blob: 96b2b2fd0fbd13f0b7385e7adc3359c6c7793503 [file] [log] [blame]
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -07001#include <linux/kernel.h>
2#include <linux/errno.h>
3#include <linux/err.h>
4#include <linux/spinlock.h>
5
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -07006#include <linux/mm.h>
Dan Williams3565fce2016-01-15 16:56:55 -08007#include <linux/memremap.h>
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -07008#include <linux/pagemap.h>
9#include <linux/rmap.h>
10#include <linux/swap.h>
11#include <linux/swapops.h>
12
Steve Capper2667f502014-10-09 15:29:14 -070013#include <linux/sched.h>
14#include <linux/rwsem.h>
Aneesh Kumar K.Vf30c59e2014-11-05 21:57:40 +053015#include <linux/hugetlb.h>
Kirill A. Shutemov1027e442015-09-04 15:47:55 -070016
Dave Hansen33a709b2016-02-12 13:02:19 -080017#include <asm/mmu_context.h>
Steve Capper2667f502014-10-09 15:29:14 -070018#include <asm/pgtable.h>
Kirill A. Shutemov1027e442015-09-04 15:47:55 -070019#include <asm/tlbflush.h>
Steve Capper2667f502014-10-09 15:29:14 -070020
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -070021#include "internal.h"
22
Kirill A. Shutemov69e68b42014-06-04 16:08:11 -070023static struct page *no_page_table(struct vm_area_struct *vma,
24 unsigned int flags)
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -070025{
Kirill A. Shutemov69e68b42014-06-04 16:08:11 -070026 /*
27 * When core dumping an enormous anonymous area that nobody
28 * has touched so far, we don't want to allocate unnecessary pages or
29 * page tables. Return error instead of NULL to skip handle_mm_fault,
30 * then get_dump_page() will return NULL to leave a hole in the dump.
31 * But we can only make this optimization where a hole would surely
32 * be zero-filled if handle_mm_fault() actually did handle it.
33 */
34 if ((flags & FOLL_DUMP) && (!vma->vm_ops || !vma->vm_ops->fault))
35 return ERR_PTR(-EFAULT);
36 return NULL;
37}
38
Kirill A. Shutemov1027e442015-09-04 15:47:55 -070039static int follow_pfn_pte(struct vm_area_struct *vma, unsigned long address,
40 pte_t *pte, unsigned int flags)
41{
42 /* No page to get reference */
43 if (flags & FOLL_GET)
44 return -EFAULT;
45
46 if (flags & FOLL_TOUCH) {
47 pte_t entry = *pte;
48
49 if (flags & FOLL_WRITE)
50 entry = pte_mkdirty(entry);
51 entry = pte_mkyoung(entry);
52
53 if (!pte_same(*pte, entry)) {
54 set_pte_at(vma->vm_mm, address, pte, entry);
55 update_mmu_cache(vma, address, pte);
56 }
57 }
58
59 /* Proper page table entry exists, but no corresponding struct page */
60 return -EEXIST;
61}
62
Kirill A. Shutemov69e68b42014-06-04 16:08:11 -070063static struct page *follow_page_pte(struct vm_area_struct *vma,
64 unsigned long address, pmd_t *pmd, unsigned int flags)
65{
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -070066 struct mm_struct *mm = vma->vm_mm;
Dan Williams3565fce2016-01-15 16:56:55 -080067 struct dev_pagemap *pgmap = NULL;
Kirill A. Shutemov69e68b42014-06-04 16:08:11 -070068 struct page *page;
69 spinlock_t *ptl;
70 pte_t *ptep, pte;
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -070071
Kirill A. Shutemov69e68b42014-06-04 16:08:11 -070072retry:
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -070073 if (unlikely(pmd_bad(*pmd)))
Kirill A. Shutemov69e68b42014-06-04 16:08:11 -070074 return no_page_table(vma, flags);
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -070075
76 ptep = pte_offset_map_lock(mm, pmd, address, &ptl);
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -070077 pte = *ptep;
78 if (!pte_present(pte)) {
79 swp_entry_t entry;
80 /*
81 * KSM's break_ksm() relies upon recognizing a ksm page
82 * even while it is being migrated, so for that case we
83 * need migration_entry_wait().
84 */
85 if (likely(!(flags & FOLL_MIGRATION)))
86 goto no_page;
Kirill A. Shutemov0661a332015-02-10 14:10:04 -080087 if (pte_none(pte))
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -070088 goto no_page;
89 entry = pte_to_swp_entry(pte);
90 if (!is_migration_entry(entry))
91 goto no_page;
92 pte_unmap_unlock(ptep, ptl);
93 migration_entry_wait(mm, pmd, address);
Kirill A. Shutemov69e68b42014-06-04 16:08:11 -070094 goto retry;
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -070095 }
Mel Gorman8a0516e2015-02-12 14:58:22 -080096 if ((flags & FOLL_NUMA) && pte_protnone(pte))
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -070097 goto no_page;
Kirill A. Shutemov69e68b42014-06-04 16:08:11 -070098 if ((flags & FOLL_WRITE) && !pte_write(pte)) {
99 pte_unmap_unlock(ptep, ptl);
100 return NULL;
101 }
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700102
103 page = vm_normal_page(vma, address, pte);
Dan Williams3565fce2016-01-15 16:56:55 -0800104 if (!page && pte_devmap(pte) && (flags & FOLL_GET)) {
105 /*
106 * Only return device mapping pages in the FOLL_GET case since
107 * they are only valid while holding the pgmap reference.
108 */
109 pgmap = get_dev_pagemap(pte_pfn(pte), NULL);
110 if (pgmap)
111 page = pte_page(pte);
112 else
113 goto no_page;
114 } else if (unlikely(!page)) {
Kirill A. Shutemov1027e442015-09-04 15:47:55 -0700115 if (flags & FOLL_DUMP) {
116 /* Avoid special (like zero) pages in core dumps */
117 page = ERR_PTR(-EFAULT);
118 goto out;
119 }
120
121 if (is_zero_pfn(pte_pfn(pte))) {
122 page = pte_page(pte);
123 } else {
124 int ret;
125
126 ret = follow_pfn_pte(vma, address, ptep, flags);
127 page = ERR_PTR(ret);
128 goto out;
129 }
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700130 }
131
Kirill A. Shutemov6742d292016-01-15 16:52:28 -0800132 if (flags & FOLL_SPLIT && PageTransCompound(page)) {
133 int ret;
134 get_page(page);
135 pte_unmap_unlock(ptep, ptl);
136 lock_page(page);
137 ret = split_huge_page(page);
138 unlock_page(page);
139 put_page(page);
140 if (ret)
141 return ERR_PTR(ret);
142 goto retry;
143 }
144
Dan Williams3565fce2016-01-15 16:56:55 -0800145 if (flags & FOLL_GET) {
Kirill A. Shutemovddc58f22016-01-15 16:52:56 -0800146 get_page(page);
Dan Williams3565fce2016-01-15 16:56:55 -0800147
148 /* drop the pgmap reference now that we hold the page */
149 if (pgmap) {
150 put_dev_pagemap(pgmap);
151 pgmap = NULL;
152 }
153 }
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700154 if (flags & FOLL_TOUCH) {
155 if ((flags & FOLL_WRITE) &&
156 !pte_dirty(pte) && !PageDirty(page))
157 set_page_dirty(page);
158 /*
159 * pte_mkyoung() would be more correct here, but atomic care
160 * is needed to avoid losing the dirty bit: it is easier to use
161 * mark_page_accessed().
162 */
163 mark_page_accessed(page);
164 }
Eric B Munsonde60f5f2015-11-05 18:51:36 -0800165 if ((flags & FOLL_MLOCK) && (vma->vm_flags & VM_LOCKED)) {
Kirill A. Shutemove90309c2016-01-15 16:54:33 -0800166 /* Do not mlock pte-mapped THP */
167 if (PageTransCompound(page))
168 goto out;
169
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700170 /*
171 * The preliminary mapping check is mainly to avoid the
172 * pointless overhead of lock_page on the ZERO_PAGE
173 * which might bounce very badly if there is contention.
174 *
175 * If the page is already locked, we don't need to
176 * handle it now - vmscan will handle it later if and
177 * when it attempts to reclaim the page.
178 */
179 if (page->mapping && trylock_page(page)) {
180 lru_add_drain(); /* push cached pages to LRU */
181 /*
182 * Because we lock page here, and migration is
183 * blocked by the pte's page reference, and we
184 * know the page is still mapped, we don't even
185 * need to check for file-cache page truncation.
186 */
187 mlock_vma_page(page);
188 unlock_page(page);
189 }
190 }
Kirill A. Shutemov1027e442015-09-04 15:47:55 -0700191out:
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700192 pte_unmap_unlock(ptep, ptl);
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700193 return page;
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700194no_page:
195 pte_unmap_unlock(ptep, ptl);
196 if (!pte_none(pte))
Kirill A. Shutemov69e68b42014-06-04 16:08:11 -0700197 return NULL;
198 return no_page_table(vma, flags);
199}
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700200
Kirill A. Shutemov69e68b42014-06-04 16:08:11 -0700201/**
202 * follow_page_mask - look up a page descriptor from a user-virtual address
203 * @vma: vm_area_struct mapping @address
204 * @address: virtual address to look up
205 * @flags: flags modifying lookup behaviour
206 * @page_mask: on output, *page_mask is set according to the size of the page
207 *
208 * @flags can have FOLL_ flags set, defined in <linux/mm.h>
209 *
210 * Returns the mapped (struct page *), %NULL if no mapping exists, or
211 * an error pointer if there is a mapping to something not represented
212 * by a page descriptor (see also vm_normal_page()).
213 */
214struct page *follow_page_mask(struct vm_area_struct *vma,
215 unsigned long address, unsigned int flags,
216 unsigned int *page_mask)
217{
218 pgd_t *pgd;
219 pud_t *pud;
220 pmd_t *pmd;
221 spinlock_t *ptl;
222 struct page *page;
223 struct mm_struct *mm = vma->vm_mm;
224
225 *page_mask = 0;
226
227 page = follow_huge_addr(mm, address, flags & FOLL_WRITE);
228 if (!IS_ERR(page)) {
229 BUG_ON(flags & FOLL_GET);
230 return page;
231 }
232
233 pgd = pgd_offset(mm, address);
234 if (pgd_none(*pgd) || unlikely(pgd_bad(*pgd)))
235 return no_page_table(vma, flags);
236
237 pud = pud_offset(pgd, address);
238 if (pud_none(*pud))
239 return no_page_table(vma, flags);
240 if (pud_huge(*pud) && vma->vm_flags & VM_HUGETLB) {
Naoya Horiguchie66f17f2015-02-11 15:25:22 -0800241 page = follow_huge_pud(mm, address, pud, flags);
242 if (page)
243 return page;
244 return no_page_table(vma, flags);
Kirill A. Shutemov69e68b42014-06-04 16:08:11 -0700245 }
246 if (unlikely(pud_bad(*pud)))
247 return no_page_table(vma, flags);
248
249 pmd = pmd_offset(pud, address);
250 if (pmd_none(*pmd))
251 return no_page_table(vma, flags);
252 if (pmd_huge(*pmd) && vma->vm_flags & VM_HUGETLB) {
Naoya Horiguchie66f17f2015-02-11 15:25:22 -0800253 page = follow_huge_pmd(mm, address, pmd, flags);
254 if (page)
255 return page;
256 return no_page_table(vma, flags);
Kirill A. Shutemov69e68b42014-06-04 16:08:11 -0700257 }
Mel Gorman8a0516e2015-02-12 14:58:22 -0800258 if ((flags & FOLL_NUMA) && pmd_protnone(*pmd))
Kirill A. Shutemov69e68b42014-06-04 16:08:11 -0700259 return no_page_table(vma, flags);
Dan Williams3565fce2016-01-15 16:56:55 -0800260 if (pmd_devmap(*pmd)) {
261 ptl = pmd_lock(mm, pmd);
262 page = follow_devmap_pmd(vma, address, pmd, flags);
263 spin_unlock(ptl);
264 if (page)
265 return page;
266 }
Kirill A. Shutemov6742d292016-01-15 16:52:28 -0800267 if (likely(!pmd_trans_huge(*pmd)))
268 return follow_page_pte(vma, address, pmd, flags);
269
270 ptl = pmd_lock(mm, pmd);
271 if (unlikely(!pmd_trans_huge(*pmd))) {
272 spin_unlock(ptl);
273 return follow_page_pte(vma, address, pmd, flags);
Kirill A. Shutemov69e68b42014-06-04 16:08:11 -0700274 }
Kirill A. Shutemov6742d292016-01-15 16:52:28 -0800275 if (flags & FOLL_SPLIT) {
276 int ret;
277 page = pmd_page(*pmd);
278 if (is_huge_zero_page(page)) {
279 spin_unlock(ptl);
280 ret = 0;
Kirill A. Shutemov78ddc532016-01-15 16:52:42 -0800281 split_huge_pmd(vma, pmd, address);
Naoya Horiguchi337d9ab2016-07-26 15:24:03 -0700282 if (pmd_trans_unstable(pmd))
283 ret = -EBUSY;
Kirill A. Shutemov6742d292016-01-15 16:52:28 -0800284 } else {
285 get_page(page);
286 spin_unlock(ptl);
287 lock_page(page);
288 ret = split_huge_page(page);
289 unlock_page(page);
290 put_page(page);
Kirill A. Shutemovbaa355f2016-07-26 15:25:51 -0700291 if (pmd_none(*pmd))
292 return no_page_table(vma, flags);
Kirill A. Shutemov6742d292016-01-15 16:52:28 -0800293 }
294
295 return ret ? ERR_PTR(ret) :
296 follow_page_pte(vma, address, pmd, flags);
297 }
298
299 page = follow_trans_huge_pmd(vma, address, pmd, flags);
300 spin_unlock(ptl);
301 *page_mask = HPAGE_PMD_NR - 1;
302 return page;
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700303}
304
Kirill A. Shutemovf2b495c2014-06-04 16:08:11 -0700305static int get_gate_page(struct mm_struct *mm, unsigned long address,
306 unsigned int gup_flags, struct vm_area_struct **vma,
307 struct page **page)
308{
309 pgd_t *pgd;
310 pud_t *pud;
311 pmd_t *pmd;
312 pte_t *pte;
313 int ret = -EFAULT;
314
315 /* user gate pages are read-only */
316 if (gup_flags & FOLL_WRITE)
317 return -EFAULT;
318 if (address > TASK_SIZE)
319 pgd = pgd_offset_k(address);
320 else
321 pgd = pgd_offset_gate(mm, address);
322 BUG_ON(pgd_none(*pgd));
323 pud = pud_offset(pgd, address);
324 BUG_ON(pud_none(*pud));
325 pmd = pmd_offset(pud, address);
326 if (pmd_none(*pmd))
327 return -EFAULT;
328 VM_BUG_ON(pmd_trans_huge(*pmd));
329 pte = pte_offset_map(pmd, address);
330 if (pte_none(*pte))
331 goto unmap;
332 *vma = get_gate_vma(mm);
333 if (!page)
334 goto out;
335 *page = vm_normal_page(*vma, address, *pte);
336 if (!*page) {
337 if ((gup_flags & FOLL_DUMP) || !is_zero_pfn(pte_pfn(*pte)))
338 goto unmap;
339 *page = pte_page(*pte);
340 }
341 get_page(*page);
342out:
343 ret = 0;
344unmap:
345 pte_unmap(pte);
346 return ret;
347}
348
Paul Cassella9a95f3c2014-08-06 16:07:24 -0700349/*
350 * mmap_sem must be held on entry. If @nonblocking != NULL and
351 * *@flags does not include FOLL_NOWAIT, the mmap_sem may be released.
352 * If it is, *@nonblocking will be set to 0 and -EBUSY returned.
353 */
Kirill A. Shutemov16744482014-06-04 16:08:12 -0700354static int faultin_page(struct task_struct *tsk, struct vm_area_struct *vma,
355 unsigned long address, unsigned int *flags, int *nonblocking)
356{
Kirill A. Shutemov16744482014-06-04 16:08:12 -0700357 unsigned int fault_flags = 0;
358 int ret;
359
Eric B Munsonde60f5f2015-11-05 18:51:36 -0800360 /* mlock all present pages, but do not fault in new pages */
361 if ((*flags & (FOLL_POPULATE | FOLL_MLOCK)) == FOLL_MLOCK)
362 return -ENOENT;
Kirill A. Shutemov84d33df2015-04-14 15:44:37 -0700363 /* For mm_populate(), just skip the stack guard page. */
364 if ((*flags & FOLL_POPULATE) &&
Kirill A. Shutemov16744482014-06-04 16:08:12 -0700365 (stack_guard_page_start(vma, address) ||
366 stack_guard_page_end(vma, address + PAGE_SIZE)))
367 return -ENOENT;
368 if (*flags & FOLL_WRITE)
369 fault_flags |= FAULT_FLAG_WRITE;
Dave Hansen1b2ee122016-02-12 13:02:21 -0800370 if (*flags & FOLL_REMOTE)
371 fault_flags |= FAULT_FLAG_REMOTE;
Kirill A. Shutemov16744482014-06-04 16:08:12 -0700372 if (nonblocking)
373 fault_flags |= FAULT_FLAG_ALLOW_RETRY;
374 if (*flags & FOLL_NOWAIT)
375 fault_flags |= FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT;
Andres Lagar-Cavilla234b2392014-09-17 10:51:48 -0700376 if (*flags & FOLL_TRIED) {
377 VM_WARN_ON_ONCE(fault_flags & FAULT_FLAG_ALLOW_RETRY);
378 fault_flags |= FAULT_FLAG_TRIED;
379 }
Kirill A. Shutemov16744482014-06-04 16:08:12 -0700380
Kirill A. Shutemovdcddffd2016-07-26 15:25:18 -0700381 ret = handle_mm_fault(vma, address, fault_flags);
Kirill A. Shutemov16744482014-06-04 16:08:12 -0700382 if (ret & VM_FAULT_ERROR) {
383 if (ret & VM_FAULT_OOM)
384 return -ENOMEM;
385 if (ret & (VM_FAULT_HWPOISON | VM_FAULT_HWPOISON_LARGE))
386 return *flags & FOLL_HWPOISON ? -EHWPOISON : -EFAULT;
Linus Torvalds33692f22015-01-29 10:51:32 -0800387 if (ret & (VM_FAULT_SIGBUS | VM_FAULT_SIGSEGV))
Kirill A. Shutemov16744482014-06-04 16:08:12 -0700388 return -EFAULT;
389 BUG();
390 }
391
392 if (tsk) {
393 if (ret & VM_FAULT_MAJOR)
394 tsk->maj_flt++;
395 else
396 tsk->min_flt++;
397 }
398
399 if (ret & VM_FAULT_RETRY) {
400 if (nonblocking)
401 *nonblocking = 0;
402 return -EBUSY;
403 }
404
405 /*
406 * The VM_FAULT_WRITE bit tells us that do_wp_page has broken COW when
407 * necessary, even if maybe_mkwrite decided not to set pte_write. We
408 * can thus safely do subsequent page lookups as if they were reads.
409 * But only do so when looping for pte_write is futile: in some cases
410 * userspace may also be wanting to write to the gotten user page,
411 * which a read fault here might prevent (a readonly page might get
412 * reCOWed by userspace write).
413 */
414 if ((ret & VM_FAULT_WRITE) && !(vma->vm_flags & VM_WRITE))
415 *flags &= ~FOLL_WRITE;
416 return 0;
417}
418
Kirill A. Shutemovfa5bb202014-06-04 16:08:13 -0700419static int check_vma_flags(struct vm_area_struct *vma, unsigned long gup_flags)
420{
421 vm_flags_t vm_flags = vma->vm_flags;
Dave Hansen1b2ee122016-02-12 13:02:21 -0800422 int write = (gup_flags & FOLL_WRITE);
423 int foreign = (gup_flags & FOLL_REMOTE);
Kirill A. Shutemovfa5bb202014-06-04 16:08:13 -0700424
425 if (vm_flags & (VM_IO | VM_PFNMAP))
426 return -EFAULT;
427
Dave Hansen1b2ee122016-02-12 13:02:21 -0800428 if (write) {
Kirill A. Shutemovfa5bb202014-06-04 16:08:13 -0700429 if (!(vm_flags & VM_WRITE)) {
430 if (!(gup_flags & FOLL_FORCE))
431 return -EFAULT;
432 /*
433 * We used to let the write,force case do COW in a
434 * VM_MAYWRITE VM_SHARED !VM_WRITE vma, so ptrace could
435 * set a breakpoint in a read-only mapping of an
436 * executable, without corrupting the file (yet only
437 * when that file had been opened for writing!).
438 * Anon pages in shared mappings are surprising: now
439 * just reject it.
440 */
Hugh Dickins46435362016-01-30 18:03:16 -0800441 if (!is_cow_mapping(vm_flags))
Kirill A. Shutemovfa5bb202014-06-04 16:08:13 -0700442 return -EFAULT;
Kirill A. Shutemovfa5bb202014-06-04 16:08:13 -0700443 }
444 } else if (!(vm_flags & VM_READ)) {
445 if (!(gup_flags & FOLL_FORCE))
446 return -EFAULT;
447 /*
448 * Is there actually any vma we can reach here which does not
449 * have VM_MAYREAD set?
450 */
451 if (!(vm_flags & VM_MAYREAD))
452 return -EFAULT;
453 }
Dave Hansend61172b2016-02-12 13:02:24 -0800454 /*
455 * gups are always data accesses, not instruction
456 * fetches, so execute=false here
457 */
458 if (!arch_vma_access_permitted(vma, write, false, foreign))
Dave Hansen33a709b2016-02-12 13:02:19 -0800459 return -EFAULT;
Kirill A. Shutemovfa5bb202014-06-04 16:08:13 -0700460 return 0;
461}
462
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700463/**
464 * __get_user_pages() - pin user pages in memory
465 * @tsk: task_struct of target task
466 * @mm: mm_struct of target mm
467 * @start: starting user address
468 * @nr_pages: number of pages from start to pin
469 * @gup_flags: flags modifying pin behaviour
470 * @pages: array that receives pointers to the pages pinned.
471 * Should be at least nr_pages long. Or NULL, if caller
472 * only intends to ensure the pages are faulted in.
473 * @vmas: array of pointers to vmas corresponding to each page.
474 * Or NULL if the caller does not require them.
475 * @nonblocking: whether waiting for disk IO or mmap_sem contention
476 *
477 * Returns number of pages pinned. This may be fewer than the number
478 * requested. If nr_pages is 0 or negative, returns 0. If no pages
479 * were pinned, returns -errno. Each page returned must be released
480 * with a put_page() call when it is finished with. vmas will only
481 * remain valid while mmap_sem is held.
482 *
Paul Cassella9a95f3c2014-08-06 16:07:24 -0700483 * Must be called with mmap_sem held. It may be released. See below.
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700484 *
485 * __get_user_pages walks a process's page tables and takes a reference to
486 * each struct page that each user address corresponds to at a given
487 * instant. That is, it takes the page that would be accessed if a user
488 * thread accesses the given user virtual address at that instant.
489 *
490 * This does not guarantee that the page exists in the user mappings when
491 * __get_user_pages returns, and there may even be a completely different
492 * page there in some cases (eg. if mmapped pagecache has been invalidated
493 * and subsequently re faulted). However it does guarantee that the page
494 * won't be freed completely. And mostly callers simply care that the page
495 * contains data that was valid *at some point in time*. Typically, an IO
496 * or similar operation cannot guarantee anything stronger anyway because
497 * locks can't be held over the syscall boundary.
498 *
499 * If @gup_flags & FOLL_WRITE == 0, the page must not be written to. If
500 * the page is written to, set_page_dirty (or set_page_dirty_lock, as
501 * appropriate) must be called after the page is finished with, and
502 * before put_page is called.
503 *
504 * If @nonblocking != NULL, __get_user_pages will not wait for disk IO
505 * or mmap_sem contention, and if waiting is needed to pin all pages,
Paul Cassella9a95f3c2014-08-06 16:07:24 -0700506 * *@nonblocking will be set to 0. Further, if @gup_flags does not
507 * include FOLL_NOWAIT, the mmap_sem will be released via up_read() in
508 * this case.
509 *
510 * A caller using such a combination of @nonblocking and @gup_flags
511 * must therefore hold the mmap_sem for reading only, and recognize
512 * when it's been released. Otherwise, it must be held for either
513 * reading or writing and will not be released.
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700514 *
515 * In most cases, get_user_pages or get_user_pages_fast should be used
516 * instead of __get_user_pages. __get_user_pages should be used only if
517 * you need some special @gup_flags.
518 */
519long __get_user_pages(struct task_struct *tsk, struct mm_struct *mm,
520 unsigned long start, unsigned long nr_pages,
521 unsigned int gup_flags, struct page **pages,
522 struct vm_area_struct **vmas, int *nonblocking)
523{
Kirill A. Shutemovfa5bb202014-06-04 16:08:13 -0700524 long i = 0;
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700525 unsigned int page_mask;
Kirill A. Shutemovfa5bb202014-06-04 16:08:13 -0700526 struct vm_area_struct *vma = NULL;
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700527
528 if (!nr_pages)
529 return 0;
530
531 VM_BUG_ON(!!pages != !!(gup_flags & FOLL_GET));
532
533 /*
534 * If FOLL_FORCE is set then do not force a full fault as the hinting
535 * fault information is unrelated to the reference behaviour of a task
536 * using the address space
537 */
538 if (!(gup_flags & FOLL_FORCE))
539 gup_flags |= FOLL_NUMA;
540
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700541 do {
Kirill A. Shutemovfa5bb202014-06-04 16:08:13 -0700542 struct page *page;
543 unsigned int foll_flags = gup_flags;
544 unsigned int page_increm;
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700545
Kirill A. Shutemovfa5bb202014-06-04 16:08:13 -0700546 /* first iteration or cross vma bound */
547 if (!vma || start >= vma->vm_end) {
548 vma = find_extend_vma(mm, start);
549 if (!vma && in_gate_area(mm, start)) {
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700550 int ret;
Kirill A. Shutemovfa5bb202014-06-04 16:08:13 -0700551 ret = get_gate_page(mm, start & PAGE_MASK,
552 gup_flags, &vma,
553 pages ? &pages[i] : NULL);
554 if (ret)
555 return i ? : ret;
556 page_mask = 0;
557 goto next_page;
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700558 }
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700559
Kirill A. Shutemovfa5bb202014-06-04 16:08:13 -0700560 if (!vma || check_vma_flags(vma, gup_flags))
561 return i ? : -EFAULT;
562 if (is_vm_hugetlb_page(vma)) {
563 i = follow_hugetlb_page(mm, vma, pages, vmas,
564 &start, &nr_pages, i,
565 gup_flags);
566 continue;
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700567 }
Kirill A. Shutemovfa5bb202014-06-04 16:08:13 -0700568 }
569retry:
570 /*
571 * If we have a pending SIGKILL, don't keep faulting pages and
572 * potentially allocating memory.
573 */
574 if (unlikely(fatal_signal_pending(current)))
575 return i ? i : -ERESTARTSYS;
576 cond_resched();
577 page = follow_page_mask(vma, start, foll_flags, &page_mask);
578 if (!page) {
579 int ret;
580 ret = faultin_page(tsk, vma, start, &foll_flags,
581 nonblocking);
582 switch (ret) {
583 case 0:
584 goto retry;
585 case -EFAULT:
586 case -ENOMEM:
587 case -EHWPOISON:
588 return i ? i : ret;
589 case -EBUSY:
590 return i;
591 case -ENOENT:
592 goto next_page;
593 }
594 BUG();
Kirill A. Shutemov1027e442015-09-04 15:47:55 -0700595 } else if (PTR_ERR(page) == -EEXIST) {
596 /*
597 * Proper page table entry exists, but no corresponding
598 * struct page.
599 */
600 goto next_page;
601 } else if (IS_ERR(page)) {
Kirill A. Shutemovfa5bb202014-06-04 16:08:13 -0700602 return i ? i : PTR_ERR(page);
Kirill A. Shutemov1027e442015-09-04 15:47:55 -0700603 }
Kirill A. Shutemovfa5bb202014-06-04 16:08:13 -0700604 if (pages) {
605 pages[i] = page;
606 flush_anon_page(vma, page, start);
607 flush_dcache_page(page);
608 page_mask = 0;
609 }
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700610next_page:
Kirill A. Shutemovfa5bb202014-06-04 16:08:13 -0700611 if (vmas) {
612 vmas[i] = vma;
613 page_mask = 0;
614 }
615 page_increm = 1 + (~(start >> PAGE_SHIFT) & page_mask);
616 if (page_increm > nr_pages)
617 page_increm = nr_pages;
618 i += page_increm;
619 start += page_increm * PAGE_SIZE;
620 nr_pages -= page_increm;
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700621 } while (nr_pages);
622 return i;
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700623}
624EXPORT_SYMBOL(__get_user_pages);
625
Dave Hansend4925e02016-02-12 13:02:16 -0800626bool vma_permits_fault(struct vm_area_struct *vma, unsigned int fault_flags)
627{
Dave Hansen1b2ee122016-02-12 13:02:21 -0800628 bool write = !!(fault_flags & FAULT_FLAG_WRITE);
629 bool foreign = !!(fault_flags & FAULT_FLAG_REMOTE);
Dave Hansen33a709b2016-02-12 13:02:19 -0800630 vm_flags_t vm_flags = write ? VM_WRITE : VM_READ;
Dave Hansend4925e02016-02-12 13:02:16 -0800631
632 if (!(vm_flags & vma->vm_flags))
633 return false;
634
Dave Hansen33a709b2016-02-12 13:02:19 -0800635 /*
636 * The architecture might have a hardware protection
Dave Hansen1b2ee122016-02-12 13:02:21 -0800637 * mechanism other than read/write that can deny access.
Dave Hansend61172b2016-02-12 13:02:24 -0800638 *
639 * gup always represents data access, not instruction
640 * fetches, so execute=false here:
Dave Hansen33a709b2016-02-12 13:02:19 -0800641 */
Dave Hansend61172b2016-02-12 13:02:24 -0800642 if (!arch_vma_access_permitted(vma, write, false, foreign))
Dave Hansen33a709b2016-02-12 13:02:19 -0800643 return false;
644
Dave Hansend4925e02016-02-12 13:02:16 -0800645 return true;
646}
647
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700648/*
649 * fixup_user_fault() - manually resolve a user page fault
650 * @tsk: the task_struct to use for page fault accounting, or
651 * NULL if faults are not to be recorded.
652 * @mm: mm_struct of target mm
653 * @address: user address
654 * @fault_flags:flags to pass down to handle_mm_fault()
Dominik Dingel4a9e1cd2016-01-15 16:57:04 -0800655 * @unlocked: did we unlock the mmap_sem while retrying, maybe NULL if caller
656 * does not allow retry
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700657 *
658 * This is meant to be called in the specific scenario where for locking reasons
659 * we try to access user memory in atomic context (within a pagefault_disable()
660 * section), this returns -EFAULT, and we want to resolve the user fault before
661 * trying again.
662 *
663 * Typically this is meant to be used by the futex code.
664 *
665 * The main difference with get_user_pages() is that this function will
666 * unconditionally call handle_mm_fault() which will in turn perform all the
667 * necessary SW fixup of the dirty and young bits in the PTE, while
Dominik Dingel4a9e1cd2016-01-15 16:57:04 -0800668 * get_user_pages() only guarantees to update these in the struct page.
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700669 *
670 * This is important for some architectures where those bits also gate the
671 * access permission to the page because they are maintained in software. On
672 * such architectures, gup() will not be enough to make a subsequent access
673 * succeed.
674 *
Dominik Dingel4a9e1cd2016-01-15 16:57:04 -0800675 * This function will not return with an unlocked mmap_sem. So it has not the
676 * same semantics wrt the @mm->mmap_sem as does filemap_fault().
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700677 */
678int fixup_user_fault(struct task_struct *tsk, struct mm_struct *mm,
Dominik Dingel4a9e1cd2016-01-15 16:57:04 -0800679 unsigned long address, unsigned int fault_flags,
680 bool *unlocked)
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700681{
682 struct vm_area_struct *vma;
Dominik Dingel4a9e1cd2016-01-15 16:57:04 -0800683 int ret, major = 0;
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700684
Dominik Dingel4a9e1cd2016-01-15 16:57:04 -0800685 if (unlocked)
686 fault_flags |= FAULT_FLAG_ALLOW_RETRY;
687
688retry:
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700689 vma = find_extend_vma(mm, address);
690 if (!vma || address < vma->vm_start)
691 return -EFAULT;
692
Dave Hansend4925e02016-02-12 13:02:16 -0800693 if (!vma_permits_fault(vma, fault_flags))
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700694 return -EFAULT;
695
Kirill A. Shutemovdcddffd2016-07-26 15:25:18 -0700696 ret = handle_mm_fault(vma, address, fault_flags);
Dominik Dingel4a9e1cd2016-01-15 16:57:04 -0800697 major |= ret & VM_FAULT_MAJOR;
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700698 if (ret & VM_FAULT_ERROR) {
699 if (ret & VM_FAULT_OOM)
700 return -ENOMEM;
701 if (ret & (VM_FAULT_HWPOISON | VM_FAULT_HWPOISON_LARGE))
702 return -EHWPOISON;
Linus Torvalds33692f22015-01-29 10:51:32 -0800703 if (ret & (VM_FAULT_SIGBUS | VM_FAULT_SIGSEGV))
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700704 return -EFAULT;
705 BUG();
706 }
Dominik Dingel4a9e1cd2016-01-15 16:57:04 -0800707
708 if (ret & VM_FAULT_RETRY) {
709 down_read(&mm->mmap_sem);
710 if (!(fault_flags & FAULT_FLAG_TRIED)) {
711 *unlocked = true;
712 fault_flags &= ~FAULT_FLAG_ALLOW_RETRY;
713 fault_flags |= FAULT_FLAG_TRIED;
714 goto retry;
715 }
716 }
717
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700718 if (tsk) {
Dominik Dingel4a9e1cd2016-01-15 16:57:04 -0800719 if (major)
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700720 tsk->maj_flt++;
721 else
722 tsk->min_flt++;
723 }
724 return 0;
725}
Paolo Bonziniadd6a0c2016-06-07 17:51:18 +0200726EXPORT_SYMBOL_GPL(fixup_user_fault);
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700727
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800728static __always_inline long __get_user_pages_locked(struct task_struct *tsk,
729 struct mm_struct *mm,
730 unsigned long start,
731 unsigned long nr_pages,
732 int write, int force,
733 struct page **pages,
734 struct vm_area_struct **vmas,
Andrea Arcangeli0fd71a52015-02-11 15:27:20 -0800735 int *locked, bool notify_drop,
736 unsigned int flags)
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800737{
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800738 long ret, pages_done;
739 bool lock_dropped;
740
741 if (locked) {
742 /* if VM_FAULT_RETRY can be returned, vmas become invalid */
743 BUG_ON(vmas);
744 /* check caller initialized locked */
745 BUG_ON(*locked != 1);
746 }
747
748 if (pages)
749 flags |= FOLL_GET;
750 if (write)
751 flags |= FOLL_WRITE;
752 if (force)
753 flags |= FOLL_FORCE;
754
755 pages_done = 0;
756 lock_dropped = false;
757 for (;;) {
758 ret = __get_user_pages(tsk, mm, start, nr_pages, flags, pages,
759 vmas, locked);
760 if (!locked)
761 /* VM_FAULT_RETRY couldn't trigger, bypass */
762 return ret;
763
764 /* VM_FAULT_RETRY cannot return errors */
765 if (!*locked) {
766 BUG_ON(ret < 0);
767 BUG_ON(ret >= nr_pages);
768 }
769
770 if (!pages)
771 /* If it's a prefault don't insist harder */
772 return ret;
773
774 if (ret > 0) {
775 nr_pages -= ret;
776 pages_done += ret;
777 if (!nr_pages)
778 break;
779 }
780 if (*locked) {
781 /* VM_FAULT_RETRY didn't trigger */
782 if (!pages_done)
783 pages_done = ret;
784 break;
785 }
786 /* VM_FAULT_RETRY triggered, so seek to the faulting offset */
787 pages += ret;
788 start += ret << PAGE_SHIFT;
789
790 /*
791 * Repeat on the address that fired VM_FAULT_RETRY
792 * without FAULT_FLAG_ALLOW_RETRY but with
793 * FAULT_FLAG_TRIED.
794 */
795 *locked = 1;
796 lock_dropped = true;
797 down_read(&mm->mmap_sem);
798 ret = __get_user_pages(tsk, mm, start, 1, flags | FOLL_TRIED,
799 pages, NULL, NULL);
800 if (ret != 1) {
801 BUG_ON(ret > 1);
802 if (!pages_done)
803 pages_done = ret;
804 break;
805 }
806 nr_pages--;
807 pages_done++;
808 if (!nr_pages)
809 break;
810 pages++;
811 start += PAGE_SIZE;
812 }
813 if (notify_drop && lock_dropped && *locked) {
814 /*
815 * We must let the caller know we temporarily dropped the lock
816 * and so the critical section protected by it was lost.
817 */
818 up_read(&mm->mmap_sem);
819 *locked = 0;
820 }
821 return pages_done;
822}
823
824/*
825 * We can leverage the VM_FAULT_RETRY functionality in the page fault
826 * paths better by using either get_user_pages_locked() or
827 * get_user_pages_unlocked().
828 *
829 * get_user_pages_locked() is suitable to replace the form:
830 *
831 * down_read(&mm->mmap_sem);
832 * do_something()
833 * get_user_pages(tsk, mm, ..., pages, NULL);
834 * up_read(&mm->mmap_sem);
835 *
836 * to:
837 *
838 * int locked = 1;
839 * down_read(&mm->mmap_sem);
840 * do_something()
841 * get_user_pages_locked(tsk, mm, ..., pages, &locked);
842 * if (locked)
843 * up_read(&mm->mmap_sem);
844 */
Ingo Molnarc12d2da2016-04-04 10:24:58 +0200845long get_user_pages_locked(unsigned long start, unsigned long nr_pages,
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800846 int write, int force, struct page **pages,
847 int *locked)
848{
Dave Hansencde70142016-02-12 13:01:55 -0800849 return __get_user_pages_locked(current, current->mm, start, nr_pages,
850 write, force, pages, NULL, locked, true,
851 FOLL_TOUCH);
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800852}
Ingo Molnarc12d2da2016-04-04 10:24:58 +0200853EXPORT_SYMBOL(get_user_pages_locked);
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800854
855/*
Andrea Arcangeli0fd71a52015-02-11 15:27:20 -0800856 * Same as get_user_pages_unlocked(...., FOLL_TOUCH) but it allows to
857 * pass additional gup_flags as last parameter (like FOLL_HWPOISON).
858 *
859 * NOTE: here FOLL_TOUCH is not set implicitly and must be set by the
860 * caller if required (just like with __get_user_pages). "FOLL_GET",
861 * "FOLL_WRITE" and "FOLL_FORCE" are set implicitly as needed
862 * according to the parameters "pages", "write", "force"
863 * respectively.
864 */
865__always_inline long __get_user_pages_unlocked(struct task_struct *tsk, struct mm_struct *mm,
866 unsigned long start, unsigned long nr_pages,
867 int write, int force, struct page **pages,
868 unsigned int gup_flags)
869{
870 long ret;
871 int locked = 1;
872 down_read(&mm->mmap_sem);
873 ret = __get_user_pages_locked(tsk, mm, start, nr_pages, write, force,
874 pages, NULL, &locked, false, gup_flags);
875 if (locked)
876 up_read(&mm->mmap_sem);
877 return ret;
878}
879EXPORT_SYMBOL(__get_user_pages_unlocked);
880
881/*
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800882 * get_user_pages_unlocked() is suitable to replace the form:
883 *
884 * down_read(&mm->mmap_sem);
885 * get_user_pages(tsk, mm, ..., pages, NULL);
886 * up_read(&mm->mmap_sem);
887 *
888 * with:
889 *
890 * get_user_pages_unlocked(tsk, mm, ..., pages);
891 *
892 * It is functionally equivalent to get_user_pages_fast so
893 * get_user_pages_fast should be used instead, if the two parameters
894 * "tsk" and "mm" are respectively equal to current and current->mm,
895 * or if "force" shall be set to 1 (get_user_pages_fast misses the
896 * "force" parameter).
897 */
Ingo Molnarc12d2da2016-04-04 10:24:58 +0200898long get_user_pages_unlocked(unsigned long start, unsigned long nr_pages,
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800899 int write, int force, struct page **pages)
900{
Dave Hansencde70142016-02-12 13:01:55 -0800901 return __get_user_pages_unlocked(current, current->mm, start, nr_pages,
902 write, force, pages, FOLL_TOUCH);
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800903}
Ingo Molnarc12d2da2016-04-04 10:24:58 +0200904EXPORT_SYMBOL(get_user_pages_unlocked);
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800905
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700906/*
Dave Hansen1e987792016-02-12 13:01:54 -0800907 * get_user_pages_remote() - pin user pages in memory
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700908 * @tsk: the task_struct to use for page fault accounting, or
909 * NULL if faults are not to be recorded.
910 * @mm: mm_struct of target mm
911 * @start: starting user address
912 * @nr_pages: number of pages from start to pin
913 * @write: whether pages will be written to by the caller
914 * @force: whether to force access even when user mapping is currently
915 * protected (but never forces write access to shared mapping).
916 * @pages: array that receives pointers to the pages pinned.
917 * Should be at least nr_pages long. Or NULL, if caller
918 * only intends to ensure the pages are faulted in.
919 * @vmas: array of pointers to vmas corresponding to each page.
920 * Or NULL if the caller does not require them.
921 *
922 * Returns number of pages pinned. This may be fewer than the number
923 * requested. If nr_pages is 0 or negative, returns 0. If no pages
924 * were pinned, returns -errno. Each page returned must be released
925 * with a put_page() call when it is finished with. vmas will only
926 * remain valid while mmap_sem is held.
927 *
928 * Must be called with mmap_sem held for read or write.
929 *
930 * get_user_pages walks a process's page tables and takes a reference to
931 * each struct page that each user address corresponds to at a given
932 * instant. That is, it takes the page that would be accessed if a user
933 * thread accesses the given user virtual address at that instant.
934 *
935 * This does not guarantee that the page exists in the user mappings when
936 * get_user_pages returns, and there may even be a completely different
937 * page there in some cases (eg. if mmapped pagecache has been invalidated
938 * and subsequently re faulted). However it does guarantee that the page
939 * won't be freed completely. And mostly callers simply care that the page
940 * contains data that was valid *at some point in time*. Typically, an IO
941 * or similar operation cannot guarantee anything stronger anyway because
942 * locks can't be held over the syscall boundary.
943 *
944 * If write=0, the page must not be written to. If the page is written to,
945 * set_page_dirty (or set_page_dirty_lock, as appropriate) must be called
946 * after the page is finished with, and before put_page is called.
947 *
948 * get_user_pages is typically used for fewer-copy IO operations, to get a
949 * handle on the memory by some means other than accesses via the user virtual
950 * addresses. The pages may be submitted for DMA to devices or accessed via
951 * their kernel linear mapping (via the kmap APIs). Care should be taken to
952 * use the correct cache flushing APIs.
953 *
954 * See also get_user_pages_fast, for performance critical applications.
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800955 *
956 * get_user_pages should be phased out in favor of
957 * get_user_pages_locked|unlocked or get_user_pages_fast. Nothing
958 * should use get_user_pages because it cannot pass
959 * FAULT_FLAG_ALLOW_RETRY to handle_mm_fault.
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700960 */
Dave Hansen1e987792016-02-12 13:01:54 -0800961long get_user_pages_remote(struct task_struct *tsk, struct mm_struct *mm,
962 unsigned long start, unsigned long nr_pages,
963 int write, int force, struct page **pages,
964 struct vm_area_struct **vmas)
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700965{
Andrea Arcangelif0818f42015-02-11 15:27:17 -0800966 return __get_user_pages_locked(tsk, mm, start, nr_pages, write, force,
Dave Hansen1e987792016-02-12 13:01:54 -0800967 pages, vmas, NULL, false,
968 FOLL_TOUCH | FOLL_REMOTE);
969}
970EXPORT_SYMBOL(get_user_pages_remote);
971
972/*
Dave Hansend4edcf02016-02-12 13:01:56 -0800973 * This is the same as get_user_pages_remote(), just with a
974 * less-flexible calling convention where we assume that the task
975 * and mm being operated on are the current task's. We also
976 * obviously don't pass FOLL_REMOTE in here.
Dave Hansen1e987792016-02-12 13:01:54 -0800977 */
Ingo Molnarc12d2da2016-04-04 10:24:58 +0200978long get_user_pages(unsigned long start, unsigned long nr_pages,
Dave Hansen1e987792016-02-12 13:01:54 -0800979 int write, int force, struct page **pages,
980 struct vm_area_struct **vmas)
981{
Dave Hansencde70142016-02-12 13:01:55 -0800982 return __get_user_pages_locked(current, current->mm, start, nr_pages,
Dave Hansen1e987792016-02-12 13:01:54 -0800983 write, force, pages, vmas, NULL, false,
984 FOLL_TOUCH);
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700985}
Ingo Molnarc12d2da2016-04-04 10:24:58 +0200986EXPORT_SYMBOL(get_user_pages);
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -0700987
988/**
Kirill A. Shutemovacc3c8d2015-04-14 15:44:45 -0700989 * populate_vma_page_range() - populate a range of pages in the vma.
990 * @vma: target vma
991 * @start: start address
992 * @end: end address
993 * @nonblocking:
994 *
995 * This takes care of mlocking the pages too if VM_LOCKED is set.
996 *
997 * return 0 on success, negative error code on error.
998 *
999 * vma->vm_mm->mmap_sem must be held.
1000 *
1001 * If @nonblocking is NULL, it may be held for read or write and will
1002 * be unperturbed.
1003 *
1004 * If @nonblocking is non-NULL, it must held for read only and may be
1005 * released. If it's released, *@nonblocking will be set to 0.
1006 */
1007long populate_vma_page_range(struct vm_area_struct *vma,
1008 unsigned long start, unsigned long end, int *nonblocking)
1009{
1010 struct mm_struct *mm = vma->vm_mm;
1011 unsigned long nr_pages = (end - start) / PAGE_SIZE;
1012 int gup_flags;
1013
1014 VM_BUG_ON(start & ~PAGE_MASK);
1015 VM_BUG_ON(end & ~PAGE_MASK);
1016 VM_BUG_ON_VMA(start < vma->vm_start, vma);
1017 VM_BUG_ON_VMA(end > vma->vm_end, vma);
1018 VM_BUG_ON_MM(!rwsem_is_locked(&mm->mmap_sem), mm);
1019
Eric B Munsonde60f5f2015-11-05 18:51:36 -08001020 gup_flags = FOLL_TOUCH | FOLL_POPULATE | FOLL_MLOCK;
1021 if (vma->vm_flags & VM_LOCKONFAULT)
1022 gup_flags &= ~FOLL_POPULATE;
Kirill A. Shutemovacc3c8d2015-04-14 15:44:45 -07001023 /*
1024 * We want to touch writable mappings with a write fault in order
1025 * to break COW, except for shared mappings because these don't COW
1026 * and we would not want to dirty them for nothing.
1027 */
1028 if ((vma->vm_flags & (VM_WRITE | VM_SHARED)) == VM_WRITE)
1029 gup_flags |= FOLL_WRITE;
1030
1031 /*
1032 * We want mlock to succeed for regions that have any permissions
1033 * other than PROT_NONE.
1034 */
1035 if (vma->vm_flags & (VM_READ | VM_WRITE | VM_EXEC))
1036 gup_flags |= FOLL_FORCE;
1037
1038 /*
1039 * We made sure addr is within a VMA, so the following will
1040 * not result in a stack expansion that recurses back here.
1041 */
1042 return __get_user_pages(current, mm, start, nr_pages, gup_flags,
1043 NULL, NULL, nonblocking);
1044}
1045
1046/*
1047 * __mm_populate - populate and/or mlock pages within a range of address space.
1048 *
1049 * This is used to implement mlock() and the MAP_POPULATE / MAP_LOCKED mmap
1050 * flags. VMAs must be already marked with the desired vm_flags, and
1051 * mmap_sem must not be held.
1052 */
1053int __mm_populate(unsigned long start, unsigned long len, int ignore_errors)
1054{
1055 struct mm_struct *mm = current->mm;
1056 unsigned long end, nstart, nend;
1057 struct vm_area_struct *vma = NULL;
1058 int locked = 0;
1059 long ret = 0;
1060
1061 VM_BUG_ON(start & ~PAGE_MASK);
1062 VM_BUG_ON(len != PAGE_ALIGN(len));
1063 end = start + len;
1064
1065 for (nstart = start; nstart < end; nstart = nend) {
1066 /*
1067 * We want to fault in pages for [nstart; end) address range.
1068 * Find first corresponding VMA.
1069 */
1070 if (!locked) {
1071 locked = 1;
1072 down_read(&mm->mmap_sem);
1073 vma = find_vma(mm, nstart);
1074 } else if (nstart >= vma->vm_end)
1075 vma = vma->vm_next;
1076 if (!vma || vma->vm_start >= end)
1077 break;
1078 /*
1079 * Set [nstart; nend) to intersection of desired address
1080 * range with the first VMA. Also, skip undesirable VMA types.
1081 */
1082 nend = min(end, vma->vm_end);
1083 if (vma->vm_flags & (VM_IO | VM_PFNMAP))
1084 continue;
1085 if (nstart < vma->vm_start)
1086 nstart = vma->vm_start;
1087 /*
1088 * Now fault in a range of pages. populate_vma_page_range()
1089 * double checks the vma flags, so that it won't mlock pages
1090 * if the vma was already munlocked.
1091 */
1092 ret = populate_vma_page_range(vma, nstart, nend, &locked);
1093 if (ret < 0) {
1094 if (ignore_errors) {
1095 ret = 0;
1096 continue; /* continue at next VMA */
1097 }
1098 break;
1099 }
1100 nend = nstart + ret * PAGE_SIZE;
1101 ret = 0;
1102 }
1103 if (locked)
1104 up_read(&mm->mmap_sem);
1105 return ret; /* 0 or negative error code */
1106}
1107
1108/**
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -07001109 * get_dump_page() - pin user page in memory while writing it to core dump
1110 * @addr: user address
1111 *
1112 * Returns struct page pointer of user page pinned for dump,
Kirill A. Shutemovea1754a2016-04-01 15:29:48 +03001113 * to be freed afterwards by put_page().
Kirill A. Shutemov4bbd4c72014-06-04 16:08:10 -07001114 *
1115 * Returns NULL on any kind of failure - a hole must then be inserted into
1116 * the corefile, to preserve alignment with its headers; and also returns
1117 * NULL wherever the ZERO_PAGE, or an anonymous pte_none, has been found -
1118 * allowing a hole to be left in the corefile to save diskspace.
1119 *
1120 * Called without mmap_sem, but after all other threads have been killed.
1121 */
1122#ifdef CONFIG_ELF_CORE
1123struct page *get_dump_page(unsigned long addr)
1124{
1125 struct vm_area_struct *vma;
1126 struct page *page;
1127
1128 if (__get_user_pages(current, current->mm, addr, 1,
1129 FOLL_FORCE | FOLL_DUMP | FOLL_GET, &page, &vma,
1130 NULL) < 1)
1131 return NULL;
1132 flush_cache_page(vma, addr, page_to_pfn(page));
1133 return page;
1134}
1135#endif /* CONFIG_ELF_CORE */
Steve Capper2667f502014-10-09 15:29:14 -07001136
1137/*
1138 * Generic RCU Fast GUP
1139 *
1140 * get_user_pages_fast attempts to pin user pages by walking the page
1141 * tables directly and avoids taking locks. Thus the walker needs to be
1142 * protected from page table pages being freed from under it, and should
1143 * block any THP splits.
1144 *
1145 * One way to achieve this is to have the walker disable interrupts, and
1146 * rely on IPIs from the TLB flushing code blocking before the page table
1147 * pages are freed. This is unsuitable for architectures that do not need
1148 * to broadcast an IPI when invalidating TLBs.
1149 *
1150 * Another way to achieve this is to batch up page table containing pages
1151 * belonging to more than one mm_user, then rcu_sched a callback to free those
1152 * pages. Disabling interrupts will allow the fast_gup walker to both block
1153 * the rcu_sched callback, and an IPI that we broadcast for splitting THPs
1154 * (which is a relatively rare event). The code below adopts this strategy.
1155 *
1156 * Before activating this code, please be aware that the following assumptions
1157 * are currently made:
1158 *
1159 * *) HAVE_RCU_TABLE_FREE is enabled, and tlb_remove_table is used to free
1160 * pages containing page tables.
1161 *
Steve Capper2667f502014-10-09 15:29:14 -07001162 * *) ptes can be read atomically by the architecture.
1163 *
1164 * *) access_ok is sufficient to validate userspace address ranges.
1165 *
1166 * The last two assumptions can be relaxed by the addition of helper functions.
1167 *
1168 * This code is based heavily on the PowerPC implementation by Nick Piggin.
1169 */
1170#ifdef CONFIG_HAVE_GENERIC_RCU_GUP
1171
1172#ifdef __HAVE_ARCH_PTE_SPECIAL
1173static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end,
1174 int write, struct page **pages, int *nr)
1175{
1176 pte_t *ptep, *ptem;
1177 int ret = 0;
1178
1179 ptem = ptep = pte_offset_map(&pmd, addr);
1180 do {
1181 /*
1182 * In the line below we are assuming that the pte can be read
1183 * atomically. If this is not the case for your architecture,
1184 * please wrap this in a helper function!
1185 *
1186 * for an example see gup_get_pte in arch/x86/mm/gup.c
1187 */
Jason Low9d8c47e2015-04-15 16:14:05 -07001188 pte_t pte = READ_ONCE(*ptep);
Kirill A. Shutemov7aef4172016-01-15 16:52:32 -08001189 struct page *head, *page;
Steve Capper2667f502014-10-09 15:29:14 -07001190
1191 /*
1192 * Similar to the PMD case below, NUMA hinting must take slow
Mel Gorman8a0516e2015-02-12 14:58:22 -08001193 * path using the pte_protnone check.
Steve Capper2667f502014-10-09 15:29:14 -07001194 */
1195 if (!pte_present(pte) || pte_special(pte) ||
Mel Gorman8a0516e2015-02-12 14:58:22 -08001196 pte_protnone(pte) || (write && !pte_write(pte)))
Steve Capper2667f502014-10-09 15:29:14 -07001197 goto pte_unmap;
1198
Dave Hansen33a709b2016-02-12 13:02:19 -08001199 if (!arch_pte_access_permitted(pte, write))
1200 goto pte_unmap;
1201
Steve Capper2667f502014-10-09 15:29:14 -07001202 VM_BUG_ON(!pfn_valid(pte_pfn(pte)));
1203 page = pte_page(pte);
Kirill A. Shutemov7aef4172016-01-15 16:52:32 -08001204 head = compound_head(page);
Steve Capper2667f502014-10-09 15:29:14 -07001205
Kirill A. Shutemov7aef4172016-01-15 16:52:32 -08001206 if (!page_cache_get_speculative(head))
Steve Capper2667f502014-10-09 15:29:14 -07001207 goto pte_unmap;
1208
1209 if (unlikely(pte_val(pte) != pte_val(*ptep))) {
Kirill A. Shutemov7aef4172016-01-15 16:52:32 -08001210 put_page(head);
Steve Capper2667f502014-10-09 15:29:14 -07001211 goto pte_unmap;
1212 }
1213
Kirill A. Shutemov7aef4172016-01-15 16:52:32 -08001214 VM_BUG_ON_PAGE(compound_head(page) != head, page);
Steve Capper2667f502014-10-09 15:29:14 -07001215 pages[*nr] = page;
1216 (*nr)++;
1217
1218 } while (ptep++, addr += PAGE_SIZE, addr != end);
1219
1220 ret = 1;
1221
1222pte_unmap:
1223 pte_unmap(ptem);
1224 return ret;
1225}
1226#else
1227
1228/*
1229 * If we can't determine whether or not a pte is special, then fail immediately
1230 * for ptes. Note, we can still pin HugeTLB and THP as these are guaranteed not
1231 * to be special.
1232 *
1233 * For a futex to be placed on a THP tail page, get_futex_key requires a
1234 * __get_user_pages_fast implementation that can pin pages. Thus it's still
1235 * useful to have gup_huge_pmd even if we can't operate on ptes.
1236 */
1237static int gup_pte_range(pmd_t pmd, unsigned long addr, unsigned long end,
1238 int write, struct page **pages, int *nr)
1239{
1240 return 0;
1241}
1242#endif /* __HAVE_ARCH_PTE_SPECIAL */
1243
1244static int gup_huge_pmd(pmd_t orig, pmd_t *pmdp, unsigned long addr,
1245 unsigned long end, int write, struct page **pages, int *nr)
1246{
Kirill A. Shutemovddc58f22016-01-15 16:52:56 -08001247 struct page *head, *page;
Steve Capper2667f502014-10-09 15:29:14 -07001248 int refs;
1249
1250 if (write && !pmd_write(orig))
1251 return 0;
1252
1253 refs = 0;
1254 head = pmd_page(orig);
1255 page = head + ((addr & ~PMD_MASK) >> PAGE_SHIFT);
Steve Capper2667f502014-10-09 15:29:14 -07001256 do {
1257 VM_BUG_ON_PAGE(compound_head(page) != head, page);
1258 pages[*nr] = page;
1259 (*nr)++;
1260 page++;
1261 refs++;
1262 } while (addr += PAGE_SIZE, addr != end);
1263
1264 if (!page_cache_add_speculative(head, refs)) {
1265 *nr -= refs;
1266 return 0;
1267 }
1268
1269 if (unlikely(pmd_val(orig) != pmd_val(*pmdp))) {
1270 *nr -= refs;
1271 while (refs--)
1272 put_page(head);
1273 return 0;
1274 }
1275
Steve Capper2667f502014-10-09 15:29:14 -07001276 return 1;
1277}
1278
1279static int gup_huge_pud(pud_t orig, pud_t *pudp, unsigned long addr,
1280 unsigned long end, int write, struct page **pages, int *nr)
1281{
Kirill A. Shutemovddc58f22016-01-15 16:52:56 -08001282 struct page *head, *page;
Steve Capper2667f502014-10-09 15:29:14 -07001283 int refs;
1284
1285 if (write && !pud_write(orig))
1286 return 0;
1287
1288 refs = 0;
1289 head = pud_page(orig);
1290 page = head + ((addr & ~PUD_MASK) >> PAGE_SHIFT);
Steve Capper2667f502014-10-09 15:29:14 -07001291 do {
1292 VM_BUG_ON_PAGE(compound_head(page) != head, page);
1293 pages[*nr] = page;
1294 (*nr)++;
1295 page++;
1296 refs++;
1297 } while (addr += PAGE_SIZE, addr != end);
1298
1299 if (!page_cache_add_speculative(head, refs)) {
1300 *nr -= refs;
1301 return 0;
1302 }
1303
1304 if (unlikely(pud_val(orig) != pud_val(*pudp))) {
1305 *nr -= refs;
1306 while (refs--)
1307 put_page(head);
1308 return 0;
1309 }
1310
Steve Capper2667f502014-10-09 15:29:14 -07001311 return 1;
1312}
1313
Aneesh Kumar K.Vf30c59e2014-11-05 21:57:40 +05301314static int gup_huge_pgd(pgd_t orig, pgd_t *pgdp, unsigned long addr,
1315 unsigned long end, int write,
1316 struct page **pages, int *nr)
1317{
1318 int refs;
Kirill A. Shutemovddc58f22016-01-15 16:52:56 -08001319 struct page *head, *page;
Aneesh Kumar K.Vf30c59e2014-11-05 21:57:40 +05301320
1321 if (write && !pgd_write(orig))
1322 return 0;
1323
1324 refs = 0;
1325 head = pgd_page(orig);
1326 page = head + ((addr & ~PGDIR_MASK) >> PAGE_SHIFT);
Aneesh Kumar K.Vf30c59e2014-11-05 21:57:40 +05301327 do {
1328 VM_BUG_ON_PAGE(compound_head(page) != head, page);
1329 pages[*nr] = page;
1330 (*nr)++;
1331 page++;
1332 refs++;
1333 } while (addr += PAGE_SIZE, addr != end);
1334
1335 if (!page_cache_add_speculative(head, refs)) {
1336 *nr -= refs;
1337 return 0;
1338 }
1339
1340 if (unlikely(pgd_val(orig) != pgd_val(*pgdp))) {
1341 *nr -= refs;
1342 while (refs--)
1343 put_page(head);
1344 return 0;
1345 }
1346
Aneesh Kumar K.Vf30c59e2014-11-05 21:57:40 +05301347 return 1;
1348}
1349
Steve Capper2667f502014-10-09 15:29:14 -07001350static int gup_pmd_range(pud_t pud, unsigned long addr, unsigned long end,
1351 int write, struct page **pages, int *nr)
1352{
1353 unsigned long next;
1354 pmd_t *pmdp;
1355
1356 pmdp = pmd_offset(&pud, addr);
1357 do {
Christian Borntraeger38c5ce92015-01-06 22:54:46 +01001358 pmd_t pmd = READ_ONCE(*pmdp);
Steve Capper2667f502014-10-09 15:29:14 -07001359
1360 next = pmd_addr_end(addr, end);
Kirill A. Shutemov4b471e82016-01-15 16:53:39 -08001361 if (pmd_none(pmd))
Steve Capper2667f502014-10-09 15:29:14 -07001362 return 0;
1363
1364 if (unlikely(pmd_trans_huge(pmd) || pmd_huge(pmd))) {
1365 /*
1366 * NUMA hinting faults need to be handled in the GUP
1367 * slowpath for accounting purposes and so that they
1368 * can be serialised against THP migration.
1369 */
Mel Gorman8a0516e2015-02-12 14:58:22 -08001370 if (pmd_protnone(pmd))
Steve Capper2667f502014-10-09 15:29:14 -07001371 return 0;
1372
1373 if (!gup_huge_pmd(pmd, pmdp, addr, next, write,
1374 pages, nr))
1375 return 0;
1376
Aneesh Kumar K.Vf30c59e2014-11-05 21:57:40 +05301377 } else if (unlikely(is_hugepd(__hugepd(pmd_val(pmd))))) {
1378 /*
1379 * architecture have different format for hugetlbfs
1380 * pmd format and THP pmd format
1381 */
1382 if (!gup_huge_pd(__hugepd(pmd_val(pmd)), addr,
1383 PMD_SHIFT, next, write, pages, nr))
1384 return 0;
Steve Capper2667f502014-10-09 15:29:14 -07001385 } else if (!gup_pte_range(pmd, addr, next, write, pages, nr))
1386 return 0;
1387 } while (pmdp++, addr = next, addr != end);
1388
1389 return 1;
1390}
1391
Aneesh Kumar K.Vf30c59e2014-11-05 21:57:40 +05301392static int gup_pud_range(pgd_t pgd, unsigned long addr, unsigned long end,
1393 int write, struct page **pages, int *nr)
Steve Capper2667f502014-10-09 15:29:14 -07001394{
1395 unsigned long next;
1396 pud_t *pudp;
1397
Aneesh Kumar K.Vf30c59e2014-11-05 21:57:40 +05301398 pudp = pud_offset(&pgd, addr);
Steve Capper2667f502014-10-09 15:29:14 -07001399 do {
Christian Borntraegere37c6982014-12-07 21:41:33 +01001400 pud_t pud = READ_ONCE(*pudp);
Steve Capper2667f502014-10-09 15:29:14 -07001401
1402 next = pud_addr_end(addr, end);
1403 if (pud_none(pud))
1404 return 0;
Aneesh Kumar K.Vf30c59e2014-11-05 21:57:40 +05301405 if (unlikely(pud_huge(pud))) {
Steve Capper2667f502014-10-09 15:29:14 -07001406 if (!gup_huge_pud(pud, pudp, addr, next, write,
Aneesh Kumar K.Vf30c59e2014-11-05 21:57:40 +05301407 pages, nr))
1408 return 0;
1409 } else if (unlikely(is_hugepd(__hugepd(pud_val(pud))))) {
1410 if (!gup_huge_pd(__hugepd(pud_val(pud)), addr,
1411 PUD_SHIFT, next, write, pages, nr))
Steve Capper2667f502014-10-09 15:29:14 -07001412 return 0;
1413 } else if (!gup_pmd_range(pud, addr, next, write, pages, nr))
1414 return 0;
1415 } while (pudp++, addr = next, addr != end);
1416
1417 return 1;
1418}
1419
1420/*
1421 * Like get_user_pages_fast() except it's IRQ-safe in that it won't fall back to
1422 * the regular GUP. It will only return non-negative values.
1423 */
1424int __get_user_pages_fast(unsigned long start, int nr_pages, int write,
1425 struct page **pages)
1426{
1427 struct mm_struct *mm = current->mm;
1428 unsigned long addr, len, end;
1429 unsigned long next, flags;
1430 pgd_t *pgdp;
1431 int nr = 0;
1432
1433 start &= PAGE_MASK;
1434 addr = start;
1435 len = (unsigned long) nr_pages << PAGE_SHIFT;
1436 end = start + len;
1437
1438 if (unlikely(!access_ok(write ? VERIFY_WRITE : VERIFY_READ,
1439 start, len)))
1440 return 0;
1441
1442 /*
1443 * Disable interrupts. We use the nested form as we can already have
1444 * interrupts disabled by get_futex_key.
1445 *
1446 * With interrupts disabled, we block page table pages from being
1447 * freed from under us. See mmu_gather_tlb in asm-generic/tlb.h
1448 * for more details.
1449 *
1450 * We do not adopt an rcu_read_lock(.) here as we also want to
1451 * block IPIs that come from THPs splitting.
1452 */
1453
1454 local_irq_save(flags);
1455 pgdp = pgd_offset(mm, addr);
1456 do {
Jason Low9d8c47e2015-04-15 16:14:05 -07001457 pgd_t pgd = READ_ONCE(*pgdp);
Aneesh Kumar K.Vf30c59e2014-11-05 21:57:40 +05301458
Steve Capper2667f502014-10-09 15:29:14 -07001459 next = pgd_addr_end(addr, end);
Aneesh Kumar K.Vf30c59e2014-11-05 21:57:40 +05301460 if (pgd_none(pgd))
Steve Capper2667f502014-10-09 15:29:14 -07001461 break;
Aneesh Kumar K.Vf30c59e2014-11-05 21:57:40 +05301462 if (unlikely(pgd_huge(pgd))) {
1463 if (!gup_huge_pgd(pgd, pgdp, addr, next, write,
1464 pages, &nr))
1465 break;
1466 } else if (unlikely(is_hugepd(__hugepd(pgd_val(pgd))))) {
1467 if (!gup_huge_pd(__hugepd(pgd_val(pgd)), addr,
1468 PGDIR_SHIFT, next, write, pages, &nr))
1469 break;
1470 } else if (!gup_pud_range(pgd, addr, next, write, pages, &nr))
Steve Capper2667f502014-10-09 15:29:14 -07001471 break;
1472 } while (pgdp++, addr = next, addr != end);
1473 local_irq_restore(flags);
1474
1475 return nr;
1476}
1477
1478/**
1479 * get_user_pages_fast() - pin user pages in memory
1480 * @start: starting user address
1481 * @nr_pages: number of pages from start to pin
1482 * @write: whether pages will be written to
1483 * @pages: array that receives pointers to the pages pinned.
1484 * Should be at least nr_pages long.
1485 *
1486 * Attempt to pin user pages in memory without taking mm->mmap_sem.
1487 * If not successful, it will fall back to taking the lock and
1488 * calling get_user_pages().
1489 *
1490 * Returns number of pages pinned. This may be fewer than the number
1491 * requested. If nr_pages is 0 or negative, returns 0. If no pages
1492 * were pinned, returns -errno.
1493 */
1494int get_user_pages_fast(unsigned long start, int nr_pages, int write,
1495 struct page **pages)
1496{
Steve Capper2667f502014-10-09 15:29:14 -07001497 int nr, ret;
1498
1499 start &= PAGE_MASK;
1500 nr = __get_user_pages_fast(start, nr_pages, write, pages);
1501 ret = nr;
1502
1503 if (nr < nr_pages) {
1504 /* Try to get the remaining pages with get_user_pages */
1505 start += nr << PAGE_SHIFT;
1506 pages += nr;
1507
Ingo Molnarc12d2da2016-04-04 10:24:58 +02001508 ret = get_user_pages_unlocked(start, nr_pages - nr, write, 0, pages);
Steve Capper2667f502014-10-09 15:29:14 -07001509
1510 /* Have to be a bit careful with return values */
1511 if (nr > 0) {
1512 if (ret < 0)
1513 ret = nr;
1514 else
1515 ret += nr;
1516 }
1517 }
1518
1519 return ret;
1520}
1521
1522#endif /* CONFIG_HAVE_GENERIC_RCU_GUP */