1 // SPDX-License-Identifier: GPL-2.0
2 #include <linux/mm.h>
3 #include <linux/rmap.h>
4 #include <linux/hugetlb.h>
5 #include <linux/swap.h>
6 #include <linux/leafops.h>
7
8 #include "internal.h"
9
not_found(struct page_vma_mapped_walk * pvmw)10 static inline bool not_found(struct page_vma_mapped_walk *pvmw)
11 {
12 page_vma_mapped_walk_done(pvmw);
13 return false;
14 }
15
map_pte(struct page_vma_mapped_walk * pvmw,pmd_t * pmdvalp,spinlock_t ** ptlp)16 static bool map_pte(struct page_vma_mapped_walk *pvmw, pmd_t *pmdvalp,
17 spinlock_t **ptlp)
18 {
19 bool is_migration;
20 pte_t ptent;
21
22 if (pvmw->flags & PVMW_SYNC) {
23 /* Use the stricter lookup */
24 pvmw->pte = pte_offset_map_lock(pvmw->vma->vm_mm, pvmw->pmd,
25 pvmw->address, &pvmw->ptl);
26 *ptlp = pvmw->ptl;
27 return !!pvmw->pte;
28 }
29
30 is_migration = pvmw->flags & PVMW_MIGRATION;
31 again:
32 /*
33 * It is important to return the ptl corresponding to pte,
34 * in case *pvmw->pmd changes underneath us; so we need to
35 * return it even when choosing not to lock, in case caller
36 * proceeds to loop over next ptes, and finds a match later.
37 * Though, in most cases, page lock already protects this.
38 */
39 pvmw->pte = pte_offset_map_rw_nolock(pvmw->vma->vm_mm, pvmw->pmd,
40 pvmw->address, pmdvalp, ptlp);
41 if (!pvmw->pte)
42 return false;
43
44 ptent = ptep_get_lockless(pvmw->pte);
45
46 if (pte_none(ptent)) {
47 return false;
48 } else if (pte_present(ptent)) {
49 if (is_migration)
50 return false;
51 } else if (!is_migration) {
52 softleaf_t entry;
53
54 /*
55 * Handle un-addressable ZONE_DEVICE memory.
56 *
57 * We get here when we are trying to unmap a private
58 * device page from the process address space. Such
59 * page is not CPU accessible and thus is mapped as
60 * a special swap entry, nonetheless it still does
61 * count as a valid regular mapping for the page
62 * (and is accounted as such in page maps count).
63 *
64 * So handle this special case as if it was a normal
65 * page mapping ie lock CPU page table and return true.
66 *
67 * For more details on device private memory see HMM
68 * (include/linux/hmm.h or mm/hmm.c).
69 */
70 entry = softleaf_from_pte(ptent);
71 if (!softleaf_is_device_private(entry) &&
72 !softleaf_is_device_exclusive(entry))
73 return false;
74 }
75 spin_lock(*ptlp);
76 if (unlikely(!pmd_same(*pmdvalp, pmdp_get_lockless(pvmw->pmd)))) {
77 pte_unmap_unlock(pvmw->pte, *ptlp);
78 goto again;
79 }
80 pvmw->ptl = *ptlp;
81
82 return true;
83 }
84
85 /**
86 * check_pte - check if [pvmw->pfn, @pvmw->pfn + @pvmw->nr_pages) is
87 * mapped at the @pvmw->pte
88 * @pvmw: page_vma_mapped_walk struct, includes a pair pte and pfn range
89 * for checking
90 * @pte_nr: the number of small pages described by @pvmw->pte.
91 *
92 * page_vma_mapped_walk() found a place where pfn range is *potentially*
93 * mapped. check_pte() has to validate this.
94 *
95 * pvmw->pte may point to empty PTE, swap PTE or PTE pointing to
96 * arbitrary page.
97 *
98 * If PVMW_MIGRATION flag is set, returns true if @pvmw->pte contains migration
99 * entry that points to [pvmw->pfn, @pvmw->pfn + @pvmw->nr_pages)
100 *
101 * If PVMW_MIGRATION flag is not set, returns true if pvmw->pte points to
102 * [pvmw->pfn, @pvmw->pfn + @pvmw->nr_pages)
103 *
104 * Otherwise, return false.
105 *
106 */
check_pte(struct page_vma_mapped_walk * pvmw,unsigned long pte_nr)107 static bool check_pte(struct page_vma_mapped_walk *pvmw, unsigned long pte_nr)
108 {
109 unsigned long pfn;
110 pte_t ptent;
111
112 if (is_vm_hugetlb_page(pvmw->vma))
113 ptent = huge_ptep_get(pvmw->vma->vm_mm, pvmw->address,
114 pvmw->pte);
115 else
116 ptent = ptep_get(pvmw->pte);
117
118 if (pvmw->flags & PVMW_MIGRATION) {
119 const softleaf_t entry = softleaf_from_pte(ptent);
120
121 if (!softleaf_is_migration(entry))
122 return false;
123
124 pfn = softleaf_to_pfn(entry);
125 } else if (pte_present(ptent)) {
126 pfn = pte_pfn(ptent);
127 } else {
128 const softleaf_t entry = softleaf_from_pte(ptent);
129
130 /* Handle un-addressable ZONE_DEVICE memory */
131 if (!softleaf_is_device_private(entry) &&
132 !softleaf_is_device_exclusive(entry))
133 return false;
134
135 pfn = softleaf_to_pfn(entry);
136 }
137
138 if ((pfn + pte_nr - 1) < pvmw->pfn)
139 return false;
140 if (pfn > (pvmw->pfn + pvmw->nr_pages - 1))
141 return false;
142 return true;
143 }
144
145 /* Returns true if the two ranges overlap. Careful to not overflow. */
check_pmd(unsigned long pfn,struct page_vma_mapped_walk * pvmw)146 static bool check_pmd(unsigned long pfn, struct page_vma_mapped_walk *pvmw)
147 {
148 if ((pfn + HPAGE_PMD_NR - 1) < pvmw->pfn)
149 return false;
150 if (pfn > pvmw->pfn + pvmw->nr_pages - 1)
151 return false;
152 return true;
153 }
154
step_forward(struct page_vma_mapped_walk * pvmw,unsigned long size)155 static void step_forward(struct page_vma_mapped_walk *pvmw, unsigned long size)
156 {
157 pvmw->address = (pvmw->address + size) & ~(size - 1);
158 if (!pvmw->address)
159 pvmw->address = ULONG_MAX;
160 }
161
162 /**
163 * page_vma_mapped_walk - check if @pvmw->pfn is mapped in @pvmw->vma at
164 * @pvmw->address
165 * @pvmw: pointer to struct page_vma_mapped_walk. page, vma, address and flags
166 * must be set. pmd, pte and ptl must be NULL.
167 *
168 * Returns true if the page is mapped in the vma. @pvmw->pmd and @pvmw->pte point
169 * to relevant page table entries. @pvmw->ptl is locked. @pvmw->address is
170 * adjusted if needed (for PTE-mapped THPs).
171 *
172 * If @pvmw->pmd is set but @pvmw->pte is not, you have found PMD-mapped page
173 * (usually THP). For PTE-mapped THP, you should run page_vma_mapped_walk() in
174 * a loop to find all PTEs that map the THP.
175 *
176 * For HugeTLB pages, @pvmw->pte is set to the relevant page table entry
177 * regardless of which page table level the page is mapped at. @pvmw->pmd is
178 * NULL.
179 *
180 * Returns false if there are no more page table entries for the page in
181 * the vma. @pvmw->ptl is unlocked and @pvmw->pte is unmapped.
182 *
183 * If you need to stop the walk before page_vma_mapped_walk() returned false,
184 * use page_vma_mapped_walk_done(). It will do the housekeeping.
185 */
page_vma_mapped_walk(struct page_vma_mapped_walk * pvmw)186 bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw)
187 {
188 struct vm_area_struct *vma = pvmw->vma;
189 struct mm_struct *mm = vma->vm_mm;
190 unsigned long end;
191 spinlock_t *ptl;
192 pte_t pteval;
193 pgd_t *pgd;
194 p4d_t *p4d;
195 pud_t *pud;
196 pmd_t pmde;
197
198 /* The only possible pmd mapping has been handled on last iteration */
199 if (pvmw->pmd && !pvmw->pte)
200 return not_found(pvmw);
201
202 if (unlikely(is_vm_hugetlb_page(vma))) {
203 struct hstate *hstate = hstate_vma(vma);
204 unsigned long size = huge_page_size(hstate);
205 /* The only possible mapping was handled on last iteration */
206 if (pvmw->pte)
207 return not_found(pvmw);
208 /*
209 * All callers that get here will already hold the
210 * i_mmap_rwsem. Therefore, no additional locks need to be
211 * taken before calling hugetlb_walk().
212 */
213 pvmw->pte = hugetlb_walk(vma, pvmw->address, size);
214 if (!pvmw->pte)
215 return false;
216
217 pvmw->ptl = huge_pte_lock(hstate, mm, pvmw->pte);
218 if (!check_pte(pvmw, pages_per_huge_page(hstate)))
219 return not_found(pvmw);
220 return true;
221 }
222
223 end = vma_address_end(pvmw);
224 if (pvmw->pte)
225 goto next_pte;
226 restart:
227 do {
228 pgd = pgd_offset(mm, pvmw->address);
229 if (!pgd_present(*pgd)) {
230 step_forward(pvmw, PGDIR_SIZE);
231 continue;
232 }
233 p4d = p4d_offset(pgd, pvmw->address);
234 if (!p4d_present(*p4d)) {
235 step_forward(pvmw, P4D_SIZE);
236 continue;
237 }
238 pud = pud_offset(p4d, pvmw->address);
239 if (!pud_present(*pud)) {
240 step_forward(pvmw, PUD_SIZE);
241 continue;
242 }
243
244 pvmw->pmd = pmd_offset(pud, pvmw->address);
245 /*
246 * Make sure the pmd value isn't cached in a register by the
247 * compiler and used as a stale value after we've observed a
248 * subsequent update.
249 */
250 pmde = pmdp_get_lockless(pvmw->pmd);
251
252 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
253 (pmd_trans_huge(pmde) || pmd_is_migration_entry(pmde) ||
254 pmd_is_device_private_entry(pmde))) {
255 pvmw->ptl = pmd_lock(mm, pvmw->pmd);
256 pmde = *pvmw->pmd;
257 if (pmd_is_migration_entry(pmde)) {
258 softleaf_t entry;
259
260 if (!(pvmw->flags & PVMW_MIGRATION))
261 return not_found(pvmw);
262 entry = softleaf_from_pmd(pmde);
263 if (!check_pmd(softleaf_to_pfn(entry), pvmw))
264 return not_found(pvmw);
265 return true;
266 } else if (pmd_is_device_private_entry(pmde)) {
267 softleaf_t entry;
268
269 if (pvmw->flags & PVMW_MIGRATION)
270 return not_found(pvmw);
271 entry = softleaf_from_pmd(pmde);
272 if (!check_pmd(softleaf_to_pfn(entry), pvmw))
273 return not_found(pvmw);
274 return true;
275 } else if (!pmd_present(pmde)) {
276 return not_found(pvmw);
277 }
278 if (likely(pmd_trans_huge(pmde))) {
279 if (pvmw->flags & PVMW_MIGRATION)
280 return not_found(pvmw);
281 if (!check_pmd(pmd_pfn(pmde), pvmw))
282 return not_found(pvmw);
283 return true;
284 }
285 /* THP/device-private pmd was split under us: handle on pte level */
286 spin_unlock(pvmw->ptl);
287 pvmw->ptl = NULL;
288 } else if (!pmd_present(pmde)) {
289 if ((pvmw->flags & PVMW_SYNC) &&
290 thp_vma_suitable_order(vma, pvmw->address,
291 PMD_ORDER) &&
292 (pvmw->nr_pages >= HPAGE_PMD_NR))
293 sync_with_folio_pmd_zap(mm, pvmw->pmd);
294
295 step_forward(pvmw, PMD_SIZE);
296 continue;
297 }
298 if (!map_pte(pvmw, &pmde, &ptl)) {
299 if (!pvmw->pte)
300 goto restart;
301 goto next_pte;
302 }
303 this_pte:
304 if (check_pte(pvmw, 1))
305 return true;
306 next_pte:
307 do {
308 pvmw->address += PAGE_SIZE;
309 if (pvmw->address >= end)
310 return not_found(pvmw);
311 /* Did we cross page table boundary? */
312 if ((pvmw->address & (PMD_SIZE - PAGE_SIZE)) == 0) {
313 if (pvmw->ptl) {
314 spin_unlock(pvmw->ptl);
315 pvmw->ptl = NULL;
316 }
317 pte_unmap(pvmw->pte);
318 pvmw->pte = NULL;
319 pvmw->flags |= PVMW_PGTABLE_CROSSED;
320 goto restart;
321 }
322 pvmw->pte++;
323 if (!pvmw->ptl)
324 pteval = ptep_get_lockless(pvmw->pte);
325 else
326 pteval = ptep_get(pvmw->pte);
327 } while (pte_none(pteval));
328
329 if (!pvmw->ptl) {
330 spin_lock(ptl);
331 if (unlikely(!pmd_same(pmde, pmdp_get_lockless(pvmw->pmd)))) {
332 pte_unmap_unlock(pvmw->pte, ptl);
333 pvmw->pte = NULL;
334 goto restart;
335 }
336 pvmw->ptl = ptl;
337 }
338 goto this_pte;
339 } while (pvmw->address < end);
340
341 return false;
342 }
343
344 #ifdef CONFIG_MEMORY_FAILURE
345 /**
346 * page_mapped_in_vma - check whether a page is really mapped in a VMA
347 * @page: the page to test
348 * @vma: the VMA to test
349 *
350 * Return: The address the page is mapped at if the page is in the range
351 * covered by the VMA and present in the page table. If the page is
352 * outside the VMA or not present, returns -EFAULT.
353 * Only valid for normal file or anonymous VMAs.
354 */
page_mapped_in_vma(const struct page * page,struct vm_area_struct * vma)355 unsigned long page_mapped_in_vma(const struct page *page,
356 struct vm_area_struct *vma)
357 {
358 const struct folio *folio = page_folio(page);
359 const pgoff_t pgoff = page_pgoff(folio, page);
360 struct page_vma_mapped_walk pvmw = {
361 .pfn = page_to_pfn(page),
362 .nr_pages = 1,
363 .vma = vma,
364 .flags = PVMW_SYNC,
365 };
366
367 if (folio_test_anon(folio))
368 pvmw.address = vma_anon_address(vma, pgoff, 1);
369 else
370 pvmw.address = vma_filebacked_address(vma, pgoff, 1);
371 if (pvmw.address == -EFAULT)
372 goto out;
373 if (!page_vma_mapped_walk(&pvmw))
374 return -EFAULT;
375 page_vma_mapped_walk_done(&pvmw);
376 out:
377 return pvmw.address;
378 }
379 #endif
380