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> 12f9fe48beSRoss 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 37f9fe48beSRoss Zwisler if (dax_mapping(mapping)) { 38ac401cc7SJan Kara dax_delete_mapping_entry(mapping, index); 39ac401cc7SJan Kara return; 40ac401cc7SJan Kara } 41ac401cc7SJan Kara spin_lock_irq(&mapping->tree_lock); 420cd6144aSJohannes Weiner /* 430cd6144aSJohannes Weiner * Regular page slots are stabilized by the page lock even 440cd6144aSJohannes Weiner * without the tree itself locked. These unlocked entries 450cd6144aSJohannes Weiner * need verification under the tree lock. 460cd6144aSJohannes Weiner */ 47*14b46879SJohannes Weiner if (!__radix_tree_lookup(&mapping->page_tree, index, &node, &slot)) 48449dd698SJohannes Weiner goto unlock; 49449dd698SJohannes Weiner if (*slot != entry) 50449dd698SJohannes Weiner goto unlock; 51*14b46879SJohannes Weiner __radix_tree_replace(&mapping->page_tree, node, slot, NULL, 52*14b46879SJohannes Weiner workingset_update_node, mapping); 53f9fe48beSRoss Zwisler mapping->nrexceptional--; 54449dd698SJohannes Weiner unlock: 550cd6144aSJohannes Weiner spin_unlock_irq(&mapping->tree_lock); 560cd6144aSJohannes Weiner } 571da177e4SLinus Torvalds 58cf9a2ae8SDavid Howells /** 5928bc44d7SFengguang Wu * do_invalidatepage - invalidate part or all of a page 60cf9a2ae8SDavid Howells * @page: the page which is affected 61d47992f8SLukas Czerner * @offset: start of the range to invalidate 62d47992f8SLukas Czerner * @length: length of the range to invalidate 63cf9a2ae8SDavid Howells * 64cf9a2ae8SDavid Howells * do_invalidatepage() is called when all or part of the page has become 65cf9a2ae8SDavid Howells * invalidated by a truncate operation. 66cf9a2ae8SDavid Howells * 67cf9a2ae8SDavid Howells * do_invalidatepage() does not have to release all buffers, but it must 68cf9a2ae8SDavid Howells * ensure that no dirty buffer is left outside @offset and that no I/O 69cf9a2ae8SDavid Howells * is underway against any of the blocks which are outside the truncation 70cf9a2ae8SDavid Howells * point. Because the caller is about to free (and possibly reuse) those 71cf9a2ae8SDavid Howells * blocks on-disk. 72cf9a2ae8SDavid Howells */ 73d47992f8SLukas Czerner void do_invalidatepage(struct page *page, unsigned int offset, 74d47992f8SLukas Czerner unsigned int length) 75cf9a2ae8SDavid Howells { 76d47992f8SLukas Czerner void (*invalidatepage)(struct page *, unsigned int, unsigned int); 77d47992f8SLukas Czerner 78cf9a2ae8SDavid Howells invalidatepage = page->mapping->a_ops->invalidatepage; 799361401eSDavid Howells #ifdef CONFIG_BLOCK 80cf9a2ae8SDavid Howells if (!invalidatepage) 81cf9a2ae8SDavid Howells invalidatepage = block_invalidatepage; 829361401eSDavid Howells #endif 83cf9a2ae8SDavid Howells if (invalidatepage) 84d47992f8SLukas Czerner (*invalidatepage)(page, offset, length); 85cf9a2ae8SDavid Howells } 86cf9a2ae8SDavid Howells 87ecdfc978SLinus Torvalds /* 881da177e4SLinus Torvalds * If truncate cannot remove the fs-private metadata from the page, the page 8962e1c553SShaohua Li * becomes orphaned. It will be left on the LRU and may even be mapped into 9054cb8821SNick Piggin * user pagetables if we're racing with filemap_fault(). 911da177e4SLinus Torvalds * 921da177e4SLinus Torvalds * We need to bale out if page->mapping is no longer equal to the original 931da177e4SLinus Torvalds * mapping. This happens a) when the VM reclaimed the page while we waited on 94fc0ecff6SAndrew Morton * its lock, b) when a concurrent invalidate_mapping_pages got there first and 951da177e4SLinus Torvalds * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. 961da177e4SLinus Torvalds */ 97750b4987SNick Piggin static int 981da177e4SLinus Torvalds truncate_complete_page(struct address_space *mapping, struct page *page) 991da177e4SLinus Torvalds { 1001da177e4SLinus Torvalds if (page->mapping != mapping) 101750b4987SNick Piggin return -EIO; 1021da177e4SLinus Torvalds 103266cf658SDavid Howells if (page_has_private(page)) 10409cbfeafSKirill A. Shutemov do_invalidatepage(page, 0, PAGE_SIZE); 1051da177e4SLinus Torvalds 106b9ea2515SKonstantin Khlebnikov /* 107b9ea2515SKonstantin Khlebnikov * Some filesystems seem to re-dirty the page even after 108b9ea2515SKonstantin Khlebnikov * the VM has canceled the dirty bit (eg ext3 journaling). 109b9ea2515SKonstantin Khlebnikov * Hence dirty accounting check is placed after invalidation. 110b9ea2515SKonstantin Khlebnikov */ 11111f81becSTejun Heo cancel_dirty_page(page); 1121da177e4SLinus Torvalds ClearPageMappedToDisk(page); 1135adc7b51SMinchan Kim delete_from_page_cache(page); 114750b4987SNick Piggin return 0; 1151da177e4SLinus Torvalds } 1161da177e4SLinus Torvalds 1171da177e4SLinus Torvalds /* 118fc0ecff6SAndrew Morton * This is for invalidate_mapping_pages(). That function can be called at 1191da177e4SLinus Torvalds * any time, and is not supposed to throw away dirty pages. But pages can 1200fd0e6b0SNick Piggin * be marked dirty at any time too, so use remove_mapping which safely 1210fd0e6b0SNick Piggin * discards clean, unused pages. 1221da177e4SLinus Torvalds * 1231da177e4SLinus Torvalds * Returns non-zero if the page was successfully invalidated. 1241da177e4SLinus Torvalds */ 1251da177e4SLinus Torvalds static int 1261da177e4SLinus Torvalds invalidate_complete_page(struct address_space *mapping, struct page *page) 1271da177e4SLinus Torvalds { 1280fd0e6b0SNick Piggin int ret; 1290fd0e6b0SNick Piggin 1301da177e4SLinus Torvalds if (page->mapping != mapping) 1311da177e4SLinus Torvalds return 0; 1321da177e4SLinus Torvalds 133266cf658SDavid Howells if (page_has_private(page) && !try_to_release_page(page, 0)) 1341da177e4SLinus Torvalds return 0; 1351da177e4SLinus Torvalds 1360fd0e6b0SNick Piggin ret = remove_mapping(mapping, page); 1370fd0e6b0SNick Piggin 1380fd0e6b0SNick Piggin return ret; 1391da177e4SLinus Torvalds } 1401da177e4SLinus Torvalds 141750b4987SNick Piggin int truncate_inode_page(struct address_space *mapping, struct page *page) 142750b4987SNick Piggin { 143fc127da0SKirill A. Shutemov loff_t holelen; 144fc127da0SKirill A. Shutemov VM_BUG_ON_PAGE(PageTail(page), page); 145fc127da0SKirill A. Shutemov 146fc127da0SKirill A. Shutemov holelen = PageTransHuge(page) ? HPAGE_PMD_SIZE : PAGE_SIZE; 147750b4987SNick Piggin if (page_mapped(page)) { 148750b4987SNick Piggin unmap_mapping_range(mapping, 14909cbfeafSKirill A. Shutemov (loff_t)page->index << PAGE_SHIFT, 150fc127da0SKirill A. Shutemov holelen, 0); 151750b4987SNick Piggin } 152750b4987SNick Piggin return truncate_complete_page(mapping, page); 153750b4987SNick Piggin } 154750b4987SNick Piggin 15583f78668SWu Fengguang /* 15625718736SAndi Kleen * Used to get rid of pages on hardware memory corruption. 15725718736SAndi Kleen */ 15825718736SAndi Kleen int generic_error_remove_page(struct address_space *mapping, struct page *page) 15925718736SAndi Kleen { 16025718736SAndi Kleen if (!mapping) 16125718736SAndi Kleen return -EINVAL; 16225718736SAndi Kleen /* 16325718736SAndi Kleen * Only punch for normal data pages for now. 16425718736SAndi Kleen * Handling other types like directories would need more auditing. 16525718736SAndi Kleen */ 16625718736SAndi Kleen if (!S_ISREG(mapping->host->i_mode)) 16725718736SAndi Kleen return -EIO; 16825718736SAndi Kleen return truncate_inode_page(mapping, page); 16925718736SAndi Kleen } 17025718736SAndi Kleen EXPORT_SYMBOL(generic_error_remove_page); 17125718736SAndi Kleen 17225718736SAndi Kleen /* 17383f78668SWu Fengguang * Safely invalidate one page from its pagecache mapping. 17483f78668SWu Fengguang * It only drops clean, unused pages. The page must be locked. 17583f78668SWu Fengguang * 17683f78668SWu Fengguang * Returns 1 if the page is successfully invalidated, otherwise 0. 17783f78668SWu Fengguang */ 17883f78668SWu Fengguang int invalidate_inode_page(struct page *page) 17983f78668SWu Fengguang { 18083f78668SWu Fengguang struct address_space *mapping = page_mapping(page); 18183f78668SWu Fengguang if (!mapping) 18283f78668SWu Fengguang return 0; 18383f78668SWu Fengguang if (PageDirty(page) || PageWriteback(page)) 18483f78668SWu Fengguang return 0; 18583f78668SWu Fengguang if (page_mapped(page)) 18683f78668SWu Fengguang return 0; 18783f78668SWu Fengguang return invalidate_complete_page(mapping, page); 18883f78668SWu Fengguang } 18983f78668SWu Fengguang 1901da177e4SLinus Torvalds /** 19173c1e204SLiu Bo * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets 1921da177e4SLinus Torvalds * @mapping: mapping to truncate 1931da177e4SLinus Torvalds * @lstart: offset from which to truncate 1945a720394SLukas Czerner * @lend: offset to which to truncate (inclusive) 1951da177e4SLinus Torvalds * 196d7339071SHans Reiser * Truncate the page cache, removing the pages that are between 1975a720394SLukas Czerner * specified offsets (and zeroing out partial pages 1985a720394SLukas Czerner * if lstart or lend + 1 is not page aligned). 1991da177e4SLinus Torvalds * 2001da177e4SLinus Torvalds * Truncate takes two passes - the first pass is nonblocking. It will not 2011da177e4SLinus Torvalds * block on page locks and it will not block on writeback. The second pass 2021da177e4SLinus Torvalds * will wait. This is to prevent as much IO as possible in the affected region. 2031da177e4SLinus Torvalds * The first pass will remove most pages, so the search cost of the second pass 2041da177e4SLinus Torvalds * is low. 2051da177e4SLinus Torvalds * 2061da177e4SLinus Torvalds * We pass down the cache-hot hint to the page freeing code. Even if the 2071da177e4SLinus Torvalds * mapping is large, it is probably the case that the final pages are the most 2081da177e4SLinus Torvalds * recently touched, and freeing happens in ascending file offset order. 2095a720394SLukas Czerner * 2105a720394SLukas Czerner * Note that since ->invalidatepage() accepts range to invalidate 2115a720394SLukas Czerner * truncate_inode_pages_range is able to handle cases where lend + 1 is not 2125a720394SLukas Czerner * page aligned properly. 2131da177e4SLinus Torvalds */ 214d7339071SHans Reiser void truncate_inode_pages_range(struct address_space *mapping, 215d7339071SHans Reiser loff_t lstart, loff_t lend) 2161da177e4SLinus Torvalds { 2175a720394SLukas Czerner pgoff_t start; /* inclusive */ 2185a720394SLukas Czerner pgoff_t end; /* exclusive */ 2195a720394SLukas Czerner unsigned int partial_start; /* inclusive */ 2205a720394SLukas Czerner unsigned int partial_end; /* exclusive */ 2211da177e4SLinus Torvalds struct pagevec pvec; 2220cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 223b85e0effSHugh Dickins pgoff_t index; 2241da177e4SLinus Torvalds int i; 2251da177e4SLinus Torvalds 2263167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 227f9fe48beSRoss Zwisler if (mapping->nrpages == 0 && mapping->nrexceptional == 0) 2281da177e4SLinus Torvalds return; 2291da177e4SLinus Torvalds 2305a720394SLukas Czerner /* Offsets within partial pages */ 23109cbfeafSKirill A. Shutemov partial_start = lstart & (PAGE_SIZE - 1); 23209cbfeafSKirill A. Shutemov partial_end = (lend + 1) & (PAGE_SIZE - 1); 2335a720394SLukas Czerner 2345a720394SLukas Czerner /* 2355a720394SLukas Czerner * 'start' and 'end' always covers the range of pages to be fully 2365a720394SLukas Czerner * truncated. Partial pages are covered with 'partial_start' at the 2375a720394SLukas Czerner * start of the range and 'partial_end' at the end of the range. 2385a720394SLukas Czerner * Note that 'end' is exclusive while 'lend' is inclusive. 2395a720394SLukas Czerner */ 24009cbfeafSKirill A. Shutemov start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; 2415a720394SLukas Czerner if (lend == -1) 2425a720394SLukas Czerner /* 2435a720394SLukas Czerner * lend == -1 indicates end-of-file so we have to set 'end' 2445a720394SLukas Czerner * to the highest possible pgoff_t and since the type is 2455a720394SLukas Czerner * unsigned we're using -1. 2465a720394SLukas Czerner */ 2475a720394SLukas Czerner end = -1; 2485a720394SLukas Czerner else 24909cbfeafSKirill A. Shutemov end = (lend + 1) >> PAGE_SHIFT; 250d7339071SHans Reiser 2511da177e4SLinus Torvalds pagevec_init(&pvec, 0); 252b85e0effSHugh Dickins index = start; 2530cd6144aSJohannes Weiner while (index < end && pagevec_lookup_entries(&pvec, mapping, index, 2540cd6144aSJohannes Weiner min(end - index, (pgoff_t)PAGEVEC_SIZE), 2550cd6144aSJohannes Weiner indices)) { 2561da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 2571da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 2581da177e4SLinus Torvalds 259b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 2600cd6144aSJohannes Weiner index = indices[i]; 2615a720394SLukas Czerner if (index >= end) 262d7339071SHans Reiser break; 263d7339071SHans Reiser 2640cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 2650cd6144aSJohannes Weiner clear_exceptional_entry(mapping, index, page); 2660cd6144aSJohannes Weiner continue; 2670cd6144aSJohannes Weiner } 2680cd6144aSJohannes Weiner 269529ae9aaSNick Piggin if (!trylock_page(page)) 2701da177e4SLinus Torvalds continue; 2715cbc198aSKirill A. Shutemov WARN_ON(page_to_index(page) != index); 2721da177e4SLinus Torvalds if (PageWriteback(page)) { 2731da177e4SLinus Torvalds unlock_page(page); 2741da177e4SLinus Torvalds continue; 2751da177e4SLinus Torvalds } 276750b4987SNick Piggin truncate_inode_page(mapping, page); 2771da177e4SLinus Torvalds unlock_page(page); 2781da177e4SLinus Torvalds } 2790cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 2801da177e4SLinus Torvalds pagevec_release(&pvec); 2811da177e4SLinus Torvalds cond_resched(); 282b85e0effSHugh Dickins index++; 2831da177e4SLinus Torvalds } 2841da177e4SLinus Torvalds 2855a720394SLukas Czerner if (partial_start) { 2861da177e4SLinus Torvalds struct page *page = find_lock_page(mapping, start - 1); 2871da177e4SLinus Torvalds if (page) { 28809cbfeafSKirill A. Shutemov unsigned int top = PAGE_SIZE; 2895a720394SLukas Czerner if (start > end) { 2905a720394SLukas Czerner /* Truncation within a single page */ 2915a720394SLukas Czerner top = partial_end; 2925a720394SLukas Czerner partial_end = 0; 2935a720394SLukas Czerner } 2941da177e4SLinus Torvalds wait_on_page_writeback(page); 2955a720394SLukas Czerner zero_user_segment(page, partial_start, top); 2965a720394SLukas Czerner cleancache_invalidate_page(mapping, page); 2975a720394SLukas Czerner if (page_has_private(page)) 2985a720394SLukas Czerner do_invalidatepage(page, partial_start, 2995a720394SLukas Czerner top - partial_start); 3001da177e4SLinus Torvalds unlock_page(page); 30109cbfeafSKirill A. Shutemov put_page(page); 3021da177e4SLinus Torvalds } 3031da177e4SLinus Torvalds } 3045a720394SLukas Czerner if (partial_end) { 3055a720394SLukas Czerner struct page *page = find_lock_page(mapping, end); 3065a720394SLukas Czerner if (page) { 3075a720394SLukas Czerner wait_on_page_writeback(page); 3085a720394SLukas Czerner zero_user_segment(page, 0, partial_end); 3095a720394SLukas Czerner cleancache_invalidate_page(mapping, page); 3105a720394SLukas Czerner if (page_has_private(page)) 3115a720394SLukas Czerner do_invalidatepage(page, 0, 3125a720394SLukas Czerner partial_end); 3135a720394SLukas Czerner unlock_page(page); 31409cbfeafSKirill A. Shutemov put_page(page); 3155a720394SLukas Czerner } 3165a720394SLukas Czerner } 3175a720394SLukas Czerner /* 3185a720394SLukas Czerner * If the truncation happened within a single page no pages 3195a720394SLukas Czerner * will be released, just zeroed, so we can bail out now. 3205a720394SLukas Czerner */ 3215a720394SLukas Czerner if (start >= end) 3225a720394SLukas Czerner return; 3231da177e4SLinus Torvalds 324b85e0effSHugh Dickins index = start; 3251da177e4SLinus Torvalds for ( ; ; ) { 3261da177e4SLinus Torvalds cond_resched(); 3270cd6144aSJohannes Weiner if (!pagevec_lookup_entries(&pvec, mapping, index, 328792ceaefSHugh Dickins min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) { 329792ceaefSHugh Dickins /* If all gone from start onwards, we're done */ 330b85e0effSHugh Dickins if (index == start) 3311da177e4SLinus Torvalds break; 332792ceaefSHugh Dickins /* Otherwise restart to make sure all gone */ 333b85e0effSHugh Dickins index = start; 3341da177e4SLinus Torvalds continue; 3351da177e4SLinus Torvalds } 3360cd6144aSJohannes Weiner if (index == start && indices[0] >= end) { 337792ceaefSHugh Dickins /* All gone out of hole to be punched, we're done */ 3380cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 339d7339071SHans Reiser pagevec_release(&pvec); 340d7339071SHans Reiser break; 341d7339071SHans Reiser } 3421da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 3431da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 3441da177e4SLinus Torvalds 345b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 3460cd6144aSJohannes Weiner index = indices[i]; 347792ceaefSHugh Dickins if (index >= end) { 348792ceaefSHugh Dickins /* Restart punch to make sure all gone */ 349792ceaefSHugh Dickins index = start - 1; 350d7339071SHans Reiser break; 351792ceaefSHugh Dickins } 352b85e0effSHugh Dickins 3530cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 3540cd6144aSJohannes Weiner clear_exceptional_entry(mapping, index, page); 3550cd6144aSJohannes Weiner continue; 3560cd6144aSJohannes Weiner } 3570cd6144aSJohannes Weiner 3581da177e4SLinus Torvalds lock_page(page); 3595cbc198aSKirill A. Shutemov WARN_ON(page_to_index(page) != index); 3601da177e4SLinus Torvalds wait_on_page_writeback(page); 361750b4987SNick Piggin truncate_inode_page(mapping, page); 3621da177e4SLinus Torvalds unlock_page(page); 3631da177e4SLinus Torvalds } 3640cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 3651da177e4SLinus Torvalds pagevec_release(&pvec); 366b85e0effSHugh Dickins index++; 3671da177e4SLinus Torvalds } 3683167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 3691da177e4SLinus Torvalds } 370d7339071SHans Reiser EXPORT_SYMBOL(truncate_inode_pages_range); 3711da177e4SLinus Torvalds 372d7339071SHans Reiser /** 373d7339071SHans Reiser * truncate_inode_pages - truncate *all* the pages from an offset 374d7339071SHans Reiser * @mapping: mapping to truncate 375d7339071SHans Reiser * @lstart: offset from which to truncate 376d7339071SHans Reiser * 3771b1dcc1bSJes Sorensen * Called under (and serialised by) inode->i_mutex. 37808142579SJan Kara * 37908142579SJan Kara * Note: When this function returns, there can be a page in the process of 38008142579SJan Kara * deletion (inside __delete_from_page_cache()) in the specified range. Thus 38108142579SJan Kara * mapping->nrpages can be non-zero when this function returns even after 38208142579SJan Kara * truncation of the whole mapping. 383d7339071SHans Reiser */ 384d7339071SHans Reiser void truncate_inode_pages(struct address_space *mapping, loff_t lstart) 385d7339071SHans Reiser { 386d7339071SHans Reiser truncate_inode_pages_range(mapping, lstart, (loff_t)-1); 387d7339071SHans Reiser } 3881da177e4SLinus Torvalds EXPORT_SYMBOL(truncate_inode_pages); 3891da177e4SLinus Torvalds 39028697355SMike Waychison /** 39191b0abe3SJohannes Weiner * truncate_inode_pages_final - truncate *all* pages before inode dies 39291b0abe3SJohannes Weiner * @mapping: mapping to truncate 39391b0abe3SJohannes Weiner * 39491b0abe3SJohannes Weiner * Called under (and serialized by) inode->i_mutex. 39591b0abe3SJohannes Weiner * 39691b0abe3SJohannes Weiner * Filesystems have to use this in the .evict_inode path to inform the 39791b0abe3SJohannes Weiner * VM that this is the final truncate and the inode is going away. 39891b0abe3SJohannes Weiner */ 39991b0abe3SJohannes Weiner void truncate_inode_pages_final(struct address_space *mapping) 40091b0abe3SJohannes Weiner { 401f9fe48beSRoss Zwisler unsigned long nrexceptional; 40291b0abe3SJohannes Weiner unsigned long nrpages; 40391b0abe3SJohannes Weiner 40491b0abe3SJohannes Weiner /* 40591b0abe3SJohannes Weiner * Page reclaim can not participate in regular inode lifetime 40691b0abe3SJohannes Weiner * management (can't call iput()) and thus can race with the 40791b0abe3SJohannes Weiner * inode teardown. Tell it when the address space is exiting, 40891b0abe3SJohannes Weiner * so that it does not install eviction information after the 40991b0abe3SJohannes Weiner * final truncate has begun. 41091b0abe3SJohannes Weiner */ 41191b0abe3SJohannes Weiner mapping_set_exiting(mapping); 41291b0abe3SJohannes Weiner 41391b0abe3SJohannes Weiner /* 41491b0abe3SJohannes Weiner * When reclaim installs eviction entries, it increases 415f9fe48beSRoss Zwisler * nrexceptional first, then decreases nrpages. Make sure we see 41691b0abe3SJohannes Weiner * this in the right order or we might miss an entry. 41791b0abe3SJohannes Weiner */ 41891b0abe3SJohannes Weiner nrpages = mapping->nrpages; 41991b0abe3SJohannes Weiner smp_rmb(); 420f9fe48beSRoss Zwisler nrexceptional = mapping->nrexceptional; 42191b0abe3SJohannes Weiner 422f9fe48beSRoss Zwisler if (nrpages || nrexceptional) { 42391b0abe3SJohannes Weiner /* 42491b0abe3SJohannes Weiner * As truncation uses a lockless tree lookup, cycle 42591b0abe3SJohannes Weiner * the tree lock to make sure any ongoing tree 42691b0abe3SJohannes Weiner * modification that does not see AS_EXITING is 42791b0abe3SJohannes Weiner * completed before starting the final truncate. 42891b0abe3SJohannes Weiner */ 42991b0abe3SJohannes Weiner spin_lock_irq(&mapping->tree_lock); 43091b0abe3SJohannes Weiner spin_unlock_irq(&mapping->tree_lock); 43191b0abe3SJohannes Weiner 43291b0abe3SJohannes Weiner truncate_inode_pages(mapping, 0); 43391b0abe3SJohannes Weiner } 43491b0abe3SJohannes Weiner } 43591b0abe3SJohannes Weiner EXPORT_SYMBOL(truncate_inode_pages_final); 43691b0abe3SJohannes Weiner 43791b0abe3SJohannes Weiner /** 43828697355SMike Waychison * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode 43928697355SMike Waychison * @mapping: the address_space which holds the pages to invalidate 44028697355SMike Waychison * @start: the offset 'from' which to invalidate 44128697355SMike Waychison * @end: the offset 'to' which to invalidate (inclusive) 44228697355SMike Waychison * 44328697355SMike Waychison * This function only removes the unlocked pages, if you want to 44428697355SMike Waychison * remove all the pages of one inode, you must call truncate_inode_pages. 44528697355SMike Waychison * 44628697355SMike Waychison * invalidate_mapping_pages() will not block on IO activity. It will not 44728697355SMike Waychison * invalidate pages which are dirty, locked, under writeback or mapped into 44828697355SMike Waychison * pagetables. 44928697355SMike Waychison */ 45028697355SMike Waychison unsigned long invalidate_mapping_pages(struct address_space *mapping, 45128697355SMike Waychison pgoff_t start, pgoff_t end) 4521da177e4SLinus Torvalds { 4530cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 4541da177e4SLinus Torvalds struct pagevec pvec; 455b85e0effSHugh Dickins pgoff_t index = start; 45631560180SMinchan Kim unsigned long ret; 45731560180SMinchan Kim unsigned long count = 0; 4581da177e4SLinus Torvalds int i; 4591da177e4SLinus Torvalds 4601da177e4SLinus Torvalds pagevec_init(&pvec, 0); 4610cd6144aSJohannes Weiner while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, 4620cd6144aSJohannes Weiner min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, 4630cd6144aSJohannes Weiner indices)) { 4641da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 4651da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 4661da177e4SLinus Torvalds 467b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 4680cd6144aSJohannes Weiner index = indices[i]; 469b85e0effSHugh Dickins if (index > end) 470b85e0effSHugh Dickins break; 471e0f23603SNeilBrown 4720cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 4730cd6144aSJohannes Weiner clear_exceptional_entry(mapping, index, page); 4740cd6144aSJohannes Weiner continue; 4750cd6144aSJohannes Weiner } 4760cd6144aSJohannes Weiner 477b85e0effSHugh Dickins if (!trylock_page(page)) 478b85e0effSHugh Dickins continue; 479fc127da0SKirill A. Shutemov 4805cbc198aSKirill A. Shutemov WARN_ON(page_to_index(page) != index); 481fc127da0SKirill A. Shutemov 482fc127da0SKirill A. Shutemov /* Middle of THP: skip */ 483fc127da0SKirill A. Shutemov if (PageTransTail(page)) { 484fc127da0SKirill A. Shutemov unlock_page(page); 485fc127da0SKirill A. Shutemov continue; 486fc127da0SKirill A. Shutemov } else if (PageTransHuge(page)) { 487fc127da0SKirill A. Shutemov index += HPAGE_PMD_NR - 1; 488fc127da0SKirill A. Shutemov i += HPAGE_PMD_NR - 1; 489fc127da0SKirill A. Shutemov /* 'end' is in the middle of THP */ 490fc127da0SKirill A. Shutemov if (index == round_down(end, HPAGE_PMD_NR)) 491fc127da0SKirill A. Shutemov continue; 492fc127da0SKirill A. Shutemov } 493fc127da0SKirill A. Shutemov 49431560180SMinchan Kim ret = invalidate_inode_page(page); 4951da177e4SLinus Torvalds unlock_page(page); 49631560180SMinchan Kim /* 49731560180SMinchan Kim * Invalidation is a hint that the page is no longer 49831560180SMinchan Kim * of interest and try to speed up its reclaim. 49931560180SMinchan Kim */ 50031560180SMinchan Kim if (!ret) 501cc5993bdSMinchan Kim deactivate_file_page(page); 50231560180SMinchan Kim count += ret; 5031da177e4SLinus Torvalds } 5040cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 5051da177e4SLinus Torvalds pagevec_release(&pvec); 506fc9a07e7SAndrew Morton cond_resched(); 507b85e0effSHugh Dickins index++; 5081da177e4SLinus Torvalds } 50931560180SMinchan Kim return count; 5101da177e4SLinus Torvalds } 51154bc4855SAnton Altaparmakov EXPORT_SYMBOL(invalidate_mapping_pages); 5121da177e4SLinus Torvalds 513bd4c8ce4SAndrew Morton /* 514bd4c8ce4SAndrew Morton * This is like invalidate_complete_page(), except it ignores the page's 515bd4c8ce4SAndrew Morton * refcount. We do this because invalidate_inode_pages2() needs stronger 516bd4c8ce4SAndrew Morton * invalidation guarantees, and cannot afford to leave pages behind because 5172706a1b8SAnderson Briglia * shrink_page_list() has a temp ref on them, or because they're transiently 5182706a1b8SAnderson Briglia * sitting in the lru_cache_add() pagevecs. 519bd4c8ce4SAndrew Morton */ 520bd4c8ce4SAndrew Morton static int 521bd4c8ce4SAndrew Morton invalidate_complete_page2(struct address_space *mapping, struct page *page) 522bd4c8ce4SAndrew Morton { 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 spin_lock_irqsave(&mapping->tree_lock, flags); 532bd4c8ce4SAndrew Morton if (PageDirty(page)) 533bd4c8ce4SAndrew Morton goto failed; 534bd4c8ce4SAndrew Morton 535266cf658SDavid Howells BUG_ON(page_has_private(page)); 53662cccb8cSJohannes Weiner __delete_from_page_cache(page, NULL); 537c4843a75SGreg Thelen spin_unlock_irqrestore(&mapping->tree_lock, flags); 5386072d13cSLinus Torvalds 5396072d13cSLinus Torvalds if (mapping->a_ops->freepage) 5406072d13cSLinus Torvalds mapping->a_ops->freepage(page); 5416072d13cSLinus Torvalds 54209cbfeafSKirill A. Shutemov put_page(page); /* pagecache ref */ 543bd4c8ce4SAndrew Morton return 1; 544bd4c8ce4SAndrew Morton failed: 545c4843a75SGreg Thelen spin_unlock_irqrestore(&mapping->tree_lock, flags); 546bd4c8ce4SAndrew Morton return 0; 547bd4c8ce4SAndrew Morton } 548bd4c8ce4SAndrew Morton 549e3db7691STrond Myklebust static int do_launder_page(struct address_space *mapping, struct page *page) 550e3db7691STrond Myklebust { 551e3db7691STrond Myklebust if (!PageDirty(page)) 552e3db7691STrond Myklebust return 0; 553e3db7691STrond Myklebust if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) 554e3db7691STrond Myklebust return 0; 555e3db7691STrond Myklebust return mapping->a_ops->launder_page(page); 556e3db7691STrond Myklebust } 557e3db7691STrond Myklebust 5581da177e4SLinus Torvalds /** 5591da177e4SLinus Torvalds * invalidate_inode_pages2_range - remove range of pages from an address_space 56067be2dd1SMartin Waitz * @mapping: the address_space 5611da177e4SLinus Torvalds * @start: the page offset 'from' which to invalidate 5621da177e4SLinus Torvalds * @end: the page offset 'to' which to invalidate (inclusive) 5631da177e4SLinus Torvalds * 5641da177e4SLinus Torvalds * Any pages which are found to be mapped into pagetables are unmapped prior to 5651da177e4SLinus Torvalds * invalidation. 5661da177e4SLinus Torvalds * 5676ccfa806SHisashi Hifumi * Returns -EBUSY if any pages could not be invalidated. 5681da177e4SLinus Torvalds */ 5691da177e4SLinus Torvalds int invalidate_inode_pages2_range(struct address_space *mapping, 5701da177e4SLinus Torvalds pgoff_t start, pgoff_t end) 5711da177e4SLinus Torvalds { 5720cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 5731da177e4SLinus Torvalds struct pagevec pvec; 574b85e0effSHugh Dickins pgoff_t index; 5751da177e4SLinus Torvalds int i; 5761da177e4SLinus Torvalds int ret = 0; 5770dd1334fSHisashi Hifumi int ret2 = 0; 5781da177e4SLinus Torvalds int did_range_unmap = 0; 5791da177e4SLinus Torvalds 5803167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 5811da177e4SLinus Torvalds pagevec_init(&pvec, 0); 582b85e0effSHugh Dickins index = start; 5830cd6144aSJohannes Weiner while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, 5840cd6144aSJohannes Weiner min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, 5850cd6144aSJohannes Weiner indices)) { 5867b965e08STrond Myklebust for (i = 0; i < pagevec_count(&pvec); i++) { 5871da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 588b85e0effSHugh Dickins 589b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 5900cd6144aSJohannes Weiner index = indices[i]; 591b85e0effSHugh Dickins if (index > end) 592b85e0effSHugh Dickins break; 5931da177e4SLinus Torvalds 5940cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 5950cd6144aSJohannes Weiner clear_exceptional_entry(mapping, index, page); 5960cd6144aSJohannes Weiner continue; 5970cd6144aSJohannes Weiner } 5980cd6144aSJohannes Weiner 5991da177e4SLinus Torvalds lock_page(page); 6005cbc198aSKirill A. Shutemov WARN_ON(page_to_index(page) != index); 6011da177e4SLinus Torvalds if (page->mapping != mapping) { 6021da177e4SLinus Torvalds unlock_page(page); 6031da177e4SLinus Torvalds continue; 6041da177e4SLinus Torvalds } 6051da177e4SLinus Torvalds wait_on_page_writeback(page); 606d00806b1SNick Piggin if (page_mapped(page)) { 6071da177e4SLinus Torvalds if (!did_range_unmap) { 6081da177e4SLinus Torvalds /* 6091da177e4SLinus Torvalds * Zap the rest of the file in one hit. 6101da177e4SLinus Torvalds */ 6111da177e4SLinus Torvalds unmap_mapping_range(mapping, 61209cbfeafSKirill A. Shutemov (loff_t)index << PAGE_SHIFT, 613b85e0effSHugh Dickins (loff_t)(1 + end - index) 61409cbfeafSKirill A. Shutemov << PAGE_SHIFT, 6151da177e4SLinus Torvalds 0); 6161da177e4SLinus Torvalds did_range_unmap = 1; 6171da177e4SLinus Torvalds } else { 6181da177e4SLinus Torvalds /* 6191da177e4SLinus Torvalds * Just zap this page 6201da177e4SLinus Torvalds */ 6211da177e4SLinus Torvalds unmap_mapping_range(mapping, 62209cbfeafSKirill A. Shutemov (loff_t)index << PAGE_SHIFT, 62309cbfeafSKirill A. Shutemov PAGE_SIZE, 0); 6241da177e4SLinus Torvalds } 6251da177e4SLinus Torvalds } 626d00806b1SNick Piggin BUG_ON(page_mapped(page)); 6270dd1334fSHisashi Hifumi ret2 = do_launder_page(mapping, page); 6280dd1334fSHisashi Hifumi if (ret2 == 0) { 6290dd1334fSHisashi Hifumi if (!invalidate_complete_page2(mapping, page)) 6306ccfa806SHisashi Hifumi ret2 = -EBUSY; 6310dd1334fSHisashi Hifumi } 6320dd1334fSHisashi Hifumi if (ret2 < 0) 6330dd1334fSHisashi Hifumi ret = ret2; 6341da177e4SLinus Torvalds unlock_page(page); 6351da177e4SLinus Torvalds } 6360cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 6371da177e4SLinus Torvalds pagevec_release(&pvec); 6381da177e4SLinus Torvalds cond_resched(); 639b85e0effSHugh Dickins index++; 6401da177e4SLinus Torvalds } 6413167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 6421da177e4SLinus Torvalds return ret; 6431da177e4SLinus Torvalds } 6441da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); 6451da177e4SLinus Torvalds 6461da177e4SLinus Torvalds /** 6471da177e4SLinus Torvalds * invalidate_inode_pages2 - remove all pages from an address_space 64867be2dd1SMartin Waitz * @mapping: the address_space 6491da177e4SLinus Torvalds * 6501da177e4SLinus Torvalds * Any pages which are found to be mapped into pagetables are unmapped prior to 6511da177e4SLinus Torvalds * invalidation. 6521da177e4SLinus Torvalds * 653e9de25ddSPeng Tao * Returns -EBUSY if any pages could not be invalidated. 6541da177e4SLinus Torvalds */ 6551da177e4SLinus Torvalds int invalidate_inode_pages2(struct address_space *mapping) 6561da177e4SLinus Torvalds { 6571da177e4SLinus Torvalds return invalidate_inode_pages2_range(mapping, 0, -1); 6581da177e4SLinus Torvalds } 6591da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(invalidate_inode_pages2); 66025d9e2d1Snpiggin@suse.de 66125d9e2d1Snpiggin@suse.de /** 66225d9e2d1Snpiggin@suse.de * truncate_pagecache - unmap and remove pagecache that has been truncated 66325d9e2d1Snpiggin@suse.de * @inode: inode 6648a549beaSHugh Dickins * @newsize: new file size 66525d9e2d1Snpiggin@suse.de * 66625d9e2d1Snpiggin@suse.de * inode's new i_size must already be written before truncate_pagecache 66725d9e2d1Snpiggin@suse.de * is called. 66825d9e2d1Snpiggin@suse.de * 66925d9e2d1Snpiggin@suse.de * This function should typically be called before the filesystem 67025d9e2d1Snpiggin@suse.de * releases resources associated with the freed range (eg. deallocates 67125d9e2d1Snpiggin@suse.de * blocks). This way, pagecache will always stay logically coherent 67225d9e2d1Snpiggin@suse.de * with on-disk format, and the filesystem would not have to deal with 67325d9e2d1Snpiggin@suse.de * situations such as writepage being called for a page that has already 67425d9e2d1Snpiggin@suse.de * had its underlying blocks deallocated. 67525d9e2d1Snpiggin@suse.de */ 6767caef267SKirill A. Shutemov void truncate_pagecache(struct inode *inode, loff_t newsize) 67725d9e2d1Snpiggin@suse.de { 67825d9e2d1Snpiggin@suse.de struct address_space *mapping = inode->i_mapping; 6798a549beaSHugh Dickins loff_t holebegin = round_up(newsize, PAGE_SIZE); 68025d9e2d1Snpiggin@suse.de 68125d9e2d1Snpiggin@suse.de /* 68225d9e2d1Snpiggin@suse.de * unmap_mapping_range is called twice, first simply for 68325d9e2d1Snpiggin@suse.de * efficiency so that truncate_inode_pages does fewer 68425d9e2d1Snpiggin@suse.de * single-page unmaps. However after this first call, and 68525d9e2d1Snpiggin@suse.de * before truncate_inode_pages finishes, it is possible for 68625d9e2d1Snpiggin@suse.de * private pages to be COWed, which remain after 68725d9e2d1Snpiggin@suse.de * truncate_inode_pages finishes, hence the second 68825d9e2d1Snpiggin@suse.de * unmap_mapping_range call must be made for correctness. 68925d9e2d1Snpiggin@suse.de */ 6908a549beaSHugh Dickins unmap_mapping_range(mapping, holebegin, 0, 1); 6918a549beaSHugh Dickins truncate_inode_pages(mapping, newsize); 6928a549beaSHugh Dickins unmap_mapping_range(mapping, holebegin, 0, 1); 69325d9e2d1Snpiggin@suse.de } 69425d9e2d1Snpiggin@suse.de EXPORT_SYMBOL(truncate_pagecache); 69525d9e2d1Snpiggin@suse.de 69625d9e2d1Snpiggin@suse.de /** 6972c27c65eSChristoph Hellwig * truncate_setsize - update inode and pagecache for a new file size 6982c27c65eSChristoph Hellwig * @inode: inode 6992c27c65eSChristoph Hellwig * @newsize: new file size 7002c27c65eSChristoph Hellwig * 701382e27daSJan Kara * truncate_setsize updates i_size and performs pagecache truncation (if 702382e27daSJan Kara * necessary) to @newsize. It will be typically be called from the filesystem's 703382e27daSJan Kara * setattr function when ATTR_SIZE is passed in. 7042c27c65eSChristoph Hellwig * 70577783d06SJan Kara * Must be called with a lock serializing truncates and writes (generally 70677783d06SJan Kara * i_mutex but e.g. xfs uses a different lock) and before all filesystem 70777783d06SJan Kara * specific block truncation has been performed. 7082c27c65eSChristoph Hellwig */ 7092c27c65eSChristoph Hellwig void truncate_setsize(struct inode *inode, loff_t newsize) 7102c27c65eSChristoph Hellwig { 71190a80202SJan Kara loff_t oldsize = inode->i_size; 71290a80202SJan Kara 7132c27c65eSChristoph Hellwig i_size_write(inode, newsize); 71490a80202SJan Kara if (newsize > oldsize) 71590a80202SJan Kara pagecache_isize_extended(inode, oldsize, newsize); 7167caef267SKirill A. Shutemov truncate_pagecache(inode, newsize); 7172c27c65eSChristoph Hellwig } 7182c27c65eSChristoph Hellwig EXPORT_SYMBOL(truncate_setsize); 7192c27c65eSChristoph Hellwig 7202c27c65eSChristoph Hellwig /** 72190a80202SJan Kara * pagecache_isize_extended - update pagecache after extension of i_size 72290a80202SJan Kara * @inode: inode for which i_size was extended 72390a80202SJan Kara * @from: original inode size 72490a80202SJan Kara * @to: new inode size 72590a80202SJan Kara * 72690a80202SJan Kara * Handle extension of inode size either caused by extending truncate or by 72790a80202SJan Kara * write starting after current i_size. We mark the page straddling current 72890a80202SJan Kara * i_size RO so that page_mkwrite() is called on the nearest write access to 72990a80202SJan Kara * the page. This way filesystem can be sure that page_mkwrite() is called on 73090a80202SJan Kara * the page before user writes to the page via mmap after the i_size has been 73190a80202SJan Kara * changed. 73290a80202SJan Kara * 73390a80202SJan Kara * The function must be called after i_size is updated so that page fault 73490a80202SJan Kara * coming after we unlock the page will already see the new i_size. 73590a80202SJan Kara * The function must be called while we still hold i_mutex - this not only 73690a80202SJan Kara * makes sure i_size is stable but also that userspace cannot observe new 73790a80202SJan Kara * i_size value before we are prepared to store mmap writes at new inode size. 73890a80202SJan Kara */ 73990a80202SJan Kara void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) 74090a80202SJan Kara { 74190a80202SJan Kara int bsize = 1 << inode->i_blkbits; 74290a80202SJan Kara loff_t rounded_from; 74390a80202SJan Kara struct page *page; 74490a80202SJan Kara pgoff_t index; 74590a80202SJan Kara 74690a80202SJan Kara WARN_ON(to > inode->i_size); 74790a80202SJan Kara 74809cbfeafSKirill A. Shutemov if (from >= to || bsize == PAGE_SIZE) 74990a80202SJan Kara return; 75090a80202SJan Kara /* Page straddling @from will not have any hole block created? */ 75190a80202SJan Kara rounded_from = round_up(from, bsize); 75209cbfeafSKirill A. Shutemov if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1))) 75390a80202SJan Kara return; 75490a80202SJan Kara 75509cbfeafSKirill A. Shutemov index = from >> PAGE_SHIFT; 75690a80202SJan Kara page = find_lock_page(inode->i_mapping, index); 75790a80202SJan Kara /* Page not cached? Nothing to do */ 75890a80202SJan Kara if (!page) 75990a80202SJan Kara return; 76090a80202SJan Kara /* 76190a80202SJan Kara * See clear_page_dirty_for_io() for details why set_page_dirty() 76290a80202SJan Kara * is needed. 76390a80202SJan Kara */ 76490a80202SJan Kara if (page_mkclean(page)) 76590a80202SJan Kara set_page_dirty(page); 76690a80202SJan Kara unlock_page(page); 76709cbfeafSKirill A. Shutemov put_page(page); 76890a80202SJan Kara } 76990a80202SJan Kara EXPORT_SYMBOL(pagecache_isize_extended); 77090a80202SJan Kara 77190a80202SJan Kara /** 772623e3db9SHugh Dickins * truncate_pagecache_range - unmap and remove pagecache that is hole-punched 773623e3db9SHugh Dickins * @inode: inode 774623e3db9SHugh Dickins * @lstart: offset of beginning of hole 775623e3db9SHugh Dickins * @lend: offset of last byte of hole 776623e3db9SHugh Dickins * 777623e3db9SHugh Dickins * This function should typically be called before the filesystem 778623e3db9SHugh Dickins * releases resources associated with the freed range (eg. deallocates 779623e3db9SHugh Dickins * blocks). This way, pagecache will always stay logically coherent 780623e3db9SHugh Dickins * with on-disk format, and the filesystem would not have to deal with 781623e3db9SHugh Dickins * situations such as writepage being called for a page that has already 782623e3db9SHugh Dickins * had its underlying blocks deallocated. 783623e3db9SHugh Dickins */ 784623e3db9SHugh Dickins void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) 785623e3db9SHugh Dickins { 786623e3db9SHugh Dickins struct address_space *mapping = inode->i_mapping; 787623e3db9SHugh Dickins loff_t unmap_start = round_up(lstart, PAGE_SIZE); 788623e3db9SHugh Dickins loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; 789623e3db9SHugh Dickins /* 790623e3db9SHugh Dickins * This rounding is currently just for example: unmap_mapping_range 791623e3db9SHugh Dickins * expands its hole outwards, whereas we want it to contract the hole 792623e3db9SHugh Dickins * inwards. However, existing callers of truncate_pagecache_range are 7935a720394SLukas Czerner * doing their own page rounding first. Note that unmap_mapping_range 7945a720394SLukas Czerner * allows holelen 0 for all, and we allow lend -1 for end of file. 795623e3db9SHugh Dickins */ 796623e3db9SHugh Dickins 797623e3db9SHugh Dickins /* 798623e3db9SHugh Dickins * Unlike in truncate_pagecache, unmap_mapping_range is called only 799623e3db9SHugh Dickins * once (before truncating pagecache), and without "even_cows" flag: 800623e3db9SHugh Dickins * hole-punching should not remove private COWed pages from the hole. 801623e3db9SHugh Dickins */ 802623e3db9SHugh Dickins if ((u64)unmap_end > (u64)unmap_start) 803623e3db9SHugh Dickins unmap_mapping_range(mapping, unmap_start, 804623e3db9SHugh Dickins 1 + unmap_end - unmap_start, 0); 805623e3db9SHugh Dickins truncate_inode_pages_range(mapping, lstart, lend); 806623e3db9SHugh Dickins } 807623e3db9SHugh Dickins EXPORT_SYMBOL(truncate_pagecache_range); 808