 unsigned long restart_addr;
1748 int need_break;
1750 again:
1751 restart_addr = vma->vm_truncate_count;
1752 if (is_restart_addr(restart_addr) && start_addr < restart_addr) {
1753 start_addr = restart_addr;
1754 if (start_addr >= end_addr) {
1755 /* Top of vma has been split off since last time */
1756 vma->vm_truncate_count = details->truncate_count;
1757 return 0;
1761 restart_addr = zap_page_range(vma, start_addr,
1762 end_addr - start_addr, details);
1763 need_break = need_resched() ||
1764 need_lockbreak(details->i_mmap_lock);
1766 if (restart_addr >= end_addr) {
1767 /* We have now completed this vma: mark it so */
1768 vma->vm_truncate_count = details->truncate_count;
1769 if (!need_break)
1770 return 0;
1771 } else {
1772 /* Note restart_addr in vma's truncate_count field */
1773 vma->vm_truncate_count = restart_addr;
1774 if (!need_break)
1775 goto again;
1778 spin_unlock(details->i_mmap_lock);
1779 cond_resched();
1780 spin_lock(details->i_mmap_lock);
1781 return -EINTR;
1784 static inline void unmap_mapping_range_tree(struct prio_tree_root *root,
1785 struct zap_details *details)
1787 struct vm_area_struct *vma;
1788 struct prio_tree_iter iter;
1789 pgoff_t vba, vea, zba, zea;
1791 restart:
1792 vma_prio_tree_foreach(vma, &iter, root,
1793 details->first_index, details->last_index) {
1794 /* Skip quickly over those we have already dealt with */
1795 if (vma->vm_truncate_count == details->truncate_count)
1796 continue;
1798 vba = vma->vm_pgoff;
1799 vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1;
1800 /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */
1801 zba = details->first_index;
1802 if (zba < vba)
1803 zba = vba;
1804 zea = details->last_index;
1805 if (zea > vea)
1806 zea = vea;
1808 if (unmap_mapping_range_vma(vma,
1809 ((zba - vba) << PAGE_SHIFT) + vma->vm_start,
1810 ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start,
1811 details) < 0)
1812 goto restart;
1816 static inline void unmap_mapping_range_list(struct list_head *head,
1817 struct zap_details *details)
1819 struct vm_area_struct *vma;
1821 /*
1822 * In nonlinear VMAs there is no correspondence between virtual address
1823 * offset and file offset. So we must perform an exhaustive search
1824 * across *all* the pages in each nonlinear VMA, not just the pages
1825 * whose virtual address lies outside the file truncation point.
1826 */
1827 restart:
1828 list_for_each_entry(vma, head, shared.vm_set.list) {
1829 /* Skip quickly over those we have already dealt with */
1830 if (vma->vm_truncate_count == details->truncate_count)
1831 continue;
1832 details->nonlinear_vma = vma;
1833 if (unmap_mapping_range_vma(vma, vma->vm_start,
1834 vma->vm_end, details) < 0)
1835 goto restart;
1839 /**
1840 * unmap_mapping_range - unmap the portion of all mmaps
1841 * in the specified address_space corresponding to the specified
1842 * page range in the underlying file.
1843 * @mapping: the address space containing mmaps to be unmapped.
1844 * @holebegin: byte in first page to unmap, relative to the start of
1845 * the underlying file. This will be rounded down to a PAGE_SIZE
1846 * boundary. Note that this is different from vmtruncate(), which
1847 * must keep the partial page. In contrast, we must get rid of
1848 * partial pages.
1849 * @holelen: size of prospective hole in bytes. This will be rounded
1850 * up to a PAGE_SIZE boundary. A holelen of zero truncates to the
1851 * end of the file.
1852 * @even_cows: 1 when truncating a file, unmap even private COWed pages;
1853 * but 0 when invalidating pagecache, don't throw away private data.
1854 */
1855 void unmap_mapping_range(struct address_space *mapping,
1856 loff_t const holebegin, loff_t const holelen, int even_cows)
1858 struct zap_details details;
1859 pgoff_t hba = holebegin >> PAGE_SHIFT;
1860 pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1862 /* Check for overflow. */
1863 if (sizeof(holelen) > sizeof(hlen)) {
1864 long long holeend =
1865 (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
1866 if (holeend & ~(long long)ULONG_MAX)
1867 hlen = ULONG_MAX - hba + 1;
1870 details.check_mapping = even_cows? NULL: mapping;
1871 details.nonlinear_vma = NULL;
1872 details.first_index = hba;
1873 details.last_index = hba + hlen - 1;
1874 if (details.last_index < details.first_index)
1875 details.last_index = ULONG_MAX;
1876 details.i_mmap_lock = &mapping->i_mmap_lock;
1878 spin_lock(&mapping->i_mmap_lock);
1880 /* serialize i_size write against truncate_count write */
1881 smp_wmb();
1882 /* Protect against page faults, and endless unmapping loops */
1883 mapping->truncate_count++;
1884 /*
1885 * For archs where spin_lock has inclusive semantics like ia64
1886 * this smp_mb() will prevent to read pagetable contents
1887 * before the truncate_count increment is visible to
1888 * other cpus.
1889 */
1890 smp_mb();
1891 if (unlikely(is_restart_addr(mapping->truncate_count))) {
1892 if (mapping->truncate_count == 0)
1893 reset_vma_truncate_counts(mapping);
1894 mapping->truncate_count++;
1896 details.truncate_count = mapping->truncate_count;
1898 if (unlikely(!prio_tree_empty(&mapping->i_mmap)))
1899 unmap_mapping_range_tree(&mapping->i_mmap, &details);
1900 if (unlikely(!list_empty(&mapping->i_mmap_nonlinear)))
1901 unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details);
1902 spin_unlock(&mapping->i_mmap_lock);
1904 EXPORT_SYMBOL(unmap_mapping_range);
1906 /*
1907 * Handle all mappings that got truncated by a "truncate()"
1908 * system call.
1910 * NOTE! We have to be ready to update the memory sharing
1911 * between the file and the memory map for a potential last
1912 * incomplete page. Ugly, but necessary.
1913 */
1914 int vmtruncate(struct inode * inode, loff_t offset)
1916 struct address_space *mapping = inode->i_mapping;
1917 unsigned long limit;
1919 if (inode->i_size < offset)
1920 goto do_expand;
1921 /*
1922 * truncation of in-use swapfiles is disallowed - it would cause
1923 * subsequent swapout to scribble on the now-freed blocks.
1924 */
1925 if (IS_SWAPFILE(inode))
1926 goto out_busy;
1927 i_size_write(inode, offset);
1928 unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
1929 truncate_inode_pages(mapping, offset);
1930 goto out_truncate;
1932 do_expand:
1933 limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur;
1934 if (limit != RLIM_INFINITY && offset > limit)
1935 goto out_sig;
1936 if (offset > inode->i_sb->s_maxbytes)
1937 goto out_big;
1938 i_size_write(inode, offset);
1940 out_truncate:
1941 if (inode->i_op && inode->i_op->truncate)
1942 inode->i_op->truncate(inode);
1943 return 0;
1944 out_sig:
1945 send_sig(SIGXFSZ, current, 0);
1946 out_big:
1947 return -EFBIG;
1948 out_busy:
1949 return -ETXTBSY;
1951 EXPORT_SYMBOL(vmtruncate);
1953 int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
1955 struct address_space *mapping = inode->i_mapping;
1957 /*
1958 * If the underlying filesystem is not going to provide
1959 * a way to truncate a range of blocks (punch a hole) -
1960 * we should return failure right now.
1961 */
1962 if (!inode->i_op || !inode->i_op->truncate_range)
1963 return -ENOSYS;
1965 mutex_lock(&inode->i_mutex);
1966 down_write(&inode->i_alloc_sem);
1967 unmap_mapping_range(mapping, offset, (end - offset), 1);
1968 truncate_inode_pages_range(mapping, offset, end);
1969 inode->i_op->truncate_range(inode, offset, end);
1970 up_write(&inode->i_alloc_sem);
1971 mutex_unlock(&inode->i_mutex);
1973 return 0;
1975 EXPORT_UNUSED_SYMBOL(vmtruncate_range); /* June 2006 */
1977 /*
1978 * Primitive swap readahead code. We simply read an aligned block of
1979 * (1 << page_cluster) entries in the swap area. This method is chosen
1980 * because it doesn't cost us any seek time. We also make sure to queue
1981 * the 'original' request together with the readahead ones...
1983 * This has been extended to use the NUMA policies from the mm triggering
1984 * the readahead.
1986 * Caller must hold down_read on the vma->vm_mm if vma is not NULL.
1987 */
1988 void swapin_readahead(swp_entry_t entry, unsigned long addr,struct vm_area_struct *vma)
1990 #ifdef CONFIG_NUMA
1991 struct vm_area_struct *next_vma = vma ? vma->vm_next : NULL;
1992 #endif
1993 int i, num;
1994 struct page *new_page;
1995 unsigned long offset;
1997 /*
1998 * Get the number of handles we should do readahead io to.
1999 */
2000 num = valid_swaphandles(entry, &offset);
2001 for (i = 0; i < num; offset++, i++) {
2002 /* Ok, do the async read-ahead now */
2003 new_page = read_swap_cache_async(swp_entry(swp_type(entry),
2004 offset), vma, addr);
2005 if (!new_page)
2006 break;
2007 page_cache_release(new_page);
2008 #ifdef CONFIG_NUMA
2009 /*
2010 * Find the next applicable VMA for the NUMA policy.
2011 */
2012 addr += PAGE_SIZE;
2013 if (addr == 0)
2014 vma = NULL;
2015 if (vma) {
2016 if (addr >= vma->vm_end) {
2017 vma = next_vma;
2018 next_vma = vma ? vma->vm_next : NULL;
2020 if (vma && addr < vma->vm_start)
2021 vma = NULL;
2022 } else {
2023 if (next_vma && addr >= next_vma->vm_start) {
2024 vma = next_vma;
2025 next_vma = vma->vm_next;
2028 #endif
2030 lru_add_drain(); /* Push any new pages onto the LRU now */
2033 /*
2034 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2035 * but allow concurrent faults), and pte mapped but not yet locked.
2036 * We return with mmap_sem still held, but pte unmapped and unlocked.
2037 */
2038 static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma,
2039 unsigned long address, pte_t *page_table, pmd_t *pmd,
2040 int write_access, pte_t orig_pte)
2042 spinlock_t *ptl;
2043 struct page *page;
2044 swp_entry_t entry;
2045 pte_t pte;
2046 int ret = VM_FAULT_MINOR;
2048 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
2049 goto out;
2051 entry = pte_to_swp_entry(orig_pte);
2052 if (is_migration_entry(entry)) {
2053 migration_entry_wait(mm, pmd, address);
2054 goto out;
2056 delayacct_set_flag(DELAYACCT_PF_SWAPIN);
2057 page = lookup_swap_cache(entry);
2058 if (!page) {
2059 swapin_readahead(entry, address, vma);
2060 page = read_swap_cache_async(entry, vma, address);
2061 if (!page) {
2062 /*
2063 * Back out if somebody else faulted in this pte
2064 * while we released the pte lock.
2065 */
2066 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2067 if (likely(pte_same(*page_table, orig_pte)))
2068 ret = VM_FAULT_OOM;
2069 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2070 goto unlock;
2073 /* Had to read the page from swap area: Major fault */
2074 ret = VM_FAULT_MAJOR;
2075 count_vm_event(PGMAJFAULT);
2076 grab_swap_token();
2079 delayacct_clear_flag(DELAYACCT_PF_SWAPIN);
2080 mark_page_accessed(page);
2081 lock_page(page);
2083 /*
2084 * Back out if somebody else already faulted in this pte.
2085 */
2086 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2087 if (unlikely(!pte_same(*page_table, orig_pte)))
2088 goto out_nomap;
2090 if (unlikely(!PageUptodate(page))) {
2091 ret = VM_FAULT_SIGBUS;
2092 goto out_nomap;
2095 /* The page isn't present yet, go ahead with the fault. */
2097 inc_mm_counter(mm, anon_rss);
2098 pte = mk_pte(page, vma->vm_page_prot);
2099 if (write_access && can_share_swap_page(page)) {
2100 pte = maybe_mkwrite(pte_mkdirty(pte), vma);
2101 write_access = 0;
2104 flush_icache_page(vma, page);
2105 set_pte_at(mm, address, page_table, pte);
2106 page_add_anon_rmap(page, vma, address);
2108 swap_free(entry);
2109 if (vm_swap_full())
2110 remove_exclusive_swap_page(page);
2111 unlock_page(page);
2113 if (write_access) {
2114 if (do_wp_page(mm, vma, address,
2115 page_table, pmd, ptl, pte) == VM_FAULT_OOM)
2116 ret = VM_FAULT_OOM;
2117 goto out;
2120 /* No need to invalidate - it was non-present before */
2121 update_mmu_cache(vma, address, pte);
2122 lazy_mmu_prot_update(pte);
2123 unlock:
2124 pte_unmap_unlock(page_table, ptl);
2125 out:
2126 return ret;
2127 out_nomap:
2128 pte_unmap_unlock(page_table, ptl);
2129 unlock_page(page);
2130 page_cache_release(page);
2131 return ret;
2134 /*
2135 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2136 * but allow concurrent faults), and pte mapped but not yet locked.
2137 * We return with mmap_sem still held, but pte unmapped and unlocked.
2138 */
2139 static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma,
2140 unsigned long address, pte_t *page_table, pmd_t *pmd,
2141 int write_access)
2143 struct page *page;
2144 spinlock_t *ptl;
2145 pte_t entry;
2147 if (write_access) {
2148 /* Allocate our own private page. */
2149 pte_unmap(page_table);
2151 if (unlikely(anon_vma_prepare(vma)))
2152 goto oom;
2153 page = alloc_zeroed_user_highpage(vma, address);
2154 if (!page)
2155 goto oom;
2157 entry = mk_pte(page, vma->vm_page_prot);
2158 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2160 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2161 if (!pte_none(*page_table))
2162 goto release;
2163 inc_mm_counter(mm, anon_rss);
2164 lru_cache_add_active(page);
2165 page_add_new_anon_rmap(page, vma, address);
2166 } else {
2167 /* Map the ZERO_PAGE - vm_page_prot is readonly */
2168 page = ZERO_PAGE(address);
2169 page_cache_get(page);
2170 entry = mk_pte(page, vma->vm_page_prot);
2172 ptl = pte_lockptr(mm, pmd);
2173 spin_lock(ptl);
2174 if (!pte_none(*page_table))
2175 goto release;
2176 inc_mm_counter(mm, file_rss);
2177 page_add_file_rmap(page);
2180 set_pte_at(mm, address, page_table, entry);
2182 /* No need to invalidate - it was non-present before */
2183 update_mmu_cache(vma, address, entry);
2184 lazy_mmu_prot_update(entry);
2185 unlock:
2186 pte_unmap_unlock(page_table, ptl);
2187 return VM_FAULT_MINOR;
2188 release:
2189 page_cache_release(page);
2190 goto unlock;
2191 oom:
2192 return VM_FAULT_OOM;
2195 /*
2196 * do_no_page() tries to create a new page mapping. It aggressively
2197 * tries to share with existing pages, but makes a separate copy if
2198 * the "write_access" parameter is true in order to avoid the next
2199 * page fault.
2201 * As this is called only for pages that do not currently exist, we
2202 * do not need to flush old virtual caches or the TLB.
2204 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2205 * but allow concurrent faults), and pte mapped but not yet locked.
2206 * We return with mmap_sem still held, but pte unmapped and unlocked.
2207 */
2208 static int do_no_page(struct mm_struct *mm, struct vm_area_struct *vma,
2209 unsigned long address, pte_t *page_table, pmd_t *pmd,
2210 int write_access)
2212 spinlock_t *ptl;
2213 struct page *new_page;
2214 struct address_space *mapping = NULL;
2215 pte_t entry;
2216 unsigned int sequence = 0;
2217 int ret = VM_FAULT_MINOR;
2218 int anon = 0;
2220 pte_unmap(page_table);
2221 BUG_ON(vma->vm_flags & VM_PFNMAP);
2223 if (vma->vm_file) {
2224 mapping = vma->vm_file->f_mapping;
2225 sequence = mapping->truncate_count;
2226 smp_rmb(); /* serializes i_size against truncate_count */
2228 retry:
2229 new_page = vma->vm_ops->nopage(vma, address & PAGE_MASK, &ret);
2230 /*
2231 * No smp_rmb is needed here as long as there's a full
2232 * spin_lock/unlock sequence inside the ->nopage callback
2233 * (for the pagecache lookup) that acts as an implicit
2234 * smp_mb() and prevents the i_size read to happen
2235 * after the next truncate_count read.
2236 */
2238 /* no page was available -- either SIGBUS or OOM */
2239 if (new_page == NOPAGE_SIGBUS)
2240 return VM_FAULT_SIGBUS;
2241 if (new_page == NOPAGE_OOM)
2242 return VM_FAULT_OOM;
2244 /*
2245 * Should we do an early C-O-W break?
2246 */
2247 if (write_access) {
2248 if (!(vma->vm_flags & VM_SHARED)) {
2249 struct page *page;
2251 if (unlikely(anon_vma_prepare(vma)))
2252 goto oom;
2253 page = alloc_page_vma(GFP_HIGHUSER, vma, address);
2254 if (!page)
2255 goto oom;
2256 copy_user_highpage(page, new_page, address);
2257 page_cache_release(new_page);
2258 new_page = page;
2259 anon = 1;
2261 } else {
2262 /* if the page will be shareable, see if the backing
2263 * address space wants to know that the page is about
2264 * to become writable */
2265 if (vma->vm_ops->page_mkwrite &&
2266 vma->vm_ops->page_mkwrite(vma, new_page) < 0
2267 ) {
2268 page_cache_release(new_page);
2269 return VM_FAULT_SIGBUS;
2274 page_table = pte_offset_map_lock(mm, pmd, address, &ptl);
2275 /*
2276 * For a file-backed vma, someone could have truncated or otherwise
2277 * invalidated this page. If unmap_mapping_range got called,
2278 * retry getting the page.
2279 */
2280 if (mapping && unlikely(sequence != mapping->truncate_count)) {
2281 pte_unmap_unlock(page_table, ptl);
2282 page_cache_release(new_page);
2283 cond_resched();
2284 sequence = mapping->truncate_count;
2285 smp_rmb();
2286 goto retry;
2289 /*
2290 * This silly early PAGE_DIRTY setting removes a race
2291 * due to the bad i386 page protection. But it's valid
2292 * for other architectures too.
2294 * Note that if write_access is true, we either now have
2295 * an exclusive copy of the page, or this is a shared mapping,
2296 * so we can make it writable and dirty to avoid having to
2297 * handle that later.
2298 */
2299 /* Only go through if we didn't race with anybody else... */
2300 if (pte_none(*page_table)) {
2301 flush_icache_page(vma, new_page);
2302 entry = mk_pte(new_page, vma->vm_page_prot);
2303 if (write_access)
2304 entry = maybe_mkwrite(pte_mkdirty(entry), vma);
2305 set_pte_at(mm, address, page_table, entry);
2306 if (anon) {
2307 inc_mm_counter(mm, anon_rss);
2308 lru_cache_add_active(new_page);
2309 page_add_new_anon_rmap(new_page, vma, address);
2310 } else {
2311 inc_mm_counter(mm, file_rss);
2312 page_add_file_rmap(new_page);
2314 } else {
2315 /* One of our sibling threads was faster, back out. */
2316 page_cache_release(new_page);
2317 goto unlock;
2320 /* no need to invalidate: a not-present page shouldn't be cached */
2321 update_mmu_cache(vma, address, entry);
2322 lazy_mmu_prot_update(entry);
2323 unlock:
2324 pte_unmap_unlock(page_table, ptl);
2325 return ret;
2326 oom:
2327 page_cache_release(new_page);
2328 return VM_FAULT_OOM;
2331 /*
2332 * Fault of a previously existing named mapping. Repopulate the pte
2333 * from the encoded file_pte if possible. This enables swappable
2334 * nonlinear vmas.
2336 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2337 * but allow concurrent faults), and pte mapped but not yet locked.
2338 * We return with mmap_sem still held, but pte unmapped and unlocked.
2339 */
2340 static int do_file_page(struct mm_struct *mm, struct vm_area_struct *vma,
2341 unsigned long address, pte_t *page_table, pmd_t *pmd,
2342 int write_access, pte_t orig_pte)
2344 pgoff_t pgoff;
2345 int err;
2347 if (!pte_unmap_same(mm, pmd, page_table, orig_pte))
2348 return VM_FAULT_MINOR;
2350 if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) {
2351 /*
2352 * Page table corrupted: show pte and kill process.
2353 */
2354 print_bad_pte(vma, orig_pte, address);
2355 return VM_FAULT_OOM;
2357 /* We can then assume vm->vm_ops && vma->vm_ops->populate */
2359 pgoff = pte_to_pgoff(orig_pte);
2360 err = vma->vm_ops->populate(vma, address & PAGE_MASK, PAGE_SIZE,
2361 vma->vm_page_prot, pgoff, 0);
2362 if (err == -ENOMEM)
2363 return VM_FAULT_OOM;
2364 if (err)
2365 return VM_FAULT_SIGBUS;
2366 return VM_FAULT_MAJOR;
2369 /*
2370 * These routines also need to handle stuff like marking pages dirty
2371 * and/or accessed for architectures that don't do it in hardware (most
2372 * RISC architectures). The early dirtying is also good on the i386.
2374 * There is also a hook called "update_mmu_cache()" that architectures
2375 * with external mmu caches can use to update those (ie the Sparc or
2376 * PowerPC hashed page tables that act as extended TLBs).
2378 * We enter with non-exclusive mmap_sem (to exclude vma changes,
2379 * but allow concurrent faults), and pte mapped but not yet locked.
2380 * We return with mmap_sem still held, but pte unmapped and unlocked.
2381 */
2382 static inline int handle_pte_fault(struct mm_struct *mm,
2383 struct vm_area_struct *vma, unsigned long address,
2384 pte_t *pte, pmd_t *pmd, int write_access)
2386 pte_t entry;
2387 pte_t old_entry;
2388 spinlock_t *ptl;
2390 old_entry = entry = *pte;
2391 if (!pte_present(entry)) {
2392 if (pte_none(entry)) {
2393 if (!vma->vm_ops || !vma->vm_ops->nopage)
2394 return do_anonymous_page(mm, vma, address,
2395 pte, pmd, write_access);
2396 return do_no_page(mm, vma, address,
2397 pte, pmd, write_access);
2399 if (pte_file(entry))
2400 return do_file_page(mm, vma, address,
2401 pte, pmd, write_access, entry);
2402 return do_swap_page(mm, vma, address,
2403 pte, pmd, write_access, entry);
2406 ptl = pte_lockptr(mm, pmd);
2407 spin_lock(ptl);
2408 if (unlikely(!pte_same(*pte, entry)))
2409 goto unlock;
2410 if (write_access) {
2411 if (!pte_write(entry))
2412 return do_wp_page(mm, vma, address,
2413 pte, pmd, ptl, entry);
2414 entry = pte_mkdirty(entry);
2416 entry = pte_mkyoung(entry);
2417 if (!pte_same(old_entry, entry)) {
2418 ptep_set_access_flags(vma, address, pte, entry, write_access);
2419 update_mmu_cache(vma, address, entry);
2420 lazy_mmu_prot_update(entry);
2421 } else {
2422 /*
2423 * This is needed only for protection faults but the arch code
2424 * is not yet telling us if this is a protection fault or not.
2425 * This still avoids useless tlb flushes for .text page faults
2426 * with threads.
2427 */
2428 if (write_access)
2429 flush_tlb_page(vma, address);
2431 unlock:
2432 pte_unmap_unlock(pte, ptl);
2433 return VM_FAULT_MINOR;
2436 /*
2437 * By the time we get here, we already hold the mm semaphore
2438 */
2439 int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
2440 unsigned long address, int write_access)
2442 pgd_t *pgd;
2443 pud_t *pud;
2444 pmd_t *pmd;
2445 pte_t *pte;
2447 __set_current_state(TASK_RUNNING);
2449 count_vm_event(PGFAULT);
2451 if (unlikely(is_vm_hugetlb_page(vma)))
2452 return hugetlb_fault(mm, vma, address, write_access);
2454 pgd = pgd_offset(mm, address);
2455 pud = pud_alloc(mm, pgd, address);
2456 if (!pud)
2457 return VM_FAULT_OOM;
2458 pmd = pmd_alloc(mm, pud, address);
2459 if (!pmd)
2460 return VM_FAULT_OOM;
2461 pte = pte_alloc_map(mm, pmd, address);
2462 if (!pte)
2463 return VM_FAULT_OOM;
2465 return handle_pte_fault(mm, vma, address, pte, pmd, write_access);
2468 EXPORT_SYMBOL_GPL(__handle_mm_fault);
2470 #ifndef __PAGETABLE_PUD_FOLDED
2471 /*
2472 * Allocate page upper directory.
2473 * We've already handled the fast-path in-line.
2474 */
2475 int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
2477 pud_t *new = pud_alloc_one(mm, address);
2478 if (!new)
2479 return -ENOMEM;
2481 spin_lock(&mm->page_table_lock);
2482 if (pgd_present(*pgd)) /* Another has populated it */
2483 pud_free(new);
2484 else
2485 pgd_populate(mm, pgd, new);
2486 spin_unlock(&mm->page_table_lock);
2487 return 0;
2489 #else
2490 /* Workaround for gcc 2.96 */
2491 int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address)
2493 return 0;
2495 #endif /* __PAGETABLE_PUD_FOLDED */
2497 #ifndef __PAGETABLE_PMD_FOLDED
2498 /*
2499 * Allocate page middle directory.
2500 * We've already handled the fast-path in-line.
2501 */
2502 int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
2504 pmd_t *new = pmd_alloc_one(mm, address);
2505 if (!new)
2506 return -ENOMEM;
2508 spin_lock(&mm->page_table_lock);
2509 #ifndef __ARCH_HAS_4LEVEL_HACK
2510 if (pud_present(*pud)) /* Another has populated it */
2511 pmd_free(new);
2512 else
2513 pud_populate(mm, pud, new);
2514 #else
2515 if (pgd_present(*pud)) /* Another has populated it */
2516 pmd_free(new);
2517 else
2518 pgd_populate(mm, pud, new);
2519 #endif /* __ARCH_HAS_4LEVEL_HACK */
2520 spin_unlock(&mm->page_table_lock);
2521 return 0;
2523 #else
2524 /* Workaround for gcc 2.96 */
2525 int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address)
2527 return 0;
2529 #endif /* __PAGETABLE_PMD_FOLDED */
2531 int make_pages_present(unsigned long addr, unsigned long end)
2533 int ret, len, write;
2534 struct vm_area_struct * vma;
2536 vma = find_vma(current->mm, addr);
2537 if (!vma)
2538 return -1;
2539 write = (vma->vm_flags & VM_WRITE) != 0;
2540 BUG_ON(addr >= end);
2541 BUG_ON(end > vma->vm_end);
2542 len = (end+PAGE_SIZE-1)/PAGE_SIZE-addr/PAGE_SIZE;
2543 ret = get_user_pages(current, current->mm, addr,
2544 len, write, 0, NULL, NULL);
2545 if (ret < 0)
2546 return ret;
2547 return ret == len ? 0 : -1;
2550 /*
2551 * Map a vmalloc()-space virtual address to the physical page.
2552 */
2553 struct page * vmalloc_to_page(void * vmalloc_addr)
2555 unsigned long addr = (unsigned long) vmalloc_addr;
2556 struct page *page = NULL;
2557 pgd_t *pgd = pgd_offset_k(addr);
2558 pud_t *pud;
2559 pmd_t *pmd;
2560 pte_t *ptep, pte;
2562 if (!pgd_none(*pgd)) {
2563 pud = pud_offset(pgd, addr);
2564 if (!pud_none(*pud)) {
2565 pmd = pmd_offset(pud, addr);
2566 if (!pmd_none(*pmd)) {
2567 ptep = pte_offset_map(pmd, addr);
2568 pte = *ptep;
2569 if (pte_present(pte))
2570 page = pte_page(pte);
2571 pte_unmap(ptep);
2575 return page;
2578 EXPORT_SYMBOL(vmalloc_to_page);
2580 /*
2581 * Map a vmalloc()-space virtual address to the physical page frame number.
2582 */
2583 unsigned long vmalloc_to_pfn(void * vmalloc_addr)
2585 return page_to_pfn(vmalloc_to_page(vmalloc_addr));
2588 EXPORT_SYMBOL(vmalloc_to_pfn);
2590 #if !defined(__HAVE_ARCH_GATE_AREA)
2592 #if defined(AT_SYSINFO_EHDR)
2593 static struct vm_area_struct gate_vma;
2595 static int __init gate_vma_init(void)
2597 gate_vma.vm_mm = NULL;
2598 gate_vma.vm_start = FIXADDR_USER_START;
2599 gate_vma.vm_end = FIXADDR_USER_END;
2600 gate_vma.vm_page_prot = PAGE_READONLY;
2601 gate_vma.vm_flags = 0;
2602 return 0;
2604 __initcall(gate_vma_init);
2605 #endif
2607 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
2609 #ifdef AT_SYSINFO_EHDR
2610 return &gate_vma;
2611 #else
2612 return NULL;
2613 #endif
2616 int in_gate_area_no_task(unsigned long addr)
2618 #ifdef AT_SYSINFO_EHDR
2619 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
2620 return 1;
2621 #endif
2622 return 0;
2625 #endif /* __HAVE_ARCH_GATE_AREA */