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