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> |
Tejun Heo | 5a0e3ad | 2010-03-24 17:04:11 +0900 | [diff] [blame] | 12 | #include <linux/gfp.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 13 | #include <linux/mm.h> |
Nick Piggin | 0fd0e6b | 2006-09-27 01:50:02 -0700 | [diff] [blame] | 14 | #include <linux/swap.h> |
Paul Gortmaker | b95f1b31 | 2011-10-16 02:01:52 -0400 | [diff] [blame] | 15 | #include <linux/export.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 16 | #include <linux/pagemap.h> |
Nate Diller | 01f2705 | 2007-05-09 02:35:07 -0700 | [diff] [blame] | 17 | #include <linux/highmem.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 18 | #include <linux/pagevec.h> |
Andrew Morton | e08748ce | 2006-12-10 02:19:31 -0800 | [diff] [blame] | 19 | #include <linux/task_io_accounting_ops.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 20 | #include <linux/buffer_head.h> /* grr. try_to_release_page, |
Jan Kara | aaa4059 | 2005-10-30 15:00:16 -0800 | [diff] [blame] | 21 | do_invalidatepage */ |
Dan Magenheimer | c515e1f | 2011-05-26 10:01:43 -0600 | [diff] [blame] | 22 | #include <linux/cleancache.h> |
Rik van Riel | ba470de | 2008-10-18 20:26:50 -0700 | [diff] [blame] | 23 | #include "internal.h" |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 24 | |
| 25 | |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 26 | /** |
Fengguang Wu | 28bc44d | 2008-02-03 18:04:10 +0200 | [diff] [blame] | 27 | * do_invalidatepage - invalidate part or all of a page |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 28 | * @page: the page which is affected |
Lukas Czerner | d47992f | 2013-05-21 23:17:23 -0400 | [diff] [blame] | 29 | * @offset: start of the range to invalidate |
| 30 | * @length: length of the range to invalidate |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 31 | * |
| 32 | * do_invalidatepage() is called when all or part of the page has become |
| 33 | * invalidated by a truncate operation. |
| 34 | * |
| 35 | * do_invalidatepage() does not have to release all buffers, but it must |
| 36 | * ensure that no dirty buffer is left outside @offset and that no I/O |
| 37 | * is underway against any of the blocks which are outside the truncation |
| 38 | * point. Because the caller is about to free (and possibly reuse) those |
| 39 | * blocks on-disk. |
| 40 | */ |
Lukas Czerner | d47992f | 2013-05-21 23:17:23 -0400 | [diff] [blame] | 41 | void do_invalidatepage(struct page *page, unsigned int offset, |
| 42 | unsigned int length) |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 43 | { |
Lukas Czerner | d47992f | 2013-05-21 23:17:23 -0400 | [diff] [blame] | 44 | void (*invalidatepage)(struct page *, unsigned int, unsigned int); |
| 45 | |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 46 | invalidatepage = page->mapping->a_ops->invalidatepage; |
David Howells | 9361401 | 2006-09-30 20:45:40 +0200 | [diff] [blame] | 47 | #ifdef CONFIG_BLOCK |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 48 | if (!invalidatepage) |
| 49 | invalidatepage = block_invalidatepage; |
David Howells | 9361401 | 2006-09-30 20:45:40 +0200 | [diff] [blame] | 50 | #endif |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 51 | if (invalidatepage) |
Lukas Czerner | d47992f | 2013-05-21 23:17:23 -0400 | [diff] [blame] | 52 | (*invalidatepage)(page, offset, length); |
David Howells | cf9a2ae | 2006-08-29 19:05:54 +0100 | [diff] [blame] | 53 | } |
| 54 | |
Linus Torvalds | ecdfc97 | 2007-01-26 12:47:06 -0800 | [diff] [blame] | 55 | /* |
| 56 | * This cancels just the dirty bit on the kernel page itself, it |
| 57 | * does NOT actually remove dirty bits on any mmap's that may be |
| 58 | * around. It also leaves the page tagged dirty, so any sync |
| 59 | * activity will still find it on the dirty lists, and in particular, |
| 60 | * clear_page_dirty_for_io() will still look at the dirty bits in |
| 61 | * the VM. |
| 62 | * |
| 63 | * Doing this should *normally* only ever be done when a page |
| 64 | * is truncated, and is not actually mapped anywhere at all. However, |
| 65 | * fs/buffer.c does this when it notices that somebody has cleaned |
| 66 | * out all the buffers on a page without actually doing it through |
| 67 | * the VM. Can you say "ext3 is horribly ugly"? Tought you could. |
| 68 | */ |
Linus Torvalds | fba2591 | 2006-12-20 13:46:42 -0800 | [diff] [blame] | 69 | void cancel_dirty_page(struct page *page, unsigned int account_size) |
| 70 | { |
Linus Torvalds | 8368e32 | 2006-12-23 09:25:04 -0800 | [diff] [blame] | 71 | if (TestClearPageDirty(page)) { |
| 72 | struct address_space *mapping = page->mapping; |
| 73 | if (mapping && mapping_cap_account_dirty(mapping)) { |
| 74 | dec_zone_page_state(page, NR_FILE_DIRTY); |
Peter Zijlstra | c9e51e4 | 2007-10-16 23:25:47 -0700 | [diff] [blame] | 75 | dec_bdi_stat(mapping->backing_dev_info, |
| 76 | BDI_RECLAIMABLE); |
Linus Torvalds | 8368e32 | 2006-12-23 09:25:04 -0800 | [diff] [blame] | 77 | if (account_size) |
| 78 | task_io_account_cancelled_write(account_size); |
| 79 | } |
Andrew Morton | 3e67c09 | 2006-12-21 11:00:33 -0800 | [diff] [blame] | 80 | } |
Linus Torvalds | fba2591 | 2006-12-20 13:46:42 -0800 | [diff] [blame] | 81 | } |
Linus Torvalds | 8368e32 | 2006-12-23 09:25:04 -0800 | [diff] [blame] | 82 | EXPORT_SYMBOL(cancel_dirty_page); |
Linus Torvalds | fba2591 | 2006-12-20 13:46:42 -0800 | [diff] [blame] | 83 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 84 | /* |
| 85 | * 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] | 86 | * 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] | 87 | * user pagetables if we're racing with filemap_fault(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 88 | * |
| 89 | * We need to bale out if page->mapping is no longer equal to the original |
| 90 | * 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] | 91 | * 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] | 92 | * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. |
| 93 | */ |
Nick Piggin | 750b498 | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 94 | static int |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 95 | truncate_complete_page(struct address_space *mapping, struct page *page) |
| 96 | { |
| 97 | if (page->mapping != mapping) |
Nick Piggin | 750b498 | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 98 | return -EIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 99 | |
David Howells | 266cf65 | 2009-04-03 16:42:36 +0100 | [diff] [blame] | 100 | if (page_has_private(page)) |
Lukas Czerner | d47992f | 2013-05-21 23:17:23 -0400 | [diff] [blame] | 101 | do_invalidatepage(page, 0, PAGE_CACHE_SIZE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 102 | |
Bjorn Steinbrink | a2b3456 | 2008-02-04 22:29:28 -0800 | [diff] [blame] | 103 | cancel_dirty_page(page, PAGE_CACHE_SIZE); |
| 104 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 105 | ClearPageMappedToDisk(page); |
Minchan Kim | 5adc7b5 | 2011-03-22 16:32:41 -0700 | [diff] [blame] | 106 | delete_from_page_cache(page); |
Nick Piggin | 750b498 | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 107 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 108 | } |
| 109 | |
| 110 | /* |
Andrew Morton | fc0ecff | 2007-02-10 01:45:39 -0800 | [diff] [blame] | 111 | * This is for invalidate_mapping_pages(). That function can be called at |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 112 | * 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] | 113 | * be marked dirty at any time too, so use remove_mapping which safely |
| 114 | * discards clean, unused pages. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 115 | * |
| 116 | * Returns non-zero if the page was successfully invalidated. |
| 117 | */ |
| 118 | static int |
| 119 | invalidate_complete_page(struct address_space *mapping, struct page *page) |
| 120 | { |
Nick Piggin | 0fd0e6b | 2006-09-27 01:50:02 -0700 | [diff] [blame] | 121 | int ret; |
| 122 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 123 | if (page->mapping != mapping) |
| 124 | return 0; |
| 125 | |
David Howells | 266cf65 | 2009-04-03 16:42:36 +0100 | [diff] [blame] | 126 | if (page_has_private(page) && !try_to_release_page(page, 0)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 127 | return 0; |
| 128 | |
Nick Piggin | 0fd0e6b | 2006-09-27 01:50:02 -0700 | [diff] [blame] | 129 | ret = remove_mapping(mapping, page); |
Nick Piggin | 0fd0e6b | 2006-09-27 01:50:02 -0700 | [diff] [blame] | 130 | |
| 131 | return ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 132 | } |
| 133 | |
Nick Piggin | 750b498 | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 134 | int truncate_inode_page(struct address_space *mapping, struct page *page) |
| 135 | { |
| 136 | if (page_mapped(page)) { |
| 137 | unmap_mapping_range(mapping, |
| 138 | (loff_t)page->index << PAGE_CACHE_SHIFT, |
| 139 | PAGE_CACHE_SIZE, 0); |
| 140 | } |
| 141 | return truncate_complete_page(mapping, page); |
| 142 | } |
| 143 | |
Wu Fengguang | 83f7866 | 2009-09-16 11:50:13 +0200 | [diff] [blame] | 144 | /* |
Andi Kleen | 2571873 | 2009-09-16 11:50:13 +0200 | [diff] [blame] | 145 | * Used to get rid of pages on hardware memory corruption. |
| 146 | */ |
| 147 | int generic_error_remove_page(struct address_space *mapping, struct page *page) |
| 148 | { |
| 149 | if (!mapping) |
| 150 | return -EINVAL; |
| 151 | /* |
| 152 | * Only punch for normal data pages for now. |
| 153 | * Handling other types like directories would need more auditing. |
| 154 | */ |
| 155 | if (!S_ISREG(mapping->host->i_mode)) |
| 156 | return -EIO; |
| 157 | return truncate_inode_page(mapping, page); |
| 158 | } |
| 159 | EXPORT_SYMBOL(generic_error_remove_page); |
| 160 | |
| 161 | /* |
Wu Fengguang | 83f7866 | 2009-09-16 11:50:13 +0200 | [diff] [blame] | 162 | * Safely invalidate one page from its pagecache mapping. |
| 163 | * It only drops clean, unused pages. The page must be locked. |
| 164 | * |
| 165 | * Returns 1 if the page is successfully invalidated, otherwise 0. |
| 166 | */ |
| 167 | int invalidate_inode_page(struct page *page) |
| 168 | { |
| 169 | struct address_space *mapping = page_mapping(page); |
| 170 | if (!mapping) |
| 171 | return 0; |
| 172 | if (PageDirty(page) || PageWriteback(page)) |
| 173 | return 0; |
| 174 | if (page_mapped(page)) |
| 175 | return 0; |
| 176 | return invalidate_complete_page(mapping, page); |
| 177 | } |
| 178 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 179 | /** |
Liu Bo | 73c1e20 | 2012-02-21 10:57:20 +0800 | [diff] [blame] | 180 | * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 181 | * @mapping: mapping to truncate |
| 182 | * @lstart: offset from which to truncate |
Lukas Czerner | 5a72039 | 2013-05-27 23:32:35 -0400 | [diff] [blame] | 183 | * @lend: offset to which to truncate (inclusive) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 184 | * |
Hans Reiser | d733907 | 2006-01-06 00:10:36 -0800 | [diff] [blame] | 185 | * Truncate the page cache, removing the pages that are between |
Lukas Czerner | 5a72039 | 2013-05-27 23:32:35 -0400 | [diff] [blame] | 186 | * specified offsets (and zeroing out partial pages |
| 187 | * if lstart or lend + 1 is not page aligned). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 188 | * |
| 189 | * Truncate takes two passes - the first pass is nonblocking. It will not |
| 190 | * block on page locks and it will not block on writeback. The second pass |
| 191 | * will wait. This is to prevent as much IO as possible in the affected region. |
| 192 | * The first pass will remove most pages, so the search cost of the second pass |
| 193 | * is low. |
| 194 | * |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 195 | * We pass down the cache-hot hint to the page freeing code. Even if the |
| 196 | * mapping is large, it is probably the case that the final pages are the most |
| 197 | * recently touched, and freeing happens in ascending file offset order. |
Lukas Czerner | 5a72039 | 2013-05-27 23:32:35 -0400 | [diff] [blame] | 198 | * |
| 199 | * Note that since ->invalidatepage() accepts range to invalidate |
| 200 | * truncate_inode_pages_range is able to handle cases where lend + 1 is not |
| 201 | * page aligned properly. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 202 | */ |
Hans Reiser | d733907 | 2006-01-06 00:10:36 -0800 | [diff] [blame] | 203 | void truncate_inode_pages_range(struct address_space *mapping, |
| 204 | loff_t lstart, loff_t lend) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 205 | { |
Lukas Czerner | 5a72039 | 2013-05-27 23:32:35 -0400 | [diff] [blame] | 206 | pgoff_t start; /* inclusive */ |
| 207 | pgoff_t end; /* exclusive */ |
| 208 | unsigned int partial_start; /* inclusive */ |
| 209 | unsigned int partial_end; /* exclusive */ |
| 210 | struct pagevec pvec; |
| 211 | pgoff_t index; |
| 212 | int i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 213 | |
Dan Magenheimer | 3167760 | 2011-09-21 11:56:28 -0400 | [diff] [blame] | 214 | cleancache_invalidate_inode(mapping); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 215 | if (mapping->nrpages == 0) |
| 216 | return; |
| 217 | |
Lukas Czerner | 5a72039 | 2013-05-27 23:32:35 -0400 | [diff] [blame] | 218 | /* Offsets within partial pages */ |
| 219 | partial_start = lstart & (PAGE_CACHE_SIZE - 1); |
| 220 | partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1); |
| 221 | |
| 222 | /* |
| 223 | * 'start' and 'end' always covers the range of pages to be fully |
| 224 | * truncated. Partial pages are covered with 'partial_start' at the |
| 225 | * start of the range and 'partial_end' at the end of the range. |
| 226 | * Note that 'end' is exclusive while 'lend' is inclusive. |
| 227 | */ |
| 228 | start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; |
| 229 | if (lend == -1) |
| 230 | /* |
| 231 | * lend == -1 indicates end-of-file so we have to set 'end' |
| 232 | * to the highest possible pgoff_t and since the type is |
| 233 | * unsigned we're using -1. |
| 234 | */ |
| 235 | end = -1; |
| 236 | else |
| 237 | end = (lend + 1) >> PAGE_CACHE_SHIFT; |
Hans Reiser | d733907 | 2006-01-06 00:10:36 -0800 | [diff] [blame] | 238 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 239 | pagevec_init(&pvec, 0); |
Hugh Dickins | b85e0ef | 2011-07-25 17:12:25 -0700 | [diff] [blame] | 240 | index = start; |
Lukas Czerner | 5a72039 | 2013-05-27 23:32:35 -0400 | [diff] [blame] | 241 | while (index < end && pagevec_lookup(&pvec, mapping, index, |
| 242 | min(end - index, (pgoff_t)PAGEVEC_SIZE))) { |
Hugh Dickins | e5598f8 | 2011-02-25 14:44:29 -0800 | [diff] [blame] | 243 | mem_cgroup_uncharge_start(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 244 | for (i = 0; i < pagevec_count(&pvec); i++) { |
| 245 | struct page *page = pvec.pages[i]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 246 | |
Hugh Dickins | b85e0ef | 2011-07-25 17:12:25 -0700 | [diff] [blame] | 247 | /* We rely upon deletion not changing page->index */ |
| 248 | index = page->index; |
Lukas Czerner | 5a72039 | 2013-05-27 23:32:35 -0400 | [diff] [blame] | 249 | if (index >= end) |
Hans Reiser | d733907 | 2006-01-06 00:10:36 -0800 | [diff] [blame] | 250 | break; |
Hans Reiser | d733907 | 2006-01-06 00:10:36 -0800 | [diff] [blame] | 251 | |
Nick Piggin | 529ae9a | 2008-08-02 12:01:03 +0200 | [diff] [blame] | 252 | if (!trylock_page(page)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 253 | continue; |
Hugh Dickins | b85e0ef | 2011-07-25 17:12:25 -0700 | [diff] [blame] | 254 | WARN_ON(page->index != index); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 255 | if (PageWriteback(page)) { |
| 256 | unlock_page(page); |
| 257 | continue; |
| 258 | } |
Nick Piggin | 750b498 | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 259 | truncate_inode_page(mapping, page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 260 | unlock_page(page); |
| 261 | } |
| 262 | pagevec_release(&pvec); |
Hugh Dickins | e5598f8 | 2011-02-25 14:44:29 -0800 | [diff] [blame] | 263 | mem_cgroup_uncharge_end(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 264 | cond_resched(); |
Hugh Dickins | b85e0ef | 2011-07-25 17:12:25 -0700 | [diff] [blame] | 265 | index++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 266 | } |
| 267 | |
Lukas Czerner | 5a72039 | 2013-05-27 23:32:35 -0400 | [diff] [blame] | 268 | if (partial_start) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 269 | struct page *page = find_lock_page(mapping, start - 1); |
| 270 | if (page) { |
Lukas Czerner | 5a72039 | 2013-05-27 23:32:35 -0400 | [diff] [blame] | 271 | unsigned int top = PAGE_CACHE_SIZE; |
| 272 | if (start > end) { |
| 273 | /* Truncation within a single page */ |
| 274 | top = partial_end; |
| 275 | partial_end = 0; |
| 276 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 277 | wait_on_page_writeback(page); |
Lukas Czerner | 5a72039 | 2013-05-27 23:32:35 -0400 | [diff] [blame] | 278 | zero_user_segment(page, partial_start, top); |
| 279 | cleancache_invalidate_page(mapping, page); |
| 280 | if (page_has_private(page)) |
| 281 | do_invalidatepage(page, partial_start, |
| 282 | top - partial_start); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 283 | unlock_page(page); |
| 284 | page_cache_release(page); |
| 285 | } |
| 286 | } |
Lukas Czerner | 5a72039 | 2013-05-27 23:32:35 -0400 | [diff] [blame] | 287 | if (partial_end) { |
| 288 | struct page *page = find_lock_page(mapping, end); |
| 289 | if (page) { |
| 290 | wait_on_page_writeback(page); |
| 291 | zero_user_segment(page, 0, partial_end); |
| 292 | cleancache_invalidate_page(mapping, page); |
| 293 | if (page_has_private(page)) |
| 294 | do_invalidatepage(page, 0, |
| 295 | partial_end); |
| 296 | unlock_page(page); |
| 297 | page_cache_release(page); |
| 298 | } |
| 299 | } |
| 300 | /* |
| 301 | * If the truncation happened within a single page no pages |
| 302 | * will be released, just zeroed, so we can bail out now. |
| 303 | */ |
| 304 | if (start >= end) |
| 305 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 306 | |
Hugh Dickins | b85e0ef | 2011-07-25 17:12:25 -0700 | [diff] [blame] | 307 | index = start; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 308 | for ( ; ; ) { |
| 309 | cond_resched(); |
Hugh Dickins | b85e0ef | 2011-07-25 17:12:25 -0700 | [diff] [blame] | 310 | if (!pagevec_lookup(&pvec, mapping, index, |
Lukas Czerner | 5a72039 | 2013-05-27 23:32:35 -0400 | [diff] [blame] | 311 | min(end - index, (pgoff_t)PAGEVEC_SIZE))) { |
Hugh Dickins | b85e0ef | 2011-07-25 17:12:25 -0700 | [diff] [blame] | 312 | if (index == start) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 313 | break; |
Hugh Dickins | b85e0ef | 2011-07-25 17:12:25 -0700 | [diff] [blame] | 314 | index = start; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 315 | continue; |
| 316 | } |
Lukas Czerner | 5a72039 | 2013-05-27 23:32:35 -0400 | [diff] [blame] | 317 | if (index == start && pvec.pages[0]->index >= end) { |
Hans Reiser | d733907 | 2006-01-06 00:10:36 -0800 | [diff] [blame] | 318 | pagevec_release(&pvec); |
| 319 | break; |
| 320 | } |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 321 | mem_cgroup_uncharge_start(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 322 | for (i = 0; i < pagevec_count(&pvec); i++) { |
| 323 | struct page *page = pvec.pages[i]; |
| 324 | |
Hugh Dickins | b85e0ef | 2011-07-25 17:12:25 -0700 | [diff] [blame] | 325 | /* We rely upon deletion not changing page->index */ |
| 326 | index = page->index; |
Lukas Czerner | 5a72039 | 2013-05-27 23:32:35 -0400 | [diff] [blame] | 327 | if (index >= end) |
Hans Reiser | d733907 | 2006-01-06 00:10:36 -0800 | [diff] [blame] | 328 | break; |
Hugh Dickins | b85e0ef | 2011-07-25 17:12:25 -0700 | [diff] [blame] | 329 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 330 | lock_page(page); |
Hugh Dickins | b85e0ef | 2011-07-25 17:12:25 -0700 | [diff] [blame] | 331 | WARN_ON(page->index != index); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 332 | wait_on_page_writeback(page); |
Nick Piggin | 750b498 | 2009-09-16 11:50:12 +0200 | [diff] [blame] | 333 | truncate_inode_page(mapping, page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 334 | unlock_page(page); |
| 335 | } |
| 336 | pagevec_release(&pvec); |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 337 | mem_cgroup_uncharge_end(); |
Hugh Dickins | b85e0ef | 2011-07-25 17:12:25 -0700 | [diff] [blame] | 338 | index++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 339 | } |
Dan Magenheimer | 3167760 | 2011-09-21 11:56:28 -0400 | [diff] [blame] | 340 | cleancache_invalidate_inode(mapping); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 341 | } |
Hans Reiser | d733907 | 2006-01-06 00:10:36 -0800 | [diff] [blame] | 342 | EXPORT_SYMBOL(truncate_inode_pages_range); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 343 | |
Hans Reiser | d733907 | 2006-01-06 00:10:36 -0800 | [diff] [blame] | 344 | /** |
| 345 | * truncate_inode_pages - truncate *all* the pages from an offset |
| 346 | * @mapping: mapping to truncate |
| 347 | * @lstart: offset from which to truncate |
| 348 | * |
Jes Sorensen | 1b1dcc1 | 2006-01-09 15:59:24 -0800 | [diff] [blame] | 349 | * Called under (and serialised by) inode->i_mutex. |
Jan Kara | 0814257 | 2011-06-27 16:18:10 -0700 | [diff] [blame] | 350 | * |
| 351 | * Note: When this function returns, there can be a page in the process of |
| 352 | * deletion (inside __delete_from_page_cache()) in the specified range. Thus |
| 353 | * mapping->nrpages can be non-zero when this function returns even after |
| 354 | * truncation of the whole mapping. |
Hans Reiser | d733907 | 2006-01-06 00:10:36 -0800 | [diff] [blame] | 355 | */ |
| 356 | void truncate_inode_pages(struct address_space *mapping, loff_t lstart) |
| 357 | { |
| 358 | truncate_inode_pages_range(mapping, lstart, (loff_t)-1); |
| 359 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 360 | EXPORT_SYMBOL(truncate_inode_pages); |
| 361 | |
Mike Waychison | 2869735 | 2009-06-16 15:32:59 -0700 | [diff] [blame] | 362 | /** |
| 363 | * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode |
| 364 | * @mapping: the address_space which holds the pages to invalidate |
| 365 | * @start: the offset 'from' which to invalidate |
| 366 | * @end: the offset 'to' which to invalidate (inclusive) |
| 367 | * |
| 368 | * This function only removes the unlocked pages, if you want to |
| 369 | * remove all the pages of one inode, you must call truncate_inode_pages. |
| 370 | * |
| 371 | * invalidate_mapping_pages() will not block on IO activity. It will not |
| 372 | * invalidate pages which are dirty, locked, under writeback or mapped into |
| 373 | * pagetables. |
| 374 | */ |
| 375 | unsigned long invalidate_mapping_pages(struct address_space *mapping, |
Minchan Kim | 3156018 | 2011-03-22 16:32:52 -0700 | [diff] [blame] | 376 | pgoff_t start, pgoff_t end) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 377 | { |
| 378 | struct pagevec pvec; |
Hugh Dickins | b85e0ef | 2011-07-25 17:12:25 -0700 | [diff] [blame] | 379 | pgoff_t index = start; |
Minchan Kim | 3156018 | 2011-03-22 16:32:52 -0700 | [diff] [blame] | 380 | unsigned long ret; |
| 381 | unsigned long count = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 382 | int i; |
| 383 | |
Hugh Dickins | 31475dd | 2011-08-03 16:21:27 -0700 | [diff] [blame] | 384 | /* |
| 385 | * Note: this function may get called on a shmem/tmpfs mapping: |
| 386 | * pagevec_lookup() might then return 0 prematurely (because it |
| 387 | * got a gangful of swap entries); but it's hardly worth worrying |
| 388 | * about - it can rarely have anything to free from such a mapping |
| 389 | * (most pages are dirty), and already skips over any difficulties. |
| 390 | */ |
| 391 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 392 | pagevec_init(&pvec, 0); |
Hugh Dickins | b85e0ef | 2011-07-25 17:12:25 -0700 | [diff] [blame] | 393 | while (index <= end && pagevec_lookup(&pvec, mapping, index, |
| 394 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) { |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 395 | mem_cgroup_uncharge_start(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 396 | for (i = 0; i < pagevec_count(&pvec); i++) { |
| 397 | struct page *page = pvec.pages[i]; |
| 398 | |
Hugh Dickins | b85e0ef | 2011-07-25 17:12:25 -0700 | [diff] [blame] | 399 | /* We rely upon deletion not changing page->index */ |
NeilBrown | e0f2360 | 2006-06-23 02:05:48 -0700 | [diff] [blame] | 400 | index = page->index; |
Hugh Dickins | b85e0ef | 2011-07-25 17:12:25 -0700 | [diff] [blame] | 401 | if (index > end) |
| 402 | break; |
NeilBrown | e0f2360 | 2006-06-23 02:05:48 -0700 | [diff] [blame] | 403 | |
Hugh Dickins | b85e0ef | 2011-07-25 17:12:25 -0700 | [diff] [blame] | 404 | if (!trylock_page(page)) |
| 405 | continue; |
| 406 | WARN_ON(page->index != index); |
Minchan Kim | 3156018 | 2011-03-22 16:32:52 -0700 | [diff] [blame] | 407 | ret = invalidate_inode_page(page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 408 | unlock_page(page); |
Minchan Kim | 3156018 | 2011-03-22 16:32:52 -0700 | [diff] [blame] | 409 | /* |
| 410 | * Invalidation is a hint that the page is no longer |
| 411 | * of interest and try to speed up its reclaim. |
| 412 | */ |
| 413 | if (!ret) |
| 414 | deactivate_page(page); |
| 415 | count += ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 416 | } |
| 417 | pagevec_release(&pvec); |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 418 | mem_cgroup_uncharge_end(); |
Mike Waychison | 2869735 | 2009-06-16 15:32:59 -0700 | [diff] [blame] | 419 | cond_resched(); |
Hugh Dickins | b85e0ef | 2011-07-25 17:12:25 -0700 | [diff] [blame] | 420 | index++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 421 | } |
Minchan Kim | 3156018 | 2011-03-22 16:32:52 -0700 | [diff] [blame] | 422 | return count; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 423 | } |
Anton Altaparmakov | 54bc485 | 2007-02-10 01:45:38 -0800 | [diff] [blame] | 424 | EXPORT_SYMBOL(invalidate_mapping_pages); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 425 | |
Andrew Morton | bd4c8ce | 2006-09-30 23:29:29 -0700 | [diff] [blame] | 426 | /* |
| 427 | * This is like invalidate_complete_page(), except it ignores the page's |
| 428 | * refcount. We do this because invalidate_inode_pages2() needs stronger |
| 429 | * invalidation guarantees, and cannot afford to leave pages behind because |
Anderson Briglia | 2706a1b | 2007-07-15 23:38:09 -0700 | [diff] [blame] | 430 | * shrink_page_list() has a temp ref on them, or because they're transiently |
| 431 | * sitting in the lru_cache_add() pagevecs. |
Andrew Morton | bd4c8ce | 2006-09-30 23:29:29 -0700 | [diff] [blame] | 432 | */ |
| 433 | static int |
| 434 | invalidate_complete_page2(struct address_space *mapping, struct page *page) |
| 435 | { |
| 436 | if (page->mapping != mapping) |
| 437 | return 0; |
| 438 | |
David Howells | 266cf65 | 2009-04-03 16:42:36 +0100 | [diff] [blame] | 439 | if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) |
Andrew Morton | bd4c8ce | 2006-09-30 23:29:29 -0700 | [diff] [blame] | 440 | return 0; |
| 441 | |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 442 | spin_lock_irq(&mapping->tree_lock); |
Andrew Morton | bd4c8ce | 2006-09-30 23:29:29 -0700 | [diff] [blame] | 443 | if (PageDirty(page)) |
| 444 | goto failed; |
| 445 | |
David Howells | 266cf65 | 2009-04-03 16:42:36 +0100 | [diff] [blame] | 446 | BUG_ON(page_has_private(page)); |
Minchan Kim | e64a782 | 2011-03-22 16:32:44 -0700 | [diff] [blame] | 447 | __delete_from_page_cache(page); |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 448 | spin_unlock_irq(&mapping->tree_lock); |
Daisuke Nishimura | e767e05 | 2009-05-28 14:34:28 -0700 | [diff] [blame] | 449 | mem_cgroup_uncharge_cache_page(page); |
Linus Torvalds | 6072d13 | 2010-12-01 13:35:19 -0500 | [diff] [blame] | 450 | |
| 451 | if (mapping->a_ops->freepage) |
| 452 | mapping->a_ops->freepage(page); |
| 453 | |
Andrew Morton | bd4c8ce | 2006-09-30 23:29:29 -0700 | [diff] [blame] | 454 | page_cache_release(page); /* pagecache ref */ |
| 455 | return 1; |
| 456 | failed: |
Nick Piggin | 19fd623 | 2008-07-25 19:45:32 -0700 | [diff] [blame] | 457 | spin_unlock_irq(&mapping->tree_lock); |
Andrew Morton | bd4c8ce | 2006-09-30 23:29:29 -0700 | [diff] [blame] | 458 | return 0; |
| 459 | } |
| 460 | |
Trond Myklebust | e3db769 | 2007-01-10 23:15:39 -0800 | [diff] [blame] | 461 | static int do_launder_page(struct address_space *mapping, struct page *page) |
| 462 | { |
| 463 | if (!PageDirty(page)) |
| 464 | return 0; |
| 465 | if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) |
| 466 | return 0; |
| 467 | return mapping->a_ops->launder_page(page); |
| 468 | } |
| 469 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 470 | /** |
| 471 | * invalidate_inode_pages2_range - remove range of pages from an address_space |
Martin Waitz | 67be2dd | 2005-05-01 08:59:26 -0700 | [diff] [blame] | 472 | * @mapping: the address_space |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 473 | * @start: the page offset 'from' which to invalidate |
| 474 | * @end: the page offset 'to' which to invalidate (inclusive) |
| 475 | * |
| 476 | * Any pages which are found to be mapped into pagetables are unmapped prior to |
| 477 | * invalidation. |
| 478 | * |
Hisashi Hifumi | 6ccfa80 | 2008-09-02 14:35:40 -0700 | [diff] [blame] | 479 | * Returns -EBUSY if any pages could not be invalidated. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 480 | */ |
| 481 | int invalidate_inode_pages2_range(struct address_space *mapping, |
| 482 | pgoff_t start, pgoff_t end) |
| 483 | { |
| 484 | struct pagevec pvec; |
Hugh Dickins | b85e0ef | 2011-07-25 17:12:25 -0700 | [diff] [blame] | 485 | pgoff_t index; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 486 | int i; |
| 487 | int ret = 0; |
Hisashi Hifumi | 0dd1334 | 2008-04-28 02:12:08 -0700 | [diff] [blame] | 488 | int ret2 = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 489 | int did_range_unmap = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 490 | |
Dan Magenheimer | 3167760 | 2011-09-21 11:56:28 -0400 | [diff] [blame] | 491 | cleancache_invalidate_inode(mapping); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 492 | pagevec_init(&pvec, 0); |
Hugh Dickins | b85e0ef | 2011-07-25 17:12:25 -0700 | [diff] [blame] | 493 | index = start; |
| 494 | while (index <= end && pagevec_lookup(&pvec, mapping, index, |
| 495 | min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) { |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 496 | mem_cgroup_uncharge_start(); |
Trond Myklebust | 7b965e0 | 2007-02-28 20:13:55 -0800 | [diff] [blame] | 497 | for (i = 0; i < pagevec_count(&pvec); i++) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 498 | struct page *page = pvec.pages[i]; |
Hugh Dickins | b85e0ef | 2011-07-25 17:12:25 -0700 | [diff] [blame] | 499 | |
| 500 | /* We rely upon deletion not changing page->index */ |
| 501 | index = page->index; |
| 502 | if (index > end) |
| 503 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 504 | |
| 505 | lock_page(page); |
Hugh Dickins | b85e0ef | 2011-07-25 17:12:25 -0700 | [diff] [blame] | 506 | WARN_ON(page->index != index); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 507 | if (page->mapping != mapping) { |
| 508 | unlock_page(page); |
| 509 | continue; |
| 510 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 511 | wait_on_page_writeback(page); |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 512 | if (page_mapped(page)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 513 | if (!did_range_unmap) { |
| 514 | /* |
| 515 | * Zap the rest of the file in one hit. |
| 516 | */ |
| 517 | unmap_mapping_range(mapping, |
Hugh Dickins | b85e0ef | 2011-07-25 17:12:25 -0700 | [diff] [blame] | 518 | (loff_t)index << PAGE_CACHE_SHIFT, |
| 519 | (loff_t)(1 + end - index) |
| 520 | << PAGE_CACHE_SHIFT, |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 521 | 0); |
| 522 | did_range_unmap = 1; |
| 523 | } else { |
| 524 | /* |
| 525 | * Just zap this page |
| 526 | */ |
| 527 | unmap_mapping_range(mapping, |
Hugh Dickins | b85e0ef | 2011-07-25 17:12:25 -0700 | [diff] [blame] | 528 | (loff_t)index << PAGE_CACHE_SHIFT, |
| 529 | PAGE_CACHE_SIZE, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 530 | } |
| 531 | } |
Nick Piggin | d00806b | 2007-07-19 01:46:57 -0700 | [diff] [blame] | 532 | BUG_ON(page_mapped(page)); |
Hisashi Hifumi | 0dd1334 | 2008-04-28 02:12:08 -0700 | [diff] [blame] | 533 | ret2 = do_launder_page(mapping, page); |
| 534 | if (ret2 == 0) { |
| 535 | if (!invalidate_complete_page2(mapping, page)) |
Hisashi Hifumi | 6ccfa80 | 2008-09-02 14:35:40 -0700 | [diff] [blame] | 536 | ret2 = -EBUSY; |
Hisashi Hifumi | 0dd1334 | 2008-04-28 02:12:08 -0700 | [diff] [blame] | 537 | } |
| 538 | if (ret2 < 0) |
| 539 | ret = ret2; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 540 | unlock_page(page); |
| 541 | } |
| 542 | pagevec_release(&pvec); |
KAMEZAWA Hiroyuki | 569b846 | 2009-12-15 16:47:03 -0800 | [diff] [blame] | 543 | mem_cgroup_uncharge_end(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 544 | cond_resched(); |
Hugh Dickins | b85e0ef | 2011-07-25 17:12:25 -0700 | [diff] [blame] | 545 | index++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 546 | } |
Dan Magenheimer | 3167760 | 2011-09-21 11:56:28 -0400 | [diff] [blame] | 547 | cleancache_invalidate_inode(mapping); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 548 | return ret; |
| 549 | } |
| 550 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); |
| 551 | |
| 552 | /** |
| 553 | * invalidate_inode_pages2 - remove all pages from an address_space |
Martin Waitz | 67be2dd | 2005-05-01 08:59:26 -0700 | [diff] [blame] | 554 | * @mapping: the address_space |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 555 | * |
| 556 | * Any pages which are found to be mapped into pagetables are unmapped prior to |
| 557 | * invalidation. |
| 558 | * |
Peng Tao | e9de25d | 2009-10-19 14:48:13 +0800 | [diff] [blame] | 559 | * Returns -EBUSY if any pages could not be invalidated. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 560 | */ |
| 561 | int invalidate_inode_pages2(struct address_space *mapping) |
| 562 | { |
| 563 | return invalidate_inode_pages2_range(mapping, 0, -1); |
| 564 | } |
| 565 | EXPORT_SYMBOL_GPL(invalidate_inode_pages2); |
npiggin@suse.de | 25d9e2d | 2009-08-21 02:35:05 +1000 | [diff] [blame] | 566 | |
| 567 | /** |
| 568 | * truncate_pagecache - unmap and remove pagecache that has been truncated |
| 569 | * @inode: inode |
Hugh Dickins | 8a549be | 2011-07-25 17:12:24 -0700 | [diff] [blame] | 570 | * @oldsize: old file size |
| 571 | * @newsize: new file size |
npiggin@suse.de | 25d9e2d | 2009-08-21 02:35:05 +1000 | [diff] [blame] | 572 | * |
| 573 | * inode's new i_size must already be written before truncate_pagecache |
| 574 | * is called. |
| 575 | * |
| 576 | * This function should typically be called before the filesystem |
| 577 | * releases resources associated with the freed range (eg. deallocates |
| 578 | * blocks). This way, pagecache will always stay logically coherent |
| 579 | * with on-disk format, and the filesystem would not have to deal with |
| 580 | * situations such as writepage being called for a page that has already |
| 581 | * had its underlying blocks deallocated. |
| 582 | */ |
Hugh Dickins | 8a549be | 2011-07-25 17:12:24 -0700 | [diff] [blame] | 583 | void truncate_pagecache(struct inode *inode, loff_t oldsize, loff_t newsize) |
npiggin@suse.de | 25d9e2d | 2009-08-21 02:35:05 +1000 | [diff] [blame] | 584 | { |
OGAWA Hirofumi | cedabed | 2010-01-13 21:14:09 +0900 | [diff] [blame] | 585 | struct address_space *mapping = inode->i_mapping; |
Hugh Dickins | 8a549be | 2011-07-25 17:12:24 -0700 | [diff] [blame] | 586 | loff_t holebegin = round_up(newsize, PAGE_SIZE); |
npiggin@suse.de | 25d9e2d | 2009-08-21 02:35:05 +1000 | [diff] [blame] | 587 | |
OGAWA Hirofumi | cedabed | 2010-01-13 21:14:09 +0900 | [diff] [blame] | 588 | /* |
| 589 | * unmap_mapping_range is called twice, first simply for |
| 590 | * efficiency so that truncate_inode_pages does fewer |
| 591 | * single-page unmaps. However after this first call, and |
| 592 | * before truncate_inode_pages finishes, it is possible for |
| 593 | * private pages to be COWed, which remain after |
| 594 | * truncate_inode_pages finishes, hence the second |
| 595 | * unmap_mapping_range call must be made for correctness. |
| 596 | */ |
Hugh Dickins | 8a549be | 2011-07-25 17:12:24 -0700 | [diff] [blame] | 597 | unmap_mapping_range(mapping, holebegin, 0, 1); |
| 598 | truncate_inode_pages(mapping, newsize); |
| 599 | unmap_mapping_range(mapping, holebegin, 0, 1); |
npiggin@suse.de | 25d9e2d | 2009-08-21 02:35:05 +1000 | [diff] [blame] | 600 | } |
| 601 | EXPORT_SYMBOL(truncate_pagecache); |
| 602 | |
| 603 | /** |
Christoph Hellwig | 2c27c65 | 2010-06-04 11:30:04 +0200 | [diff] [blame] | 604 | * truncate_setsize - update inode and pagecache for a new file size |
| 605 | * @inode: inode |
| 606 | * @newsize: new file size |
| 607 | * |
Jan Kara | 382e27d | 2011-01-20 14:44:26 -0800 | [diff] [blame] | 608 | * truncate_setsize updates i_size and performs pagecache truncation (if |
| 609 | * necessary) to @newsize. It will be typically be called from the filesystem's |
| 610 | * setattr function when ATTR_SIZE is passed in. |
Christoph Hellwig | 2c27c65 | 2010-06-04 11:30:04 +0200 | [diff] [blame] | 611 | * |
Jan Kara | 382e27d | 2011-01-20 14:44:26 -0800 | [diff] [blame] | 612 | * Must be called with inode_mutex held and before all filesystem specific |
| 613 | * block truncation has been performed. |
Christoph Hellwig | 2c27c65 | 2010-06-04 11:30:04 +0200 | [diff] [blame] | 614 | */ |
| 615 | void truncate_setsize(struct inode *inode, loff_t newsize) |
| 616 | { |
| 617 | loff_t oldsize; |
| 618 | |
| 619 | oldsize = inode->i_size; |
| 620 | i_size_write(inode, newsize); |
| 621 | |
| 622 | truncate_pagecache(inode, oldsize, newsize); |
| 623 | } |
| 624 | EXPORT_SYMBOL(truncate_setsize); |
| 625 | |
| 626 | /** |
Hugh Dickins | 623e3db | 2012-03-28 14:42:40 -0700 | [diff] [blame] | 627 | * truncate_pagecache_range - unmap and remove pagecache that is hole-punched |
| 628 | * @inode: inode |
| 629 | * @lstart: offset of beginning of hole |
| 630 | * @lend: offset of last byte of hole |
| 631 | * |
| 632 | * This function should typically be called before the filesystem |
| 633 | * releases resources associated with the freed range (eg. deallocates |
| 634 | * blocks). This way, pagecache will always stay logically coherent |
| 635 | * with on-disk format, and the filesystem would not have to deal with |
| 636 | * situations such as writepage being called for a page that has already |
| 637 | * had its underlying blocks deallocated. |
| 638 | */ |
| 639 | void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) |
| 640 | { |
| 641 | struct address_space *mapping = inode->i_mapping; |
| 642 | loff_t unmap_start = round_up(lstart, PAGE_SIZE); |
| 643 | loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; |
| 644 | /* |
| 645 | * This rounding is currently just for example: unmap_mapping_range |
| 646 | * expands its hole outwards, whereas we want it to contract the hole |
| 647 | * inwards. However, existing callers of truncate_pagecache_range are |
Lukas Czerner | 5a72039 | 2013-05-27 23:32:35 -0400 | [diff] [blame] | 648 | * doing their own page rounding first. Note that unmap_mapping_range |
| 649 | * allows holelen 0 for all, and we allow lend -1 for end of file. |
Hugh Dickins | 623e3db | 2012-03-28 14:42:40 -0700 | [diff] [blame] | 650 | */ |
| 651 | |
| 652 | /* |
| 653 | * Unlike in truncate_pagecache, unmap_mapping_range is called only |
| 654 | * once (before truncating pagecache), and without "even_cows" flag: |
| 655 | * hole-punching should not remove private COWed pages from the hole. |
| 656 | */ |
| 657 | if ((u64)unmap_end > (u64)unmap_start) |
| 658 | unmap_mapping_range(mapping, unmap_start, |
| 659 | 1 + unmap_end - unmap_start, 0); |
| 660 | truncate_inode_pages_range(mapping, lstart, lend); |
| 661 | } |
| 662 | EXPORT_SYMBOL(truncate_pagecache_range); |