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 27*c6dcf52cSJan Kara static void clear_shadow_entry(struct address_space *mapping, pgoff_t index, 28*c6dcf52cSJan Kara void *entry) 290cd6144aSJohannes Weiner { 30449dd698SJohannes Weiner struct radix_tree_node *node; 31449dd698SJohannes Weiner void **slot; 32449dd698SJohannes Weiner 33ac401cc7SJan Kara spin_lock_irq(&mapping->tree_lock); 340cd6144aSJohannes Weiner /* 350cd6144aSJohannes Weiner * Regular page slots are stabilized by the page lock even 360cd6144aSJohannes Weiner * without the tree itself locked. These unlocked entries 370cd6144aSJohannes Weiner * need verification under the tree lock. 380cd6144aSJohannes Weiner */ 3914b46879SJohannes Weiner if (!__radix_tree_lookup(&mapping->page_tree, index, &node, &slot)) 40449dd698SJohannes Weiner goto unlock; 41449dd698SJohannes Weiner if (*slot != entry) 42449dd698SJohannes Weiner goto unlock; 4314b46879SJohannes Weiner __radix_tree_replace(&mapping->page_tree, node, slot, NULL, 4414b46879SJohannes Weiner workingset_update_node, mapping); 45f9fe48beSRoss Zwisler mapping->nrexceptional--; 46449dd698SJohannes Weiner unlock: 470cd6144aSJohannes Weiner spin_unlock_irq(&mapping->tree_lock); 480cd6144aSJohannes Weiner } 491da177e4SLinus Torvalds 50*c6dcf52cSJan Kara /* 51*c6dcf52cSJan Kara * Unconditionally remove exceptional entry. Usually called from truncate path. 52*c6dcf52cSJan Kara */ 53*c6dcf52cSJan Kara static void truncate_exceptional_entry(struct address_space *mapping, 54*c6dcf52cSJan Kara pgoff_t index, void *entry) 55*c6dcf52cSJan Kara { 56*c6dcf52cSJan Kara /* Handled by shmem itself */ 57*c6dcf52cSJan Kara if (shmem_mapping(mapping)) 58*c6dcf52cSJan Kara return; 59*c6dcf52cSJan Kara 60*c6dcf52cSJan Kara if (dax_mapping(mapping)) { 61*c6dcf52cSJan Kara dax_delete_mapping_entry(mapping, index); 62*c6dcf52cSJan Kara return; 63*c6dcf52cSJan Kara } 64*c6dcf52cSJan Kara clear_shadow_entry(mapping, index, entry); 65*c6dcf52cSJan Kara } 66*c6dcf52cSJan Kara 67*c6dcf52cSJan Kara /* 68*c6dcf52cSJan Kara * Invalidate exceptional entry if easily possible. This handles exceptional 69*c6dcf52cSJan Kara * entries for invalidate_inode_pages() so for DAX it evicts only unlocked and 70*c6dcf52cSJan Kara * clean entries. 71*c6dcf52cSJan Kara */ 72*c6dcf52cSJan Kara static int invalidate_exceptional_entry(struct address_space *mapping, 73*c6dcf52cSJan Kara pgoff_t index, void *entry) 74*c6dcf52cSJan Kara { 75*c6dcf52cSJan Kara /* Handled by shmem itself */ 76*c6dcf52cSJan Kara if (shmem_mapping(mapping)) 77*c6dcf52cSJan Kara return 1; 78*c6dcf52cSJan Kara if (dax_mapping(mapping)) 79*c6dcf52cSJan Kara return dax_invalidate_mapping_entry(mapping, index); 80*c6dcf52cSJan Kara clear_shadow_entry(mapping, index, entry); 81*c6dcf52cSJan Kara return 1; 82*c6dcf52cSJan Kara } 83*c6dcf52cSJan Kara 84*c6dcf52cSJan Kara /* 85*c6dcf52cSJan Kara * Invalidate exceptional entry if clean. This handles exceptional entries for 86*c6dcf52cSJan Kara * invalidate_inode_pages2() so for DAX it evicts only clean entries. 87*c6dcf52cSJan Kara */ 88*c6dcf52cSJan Kara static int invalidate_exceptional_entry2(struct address_space *mapping, 89*c6dcf52cSJan Kara pgoff_t index, void *entry) 90*c6dcf52cSJan Kara { 91*c6dcf52cSJan Kara /* Handled by shmem itself */ 92*c6dcf52cSJan Kara if (shmem_mapping(mapping)) 93*c6dcf52cSJan Kara return 1; 94*c6dcf52cSJan Kara if (dax_mapping(mapping)) 95*c6dcf52cSJan Kara return dax_invalidate_mapping_entry_sync(mapping, index); 96*c6dcf52cSJan Kara clear_shadow_entry(mapping, index, entry); 97*c6dcf52cSJan Kara return 1; 98*c6dcf52cSJan Kara } 99*c6dcf52cSJan Kara 100cf9a2ae8SDavid Howells /** 10128bc44d7SFengguang Wu * do_invalidatepage - invalidate part or all of a page 102cf9a2ae8SDavid Howells * @page: the page which is affected 103d47992f8SLukas Czerner * @offset: start of the range to invalidate 104d47992f8SLukas Czerner * @length: length of the range to invalidate 105cf9a2ae8SDavid Howells * 106cf9a2ae8SDavid Howells * do_invalidatepage() is called when all or part of the page has become 107cf9a2ae8SDavid Howells * invalidated by a truncate operation. 108cf9a2ae8SDavid Howells * 109cf9a2ae8SDavid Howells * do_invalidatepage() does not have to release all buffers, but it must 110cf9a2ae8SDavid Howells * ensure that no dirty buffer is left outside @offset and that no I/O 111cf9a2ae8SDavid Howells * is underway against any of the blocks which are outside the truncation 112cf9a2ae8SDavid Howells * point. Because the caller is about to free (and possibly reuse) those 113cf9a2ae8SDavid Howells * blocks on-disk. 114cf9a2ae8SDavid Howells */ 115d47992f8SLukas Czerner void do_invalidatepage(struct page *page, unsigned int offset, 116d47992f8SLukas Czerner unsigned int length) 117cf9a2ae8SDavid Howells { 118d47992f8SLukas Czerner void (*invalidatepage)(struct page *, unsigned int, unsigned int); 119d47992f8SLukas Czerner 120cf9a2ae8SDavid Howells invalidatepage = page->mapping->a_ops->invalidatepage; 1219361401eSDavid Howells #ifdef CONFIG_BLOCK 122cf9a2ae8SDavid Howells if (!invalidatepage) 123cf9a2ae8SDavid Howells invalidatepage = block_invalidatepage; 1249361401eSDavid Howells #endif 125cf9a2ae8SDavid Howells if (invalidatepage) 126d47992f8SLukas Czerner (*invalidatepage)(page, offset, length); 127cf9a2ae8SDavid Howells } 128cf9a2ae8SDavid Howells 129ecdfc978SLinus Torvalds /* 1301da177e4SLinus Torvalds * If truncate cannot remove the fs-private metadata from the page, the page 13162e1c553SShaohua Li * becomes orphaned. It will be left on the LRU and may even be mapped into 13254cb8821SNick Piggin * user pagetables if we're racing with filemap_fault(). 1331da177e4SLinus Torvalds * 1341da177e4SLinus Torvalds * We need to bale out if page->mapping is no longer equal to the original 1351da177e4SLinus Torvalds * mapping. This happens a) when the VM reclaimed the page while we waited on 136fc0ecff6SAndrew Morton * its lock, b) when a concurrent invalidate_mapping_pages got there first and 1371da177e4SLinus Torvalds * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. 1381da177e4SLinus Torvalds */ 139750b4987SNick Piggin static int 1401da177e4SLinus Torvalds truncate_complete_page(struct address_space *mapping, struct page *page) 1411da177e4SLinus Torvalds { 1421da177e4SLinus Torvalds if (page->mapping != mapping) 143750b4987SNick Piggin return -EIO; 1441da177e4SLinus Torvalds 145266cf658SDavid Howells if (page_has_private(page)) 14609cbfeafSKirill A. Shutemov do_invalidatepage(page, 0, PAGE_SIZE); 1471da177e4SLinus Torvalds 148b9ea2515SKonstantin Khlebnikov /* 149b9ea2515SKonstantin Khlebnikov * Some filesystems seem to re-dirty the page even after 150b9ea2515SKonstantin Khlebnikov * the VM has canceled the dirty bit (eg ext3 journaling). 151b9ea2515SKonstantin Khlebnikov * Hence dirty accounting check is placed after invalidation. 152b9ea2515SKonstantin Khlebnikov */ 15311f81becSTejun Heo cancel_dirty_page(page); 1541da177e4SLinus Torvalds ClearPageMappedToDisk(page); 1555adc7b51SMinchan Kim delete_from_page_cache(page); 156750b4987SNick Piggin return 0; 1571da177e4SLinus Torvalds } 1581da177e4SLinus Torvalds 1591da177e4SLinus Torvalds /* 160fc0ecff6SAndrew Morton * This is for invalidate_mapping_pages(). That function can be called at 1611da177e4SLinus Torvalds * any time, and is not supposed to throw away dirty pages. But pages can 1620fd0e6b0SNick Piggin * be marked dirty at any time too, so use remove_mapping which safely 1630fd0e6b0SNick Piggin * discards clean, unused pages. 1641da177e4SLinus Torvalds * 1651da177e4SLinus Torvalds * Returns non-zero if the page was successfully invalidated. 1661da177e4SLinus Torvalds */ 1671da177e4SLinus Torvalds static int 1681da177e4SLinus Torvalds invalidate_complete_page(struct address_space *mapping, struct page *page) 1691da177e4SLinus Torvalds { 1700fd0e6b0SNick Piggin int ret; 1710fd0e6b0SNick Piggin 1721da177e4SLinus Torvalds if (page->mapping != mapping) 1731da177e4SLinus Torvalds return 0; 1741da177e4SLinus Torvalds 175266cf658SDavid Howells if (page_has_private(page) && !try_to_release_page(page, 0)) 1761da177e4SLinus Torvalds return 0; 1771da177e4SLinus Torvalds 1780fd0e6b0SNick Piggin ret = remove_mapping(mapping, page); 1790fd0e6b0SNick Piggin 1800fd0e6b0SNick Piggin return ret; 1811da177e4SLinus Torvalds } 1821da177e4SLinus Torvalds 183750b4987SNick Piggin int truncate_inode_page(struct address_space *mapping, struct page *page) 184750b4987SNick Piggin { 185fc127da0SKirill A. Shutemov loff_t holelen; 186fc127da0SKirill A. Shutemov VM_BUG_ON_PAGE(PageTail(page), page); 187fc127da0SKirill A. Shutemov 188fc127da0SKirill A. Shutemov holelen = PageTransHuge(page) ? HPAGE_PMD_SIZE : PAGE_SIZE; 189750b4987SNick Piggin if (page_mapped(page)) { 190750b4987SNick Piggin unmap_mapping_range(mapping, 19109cbfeafSKirill A. Shutemov (loff_t)page->index << PAGE_SHIFT, 192fc127da0SKirill A. Shutemov holelen, 0); 193750b4987SNick Piggin } 194750b4987SNick Piggin return truncate_complete_page(mapping, page); 195750b4987SNick Piggin } 196750b4987SNick Piggin 19783f78668SWu Fengguang /* 19825718736SAndi Kleen * Used to get rid of pages on hardware memory corruption. 19925718736SAndi Kleen */ 20025718736SAndi Kleen int generic_error_remove_page(struct address_space *mapping, struct page *page) 20125718736SAndi Kleen { 20225718736SAndi Kleen if (!mapping) 20325718736SAndi Kleen return -EINVAL; 20425718736SAndi Kleen /* 20525718736SAndi Kleen * Only punch for normal data pages for now. 20625718736SAndi Kleen * Handling other types like directories would need more auditing. 20725718736SAndi Kleen */ 20825718736SAndi Kleen if (!S_ISREG(mapping->host->i_mode)) 20925718736SAndi Kleen return -EIO; 21025718736SAndi Kleen return truncate_inode_page(mapping, page); 21125718736SAndi Kleen } 21225718736SAndi Kleen EXPORT_SYMBOL(generic_error_remove_page); 21325718736SAndi Kleen 21425718736SAndi Kleen /* 21583f78668SWu Fengguang * Safely invalidate one page from its pagecache mapping. 21683f78668SWu Fengguang * It only drops clean, unused pages. The page must be locked. 21783f78668SWu Fengguang * 21883f78668SWu Fengguang * Returns 1 if the page is successfully invalidated, otherwise 0. 21983f78668SWu Fengguang */ 22083f78668SWu Fengguang int invalidate_inode_page(struct page *page) 22183f78668SWu Fengguang { 22283f78668SWu Fengguang struct address_space *mapping = page_mapping(page); 22383f78668SWu Fengguang if (!mapping) 22483f78668SWu Fengguang return 0; 22583f78668SWu Fengguang if (PageDirty(page) || PageWriteback(page)) 22683f78668SWu Fengguang return 0; 22783f78668SWu Fengguang if (page_mapped(page)) 22883f78668SWu Fengguang return 0; 22983f78668SWu Fengguang return invalidate_complete_page(mapping, page); 23083f78668SWu Fengguang } 23183f78668SWu Fengguang 2321da177e4SLinus Torvalds /** 23373c1e204SLiu Bo * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets 2341da177e4SLinus Torvalds * @mapping: mapping to truncate 2351da177e4SLinus Torvalds * @lstart: offset from which to truncate 2365a720394SLukas Czerner * @lend: offset to which to truncate (inclusive) 2371da177e4SLinus Torvalds * 238d7339071SHans Reiser * Truncate the page cache, removing the pages that are between 2395a720394SLukas Czerner * specified offsets (and zeroing out partial pages 2405a720394SLukas Czerner * if lstart or lend + 1 is not page aligned). 2411da177e4SLinus Torvalds * 2421da177e4SLinus Torvalds * Truncate takes two passes - the first pass is nonblocking. It will not 2431da177e4SLinus Torvalds * block on page locks and it will not block on writeback. The second pass 2441da177e4SLinus Torvalds * will wait. This is to prevent as much IO as possible in the affected region. 2451da177e4SLinus Torvalds * The first pass will remove most pages, so the search cost of the second pass 2461da177e4SLinus Torvalds * is low. 2471da177e4SLinus Torvalds * 2481da177e4SLinus Torvalds * We pass down the cache-hot hint to the page freeing code. Even if the 2491da177e4SLinus Torvalds * mapping is large, it is probably the case that the final pages are the most 2501da177e4SLinus Torvalds * recently touched, and freeing happens in ascending file offset order. 2515a720394SLukas Czerner * 2525a720394SLukas Czerner * Note that since ->invalidatepage() accepts range to invalidate 2535a720394SLukas Czerner * truncate_inode_pages_range is able to handle cases where lend + 1 is not 2545a720394SLukas Czerner * page aligned properly. 2551da177e4SLinus Torvalds */ 256d7339071SHans Reiser void truncate_inode_pages_range(struct address_space *mapping, 257d7339071SHans Reiser loff_t lstart, loff_t lend) 2581da177e4SLinus Torvalds { 2595a720394SLukas Czerner pgoff_t start; /* inclusive */ 2605a720394SLukas Czerner pgoff_t end; /* exclusive */ 2615a720394SLukas Czerner unsigned int partial_start; /* inclusive */ 2625a720394SLukas Czerner unsigned int partial_end; /* exclusive */ 2631da177e4SLinus Torvalds struct pagevec pvec; 2640cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 265b85e0effSHugh Dickins pgoff_t index; 2661da177e4SLinus Torvalds int i; 2671da177e4SLinus Torvalds 2683167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 269f9fe48beSRoss Zwisler if (mapping->nrpages == 0 && mapping->nrexceptional == 0) 2701da177e4SLinus Torvalds return; 2711da177e4SLinus Torvalds 2725a720394SLukas Czerner /* Offsets within partial pages */ 27309cbfeafSKirill A. Shutemov partial_start = lstart & (PAGE_SIZE - 1); 27409cbfeafSKirill A. Shutemov partial_end = (lend + 1) & (PAGE_SIZE - 1); 2755a720394SLukas Czerner 2765a720394SLukas Czerner /* 2775a720394SLukas Czerner * 'start' and 'end' always covers the range of pages to be fully 2785a720394SLukas Czerner * truncated. Partial pages are covered with 'partial_start' at the 2795a720394SLukas Czerner * start of the range and 'partial_end' at the end of the range. 2805a720394SLukas Czerner * Note that 'end' is exclusive while 'lend' is inclusive. 2815a720394SLukas Czerner */ 28209cbfeafSKirill A. Shutemov start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; 2835a720394SLukas Czerner if (lend == -1) 2845a720394SLukas Czerner /* 2855a720394SLukas Czerner * lend == -1 indicates end-of-file so we have to set 'end' 2865a720394SLukas Czerner * to the highest possible pgoff_t and since the type is 2875a720394SLukas Czerner * unsigned we're using -1. 2885a720394SLukas Czerner */ 2895a720394SLukas Czerner end = -1; 2905a720394SLukas Czerner else 29109cbfeafSKirill A. Shutemov end = (lend + 1) >> PAGE_SHIFT; 292d7339071SHans Reiser 2931da177e4SLinus Torvalds pagevec_init(&pvec, 0); 294b85e0effSHugh Dickins index = start; 2950cd6144aSJohannes Weiner while (index < end && pagevec_lookup_entries(&pvec, mapping, index, 2960cd6144aSJohannes Weiner min(end - index, (pgoff_t)PAGEVEC_SIZE), 2970cd6144aSJohannes Weiner indices)) { 2981da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 2991da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 3001da177e4SLinus Torvalds 301b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 3020cd6144aSJohannes Weiner index = indices[i]; 3035a720394SLukas Czerner if (index >= end) 304d7339071SHans Reiser break; 305d7339071SHans Reiser 3060cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 307*c6dcf52cSJan Kara truncate_exceptional_entry(mapping, index, 308*c6dcf52cSJan Kara page); 3090cd6144aSJohannes Weiner continue; 3100cd6144aSJohannes Weiner } 3110cd6144aSJohannes Weiner 312529ae9aaSNick Piggin if (!trylock_page(page)) 3131da177e4SLinus Torvalds continue; 3145cbc198aSKirill A. Shutemov WARN_ON(page_to_index(page) != index); 3151da177e4SLinus Torvalds if (PageWriteback(page)) { 3161da177e4SLinus Torvalds unlock_page(page); 3171da177e4SLinus Torvalds continue; 3181da177e4SLinus Torvalds } 319750b4987SNick Piggin truncate_inode_page(mapping, page); 3201da177e4SLinus Torvalds unlock_page(page); 3211da177e4SLinus Torvalds } 3220cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 3231da177e4SLinus Torvalds pagevec_release(&pvec); 3241da177e4SLinus Torvalds cond_resched(); 325b85e0effSHugh Dickins index++; 3261da177e4SLinus Torvalds } 3271da177e4SLinus Torvalds 3285a720394SLukas Czerner if (partial_start) { 3291da177e4SLinus Torvalds struct page *page = find_lock_page(mapping, start - 1); 3301da177e4SLinus Torvalds if (page) { 33109cbfeafSKirill A. Shutemov unsigned int top = PAGE_SIZE; 3325a720394SLukas Czerner if (start > end) { 3335a720394SLukas Czerner /* Truncation within a single page */ 3345a720394SLukas Czerner top = partial_end; 3355a720394SLukas Czerner partial_end = 0; 3365a720394SLukas Czerner } 3371da177e4SLinus Torvalds wait_on_page_writeback(page); 3385a720394SLukas Czerner zero_user_segment(page, partial_start, top); 3395a720394SLukas Czerner cleancache_invalidate_page(mapping, page); 3405a720394SLukas Czerner if (page_has_private(page)) 3415a720394SLukas Czerner do_invalidatepage(page, partial_start, 3425a720394SLukas Czerner top - partial_start); 3431da177e4SLinus Torvalds unlock_page(page); 34409cbfeafSKirill A. Shutemov put_page(page); 3451da177e4SLinus Torvalds } 3461da177e4SLinus Torvalds } 3475a720394SLukas Czerner if (partial_end) { 3485a720394SLukas Czerner struct page *page = find_lock_page(mapping, end); 3495a720394SLukas Czerner if (page) { 3505a720394SLukas Czerner wait_on_page_writeback(page); 3515a720394SLukas Czerner zero_user_segment(page, 0, partial_end); 3525a720394SLukas Czerner cleancache_invalidate_page(mapping, page); 3535a720394SLukas Czerner if (page_has_private(page)) 3545a720394SLukas Czerner do_invalidatepage(page, 0, 3555a720394SLukas Czerner partial_end); 3565a720394SLukas Czerner unlock_page(page); 35709cbfeafSKirill A. Shutemov put_page(page); 3585a720394SLukas Czerner } 3595a720394SLukas Czerner } 3605a720394SLukas Czerner /* 3615a720394SLukas Czerner * If the truncation happened within a single page no pages 3625a720394SLukas Czerner * will be released, just zeroed, so we can bail out now. 3635a720394SLukas Czerner */ 3645a720394SLukas Czerner if (start >= end) 3655a720394SLukas Czerner return; 3661da177e4SLinus Torvalds 367b85e0effSHugh Dickins index = start; 3681da177e4SLinus Torvalds for ( ; ; ) { 3691da177e4SLinus Torvalds cond_resched(); 3700cd6144aSJohannes Weiner if (!pagevec_lookup_entries(&pvec, mapping, index, 371792ceaefSHugh Dickins min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) { 372792ceaefSHugh Dickins /* If all gone from start onwards, we're done */ 373b85e0effSHugh Dickins if (index == start) 3741da177e4SLinus Torvalds break; 375792ceaefSHugh Dickins /* Otherwise restart to make sure all gone */ 376b85e0effSHugh Dickins index = start; 3771da177e4SLinus Torvalds continue; 3781da177e4SLinus Torvalds } 3790cd6144aSJohannes Weiner if (index == start && indices[0] >= end) { 380792ceaefSHugh Dickins /* All gone out of hole to be punched, we're done */ 3810cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 382d7339071SHans Reiser pagevec_release(&pvec); 383d7339071SHans Reiser break; 384d7339071SHans Reiser } 3851da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 3861da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 3871da177e4SLinus Torvalds 388b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 3890cd6144aSJohannes Weiner index = indices[i]; 390792ceaefSHugh Dickins if (index >= end) { 391792ceaefSHugh Dickins /* Restart punch to make sure all gone */ 392792ceaefSHugh Dickins index = start - 1; 393d7339071SHans Reiser break; 394792ceaefSHugh Dickins } 395b85e0effSHugh Dickins 3960cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 397*c6dcf52cSJan Kara truncate_exceptional_entry(mapping, index, 398*c6dcf52cSJan Kara page); 3990cd6144aSJohannes Weiner continue; 4000cd6144aSJohannes Weiner } 4010cd6144aSJohannes Weiner 4021da177e4SLinus Torvalds lock_page(page); 4035cbc198aSKirill A. Shutemov WARN_ON(page_to_index(page) != index); 4041da177e4SLinus Torvalds wait_on_page_writeback(page); 405750b4987SNick Piggin truncate_inode_page(mapping, page); 4061da177e4SLinus Torvalds unlock_page(page); 4071da177e4SLinus Torvalds } 4080cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 4091da177e4SLinus Torvalds pagevec_release(&pvec); 410b85e0effSHugh Dickins index++; 4111da177e4SLinus Torvalds } 4123167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 4131da177e4SLinus Torvalds } 414d7339071SHans Reiser EXPORT_SYMBOL(truncate_inode_pages_range); 4151da177e4SLinus Torvalds 416d7339071SHans Reiser /** 417d7339071SHans Reiser * truncate_inode_pages - truncate *all* the pages from an offset 418d7339071SHans Reiser * @mapping: mapping to truncate 419d7339071SHans Reiser * @lstart: offset from which to truncate 420d7339071SHans Reiser * 4211b1dcc1bSJes Sorensen * Called under (and serialised by) inode->i_mutex. 42208142579SJan Kara * 42308142579SJan Kara * Note: When this function returns, there can be a page in the process of 42408142579SJan Kara * deletion (inside __delete_from_page_cache()) in the specified range. Thus 42508142579SJan Kara * mapping->nrpages can be non-zero when this function returns even after 42608142579SJan Kara * truncation of the whole mapping. 427d7339071SHans Reiser */ 428d7339071SHans Reiser void truncate_inode_pages(struct address_space *mapping, loff_t lstart) 429d7339071SHans Reiser { 430d7339071SHans Reiser truncate_inode_pages_range(mapping, lstart, (loff_t)-1); 431d7339071SHans Reiser } 4321da177e4SLinus Torvalds EXPORT_SYMBOL(truncate_inode_pages); 4331da177e4SLinus Torvalds 43428697355SMike Waychison /** 43591b0abe3SJohannes Weiner * truncate_inode_pages_final - truncate *all* pages before inode dies 43691b0abe3SJohannes Weiner * @mapping: mapping to truncate 43791b0abe3SJohannes Weiner * 43891b0abe3SJohannes Weiner * Called under (and serialized by) inode->i_mutex. 43991b0abe3SJohannes Weiner * 44091b0abe3SJohannes Weiner * Filesystems have to use this in the .evict_inode path to inform the 44191b0abe3SJohannes Weiner * VM that this is the final truncate and the inode is going away. 44291b0abe3SJohannes Weiner */ 44391b0abe3SJohannes Weiner void truncate_inode_pages_final(struct address_space *mapping) 44491b0abe3SJohannes Weiner { 445f9fe48beSRoss Zwisler unsigned long nrexceptional; 44691b0abe3SJohannes Weiner unsigned long nrpages; 44791b0abe3SJohannes Weiner 44891b0abe3SJohannes Weiner /* 44991b0abe3SJohannes Weiner * Page reclaim can not participate in regular inode lifetime 45091b0abe3SJohannes Weiner * management (can't call iput()) and thus can race with the 45191b0abe3SJohannes Weiner * inode teardown. Tell it when the address space is exiting, 45291b0abe3SJohannes Weiner * so that it does not install eviction information after the 45391b0abe3SJohannes Weiner * final truncate has begun. 45491b0abe3SJohannes Weiner */ 45591b0abe3SJohannes Weiner mapping_set_exiting(mapping); 45691b0abe3SJohannes Weiner 45791b0abe3SJohannes Weiner /* 45891b0abe3SJohannes Weiner * When reclaim installs eviction entries, it increases 459f9fe48beSRoss Zwisler * nrexceptional first, then decreases nrpages. Make sure we see 46091b0abe3SJohannes Weiner * this in the right order or we might miss an entry. 46191b0abe3SJohannes Weiner */ 46291b0abe3SJohannes Weiner nrpages = mapping->nrpages; 46391b0abe3SJohannes Weiner smp_rmb(); 464f9fe48beSRoss Zwisler nrexceptional = mapping->nrexceptional; 46591b0abe3SJohannes Weiner 466f9fe48beSRoss Zwisler if (nrpages || nrexceptional) { 46791b0abe3SJohannes Weiner /* 46891b0abe3SJohannes Weiner * As truncation uses a lockless tree lookup, cycle 46991b0abe3SJohannes Weiner * the tree lock to make sure any ongoing tree 47091b0abe3SJohannes Weiner * modification that does not see AS_EXITING is 47191b0abe3SJohannes Weiner * completed before starting the final truncate. 47291b0abe3SJohannes Weiner */ 47391b0abe3SJohannes Weiner spin_lock_irq(&mapping->tree_lock); 47491b0abe3SJohannes Weiner spin_unlock_irq(&mapping->tree_lock); 47591b0abe3SJohannes Weiner 47691b0abe3SJohannes Weiner truncate_inode_pages(mapping, 0); 47791b0abe3SJohannes Weiner } 47891b0abe3SJohannes Weiner } 47991b0abe3SJohannes Weiner EXPORT_SYMBOL(truncate_inode_pages_final); 48091b0abe3SJohannes Weiner 48191b0abe3SJohannes Weiner /** 48228697355SMike Waychison * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode 48328697355SMike Waychison * @mapping: the address_space which holds the pages to invalidate 48428697355SMike Waychison * @start: the offset 'from' which to invalidate 48528697355SMike Waychison * @end: the offset 'to' which to invalidate (inclusive) 48628697355SMike Waychison * 48728697355SMike Waychison * This function only removes the unlocked pages, if you want to 48828697355SMike Waychison * remove all the pages of one inode, you must call truncate_inode_pages. 48928697355SMike Waychison * 49028697355SMike Waychison * invalidate_mapping_pages() will not block on IO activity. It will not 49128697355SMike Waychison * invalidate pages which are dirty, locked, under writeback or mapped into 49228697355SMike Waychison * pagetables. 49328697355SMike Waychison */ 49428697355SMike Waychison unsigned long invalidate_mapping_pages(struct address_space *mapping, 49528697355SMike Waychison pgoff_t start, pgoff_t end) 4961da177e4SLinus Torvalds { 4970cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 4981da177e4SLinus Torvalds struct pagevec pvec; 499b85e0effSHugh Dickins pgoff_t index = start; 50031560180SMinchan Kim unsigned long ret; 50131560180SMinchan Kim unsigned long count = 0; 5021da177e4SLinus Torvalds int i; 5031da177e4SLinus Torvalds 5041da177e4SLinus Torvalds pagevec_init(&pvec, 0); 5050cd6144aSJohannes Weiner while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, 5060cd6144aSJohannes Weiner min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, 5070cd6144aSJohannes Weiner indices)) { 5081da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 5091da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 5101da177e4SLinus Torvalds 511b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 5120cd6144aSJohannes Weiner index = indices[i]; 513b85e0effSHugh Dickins if (index > end) 514b85e0effSHugh Dickins break; 515e0f23603SNeilBrown 5160cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 517*c6dcf52cSJan Kara invalidate_exceptional_entry(mapping, index, 518*c6dcf52cSJan Kara page); 5190cd6144aSJohannes Weiner continue; 5200cd6144aSJohannes Weiner } 5210cd6144aSJohannes Weiner 522b85e0effSHugh Dickins if (!trylock_page(page)) 523b85e0effSHugh Dickins continue; 524fc127da0SKirill A. Shutemov 5255cbc198aSKirill A. Shutemov WARN_ON(page_to_index(page) != index); 526fc127da0SKirill A. Shutemov 527fc127da0SKirill A. Shutemov /* Middle of THP: skip */ 528fc127da0SKirill A. Shutemov if (PageTransTail(page)) { 529fc127da0SKirill A. Shutemov unlock_page(page); 530fc127da0SKirill A. Shutemov continue; 531fc127da0SKirill A. Shutemov } else if (PageTransHuge(page)) { 532fc127da0SKirill A. Shutemov index += HPAGE_PMD_NR - 1; 533fc127da0SKirill A. Shutemov i += HPAGE_PMD_NR - 1; 534fc127da0SKirill A. Shutemov /* 'end' is in the middle of THP */ 535fc127da0SKirill A. Shutemov if (index == round_down(end, HPAGE_PMD_NR)) 536fc127da0SKirill A. Shutemov continue; 537fc127da0SKirill A. Shutemov } 538fc127da0SKirill A. Shutemov 53931560180SMinchan Kim ret = invalidate_inode_page(page); 5401da177e4SLinus Torvalds unlock_page(page); 54131560180SMinchan Kim /* 54231560180SMinchan Kim * Invalidation is a hint that the page is no longer 54331560180SMinchan Kim * of interest and try to speed up its reclaim. 54431560180SMinchan Kim */ 54531560180SMinchan Kim if (!ret) 546cc5993bdSMinchan Kim deactivate_file_page(page); 54731560180SMinchan Kim count += ret; 5481da177e4SLinus Torvalds } 5490cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 5501da177e4SLinus Torvalds pagevec_release(&pvec); 551fc9a07e7SAndrew Morton cond_resched(); 552b85e0effSHugh Dickins index++; 5531da177e4SLinus Torvalds } 55431560180SMinchan Kim return count; 5551da177e4SLinus Torvalds } 55654bc4855SAnton Altaparmakov EXPORT_SYMBOL(invalidate_mapping_pages); 5571da177e4SLinus Torvalds 558bd4c8ce4SAndrew Morton /* 559bd4c8ce4SAndrew Morton * This is like invalidate_complete_page(), except it ignores the page's 560bd4c8ce4SAndrew Morton * refcount. We do this because invalidate_inode_pages2() needs stronger 561bd4c8ce4SAndrew Morton * invalidation guarantees, and cannot afford to leave pages behind because 5622706a1b8SAnderson Briglia * shrink_page_list() has a temp ref on them, or because they're transiently 5632706a1b8SAnderson Briglia * sitting in the lru_cache_add() pagevecs. 564bd4c8ce4SAndrew Morton */ 565bd4c8ce4SAndrew Morton static int 566bd4c8ce4SAndrew Morton invalidate_complete_page2(struct address_space *mapping, struct page *page) 567bd4c8ce4SAndrew Morton { 568c4843a75SGreg Thelen unsigned long flags; 569c4843a75SGreg Thelen 570bd4c8ce4SAndrew Morton if (page->mapping != mapping) 571bd4c8ce4SAndrew Morton return 0; 572bd4c8ce4SAndrew Morton 573266cf658SDavid Howells if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) 574bd4c8ce4SAndrew Morton return 0; 575bd4c8ce4SAndrew Morton 576c4843a75SGreg Thelen spin_lock_irqsave(&mapping->tree_lock, flags); 577bd4c8ce4SAndrew Morton if (PageDirty(page)) 578bd4c8ce4SAndrew Morton goto failed; 579bd4c8ce4SAndrew Morton 580266cf658SDavid Howells BUG_ON(page_has_private(page)); 58162cccb8cSJohannes Weiner __delete_from_page_cache(page, NULL); 582c4843a75SGreg Thelen spin_unlock_irqrestore(&mapping->tree_lock, flags); 5836072d13cSLinus Torvalds 5846072d13cSLinus Torvalds if (mapping->a_ops->freepage) 5856072d13cSLinus Torvalds mapping->a_ops->freepage(page); 5866072d13cSLinus Torvalds 58709cbfeafSKirill A. Shutemov put_page(page); /* pagecache ref */ 588bd4c8ce4SAndrew Morton return 1; 589bd4c8ce4SAndrew Morton failed: 590c4843a75SGreg Thelen spin_unlock_irqrestore(&mapping->tree_lock, flags); 591bd4c8ce4SAndrew Morton return 0; 592bd4c8ce4SAndrew Morton } 593bd4c8ce4SAndrew Morton 594e3db7691STrond Myklebust static int do_launder_page(struct address_space *mapping, struct page *page) 595e3db7691STrond Myklebust { 596e3db7691STrond Myklebust if (!PageDirty(page)) 597e3db7691STrond Myklebust return 0; 598e3db7691STrond Myklebust if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) 599e3db7691STrond Myklebust return 0; 600e3db7691STrond Myklebust return mapping->a_ops->launder_page(page); 601e3db7691STrond Myklebust } 602e3db7691STrond Myklebust 6031da177e4SLinus Torvalds /** 6041da177e4SLinus Torvalds * invalidate_inode_pages2_range - remove range of pages from an address_space 60567be2dd1SMartin Waitz * @mapping: the address_space 6061da177e4SLinus Torvalds * @start: the page offset 'from' which to invalidate 6071da177e4SLinus Torvalds * @end: the page offset 'to' which to invalidate (inclusive) 6081da177e4SLinus Torvalds * 6091da177e4SLinus Torvalds * Any pages which are found to be mapped into pagetables are unmapped prior to 6101da177e4SLinus Torvalds * invalidation. 6111da177e4SLinus Torvalds * 6126ccfa806SHisashi Hifumi * Returns -EBUSY if any pages could not be invalidated. 6131da177e4SLinus Torvalds */ 6141da177e4SLinus Torvalds int invalidate_inode_pages2_range(struct address_space *mapping, 6151da177e4SLinus Torvalds pgoff_t start, pgoff_t end) 6161da177e4SLinus Torvalds { 6170cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 6181da177e4SLinus Torvalds struct pagevec pvec; 619b85e0effSHugh Dickins pgoff_t index; 6201da177e4SLinus Torvalds int i; 6211da177e4SLinus Torvalds int ret = 0; 6220dd1334fSHisashi Hifumi int ret2 = 0; 6231da177e4SLinus Torvalds int did_range_unmap = 0; 6241da177e4SLinus Torvalds 6253167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 6261da177e4SLinus Torvalds pagevec_init(&pvec, 0); 627b85e0effSHugh Dickins index = start; 6280cd6144aSJohannes Weiner while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, 6290cd6144aSJohannes Weiner min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, 6300cd6144aSJohannes Weiner indices)) { 6317b965e08STrond Myklebust for (i = 0; i < pagevec_count(&pvec); i++) { 6321da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 633b85e0effSHugh Dickins 634b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 6350cd6144aSJohannes Weiner index = indices[i]; 636b85e0effSHugh Dickins if (index > end) 637b85e0effSHugh Dickins break; 6381da177e4SLinus Torvalds 6390cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 640*c6dcf52cSJan Kara if (!invalidate_exceptional_entry2(mapping, 641*c6dcf52cSJan Kara index, page)) 642*c6dcf52cSJan Kara ret = -EBUSY; 6430cd6144aSJohannes Weiner continue; 6440cd6144aSJohannes Weiner } 6450cd6144aSJohannes Weiner 6461da177e4SLinus Torvalds lock_page(page); 6475cbc198aSKirill A. Shutemov WARN_ON(page_to_index(page) != index); 6481da177e4SLinus Torvalds if (page->mapping != mapping) { 6491da177e4SLinus Torvalds unlock_page(page); 6501da177e4SLinus Torvalds continue; 6511da177e4SLinus Torvalds } 6521da177e4SLinus Torvalds wait_on_page_writeback(page); 653d00806b1SNick Piggin if (page_mapped(page)) { 6541da177e4SLinus Torvalds if (!did_range_unmap) { 6551da177e4SLinus Torvalds /* 6561da177e4SLinus Torvalds * Zap the rest of the file in one hit. 6571da177e4SLinus Torvalds */ 6581da177e4SLinus Torvalds unmap_mapping_range(mapping, 65909cbfeafSKirill A. Shutemov (loff_t)index << PAGE_SHIFT, 660b85e0effSHugh Dickins (loff_t)(1 + end - index) 66109cbfeafSKirill A. Shutemov << PAGE_SHIFT, 6621da177e4SLinus Torvalds 0); 6631da177e4SLinus Torvalds did_range_unmap = 1; 6641da177e4SLinus Torvalds } else { 6651da177e4SLinus Torvalds /* 6661da177e4SLinus Torvalds * Just zap this page 6671da177e4SLinus Torvalds */ 6681da177e4SLinus Torvalds unmap_mapping_range(mapping, 66909cbfeafSKirill A. Shutemov (loff_t)index << PAGE_SHIFT, 67009cbfeafSKirill A. Shutemov PAGE_SIZE, 0); 6711da177e4SLinus Torvalds } 6721da177e4SLinus Torvalds } 673d00806b1SNick Piggin BUG_ON(page_mapped(page)); 6740dd1334fSHisashi Hifumi ret2 = do_launder_page(mapping, page); 6750dd1334fSHisashi Hifumi if (ret2 == 0) { 6760dd1334fSHisashi Hifumi if (!invalidate_complete_page2(mapping, page)) 6776ccfa806SHisashi Hifumi ret2 = -EBUSY; 6780dd1334fSHisashi Hifumi } 6790dd1334fSHisashi Hifumi if (ret2 < 0) 6800dd1334fSHisashi Hifumi ret = ret2; 6811da177e4SLinus Torvalds unlock_page(page); 6821da177e4SLinus Torvalds } 6830cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 6841da177e4SLinus Torvalds pagevec_release(&pvec); 6851da177e4SLinus Torvalds cond_resched(); 686b85e0effSHugh Dickins index++; 6871da177e4SLinus Torvalds } 6883167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 6891da177e4SLinus Torvalds return ret; 6901da177e4SLinus Torvalds } 6911da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); 6921da177e4SLinus Torvalds 6931da177e4SLinus Torvalds /** 6941da177e4SLinus Torvalds * invalidate_inode_pages2 - remove all pages from an address_space 69567be2dd1SMartin Waitz * @mapping: the address_space 6961da177e4SLinus Torvalds * 6971da177e4SLinus Torvalds * Any pages which are found to be mapped into pagetables are unmapped prior to 6981da177e4SLinus Torvalds * invalidation. 6991da177e4SLinus Torvalds * 700e9de25ddSPeng Tao * Returns -EBUSY if any pages could not be invalidated. 7011da177e4SLinus Torvalds */ 7021da177e4SLinus Torvalds int invalidate_inode_pages2(struct address_space *mapping) 7031da177e4SLinus Torvalds { 7041da177e4SLinus Torvalds return invalidate_inode_pages2_range(mapping, 0, -1); 7051da177e4SLinus Torvalds } 7061da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(invalidate_inode_pages2); 70725d9e2d1Snpiggin@suse.de 70825d9e2d1Snpiggin@suse.de /** 70925d9e2d1Snpiggin@suse.de * truncate_pagecache - unmap and remove pagecache that has been truncated 71025d9e2d1Snpiggin@suse.de * @inode: inode 7118a549beaSHugh Dickins * @newsize: new file size 71225d9e2d1Snpiggin@suse.de * 71325d9e2d1Snpiggin@suse.de * inode's new i_size must already be written before truncate_pagecache 71425d9e2d1Snpiggin@suse.de * is called. 71525d9e2d1Snpiggin@suse.de * 71625d9e2d1Snpiggin@suse.de * This function should typically be called before the filesystem 71725d9e2d1Snpiggin@suse.de * releases resources associated with the freed range (eg. deallocates 71825d9e2d1Snpiggin@suse.de * blocks). This way, pagecache will always stay logically coherent 71925d9e2d1Snpiggin@suse.de * with on-disk format, and the filesystem would not have to deal with 72025d9e2d1Snpiggin@suse.de * situations such as writepage being called for a page that has already 72125d9e2d1Snpiggin@suse.de * had its underlying blocks deallocated. 72225d9e2d1Snpiggin@suse.de */ 7237caef267SKirill A. Shutemov void truncate_pagecache(struct inode *inode, loff_t newsize) 72425d9e2d1Snpiggin@suse.de { 72525d9e2d1Snpiggin@suse.de struct address_space *mapping = inode->i_mapping; 7268a549beaSHugh Dickins loff_t holebegin = round_up(newsize, PAGE_SIZE); 72725d9e2d1Snpiggin@suse.de 72825d9e2d1Snpiggin@suse.de /* 72925d9e2d1Snpiggin@suse.de * unmap_mapping_range is called twice, first simply for 73025d9e2d1Snpiggin@suse.de * efficiency so that truncate_inode_pages does fewer 73125d9e2d1Snpiggin@suse.de * single-page unmaps. However after this first call, and 73225d9e2d1Snpiggin@suse.de * before truncate_inode_pages finishes, it is possible for 73325d9e2d1Snpiggin@suse.de * private pages to be COWed, which remain after 73425d9e2d1Snpiggin@suse.de * truncate_inode_pages finishes, hence the second 73525d9e2d1Snpiggin@suse.de * unmap_mapping_range call must be made for correctness. 73625d9e2d1Snpiggin@suse.de */ 7378a549beaSHugh Dickins unmap_mapping_range(mapping, holebegin, 0, 1); 7388a549beaSHugh Dickins truncate_inode_pages(mapping, newsize); 7398a549beaSHugh Dickins unmap_mapping_range(mapping, holebegin, 0, 1); 74025d9e2d1Snpiggin@suse.de } 74125d9e2d1Snpiggin@suse.de EXPORT_SYMBOL(truncate_pagecache); 74225d9e2d1Snpiggin@suse.de 74325d9e2d1Snpiggin@suse.de /** 7442c27c65eSChristoph Hellwig * truncate_setsize - update inode and pagecache for a new file size 7452c27c65eSChristoph Hellwig * @inode: inode 7462c27c65eSChristoph Hellwig * @newsize: new file size 7472c27c65eSChristoph Hellwig * 748382e27daSJan Kara * truncate_setsize updates i_size and performs pagecache truncation (if 749382e27daSJan Kara * necessary) to @newsize. It will be typically be called from the filesystem's 750382e27daSJan Kara * setattr function when ATTR_SIZE is passed in. 7512c27c65eSChristoph Hellwig * 75277783d06SJan Kara * Must be called with a lock serializing truncates and writes (generally 75377783d06SJan Kara * i_mutex but e.g. xfs uses a different lock) and before all filesystem 75477783d06SJan Kara * specific block truncation has been performed. 7552c27c65eSChristoph Hellwig */ 7562c27c65eSChristoph Hellwig void truncate_setsize(struct inode *inode, loff_t newsize) 7572c27c65eSChristoph Hellwig { 75890a80202SJan Kara loff_t oldsize = inode->i_size; 75990a80202SJan Kara 7602c27c65eSChristoph Hellwig i_size_write(inode, newsize); 76190a80202SJan Kara if (newsize > oldsize) 76290a80202SJan Kara pagecache_isize_extended(inode, oldsize, newsize); 7637caef267SKirill A. Shutemov truncate_pagecache(inode, newsize); 7642c27c65eSChristoph Hellwig } 7652c27c65eSChristoph Hellwig EXPORT_SYMBOL(truncate_setsize); 7662c27c65eSChristoph Hellwig 7672c27c65eSChristoph Hellwig /** 76890a80202SJan Kara * pagecache_isize_extended - update pagecache after extension of i_size 76990a80202SJan Kara * @inode: inode for which i_size was extended 77090a80202SJan Kara * @from: original inode size 77190a80202SJan Kara * @to: new inode size 77290a80202SJan Kara * 77390a80202SJan Kara * Handle extension of inode size either caused by extending truncate or by 77490a80202SJan Kara * write starting after current i_size. We mark the page straddling current 77590a80202SJan Kara * i_size RO so that page_mkwrite() is called on the nearest write access to 77690a80202SJan Kara * the page. This way filesystem can be sure that page_mkwrite() is called on 77790a80202SJan Kara * the page before user writes to the page via mmap after the i_size has been 77890a80202SJan Kara * changed. 77990a80202SJan Kara * 78090a80202SJan Kara * The function must be called after i_size is updated so that page fault 78190a80202SJan Kara * coming after we unlock the page will already see the new i_size. 78290a80202SJan Kara * The function must be called while we still hold i_mutex - this not only 78390a80202SJan Kara * makes sure i_size is stable but also that userspace cannot observe new 78490a80202SJan Kara * i_size value before we are prepared to store mmap writes at new inode size. 78590a80202SJan Kara */ 78690a80202SJan Kara void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) 78790a80202SJan Kara { 78890a80202SJan Kara int bsize = 1 << inode->i_blkbits; 78990a80202SJan Kara loff_t rounded_from; 79090a80202SJan Kara struct page *page; 79190a80202SJan Kara pgoff_t index; 79290a80202SJan Kara 79390a80202SJan Kara WARN_ON(to > inode->i_size); 79490a80202SJan Kara 79509cbfeafSKirill A. Shutemov if (from >= to || bsize == PAGE_SIZE) 79690a80202SJan Kara return; 79790a80202SJan Kara /* Page straddling @from will not have any hole block created? */ 79890a80202SJan Kara rounded_from = round_up(from, bsize); 79909cbfeafSKirill A. Shutemov if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1))) 80090a80202SJan Kara return; 80190a80202SJan Kara 80209cbfeafSKirill A. Shutemov index = from >> PAGE_SHIFT; 80390a80202SJan Kara page = find_lock_page(inode->i_mapping, index); 80490a80202SJan Kara /* Page not cached? Nothing to do */ 80590a80202SJan Kara if (!page) 80690a80202SJan Kara return; 80790a80202SJan Kara /* 80890a80202SJan Kara * See clear_page_dirty_for_io() for details why set_page_dirty() 80990a80202SJan Kara * is needed. 81090a80202SJan Kara */ 81190a80202SJan Kara if (page_mkclean(page)) 81290a80202SJan Kara set_page_dirty(page); 81390a80202SJan Kara unlock_page(page); 81409cbfeafSKirill A. Shutemov put_page(page); 81590a80202SJan Kara } 81690a80202SJan Kara EXPORT_SYMBOL(pagecache_isize_extended); 81790a80202SJan Kara 81890a80202SJan Kara /** 819623e3db9SHugh Dickins * truncate_pagecache_range - unmap and remove pagecache that is hole-punched 820623e3db9SHugh Dickins * @inode: inode 821623e3db9SHugh Dickins * @lstart: offset of beginning of hole 822623e3db9SHugh Dickins * @lend: offset of last byte of hole 823623e3db9SHugh Dickins * 824623e3db9SHugh Dickins * This function should typically be called before the filesystem 825623e3db9SHugh Dickins * releases resources associated with the freed range (eg. deallocates 826623e3db9SHugh Dickins * blocks). This way, pagecache will always stay logically coherent 827623e3db9SHugh Dickins * with on-disk format, and the filesystem would not have to deal with 828623e3db9SHugh Dickins * situations such as writepage being called for a page that has already 829623e3db9SHugh Dickins * had its underlying blocks deallocated. 830623e3db9SHugh Dickins */ 831623e3db9SHugh Dickins void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) 832623e3db9SHugh Dickins { 833623e3db9SHugh Dickins struct address_space *mapping = inode->i_mapping; 834623e3db9SHugh Dickins loff_t unmap_start = round_up(lstart, PAGE_SIZE); 835623e3db9SHugh Dickins loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; 836623e3db9SHugh Dickins /* 837623e3db9SHugh Dickins * This rounding is currently just for example: unmap_mapping_range 838623e3db9SHugh Dickins * expands its hole outwards, whereas we want it to contract the hole 839623e3db9SHugh Dickins * inwards. However, existing callers of truncate_pagecache_range are 8405a720394SLukas Czerner * doing their own page rounding first. Note that unmap_mapping_range 8415a720394SLukas Czerner * allows holelen 0 for all, and we allow lend -1 for end of file. 842623e3db9SHugh Dickins */ 843623e3db9SHugh Dickins 844623e3db9SHugh Dickins /* 845623e3db9SHugh Dickins * Unlike in truncate_pagecache, unmap_mapping_range is called only 846623e3db9SHugh Dickins * once (before truncating pagecache), and without "even_cows" flag: 847623e3db9SHugh Dickins * hole-punching should not remove private COWed pages from the hole. 848623e3db9SHugh Dickins */ 849623e3db9SHugh Dickins if ((u64)unmap_end > (u64)unmap_start) 850623e3db9SHugh Dickins unmap_mapping_range(mapping, unmap_start, 851623e3db9SHugh Dickins 1 + unmap_end - unmap_start, 0); 852623e3db9SHugh Dickins truncate_inode_pages_range(mapping, lstart, lend); 853623e3db9SHugh Dickins } 854623e3db9SHugh Dickins EXPORT_SYMBOL(truncate_pagecache_range); 855