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 <asm/tlbflush.h>
  17 #include "filemap.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 ssize_t
 162 xip_file_sendfile(struct file *in_file, loff_t *ppos,
 163              size_t count, read_actor_t actor, void *target)
 164 {
 165         read_descriptor_t desc;
 166
 167         if (!count)
 168                 return 0;
 169
 170         desc.written = 0;
 171         desc.count = count;
 172         desc.arg.data = target;
 173         desc.error = 0;
 174
 175         do_xip_mapping_read(in_file->f_mapping, &in_file->f_ra, in_file,
 176                             ppos, &desc, actor);
 177         if (desc.written)
 178                 return desc.written;
 179         return desc.error;
 180 }
 181 EXPORT_SYMBOL_GPL(xip_file_sendfile);
 182
 183 /*
 184  * __xip_unmap is invoked from xip_unmap and
 185  * xip_write
 186  *
 187  * This function walks all vmas of the address_space and unmaps the
 188  * __xip_sparse_page when found at pgoff.
 189  */
 190 static void
 191 __xip_unmap (struct address_space * mapping,
 192                      unsigned long pgoff)
 193 {
 194         struct vm_area_struct *vma;
 195         struct mm_struct *mm;
 196         struct prio_tree_iter iter;
 197         unsigned long address;
 198         pte_t *pte;
 199         pte_t pteval;
 200         spinlock_t *ptl;
 201         struct page *page;
 202
 203         page = __xip_sparse_page;
 204         if (!page)
 205                 return;
 206
 207         spin_lock(&mapping->i_mmap_lock);
 208         vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
 209                 mm = vma->vm_mm;
 210                 address = vma->vm_start +
 211                         ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
 212                 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
 213                 pte = page_check_address(page, mm, address, &ptl);
 214                 if (pte) {
 215                         /* Nuke the page table entry. */
 216                         flush_cache_page(vma, address, pte_pfn(*pte));
 217                         pteval = ptep_clear_flush(vma, address, pte);
 218                         page_remove_rmap(page, vma);
 219                         dec_mm_counter(mm, file_rss);
 220                         BUG_ON(pte_dirty(pteval));
 221                         pte_unmap_unlock(pte, ptl);
 222                         page_cache_release(page);
 223                 }
 224         }
 225         spin_unlock(&mapping->i_mmap_lock);
 226 }
 227
 228 /*
 229  * xip_nopage() is invoked via the vma operations vector for a
 230  * mapped memory region to read in file data during a page fault.
 231  *
 232  * This function is derived from filemap_nopage, but used for execute in place
 233  */
 234 static struct page *
 235 xip_file_nopage(struct vm_area_struct * area,
 236                    unsigned long address,
 237                    int *type)
 238 {
 239         struct file *file = area->vm_file;
 240         struct address_space *mapping = file->f_mapping;
 241         struct inode *inode = mapping->host;
 242         struct page *page;
 243         unsigned long size, pgoff, endoff;
 244
 245         pgoff = ((address - area->vm_start) >> PAGE_CACHE_SHIFT)
 246                 + area->vm_pgoff;
 247         endoff = ((area->vm_end - area->vm_start) >> PAGE_CACHE_SHIFT)
 248                 + area->vm_pgoff;
 249
 250         size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 251         if (pgoff >= size)
 252                 return NOPAGE_SIGBUS;
 253
 254         page = mapping->a_ops->get_xip_page(mapping, pgoff*(PAGE_SIZE/512), 0);
 255         if (!IS_ERR(page))
 256                 goto out;
 257         if (PTR_ERR(page) != -ENODATA)
 258                 return NOPAGE_SIGBUS;
 259
 260         /* sparse block */
 261         if ((area->vm_flags & (VM_WRITE | VM_MAYWRITE)) &&
 262             (area->vm_flags & (VM_SHARED| VM_MAYSHARE)) &&
 263             (!(mapping->host->i_sb->s_flags & MS_RDONLY))) {
 264                 /* maybe shared writable, allocate new block */
 265                 page = mapping->a_ops->get_xip_page (mapping,
 266                         pgoff*(PAGE_SIZE/512), 1);
 267                 if (IS_ERR(page))
 268                         return NOPAGE_SIGBUS;
 269                 /* unmap page at pgoff from all other vmas */
 270                 __xip_unmap(mapping, pgoff);
 271         } else {
 272                 /* not shared and writable, use xip_sparse_page() */
 273                 page = xip_sparse_page();
 274                 if (!page)
 275                         return NOPAGE_OOM;
 276         }
 277
 278 out:
 279         page_cache_get(page);
 280         return page;
 281 }
 282
 283 static struct vm_operations_struct xip_file_vm_ops = {
 284         .nopage         = xip_file_nopage,
 285 };
 286
 287 int xip_file_mmap(struct file * file, struct vm_area_struct * vma)
 288 {
 289         BUG_ON(!file->f_mapping->a_ops->get_xip_page);
 290
 291         file_accessed(file);
 292         vma->vm_ops = &xip_file_vm_ops;
 293         return 0;
 294 }
 295 EXPORT_SYMBOL_GPL(xip_file_mmap);
 296
 297 static ssize_t
 298 __xip_file_write(struct file *filp, const char __user *buf,
 299                   size_t count, loff_t pos, loff_t *ppos)
 300 {
 301         struct address_space * mapping = filp->f_mapping;
 302         const struct address_space_operations *a_ops = mapping->a_ops;
 303         struct inode    *inode = mapping->host;
 304         long            status = 0;
 305         struct page     *page;
 306         size_t          bytes;
 307         ssize_t         written = 0;
 308
 309         BUG_ON(!mapping->a_ops->get_xip_page);
 310
 311         do {
 312                 unsigned long index;
 313                 unsigned long offset;
 314                 size_t copied;
 315
 316                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
 317                 index = pos >> PAGE_CACHE_SHIFT;
 318                 bytes = PAGE_CACHE_SIZE - offset;
 319                 if (bytes > count)
 320                         bytes = count;
 321
 322                 /*
 323                  * Bring in the user page that we will copy from _first_.
 324                  * Otherwise there's a nasty deadlock on copying from the
 325                  * same page as we're writing to, without it being marked
 326                  * up-to-date.
 327                  */
 328                 fault_in_pages_readable(buf, bytes);
 329
 330                 page = a_ops->get_xip_page(mapping,
 331                                            index*(PAGE_SIZE/512), 0);
 332                 if (IS_ERR(page) && (PTR_ERR(page) == -ENODATA)) {
 333                         /* we allocate a new page unmap it */
 334                         page = a_ops->get_xip_page(mapping,
 335                                                    index*(PAGE_SIZE/512), 1);
 336                         if (!IS_ERR(page))
 337                                 /* unmap page at pgoff from all other vmas */
 338                                 __xip_unmap(mapping, index);
 339                 }
 340
 341                 if (IS_ERR(page)) {
 342                         status = PTR_ERR(page);
 343                         break;
 344                 }
 345
 346                 copied = filemap_copy_from_user(page, offset, buf, bytes);
 347                 flush_dcache_page(page);
 348                 if (likely(copied > 0)) {
 349                         status = copied;
 350
 351                         if (status >= 0) {
 352                                 written += status;
 353                                 count -= status;
 354                                 pos += status;
 355                                 buf += status;
 356                         }
 357                 }
 358                 if (unlikely(copied != bytes))
 359                         if (status >= 0)
 360                                 status = -EFAULT;
 361                 if (status < 0)
 362                         break;
 363         } while (count);
 364         *ppos = pos;
 365         /*
 366          * No need to use i_size_read() here, the i_size
 367          * cannot change under us because we hold i_mutex.
 368          */
 369         if (pos > inode->i_size) {
 370                 i_size_write(inode, pos);
 371                 mark_inode_dirty(inode);
 372         }
 373
 374         return written ? written : status;
 375 }
 376
 377 ssize_t
 378 xip_file_write(struct file *filp, const char __user *buf, size_t len,
 379                loff_t *ppos)
 380 {
 381         struct address_space *mapping = filp->f_mapping;
 382         struct inode *inode = mapping->host;
 383         size_t count;
 384         loff_t pos;
 385         ssize_t ret;
 386
 387         mutex_lock(&inode->i_mutex);
 388
 389         if (!access_ok(VERIFY_READ, buf, len)) {
 390                 ret=-EFAULT;
 391                 goto out_up;
 392         }
 393
 394         pos = *ppos;
 395         count = len;
 396
 397         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 398
 399         /* We can write back this queue in page reclaim */
 400         current->backing_dev_info = mapping->backing_dev_info;
 401
 402         ret = generic_write_checks(filp, &pos, &count, S_ISBLK(inode->i_mode));
 403         if (ret)
 404                 goto out_backing;
 405         if (count == 0)
 406                 goto out_backing;
 407
 408         ret = remove_suid(filp->f_path.dentry);
 409         if (ret)
 410                 goto out_backing;
 411
 412         file_update_time(filp);
 413
 414         ret = __xip_file_write (filp, buf, count, pos, ppos);
 415
 416  out_backing:
 417         current->backing_dev_info = NULL;
 418  out_up:
 419         mutex_unlock(&inode->i_mutex);
 420         return ret;
 421 }
 422 EXPORT_SYMBOL_GPL(xip_file_write);
 423
 424 /*
 425  * truncate a page used for execute in place
 426  * functionality is analog to block_truncate_page but does use get_xip_page
 427  * to get the page instead of page cache
 428  */
 429 int
 430 xip_truncate_page(struct address_space *mapping, loff_t from)
 431 {
 432         pgoff_t index = from >> PAGE_CACHE_SHIFT;
 433         unsigned offset = from & (PAGE_CACHE_SIZE-1);
 434         unsigned blocksize;
 435         unsigned length;
 436         struct page *page;
 437         void *kaddr;
 438
 439         BUG_ON(!mapping->a_ops->get_xip_page);
 440
 441         blocksize = 1 << mapping->host->i_blkbits;
 442         length = offset & (blocksize - 1);
 443
 444         /* Block boundary? Nothing to do */
 445         if (!length)
 446                 return 0;
 447
 448         length = blocksize - length;
 449
 450         page = mapping->a_ops->get_xip_page(mapping,
 451                                             index*(PAGE_SIZE/512), 0);
 452         if (!page)
 453                 return -ENOMEM;
 454         if (unlikely(IS_ERR(page))) {
 455                 if (PTR_ERR(page) == -ENODATA)
 456                         /* Hole? No need to truncate */
 457                         return 0;
 458                 else
 459                         return PTR_ERR(page);
 460         }
 461         kaddr = kmap_atomic(page, KM_USER0);
 462         memset(kaddr + offset, 0, length);
 463         kunmap_atomic(kaddr, KM_USER0);
 464
 465         flush_dcache_page(page);
 466         return 0;
 467 }
 468 EXPORT_SYMBOL_GPL(xip_truncate_page);