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 <linux/seqlock.h>
  19 #include <linux/mutex.h>
  20 #include <linux/gfp.h>
  21 #include <asm/tlbflush.h>
  22 #include <asm/io.h>
  23
  24 /*
  25  * We do use our own empty page to avoid interference with other users
  26  * of ZERO_PAGE(), such as /dev/zero
  27  */
  28 static DEFINE_MUTEX(xip_sparse_mutex);
  29 static seqcount_t xip_sparse_seq = SEQCNT_ZERO;
  30 static struct page *__xip_sparse_page;
  31
  32 /* called under xip_sparse_mutex */
  33 static struct page *xip_sparse_page(void)
  34 {
  35         if (!__xip_sparse_page) {
  36                 struct page *page = alloc_page(GFP_HIGHUSER | __GFP_ZERO);
  37
  38                 if (page)
  39                         __xip_sparse_page = page;
  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 - copied)
  94                         nr = len - copied;
  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         unsigned count;
 177         int locked = 0;
 178
 179         count = read_seqcount_begin(&xip_sparse_seq);
 180
 181         page = __xip_sparse_page;
 182         if (!page)
 183                 return;
 184
 185 retry:
 186         mutex_lock(&mapping->i_mmap_mutex);
 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, 1);
 193                 if (pte) {
 194                         /* Nuke the page table entry. */
 195                         flush_cache_page(vma, address, pte_pfn(*pte));
 196                         pteval = ptep_clear_flush_notify(vma, address, pte);
 197                         page_remove_rmap(page);
 198                         dec_mm_counter(mm, MM_FILEPAGES);
 199                         BUG_ON(pte_dirty(pteval));
 200                         pte_unmap_unlock(pte, ptl);
 201                         page_cache_release(page);
 202                 }
 203         }
 204         mutex_unlock(&mapping->i_mmap_mutex);
 205
 206         if (locked) {
 207                 mutex_unlock(&xip_sparse_mutex);
 208         } else if (read_seqcount_retry(&xip_sparse_seq, count)) {
 209                 mutex_lock(&xip_sparse_mutex);
 210                 locked = 1;
 211                 goto retry;
 212         }
 213 }
 214
 215 /*
 216  * xip_fault() is invoked via the vma operations vector for a
 217  * mapped memory region to read in file data during a page fault.
 218  *
 219  * This function is derived from filemap_fault, but used for execute in place
 220  */
 221 static int xip_file_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 222 {
 223         struct file *file = vma->vm_file;
 224         struct address_space *mapping = file->f_mapping;
 225         struct inode *inode = mapping->host;
 226         pgoff_t size;
 227         void *xip_mem;
 228         unsigned long xip_pfn;
 229         struct page *page;
 230         int error;
 231
 232         /* XXX: are VM_FAULT_ codes OK? */
 233 again:
 234         size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 235         if (vmf->pgoff >= size)
 236                 return VM_FAULT_SIGBUS;
 237
 238         error = mapping->a_ops->get_xip_mem(mapping, vmf->pgoff, 0,
 239                                                 &xip_mem, &xip_pfn);
 240         if (likely(!error))
 241                 goto found;
 242         if (error != -ENODATA)
 243                 return VM_FAULT_OOM;
 244
 245         /* sparse block */
 246         if ((vma->vm_flags & (VM_WRITE | VM_MAYWRITE)) &&
 247             (vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) &&
 248             (!(mapping->host->i_sb->s_flags & MS_RDONLY))) {
 249                 int err;
 250
 251                 /* maybe shared writable, allocate new block */
 252                 mutex_lock(&xip_sparse_mutex);
 253                 error = mapping->a_ops->get_xip_mem(mapping, vmf->pgoff, 1,
 254                                                         &xip_mem, &xip_pfn);
 255                 mutex_unlock(&xip_sparse_mutex);
 256                 if (error)
 257                         return VM_FAULT_SIGBUS;
 258                 /* unmap sparse mappings at pgoff from all other vmas */
 259                 __xip_unmap(mapping, vmf->pgoff);
 260
 261 found:
 262                 err = vm_insert_mixed(vma, (unsigned long)vmf->virtual_address,
 263                                                         xip_pfn);
 264                 if (err == -ENOMEM)
 265                         return VM_FAULT_OOM;
 266                 /*
 267                  * err == -EBUSY is fine, we've raced against another thread
 268                  * that faulted-in the same page
 269                  */
 270                 if (err != -EBUSY)
 271                         BUG_ON(err);
 272                 return VM_FAULT_NOPAGE;
 273         } else {
 274                 int err, ret = VM_FAULT_OOM;
 275
 276                 mutex_lock(&xip_sparse_mutex);
 277                 write_seqcount_begin(&xip_sparse_seq);
 278                 error = mapping->a_ops->get_xip_mem(mapping, vmf->pgoff, 0,
 279                                                         &xip_mem, &xip_pfn);
 280                 if (unlikely(!error)) {
 281                         write_seqcount_end(&xip_sparse_seq);
 282                         mutex_unlock(&xip_sparse_mutex);
 283                         goto again;
 284                 }
 285                 if (error != -ENODATA)
 286                         goto out;
 287                 /* not shared and writable, use xip_sparse_page() */
 288                 page = xip_sparse_page();
 289                 if (!page)
 290                         goto out;
 291                 err = vm_insert_page(vma, (unsigned long)vmf->virtual_address,
 292                                                         page);
 293                 if (err == -ENOMEM)
 294                         goto out;
 295
 296                 ret = VM_FAULT_NOPAGE;
 297 out:
 298                 write_seqcount_end(&xip_sparse_seq);
 299                 mutex_unlock(&xip_sparse_mutex);
 300
 301                 return ret;
 302         }
 303 }
 304
 305 static const struct vm_operations_struct xip_file_vm_ops = {
 306         .fault  = xip_file_fault,
 307 };
 308
 309 int xip_file_mmap(struct file * file, struct vm_area_struct * vma)
 310 {
 311         BUG_ON(!file->f_mapping->a_ops->get_xip_mem);
 312
 313         file_accessed(file);
 314         vma->vm_ops = &xip_file_vm_ops;
 315         vma->vm_flags |= VM_CAN_NONLINEAR | VM_MIXEDMAP;
 316         return 0;
 317 }
 318 EXPORT_SYMBOL_GPL(xip_file_mmap);
 319
 320 static ssize_t
 321 __xip_file_write(struct file *filp, const char __user *buf,
 322                   size_t count, loff_t pos, loff_t *ppos)
 323 {
 324         struct address_space * mapping = filp->f_mapping;
 325         const struct address_space_operations *a_ops = mapping->a_ops;
 326         struct inode    *inode = mapping->host;
 327         long            status = 0;
 328         size_t          bytes;
 329         ssize_t         written = 0;
 330
 331         BUG_ON(!mapping->a_ops->get_xip_mem);
 332
 333         do {
 334                 unsigned long index;
 335                 unsigned long offset;
 336                 size_t copied;
 337                 void *xip_mem;
 338                 unsigned long xip_pfn;
 339
 340                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
 341                 index = pos >> PAGE_CACHE_SHIFT;
 342                 bytes = PAGE_CACHE_SIZE - offset;
 343                 if (bytes > count)
 344                         bytes = count;
 345
 346                 status = a_ops->get_xip_mem(mapping, index, 0,
 347                                                 &xip_mem, &xip_pfn);
 348                 if (status == -ENODATA) {
 349                         /* we allocate a new page unmap it */
 350                         mutex_lock(&xip_sparse_mutex);
 351                         status = a_ops->get_xip_mem(mapping, index, 1,
 352                                                         &xip_mem, &xip_pfn);
 353                         mutex_unlock(&xip_sparse_mutex);
 354                         if (!status)
 355                                 /* unmap page at pgoff from all other vmas */
 356                                 __xip_unmap(mapping, index);
 357                 }
 358
 359                 if (status)
 360                         break;
 361
 362                 copied = bytes -
 363                         __copy_from_user_nocache(xip_mem + offset, buf, bytes);
 364
 365                 if (likely(copied > 0)) {
 366                         status = copied;
 367
 368                         if (status >= 0) {
 369                                 written += status;
 370                                 count -= status;
 371                                 pos += status;
 372                                 buf += status;
 373                         }
 374                 }
 375                 if (unlikely(copied != bytes))
 376                         if (status >= 0)
 377                                 status = -EFAULT;
 378                 if (status < 0)
 379                         break;
 380         } while (count);
 381         *ppos = pos;
 382         /*
 383          * No need to use i_size_read() here, the i_size
 384          * cannot change under us because we hold i_mutex.
 385          */
 386         if (pos > inode->i_size) {
 387                 i_size_write(inode, pos);
 388                 mark_inode_dirty(inode);
 389         }
 390
 391         return written ? written : status;
 392 }
 393
 394 ssize_t
 395 xip_file_write(struct file *filp, const char __user *buf, size_t len,
 396                loff_t *ppos)
 397 {
 398         struct address_space *mapping = filp->f_mapping;
 399         struct inode *inode = mapping->host;
 400         size_t count;
 401         loff_t pos;
 402         ssize_t ret;
 403
 404         mutex_lock(&inode->i_mutex);
 405
 406         if (!access_ok(VERIFY_READ, buf, len)) {
 407                 ret=-EFAULT;
 408                 goto out_up;
 409         }
 410
 411         pos = *ppos;
 412         count = len;
 413
 414         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 415
 416         /* We can write back this queue in page reclaim */
 417         current->backing_dev_info = mapping->backing_dev_info;
 418
 419         ret = generic_write_checks(filp, &pos, &count, S_ISBLK(inode->i_mode));
 420         if (ret)
 421                 goto out_backing;
 422         if (count == 0)
 423                 goto out_backing;
 424
 425         ret = file_remove_suid(filp);
 426         if (ret)
 427                 goto out_backing;
 428
 429         file_update_time(filp);
 430
 431         ret = __xip_file_write (filp, buf, count, pos, ppos);
 432
 433  out_backing:
 434         current->backing_dev_info = NULL;
 435  out_up:
 436         mutex_unlock(&inode->i_mutex);
 437         return ret;
 438 }
 439 EXPORT_SYMBOL_GPL(xip_file_write);
 440
 441 /*
 442  * truncate a page used for execute in place
 443  * functionality is analog to block_truncate_page but does use get_xip_mem
 444  * to get the page instead of page cache
 445  */
 446 int
 447 xip_truncate_page(struct address_space *mapping, loff_t from)
 448 {
 449         pgoff_t index = from >> PAGE_CACHE_SHIFT;
 450         unsigned offset = from & (PAGE_CACHE_SIZE-1);
 451         unsigned blocksize;
 452         unsigned length;
 453         void *xip_mem;
 454         unsigned long xip_pfn;
 455         int err;
 456
 457         BUG_ON(!mapping->a_ops->get_xip_mem);
 458
 459         blocksize = 1 << mapping->host->i_blkbits;
 460         length = offset & (blocksize - 1);
 461
 462         /* Block boundary? Nothing to do */
 463         if (!length)
 464                 return 0;
 465
 466         length = blocksize - length;
 467
 468         err = mapping->a_ops->get_xip_mem(mapping, index, 0,
 469                                                 &xip_mem, &xip_pfn);
 470         if (unlikely(err)) {
 471                 if (err == -ENODATA)
 472                         /* Hole? No need to truncate */
 473                         return 0;
 474                 else
 475                         return err;
 476         }
 477         memset(xip_mem + offset, 0, length);
 478         return 0;
 479 }
 480 EXPORT_SYMBOL_GPL(xip_truncate_page);