mm/page_vma_mapped.c

   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/swapops.h>
   7
   8 #include "internal.h"
   9
  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
  16 static 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 {
  24                         if (!pte_present(*pvmw->pte))
  25                                 return false;
  26                 }
  27         }
  28         pvmw->ptl = pte_lockptr(pvmw->vma->vm_mm, pvmw->pmd);
  29         spin_lock(pvmw->ptl);
  30         return true;
  31 }
  32
  33 static inline bool pfn_in_hpage(struct page *hpage, unsigned long pfn)
  34 {
  35         unsigned long hpage_pfn = page_to_pfn(hpage);
  36
  37         /* THP can be referenced by any subpage */
  38         return pfn >= hpage_pfn && pfn - hpage_pfn < hpage_nr_pages(hpage);
  39 }
  40
  41 /**
  42  * check_pte - check if @pvmw->page is mapped at the @pvmw->pte
  43  *
  44  * page_vma_mapped_walk() found a place where @pvmw->page is *potentially*
  45  * mapped. check_pte() has to validate this.
  46  *
  47  * @pvmw->pte may point to empty PTE, swap PTE or PTE pointing to arbitrary
  48  * page.
  49  *
  50  * If PVMW_MIGRATION flag is set, returns true if @pvmw->pte contains migration
  51  * entry that points to @pvmw->page or any subpage in case of THP.
  52  *
  53  * If PVMW_MIGRATION flag is not set, returns true if @pvmw->pte points to
  54  * @pvmw->page or any subpage in case of THP.
  55  *
  56  * Otherwise, return false.
  57  *
  58  */
  59 static bool check_pte(struct page_vma_mapped_walk *pvmw)
  60 {
  61         unsigned long pfn;
  62
  63         if (pvmw->flags & PVMW_MIGRATION) {
  64                 swp_entry_t entry;
  65                 if (!is_swap_pte(*pvmw->pte))
  66                         return false;
  67                 entry = pte_to_swp_entry(*pvmw->pte);
  68
  69                 if (!is_migration_entry(entry))
  70                         return false;
  71
  72                 pfn = migration_entry_to_pfn(entry);
  73         } else if (is_swap_pte(*pvmw->pte)) {
  74                 swp_entry_t entry;
  75
  76                 /* Handle un-addressable ZONE_DEVICE memory */
  77                 entry = pte_to_swp_entry(*pvmw->pte);
  78                 if (!is_device_private_entry(entry))
  79                         return false;
  80
  81                 pfn = device_private_entry_to_pfn(entry);
  82         } else {
  83                 if (!pte_present(*pvmw->pte))
  84                         return false;
  85
  86                 pfn = pte_pfn(*pvmw->pte);
  87         }
  88
  89         return pfn_in_hpage(pvmw->page, pfn);
  90 }
  91
  92 /**
  93  * page_vma_mapped_walk - check if @pvmw->page is mapped in @pvmw->vma at
  94  * @pvmw->address
  95  * @pvmw: pointer to struct page_vma_mapped_walk. page, vma, address and flags
  96  * must be set. pmd, pte and ptl must be NULL.
  97  *
  98  * Returns true if the page is mapped in the vma. @pvmw->pmd and @pvmw->pte point
  99  * to relevant page table entries. @pvmw->ptl is locked. @pvmw->address is
 100  * adjusted if needed (for PTE-mapped THPs).
 101  *
 102  * If @pvmw->pmd is set but @pvmw->pte is not, you have found PMD-mapped page
 103  * (usually THP). For PTE-mapped THP, you should run page_vma_mapped_walk() in
 104  * a loop to find all PTEs that map the THP.
 105  *
 106  * For HugeTLB pages, @pvmw->pte is set to the relevant page table entry
 107  * regardless of which page table level the page is mapped at. @pvmw->pmd is
 108  * NULL.
 109  *
 110  * Retruns false if there are no more page table entries for the page in
 111  * the vma. @pvmw->ptl is unlocked and @pvmw->pte is unmapped.
 112  *
 113  * If you need to stop the walk before page_vma_mapped_walk() returned false,
 114  * use page_vma_mapped_walk_done(). It will do the housekeeping.
 115  */
 116 bool page_vma_mapped_walk(struct page_vma_mapped_walk *pvmw)
 117 {
 118         struct mm_struct *mm = pvmw->vma->vm_mm;
 119         struct page *page = pvmw->page;
 120         pgd_t *pgd;
 121         p4d_t *p4d;
 122         pud_t *pud;
 123         pmd_t pmde;
 124
 125         /* The only possible pmd mapping has been handled on last iteration */
 126         if (pvmw->pmd && !pvmw->pte)
 127                 return not_found(pvmw);
 128
 129         if (pvmw->pte)
 130                 goto next_pte;
 131
 132         if (unlikely(PageHuge(pvmw->page))) {
 133                 /* when pud is not present, pte will be NULL */
 134                 pvmw->pte = huge_pte_offset(mm, pvmw->address,
 135                                             PAGE_SIZE << compound_order(page));
 136                 if (!pvmw->pte)
 137                         return false;
 138
 139                 pvmw->ptl = huge_pte_lockptr(page_hstate(page), mm, pvmw->pte);
 140                 spin_lock(pvmw->ptl);
 141                 if (!check_pte(pvmw))
 142                         return not_found(pvmw);
 143                 return true;
 144         }
 145 restart:
 146         pgd = pgd_offset(mm, pvmw->address);
 147         if (!pgd_present(*pgd))
 148                 return false;
 149         p4d = p4d_offset(pgd, pvmw->address);
 150         if (!p4d_present(*p4d))
 151                 return false;
 152         pud = pud_offset(p4d, pvmw->address);
 153         if (!pud_present(*pud))
 154                 return false;
 155         pvmw->pmd = pmd_offset(pud, pvmw->address);
 156         /*
 157          * Make sure the pmd value isn't cached in a register by the
 158          * compiler and used as a stale value after we've observed a
 159          * subsequent update.
 160          */
 161         pmde = READ_ONCE(*pvmw->pmd);
 162         if (pmd_trans_huge(pmde) || is_pmd_migration_entry(pmde)) {
 163                 pvmw->ptl = pmd_lock(mm, pvmw->pmd);
 164                 if (likely(pmd_trans_huge(*pvmw->pmd))) {
 165                         if (pvmw->flags & PVMW_MIGRATION)
 166                                 return not_found(pvmw);
 167                         if (pmd_page(*pvmw->pmd) != page)
 168                                 return not_found(pvmw);
 169                         return true;
 170                 } else if (!pmd_present(*pvmw->pmd)) {
 171                         if (thp_migration_supported()) {
 172                                 if (!(pvmw->flags & PVMW_MIGRATION))
 173                                         return not_found(pvmw);
 174                                 if (is_migration_entry(pmd_to_swp_entry(*pvmw->pmd))) {
 175                                         swp_entry_t entry = pmd_to_swp_entry(*pvmw->pmd);
 176
 177                                         if (migration_entry_to_page(entry) != page)
 178                                                 return not_found(pvmw);
 179                                         return true;
 180                                 }
 181                         }
 182                         return not_found(pvmw);
 183                 } else {
 184                         /* THP pmd was split under us: handle on pte level */
 185                         spin_unlock(pvmw->ptl);
 186                         pvmw->ptl = NULL;
 187                 }
 188         } else if (!pmd_present(pmde)) {
 189                 return false;
 190         }
 191         if (!map_pte(pvmw))
 192                 goto next_pte;
 193         while (1) {
 194                 if (check_pte(pvmw))
 195                         return true;
 196 next_pte:
 197                 /* Seek to next pte only makes sense for THP */
 198                 if (!PageTransHuge(pvmw->page) || PageHuge(pvmw->page))
 199                         return not_found(pvmw);
 200                 do {
 201                         pvmw->address += PAGE_SIZE;
 202                         if (pvmw->address >= pvmw->vma->vm_end ||
 203                             pvmw->address >=
 204                                         __vma_address(pvmw->page, pvmw->vma) +
 205                                         hpage_nr_pages(pvmw->page) * PAGE_SIZE)
 206                                 return not_found(pvmw);
 207                         /* Did we cross page table boundary? */
 208                         if (pvmw->address % PMD_SIZE == 0) {
 209                                 pte_unmap(pvmw->pte);
 210                                 if (pvmw->ptl) {
 211                                         spin_unlock(pvmw->ptl);
 212                                         pvmw->ptl = NULL;
 213                                 }
 214                                 goto restart;
 215                         } else {
 216                                 pvmw->pte++;
 217                         }
 218                 } while (pte_none(*pvmw->pte));
 219
 220                 if (!pvmw->ptl) {
 221                         pvmw->ptl = pte_lockptr(mm, pvmw->pmd);
 222                         spin_lock(pvmw->ptl);
 223                 }
 224         }
 225 }
 226
 227 /**
 228  * page_mapped_in_vma - check whether a page is really mapped in a VMA
 229  * @page: the page to test
 230  * @vma: the VMA to test
 231  *
 232  * Returns 1 if the page is mapped into the page tables of the VMA, 0
 233  * if the page is not mapped into the page tables of this VMA.  Only
 234  * valid for normal file or anonymous VMAs.
 235  */
 236 int page_mapped_in_vma(struct page *page, struct vm_area_struct *vma)
 237 {
 238         struct page_vma_mapped_walk pvmw = {
 239                 .page = page,
 240                 .vma = vma,
 241                 .flags = PVMW_SYNC,
 242         };
 243         unsigned long start, end;
 244
 245         start = __vma_address(page, vma);
 246         end = start + PAGE_SIZE * (hpage_nr_pages(page) - 1);
 247
 248         if (unlikely(end < vma->vm_start || start >= vma->vm_end))
 249                 return 0;
 250         pvmw.address = max(start, vma->vm_start);
 251         if (!page_vma_mapped_walk(&pvmw))
 252                 return 0;
 253         page_vma_mapped_walk_done(&pvmw);
 254         return 1;
 255 }