11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * mm/truncate.c - code for taking down pages from address_spaces 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 2002, Linus Torvalds 51da177e4SLinus Torvalds * 6e1f8e874SFrancois Cami * 10Sep2002 Andrew Morton 71da177e4SLinus Torvalds * Initial version. 81da177e4SLinus Torvalds */ 91da177e4SLinus Torvalds 101da177e4SLinus Torvalds #include <linux/kernel.h> 114af3c9ccSAlexey Dobriyan #include <linux/backing-dev.h> 12*f9fe48beSRoss Zwisler #include <linux/dax.h> 135a0e3ad6STejun Heo #include <linux/gfp.h> 141da177e4SLinus Torvalds #include <linux/mm.h> 150fd0e6b0SNick Piggin #include <linux/swap.h> 16b95f1b31SPaul Gortmaker #include <linux/export.h> 171da177e4SLinus Torvalds #include <linux/pagemap.h> 1801f2705dSNate Diller #include <linux/highmem.h> 191da177e4SLinus Torvalds #include <linux/pagevec.h> 20e08748ceSAndrew Morton #include <linux/task_io_accounting_ops.h> 211da177e4SLinus Torvalds #include <linux/buffer_head.h> /* grr. try_to_release_page, 22aaa4059bSJan Kara do_invalidatepage */ 23c515e1fdSDan Magenheimer #include <linux/cleancache.h> 2490a80202SJan Kara #include <linux/rmap.h> 25ba470de4SRik van Riel #include "internal.h" 261da177e4SLinus Torvalds 270cd6144aSJohannes Weiner static void clear_exceptional_entry(struct address_space *mapping, 280cd6144aSJohannes Weiner pgoff_t index, void *entry) 290cd6144aSJohannes Weiner { 30449dd698SJohannes Weiner struct radix_tree_node *node; 31449dd698SJohannes Weiner void **slot; 32449dd698SJohannes Weiner 330cd6144aSJohannes Weiner /* Handled by shmem itself */ 340cd6144aSJohannes Weiner if (shmem_mapping(mapping)) 350cd6144aSJohannes Weiner return; 360cd6144aSJohannes Weiner 370cd6144aSJohannes Weiner spin_lock_irq(&mapping->tree_lock); 38*f9fe48beSRoss Zwisler 39*f9fe48beSRoss Zwisler if (dax_mapping(mapping)) { 40*f9fe48beSRoss Zwisler if (radix_tree_delete_item(&mapping->page_tree, index, entry)) 41*f9fe48beSRoss Zwisler mapping->nrexceptional--; 42*f9fe48beSRoss Zwisler } else { 430cd6144aSJohannes Weiner /* 440cd6144aSJohannes Weiner * Regular page slots are stabilized by the page lock even 450cd6144aSJohannes Weiner * without the tree itself locked. These unlocked entries 460cd6144aSJohannes Weiner * need verification under the tree lock. 470cd6144aSJohannes Weiner */ 48*f9fe48beSRoss Zwisler if (!__radix_tree_lookup(&mapping->page_tree, index, &node, 49*f9fe48beSRoss Zwisler &slot)) 50449dd698SJohannes Weiner goto unlock; 51449dd698SJohannes Weiner if (*slot != entry) 52449dd698SJohannes Weiner goto unlock; 53449dd698SJohannes Weiner radix_tree_replace_slot(slot, NULL); 54*f9fe48beSRoss Zwisler mapping->nrexceptional--; 55449dd698SJohannes Weiner if (!node) 56449dd698SJohannes Weiner goto unlock; 57449dd698SJohannes Weiner workingset_node_shadows_dec(node); 58449dd698SJohannes Weiner /* 59449dd698SJohannes Weiner * Don't track node without shadow entries. 60449dd698SJohannes Weiner * 61449dd698SJohannes Weiner * Avoid acquiring the list_lru lock if already untracked. 62449dd698SJohannes Weiner * The list_empty() test is safe as node->private_list is 63449dd698SJohannes Weiner * protected by mapping->tree_lock. 64449dd698SJohannes Weiner */ 65449dd698SJohannes Weiner if (!workingset_node_shadows(node) && 66449dd698SJohannes Weiner !list_empty(&node->private_list)) 67*f9fe48beSRoss Zwisler list_lru_del(&workingset_shadow_nodes, 68*f9fe48beSRoss Zwisler &node->private_list); 69449dd698SJohannes Weiner __radix_tree_delete_node(&mapping->page_tree, node); 70*f9fe48beSRoss Zwisler } 71449dd698SJohannes Weiner unlock: 720cd6144aSJohannes Weiner spin_unlock_irq(&mapping->tree_lock); 730cd6144aSJohannes Weiner } 741da177e4SLinus Torvalds 75cf9a2ae8SDavid Howells /** 7628bc44d7SFengguang Wu * do_invalidatepage - invalidate part or all of a page 77cf9a2ae8SDavid Howells * @page: the page which is affected 78d47992f8SLukas Czerner * @offset: start of the range to invalidate 79d47992f8SLukas Czerner * @length: length of the range to invalidate 80cf9a2ae8SDavid Howells * 81cf9a2ae8SDavid Howells * do_invalidatepage() is called when all or part of the page has become 82cf9a2ae8SDavid Howells * invalidated by a truncate operation. 83cf9a2ae8SDavid Howells * 84cf9a2ae8SDavid Howells * do_invalidatepage() does not have to release all buffers, but it must 85cf9a2ae8SDavid Howells * ensure that no dirty buffer is left outside @offset and that no I/O 86cf9a2ae8SDavid Howells * is underway against any of the blocks which are outside the truncation 87cf9a2ae8SDavid Howells * point. Because the caller is about to free (and possibly reuse) those 88cf9a2ae8SDavid Howells * blocks on-disk. 89cf9a2ae8SDavid Howells */ 90d47992f8SLukas Czerner void do_invalidatepage(struct page *page, unsigned int offset, 91d47992f8SLukas Czerner unsigned int length) 92cf9a2ae8SDavid Howells { 93d47992f8SLukas Czerner void (*invalidatepage)(struct page *, unsigned int, unsigned int); 94d47992f8SLukas Czerner 95cf9a2ae8SDavid Howells invalidatepage = page->mapping->a_ops->invalidatepage; 969361401eSDavid Howells #ifdef CONFIG_BLOCK 97cf9a2ae8SDavid Howells if (!invalidatepage) 98cf9a2ae8SDavid Howells invalidatepage = block_invalidatepage; 999361401eSDavid Howells #endif 100cf9a2ae8SDavid Howells if (invalidatepage) 101d47992f8SLukas Czerner (*invalidatepage)(page, offset, length); 102cf9a2ae8SDavid Howells } 103cf9a2ae8SDavid Howells 104ecdfc978SLinus Torvalds /* 1051da177e4SLinus Torvalds * If truncate cannot remove the fs-private metadata from the page, the page 10662e1c553SShaohua Li * becomes orphaned. It will be left on the LRU and may even be mapped into 10754cb8821SNick Piggin * user pagetables if we're racing with filemap_fault(). 1081da177e4SLinus Torvalds * 1091da177e4SLinus Torvalds * We need to bale out if page->mapping is no longer equal to the original 1101da177e4SLinus Torvalds * mapping. This happens a) when the VM reclaimed the page while we waited on 111fc0ecff6SAndrew Morton * its lock, b) when a concurrent invalidate_mapping_pages got there first and 1121da177e4SLinus Torvalds * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. 1131da177e4SLinus Torvalds */ 114750b4987SNick Piggin static int 1151da177e4SLinus Torvalds truncate_complete_page(struct address_space *mapping, struct page *page) 1161da177e4SLinus Torvalds { 1171da177e4SLinus Torvalds if (page->mapping != mapping) 118750b4987SNick Piggin return -EIO; 1191da177e4SLinus Torvalds 120266cf658SDavid Howells if (page_has_private(page)) 121d47992f8SLukas Czerner do_invalidatepage(page, 0, PAGE_CACHE_SIZE); 1221da177e4SLinus Torvalds 123b9ea2515SKonstantin Khlebnikov /* 124b9ea2515SKonstantin Khlebnikov * Some filesystems seem to re-dirty the page even after 125b9ea2515SKonstantin Khlebnikov * the VM has canceled the dirty bit (eg ext3 journaling). 126b9ea2515SKonstantin Khlebnikov * Hence dirty accounting check is placed after invalidation. 127b9ea2515SKonstantin Khlebnikov */ 12811f81becSTejun Heo cancel_dirty_page(page); 1291da177e4SLinus Torvalds ClearPageMappedToDisk(page); 1305adc7b51SMinchan Kim delete_from_page_cache(page); 131750b4987SNick Piggin return 0; 1321da177e4SLinus Torvalds } 1331da177e4SLinus Torvalds 1341da177e4SLinus Torvalds /* 135fc0ecff6SAndrew Morton * This is for invalidate_mapping_pages(). That function can be called at 1361da177e4SLinus Torvalds * any time, and is not supposed to throw away dirty pages. But pages can 1370fd0e6b0SNick Piggin * be marked dirty at any time too, so use remove_mapping which safely 1380fd0e6b0SNick Piggin * discards clean, unused pages. 1391da177e4SLinus Torvalds * 1401da177e4SLinus Torvalds * Returns non-zero if the page was successfully invalidated. 1411da177e4SLinus Torvalds */ 1421da177e4SLinus Torvalds static int 1431da177e4SLinus Torvalds invalidate_complete_page(struct address_space *mapping, struct page *page) 1441da177e4SLinus Torvalds { 1450fd0e6b0SNick Piggin int ret; 1460fd0e6b0SNick Piggin 1471da177e4SLinus Torvalds if (page->mapping != mapping) 1481da177e4SLinus Torvalds return 0; 1491da177e4SLinus Torvalds 150266cf658SDavid Howells if (page_has_private(page) && !try_to_release_page(page, 0)) 1511da177e4SLinus Torvalds return 0; 1521da177e4SLinus Torvalds 1530fd0e6b0SNick Piggin ret = remove_mapping(mapping, page); 1540fd0e6b0SNick Piggin 1550fd0e6b0SNick Piggin return ret; 1561da177e4SLinus Torvalds } 1571da177e4SLinus Torvalds 158750b4987SNick Piggin int truncate_inode_page(struct address_space *mapping, struct page *page) 159750b4987SNick Piggin { 160750b4987SNick Piggin if (page_mapped(page)) { 161750b4987SNick Piggin unmap_mapping_range(mapping, 162750b4987SNick Piggin (loff_t)page->index << PAGE_CACHE_SHIFT, 163750b4987SNick Piggin PAGE_CACHE_SIZE, 0); 164750b4987SNick Piggin } 165750b4987SNick Piggin return truncate_complete_page(mapping, page); 166750b4987SNick Piggin } 167750b4987SNick Piggin 16883f78668SWu Fengguang /* 16925718736SAndi Kleen * Used to get rid of pages on hardware memory corruption. 17025718736SAndi Kleen */ 17125718736SAndi Kleen int generic_error_remove_page(struct address_space *mapping, struct page *page) 17225718736SAndi Kleen { 17325718736SAndi Kleen if (!mapping) 17425718736SAndi Kleen return -EINVAL; 17525718736SAndi Kleen /* 17625718736SAndi Kleen * Only punch for normal data pages for now. 17725718736SAndi Kleen * Handling other types like directories would need more auditing. 17825718736SAndi Kleen */ 17925718736SAndi Kleen if (!S_ISREG(mapping->host->i_mode)) 18025718736SAndi Kleen return -EIO; 18125718736SAndi Kleen return truncate_inode_page(mapping, page); 18225718736SAndi Kleen } 18325718736SAndi Kleen EXPORT_SYMBOL(generic_error_remove_page); 18425718736SAndi Kleen 18525718736SAndi Kleen /* 18683f78668SWu Fengguang * Safely invalidate one page from its pagecache mapping. 18783f78668SWu Fengguang * It only drops clean, unused pages. The page must be locked. 18883f78668SWu Fengguang * 18983f78668SWu Fengguang * Returns 1 if the page is successfully invalidated, otherwise 0. 19083f78668SWu Fengguang */ 19183f78668SWu Fengguang int invalidate_inode_page(struct page *page) 19283f78668SWu Fengguang { 19383f78668SWu Fengguang struct address_space *mapping = page_mapping(page); 19483f78668SWu Fengguang if (!mapping) 19583f78668SWu Fengguang return 0; 19683f78668SWu Fengguang if (PageDirty(page) || PageWriteback(page)) 19783f78668SWu Fengguang return 0; 19883f78668SWu Fengguang if (page_mapped(page)) 19983f78668SWu Fengguang return 0; 20083f78668SWu Fengguang return invalidate_complete_page(mapping, page); 20183f78668SWu Fengguang } 20283f78668SWu Fengguang 2031da177e4SLinus Torvalds /** 20473c1e204SLiu Bo * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets 2051da177e4SLinus Torvalds * @mapping: mapping to truncate 2061da177e4SLinus Torvalds * @lstart: offset from which to truncate 2075a720394SLukas Czerner * @lend: offset to which to truncate (inclusive) 2081da177e4SLinus Torvalds * 209d7339071SHans Reiser * Truncate the page cache, removing the pages that are between 2105a720394SLukas Czerner * specified offsets (and zeroing out partial pages 2115a720394SLukas Czerner * if lstart or lend + 1 is not page aligned). 2121da177e4SLinus Torvalds * 2131da177e4SLinus Torvalds * Truncate takes two passes - the first pass is nonblocking. It will not 2141da177e4SLinus Torvalds * block on page locks and it will not block on writeback. The second pass 2151da177e4SLinus Torvalds * will wait. This is to prevent as much IO as possible in the affected region. 2161da177e4SLinus Torvalds * The first pass will remove most pages, so the search cost of the second pass 2171da177e4SLinus Torvalds * is low. 2181da177e4SLinus Torvalds * 2191da177e4SLinus Torvalds * We pass down the cache-hot hint to the page freeing code. Even if the 2201da177e4SLinus Torvalds * mapping is large, it is probably the case that the final pages are the most 2211da177e4SLinus Torvalds * recently touched, and freeing happens in ascending file offset order. 2225a720394SLukas Czerner * 2235a720394SLukas Czerner * Note that since ->invalidatepage() accepts range to invalidate 2245a720394SLukas Czerner * truncate_inode_pages_range is able to handle cases where lend + 1 is not 2255a720394SLukas Czerner * page aligned properly. 2261da177e4SLinus Torvalds */ 227d7339071SHans Reiser void truncate_inode_pages_range(struct address_space *mapping, 228d7339071SHans Reiser loff_t lstart, loff_t lend) 2291da177e4SLinus Torvalds { 2305a720394SLukas Czerner pgoff_t start; /* inclusive */ 2315a720394SLukas Czerner pgoff_t end; /* exclusive */ 2325a720394SLukas Czerner unsigned int partial_start; /* inclusive */ 2335a720394SLukas Czerner unsigned int partial_end; /* exclusive */ 2341da177e4SLinus Torvalds struct pagevec pvec; 2350cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 236b85e0effSHugh Dickins pgoff_t index; 2371da177e4SLinus Torvalds int i; 2381da177e4SLinus Torvalds 2393167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 240*f9fe48beSRoss Zwisler if (mapping->nrpages == 0 && mapping->nrexceptional == 0) 2411da177e4SLinus Torvalds return; 2421da177e4SLinus Torvalds 2435a720394SLukas Czerner /* Offsets within partial pages */ 2445a720394SLukas Czerner partial_start = lstart & (PAGE_CACHE_SIZE - 1); 2455a720394SLukas Czerner partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1); 2465a720394SLukas Czerner 2475a720394SLukas Czerner /* 2485a720394SLukas Czerner * 'start' and 'end' always covers the range of pages to be fully 2495a720394SLukas Czerner * truncated. Partial pages are covered with 'partial_start' at the 2505a720394SLukas Czerner * start of the range and 'partial_end' at the end of the range. 2515a720394SLukas Czerner * Note that 'end' is exclusive while 'lend' is inclusive. 2525a720394SLukas Czerner */ 2535a720394SLukas Czerner start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 2545a720394SLukas Czerner if (lend == -1) 2555a720394SLukas Czerner /* 2565a720394SLukas Czerner * lend == -1 indicates end-of-file so we have to set 'end' 2575a720394SLukas Czerner * to the highest possible pgoff_t and since the type is 2585a720394SLukas Czerner * unsigned we're using -1. 2595a720394SLukas Czerner */ 2605a720394SLukas Czerner end = -1; 2615a720394SLukas Czerner else 2625a720394SLukas Czerner end = (lend + 1) >> PAGE_CACHE_SHIFT; 263d7339071SHans Reiser 2641da177e4SLinus Torvalds pagevec_init(&pvec, 0); 265b85e0effSHugh Dickins index = start; 2660cd6144aSJohannes Weiner while (index < end && pagevec_lookup_entries(&pvec, mapping, index, 2670cd6144aSJohannes Weiner min(end - index, (pgoff_t)PAGEVEC_SIZE), 2680cd6144aSJohannes Weiner indices)) { 2691da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 2701da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 2711da177e4SLinus Torvalds 272b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 2730cd6144aSJohannes Weiner index = indices[i]; 2745a720394SLukas Czerner if (index >= end) 275d7339071SHans Reiser break; 276d7339071SHans Reiser 2770cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 2780cd6144aSJohannes Weiner clear_exceptional_entry(mapping, index, page); 2790cd6144aSJohannes Weiner continue; 2800cd6144aSJohannes Weiner } 2810cd6144aSJohannes Weiner 282529ae9aaSNick Piggin if (!trylock_page(page)) 2831da177e4SLinus Torvalds continue; 284b85e0effSHugh Dickins WARN_ON(page->index != index); 2851da177e4SLinus Torvalds if (PageWriteback(page)) { 2861da177e4SLinus Torvalds unlock_page(page); 2871da177e4SLinus Torvalds continue; 2881da177e4SLinus Torvalds } 289750b4987SNick Piggin truncate_inode_page(mapping, page); 2901da177e4SLinus Torvalds unlock_page(page); 2911da177e4SLinus Torvalds } 2920cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 2931da177e4SLinus Torvalds pagevec_release(&pvec); 2941da177e4SLinus Torvalds cond_resched(); 295b85e0effSHugh Dickins index++; 2961da177e4SLinus Torvalds } 2971da177e4SLinus Torvalds 2985a720394SLukas Czerner if (partial_start) { 2991da177e4SLinus Torvalds struct page *page = find_lock_page(mapping, start - 1); 3001da177e4SLinus Torvalds if (page) { 3015a720394SLukas Czerner unsigned int top = PAGE_CACHE_SIZE; 3025a720394SLukas Czerner if (start > end) { 3035a720394SLukas Czerner /* Truncation within a single page */ 3045a720394SLukas Czerner top = partial_end; 3055a720394SLukas Czerner partial_end = 0; 3065a720394SLukas Czerner } 3071da177e4SLinus Torvalds wait_on_page_writeback(page); 3085a720394SLukas Czerner zero_user_segment(page, partial_start, top); 3095a720394SLukas Czerner cleancache_invalidate_page(mapping, page); 3105a720394SLukas Czerner if (page_has_private(page)) 3115a720394SLukas Czerner do_invalidatepage(page, partial_start, 3125a720394SLukas Czerner top - partial_start); 3131da177e4SLinus Torvalds unlock_page(page); 3141da177e4SLinus Torvalds page_cache_release(page); 3151da177e4SLinus Torvalds } 3161da177e4SLinus Torvalds } 3175a720394SLukas Czerner if (partial_end) { 3185a720394SLukas Czerner struct page *page = find_lock_page(mapping, end); 3195a720394SLukas Czerner if (page) { 3205a720394SLukas Czerner wait_on_page_writeback(page); 3215a720394SLukas Czerner zero_user_segment(page, 0, partial_end); 3225a720394SLukas Czerner cleancache_invalidate_page(mapping, page); 3235a720394SLukas Czerner if (page_has_private(page)) 3245a720394SLukas Czerner do_invalidatepage(page, 0, 3255a720394SLukas Czerner partial_end); 3265a720394SLukas Czerner unlock_page(page); 3275a720394SLukas Czerner page_cache_release(page); 3285a720394SLukas Czerner } 3295a720394SLukas Czerner } 3305a720394SLukas Czerner /* 3315a720394SLukas Czerner * If the truncation happened within a single page no pages 3325a720394SLukas Czerner * will be released, just zeroed, so we can bail out now. 3335a720394SLukas Czerner */ 3345a720394SLukas Czerner if (start >= end) 3355a720394SLukas Czerner return; 3361da177e4SLinus Torvalds 337b85e0effSHugh Dickins index = start; 3381da177e4SLinus Torvalds for ( ; ; ) { 3391da177e4SLinus Torvalds cond_resched(); 3400cd6144aSJohannes Weiner if (!pagevec_lookup_entries(&pvec, mapping, index, 341792ceaefSHugh Dickins min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) { 342792ceaefSHugh Dickins /* If all gone from start onwards, we're done */ 343b85e0effSHugh Dickins if (index == start) 3441da177e4SLinus Torvalds break; 345792ceaefSHugh Dickins /* Otherwise restart to make sure all gone */ 346b85e0effSHugh Dickins index = start; 3471da177e4SLinus Torvalds continue; 3481da177e4SLinus Torvalds } 3490cd6144aSJohannes Weiner if (index == start && indices[0] >= end) { 350792ceaefSHugh Dickins /* All gone out of hole to be punched, we're done */ 3510cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 352d7339071SHans Reiser pagevec_release(&pvec); 353d7339071SHans Reiser break; 354d7339071SHans Reiser } 3551da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 3561da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 3571da177e4SLinus Torvalds 358b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 3590cd6144aSJohannes Weiner index = indices[i]; 360792ceaefSHugh Dickins if (index >= end) { 361792ceaefSHugh Dickins /* Restart punch to make sure all gone */ 362792ceaefSHugh Dickins index = start - 1; 363d7339071SHans Reiser break; 364792ceaefSHugh Dickins } 365b85e0effSHugh Dickins 3660cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 3670cd6144aSJohannes Weiner clear_exceptional_entry(mapping, index, page); 3680cd6144aSJohannes Weiner continue; 3690cd6144aSJohannes Weiner } 3700cd6144aSJohannes Weiner 3711da177e4SLinus Torvalds lock_page(page); 372b85e0effSHugh Dickins WARN_ON(page->index != index); 3731da177e4SLinus Torvalds wait_on_page_writeback(page); 374750b4987SNick Piggin truncate_inode_page(mapping, page); 3751da177e4SLinus Torvalds unlock_page(page); 3761da177e4SLinus Torvalds } 3770cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 3781da177e4SLinus Torvalds pagevec_release(&pvec); 379b85e0effSHugh Dickins index++; 3801da177e4SLinus Torvalds } 3813167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 3821da177e4SLinus Torvalds } 383d7339071SHans Reiser EXPORT_SYMBOL(truncate_inode_pages_range); 3841da177e4SLinus Torvalds 385d7339071SHans Reiser /** 386d7339071SHans Reiser * truncate_inode_pages - truncate *all* the pages from an offset 387d7339071SHans Reiser * @mapping: mapping to truncate 388d7339071SHans Reiser * @lstart: offset from which to truncate 389d7339071SHans Reiser * 3901b1dcc1bSJes Sorensen * Called under (and serialised by) inode->i_mutex. 39108142579SJan Kara * 39208142579SJan Kara * Note: When this function returns, there can be a page in the process of 39308142579SJan Kara * deletion (inside __delete_from_page_cache()) in the specified range. Thus 39408142579SJan Kara * mapping->nrpages can be non-zero when this function returns even after 39508142579SJan Kara * truncation of the whole mapping. 396d7339071SHans Reiser */ 397d7339071SHans Reiser void truncate_inode_pages(struct address_space *mapping, loff_t lstart) 398d7339071SHans Reiser { 399d7339071SHans Reiser truncate_inode_pages_range(mapping, lstart, (loff_t)-1); 400d7339071SHans Reiser } 4011da177e4SLinus Torvalds EXPORT_SYMBOL(truncate_inode_pages); 4021da177e4SLinus Torvalds 40328697355SMike Waychison /** 40491b0abe3SJohannes Weiner * truncate_inode_pages_final - truncate *all* pages before inode dies 40591b0abe3SJohannes Weiner * @mapping: mapping to truncate 40691b0abe3SJohannes Weiner * 40791b0abe3SJohannes Weiner * Called under (and serialized by) inode->i_mutex. 40891b0abe3SJohannes Weiner * 40991b0abe3SJohannes Weiner * Filesystems have to use this in the .evict_inode path to inform the 41091b0abe3SJohannes Weiner * VM that this is the final truncate and the inode is going away. 41191b0abe3SJohannes Weiner */ 41291b0abe3SJohannes Weiner void truncate_inode_pages_final(struct address_space *mapping) 41391b0abe3SJohannes Weiner { 414*f9fe48beSRoss Zwisler unsigned long nrexceptional; 41591b0abe3SJohannes Weiner unsigned long nrpages; 41691b0abe3SJohannes Weiner 41791b0abe3SJohannes Weiner /* 41891b0abe3SJohannes Weiner * Page reclaim can not participate in regular inode lifetime 41991b0abe3SJohannes Weiner * management (can't call iput()) and thus can race with the 42091b0abe3SJohannes Weiner * inode teardown. Tell it when the address space is exiting, 42191b0abe3SJohannes Weiner * so that it does not install eviction information after the 42291b0abe3SJohannes Weiner * final truncate has begun. 42391b0abe3SJohannes Weiner */ 42491b0abe3SJohannes Weiner mapping_set_exiting(mapping); 42591b0abe3SJohannes Weiner 42691b0abe3SJohannes Weiner /* 42791b0abe3SJohannes Weiner * When reclaim installs eviction entries, it increases 428*f9fe48beSRoss Zwisler * nrexceptional first, then decreases nrpages. Make sure we see 42991b0abe3SJohannes Weiner * this in the right order or we might miss an entry. 43091b0abe3SJohannes Weiner */ 43191b0abe3SJohannes Weiner nrpages = mapping->nrpages; 43291b0abe3SJohannes Weiner smp_rmb(); 433*f9fe48beSRoss Zwisler nrexceptional = mapping->nrexceptional; 43491b0abe3SJohannes Weiner 435*f9fe48beSRoss Zwisler if (nrpages || nrexceptional) { 43691b0abe3SJohannes Weiner /* 43791b0abe3SJohannes Weiner * As truncation uses a lockless tree lookup, cycle 43891b0abe3SJohannes Weiner * the tree lock to make sure any ongoing tree 43991b0abe3SJohannes Weiner * modification that does not see AS_EXITING is 44091b0abe3SJohannes Weiner * completed before starting the final truncate. 44191b0abe3SJohannes Weiner */ 44291b0abe3SJohannes Weiner spin_lock_irq(&mapping->tree_lock); 44391b0abe3SJohannes Weiner spin_unlock_irq(&mapping->tree_lock); 44491b0abe3SJohannes Weiner 44591b0abe3SJohannes Weiner truncate_inode_pages(mapping, 0); 44691b0abe3SJohannes Weiner } 44791b0abe3SJohannes Weiner } 44891b0abe3SJohannes Weiner EXPORT_SYMBOL(truncate_inode_pages_final); 44991b0abe3SJohannes Weiner 45091b0abe3SJohannes Weiner /** 45128697355SMike Waychison * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode 45228697355SMike Waychison * @mapping: the address_space which holds the pages to invalidate 45328697355SMike Waychison * @start: the offset 'from' which to invalidate 45428697355SMike Waychison * @end: the offset 'to' which to invalidate (inclusive) 45528697355SMike Waychison * 45628697355SMike Waychison * This function only removes the unlocked pages, if you want to 45728697355SMike Waychison * remove all the pages of one inode, you must call truncate_inode_pages. 45828697355SMike Waychison * 45928697355SMike Waychison * invalidate_mapping_pages() will not block on IO activity. It will not 46028697355SMike Waychison * invalidate pages which are dirty, locked, under writeback or mapped into 46128697355SMike Waychison * pagetables. 46228697355SMike Waychison */ 46328697355SMike Waychison unsigned long invalidate_mapping_pages(struct address_space *mapping, 46428697355SMike Waychison pgoff_t start, pgoff_t end) 4651da177e4SLinus Torvalds { 4660cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 4671da177e4SLinus Torvalds struct pagevec pvec; 468b85e0effSHugh Dickins pgoff_t index = start; 46931560180SMinchan Kim unsigned long ret; 47031560180SMinchan Kim unsigned long count = 0; 4711da177e4SLinus Torvalds int i; 4721da177e4SLinus Torvalds 4731da177e4SLinus Torvalds pagevec_init(&pvec, 0); 4740cd6144aSJohannes Weiner while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, 4750cd6144aSJohannes Weiner min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, 4760cd6144aSJohannes Weiner indices)) { 4771da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 4781da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 4791da177e4SLinus Torvalds 480b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 4810cd6144aSJohannes Weiner index = indices[i]; 482b85e0effSHugh Dickins if (index > end) 483b85e0effSHugh Dickins break; 484e0f23603SNeilBrown 4850cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 4860cd6144aSJohannes Weiner clear_exceptional_entry(mapping, index, page); 4870cd6144aSJohannes Weiner continue; 4880cd6144aSJohannes Weiner } 4890cd6144aSJohannes Weiner 490b85e0effSHugh Dickins if (!trylock_page(page)) 491b85e0effSHugh Dickins continue; 492b85e0effSHugh Dickins WARN_ON(page->index != index); 49331560180SMinchan Kim ret = invalidate_inode_page(page); 4941da177e4SLinus Torvalds unlock_page(page); 49531560180SMinchan Kim /* 49631560180SMinchan Kim * Invalidation is a hint that the page is no longer 49731560180SMinchan Kim * of interest and try to speed up its reclaim. 49831560180SMinchan Kim */ 49931560180SMinchan Kim if (!ret) 500cc5993bdSMinchan Kim deactivate_file_page(page); 50131560180SMinchan Kim count += ret; 5021da177e4SLinus Torvalds } 5030cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 5041da177e4SLinus Torvalds pagevec_release(&pvec); 505fc9a07e7SAndrew Morton cond_resched(); 506b85e0effSHugh Dickins index++; 5071da177e4SLinus Torvalds } 50831560180SMinchan Kim return count; 5091da177e4SLinus Torvalds } 51054bc4855SAnton Altaparmakov EXPORT_SYMBOL(invalidate_mapping_pages); 5111da177e4SLinus Torvalds 512bd4c8ce4SAndrew Morton /* 513bd4c8ce4SAndrew Morton * This is like invalidate_complete_page(), except it ignores the page's 514bd4c8ce4SAndrew Morton * refcount. We do this because invalidate_inode_pages2() needs stronger 515bd4c8ce4SAndrew Morton * invalidation guarantees, and cannot afford to leave pages behind because 5162706a1b8SAnderson Briglia * shrink_page_list() has a temp ref on them, or because they're transiently 5172706a1b8SAnderson Briglia * sitting in the lru_cache_add() pagevecs. 518bd4c8ce4SAndrew Morton */ 519bd4c8ce4SAndrew Morton static int 520bd4c8ce4SAndrew Morton invalidate_complete_page2(struct address_space *mapping, struct page *page) 521bd4c8ce4SAndrew Morton { 522c4843a75SGreg Thelen struct mem_cgroup *memcg; 523c4843a75SGreg Thelen unsigned long flags; 524c4843a75SGreg Thelen 525bd4c8ce4SAndrew Morton if (page->mapping != mapping) 526bd4c8ce4SAndrew Morton return 0; 527bd4c8ce4SAndrew Morton 528266cf658SDavid Howells if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) 529bd4c8ce4SAndrew Morton return 0; 530bd4c8ce4SAndrew Morton 531c4843a75SGreg Thelen memcg = mem_cgroup_begin_page_stat(page); 532c4843a75SGreg Thelen spin_lock_irqsave(&mapping->tree_lock, flags); 533bd4c8ce4SAndrew Morton if (PageDirty(page)) 534bd4c8ce4SAndrew Morton goto failed; 535bd4c8ce4SAndrew Morton 536266cf658SDavid Howells BUG_ON(page_has_private(page)); 537c4843a75SGreg Thelen __delete_from_page_cache(page, NULL, memcg); 538c4843a75SGreg Thelen spin_unlock_irqrestore(&mapping->tree_lock, flags); 539c4843a75SGreg Thelen mem_cgroup_end_page_stat(memcg); 5406072d13cSLinus Torvalds 5416072d13cSLinus Torvalds if (mapping->a_ops->freepage) 5426072d13cSLinus Torvalds mapping->a_ops->freepage(page); 5436072d13cSLinus Torvalds 544bd4c8ce4SAndrew Morton page_cache_release(page); /* pagecache ref */ 545bd4c8ce4SAndrew Morton return 1; 546bd4c8ce4SAndrew Morton failed: 547c4843a75SGreg Thelen spin_unlock_irqrestore(&mapping->tree_lock, flags); 548c4843a75SGreg Thelen mem_cgroup_end_page_stat(memcg); 549bd4c8ce4SAndrew Morton return 0; 550bd4c8ce4SAndrew Morton } 551bd4c8ce4SAndrew Morton 552e3db7691STrond Myklebust static int do_launder_page(struct address_space *mapping, struct page *page) 553e3db7691STrond Myklebust { 554e3db7691STrond Myklebust if (!PageDirty(page)) 555e3db7691STrond Myklebust return 0; 556e3db7691STrond Myklebust if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) 557e3db7691STrond Myklebust return 0; 558e3db7691STrond Myklebust return mapping->a_ops->launder_page(page); 559e3db7691STrond Myklebust } 560e3db7691STrond Myklebust 5611da177e4SLinus Torvalds /** 5621da177e4SLinus Torvalds * invalidate_inode_pages2_range - remove range of pages from an address_space 56367be2dd1SMartin Waitz * @mapping: the address_space 5641da177e4SLinus Torvalds * @start: the page offset 'from' which to invalidate 5651da177e4SLinus Torvalds * @end: the page offset 'to' which to invalidate (inclusive) 5661da177e4SLinus Torvalds * 5671da177e4SLinus Torvalds * Any pages which are found to be mapped into pagetables are unmapped prior to 5681da177e4SLinus Torvalds * invalidation. 5691da177e4SLinus Torvalds * 5706ccfa806SHisashi Hifumi * Returns -EBUSY if any pages could not be invalidated. 5711da177e4SLinus Torvalds */ 5721da177e4SLinus Torvalds int invalidate_inode_pages2_range(struct address_space *mapping, 5731da177e4SLinus Torvalds pgoff_t start, pgoff_t end) 5741da177e4SLinus Torvalds { 5750cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 5761da177e4SLinus Torvalds struct pagevec pvec; 577b85e0effSHugh Dickins pgoff_t index; 5781da177e4SLinus Torvalds int i; 5791da177e4SLinus Torvalds int ret = 0; 5800dd1334fSHisashi Hifumi int ret2 = 0; 5811da177e4SLinus Torvalds int did_range_unmap = 0; 5821da177e4SLinus Torvalds 5833167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 5841da177e4SLinus Torvalds pagevec_init(&pvec, 0); 585b85e0effSHugh Dickins index = start; 5860cd6144aSJohannes Weiner while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, 5870cd6144aSJohannes Weiner min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, 5880cd6144aSJohannes Weiner indices)) { 5897b965e08STrond Myklebust for (i = 0; i < pagevec_count(&pvec); i++) { 5901da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 591b85e0effSHugh Dickins 592b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 5930cd6144aSJohannes Weiner index = indices[i]; 594b85e0effSHugh Dickins if (index > end) 595b85e0effSHugh Dickins break; 5961da177e4SLinus Torvalds 5970cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 5980cd6144aSJohannes Weiner clear_exceptional_entry(mapping, index, page); 5990cd6144aSJohannes Weiner continue; 6000cd6144aSJohannes Weiner } 6010cd6144aSJohannes Weiner 6021da177e4SLinus Torvalds lock_page(page); 603b85e0effSHugh Dickins WARN_ON(page->index != index); 6041da177e4SLinus Torvalds if (page->mapping != mapping) { 6051da177e4SLinus Torvalds unlock_page(page); 6061da177e4SLinus Torvalds continue; 6071da177e4SLinus Torvalds } 6081da177e4SLinus Torvalds wait_on_page_writeback(page); 609d00806b1SNick Piggin if (page_mapped(page)) { 6101da177e4SLinus Torvalds if (!did_range_unmap) { 6111da177e4SLinus Torvalds /* 6121da177e4SLinus Torvalds * Zap the rest of the file in one hit. 6131da177e4SLinus Torvalds */ 6141da177e4SLinus Torvalds unmap_mapping_range(mapping, 615b85e0effSHugh Dickins (loff_t)index << PAGE_CACHE_SHIFT, 616b85e0effSHugh Dickins (loff_t)(1 + end - index) 6171da177e4SLinus Torvalds << PAGE_CACHE_SHIFT, 6181da177e4SLinus Torvalds 0); 6191da177e4SLinus Torvalds did_range_unmap = 1; 6201da177e4SLinus Torvalds } else { 6211da177e4SLinus Torvalds /* 6221da177e4SLinus Torvalds * Just zap this page 6231da177e4SLinus Torvalds */ 6241da177e4SLinus Torvalds unmap_mapping_range(mapping, 625b85e0effSHugh Dickins (loff_t)index << PAGE_CACHE_SHIFT, 6261da177e4SLinus Torvalds PAGE_CACHE_SIZE, 0); 6271da177e4SLinus Torvalds } 6281da177e4SLinus Torvalds } 629d00806b1SNick Piggin BUG_ON(page_mapped(page)); 6300dd1334fSHisashi Hifumi ret2 = do_launder_page(mapping, page); 6310dd1334fSHisashi Hifumi if (ret2 == 0) { 6320dd1334fSHisashi Hifumi if (!invalidate_complete_page2(mapping, page)) 6336ccfa806SHisashi Hifumi ret2 = -EBUSY; 6340dd1334fSHisashi Hifumi } 6350dd1334fSHisashi Hifumi if (ret2 < 0) 6360dd1334fSHisashi Hifumi ret = ret2; 6371da177e4SLinus Torvalds unlock_page(page); 6381da177e4SLinus Torvalds } 6390cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 6401da177e4SLinus Torvalds pagevec_release(&pvec); 6411da177e4SLinus Torvalds cond_resched(); 642b85e0effSHugh Dickins index++; 6431da177e4SLinus Torvalds } 6443167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 6451da177e4SLinus Torvalds return ret; 6461da177e4SLinus Torvalds } 6471da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); 6481da177e4SLinus Torvalds 6491da177e4SLinus Torvalds /** 6501da177e4SLinus Torvalds * invalidate_inode_pages2 - remove all pages from an address_space 65167be2dd1SMartin Waitz * @mapping: the address_space 6521da177e4SLinus Torvalds * 6531da177e4SLinus Torvalds * Any pages which are found to be mapped into pagetables are unmapped prior to 6541da177e4SLinus Torvalds * invalidation. 6551da177e4SLinus Torvalds * 656e9de25ddSPeng Tao * Returns -EBUSY if any pages could not be invalidated. 6571da177e4SLinus Torvalds */ 6581da177e4SLinus Torvalds int invalidate_inode_pages2(struct address_space *mapping) 6591da177e4SLinus Torvalds { 6601da177e4SLinus Torvalds return invalidate_inode_pages2_range(mapping, 0, -1); 6611da177e4SLinus Torvalds } 6621da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(invalidate_inode_pages2); 66325d9e2d1Snpiggin@suse.de 66425d9e2d1Snpiggin@suse.de /** 66525d9e2d1Snpiggin@suse.de * truncate_pagecache - unmap and remove pagecache that has been truncated 66625d9e2d1Snpiggin@suse.de * @inode: inode 6678a549beaSHugh Dickins * @newsize: new file size 66825d9e2d1Snpiggin@suse.de * 66925d9e2d1Snpiggin@suse.de * inode's new i_size must already be written before truncate_pagecache 67025d9e2d1Snpiggin@suse.de * is called. 67125d9e2d1Snpiggin@suse.de * 67225d9e2d1Snpiggin@suse.de * This function should typically be called before the filesystem 67325d9e2d1Snpiggin@suse.de * releases resources associated with the freed range (eg. deallocates 67425d9e2d1Snpiggin@suse.de * blocks). This way, pagecache will always stay logically coherent 67525d9e2d1Snpiggin@suse.de * with on-disk format, and the filesystem would not have to deal with 67625d9e2d1Snpiggin@suse.de * situations such as writepage being called for a page that has already 67725d9e2d1Snpiggin@suse.de * had its underlying blocks deallocated. 67825d9e2d1Snpiggin@suse.de */ 6797caef267SKirill A. Shutemov void truncate_pagecache(struct inode *inode, loff_t newsize) 68025d9e2d1Snpiggin@suse.de { 68125d9e2d1Snpiggin@suse.de struct address_space *mapping = inode->i_mapping; 6828a549beaSHugh Dickins loff_t holebegin = round_up(newsize, PAGE_SIZE); 68325d9e2d1Snpiggin@suse.de 68425d9e2d1Snpiggin@suse.de /* 68525d9e2d1Snpiggin@suse.de * unmap_mapping_range is called twice, first simply for 68625d9e2d1Snpiggin@suse.de * efficiency so that truncate_inode_pages does fewer 68725d9e2d1Snpiggin@suse.de * single-page unmaps. However after this first call, and 68825d9e2d1Snpiggin@suse.de * before truncate_inode_pages finishes, it is possible for 68925d9e2d1Snpiggin@suse.de * private pages to be COWed, which remain after 69025d9e2d1Snpiggin@suse.de * truncate_inode_pages finishes, hence the second 69125d9e2d1Snpiggin@suse.de * unmap_mapping_range call must be made for correctness. 69225d9e2d1Snpiggin@suse.de */ 6938a549beaSHugh Dickins unmap_mapping_range(mapping, holebegin, 0, 1); 6948a549beaSHugh Dickins truncate_inode_pages(mapping, newsize); 6958a549beaSHugh Dickins unmap_mapping_range(mapping, holebegin, 0, 1); 69625d9e2d1Snpiggin@suse.de } 69725d9e2d1Snpiggin@suse.de EXPORT_SYMBOL(truncate_pagecache); 69825d9e2d1Snpiggin@suse.de 69925d9e2d1Snpiggin@suse.de /** 7002c27c65eSChristoph Hellwig * truncate_setsize - update inode and pagecache for a new file size 7012c27c65eSChristoph Hellwig * @inode: inode 7022c27c65eSChristoph Hellwig * @newsize: new file size 7032c27c65eSChristoph Hellwig * 704382e27daSJan Kara * truncate_setsize updates i_size and performs pagecache truncation (if 705382e27daSJan Kara * necessary) to @newsize. It will be typically be called from the filesystem's 706382e27daSJan Kara * setattr function when ATTR_SIZE is passed in. 7072c27c65eSChristoph Hellwig * 70877783d06SJan Kara * Must be called with a lock serializing truncates and writes (generally 70977783d06SJan Kara * i_mutex but e.g. xfs uses a different lock) and before all filesystem 71077783d06SJan Kara * specific block truncation has been performed. 7112c27c65eSChristoph Hellwig */ 7122c27c65eSChristoph Hellwig void truncate_setsize(struct inode *inode, loff_t newsize) 7132c27c65eSChristoph Hellwig { 71490a80202SJan Kara loff_t oldsize = inode->i_size; 71590a80202SJan Kara 7162c27c65eSChristoph Hellwig i_size_write(inode, newsize); 71790a80202SJan Kara if (newsize > oldsize) 71890a80202SJan Kara pagecache_isize_extended(inode, oldsize, newsize); 7197caef267SKirill A. Shutemov truncate_pagecache(inode, newsize); 7202c27c65eSChristoph Hellwig } 7212c27c65eSChristoph Hellwig EXPORT_SYMBOL(truncate_setsize); 7222c27c65eSChristoph Hellwig 7232c27c65eSChristoph Hellwig /** 72490a80202SJan Kara * pagecache_isize_extended - update pagecache after extension of i_size 72590a80202SJan Kara * @inode: inode for which i_size was extended 72690a80202SJan Kara * @from: original inode size 72790a80202SJan Kara * @to: new inode size 72890a80202SJan Kara * 72990a80202SJan Kara * Handle extension of inode size either caused by extending truncate or by 73090a80202SJan Kara * write starting after current i_size. We mark the page straddling current 73190a80202SJan Kara * i_size RO so that page_mkwrite() is called on the nearest write access to 73290a80202SJan Kara * the page. This way filesystem can be sure that page_mkwrite() is called on 73390a80202SJan Kara * the page before user writes to the page via mmap after the i_size has been 73490a80202SJan Kara * changed. 73590a80202SJan Kara * 73690a80202SJan Kara * The function must be called after i_size is updated so that page fault 73790a80202SJan Kara * coming after we unlock the page will already see the new i_size. 73890a80202SJan Kara * The function must be called while we still hold i_mutex - this not only 73990a80202SJan Kara * makes sure i_size is stable but also that userspace cannot observe new 74090a80202SJan Kara * i_size value before we are prepared to store mmap writes at new inode size. 74190a80202SJan Kara */ 74290a80202SJan Kara void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) 74390a80202SJan Kara { 74490a80202SJan Kara int bsize = 1 << inode->i_blkbits; 74590a80202SJan Kara loff_t rounded_from; 74690a80202SJan Kara struct page *page; 74790a80202SJan Kara pgoff_t index; 74890a80202SJan Kara 74990a80202SJan Kara WARN_ON(to > inode->i_size); 75090a80202SJan Kara 75190a80202SJan Kara if (from >= to || bsize == PAGE_CACHE_SIZE) 75290a80202SJan Kara return; 75390a80202SJan Kara /* Page straddling @from will not have any hole block created? */ 75490a80202SJan Kara rounded_from = round_up(from, bsize); 75590a80202SJan Kara if (to <= rounded_from || !(rounded_from & (PAGE_CACHE_SIZE - 1))) 75690a80202SJan Kara return; 75790a80202SJan Kara 75890a80202SJan Kara index = from >> PAGE_CACHE_SHIFT; 75990a80202SJan Kara page = find_lock_page(inode->i_mapping, index); 76090a80202SJan Kara /* Page not cached? Nothing to do */ 76190a80202SJan Kara if (!page) 76290a80202SJan Kara return; 76390a80202SJan Kara /* 76490a80202SJan Kara * See clear_page_dirty_for_io() for details why set_page_dirty() 76590a80202SJan Kara * is needed. 76690a80202SJan Kara */ 76790a80202SJan Kara if (page_mkclean(page)) 76890a80202SJan Kara set_page_dirty(page); 76990a80202SJan Kara unlock_page(page); 77090a80202SJan Kara page_cache_release(page); 77190a80202SJan Kara } 77290a80202SJan Kara EXPORT_SYMBOL(pagecache_isize_extended); 77390a80202SJan Kara 77490a80202SJan Kara /** 775623e3db9SHugh Dickins * truncate_pagecache_range - unmap and remove pagecache that is hole-punched 776623e3db9SHugh Dickins * @inode: inode 777623e3db9SHugh Dickins * @lstart: offset of beginning of hole 778623e3db9SHugh Dickins * @lend: offset of last byte of hole 779623e3db9SHugh Dickins * 780623e3db9SHugh Dickins * This function should typically be called before the filesystem 781623e3db9SHugh Dickins * releases resources associated with the freed range (eg. deallocates 782623e3db9SHugh Dickins * blocks). This way, pagecache will always stay logically coherent 783623e3db9SHugh Dickins * with on-disk format, and the filesystem would not have to deal with 784623e3db9SHugh Dickins * situations such as writepage being called for a page that has already 785623e3db9SHugh Dickins * had its underlying blocks deallocated. 786623e3db9SHugh Dickins */ 787623e3db9SHugh Dickins void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) 788623e3db9SHugh Dickins { 789623e3db9SHugh Dickins struct address_space *mapping = inode->i_mapping; 790623e3db9SHugh Dickins loff_t unmap_start = round_up(lstart, PAGE_SIZE); 791623e3db9SHugh Dickins loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; 792623e3db9SHugh Dickins /* 793623e3db9SHugh Dickins * This rounding is currently just for example: unmap_mapping_range 794623e3db9SHugh Dickins * expands its hole outwards, whereas we want it to contract the hole 795623e3db9SHugh Dickins * inwards. However, existing callers of truncate_pagecache_range are 7965a720394SLukas Czerner * doing their own page rounding first. Note that unmap_mapping_range 7975a720394SLukas Czerner * allows holelen 0 for all, and we allow lend -1 for end of file. 798623e3db9SHugh Dickins */ 799623e3db9SHugh Dickins 800623e3db9SHugh Dickins /* 801623e3db9SHugh Dickins * Unlike in truncate_pagecache, unmap_mapping_range is called only 802623e3db9SHugh Dickins * once (before truncating pagecache), and without "even_cows" flag: 803623e3db9SHugh Dickins * hole-punching should not remove private COWed pages from the hole. 804623e3db9SHugh Dickins */ 805623e3db9SHugh Dickins if ((u64)unmap_end > (u64)unmap_start) 806623e3db9SHugh Dickins unmap_mapping_range(mapping, unmap_start, 807623e3db9SHugh Dickins 1 + unmap_end - unmap_start, 0); 808623e3db9SHugh Dickins truncate_inode_pages_range(mapping, lstart, lend); 809623e3db9SHugh Dickins } 810623e3db9SHugh Dickins EXPORT_SYMBOL(truncate_pagecache_range); 811