mm/truncate.c

   1 /*
   2  * mm/truncate.c - code for taking down pages from address_spaces
   3  *
   4  * Copyright (C) 2002, Linus Torvalds
   5  *
   6  * 10Sep2002    akpm@zip.com.au
   7  *              Initial version.
   8  */
   9
  10 #include <linux/kernel.h>
  11 #include <linux/mm.h>
  12 #include <linux/module.h>
  13 #include <linux/pagemap.h>
  14 #include <linux/pagevec.h>
  15 #include <linux/buffer_head.h>  /* grr. try_to_release_page,
  16                                    block_invalidatepage */
  17
  18
  19 static int do_invalidatepage(struct page *page, unsigned long offset)
  20 {
  21         int (*invalidatepage)(struct page *, unsigned long);
  22         invalidatepage = page->mapping->a_ops->invalidatepage;
  23         if (invalidatepage == NULL)
  24                 invalidatepage = block_invalidatepage;
  25         return (*invalidatepage)(page, offset);
  26 }
  27
  28 static inline void truncate_partial_page(struct page *page, unsigned partial)
  29 {
  30         memclear_highpage_flush(page, partial, PAGE_CACHE_SIZE-partial);
  31         if (PagePrivate(page))
  32                 do_invalidatepage(page, partial);
  33 }
  34
  35 /*
  36  * If truncate cannot remove the fs-private metadata from the page, the page
  37  * becomes anonymous.  It will be left on the LRU and may even be mapped into
  38  * user pagetables if we're racing with filemap_nopage().
  39  *
  40  * We need to bale out if page->mapping is no longer equal to the original
  41  * mapping.  This happens a) when the VM reclaimed the page while we waited on
  42  * its lock, b) when a concurrent invalidate_inode_pages got there first and
  43  * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
  44  */
  45 static void
  46 truncate_complete_page(struct address_space *mapping, struct page *page)
  47 {
  48         if (page->mapping != mapping)
  49                 return;
  50
  51         if (PagePrivate(page))
  52                 do_invalidatepage(page, 0);
  53
  54         clear_page_dirty(page);
  55         ClearPageUptodate(page);
  56         ClearPageMappedToDisk(page);
  57         remove_from_page_cache(page);
  58         page_cache_release(page);       /* pagecache ref */
  59 }
  60
  61 /*
  62  * This is for invalidate_inode_pages().  That function can be called at
  63  * any time, and is not supposed to throw away dirty pages.  But pages can
  64  * be marked dirty at any time too.  So we re-check the dirtiness inside
  65  * ->tree_lock.  That provides exclusion against the __set_page_dirty
  66  * functions.
  67  */
  68 static int
  69 invalidate_complete_page(struct address_space *mapping, struct page *page)
  70 {
  71         if (page->mapping != mapping)
  72                 return 0;
  73
  74         if (PagePrivate(page) && !try_to_release_page(page, 0))
  75                 return 0;
  76
  77         spin_lock_irq(&mapping->tree_lock);
  78         if (PageDirty(page)) {
  79                 spin_unlock_irq(&mapping->tree_lock);
  80                 return 0;
  81         }
  82         __remove_from_page_cache(page);
  83         spin_unlock_irq(&mapping->tree_lock);
  84         ClearPageUptodate(page);
  85         page_cache_release(page);       /* pagecache ref */
  86         return 1;
  87 }
  88
  89 /**
  90  * truncate_inode_pages - truncate *all* the pages from an offset
  91  * @mapping: mapping to truncate
  92  * @lstart: offset from which to truncate
  93  *
  94  * Truncate the page cache at a set offset, removing the pages that are beyond
  95  * that offset (and zeroing out partial pages).
  96  *
  97  * Truncate takes two passes - the first pass is nonblocking.  It will not
  98  * block on page locks and it will not block on writeback.  The second pass
  99  * will wait.  This is to prevent as much IO as possible in the affected region.
 100  * The first pass will remove most pages, so the search cost of the second pass
 101  * is low.
 102  *
 103  * When looking at page->index outside the page lock we need to be careful to
 104  * copy it into a local to avoid races (it could change at any time).
 105  *
 106  * We pass down the cache-hot hint to the page freeing code.  Even if the
 107  * mapping is large, it is probably the case that the final pages are the most
 108  * recently touched, and freeing happens in ascending file offset order.
 109  *
 110  * Called under (and serialised by) inode->i_sem.
 111  */
 112 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
 113 {
 114         const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
 115         const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
 116         struct pagevec pvec;
 117         pgoff_t next;
 118         int i;
 119
 120         if (mapping->nrpages == 0)
 121                 return;
 122
 123         pagevec_init(&pvec, 0);
 124         next = start;
 125         while (pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 126                 for (i = 0; i < pagevec_count(&pvec); i++) {
 127                         struct page *page = pvec.pages[i];
 128                         pgoff_t page_index = page->index;
 129
 130                         if (page_index > next)
 131                                 next = page_index;
 132                         next++;
 133                         if (TestSetPageLocked(page))
 134                                 continue;
 135                         if (PageWriteback(page)) {
 136                                 unlock_page(page);
 137                                 continue;
 138                         }
 139                         truncate_complete_page(mapping, page);
 140                         unlock_page(page);
 141                 }
 142                 pagevec_release(&pvec);
 143                 cond_resched();
 144         }
 145
 146         if (partial) {
 147                 struct page *page = find_lock_page(mapping, start - 1);
 148                 if (page) {
 149                         wait_on_page_writeback(page);
 150                         truncate_partial_page(page, partial);
 151                         unlock_page(page);
 152                         page_cache_release(page);
 153                 }
 154         }
 155
 156         next = start;
 157         for ( ; ; ) {
 158                 cond_resched();
 159                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 160                         if (next == start)
 161                                 break;
 162                         next = start;
 163                         continue;
 164                 }
 165                 for (i = 0; i < pagevec_count(&pvec); i++) {
 166                         struct page *page = pvec.pages[i];
 167
 168                         lock_page(page);
 169                         wait_on_page_writeback(page);
 170                         if (page->index > next)
 171                                 next = page->index;
 172                         next++;
 173                         truncate_complete_page(mapping, page);
 174                         unlock_page(page);
 175                 }
 176                 pagevec_release(&pvec);
 177         }
 178 }
 179
 180 EXPORT_SYMBOL(truncate_inode_pages);
 181
 182 /**
 183  * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
 184  * @mapping: the address_space which holds the pages to invalidate
 185  * @start: the offset 'from' which to invalidate
 186  * @end: the offset 'to' which to invalidate (inclusive)
 187  *
 188  * This function only removes the unlocked pages, if you want to
 189  * remove all the pages of one inode, you must call truncate_inode_pages.
 190  *
 191  * invalidate_mapping_pages() will not block on IO activity. It will not
 192  * invalidate pages which are dirty, locked, under writeback or mapped into
 193  * pagetables.
 194  */
 195 unsigned long invalidate_mapping_pages(struct address_space *mapping,
 196                                 pgoff_t start, pgoff_t end)
 197 {
 198         struct pagevec pvec;
 199         pgoff_t next = start;
 200         unsigned long ret = 0;
 201         int i;
 202
 203         pagevec_init(&pvec, 0);
 204         while (next <= end &&
 205                         pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 206                 for (i = 0; i < pagevec_count(&pvec); i++) {
 207                         struct page *page = pvec.pages[i];
 208
 209                         if (TestSetPageLocked(page)) {
 210                                 next++;
 211                                 continue;
 212                         }
 213                         if (page->index > next)
 214                                 next = page->index;
 215                         next++;
 216                         if (PageDirty(page) || PageWriteback(page))
 217                                 goto unlock;
 218                         if (page_mapped(page))
 219                                 goto unlock;
 220                         ret += invalidate_complete_page(mapping, page);
 221 unlock:
 222                         unlock_page(page);
 223                         if (next > end)
 224                                 break;
 225                 }
 226                 pagevec_release(&pvec);
 227                 cond_resched();
 228         }
 229         return ret;
 230 }
 231
 232 unsigned long invalidate_inode_pages(struct address_space *mapping)
 233 {
 234         return invalidate_mapping_pages(mapping, 0, ~0UL);
 235 }
 236
 237 EXPORT_SYMBOL(invalidate_inode_pages);
 238
 239 /**
 240  * invalidate_inode_pages2 - remove all unmapped pages from an address_space
 241  * @mapping - the address_space
 242  *
 243  * invalidate_inode_pages2() is like truncate_inode_pages(), except for the case
 244  * where the page is seen to be mapped into process pagetables.  In that case,
 245  * the page is marked clean but is left attached to its address_space.
 246  *
 247  * The page is also marked not uptodate so that a subsequent pagefault will
 248  * perform I/O to bringthe page's contents back into sync with its backing
 249  * store.
 250  *
 251  * FIXME: invalidate_inode_pages2() is probably trivially livelockable.
 252  */
 253 void invalidate_inode_pages2(struct address_space *mapping)
 254 {
 255         struct pagevec pvec;
 256         pgoff_t next = 0;
 257         int i;
 258
 259         pagevec_init(&pvec, 0);
 260         while (pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 261                 for (i = 0; i < pagevec_count(&pvec); i++) {
 262                         struct page *page = pvec.pages[i];
 263
 264                         lock_page(page);
 265                         if (page->mapping == mapping) { /* truncate race? */
 266                                 wait_on_page_writeback(page);
 267                                 next = page->index + 1;
 268                                 if (page_mapped(page)) {
 269                                         clear_page_dirty(page);
 270                                         ClearPageUptodate(page);
 271                                 } else {
 272                                         invalidate_complete_page(mapping, page);
 273                                 }
 274                         }
 275                         unlock_page(page);
 276                 }
 277                 pagevec_release(&pvec);
 278                 cond_resched();
 279         }
 280 }
 281
 282 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);