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 <asm/io.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                 struct page *page = alloc_page(GFP_HIGHUSER | __GFP_ZERO);
  30
  31                 if (page) {
  32                         static DEFINE_SPINLOCK(xip_alloc_lock);
  33                         spin_lock(&xip_alloc_lock);
  34                         if (!__xip_sparse_page)
  35                                 __xip_sparse_page = page;
  36                         else
  37                                 __free_page(page);
  38                         spin_unlock(&xip_alloc_lock);
  39                 }
  40         }
  41         return __xip_sparse_page;
  42 }
  43
  44 /*
  45  * This is a file read routine for execute in place files, and uses
  46  * the mapping->a_ops->get_xip_mem() function for the actual low-level
  47  * stuff.
  48  *
  49  * Note the struct file* is not used at all.  It may be NULL.
  50  */
  51 static ssize_t
  52 do_xip_mapping_read(struct address_space *mapping,
  53                     struct file_ra_state *_ra,
  54                     struct file *filp,
  55                     char __user *buf,
  56                     size_t len,
  57                     loff_t *ppos)
  58 {
  59         struct inode *inode = mapping->host;
  60         pgoff_t index, end_index;
  61         unsigned long offset;
  62         loff_t isize, pos;
  63         size_t copied = 0, error = 0;
  64
  65         BUG_ON(!mapping->a_ops->get_xip_mem);
  66
  67         pos = *ppos;
  68         index = pos >> PAGE_CACHE_SHIFT;
  69         offset = pos & ~PAGE_CACHE_MASK;
  70
  71         isize = i_size_read(inode);
  72         if (!isize)
  73                 goto out;
  74
  75         end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
  76         do {
  77                 unsigned long nr, left;
  78                 void *xip_mem;
  79                 unsigned long xip_pfn;
  80                 int zero = 0;
  81
  82                 /* nr is the maximum number of bytes to copy from this page */
  83                 nr = PAGE_CACHE_SIZE;
  84                 if (index >= end_index) {
  85                         if (index > end_index)
  86                                 goto out;
  87                         nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
  88                         if (nr <= offset) {
  89                                 goto out;
  90                         }
  91                 }
  92                 nr = nr - offset;
  93                 if (nr > len)
  94                         nr = len;
  95
  96                 error = mapping->a_ops->get_xip_mem(mapping, index, 0,
  97                                                         &xip_mem, &xip_pfn);
  98                 if (unlikely(error)) {
  99                         if (error == -ENODATA) {
 100                                 /* sparse */
 101                                 zero = 1;
 102                         } else
 103                                 goto out;
 104                 }
 105
 106                 /* If users can be writing to this page using arbitrary
 107                  * virtual addresses, take care about potential aliasing
 108                  * before reading the page on the kernel side.
 109                  */
 110                 if (mapping_writably_mapped(mapping))
 111                         /* address based flush */ ;
 112
 113                 /*
 114                  * Ok, we have the mem, so now we can copy it to user space...
 115                  *
 116                  * The actor routine returns how many bytes were actually used..
 117                  * NOTE! This may not be the same as how much of a user buffer
 118                  * we filled up (we may be padding etc), so we can only update
 119                  * "pos" here (the actor routine has to update the user buffer
 120                  * pointers and the remaining count).
 121                  */
 122                 if (!zero)
 123                         left = __copy_to_user(buf+copied, xip_mem+offset, nr);
 124                 else
 125                         left = __clear_user(buf + copied, nr);
 126
 127                 if (left) {
 128                         error = -EFAULT;
 129                         goto out;
 130                 }
 131
 132                 copied += (nr - left);
 133                 offset += (nr - left);
 134                 index += offset >> PAGE_CACHE_SHIFT;
 135                 offset &= ~PAGE_CACHE_MASK;
 136         } while (copied < len);
 137
 138 out:
 139         *ppos = pos + copied;
 140         if (filp)
 141                 file_accessed(filp);
 142
 143         return (copied ? copied : error);
 144 }
 145
 146 ssize_t
 147 xip_file_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
 148 {
 149         if (!access_ok(VERIFY_WRITE, buf, len))
 150                 return -EFAULT;
 151
 152         return do_xip_mapping_read(filp->f_mapping, &filp->f_ra, filp,
 153                             buf, len, ppos);
 154 }
 155 EXPORT_SYMBOL_GPL(xip_file_read);
 156
 157 /*
 158  * __xip_unmap is invoked from xip_unmap and
 159  * xip_write
 160  *
 161  * This function walks all vmas of the address_space and unmaps the
 162  * __xip_sparse_page when found at pgoff.
 163  */
 164 static void
 165 __xip_unmap (struct address_space * mapping,
 166                      unsigned long pgoff)
 167 {
 168         struct vm_area_struct *vma;
 169         struct mm_struct *mm;
 170         struct prio_tree_iter iter;
 171         unsigned long address;
 172         pte_t *pte;
 173         pte_t pteval;
 174         spinlock_t *ptl;
 175         struct page *page;
 176
 177         page = __xip_sparse_page;
 178         if (!page)
 179                 return;
 180
 181         spin_lock(&mapping->i_mmap_lock);
 182         vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
 183                 mm = vma->vm_mm;
 184                 address = vma->vm_start +
 185                         ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
 186                 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
 187                 pte = page_check_address(page, mm, address, &ptl, 1);
 188                 if (pte) {
 189                         /* Nuke the page table entry. */
 190                         flush_cache_page(vma, address, pte_pfn(*pte));
 191                         pteval = ptep_clear_flush(vma, address, pte);
 192                         page_remove_rmap(page, vma);
 193                         dec_mm_counter(mm, file_rss);
 194                         BUG_ON(pte_dirty(pteval));
 195                         pte_unmap_unlock(pte, ptl);
 196                         page_cache_release(page);
 197                 }
 198         }
 199         spin_unlock(&mapping->i_mmap_lock);
 200 }
 201
 202 /*
 203  * xip_fault() is invoked via the vma operations vector for a
 204  * mapped memory region to read in file data during a page fault.
 205  *
 206  * This function is derived from filemap_fault, but used for execute in place
 207  */
 208 static int xip_file_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 209 {
 210         struct file *file = vma->vm_file;
 211         struct address_space *mapping = file->f_mapping;
 212         struct inode *inode = mapping->host;
 213         pgoff_t size;
 214         void *xip_mem;
 215         unsigned long xip_pfn;
 216         struct page *page;
 217         int error;
 218
 219         /* XXX: are VM_FAULT_ codes OK? */
 220
 221         size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 222         if (vmf->pgoff >= size)
 223                 return VM_FAULT_SIGBUS;
 224
 225         error = mapping->a_ops->get_xip_mem(mapping, vmf->pgoff, 0,
 226                                                 &xip_mem, &xip_pfn);
 227         if (likely(!error))
 228                 goto found;
 229         if (error != -ENODATA)
 230                 return VM_FAULT_OOM;
 231
 232         /* sparse block */
 233         if ((vma->vm_flags & (VM_WRITE | VM_MAYWRITE)) &&
 234             (vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) &&
 235             (!(mapping->host->i_sb->s_flags & MS_RDONLY))) {
 236                 int err;
 237
 238                 /* maybe shared writable, allocate new block */
 239                 error = mapping->a_ops->get_xip_mem(mapping, vmf->pgoff, 1,
 240                                                         &xip_mem, &xip_pfn);
 241                 if (error)
 242                         return VM_FAULT_SIGBUS;
 243                 /* unmap sparse mappings at pgoff from all other vmas */
 244                 __xip_unmap(mapping, vmf->pgoff);
 245
 246 found:
 247                 err = vm_insert_mixed(vma, (unsigned long)vmf->virtual_address,
 248                                                         xip_pfn);
 249                 if (err == -ENOMEM)
 250                         return VM_FAULT_OOM;
 251                 BUG_ON(err);
 252                 return VM_FAULT_NOPAGE;
 253         } else {
 254                 /* not shared and writable, use xip_sparse_page() */
 255                 page = xip_sparse_page();
 256                 if (!page)
 257                         return VM_FAULT_OOM;
 258
 259                 page_cache_get(page);
 260                 vmf->page = page;
 261                 return 0;
 262         }
 263 }
 264
 265 static struct vm_operations_struct xip_file_vm_ops = {
 266         .fault  = xip_file_fault,
 267 };
 268
 269 int xip_file_mmap(struct file * file, struct vm_area_struct * vma)
 270 {
 271         BUG_ON(!file->f_mapping->a_ops->get_xip_mem);
 272
 273         file_accessed(file);
 274         vma->vm_ops = &xip_file_vm_ops;
 275         vma->vm_flags |= VM_CAN_NONLINEAR | VM_MIXEDMAP;
 276         return 0;
 277 }
 278 EXPORT_SYMBOL_GPL(xip_file_mmap);
 279
 280 static ssize_t
 281 __xip_file_write(struct file *filp, const char __user *buf,
 282                   size_t count, loff_t pos, loff_t *ppos)
 283 {
 284         struct address_space * mapping = filp->f_mapping;
 285         const struct address_space_operations *a_ops = mapping->a_ops;
 286         struct inode    *inode = mapping->host;
 287         long            status = 0;
 288         size_t          bytes;
 289         ssize_t         written = 0;
 290
 291         BUG_ON(!mapping->a_ops->get_xip_mem);
 292
 293         do {
 294                 unsigned long index;
 295                 unsigned long offset;
 296                 size_t copied;
 297                 void *xip_mem;
 298                 unsigned long xip_pfn;
 299
 300                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
 301                 index = pos >> PAGE_CACHE_SHIFT;
 302                 bytes = PAGE_CACHE_SIZE - offset;
 303                 if (bytes > count)
 304                         bytes = count;
 305
 306                 status = a_ops->get_xip_mem(mapping, index, 0,
 307                                                 &xip_mem, &xip_pfn);
 308                 if (status == -ENODATA) {
 309                         /* we allocate a new page unmap it */
 310                         status = a_ops->get_xip_mem(mapping, index, 1,
 311                                                         &xip_mem, &xip_pfn);
 312                         if (!status)
 313                                 /* unmap page at pgoff from all other vmas */
 314                                 __xip_unmap(mapping, index);
 315                 }
 316
 317                 if (status)
 318                         break;
 319
 320                 copied = bytes -
 321                         __copy_from_user_nocache(xip_mem + offset, buf, bytes);
 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_mem
 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         void *xip_mem;
 412         unsigned long xip_pfn;
 413         int err;
 414
 415         BUG_ON(!mapping->a_ops->get_xip_mem);
 416
 417         blocksize = 1 << mapping->host->i_blkbits;
 418         length = offset & (blocksize - 1);
 419
 420         /* Block boundary? Nothing to do */
 421         if (!length)
 422                 return 0;
 423
 424         length = blocksize - length;
 425
 426         err = mapping->a_ops->get_xip_mem(mapping, index, 0,
 427                                                 &xip_mem, &xip_pfn);
 428         if (unlikely(err)) {
 429                 if (err == -ENODATA)
 430                         /* Hole? No need to truncate */
 431                         return 0;
 432                 else
 433                         return err;
 434         }
 435         memset(xip_mem + offset, 0, length);
 436         return 0;
 437 }
 438 EXPORT_SYMBOL_GPL(xip_truncate_page);