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 <asm/tlbflush.h>
  21 #include <asm/io.h>
  22
  23 /*
  24  * We do use our own empty page to avoid interference with other users
  25  * of ZERO_PAGE(), such as /dev/zero
  26  */
  27 static DEFINE_MUTEX(xip_sparse_mutex);
  28 static seqcount_t xip_sparse_seq = SEQCNT_ZERO;
  29 static struct page *__xip_sparse_page;
  30
  31 /* called under xip_sparse_mutex */
  32 static struct page *xip_sparse_page(void)
  33 {
  34         if (!__xip_sparse_page) {
  35                 struct page *page = alloc_page(GFP_HIGHUSER | __GFP_ZERO);
  36
  37                 if (page)
  38                         __xip_sparse_page = page;
  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_mem() 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 ssize_t
  51 do_xip_mapping_read(struct address_space *mapping,
  52                     struct file_ra_state *_ra,
  53                     struct file *filp,
  54                     char __user *buf,
  55                     size_t len,
  56                     loff_t *ppos)
  57 {
  58         struct inode *inode = mapping->host;
  59         pgoff_t index, end_index;
  60         unsigned long offset;
  61         loff_t isize, pos;
  62         size_t copied = 0, error = 0;
  63
  64         BUG_ON(!mapping->a_ops->get_xip_mem);
  65
  66         pos = *ppos;
  67         index = pos >> PAGE_CACHE_SHIFT;
  68         offset = pos & ~PAGE_CACHE_MASK;
  69
  70         isize = i_size_read(inode);
  71         if (!isize)
  72                 goto out;
  73
  74         end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
  75         do {
  76                 unsigned long nr, left;
  77                 void *xip_mem;
  78                 unsigned long xip_pfn;
  79                 int zero = 0;
  80
  81                 /* nr is the maximum number of bytes to copy from this page */
  82                 nr = PAGE_CACHE_SIZE;
  83                 if (index >= end_index) {
  84                         if (index > end_index)
  85                                 goto out;
  86                         nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
  87                         if (nr <= offset) {
  88                                 goto out;
  89                         }
  90                 }
  91                 nr = nr - offset;
  92                 if (nr > len - copied)
  93                         nr = len - copied;
  94
  95                 error = mapping->a_ops->get_xip_mem(mapping, index, 0,
  96                                                         &xip_mem, &xip_pfn);
  97                 if (unlikely(error)) {
  98                         if (error == -ENODATA) {
  99                                 /* sparse */
 100                                 zero = 1;
 101                         } else
 102                                 goto out;
 103                 }
 104
 105                 /* If users can be writing to this page using arbitrary
 106                  * virtual addresses, take care about potential aliasing
 107                  * before reading the page on the kernel side.
 108                  */
 109                 if (mapping_writably_mapped(mapping))
 110                         /* address based flush */ ;
 111
 112                 /*
 113                  * Ok, we have the mem, so now we can copy it to user space...
 114                  *
 115                  * The actor routine returns how many bytes were actually used..
 116                  * NOTE! This may not be the same as how much of a user buffer
 117                  * we filled up (we may be padding etc), so we can only update
 118                  * "pos" here (the actor routine has to update the user buffer
 119                  * pointers and the remaining count).
 120                  */
 121                 if (!zero)
 122                         left = __copy_to_user(buf+copied, xip_mem+offset, nr);
 123                 else
 124                         left = __clear_user(buf + copied, nr);
 125
 126                 if (left) {
 127                         error = -EFAULT;
 128                         goto out;
 129                 }
 130
 131                 copied += (nr - left);
 132                 offset += (nr - left);
 133                 index += offset >> PAGE_CACHE_SHIFT;
 134                 offset &= ~PAGE_CACHE_MASK;
 135         } while (copied < len);
 136
 137 out:
 138         *ppos = pos + copied;
 139         if (filp)
 140                 file_accessed(filp);
 141
 142         return (copied ? copied : error);
 143 }
 144
 145 ssize_t
 146 xip_file_read(struct file *filp, char __user *buf, size_t len, loff_t *ppos)
 147 {
 148         if (!access_ok(VERIFY_WRITE, buf, len))
 149                 return -EFAULT;
 150
 151         return do_xip_mapping_read(filp->f_mapping, &filp->f_ra, filp,
 152                             buf, len, ppos);
 153 }
 154 EXPORT_SYMBOL_GPL(xip_file_read);
 155
 156 /*
 157  * __xip_unmap is invoked from xip_unmap and
 158  * xip_write
 159  *
 160  * This function walks all vmas of the address_space and unmaps the
 161  * __xip_sparse_page when found at pgoff.
 162  */
 163 static void
 164 __xip_unmap (struct address_space * mapping,
 165                      unsigned long pgoff)
 166 {
 167         struct vm_area_struct *vma;
 168         struct mm_struct *mm;
 169         struct prio_tree_iter iter;
 170         unsigned long address;
 171         pte_t *pte;
 172         pte_t pteval;
 173         spinlock_t *ptl;
 174         struct page *page;
 175         unsigned count;
 176         int locked = 0;
 177
 178         count = read_seqcount_begin(&xip_sparse_seq);
 179
 180         page = __xip_sparse_page;
 181         if (!page)
 182                 return;
 183
 184 retry:
 185         spin_lock(&mapping->i_mmap_lock);
 186         vma_prio_tree_foreach(vma, &iter, &mapping->i_mmap, pgoff, pgoff) {
 187                 mm = vma->vm_mm;
 188                 address = vma->vm_start +
 189                         ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
 190                 BUG_ON(address < vma->vm_start || address >= vma->vm_end);
 191                 pte = page_check_address(page, mm, address, &ptl, 1);
 192                 if (pte) {
 193                         /* Nuke the page table entry. */
 194                         flush_cache_page(vma, address, pte_pfn(*pte));
 195                         pteval = ptep_clear_flush_notify(vma, address, pte);
 196                         page_remove_rmap(page);
 197                         dec_mm_counter(mm, file_rss);
 198                         BUG_ON(pte_dirty(pteval));
 199                         pte_unmap_unlock(pte, ptl);
 200                         page_cache_release(page);
 201                 }
 202         }
 203         spin_unlock(&mapping->i_mmap_lock);
 204
 205         if (locked) {
 206                 mutex_unlock(&xip_sparse_mutex);
 207         } else if (read_seqcount_retry(&xip_sparse_seq, count)) {
 208                 mutex_lock(&xip_sparse_mutex);
 209                 locked = 1;
 210                 goto retry;
 211         }
 212 }
 213
 214 /*
 215  * xip_fault() is invoked via the vma operations vector for a
 216  * mapped memory region to read in file data during a page fault.
 217  *
 218  * This function is derived from filemap_fault, but used for execute in place
 219  */
 220 static int xip_file_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 221 {
 222         struct file *file = vma->vm_file;
 223         struct address_space *mapping = file->f_mapping;
 224         struct inode *inode = mapping->host;
 225         pgoff_t size;
 226         void *xip_mem;
 227         unsigned long xip_pfn;
 228         struct page *page;
 229         int error;
 230
 231         /* XXX: are VM_FAULT_ codes OK? */
 232 again:
 233         size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 234         if (vmf->pgoff >= size)
 235                 return VM_FAULT_SIGBUS;
 236
 237         error = mapping->a_ops->get_xip_mem(mapping, vmf->pgoff, 0,
 238                                                 &xip_mem, &xip_pfn);
 239         if (likely(!error))
 240                 goto found;
 241         if (error != -ENODATA)
 242                 return VM_FAULT_OOM;
 243
 244         /* sparse block */
 245         if ((vma->vm_flags & (VM_WRITE | VM_MAYWRITE)) &&
 246             (vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) &&
 247             (!(mapping->host->i_sb->s_flags & MS_RDONLY))) {
 248                 int err;
 249
 250                 /* maybe shared writable, allocate new block */
 251                 mutex_lock(&xip_sparse_mutex);
 252                 error = mapping->a_ops->get_xip_mem(mapping, vmf->pgoff, 1,
 253                                                         &xip_mem, &xip_pfn);
 254                 mutex_unlock(&xip_sparse_mutex);
 255                 if (error)
 256                         return VM_FAULT_SIGBUS;
 257                 /* unmap sparse mappings at pgoff from all other vmas */
 258                 __xip_unmap(mapping, vmf->pgoff);
 259
 260 found:
 261                 err = vm_insert_mixed(vma, (unsigned long)vmf->virtual_address,
 262                                                         xip_pfn);
 263                 if (err == -ENOMEM)
 264                         return VM_FAULT_OOM;
 265                 /*
 266                  * err == -EBUSY is fine, we've raced against another thread
 267                  * that faulted-in the same page
 268                  */
 269                 if (err != -EBUSY)
 270                         BUG_ON(err);
 271                 return VM_FAULT_NOPAGE;
 272         } else {
 273                 int err, ret = VM_FAULT_OOM;
 274
 275                 mutex_lock(&xip_sparse_mutex);
 276                 write_seqcount_begin(&xip_sparse_seq);
 277                 error = mapping->a_ops->get_xip_mem(mapping, vmf->pgoff, 0,
 278                                                         &xip_mem, &xip_pfn);
 279                 if (unlikely(!error)) {
 280                         write_seqcount_end(&xip_sparse_seq);
 281                         mutex_unlock(&xip_sparse_mutex);
 282                         goto again;
 283                 }
 284                 if (error != -ENODATA)
 285                         goto out;
 286                 /* not shared and writable, use xip_sparse_page() */
 287                 page = xip_sparse_page();
 288                 if (!page)
 289                         goto out;
 290                 err = vm_insert_page(vma, (unsigned long)vmf->virtual_address,
 291                                                         page);
 292                 if (err == -ENOMEM)
 293                         goto out;
 294
 295                 ret = VM_FAULT_NOPAGE;
 296 out:
 297                 write_seqcount_end(&xip_sparse_seq);
 298                 mutex_unlock(&xip_sparse_mutex);
 299
 300                 return ret;
 301         }
 302 }
 303
 304 static const struct vm_operations_struct xip_file_vm_ops = {
 305         .fault  = xip_file_fault,
 306 };
 307
 308 int xip_file_mmap(struct file * file, struct vm_area_struct * vma)
 309 {
 310         BUG_ON(!file->f_mapping->a_ops->get_xip_mem);
 311
 312         file_accessed(file);
 313         vma->vm_ops = &xip_file_vm_ops;
 314         vma->vm_flags |= VM_CAN_NONLINEAR | VM_MIXEDMAP;
 315         return 0;
 316 }
 317 EXPORT_SYMBOL_GPL(xip_file_mmap);
 318
 319 static ssize_t
 320 __xip_file_write(struct file *filp, const char __user *buf,
 321                   size_t count, loff_t pos, loff_t *ppos)
 322 {
 323         struct address_space * mapping = filp->f_mapping;
 324         const struct address_space_operations *a_ops = mapping->a_ops;
 325         struct inode    *inode = mapping->host;
 326         long            status = 0;
 327         size_t          bytes;
 328         ssize_t         written = 0;
 329
 330         BUG_ON(!mapping->a_ops->get_xip_mem);
 331
 332         do {
 333                 unsigned long index;
 334                 unsigned long offset;
 335                 size_t copied;
 336                 void *xip_mem;
 337                 unsigned long xip_pfn;
 338
 339                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
 340                 index = pos >> PAGE_CACHE_SHIFT;
 341                 bytes = PAGE_CACHE_SIZE - offset;
 342                 if (bytes > count)
 343                         bytes = count;
 344
 345                 status = a_ops->get_xip_mem(mapping, index, 0,
 346                                                 &xip_mem, &xip_pfn);
 347                 if (status == -ENODATA) {
 348                         /* we allocate a new page unmap it */
 349                         mutex_lock(&xip_sparse_mutex);
 350                         status = a_ops->get_xip_mem(mapping, index, 1,
 351                                                         &xip_mem, &xip_pfn);
 352                         mutex_unlock(&xip_sparse_mutex);
 353                         if (!status)
 354                                 /* unmap page at pgoff from all other vmas */
 355                                 __xip_unmap(mapping, index);
 356                 }
 357
 358                 if (status)
 359                         break;
 360
 361                 copied = bytes -
 362                         __copy_from_user_nocache(xip_mem + offset, buf, bytes);
 363
 364                 if (likely(copied > 0)) {
 365                         status = copied;
 366
 367                         if (status >= 0) {
 368                                 written += status;
 369                                 count -= status;
 370                                 pos += status;
 371                                 buf += status;
 372                         }
 373                 }
 374                 if (unlikely(copied != bytes))
 375                         if (status >= 0)
 376                                 status = -EFAULT;
 377                 if (status < 0)
 378                         break;
 379         } while (count);
 380         *ppos = pos;
 381         /*
 382          * No need to use i_size_read() here, the i_size
 383          * cannot change under us because we hold i_mutex.
 384          */
 385         if (pos > inode->i_size) {
 386                 i_size_write(inode, pos);
 387                 mark_inode_dirty(inode);
 388         }
 389
 390         return written ? written : status;
 391 }
 392
 393 ssize_t
 394 xip_file_write(struct file *filp, const char __user *buf, size_t len,
 395                loff_t *ppos)
 396 {
 397         struct address_space *mapping = filp->f_mapping;
 398         struct inode *inode = mapping->host;
 399         size_t count;
 400         loff_t pos;
 401         ssize_t ret;
 402
 403         mutex_lock(&inode->i_mutex);
 404
 405         if (!access_ok(VERIFY_READ, buf, len)) {
 406                 ret=-EFAULT;
 407                 goto out_up;
 408         }
 409
 410         pos = *ppos;
 411         count = len;
 412
 413         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 414
 415         /* We can write back this queue in page reclaim */
 416         current->backing_dev_info = mapping->backing_dev_info;
 417
 418         ret = generic_write_checks(filp, &pos, &count, S_ISBLK(inode->i_mode));
 419         if (ret)
 420                 goto out_backing;
 421         if (count == 0)
 422                 goto out_backing;
 423
 424         ret = file_remove_suid(filp);
 425         if (ret)
 426                 goto out_backing;
 427
 428         file_update_time(filp);
 429
 430         ret = __xip_file_write (filp, buf, count, pos, ppos);
 431
 432  out_backing:
 433         current->backing_dev_info = NULL;
 434  out_up:
 435         mutex_unlock(&inode->i_mutex);
 436         return ret;
 437 }
 438 EXPORT_SYMBOL_GPL(xip_file_write);
 439
 440 /*
 441  * truncate a page used for execute in place
 442  * functionality is analog to block_truncate_page but does use get_xip_mem
 443  * to get the page instead of page cache
 444  */
 445 int
 446 xip_truncate_page(struct address_space *mapping, loff_t from)
 447 {
 448         pgoff_t index = from >> PAGE_CACHE_SHIFT;
 449         unsigned offset = from & (PAGE_CACHE_SIZE-1);
 450         unsigned blocksize;
 451         unsigned length;
 452         void *xip_mem;
 453         unsigned long xip_pfn;
 454         int err;
 455
 456         BUG_ON(!mapping->a_ops->get_xip_mem);
 457
 458         blocksize = 1 << mapping->host->i_blkbits;
 459         length = offset & (blocksize - 1);
 460
 461         /* Block boundary? Nothing to do */
 462         if (!length)
 463                 return 0;
 464
 465         length = blocksize - length;
 466
 467         err = mapping->a_ops->get_xip_mem(mapping, index, 0,
 468                                                 &xip_mem, &xip_pfn);
 469         if (unlikely(err)) {
 470                 if (err == -ENODATA)
 471                         /* Hole? No need to truncate */
 472                         return 0;
 473                 else
 474                         return err;
 475         }
 476         memset(xip_mem + offset, 0, length);
 477         return 0;
 478 }
 479 EXPORT_SYMBOL_GPL(xip_truncate_page);