mm/frame_vector.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <linux/kernel.h>
   3 #include <linux/errno.h>
   4 #include <linux/err.h>
   5 #include <linux/mm.h>
   6 #include <linux/slab.h>
   7 #include <linux/vmalloc.h>
   8 #include <linux/pagemap.h>
   9 #include <linux/sched.h>
  10
  11 /**
  12  * get_vaddr_frames() - map virtual addresses to pfns
  13  * @start:      starting user address
  14  * @nr_frames:  number of pages / pfns from start to map
  15  * @gup_flags:  flags modifying lookup behaviour
  16  * @vec:        structure which receives pages / pfns of the addresses mapped.
  17  *              It should have space for at least nr_frames entries.
  18  *
  19  * This function maps virtual addresses from @start and fills @vec structure
  20  * with page frame numbers or page pointers to corresponding pages (choice
  21  * depends on the type of the vma underlying the virtual address). If @start
  22  * belongs to a normal vma, the function grabs reference to each of the pages
  23  * to pin them in memory. If @start belongs to VM_IO | VM_PFNMAP vma, we don't
  24  * touch page structures and the caller must make sure pfns aren't reused for
  25  * anything else while he is using them.
  26  *
  27  * The function returns number of pages mapped which may be less than
  28  * @nr_frames. In particular we stop mapping if there are more vmas of
  29  * different type underlying the specified range of virtual addresses.
  30  * When the function isn't able to map a single page, it returns error.
  31  *
  32  * This function takes care of grabbing mmap_lock as necessary.
  33  */
  34 int get_vaddr_frames(unsigned long start, unsigned int nr_frames,
  35                      unsigned int gup_flags, struct frame_vector *vec)
  36 {
  37         struct mm_struct *mm = current->mm;
  38         struct vm_area_struct *vma;
  39         int ret = 0;
  40         int err;
  41         int locked;
  42
  43         if (nr_frames == 0)
  44                 return 0;
  45
  46         if (WARN_ON_ONCE(nr_frames > vec->nr_allocated))
  47                 nr_frames = vec->nr_allocated;
  48
  49         start = untagged_addr(start);
  50
  51         mmap_read_lock(mm);
  52         locked = 1;
  53         vma = find_vma_intersection(mm, start, start + 1);
  54         if (!vma) {
  55                 ret = -EFAULT;
  56                 goto out;
  57         }
  58
  59         /*
  60          * While get_vaddr_frames() could be used for transient (kernel
  61          * controlled lifetime) pinning of memory pages all current
  62          * users establish long term (userspace controlled lifetime)
  63          * page pinning. Treat get_vaddr_frames() like
  64          * get_user_pages_longterm() and disallow it for filesystem-dax
  65          * mappings.
  66          */
  67         if (vma_is_fsdax(vma)) {
  68                 ret = -EOPNOTSUPP;
  69                 goto out;
  70         }
  71
  72         if (!(vma->vm_flags & (VM_IO | VM_PFNMAP))) {
  73                 vec->got_ref = true;
  74                 vec->is_pfns = false;
  75                 ret = pin_user_pages_locked(start, nr_frames,
  76                         gup_flags, (struct page **)(vec->ptrs), &locked);
  77                 goto out;
  78         }
  79
  80         vec->got_ref = false;
  81         vec->is_pfns = true;
  82         do {
  83                 unsigned long *nums = frame_vector_pfns(vec);
  84
  85                 while (ret < nr_frames && start + PAGE_SIZE <= vma->vm_end) {
  86                         err = follow_pfn(vma, start, &nums[ret]);
  87                         if (err) {
  88                                 if (ret == 0)
  89                                         ret = err;
  90                                 goto out;
  91                         }
  92                         start += PAGE_SIZE;
  93                         ret++;
  94                 }
  95                 /*
  96                  * We stop if we have enough pages or if VMA doesn't completely
  97                  * cover the tail page.
  98                  */
  99                 if (ret >= nr_frames || start < vma->vm_end)
 100                         break;
 101                 vma = find_vma_intersection(mm, start, start + 1);
 102         } while (vma && vma->vm_flags & (VM_IO | VM_PFNMAP));
 103 out:
 104         if (locked)
 105                 mmap_read_unlock(mm);
 106         if (!ret)
 107                 ret = -EFAULT;
 108         if (ret > 0)
 109                 vec->nr_frames = ret;
 110         return ret;
 111 }
 112 EXPORT_SYMBOL(get_vaddr_frames);
 113
 114 /**
 115  * put_vaddr_frames() - drop references to pages if get_vaddr_frames() acquired
 116  *                      them
 117  * @vec:        frame vector to put
 118  *
 119  * Drop references to pages if get_vaddr_frames() acquired them. We also
 120  * invalidate the frame vector so that it is prepared for the next call into
 121  * get_vaddr_frames().
 122  */
 123 void put_vaddr_frames(struct frame_vector *vec)
 124 {
 125         struct page **pages;
 126
 127         if (!vec->got_ref)
 128                 goto out;
 129         pages = frame_vector_pages(vec);
 130         /*
 131          * frame_vector_pages() might needed to do a conversion when
 132          * get_vaddr_frames() got pages but vec was later converted to pfns.
 133          * But it shouldn't really fail to convert pfns back...
 134          */
 135         if (WARN_ON(IS_ERR(pages)))
 136                 goto out;
 137
 138         unpin_user_pages(pages, vec->nr_frames);
 139         vec->got_ref = false;
 140 out:
 141         vec->nr_frames = 0;
 142 }
 143 EXPORT_SYMBOL(put_vaddr_frames);
 144
 145 /**
 146  * frame_vector_to_pages - convert frame vector to contain page pointers
 147  * @vec:        frame vector to convert
 148  *
 149  * Convert @vec to contain array of page pointers.  If the conversion is
 150  * successful, return 0. Otherwise return an error. Note that we do not grab
 151  * page references for the page structures.
 152  */
 153 int frame_vector_to_pages(struct frame_vector *vec)
 154 {
 155         int i;
 156         unsigned long *nums;
 157         struct page **pages;
 158
 159         if (!vec->is_pfns)
 160                 return 0;
 161         nums = frame_vector_pfns(vec);
 162         for (i = 0; i < vec->nr_frames; i++)
 163                 if (!pfn_valid(nums[i]))
 164                         return -EINVAL;
 165         pages = (struct page **)nums;
 166         for (i = 0; i < vec->nr_frames; i++)
 167                 pages[i] = pfn_to_page(nums[i]);
 168         vec->is_pfns = false;
 169         return 0;
 170 }
 171 EXPORT_SYMBOL(frame_vector_to_pages);
 172
 173 /**
 174  * frame_vector_to_pfns - convert frame vector to contain pfns
 175  * @vec:        frame vector to convert
 176  *
 177  * Convert @vec to contain array of pfns.
 178  */
 179 void frame_vector_to_pfns(struct frame_vector *vec)
 180 {
 181         int i;
 182         unsigned long *nums;
 183         struct page **pages;
 184
 185         if (vec->is_pfns)
 186                 return;
 187         pages = (struct page **)(vec->ptrs);
 188         nums = (unsigned long *)pages;
 189         for (i = 0; i < vec->nr_frames; i++)
 190                 nums[i] = page_to_pfn(pages[i]);
 191         vec->is_pfns = true;
 192 }
 193 EXPORT_SYMBOL(frame_vector_to_pfns);
 194
 195 /**
 196  * frame_vector_create() - allocate & initialize structure for pinned pfns
 197  * @nr_frames:  number of pfns slots we should reserve
 198  *
 199  * Allocate and initialize struct pinned_pfns to be able to hold @nr_pfns
 200  * pfns.
 201  */
 202 struct frame_vector *frame_vector_create(unsigned int nr_frames)
 203 {
 204         struct frame_vector *vec;
 205         int size = sizeof(struct frame_vector) + sizeof(void *) * nr_frames;
 206
 207         if (WARN_ON_ONCE(nr_frames == 0))
 208                 return NULL;
 209         /*
 210          * This is absurdly high. It's here just to avoid strange effects when
 211          * arithmetics overflows.
 212          */
 213         if (WARN_ON_ONCE(nr_frames > INT_MAX / sizeof(void *) / 2))
 214                 return NULL;
 215         /*
 216          * Avoid higher order allocations, use vmalloc instead. It should
 217          * be rare anyway.
 218          */
 219         vec = kvmalloc(size, GFP_KERNEL);
 220         if (!vec)
 221                 return NULL;
 222         vec->nr_allocated = nr_frames;
 223         vec->nr_frames = 0;
 224         return vec;
 225 }
 226 EXPORT_SYMBOL(frame_vector_create);
 227
 228 /**
 229  * frame_vector_destroy() - free memory allocated to carry frame vector
 230  * @vec:        Frame vector to free
 231  *
 232  * Free structure allocated by frame_vector_create() to carry frames.
 233  */
 234 void frame_vector_destroy(struct frame_vector *vec)
 235 {
 236         /* Make sure put_vaddr_frames() got called properly... */
 237         VM_BUG_ON(vec->nr_frames > 0);
 238         kvfree(vec);
 239 }
 240 EXPORT_SYMBOL(frame_vector_destroy);