mm/filemap_xip.c

   1 /*
   2  *      linux/mm/filemap_xip.c
   3  *
   4  * Copyright (C) 2005 IBM Corporation
   5  * Author: Carsten Otte <cotte@de.ibm.com>
   6  *
   7  * derived from linux/mm/filemap.c - Copyright (C) Linus Torvalds
   8  *
   9  */
  10
  11 #include <linux/fs.h>
  12 #include <linux/pagemap.h>
  13 #include <linux/module.h>
  14 #include <linux/uio.h>
  15 #include <linux/rmap.h>
  16 #include <linux/sched.h>
  17 #include <asm/tlbflush.h>
  18
  19 /*
  20  * We do use our own empty page to avoid interference with other users
  21  * of ZERO_PAGE(), such as /dev/zero
  22  */
  23 static struct page *__xip_sparse_page;
  24
  25 static struct page *xip_sparse_page(void)
  26 {
  27         if (!__xip_sparse_page) {
  28                 unsigned long zeroes = get_zeroed_page(GFP_HIGHUSER);
  29                 if (zeroes) {
  30                         static DEFINE_SPINLOCK(xip_alloc_lock);
  31                         spin_lock(&xip_alloc_lock);
  32                         if (!__xip_sparse_page)
  33                                 __xip_sparse_page = virt_to_page(zeroes);
  34                         else
  35                                 free_page(zeroes);
  36                         spin_unlock(&xip_alloc_lock);
  37                 }
  38         }
  39         return __xip_sparse_page;
  40 }
  41
  42 /*
  43  * This is a file read routine for execute in place files, and uses
  44  * the mapping->a_ops->get_xip_page() function for the actual low-level
  45  * stuff.
  46  *
  47  * Note the struct file* is not used at all.  It may be NULL.
  48  */
  49 static void
  50 do_xip_mapping_read(struct address_space *mapping,
  51                     struct file_ra_state *_ra,
  52                     struct file *filp,
  53                     loff_t *ppos,
  54                     read_descriptor_t *desc,
  55                     read_actor_t actor)
  56 {
  57         struct inode *inode = mapping->host;
  58         unsigned long index, end_index, offset;
  59         loff_t isize;
  60
  61         BUG_ON(!mapping->a_ops->get_xip_page);
  62
  63         index = *ppos >> PAGE_CACHE_SHIFT;
  64         offset = *ppos & ~PAGE_CACHE_MASK;
  65
  66         isize = i_size_read(inode);
  67         if (!isize)
  68                 goto out;
  69
  70         end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
  71         for (;;) {
  72                 struct page *page;
  73                 unsigned long nr, ret;
  74
  75                 /* nr is the maximum number of bytes to copy from this page */
  76                 nr = PAGE_CACHE_SIZE;
  77                 if (index >= end_index) {
  78                         if (index > end_index)
  79                                 goto out;
  80                         nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
  81                         if (nr <= offset) {
  82                                 goto out;
  83                         }
  84                 }
  85                 nr = nr - offset;
  86
  87                 page = mapping->a_ops->get_xip_page(mapping,
  88                         index*(PAGE_SIZE/512), 0);
  89                 if (!page)
  90                         goto no_xip_page;
  91                 if (unlikely(IS_ERR(page))) {
  92                         if (PTR_ERR(page) == -ENODATA) {
  93                                 /* sparse */
  94                                 page = ZERO_PAGE(0);
  95                         } else {
  96                                 desc->error = PTR_ERR(page);
  97                                 goto out;
  98                         }
  99                 }
 100
 101                 /* If users can be writing to this page using arbitrary
 102                  * virtual addresses, take care about potential aliasing
 103                  * before reading the page on the kernel side.
 104                  */
 105                 if (mapping_writably_mapped(mapping))
 106                         flush_dcache_page(page);
 107
 108                 /*
 109                  * Ok, we have the page, so now we can copy it to user space...
 110                  *
 111                  * The actor routine returns how many bytes were actually used..
 112                  * NOTE! This may not be the same as how much of a user buffer
 113                  * we filled up (we may be padding etc), so we can only update
 114                  * "pos" here (the actor routine has to update the user buffer
 115                  * pointers and the remaining count).
 116                  */
 117                 ret = actor(desc, page, offset, nr);
 118                 offset += ret;
 119                 index += offset >> PAGE_CACHE_SHIFT;
 120                 offset &= ~PAGE_CACHE_MASK;
 121
 122                 if (ret == nr && desc->count)
 123                         continue;
 124                 goto out;
 125
 126 no_xip_page:
 127                 /* Did not get the page. Report it */
 128                 desc->error = -EIO;
 129                 goto out;
 130         }
 131
 132 out:
 133         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
 134         if (filp)
 135                 file_accessed(filp);
 136 }
 137
 138 ssize_t
 139 xip_file_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
 140 {
 141         read_descriptor_t desc;
 142
 143         if (!access_ok(VERIFY_WRITE, buf, len))
 144                 return -EFAULT;
 145
 146         desc.written = 0;
 147         desc.arg.buf = buf;
 148         desc.count = len;
 149         desc.error = 0;
 150
 151         do_xip_mapping_read(filp->f_mapping, &filp->f_ra, filp,
 152                             ppos, &desc, file_read_actor);
 153
 154         if (desc.written)
 155                 return desc.written;
 156         else
 157                 return desc.error;
 158 }
 159 EXPORT_SYMBOL_GPL(xip_file_read);
 160
 161 /*
 162  * __xip_unmap is invoked from xip_unmap and
 163  * xip_write
 164  *
 165  * This function walks all vmas of the address_space and unmaps the
 166  * __xip_sparse_page when found at pgoff.
 167  */
 168 static void
 169 __xip_unmap (struct address_space * mapping,
 170                      unsigned long pgoff)
 171 {
 172         struct vm_area_struct *vma;
 173         struct mm_struct *mm;
 174         struct prio_tree_iter iter;
 175         unsigned long address;
 176         pte_t *pte;
 177         pte_t pteval;
 178         spinlock_t *ptl;
 179         struct page *page;
 180
 181         page = __xip_sparse_page;
 182         if (!page)
 183                 return;
 184
 185         spin_lock(&mapping->i_mmap_lock);
 186         vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
 187                 mm = vma->vm_mm;
 188                 address = vma->vm_start +
 189                         ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
 190                 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
 191                 pte = page_check_address(page, mm, address, &ptl);
 192                 if (pte) {
 193                         /* Nuke the page table entry. */
 194                         flush_cache_page(vma, address, pte_pfn(*pte));
 195                         pteval = ptep_clear_flush(vma, address, pte);
 196                         page_remove_rmap(page, vma);
 197                         dec_mm_counter(mm, file_rss);
 198                         BUG_ON(pte_dirty(pteval));
 199                         pte_unmap_unlock(pte, ptl);
 200                         page_cache_release(page);
 201                 }
 202         }
 203         spin_unlock(&mapping->i_mmap_lock);
 204 }
 205
 206 /*
 207  * xip_fault() is invoked via the vma operations vector for a
 208  * mapped memory region to read in file data during a page fault.
 209  *
 210  * This function is derived from filemap_fault, but used for execute in place
 211  */
 212 static int xip_file_fault(struct vm_area_struct *area, struct vm_fault *vmf)
 213 {
 214         struct file *file = area->vm_file;
 215         struct address_space *mapping = file->f_mapping;
 216         struct inode *inode = mapping->host;
 217         struct page *page;
 218         pgoff_t size;
 219
 220         /* XXX: are VM_FAULT_ codes OK? */
 221
 222         size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 223         if (vmf->pgoff >= size)
 224                 return VM_FAULT_SIGBUS;
 225
 226         page = mapping->a_ops->get_xip_page(mapping,
 227                                         vmf->pgoff*(PAGE_SIZE/512), 0);
 228         if (!IS_ERR(page))
 229                 goto out;
 230         if (PTR_ERR(page) != -ENODATA)
 231                 return VM_FAULT_OOM;
 232
 233         /* sparse block */
 234         if ((area->vm_flags & (VM_WRITE | VM_MAYWRITE)) &&
 235             (area->vm_flags & (VM_SHARED| VM_MAYSHARE)) &&
 236             (!(mapping->host->i_sb->s_flags & MS_RDONLY))) {
 237                 /* maybe shared writable, allocate new block */
 238                 page = mapping->a_ops->get_xip_page(mapping,
 239                                         vmf->pgoff*(PAGE_SIZE/512), 1);
 240                 if (IS_ERR(page))
 241                         return VM_FAULT_SIGBUS;
 242                 /* unmap page at pgoff from all other vmas */
 243                 __xip_unmap(mapping, vmf->pgoff);
 244         } else {
 245                 /* not shared and writable, use xip_sparse_page() */
 246                 page = xip_sparse_page();
 247                 if (!page)
 248                         return VM_FAULT_OOM;
 249         }
 250
 251 out:
 252         page_cache_get(page);
 253         vmf->page = page;
 254         return 0;
 255 }
 256
 257 static struct vm_operations_struct xip_file_vm_ops = {
 258         .fault  = xip_file_fault,
 259 };
 260
 261 int xip_file_mmap(struct file * file, struct vm_area_struct * vma)
 262 {
 263         BUG_ON(!file->f_mapping->a_ops->get_xip_page);
 264
 265         file_accessed(file);
 266         vma->vm_ops = &xip_file_vm_ops;
 267         vma->vm_flags |= VM_CAN_NONLINEAR;
 268         return 0;
 269 }
 270 EXPORT_SYMBOL_GPL(xip_file_mmap);
 271
 272 static ssize_t
 273 __xip_file_write(struct file *filp, const char __user *buf,
 274                   size_t count, loff_t pos, loff_t *ppos)
 275 {
 276         struct address_space * mapping = filp->f_mapping;
 277         const struct address_space_operations *a_ops = mapping->a_ops;
 278         struct inode    *inode = mapping->host;
 279         long            status = 0;
 280         struct page     *page;
 281         size_t          bytes;
 282         ssize_t         written = 0;
 283
 284         BUG_ON(!mapping->a_ops->get_xip_page);
 285
 286         do {
 287                 unsigned long index;
 288                 unsigned long offset;
 289                 size_t copied;
 290                 char *kaddr;
 291
 292                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
 293                 index = pos >> PAGE_CACHE_SHIFT;
 294                 bytes = PAGE_CACHE_SIZE - offset;
 295                 if (bytes > count)
 296                         bytes = count;
 297
 298                 page = a_ops->get_xip_page(mapping,
 299                                            index*(PAGE_SIZE/512), 0);
 300                 if (IS_ERR(page) && (PTR_ERR(page) == -ENODATA)) {
 301                         /* we allocate a new page unmap it */
 302                         page = a_ops->get_xip_page(mapping,
 303                                                    index*(PAGE_SIZE/512), 1);
 304                         if (!IS_ERR(page))
 305                                 /* unmap page at pgoff from all other vmas */
 306                                 __xip_unmap(mapping, index);
 307                 }
 308
 309                 if (IS_ERR(page)) {
 310                         status = PTR_ERR(page);
 311                         break;
 312                 }
 313
 314                 fault_in_pages_readable(buf, bytes);
 315                 kaddr = kmap_atomic(page, KM_USER0);
 316                 copied = bytes -
 317                         __copy_from_user_inatomic_nocache(kaddr + offset, buf, bytes);
 318                 kunmap_atomic(kaddr, KM_USER0);
 319                 flush_dcache_page(page);
 320
 321                 if (likely(copied > 0)) {
 322                         status = copied;
 323
 324                         if (status >= 0) {
 325                                 written += status;
 326                                 count -= status;
 327                                 pos += status;
 328                                 buf += status;
 329                         }
 330                 }
 331                 if (unlikely(copied != bytes))
 332                         if (status >= 0)
 333                                 status = -EFAULT;
 334                 if (status < 0)
 335                         break;
 336         } while (count);
 337         *ppos = pos;
 338         /*
 339          * No need to use i_size_read() here, the i_size
 340          * cannot change under us because we hold i_mutex.
 341          */
 342         if (pos > inode->i_size) {
 343                 i_size_write(inode, pos);
 344                 mark_inode_dirty(inode);
 345         }
 346
 347         return written ? written : status;
 348 }
 349
 350 ssize_t
 351 xip_file_write(struct file *filp, const char __user *buf, size_t len,
 352                loff_t *ppos)
 353 {
 354         struct address_space *mapping = filp->f_mapping;
 355         struct inode *inode = mapping->host;
 356         size_t count;
 357         loff_t pos;
 358         ssize_t ret;
 359
 360         mutex_lock(&inode->i_mutex);
 361
 362         if (!access_ok(VERIFY_READ, buf, len)) {
 363                 ret=-EFAULT;
 364                 goto out_up;
 365         }
 366
 367         pos = *ppos;
 368         count = len;
 369
 370         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 371
 372         /* We can write back this queue in page reclaim */
 373         current->backing_dev_info = mapping->backing_dev_info;
 374
 375         ret = generic_write_checks(filp, &pos, &count, S_ISBLK(inode->i_mode));
 376         if (ret)
 377                 goto out_backing;
 378         if (count == 0)
 379                 goto out_backing;
 380
 381         ret = remove_suid(filp->f_path.dentry);
 382         if (ret)
 383                 goto out_backing;
 384
 385         file_update_time(filp);
 386
 387         ret = __xip_file_write (filp, buf, count, pos, ppos);
 388
 389  out_backing:
 390         current->backing_dev_info = NULL;
 391  out_up:
 392         mutex_unlock(&inode->i_mutex);
 393         return ret;
 394 }
 395 EXPORT_SYMBOL_GPL(xip_file_write);
 396
 397 /*
 398  * truncate a page used for execute in place
 399  * functionality is analog to block_truncate_page but does use get_xip_page
 400  * to get the page instead of page cache
 401  */
 402 int
 403 xip_truncate_page(struct address_space *mapping, loff_t from)
 404 {
 405         pgoff_t index = from >> PAGE_CACHE_SHIFT;
 406         unsigned offset = from & (PAGE_CACHE_SIZE-1);
 407         unsigned blocksize;
 408         unsigned length;
 409         struct page *page;
 410
 411         BUG_ON(!mapping->a_ops->get_xip_page);
 412
 413         blocksize = 1 << mapping->host->i_blkbits;
 414         length = offset & (blocksize - 1);
 415
 416         /* Block boundary? Nothing to do */
 417         if (!length)
 418                 return 0;
 419
 420         length = blocksize - length;
 421
 422         page = mapping->a_ops->get_xip_page(mapping,
 423                                             index*(PAGE_SIZE/512), 0);
 424         if (!page)
 425                 return -ENOMEM;
 426         if (unlikely(IS_ERR(page))) {
 427                 if (PTR_ERR(page) == -ENODATA)
 428                         /* Hole? No need to truncate */
 429                         return 0;
 430                 else
 431                         return PTR_ERR(page);
 432         }
 433         zero_user_page(page, offset, length, KM_USER0);
 434         return 0;
 435 }
 436 EXPORT_SYMBOL_GPL(xip_truncate_page);