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/backing-dev.h>
  13 #include <linux/pagemap.h>
  14 #include <linux/export.h>
  15 #include <linux/uio.h>
  16 #include <linux/rmap.h>
  17 #include <linux/mmu_notifier.h>
  18 #include <linux/sched.h>
  19 #include <linux/seqlock.h>
  20 #include <linux/mutex.h>
  21 #include <linux/gfp.h>
  22 #include <asm/tlbflush.h>
  23 #include <asm/io.h>
  24
  25 /*
  26  * We do use our own empty page to avoid interference with other users
  27  * of ZERO_PAGE(), such as /dev/zero
  28  */
  29 static DEFINE_MUTEX(xip_sparse_mutex);
  30 static seqcount_t xip_sparse_seq = SEQCNT_ZERO(xip_sparse_seq);
  31 static struct page *__xip_sparse_page;
  32
  33 /* called under xip_sparse_mutex */
  34 static struct page *xip_sparse_page(void)
  35 {
  36         if (!__xip_sparse_page) {
  37                 struct page *page = alloc_page(GFP_HIGHUSER | __GFP_ZERO);
  38
  39                 if (page)
  40                         __xip_sparse_page = page;
  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 - copied)
  95                         nr = len - copied;
  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 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 __xip_unmap(struct address_space * mapping, unsigned long pgoff)
 165 {
 166         struct vm_area_struct *vma;
 167         struct page *page;
 168         unsigned count;
 169         int locked = 0;
 170
 171         count = read_seqcount_begin(&xip_sparse_seq);
 172
 173         page = __xip_sparse_page;
 174         if (!page)
 175                 return;
 176
 177 retry:
 178         i_mmap_lock_read(mapping);
 179         vma_interval_tree_foreach(vma, &mapping->i_mmap, pgoff, pgoff) {
 180                 pte_t *pte, pteval;
 181                 spinlock_t *ptl;
 182                 struct mm_struct *mm = vma->vm_mm;
 183                 unsigned long address = vma->vm_start +
 184                         ((pgoff - vma->vm_pgoff) << PAGE_SHIFT);
 185
 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);
 193                         dec_mm_counter(mm, MM_FILEPAGES);
 194                         BUG_ON(pte_dirty(pteval));
 195                         pte_unmap_unlock(pte, ptl);
 196                         /* must invalidate_page _before_ freeing the page */
 197                         mmu_notifier_invalidate_page(mm, address);
 198                         page_cache_release(page);
 199                 }
 200         }
 201         i_mmap_unlock_read(mapping);
 202
 203         if (locked) {
 204                 mutex_unlock(&xip_sparse_mutex);
 205         } else if (read_seqcount_retry(&xip_sparse_seq, count)) {
 206                 mutex_lock(&xip_sparse_mutex);
 207                 locked = 1;
 208                 goto retry;
 209         }
 210 }
 211
 212 /*
 213  * xip_fault() is invoked via the vma operations vector for a
 214  * mapped memory region to read in file data during a page fault.
 215  *
 216  * This function is derived from filemap_fault, but used for execute in place
 217  */
 218 static int xip_file_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 219 {
 220         struct file *file = vma->vm_file;
 221         struct address_space *mapping = file->f_mapping;
 222         struct inode *inode = mapping->host;
 223         pgoff_t size;
 224         void *xip_mem;
 225         unsigned long xip_pfn;
 226         struct page *page;
 227         int error;
 228
 229         /* XXX: are VM_FAULT_ codes OK? */
 230 again:
 231         size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 232         if (vmf->pgoff >= size)
 233                 return VM_FAULT_SIGBUS;
 234
 235         error = mapping->a_ops->get_xip_mem(mapping, vmf->pgoff, 0,
 236                                                 &xip_mem, &xip_pfn);
 237         if (likely(!error))
 238                 goto found;
 239         if (error != -ENODATA)
 240                 return VM_FAULT_OOM;
 241
 242         /* sparse block */
 243         if ((vma->vm_flags & (VM_WRITE | VM_MAYWRITE)) &&
 244             (vma->vm_flags & (VM_SHARED | VM_MAYSHARE)) &&
 245             (!(mapping->host->i_sb->s_flags & MS_RDONLY))) {
 246                 int err;
 247
 248                 /* maybe shared writable, allocate new block */
 249                 mutex_lock(&xip_sparse_mutex);
 250                 error = mapping->a_ops->get_xip_mem(mapping, vmf->pgoff, 1,
 251                                                         &xip_mem, &xip_pfn);
 252                 mutex_unlock(&xip_sparse_mutex);
 253                 if (error)
 254                         return VM_FAULT_SIGBUS;
 255                 /* unmap sparse mappings at pgoff from all other vmas */
 256                 __xip_unmap(mapping, vmf->pgoff);
 257
 258 found:
 259                 err = vm_insert_mixed(vma, (unsigned long)vmf->virtual_address,
 260                                                         xip_pfn);
 261                 if (err == -ENOMEM)
 262                         return VM_FAULT_OOM;
 263                 /*
 264                  * err == -EBUSY is fine, we've raced against another thread
 265                  * that faulted-in the same page
 266                  */
 267                 if (err != -EBUSY)
 268                         BUG_ON(err);
 269                 return VM_FAULT_NOPAGE;
 270         } else {
 271                 int err, ret = VM_FAULT_OOM;
 272
 273                 mutex_lock(&xip_sparse_mutex);
 274                 write_seqcount_begin(&xip_sparse_seq);
 275                 error = mapping->a_ops->get_xip_mem(mapping, vmf->pgoff, 0,
 276                                                         &xip_mem, &xip_pfn);
 277                 if (unlikely(!error)) {
 278                         write_seqcount_end(&xip_sparse_seq);
 279                         mutex_unlock(&xip_sparse_mutex);
 280                         goto again;
 281                 }
 282                 if (error != -ENODATA)
 283                         goto out;
 284                 /* not shared and writable, use xip_sparse_page() */
 285                 page = xip_sparse_page();
 286                 if (!page)
 287                         goto out;
 288                 err = vm_insert_page(vma, (unsigned long)vmf->virtual_address,
 289                                                         page);
 290                 if (err == -ENOMEM)
 291                         goto out;
 292
 293                 ret = VM_FAULT_NOPAGE;
 294 out:
 295                 write_seqcount_end(&xip_sparse_seq);
 296                 mutex_unlock(&xip_sparse_mutex);
 297
 298                 return ret;
 299         }
 300 }
 301
 302 static const struct vm_operations_struct xip_file_vm_ops = {
 303         .fault  = xip_file_fault,
 304         .page_mkwrite   = filemap_page_mkwrite,
 305 };
 306
 307 int xip_file_mmap(struct file * file, struct vm_area_struct * vma)
 308 {
 309         BUG_ON(!file->f_mapping->a_ops->get_xip_mem);
 310
 311         file_accessed(file);
 312         vma->vm_ops = &xip_file_vm_ops;
 313         vma->vm_flags |= VM_MIXEDMAP;
 314         return 0;
 315 }
 316 EXPORT_SYMBOL_GPL(xip_file_mmap);
 317
 318 static ssize_t
 319 __xip_file_write(struct file *filp, const char __user *buf,
 320                   size_t count, loff_t pos, loff_t *ppos)
 321 {
 322         struct address_space * mapping = filp->f_mapping;
 323         const struct address_space_operations *a_ops = mapping->a_ops;
 324         struct inode    *inode = mapping->host;
 325         long            status = 0;
 326         size_t          bytes;
 327         ssize_t         written = 0;
 328
 329         BUG_ON(!mapping->a_ops->get_xip_mem);
 330
 331         do {
 332                 unsigned long index;
 333                 unsigned long offset;
 334                 size_t copied;
 335                 void *xip_mem;
 336                 unsigned long xip_pfn;
 337
 338                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
 339                 index = pos >> PAGE_CACHE_SHIFT;
 340                 bytes = PAGE_CACHE_SIZE - offset;
 341                 if (bytes > count)
 342                         bytes = count;
 343
 344                 status = a_ops->get_xip_mem(mapping, index, 0,
 345                                                 &xip_mem, &xip_pfn);
 346                 if (status == -ENODATA) {
 347                         /* we allocate a new page unmap it */
 348                         mutex_lock(&xip_sparse_mutex);
 349                         status = a_ops->get_xip_mem(mapping, index, 1,
 350                                                         &xip_mem, &xip_pfn);
 351                         mutex_unlock(&xip_sparse_mutex);
 352                         if (!status)
 353                                 /* unmap page at pgoff from all other vmas */
 354                                 __xip_unmap(mapping, index);
 355                 }
 356
 357                 if (status)
 358                         break;
 359
 360                 copied = bytes -
 361                         __copy_from_user_nocache(xip_mem + offset, buf, bytes);
 362
 363                 if (likely(copied > 0)) {
 364                         status = copied;
 365
 366                         if (status >= 0) {
 367                                 written += status;
 368                                 count -= status;
 369                                 pos += status;
 370                                 buf += status;
 371                         }
 372                 }
 373                 if (unlikely(copied != bytes))
 374                         if (status >= 0)
 375                                 status = -EFAULT;
 376                 if (status < 0)
 377                         break;
 378         } while (count);
 379         *ppos = pos;
 380         /*
 381          * No need to use i_size_read() here, the i_size
 382          * cannot change under us because we hold i_mutex.
 383          */
 384         if (pos > inode->i_size) {
 385                 i_size_write(inode, pos);
 386                 mark_inode_dirty(inode);
 387         }
 388
 389         return written ? written : status;
 390 }
 391
 392 ssize_t
 393 xip_file_write(struct file *filp, const char __user *buf, size_t len,
 394                loff_t *ppos)
 395 {
 396         struct address_space *mapping = filp->f_mapping;
 397         struct inode *inode = mapping->host;
 398         size_t count;
 399         loff_t pos;
 400         ssize_t ret;
 401
 402         mutex_lock(&inode->i_mutex);
 403
 404         if (!access_ok(VERIFY_READ, buf, len)) {
 405                 ret=-EFAULT;
 406                 goto out_up;
 407         }
 408
 409         pos = *ppos;
 410         count = len;
 411
 412         /* We can write back this queue in page reclaim */
 413         current->backing_dev_info = inode_to_bdi(inode);
 414
 415         ret = generic_write_checks(filp, &pos, &count, S_ISBLK(inode->i_mode));
 416         if (ret)
 417                 goto out_backing;
 418         if (count == 0)
 419                 goto out_backing;
 420
 421         ret = file_remove_suid(filp);
 422         if (ret)
 423                 goto out_backing;
 424
 425         ret = file_update_time(filp);
 426         if (ret)
 427                 goto out_backing;
 428
 429         ret = __xip_file_write (filp, buf, count, pos, ppos);
 430
 431  out_backing:
 432         current->backing_dev_info = NULL;
 433  out_up:
 434         mutex_unlock(&inode->i_mutex);
 435         return ret;
 436 }
 437 EXPORT_SYMBOL_GPL(xip_file_write);
 438
 439 /*
 440  * truncate a page used for execute in place
 441  * functionality is analog to block_truncate_page but does use get_xip_mem
 442  * to get the page instead of page cache
 443  */
 444 int
 445 xip_truncate_page(struct address_space *mapping, loff_t from)
 446 {
 447         pgoff_t index = from >> PAGE_CACHE_SHIFT;
 448         unsigned offset = from & (PAGE_CACHE_SIZE-1);
 449         unsigned blocksize;
 450         unsigned length;
 451         void *xip_mem;
 452         unsigned long xip_pfn;
 453         int err;
 454
 455         BUG_ON(!mapping->a_ops->get_xip_mem);
 456
 457         blocksize = 1 << mapping->host->i_blkbits;
 458         length = offset & (blocksize - 1);
 459
 460         /* Block boundary? Nothing to do */
 461         if (!length)
 462                 return 0;
 463
 464         length = blocksize - length;
 465
 466         err = mapping->a_ops->get_xip_mem(mapping, index, 0,
 467                                                 &xip_mem, &xip_pfn);
 468         if (unlikely(err)) {
 469                 if (err == -ENODATA)
 470                         /* Hole? No need to truncate */
 471                         return 0;
 472                 else
 473                         return err;
 474         }
 475         memset(xip_mem + offset, 0, length);
 476         return 0;
 477 }
 478 EXPORT_SYMBOL_GPL(xip_truncate_page);