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> 125a0e3ad6STejun Heo #include <linux/gfp.h> 131da177e4SLinus Torvalds #include <linux/mm.h> 140fd0e6b0SNick Piggin #include <linux/swap.h> 15b95f1b31SPaul Gortmaker #include <linux/export.h> 161da177e4SLinus Torvalds #include <linux/pagemap.h> 1701f2705dSNate Diller #include <linux/highmem.h> 181da177e4SLinus Torvalds #include <linux/pagevec.h> 19e08748ceSAndrew Morton #include <linux/task_io_accounting_ops.h> 201da177e4SLinus Torvalds #include <linux/buffer_head.h> /* grr. try_to_release_page, 21aaa4059bSJan Kara do_invalidatepage */ 22c515e1fdSDan Magenheimer #include <linux/cleancache.h> 2390a80202SJan Kara #include <linux/rmap.h> 24ba470de4SRik van Riel #include "internal.h" 251da177e4SLinus Torvalds 260cd6144aSJohannes Weiner static void clear_exceptional_entry(struct address_space *mapping, 270cd6144aSJohannes Weiner pgoff_t index, void *entry) 280cd6144aSJohannes Weiner { 29449dd698SJohannes Weiner struct radix_tree_node *node; 30449dd698SJohannes Weiner void **slot; 31449dd698SJohannes Weiner 320cd6144aSJohannes Weiner /* Handled by shmem itself */ 330cd6144aSJohannes Weiner if (shmem_mapping(mapping)) 340cd6144aSJohannes Weiner return; 350cd6144aSJohannes Weiner 360cd6144aSJohannes Weiner spin_lock_irq(&mapping->tree_lock); 370cd6144aSJohannes Weiner /* 380cd6144aSJohannes Weiner * Regular page slots are stabilized by the page lock even 390cd6144aSJohannes Weiner * without the tree itself locked. These unlocked entries 400cd6144aSJohannes Weiner * need verification under the tree lock. 410cd6144aSJohannes Weiner */ 42449dd698SJohannes Weiner if (!__radix_tree_lookup(&mapping->page_tree, index, &node, &slot)) 43449dd698SJohannes Weiner goto unlock; 44449dd698SJohannes Weiner if (*slot != entry) 45449dd698SJohannes Weiner goto unlock; 46449dd698SJohannes Weiner radix_tree_replace_slot(slot, NULL); 4791b0abe3SJohannes Weiner mapping->nrshadows--; 48449dd698SJohannes Weiner if (!node) 49449dd698SJohannes Weiner goto unlock; 50449dd698SJohannes Weiner workingset_node_shadows_dec(node); 51449dd698SJohannes Weiner /* 52449dd698SJohannes Weiner * Don't track node without shadow entries. 53449dd698SJohannes Weiner * 54449dd698SJohannes Weiner * Avoid acquiring the list_lru lock if already untracked. 55449dd698SJohannes Weiner * The list_empty() test is safe as node->private_list is 56449dd698SJohannes Weiner * protected by mapping->tree_lock. 57449dd698SJohannes Weiner */ 58449dd698SJohannes Weiner if (!workingset_node_shadows(node) && 59449dd698SJohannes Weiner !list_empty(&node->private_list)) 60449dd698SJohannes Weiner list_lru_del(&workingset_shadow_nodes, &node->private_list); 61449dd698SJohannes Weiner __radix_tree_delete_node(&mapping->page_tree, node); 62449dd698SJohannes Weiner unlock: 630cd6144aSJohannes Weiner spin_unlock_irq(&mapping->tree_lock); 640cd6144aSJohannes Weiner } 651da177e4SLinus Torvalds 66cf9a2ae8SDavid Howells /** 6728bc44d7SFengguang Wu * do_invalidatepage - invalidate part or all of a page 68cf9a2ae8SDavid Howells * @page: the page which is affected 69d47992f8SLukas Czerner * @offset: start of the range to invalidate 70d47992f8SLukas Czerner * @length: length of the range to invalidate 71cf9a2ae8SDavid Howells * 72cf9a2ae8SDavid Howells * do_invalidatepage() is called when all or part of the page has become 73cf9a2ae8SDavid Howells * invalidated by a truncate operation. 74cf9a2ae8SDavid Howells * 75cf9a2ae8SDavid Howells * do_invalidatepage() does not have to release all buffers, but it must 76cf9a2ae8SDavid Howells * ensure that no dirty buffer is left outside @offset and that no I/O 77cf9a2ae8SDavid Howells * is underway against any of the blocks which are outside the truncation 78cf9a2ae8SDavid Howells * point. Because the caller is about to free (and possibly reuse) those 79cf9a2ae8SDavid Howells * blocks on-disk. 80cf9a2ae8SDavid Howells */ 81d47992f8SLukas Czerner void do_invalidatepage(struct page *page, unsigned int offset, 82d47992f8SLukas Czerner unsigned int length) 83cf9a2ae8SDavid Howells { 84d47992f8SLukas Czerner void (*invalidatepage)(struct page *, unsigned int, unsigned int); 85d47992f8SLukas Czerner 86cf9a2ae8SDavid Howells invalidatepage = page->mapping->a_ops->invalidatepage; 879361401eSDavid Howells #ifdef CONFIG_BLOCK 88cf9a2ae8SDavid Howells if (!invalidatepage) 89cf9a2ae8SDavid Howells invalidatepage = block_invalidatepage; 909361401eSDavid Howells #endif 91cf9a2ae8SDavid Howells if (invalidatepage) 92d47992f8SLukas Czerner (*invalidatepage)(page, offset, length); 93cf9a2ae8SDavid Howells } 94cf9a2ae8SDavid Howells 95ecdfc978SLinus Torvalds /* 961da177e4SLinus Torvalds * If truncate cannot remove the fs-private metadata from the page, the page 9762e1c553SShaohua Li * becomes orphaned. It will be left on the LRU and may even be mapped into 9854cb8821SNick Piggin * user pagetables if we're racing with filemap_fault(). 991da177e4SLinus Torvalds * 1001da177e4SLinus Torvalds * We need to bale out if page->mapping is no longer equal to the original 1011da177e4SLinus Torvalds * mapping. This happens a) when the VM reclaimed the page while we waited on 102fc0ecff6SAndrew Morton * its lock, b) when a concurrent invalidate_mapping_pages got there first and 1031da177e4SLinus Torvalds * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. 1041da177e4SLinus Torvalds */ 105750b4987SNick Piggin static int 1061da177e4SLinus Torvalds truncate_complete_page(struct address_space *mapping, struct page *page) 1071da177e4SLinus Torvalds { 1081da177e4SLinus Torvalds if (page->mapping != mapping) 109750b4987SNick Piggin return -EIO; 1101da177e4SLinus Torvalds 111266cf658SDavid Howells if (page_has_private(page)) 112d47992f8SLukas Czerner do_invalidatepage(page, 0, PAGE_CACHE_SIZE); 1131da177e4SLinus Torvalds 114*b9ea2515SKonstantin Khlebnikov /* 115*b9ea2515SKonstantin Khlebnikov * Some filesystems seem to re-dirty the page even after 116*b9ea2515SKonstantin Khlebnikov * the VM has canceled the dirty bit (eg ext3 journaling). 117*b9ea2515SKonstantin Khlebnikov * Hence dirty accounting check is placed after invalidation. 118*b9ea2515SKonstantin Khlebnikov */ 119*b9ea2515SKonstantin Khlebnikov if (TestClearPageDirty(page)) 120*b9ea2515SKonstantin Khlebnikov account_page_cleaned(page, mapping); 121a2b34564SBjorn Steinbrink 1221da177e4SLinus Torvalds ClearPageMappedToDisk(page); 1235adc7b51SMinchan Kim delete_from_page_cache(page); 124750b4987SNick Piggin return 0; 1251da177e4SLinus Torvalds } 1261da177e4SLinus Torvalds 1271da177e4SLinus Torvalds /* 128fc0ecff6SAndrew Morton * This is for invalidate_mapping_pages(). That function can be called at 1291da177e4SLinus Torvalds * any time, and is not supposed to throw away dirty pages. But pages can 1300fd0e6b0SNick Piggin * be marked dirty at any time too, so use remove_mapping which safely 1310fd0e6b0SNick Piggin * discards clean, unused pages. 1321da177e4SLinus Torvalds * 1331da177e4SLinus Torvalds * Returns non-zero if the page was successfully invalidated. 1341da177e4SLinus Torvalds */ 1351da177e4SLinus Torvalds static int 1361da177e4SLinus Torvalds invalidate_complete_page(struct address_space *mapping, struct page *page) 1371da177e4SLinus Torvalds { 1380fd0e6b0SNick Piggin int ret; 1390fd0e6b0SNick Piggin 1401da177e4SLinus Torvalds if (page->mapping != mapping) 1411da177e4SLinus Torvalds return 0; 1421da177e4SLinus Torvalds 143266cf658SDavid Howells if (page_has_private(page) && !try_to_release_page(page, 0)) 1441da177e4SLinus Torvalds return 0; 1451da177e4SLinus Torvalds 1460fd0e6b0SNick Piggin ret = remove_mapping(mapping, page); 1470fd0e6b0SNick Piggin 1480fd0e6b0SNick Piggin return ret; 1491da177e4SLinus Torvalds } 1501da177e4SLinus Torvalds 151750b4987SNick Piggin int truncate_inode_page(struct address_space *mapping, struct page *page) 152750b4987SNick Piggin { 153750b4987SNick Piggin if (page_mapped(page)) { 154750b4987SNick Piggin unmap_mapping_range(mapping, 155750b4987SNick Piggin (loff_t)page->index << PAGE_CACHE_SHIFT, 156750b4987SNick Piggin PAGE_CACHE_SIZE, 0); 157750b4987SNick Piggin } 158750b4987SNick Piggin return truncate_complete_page(mapping, page); 159750b4987SNick Piggin } 160750b4987SNick Piggin 16183f78668SWu Fengguang /* 16225718736SAndi Kleen * Used to get rid of pages on hardware memory corruption. 16325718736SAndi Kleen */ 16425718736SAndi Kleen int generic_error_remove_page(struct address_space *mapping, struct page *page) 16525718736SAndi Kleen { 16625718736SAndi Kleen if (!mapping) 16725718736SAndi Kleen return -EINVAL; 16825718736SAndi Kleen /* 16925718736SAndi Kleen * Only punch for normal data pages for now. 17025718736SAndi Kleen * Handling other types like directories would need more auditing. 17125718736SAndi Kleen */ 17225718736SAndi Kleen if (!S_ISREG(mapping->host->i_mode)) 17325718736SAndi Kleen return -EIO; 17425718736SAndi Kleen return truncate_inode_page(mapping, page); 17525718736SAndi Kleen } 17625718736SAndi Kleen EXPORT_SYMBOL(generic_error_remove_page); 17725718736SAndi Kleen 17825718736SAndi Kleen /* 17983f78668SWu Fengguang * Safely invalidate one page from its pagecache mapping. 18083f78668SWu Fengguang * It only drops clean, unused pages. The page must be locked. 18183f78668SWu Fengguang * 18283f78668SWu Fengguang * Returns 1 if the page is successfully invalidated, otherwise 0. 18383f78668SWu Fengguang */ 18483f78668SWu Fengguang int invalidate_inode_page(struct page *page) 18583f78668SWu Fengguang { 18683f78668SWu Fengguang struct address_space *mapping = page_mapping(page); 18783f78668SWu Fengguang if (!mapping) 18883f78668SWu Fengguang return 0; 18983f78668SWu Fengguang if (PageDirty(page) || PageWriteback(page)) 19083f78668SWu Fengguang return 0; 19183f78668SWu Fengguang if (page_mapped(page)) 19283f78668SWu Fengguang return 0; 19383f78668SWu Fengguang return invalidate_complete_page(mapping, page); 19483f78668SWu Fengguang } 19583f78668SWu Fengguang 1961da177e4SLinus Torvalds /** 19773c1e204SLiu Bo * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets 1981da177e4SLinus Torvalds * @mapping: mapping to truncate 1991da177e4SLinus Torvalds * @lstart: offset from which to truncate 2005a720394SLukas Czerner * @lend: offset to which to truncate (inclusive) 2011da177e4SLinus Torvalds * 202d7339071SHans Reiser * Truncate the page cache, removing the pages that are between 2035a720394SLukas Czerner * specified offsets (and zeroing out partial pages 2045a720394SLukas Czerner * if lstart or lend + 1 is not page aligned). 2051da177e4SLinus Torvalds * 2061da177e4SLinus Torvalds * Truncate takes two passes - the first pass is nonblocking. It will not 2071da177e4SLinus Torvalds * block on page locks and it will not block on writeback. The second pass 2081da177e4SLinus Torvalds * will wait. This is to prevent as much IO as possible in the affected region. 2091da177e4SLinus Torvalds * The first pass will remove most pages, so the search cost of the second pass 2101da177e4SLinus Torvalds * is low. 2111da177e4SLinus Torvalds * 2121da177e4SLinus Torvalds * We pass down the cache-hot hint to the page freeing code. Even if the 2131da177e4SLinus Torvalds * mapping is large, it is probably the case that the final pages are the most 2141da177e4SLinus Torvalds * recently touched, and freeing happens in ascending file offset order. 2155a720394SLukas Czerner * 2165a720394SLukas Czerner * Note that since ->invalidatepage() accepts range to invalidate 2175a720394SLukas Czerner * truncate_inode_pages_range is able to handle cases where lend + 1 is not 2185a720394SLukas Czerner * page aligned properly. 2191da177e4SLinus Torvalds */ 220d7339071SHans Reiser void truncate_inode_pages_range(struct address_space *mapping, 221d7339071SHans Reiser loff_t lstart, loff_t lend) 2221da177e4SLinus Torvalds { 2235a720394SLukas Czerner pgoff_t start; /* inclusive */ 2245a720394SLukas Czerner pgoff_t end; /* exclusive */ 2255a720394SLukas Czerner unsigned int partial_start; /* inclusive */ 2265a720394SLukas Czerner unsigned int partial_end; /* exclusive */ 2271da177e4SLinus Torvalds struct pagevec pvec; 2280cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 229b85e0effSHugh Dickins pgoff_t index; 2301da177e4SLinus Torvalds int i; 2311da177e4SLinus Torvalds 2323167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 23391b0abe3SJohannes Weiner if (mapping->nrpages == 0 && mapping->nrshadows == 0) 2341da177e4SLinus Torvalds return; 2351da177e4SLinus Torvalds 2365a720394SLukas Czerner /* Offsets within partial pages */ 2375a720394SLukas Czerner partial_start = lstart & (PAGE_CACHE_SIZE - 1); 2385a720394SLukas Czerner partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1); 2395a720394SLukas Czerner 2405a720394SLukas Czerner /* 2415a720394SLukas Czerner * 'start' and 'end' always covers the range of pages to be fully 2425a720394SLukas Czerner * truncated. Partial pages are covered with 'partial_start' at the 2435a720394SLukas Czerner * start of the range and 'partial_end' at the end of the range. 2445a720394SLukas Czerner * Note that 'end' is exclusive while 'lend' is inclusive. 2455a720394SLukas Czerner */ 2465a720394SLukas Czerner start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 2475a720394SLukas Czerner if (lend == -1) 2485a720394SLukas Czerner /* 2495a720394SLukas Czerner * lend == -1 indicates end-of-file so we have to set 'end' 2505a720394SLukas Czerner * to the highest possible pgoff_t and since the type is 2515a720394SLukas Czerner * unsigned we're using -1. 2525a720394SLukas Czerner */ 2535a720394SLukas Czerner end = -1; 2545a720394SLukas Czerner else 2555a720394SLukas Czerner end = (lend + 1) >> PAGE_CACHE_SHIFT; 256d7339071SHans Reiser 2571da177e4SLinus Torvalds pagevec_init(&pvec, 0); 258b85e0effSHugh Dickins index = start; 2590cd6144aSJohannes Weiner while (index < end && pagevec_lookup_entries(&pvec, mapping, index, 2600cd6144aSJohannes Weiner min(end - index, (pgoff_t)PAGEVEC_SIZE), 2610cd6144aSJohannes Weiner indices)) { 2621da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 2631da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 2641da177e4SLinus Torvalds 265b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 2660cd6144aSJohannes Weiner index = indices[i]; 2675a720394SLukas Czerner if (index >= end) 268d7339071SHans Reiser break; 269d7339071SHans Reiser 2700cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 2710cd6144aSJohannes Weiner clear_exceptional_entry(mapping, index, page); 2720cd6144aSJohannes Weiner continue; 2730cd6144aSJohannes Weiner } 2740cd6144aSJohannes Weiner 275529ae9aaSNick Piggin if (!trylock_page(page)) 2761da177e4SLinus Torvalds continue; 277b85e0effSHugh Dickins WARN_ON(page->index != index); 2781da177e4SLinus Torvalds if (PageWriteback(page)) { 2791da177e4SLinus Torvalds unlock_page(page); 2801da177e4SLinus Torvalds continue; 2811da177e4SLinus Torvalds } 282750b4987SNick Piggin truncate_inode_page(mapping, page); 2831da177e4SLinus Torvalds unlock_page(page); 2841da177e4SLinus Torvalds } 2850cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 2861da177e4SLinus Torvalds pagevec_release(&pvec); 2871da177e4SLinus Torvalds cond_resched(); 288b85e0effSHugh Dickins index++; 2891da177e4SLinus Torvalds } 2901da177e4SLinus Torvalds 2915a720394SLukas Czerner if (partial_start) { 2921da177e4SLinus Torvalds struct page *page = find_lock_page(mapping, start - 1); 2931da177e4SLinus Torvalds if (page) { 2945a720394SLukas Czerner unsigned int top = PAGE_CACHE_SIZE; 2955a720394SLukas Czerner if (start > end) { 2965a720394SLukas Czerner /* Truncation within a single page */ 2975a720394SLukas Czerner top = partial_end; 2985a720394SLukas Czerner partial_end = 0; 2995a720394SLukas Czerner } 3001da177e4SLinus Torvalds wait_on_page_writeback(page); 3015a720394SLukas Czerner zero_user_segment(page, partial_start, top); 3025a720394SLukas Czerner cleancache_invalidate_page(mapping, page); 3035a720394SLukas Czerner if (page_has_private(page)) 3045a720394SLukas Czerner do_invalidatepage(page, partial_start, 3055a720394SLukas Czerner top - partial_start); 3061da177e4SLinus Torvalds unlock_page(page); 3071da177e4SLinus Torvalds page_cache_release(page); 3081da177e4SLinus Torvalds } 3091da177e4SLinus Torvalds } 3105a720394SLukas Czerner if (partial_end) { 3115a720394SLukas Czerner struct page *page = find_lock_page(mapping, end); 3125a720394SLukas Czerner if (page) { 3135a720394SLukas Czerner wait_on_page_writeback(page); 3145a720394SLukas Czerner zero_user_segment(page, 0, partial_end); 3155a720394SLukas Czerner cleancache_invalidate_page(mapping, page); 3165a720394SLukas Czerner if (page_has_private(page)) 3175a720394SLukas Czerner do_invalidatepage(page, 0, 3185a720394SLukas Czerner partial_end); 3195a720394SLukas Czerner unlock_page(page); 3205a720394SLukas Czerner page_cache_release(page); 3215a720394SLukas Czerner } 3225a720394SLukas Czerner } 3235a720394SLukas Czerner /* 3245a720394SLukas Czerner * If the truncation happened within a single page no pages 3255a720394SLukas Czerner * will be released, just zeroed, so we can bail out now. 3265a720394SLukas Czerner */ 3275a720394SLukas Czerner if (start >= end) 3285a720394SLukas Czerner return; 3291da177e4SLinus Torvalds 330b85e0effSHugh Dickins index = start; 3311da177e4SLinus Torvalds for ( ; ; ) { 3321da177e4SLinus Torvalds cond_resched(); 3330cd6144aSJohannes Weiner if (!pagevec_lookup_entries(&pvec, mapping, index, 334792ceaefSHugh Dickins min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) { 335792ceaefSHugh Dickins /* If all gone from start onwards, we're done */ 336b85e0effSHugh Dickins if (index == start) 3371da177e4SLinus Torvalds break; 338792ceaefSHugh Dickins /* Otherwise restart to make sure all gone */ 339b85e0effSHugh Dickins index = start; 3401da177e4SLinus Torvalds continue; 3411da177e4SLinus Torvalds } 3420cd6144aSJohannes Weiner if (index == start && indices[0] >= end) { 343792ceaefSHugh Dickins /* All gone out of hole to be punched, we're done */ 3440cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 345d7339071SHans Reiser pagevec_release(&pvec); 346d7339071SHans Reiser break; 347d7339071SHans Reiser } 3481da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 3491da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 3501da177e4SLinus Torvalds 351b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 3520cd6144aSJohannes Weiner index = indices[i]; 353792ceaefSHugh Dickins if (index >= end) { 354792ceaefSHugh Dickins /* Restart punch to make sure all gone */ 355792ceaefSHugh Dickins index = start - 1; 356d7339071SHans Reiser break; 357792ceaefSHugh Dickins } 358b85e0effSHugh Dickins 3590cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 3600cd6144aSJohannes Weiner clear_exceptional_entry(mapping, index, page); 3610cd6144aSJohannes Weiner continue; 3620cd6144aSJohannes Weiner } 3630cd6144aSJohannes Weiner 3641da177e4SLinus Torvalds lock_page(page); 365b85e0effSHugh Dickins WARN_ON(page->index != index); 3661da177e4SLinus Torvalds wait_on_page_writeback(page); 367750b4987SNick Piggin truncate_inode_page(mapping, page); 3681da177e4SLinus Torvalds unlock_page(page); 3691da177e4SLinus Torvalds } 3700cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 3711da177e4SLinus Torvalds pagevec_release(&pvec); 372b85e0effSHugh Dickins index++; 3731da177e4SLinus Torvalds } 3743167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 3751da177e4SLinus Torvalds } 376d7339071SHans Reiser EXPORT_SYMBOL(truncate_inode_pages_range); 3771da177e4SLinus Torvalds 378d7339071SHans Reiser /** 379d7339071SHans Reiser * truncate_inode_pages - truncate *all* the pages from an offset 380d7339071SHans Reiser * @mapping: mapping to truncate 381d7339071SHans Reiser * @lstart: offset from which to truncate 382d7339071SHans Reiser * 3831b1dcc1bSJes Sorensen * Called under (and serialised by) inode->i_mutex. 38408142579SJan Kara * 38508142579SJan Kara * Note: When this function returns, there can be a page in the process of 38608142579SJan Kara * deletion (inside __delete_from_page_cache()) in the specified range. Thus 38708142579SJan Kara * mapping->nrpages can be non-zero when this function returns even after 38808142579SJan Kara * truncation of the whole mapping. 389d7339071SHans Reiser */ 390d7339071SHans Reiser void truncate_inode_pages(struct address_space *mapping, loff_t lstart) 391d7339071SHans Reiser { 392d7339071SHans Reiser truncate_inode_pages_range(mapping, lstart, (loff_t)-1); 393d7339071SHans Reiser } 3941da177e4SLinus Torvalds EXPORT_SYMBOL(truncate_inode_pages); 3951da177e4SLinus Torvalds 39628697355SMike Waychison /** 39791b0abe3SJohannes Weiner * truncate_inode_pages_final - truncate *all* pages before inode dies 39891b0abe3SJohannes Weiner * @mapping: mapping to truncate 39991b0abe3SJohannes Weiner * 40091b0abe3SJohannes Weiner * Called under (and serialized by) inode->i_mutex. 40191b0abe3SJohannes Weiner * 40291b0abe3SJohannes Weiner * Filesystems have to use this in the .evict_inode path to inform the 40391b0abe3SJohannes Weiner * VM that this is the final truncate and the inode is going away. 40491b0abe3SJohannes Weiner */ 40591b0abe3SJohannes Weiner void truncate_inode_pages_final(struct address_space *mapping) 40691b0abe3SJohannes Weiner { 40791b0abe3SJohannes Weiner unsigned long nrshadows; 40891b0abe3SJohannes Weiner unsigned long nrpages; 40991b0abe3SJohannes Weiner 41091b0abe3SJohannes Weiner /* 41191b0abe3SJohannes Weiner * Page reclaim can not participate in regular inode lifetime 41291b0abe3SJohannes Weiner * management (can't call iput()) and thus can race with the 41391b0abe3SJohannes Weiner * inode teardown. Tell it when the address space is exiting, 41491b0abe3SJohannes Weiner * so that it does not install eviction information after the 41591b0abe3SJohannes Weiner * final truncate has begun. 41691b0abe3SJohannes Weiner */ 41791b0abe3SJohannes Weiner mapping_set_exiting(mapping); 41891b0abe3SJohannes Weiner 41991b0abe3SJohannes Weiner /* 42091b0abe3SJohannes Weiner * When reclaim installs eviction entries, it increases 42191b0abe3SJohannes Weiner * nrshadows first, then decreases nrpages. Make sure we see 42291b0abe3SJohannes Weiner * this in the right order or we might miss an entry. 42391b0abe3SJohannes Weiner */ 42491b0abe3SJohannes Weiner nrpages = mapping->nrpages; 42591b0abe3SJohannes Weiner smp_rmb(); 42691b0abe3SJohannes Weiner nrshadows = mapping->nrshadows; 42791b0abe3SJohannes Weiner 42891b0abe3SJohannes Weiner if (nrpages || nrshadows) { 42991b0abe3SJohannes Weiner /* 43091b0abe3SJohannes Weiner * As truncation uses a lockless tree lookup, cycle 43191b0abe3SJohannes Weiner * the tree lock to make sure any ongoing tree 43291b0abe3SJohannes Weiner * modification that does not see AS_EXITING is 43391b0abe3SJohannes Weiner * completed before starting the final truncate. 43491b0abe3SJohannes Weiner */ 43591b0abe3SJohannes Weiner spin_lock_irq(&mapping->tree_lock); 43691b0abe3SJohannes Weiner spin_unlock_irq(&mapping->tree_lock); 43791b0abe3SJohannes Weiner 43891b0abe3SJohannes Weiner truncate_inode_pages(mapping, 0); 43991b0abe3SJohannes Weiner } 44091b0abe3SJohannes Weiner } 44191b0abe3SJohannes Weiner EXPORT_SYMBOL(truncate_inode_pages_final); 44291b0abe3SJohannes Weiner 44391b0abe3SJohannes Weiner /** 44428697355SMike Waychison * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode 44528697355SMike Waychison * @mapping: the address_space which holds the pages to invalidate 44628697355SMike Waychison * @start: the offset 'from' which to invalidate 44728697355SMike Waychison * @end: the offset 'to' which to invalidate (inclusive) 44828697355SMike Waychison * 44928697355SMike Waychison * This function only removes the unlocked pages, if you want to 45028697355SMike Waychison * remove all the pages of one inode, you must call truncate_inode_pages. 45128697355SMike Waychison * 45228697355SMike Waychison * invalidate_mapping_pages() will not block on IO activity. It will not 45328697355SMike Waychison * invalidate pages which are dirty, locked, under writeback or mapped into 45428697355SMike Waychison * pagetables. 45528697355SMike Waychison */ 45628697355SMike Waychison unsigned long invalidate_mapping_pages(struct address_space *mapping, 45728697355SMike Waychison pgoff_t start, pgoff_t end) 4581da177e4SLinus Torvalds { 4590cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 4601da177e4SLinus Torvalds struct pagevec pvec; 461b85e0effSHugh Dickins pgoff_t index = start; 46231560180SMinchan Kim unsigned long ret; 46331560180SMinchan Kim unsigned long count = 0; 4641da177e4SLinus Torvalds int i; 4651da177e4SLinus Torvalds 4661da177e4SLinus Torvalds pagevec_init(&pvec, 0); 4670cd6144aSJohannes Weiner while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, 4680cd6144aSJohannes Weiner min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, 4690cd6144aSJohannes Weiner indices)) { 4701da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 4711da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 4721da177e4SLinus Torvalds 473b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 4740cd6144aSJohannes Weiner index = indices[i]; 475b85e0effSHugh Dickins if (index > end) 476b85e0effSHugh Dickins break; 477e0f23603SNeilBrown 4780cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 4790cd6144aSJohannes Weiner clear_exceptional_entry(mapping, index, page); 4800cd6144aSJohannes Weiner continue; 4810cd6144aSJohannes Weiner } 4820cd6144aSJohannes Weiner 483b85e0effSHugh Dickins if (!trylock_page(page)) 484b85e0effSHugh Dickins continue; 485b85e0effSHugh Dickins WARN_ON(page->index != index); 48631560180SMinchan Kim ret = invalidate_inode_page(page); 4871da177e4SLinus Torvalds unlock_page(page); 48831560180SMinchan Kim /* 48931560180SMinchan Kim * Invalidation is a hint that the page is no longer 49031560180SMinchan Kim * of interest and try to speed up its reclaim. 49131560180SMinchan Kim */ 49231560180SMinchan Kim if (!ret) 49331560180SMinchan Kim deactivate_page(page); 49431560180SMinchan Kim count += ret; 4951da177e4SLinus Torvalds } 4960cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 4971da177e4SLinus Torvalds pagevec_release(&pvec); 498fc9a07e7SAndrew Morton cond_resched(); 499b85e0effSHugh Dickins index++; 5001da177e4SLinus Torvalds } 50131560180SMinchan Kim return count; 5021da177e4SLinus Torvalds } 50354bc4855SAnton Altaparmakov EXPORT_SYMBOL(invalidate_mapping_pages); 5041da177e4SLinus Torvalds 505bd4c8ce4SAndrew Morton /* 506bd4c8ce4SAndrew Morton * This is like invalidate_complete_page(), except it ignores the page's 507bd4c8ce4SAndrew Morton * refcount. We do this because invalidate_inode_pages2() needs stronger 508bd4c8ce4SAndrew Morton * invalidation guarantees, and cannot afford to leave pages behind because 5092706a1b8SAnderson Briglia * shrink_page_list() has a temp ref on them, or because they're transiently 5102706a1b8SAnderson Briglia * sitting in the lru_cache_add() pagevecs. 511bd4c8ce4SAndrew Morton */ 512bd4c8ce4SAndrew Morton static int 513bd4c8ce4SAndrew Morton invalidate_complete_page2(struct address_space *mapping, struct page *page) 514bd4c8ce4SAndrew Morton { 515bd4c8ce4SAndrew Morton if (page->mapping != mapping) 516bd4c8ce4SAndrew Morton return 0; 517bd4c8ce4SAndrew Morton 518266cf658SDavid Howells if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) 519bd4c8ce4SAndrew Morton return 0; 520bd4c8ce4SAndrew Morton 52119fd6231SNick Piggin spin_lock_irq(&mapping->tree_lock); 522bd4c8ce4SAndrew Morton if (PageDirty(page)) 523bd4c8ce4SAndrew Morton goto failed; 524bd4c8ce4SAndrew Morton 525266cf658SDavid Howells BUG_ON(page_has_private(page)); 52691b0abe3SJohannes Weiner __delete_from_page_cache(page, NULL); 52719fd6231SNick Piggin spin_unlock_irq(&mapping->tree_lock); 5286072d13cSLinus Torvalds 5296072d13cSLinus Torvalds if (mapping->a_ops->freepage) 5306072d13cSLinus Torvalds mapping->a_ops->freepage(page); 5316072d13cSLinus Torvalds 532bd4c8ce4SAndrew Morton page_cache_release(page); /* pagecache ref */ 533bd4c8ce4SAndrew Morton return 1; 534bd4c8ce4SAndrew Morton failed: 53519fd6231SNick Piggin spin_unlock_irq(&mapping->tree_lock); 536bd4c8ce4SAndrew Morton return 0; 537bd4c8ce4SAndrew Morton } 538bd4c8ce4SAndrew Morton 539e3db7691STrond Myklebust static int do_launder_page(struct address_space *mapping, struct page *page) 540e3db7691STrond Myklebust { 541e3db7691STrond Myklebust if (!PageDirty(page)) 542e3db7691STrond Myklebust return 0; 543e3db7691STrond Myklebust if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) 544e3db7691STrond Myklebust return 0; 545e3db7691STrond Myklebust return mapping->a_ops->launder_page(page); 546e3db7691STrond Myklebust } 547e3db7691STrond Myklebust 5481da177e4SLinus Torvalds /** 5491da177e4SLinus Torvalds * invalidate_inode_pages2_range - remove range of pages from an address_space 55067be2dd1SMartin Waitz * @mapping: the address_space 5511da177e4SLinus Torvalds * @start: the page offset 'from' which to invalidate 5521da177e4SLinus Torvalds * @end: the page offset 'to' which to invalidate (inclusive) 5531da177e4SLinus Torvalds * 5541da177e4SLinus Torvalds * Any pages which are found to be mapped into pagetables are unmapped prior to 5551da177e4SLinus Torvalds * invalidation. 5561da177e4SLinus Torvalds * 5576ccfa806SHisashi Hifumi * Returns -EBUSY if any pages could not be invalidated. 5581da177e4SLinus Torvalds */ 5591da177e4SLinus Torvalds int invalidate_inode_pages2_range(struct address_space *mapping, 5601da177e4SLinus Torvalds pgoff_t start, pgoff_t end) 5611da177e4SLinus Torvalds { 5620cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 5631da177e4SLinus Torvalds struct pagevec pvec; 564b85e0effSHugh Dickins pgoff_t index; 5651da177e4SLinus Torvalds int i; 5661da177e4SLinus Torvalds int ret = 0; 5670dd1334fSHisashi Hifumi int ret2 = 0; 5681da177e4SLinus Torvalds int did_range_unmap = 0; 5691da177e4SLinus Torvalds 5703167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 5711da177e4SLinus Torvalds pagevec_init(&pvec, 0); 572b85e0effSHugh Dickins index = start; 5730cd6144aSJohannes Weiner while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, 5740cd6144aSJohannes Weiner min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, 5750cd6144aSJohannes Weiner indices)) { 5767b965e08STrond Myklebust for (i = 0; i < pagevec_count(&pvec); i++) { 5771da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 578b85e0effSHugh Dickins 579b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 5800cd6144aSJohannes Weiner index = indices[i]; 581b85e0effSHugh Dickins if (index > end) 582b85e0effSHugh Dickins break; 5831da177e4SLinus Torvalds 5840cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 5850cd6144aSJohannes Weiner clear_exceptional_entry(mapping, index, page); 5860cd6144aSJohannes Weiner continue; 5870cd6144aSJohannes Weiner } 5880cd6144aSJohannes Weiner 5891da177e4SLinus Torvalds lock_page(page); 590b85e0effSHugh Dickins WARN_ON(page->index != index); 5911da177e4SLinus Torvalds if (page->mapping != mapping) { 5921da177e4SLinus Torvalds unlock_page(page); 5931da177e4SLinus Torvalds continue; 5941da177e4SLinus Torvalds } 5951da177e4SLinus Torvalds wait_on_page_writeback(page); 596d00806b1SNick Piggin if (page_mapped(page)) { 5971da177e4SLinus Torvalds if (!did_range_unmap) { 5981da177e4SLinus Torvalds /* 5991da177e4SLinus Torvalds * Zap the rest of the file in one hit. 6001da177e4SLinus Torvalds */ 6011da177e4SLinus Torvalds unmap_mapping_range(mapping, 602b85e0effSHugh Dickins (loff_t)index << PAGE_CACHE_SHIFT, 603b85e0effSHugh Dickins (loff_t)(1 + end - index) 6041da177e4SLinus Torvalds << PAGE_CACHE_SHIFT, 6051da177e4SLinus Torvalds 0); 6061da177e4SLinus Torvalds did_range_unmap = 1; 6071da177e4SLinus Torvalds } else { 6081da177e4SLinus Torvalds /* 6091da177e4SLinus Torvalds * Just zap this page 6101da177e4SLinus Torvalds */ 6111da177e4SLinus Torvalds unmap_mapping_range(mapping, 612b85e0effSHugh Dickins (loff_t)index << PAGE_CACHE_SHIFT, 6131da177e4SLinus Torvalds PAGE_CACHE_SIZE, 0); 6141da177e4SLinus Torvalds } 6151da177e4SLinus Torvalds } 616d00806b1SNick Piggin BUG_ON(page_mapped(page)); 6170dd1334fSHisashi Hifumi ret2 = do_launder_page(mapping, page); 6180dd1334fSHisashi Hifumi if (ret2 == 0) { 6190dd1334fSHisashi Hifumi if (!invalidate_complete_page2(mapping, page)) 6206ccfa806SHisashi Hifumi ret2 = -EBUSY; 6210dd1334fSHisashi Hifumi } 6220dd1334fSHisashi Hifumi if (ret2 < 0) 6230dd1334fSHisashi Hifumi ret = ret2; 6241da177e4SLinus Torvalds unlock_page(page); 6251da177e4SLinus Torvalds } 6260cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 6271da177e4SLinus Torvalds pagevec_release(&pvec); 6281da177e4SLinus Torvalds cond_resched(); 629b85e0effSHugh Dickins index++; 6301da177e4SLinus Torvalds } 6313167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 6321da177e4SLinus Torvalds return ret; 6331da177e4SLinus Torvalds } 6341da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); 6351da177e4SLinus Torvalds 6361da177e4SLinus Torvalds /** 6371da177e4SLinus Torvalds * invalidate_inode_pages2 - remove all pages from an address_space 63867be2dd1SMartin Waitz * @mapping: the address_space 6391da177e4SLinus Torvalds * 6401da177e4SLinus Torvalds * Any pages which are found to be mapped into pagetables are unmapped prior to 6411da177e4SLinus Torvalds * invalidation. 6421da177e4SLinus Torvalds * 643e9de25ddSPeng Tao * Returns -EBUSY if any pages could not be invalidated. 6441da177e4SLinus Torvalds */ 6451da177e4SLinus Torvalds int invalidate_inode_pages2(struct address_space *mapping) 6461da177e4SLinus Torvalds { 6471da177e4SLinus Torvalds return invalidate_inode_pages2_range(mapping, 0, -1); 6481da177e4SLinus Torvalds } 6491da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(invalidate_inode_pages2); 65025d9e2d1Snpiggin@suse.de 65125d9e2d1Snpiggin@suse.de /** 65225d9e2d1Snpiggin@suse.de * truncate_pagecache - unmap and remove pagecache that has been truncated 65325d9e2d1Snpiggin@suse.de * @inode: inode 6548a549beaSHugh Dickins * @newsize: new file size 65525d9e2d1Snpiggin@suse.de * 65625d9e2d1Snpiggin@suse.de * inode's new i_size must already be written before truncate_pagecache 65725d9e2d1Snpiggin@suse.de * is called. 65825d9e2d1Snpiggin@suse.de * 65925d9e2d1Snpiggin@suse.de * This function should typically be called before the filesystem 66025d9e2d1Snpiggin@suse.de * releases resources associated with the freed range (eg. deallocates 66125d9e2d1Snpiggin@suse.de * blocks). This way, pagecache will always stay logically coherent 66225d9e2d1Snpiggin@suse.de * with on-disk format, and the filesystem would not have to deal with 66325d9e2d1Snpiggin@suse.de * situations such as writepage being called for a page that has already 66425d9e2d1Snpiggin@suse.de * had its underlying blocks deallocated. 66525d9e2d1Snpiggin@suse.de */ 6667caef267SKirill A. Shutemov void truncate_pagecache(struct inode *inode, loff_t newsize) 66725d9e2d1Snpiggin@suse.de { 66825d9e2d1Snpiggin@suse.de struct address_space *mapping = inode->i_mapping; 6698a549beaSHugh Dickins loff_t holebegin = round_up(newsize, PAGE_SIZE); 67025d9e2d1Snpiggin@suse.de 67125d9e2d1Snpiggin@suse.de /* 67225d9e2d1Snpiggin@suse.de * unmap_mapping_range is called twice, first simply for 67325d9e2d1Snpiggin@suse.de * efficiency so that truncate_inode_pages does fewer 67425d9e2d1Snpiggin@suse.de * single-page unmaps. However after this first call, and 67525d9e2d1Snpiggin@suse.de * before truncate_inode_pages finishes, it is possible for 67625d9e2d1Snpiggin@suse.de * private pages to be COWed, which remain after 67725d9e2d1Snpiggin@suse.de * truncate_inode_pages finishes, hence the second 67825d9e2d1Snpiggin@suse.de * unmap_mapping_range call must be made for correctness. 67925d9e2d1Snpiggin@suse.de */ 6808a549beaSHugh Dickins unmap_mapping_range(mapping, holebegin, 0, 1); 6818a549beaSHugh Dickins truncate_inode_pages(mapping, newsize); 6828a549beaSHugh Dickins unmap_mapping_range(mapping, holebegin, 0, 1); 68325d9e2d1Snpiggin@suse.de } 68425d9e2d1Snpiggin@suse.de EXPORT_SYMBOL(truncate_pagecache); 68525d9e2d1Snpiggin@suse.de 68625d9e2d1Snpiggin@suse.de /** 6872c27c65eSChristoph Hellwig * truncate_setsize - update inode and pagecache for a new file size 6882c27c65eSChristoph Hellwig * @inode: inode 6892c27c65eSChristoph Hellwig * @newsize: new file size 6902c27c65eSChristoph Hellwig * 691382e27daSJan Kara * truncate_setsize updates i_size and performs pagecache truncation (if 692382e27daSJan Kara * necessary) to @newsize. It will be typically be called from the filesystem's 693382e27daSJan Kara * setattr function when ATTR_SIZE is passed in. 6942c27c65eSChristoph Hellwig * 69577783d06SJan Kara * Must be called with a lock serializing truncates and writes (generally 69677783d06SJan Kara * i_mutex but e.g. xfs uses a different lock) and before all filesystem 69777783d06SJan Kara * specific block truncation has been performed. 6982c27c65eSChristoph Hellwig */ 6992c27c65eSChristoph Hellwig void truncate_setsize(struct inode *inode, loff_t newsize) 7002c27c65eSChristoph Hellwig { 70190a80202SJan Kara loff_t oldsize = inode->i_size; 70290a80202SJan Kara 7032c27c65eSChristoph Hellwig i_size_write(inode, newsize); 70490a80202SJan Kara if (newsize > oldsize) 70590a80202SJan Kara pagecache_isize_extended(inode, oldsize, newsize); 7067caef267SKirill A. Shutemov truncate_pagecache(inode, newsize); 7072c27c65eSChristoph Hellwig } 7082c27c65eSChristoph Hellwig EXPORT_SYMBOL(truncate_setsize); 7092c27c65eSChristoph Hellwig 7102c27c65eSChristoph Hellwig /** 71190a80202SJan Kara * pagecache_isize_extended - update pagecache after extension of i_size 71290a80202SJan Kara * @inode: inode for which i_size was extended 71390a80202SJan Kara * @from: original inode size 71490a80202SJan Kara * @to: new inode size 71590a80202SJan Kara * 71690a80202SJan Kara * Handle extension of inode size either caused by extending truncate or by 71790a80202SJan Kara * write starting after current i_size. We mark the page straddling current 71890a80202SJan Kara * i_size RO so that page_mkwrite() is called on the nearest write access to 71990a80202SJan Kara * the page. This way filesystem can be sure that page_mkwrite() is called on 72090a80202SJan Kara * the page before user writes to the page via mmap after the i_size has been 72190a80202SJan Kara * changed. 72290a80202SJan Kara * 72390a80202SJan Kara * The function must be called after i_size is updated so that page fault 72490a80202SJan Kara * coming after we unlock the page will already see the new i_size. 72590a80202SJan Kara * The function must be called while we still hold i_mutex - this not only 72690a80202SJan Kara * makes sure i_size is stable but also that userspace cannot observe new 72790a80202SJan Kara * i_size value before we are prepared to store mmap writes at new inode size. 72890a80202SJan Kara */ 72990a80202SJan Kara void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) 73090a80202SJan Kara { 73190a80202SJan Kara int bsize = 1 << inode->i_blkbits; 73290a80202SJan Kara loff_t rounded_from; 73390a80202SJan Kara struct page *page; 73490a80202SJan Kara pgoff_t index; 73590a80202SJan Kara 73690a80202SJan Kara WARN_ON(to > inode->i_size); 73790a80202SJan Kara 73890a80202SJan Kara if (from >= to || bsize == PAGE_CACHE_SIZE) 73990a80202SJan Kara return; 74090a80202SJan Kara /* Page straddling @from will not have any hole block created? */ 74190a80202SJan Kara rounded_from = round_up(from, bsize); 74290a80202SJan Kara if (to <= rounded_from || !(rounded_from & (PAGE_CACHE_SIZE - 1))) 74390a80202SJan Kara return; 74490a80202SJan Kara 74590a80202SJan Kara index = from >> PAGE_CACHE_SHIFT; 74690a80202SJan Kara page = find_lock_page(inode->i_mapping, index); 74790a80202SJan Kara /* Page not cached? Nothing to do */ 74890a80202SJan Kara if (!page) 74990a80202SJan Kara return; 75090a80202SJan Kara /* 75190a80202SJan Kara * See clear_page_dirty_for_io() for details why set_page_dirty() 75290a80202SJan Kara * is needed. 75390a80202SJan Kara */ 75490a80202SJan Kara if (page_mkclean(page)) 75590a80202SJan Kara set_page_dirty(page); 75690a80202SJan Kara unlock_page(page); 75790a80202SJan Kara page_cache_release(page); 75890a80202SJan Kara } 75990a80202SJan Kara EXPORT_SYMBOL(pagecache_isize_extended); 76090a80202SJan Kara 76190a80202SJan Kara /** 762623e3db9SHugh Dickins * truncate_pagecache_range - unmap and remove pagecache that is hole-punched 763623e3db9SHugh Dickins * @inode: inode 764623e3db9SHugh Dickins * @lstart: offset of beginning of hole 765623e3db9SHugh Dickins * @lend: offset of last byte of hole 766623e3db9SHugh Dickins * 767623e3db9SHugh Dickins * This function should typically be called before the filesystem 768623e3db9SHugh Dickins * releases resources associated with the freed range (eg. deallocates 769623e3db9SHugh Dickins * blocks). This way, pagecache will always stay logically coherent 770623e3db9SHugh Dickins * with on-disk format, and the filesystem would not have to deal with 771623e3db9SHugh Dickins * situations such as writepage being called for a page that has already 772623e3db9SHugh Dickins * had its underlying blocks deallocated. 773623e3db9SHugh Dickins */ 774623e3db9SHugh Dickins void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) 775623e3db9SHugh Dickins { 776623e3db9SHugh Dickins struct address_space *mapping = inode->i_mapping; 777623e3db9SHugh Dickins loff_t unmap_start = round_up(lstart, PAGE_SIZE); 778623e3db9SHugh Dickins loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; 779623e3db9SHugh Dickins /* 780623e3db9SHugh Dickins * This rounding is currently just for example: unmap_mapping_range 781623e3db9SHugh Dickins * expands its hole outwards, whereas we want it to contract the hole 782623e3db9SHugh Dickins * inwards. However, existing callers of truncate_pagecache_range are 7835a720394SLukas Czerner * doing their own page rounding first. Note that unmap_mapping_range 7845a720394SLukas Czerner * allows holelen 0 for all, and we allow lend -1 for end of file. 785623e3db9SHugh Dickins */ 786623e3db9SHugh Dickins 787623e3db9SHugh Dickins /* 788623e3db9SHugh Dickins * Unlike in truncate_pagecache, unmap_mapping_range is called only 789623e3db9SHugh Dickins * once (before truncating pagecache), and without "even_cows" flag: 790623e3db9SHugh Dickins * hole-punching should not remove private COWed pages from the hole. 791623e3db9SHugh Dickins */ 792623e3db9SHugh Dickins if ((u64)unmap_end > (u64)unmap_start) 793623e3db9SHugh Dickins unmap_mapping_range(mapping, unmap_start, 794623e3db9SHugh Dickins 1 + unmap_end - unmap_start, 0); 795623e3db9SHugh Dickins truncate_inode_pages_range(mapping, lstart, lend); 796623e3db9SHugh Dickins } 797623e3db9SHugh Dickins EXPORT_SYMBOL(truncate_pagecache_range); 798