blob: 4e448cfbc6efa715ee9805caa980bc2fc15522bc [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
Kirill A. Shutemovace71a12017-02-24 14:57:45 -08002#include <linux/mm.h>
3#include <linux/rmap.h>
4#include <linux/hugetlb.h>
5#include <linux/swap.h>
6#include <linux/swapops.h>
7
8#include "internal.h"
9
Kirill A. Shutemovace71a12017-02-24 14:57:45 -080010static inline bool not_found(struct page_vma_mapped_walk *pvmw)
11{
12 page_vma_mapped_walk_done(pvmw);
13 return false;
14}
15
16static bool map_pte(struct page_vma_mapped_walk *pvmw)
17{
18 pvmw->pte = pte_offset_map(pvmw->pmd, pvmw->address);
19 if (!(pvmw->flags & PVMW_SYNC)) {
20 if (pvmw->flags & PVMW_MIGRATION) {
21 if (!is_swap_pte(*pvmw->pte))
22 return false;
23 } else {
Ralph Campbellaab8d052018-10-30 15:04:11 -070024 /*
25 * We get here when we are trying to unmap a private
26 * device page from the process address space. Such
27 * page is not CPU accessible and thus is mapped as
28 * a special swap entry, nonetheless it still does
29 * count as a valid regular mapping for the page (and
30 * is accounted as such in page maps count).
31 *
32 * So handle this special case as if it was a normal
33 * page mapping ie lock CPU page table and returns
34 * true.
35 *
36 * For more details on device private memory see HMM
37 * (include/linux/hmm.h or mm/hmm.c).
38 */
39 if (is_swap_pte(*pvmw->pte)) {
40 swp_entry_t entry;
41
42 /* Handle un-addressable ZONE_DEVICE memory */
43 entry = pte_to_swp_entry(*pvmw->pte);
Alistair Poppleb756a3b2021-06-30 18:54:25 -070044 if (!is_device_private_entry(entry) &&
45 !is_device_exclusive_entry(entry))
Ralph Campbellaab8d052018-10-30 15:04:11 -070046 return false;
47 } else if (!pte_present(*pvmw->pte))
Kirill A. Shutemovace71a12017-02-24 14:57:45 -080048 return false;
49 }
50 }
51 pvmw->ptl = pte_lockptr(pvmw->vma->vm_mm, pvmw->pmd);
52 spin_lock(pvmw->ptl);
53 return true;
54}
55
Kirill A. Shutemov0d665e72018-01-19 15:49:24 +030056/**
57 * check_pte - check if @pvmw->page is mapped at the @pvmw->pte
Alex Shi777f3032020-12-14 19:07:51 -080058 * @pvmw: page_vma_mapped_walk struct, includes a pair pte and page for checking
Kirill A. Shutemov0d665e72018-01-19 15:49:24 +030059 *
60 * page_vma_mapped_walk() found a place where @pvmw->page is *potentially*
61 * mapped. check_pte() has to validate this.
62 *
Alex Shi777f3032020-12-14 19:07:51 -080063 * pvmw->pte may point to empty PTE, swap PTE or PTE pointing to
64 * arbitrary page.
Kirill A. Shutemov0d665e72018-01-19 15:49:24 +030065 *
66 * If PVMW_MIGRATION flag is set, returns true if @pvmw->pte contains migration
67 * entry that points to @pvmw->page or any subpage in case of THP.
68 *
Alex Shi777f3032020-12-14 19:07:51 -080069 * If PVMW_MIGRATION flag is not set, returns true if pvmw->pte points to
70 * pvmw->page or any subpage in case of THP.
Kirill A. Shutemov0d665e72018-01-19 15:49:24 +030071 *
72 * Otherwise, return false.
73 *
74 */
Kirill A. Shutemovace71a12017-02-24 14:57:45 -080075static bool check_pte(struct page_vma_mapped_walk *pvmw)
76{
Kirill A. Shutemov0d665e72018-01-19 15:49:24 +030077 unsigned long pfn;
78
Kirill A. Shutemovace71a12017-02-24 14:57:45 -080079 if (pvmw->flags & PVMW_MIGRATION) {
Kirill A. Shutemovace71a12017-02-24 14:57:45 -080080 swp_entry_t entry;
81 if (!is_swap_pte(*pvmw->pte))
82 return false;
83 entry = pte_to_swp_entry(*pvmw->pte);
Jérôme Glissea5430dd2017-09-08 16:12:17 -070084
Alistair Poppleb756a3b2021-06-30 18:54:25 -070085 if (!is_migration_entry(entry) &&
86 !is_device_exclusive_entry(entry))
Kirill A. Shutemovace71a12017-02-24 14:57:45 -080087 return false;
Kirill A. Shutemov0d665e72018-01-19 15:49:24 +030088
Peter Xu0d206b52022-08-11 12:13:27 -040089 pfn = swp_offset_pfn(entry);
Kirill A. Shutemov0d665e72018-01-19 15:49:24 +030090 } else if (is_swap_pte(*pvmw->pte)) {
91 swp_entry_t entry;
92
93 /* Handle un-addressable ZONE_DEVICE memory */
94 entry = pte_to_swp_entry(*pvmw->pte);
Alistair Poppleb756a3b2021-06-30 18:54:25 -070095 if (!is_device_private_entry(entry) &&
96 !is_device_exclusive_entry(entry))
Kirill A. Shutemovace71a12017-02-24 14:57:45 -080097 return false;
Kirill A. Shutemov0d665e72018-01-19 15:49:24 +030098
Peter Xu0d206b52022-08-11 12:13:27 -040099 pfn = swp_offset_pfn(entry);
Kirill A. Shutemovace71a12017-02-24 14:57:45 -0800100 } else {
101 if (!pte_present(*pvmw->pte))
102 return false;
103
Kirill A. Shutemov0d665e72018-01-19 15:49:24 +0300104 pfn = pte_pfn(*pvmw->pte);
Kirill A. Shutemovace71a12017-02-24 14:57:45 -0800105 }
106
Matthew Wilcox (Oracle)2aff7a42022-02-03 11:40:17 -0500107 return (pfn - pvmw->pfn) < pvmw->nr_pages;
108}
109
110/* Returns true if the two ranges overlap. Careful to not overflow. */
111static bool check_pmd(unsigned long pfn, struct page_vma_mapped_walk *pvmw)
112{
113 if ((pfn + HPAGE_PMD_NR - 1) < pvmw->pfn)
114 return false;
115 if (pfn > pvmw->pfn + pvmw->nr_pages - 1)
116 return false;
117 return true;
Kirill A. Shutemovace71a12017-02-24 14:57:45 -0800118}
119
Hugh Dickinsa9a75042021-06-24 18:39:26 -0700120static void step_forward(struct page_vma_mapped_walk *pvmw, unsigned long size)
121{
122 pvmw->address = (pvmw->address + size) & ~(size - 1);
123 if (!pvmw->address)
124 pvmw->address = ULONG_MAX;
125}
126
Kirill A. Shutemovace71a12017-02-24 14:57:45 -0800127/**
Matthew Wilcox (Oracle)2aff7a42022-02-03 11:40:17 -0500128 * page_vma_mapped_walk - check if @pvmw->pfn is mapped in @pvmw->vma at
Kirill A. Shutemovace71a12017-02-24 14:57:45 -0800129 * @pvmw->address
130 * @pvmw: pointer to struct page_vma_mapped_walk. page, vma, address and flags
131 * must be set. pmd, pte and ptl must be NULL.
132 *
133 * Returns true if the page is mapped in the vma. @pvmw->pmd and @pvmw->pte point
134 * to relevant page table entries. @pvmw->ptl is locked. @pvmw->address is
135 * adjusted if needed (for PTE-mapped THPs).
136 *
137 * If @pvmw->pmd is set but @pvmw->pte is not, you have found PMD-mapped page
138 * (usually THP). For PTE-mapped THP, you should run page_vma_mapped_walk() in
139 * a loop to find all PTEs that map the THP.
140 *
141 * For HugeTLB pages, @pvmw->pte is set to the relevant page table entry
142 * regardless of which page table level the page is mapped at. @pvmw->pmd is
143 * NULL.
144 *
Lu Jialinbaf2f902021-05-06 18:06:50 -0700145 * Returns false if there are no more page table entries for the page in
Kirill A. Shutemovace71a12017-02-24 14:57:45 -0800146 * the vma. @pvmw->ptl is unlocked and @pvmw->pte is unmapped.
147 *
148 * If you need to stop the walk before page_vma_mapped_walk() returned false,
149 * use page_vma_mapped_walk_done(). It will do the housekeeping.
150 */
151bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw)
152{
Matthew Wilcox (Oracle)2aff7a42022-02-03 11:40:17 -0500153 struct vm_area_struct *vma = pvmw->vma;
154 struct mm_struct *mm = vma->vm_mm;
Hugh Dickins47446632021-06-24 18:39:20 -0700155 unsigned long end;
Kirill A. Shutemovace71a12017-02-24 14:57:45 -0800156 pgd_t *pgd;
Kirill A. Shutemovc2febaf2017-03-09 17:24:07 +0300157 p4d_t *p4d;
Kirill A. Shutemovace71a12017-02-24 14:57:45 -0800158 pud_t *pud;
Will Deacona7b10092017-10-13 15:58:25 -0700159 pmd_t pmde;
Kirill A. Shutemovace71a12017-02-24 14:57:45 -0800160
161 /* The only possible pmd mapping has been handled on last iteration */
162 if (pvmw->pmd && !pvmw->pte)
163 return not_found(pvmw);
164
Matthew Wilcox (Oracle)2aff7a42022-02-03 11:40:17 -0500165 if (unlikely(is_vm_hugetlb_page(vma))) {
zhenwei pi98ea0252022-04-07 14:40:08 +0800166 struct hstate *hstate = hstate_vma(vma);
167 unsigned long size = huge_page_size(hstate);
Hugh Dickins6d0fd592021-06-24 18:39:04 -0700168 /* The only possible mapping was handled on last iteration */
169 if (pvmw->pte)
170 return not_found(pvmw);
Peter Xu9c67a202022-12-16 10:52:29 -0500171 /*
172 * All callers that get here will already hold the
173 * i_mmap_rwsem. Therefore, no additional locks need to be
174 * taken before calling hugetlb_walk().
175 */
176 pvmw->pte = hugetlb_walk(vma, pvmw->address, size);
Kirill A. Shutemovace71a12017-02-24 14:57:45 -0800177 if (!pvmw->pte)
178 return false;
179
Miaohe Lin8f0b7472022-07-09 17:24:40 +0800180 pvmw->ptl = huge_pte_lock(hstate, mm, pvmw->pte);
Kirill A. Shutemovace71a12017-02-24 14:57:45 -0800181 if (!check_pte(pvmw))
182 return not_found(pvmw);
183 return true;
184 }
Hugh Dickins6d0fd592021-06-24 18:39:04 -0700185
Matthew Wilcox (Oracle)2aff7a42022-02-03 11:40:17 -0500186 end = vma_address_end(pvmw);
Hugh Dickins6d0fd592021-06-24 18:39:04 -0700187 if (pvmw->pte)
188 goto next_pte;
Kirill A. Shutemovace71a12017-02-24 14:57:45 -0800189restart:
Hugh Dickinsa9a75042021-06-24 18:39:26 -0700190 do {
Hugh Dickinsb3807a92021-06-24 18:39:17 -0700191 pgd = pgd_offset(mm, pvmw->address);
Hugh Dickinsa9a75042021-06-24 18:39:26 -0700192 if (!pgd_present(*pgd)) {
193 step_forward(pvmw, PGDIR_SIZE);
194 continue;
195 }
Hugh Dickinsb3807a92021-06-24 18:39:17 -0700196 p4d = p4d_offset(pgd, pvmw->address);
Hugh Dickinsa9a75042021-06-24 18:39:26 -0700197 if (!p4d_present(*p4d)) {
198 step_forward(pvmw, P4D_SIZE);
199 continue;
200 }
Hugh Dickinsb3807a92021-06-24 18:39:17 -0700201 pud = pud_offset(p4d, pvmw->address);
Hugh Dickinsa9a75042021-06-24 18:39:26 -0700202 if (!pud_present(*pud)) {
203 step_forward(pvmw, PUD_SIZE);
204 continue;
205 }
Hugh Dickinse2e1d402021-06-24 18:39:10 -0700206
Hugh Dickinsb3807a92021-06-24 18:39:17 -0700207 pvmw->pmd = pmd_offset(pud, pvmw->address);
Hugh Dickins732ed552021-06-15 18:23:53 -0700208 /*
Hugh Dickinsb3807a92021-06-24 18:39:17 -0700209 * Make sure the pmd value isn't cached in a register by the
210 * compiler and used as a stale value after we've observed a
211 * subsequent update.
Hugh Dickins732ed552021-06-15 18:23:53 -0700212 */
Hugh Dickinsb3807a92021-06-24 18:39:17 -0700213 pmde = READ_ONCE(*pvmw->pmd);
Hugh Dickins732ed552021-06-15 18:23:53 -0700214
Muchun Song6472f6d2022-04-28 23:16:10 -0700215 if (pmd_trans_huge(pmde) || is_pmd_migration_entry(pmde) ||
216 (pmd_present(pmde) && pmd_devmap(pmde))) {
Hugh Dickinsb3807a92021-06-24 18:39:17 -0700217 pvmw->ptl = pmd_lock(mm, pvmw->pmd);
218 pmde = *pvmw->pmd;
Hugh Dickinsb3807a92021-06-24 18:39:17 -0700219 if (!pmd_present(pmde)) {
220 swp_entry_t entry;
221
222 if (!thp_migration_supported() ||
223 !(pvmw->flags & PVMW_MIGRATION))
224 return not_found(pvmw);
225 entry = pmd_to_swp_entry(pmde);
226 if (!is_migration_entry(entry) ||
Peter Xu0d206b52022-08-11 12:13:27 -0400227 !check_pmd(swp_offset_pfn(entry), pvmw))
Hugh Dickinsb3807a92021-06-24 18:39:17 -0700228 return not_found(pvmw);
229 return true;
230 }
Muchun Song6472f6d2022-04-28 23:16:10 -0700231 if (likely(pmd_trans_huge(pmde) || pmd_devmap(pmde))) {
232 if (pvmw->flags & PVMW_MIGRATION)
233 return not_found(pvmw);
234 if (!check_pmd(pmd_pfn(pmde), pvmw))
235 return not_found(pvmw);
236 return true;
237 }
Hugh Dickinsb3807a92021-06-24 18:39:17 -0700238 /* THP pmd was split under us: handle on pte level */
239 spin_unlock(pvmw->ptl);
240 pvmw->ptl = NULL;
241 } else if (!pmd_present(pmde)) {
242 /*
243 * If PVMW_SYNC, take and drop THP pmd lock so that we
244 * cannot return prematurely, while zap_huge_pmd() has
245 * cleared *pmd but not decremented compound_mapcount().
246 */
247 if ((pvmw->flags & PVMW_SYNC) &&
Yang Shic453d8c2022-05-13 12:17:05 -0700248 transhuge_vma_suitable(vma, pvmw->address) &&
Matthew Wilcox (Oracle)2aff7a42022-02-03 11:40:17 -0500249 (pvmw->nr_pages >= HPAGE_PMD_NR)) {
Hugh Dickinsb3807a92021-06-24 18:39:17 -0700250 spinlock_t *ptl = pmd_lock(mm, pvmw->pmd);
251
252 spin_unlock(ptl);
253 }
Hugh Dickinsa9a75042021-06-24 18:39:26 -0700254 step_forward(pvmw, PMD_SIZE);
255 continue;
Hugh Dickins732ed552021-06-15 18:23:53 -0700256 }
Hugh Dickinsb3807a92021-06-24 18:39:17 -0700257 if (!map_pte(pvmw))
258 goto next_pte;
Hugh Dickins47446632021-06-24 18:39:20 -0700259this_pte:
Kirill A. Shutemovace71a12017-02-24 14:57:45 -0800260 if (check_pte(pvmw))
261 return true;
Hugh Dickinsd75450f2017-04-07 16:04:39 -0700262next_pte:
Hugh Dickinsd75450f2017-04-07 16:04:39 -0700263 do {
Kirill A. Shutemovace71a12017-02-24 14:57:45 -0800264 pvmw->address += PAGE_SIZE;
Hugh Dickins494334e2021-06-15 18:23:56 -0700265 if (pvmw->address >= end)
Kirill A. Shutemovace71a12017-02-24 14:57:45 -0800266 return not_found(pvmw);
267 /* Did we cross page table boundary? */
Hugh Dickins44828242021-06-24 18:39:14 -0700268 if ((pvmw->address & (PMD_SIZE - PAGE_SIZE)) == 0) {
Kirill A. Shutemovace71a12017-02-24 14:57:45 -0800269 if (pvmw->ptl) {
270 spin_unlock(pvmw->ptl);
271 pvmw->ptl = NULL;
272 }
Hugh Dickins44828242021-06-24 18:39:14 -0700273 pte_unmap(pvmw->pte);
274 pvmw->pte = NULL;
Kirill A. Shutemovace71a12017-02-24 14:57:45 -0800275 goto restart;
Kirill A. Shutemovace71a12017-02-24 14:57:45 -0800276 }
Hugh Dickins44828242021-06-24 18:39:14 -0700277 pvmw->pte++;
Hugh Dickinsa7a69d82021-06-24 18:39:30 -0700278 if ((pvmw->flags & PVMW_SYNC) && !pvmw->ptl) {
279 pvmw->ptl = pte_lockptr(mm, pvmw->pmd);
280 spin_lock(pvmw->ptl);
281 }
Kirill A. Shutemovace71a12017-02-24 14:57:45 -0800282 } while (pte_none(*pvmw->pte));
283
284 if (!pvmw->ptl) {
285 pvmw->ptl = pte_lockptr(mm, pvmw->pmd);
286 spin_lock(pvmw->ptl);
287 }
Hugh Dickins47446632021-06-24 18:39:20 -0700288 goto this_pte;
Hugh Dickinsa9a75042021-06-24 18:39:26 -0700289 } while (pvmw->address < end);
290
291 return false;
Kirill A. Shutemovace71a12017-02-24 14:57:45 -0800292}
Kirill A. Shutemov6a328a62017-02-24 14:58:10 -0800293
294/**
295 * page_mapped_in_vma - check whether a page is really mapped in a VMA
296 * @page: the page to test
297 * @vma: the VMA to test
298 *
299 * Returns 1 if the page is mapped into the page tables of the VMA, 0
300 * if the page is not mapped into the page tables of this VMA. Only
301 * valid for normal file or anonymous VMAs.
302 */
303int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
304{
305 struct page_vma_mapped_walk pvmw = {
Matthew Wilcox (Oracle)2aff7a42022-02-03 11:40:17 -0500306 .pfn = page_to_pfn(page),
307 .nr_pages = 1,
Kirill A. Shutemov6a328a62017-02-24 14:58:10 -0800308 .vma = vma,
309 .flags = PVMW_SYNC,
310 };
Kirill A. Shutemov6a328a62017-02-24 14:58:10 -0800311
Hugh Dickins494334e2021-06-15 18:23:56 -0700312 pvmw.address = vma_address(page, vma);
313 if (pvmw.address == -EFAULT)
Kirill A. Shutemov6a328a62017-02-24 14:58:10 -0800314 return 0;
Kirill A. Shutemov6a328a62017-02-24 14:58:10 -0800315 if (!page_vma_mapped_walk(&pvmw))
316 return 0;
317 page_vma_mapped_walk_done(&pvmw);
318 return 1;
319}