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 #include "filemap.h"
  19
  20 /*
  21  * We do use our own empty page to avoid interference with other users
  22  * of ZERO_PAGE(), such as /dev/zero
  23  */
  24 static struct page *__xip_sparse_page;
  25
  26 static struct page *xip_sparse_page(void)
  27 {
  28         if (!__xip_sparse_page) {
  29                 unsigned long zeroes = get_zeroed_page(GFP_HIGHUSER);
  30                 if (zeroes) {
  31                         static DEFINE_SPINLOCK(xip_alloc_lock);
  32                         spin_lock(&xip_alloc_lock);
  33                         if (!__xip_sparse_page)
  34                                 __xip_sparse_page = virt_to_page(zeroes);
  35                         else
  36                                 free_page(zeroes);
  37                         spin_unlock(&xip_alloc_lock);
  38                 }
  39         }
  40         return __xip_sparse_page;
  41 }
  42
  43 /*
  44  * This is a file read routine for execute in place files, and uses
  45  * the mapping->a_ops->get_xip_page() function for the actual low-level
  46  * stuff.
  47  *
  48  * Note the struct file* is not used at all.  It may be NULL.
  49  */
  50 static void
  51 do_xip_mapping_read(struct address_space *mapping,
  52                     struct file_ra_state *_ra,
  53                     struct file *filp,
  54                     loff_t *ppos,
  55                     read_descriptor_t *desc,
  56                     read_actor_t actor)
  57 {
  58         struct inode *inode = mapping->host;
  59         unsigned long index, end_index, offset;
  60         loff_t isize;
  61
  62         BUG_ON(!mapping->a_ops->get_xip_page);
  63
  64         index = *ppos >> PAGE_CACHE_SHIFT;
  65         offset = *ppos & ~PAGE_CACHE_MASK;
  66
  67         isize = i_size_read(inode);
  68         if (!isize)
  69                 goto out;
  70
  71         end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
  72         for (;;) {
  73                 struct page *page;
  74                 unsigned long nr, ret;
  75
  76                 /* nr is the maximum number of bytes to copy from this page */
  77                 nr = PAGE_CACHE_SIZE;
  78                 if (index >= end_index) {
  79                         if (index > end_index)
  80                                 goto out;
  81                         nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
  82                         if (nr <= offset) {
  83                                 goto out;
  84                         }
  85                 }
  86                 nr = nr - offset;
  87
  88                 page = mapping->a_ops->get_xip_page(mapping,
  89                         index*(PAGE_SIZE/512), 0);
  90                 if (!page)
  91                         goto no_xip_page;
  92                 if (unlikely(IS_ERR(page))) {
  93                         if (PTR_ERR(page) == -ENODATA) {
  94                                 /* sparse */
  95                                 page = ZERO_PAGE(0);
  96                         } else {
  97                                 desc->error = PTR_ERR(page);
  98                                 goto out;
  99                         }
 100                 }
 101
 102                 /* If users can be writing to this page using arbitrary
 103                  * virtual addresses, take care about potential aliasing
 104                  * before reading the page on the kernel side.
 105                  */
 106                 if (mapping_writably_mapped(mapping))
 107                         flush_dcache_page(page);
 108
 109                 /*
 110                  * Ok, we have the page, so now we can copy it to user space...
 111                  *
 112                  * The actor routine returns how many bytes were actually used..
 113                  * NOTE! This may not be the same as how much of a user buffer
 114                  * we filled up (we may be padding etc), so we can only update
 115                  * "pos" here (the actor routine has to update the user buffer
 116                  * pointers and the remaining count).
 117                  */
 118                 ret = actor(desc, page, offset, nr);
 119                 offset += ret;
 120                 index += offset >> PAGE_CACHE_SHIFT;
 121                 offset &= ~PAGE_CACHE_MASK;
 122
 123                 if (ret == nr && desc->count)
 124                         continue;
 125                 goto out;
 126
 127 no_xip_page:
 128                 /* Did not get the page. Report it */
 129                 desc->error = -EIO;
 130                 goto out;
 131         }
 132
 133 out:
 134         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
 135         if (filp)
 136                 file_accessed(filp);
 137 }
 138
 139 ssize_t
 140 xip_file_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
 141 {
 142         read_descriptor_t desc;
 143
 144         if (!access_ok(VERIFY_WRITE, buf, len))
 145                 return -EFAULT;
 146
 147         desc.written = 0;
 148         desc.arg.buf = buf;
 149         desc.count = len;
 150         desc.error = 0;
 151
 152         do_xip_mapping_read(filp->f_mapping, &filp->f_ra, filp,
 153                             ppos, &desc, file_read_actor);
 154
 155         if (desc.written)
 156                 return desc.written;
 157         else
 158                 return desc.error;
 159 }
 160 EXPORT_SYMBOL_GPL(xip_file_read);
 161
 162 ssize_t
 163 xip_file_sendfile(struct file *in_file, loff_t *ppos,
 164              size_t count, read_actor_t actor, void *target)
 165 {
 166         read_descriptor_t desc;
 167
 168         if (!count)
 169                 return 0;
 170
 171         desc.written = 0;
 172         desc.count = count;
 173         desc.arg.data = target;
 174         desc.error = 0;
 175
 176         do_xip_mapping_read(in_file->f_mapping, &in_file->f_ra, in_file,
 177                             ppos, &desc, actor);
 178         if (desc.written)
 179                 return desc.written;
 180         return desc.error;
 181 }
 182 EXPORT_SYMBOL_GPL(xip_file_sendfile);
 183
 184 /*
 185  * __xip_unmap is invoked from xip_unmap and
 186  * xip_write
 187  *
 188  * This function walks all vmas of the address_space and unmaps the
 189  * __xip_sparse_page when found at pgoff.
 190  */
 191 static void
 192 __xip_unmap (struct address_space * mapping,
 193                      unsigned long pgoff)
 194 {
 195         struct vm_area_struct *vma;
 196         struct mm_struct *mm;
 197         struct prio_tree_iter iter;
 198         unsigned long address;
 199         pte_t *pte;
 200         pte_t pteval;
 201         spinlock_t *ptl;
 202         struct page *page;
 203
 204         page = __xip_sparse_page;
 205         if (!page)
 206                 return;
 207
 208         spin_lock(&mapping->i_mmap_lock);
 209         vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
 210                 mm = vma->vm_mm;
 211                 address = vma->vm_start +
 212                         ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
 213                 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
 214                 pte = page_check_address(page, mm, address, &ptl);
 215                 if (pte) {
 216                         /* Nuke the page table entry. */
 217                         flush_cache_page(vma, address, pte_pfn(*pte));
 218                         pteval = ptep_clear_flush(vma, address, pte);
 219                         page_remove_rmap(page, vma);
 220                         dec_mm_counter(mm, file_rss);
 221                         BUG_ON(pte_dirty(pteval));
 222                         pte_unmap_unlock(pte, ptl);
 223                         page_cache_release(page);
 224                 }
 225         }
 226         spin_unlock(&mapping->i_mmap_lock);
 227 }
 228
 229 /*
 230  * xip_nopage() is invoked via the vma operations vector for a
 231  * mapped memory region to read in file data during a page fault.
 232  *
 233  * This function is derived from filemap_nopage, but used for execute in place
 234  */
 235 static struct page *
 236 xip_file_nopage(struct vm_area_struct * area,
 237                    unsigned long address,
 238                    int *type)
 239 {
 240         struct file *file = area->vm_file;
 241         struct address_space *mapping = file->f_mapping;
 242         struct inode *inode = mapping->host;
 243         struct page *page;
 244         unsigned long size, pgoff, endoff;
 245
 246         pgoff = ((address - area->vm_start) >> PAGE_CACHE_SHIFT)
 247                 + area->vm_pgoff;
 248         endoff = ((area->vm_end - area->vm_start) >> PAGE_CACHE_SHIFT)
 249                 + area->vm_pgoff;
 250
 251         size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 252         if (pgoff >= size)
 253                 return NOPAGE_SIGBUS;
 254
 255         page = mapping->a_ops->get_xip_page(mapping, pgoff*(PAGE_SIZE/512), 0);
 256         if (!IS_ERR(page))
 257                 goto out;
 258         if (PTR_ERR(page) != -ENODATA)
 259                 return NOPAGE_SIGBUS;
 260
 261         /* sparse block */
 262         if ((area->vm_flags & (VM_WRITE | VM_MAYWRITE)) &&
 263             (area->vm_flags & (VM_SHARED| VM_MAYSHARE)) &&
 264             (!(mapping->host->i_sb->s_flags & MS_RDONLY))) {
 265                 /* maybe shared writable, allocate new block */
 266                 page = mapping->a_ops->get_xip_page (mapping,
 267                         pgoff*(PAGE_SIZE/512), 1);
 268                 if (IS_ERR(page))
 269                         return NOPAGE_SIGBUS;
 270                 /* unmap page at pgoff from all other vmas */
 271                 __xip_unmap(mapping, pgoff);
 272         } else {
 273                 /* not shared and writable, use xip_sparse_page() */
 274                 page = xip_sparse_page();
 275                 if (!page)
 276                         return NOPAGE_OOM;
 277         }
 278
 279 out:
 280         page_cache_get(page);
 281         return page;
 282 }
 283
 284 static struct vm_operations_struct xip_file_vm_ops = {
 285         .nopage         = xip_file_nopage,
 286 };
 287
 288 int xip_file_mmap(struct file * file, struct vm_area_struct * vma)
 289 {
 290         BUG_ON(!file->f_mapping->a_ops->get_xip_page);
 291
 292         file_accessed(file);
 293         vma->vm_ops = &xip_file_vm_ops;
 294         return 0;
 295 }
 296 EXPORT_SYMBOL_GPL(xip_file_mmap);
 297
 298 static ssize_t
 299 __xip_file_write(struct file *filp, const char __user *buf,
 300                   size_t count, loff_t pos, loff_t *ppos)
 301 {
 302         struct address_space * mapping = filp->f_mapping;
 303         const struct address_space_operations *a_ops = mapping->a_ops;
 304         struct inode    *inode = mapping->host;
 305         long            status = 0;
 306         struct page     *page;
 307         size_t          bytes;
 308         ssize_t         written = 0;
 309
 310         BUG_ON(!mapping->a_ops->get_xip_page);
 311
 312         do {
 313                 unsigned long index;
 314                 unsigned long offset;
 315                 size_t copied;
 316
 317                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
 318                 index = pos >> PAGE_CACHE_SHIFT;
 319                 bytes = PAGE_CACHE_SIZE - offset;
 320                 if (bytes > count)
 321                         bytes = count;
 322
 323                 /*
 324                  * Bring in the user page that we will copy from _first_.
 325                  * Otherwise there's a nasty deadlock on copying from the
 326                  * same page as we're writing to, without it being marked
 327                  * up-to-date.
 328                  */
 329                 fault_in_pages_readable(buf, bytes);
 330
 331                 page = a_ops->get_xip_page(mapping,
 332                                            index*(PAGE_SIZE/512), 0);
 333                 if (IS_ERR(page) && (PTR_ERR(page) == -ENODATA)) {
 334                         /* we allocate a new page unmap it */
 335                         page = a_ops->get_xip_page(mapping,
 336                                                    index*(PAGE_SIZE/512), 1);
 337                         if (!IS_ERR(page))
 338                                 /* unmap page at pgoff from all other vmas */
 339                                 __xip_unmap(mapping, index);
 340                 }
 341
 342                 if (IS_ERR(page)) {
 343                         status = PTR_ERR(page);
 344                         break;
 345                 }
 346
 347                 copied = filemap_copy_from_user(page, offset, buf, bytes);
 348                 flush_dcache_page(page);
 349                 if (likely(copied > 0)) {
 350                         status = copied;
 351
 352                         if (status >= 0) {
 353                                 written += status;
 354                                 count -= status;
 355                                 pos += status;
 356                                 buf += status;
 357                         }
 358                 }
 359                 if (unlikely(copied != bytes))
 360                         if (status >= 0)
 361                                 status = -EFAULT;
 362                 if (status < 0)
 363                         break;
 364         } while (count);
 365         *ppos = pos;
 366         /*
 367          * No need to use i_size_read() here, the i_size
 368          * cannot change under us because we hold i_mutex.
 369          */
 370         if (pos > inode->i_size) {
 371                 i_size_write(inode, pos);
 372                 mark_inode_dirty(inode);
 373         }
 374
 375         return written ? written : status;
 376 }
 377
 378 ssize_t
 379 xip_file_write(struct file *filp, const char __user *buf, size_t len,
 380                loff_t *ppos)
 381 {
 382         struct address_space *mapping = filp->f_mapping;
 383         struct inode *inode = mapping->host;
 384         size_t count;
 385         loff_t pos;
 386         ssize_t ret;
 387
 388         mutex_lock(&inode->i_mutex);
 389
 390         if (!access_ok(VERIFY_READ, buf, len)) {
 391                 ret=-EFAULT;
 392                 goto out_up;
 393         }
 394
 395         pos = *ppos;
 396         count = len;
 397
 398         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 399
 400         /* We can write back this queue in page reclaim */
 401         current->backing_dev_info = mapping->backing_dev_info;
 402
 403         ret = generic_write_checks(filp, &pos, &count, S_ISBLK(inode->i_mode));
 404         if (ret)
 405                 goto out_backing;
 406         if (count == 0)
 407                 goto out_backing;
 408
 409         ret = remove_suid(filp->f_path.dentry);
 410         if (ret)
 411                 goto out_backing;
 412
 413         file_update_time(filp);
 414
 415         ret = __xip_file_write (filp, buf, count, pos, ppos);
 416
 417  out_backing:
 418         current->backing_dev_info = NULL;
 419  out_up:
 420         mutex_unlock(&inode->i_mutex);
 421         return ret;
 422 }
 423 EXPORT_SYMBOL_GPL(xip_file_write);
 424
 425 /*
 426  * truncate a page used for execute in place
 427  * functionality is analog to block_truncate_page but does use get_xip_page
 428  * to get the page instead of page cache
 429  */
 430 int
 431 xip_truncate_page(struct address_space *mapping, loff_t from)
 432 {
 433         pgoff_t index = from >> PAGE_CACHE_SHIFT;
 434         unsigned offset = from & (PAGE_CACHE_SIZE-1);
 435         unsigned blocksize;
 436         unsigned length;
 437         struct page *page;
 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         zero_user_page(page, offset, length, KM_USER0);
 462         return 0;
 463 }
 464 EXPORT_SYMBOL_GPL(xip_truncate_page);