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 <asm/tlbflush.h>
  17 #include "filemap.h"
  18
  19 /*
  20  * This is a file read routine for execute in place files, and uses
  21  * the mapping->a_ops->get_xip_page() function for the actual low-level
  22  * stuff.
  23  *
  24  * Note the struct file* is not used at all.  It may be NULL.
  25  */
  26 static void
  27 do_xip_mapping_read(struct address_space *mapping,
  28                     struct file_ra_state *_ra,
  29                     struct file *filp,
  30                     loff_t *ppos,
  31                     read_descriptor_t *desc,
  32                     read_actor_t actor)
  33 {
  34         struct inode *inode = mapping->host;
  35         unsigned long index, end_index, offset;
  36         loff_t isize;
  37
  38         BUG_ON(!mapping->a_ops->get_xip_page);
  39
  40         index = *ppos >> PAGE_CACHE_SHIFT;
  41         offset = *ppos & ~PAGE_CACHE_MASK;
  42
  43         isize = i_size_read(inode);
  44         if (!isize)
  45                 goto out;
  46
  47         end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
  48         for (;;) {
  49                 struct page *page;
  50                 unsigned long nr, ret;
  51
  52                 /* nr is the maximum number of bytes to copy from this page */
  53                 nr = PAGE_CACHE_SIZE;
  54                 if (index >= end_index) {
  55                         if (index > end_index)
  56                                 goto out;
  57                         nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
  58                         if (nr <= offset) {
  59                                 goto out;
  60                         }
  61                 }
  62                 nr = nr - offset;
  63
  64                 page = mapping->a_ops->get_xip_page(mapping,
  65                         index*(PAGE_SIZE/512), 0);
  66                 if (!page)
  67                         goto no_xip_page;
  68                 if (unlikely(IS_ERR(page))) {
  69                         if (PTR_ERR(page) == -ENODATA) {
  70                                 /* sparse */
  71                                 page = ZERO_PAGE(0);
  72                         } else {
  73                                 desc->error = PTR_ERR(page);
  74                                 goto out;
  75                         }
  76                 }
  77
  78                 /* If users can be writing to this page using arbitrary
  79                  * virtual addresses, take care about potential aliasing
  80                  * before reading the page on the kernel side.
  81                  */
  82                 if (mapping_writably_mapped(mapping))
  83                         flush_dcache_page(page);
  84
  85                 /*
  86                  * Ok, we have the page, so now we can copy it to user space...
  87                  *
  88                  * The actor routine returns how many bytes were actually used..
  89                  * NOTE! This may not be the same as how much of a user buffer
  90                  * we filled up (we may be padding etc), so we can only update
  91                  * "pos" here (the actor routine has to update the user buffer
  92                  * pointers and the remaining count).
  93                  */
  94                 ret = actor(desc, page, offset, nr);
  95                 offset += ret;
  96                 index += offset >> PAGE_CACHE_SHIFT;
  97                 offset &= ~PAGE_CACHE_MASK;
  98
  99                 if (ret == nr && desc->count)
 100                         continue;
 101                 goto out;
 102
 103 no_xip_page:
 104                 /* Did not get the page. Report it */
 105                 desc->error = -EIO;
 106                 goto out;
 107         }
 108
 109 out:
 110         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
 111         if (filp)
 112                 file_accessed(filp);
 113 }
 114
 115 ssize_t
 116 xip_file_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
 117 {
 118         read_descriptor_t desc;
 119
 120         if (!access_ok(VERIFY_WRITE, buf, len))
 121                 return -EFAULT;
 122
 123         desc.written = 0;
 124         desc.arg.buf = buf;
 125         desc.count = len;
 126         desc.error = 0;
 127
 128         do_xip_mapping_read(filp->f_mapping, &filp->f_ra, filp,
 129                             ppos, &desc, file_read_actor);
 130
 131         if (desc.written)
 132                 return desc.written;
 133         else
 134                 return desc.error;
 135 }
 136 EXPORT_SYMBOL_GPL(xip_file_read);
 137
 138 ssize_t
 139 xip_file_sendfile(struct file *in_file, loff_t *ppos,
 140              size_t count, read_actor_t actor, void *target)
 141 {
 142         read_descriptor_t desc;
 143
 144         if (!count)
 145                 return 0;
 146
 147         desc.written = 0;
 148         desc.count = count;
 149         desc.arg.data = target;
 150         desc.error = 0;
 151
 152         do_xip_mapping_read(in_file->f_mapping, &in_file->f_ra, in_file,
 153                             ppos, &desc, actor);
 154         if (desc.written)
 155                 return desc.written;
 156         return desc.error;
 157 }
 158 EXPORT_SYMBOL_GPL(xip_file_sendfile);
 159
 160 /*
 161  * __xip_unmap is invoked from xip_unmap and
 162  * xip_write
 163  *
 164  * This function walks all vmas of the address_space and unmaps the
 165  * ZERO_PAGE when found at pgoff. Should it go in rmap.c?
 166  */
 167 static void
 168 __xip_unmap (struct address_space * mapping,
 169                      unsigned long pgoff)
 170 {
 171         struct vm_area_struct *vma;
 172         struct mm_struct *mm;
 173         struct prio_tree_iter iter;
 174         unsigned long address;
 175         pte_t *pte;
 176         pte_t pteval;
 177         spinlock_t *ptl;
 178         struct page *page;
 179
 180         spin_lock(&mapping->i_mmap_lock);
 181         vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
 182                 mm = vma->vm_mm;
 183                 address = vma->vm_start +
 184                         ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
 185                 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
 186                 page = ZERO_PAGE(address);
 187                 pte = page_check_address(page, mm, address, &ptl);
 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);
 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_nopage() 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_nopage, but used for execute in place
 207  */
 208 static struct page *
 209 xip_file_nopage(struct vm_area_struct * area,
 210                    unsigned long address,
 211                    int *type)
 212 {
 213         struct file *file = area->vm_file;
 214         struct address_space *mapping = file->f_mapping;
 215         struct inode *inode = mapping->host;
 216         struct page *page;
 217         unsigned long size, pgoff, endoff;
 218
 219         pgoff = ((address - area->vm_start) >> PAGE_CACHE_SHIFT)
 220                 + area->vm_pgoff;
 221         endoff = ((area->vm_end - area->vm_start) >> PAGE_CACHE_SHIFT)
 222                 + area->vm_pgoff;
 223
 224         size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 225         if (pgoff >= size) {
 226                 return NULL;
 227         }
 228
 229         page = mapping->a_ops->get_xip_page(mapping, pgoff*(PAGE_SIZE/512), 0);
 230         if (!IS_ERR(page)) {
 231                 goto out;
 232         }
 233         if (PTR_ERR(page) != -ENODATA)
 234                 return NULL;
 235
 236         /* sparse block */
 237         if ((area->vm_flags & (VM_WRITE | VM_MAYWRITE)) &&
 238             (area->vm_flags & (VM_SHARED| VM_MAYSHARE)) &&
 239             (!(mapping->host->i_sb->s_flags & MS_RDONLY))) {
 240                 /* maybe shared writable, allocate new block */
 241                 page = mapping->a_ops->get_xip_page (mapping,
 242                         pgoff*(PAGE_SIZE/512), 1);
 243                 if (IS_ERR(page))
 244                         return NULL;
 245                 /* unmap page at pgoff from all other vmas */
 246                 __xip_unmap(mapping, pgoff);
 247         } else {
 248                 /* not shared and writable, use ZERO_PAGE() */
 249                 page = ZERO_PAGE(address);
 250         }
 251
 252 out:
 253         page_cache_get(page);
 254         return page;
 255 }
 256
 257 static struct vm_operations_struct xip_file_vm_ops = {
 258         .nopage         = xip_file_nopage,
 259 };
 260
 261 int xip_file_mmap(struct file * file, struct vm_area_struct * vma)
 262 {
 263         BUG_ON(!file->f_mapping->a_ops->get_xip_page);
 264
 265         file_accessed(file);
 266         vma->vm_ops = &xip_file_vm_ops;
 267         return 0;
 268 }
 269 EXPORT_SYMBOL_GPL(xip_file_mmap);
 270
 271 static ssize_t
 272 __xip_file_write(struct file *filp, const char __user *buf,
 273                   size_t count, loff_t pos, loff_t *ppos)
 274 {
 275         struct address_space * mapping = filp->f_mapping;
 276         struct address_space_operations *a_ops = mapping->a_ops;
 277         struct inode    *inode = mapping->host;
 278         long            status = 0;
 279         struct page     *page;
 280         size_t          bytes;
 281         ssize_t         written = 0;
 282
 283         BUG_ON(!mapping->a_ops->get_xip_page);
 284
 285         do {
 286                 unsigned long index;
 287                 unsigned long offset;
 288                 size_t copied;
 289
 290                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
 291                 index = pos >> PAGE_CACHE_SHIFT;
 292                 bytes = PAGE_CACHE_SIZE - offset;
 293                 if (bytes > count)
 294                         bytes = count;
 295
 296                 /*
 297                  * Bring in the user page that we will copy from _first_.
 298                  * Otherwise there's a nasty deadlock on copying from the
 299                  * same page as we're writing to, without it being marked
 300                  * up-to-date.
 301                  */
 302                 fault_in_pages_readable(buf, bytes);
 303
 304                 page = a_ops->get_xip_page(mapping,
 305                                            index*(PAGE_SIZE/512), 0);
 306                 if (IS_ERR(page) && (PTR_ERR(page) == -ENODATA)) {
 307                         /* we allocate a new page unmap it */
 308                         page = a_ops->get_xip_page(mapping,
 309                                                    index*(PAGE_SIZE/512), 1);
 310                         if (!IS_ERR(page))
 311                                 /* unmap page at pgoff from all other vmas */
 312                                 __xip_unmap(mapping, index);
 313                 }
 314
 315                 if (IS_ERR(page)) {
 316                         status = PTR_ERR(page);
 317                         break;
 318                 }
 319
 320                 copied = filemap_copy_from_user(page, offset, buf, bytes);
 321                 flush_dcache_page(page);
 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_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         void *kaddr;
 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         kaddr = kmap_atomic(page, KM_USER0);
 436         memset(kaddr + offset, 0, length);
 437         kunmap_atomic(kaddr, KM_USER0);
 438
 439         flush_dcache_page(page);
 440         return 0;
 441 }
 442 EXPORT_SYMBOL_GPL(xip_truncate_page);