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