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/mmu_notifier.h>
  17 #include <linux/sched.h>
  18 #include <asm/tlbflush.h>
  19 #include <asm/io.h>
  20
  21 /*
  22  * We do use our own empty page to avoid interference with other users
  23  * of ZERO_PAGE(), such as /dev/zero
  24  */
  25 static struct page *__xip_sparse_page;
  26
  27 static struct page *xip_sparse_page(void)
  28 {
  29         if (!__xip_sparse_page) {
  30                 struct page *page = alloc_page(GFP_HIGHUSER | __GFP_ZERO);
  31
  32                 if (page) {
  33                         static DEFINE_SPINLOCK(xip_alloc_lock);
  34                         spin_lock(&xip_alloc_lock);
  35                         if (!__xip_sparse_page)
  36                                 __xip_sparse_page = page;
  37                         else
  38                                 __free_page(page);
  39                         spin_unlock(&xip_alloc_lock);
  40                 }
  41         }
  42         return __xip_sparse_page;
  43 }
  44
  45 /*
  46  * This is a file read routine for execute in place files, and uses
  47  * the mapping->a_ops->get_xip_mem() function for the actual low-level
  48  * stuff.
  49  *
  50  * Note the struct file* is not used at all.  It may be NULL.
  51  */
  52 static ssize_t
  53 do_xip_mapping_read(struct address_space *mapping,
  54                     struct file_ra_state *_ra,
  55                     struct file *filp,
  56                     char __user *buf,
  57                     size_t len,
  58                     loff_t *ppos)
  59 {
  60         struct inode *inode = mapping->host;
  61         pgoff_t index, end_index;
  62         unsigned long offset;
  63         loff_t isize, pos;
  64         size_t copied = 0, error = 0;
  65
  66         BUG_ON(!mapping->a_ops->get_xip_mem);
  67
  68         pos = *ppos;
  69         index = pos >> PAGE_CACHE_SHIFT;
  70         offset = pos & ~PAGE_CACHE_MASK;
  71
  72         isize = i_size_read(inode);
  73         if (!isize)
  74                 goto out;
  75
  76         end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
  77         do {
  78                 unsigned long nr, left;
  79                 void *xip_mem;
  80                 unsigned long xip_pfn;
  81                 int zero = 0;
  82
  83                 /* nr is the maximum number of bytes to copy from this page */
  84                 nr = PAGE_CACHE_SIZE;
  85                 if (index >= end_index) {
  86                         if (index > end_index)
  87                                 goto out;
  88                         nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
  89                         if (nr <= offset) {
  90                                 goto out;
  91                         }
  92                 }
  93                 nr = nr - offset;
  94                 if (nr > len)
  95                         nr = len;
  96
  97                 error = mapping->a_ops->get_xip_mem(mapping, index, 0,
  98                                                         &xip_mem, &xip_pfn);
  99                 if (unlikely(error)) {
 100                         if (error == -ENODATA) {
 101                                 /* sparse */
 102                                 zero = 1;
 103                         } else
 104                                 goto out;
 105                 }
 106
 107                 /* If users can be writing to this page using arbitrary
 108                  * virtual addresses, take care about potential aliasing
 109                  * before reading the page on the kernel side.
 110                  */
 111                 if (mapping_writably_mapped(mapping))
 112                         /* address based flush */ ;
 113
 114                 /*
 115                  * Ok, we have the mem, so now we can copy it to user space...
 116                  *
 117                  * The actor routine returns how many bytes were actually used..
 118                  * NOTE! This may not be the same as how much of a user buffer
 119                  * we filled up (we may be padding etc), so we can only update
 120                  * "pos" here (the actor routine has to update the user buffer
 121                  * pointers and the remaining count).
 122                  */
 123                 if (!zero)
 124                         left = __copy_to_user(buf+copied, xip_mem+offset, nr);
 125                 else
 126                         left = __clear_user(buf + copied, nr);
 127
 128                 if (left) {
 129                         error = -EFAULT;
 130                         goto out;
 131                 }
 132
 133                 copied += (nr - left);
 134                 offset += (nr - left);
 135                 index += offset >> PAGE_CACHE_SHIFT;
 136                 offset &= ~PAGE_CACHE_MASK;
 137         } while (copied < len);
 138
 139 out:
 140         *ppos = pos + copied;
 141         if (filp)
 142                 file_accessed(filp);
 143
 144         return (copied ? copied : error);
 145 }
 146
 147 ssize_t
 148 xip_file_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
 149 {
 150         if (!access_ok(VERIFY_WRITE, buf, len))
 151                 return -EFAULT;
 152
 153         return do_xip_mapping_read(filp->f_mapping, &filp->f_ra, filp,
 154                             buf, len, ppos);
 155 }
 156 EXPORT_SYMBOL_GPL(xip_file_read);
 157
 158 /*
 159  * __xip_unmap is invoked from xip_unmap and
 160  * xip_write
 161  *
 162  * This function walks all vmas of the address_space and unmaps the
 163  * __xip_sparse_page when found at pgoff.
 164  */
 165 static void
 166 __xip_unmap (struct address_space * mapping,
 167                      unsigned long pgoff)
 168 {
 169         struct vm_area_struct *vma;
 170         struct mm_struct *mm;
 171         struct prio_tree_iter iter;
 172         unsigned long address;
 173         pte_t *pte;
 174         pte_t pteval;
 175         spinlock_t *ptl;
 176         struct page *page;
 177
 178         page = __xip_sparse_page;
 179         if (!page)
 180                 return;
 181
 182         spin_lock(&mapping->i_mmap_lock);
 183         vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
 184                 mm = vma->vm_mm;
 185                 address = vma->vm_start +
 186                         ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
 187                 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
 188                 pte = page_check_address(page, mm, address, &ptl);
 189                 if (pte) {
 190                         /* Nuke the page table entry. */
 191                         flush_cache_page(vma, address, pte_pfn(*pte));
 192                         pteval = ptep_clear_flush_notify(vma, address, pte);
 193                         page_remove_rmap(page, vma);
 194                         dec_mm_counter(mm, file_rss);
 195                         BUG_ON(pte_dirty(pteval));
 196                         pte_unmap_unlock(pte, ptl);
 197                         page_cache_release(page);
 198                 }
 199         }
 200         spin_unlock(&mapping->i_mmap_lock);
 201 }
 202
 203 /*
 204  * xip_fault() is invoked via the vma operations vector for a
 205  * mapped memory region to read in file data during a page fault.
 206  *
 207  * This function is derived from filemap_fault, but used for execute in place
 208  */
 209 static int xip_file_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 210 {
 211         struct file *file = vma->vm_file;
 212         struct address_space *mapping = file->f_mapping;
 213         struct inode *inode = mapping->host;
 214         pgoff_t size;
 215         void *xip_mem;
 216         unsigned long xip_pfn;
 217         struct page *page;
 218         int error;
 219
 220         /* XXX: are VM_FAULT_ codes OK? */
 221
 222         size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 223         if (vmf->pgoff >= size)
 224                 return VM_FAULT_SIGBUS;
 225
 226         error = mapping->a_ops->get_xip_mem(mapping, vmf->pgoff, 0,
 227                                                 &xip_mem, &xip_pfn);
 228         if (likely(!error))
 229                 goto found;
 230         if (error != -ENODATA)
 231                 return VM_FAULT_OOM;
 232
 233         /* sparse block */
 234         if ((vma->vm_flags & (VM_WRITE | VM_MAYWRITE)) &&
 235             (vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) &&
 236             (!(mapping->host->i_sb->s_flags & MS_RDONLY))) {
 237                 int err;
 238
 239                 /* maybe shared writable, allocate new block */
 240                 error = mapping->a_ops->get_xip_mem(mapping, vmf->pgoff, 1,
 241                                                         &xip_mem, &xip_pfn);
 242                 if (error)
 243                         return VM_FAULT_SIGBUS;
 244                 /* unmap sparse mappings at pgoff from all other vmas */
 245                 __xip_unmap(mapping, vmf->pgoff);
 246
 247 found:
 248                 err = vm_insert_mixed(vma, (unsigned long)vmf->virtual_address,
 249                                                         xip_pfn);
 250                 if (err == -ENOMEM)
 251                         return VM_FAULT_OOM;
 252                 BUG_ON(err);
 253                 return VM_FAULT_NOPAGE;
 254         } else {
 255                 /* not shared and writable, use xip_sparse_page() */
 256                 page = xip_sparse_page();
 257                 if (!page)
 258                         return VM_FAULT_OOM;
 259
 260                 page_cache_get(page);
 261                 vmf->page = page;
 262                 return 0;
 263         }
 264 }
 265
 266 static struct vm_operations_struct xip_file_vm_ops = {
 267         .fault  = xip_file_fault,
 268 };
 269
 270 int xip_file_mmap(struct file * file, struct vm_area_struct * vma)
 271 {
 272         BUG_ON(!file->f_mapping->a_ops->get_xip_mem);
 273
 274         file_accessed(file);
 275         vma->vm_ops = &xip_file_vm_ops;
 276         vma->vm_flags |= VM_CAN_NONLINEAR | VM_MIXEDMAP;
 277         return 0;
 278 }
 279 EXPORT_SYMBOL_GPL(xip_file_mmap);
 280
 281 static ssize_t
 282 __xip_file_write(struct file *filp, const char __user *buf,
 283                   size_t count, loff_t pos, loff_t *ppos)
 284 {
 285         struct address_space * mapping = filp->f_mapping;
 286         const struct address_space_operations *a_ops = mapping->a_ops;
 287         struct inode    *inode = mapping->host;
 288         long            status = 0;
 289         size_t          bytes;
 290         ssize_t         written = 0;
 291
 292         BUG_ON(!mapping->a_ops->get_xip_mem);
 293
 294         do {
 295                 unsigned long index;
 296                 unsigned long offset;
 297                 size_t copied;
 298                 void *xip_mem;
 299                 unsigned long xip_pfn;
 300
 301                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
 302                 index = pos >> PAGE_CACHE_SHIFT;
 303                 bytes = PAGE_CACHE_SIZE - offset;
 304                 if (bytes > count)
 305                         bytes = count;
 306
 307                 status = a_ops->get_xip_mem(mapping, index, 0,
 308                                                 &xip_mem, &xip_pfn);
 309                 if (status == -ENODATA) {
 310                         /* we allocate a new page unmap it */
 311                         status = a_ops->get_xip_mem(mapping, index, 1,
 312                                                         &xip_mem, &xip_pfn);
 313                         if (!status)
 314                                 /* unmap page at pgoff from all other vmas */
 315                                 __xip_unmap(mapping, index);
 316                 }
 317
 318                 if (status)
 319                         break;
 320
 321                 copied = bytes -
 322                         __copy_from_user_nocache(xip_mem + offset, buf, bytes);
 323
 324                 if (likely(copied > 0)) {
 325                         status = copied;
 326
 327                         if (status >= 0) {
 328                                 written += status;
 329                                 count -= status;
 330                                 pos += status;
 331                                 buf += status;
 332                         }
 333                 }
 334                 if (unlikely(copied != bytes))
 335                         if (status >= 0)
 336                                 status = -EFAULT;
 337                 if (status < 0)
 338                         break;
 339         } while (count);
 340         *ppos = pos;
 341         /*
 342          * No need to use i_size_read() here, the i_size
 343          * cannot change under us because we hold i_mutex.
 344          */
 345         if (pos > inode->i_size) {
 346                 i_size_write(inode, pos);
 347                 mark_inode_dirty(inode);
 348         }
 349
 350         return written ? written : status;
 351 }
 352
 353 ssize_t
 354 xip_file_write(struct file *filp, const char __user *buf, size_t len,
 355                loff_t *ppos)
 356 {
 357         struct address_space *mapping = filp->f_mapping;
 358         struct inode *inode = mapping->host;
 359         size_t count;
 360         loff_t pos;
 361         ssize_t ret;
 362
 363         mutex_lock(&inode->i_mutex);
 364
 365         if (!access_ok(VERIFY_READ, buf, len)) {
 366                 ret=-EFAULT;
 367                 goto out_up;
 368         }
 369
 370         pos = *ppos;
 371         count = len;
 372
 373         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 374
 375         /* We can write back this queue in page reclaim */
 376         current->backing_dev_info = mapping->backing_dev_info;
 377
 378         ret = generic_write_checks(filp, &pos, &count, S_ISBLK(inode->i_mode));
 379         if (ret)
 380                 goto out_backing;
 381         if (count == 0)
 382                 goto out_backing;
 383
 384         ret = file_remove_suid(filp);
 385         if (ret)
 386                 goto out_backing;
 387
 388         file_update_time(filp);
 389
 390         ret = __xip_file_write (filp, buf, count, pos, ppos);
 391
 392  out_backing:
 393         current->backing_dev_info = NULL;
 394  out_up:
 395         mutex_unlock(&inode->i_mutex);
 396         return ret;
 397 }
 398 EXPORT_SYMBOL_GPL(xip_file_write);
 399
 400 /*
 401  * truncate a page used for execute in place
 402  * functionality is analog to block_truncate_page but does use get_xip_mem
 403  * to get the page instead of page cache
 404  */
 405 int
 406 xip_truncate_page(struct address_space *mapping, loff_t from)
 407 {
 408         pgoff_t index = from >> PAGE_CACHE_SHIFT;
 409         unsigned offset = from & (PAGE_CACHE_SIZE-1);
 410         unsigned blocksize;
 411         unsigned length;
 412         void *xip_mem;
 413         unsigned long xip_pfn;
 414         int err;
 415
 416         BUG_ON(!mapping->a_ops->get_xip_mem);
 417
 418         blocksize = 1 << mapping->host->i_blkbits;
 419         length = offset & (blocksize - 1);
 420
 421         /* Block boundary? Nothing to do */
 422         if (!length)
 423                 return 0;
 424
 425         length = blocksize - length;
 426
 427         err = mapping->a_ops->get_xip_mem(mapping, index, 0,
 428                                                 &xip_mem, &xip_pfn);
 429         if (unlikely(err)) {
 430                 if (err == -ENODATA)
 431                         /* Hole? No need to truncate */
 432                         return 0;
 433                 else
 434                         return err;
 435         }
 436         memset(xip_mem + offset, 0, length);
 437         return 0;
 438 }
 439 EXPORT_SYMBOL_GPL(xip_truncate_page);