]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blobdiff - mm/memory.c
oom, oom_reaper: protect oom_reaper_list using simpler way
[mirror_ubuntu-bionic-kernel.git] / mm / memory.c
index 8132787ae4d509d475ed6a77705076ddfee30630..098f00d0546170d6a969d4994b27cfab7a4955dc 100644 (file)
@@ -65,6 +65,7 @@
 #include <linux/userfaultfd_k.h>
 
 #include <asm/io.h>
+#include <asm/mmu_context.h>
 #include <asm/pgalloc.h>
 #include <asm/uaccess.h>
 #include <asm/tlb.h>
@@ -562,8 +563,7 @@ void free_pgtables(struct mmu_gather *tlb, struct vm_area_struct *vma,
        }
 }
 
-int __pte_alloc(struct mm_struct *mm, struct vm_area_struct *vma,
-               pmd_t *pmd, unsigned long address)
+int __pte_alloc(struct mm_struct *mm, pmd_t *pmd, unsigned long address)
 {
        spinlock_t *ptl;
        pgtable_t new = pte_alloc_one(mm, address);
@@ -661,9 +661,8 @@ static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
                        return;
                }
                if (nr_unshown) {
-                       printk(KERN_ALERT
-                               "BUG: Bad page map: %lu messages suppressed\n",
-                               nr_unshown);
+                       pr_alert("BUG: Bad page map: %lu messages suppressed\n",
+                                nr_unshown);
                        nr_unshown = 0;
                }
                nr_shown = 0;
@@ -674,15 +673,13 @@ static void print_bad_pte(struct vm_area_struct *vma, unsigned long addr,
        mapping = vma->vm_file ? vma->vm_file->f_mapping : NULL;
        index = linear_page_index(vma, addr);
 
-       printk(KERN_ALERT
-               "BUG: Bad page map in process %s  pte:%08llx pmd:%08llx\n",
-               current->comm,
-               (long long)pte_val(pte), (long long)pmd_val(*pmd));
+       pr_alert("BUG: Bad page map in process %s  pte:%08llx pmd:%08llx\n",
+                current->comm,
+                (long long)pte_val(pte), (long long)pmd_val(*pmd));
        if (page)
                dump_page(page, "bad pte");
-       printk(KERN_ALERT
-               "addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
-               (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
+       pr_alert("addr:%p vm_flags:%08lx anon_vma:%p mapping:%p index:%lx\n",
+                (void *)addr, vma->vm_flags, vma->anon_vma, mapping, index);
        /*
         * Choose text because data symbols depend on CONFIG_KALLSYMS_ALL=y
         */
@@ -1105,6 +1102,12 @@ again:
 
                        if (!PageAnon(page)) {
                                if (pte_dirty(ptent)) {
+                                       /*
+                                        * oom_reaper cannot tear down dirty
+                                        * pages
+                                        */
+                                       if (unlikely(details && details->ignore_dirty))
+                                               continue;
                                        force_flush = 1;
                                        set_page_dirty(page);
                                }
@@ -1123,8 +1126,8 @@ again:
                        }
                        continue;
                }
-               /* If details->check_mapping, we leave swap entries. */
-               if (unlikely(details))
+               /* only check swap_entries if explicitly asked for in details */
+               if (unlikely(details && !details->check_swap_entries))
                        continue;
 
                entry = pte_to_swp_entry(ptent);
@@ -1229,7 +1232,7 @@ static inline unsigned long zap_pud_range(struct mmu_gather *tlb,
        return addr;
 }
 
-static void unmap_page_range(struct mmu_gather *tlb,
+void unmap_page_range(struct mmu_gather *tlb,
                             struct vm_area_struct *vma,
                             unsigned long addr, unsigned long end,
                             struct zap_details *details)
@@ -1237,9 +1240,6 @@ static void unmap_page_range(struct mmu_gather *tlb,
        pgd_t *pgd;
        unsigned long next;
 
-       if (details && !details->check_mapping)
-               details = NULL;
-
        BUG_ON(addr >= end);
        tlb_start_vma(tlb, vma);
        pgd = pgd_offset(vma->vm_mm, addr);
@@ -1550,9 +1550,30 @@ out:
  */
 int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
                        unsigned long pfn)
+{
+       return vm_insert_pfn_prot(vma, addr, pfn, vma->vm_page_prot);
+}
+EXPORT_SYMBOL(vm_insert_pfn);
+
+/**
+ * vm_insert_pfn_prot - insert single pfn into user vma with specified pgprot
+ * @vma: user vma to map to
+ * @addr: target user address of this page
+ * @pfn: source kernel pfn
+ * @pgprot: pgprot flags for the inserted page
+ *
+ * This is exactly like vm_insert_pfn, except that it allows drivers to
+ * to override pgprot on a per-page basis.
+ *
+ * This only makes sense for IO mappings, and it makes no sense for
+ * cow mappings.  In general, using multiple vmas is preferable;
+ * vm_insert_pfn_prot should only be used if using multiple VMAs is
+ * impractical.
+ */
+int vm_insert_pfn_prot(struct vm_area_struct *vma, unsigned long addr,
+                       unsigned long pfn, pgprot_t pgprot)
 {
        int ret;
-       pgprot_t pgprot = vma->vm_page_prot;
        /*
         * Technically, architectures with pte_special can avoid all these
         * restrictions (same for remap_pfn_range).  However we would like
@@ -1574,7 +1595,7 @@ int vm_insert_pfn(struct vm_area_struct *vma, unsigned long addr,
 
        return ret;
 }
-EXPORT_SYMBOL(vm_insert_pfn);
+EXPORT_SYMBOL(vm_insert_pfn_prot);
 
 int vm_insert_mixed(struct vm_area_struct *vma, unsigned long addr,
                        pfn_t pfn)
@@ -1876,7 +1897,9 @@ int apply_to_page_range(struct mm_struct *mm, unsigned long addr,
        unsigned long end = addr + size;
        int err;
 
-       BUG_ON(addr >= end);
+       if (WARN_ON(addr >= end))
+               return -EINVAL;
+
        pgd = pgd_offset(mm, addr);
        do {
                next = pgd_addr_end(addr, end);
@@ -2412,7 +2435,7 @@ static inline void unmap_mapping_range_tree(struct rb_root *root,
 void unmap_mapping_range(struct address_space *mapping,
                loff_t const holebegin, loff_t const holelen, int even_cows)
 {
-       struct zap_details details;
+       struct zap_details details = { };
        pgoff_t hba = holebegin >> PAGE_SHIFT;
        pgoff_t hlen = (holelen + PAGE_SIZE - 1) >> PAGE_SHIFT;
 
@@ -3122,8 +3145,7 @@ static int do_fault(struct mm_struct *mm, struct vm_area_struct *vma,
                unsigned long address, pte_t *page_table, pmd_t *pmd,
                unsigned int flags, pte_t orig_pte)
 {
-       pgoff_t pgoff = (((address & PAGE_MASK)
-                       - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
+       pgoff_t pgoff = linear_page_index(vma, address);
 
        pte_unmap(page_table);
        /* The VMA was not fully populated on mmap() or missing VM_DONTEXPAND */
@@ -3357,6 +3379,11 @@ static int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
        pmd_t *pmd;
        pte_t *pte;
 
+       if (!arch_vma_access_permitted(vma, flags & FAULT_FLAG_WRITE,
+                                           flags & FAULT_FLAG_INSTRUCTION,
+                                           flags & FAULT_FLAG_REMOTE))
+               return VM_FAULT_SIGSEGV;
+
        if (unlikely(is_vm_hugetlb_page(vma)))
                return hugetlb_fault(mm, vma, address, flags);
 
@@ -3397,12 +3424,11 @@ static int __handle_mm_fault(struct mm_struct *mm, struct vm_area_struct *vma,
        }
 
        /*
-        * Use __pte_alloc instead of pte_alloc_map, because we can't
+        * Use pte_alloc() instead of pte_alloc_map, because we can't
         * run pte_offset_map on the pmd, if an huge pmd could
         * materialize from under us from a different thread.
         */
-       if (unlikely(pmd_none(*pmd)) &&
-           unlikely(__pte_alloc(mm, vma, pmd, address)))
+       if (unlikely(pte_alloc(mm, pmd, address)))
                return VM_FAULT_OOM;
        /*
         * If a huge pmd materialized under us just retry later.  Use
@@ -3674,7 +3700,7 @@ static int __access_remote_vm(struct task_struct *tsk, struct mm_struct *mm,
                void *maddr;
                struct page *page = NULL;
 
-               ret = get_user_pages(tsk, mm, addr, 1,
+               ret = get_user_pages_remote(tsk, mm, addr, 1,
                                write, 1, &page, &vma);
                if (ret <= 0) {
 #ifndef CONFIG_HAVE_IOREMAP_PROT