Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1 | /* |
| 2 | * mm/truncate.c - code for taking down pages from address_spaces |
| 3 | * |
| 4 | * Copyright (C) 2002, Linus Torvalds |
| 5 | * |
Francois Cami | e1f8e87 | 2008-10-15 22:01:59 -0700 | [diff] [blame] | 6 | * 10Sep2002 Andrew Morton |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 7 | * Initial version. |
| 8 | */ |
| 9 | |
| 10 | #include <linux/kernel.h> |
Alexey Dobriyan | 4af3c9c | 2007-10-16 23:29:23 -0700 | [diff] [blame] | 11 | #include <linux/backing-dev.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 12 | #include <linux/mm.h> |
Nick Piggin | 0fd0e6b | 2006-09-27 01:50:02 -0700 | [diff] [blame] | 13 | #include <linux/swap.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 14 | #include <linux/module.h> |
| 15 | #include <linux/pagemap.h> |
Nate Diller | 01f2705d | 2007-05-09 02:35:07 -0700 | [diff] [blame] | 16 | #include <linux/highmem.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 17 | #include <linux/pagevec.h> |
Andrew Morton | e08748ce | 2006-12-10 02:19:31 -0800 | [diff] [blame] | 18 | #include <linux/task_io_accounting_ops.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 19 | #include <linux/buffer_head.h> /* grr. try_to_release_page, |
Jan Kara | aaa4059 | 2005-10-30 15:00:16 -0800 | [diff] [blame] | 20 | do_invalidatepage */ |
Rik van Riel | ba470de | 2008-10-18 20:26:50 -0700 | [diff] [blame] | 21 | #include "internal.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 22 | |
| 23 | |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 24 | /** |
Fengguang Wu | 28bc44d | 2008-02-03 18:04:10 +0200 | [diff] [blame] | 25 | * do_invalidatepage - invalidate part or all of a page |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 26 | * @page: the page which is affected |
| 27 | * @offset: the index of the truncation point |
| 28 | * |
| 29 | * do_invalidatepage() is called when all or part of the page has become |
| 30 | * invalidated by a truncate operation. |
| 31 | * |
| 32 | * do_invalidatepage() does not have to release all buffers, but it must |
| 33 | * ensure that no dirty buffer is left outside @offset and that no I/O |
| 34 | * is underway against any of the blocks which are outside the truncation |
| 35 | * point. Because the caller is about to free (and possibly reuse) those |
| 36 | * blocks on-disk. |
| 37 | */ |
| 38 | void do_invalidatepage(struct page *page, unsigned long offset) |
| 39 | { |
| 40 | void (*invalidatepage)(struct page *, unsigned long); |
| 41 | invalidatepage = page->mapping->a_ops->invalidatepage; |
David Howells | 9361401 | 2006-09-30 20:45:40 +0200 | [diff] [blame] | 42 | #ifdef CONFIG_BLOCK |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 43 | if (!invalidatepage) |
| 44 | invalidatepage = block_invalidatepage; |
David Howells | 9361401 | 2006-09-30 20:45:40 +0200 | [diff] [blame] | 45 | #endif |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 46 | if (invalidatepage) |
| 47 | (*invalidatepage)(page, offset); |
| 48 | } |
| 49 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 50 | static inline void truncate_partial_page(struct page *page, unsigned partial) |
| 51 | { |
Christoph Lameter | eebd2aa | 2008-02-04 22:28:29 -0800 | [diff] [blame] | 52 | zero_user_segment(page, partial, PAGE_CACHE_SIZE); |
David Howells | 266cf65 | 2009-04-03 16:42:36 +0100 | [diff] [blame] | 53 | if (page_has_private(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 54 | do_invalidatepage(page, partial); |
| 55 | } |
| 56 | |
Linus Torvalds | ecdfc97 | 2007-01-26 12:47:06 -0800 | [diff] [blame] | 57 | /* |
| 58 | * This cancels just the dirty bit on the kernel page itself, it |
| 59 | * does NOT actually remove dirty bits on any mmap's that may be |
| 60 | * around. It also leaves the page tagged dirty, so any sync |
| 61 | * activity will still find it on the dirty lists, and in particular, |
| 62 | * clear_page_dirty_for_io() will still look at the dirty bits in |
| 63 | * the VM. |
| 64 | * |
| 65 | * Doing this should *normally* only ever be done when a page |
| 66 | * is truncated, and is not actually mapped anywhere at all. However, |
| 67 | * fs/buffer.c does this when it notices that somebody has cleaned |
| 68 | * out all the buffers on a page without actually doing it through |
| 69 | * the VM. Can you say "ext3 is horribly ugly"? Tought you could. |
| 70 | */ |
Linus Torvalds | fba2591 | 2006-12-20 13:46:42 -0800 | [diff] [blame] | 71 | void cancel_dirty_page(struct page *page, unsigned int account_size) |
| 72 | { |
Linus Torvalds | 8368e32 | 2006-12-23 09:25:04 -0800 | [diff] [blame] | 73 | if (TestClearPageDirty(page)) { |
| 74 | struct address_space *mapping = page->mapping; |
| 75 | if (mapping && mapping_cap_account_dirty(mapping)) { |
| 76 | dec_zone_page_state(page, NR_FILE_DIRTY); |
Peter Zijlstra | c9e51e4 | 2007-10-16 23:25:47 -0700 | [diff] [blame] | 77 | dec_bdi_stat(mapping->backing_dev_info, |
| 78 | BDI_RECLAIMABLE); |
Linus Torvalds | 8368e32 | 2006-12-23 09:25:04 -0800 | [diff] [blame] | 79 | if (account_size) |
| 80 | task_io_account_cancelled_write(account_size); |
| 81 | } |
Andrew Morton | 3e67c09 | 2006-12-21 11:00:33 -0800 | [diff] [blame] | 82 | } |
Linus Torvalds | fba2591 | 2006-12-20 13:46:42 -0800 | [diff] [blame] | 83 | } |
Linus Torvalds | 8368e32 | 2006-12-23 09:25:04 -0800 | [diff] [blame] | 84 | EXPORT_SYMBOL(cancel_dirty_page); |
Linus Torvalds | fba2591 | 2006-12-20 13:46:42 -0800 | [diff] [blame] | 85 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 86 | /* |
| 87 | * If truncate cannot remove the fs-private metadata from the page, the page |
Shaohua Li | 62e1c55 | 2008-02-04 22:29:33 -0800 | [diff] [blame] | 88 | * becomes orphaned. It will be left on the LRU and may even be mapped into |
Nick Piggin | 54cb882 | 2007-07-19 01:46:59 -0700 | [diff] [blame] | 89 | * user pagetables if we're racing with filemap_fault(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 90 | * |
| 91 | * We need to bale out if page->mapping is no longer equal to the original |
| 92 | * mapping. This happens a) when the VM reclaimed the page while we waited on |
Andrew Morton | fc0ecff | 2007-02-10 01:45:39 -0800 | [diff] [blame] | 93 | * its lock, b) when a concurrent invalidate_mapping_pages got there first and |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 94 | * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. |
| 95 | */ |
Nick Piggin | 750b498 | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 96 | static int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 97 | truncate_complete_page(struct address_space *mapping, struct page *page) |
| 98 | { |
| 99 | if (page->mapping != mapping) |
Nick Piggin | 750b498 | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 100 | return -EIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 101 | |
David Howells | 266cf65 | 2009-04-03 16:42:36 +0100 | [diff] [blame] | 102 | if (page_has_private(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 103 | do_invalidatepage(page, 0); |
| 104 | |
Bjorn Steinbrink | a2b3456 | 2008-02-04 22:29:28 -0800 | [diff] [blame] | 105 | cancel_dirty_page(page, PAGE_CACHE_SIZE); |
| 106 | |
Rik van Riel | ba470de | 2008-10-18 20:26:50 -0700 | [diff] [blame] | 107 | clear_page_mlock(page); |
Nick Piggin | 787d2214 | 2007-07-17 04:03:34 -0700 | [diff] [blame] | 108 | remove_from_page_cache(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 109 | ClearPageMappedToDisk(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 110 | page_cache_release(page); /* pagecache ref */ |
Nick Piggin | 750b498 | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 111 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 112 | } |
| 113 | |
| 114 | /* |
Andrew Morton | fc0ecff | 2007-02-10 01:45:39 -0800 | [diff] [blame] | 115 | * This is for invalidate_mapping_pages(). That function can be called at |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 116 | * any time, and is not supposed to throw away dirty pages. But pages can |
Nick Piggin | 0fd0e6b | 2006-09-27 01:50:02 -0700 | [diff] [blame] | 117 | * be marked dirty at any time too, so use remove_mapping which safely |
| 118 | * discards clean, unused pages. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 119 | * |
| 120 | * Returns non-zero if the page was successfully invalidated. |
| 121 | */ |
| 122 | static int |
| 123 | invalidate_complete_page(struct address_space *mapping, struct page *page) |
| 124 | { |
Nick Piggin | 0fd0e6b | 2006-09-27 01:50:02 -0700 | [diff] [blame] | 125 | int ret; |
| 126 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 127 | if (page->mapping != mapping) |
| 128 | return 0; |
| 129 | |
David Howells | 266cf65 | 2009-04-03 16:42:36 +0100 | [diff] [blame] | 130 | if (page_has_private(page) && !try_to_release_page(page, 0)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 131 | return 0; |
| 132 | |
Rik van Riel | ba470de | 2008-10-18 20:26:50 -0700 | [diff] [blame] | 133 | clear_page_mlock(page); |
Nick Piggin | 0fd0e6b | 2006-09-27 01:50:02 -0700 | [diff] [blame] | 134 | ret = remove_mapping(mapping, page); |
Nick Piggin | 0fd0e6b | 2006-09-27 01:50:02 -0700 | [diff] [blame] | 135 | |
| 136 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 137 | } |
| 138 | |
Nick Piggin | 750b498 | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 139 | int truncate_inode_page(struct address_space *mapping, struct page *page) |
| 140 | { |
| 141 | if (page_mapped(page)) { |
| 142 | unmap_mapping_range(mapping, |
| 143 | (loff_t)page->index << PAGE_CACHE_SHIFT, |
| 144 | PAGE_CACHE_SIZE, 0); |
| 145 | } |
| 146 | return truncate_complete_page(mapping, page); |
| 147 | } |
| 148 | |
Wu Fengguang | 83f7866 | 2009-09-16 11:50:13 +0200 | [diff] [blame] | 149 | /* |
Andi Kleen | 2571873 | 2009-09-16 11:50:13 +0200 | [diff] [blame] | 150 | * Used to get rid of pages on hardware memory corruption. |
| 151 | */ |
| 152 | int generic_error_remove_page(struct address_space *mapping, struct page *page) |
| 153 | { |
| 154 | if (!mapping) |
| 155 | return -EINVAL; |
| 156 | /* |
| 157 | * Only punch for normal data pages for now. |
| 158 | * Handling other types like directories would need more auditing. |
| 159 | */ |
| 160 | if (!S_ISREG(mapping->host->i_mode)) |
| 161 | return -EIO; |
| 162 | return truncate_inode_page(mapping, page); |
| 163 | } |
| 164 | EXPORT_SYMBOL(generic_error_remove_page); |
| 165 | |
| 166 | /* |
Wu Fengguang | 83f7866 | 2009-09-16 11:50:13 +0200 | [diff] [blame] | 167 | * Safely invalidate one page from its pagecache mapping. |
| 168 | * It only drops clean, unused pages. The page must be locked. |
| 169 | * |
| 170 | * Returns 1 if the page is successfully invalidated, otherwise 0. |
| 171 | */ |
| 172 | int invalidate_inode_page(struct page *page) |
| 173 | { |
| 174 | struct address_space *mapping = page_mapping(page); |
| 175 | if (!mapping) |
| 176 | return 0; |
| 177 | if (PageDirty(page) || PageWriteback(page)) |
| 178 | return 0; |
| 179 | if (page_mapped(page)) |
| 180 | return 0; |
| 181 | return invalidate_complete_page(mapping, page); |
| 182 | } |
| 183 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 184 | /** |
Randy Dunlap | 0643245 | 2008-02-29 22:03:15 -0800 | [diff] [blame] | 185 | * truncate_inode_pages - truncate range of pages specified by start & end byte offsets |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 186 | * @mapping: mapping to truncate |
| 187 | * @lstart: offset from which to truncate |
Hans Reiser | d733907 | 2006-01-06 00:10:36 -0800 | [diff] [blame] | 188 | * @lend: offset to which to truncate |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 189 | * |
Hans Reiser | d733907 | 2006-01-06 00:10:36 -0800 | [diff] [blame] | 190 | * Truncate the page cache, removing the pages that are between |
| 191 | * specified offsets (and zeroing out partial page |
| 192 | * (if lstart is not page aligned)). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 193 | * |
| 194 | * Truncate takes two passes - the first pass is nonblocking. It will not |
| 195 | * block on page locks and it will not block on writeback. The second pass |
| 196 | * will wait. This is to prevent as much IO as possible in the affected region. |
| 197 | * The first pass will remove most pages, so the search cost of the second pass |
| 198 | * is low. |
| 199 | * |
| 200 | * When looking at page->index outside the page lock we need to be careful to |
| 201 | * copy it into a local to avoid races (it could change at any time). |
| 202 | * |
| 203 | * We pass down the cache-hot hint to the page freeing code. Even if the |
| 204 | * mapping is large, it is probably the case that the final pages are the most |
| 205 | * recently touched, and freeing happens in ascending file offset order. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 206 | */ |
Hans Reiser | d733907 | 2006-01-06 00:10:36 -0800 | [diff] [blame] | 207 | void truncate_inode_pages_range(struct address_space *mapping, |
| 208 | loff_t lstart, loff_t lend) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 209 | { |
| 210 | const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; |
Hans Reiser | d733907 | 2006-01-06 00:10:36 -0800 | [diff] [blame] | 211 | pgoff_t end; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 212 | const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1); |
| 213 | struct pagevec pvec; |
| 214 | pgoff_t next; |
| 215 | int i; |
| 216 | |
| 217 | if (mapping->nrpages == 0) |
| 218 | return; |
| 219 | |
Hans Reiser | d733907 | 2006-01-06 00:10:36 -0800 | [diff] [blame] | 220 | BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1)); |
| 221 | end = (lend >> PAGE_CACHE_SHIFT); |
| 222 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 223 | pagevec_init(&pvec, 0); |
| 224 | next = start; |
Hans Reiser | d733907 | 2006-01-06 00:10:36 -0800 | [diff] [blame] | 225 | while (next <= end && |
| 226 | pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 227 | for (i = 0; i < pagevec_count(&pvec); i++) { |
| 228 | struct page *page = pvec.pages[i]; |
| 229 | pgoff_t page_index = page->index; |
| 230 | |
Hans Reiser | d733907 | 2006-01-06 00:10:36 -0800 | [diff] [blame] | 231 | if (page_index > end) { |
| 232 | next = page_index; |
| 233 | break; |
| 234 | } |
| 235 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 236 | if (page_index > next) |
| 237 | next = page_index; |
| 238 | next++; |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 239 | if (!trylock_page(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 240 | continue; |
| 241 | if (PageWriteback(page)) { |
| 242 | unlock_page(page); |
| 243 | continue; |
| 244 | } |
Nick Piggin | 750b498 | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 245 | truncate_inode_page(mapping, page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 246 | unlock_page(page); |
| 247 | } |
| 248 | pagevec_release(&pvec); |
| 249 | cond_resched(); |
| 250 | } |
| 251 | |
| 252 | if (partial) { |
| 253 | struct page *page = find_lock_page(mapping, start - 1); |
| 254 | if (page) { |
| 255 | wait_on_page_writeback(page); |
| 256 | truncate_partial_page(page, partial); |
| 257 | unlock_page(page); |
| 258 | page_cache_release(page); |
| 259 | } |
| 260 | } |
| 261 | |
| 262 | next = start; |
| 263 | for ( ; ; ) { |
| 264 | cond_resched(); |
| 265 | if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { |
| 266 | if (next == start) |
| 267 | break; |
| 268 | next = start; |
| 269 | continue; |
| 270 | } |
Hans Reiser | d733907 | 2006-01-06 00:10:36 -0800 | [diff] [blame] | 271 | if (pvec.pages[0]->index > end) { |
| 272 | pagevec_release(&pvec); |
| 273 | break; |
| 274 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 275 | for (i = 0; i < pagevec_count(&pvec); i++) { |
| 276 | struct page *page = pvec.pages[i]; |
| 277 | |
Hans Reiser | d733907 | 2006-01-06 00:10:36 -0800 | [diff] [blame] | 278 | if (page->index > end) |
| 279 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 280 | lock_page(page); |
| 281 | wait_on_page_writeback(page); |
Nick Piggin | 750b498 | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 282 | truncate_inode_page(mapping, page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 283 | if (page->index > next) |
| 284 | next = page->index; |
| 285 | next++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 286 | unlock_page(page); |
| 287 | } |
| 288 | pagevec_release(&pvec); |
| 289 | } |
| 290 | } |
Hans Reiser | d733907 | 2006-01-06 00:10:36 -0800 | [diff] [blame] | 291 | EXPORT_SYMBOL(truncate_inode_pages_range); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 292 | |
Hans Reiser | d733907 | 2006-01-06 00:10:36 -0800 | [diff] [blame] | 293 | /** |
| 294 | * truncate_inode_pages - truncate *all* the pages from an offset |
| 295 | * @mapping: mapping to truncate |
| 296 | * @lstart: offset from which to truncate |
| 297 | * |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 298 | * Called under (and serialised by) inode->i_mutex. |
Hans Reiser | d733907 | 2006-01-06 00:10:36 -0800 | [diff] [blame] | 299 | */ |
| 300 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) |
| 301 | { |
| 302 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); |
| 303 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 304 | EXPORT_SYMBOL(truncate_inode_pages); |
| 305 | |
Mike Waychison | 2869735 | 2009-06-16 15:32:59 -0700 | [diff] [blame] | 306 | /** |
| 307 | * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode |
| 308 | * @mapping: the address_space which holds the pages to invalidate |
| 309 | * @start: the offset 'from' which to invalidate |
| 310 | * @end: the offset 'to' which to invalidate (inclusive) |
| 311 | * |
| 312 | * This function only removes the unlocked pages, if you want to |
| 313 | * remove all the pages of one inode, you must call truncate_inode_pages. |
| 314 | * |
| 315 | * invalidate_mapping_pages() will not block on IO activity. It will not |
| 316 | * invalidate pages which are dirty, locked, under writeback or mapped into |
| 317 | * pagetables. |
| 318 | */ |
| 319 | unsigned long invalidate_mapping_pages(struct address_space *mapping, |
| 320 | pgoff_t start, pgoff_t end) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 321 | { |
| 322 | struct pagevec pvec; |
| 323 | pgoff_t next = start; |
| 324 | unsigned long ret = 0; |
| 325 | int i; |
| 326 | |
| 327 | pagevec_init(&pvec, 0); |
| 328 | while (next <= end && |
| 329 | pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { |
| 330 | for (i = 0; i < pagevec_count(&pvec); i++) { |
| 331 | struct page *page = pvec.pages[i]; |
NeilBrown | e0f23603 | 2006-06-23 02:05:48 -0700 | [diff] [blame] | 332 | pgoff_t index; |
| 333 | int lock_failed; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 334 | |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 335 | lock_failed = !trylock_page(page); |
NeilBrown | e0f23603 | 2006-06-23 02:05:48 -0700 | [diff] [blame] | 336 | |
| 337 | /* |
| 338 | * We really shouldn't be looking at the ->index of an |
| 339 | * unlocked page. But we're not allowed to lock these |
| 340 | * pages. So we rely upon nobody altering the ->index |
| 341 | * of this (pinned-by-us) page. |
| 342 | */ |
| 343 | index = page->index; |
| 344 | if (index > next) |
| 345 | next = index; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 346 | next++; |
NeilBrown | e0f23603 | 2006-06-23 02:05:48 -0700 | [diff] [blame] | 347 | if (lock_failed) |
| 348 | continue; |
| 349 | |
Wu Fengguang | 83f7866 | 2009-09-16 11:50:13 +0200 | [diff] [blame] | 350 | ret += invalidate_inode_page(page); |
| 351 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 352 | unlock_page(page); |
| 353 | if (next > end) |
| 354 | break; |
| 355 | } |
| 356 | pagevec_release(&pvec); |
Mike Waychison | 2869735 | 2009-06-16 15:32:59 -0700 | [diff] [blame] | 357 | cond_resched(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 358 | } |
| 359 | return ret; |
| 360 | } |
Anton Altaparmakov | 54bc485 | 2007-02-10 01:45:38 -0800 | [diff] [blame] | 361 | EXPORT_SYMBOL(invalidate_mapping_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 362 | |
Andrew Morton | bd4c8ce | 2006-09-30 23:29:29 -0700 | [diff] [blame] | 363 | /* |
| 364 | * This is like invalidate_complete_page(), except it ignores the page's |
| 365 | * refcount. We do this because invalidate_inode_pages2() needs stronger |
| 366 | * invalidation guarantees, and cannot afford to leave pages behind because |
Anderson Briglia | 2706a1b | 2007-07-15 23:38:09 -0700 | [diff] [blame] | 367 | * shrink_page_list() has a temp ref on them, or because they're transiently |
| 368 | * sitting in the lru_cache_add() pagevecs. |
Andrew Morton | bd4c8ce | 2006-09-30 23:29:29 -0700 | [diff] [blame] | 369 | */ |
| 370 | static int |
| 371 | invalidate_complete_page2(struct address_space *mapping, struct page *page) |
| 372 | { |
| 373 | if (page->mapping != mapping) |
| 374 | return 0; |
| 375 | |
David Howells | 266cf65 | 2009-04-03 16:42:36 +0100 | [diff] [blame] | 376 | if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) |
Andrew Morton | bd4c8ce | 2006-09-30 23:29:29 -0700 | [diff] [blame] | 377 | return 0; |
| 378 | |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 379 | spin_lock_irq(&mapping->tree_lock); |
Andrew Morton | bd4c8ce | 2006-09-30 23:29:29 -0700 | [diff] [blame] | 380 | if (PageDirty(page)) |
| 381 | goto failed; |
| 382 | |
Rik van Riel | ba470de | 2008-10-18 20:26:50 -0700 | [diff] [blame] | 383 | clear_page_mlock(page); |
David Howells | 266cf65 | 2009-04-03 16:42:36 +0100 | [diff] [blame] | 384 | BUG_ON(page_has_private(page)); |
Andrew Morton | bd4c8ce | 2006-09-30 23:29:29 -0700 | [diff] [blame] | 385 | __remove_from_page_cache(page); |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 386 | spin_unlock_irq(&mapping->tree_lock); |
Daisuke Nishimura | e767e05 | 2009-05-28 14:34:28 -0700 | [diff] [blame] | 387 | mem_cgroup_uncharge_cache_page(page); |
Andrew Morton | bd4c8ce | 2006-09-30 23:29:29 -0700 | [diff] [blame] | 388 | page_cache_release(page); /* pagecache ref */ |
| 389 | return 1; |
| 390 | failed: |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 391 | spin_unlock_irq(&mapping->tree_lock); |
Andrew Morton | bd4c8ce | 2006-09-30 23:29:29 -0700 | [diff] [blame] | 392 | return 0; |
| 393 | } |
| 394 | |
Trond Myklebust | e3db769 | 2007-01-10 23:15:39 -0800 | [diff] [blame] | 395 | static int do_launder_page(struct address_space *mapping, struct page *page) |
| 396 | { |
| 397 | if (!PageDirty(page)) |
| 398 | return 0; |
| 399 | if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) |
| 400 | return 0; |
| 401 | return mapping->a_ops->launder_page(page); |
| 402 | } |
| 403 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 404 | /** |
| 405 | * invalidate_inode_pages2_range - remove range of pages from an address_space |
Martin Waitz | 67be2dd | 2005-05-01 08:59:26 -0700 | [diff] [blame] | 406 | * @mapping: the address_space |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 407 | * @start: the page offset 'from' which to invalidate |
| 408 | * @end: the page offset 'to' which to invalidate (inclusive) |
| 409 | * |
| 410 | * Any pages which are found to be mapped into pagetables are unmapped prior to |
| 411 | * invalidation. |
| 412 | * |
Hisashi Hifumi | 6ccfa80 | 2008-09-02 14:35:40 -0700 | [diff] [blame] | 413 | * Returns -EBUSY if any pages could not be invalidated. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 414 | */ |
| 415 | int invalidate_inode_pages2_range(struct address_space *mapping, |
| 416 | pgoff_t start, pgoff_t end) |
| 417 | { |
| 418 | struct pagevec pvec; |
| 419 | pgoff_t next; |
| 420 | int i; |
| 421 | int ret = 0; |
Hisashi Hifumi | 0dd1334 | 2008-04-28 02:12:08 -0700 | [diff] [blame] | 422 | int ret2 = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 423 | int did_range_unmap = 0; |
| 424 | int wrapped = 0; |
| 425 | |
| 426 | pagevec_init(&pvec, 0); |
| 427 | next = start; |
Trond Myklebust | 7b965e0 | 2007-02-28 20:13:55 -0800 | [diff] [blame] | 428 | while (next <= end && !wrapped && |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 429 | pagevec_lookup(&pvec, mapping, next, |
| 430 | min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) { |
Trond Myklebust | 7b965e0 | 2007-02-28 20:13:55 -0800 | [diff] [blame] | 431 | for (i = 0; i < pagevec_count(&pvec); i++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 432 | struct page *page = pvec.pages[i]; |
| 433 | pgoff_t page_index; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 434 | |
| 435 | lock_page(page); |
| 436 | if (page->mapping != mapping) { |
| 437 | unlock_page(page); |
| 438 | continue; |
| 439 | } |
| 440 | page_index = page->index; |
| 441 | next = page_index + 1; |
| 442 | if (next == 0) |
| 443 | wrapped = 1; |
| 444 | if (page_index > end) { |
| 445 | unlock_page(page); |
| 446 | break; |
| 447 | } |
| 448 | wait_on_page_writeback(page); |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 449 | if (page_mapped(page)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 450 | if (!did_range_unmap) { |
| 451 | /* |
| 452 | * Zap the rest of the file in one hit. |
| 453 | */ |
| 454 | unmap_mapping_range(mapping, |
Oleg Drokin | 479ef59 | 2005-11-23 13:37:47 -0800 | [diff] [blame] | 455 | (loff_t)page_index<<PAGE_CACHE_SHIFT, |
| 456 | (loff_t)(end - page_index + 1) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 457 | << PAGE_CACHE_SHIFT, |
| 458 | 0); |
| 459 | did_range_unmap = 1; |
| 460 | } else { |
| 461 | /* |
| 462 | * Just zap this page |
| 463 | */ |
| 464 | unmap_mapping_range(mapping, |
Oleg Drokin | 479ef59 | 2005-11-23 13:37:47 -0800 | [diff] [blame] | 465 | (loff_t)page_index<<PAGE_CACHE_SHIFT, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 466 | PAGE_CACHE_SIZE, 0); |
| 467 | } |
| 468 | } |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 469 | BUG_ON(page_mapped(page)); |
Hisashi Hifumi | 0dd1334 | 2008-04-28 02:12:08 -0700 | [diff] [blame] | 470 | ret2 = do_launder_page(mapping, page); |
| 471 | if (ret2 == 0) { |
| 472 | if (!invalidate_complete_page2(mapping, page)) |
Hisashi Hifumi | 6ccfa80 | 2008-09-02 14:35:40 -0700 | [diff] [blame] | 473 | ret2 = -EBUSY; |
Hisashi Hifumi | 0dd1334 | 2008-04-28 02:12:08 -0700 | [diff] [blame] | 474 | } |
| 475 | if (ret2 < 0) |
| 476 | ret = ret2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 477 | unlock_page(page); |
| 478 | } |
| 479 | pagevec_release(&pvec); |
| 480 | cond_resched(); |
| 481 | } |
| 482 | return ret; |
| 483 | } |
| 484 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); |
| 485 | |
| 486 | /** |
| 487 | * invalidate_inode_pages2 - remove all pages from an address_space |
Martin Waitz | 67be2dd | 2005-05-01 08:59:26 -0700 | [diff] [blame] | 488 | * @mapping: the address_space |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 489 | * |
| 490 | * Any pages which are found to be mapped into pagetables are unmapped prior to |
| 491 | * invalidation. |
| 492 | * |
Peng Tao | e9de25d | 2009-10-19 14:48:13 +0800 | [diff] [blame] | 493 | * Returns -EBUSY if any pages could not be invalidated. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 494 | */ |
| 495 | int invalidate_inode_pages2(struct address_space *mapping) |
| 496 | { |
| 497 | return invalidate_inode_pages2_range(mapping, 0, -1); |
| 498 | } |
| 499 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); |
npiggin@suse.de | 25d9e2d | 2009-08-21 02:35:05 +1000 | [diff] [blame] | 500 | |
| 501 | /** |
| 502 | * truncate_pagecache - unmap and remove pagecache that has been truncated |
| 503 | * @inode: inode |
| 504 | * @old: old file offset |
| 505 | * @new: new file offset |
| 506 | * |
| 507 | * inode's new i_size must already be written before truncate_pagecache |
| 508 | * is called. |
| 509 | * |
| 510 | * This function should typically be called before the filesystem |
| 511 | * releases resources associated with the freed range (eg. deallocates |
| 512 | * blocks). This way, pagecache will always stay logically coherent |
| 513 | * with on-disk format, and the filesystem would not have to deal with |
| 514 | * situations such as writepage being called for a page that has already |
| 515 | * had its underlying blocks deallocated. |
| 516 | */ |
| 517 | void truncate_pagecache(struct inode *inode, loff_t old, loff_t new) |
| 518 | { |
| 519 | if (new < old) { |
| 520 | struct address_space *mapping = inode->i_mapping; |
| 521 | |
| 522 | /* |
| 523 | * unmap_mapping_range is called twice, first simply for |
| 524 | * efficiency so that truncate_inode_pages does fewer |
| 525 | * single-page unmaps. However after this first call, and |
| 526 | * before truncate_inode_pages finishes, it is possible for |
| 527 | * private pages to be COWed, which remain after |
| 528 | * truncate_inode_pages finishes, hence the second |
| 529 | * unmap_mapping_range call must be made for correctness. |
| 530 | */ |
| 531 | unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1); |
| 532 | truncate_inode_pages(mapping, new); |
| 533 | unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1); |
| 534 | } |
| 535 | } |
| 536 | EXPORT_SYMBOL(truncate_pagecache); |
| 537 | |
| 538 | /** |
| 539 | * vmtruncate - unmap mappings "freed" by truncate() syscall |
| 540 | * @inode: inode of the file used |
| 541 | * @offset: file offset to start truncating |
| 542 | * |
| 543 | * NOTE! We have to be ready to update the memory sharing |
| 544 | * between the file and the memory map for a potential last |
| 545 | * incomplete page. Ugly, but necessary. |
| 546 | */ |
| 547 | int vmtruncate(struct inode *inode, loff_t offset) |
| 548 | { |
| 549 | loff_t oldsize; |
| 550 | int error; |
| 551 | |
| 552 | error = inode_newsize_ok(inode, offset); |
| 553 | if (error) |
| 554 | return error; |
| 555 | oldsize = inode->i_size; |
| 556 | i_size_write(inode, offset); |
| 557 | truncate_pagecache(inode, oldsize, offset); |
| 558 | if (inode->i_op->truncate) |
| 559 | inode->i_op->truncate(inode); |
| 560 | |
| 561 | return error; |
| 562 | } |
| 563 | EXPORT_SYMBOL(vmtruncate); |