• source navigation  • diff markup  • identifier search  • freetext search  • file search  • 


Version: 2.6.8   2.6.16   2.6.25   2.6.30   2.6.34  

Architecture: i386   arm   mips   ppc   alpha   m68k   sparc   sparc64  

linux/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                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
159                         if (next == start)
160                                 break;
161                         next = start;
162                         continue;
163                 }
164                 for (i = 0; i < pagevec_count(&pvec); i++) {
165                         struct page *page = pvec.pages[i];
166 
167                         lock_page(page);
168                         wait_on_page_writeback(page);
169                         if (page->index > next)
170                                 next = page->index;
171                         next++;
172                         truncate_complete_page(mapping, page);
173                         unlock_page(page);
174                 }
175                 pagevec_release(&pvec);
176         }
177 }
178 
179 EXPORT_SYMBOL(truncate_inode_pages);
180 
181 /**
182  * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
183  * @mapping: the address_space which holds the pages to invalidate
184  * @start: the offset 'from' which to invalidate
185  * @end: the offset 'to' which to invalidate (inclusive)
186  *
187  * This function only removes the unlocked pages, if you want to
188  * remove all the pages of one inode, you must call truncate_inode_pages.
189  *
190  * invalidate_mapping_pages() will not block on IO activity. It will not
191  * invalidate pages which are dirty, locked, under writeback or mapped into
192  * pagetables.
193  */
194 unsigned long invalidate_mapping_pages(struct address_space *mapping,
195                                 pgoff_t start, pgoff_t end)
196 {
197         struct pagevec pvec;
198         pgoff_t next = start;
199         unsigned long ret = 0;
200         int i;
201 
202         pagevec_init(&pvec, 0);
203         while (next <= end &&
204                         pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
205                 for (i = 0; i < pagevec_count(&pvec); i++) {
206                         struct page *page = pvec.pages[i];
207 
208                         if (TestSetPageLocked(page)) {
209                                 next++;
210                                 continue;
211                         }
212                         if (page->index > next)
213                                 next = page->index;
214                         next++;
215                         if (PageDirty(page) || PageWriteback(page))
216                                 goto unlock;
217                         if (page_mapped(page))
218                                 goto unlock;
219                         ret += invalidate_complete_page(mapping, page);
220 unlock:
221                         unlock_page(page);
222                         if (next > end)
223                                 break;
224                 }
225                 pagevec_release(&pvec);
226                 cond_resched();
227         }
228         return ret;
229 }
230 
231 unsigned long invalidate_inode_pages(struct address_space *mapping)
232 {
233         return invalidate_mapping_pages(mapping, 0, ~0UL);
234 }
235 
236 EXPORT_SYMBOL(invalidate_inode_pages);
237 
238 /**
239  * invalidate_inode_pages2 - remove all unmapped pages from an address_space
240  * @mapping - the address_space
241  *
242  * invalidate_inode_pages2() is like truncate_inode_pages(), except for the case
243  * where the page is seen to be mapped into process pagetables.  In that case,
244  * the page is marked clean but is left attached to its address_space.
245  *
246  * The page is also marked not uptodate so that a subsequent pagefault will
247  * perform I/O to bringthe page's contents back into sync with its backing
248  * store.
249  *
250  * FIXME: invalidate_inode_pages2() is probably trivially livelockable.
251  */
252 void invalidate_inode_pages2(struct address_space *mapping)
253 {
254         struct pagevec pvec;
255         pgoff_t next = 0;
256         int i;
257 
258         pagevec_init(&pvec, 0);
259         while (pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
260                 for (i = 0; i < pagevec_count(&pvec); i++) {
261                         struct page *page = pvec.pages[i];
262 
263                         lock_page(page);
264                         if (page->mapping == mapping) { /* truncate race? */
265                                 wait_on_page_writeback(page);
266                                 next = page->index + 1;
267                                 if (page_mapped(page)) {
268                                         clear_page_dirty(page);
269                                         ClearPageUptodate(page);
270                                 } else {
271                                         invalidate_complete_page(mapping, page);
272                                 }
273                         }
274                         unlock_page(page);
275                 }
276                 pagevec_release(&pvec);
277                 cond_resched();
278         }
279 }
280 
281 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
282 

This page was automatically generated by LXR 0.3.1. •  Linux is a registered trademark of Linus Torvalds