Root/
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> |
50 | #include <linux/delayacct.h> |
51 | #include <linux/init.h> |
52 | #include <linux/writeback.h> |
53 | #include <linux/memcontrol.h> |
54 | #include <linux/mmu_notifier.h> |
55 | #include <linux/kallsyms.h> |
56 | #include <linux/swapops.h> |
57 | #include <linux/elf.h> |
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 | |
65 | #include "internal.h" |
66 | |
67 | #ifndef CONFIG_NEED_MULTIPLE_NODES |
68 | /* use the per-pgdat data instead for discontigmem - mbligh */ |
69 | unsigned long max_mapnr; |
70 | struct page *mem_map; |
71 | |
72 | EXPORT_SYMBOL(max_mapnr); |
73 | EXPORT_SYMBOL(mem_map); |
74 | #endif |
75 | |
76 | unsigned 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 | */ |
84 | void * high_memory; |
85 | |
86 | EXPORT_SYMBOL(num_physpages); |
87 | EXPORT_SYMBOL(high_memory); |
88 | |
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 | */ |
95 | int randomize_va_space __read_mostly = |
96 | #ifdef CONFIG_COMPAT_BRK |
97 | 1; |
98 | #else |
99 | 2; |
100 | #endif |
101 | |
102 | static int __init disable_randmaps(char *s) |
103 | { |
104 | randomize_va_space = 0; |
105 | return 1; |
106 | } |
107 | __setup("norandmaps", disable_randmaps); |
108 | |
109 | |
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 | |
116 | void pgd_clear_bad(pgd_t *pgd) |
117 | { |
118 | pgd_ERROR(*pgd); |
119 | pgd_clear(pgd); |
120 | } |
121 | |
122 | void pud_clear_bad(pud_t *pud) |
123 | { |
124 | pud_ERROR(*pud); |
125 | pud_clear(pud); |
126 | } |
127 | |
128 | void 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 | */ |
138 | static void free_pte_range(struct mmu_gather *tlb, pmd_t *pmd, |
139 | unsigned long addr) |
140 | { |
141 | pgtable_t token = pmd_pgtable(*pmd); |
142 | pmd_clear(pmd); |
143 | pte_free_tlb(tlb, token, addr); |
144 | tlb->mm->nr_ptes--; |
145 | } |
146 | |
147 | static 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) |
150 | { |
151 | pmd_t *pmd; |
152 | unsigned long next; |
153 | unsigned long start; |
154 | |
155 | start = addr; |
156 | pmd = pmd_offset(pud, addr); |
157 | do { |
158 | next = pmd_addr_end(addr, end); |
159 | if (pmd_none_or_clear_bad(pmd)) |
160 | continue; |
161 | free_pte_range(tlb, pmd, addr); |
162 | } while (pmd++, addr = next, addr != end); |
163 | |
164 | start &= PUD_MASK; |
165 | if (start < floor) |
166 | return; |
167 | if (ceiling) { |
168 | ceiling &= PUD_MASK; |
169 | if (!ceiling) |
170 | return; |
171 | } |
172 | if (end - 1 > ceiling - 1) |
173 | return; |
174 | |
175 | pmd = pmd_offset(pud, start); |
176 | pud_clear(pud); |
177 | pmd_free_tlb(tlb, pmd, start); |
178 | } |
179 | |
180 | static 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) |
183 | { |
184 | pud_t *pud; |
185 | unsigned long next; |
186 | unsigned long start; |
187 | |
188 | start = addr; |
189 | pud = pud_offset(pgd, addr); |
190 | do { |
191 | next = pud_addr_end(addr, end); |
192 | if (pud_none_or_clear_bad(pud)) |
193 | continue; |
194 | free_pmd_range(tlb, pud, addr, next, floor, ceiling); |
195 | } while (pud++, addr = next, addr != end); |
196 | |
197 | start &= PGDIR_MASK; |
198 | if (start < floor) |
199 | return; |
200 | if (ceiling) { |
201 | ceiling &= PGDIR_MASK; |
202 | if (!ceiling) |
203 | return; |
204 | } |
205 | if (end - 1 > ceiling - 1) |
206 | return; |
207 | |
208 | pud = pud_offset(pgd, start); |
209 | pgd_clear(pgd); |
210 | pud_free_tlb(tlb, pud, start); |
211 | } |
212 | |
213 | /* |
214 | * This function frees user-level page tables of a process. |
215 | * |
216 | * Must be called with pagetable lock held. |
217 | */ |
218 | void free_pgd_range(struct mmu_gather *tlb, |
219 | unsigned long addr, unsigned long end, |
220 | unsigned long floor, unsigned long ceiling) |
221 | { |
222 | pgd_t *pgd; |
223 | unsigned long next; |
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 | */ |
251 | |
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; |
269 | pgd = pgd_offset(tlb->mm, addr); |
270 | do { |
271 | next = pgd_addr_end(addr, end); |
272 | if (pgd_none_or_clear_bad(pgd)) |
273 | continue; |
274 | free_pud_range(tlb, pgd, addr, next, floor, ceiling); |
275 | } while (pgd++, addr = next, addr != end); |
276 | } |
277 | |
278 | void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma, |
279 | unsigned long floor, unsigned long ceiling) |
280 | { |
281 | while (vma) { |
282 | struct vm_area_struct *next = vma->vm_next; |
283 | unsigned long addr = vma->vm_start; |
284 | |
285 | /* |
286 | * Hide vma from rmap and vmtruncate before freeing pgtables |
287 | */ |
288 | anon_vma_unlink(vma); |
289 | unlink_file_vma(vma); |
290 | |
291 | if (is_vm_hugetlb_page(vma)) { |
292 | hugetlb_free_pgd_range(tlb, addr, vma->vm_end, |
293 | floor, next? next->vm_start: ceiling); |
294 | } else { |
295 | /* |
296 | * Optimization: gather nearby vmas into one call down |
297 | */ |
298 | while (next && next->vm_start <= vma->vm_end + PMD_SIZE |
299 | && !is_vm_hugetlb_page(next)) { |
300 | vma = next; |
301 | next = vma->vm_next; |
302 | anon_vma_unlink(vma); |
303 | unlink_file_vma(vma); |
304 | } |
305 | free_pgd_range(tlb, addr, vma->vm_end, |
306 | floor, next? next->vm_start: ceiling); |
307 | } |
308 | vma = next; |
309 | } |
310 | } |
311 | |
312 | int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address) |
313 | { |
314 | pgtable_t new = pte_alloc_one(mm, address); |
315 | if (!new) |
316 | return -ENOMEM; |
317 | |
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 | |
333 | spin_lock(&mm->page_table_lock); |
334 | if (!pmd_present(*pmd)) { /* Has another populated it ? */ |
335 | mm->nr_ptes++; |
336 | pmd_populate(mm, pmd, new); |
337 | new = NULL; |
338 | } |
339 | spin_unlock(&mm->page_table_lock); |
340 | if (new) |
341 | pte_free(mm, new); |
342 | return 0; |
343 | } |
344 | |
345 | int __pte_alloc_kernel(pmd_t *pmd, unsigned long address) |
346 | { |
347 | pte_t *new = pte_alloc_one_kernel(&init_mm, address); |
348 | if (!new) |
349 | return -ENOMEM; |
350 | |
351 | smp_wmb(); /* See comment in __pte_alloc */ |
352 | |
353 | spin_lock(&init_mm.page_table_lock); |
354 | if (!pmd_present(*pmd)) { /* Has another populated it ? */ |
355 | pmd_populate_kernel(&init_mm, pmd, new); |
356 | new = NULL; |
357 | } |
358 | spin_unlock(&init_mm.page_table_lock); |
359 | if (new) |
360 | pte_free_kernel(&init_mm, new); |
361 | return 0; |
362 | } |
363 | |
364 | static 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 | |
372 | /* |
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. |
376 | * |
377 | * The calling function must still handle the error. |
378 | */ |
379 | static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr, |
380 | pte_t pte, struct page *page) |
381 | { |
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; |
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) { |
401 | printk(KERN_ALERT |
402 | "BUG: Bad page map: %lu messages suppressed\n", |
403 | nr_unshown); |
404 | nr_unshown = 0; |
405 | } |
406 | nr_shown = 0; |
407 | } |
408 | if (nr_shown++ == 0) |
409 | resume = jiffies + 60 * HZ; |
410 | |
411 | mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL; |
412 | index = linear_page_index(vma, addr); |
413 | |
414 | printk(KERN_ALERT |
415 | "BUG: Bad page map in process %s pte:%08llx pmd:%08llx\n", |
416 | current->comm, |
417 | (long long)pte_val(pte), (long long)pmd_val(*pmd)); |
418 | if (page) { |
419 | printk(KERN_ALERT |
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 | } |
424 | printk(KERN_ALERT |
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) |
431 | print_symbol(KERN_ALERT "vma->vm_ops->fault: %s\n", |
432 | (unsigned long)vma->vm_ops->fault); |
433 | if (vma->vm_file && vma->vm_file->f_op) |
434 | print_symbol(KERN_ALERT "vma->vm_file->f_op->mmap: %s\n", |
435 | (unsigned long)vma->vm_file->f_op->mmap); |
436 | dump_stack(); |
437 | add_taint(TAINT_BAD_PAGE); |
438 | } |
439 | |
440 | static inline int is_cow_mapping(unsigned int flags) |
441 | { |
442 | return (flags & (VM_SHARED | VM_MAYWRITE)) == VM_MAYWRITE; |
443 | } |
444 | |
445 | /* |
446 | * vm_normal_page -- This function gets the "struct page" associated with a pte. |
447 | * |
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. |
451 | * |
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. |
460 | * |
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 |
463 | * set, and the vm_pgoff will point to the first PFN mapped: thus every special |
464 | * mapping will always honor the rule |
465 | * |
466 | * pfn_of_page == vma->vm_pgoff + ((addr - vma->vm_start) >> PAGE_SHIFT) |
467 | * |
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). |
474 | * |
475 | * |
476 | * In order to support COW of arbitrary special mappings, we have VM_MIXEDMAP. |
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 | * |
486 | */ |
487 | #ifdef __HAVE_ARCH_PTE_SPECIAL |
488 | # define HAVE_PTE_SPECIAL 1 |
489 | #else |
490 | # define HAVE_PTE_SPECIAL 0 |
491 | #endif |
492 | struct page *vm_normal_page(struct vm_area_struct *vma, unsigned long addr, |
493 | pte_t pte) |
494 | { |
495 | unsigned long pfn = pte_pfn(pte); |
496 | |
497 | if (HAVE_PTE_SPECIAL) { |
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); |
502 | return NULL; |
503 | } |
504 | |
505 | /* !HAVE_PTE_SPECIAL case follows: */ |
506 | |
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 { |
513 | unsigned long off; |
514 | off = (addr - vma->vm_start) >> PAGE_SHIFT; |
515 | if (pfn == vma->vm_pgoff + off) |
516 | return NULL; |
517 | if (!is_cow_mapping(vma->vm_flags)) |
518 | return NULL; |
519 | } |
520 | } |
521 | |
522 | check_pfn: |
523 | if (unlikely(pfn > highest_memmap_pfn)) { |
524 | print_bad_pte(vma, addr, pte, NULL); |
525 | return NULL; |
526 | } |
527 | |
528 | /* |
529 | * NOTE! We still have PageReserved() pages in the page tables. |
530 | * eg. VDSO mappings can cause them to exist. |
531 | */ |
532 | out: |
533 | return pfn_to_page(pfn); |
534 | } |
535 | |
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. |
540 | */ |
541 | |
542 | static inline void |
543 | copy_one_pte(struct mm_struct *dst_mm, struct mm_struct *src_mm, |
544 | pte_t *dst_pte, pte_t *src_pte, struct vm_area_struct *vma, |
545 | unsigned long addr, int *rss) |
546 | { |
547 | unsigned long vm_flags = vma->vm_flags; |
548 | pte_t pte = *src_pte; |
549 | struct page *page; |
550 | |
551 | /* pte contains position in swap or file, so copy. */ |
552 | if (unlikely(!pte_present(pte))) { |
553 | if (!pte_file(pte)) { |
554 | swp_entry_t entry = pte_to_swp_entry(pte); |
555 | |
556 | swap_duplicate(entry); |
557 | /* make sure dst_mm is on swapoff's mmlist. */ |
558 | if (unlikely(list_empty(&dst_mm->mmlist))) { |
559 | spin_lock(&mmlist_lock); |
560 | if (list_empty(&dst_mm->mmlist)) |
561 | list_add(&dst_mm->mmlist, |
562 | &src_mm->mmlist); |
563 | spin_unlock(&mmlist_lock); |
564 | } |
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 | } |
575 | } |
576 | goto out_set_pte; |
577 | } |
578 | |
579 | /* |
580 | * If it's a COW mapping, write protect it both |
581 | * in the parent and the child |
582 | */ |
583 | if (is_cow_mapping(vm_flags)) { |
584 | ptep_set_wrprotect(src_mm, addr, src_pte); |
585 | pte = pte_wrprotect(pte); |
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); |
595 | |
596 | page = vm_normal_page(vma, addr, pte); |
597 | if (page) { |
598 | get_page(page); |
599 | page_dup_rmap(page, vma, addr); |
600 | rss[!!PageAnon(page)]++; |
601 | } |
602 | |
603 | out_set_pte: |
604 | set_pte_at(dst_mm, addr, dst_pte, pte); |
605 | } |
606 | |
607 | static 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; |
612 | spinlock_t *src_ptl, *dst_ptl; |
613 | int progress = 0; |
614 | int rss[2]; |
615 | |
616 | again: |
617 | rss[1] = rss[0] = 0; |
618 | dst_pte = pte_alloc_map_lock(dst_mm, dst_pmd, addr, &dst_ptl); |
619 | if (!dst_pte) |
620 | return -ENOMEM; |
621 | src_pte = pte_offset_map_nested(src_pmd, addr); |
622 | src_ptl = pte_lockptr(src_mm, src_pmd); |
623 | spin_lock_nested(src_ptl, SINGLE_DEPTH_NESTING); |
624 | arch_enter_lazy_mmu_mode(); |
625 | |
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 | */ |
631 | if (progress >= 32) { |
632 | progress = 0; |
633 | if (need_resched() || |
634 | spin_needbreak(src_ptl) || spin_needbreak(dst_ptl)) |
635 | break; |
636 | } |
637 | if (pte_none(*src_pte)) { |
638 | progress++; |
639 | continue; |
640 | } |
641 | copy_one_pte(dst_mm, src_mm, dst_pte, src_pte, vma, addr, rss); |
642 | progress += 8; |
643 | } while (dst_pte++, src_pte++, addr += PAGE_SIZE, addr != end); |
644 | |
645 | arch_leave_lazy_mmu_mode(); |
646 | spin_unlock(src_ptl); |
647 | pte_unmap_nested(src_pte - 1); |
648 | add_mm_rss(dst_mm, rss[0], rss[1]); |
649 | pte_unmap_unlock(dst_pte - 1, dst_ptl); |
650 | cond_resched(); |
651 | if (addr != end) |
652 | goto again; |
653 | return 0; |
654 | } |
655 | |
656 | static 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 | |
678 | static 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 | |
700 | int 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; |
707 | int ret; |
708 | |
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 | */ |
715 | if (!(vma->vm_flags & (VM_HUGETLB|VM_NONLINEAR|VM_PFNMAP|VM_INSERTPAGE))) { |
716 | if (!vma->anon_vma) |
717 | return 0; |
718 | } |
719 | |
720 | if (is_vm_hugetlb_page(vma)) |
721 | return copy_hugetlb_page_range(dst_mm, src_mm, vma); |
722 | |
723 | if (unlikely(is_pfn_mapping(vma))) { |
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 | |
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; |
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; |
749 | if (unlikely(copy_pud_range(dst_mm, src_mm, dst_pgd, src_pgd, |
750 | vma, addr, next))) { |
751 | ret = -ENOMEM; |
752 | break; |
753 | } |
754 | } while (dst_pgd++, src_pgd++, addr = next, addr != end); |
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; |
760 | } |
761 | |
762 | static unsigned long zap_pte_range(struct mmu_gather *tlb, |
763 | struct vm_area_struct *vma, pmd_t *pmd, |
764 | unsigned long addr, unsigned long end, |
765 | long *zap_work, struct zap_details *details) |
766 | { |
767 | struct mm_struct *mm = tlb->mm; |
768 | pte_t *pte; |
769 | spinlock_t *ptl; |
770 | int file_rss = 0; |
771 | int anon_rss = 0; |
772 | |
773 | pte = pte_offset_map_lock(mm, pmd, addr, &ptl); |
774 | arch_enter_lazy_mmu_mode(); |
775 | do { |
776 | pte_t ptent = *pte; |
777 | if (pte_none(ptent)) { |
778 | (*zap_work)--; |
779 | continue; |
780 | } |
781 | |
782 | (*zap_work) -= PAGE_SIZE; |
783 | |
784 | if (pte_present(ptent)) { |
785 | struct page *page; |
786 | |
787 | page = vm_normal_page(vma, addr, ptent); |
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 | } |
806 | ptent = ptep_get_and_clear_full(mm, addr, pte, |
807 | tlb->fullmm); |
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) |
814 | set_pte_at(mm, addr, pte, |
815 | pgoff_to_pte(page->index)); |
816 | if (PageAnon(page)) |
817 | anon_rss--; |
818 | else { |
819 | if (pte_dirty(ptent)) |
820 | set_page_dirty(page); |
821 | if (pte_young(ptent) && |
822 | likely(!VM_SequentialReadHint(vma))) |
823 | mark_page_accessed(page); |
824 | file_rss--; |
825 | } |
826 | page_remove_rmap(page); |
827 | if (unlikely(page_mapcount(page) < 0)) |
828 | print_bad_pte(vma, addr, ptent, page); |
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; |
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); |
844 | pte_clear_not_present_full(mm, addr, pte, tlb->fullmm); |
845 | } while (pte++, addr += PAGE_SIZE, (addr != end && *zap_work > 0)); |
846 | |
847 | add_mm_rss(mm, file_rss, anon_rss); |
848 | arch_leave_lazy_mmu_mode(); |
849 | pte_unmap_unlock(pte - 1, ptl); |
850 | |
851 | return addr; |
852 | } |
853 | |
854 | static inline unsigned long zap_pmd_range(struct mmu_gather *tlb, |
855 | struct vm_area_struct *vma, pud_t *pud, |
856 | unsigned long addr, unsigned long end, |
857 | long *zap_work, struct zap_details *details) |
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); |
865 | if (pmd_none_or_clear_bad(pmd)) { |
866 | (*zap_work)--; |
867 | continue; |
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; |
874 | } |
875 | |
876 | static inline unsigned long zap_pud_range(struct mmu_gather *tlb, |
877 | struct vm_area_struct *vma, pgd_t *pgd, |
878 | unsigned long addr, unsigned long end, |
879 | long *zap_work, struct zap_details *details) |
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); |
887 | if (pud_none_or_clear_bad(pud)) { |
888 | (*zap_work)--; |
889 | continue; |
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; |
896 | } |
897 | |
898 | static unsigned long unmap_page_range(struct mmu_gather *tlb, |
899 | struct vm_area_struct *vma, |
900 | unsigned long addr, unsigned long end, |
901 | long *zap_work, struct zap_details *details) |
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); |
914 | if (pgd_none_or_clear_bad(pgd)) { |
915 | (*zap_work)--; |
916 | continue; |
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)); |
921 | tlb_end_vma(tlb, vma); |
922 | |
923 | return addr; |
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 |
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 | * |
942 | * Returns the end address of the unmapping (restart addr if interrupted). |
943 | * |
944 | * Unmap all pages in the vma list. |
945 | * |
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 |
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 | */ |
959 | unsigned long unmap_vmas(struct mmu_gather **tlbp, |
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 | { |
964 | long zap_work = ZAP_BLOCK_SIZE; |
965 | unsigned long tlb_start = 0; /* For tlb_finish_mmu */ |
966 | int tlb_start_valid = 0; |
967 | unsigned long start = start_addr; |
968 | spinlock_t *i_mmap_lock = details? details->i_mmap_lock: NULL; |
969 | int fullmm = (*tlbp)->fullmm; |
970 | struct mm_struct *mm = vma->vm_mm; |
971 | |
972 | mmu_notifier_invalidate_range_start(mm, start_addr, end_addr); |
973 | for ( ; vma && vma->vm_start < end_addr; vma = vma->vm_next) { |
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 | |
986 | if (unlikely(is_pfn_mapping(vma))) |
987 | untrack_pfn_vma(vma, 0, 0); |
988 | |
989 | while (start != end) { |
990 | if (!tlb_start_valid) { |
991 | tlb_start = start; |
992 | tlb_start_valid = 1; |
993 | } |
994 | |
995 | if (unlikely(is_vm_hugetlb_page(vma))) { |
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) / |
1010 | pages_per_huge_page(hstate_vma(vma)); |
1011 | } |
1012 | |
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; |
1021 | } |
1022 | |
1023 | tlb_finish_mmu(*tlbp, tlb_start, start); |
1024 | |
1025 | if (need_resched() || |
1026 | (i_mmap_lock && spin_needbreak(i_mmap_lock))) { |
1027 | if (i_mmap_lock) { |
1028 | *tlbp = NULL; |
1029 | goto out; |
1030 | } |
1031 | cond_resched(); |
1032 | } |
1033 | |
1034 | *tlbp = tlb_gather_mmu(vma->vm_mm, fullmm); |
1035 | tlb_start_valid = 0; |
1036 | zap_work = ZAP_BLOCK_SIZE; |
1037 | } |
1038 | } |
1039 | out: |
1040 | mmu_notifier_invalidate_range_end(mm, start_addr, end_addr); |
1041 | return start; /* which is now the end (or restart) address */ |
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 | */ |
1051 | unsigned long zap_page_range(struct vm_area_struct *vma, unsigned long address, |
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 | |
1059 | lru_add_drain(); |
1060 | tlb = tlb_gather_mmu(mm, 0); |
1061 | update_hiwater_rss(mm); |
1062 | end = unmap_vmas(&tlb, vma, address, end, &nr_accounted, details); |
1063 | if (tlb) |
1064 | tlb_finish_mmu(tlb, address, end); |
1065 | return end; |
1066 | } |
1067 | EXPORT_SYMBOL_GPL(zap_page_range); |
1068 | |
1069 | /** |
1070 | * zap_vma_ptes - remove ptes mapping the vma |
1071 | * @vma: vm_area_struct holding ptes to be zapped |
1072 | * @address: starting address of pages to zap |
1073 | * @size: number of bytes to zap |
1074 | * |
1075 | * This function only unmaps ptes assigned to VM_PFNMAP vmas. |
1076 | * |
1077 | * The entire address range must be fully contained within the vma. |
1078 | * |
1079 | * Returns 0 if successful. |
1080 | */ |
1081 | int zap_vma_ptes(struct vm_area_struct *vma, unsigned long address, |
1082 | unsigned long size) |
1083 | { |
1084 | if (address < vma->vm_start || address + size > vma->vm_end || |
1085 | !(vma->vm_flags & VM_PFNMAP)) |
1086 | return -1; |
1087 | zap_page_range(vma, address, size, NULL); |
1088 | return 0; |
1089 | } |
1090 | EXPORT_SYMBOL_GPL(zap_vma_ptes); |
1091 | |
1092 | /* |
1093 | * Do a quick page-table lookup for a single page. |
1094 | */ |
1095 | struct page *follow_page(struct vm_area_struct *vma, unsigned long address, |
1096 | unsigned int flags) |
1097 | { |
1098 | pgd_t *pgd; |
1099 | pud_t *pud; |
1100 | pmd_t *pmd; |
1101 | pte_t *ptep, pte; |
1102 | spinlock_t *ptl; |
1103 | struct page *page; |
1104 | struct mm_struct *mm = vma->vm_mm; |
1105 | |
1106 | page = follow_huge_addr(mm, address, flags & FOLL_WRITE); |
1107 | if (!IS_ERR(page)) { |
1108 | BUG_ON(flags & FOLL_GET); |
1109 | goto out; |
1110 | } |
1111 | |
1112 | page = NULL; |
1113 | pgd = pgd_offset(mm, address); |
1114 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) |
1115 | goto no_page_table; |
1116 | |
1117 | pud = pud_offset(pgd, address); |
1118 | if (pud_none(*pud)) |
1119 | goto no_page_table; |
1120 | if (pud_huge(*pud)) { |
1121 | BUG_ON(flags & FOLL_GET); |
1122 | page = follow_huge_pud(mm, address, pud, flags & FOLL_WRITE); |
1123 | goto out; |
1124 | } |
1125 | if (unlikely(pud_bad(*pud))) |
1126 | goto no_page_table; |
1127 | |
1128 | pmd = pmd_offset(pud, address); |
1129 | if (pmd_none(*pmd)) |
1130 | goto no_page_table; |
1131 | if (pmd_huge(*pmd)) { |
1132 | BUG_ON(flags & FOLL_GET); |
1133 | page = follow_huge_pmd(mm, address, pmd, flags & FOLL_WRITE); |
1134 | goto out; |
1135 | } |
1136 | if (unlikely(pmd_bad(*pmd))) |
1137 | goto no_page_table; |
1138 | |
1139 | ptep = pte_offset_map_lock(mm, pmd, address, &ptl); |
1140 | |
1141 | pte = *ptep; |
1142 | if (!pte_present(pte)) |
1143 | goto no_page; |
1144 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
1145 | goto unlock; |
1146 | page = vm_normal_page(vma, address, pte); |
1147 | if (unlikely(!page)) |
1148 | goto bad_page; |
1149 | |
1150 | if (flags & FOLL_GET) |
1151 | get_page(page); |
1152 | if (flags & FOLL_TOUCH) { |
1153 | if ((flags & FOLL_WRITE) && |
1154 | !pte_dirty(pte) && !PageDirty(page)) |
1155 | set_page_dirty(page); |
1156 | /* |
1157 | * pte_mkyoung() would be more correct here, but atomic care |
1158 | * is needed to avoid losing the dirty bit: it is easier to use |
1159 | * mark_page_accessed(). |
1160 | */ |
1161 | mark_page_accessed(page); |
1162 | } |
1163 | unlock: |
1164 | pte_unmap_unlock(ptep, ptl); |
1165 | out: |
1166 | return page; |
1167 | |
1168 | bad_page: |
1169 | pte_unmap_unlock(ptep, ptl); |
1170 | return ERR_PTR(-EFAULT); |
1171 | |
1172 | no_page: |
1173 | pte_unmap_unlock(ptep, ptl); |
1174 | if (!pte_none(pte)) |
1175 | return page; |
1176 | /* Fall through to ZERO_PAGE handling */ |
1177 | no_page_table: |
1178 | /* |
1179 | * When core dumping an enormous anonymous area that nobody |
1180 | * has touched so far, we don't want to allocate page tables. |
1181 | */ |
1182 | if (flags & FOLL_ANON) { |
1183 | page = ZERO_PAGE(0); |
1184 | if (flags & FOLL_GET) |
1185 | get_page(page); |
1186 | BUG_ON(flags & FOLL_WRITE); |
1187 | } |
1188 | return page; |
1189 | } |
1190 | |
1191 | /* Can we do the FOLL_ANON optimization? */ |
1192 | static inline int use_zero_page(struct vm_area_struct *vma) |
1193 | { |
1194 | /* |
1195 | * We don't want to optimize FOLL_ANON for make_pages_present() |
1196 | * when it tries to page in a VM_LOCKED region. As to VM_SHARED, |
1197 | * we want to get the page from the page tables to make sure |
1198 | * that we serialize and update with any other user of that |
1199 | * mapping. |
1200 | */ |
1201 | if (vma->vm_flags & (VM_LOCKED | VM_SHARED)) |
1202 | return 0; |
1203 | /* |
1204 | * And if we have a fault routine, it's not an anonymous region. |
1205 | */ |
1206 | return !vma->vm_ops || !vma->vm_ops->fault; |
1207 | } |
1208 | |
1209 | |
1210 | |
1211 | int __get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
1212 | unsigned long start, int nr_pages, int flags, |
1213 | struct page **pages, struct vm_area_struct **vmas) |
1214 | { |
1215 | int i; |
1216 | unsigned int vm_flags = 0; |
1217 | int write = !!(flags & GUP_FLAGS_WRITE); |
1218 | int force = !!(flags & GUP_FLAGS_FORCE); |
1219 | int ignore = !!(flags & GUP_FLAGS_IGNORE_VMA_PERMISSIONS); |
1220 | int ignore_sigkill = !!(flags & GUP_FLAGS_IGNORE_SIGKILL); |
1221 | |
1222 | if (nr_pages <= 0) |
1223 | return 0; |
1224 | /* |
1225 | * Require read or write permissions. |
1226 | * If 'force' is set, we only require the "MAY" flags. |
1227 | */ |
1228 | vm_flags = write ? (VM_WRITE | VM_MAYWRITE) : (VM_READ | VM_MAYREAD); |
1229 | vm_flags &= force ? (VM_MAYREAD | VM_MAYWRITE) : (VM_READ | VM_WRITE); |
1230 | i = 0; |
1231 | |
1232 | do { |
1233 | struct vm_area_struct *vma; |
1234 | unsigned int foll_flags; |
1235 | |
1236 | vma = find_extend_vma(mm, start); |
1237 | if (!vma && in_gate_area(tsk, start)) { |
1238 | unsigned long pg = start & PAGE_MASK; |
1239 | struct vm_area_struct *gate_vma = get_gate_vma(tsk); |
1240 | pgd_t *pgd; |
1241 | pud_t *pud; |
1242 | pmd_t *pmd; |
1243 | pte_t *pte; |
1244 | |
1245 | /* user gate pages are read-only */ |
1246 | if (!ignore && write) |
1247 | return i ? : -EFAULT; |
1248 | if (pg > TASK_SIZE) |
1249 | pgd = pgd_offset_k(pg); |
1250 | else |
1251 | pgd = pgd_offset_gate(mm, pg); |
1252 | BUG_ON(pgd_none(*pgd)); |
1253 | pud = pud_offset(pgd, pg); |
1254 | BUG_ON(pud_none(*pud)); |
1255 | pmd = pmd_offset(pud, pg); |
1256 | if (pmd_none(*pmd)) |
1257 | return i ? : -EFAULT; |
1258 | pte = pte_offset_map(pmd, pg); |
1259 | if (pte_none(*pte)) { |
1260 | pte_unmap(pte); |
1261 | return i ? : -EFAULT; |
1262 | } |
1263 | if (pages) { |
1264 | struct page *page = vm_normal_page(gate_vma, start, *pte); |
1265 | pages[i] = page; |
1266 | if (page) |
1267 | get_page(page); |
1268 | } |
1269 | pte_unmap(pte); |
1270 | if (vmas) |
1271 | vmas[i] = gate_vma; |
1272 | i++; |
1273 | start += PAGE_SIZE; |
1274 | nr_pages--; |
1275 | continue; |
1276 | } |
1277 | |
1278 | if (!vma || |
1279 | (vma->vm_flags & (VM_IO | VM_PFNMAP)) || |
1280 | (!ignore && !(vm_flags & vma->vm_flags))) |
1281 | return i ? : -EFAULT; |
1282 | |
1283 | if (is_vm_hugetlb_page(vma)) { |
1284 | i = follow_hugetlb_page(mm, vma, pages, vmas, |
1285 | &start, &nr_pages, i, write); |
1286 | continue; |
1287 | } |
1288 | |
1289 | foll_flags = FOLL_TOUCH; |
1290 | if (pages) |
1291 | foll_flags |= FOLL_GET; |
1292 | if (!write && use_zero_page(vma)) |
1293 | foll_flags |= FOLL_ANON; |
1294 | |
1295 | do { |
1296 | struct page *page; |
1297 | |
1298 | /* |
1299 | * If we have a pending SIGKILL, don't keep faulting |
1300 | * pages and potentially allocating memory, unless |
1301 | * current is handling munlock--e.g., on exit. In |
1302 | * that case, we are not allocating memory. Rather, |
1303 | * we're only unlocking already resident/mapped pages. |
1304 | */ |
1305 | if (unlikely(!ignore_sigkill && |
1306 | fatal_signal_pending(current))) |
1307 | return i ? i : -ERESTARTSYS; |
1308 | |
1309 | if (write) |
1310 | foll_flags |= FOLL_WRITE; |
1311 | |
1312 | cond_resched(); |
1313 | while (!(page = follow_page(vma, start, foll_flags))) { |
1314 | int ret; |
1315 | |
1316 | ret = handle_mm_fault(mm, vma, start, |
1317 | (foll_flags & FOLL_WRITE) ? |
1318 | FAULT_FLAG_WRITE : 0); |
1319 | |
1320 | if (ret & VM_FAULT_ERROR) { |
1321 | if (ret & VM_FAULT_OOM) |
1322 | return i ? i : -ENOMEM; |
1323 | else if (ret & VM_FAULT_SIGBUS) |
1324 | return i ? i : -EFAULT; |
1325 | BUG(); |
1326 | } |
1327 | if (ret & VM_FAULT_MAJOR) |
1328 | tsk->maj_flt++; |
1329 | else |
1330 | tsk->min_flt++; |
1331 | |
1332 | /* |
1333 | * The VM_FAULT_WRITE bit tells us that |
1334 | * do_wp_page has broken COW when necessary, |
1335 | * even if maybe_mkwrite decided not to set |
1336 | * pte_write. We can thus safely do subsequent |
1337 | * page lookups as if they were reads. But only |
1338 | * do so when looping for pte_write is futile: |
1339 | * in some cases userspace may also be wanting |
1340 | * to write to the gotten user page, which a |
1341 | * read fault here might prevent (a readonly |
1342 | * page might get reCOWed by userspace write). |
1343 | */ |
1344 | if ((ret & VM_FAULT_WRITE) && |
1345 | !(vma->vm_flags & VM_WRITE)) |
1346 | foll_flags &= ~FOLL_WRITE; |
1347 | |
1348 | cond_resched(); |
1349 | } |
1350 | if (IS_ERR(page)) |
1351 | return i ? i : PTR_ERR(page); |
1352 | if (pages) { |
1353 | pages[i] = page; |
1354 | |
1355 | flush_anon_page(vma, page, start); |
1356 | flush_dcache_page(page); |
1357 | } |
1358 | if (vmas) |
1359 | vmas[i] = vma; |
1360 | i++; |
1361 | start += PAGE_SIZE; |
1362 | nr_pages--; |
1363 | } while (nr_pages && start < vma->vm_end); |
1364 | } while (nr_pages); |
1365 | return i; |
1366 | } |
1367 | |
1368 | /** |
1369 | * get_user_pages() - pin user pages in memory |
1370 | * @tsk: task_struct of target task |
1371 | * @mm: mm_struct of target mm |
1372 | * @start: starting user address |
1373 | * @nr_pages: number of pages from start to pin |
1374 | * @write: whether pages will be written to by the caller |
1375 | * @force: whether to force write access even if user mapping is |
1376 | * readonly. This will result in the page being COWed even |
1377 | * in MAP_SHARED mappings. You do not want this. |
1378 | * @pages: array that receives pointers to the pages pinned. |
1379 | * Should be at least nr_pages long. Or NULL, if caller |
1380 | * only intends to ensure the pages are faulted in. |
1381 | * @vmas: array of pointers to vmas corresponding to each page. |
1382 | * Or NULL if the caller does not require them. |
1383 | * |
1384 | * Returns number of pages pinned. This may be fewer than the number |
1385 | * requested. If nr_pages is 0 or negative, returns 0. If no pages |
1386 | * were pinned, returns -errno. Each page returned must be released |
1387 | * with a put_page() call when it is finished with. vmas will only |
1388 | * remain valid while mmap_sem is held. |
1389 | * |
1390 | * Must be called with mmap_sem held for read or write. |
1391 | * |
1392 | * get_user_pages walks a process's page tables and takes a reference to |
1393 | * each struct page that each user address corresponds to at a given |
1394 | * instant. That is, it takes the page that would be accessed if a user |
1395 | * thread accesses the given user virtual address at that instant. |
1396 | * |
1397 | * This does not guarantee that the page exists in the user mappings when |
1398 | * get_user_pages returns, and there may even be a completely different |
1399 | * page there in some cases (eg. if mmapped pagecache has been invalidated |
1400 | * and subsequently re faulted). However it does guarantee that the page |
1401 | * won't be freed completely. And mostly callers simply care that the page |
1402 | * contains data that was valid *at some point in time*. Typically, an IO |
1403 | * or similar operation cannot guarantee anything stronger anyway because |
1404 | * locks can't be held over the syscall boundary. |
1405 | * |
1406 | * If write=0, the page must not be written to. If the page is written to, |
1407 | * set_page_dirty (or set_page_dirty_lock, as appropriate) must be called |
1408 | * after the page is finished with, and before put_page is called. |
1409 | * |
1410 | * get_user_pages is typically used for fewer-copy IO operations, to get a |
1411 | * handle on the memory by some means other than accesses via the user virtual |
1412 | * addresses. The pages may be submitted for DMA to devices or accessed via |
1413 | * their kernel linear mapping (via the kmap APIs). Care should be taken to |
1414 | * use the correct cache flushing APIs. |
1415 | * |
1416 | * See also get_user_pages_fast, for performance critical applications. |
1417 | */ |
1418 | int get_user_pages(struct task_struct *tsk, struct mm_struct *mm, |
1419 | unsigned long start, int nr_pages, int write, int force, |
1420 | struct page **pages, struct vm_area_struct **vmas) |
1421 | { |
1422 | int flags = 0; |
1423 | |
1424 | if (write) |
1425 | flags |= GUP_FLAGS_WRITE; |
1426 | if (force) |
1427 | flags |= GUP_FLAGS_FORCE; |
1428 | |
1429 | return __get_user_pages(tsk, mm, start, nr_pages, flags, pages, vmas); |
1430 | } |
1431 | |
1432 | EXPORT_SYMBOL(get_user_pages); |
1433 | |
1434 | pte_t *get_locked_pte(struct mm_struct *mm, unsigned long addr, |
1435 | spinlock_t **ptl) |
1436 | { |
1437 | pgd_t * pgd = pgd_offset(mm, addr); |
1438 | pud_t * pud = pud_alloc(mm, pgd, addr); |
1439 | if (pud) { |
1440 | pmd_t * pmd = pmd_alloc(mm, pud, addr); |
1441 | if (pmd) |
1442 | return pte_alloc_map_lock(mm, pmd, addr, ptl); |
1443 | } |
1444 | return NULL; |
1445 | } |
1446 | |
1447 | /* |
1448 | * This is the old fallback for page remapping. |
1449 | * |
1450 | * For historical reasons, it only allows reserved pages. Only |
1451 | * old drivers should use this, and they needed to mark their |
1452 | * pages reserved for the old functions anyway. |
1453 | */ |
1454 | static int insert_page(struct vm_area_struct *vma, unsigned long addr, |
1455 | struct page *page, pgprot_t prot) |
1456 | { |
1457 | struct mm_struct *mm = vma->vm_mm; |
1458 | int retval; |
1459 | pte_t *pte; |
1460 | spinlock_t *ptl; |
1461 | |
1462 | retval = -EINVAL; |
1463 | if (PageAnon(page)) |
1464 | goto out; |
1465 | retval = -ENOMEM; |
1466 | flush_dcache_page(page); |
1467 | pte = get_locked_pte(mm, addr, &ptl); |
1468 | if (!pte) |
1469 | goto out; |
1470 | retval = -EBUSY; |
1471 | if (!pte_none(*pte)) |
1472 | goto out_unlock; |
1473 | |
1474 | /* Ok, finally just insert the thing.. */ |
1475 | get_page(page); |
1476 | inc_mm_counter(mm, file_rss); |
1477 | page_add_file_rmap(page); |
1478 | set_pte_at(mm, addr, pte, mk_pte(page, prot)); |
1479 | |
1480 | retval = 0; |
1481 | pte_unmap_unlock(pte, ptl); |
1482 | return retval; |
1483 | out_unlock: |
1484 | pte_unmap_unlock(pte, ptl); |
1485 | out: |
1486 | return retval; |
1487 | } |
1488 | |
1489 | /** |
1490 | * vm_insert_page - insert single page into user vma |
1491 | * @vma: user vma to map to |
1492 | * @addr: target user address of this page |
1493 | * @page: source kernel page |
1494 | * |
1495 | * This allows drivers to insert individual pages they've allocated |
1496 | * into a user vma. |
1497 | * |
1498 | * The page has to be a nice clean _individual_ kernel allocation. |
1499 | * If you allocate a compound page, you need to have marked it as |
1500 | * such (__GFP_COMP), or manually just split the page up yourself |
1501 | * (see split_page()). |
1502 | * |
1503 | * NOTE! Traditionally this was done with "remap_pfn_range()" which |
1504 | * took an arbitrary page protection parameter. This doesn't allow |
1505 | * that. Your vma protection will have to be set up correctly, which |
1506 | * means that if you want a shared writable mapping, you'd better |
1507 | * ask for a shared writable mapping! |
1508 | * |
1509 | * The page does not need to be reserved. |
1510 | */ |
1511 | int vm_insert_page(struct vm_area_struct *vma, unsigned long addr, |
1512 | struct page *page) |
1513 | { |
1514 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1515 | return -EFAULT; |
1516 | if (!page_count(page)) |
1517 | return -EINVAL; |
1518 | vma->vm_flags |= VM_INSERTPAGE; |
1519 | return insert_page(vma, addr, page, vma->vm_page_prot); |
1520 | } |
1521 | EXPORT_SYMBOL(vm_insert_page); |
1522 | |
1523 | static int insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
1524 | unsigned long pfn, pgprot_t prot) |
1525 | { |
1526 | struct mm_struct *mm = vma->vm_mm; |
1527 | int retval; |
1528 | pte_t *pte, entry; |
1529 | spinlock_t *ptl; |
1530 | |
1531 | retval = -ENOMEM; |
1532 | pte = get_locked_pte(mm, addr, &ptl); |
1533 | if (!pte) |
1534 | goto out; |
1535 | retval = -EBUSY; |
1536 | if (!pte_none(*pte)) |
1537 | goto out_unlock; |
1538 | |
1539 | /* Ok, finally just insert the thing.. */ |
1540 | entry = pte_mkspecial(pfn_pte(pfn, prot)); |
1541 | set_pte_at(mm, addr, pte, entry); |
1542 | update_mmu_cache(vma, addr, entry); /* XXX: why not for insert_page? */ |
1543 | |
1544 | retval = 0; |
1545 | out_unlock: |
1546 | pte_unmap_unlock(pte, ptl); |
1547 | out: |
1548 | return retval; |
1549 | } |
1550 | |
1551 | /** |
1552 | * vm_insert_pfn - insert single pfn into user vma |
1553 | * @vma: user vma to map to |
1554 | * @addr: target user address of this page |
1555 | * @pfn: source kernel pfn |
1556 | * |
1557 | * Similar to vm_inert_page, this allows drivers to insert individual pages |
1558 | * they've allocated into a user vma. Same comments apply. |
1559 | * |
1560 | * This function should only be called from a vm_ops->fault handler, and |
1561 | * in that case the handler should return NULL. |
1562 | * |
1563 | * vma cannot be a COW mapping. |
1564 | * |
1565 | * As this is called only for pages that do not currently exist, we |
1566 | * do not need to flush old virtual caches or the TLB. |
1567 | */ |
1568 | int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr, |
1569 | unsigned long pfn) |
1570 | { |
1571 | int ret; |
1572 | pgprot_t pgprot = vma->vm_page_prot; |
1573 | /* |
1574 | * Technically, architectures with pte_special can avoid all these |
1575 | * restrictions (same for remap_pfn_range). However we would like |
1576 | * consistency in testing and feature parity among all, so we should |
1577 | * try to keep these invariants in place for everybody. |
1578 | */ |
1579 | BUG_ON(!(vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP))); |
1580 | BUG_ON((vma->vm_flags & (VM_PFNMAP|VM_MIXEDMAP)) == |
1581 | (VM_PFNMAP|VM_MIXEDMAP)); |
1582 | BUG_ON((vma->vm_flags & VM_PFNMAP) && is_cow_mapping(vma->vm_flags)); |
1583 | BUG_ON((vma->vm_flags & VM_MIXEDMAP) && pfn_valid(pfn)); |
1584 | |
1585 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1586 | return -EFAULT; |
1587 | if (track_pfn_vma_new(vma, &pgprot, pfn, PAGE_SIZE)) |
1588 | return -EINVAL; |
1589 | |
1590 | ret = insert_pfn(vma, addr, pfn, pgprot); |
1591 | |
1592 | if (ret) |
1593 | untrack_pfn_vma(vma, pfn, PAGE_SIZE); |
1594 | |
1595 | return ret; |
1596 | } |
1597 | EXPORT_SYMBOL(vm_insert_pfn); |
1598 | |
1599 | int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr, |
1600 | unsigned long pfn) |
1601 | { |
1602 | BUG_ON(!(vma->vm_flags & VM_MIXEDMAP)); |
1603 | |
1604 | if (addr < vma->vm_start || addr >= vma->vm_end) |
1605 | return -EFAULT; |
1606 | |
1607 | /* |
1608 | * If we don't have pte special, then we have to use the pfn_valid() |
1609 | * based VM_MIXEDMAP scheme (see vm_normal_page), and thus we *must* |
1610 | * refcount the page if pfn_valid is true (hence insert_page rather |
1611 | * than insert_pfn). |
1612 | */ |
1613 | if (!HAVE_PTE_SPECIAL && pfn_valid(pfn)) { |
1614 | struct page *page; |
1615 | |
1616 | page = pfn_to_page(pfn); |
1617 | return insert_page(vma, addr, page, vma->vm_page_prot); |
1618 | } |
1619 | return insert_pfn(vma, addr, pfn, vma->vm_page_prot); |
1620 | } |
1621 | EXPORT_SYMBOL(vm_insert_mixed); |
1622 | |
1623 | /* |
1624 | * maps a range of physical memory into the requested pages. the old |
1625 | * mappings are removed. any references to nonexistent pages results |
1626 | * in null mappings (currently treated as "copy-on-access") |
1627 | */ |
1628 | static int remap_pte_range(struct mm_struct *mm, pmd_t *pmd, |
1629 | unsigned long addr, unsigned long end, |
1630 | unsigned long pfn, pgprot_t prot) |
1631 | { |
1632 | pte_t *pte; |
1633 | spinlock_t *ptl; |
1634 | |
1635 | pte = pte_alloc_map_lock(mm, pmd, addr, &ptl); |
1636 | if (!pte) |
1637 | return -ENOMEM; |
1638 | arch_enter_lazy_mmu_mode(); |
1639 | do { |
1640 | BUG_ON(!pte_none(*pte)); |
1641 | set_pte_at(mm, addr, pte, pte_mkspecial(pfn_pte(pfn, prot))); |
1642 | pfn++; |
1643 | } while (pte++, addr += PAGE_SIZE, addr != end); |
1644 | arch_leave_lazy_mmu_mode(); |
1645 | pte_unmap_unlock(pte - 1, ptl); |
1646 | return 0; |
1647 | } |
1648 | |
1649 | static inline int remap_pmd_range(struct mm_struct *mm, pud_t *pud, |
1650 | unsigned long addr, unsigned long end, |
1651 | unsigned long pfn, pgprot_t prot) |
1652 | { |
1653 | pmd_t *pmd; |
1654 | unsigned long next; |
1655 | |
1656 | pfn -= addr >> PAGE_SHIFT; |
1657 | pmd = pmd_alloc(mm, pud, addr); |
1658 | if (!pmd) |
1659 | return -ENOMEM; |
1660 | do { |
1661 | next = pmd_addr_end(addr, end); |
1662 | if (remap_pte_range(mm, pmd, addr, next, |
1663 | pfn + (addr >> PAGE_SHIFT), prot)) |
1664 | return -ENOMEM; |
1665 | } while (pmd++, addr = next, addr != end); |
1666 | return 0; |
1667 | } |
1668 | |
1669 | static inline int remap_pud_range(struct mm_struct *mm, pgd_t *pgd, |
1670 | unsigned long addr, unsigned long end, |
1671 | unsigned long pfn, pgprot_t prot) |
1672 | { |
1673 | pud_t *pud; |
1674 | unsigned long next; |
1675 | |
1676 | pfn -= addr >> PAGE_SHIFT; |
1677 | pud = pud_alloc(mm, pgd, addr); |
1678 | if (!pud) |
1679 | return -ENOMEM; |
1680 | do { |
1681 | next = pud_addr_end(addr, end); |
1682 | if (remap_pmd_range(mm, pud, addr, next, |
1683 | pfn + (addr >> PAGE_SHIFT), prot)) |
1684 | return -ENOMEM; |
1685 | } while (pud++, addr = next, addr != end); |
1686 | return 0; |
1687 | } |
1688 | |
1689 | /** |
1690 | * remap_pfn_range - remap kernel memory to userspace |
1691 | * @vma: user vma to map to |
1692 | * @addr: target user address to start at |
1693 | * @pfn: physical address of kernel memory |
1694 | * @size: size of map area |
1695 | * @prot: page protection flags for this mapping |
1696 | * |
1697 | * Note: this is only safe if the mm semaphore is held when called. |
1698 | */ |
1699 | int remap_pfn_range(struct vm_area_struct *vma, unsigned long addr, |
1700 | unsigned long pfn, unsigned long size, pgprot_t prot) |
1701 | { |
1702 | pgd_t *pgd; |
1703 | unsigned long next; |
1704 | unsigned long end = addr + PAGE_ALIGN(size); |
1705 | struct mm_struct *mm = vma->vm_mm; |
1706 | int err; |
1707 | |
1708 | /* |
1709 | * Physically remapped pages are special. Tell the |
1710 | * rest of the world about it: |
1711 | * VM_IO tells people not to look at these pages |
1712 | * (accesses can have side effects). |
1713 | * VM_RESERVED is specified all over the place, because |
1714 | * in 2.4 it kept swapout's vma scan off this vma; but |
1715 | * in 2.6 the LRU scan won't even find its pages, so this |
1716 | * flag means no more than count its pages in reserved_vm, |
1717 | * and omit it from core dump, even when VM_IO turned off. |
1718 | * VM_PFNMAP tells the core MM that the base pages are just |
1719 | * raw PFN mappings, and do not have a "struct page" associated |
1720 | * with them. |
1721 | * |
1722 | * There's a horrible special case to handle copy-on-write |
1723 | * behaviour that some programs depend on. We mark the "original" |
1724 | * un-COW'ed pages by matching them up with "vma->vm_pgoff". |
1725 | */ |
1726 | if (addr == vma->vm_start && end == vma->vm_end) { |
1727 | vma->vm_pgoff = pfn; |
1728 | vma->vm_flags |= VM_PFN_AT_MMAP; |
1729 | } else if (is_cow_mapping(vma->vm_flags)) |
1730 | return -EINVAL; |
1731 | |
1732 | vma->vm_flags |= VM_IO | VM_RESERVED | VM_PFNMAP; |
1733 | |
1734 | err = track_pfn_vma_new(vma, &prot, pfn, PAGE_ALIGN(size)); |
1735 | if (err) { |
1736 | /* |
1737 | * To indicate that track_pfn related cleanup is not |
1738 | * needed from higher level routine calling unmap_vmas |
1739 | */ |
1740 | vma->vm_flags &= ~(VM_IO | VM_RESERVED | VM_PFNMAP); |
1741 | vma->vm_flags &= ~VM_PFN_AT_MMAP; |
1742 | return -EINVAL; |
1743 | } |
1744 | |
1745 | BUG_ON(addr >= end); |
1746 | pfn -= addr >> PAGE_SHIFT; |
1747 | pgd = pgd_offset(mm, addr); |
1748 | flush_cache_range(vma, addr, end); |
1749 | do { |
1750 | next = pgd_addr_end(addr, end); |
1751 | err = remap_pud_range(mm, pgd, addr, next, |
1752 | pfn + (addr >> PAGE_SHIFT), prot); |
1753 | if (err) |
1754 | break; |
1755 | } while (pgd++, addr = next, addr != end); |
1756 | |
1757 | if (err) |
1758 | untrack_pfn_vma(vma, pfn, PAGE_ALIGN(size)); |
1759 | |
1760 | return err; |
1761 | } |
1762 | EXPORT_SYMBOL(remap_pfn_range); |
1763 | |
1764 | static int apply_to_pte_range(struct mm_struct *mm, pmd_t *pmd, |
1765 | unsigned long addr, unsigned long end, |
1766 | pte_fn_t fn, void *data) |
1767 | { |
1768 | pte_t *pte; |
1769 | int err; |
1770 | pgtable_t token; |
1771 | spinlock_t *uninitialized_var(ptl); |
1772 | |
1773 | pte = (mm == &init_mm) ? |
1774 | pte_alloc_kernel(pmd, addr) : |
1775 | pte_alloc_map_lock(mm, pmd, addr, &ptl); |
1776 | if (!pte) |
1777 | return -ENOMEM; |
1778 | |
1779 | BUG_ON(pmd_huge(*pmd)); |
1780 | |
1781 | arch_enter_lazy_mmu_mode(); |
1782 | |
1783 | token = pmd_pgtable(*pmd); |
1784 | |
1785 | do { |
1786 | err = fn(pte, token, addr, data); |
1787 | if (err) |
1788 | break; |
1789 | } while (pte++, addr += PAGE_SIZE, addr != end); |
1790 | |
1791 | arch_leave_lazy_mmu_mode(); |
1792 | |
1793 | if (mm != &init_mm) |
1794 | pte_unmap_unlock(pte-1, ptl); |
1795 | return err; |
1796 | } |
1797 | |
1798 | static int apply_to_pmd_range(struct mm_struct *mm, pud_t *pud, |
1799 | unsigned long addr, unsigned long end, |
1800 | pte_fn_t fn, void *data) |
1801 | { |
1802 | pmd_t *pmd; |
1803 | unsigned long next; |
1804 | int err; |
1805 | |
1806 | BUG_ON(pud_huge(*pud)); |
1807 | |
1808 | pmd = pmd_alloc(mm, pud, addr); |
1809 | if (!pmd) |
1810 | return -ENOMEM; |
1811 | do { |
1812 | next = pmd_addr_end(addr, end); |
1813 | err = apply_to_pte_range(mm, pmd, addr, next, fn, data); |
1814 | if (err) |
1815 | break; |
1816 | } while (pmd++, addr = next, addr != end); |
1817 | return err; |
1818 | } |
1819 | |
1820 | static int apply_to_pud_range(struct mm_struct *mm, pgd_t *pgd, |
1821 | unsigned long addr, unsigned long end, |
1822 | pte_fn_t fn, void *data) |
1823 | { |
1824 | pud_t *pud; |
1825 | unsigned long next; |
1826 | int err; |
1827 | |
1828 | pud = pud_alloc(mm, pgd, addr); |
1829 | if (!pud) |
1830 | return -ENOMEM; |
1831 | do { |
1832 | next = pud_addr_end(addr, end); |
1833 | err = apply_to_pmd_range(mm, pud, addr, next, fn, data); |
1834 | if (err) |
1835 | break; |
1836 | } while (pud++, addr = next, addr != end); |
1837 | return err; |
1838 | } |
1839 | |
1840 | /* |
1841 | * Scan a region of virtual memory, filling in page tables as necessary |
1842 | * and calling a provided function on each leaf page table. |
1843 | */ |
1844 | int apply_to_page_range(struct mm_struct *mm, unsigned long addr, |
1845 | unsigned long size, pte_fn_t fn, void *data) |
1846 | { |
1847 | pgd_t *pgd; |
1848 | unsigned long next; |
1849 | unsigned long start = addr, end = addr + size; |
1850 | int err; |
1851 | |
1852 | BUG_ON(addr >= end); |
1853 | mmu_notifier_invalidate_range_start(mm, start, end); |
1854 | pgd = pgd_offset(mm, addr); |
1855 | do { |
1856 | next = pgd_addr_end(addr, end); |
1857 | err = apply_to_pud_range(mm, pgd, addr, next, fn, data); |
1858 | if (err) |
1859 | break; |
1860 | } while (pgd++, addr = next, addr != end); |
1861 | mmu_notifier_invalidate_range_end(mm, start, end); |
1862 | return err; |
1863 | } |
1864 | EXPORT_SYMBOL_GPL(apply_to_page_range); |
1865 | |
1866 | /* |
1867 | * handle_pte_fault chooses page fault handler according to an entry |
1868 | * which was read non-atomically. Before making any commitment, on |
1869 | * those architectures or configurations (e.g. i386 with PAE) which |
1870 | * might give a mix of unmatched parts, do_swap_page and do_file_page |
1871 | * must check under lock before unmapping the pte and proceeding |
1872 | * (but do_wp_page is only called after already making such a check; |
1873 | * and do_anonymous_page and do_no_page can safely check later on). |
1874 | */ |
1875 | static inline int pte_unmap_same(struct mm_struct *mm, pmd_t *pmd, |
1876 | pte_t *page_table, pte_t orig_pte) |
1877 | { |
1878 | int same = 1; |
1879 | #if defined(CONFIG_SMP) || defined(CONFIG_PREEMPT) |
1880 | if (sizeof(pte_t) > sizeof(unsigned long)) { |
1881 | spinlock_t *ptl = pte_lockptr(mm, pmd); |
1882 | spin_lock(ptl); |
1883 | same = pte_same(*page_table, orig_pte); |
1884 | spin_unlock(ptl); |
1885 | } |
1886 | #endif |
1887 | pte_unmap(page_table); |
1888 | return same; |
1889 | } |
1890 | |
1891 | /* |
1892 | * Do pte_mkwrite, but only if the vma says VM_WRITE. We do this when |
1893 | * servicing faults for write access. In the normal case, do always want |
1894 | * pte_mkwrite. But get_user_pages can cause write faults for mappings |
1895 | * that do not have writing enabled, when used by access_process_vm. |
1896 | */ |
1897 | static inline pte_t maybe_mkwrite(pte_t pte, struct vm_area_struct *vma) |
1898 | { |
1899 | if (likely(vma->vm_flags & VM_WRITE)) |
1900 | pte = pte_mkwrite(pte); |
1901 | return pte; |
1902 | } |
1903 | |
1904 | static inline void cow_user_page(struct page *dst, struct page *src, unsigned long va, struct vm_area_struct *vma) |
1905 | { |
1906 | /* |
1907 | * If the source page was a PFN mapping, we don't have |
1908 | * a "struct page" for it. We do a best-effort copy by |
1909 | * just copying from the original user address. If that |
1910 | * fails, we just zero-fill it. Live with it. |
1911 | */ |
1912 | if (unlikely(!src)) { |
1913 | void *kaddr = kmap_atomic(dst, KM_USER0); |
1914 | void __user *uaddr = (void __user *)(va & PAGE_MASK); |
1915 | |
1916 | /* |
1917 | * This really shouldn't fail, because the page is there |
1918 | * in the page tables. But it might just be unreadable, |
1919 | * in which case we just give up and fill the result with |
1920 | * zeroes. |
1921 | */ |
1922 | if (__copy_from_user_inatomic(kaddr, uaddr, PAGE_SIZE)) |
1923 | memset(kaddr, 0, PAGE_SIZE); |
1924 | kunmap_atomic(kaddr, KM_USER0); |
1925 | flush_dcache_page(dst); |
1926 | } else |
1927 | copy_user_highpage(dst, src, va, vma); |
1928 | } |
1929 | |
1930 | /* |
1931 | * This routine handles present pages, when users try to write |
1932 | * to a shared page. It is done by copying the page to a new address |
1933 | * and decrementing the shared-page counter for the old page. |
1934 | * |
1935 | * Note that this routine assumes that the protection checks have been |
1936 | * done by the caller (the low-level page fault routine in most cases). |
1937 | * Thus we can safely just mark it writable once we've done any necessary |
1938 | * COW. |
1939 | * |
1940 | * We also mark the page dirty at this point even though the page will |
1941 | * change only once the write actually happens. This avoids a few races, |
1942 | * and potentially makes it more efficient. |
1943 | * |
1944 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
1945 | * but allow concurrent faults), with pte both mapped and locked. |
1946 | * We return with mmap_sem still held, but pte unmapped and unlocked. |
1947 | */ |
1948 | static int do_wp_page(struct mm_struct *mm, struct vm_area_struct *vma, |
1949 | unsigned long address, pte_t *page_table, pmd_t *pmd, |
1950 | spinlock_t *ptl, pte_t orig_pte) |
1951 | { |
1952 | struct page *old_page, *new_page; |
1953 | pte_t entry; |
1954 | int reuse = 0, ret = 0; |
1955 | int page_mkwrite = 0; |
1956 | struct page *dirty_page = NULL; |
1957 | |
1958 | old_page = vm_normal_page(vma, address, orig_pte); |
1959 | if (!old_page) { |
1960 | /* |
1961 | * VM_MIXEDMAP !pfn_valid() case |
1962 | * |
1963 | * We should not cow pages in a shared writeable mapping. |
1964 | * Just mark the pages writable as we can't do any dirty |
1965 | * accounting on raw pfn maps. |
1966 | */ |
1967 | if ((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
1968 | (VM_WRITE|VM_SHARED)) |
1969 | goto reuse; |
1970 | goto gotten; |
1971 | } |
1972 | |
1973 | /* |
1974 | * Take out anonymous pages first, anonymous shared vmas are |
1975 | * not dirty accountable. |
1976 | */ |
1977 | if (PageAnon(old_page)) { |
1978 | if (!trylock_page(old_page)) { |
1979 | page_cache_get(old_page); |
1980 | pte_unmap_unlock(page_table, ptl); |
1981 | lock_page(old_page); |
1982 | page_table = pte_offset_map_lock(mm, pmd, address, |
1983 | &ptl); |
1984 | if (!pte_same(*page_table, orig_pte)) { |
1985 | unlock_page(old_page); |
1986 | page_cache_release(old_page); |
1987 | goto unlock; |
1988 | } |
1989 | page_cache_release(old_page); |
1990 | } |
1991 | reuse = reuse_swap_page(old_page); |
1992 | unlock_page(old_page); |
1993 | } else if (unlikely((vma->vm_flags & (VM_WRITE|VM_SHARED)) == |
1994 | (VM_WRITE|VM_SHARED))) { |
1995 | /* |
1996 | * Only catch write-faults on shared writable pages, |
1997 | * read-only shared pages can get COWed by |
1998 | * get_user_pages(.write=1, .force=1). |
1999 | */ |
2000 | if (vma->vm_ops && vma->vm_ops->page_mkwrite) { |
2001 | struct vm_fault vmf; |
2002 | int tmp; |
2003 | |
2004 | vmf.virtual_address = (void __user *)(address & |
2005 | PAGE_MASK); |
2006 | vmf.pgoff = old_page->index; |
2007 | vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; |
2008 | vmf.page = old_page; |
2009 | |
2010 | /* |
2011 | * Notify the address space that the page is about to |
2012 | * become writable so that it can prohibit this or wait |
2013 | * for the page to get into an appropriate state. |
2014 | * |
2015 | * We do this without the lock held, so that it can |
2016 | * sleep if it needs to. |
2017 | */ |
2018 | page_cache_get(old_page); |
2019 | pte_unmap_unlock(page_table, ptl); |
2020 | |
2021 | tmp = vma->vm_ops->page_mkwrite(vma, &vmf); |
2022 | if (unlikely(tmp & |
2023 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { |
2024 | ret = tmp; |
2025 | goto unwritable_page; |
2026 | } |
2027 | if (unlikely(!(tmp & VM_FAULT_LOCKED))) { |
2028 | lock_page(old_page); |
2029 | if (!old_page->mapping) { |
2030 | ret = 0; /* retry the fault */ |
2031 | unlock_page(old_page); |
2032 | goto unwritable_page; |
2033 | } |
2034 | } else |
2035 | VM_BUG_ON(!PageLocked(old_page)); |
2036 | |
2037 | /* |
2038 | * Since we dropped the lock we need to revalidate |
2039 | * the PTE as someone else may have changed it. If |
2040 | * they did, we just return, as we can count on the |
2041 | * MMU to tell us if they didn't also make it writable. |
2042 | */ |
2043 | page_table = pte_offset_map_lock(mm, pmd, address, |
2044 | &ptl); |
2045 | if (!pte_same(*page_table, orig_pte)) { |
2046 | unlock_page(old_page); |
2047 | page_cache_release(old_page); |
2048 | goto unlock; |
2049 | } |
2050 | |
2051 | page_mkwrite = 1; |
2052 | } |
2053 | dirty_page = old_page; |
2054 | get_page(dirty_page); |
2055 | reuse = 1; |
2056 | } |
2057 | |
2058 | if (reuse) { |
2059 | reuse: |
2060 | flush_cache_page(vma, address, pte_pfn(orig_pte)); |
2061 | entry = pte_mkyoung(orig_pte); |
2062 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
2063 | if (ptep_set_access_flags(vma, address, page_table, entry,1)) |
2064 | update_mmu_cache(vma, address, entry); |
2065 | ret |= VM_FAULT_WRITE; |
2066 | goto unlock; |
2067 | } |
2068 | |
2069 | /* |
2070 | * Ok, we need to copy. Oh, well.. |
2071 | */ |
2072 | page_cache_get(old_page); |
2073 | gotten: |
2074 | pte_unmap_unlock(page_table, ptl); |
2075 | |
2076 | if (unlikely(anon_vma_prepare(vma))) |
2077 | goto oom; |
2078 | VM_BUG_ON(old_page == ZERO_PAGE(0)); |
2079 | new_page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, vma, address); |
2080 | if (!new_page) |
2081 | goto oom; |
2082 | /* |
2083 | * Don't let another task, with possibly unlocked vma, |
2084 | * keep the mlocked page. |
2085 | */ |
2086 | if ((vma->vm_flags & VM_LOCKED) && old_page) { |
2087 | lock_page(old_page); /* for LRU manipulation */ |
2088 | clear_page_mlock(old_page); |
2089 | unlock_page(old_page); |
2090 | } |
2091 | cow_user_page(new_page, old_page, address, vma); |
2092 | __SetPageUptodate(new_page); |
2093 | |
2094 | if (mem_cgroup_newpage_charge(new_page, mm, GFP_KERNEL)) |
2095 | goto oom_free_new; |
2096 | |
2097 | /* |
2098 | * Re-check the pte - we dropped the lock |
2099 | */ |
2100 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
2101 | if (likely(pte_same(*page_table, orig_pte))) { |
2102 | if (old_page) { |
2103 | if (!PageAnon(old_page)) { |
2104 | dec_mm_counter(mm, file_rss); |
2105 | inc_mm_counter(mm, anon_rss); |
2106 | } |
2107 | } else |
2108 | inc_mm_counter(mm, anon_rss); |
2109 | flush_cache_page(vma, address, pte_pfn(orig_pte)); |
2110 | entry = mk_pte(new_page, vma->vm_page_prot); |
2111 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
2112 | /* |
2113 | * Clear the pte entry and flush it first, before updating the |
2114 | * pte with the new entry. This will avoid a race condition |
2115 | * seen in the presence of one thread doing SMC and another |
2116 | * thread doing COW. |
2117 | */ |
2118 | ptep_clear_flush_notify(vma, address, page_table); |
2119 | page_add_new_anon_rmap(new_page, vma, address); |
2120 | set_pte_at(mm, address, page_table, entry); |
2121 | update_mmu_cache(vma, address, entry); |
2122 | if (old_page) { |
2123 | /* |
2124 | * Only after switching the pte to the new page may |
2125 | * we remove the mapcount here. Otherwise another |
2126 | * process may come and find the rmap count decremented |
2127 | * before the pte is switched to the new page, and |
2128 | * "reuse" the old page writing into it while our pte |
2129 | * here still points into it and can be read by other |
2130 | * threads. |
2131 | * |
2132 | * The critical issue is to order this |
2133 | * page_remove_rmap with the ptp_clear_flush above. |
2134 | * Those stores are ordered by (if nothing else,) |
2135 | * the barrier present in the atomic_add_negative |
2136 | * in page_remove_rmap. |
2137 | * |
2138 | * Then the TLB flush in ptep_clear_flush ensures that |
2139 | * no process can access the old page before the |
2140 | * decremented mapcount is visible. And the old page |
2141 | * cannot be reused until after the decremented |
2142 | * mapcount is visible. So transitively, TLBs to |
2143 | * old page will be flushed before it can be reused. |
2144 | */ |
2145 | page_remove_rmap(old_page); |
2146 | } |
2147 | |
2148 | /* Free the old page.. */ |
2149 | new_page = old_page; |
2150 | ret |= VM_FAULT_WRITE; |
2151 | } else |
2152 | mem_cgroup_uncharge_page(new_page); |
2153 | |
2154 | if (new_page) |
2155 | page_cache_release(new_page); |
2156 | if (old_page) |
2157 | page_cache_release(old_page); |
2158 | unlock: |
2159 | pte_unmap_unlock(page_table, ptl); |
2160 | if (dirty_page) { |
2161 | /* |
2162 | * Yes, Virginia, this is actually required to prevent a race |
2163 | * with clear_page_dirty_for_io() from clearing the page dirty |
2164 | * bit after it clear all dirty ptes, but before a racing |
2165 | * do_wp_page installs a dirty pte. |
2166 | * |
2167 | * do_no_page is protected similarly. |
2168 | */ |
2169 | if (!page_mkwrite) { |
2170 | wait_on_page_locked(dirty_page); |
2171 | set_page_dirty_balance(dirty_page, page_mkwrite); |
2172 | } |
2173 | put_page(dirty_page); |
2174 | if (page_mkwrite) { |
2175 | struct address_space *mapping = dirty_page->mapping; |
2176 | |
2177 | set_page_dirty(dirty_page); |
2178 | unlock_page(dirty_page); |
2179 | page_cache_release(dirty_page); |
2180 | if (mapping) { |
2181 | /* |
2182 | * Some device drivers do not set page.mapping |
2183 | * but still dirty their pages |
2184 | */ |
2185 | balance_dirty_pages_ratelimited(mapping); |
2186 | } |
2187 | } |
2188 | |
2189 | /* file_update_time outside page_lock */ |
2190 | if (vma->vm_file) |
2191 | file_update_time(vma->vm_file); |
2192 | } |
2193 | return ret; |
2194 | oom_free_new: |
2195 | page_cache_release(new_page); |
2196 | oom: |
2197 | if (old_page) { |
2198 | if (page_mkwrite) { |
2199 | unlock_page(old_page); |
2200 | page_cache_release(old_page); |
2201 | } |
2202 | page_cache_release(old_page); |
2203 | } |
2204 | return VM_FAULT_OOM; |
2205 | |
2206 | unwritable_page: |
2207 | page_cache_release(old_page); |
2208 | return ret; |
2209 | } |
2210 | |
2211 | /* |
2212 | * Helper functions for unmap_mapping_range(). |
2213 | * |
2214 | * __ Notes on dropping i_mmap_lock to reduce latency while unmapping __ |
2215 | * |
2216 | * We have to restart searching the prio_tree whenever we drop the lock, |
2217 | * since the iterator is only valid while the lock is held, and anyway |
2218 | * a later vma might be split and reinserted earlier while lock dropped. |
2219 | * |
2220 | * The list of nonlinear vmas could be handled more efficiently, using |
2221 | * a placeholder, but handle it in the same way until a need is shown. |
2222 | * It is important to search the prio_tree before nonlinear list: a vma |
2223 | * may become nonlinear and be shifted from prio_tree to nonlinear list |
2224 | * while the lock is dropped; but never shifted from list to prio_tree. |
2225 | * |
2226 | * In order to make forward progress despite restarting the search, |
2227 | * vm_truncate_count is used to mark a vma as now dealt with, so we can |
2228 | * quickly skip it next time around. Since the prio_tree search only |
2229 | * shows us those vmas affected by unmapping the range in question, we |
2230 | * can't efficiently keep all vmas in step with mapping->truncate_count: |
2231 | * so instead reset them all whenever it wraps back to 0 (then go to 1). |
2232 | * mapping->truncate_count and vma->vm_truncate_count are protected by |
2233 | * i_mmap_lock. |
2234 | * |
2235 | * In order to make forward progress despite repeatedly restarting some |
2236 | * large vma, note the restart_addr from unmap_vmas when it breaks out: |
2237 | * and restart from that address when we reach that vma again. It might |
2238 | * have been split or merged, shrunk or extended, but never shifted: so |
2239 | * restart_addr remains valid so long as it remains in the vma's range. |
2240 | * unmap_mapping_range forces truncate_count to leap over page-aligned |
2241 | * values so we can save vma's restart_addr in its truncate_count field. |
2242 | */ |
2243 | #define is_restart_addr(truncate_count) (!((truncate_count) & ~PAGE_MASK)) |
2244 | |
2245 | static void reset_vma_truncate_counts(struct address_space *mapping) |
2246 | { |
2247 | struct vm_area_struct *vma; |
2248 | struct prio_tree_iter iter; |
2249 | |
2250 | vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, 0, ULONG_MAX) |
2251 | vma->vm_truncate_count = 0; |
2252 | list_for_each_entry(vma, &mapping->i_mmap_nonlinear, shared.vm_set.list) |
2253 | vma->vm_truncate_count = 0; |
2254 | } |
2255 | |
2256 | static int unmap_mapping_range_vma(struct vm_area_struct *vma, |
2257 | unsigned long start_addr, unsigned long end_addr, |
2258 | struct zap_details *details) |
2259 | { |
2260 | unsigned long restart_addr; |
2261 | int need_break; |
2262 | |
2263 | /* |
2264 | * files that support invalidating or truncating portions of the |
2265 | * file from under mmaped areas must have their ->fault function |
2266 | * return a locked page (and set VM_FAULT_LOCKED in the return). |
2267 | * This provides synchronisation against concurrent unmapping here. |
2268 | */ |
2269 | |
2270 | again: |
2271 | restart_addr = vma->vm_truncate_count; |
2272 | if (is_restart_addr(restart_addr) && start_addr < restart_addr) { |
2273 | start_addr = restart_addr; |
2274 | if (start_addr >= end_addr) { |
2275 | /* Top of vma has been split off since last time */ |
2276 | vma->vm_truncate_count = details->truncate_count; |
2277 | return 0; |
2278 | } |
2279 | } |
2280 | |
2281 | restart_addr = zap_page_range(vma, start_addr, |
2282 | end_addr - start_addr, details); |
2283 | need_break = need_resched() || spin_needbreak(details->i_mmap_lock); |
2284 | |
2285 | if (restart_addr >= end_addr) { |
2286 | /* We have now completed this vma: mark it so */ |
2287 | vma->vm_truncate_count = details->truncate_count; |
2288 | if (!need_break) |
2289 | return 0; |
2290 | } else { |
2291 | /* Note restart_addr in vma's truncate_count field */ |
2292 | vma->vm_truncate_count = restart_addr; |
2293 | if (!need_break) |
2294 | goto again; |
2295 | } |
2296 | |
2297 | spin_unlock(details->i_mmap_lock); |
2298 | cond_resched(); |
2299 | spin_lock(details->i_mmap_lock); |
2300 | return -EINTR; |
2301 | } |
2302 | |
2303 | static inline void unmap_mapping_range_tree(struct prio_tree_root *root, |
2304 | struct zap_details *details) |
2305 | { |
2306 | struct vm_area_struct *vma; |
2307 | struct prio_tree_iter iter; |
2308 | pgoff_t vba, vea, zba, zea; |
2309 | |
2310 | restart: |
2311 | vma_prio_tree_foreach(vma, &iter, root, |
2312 | details->first_index, details->last_index) { |
2313 | /* Skip quickly over those we have already dealt with */ |
2314 | if (vma->vm_truncate_count == details->truncate_count) |
2315 | continue; |
2316 | |
2317 | vba = vma->vm_pgoff; |
2318 | vea = vba + ((vma->vm_end - vma->vm_start) >> PAGE_SHIFT) - 1; |
2319 | /* Assume for now that PAGE_CACHE_SHIFT == PAGE_SHIFT */ |
2320 | zba = details->first_index; |
2321 | if (zba < vba) |
2322 | zba = vba; |
2323 | zea = details->last_index; |
2324 | if (zea > vea) |
2325 | zea = vea; |
2326 | |
2327 | if (unmap_mapping_range_vma(vma, |
2328 | ((zba - vba) << PAGE_SHIFT) + vma->vm_start, |
2329 | ((zea - vba + 1) << PAGE_SHIFT) + vma->vm_start, |
2330 | details) < 0) |
2331 | goto restart; |
2332 | } |
2333 | } |
2334 | |
2335 | static inline void unmap_mapping_range_list(struct list_head *head, |
2336 | struct zap_details *details) |
2337 | { |
2338 | struct vm_area_struct *vma; |
2339 | |
2340 | /* |
2341 | * In nonlinear VMAs there is no correspondence between virtual address |
2342 | * offset and file offset. So we must perform an exhaustive search |
2343 | * across *all* the pages in each nonlinear VMA, not just the pages |
2344 | * whose virtual address lies outside the file truncation point. |
2345 | */ |
2346 | restart: |
2347 | list_for_each_entry(vma, head, shared.vm_set.list) { |
2348 | /* Skip quickly over those we have already dealt with */ |
2349 | if (vma->vm_truncate_count == details->truncate_count) |
2350 | continue; |
2351 | details->nonlinear_vma = vma; |
2352 | if (unmap_mapping_range_vma(vma, vma->vm_start, |
2353 | vma->vm_end, details) < 0) |
2354 | goto restart; |
2355 | } |
2356 | } |
2357 | |
2358 | /** |
2359 | * unmap_mapping_range - unmap the portion of all mmaps in the specified address_space corresponding to the specified page range in the underlying file. |
2360 | * @mapping: the address space containing mmaps to be unmapped. |
2361 | * @holebegin: byte in first page to unmap, relative to the start of |
2362 | * the underlying file. This will be rounded down to a PAGE_SIZE |
2363 | * boundary. Note that this is different from vmtruncate(), which |
2364 | * must keep the partial page. In contrast, we must get rid of |
2365 | * partial pages. |
2366 | * @holelen: size of prospective hole in bytes. This will be rounded |
2367 | * up to a PAGE_SIZE boundary. A holelen of zero truncates to the |
2368 | * end of the file. |
2369 | * @even_cows: 1 when truncating a file, unmap even private COWed pages; |
2370 | * but 0 when invalidating pagecache, don't throw away private data. |
2371 | */ |
2372 | void unmap_mapping_range(struct address_space *mapping, |
2373 | loff_t const holebegin, loff_t const holelen, int even_cows) |
2374 | { |
2375 | struct zap_details details; |
2376 | pgoff_t hba = holebegin >> PAGE_SHIFT; |
2377 | pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; |
2378 | |
2379 | /* Check for overflow. */ |
2380 | if (sizeof(holelen) > sizeof(hlen)) { |
2381 | long long holeend = |
2382 | (holebegin + holelen + PAGE_SIZE - 1) >> PAGE_SHIFT; |
2383 | if (holeend & ~(long long)ULONG_MAX) |
2384 | hlen = ULONG_MAX - hba + 1; |
2385 | } |
2386 | |
2387 | details.check_mapping = even_cows? NULL: mapping; |
2388 | details.nonlinear_vma = NULL; |
2389 | details.first_index = hba; |
2390 | details.last_index = hba + hlen - 1; |
2391 | if (details.last_index < details.first_index) |
2392 | details.last_index = ULONG_MAX; |
2393 | details.i_mmap_lock = &mapping->i_mmap_lock; |
2394 | |
2395 | spin_lock(&mapping->i_mmap_lock); |
2396 | |
2397 | /* Protect against endless unmapping loops */ |
2398 | mapping->truncate_count++; |
2399 | if (unlikely(is_restart_addr(mapping->truncate_count))) { |
2400 | if (mapping->truncate_count == 0) |
2401 | reset_vma_truncate_counts(mapping); |
2402 | mapping->truncate_count++; |
2403 | } |
2404 | details.truncate_count = mapping->truncate_count; |
2405 | |
2406 | if (unlikely(!prio_tree_empty(&mapping->i_mmap))) |
2407 | unmap_mapping_range_tree(&mapping->i_mmap, &details); |
2408 | if (unlikely(!list_empty(&mapping->i_mmap_nonlinear))) |
2409 | unmap_mapping_range_list(&mapping->i_mmap_nonlinear, &details); |
2410 | spin_unlock(&mapping->i_mmap_lock); |
2411 | } |
2412 | EXPORT_SYMBOL(unmap_mapping_range); |
2413 | |
2414 | /** |
2415 | * vmtruncate - unmap mappings "freed" by truncate() syscall |
2416 | * @inode: inode of the file used |
2417 | * @offset: file offset to start truncating |
2418 | * |
2419 | * NOTE! We have to be ready to update the memory sharing |
2420 | * between the file and the memory map for a potential last |
2421 | * incomplete page. Ugly, but necessary. |
2422 | */ |
2423 | int vmtruncate(struct inode * inode, loff_t offset) |
2424 | { |
2425 | if (inode->i_size < offset) { |
2426 | unsigned long limit; |
2427 | |
2428 | limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; |
2429 | if (limit != RLIM_INFINITY && offset > limit) |
2430 | goto out_sig; |
2431 | if (offset > inode->i_sb->s_maxbytes) |
2432 | goto out_big; |
2433 | i_size_write(inode, offset); |
2434 | } else { |
2435 | struct address_space *mapping = inode->i_mapping; |
2436 | |
2437 | /* |
2438 | * truncation of in-use swapfiles is disallowed - it would |
2439 | * cause subsequent swapout to scribble on the now-freed |
2440 | * blocks. |
2441 | */ |
2442 | if (IS_SWAPFILE(inode)) |
2443 | return -ETXTBSY; |
2444 | i_size_write(inode, offset); |
2445 | |
2446 | /* |
2447 | * unmap_mapping_range is called twice, first simply for |
2448 | * efficiency so that truncate_inode_pages does fewer |
2449 | * single-page unmaps. However after this first call, and |
2450 | * before truncate_inode_pages finishes, it is possible for |
2451 | * private pages to be COWed, which remain after |
2452 | * truncate_inode_pages finishes, hence the second |
2453 | * unmap_mapping_range call must be made for correctness. |
2454 | */ |
2455 | unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1); |
2456 | truncate_inode_pages(mapping, offset); |
2457 | unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1); |
2458 | } |
2459 | |
2460 | if (inode->i_op->truncate) |
2461 | inode->i_op->truncate(inode); |
2462 | return 0; |
2463 | |
2464 | out_sig: |
2465 | send_sig(SIGXFSZ, current, 0); |
2466 | out_big: |
2467 | return -EFBIG; |
2468 | } |
2469 | EXPORT_SYMBOL(vmtruncate); |
2470 | |
2471 | int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end) |
2472 | { |
2473 | struct address_space *mapping = inode->i_mapping; |
2474 | |
2475 | /* |
2476 | * If the underlying filesystem is not going to provide |
2477 | * a way to truncate a range of blocks (punch a hole) - |
2478 | * we should return failure right now. |
2479 | */ |
2480 | if (!inode->i_op->truncate_range) |
2481 | return -ENOSYS; |
2482 | |
2483 | mutex_lock(&inode->i_mutex); |
2484 | down_write(&inode->i_alloc_sem); |
2485 | unmap_mapping_range(mapping, offset, (end - offset), 1); |
2486 | truncate_inode_pages_range(mapping, offset, end); |
2487 | unmap_mapping_range(mapping, offset, (end - offset), 1); |
2488 | inode->i_op->truncate_range(inode, offset, end); |
2489 | up_write(&inode->i_alloc_sem); |
2490 | mutex_unlock(&inode->i_mutex); |
2491 | |
2492 | return 0; |
2493 | } |
2494 | EXPORT_SYMBOL_GPL(vmtruncate_range); |
2495 | |
2496 | /* |
2497 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2498 | * but allow concurrent faults), and pte mapped but not yet locked. |
2499 | * We return with mmap_sem still held, but pte unmapped and unlocked. |
2500 | */ |
2501 | static int do_swap_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2502 | unsigned long address, pte_t *page_table, pmd_t *pmd, |
2503 | unsigned int flags, pte_t orig_pte) |
2504 | { |
2505 | spinlock_t *ptl; |
2506 | struct page *page; |
2507 | swp_entry_t entry; |
2508 | pte_t pte; |
2509 | struct mem_cgroup *ptr = NULL; |
2510 | int ret = 0; |
2511 | |
2512 | if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) |
2513 | goto out; |
2514 | |
2515 | entry = pte_to_swp_entry(orig_pte); |
2516 | if (is_migration_entry(entry)) { |
2517 | migration_entry_wait(mm, pmd, address); |
2518 | goto out; |
2519 | } |
2520 | delayacct_set_flag(DELAYACCT_PF_SWAPIN); |
2521 | page = lookup_swap_cache(entry); |
2522 | if (!page) { |
2523 | grab_swap_token(mm); /* Contend for token _before_ read-in */ |
2524 | page = swapin_readahead(entry, |
2525 | GFP_HIGHUSER_MOVABLE, vma, address); |
2526 | if (!page) { |
2527 | /* |
2528 | * Back out if somebody else faulted in this pte |
2529 | * while we released the pte lock. |
2530 | */ |
2531 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
2532 | if (likely(pte_same(*page_table, orig_pte))) |
2533 | ret = VM_FAULT_OOM; |
2534 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
2535 | goto unlock; |
2536 | } |
2537 | |
2538 | /* Had to read the page from swap area: Major fault */ |
2539 | ret = VM_FAULT_MAJOR; |
2540 | count_vm_event(PGMAJFAULT); |
2541 | } |
2542 | |
2543 | lock_page(page); |
2544 | delayacct_clear_flag(DELAYACCT_PF_SWAPIN); |
2545 | |
2546 | if (mem_cgroup_try_charge_swapin(mm, page, GFP_KERNEL, &ptr)) { |
2547 | ret = VM_FAULT_OOM; |
2548 | goto out_page; |
2549 | } |
2550 | |
2551 | /* |
2552 | * Back out if somebody else already faulted in this pte. |
2553 | */ |
2554 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
2555 | if (unlikely(!pte_same(*page_table, orig_pte))) |
2556 | goto out_nomap; |
2557 | |
2558 | if (unlikely(!PageUptodate(page))) { |
2559 | ret = VM_FAULT_SIGBUS; |
2560 | goto out_nomap; |
2561 | } |
2562 | |
2563 | /* |
2564 | * The page isn't present yet, go ahead with the fault. |
2565 | * |
2566 | * Be careful about the sequence of operations here. |
2567 | * To get its accounting right, reuse_swap_page() must be called |
2568 | * while the page is counted on swap but not yet in mapcount i.e. |
2569 | * before page_add_anon_rmap() and swap_free(); try_to_free_swap() |
2570 | * must be called after the swap_free(), or it will never succeed. |
2571 | * Because delete_from_swap_page() may be called by reuse_swap_page(), |
2572 | * mem_cgroup_commit_charge_swapin() may not be able to find swp_entry |
2573 | * in page->private. In this case, a record in swap_cgroup is silently |
2574 | * discarded at swap_free(). |
2575 | */ |
2576 | |
2577 | inc_mm_counter(mm, anon_rss); |
2578 | pte = mk_pte(page, vma->vm_page_prot); |
2579 | if ((flags & FAULT_FLAG_WRITE) && reuse_swap_page(page)) { |
2580 | pte = maybe_mkwrite(pte_mkdirty(pte), vma); |
2581 | flags &= ~FAULT_FLAG_WRITE; |
2582 | } |
2583 | flush_icache_page(vma, page); |
2584 | set_pte_at(mm, address, page_table, pte); |
2585 | page_add_anon_rmap(page, vma, address); |
2586 | /* It's better to call commit-charge after rmap is established */ |
2587 | mem_cgroup_commit_charge_swapin(page, ptr); |
2588 | |
2589 | swap_free(entry); |
2590 | if (vm_swap_full() || (vma->vm_flags & VM_LOCKED) || PageMlocked(page)) |
2591 | try_to_free_swap(page); |
2592 | unlock_page(page); |
2593 | |
2594 | if (flags & FAULT_FLAG_WRITE) { |
2595 | ret |= do_wp_page(mm, vma, address, page_table, pmd, ptl, pte); |
2596 | if (ret & VM_FAULT_ERROR) |
2597 | ret &= VM_FAULT_ERROR; |
2598 | goto out; |
2599 | } |
2600 | |
2601 | /* No need to invalidate - it was non-present before */ |
2602 | update_mmu_cache(vma, address, pte); |
2603 | unlock: |
2604 | pte_unmap_unlock(page_table, ptl); |
2605 | out: |
2606 | return ret; |
2607 | out_nomap: |
2608 | mem_cgroup_cancel_charge_swapin(ptr); |
2609 | pte_unmap_unlock(page_table, ptl); |
2610 | out_page: |
2611 | unlock_page(page); |
2612 | page_cache_release(page); |
2613 | return ret; |
2614 | } |
2615 | |
2616 | /* |
2617 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2618 | * but allow concurrent faults), and pte mapped but not yet locked. |
2619 | * We return with mmap_sem still held, but pte unmapped and unlocked. |
2620 | */ |
2621 | static int do_anonymous_page(struct mm_struct *mm, struct vm_area_struct *vma, |
2622 | unsigned long address, pte_t *page_table, pmd_t *pmd, |
2623 | unsigned int flags) |
2624 | { |
2625 | struct page *page; |
2626 | spinlock_t *ptl; |
2627 | pte_t entry; |
2628 | |
2629 | /* Allocate our own private page. */ |
2630 | pte_unmap(page_table); |
2631 | |
2632 | if (unlikely(anon_vma_prepare(vma))) |
2633 | goto oom; |
2634 | page = alloc_zeroed_user_highpage_movable(vma, address); |
2635 | if (!page) |
2636 | goto oom; |
2637 | __SetPageUptodate(page); |
2638 | |
2639 | if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL)) |
2640 | goto oom_free_page; |
2641 | |
2642 | entry = mk_pte(page, vma->vm_page_prot); |
2643 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
2644 | |
2645 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
2646 | if (!pte_none(*page_table)) |
2647 | goto release; |
2648 | inc_mm_counter(mm, anon_rss); |
2649 | page_add_new_anon_rmap(page, vma, address); |
2650 | set_pte_at(mm, address, page_table, entry); |
2651 | |
2652 | /* No need to invalidate - it was non-present before */ |
2653 | update_mmu_cache(vma, address, entry); |
2654 | unlock: |
2655 | pte_unmap_unlock(page_table, ptl); |
2656 | return 0; |
2657 | release: |
2658 | mem_cgroup_uncharge_page(page); |
2659 | page_cache_release(page); |
2660 | goto unlock; |
2661 | oom_free_page: |
2662 | page_cache_release(page); |
2663 | oom: |
2664 | return VM_FAULT_OOM; |
2665 | } |
2666 | |
2667 | /* |
2668 | * __do_fault() tries to create a new page mapping. It aggressively |
2669 | * tries to share with existing pages, but makes a separate copy if |
2670 | * the FAULT_FLAG_WRITE is set in the flags parameter in order to avoid |
2671 | * the next page fault. |
2672 | * |
2673 | * As this is called only for pages that do not currently exist, we |
2674 | * do not need to flush old virtual caches or the TLB. |
2675 | * |
2676 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2677 | * but allow concurrent faults), and pte neither mapped nor locked. |
2678 | * We return with mmap_sem still held, but pte unmapped and unlocked. |
2679 | */ |
2680 | static int __do_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
2681 | unsigned long address, pmd_t *pmd, |
2682 | pgoff_t pgoff, unsigned int flags, pte_t orig_pte) |
2683 | { |
2684 | pte_t *page_table; |
2685 | spinlock_t *ptl; |
2686 | struct page *page; |
2687 | pte_t entry; |
2688 | int anon = 0; |
2689 | int charged = 0; |
2690 | struct page *dirty_page = NULL; |
2691 | struct vm_fault vmf; |
2692 | int ret; |
2693 | int page_mkwrite = 0; |
2694 | |
2695 | vmf.virtual_address = (void __user *)(address & PAGE_MASK); |
2696 | vmf.pgoff = pgoff; |
2697 | vmf.flags = flags; |
2698 | vmf.page = NULL; |
2699 | |
2700 | ret = vma->vm_ops->fault(vma, &vmf); |
2701 | if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) |
2702 | return ret; |
2703 | |
2704 | /* |
2705 | * For consistency in subsequent calls, make the faulted page always |
2706 | * locked. |
2707 | */ |
2708 | if (unlikely(!(ret & VM_FAULT_LOCKED))) |
2709 | lock_page(vmf.page); |
2710 | else |
2711 | VM_BUG_ON(!PageLocked(vmf.page)); |
2712 | |
2713 | /* |
2714 | * Should we do an early C-O-W break? |
2715 | */ |
2716 | page = vmf.page; |
2717 | if (flags & FAULT_FLAG_WRITE) { |
2718 | if (!(vma->vm_flags & VM_SHARED)) { |
2719 | anon = 1; |
2720 | if (unlikely(anon_vma_prepare(vma))) { |
2721 | ret = VM_FAULT_OOM; |
2722 | goto out; |
2723 | } |
2724 | page = alloc_page_vma(GFP_HIGHUSER_MOVABLE, |
2725 | vma, address); |
2726 | if (!page) { |
2727 | ret = VM_FAULT_OOM; |
2728 | goto out; |
2729 | } |
2730 | if (mem_cgroup_newpage_charge(page, mm, GFP_KERNEL)) { |
2731 | ret = VM_FAULT_OOM; |
2732 | page_cache_release(page); |
2733 | goto out; |
2734 | } |
2735 | charged = 1; |
2736 | /* |
2737 | * Don't let another task, with possibly unlocked vma, |
2738 | * keep the mlocked page. |
2739 | */ |
2740 | if (vma->vm_flags & VM_LOCKED) |
2741 | clear_page_mlock(vmf.page); |
2742 | copy_user_highpage(page, vmf.page, address, vma); |
2743 | __SetPageUptodate(page); |
2744 | } else { |
2745 | /* |
2746 | * If the page will be shareable, see if the backing |
2747 | * address space wants to know that the page is about |
2748 | * to become writable |
2749 | */ |
2750 | if (vma->vm_ops->page_mkwrite) { |
2751 | int tmp; |
2752 | |
2753 | unlock_page(page); |
2754 | vmf.flags = FAULT_FLAG_WRITE|FAULT_FLAG_MKWRITE; |
2755 | tmp = vma->vm_ops->page_mkwrite(vma, &vmf); |
2756 | if (unlikely(tmp & |
2757 | (VM_FAULT_ERROR | VM_FAULT_NOPAGE))) { |
2758 | ret = tmp; |
2759 | goto unwritable_page; |
2760 | } |
2761 | if (unlikely(!(tmp & VM_FAULT_LOCKED))) { |
2762 | lock_page(page); |
2763 | if (!page->mapping) { |
2764 | ret = 0; /* retry the fault */ |
2765 | unlock_page(page); |
2766 | goto unwritable_page; |
2767 | } |
2768 | } else |
2769 | VM_BUG_ON(!PageLocked(page)); |
2770 | page_mkwrite = 1; |
2771 | } |
2772 | } |
2773 | |
2774 | } |
2775 | |
2776 | page_table = pte_offset_map_lock(mm, pmd, address, &ptl); |
2777 | |
2778 | /* |
2779 | * This silly early PAGE_DIRTY setting removes a race |
2780 | * due to the bad i386 page protection. But it's valid |
2781 | * for other architectures too. |
2782 | * |
2783 | * Note that if FAULT_FLAG_WRITE is set, we either now have |
2784 | * an exclusive copy of the page, or this is a shared mapping, |
2785 | * so we can make it writable and dirty to avoid having to |
2786 | * handle that later. |
2787 | */ |
2788 | /* Only go through if we didn't race with anybody else... */ |
2789 | if (likely(pte_same(*page_table, orig_pte))) { |
2790 | flush_icache_page(vma, page); |
2791 | entry = mk_pte(page, vma->vm_page_prot); |
2792 | if (flags & FAULT_FLAG_WRITE) |
2793 | entry = maybe_mkwrite(pte_mkdirty(entry), vma); |
2794 | if (anon) { |
2795 | inc_mm_counter(mm, anon_rss); |
2796 | page_add_new_anon_rmap(page, vma, address); |
2797 | } else { |
2798 | inc_mm_counter(mm, file_rss); |
2799 | page_add_file_rmap(page); |
2800 | if (flags & FAULT_FLAG_WRITE) { |
2801 | dirty_page = page; |
2802 | get_page(dirty_page); |
2803 | } |
2804 | } |
2805 | set_pte_at(mm, address, page_table, entry); |
2806 | |
2807 | /* no need to invalidate: a not-present page won't be cached */ |
2808 | update_mmu_cache(vma, address, entry); |
2809 | } else { |
2810 | if (charged) |
2811 | mem_cgroup_uncharge_page(page); |
2812 | if (anon) |
2813 | page_cache_release(page); |
2814 | else |
2815 | anon = 1; /* no anon but release faulted_page */ |
2816 | } |
2817 | |
2818 | pte_unmap_unlock(page_table, ptl); |
2819 | |
2820 | out: |
2821 | if (dirty_page) { |
2822 | struct address_space *mapping = page->mapping; |
2823 | |
2824 | if (set_page_dirty(dirty_page)) |
2825 | page_mkwrite = 1; |
2826 | unlock_page(dirty_page); |
2827 | put_page(dirty_page); |
2828 | if (page_mkwrite && mapping) { |
2829 | /* |
2830 | * Some device drivers do not set page.mapping but still |
2831 | * dirty their pages |
2832 | */ |
2833 | balance_dirty_pages_ratelimited(mapping); |
2834 | } |
2835 | |
2836 | /* file_update_time outside page_lock */ |
2837 | if (vma->vm_file) |
2838 | file_update_time(vma->vm_file); |
2839 | } else { |
2840 | unlock_page(vmf.page); |
2841 | if (anon) |
2842 | page_cache_release(vmf.page); |
2843 | } |
2844 | |
2845 | return ret; |
2846 | |
2847 | unwritable_page: |
2848 | page_cache_release(page); |
2849 | return ret; |
2850 | } |
2851 | |
2852 | static int do_linear_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
2853 | unsigned long address, pte_t *page_table, pmd_t *pmd, |
2854 | unsigned int flags, pte_t orig_pte) |
2855 | { |
2856 | pgoff_t pgoff = (((address & PAGE_MASK) |
2857 | - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; |
2858 | |
2859 | pte_unmap(page_table); |
2860 | return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); |
2861 | } |
2862 | |
2863 | /* |
2864 | * Fault of a previously existing named mapping. Repopulate the pte |
2865 | * from the encoded file_pte if possible. This enables swappable |
2866 | * nonlinear vmas. |
2867 | * |
2868 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2869 | * but allow concurrent faults), and pte mapped but not yet locked. |
2870 | * We return with mmap_sem still held, but pte unmapped and unlocked. |
2871 | */ |
2872 | static int do_nonlinear_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
2873 | unsigned long address, pte_t *page_table, pmd_t *pmd, |
2874 | unsigned int flags, pte_t orig_pte) |
2875 | { |
2876 | pgoff_t pgoff; |
2877 | |
2878 | flags |= FAULT_FLAG_NONLINEAR; |
2879 | |
2880 | if (!pte_unmap_same(mm, pmd, page_table, orig_pte)) |
2881 | return 0; |
2882 | |
2883 | if (unlikely(!(vma->vm_flags & VM_NONLINEAR))) { |
2884 | /* |
2885 | * Page table corrupted: show pte and kill process. |
2886 | */ |
2887 | print_bad_pte(vma, address, orig_pte, NULL); |
2888 | return VM_FAULT_OOM; |
2889 | } |
2890 | |
2891 | pgoff = pte_to_pgoff(orig_pte); |
2892 | return __do_fault(mm, vma, address, pmd, pgoff, flags, orig_pte); |
2893 | } |
2894 | |
2895 | /* |
2896 | * These routines also need to handle stuff like marking pages dirty |
2897 | * and/or accessed for architectures that don't do it in hardware (most |
2898 | * RISC architectures). The early dirtying is also good on the i386. |
2899 | * |
2900 | * There is also a hook called "update_mmu_cache()" that architectures |
2901 | * with external mmu caches can use to update those (ie the Sparc or |
2902 | * PowerPC hashed page tables that act as extended TLBs). |
2903 | * |
2904 | * We enter with non-exclusive mmap_sem (to exclude vma changes, |
2905 | * but allow concurrent faults), and pte mapped but not yet locked. |
2906 | * We return with mmap_sem still held, but pte unmapped and unlocked. |
2907 | */ |
2908 | static inline int handle_pte_fault(struct mm_struct *mm, |
2909 | struct vm_area_struct *vma, unsigned long address, |
2910 | pte_t *pte, pmd_t *pmd, unsigned int flags) |
2911 | { |
2912 | pte_t entry; |
2913 | spinlock_t *ptl; |
2914 | |
2915 | entry = *pte; |
2916 | if (!pte_present(entry)) { |
2917 | if (pte_none(entry)) { |
2918 | if (vma->vm_ops) { |
2919 | if (likely(vma->vm_ops->fault)) |
2920 | return do_linear_fault(mm, vma, address, |
2921 | pte, pmd, flags, entry); |
2922 | } |
2923 | return do_anonymous_page(mm, vma, address, |
2924 | pte, pmd, flags); |
2925 | } |
2926 | if (pte_file(entry)) |
2927 | return do_nonlinear_fault(mm, vma, address, |
2928 | pte, pmd, flags, entry); |
2929 | return do_swap_page(mm, vma, address, |
2930 | pte, pmd, flags, entry); |
2931 | } |
2932 | |
2933 | ptl = pte_lockptr(mm, pmd); |
2934 | spin_lock(ptl); |
2935 | if (unlikely(!pte_same(*pte, entry))) |
2936 | goto unlock; |
2937 | if (flags & FAULT_FLAG_WRITE) { |
2938 | if (!pte_write(entry)) |
2939 | return do_wp_page(mm, vma, address, |
2940 | pte, pmd, ptl, entry); |
2941 | entry = pte_mkdirty(entry); |
2942 | } |
2943 | entry = pte_mkyoung(entry); |
2944 | if (ptep_set_access_flags(vma, address, pte, entry, flags & FAULT_FLAG_WRITE)) { |
2945 | update_mmu_cache(vma, address, entry); |
2946 | } else { |
2947 | /* |
2948 | * This is needed only for protection faults but the arch code |
2949 | * is not yet telling us if this is a protection fault or not. |
2950 | * This still avoids useless tlb flushes for .text page faults |
2951 | * with threads. |
2952 | */ |
2953 | if (flags & FAULT_FLAG_WRITE) |
2954 | flush_tlb_page(vma, address); |
2955 | } |
2956 | unlock: |
2957 | pte_unmap_unlock(pte, ptl); |
2958 | return 0; |
2959 | } |
2960 | |
2961 | /* |
2962 | * By the time we get here, we already hold the mm semaphore |
2963 | */ |
2964 | int handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma, |
2965 | unsigned long address, unsigned int flags) |
2966 | { |
2967 | pgd_t *pgd; |
2968 | pud_t *pud; |
2969 | pmd_t *pmd; |
2970 | pte_t *pte; |
2971 | |
2972 | __set_current_state(TASK_RUNNING); |
2973 | |
2974 | count_vm_event(PGFAULT); |
2975 | |
2976 | if (unlikely(is_vm_hugetlb_page(vma))) |
2977 | return hugetlb_fault(mm, vma, address, flags); |
2978 | |
2979 | pgd = pgd_offset(mm, address); |
2980 | pud = pud_alloc(mm, pgd, address); |
2981 | if (!pud) |
2982 | return VM_FAULT_OOM; |
2983 | pmd = pmd_alloc(mm, pud, address); |
2984 | if (!pmd) |
2985 | return VM_FAULT_OOM; |
2986 | pte = pte_alloc_map(mm, pmd, address); |
2987 | if (!pte) |
2988 | return VM_FAULT_OOM; |
2989 | |
2990 | return handle_pte_fault(mm, vma, address, pte, pmd, flags); |
2991 | } |
2992 | |
2993 | #ifndef __PAGETABLE_PUD_FOLDED |
2994 | /* |
2995 | * Allocate page upper directory. |
2996 | * We've already handled the fast-path in-line. |
2997 | */ |
2998 | int __pud_alloc(struct mm_struct *mm, pgd_t *pgd, unsigned long address) |
2999 | { |
3000 | pud_t *new = pud_alloc_one(mm, address); |
3001 | if (!new) |
3002 | return -ENOMEM; |
3003 | |
3004 | smp_wmb(); /* See comment in __pte_alloc */ |
3005 | |
3006 | spin_lock(&mm->page_table_lock); |
3007 | if (pgd_present(*pgd)) /* Another has populated it */ |
3008 | pud_free(mm, new); |
3009 | else |
3010 | pgd_populate(mm, pgd, new); |
3011 | spin_unlock(&mm->page_table_lock); |
3012 | return 0; |
3013 | } |
3014 | #endif /* __PAGETABLE_PUD_FOLDED */ |
3015 | |
3016 | #ifndef __PAGETABLE_PMD_FOLDED |
3017 | /* |
3018 | * Allocate page middle directory. |
3019 | * We've already handled the fast-path in-line. |
3020 | */ |
3021 | int __pmd_alloc(struct mm_struct *mm, pud_t *pud, unsigned long address) |
3022 | { |
3023 | pmd_t *new = pmd_alloc_one(mm, address); |
3024 | if (!new) |
3025 | return -ENOMEM; |
3026 | |
3027 | smp_wmb(); /* See comment in __pte_alloc */ |
3028 | |
3029 | spin_lock(&mm->page_table_lock); |
3030 | #ifndef __ARCH_HAS_4LEVEL_HACK |
3031 | if (pud_present(*pud)) /* Another has populated it */ |
3032 | pmd_free(mm, new); |
3033 | else |
3034 | pud_populate(mm, pud, new); |
3035 | #else |
3036 | if (pgd_present(*pud)) /* Another has populated it */ |
3037 | pmd_free(mm, new); |
3038 | else |
3039 | pgd_populate(mm, pud, new); |
3040 | #endif /* __ARCH_HAS_4LEVEL_HACK */ |
3041 | spin_unlock(&mm->page_table_lock); |
3042 | return 0; |
3043 | } |
3044 | #endif /* __PAGETABLE_PMD_FOLDED */ |
3045 | |
3046 | int make_pages_present(unsigned long addr, unsigned long end) |
3047 | { |
3048 | int ret, len, write; |
3049 | struct vm_area_struct * vma; |
3050 | |
3051 | vma = find_vma(current->mm, addr); |
3052 | if (!vma) |
3053 | return -ENOMEM; |
3054 | write = (vma->vm_flags & VM_WRITE) != 0; |
3055 | BUG_ON(addr >= end); |
3056 | BUG_ON(end > vma->vm_end); |
3057 | len = DIV_ROUND_UP(end, PAGE_SIZE) - addr/PAGE_SIZE; |
3058 | ret = get_user_pages(current, current->mm, addr, |
3059 | len, write, 0, NULL, NULL); |
3060 | if (ret < 0) |
3061 | return ret; |
3062 | return ret == len ? 0 : -EFAULT; |
3063 | } |
3064 | |
3065 | #if !defined(__HAVE_ARCH_GATE_AREA) |
3066 | |
3067 | #if defined(AT_SYSINFO_EHDR) |
3068 | static struct vm_area_struct gate_vma; |
3069 | |
3070 | static int __init gate_vma_init(void) |
3071 | { |
3072 | gate_vma.vm_mm = NULL; |
3073 | gate_vma.vm_start = FIXADDR_USER_START; |
3074 | gate_vma.vm_end = FIXADDR_USER_END; |
3075 | gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC; |
3076 | gate_vma.vm_page_prot = __P101; |
3077 | /* |
3078 | * Make sure the vDSO gets into every core dump. |
3079 | * Dumping its contents makes post-mortem fully interpretable later |
3080 | * without matching up the same kernel and hardware config to see |
3081 | * what PC values meant. |
3082 | */ |
3083 | gate_vma.vm_flags |= VM_ALWAYSDUMP; |
3084 | return 0; |
3085 | } |
3086 | __initcall(gate_vma_init); |
3087 | #endif |
3088 | |
3089 | struct vm_area_struct *get_gate_vma(struct task_struct *tsk) |
3090 | { |
3091 | #ifdef AT_SYSINFO_EHDR |
3092 | return &gate_vma; |
3093 | #else |
3094 | return NULL; |
3095 | #endif |
3096 | } |
3097 | |
3098 | int in_gate_area_no_task(unsigned long addr) |
3099 | { |
3100 | #ifdef AT_SYSINFO_EHDR |
3101 | if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END)) |
3102 | return 1; |
3103 | #endif |
3104 | return 0; |
3105 | } |
3106 | |
3107 | #endif /* __HAVE_ARCH_GATE_AREA */ |
3108 | |
3109 | static int follow_pte(struct mm_struct *mm, unsigned long address, |
3110 | pte_t **ptepp, spinlock_t **ptlp) |
3111 | { |
3112 | pgd_t *pgd; |
3113 | pud_t *pud; |
3114 | pmd_t *pmd; |
3115 | pte_t *ptep; |
3116 | |
3117 | pgd = pgd_offset(mm, address); |
3118 | if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd))) |
3119 | goto out; |
3120 | |
3121 | pud = pud_offset(pgd, address); |
3122 | if (pud_none(*pud) || unlikely(pud_bad(*pud))) |
3123 | goto out; |
3124 | |
3125 | pmd = pmd_offset(pud, address); |
3126 | if (pmd_none(*pmd) || unlikely(pmd_bad(*pmd))) |
3127 | goto out; |
3128 | |
3129 | /* We cannot handle huge page PFN maps. Luckily they don't exist. */ |
3130 | if (pmd_huge(*pmd)) |
3131 | goto out; |
3132 | |
3133 | ptep = pte_offset_map_lock(mm, pmd, address, ptlp); |
3134 | if (!ptep) |
3135 | goto out; |
3136 | if (!pte_present(*ptep)) |
3137 | goto unlock; |
3138 | *ptepp = ptep; |
3139 | return 0; |
3140 | unlock: |
3141 | pte_unmap_unlock(ptep, *ptlp); |
3142 | out: |
3143 | return -EINVAL; |
3144 | } |
3145 | |
3146 | /** |
3147 | * follow_pfn - look up PFN at a user virtual address |
3148 | * @vma: memory mapping |
3149 | * @address: user virtual address |
3150 | * @pfn: location to store found PFN |
3151 | * |
3152 | * Only IO mappings and raw PFN mappings are allowed. |
3153 | * |
3154 | * Returns zero and the pfn at @pfn on success, -ve otherwise. |
3155 | */ |
3156 | int follow_pfn(struct vm_area_struct *vma, unsigned long address, |
3157 | unsigned long *pfn) |
3158 | { |
3159 | int ret = -EINVAL; |
3160 | spinlock_t *ptl; |
3161 | pte_t *ptep; |
3162 | |
3163 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
3164 | return ret; |
3165 | |
3166 | ret = follow_pte(vma->vm_mm, address, &ptep, &ptl); |
3167 | if (ret) |
3168 | return ret; |
3169 | *pfn = pte_pfn(*ptep); |
3170 | pte_unmap_unlock(ptep, ptl); |
3171 | return 0; |
3172 | } |
3173 | EXPORT_SYMBOL(follow_pfn); |
3174 | |
3175 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
3176 | int follow_phys(struct vm_area_struct *vma, |
3177 | unsigned long address, unsigned int flags, |
3178 | unsigned long *prot, resource_size_t *phys) |
3179 | { |
3180 | int ret = -EINVAL; |
3181 | pte_t *ptep, pte; |
3182 | spinlock_t *ptl; |
3183 | |
3184 | if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) |
3185 | goto out; |
3186 | |
3187 | if (follow_pte(vma->vm_mm, address, &ptep, &ptl)) |
3188 | goto out; |
3189 | pte = *ptep; |
3190 | |
3191 | if ((flags & FOLL_WRITE) && !pte_write(pte)) |
3192 | goto unlock; |
3193 | |
3194 | *prot = pgprot_val(pte_pgprot(pte)); |
3195 | *phys = (resource_size_t)pte_pfn(pte) << PAGE_SHIFT; |
3196 | |
3197 | ret = 0; |
3198 | unlock: |
3199 | pte_unmap_unlock(ptep, ptl); |
3200 | out: |
3201 | return ret; |
3202 | } |
3203 | |
3204 | int generic_access_phys(struct vm_area_struct *vma, unsigned long addr, |
3205 | void *buf, int len, int write) |
3206 | { |
3207 | resource_size_t phys_addr; |
3208 | unsigned long prot = 0; |
3209 | void __iomem *maddr; |
3210 | int offset = addr & (PAGE_SIZE-1); |
3211 | |
3212 | if (follow_phys(vma, addr, write, &prot, &phys_addr)) |
3213 | return -EINVAL; |
3214 | |
3215 | maddr = ioremap_prot(phys_addr, PAGE_SIZE, prot); |
3216 | if (write) |
3217 | memcpy_toio(maddr + offset, buf, len); |
3218 | else |
3219 | memcpy_fromio(buf, maddr + offset, len); |
3220 | iounmap(maddr); |
3221 | |
3222 | return len; |
3223 | } |
3224 | #endif |
3225 | |
3226 | /* |
3227 | * Access another process' address space. |
3228 | * Source/target buffer must be kernel space, |
3229 | * Do not walk the page table directly, use get_user_pages |
3230 | */ |
3231 | int access_process_vm(struct task_struct *tsk, unsigned long addr, void *buf, int len, int write) |
3232 | { |
3233 | struct mm_struct *mm; |
3234 | struct vm_area_struct *vma; |
3235 | void *old_buf = buf; |
3236 | |
3237 | mm = get_task_mm(tsk); |
3238 | if (!mm) |
3239 | return 0; |
3240 | |
3241 | down_read(&mm->mmap_sem); |
3242 | /* ignore errors, just check how much was successfully transferred */ |
3243 | while (len) { |
3244 | int bytes, ret, offset; |
3245 | void *maddr; |
3246 | struct page *page = NULL; |
3247 | |
3248 | ret = get_user_pages(tsk, mm, addr, 1, |
3249 | write, 1, &page, &vma); |
3250 | if (ret <= 0) { |
3251 | /* |
3252 | * Check if this is a VM_IO | VM_PFNMAP VMA, which |
3253 | * we can access using slightly different code. |
3254 | */ |
3255 | #ifdef CONFIG_HAVE_IOREMAP_PROT |
3256 | vma = find_vma(mm, addr); |
3257 | if (!vma) |
3258 | break; |
3259 | if (vma->vm_ops && vma->vm_ops->access) |
3260 | ret = vma->vm_ops->access(vma, addr, buf, |
3261 | len, write); |
3262 | if (ret <= 0) |
3263 | #endif |
3264 | break; |
3265 | bytes = ret; |
3266 | } else { |
3267 | bytes = len; |
3268 | offset = addr & (PAGE_SIZE-1); |
3269 | if (bytes > PAGE_SIZE-offset) |
3270 | bytes = PAGE_SIZE-offset; |
3271 | |
3272 | maddr = kmap(page); |
3273 | if (write) { |
3274 | copy_to_user_page(vma, page, addr, |
3275 | maddr + offset, buf, bytes); |
3276 | set_page_dirty_lock(page); |
3277 | } else { |
3278 | copy_from_user_page(vma, page, addr, |
3279 | buf, maddr + offset, bytes); |
3280 | } |
3281 | kunmap(page); |
3282 | page_cache_release(page); |
3283 | } |
3284 | len -= bytes; |
3285 | buf += bytes; |
3286 | addr += bytes; |
3287 | } |
3288 | up_read(&mm->mmap_sem); |
3289 | mmput(mm); |
3290 | |
3291 | return buf - old_buf; |
3292 | } |
3293 | |
3294 | /* |
3295 | * Print the name of a VMA. |
3296 | */ |
3297 | void print_vma_addr(char *prefix, unsigned long ip) |
3298 | { |
3299 | struct mm_struct *mm = current->mm; |
3300 | struct vm_area_struct *vma; |
3301 | |
3302 | /* |
3303 | * Do not print if we are in atomic |
3304 | * contexts (in exception stacks, etc.): |
3305 | */ |
3306 | if (preempt_count()) |
3307 | return; |
3308 | |
3309 | down_read(&mm->mmap_sem); |
3310 | vma = find_vma(mm, ip); |
3311 | if (vma && vma->vm_file) { |
3312 | struct file *f = vma->vm_file; |
3313 | char *buf = (char *)__get_free_page(GFP_KERNEL); |
3314 | if (buf) { |
3315 | char *p, *s; |
3316 | |
3317 | p = d_path(&f->f_path, buf, PAGE_SIZE); |
3318 | if (IS_ERR(p)) |
3319 | p = "?"; |
3320 | s = strrchr(p, '/'); |
3321 | if (s) |
3322 | p = s+1; |
3323 | printk("%s%s[%lx+%lx]", prefix, p, |
3324 | vma->vm_start, |
3325 | vma->vm_end - vma->vm_start); |
3326 | free_page((unsigned long)buf); |
3327 | } |
3328 | } |
3329 | up_read(¤t->mm->mmap_sem); |
3330 | } |
3331 | |
3332 | #ifdef CONFIG_PROVE_LOCKING |
3333 | void might_fault(void) |
3334 | { |
3335 | /* |
3336 | * Some code (nfs/sunrpc) uses socket ops on kernel memory while |
3337 | * holding the mmap_sem, this is safe because kernel memory doesn't |
3338 | * get paged out, therefore we'll never actually fault, and the |
3339 | * below annotations will generate false positives. |
3340 | */ |
3341 | if (segment_eq(get_fs(), KERNEL_DS)) |
3342 | return; |
3343 | |
3344 | might_sleep(); |
3345 | /* |
3346 | * it would be nicer only to annotate paths which are not under |
3347 | * pagefault_disable, however that requires a larger audit and |
3348 | * providing helpers like get_user_atomic. |
3349 | */ |
3350 | if (!in_atomic() && current->mm) |
3351 | might_lock_read(¤t->mm->mmap_sem); |
3352 | } |
3353 | EXPORT_SYMBOL(might_fault); |
3354 | #endif |
3355 |
Branches:
ben-wpan
ben-wpan-stefan
javiroman/ks7010
jz-2.6.34
jz-2.6.34-rc5
jz-2.6.34-rc6
jz-2.6.34-rc7
jz-2.6.35
jz-2.6.36
jz-2.6.37
jz-2.6.38
jz-2.6.39
jz-3.0
jz-3.1
jz-3.11
jz-3.12
jz-3.13
jz-3.15
jz-3.16
jz-3.18-dt
jz-3.2
jz-3.3
jz-3.4
jz-3.5
jz-3.6
jz-3.6-rc2-pwm
jz-3.9
jz-3.9-clk
jz-3.9-rc8
jz47xx
jz47xx-2.6.38
master
Tags:
od-2011-09-04
od-2011-09-18
v2.6.34-rc5
v2.6.34-rc6
v2.6.34-rc7
v3.9