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 */ 47f9fe48beSRoss Zwisler if (!__radix_tree_lookup(&mapping->page_tree, index, &node, 48f9fe48beSRoss Zwisler &slot)) 49449dd698SJohannes Weiner goto unlock; 50449dd698SJohannes Weiner if (*slot != entry) 51449dd698SJohannes Weiner goto unlock; 52449dd698SJohannes Weiner radix_tree_replace_slot(slot, NULL); 53f9fe48beSRoss Zwisler mapping->nrexceptional--; 54449dd698SJohannes Weiner if (!node) 55449dd698SJohannes Weiner goto unlock; 56449dd698SJohannes Weiner workingset_node_shadows_dec(node); 57449dd698SJohannes Weiner /* 58449dd698SJohannes Weiner * Don't track node without shadow entries. 59449dd698SJohannes Weiner * 60449dd698SJohannes Weiner * Avoid acquiring the list_lru lock if already untracked. 61449dd698SJohannes Weiner * The list_empty() test is safe as node->private_list is 62449dd698SJohannes Weiner * protected by mapping->tree_lock. 63449dd698SJohannes Weiner */ 64449dd698SJohannes Weiner if (!workingset_node_shadows(node) && 65449dd698SJohannes Weiner !list_empty(&node->private_list)) 66f9fe48beSRoss Zwisler list_lru_del(&workingset_shadow_nodes, 67f9fe48beSRoss Zwisler &node->private_list); 68449dd698SJohannes Weiner __radix_tree_delete_node(&mapping->page_tree, node); 69449dd698SJohannes Weiner unlock: 700cd6144aSJohannes Weiner spin_unlock_irq(&mapping->tree_lock); 710cd6144aSJohannes Weiner } 721da177e4SLinus Torvalds 73cf9a2ae8SDavid Howells /** 7428bc44d7SFengguang Wu * do_invalidatepage - invalidate part or all of a page 75cf9a2ae8SDavid Howells * @page: the page which is affected 76d47992f8SLukas Czerner * @offset: start of the range to invalidate 77d47992f8SLukas Czerner * @length: length of the range to invalidate 78cf9a2ae8SDavid Howells * 79cf9a2ae8SDavid Howells * do_invalidatepage() is called when all or part of the page has become 80cf9a2ae8SDavid Howells * invalidated by a truncate operation. 81cf9a2ae8SDavid Howells * 82cf9a2ae8SDavid Howells * do_invalidatepage() does not have to release all buffers, but it must 83cf9a2ae8SDavid Howells * ensure that no dirty buffer is left outside @offset and that no I/O 84cf9a2ae8SDavid Howells * is underway against any of the blocks which are outside the truncation 85cf9a2ae8SDavid Howells * point. Because the caller is about to free (and possibly reuse) those 86cf9a2ae8SDavid Howells * blocks on-disk. 87cf9a2ae8SDavid Howells */ 88d47992f8SLukas Czerner void do_invalidatepage(struct page *page, unsigned int offset, 89d47992f8SLukas Czerner unsigned int length) 90cf9a2ae8SDavid Howells { 91d47992f8SLukas Czerner void (*invalidatepage)(struct page *, unsigned int, unsigned int); 92d47992f8SLukas Czerner 93cf9a2ae8SDavid Howells invalidatepage = page->mapping->a_ops->invalidatepage; 949361401eSDavid Howells #ifdef CONFIG_BLOCK 95cf9a2ae8SDavid Howells if (!invalidatepage) 96cf9a2ae8SDavid Howells invalidatepage = block_invalidatepage; 979361401eSDavid Howells #endif 98cf9a2ae8SDavid Howells if (invalidatepage) 99d47992f8SLukas Czerner (*invalidatepage)(page, offset, length); 100cf9a2ae8SDavid Howells } 101cf9a2ae8SDavid Howells 102ecdfc978SLinus Torvalds /* 1031da177e4SLinus Torvalds * If truncate cannot remove the fs-private metadata from the page, the page 10462e1c553SShaohua Li * becomes orphaned. It will be left on the LRU and may even be mapped into 10554cb8821SNick Piggin * user pagetables if we're racing with filemap_fault(). 1061da177e4SLinus Torvalds * 1071da177e4SLinus Torvalds * We need to bale out if page->mapping is no longer equal to the original 1081da177e4SLinus Torvalds * mapping. This happens a) when the VM reclaimed the page while we waited on 109fc0ecff6SAndrew Morton * its lock, b) when a concurrent invalidate_mapping_pages got there first and 1101da177e4SLinus Torvalds * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. 1111da177e4SLinus Torvalds */ 112750b4987SNick Piggin static int 1131da177e4SLinus Torvalds truncate_complete_page(struct address_space *mapping, struct page *page) 1141da177e4SLinus Torvalds { 1151da177e4SLinus Torvalds if (page->mapping != mapping) 116750b4987SNick Piggin return -EIO; 1171da177e4SLinus Torvalds 118266cf658SDavid Howells if (page_has_private(page)) 11909cbfeafSKirill A. Shutemov do_invalidatepage(page, 0, PAGE_SIZE); 1201da177e4SLinus Torvalds 121b9ea2515SKonstantin Khlebnikov /* 122b9ea2515SKonstantin Khlebnikov * Some filesystems seem to re-dirty the page even after 123b9ea2515SKonstantin Khlebnikov * the VM has canceled the dirty bit (eg ext3 journaling). 124b9ea2515SKonstantin Khlebnikov * Hence dirty accounting check is placed after invalidation. 125b9ea2515SKonstantin Khlebnikov */ 12611f81becSTejun Heo cancel_dirty_page(page); 1271da177e4SLinus Torvalds ClearPageMappedToDisk(page); 1285adc7b51SMinchan Kim delete_from_page_cache(page); 129750b4987SNick Piggin return 0; 1301da177e4SLinus Torvalds } 1311da177e4SLinus Torvalds 1321da177e4SLinus Torvalds /* 133fc0ecff6SAndrew Morton * This is for invalidate_mapping_pages(). That function can be called at 1341da177e4SLinus Torvalds * any time, and is not supposed to throw away dirty pages. But pages can 1350fd0e6b0SNick Piggin * be marked dirty at any time too, so use remove_mapping which safely 1360fd0e6b0SNick Piggin * discards clean, unused pages. 1371da177e4SLinus Torvalds * 1381da177e4SLinus Torvalds * Returns non-zero if the page was successfully invalidated. 1391da177e4SLinus Torvalds */ 1401da177e4SLinus Torvalds static int 1411da177e4SLinus Torvalds invalidate_complete_page(struct address_space *mapping, struct page *page) 1421da177e4SLinus Torvalds { 1430fd0e6b0SNick Piggin int ret; 1440fd0e6b0SNick Piggin 1451da177e4SLinus Torvalds if (page->mapping != mapping) 1461da177e4SLinus Torvalds return 0; 1471da177e4SLinus Torvalds 148266cf658SDavid Howells if (page_has_private(page) && !try_to_release_page(page, 0)) 1491da177e4SLinus Torvalds return 0; 1501da177e4SLinus Torvalds 1510fd0e6b0SNick Piggin ret = remove_mapping(mapping, page); 1520fd0e6b0SNick Piggin 1530fd0e6b0SNick Piggin return ret; 1541da177e4SLinus Torvalds } 1551da177e4SLinus Torvalds 156750b4987SNick Piggin int truncate_inode_page(struct address_space *mapping, struct page *page) 157750b4987SNick Piggin { 158fc127da0SKirill A. Shutemov loff_t holelen; 159fc127da0SKirill A. Shutemov VM_BUG_ON_PAGE(PageTail(page), page); 160fc127da0SKirill A. Shutemov 161fc127da0SKirill A. Shutemov holelen = PageTransHuge(page) ? HPAGE_PMD_SIZE : PAGE_SIZE; 162750b4987SNick Piggin if (page_mapped(page)) { 163750b4987SNick Piggin unmap_mapping_range(mapping, 16409cbfeafSKirill A. Shutemov (loff_t)page->index << PAGE_SHIFT, 165fc127da0SKirill A. Shutemov holelen, 0); 166750b4987SNick Piggin } 167750b4987SNick Piggin return truncate_complete_page(mapping, page); 168750b4987SNick Piggin } 169750b4987SNick Piggin 17083f78668SWu Fengguang /* 17125718736SAndi Kleen * Used to get rid of pages on hardware memory corruption. 17225718736SAndi Kleen */ 17325718736SAndi Kleen int generic_error_remove_page(struct address_space *mapping, struct page *page) 17425718736SAndi Kleen { 17525718736SAndi Kleen if (!mapping) 17625718736SAndi Kleen return -EINVAL; 17725718736SAndi Kleen /* 17825718736SAndi Kleen * Only punch for normal data pages for now. 17925718736SAndi Kleen * Handling other types like directories would need more auditing. 18025718736SAndi Kleen */ 18125718736SAndi Kleen if (!S_ISREG(mapping->host->i_mode)) 18225718736SAndi Kleen return -EIO; 18325718736SAndi Kleen return truncate_inode_page(mapping, page); 18425718736SAndi Kleen } 18525718736SAndi Kleen EXPORT_SYMBOL(generic_error_remove_page); 18625718736SAndi Kleen 18725718736SAndi Kleen /* 18883f78668SWu Fengguang * Safely invalidate one page from its pagecache mapping. 18983f78668SWu Fengguang * It only drops clean, unused pages. The page must be locked. 19083f78668SWu Fengguang * 19183f78668SWu Fengguang * Returns 1 if the page is successfully invalidated, otherwise 0. 19283f78668SWu Fengguang */ 19383f78668SWu Fengguang int invalidate_inode_page(struct page *page) 19483f78668SWu Fengguang { 19583f78668SWu Fengguang struct address_space *mapping = page_mapping(page); 19683f78668SWu Fengguang if (!mapping) 19783f78668SWu Fengguang return 0; 19883f78668SWu Fengguang if (PageDirty(page) || PageWriteback(page)) 19983f78668SWu Fengguang return 0; 20083f78668SWu Fengguang if (page_mapped(page)) 20183f78668SWu Fengguang return 0; 20283f78668SWu Fengguang return invalidate_complete_page(mapping, page); 20383f78668SWu Fengguang } 20483f78668SWu Fengguang 2051da177e4SLinus Torvalds /** 20673c1e204SLiu Bo * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets 2071da177e4SLinus Torvalds * @mapping: mapping to truncate 2081da177e4SLinus Torvalds * @lstart: offset from which to truncate 2095a720394SLukas Czerner * @lend: offset to which to truncate (inclusive) 2101da177e4SLinus Torvalds * 211d7339071SHans Reiser * Truncate the page cache, removing the pages that are between 2125a720394SLukas Czerner * specified offsets (and zeroing out partial pages 2135a720394SLukas Czerner * if lstart or lend + 1 is not page aligned). 2141da177e4SLinus Torvalds * 2151da177e4SLinus Torvalds * Truncate takes two passes - the first pass is nonblocking. It will not 2161da177e4SLinus Torvalds * block on page locks and it will not block on writeback. The second pass 2171da177e4SLinus Torvalds * will wait. This is to prevent as much IO as possible in the affected region. 2181da177e4SLinus Torvalds * The first pass will remove most pages, so the search cost of the second pass 2191da177e4SLinus Torvalds * is low. 2201da177e4SLinus Torvalds * 2211da177e4SLinus Torvalds * We pass down the cache-hot hint to the page freeing code. Even if the 2221da177e4SLinus Torvalds * mapping is large, it is probably the case that the final pages are the most 2231da177e4SLinus Torvalds * recently touched, and freeing happens in ascending file offset order. 2245a720394SLukas Czerner * 2255a720394SLukas Czerner * Note that since ->invalidatepage() accepts range to invalidate 2265a720394SLukas Czerner * truncate_inode_pages_range is able to handle cases where lend + 1 is not 2275a720394SLukas Czerner * page aligned properly. 2281da177e4SLinus Torvalds */ 229d7339071SHans Reiser void truncate_inode_pages_range(struct address_space *mapping, 230d7339071SHans Reiser loff_t lstart, loff_t lend) 2311da177e4SLinus Torvalds { 2325a720394SLukas Czerner pgoff_t start; /* inclusive */ 2335a720394SLukas Czerner pgoff_t end; /* exclusive */ 2345a720394SLukas Czerner unsigned int partial_start; /* inclusive */ 2355a720394SLukas Czerner unsigned int partial_end; /* exclusive */ 2361da177e4SLinus Torvalds struct pagevec pvec; 2370cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 238b85e0effSHugh Dickins pgoff_t index; 2391da177e4SLinus Torvalds int i; 2401da177e4SLinus Torvalds 2413167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 242f9fe48beSRoss Zwisler if (mapping->nrpages == 0 && mapping->nrexceptional == 0) 2431da177e4SLinus Torvalds return; 2441da177e4SLinus Torvalds 2455a720394SLukas Czerner /* Offsets within partial pages */ 24609cbfeafSKirill A. Shutemov partial_start = lstart & (PAGE_SIZE - 1); 24709cbfeafSKirill A. Shutemov partial_end = (lend + 1) & (PAGE_SIZE - 1); 2485a720394SLukas Czerner 2495a720394SLukas Czerner /* 2505a720394SLukas Czerner * 'start' and 'end' always covers the range of pages to be fully 2515a720394SLukas Czerner * truncated. Partial pages are covered with 'partial_start' at the 2525a720394SLukas Czerner * start of the range and 'partial_end' at the end of the range. 2535a720394SLukas Czerner * Note that 'end' is exclusive while 'lend' is inclusive. 2545a720394SLukas Czerner */ 25509cbfeafSKirill A. Shutemov start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; 2565a720394SLukas Czerner if (lend == -1) 2575a720394SLukas Czerner /* 2585a720394SLukas Czerner * lend == -1 indicates end-of-file so we have to set 'end' 2595a720394SLukas Czerner * to the highest possible pgoff_t and since the type is 2605a720394SLukas Czerner * unsigned we're using -1. 2615a720394SLukas Czerner */ 2625a720394SLukas Czerner end = -1; 2635a720394SLukas Czerner else 26409cbfeafSKirill A. Shutemov end = (lend + 1) >> PAGE_SHIFT; 265d7339071SHans Reiser 2661da177e4SLinus Torvalds pagevec_init(&pvec, 0); 267b85e0effSHugh Dickins index = start; 2680cd6144aSJohannes Weiner while (index < end && pagevec_lookup_entries(&pvec, mapping, index, 2690cd6144aSJohannes Weiner min(end - index, (pgoff_t)PAGEVEC_SIZE), 2700cd6144aSJohannes Weiner indices)) { 2711da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 2721da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 2731da177e4SLinus Torvalds 274b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 2750cd6144aSJohannes Weiner index = indices[i]; 2765a720394SLukas Czerner if (index >= end) 277d7339071SHans Reiser break; 278d7339071SHans Reiser 2790cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 2800cd6144aSJohannes Weiner clear_exceptional_entry(mapping, index, page); 2810cd6144aSJohannes Weiner continue; 2820cd6144aSJohannes Weiner } 2830cd6144aSJohannes Weiner 284529ae9aaSNick Piggin if (!trylock_page(page)) 2851da177e4SLinus Torvalds continue; 286*5cbc198aSKirill A. Shutemov WARN_ON(page_to_index(page) != index); 2871da177e4SLinus Torvalds if (PageWriteback(page)) { 2881da177e4SLinus Torvalds unlock_page(page); 2891da177e4SLinus Torvalds continue; 2901da177e4SLinus Torvalds } 291750b4987SNick Piggin truncate_inode_page(mapping, page); 2921da177e4SLinus Torvalds unlock_page(page); 2931da177e4SLinus Torvalds } 2940cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 2951da177e4SLinus Torvalds pagevec_release(&pvec); 2961da177e4SLinus Torvalds cond_resched(); 297b85e0effSHugh Dickins index++; 2981da177e4SLinus Torvalds } 2991da177e4SLinus Torvalds 3005a720394SLukas Czerner if (partial_start) { 3011da177e4SLinus Torvalds struct page *page = find_lock_page(mapping, start - 1); 3021da177e4SLinus Torvalds if (page) { 30309cbfeafSKirill A. Shutemov unsigned int top = PAGE_SIZE; 3045a720394SLukas Czerner if (start > end) { 3055a720394SLukas Czerner /* Truncation within a single page */ 3065a720394SLukas Czerner top = partial_end; 3075a720394SLukas Czerner partial_end = 0; 3085a720394SLukas Czerner } 3091da177e4SLinus Torvalds wait_on_page_writeback(page); 3105a720394SLukas Czerner zero_user_segment(page, partial_start, top); 3115a720394SLukas Czerner cleancache_invalidate_page(mapping, page); 3125a720394SLukas Czerner if (page_has_private(page)) 3135a720394SLukas Czerner do_invalidatepage(page, partial_start, 3145a720394SLukas Czerner top - partial_start); 3151da177e4SLinus Torvalds unlock_page(page); 31609cbfeafSKirill A. Shutemov put_page(page); 3171da177e4SLinus Torvalds } 3181da177e4SLinus Torvalds } 3195a720394SLukas Czerner if (partial_end) { 3205a720394SLukas Czerner struct page *page = find_lock_page(mapping, end); 3215a720394SLukas Czerner if (page) { 3225a720394SLukas Czerner wait_on_page_writeback(page); 3235a720394SLukas Czerner zero_user_segment(page, 0, partial_end); 3245a720394SLukas Czerner cleancache_invalidate_page(mapping, page); 3255a720394SLukas Czerner if (page_has_private(page)) 3265a720394SLukas Czerner do_invalidatepage(page, 0, 3275a720394SLukas Czerner partial_end); 3285a720394SLukas Czerner unlock_page(page); 32909cbfeafSKirill A. Shutemov put_page(page); 3305a720394SLukas Czerner } 3315a720394SLukas Czerner } 3325a720394SLukas Czerner /* 3335a720394SLukas Czerner * If the truncation happened within a single page no pages 3345a720394SLukas Czerner * will be released, just zeroed, so we can bail out now. 3355a720394SLukas Czerner */ 3365a720394SLukas Czerner if (start >= end) 3375a720394SLukas Czerner return; 3381da177e4SLinus Torvalds 339b85e0effSHugh Dickins index = start; 3401da177e4SLinus Torvalds for ( ; ; ) { 3411da177e4SLinus Torvalds cond_resched(); 3420cd6144aSJohannes Weiner if (!pagevec_lookup_entries(&pvec, mapping, index, 343792ceaefSHugh Dickins min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) { 344792ceaefSHugh Dickins /* If all gone from start onwards, we're done */ 345b85e0effSHugh Dickins if (index == start) 3461da177e4SLinus Torvalds break; 347792ceaefSHugh Dickins /* Otherwise restart to make sure all gone */ 348b85e0effSHugh Dickins index = start; 3491da177e4SLinus Torvalds continue; 3501da177e4SLinus Torvalds } 3510cd6144aSJohannes Weiner if (index == start && indices[0] >= end) { 352792ceaefSHugh Dickins /* All gone out of hole to be punched, we're done */ 3530cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 354d7339071SHans Reiser pagevec_release(&pvec); 355d7339071SHans Reiser break; 356d7339071SHans Reiser } 3571da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 3581da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 3591da177e4SLinus Torvalds 360b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 3610cd6144aSJohannes Weiner index = indices[i]; 362792ceaefSHugh Dickins if (index >= end) { 363792ceaefSHugh Dickins /* Restart punch to make sure all gone */ 364792ceaefSHugh Dickins index = start - 1; 365d7339071SHans Reiser break; 366792ceaefSHugh Dickins } 367b85e0effSHugh Dickins 3680cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 3690cd6144aSJohannes Weiner clear_exceptional_entry(mapping, index, page); 3700cd6144aSJohannes Weiner continue; 3710cd6144aSJohannes Weiner } 3720cd6144aSJohannes Weiner 3731da177e4SLinus Torvalds lock_page(page); 374*5cbc198aSKirill A. Shutemov WARN_ON(page_to_index(page) != index); 3751da177e4SLinus Torvalds wait_on_page_writeback(page); 376750b4987SNick Piggin truncate_inode_page(mapping, page); 3771da177e4SLinus Torvalds unlock_page(page); 3781da177e4SLinus Torvalds } 3790cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 3801da177e4SLinus Torvalds pagevec_release(&pvec); 381b85e0effSHugh Dickins index++; 3821da177e4SLinus Torvalds } 3833167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 3841da177e4SLinus Torvalds } 385d7339071SHans Reiser EXPORT_SYMBOL(truncate_inode_pages_range); 3861da177e4SLinus Torvalds 387d7339071SHans Reiser /** 388d7339071SHans Reiser * truncate_inode_pages - truncate *all* the pages from an offset 389d7339071SHans Reiser * @mapping: mapping to truncate 390d7339071SHans Reiser * @lstart: offset from which to truncate 391d7339071SHans Reiser * 3921b1dcc1bSJes Sorensen * Called under (and serialised by) inode->i_mutex. 39308142579SJan Kara * 39408142579SJan Kara * Note: When this function returns, there can be a page in the process of 39508142579SJan Kara * deletion (inside __delete_from_page_cache()) in the specified range. Thus 39608142579SJan Kara * mapping->nrpages can be non-zero when this function returns even after 39708142579SJan Kara * truncation of the whole mapping. 398d7339071SHans Reiser */ 399d7339071SHans Reiser void truncate_inode_pages(struct address_space *mapping, loff_t lstart) 400d7339071SHans Reiser { 401d7339071SHans Reiser truncate_inode_pages_range(mapping, lstart, (loff_t)-1); 402d7339071SHans Reiser } 4031da177e4SLinus Torvalds EXPORT_SYMBOL(truncate_inode_pages); 4041da177e4SLinus Torvalds 40528697355SMike Waychison /** 40691b0abe3SJohannes Weiner * truncate_inode_pages_final - truncate *all* pages before inode dies 40791b0abe3SJohannes Weiner * @mapping: mapping to truncate 40891b0abe3SJohannes Weiner * 40991b0abe3SJohannes Weiner * Called under (and serialized by) inode->i_mutex. 41091b0abe3SJohannes Weiner * 41191b0abe3SJohannes Weiner * Filesystems have to use this in the .evict_inode path to inform the 41291b0abe3SJohannes Weiner * VM that this is the final truncate and the inode is going away. 41391b0abe3SJohannes Weiner */ 41491b0abe3SJohannes Weiner void truncate_inode_pages_final(struct address_space *mapping) 41591b0abe3SJohannes Weiner { 416f9fe48beSRoss Zwisler unsigned long nrexceptional; 41791b0abe3SJohannes Weiner unsigned long nrpages; 41891b0abe3SJohannes Weiner 41991b0abe3SJohannes Weiner /* 42091b0abe3SJohannes Weiner * Page reclaim can not participate in regular inode lifetime 42191b0abe3SJohannes Weiner * management (can't call iput()) and thus can race with the 42291b0abe3SJohannes Weiner * inode teardown. Tell it when the address space is exiting, 42391b0abe3SJohannes Weiner * so that it does not install eviction information after the 42491b0abe3SJohannes Weiner * final truncate has begun. 42591b0abe3SJohannes Weiner */ 42691b0abe3SJohannes Weiner mapping_set_exiting(mapping); 42791b0abe3SJohannes Weiner 42891b0abe3SJohannes Weiner /* 42991b0abe3SJohannes Weiner * When reclaim installs eviction entries, it increases 430f9fe48beSRoss Zwisler * nrexceptional first, then decreases nrpages. Make sure we see 43191b0abe3SJohannes Weiner * this in the right order or we might miss an entry. 43291b0abe3SJohannes Weiner */ 43391b0abe3SJohannes Weiner nrpages = mapping->nrpages; 43491b0abe3SJohannes Weiner smp_rmb(); 435f9fe48beSRoss Zwisler nrexceptional = mapping->nrexceptional; 43691b0abe3SJohannes Weiner 437f9fe48beSRoss Zwisler if (nrpages || nrexceptional) { 43891b0abe3SJohannes Weiner /* 43991b0abe3SJohannes Weiner * As truncation uses a lockless tree lookup, cycle 44091b0abe3SJohannes Weiner * the tree lock to make sure any ongoing tree 44191b0abe3SJohannes Weiner * modification that does not see AS_EXITING is 44291b0abe3SJohannes Weiner * completed before starting the final truncate. 44391b0abe3SJohannes Weiner */ 44491b0abe3SJohannes Weiner spin_lock_irq(&mapping->tree_lock); 44591b0abe3SJohannes Weiner spin_unlock_irq(&mapping->tree_lock); 44691b0abe3SJohannes Weiner 44791b0abe3SJohannes Weiner truncate_inode_pages(mapping, 0); 44891b0abe3SJohannes Weiner } 44991b0abe3SJohannes Weiner } 45091b0abe3SJohannes Weiner EXPORT_SYMBOL(truncate_inode_pages_final); 45191b0abe3SJohannes Weiner 45291b0abe3SJohannes Weiner /** 45328697355SMike Waychison * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode 45428697355SMike Waychison * @mapping: the address_space which holds the pages to invalidate 45528697355SMike Waychison * @start: the offset 'from' which to invalidate 45628697355SMike Waychison * @end: the offset 'to' which to invalidate (inclusive) 45728697355SMike Waychison * 45828697355SMike Waychison * This function only removes the unlocked pages, if you want to 45928697355SMike Waychison * remove all the pages of one inode, you must call truncate_inode_pages. 46028697355SMike Waychison * 46128697355SMike Waychison * invalidate_mapping_pages() will not block on IO activity. It will not 46228697355SMike Waychison * invalidate pages which are dirty, locked, under writeback or mapped into 46328697355SMike Waychison * pagetables. 46428697355SMike Waychison */ 46528697355SMike Waychison unsigned long invalidate_mapping_pages(struct address_space *mapping, 46628697355SMike Waychison pgoff_t start, pgoff_t end) 4671da177e4SLinus Torvalds { 4680cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 4691da177e4SLinus Torvalds struct pagevec pvec; 470b85e0effSHugh Dickins pgoff_t index = start; 47131560180SMinchan Kim unsigned long ret; 47231560180SMinchan Kim unsigned long count = 0; 4731da177e4SLinus Torvalds int i; 4741da177e4SLinus Torvalds 4751da177e4SLinus Torvalds pagevec_init(&pvec, 0); 4760cd6144aSJohannes Weiner while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, 4770cd6144aSJohannes Weiner min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, 4780cd6144aSJohannes Weiner indices)) { 4791da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 4801da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 4811da177e4SLinus Torvalds 482b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 4830cd6144aSJohannes Weiner index = indices[i]; 484b85e0effSHugh Dickins if (index > end) 485b85e0effSHugh Dickins break; 486e0f23603SNeilBrown 4870cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 4880cd6144aSJohannes Weiner clear_exceptional_entry(mapping, index, page); 4890cd6144aSJohannes Weiner continue; 4900cd6144aSJohannes Weiner } 4910cd6144aSJohannes Weiner 492b85e0effSHugh Dickins if (!trylock_page(page)) 493b85e0effSHugh Dickins continue; 494fc127da0SKirill A. Shutemov 495*5cbc198aSKirill A. Shutemov WARN_ON(page_to_index(page) != index); 496fc127da0SKirill A. Shutemov 497fc127da0SKirill A. Shutemov /* Middle of THP: skip */ 498fc127da0SKirill A. Shutemov if (PageTransTail(page)) { 499fc127da0SKirill A. Shutemov unlock_page(page); 500fc127da0SKirill A. Shutemov continue; 501fc127da0SKirill A. Shutemov } else if (PageTransHuge(page)) { 502fc127da0SKirill A. Shutemov index += HPAGE_PMD_NR - 1; 503fc127da0SKirill A. Shutemov i += HPAGE_PMD_NR - 1; 504fc127da0SKirill A. Shutemov /* 'end' is in the middle of THP */ 505fc127da0SKirill A. Shutemov if (index == round_down(end, HPAGE_PMD_NR)) 506fc127da0SKirill A. Shutemov continue; 507fc127da0SKirill A. Shutemov } 508fc127da0SKirill A. Shutemov 50931560180SMinchan Kim ret = invalidate_inode_page(page); 5101da177e4SLinus Torvalds unlock_page(page); 51131560180SMinchan Kim /* 51231560180SMinchan Kim * Invalidation is a hint that the page is no longer 51331560180SMinchan Kim * of interest and try to speed up its reclaim. 51431560180SMinchan Kim */ 51531560180SMinchan Kim if (!ret) 516cc5993bdSMinchan Kim deactivate_file_page(page); 51731560180SMinchan Kim count += ret; 5181da177e4SLinus Torvalds } 5190cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 5201da177e4SLinus Torvalds pagevec_release(&pvec); 521fc9a07e7SAndrew Morton cond_resched(); 522b85e0effSHugh Dickins index++; 5231da177e4SLinus Torvalds } 52431560180SMinchan Kim return count; 5251da177e4SLinus Torvalds } 52654bc4855SAnton Altaparmakov EXPORT_SYMBOL(invalidate_mapping_pages); 5271da177e4SLinus Torvalds 528bd4c8ce4SAndrew Morton /* 529bd4c8ce4SAndrew Morton * This is like invalidate_complete_page(), except it ignores the page's 530bd4c8ce4SAndrew Morton * refcount. We do this because invalidate_inode_pages2() needs stronger 531bd4c8ce4SAndrew Morton * invalidation guarantees, and cannot afford to leave pages behind because 5322706a1b8SAnderson Briglia * shrink_page_list() has a temp ref on them, or because they're transiently 5332706a1b8SAnderson Briglia * sitting in the lru_cache_add() pagevecs. 534bd4c8ce4SAndrew Morton */ 535bd4c8ce4SAndrew Morton static int 536bd4c8ce4SAndrew Morton invalidate_complete_page2(struct address_space *mapping, struct page *page) 537bd4c8ce4SAndrew Morton { 538c4843a75SGreg Thelen unsigned long flags; 539c4843a75SGreg Thelen 540bd4c8ce4SAndrew Morton if (page->mapping != mapping) 541bd4c8ce4SAndrew Morton return 0; 542bd4c8ce4SAndrew Morton 543266cf658SDavid Howells if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) 544bd4c8ce4SAndrew Morton return 0; 545bd4c8ce4SAndrew Morton 546c4843a75SGreg Thelen spin_lock_irqsave(&mapping->tree_lock, flags); 547bd4c8ce4SAndrew Morton if (PageDirty(page)) 548bd4c8ce4SAndrew Morton goto failed; 549bd4c8ce4SAndrew Morton 550266cf658SDavid Howells BUG_ON(page_has_private(page)); 55162cccb8cSJohannes Weiner __delete_from_page_cache(page, NULL); 552c4843a75SGreg Thelen spin_unlock_irqrestore(&mapping->tree_lock, flags); 5536072d13cSLinus Torvalds 5546072d13cSLinus Torvalds if (mapping->a_ops->freepage) 5556072d13cSLinus Torvalds mapping->a_ops->freepage(page); 5566072d13cSLinus Torvalds 55709cbfeafSKirill A. Shutemov put_page(page); /* pagecache ref */ 558bd4c8ce4SAndrew Morton return 1; 559bd4c8ce4SAndrew Morton failed: 560c4843a75SGreg Thelen spin_unlock_irqrestore(&mapping->tree_lock, flags); 561bd4c8ce4SAndrew Morton return 0; 562bd4c8ce4SAndrew Morton } 563bd4c8ce4SAndrew Morton 564e3db7691STrond Myklebust static int do_launder_page(struct address_space *mapping, struct page *page) 565e3db7691STrond Myklebust { 566e3db7691STrond Myklebust if (!PageDirty(page)) 567e3db7691STrond Myklebust return 0; 568e3db7691STrond Myklebust if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) 569e3db7691STrond Myklebust return 0; 570e3db7691STrond Myklebust return mapping->a_ops->launder_page(page); 571e3db7691STrond Myklebust } 572e3db7691STrond Myklebust 5731da177e4SLinus Torvalds /** 5741da177e4SLinus Torvalds * invalidate_inode_pages2_range - remove range of pages from an address_space 57567be2dd1SMartin Waitz * @mapping: the address_space 5761da177e4SLinus Torvalds * @start: the page offset 'from' which to invalidate 5771da177e4SLinus Torvalds * @end: the page offset 'to' which to invalidate (inclusive) 5781da177e4SLinus Torvalds * 5791da177e4SLinus Torvalds * Any pages which are found to be mapped into pagetables are unmapped prior to 5801da177e4SLinus Torvalds * invalidation. 5811da177e4SLinus Torvalds * 5826ccfa806SHisashi Hifumi * Returns -EBUSY if any pages could not be invalidated. 5831da177e4SLinus Torvalds */ 5841da177e4SLinus Torvalds int invalidate_inode_pages2_range(struct address_space *mapping, 5851da177e4SLinus Torvalds pgoff_t start, pgoff_t end) 5861da177e4SLinus Torvalds { 5870cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 5881da177e4SLinus Torvalds struct pagevec pvec; 589b85e0effSHugh Dickins pgoff_t index; 5901da177e4SLinus Torvalds int i; 5911da177e4SLinus Torvalds int ret = 0; 5920dd1334fSHisashi Hifumi int ret2 = 0; 5931da177e4SLinus Torvalds int did_range_unmap = 0; 5941da177e4SLinus Torvalds 5953167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 5961da177e4SLinus Torvalds pagevec_init(&pvec, 0); 597b85e0effSHugh Dickins index = start; 5980cd6144aSJohannes Weiner while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, 5990cd6144aSJohannes Weiner min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, 6000cd6144aSJohannes Weiner indices)) { 6017b965e08STrond Myklebust for (i = 0; i < pagevec_count(&pvec); i++) { 6021da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 603b85e0effSHugh Dickins 604b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 6050cd6144aSJohannes Weiner index = indices[i]; 606b85e0effSHugh Dickins if (index > end) 607b85e0effSHugh Dickins break; 6081da177e4SLinus Torvalds 6090cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 6100cd6144aSJohannes Weiner clear_exceptional_entry(mapping, index, page); 6110cd6144aSJohannes Weiner continue; 6120cd6144aSJohannes Weiner } 6130cd6144aSJohannes Weiner 6141da177e4SLinus Torvalds lock_page(page); 615*5cbc198aSKirill A. Shutemov WARN_ON(page_to_index(page) != index); 6161da177e4SLinus Torvalds if (page->mapping != mapping) { 6171da177e4SLinus Torvalds unlock_page(page); 6181da177e4SLinus Torvalds continue; 6191da177e4SLinus Torvalds } 6201da177e4SLinus Torvalds wait_on_page_writeback(page); 621d00806b1SNick Piggin if (page_mapped(page)) { 6221da177e4SLinus Torvalds if (!did_range_unmap) { 6231da177e4SLinus Torvalds /* 6241da177e4SLinus Torvalds * Zap the rest of the file in one hit. 6251da177e4SLinus Torvalds */ 6261da177e4SLinus Torvalds unmap_mapping_range(mapping, 62709cbfeafSKirill A. Shutemov (loff_t)index << PAGE_SHIFT, 628b85e0effSHugh Dickins (loff_t)(1 + end - index) 62909cbfeafSKirill A. Shutemov << PAGE_SHIFT, 6301da177e4SLinus Torvalds 0); 6311da177e4SLinus Torvalds did_range_unmap = 1; 6321da177e4SLinus Torvalds } else { 6331da177e4SLinus Torvalds /* 6341da177e4SLinus Torvalds * Just zap this page 6351da177e4SLinus Torvalds */ 6361da177e4SLinus Torvalds unmap_mapping_range(mapping, 63709cbfeafSKirill A. Shutemov (loff_t)index << PAGE_SHIFT, 63809cbfeafSKirill A. Shutemov PAGE_SIZE, 0); 6391da177e4SLinus Torvalds } 6401da177e4SLinus Torvalds } 641d00806b1SNick Piggin BUG_ON(page_mapped(page)); 6420dd1334fSHisashi Hifumi ret2 = do_launder_page(mapping, page); 6430dd1334fSHisashi Hifumi if (ret2 == 0) { 6440dd1334fSHisashi Hifumi if (!invalidate_complete_page2(mapping, page)) 6456ccfa806SHisashi Hifumi ret2 = -EBUSY; 6460dd1334fSHisashi Hifumi } 6470dd1334fSHisashi Hifumi if (ret2 < 0) 6480dd1334fSHisashi Hifumi ret = ret2; 6491da177e4SLinus Torvalds unlock_page(page); 6501da177e4SLinus Torvalds } 6510cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 6521da177e4SLinus Torvalds pagevec_release(&pvec); 6531da177e4SLinus Torvalds cond_resched(); 654b85e0effSHugh Dickins index++; 6551da177e4SLinus Torvalds } 6563167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 6571da177e4SLinus Torvalds return ret; 6581da177e4SLinus Torvalds } 6591da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); 6601da177e4SLinus Torvalds 6611da177e4SLinus Torvalds /** 6621da177e4SLinus Torvalds * invalidate_inode_pages2 - remove all pages from an address_space 66367be2dd1SMartin Waitz * @mapping: the address_space 6641da177e4SLinus Torvalds * 6651da177e4SLinus Torvalds * Any pages which are found to be mapped into pagetables are unmapped prior to 6661da177e4SLinus Torvalds * invalidation. 6671da177e4SLinus Torvalds * 668e9de25ddSPeng Tao * Returns -EBUSY if any pages could not be invalidated. 6691da177e4SLinus Torvalds */ 6701da177e4SLinus Torvalds int invalidate_inode_pages2(struct address_space *mapping) 6711da177e4SLinus Torvalds { 6721da177e4SLinus Torvalds return invalidate_inode_pages2_range(mapping, 0, -1); 6731da177e4SLinus Torvalds } 6741da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(invalidate_inode_pages2); 67525d9e2d1Snpiggin@suse.de 67625d9e2d1Snpiggin@suse.de /** 67725d9e2d1Snpiggin@suse.de * truncate_pagecache - unmap and remove pagecache that has been truncated 67825d9e2d1Snpiggin@suse.de * @inode: inode 6798a549beaSHugh Dickins * @newsize: new file size 68025d9e2d1Snpiggin@suse.de * 68125d9e2d1Snpiggin@suse.de * inode's new i_size must already be written before truncate_pagecache 68225d9e2d1Snpiggin@suse.de * is called. 68325d9e2d1Snpiggin@suse.de * 68425d9e2d1Snpiggin@suse.de * This function should typically be called before the filesystem 68525d9e2d1Snpiggin@suse.de * releases resources associated with the freed range (eg. deallocates 68625d9e2d1Snpiggin@suse.de * blocks). This way, pagecache will always stay logically coherent 68725d9e2d1Snpiggin@suse.de * with on-disk format, and the filesystem would not have to deal with 68825d9e2d1Snpiggin@suse.de * situations such as writepage being called for a page that has already 68925d9e2d1Snpiggin@suse.de * had its underlying blocks deallocated. 69025d9e2d1Snpiggin@suse.de */ 6917caef267SKirill A. Shutemov void truncate_pagecache(struct inode *inode, loff_t newsize) 69225d9e2d1Snpiggin@suse.de { 69325d9e2d1Snpiggin@suse.de struct address_space *mapping = inode->i_mapping; 6948a549beaSHugh Dickins loff_t holebegin = round_up(newsize, PAGE_SIZE); 69525d9e2d1Snpiggin@suse.de 69625d9e2d1Snpiggin@suse.de /* 69725d9e2d1Snpiggin@suse.de * unmap_mapping_range is called twice, first simply for 69825d9e2d1Snpiggin@suse.de * efficiency so that truncate_inode_pages does fewer 69925d9e2d1Snpiggin@suse.de * single-page unmaps. However after this first call, and 70025d9e2d1Snpiggin@suse.de * before truncate_inode_pages finishes, it is possible for 70125d9e2d1Snpiggin@suse.de * private pages to be COWed, which remain after 70225d9e2d1Snpiggin@suse.de * truncate_inode_pages finishes, hence the second 70325d9e2d1Snpiggin@suse.de * unmap_mapping_range call must be made for correctness. 70425d9e2d1Snpiggin@suse.de */ 7058a549beaSHugh Dickins unmap_mapping_range(mapping, holebegin, 0, 1); 7068a549beaSHugh Dickins truncate_inode_pages(mapping, newsize); 7078a549beaSHugh Dickins unmap_mapping_range(mapping, holebegin, 0, 1); 70825d9e2d1Snpiggin@suse.de } 70925d9e2d1Snpiggin@suse.de EXPORT_SYMBOL(truncate_pagecache); 71025d9e2d1Snpiggin@suse.de 71125d9e2d1Snpiggin@suse.de /** 7122c27c65eSChristoph Hellwig * truncate_setsize - update inode and pagecache for a new file size 7132c27c65eSChristoph Hellwig * @inode: inode 7142c27c65eSChristoph Hellwig * @newsize: new file size 7152c27c65eSChristoph Hellwig * 716382e27daSJan Kara * truncate_setsize updates i_size and performs pagecache truncation (if 717382e27daSJan Kara * necessary) to @newsize. It will be typically be called from the filesystem's 718382e27daSJan Kara * setattr function when ATTR_SIZE is passed in. 7192c27c65eSChristoph Hellwig * 72077783d06SJan Kara * Must be called with a lock serializing truncates and writes (generally 72177783d06SJan Kara * i_mutex but e.g. xfs uses a different lock) and before all filesystem 72277783d06SJan Kara * specific block truncation has been performed. 7232c27c65eSChristoph Hellwig */ 7242c27c65eSChristoph Hellwig void truncate_setsize(struct inode *inode, loff_t newsize) 7252c27c65eSChristoph Hellwig { 72690a80202SJan Kara loff_t oldsize = inode->i_size; 72790a80202SJan Kara 7282c27c65eSChristoph Hellwig i_size_write(inode, newsize); 72990a80202SJan Kara if (newsize > oldsize) 73090a80202SJan Kara pagecache_isize_extended(inode, oldsize, newsize); 7317caef267SKirill A. Shutemov truncate_pagecache(inode, newsize); 7322c27c65eSChristoph Hellwig } 7332c27c65eSChristoph Hellwig EXPORT_SYMBOL(truncate_setsize); 7342c27c65eSChristoph Hellwig 7352c27c65eSChristoph Hellwig /** 73690a80202SJan Kara * pagecache_isize_extended - update pagecache after extension of i_size 73790a80202SJan Kara * @inode: inode for which i_size was extended 73890a80202SJan Kara * @from: original inode size 73990a80202SJan Kara * @to: new inode size 74090a80202SJan Kara * 74190a80202SJan Kara * Handle extension of inode size either caused by extending truncate or by 74290a80202SJan Kara * write starting after current i_size. We mark the page straddling current 74390a80202SJan Kara * i_size RO so that page_mkwrite() is called on the nearest write access to 74490a80202SJan Kara * the page. This way filesystem can be sure that page_mkwrite() is called on 74590a80202SJan Kara * the page before user writes to the page via mmap after the i_size has been 74690a80202SJan Kara * changed. 74790a80202SJan Kara * 74890a80202SJan Kara * The function must be called after i_size is updated so that page fault 74990a80202SJan Kara * coming after we unlock the page will already see the new i_size. 75090a80202SJan Kara * The function must be called while we still hold i_mutex - this not only 75190a80202SJan Kara * makes sure i_size is stable but also that userspace cannot observe new 75290a80202SJan Kara * i_size value before we are prepared to store mmap writes at new inode size. 75390a80202SJan Kara */ 75490a80202SJan Kara void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) 75590a80202SJan Kara { 75690a80202SJan Kara int bsize = 1 << inode->i_blkbits; 75790a80202SJan Kara loff_t rounded_from; 75890a80202SJan Kara struct page *page; 75990a80202SJan Kara pgoff_t index; 76090a80202SJan Kara 76190a80202SJan Kara WARN_ON(to > inode->i_size); 76290a80202SJan Kara 76309cbfeafSKirill A. Shutemov if (from >= to || bsize == PAGE_SIZE) 76490a80202SJan Kara return; 76590a80202SJan Kara /* Page straddling @from will not have any hole block created? */ 76690a80202SJan Kara rounded_from = round_up(from, bsize); 76709cbfeafSKirill A. Shutemov if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1))) 76890a80202SJan Kara return; 76990a80202SJan Kara 77009cbfeafSKirill A. Shutemov index = from >> PAGE_SHIFT; 77190a80202SJan Kara page = find_lock_page(inode->i_mapping, index); 77290a80202SJan Kara /* Page not cached? Nothing to do */ 77390a80202SJan Kara if (!page) 77490a80202SJan Kara return; 77590a80202SJan Kara /* 77690a80202SJan Kara * See clear_page_dirty_for_io() for details why set_page_dirty() 77790a80202SJan Kara * is needed. 77890a80202SJan Kara */ 77990a80202SJan Kara if (page_mkclean(page)) 78090a80202SJan Kara set_page_dirty(page); 78190a80202SJan Kara unlock_page(page); 78209cbfeafSKirill A. Shutemov put_page(page); 78390a80202SJan Kara } 78490a80202SJan Kara EXPORT_SYMBOL(pagecache_isize_extended); 78590a80202SJan Kara 78690a80202SJan Kara /** 787623e3db9SHugh Dickins * truncate_pagecache_range - unmap and remove pagecache that is hole-punched 788623e3db9SHugh Dickins * @inode: inode 789623e3db9SHugh Dickins * @lstart: offset of beginning of hole 790623e3db9SHugh Dickins * @lend: offset of last byte of hole 791623e3db9SHugh Dickins * 792623e3db9SHugh Dickins * This function should typically be called before the filesystem 793623e3db9SHugh Dickins * releases resources associated with the freed range (eg. deallocates 794623e3db9SHugh Dickins * blocks). This way, pagecache will always stay logically coherent 795623e3db9SHugh Dickins * with on-disk format, and the filesystem would not have to deal with 796623e3db9SHugh Dickins * situations such as writepage being called for a page that has already 797623e3db9SHugh Dickins * had its underlying blocks deallocated. 798623e3db9SHugh Dickins */ 799623e3db9SHugh Dickins void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) 800623e3db9SHugh Dickins { 801623e3db9SHugh Dickins struct address_space *mapping = inode->i_mapping; 802623e3db9SHugh Dickins loff_t unmap_start = round_up(lstart, PAGE_SIZE); 803623e3db9SHugh Dickins loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; 804623e3db9SHugh Dickins /* 805623e3db9SHugh Dickins * This rounding is currently just for example: unmap_mapping_range 806623e3db9SHugh Dickins * expands its hole outwards, whereas we want it to contract the hole 807623e3db9SHugh Dickins * inwards. However, existing callers of truncate_pagecache_range are 8085a720394SLukas Czerner * doing their own page rounding first. Note that unmap_mapping_range 8095a720394SLukas Czerner * allows holelen 0 for all, and we allow lend -1 for end of file. 810623e3db9SHugh Dickins */ 811623e3db9SHugh Dickins 812623e3db9SHugh Dickins /* 813623e3db9SHugh Dickins * Unlike in truncate_pagecache, unmap_mapping_range is called only 814623e3db9SHugh Dickins * once (before truncating pagecache), and without "even_cows" flag: 815623e3db9SHugh Dickins * hole-punching should not remove private COWed pages from the hole. 816623e3db9SHugh Dickins */ 817623e3db9SHugh Dickins if ((u64)unmap_end > (u64)unmap_start) 818623e3db9SHugh Dickins unmap_mapping_range(mapping, unmap_start, 819623e3db9SHugh Dickins 1 + unmap_end - unmap_start, 0); 820623e3db9SHugh Dickins truncate_inode_pages_range(mapping, lstart, lend); 821623e3db9SHugh Dickins } 822623e3db9SHugh Dickins EXPORT_SYMBOL(truncate_pagecache_range); 823