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 */ 233a4f8a0bSHugh Dickins #include <linux/shmem_fs.h> 24c515e1fdSDan Magenheimer #include <linux/cleancache.h> 2590a80202SJan Kara #include <linux/rmap.h> 26ba470de4SRik van Riel #include "internal.h" 271da177e4SLinus Torvalds 28c6dcf52cSJan Kara static void clear_shadow_entry(struct address_space *mapping, pgoff_t index, 29c6dcf52cSJan Kara void *entry) 300cd6144aSJohannes Weiner { 31449dd698SJohannes Weiner struct radix_tree_node *node; 32449dd698SJohannes Weiner void **slot; 33449dd698SJohannes Weiner 34ac401cc7SJan Kara spin_lock_irq(&mapping->tree_lock); 350cd6144aSJohannes Weiner /* 360cd6144aSJohannes Weiner * Regular page slots are stabilized by the page lock even 370cd6144aSJohannes Weiner * without the tree itself locked. These unlocked entries 380cd6144aSJohannes Weiner * need verification under the tree lock. 390cd6144aSJohannes Weiner */ 4014b46879SJohannes Weiner if (!__radix_tree_lookup(&mapping->page_tree, index, &node, &slot)) 41449dd698SJohannes Weiner goto unlock; 42449dd698SJohannes Weiner if (*slot != entry) 43449dd698SJohannes Weiner goto unlock; 4414b46879SJohannes Weiner __radix_tree_replace(&mapping->page_tree, node, slot, NULL, 4514b46879SJohannes Weiner workingset_update_node, mapping); 46f9fe48beSRoss Zwisler mapping->nrexceptional--; 47449dd698SJohannes Weiner unlock: 480cd6144aSJohannes Weiner spin_unlock_irq(&mapping->tree_lock); 490cd6144aSJohannes Weiner } 501da177e4SLinus Torvalds 51c6dcf52cSJan Kara /* 52c6dcf52cSJan Kara * Unconditionally remove exceptional entry. Usually called from truncate path. 53c6dcf52cSJan Kara */ 54c6dcf52cSJan Kara static void truncate_exceptional_entry(struct address_space *mapping, 55c6dcf52cSJan Kara pgoff_t index, void *entry) 56c6dcf52cSJan Kara { 57c6dcf52cSJan Kara /* Handled by shmem itself */ 58c6dcf52cSJan Kara if (shmem_mapping(mapping)) 59c6dcf52cSJan Kara return; 60c6dcf52cSJan Kara 61c6dcf52cSJan Kara if (dax_mapping(mapping)) { 62c6dcf52cSJan Kara dax_delete_mapping_entry(mapping, index); 63c6dcf52cSJan Kara return; 64c6dcf52cSJan Kara } 65c6dcf52cSJan Kara clear_shadow_entry(mapping, index, entry); 66c6dcf52cSJan Kara } 67c6dcf52cSJan Kara 68c6dcf52cSJan Kara /* 69c6dcf52cSJan Kara * Invalidate exceptional entry if easily possible. This handles exceptional 70c6dcf52cSJan Kara * entries for invalidate_inode_pages() so for DAX it evicts only unlocked and 71c6dcf52cSJan Kara * clean entries. 72c6dcf52cSJan Kara */ 73c6dcf52cSJan Kara static int invalidate_exceptional_entry(struct address_space *mapping, 74c6dcf52cSJan Kara pgoff_t index, void *entry) 75c6dcf52cSJan Kara { 76c6dcf52cSJan Kara /* Handled by shmem itself */ 77c6dcf52cSJan Kara if (shmem_mapping(mapping)) 78c6dcf52cSJan Kara return 1; 79c6dcf52cSJan Kara if (dax_mapping(mapping)) 80c6dcf52cSJan Kara return dax_invalidate_mapping_entry(mapping, index); 81c6dcf52cSJan Kara clear_shadow_entry(mapping, index, entry); 82c6dcf52cSJan Kara return 1; 83c6dcf52cSJan Kara } 84c6dcf52cSJan Kara 85c6dcf52cSJan Kara /* 86c6dcf52cSJan Kara * Invalidate exceptional entry if clean. This handles exceptional entries for 87c6dcf52cSJan Kara * invalidate_inode_pages2() so for DAX it evicts only clean entries. 88c6dcf52cSJan Kara */ 89c6dcf52cSJan Kara static int invalidate_exceptional_entry2(struct address_space *mapping, 90c6dcf52cSJan Kara pgoff_t index, void *entry) 91c6dcf52cSJan Kara { 92c6dcf52cSJan Kara /* Handled by shmem itself */ 93c6dcf52cSJan Kara if (shmem_mapping(mapping)) 94c6dcf52cSJan Kara return 1; 95c6dcf52cSJan Kara if (dax_mapping(mapping)) 96c6dcf52cSJan Kara return dax_invalidate_mapping_entry_sync(mapping, index); 97c6dcf52cSJan Kara clear_shadow_entry(mapping, index, entry); 98c6dcf52cSJan Kara return 1; 99c6dcf52cSJan Kara } 100c6dcf52cSJan Kara 101cf9a2ae8SDavid Howells /** 10228bc44d7SFengguang Wu * do_invalidatepage - invalidate part or all of a page 103cf9a2ae8SDavid Howells * @page: the page which is affected 104d47992f8SLukas Czerner * @offset: start of the range to invalidate 105d47992f8SLukas Czerner * @length: length of the range to invalidate 106cf9a2ae8SDavid Howells * 107cf9a2ae8SDavid Howells * do_invalidatepage() is called when all or part of the page has become 108cf9a2ae8SDavid Howells * invalidated by a truncate operation. 109cf9a2ae8SDavid Howells * 110cf9a2ae8SDavid Howells * do_invalidatepage() does not have to release all buffers, but it must 111cf9a2ae8SDavid Howells * ensure that no dirty buffer is left outside @offset and that no I/O 112cf9a2ae8SDavid Howells * is underway against any of the blocks which are outside the truncation 113cf9a2ae8SDavid Howells * point. Because the caller is about to free (and possibly reuse) those 114cf9a2ae8SDavid Howells * blocks on-disk. 115cf9a2ae8SDavid Howells */ 116d47992f8SLukas Czerner void do_invalidatepage(struct page *page, unsigned int offset, 117d47992f8SLukas Czerner unsigned int length) 118cf9a2ae8SDavid Howells { 119d47992f8SLukas Czerner void (*invalidatepage)(struct page *, unsigned int, unsigned int); 120d47992f8SLukas Czerner 121cf9a2ae8SDavid Howells invalidatepage = page->mapping->a_ops->invalidatepage; 1229361401eSDavid Howells #ifdef CONFIG_BLOCK 123cf9a2ae8SDavid Howells if (!invalidatepage) 124cf9a2ae8SDavid Howells invalidatepage = block_invalidatepage; 1259361401eSDavid Howells #endif 126cf9a2ae8SDavid Howells if (invalidatepage) 127d47992f8SLukas Czerner (*invalidatepage)(page, offset, length); 128cf9a2ae8SDavid Howells } 129cf9a2ae8SDavid Howells 130ecdfc978SLinus Torvalds /* 1311da177e4SLinus Torvalds * If truncate cannot remove the fs-private metadata from the page, the page 13262e1c553SShaohua Li * becomes orphaned. It will be left on the LRU and may even be mapped into 13354cb8821SNick Piggin * user pagetables if we're racing with filemap_fault(). 1341da177e4SLinus Torvalds * 1351da177e4SLinus Torvalds * We need to bale out if page->mapping is no longer equal to the original 1361da177e4SLinus Torvalds * mapping. This happens a) when the VM reclaimed the page while we waited on 137fc0ecff6SAndrew Morton * its lock, b) when a concurrent invalidate_mapping_pages got there first and 1381da177e4SLinus Torvalds * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. 1391da177e4SLinus Torvalds */ 140750b4987SNick Piggin static int 1411da177e4SLinus Torvalds truncate_complete_page(struct address_space *mapping, struct page *page) 1421da177e4SLinus Torvalds { 1431da177e4SLinus Torvalds if (page->mapping != mapping) 144750b4987SNick Piggin return -EIO; 1451da177e4SLinus Torvalds 146266cf658SDavid Howells if (page_has_private(page)) 14709cbfeafSKirill A. Shutemov do_invalidatepage(page, 0, PAGE_SIZE); 1481da177e4SLinus Torvalds 149b9ea2515SKonstantin Khlebnikov /* 150b9ea2515SKonstantin Khlebnikov * Some filesystems seem to re-dirty the page even after 151b9ea2515SKonstantin Khlebnikov * the VM has canceled the dirty bit (eg ext3 journaling). 152b9ea2515SKonstantin Khlebnikov * Hence dirty accounting check is placed after invalidation. 153b9ea2515SKonstantin Khlebnikov */ 15411f81becSTejun Heo cancel_dirty_page(page); 1551da177e4SLinus Torvalds ClearPageMappedToDisk(page); 1565adc7b51SMinchan Kim delete_from_page_cache(page); 157750b4987SNick Piggin return 0; 1581da177e4SLinus Torvalds } 1591da177e4SLinus Torvalds 1601da177e4SLinus Torvalds /* 161fc0ecff6SAndrew Morton * This is for invalidate_mapping_pages(). That function can be called at 1621da177e4SLinus Torvalds * any time, and is not supposed to throw away dirty pages. But pages can 1630fd0e6b0SNick Piggin * be marked dirty at any time too, so use remove_mapping which safely 1640fd0e6b0SNick Piggin * discards clean, unused pages. 1651da177e4SLinus Torvalds * 1661da177e4SLinus Torvalds * Returns non-zero if the page was successfully invalidated. 1671da177e4SLinus Torvalds */ 1681da177e4SLinus Torvalds static int 1691da177e4SLinus Torvalds invalidate_complete_page(struct address_space *mapping, struct page *page) 1701da177e4SLinus Torvalds { 1710fd0e6b0SNick Piggin int ret; 1720fd0e6b0SNick Piggin 1731da177e4SLinus Torvalds if (page->mapping != mapping) 1741da177e4SLinus Torvalds return 0; 1751da177e4SLinus Torvalds 176266cf658SDavid Howells if (page_has_private(page) && !try_to_release_page(page, 0)) 1771da177e4SLinus Torvalds return 0; 1781da177e4SLinus Torvalds 1790fd0e6b0SNick Piggin ret = remove_mapping(mapping, page); 1800fd0e6b0SNick Piggin 1810fd0e6b0SNick Piggin return ret; 1821da177e4SLinus Torvalds } 1831da177e4SLinus Torvalds 184750b4987SNick Piggin int truncate_inode_page(struct address_space *mapping, struct page *page) 185750b4987SNick Piggin { 186fc127da0SKirill A. Shutemov loff_t holelen; 187fc127da0SKirill A. Shutemov VM_BUG_ON_PAGE(PageTail(page), page); 188fc127da0SKirill A. Shutemov 189fc127da0SKirill A. Shutemov holelen = PageTransHuge(page) ? HPAGE_PMD_SIZE : PAGE_SIZE; 190750b4987SNick Piggin if (page_mapped(page)) { 191750b4987SNick Piggin unmap_mapping_range(mapping, 19209cbfeafSKirill A. Shutemov (loff_t)page->index << PAGE_SHIFT, 193fc127da0SKirill A. Shutemov holelen, 0); 194750b4987SNick Piggin } 195750b4987SNick Piggin return truncate_complete_page(mapping, page); 196750b4987SNick Piggin } 197750b4987SNick Piggin 19883f78668SWu Fengguang /* 19925718736SAndi Kleen * Used to get rid of pages on hardware memory corruption. 20025718736SAndi Kleen */ 20125718736SAndi Kleen int generic_error_remove_page(struct address_space *mapping, struct page *page) 20225718736SAndi Kleen { 20325718736SAndi Kleen if (!mapping) 20425718736SAndi Kleen return -EINVAL; 20525718736SAndi Kleen /* 20625718736SAndi Kleen * Only punch for normal data pages for now. 20725718736SAndi Kleen * Handling other types like directories would need more auditing. 20825718736SAndi Kleen */ 20925718736SAndi Kleen if (!S_ISREG(mapping->host->i_mode)) 21025718736SAndi Kleen return -EIO; 21125718736SAndi Kleen return truncate_inode_page(mapping, page); 21225718736SAndi Kleen } 21325718736SAndi Kleen EXPORT_SYMBOL(generic_error_remove_page); 21425718736SAndi Kleen 21525718736SAndi Kleen /* 21683f78668SWu Fengguang * Safely invalidate one page from its pagecache mapping. 21783f78668SWu Fengguang * It only drops clean, unused pages. The page must be locked. 21883f78668SWu Fengguang * 21983f78668SWu Fengguang * Returns 1 if the page is successfully invalidated, otherwise 0. 22083f78668SWu Fengguang */ 22183f78668SWu Fengguang int invalidate_inode_page(struct page *page) 22283f78668SWu Fengguang { 22383f78668SWu Fengguang struct address_space *mapping = page_mapping(page); 22483f78668SWu Fengguang if (!mapping) 22583f78668SWu Fengguang return 0; 22683f78668SWu Fengguang if (PageDirty(page) || PageWriteback(page)) 22783f78668SWu Fengguang return 0; 22883f78668SWu Fengguang if (page_mapped(page)) 22983f78668SWu Fengguang return 0; 23083f78668SWu Fengguang return invalidate_complete_page(mapping, page); 23183f78668SWu Fengguang } 23283f78668SWu Fengguang 2331da177e4SLinus Torvalds /** 23473c1e204SLiu Bo * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets 2351da177e4SLinus Torvalds * @mapping: mapping to truncate 2361da177e4SLinus Torvalds * @lstart: offset from which to truncate 2375a720394SLukas Czerner * @lend: offset to which to truncate (inclusive) 2381da177e4SLinus Torvalds * 239d7339071SHans Reiser * Truncate the page cache, removing the pages that are between 2405a720394SLukas Czerner * specified offsets (and zeroing out partial pages 2415a720394SLukas Czerner * if lstart or lend + 1 is not page aligned). 2421da177e4SLinus Torvalds * 2431da177e4SLinus Torvalds * Truncate takes two passes - the first pass is nonblocking. It will not 2441da177e4SLinus Torvalds * block on page locks and it will not block on writeback. The second pass 2451da177e4SLinus Torvalds * will wait. This is to prevent as much IO as possible in the affected region. 2461da177e4SLinus Torvalds * The first pass will remove most pages, so the search cost of the second pass 2471da177e4SLinus Torvalds * is low. 2481da177e4SLinus Torvalds * 2491da177e4SLinus Torvalds * We pass down the cache-hot hint to the page freeing code. Even if the 2501da177e4SLinus Torvalds * mapping is large, it is probably the case that the final pages are the most 2511da177e4SLinus Torvalds * recently touched, and freeing happens in ascending file offset order. 2525a720394SLukas Czerner * 2535a720394SLukas Czerner * Note that since ->invalidatepage() accepts range to invalidate 2545a720394SLukas Czerner * truncate_inode_pages_range is able to handle cases where lend + 1 is not 2555a720394SLukas Czerner * page aligned properly. 2561da177e4SLinus Torvalds */ 257d7339071SHans Reiser void truncate_inode_pages_range(struct address_space *mapping, 258d7339071SHans Reiser loff_t lstart, loff_t lend) 2591da177e4SLinus Torvalds { 2605a720394SLukas Czerner pgoff_t start; /* inclusive */ 2615a720394SLukas Czerner pgoff_t end; /* exclusive */ 2625a720394SLukas Czerner unsigned int partial_start; /* inclusive */ 2635a720394SLukas Czerner unsigned int partial_end; /* exclusive */ 2641da177e4SLinus Torvalds struct pagevec pvec; 2650cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 266b85e0effSHugh Dickins pgoff_t index; 2671da177e4SLinus Torvalds int i; 2681da177e4SLinus Torvalds 2693167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 270f9fe48beSRoss Zwisler if (mapping->nrpages == 0 && mapping->nrexceptional == 0) 2711da177e4SLinus Torvalds return; 2721da177e4SLinus Torvalds 2735a720394SLukas Czerner /* Offsets within partial pages */ 27409cbfeafSKirill A. Shutemov partial_start = lstart & (PAGE_SIZE - 1); 27509cbfeafSKirill A. Shutemov partial_end = (lend + 1) & (PAGE_SIZE - 1); 2765a720394SLukas Czerner 2775a720394SLukas Czerner /* 2785a720394SLukas Czerner * 'start' and 'end' always covers the range of pages to be fully 2795a720394SLukas Czerner * truncated. Partial pages are covered with 'partial_start' at the 2805a720394SLukas Czerner * start of the range and 'partial_end' at the end of the range. 2815a720394SLukas Czerner * Note that 'end' is exclusive while 'lend' is inclusive. 2825a720394SLukas Czerner */ 28309cbfeafSKirill A. Shutemov start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; 2845a720394SLukas Czerner if (lend == -1) 2855a720394SLukas Czerner /* 2865a720394SLukas Czerner * lend == -1 indicates end-of-file so we have to set 'end' 2875a720394SLukas Czerner * to the highest possible pgoff_t and since the type is 2885a720394SLukas Czerner * unsigned we're using -1. 2895a720394SLukas Czerner */ 2905a720394SLukas Czerner end = -1; 2915a720394SLukas Czerner else 29209cbfeafSKirill A. Shutemov end = (lend + 1) >> PAGE_SHIFT; 293d7339071SHans Reiser 2941da177e4SLinus Torvalds pagevec_init(&pvec, 0); 295b85e0effSHugh Dickins index = start; 2960cd6144aSJohannes Weiner while (index < end && pagevec_lookup_entries(&pvec, mapping, index, 2970cd6144aSJohannes Weiner min(end - index, (pgoff_t)PAGEVEC_SIZE), 2980cd6144aSJohannes Weiner indices)) { 2991da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 3001da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 3011da177e4SLinus Torvalds 302b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 3030cd6144aSJohannes Weiner index = indices[i]; 3045a720394SLukas Czerner if (index >= end) 305d7339071SHans Reiser break; 306d7339071SHans Reiser 3070cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 308c6dcf52cSJan Kara truncate_exceptional_entry(mapping, index, 309c6dcf52cSJan Kara page); 3100cd6144aSJohannes Weiner continue; 3110cd6144aSJohannes Weiner } 3120cd6144aSJohannes Weiner 313529ae9aaSNick Piggin if (!trylock_page(page)) 3141da177e4SLinus Torvalds continue; 3155cbc198aSKirill A. Shutemov WARN_ON(page_to_index(page) != index); 3161da177e4SLinus Torvalds if (PageWriteback(page)) { 3171da177e4SLinus Torvalds unlock_page(page); 3181da177e4SLinus Torvalds continue; 3191da177e4SLinus Torvalds } 320750b4987SNick Piggin truncate_inode_page(mapping, page); 3211da177e4SLinus Torvalds unlock_page(page); 3221da177e4SLinus Torvalds } 3230cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 3241da177e4SLinus Torvalds pagevec_release(&pvec); 3251da177e4SLinus Torvalds cond_resched(); 326b85e0effSHugh Dickins index++; 3271da177e4SLinus Torvalds } 3281da177e4SLinus Torvalds 3295a720394SLukas Czerner if (partial_start) { 3301da177e4SLinus Torvalds struct page *page = find_lock_page(mapping, start - 1); 3311da177e4SLinus Torvalds if (page) { 33209cbfeafSKirill A. Shutemov unsigned int top = PAGE_SIZE; 3335a720394SLukas Czerner if (start > end) { 3345a720394SLukas Czerner /* Truncation within a single page */ 3355a720394SLukas Czerner top = partial_end; 3365a720394SLukas Czerner partial_end = 0; 3375a720394SLukas Czerner } 3381da177e4SLinus Torvalds wait_on_page_writeback(page); 3395a720394SLukas Czerner zero_user_segment(page, partial_start, top); 3405a720394SLukas Czerner cleancache_invalidate_page(mapping, page); 3415a720394SLukas Czerner if (page_has_private(page)) 3425a720394SLukas Czerner do_invalidatepage(page, partial_start, 3435a720394SLukas Czerner top - partial_start); 3441da177e4SLinus Torvalds unlock_page(page); 34509cbfeafSKirill A. Shutemov put_page(page); 3461da177e4SLinus Torvalds } 3471da177e4SLinus Torvalds } 3485a720394SLukas Czerner if (partial_end) { 3495a720394SLukas Czerner struct page *page = find_lock_page(mapping, end); 3505a720394SLukas Czerner if (page) { 3515a720394SLukas Czerner wait_on_page_writeback(page); 3525a720394SLukas Czerner zero_user_segment(page, 0, partial_end); 3535a720394SLukas Czerner cleancache_invalidate_page(mapping, page); 3545a720394SLukas Czerner if (page_has_private(page)) 3555a720394SLukas Czerner do_invalidatepage(page, 0, 3565a720394SLukas Czerner partial_end); 3575a720394SLukas Czerner unlock_page(page); 35809cbfeafSKirill A. Shutemov put_page(page); 3595a720394SLukas Czerner } 3605a720394SLukas Czerner } 3615a720394SLukas Czerner /* 3625a720394SLukas Czerner * If the truncation happened within a single page no pages 3635a720394SLukas Czerner * will be released, just zeroed, so we can bail out now. 3645a720394SLukas Czerner */ 3655a720394SLukas Czerner if (start >= end) 3665a720394SLukas Czerner return; 3671da177e4SLinus Torvalds 368b85e0effSHugh Dickins index = start; 3691da177e4SLinus Torvalds for ( ; ; ) { 3701da177e4SLinus Torvalds cond_resched(); 3710cd6144aSJohannes Weiner if (!pagevec_lookup_entries(&pvec, mapping, index, 372792ceaefSHugh Dickins min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) { 373792ceaefSHugh Dickins /* If all gone from start onwards, we're done */ 374b85e0effSHugh Dickins if (index == start) 3751da177e4SLinus Torvalds break; 376792ceaefSHugh Dickins /* Otherwise restart to make sure all gone */ 377b85e0effSHugh Dickins index = start; 3781da177e4SLinus Torvalds continue; 3791da177e4SLinus Torvalds } 3800cd6144aSJohannes Weiner if (index == start && indices[0] >= end) { 381792ceaefSHugh Dickins /* All gone out of hole to be punched, we're done */ 3820cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 383d7339071SHans Reiser pagevec_release(&pvec); 384d7339071SHans Reiser break; 385d7339071SHans Reiser } 3861da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 3871da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 3881da177e4SLinus Torvalds 389b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 3900cd6144aSJohannes Weiner index = indices[i]; 391792ceaefSHugh Dickins if (index >= end) { 392792ceaefSHugh Dickins /* Restart punch to make sure all gone */ 393792ceaefSHugh Dickins index = start - 1; 394d7339071SHans Reiser break; 395792ceaefSHugh Dickins } 396b85e0effSHugh Dickins 3970cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 398c6dcf52cSJan Kara truncate_exceptional_entry(mapping, index, 399c6dcf52cSJan Kara page); 4000cd6144aSJohannes Weiner continue; 4010cd6144aSJohannes Weiner } 4020cd6144aSJohannes Weiner 4031da177e4SLinus Torvalds lock_page(page); 4045cbc198aSKirill A. Shutemov WARN_ON(page_to_index(page) != index); 4051da177e4SLinus Torvalds wait_on_page_writeback(page); 406750b4987SNick Piggin truncate_inode_page(mapping, page); 4071da177e4SLinus Torvalds unlock_page(page); 4081da177e4SLinus Torvalds } 4090cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 4101da177e4SLinus Torvalds pagevec_release(&pvec); 411b85e0effSHugh Dickins index++; 4121da177e4SLinus Torvalds } 4133167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 4141da177e4SLinus Torvalds } 415d7339071SHans Reiser EXPORT_SYMBOL(truncate_inode_pages_range); 4161da177e4SLinus Torvalds 417d7339071SHans Reiser /** 418d7339071SHans Reiser * truncate_inode_pages - truncate *all* the pages from an offset 419d7339071SHans Reiser * @mapping: mapping to truncate 420d7339071SHans Reiser * @lstart: offset from which to truncate 421d7339071SHans Reiser * 4221b1dcc1bSJes Sorensen * Called under (and serialised by) inode->i_mutex. 42308142579SJan Kara * 42408142579SJan Kara * Note: When this function returns, there can be a page in the process of 42508142579SJan Kara * deletion (inside __delete_from_page_cache()) in the specified range. Thus 42608142579SJan Kara * mapping->nrpages can be non-zero when this function returns even after 42708142579SJan Kara * truncation of the whole mapping. 428d7339071SHans Reiser */ 429d7339071SHans Reiser void truncate_inode_pages(struct address_space *mapping, loff_t lstart) 430d7339071SHans Reiser { 431d7339071SHans Reiser truncate_inode_pages_range(mapping, lstart, (loff_t)-1); 432d7339071SHans Reiser } 4331da177e4SLinus Torvalds EXPORT_SYMBOL(truncate_inode_pages); 4341da177e4SLinus Torvalds 43528697355SMike Waychison /** 43691b0abe3SJohannes Weiner * truncate_inode_pages_final - truncate *all* pages before inode dies 43791b0abe3SJohannes Weiner * @mapping: mapping to truncate 43891b0abe3SJohannes Weiner * 43991b0abe3SJohannes Weiner * Called under (and serialized by) inode->i_mutex. 44091b0abe3SJohannes Weiner * 44191b0abe3SJohannes Weiner * Filesystems have to use this in the .evict_inode path to inform the 44291b0abe3SJohannes Weiner * VM that this is the final truncate and the inode is going away. 44391b0abe3SJohannes Weiner */ 44491b0abe3SJohannes Weiner void truncate_inode_pages_final(struct address_space *mapping) 44591b0abe3SJohannes Weiner { 446f9fe48beSRoss Zwisler unsigned long nrexceptional; 44791b0abe3SJohannes Weiner unsigned long nrpages; 44891b0abe3SJohannes Weiner 44991b0abe3SJohannes Weiner /* 45091b0abe3SJohannes Weiner * Page reclaim can not participate in regular inode lifetime 45191b0abe3SJohannes Weiner * management (can't call iput()) and thus can race with the 45291b0abe3SJohannes Weiner * inode teardown. Tell it when the address space is exiting, 45391b0abe3SJohannes Weiner * so that it does not install eviction information after the 45491b0abe3SJohannes Weiner * final truncate has begun. 45591b0abe3SJohannes Weiner */ 45691b0abe3SJohannes Weiner mapping_set_exiting(mapping); 45791b0abe3SJohannes Weiner 45891b0abe3SJohannes Weiner /* 45991b0abe3SJohannes Weiner * When reclaim installs eviction entries, it increases 460f9fe48beSRoss Zwisler * nrexceptional first, then decreases nrpages. Make sure we see 46191b0abe3SJohannes Weiner * this in the right order or we might miss an entry. 46291b0abe3SJohannes Weiner */ 46391b0abe3SJohannes Weiner nrpages = mapping->nrpages; 46491b0abe3SJohannes Weiner smp_rmb(); 465f9fe48beSRoss Zwisler nrexceptional = mapping->nrexceptional; 46691b0abe3SJohannes Weiner 467f9fe48beSRoss Zwisler if (nrpages || nrexceptional) { 46891b0abe3SJohannes Weiner /* 46991b0abe3SJohannes Weiner * As truncation uses a lockless tree lookup, cycle 47091b0abe3SJohannes Weiner * the tree lock to make sure any ongoing tree 47191b0abe3SJohannes Weiner * modification that does not see AS_EXITING is 47291b0abe3SJohannes Weiner * completed before starting the final truncate. 47391b0abe3SJohannes Weiner */ 47491b0abe3SJohannes Weiner spin_lock_irq(&mapping->tree_lock); 47591b0abe3SJohannes Weiner spin_unlock_irq(&mapping->tree_lock); 47691b0abe3SJohannes Weiner 47791b0abe3SJohannes Weiner truncate_inode_pages(mapping, 0); 47891b0abe3SJohannes Weiner } 47991b0abe3SJohannes Weiner } 48091b0abe3SJohannes Weiner EXPORT_SYMBOL(truncate_inode_pages_final); 48191b0abe3SJohannes Weiner 48291b0abe3SJohannes Weiner /** 48328697355SMike Waychison * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode 48428697355SMike Waychison * @mapping: the address_space which holds the pages to invalidate 48528697355SMike Waychison * @start: the offset 'from' which to invalidate 48628697355SMike Waychison * @end: the offset 'to' which to invalidate (inclusive) 48728697355SMike Waychison * 48828697355SMike Waychison * This function only removes the unlocked pages, if you want to 48928697355SMike Waychison * remove all the pages of one inode, you must call truncate_inode_pages. 49028697355SMike Waychison * 49128697355SMike Waychison * invalidate_mapping_pages() will not block on IO activity. It will not 49228697355SMike Waychison * invalidate pages which are dirty, locked, under writeback or mapped into 49328697355SMike Waychison * pagetables. 49428697355SMike Waychison */ 49528697355SMike Waychison unsigned long invalidate_mapping_pages(struct address_space *mapping, 49628697355SMike Waychison pgoff_t start, pgoff_t end) 4971da177e4SLinus Torvalds { 4980cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 4991da177e4SLinus Torvalds struct pagevec pvec; 500b85e0effSHugh Dickins pgoff_t index = start; 50131560180SMinchan Kim unsigned long ret; 50231560180SMinchan Kim unsigned long count = 0; 5031da177e4SLinus Torvalds int i; 5041da177e4SLinus Torvalds 5051da177e4SLinus Torvalds pagevec_init(&pvec, 0); 5060cd6144aSJohannes Weiner while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, 5070cd6144aSJohannes Weiner min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, 5080cd6144aSJohannes Weiner indices)) { 5091da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 5101da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 5111da177e4SLinus Torvalds 512b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 5130cd6144aSJohannes Weiner index = indices[i]; 514b85e0effSHugh Dickins if (index > end) 515b85e0effSHugh Dickins break; 516e0f23603SNeilBrown 5170cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 518c6dcf52cSJan Kara invalidate_exceptional_entry(mapping, index, 519c6dcf52cSJan Kara page); 5200cd6144aSJohannes Weiner continue; 5210cd6144aSJohannes Weiner } 5220cd6144aSJohannes Weiner 523b85e0effSHugh Dickins if (!trylock_page(page)) 524b85e0effSHugh Dickins continue; 525fc127da0SKirill A. Shutemov 5265cbc198aSKirill A. Shutemov WARN_ON(page_to_index(page) != index); 527fc127da0SKirill A. Shutemov 528fc127da0SKirill A. Shutemov /* Middle of THP: skip */ 529fc127da0SKirill A. Shutemov if (PageTransTail(page)) { 530fc127da0SKirill A. Shutemov unlock_page(page); 531fc127da0SKirill A. Shutemov continue; 532fc127da0SKirill A. Shutemov } else if (PageTransHuge(page)) { 533fc127da0SKirill A. Shutemov index += HPAGE_PMD_NR - 1; 534fc127da0SKirill A. Shutemov i += HPAGE_PMD_NR - 1; 535fc127da0SKirill A. Shutemov /* 'end' is in the middle of THP */ 536fc127da0SKirill A. Shutemov if (index == round_down(end, HPAGE_PMD_NR)) 537fc127da0SKirill A. Shutemov continue; 538fc127da0SKirill A. Shutemov } 539fc127da0SKirill A. Shutemov 54031560180SMinchan Kim ret = invalidate_inode_page(page); 5411da177e4SLinus Torvalds unlock_page(page); 54231560180SMinchan Kim /* 54331560180SMinchan Kim * Invalidation is a hint that the page is no longer 54431560180SMinchan Kim * of interest and try to speed up its reclaim. 54531560180SMinchan Kim */ 54631560180SMinchan Kim if (!ret) 547cc5993bdSMinchan Kim deactivate_file_page(page); 54831560180SMinchan Kim count += ret; 5491da177e4SLinus Torvalds } 5500cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 5511da177e4SLinus Torvalds pagevec_release(&pvec); 552fc9a07e7SAndrew Morton cond_resched(); 553b85e0effSHugh Dickins index++; 5541da177e4SLinus Torvalds } 55531560180SMinchan Kim return count; 5561da177e4SLinus Torvalds } 55754bc4855SAnton Altaparmakov EXPORT_SYMBOL(invalidate_mapping_pages); 5581da177e4SLinus Torvalds 559bd4c8ce4SAndrew Morton /* 560bd4c8ce4SAndrew Morton * This is like invalidate_complete_page(), except it ignores the page's 561bd4c8ce4SAndrew Morton * refcount. We do this because invalidate_inode_pages2() needs stronger 562bd4c8ce4SAndrew Morton * invalidation guarantees, and cannot afford to leave pages behind because 5632706a1b8SAnderson Briglia * shrink_page_list() has a temp ref on them, or because they're transiently 5642706a1b8SAnderson Briglia * sitting in the lru_cache_add() pagevecs. 565bd4c8ce4SAndrew Morton */ 566bd4c8ce4SAndrew Morton static int 567bd4c8ce4SAndrew Morton invalidate_complete_page2(struct address_space *mapping, struct page *page) 568bd4c8ce4SAndrew Morton { 569c4843a75SGreg Thelen unsigned long flags; 570c4843a75SGreg Thelen 571bd4c8ce4SAndrew Morton if (page->mapping != mapping) 572bd4c8ce4SAndrew Morton return 0; 573bd4c8ce4SAndrew Morton 574266cf658SDavid Howells if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) 575bd4c8ce4SAndrew Morton return 0; 576bd4c8ce4SAndrew Morton 577c4843a75SGreg Thelen spin_lock_irqsave(&mapping->tree_lock, flags); 578bd4c8ce4SAndrew Morton if (PageDirty(page)) 579bd4c8ce4SAndrew Morton goto failed; 580bd4c8ce4SAndrew Morton 581266cf658SDavid Howells BUG_ON(page_has_private(page)); 58262cccb8cSJohannes Weiner __delete_from_page_cache(page, NULL); 583c4843a75SGreg Thelen spin_unlock_irqrestore(&mapping->tree_lock, flags); 5846072d13cSLinus Torvalds 5856072d13cSLinus Torvalds if (mapping->a_ops->freepage) 5866072d13cSLinus Torvalds mapping->a_ops->freepage(page); 5876072d13cSLinus Torvalds 58809cbfeafSKirill A. Shutemov put_page(page); /* pagecache ref */ 589bd4c8ce4SAndrew Morton return 1; 590bd4c8ce4SAndrew Morton failed: 591c4843a75SGreg Thelen spin_unlock_irqrestore(&mapping->tree_lock, flags); 592bd4c8ce4SAndrew Morton return 0; 593bd4c8ce4SAndrew Morton } 594bd4c8ce4SAndrew Morton 595e3db7691STrond Myklebust static int do_launder_page(struct address_space *mapping, struct page *page) 596e3db7691STrond Myklebust { 597e3db7691STrond Myklebust if (!PageDirty(page)) 598e3db7691STrond Myklebust return 0; 599e3db7691STrond Myklebust if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) 600e3db7691STrond Myklebust return 0; 601e3db7691STrond Myklebust return mapping->a_ops->launder_page(page); 602e3db7691STrond Myklebust } 603e3db7691STrond Myklebust 6041da177e4SLinus Torvalds /** 6051da177e4SLinus Torvalds * invalidate_inode_pages2_range - remove range of pages from an address_space 60667be2dd1SMartin Waitz * @mapping: the address_space 6071da177e4SLinus Torvalds * @start: the page offset 'from' which to invalidate 6081da177e4SLinus Torvalds * @end: the page offset 'to' which to invalidate (inclusive) 6091da177e4SLinus Torvalds * 6101da177e4SLinus Torvalds * Any pages which are found to be mapped into pagetables are unmapped prior to 6111da177e4SLinus Torvalds * invalidation. 6121da177e4SLinus Torvalds * 6136ccfa806SHisashi Hifumi * Returns -EBUSY if any pages could not be invalidated. 6141da177e4SLinus Torvalds */ 6151da177e4SLinus Torvalds int invalidate_inode_pages2_range(struct address_space *mapping, 6161da177e4SLinus Torvalds pgoff_t start, pgoff_t end) 6171da177e4SLinus Torvalds { 6180cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 6191da177e4SLinus Torvalds struct pagevec pvec; 620b85e0effSHugh Dickins pgoff_t index; 6211da177e4SLinus Torvalds int i; 6221da177e4SLinus Torvalds int ret = 0; 6230dd1334fSHisashi Hifumi int ret2 = 0; 6241da177e4SLinus Torvalds int did_range_unmap = 0; 6251da177e4SLinus Torvalds 6263167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 627*32691f0fSAndrey Ryabinin if (mapping->nrpages == 0 && mapping->nrexceptional == 0) 628*32691f0fSAndrey Ryabinin return 0; 629*32691f0fSAndrey Ryabinin 6301da177e4SLinus Torvalds pagevec_init(&pvec, 0); 631b85e0effSHugh Dickins index = start; 6320cd6144aSJohannes Weiner while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, 6330cd6144aSJohannes Weiner min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, 6340cd6144aSJohannes Weiner indices)) { 6357b965e08STrond Myklebust for (i = 0; i < pagevec_count(&pvec); i++) { 6361da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 637b85e0effSHugh Dickins 638b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 6390cd6144aSJohannes Weiner index = indices[i]; 640b85e0effSHugh Dickins if (index > end) 641b85e0effSHugh Dickins break; 6421da177e4SLinus Torvalds 6430cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 644c6dcf52cSJan Kara if (!invalidate_exceptional_entry2(mapping, 645c6dcf52cSJan Kara index, page)) 646c6dcf52cSJan Kara ret = -EBUSY; 6470cd6144aSJohannes Weiner continue; 6480cd6144aSJohannes Weiner } 6490cd6144aSJohannes Weiner 6501da177e4SLinus Torvalds lock_page(page); 6515cbc198aSKirill A. Shutemov WARN_ON(page_to_index(page) != index); 6521da177e4SLinus Torvalds if (page->mapping != mapping) { 6531da177e4SLinus Torvalds unlock_page(page); 6541da177e4SLinus Torvalds continue; 6551da177e4SLinus Torvalds } 6561da177e4SLinus Torvalds wait_on_page_writeback(page); 657d00806b1SNick Piggin if (page_mapped(page)) { 6581da177e4SLinus Torvalds if (!did_range_unmap) { 6591da177e4SLinus Torvalds /* 6601da177e4SLinus Torvalds * Zap the rest of the file in one hit. 6611da177e4SLinus Torvalds */ 6621da177e4SLinus Torvalds unmap_mapping_range(mapping, 66309cbfeafSKirill A. Shutemov (loff_t)index << PAGE_SHIFT, 664b85e0effSHugh Dickins (loff_t)(1 + end - index) 66509cbfeafSKirill A. Shutemov << PAGE_SHIFT, 6661da177e4SLinus Torvalds 0); 6671da177e4SLinus Torvalds did_range_unmap = 1; 6681da177e4SLinus Torvalds } else { 6691da177e4SLinus Torvalds /* 6701da177e4SLinus Torvalds * Just zap this page 6711da177e4SLinus Torvalds */ 6721da177e4SLinus Torvalds unmap_mapping_range(mapping, 67309cbfeafSKirill A. Shutemov (loff_t)index << PAGE_SHIFT, 67409cbfeafSKirill A. Shutemov PAGE_SIZE, 0); 6751da177e4SLinus Torvalds } 6761da177e4SLinus Torvalds } 677d00806b1SNick Piggin BUG_ON(page_mapped(page)); 6780dd1334fSHisashi Hifumi ret2 = do_launder_page(mapping, page); 6790dd1334fSHisashi Hifumi if (ret2 == 0) { 6800dd1334fSHisashi Hifumi if (!invalidate_complete_page2(mapping, page)) 6816ccfa806SHisashi Hifumi ret2 = -EBUSY; 6820dd1334fSHisashi Hifumi } 6830dd1334fSHisashi Hifumi if (ret2 < 0) 6840dd1334fSHisashi Hifumi ret = ret2; 6851da177e4SLinus Torvalds unlock_page(page); 6861da177e4SLinus Torvalds } 6870cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 6881da177e4SLinus Torvalds pagevec_release(&pvec); 6891da177e4SLinus Torvalds cond_resched(); 690b85e0effSHugh Dickins index++; 6911da177e4SLinus Torvalds } 6923167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 6931da177e4SLinus Torvalds return ret; 6941da177e4SLinus Torvalds } 6951da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); 6961da177e4SLinus Torvalds 6971da177e4SLinus Torvalds /** 6981da177e4SLinus Torvalds * invalidate_inode_pages2 - remove all pages from an address_space 69967be2dd1SMartin Waitz * @mapping: the address_space 7001da177e4SLinus Torvalds * 7011da177e4SLinus Torvalds * Any pages which are found to be mapped into pagetables are unmapped prior to 7021da177e4SLinus Torvalds * invalidation. 7031da177e4SLinus Torvalds * 704e9de25ddSPeng Tao * Returns -EBUSY if any pages could not be invalidated. 7051da177e4SLinus Torvalds */ 7061da177e4SLinus Torvalds int invalidate_inode_pages2(struct address_space *mapping) 7071da177e4SLinus Torvalds { 7081da177e4SLinus Torvalds return invalidate_inode_pages2_range(mapping, 0, -1); 7091da177e4SLinus Torvalds } 7101da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(invalidate_inode_pages2); 71125d9e2d1Snpiggin@suse.de 71225d9e2d1Snpiggin@suse.de /** 71325d9e2d1Snpiggin@suse.de * truncate_pagecache - unmap and remove pagecache that has been truncated 71425d9e2d1Snpiggin@suse.de * @inode: inode 7158a549beaSHugh Dickins * @newsize: new file size 71625d9e2d1Snpiggin@suse.de * 71725d9e2d1Snpiggin@suse.de * inode's new i_size must already be written before truncate_pagecache 71825d9e2d1Snpiggin@suse.de * is called. 71925d9e2d1Snpiggin@suse.de * 72025d9e2d1Snpiggin@suse.de * This function should typically be called before the filesystem 72125d9e2d1Snpiggin@suse.de * releases resources associated with the freed range (eg. deallocates 72225d9e2d1Snpiggin@suse.de * blocks). This way, pagecache will always stay logically coherent 72325d9e2d1Snpiggin@suse.de * with on-disk format, and the filesystem would not have to deal with 72425d9e2d1Snpiggin@suse.de * situations such as writepage being called for a page that has already 72525d9e2d1Snpiggin@suse.de * had its underlying blocks deallocated. 72625d9e2d1Snpiggin@suse.de */ 7277caef267SKirill A. Shutemov void truncate_pagecache(struct inode *inode, loff_t newsize) 72825d9e2d1Snpiggin@suse.de { 72925d9e2d1Snpiggin@suse.de struct address_space *mapping = inode->i_mapping; 7308a549beaSHugh Dickins loff_t holebegin = round_up(newsize, PAGE_SIZE); 73125d9e2d1Snpiggin@suse.de 73225d9e2d1Snpiggin@suse.de /* 73325d9e2d1Snpiggin@suse.de * unmap_mapping_range is called twice, first simply for 73425d9e2d1Snpiggin@suse.de * efficiency so that truncate_inode_pages does fewer 73525d9e2d1Snpiggin@suse.de * single-page unmaps. However after this first call, and 73625d9e2d1Snpiggin@suse.de * before truncate_inode_pages finishes, it is possible for 73725d9e2d1Snpiggin@suse.de * private pages to be COWed, which remain after 73825d9e2d1Snpiggin@suse.de * truncate_inode_pages finishes, hence the second 73925d9e2d1Snpiggin@suse.de * unmap_mapping_range call must be made for correctness. 74025d9e2d1Snpiggin@suse.de */ 7418a549beaSHugh Dickins unmap_mapping_range(mapping, holebegin, 0, 1); 7428a549beaSHugh Dickins truncate_inode_pages(mapping, newsize); 7438a549beaSHugh Dickins unmap_mapping_range(mapping, holebegin, 0, 1); 74425d9e2d1Snpiggin@suse.de } 74525d9e2d1Snpiggin@suse.de EXPORT_SYMBOL(truncate_pagecache); 74625d9e2d1Snpiggin@suse.de 74725d9e2d1Snpiggin@suse.de /** 7482c27c65eSChristoph Hellwig * truncate_setsize - update inode and pagecache for a new file size 7492c27c65eSChristoph Hellwig * @inode: inode 7502c27c65eSChristoph Hellwig * @newsize: new file size 7512c27c65eSChristoph Hellwig * 752382e27daSJan Kara * truncate_setsize updates i_size and performs pagecache truncation (if 753382e27daSJan Kara * necessary) to @newsize. It will be typically be called from the filesystem's 754382e27daSJan Kara * setattr function when ATTR_SIZE is passed in. 7552c27c65eSChristoph Hellwig * 75677783d06SJan Kara * Must be called with a lock serializing truncates and writes (generally 75777783d06SJan Kara * i_mutex but e.g. xfs uses a different lock) and before all filesystem 75877783d06SJan Kara * specific block truncation has been performed. 7592c27c65eSChristoph Hellwig */ 7602c27c65eSChristoph Hellwig void truncate_setsize(struct inode *inode, loff_t newsize) 7612c27c65eSChristoph Hellwig { 76290a80202SJan Kara loff_t oldsize = inode->i_size; 76390a80202SJan Kara 7642c27c65eSChristoph Hellwig i_size_write(inode, newsize); 76590a80202SJan Kara if (newsize > oldsize) 76690a80202SJan Kara pagecache_isize_extended(inode, oldsize, newsize); 7677caef267SKirill A. Shutemov truncate_pagecache(inode, newsize); 7682c27c65eSChristoph Hellwig } 7692c27c65eSChristoph Hellwig EXPORT_SYMBOL(truncate_setsize); 7702c27c65eSChristoph Hellwig 7712c27c65eSChristoph Hellwig /** 77290a80202SJan Kara * pagecache_isize_extended - update pagecache after extension of i_size 77390a80202SJan Kara * @inode: inode for which i_size was extended 77490a80202SJan Kara * @from: original inode size 77590a80202SJan Kara * @to: new inode size 77690a80202SJan Kara * 77790a80202SJan Kara * Handle extension of inode size either caused by extending truncate or by 77890a80202SJan Kara * write starting after current i_size. We mark the page straddling current 77990a80202SJan Kara * i_size RO so that page_mkwrite() is called on the nearest write access to 78090a80202SJan Kara * the page. This way filesystem can be sure that page_mkwrite() is called on 78190a80202SJan Kara * the page before user writes to the page via mmap after the i_size has been 78290a80202SJan Kara * changed. 78390a80202SJan Kara * 78490a80202SJan Kara * The function must be called after i_size is updated so that page fault 78590a80202SJan Kara * coming after we unlock the page will already see the new i_size. 78690a80202SJan Kara * The function must be called while we still hold i_mutex - this not only 78790a80202SJan Kara * makes sure i_size is stable but also that userspace cannot observe new 78890a80202SJan Kara * i_size value before we are prepared to store mmap writes at new inode size. 78990a80202SJan Kara */ 79090a80202SJan Kara void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) 79190a80202SJan Kara { 79293407472SFabian Frederick int bsize = i_blocksize(inode); 79390a80202SJan Kara loff_t rounded_from; 79490a80202SJan Kara struct page *page; 79590a80202SJan Kara pgoff_t index; 79690a80202SJan Kara 79790a80202SJan Kara WARN_ON(to > inode->i_size); 79890a80202SJan Kara 79909cbfeafSKirill A. Shutemov if (from >= to || bsize == PAGE_SIZE) 80090a80202SJan Kara return; 80190a80202SJan Kara /* Page straddling @from will not have any hole block created? */ 80290a80202SJan Kara rounded_from = round_up(from, bsize); 80309cbfeafSKirill A. Shutemov if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1))) 80490a80202SJan Kara return; 80590a80202SJan Kara 80609cbfeafSKirill A. Shutemov index = from >> PAGE_SHIFT; 80790a80202SJan Kara page = find_lock_page(inode->i_mapping, index); 80890a80202SJan Kara /* Page not cached? Nothing to do */ 80990a80202SJan Kara if (!page) 81090a80202SJan Kara return; 81190a80202SJan Kara /* 81290a80202SJan Kara * See clear_page_dirty_for_io() for details why set_page_dirty() 81390a80202SJan Kara * is needed. 81490a80202SJan Kara */ 81590a80202SJan Kara if (page_mkclean(page)) 81690a80202SJan Kara set_page_dirty(page); 81790a80202SJan Kara unlock_page(page); 81809cbfeafSKirill A. Shutemov put_page(page); 81990a80202SJan Kara } 82090a80202SJan Kara EXPORT_SYMBOL(pagecache_isize_extended); 82190a80202SJan Kara 82290a80202SJan Kara /** 823623e3db9SHugh Dickins * truncate_pagecache_range - unmap and remove pagecache that is hole-punched 824623e3db9SHugh Dickins * @inode: inode 825623e3db9SHugh Dickins * @lstart: offset of beginning of hole 826623e3db9SHugh Dickins * @lend: offset of last byte of hole 827623e3db9SHugh Dickins * 828623e3db9SHugh Dickins * This function should typically be called before the filesystem 829623e3db9SHugh Dickins * releases resources associated with the freed range (eg. deallocates 830623e3db9SHugh Dickins * blocks). This way, pagecache will always stay logically coherent 831623e3db9SHugh Dickins * with on-disk format, and the filesystem would not have to deal with 832623e3db9SHugh Dickins * situations such as writepage being called for a page that has already 833623e3db9SHugh Dickins * had its underlying blocks deallocated. 834623e3db9SHugh Dickins */ 835623e3db9SHugh Dickins void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) 836623e3db9SHugh Dickins { 837623e3db9SHugh Dickins struct address_space *mapping = inode->i_mapping; 838623e3db9SHugh Dickins loff_t unmap_start = round_up(lstart, PAGE_SIZE); 839623e3db9SHugh Dickins loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; 840623e3db9SHugh Dickins /* 841623e3db9SHugh Dickins * This rounding is currently just for example: unmap_mapping_range 842623e3db9SHugh Dickins * expands its hole outwards, whereas we want it to contract the hole 843623e3db9SHugh Dickins * inwards. However, existing callers of truncate_pagecache_range are 8445a720394SLukas Czerner * doing their own page rounding first. Note that unmap_mapping_range 8455a720394SLukas Czerner * allows holelen 0 for all, and we allow lend -1 for end of file. 846623e3db9SHugh Dickins */ 847623e3db9SHugh Dickins 848623e3db9SHugh Dickins /* 849623e3db9SHugh Dickins * Unlike in truncate_pagecache, unmap_mapping_range is called only 850623e3db9SHugh Dickins * once (before truncating pagecache), and without "even_cows" flag: 851623e3db9SHugh Dickins * hole-punching should not remove private COWed pages from the hole. 852623e3db9SHugh Dickins */ 853623e3db9SHugh Dickins if ((u64)unmap_end > (u64)unmap_start) 854623e3db9SHugh Dickins unmap_mapping_range(mapping, unmap_start, 855623e3db9SHugh Dickins 1 + unmap_end - unmap_start, 0); 856623e3db9SHugh Dickins truncate_inode_pages_range(mapping, lstart, lend); 857623e3db9SHugh Dickins } 858623e3db9SHugh Dickins EXPORT_SYMBOL(truncate_pagecache_range); 859