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