1457c8996SThomas Gleixner // SPDX-License-Identifier: GPL-2.0-only 21da177e4SLinus Torvalds /* 31da177e4SLinus Torvalds * mm/truncate.c - code for taking down pages from address_spaces 41da177e4SLinus Torvalds * 51da177e4SLinus Torvalds * Copyright (C) 2002, Linus Torvalds 61da177e4SLinus Torvalds * 7e1f8e874SFrancois Cami * 10Sep2002 Andrew Morton 81da177e4SLinus Torvalds * Initial version. 91da177e4SLinus Torvalds */ 101da177e4SLinus Torvalds 111da177e4SLinus Torvalds #include <linux/kernel.h> 124af3c9ccSAlexey Dobriyan #include <linux/backing-dev.h> 13f9fe48beSRoss Zwisler #include <linux/dax.h> 145a0e3ad6STejun Heo #include <linux/gfp.h> 151da177e4SLinus Torvalds #include <linux/mm.h> 160fd0e6b0SNick Piggin #include <linux/swap.h> 17b95f1b31SPaul Gortmaker #include <linux/export.h> 181da177e4SLinus Torvalds #include <linux/pagemap.h> 1901f2705dSNate Diller #include <linux/highmem.h> 201da177e4SLinus Torvalds #include <linux/pagevec.h> 21e08748ceSAndrew Morton #include <linux/task_io_accounting_ops.h> 221da177e4SLinus Torvalds #include <linux/buffer_head.h> /* grr. try_to_release_page, 23aaa4059bSJan Kara do_invalidatepage */ 243a4f8a0bSHugh Dickins #include <linux/shmem_fs.h> 25c515e1fdSDan Magenheimer #include <linux/cleancache.h> 2690a80202SJan Kara #include <linux/rmap.h> 27ba470de4SRik van Riel #include "internal.h" 281da177e4SLinus Torvalds 29f2187599SMel Gorman /* 30f2187599SMel Gorman * Regular page slots are stabilized by the page lock even without the tree 31f2187599SMel Gorman * itself locked. These unlocked entries need verification under the tree 32f2187599SMel Gorman * lock. 33f2187599SMel Gorman */ 34f2187599SMel Gorman static inline void __clear_shadow_entry(struct address_space *mapping, 35f2187599SMel Gorman pgoff_t index, void *entry) 360cd6144aSJohannes Weiner { 3769b6c131SMatthew Wilcox XA_STATE(xas, &mapping->i_pages, index); 38449dd698SJohannes Weiner 3969b6c131SMatthew Wilcox xas_set_update(&xas, workingset_update_node); 4069b6c131SMatthew Wilcox if (xas_load(&xas) != entry) 41f2187599SMel Gorman return; 4269b6c131SMatthew Wilcox xas_store(&xas, NULL); 43f9fe48beSRoss Zwisler mapping->nrexceptional--; 44f2187599SMel Gorman } 45f2187599SMel Gorman 46f2187599SMel Gorman static void clear_shadow_entry(struct address_space *mapping, pgoff_t index, 47f2187599SMel Gorman void *entry) 48f2187599SMel Gorman { 49b93b0163SMatthew Wilcox xa_lock_irq(&mapping->i_pages); 50f2187599SMel Gorman __clear_shadow_entry(mapping, index, entry); 51b93b0163SMatthew Wilcox xa_unlock_irq(&mapping->i_pages); 520cd6144aSJohannes Weiner } 531da177e4SLinus Torvalds 54c6dcf52cSJan Kara /* 55f2187599SMel Gorman * Unconditionally remove exceptional entries. Usually called from truncate 56f2187599SMel Gorman * path. Note that the pagevec may be altered by this function by removing 57f2187599SMel Gorman * exceptional entries similar to what pagevec_remove_exceptionals does. 58c6dcf52cSJan Kara */ 59f2187599SMel Gorman static void truncate_exceptional_pvec_entries(struct address_space *mapping, 60f2187599SMel Gorman struct pagevec *pvec, pgoff_t *indices, 61f2187599SMel Gorman pgoff_t end) 62c6dcf52cSJan Kara { 63f2187599SMel Gorman int i, j; 64f2187599SMel Gorman bool dax, lock; 65f2187599SMel Gorman 66c6dcf52cSJan Kara /* Handled by shmem itself */ 67c6dcf52cSJan Kara if (shmem_mapping(mapping)) 68c6dcf52cSJan Kara return; 69c6dcf52cSJan Kara 70f2187599SMel Gorman for (j = 0; j < pagevec_count(pvec); j++) 713159f943SMatthew Wilcox if (xa_is_value(pvec->pages[j])) 72f2187599SMel Gorman break; 73f2187599SMel Gorman 74f2187599SMel Gorman if (j == pagevec_count(pvec)) 75c6dcf52cSJan Kara return; 76f2187599SMel Gorman 77f2187599SMel Gorman dax = dax_mapping(mapping); 78f2187599SMel Gorman lock = !dax && indices[j] < end; 79f2187599SMel Gorman if (lock) 80b93b0163SMatthew Wilcox xa_lock_irq(&mapping->i_pages); 81f2187599SMel Gorman 82f2187599SMel Gorman for (i = j; i < pagevec_count(pvec); i++) { 83f2187599SMel Gorman struct page *page = pvec->pages[i]; 84f2187599SMel Gorman pgoff_t index = indices[i]; 85f2187599SMel Gorman 863159f943SMatthew Wilcox if (!xa_is_value(page)) { 87f2187599SMel Gorman pvec->pages[j++] = page; 88f2187599SMel Gorman continue; 89c6dcf52cSJan Kara } 90f2187599SMel Gorman 91f2187599SMel Gorman if (index >= end) 92f2187599SMel Gorman continue; 93f2187599SMel Gorman 94f2187599SMel Gorman if (unlikely(dax)) { 95f2187599SMel Gorman dax_delete_mapping_entry(mapping, index); 96f2187599SMel Gorman continue; 97f2187599SMel Gorman } 98f2187599SMel Gorman 99f2187599SMel Gorman __clear_shadow_entry(mapping, index, page); 100f2187599SMel Gorman } 101f2187599SMel Gorman 102f2187599SMel Gorman if (lock) 103b93b0163SMatthew Wilcox xa_unlock_irq(&mapping->i_pages); 104f2187599SMel Gorman pvec->nr = j; 105c6dcf52cSJan Kara } 106c6dcf52cSJan Kara 107c6dcf52cSJan Kara /* 108c6dcf52cSJan Kara * Invalidate exceptional entry if easily possible. This handles exceptional 1094636e70bSRoss Zwisler * entries for invalidate_inode_pages(). 110c6dcf52cSJan Kara */ 111c6dcf52cSJan Kara static int invalidate_exceptional_entry(struct address_space *mapping, 112c6dcf52cSJan Kara pgoff_t index, void *entry) 113c6dcf52cSJan Kara { 1144636e70bSRoss Zwisler /* Handled by shmem itself, or for DAX we do nothing. */ 1154636e70bSRoss Zwisler if (shmem_mapping(mapping) || dax_mapping(mapping)) 116c6dcf52cSJan Kara return 1; 117c6dcf52cSJan Kara clear_shadow_entry(mapping, index, entry); 118c6dcf52cSJan Kara return 1; 119c6dcf52cSJan Kara } 120c6dcf52cSJan Kara 121c6dcf52cSJan Kara /* 122c6dcf52cSJan Kara * Invalidate exceptional entry if clean. This handles exceptional entries for 123c6dcf52cSJan Kara * invalidate_inode_pages2() so for DAX it evicts only clean entries. 124c6dcf52cSJan Kara */ 125c6dcf52cSJan Kara static int invalidate_exceptional_entry2(struct address_space *mapping, 126c6dcf52cSJan Kara pgoff_t index, void *entry) 127c6dcf52cSJan Kara { 128c6dcf52cSJan Kara /* Handled by shmem itself */ 129c6dcf52cSJan Kara if (shmem_mapping(mapping)) 130c6dcf52cSJan Kara return 1; 131c6dcf52cSJan Kara if (dax_mapping(mapping)) 132c6dcf52cSJan Kara return dax_invalidate_mapping_entry_sync(mapping, index); 133c6dcf52cSJan Kara clear_shadow_entry(mapping, index, entry); 134c6dcf52cSJan Kara return 1; 135c6dcf52cSJan Kara } 136c6dcf52cSJan Kara 137cf9a2ae8SDavid Howells /** 13828bc44d7SFengguang Wu * do_invalidatepage - invalidate part or all of a page 139cf9a2ae8SDavid Howells * @page: the page which is affected 140d47992f8SLukas Czerner * @offset: start of the range to invalidate 141d47992f8SLukas Czerner * @length: length of the range to invalidate 142cf9a2ae8SDavid Howells * 143cf9a2ae8SDavid Howells * do_invalidatepage() is called when all or part of the page has become 144cf9a2ae8SDavid Howells * invalidated by a truncate operation. 145cf9a2ae8SDavid Howells * 146cf9a2ae8SDavid Howells * do_invalidatepage() does not have to release all buffers, but it must 147cf9a2ae8SDavid Howells * ensure that no dirty buffer is left outside @offset and that no I/O 148cf9a2ae8SDavid Howells * is underway against any of the blocks which are outside the truncation 149cf9a2ae8SDavid Howells * point. Because the caller is about to free (and possibly reuse) those 150cf9a2ae8SDavid Howells * blocks on-disk. 151cf9a2ae8SDavid Howells */ 152d47992f8SLukas Czerner void do_invalidatepage(struct page *page, unsigned int offset, 153d47992f8SLukas Czerner unsigned int length) 154cf9a2ae8SDavid Howells { 155d47992f8SLukas Czerner void (*invalidatepage)(struct page *, unsigned int, unsigned int); 156d47992f8SLukas Czerner 157cf9a2ae8SDavid Howells invalidatepage = page->mapping->a_ops->invalidatepage; 1589361401eSDavid Howells #ifdef CONFIG_BLOCK 159cf9a2ae8SDavid Howells if (!invalidatepage) 160cf9a2ae8SDavid Howells invalidatepage = block_invalidatepage; 1619361401eSDavid Howells #endif 162cf9a2ae8SDavid Howells if (invalidatepage) 163d47992f8SLukas Czerner (*invalidatepage)(page, offset, length); 164cf9a2ae8SDavid Howells } 165cf9a2ae8SDavid Howells 166ecdfc978SLinus Torvalds /* 1671da177e4SLinus Torvalds * If truncate cannot remove the fs-private metadata from the page, the page 16862e1c553SShaohua Li * becomes orphaned. It will be left on the LRU and may even be mapped into 16954cb8821SNick Piggin * user pagetables if we're racing with filemap_fault(). 1701da177e4SLinus Torvalds * 171fc3a5ac5SMatthew Wilcox (Oracle) * We need to bail out if page->mapping is no longer equal to the original 1721da177e4SLinus Torvalds * mapping. This happens a) when the VM reclaimed the page while we waited on 173fc0ecff6SAndrew Morton * its lock, b) when a concurrent invalidate_mapping_pages got there first and 1741da177e4SLinus Torvalds * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. 1751da177e4SLinus Torvalds */ 1769f4e41f4SJan Kara static void 1779f4e41f4SJan Kara truncate_cleanup_page(struct address_space *mapping, struct page *page) 1781da177e4SLinus Torvalds { 1799f4e41f4SJan Kara if (page_mapped(page)) { 180fc3a5ac5SMatthew Wilcox (Oracle) unsigned int nr = thp_nr_pages(page); 181977fbdcdSMatthew Wilcox unmap_mapping_pages(mapping, page->index, nr, false); 1829f4e41f4SJan Kara } 1831da177e4SLinus Torvalds 184266cf658SDavid Howells if (page_has_private(page)) 185fc3a5ac5SMatthew Wilcox (Oracle) do_invalidatepage(page, 0, thp_size(page)); 1861da177e4SLinus Torvalds 187b9ea2515SKonstantin Khlebnikov /* 188b9ea2515SKonstantin Khlebnikov * Some filesystems seem to re-dirty the page even after 189b9ea2515SKonstantin Khlebnikov * the VM has canceled the dirty bit (eg ext3 journaling). 190b9ea2515SKonstantin Khlebnikov * Hence dirty accounting check is placed after invalidation. 191b9ea2515SKonstantin Khlebnikov */ 19211f81becSTejun Heo cancel_dirty_page(page); 1931da177e4SLinus Torvalds ClearPageMappedToDisk(page); 1941da177e4SLinus Torvalds } 1951da177e4SLinus Torvalds 1961da177e4SLinus Torvalds /* 197fc0ecff6SAndrew Morton * This is for invalidate_mapping_pages(). That function can be called at 1981da177e4SLinus Torvalds * any time, and is not supposed to throw away dirty pages. But pages can 1990fd0e6b0SNick Piggin * be marked dirty at any time too, so use remove_mapping which safely 2000fd0e6b0SNick Piggin * discards clean, unused pages. 2011da177e4SLinus Torvalds * 2021da177e4SLinus Torvalds * Returns non-zero if the page was successfully invalidated. 2031da177e4SLinus Torvalds */ 2041da177e4SLinus Torvalds static int 2051da177e4SLinus Torvalds invalidate_complete_page(struct address_space *mapping, struct page *page) 2061da177e4SLinus Torvalds { 2070fd0e6b0SNick Piggin int ret; 2080fd0e6b0SNick Piggin 2091da177e4SLinus Torvalds if (page->mapping != mapping) 2101da177e4SLinus Torvalds return 0; 2111da177e4SLinus Torvalds 212266cf658SDavid Howells if (page_has_private(page) && !try_to_release_page(page, 0)) 2131da177e4SLinus Torvalds return 0; 2141da177e4SLinus Torvalds 2150fd0e6b0SNick Piggin ret = remove_mapping(mapping, page); 2160fd0e6b0SNick Piggin 2170fd0e6b0SNick Piggin return ret; 2181da177e4SLinus Torvalds } 2191da177e4SLinus Torvalds 220750b4987SNick Piggin int truncate_inode_page(struct address_space *mapping, struct page *page) 221750b4987SNick Piggin { 222fc127da0SKirill A. Shutemov VM_BUG_ON_PAGE(PageTail(page), page); 223fc127da0SKirill A. Shutemov 2249f4e41f4SJan Kara if (page->mapping != mapping) 2259f4e41f4SJan Kara return -EIO; 2269f4e41f4SJan Kara 2279f4e41f4SJan Kara truncate_cleanup_page(mapping, page); 2289f4e41f4SJan Kara delete_from_page_cache(page); 2299f4e41f4SJan Kara return 0; 230750b4987SNick Piggin } 231750b4987SNick Piggin 23283f78668SWu Fengguang /* 23325718736SAndi Kleen * Used to get rid of pages on hardware memory corruption. 23425718736SAndi Kleen */ 23525718736SAndi Kleen int generic_error_remove_page(struct address_space *mapping, struct page *page) 23625718736SAndi Kleen { 23725718736SAndi Kleen if (!mapping) 23825718736SAndi Kleen return -EINVAL; 23925718736SAndi Kleen /* 24025718736SAndi Kleen * Only punch for normal data pages for now. 24125718736SAndi Kleen * Handling other types like directories would need more auditing. 24225718736SAndi Kleen */ 24325718736SAndi Kleen if (!S_ISREG(mapping->host->i_mode)) 24425718736SAndi Kleen return -EIO; 24525718736SAndi Kleen return truncate_inode_page(mapping, page); 24625718736SAndi Kleen } 24725718736SAndi Kleen EXPORT_SYMBOL(generic_error_remove_page); 24825718736SAndi Kleen 24925718736SAndi Kleen /* 25083f78668SWu Fengguang * Safely invalidate one page from its pagecache mapping. 25183f78668SWu Fengguang * It only drops clean, unused pages. The page must be locked. 25283f78668SWu Fengguang * 25383f78668SWu Fengguang * Returns 1 if the page is successfully invalidated, otherwise 0. 25483f78668SWu Fengguang */ 25583f78668SWu Fengguang int invalidate_inode_page(struct page *page) 25683f78668SWu Fengguang { 25783f78668SWu Fengguang struct address_space *mapping = page_mapping(page); 25883f78668SWu Fengguang if (!mapping) 25983f78668SWu Fengguang return 0; 26083f78668SWu Fengguang if (PageDirty(page) || PageWriteback(page)) 26183f78668SWu Fengguang return 0; 26283f78668SWu Fengguang if (page_mapped(page)) 26383f78668SWu Fengguang return 0; 26483f78668SWu Fengguang return invalidate_complete_page(mapping, page); 26583f78668SWu Fengguang } 26683f78668SWu Fengguang 2671da177e4SLinus Torvalds /** 26873c1e204SLiu Bo * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets 2691da177e4SLinus Torvalds * @mapping: mapping to truncate 2701da177e4SLinus Torvalds * @lstart: offset from which to truncate 2715a720394SLukas Czerner * @lend: offset to which to truncate (inclusive) 2721da177e4SLinus Torvalds * 273d7339071SHans Reiser * Truncate the page cache, removing the pages that are between 2745a720394SLukas Czerner * specified offsets (and zeroing out partial pages 2755a720394SLukas Czerner * if lstart or lend + 1 is not page aligned). 2761da177e4SLinus Torvalds * 2771da177e4SLinus Torvalds * Truncate takes two passes - the first pass is nonblocking. It will not 2781da177e4SLinus Torvalds * block on page locks and it will not block on writeback. The second pass 2791da177e4SLinus Torvalds * will wait. This is to prevent as much IO as possible in the affected region. 2801da177e4SLinus Torvalds * The first pass will remove most pages, so the search cost of the second pass 2811da177e4SLinus Torvalds * is low. 2821da177e4SLinus Torvalds * 2831da177e4SLinus Torvalds * We pass down the cache-hot hint to the page freeing code. Even if the 2841da177e4SLinus Torvalds * mapping is large, it is probably the case that the final pages are the most 2851da177e4SLinus Torvalds * recently touched, and freeing happens in ascending file offset order. 2865a720394SLukas Czerner * 2875a720394SLukas Czerner * Note that since ->invalidatepage() accepts range to invalidate 2885a720394SLukas Czerner * truncate_inode_pages_range is able to handle cases where lend + 1 is not 2895a720394SLukas Czerner * page aligned properly. 2901da177e4SLinus Torvalds */ 291d7339071SHans Reiser void truncate_inode_pages_range(struct address_space *mapping, 292d7339071SHans Reiser loff_t lstart, loff_t lend) 2931da177e4SLinus Torvalds { 2945a720394SLukas Czerner pgoff_t start; /* inclusive */ 2955a720394SLukas Czerner pgoff_t end; /* exclusive */ 2965a720394SLukas Czerner unsigned int partial_start; /* inclusive */ 2975a720394SLukas Czerner unsigned int partial_end; /* exclusive */ 2981da177e4SLinus Torvalds struct pagevec pvec; 2990cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 300b85e0effSHugh Dickins pgoff_t index; 3011da177e4SLinus Torvalds int i; 3021da177e4SLinus Torvalds 303f9fe48beSRoss Zwisler if (mapping->nrpages == 0 && mapping->nrexceptional == 0) 30434ccb69eSAndrey Ryabinin goto out; 3051da177e4SLinus Torvalds 3065a720394SLukas Czerner /* Offsets within partial pages */ 30709cbfeafSKirill A. Shutemov partial_start = lstart & (PAGE_SIZE - 1); 30809cbfeafSKirill A. Shutemov partial_end = (lend + 1) & (PAGE_SIZE - 1); 3095a720394SLukas Czerner 3105a720394SLukas Czerner /* 3115a720394SLukas Czerner * 'start' and 'end' always covers the range of pages to be fully 3125a720394SLukas Czerner * truncated. Partial pages are covered with 'partial_start' at the 3135a720394SLukas Czerner * start of the range and 'partial_end' at the end of the range. 3145a720394SLukas Czerner * Note that 'end' is exclusive while 'lend' is inclusive. 3155a720394SLukas Czerner */ 31609cbfeafSKirill A. Shutemov start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; 3175a720394SLukas Czerner if (lend == -1) 3185a720394SLukas Czerner /* 3195a720394SLukas Czerner * lend == -1 indicates end-of-file so we have to set 'end' 3205a720394SLukas Czerner * to the highest possible pgoff_t and since the type is 3215a720394SLukas Czerner * unsigned we're using -1. 3225a720394SLukas Czerner */ 3235a720394SLukas Czerner end = -1; 3245a720394SLukas Czerner else 32509cbfeafSKirill A. Shutemov end = (lend + 1) >> PAGE_SHIFT; 326d7339071SHans Reiser 32786679820SMel Gorman pagevec_init(&pvec); 328b85e0effSHugh Dickins index = start; 3290cd6144aSJohannes Weiner while (index < end && pagevec_lookup_entries(&pvec, mapping, index, 3300cd6144aSJohannes Weiner min(end - index, (pgoff_t)PAGEVEC_SIZE), 3310cd6144aSJohannes Weiner indices)) { 332aa65c29cSJan Kara /* 333aa65c29cSJan Kara * Pagevec array has exceptional entries and we may also fail 334aa65c29cSJan Kara * to lock some pages. So we store pages that can be deleted 335aa65c29cSJan Kara * in a new pagevec. 336aa65c29cSJan Kara */ 337aa65c29cSJan Kara struct pagevec locked_pvec; 338aa65c29cSJan Kara 33986679820SMel Gorman pagevec_init(&locked_pvec); 3401da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 3411da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 3421da177e4SLinus Torvalds 343b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 3440cd6144aSJohannes Weiner index = indices[i]; 3455a720394SLukas Czerner if (index >= end) 346d7339071SHans Reiser break; 347d7339071SHans Reiser 3483159f943SMatthew Wilcox if (xa_is_value(page)) 3490cd6144aSJohannes Weiner continue; 3500cd6144aSJohannes Weiner 351529ae9aaSNick Piggin if (!trylock_page(page)) 3521da177e4SLinus Torvalds continue; 3535cbc198aSKirill A. Shutemov WARN_ON(page_to_index(page) != index); 3541da177e4SLinus Torvalds if (PageWriteback(page)) { 3551da177e4SLinus Torvalds unlock_page(page); 3561da177e4SLinus Torvalds continue; 3571da177e4SLinus Torvalds } 358aa65c29cSJan Kara if (page->mapping != mapping) { 3591da177e4SLinus Torvalds unlock_page(page); 360aa65c29cSJan Kara continue; 3611da177e4SLinus Torvalds } 362aa65c29cSJan Kara pagevec_add(&locked_pvec, page); 363aa65c29cSJan Kara } 364aa65c29cSJan Kara for (i = 0; i < pagevec_count(&locked_pvec); i++) 365aa65c29cSJan Kara truncate_cleanup_page(mapping, locked_pvec.pages[i]); 366aa65c29cSJan Kara delete_from_page_cache_batch(mapping, &locked_pvec); 367aa65c29cSJan Kara for (i = 0; i < pagevec_count(&locked_pvec); i++) 368aa65c29cSJan Kara unlock_page(locked_pvec.pages[i]); 369f2187599SMel Gorman truncate_exceptional_pvec_entries(mapping, &pvec, indices, end); 3701da177e4SLinus Torvalds pagevec_release(&pvec); 3711da177e4SLinus Torvalds cond_resched(); 372b85e0effSHugh Dickins index++; 3731da177e4SLinus Torvalds } 3745a720394SLukas Czerner if (partial_start) { 3751da177e4SLinus Torvalds struct page *page = find_lock_page(mapping, start - 1); 3761da177e4SLinus Torvalds if (page) { 37709cbfeafSKirill A. Shutemov unsigned int top = PAGE_SIZE; 3785a720394SLukas Czerner if (start > end) { 3795a720394SLukas Czerner /* Truncation within a single page */ 3805a720394SLukas Czerner top = partial_end; 3815a720394SLukas Czerner partial_end = 0; 3825a720394SLukas Czerner } 3831da177e4SLinus Torvalds wait_on_page_writeback(page); 3845a720394SLukas Czerner zero_user_segment(page, partial_start, top); 3855a720394SLukas Czerner cleancache_invalidate_page(mapping, page); 3865a720394SLukas Czerner if (page_has_private(page)) 3875a720394SLukas Czerner do_invalidatepage(page, partial_start, 3885a720394SLukas Czerner top - partial_start); 3891da177e4SLinus Torvalds unlock_page(page); 39009cbfeafSKirill A. Shutemov put_page(page); 3911da177e4SLinus Torvalds } 3921da177e4SLinus Torvalds } 3935a720394SLukas Czerner if (partial_end) { 3945a720394SLukas Czerner struct page *page = find_lock_page(mapping, end); 3955a720394SLukas Czerner if (page) { 3965a720394SLukas Czerner wait_on_page_writeback(page); 3975a720394SLukas Czerner zero_user_segment(page, 0, partial_end); 3985a720394SLukas Czerner cleancache_invalidate_page(mapping, page); 3995a720394SLukas Czerner if (page_has_private(page)) 4005a720394SLukas Czerner do_invalidatepage(page, 0, 4015a720394SLukas Czerner partial_end); 4025a720394SLukas Czerner unlock_page(page); 40309cbfeafSKirill A. Shutemov put_page(page); 4045a720394SLukas Czerner } 4055a720394SLukas Czerner } 4065a720394SLukas Czerner /* 4075a720394SLukas Czerner * If the truncation happened within a single page no pages 4085a720394SLukas Czerner * will be released, just zeroed, so we can bail out now. 4095a720394SLukas Czerner */ 4105a720394SLukas Czerner if (start >= end) 41134ccb69eSAndrey Ryabinin goto out; 4121da177e4SLinus Torvalds 413b85e0effSHugh Dickins index = start; 4141da177e4SLinus Torvalds for ( ; ; ) { 4151da177e4SLinus Torvalds cond_resched(); 4160cd6144aSJohannes Weiner if (!pagevec_lookup_entries(&pvec, mapping, index, 417792ceaefSHugh Dickins min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) { 418792ceaefSHugh Dickins /* If all gone from start onwards, we're done */ 419b85e0effSHugh Dickins if (index == start) 4201da177e4SLinus Torvalds break; 421792ceaefSHugh Dickins /* Otherwise restart to make sure all gone */ 422b85e0effSHugh Dickins index = start; 4231da177e4SLinus Torvalds continue; 4241da177e4SLinus Torvalds } 4250cd6144aSJohannes Weiner if (index == start && indices[0] >= end) { 426792ceaefSHugh Dickins /* All gone out of hole to be punched, we're done */ 4270cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 428d7339071SHans Reiser pagevec_release(&pvec); 429d7339071SHans Reiser break; 430d7339071SHans Reiser } 431f2187599SMel Gorman 4321da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 4331da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 4341da177e4SLinus Torvalds 435b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 4360cd6144aSJohannes Weiner index = indices[i]; 437792ceaefSHugh Dickins if (index >= end) { 438792ceaefSHugh Dickins /* Restart punch to make sure all gone */ 439792ceaefSHugh Dickins index = start - 1; 440d7339071SHans Reiser break; 441792ceaefSHugh Dickins } 442b85e0effSHugh Dickins 4433159f943SMatthew Wilcox if (xa_is_value(page)) 4440cd6144aSJohannes Weiner continue; 4450cd6144aSJohannes Weiner 4461da177e4SLinus Torvalds lock_page(page); 4475cbc198aSKirill A. Shutemov WARN_ON(page_to_index(page) != index); 4481da177e4SLinus Torvalds wait_on_page_writeback(page); 449750b4987SNick Piggin truncate_inode_page(mapping, page); 4501da177e4SLinus Torvalds unlock_page(page); 4511da177e4SLinus Torvalds } 452f2187599SMel Gorman truncate_exceptional_pvec_entries(mapping, &pvec, indices, end); 4531da177e4SLinus Torvalds pagevec_release(&pvec); 454b85e0effSHugh Dickins index++; 4551da177e4SLinus Torvalds } 45634ccb69eSAndrey Ryabinin 45734ccb69eSAndrey Ryabinin out: 4583167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 4591da177e4SLinus Torvalds } 460d7339071SHans Reiser EXPORT_SYMBOL(truncate_inode_pages_range); 4611da177e4SLinus Torvalds 462d7339071SHans Reiser /** 463d7339071SHans Reiser * truncate_inode_pages - truncate *all* the pages from an offset 464d7339071SHans Reiser * @mapping: mapping to truncate 465d7339071SHans Reiser * @lstart: offset from which to truncate 466d7339071SHans Reiser * 4671b1dcc1bSJes Sorensen * Called under (and serialised by) inode->i_mutex. 46808142579SJan Kara * 46908142579SJan Kara * Note: When this function returns, there can be a page in the process of 47008142579SJan Kara * deletion (inside __delete_from_page_cache()) in the specified range. Thus 47108142579SJan Kara * mapping->nrpages can be non-zero when this function returns even after 47208142579SJan Kara * truncation of the whole mapping. 473d7339071SHans Reiser */ 474d7339071SHans Reiser void truncate_inode_pages(struct address_space *mapping, loff_t lstart) 475d7339071SHans Reiser { 476d7339071SHans Reiser truncate_inode_pages_range(mapping, lstart, (loff_t)-1); 477d7339071SHans Reiser } 4781da177e4SLinus Torvalds EXPORT_SYMBOL(truncate_inode_pages); 4791da177e4SLinus Torvalds 48028697355SMike Waychison /** 48191b0abe3SJohannes Weiner * truncate_inode_pages_final - truncate *all* pages before inode dies 48291b0abe3SJohannes Weiner * @mapping: mapping to truncate 48391b0abe3SJohannes Weiner * 48491b0abe3SJohannes Weiner * Called under (and serialized by) inode->i_mutex. 48591b0abe3SJohannes Weiner * 48691b0abe3SJohannes Weiner * Filesystems have to use this in the .evict_inode path to inform the 48791b0abe3SJohannes Weiner * VM that this is the final truncate and the inode is going away. 48891b0abe3SJohannes Weiner */ 48991b0abe3SJohannes Weiner void truncate_inode_pages_final(struct address_space *mapping) 49091b0abe3SJohannes Weiner { 491f9fe48beSRoss Zwisler unsigned long nrexceptional; 49291b0abe3SJohannes Weiner unsigned long nrpages; 49391b0abe3SJohannes Weiner 49491b0abe3SJohannes Weiner /* 49591b0abe3SJohannes Weiner * Page reclaim can not participate in regular inode lifetime 49691b0abe3SJohannes Weiner * management (can't call iput()) and thus can race with the 49791b0abe3SJohannes Weiner * inode teardown. Tell it when the address space is exiting, 49891b0abe3SJohannes Weiner * so that it does not install eviction information after the 49991b0abe3SJohannes Weiner * final truncate has begun. 50091b0abe3SJohannes Weiner */ 50191b0abe3SJohannes Weiner mapping_set_exiting(mapping); 50291b0abe3SJohannes Weiner 50391b0abe3SJohannes Weiner /* 50491b0abe3SJohannes Weiner * When reclaim installs eviction entries, it increases 505f9fe48beSRoss Zwisler * nrexceptional first, then decreases nrpages. Make sure we see 50691b0abe3SJohannes Weiner * this in the right order or we might miss an entry. 50791b0abe3SJohannes Weiner */ 50891b0abe3SJohannes Weiner nrpages = mapping->nrpages; 50991b0abe3SJohannes Weiner smp_rmb(); 510f9fe48beSRoss Zwisler nrexceptional = mapping->nrexceptional; 51191b0abe3SJohannes Weiner 512f9fe48beSRoss Zwisler if (nrpages || nrexceptional) { 51391b0abe3SJohannes Weiner /* 51491b0abe3SJohannes Weiner * As truncation uses a lockless tree lookup, cycle 51591b0abe3SJohannes Weiner * the tree lock to make sure any ongoing tree 51691b0abe3SJohannes Weiner * modification that does not see AS_EXITING is 51791b0abe3SJohannes Weiner * completed before starting the final truncate. 51891b0abe3SJohannes Weiner */ 519b93b0163SMatthew Wilcox xa_lock_irq(&mapping->i_pages); 520b93b0163SMatthew Wilcox xa_unlock_irq(&mapping->i_pages); 52191b0abe3SJohannes Weiner } 5226ff38bd4SPavel Tikhomirov 5236ff38bd4SPavel Tikhomirov /* 5246ff38bd4SPavel Tikhomirov * Cleancache needs notification even if there are no pages or shadow 5256ff38bd4SPavel Tikhomirov * entries. 5266ff38bd4SPavel Tikhomirov */ 5276ff38bd4SPavel Tikhomirov truncate_inode_pages(mapping, 0); 52891b0abe3SJohannes Weiner } 52991b0abe3SJohannes Weiner EXPORT_SYMBOL(truncate_inode_pages_final); 53091b0abe3SJohannes Weiner 531a77eedbcSJason Yan static unsigned long __invalidate_mapping_pages(struct address_space *mapping, 532eb1d7a65SYafang Shao pgoff_t start, pgoff_t end, unsigned long *nr_pagevec) 5331da177e4SLinus Torvalds { 5340cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 5351da177e4SLinus Torvalds struct pagevec pvec; 536b85e0effSHugh Dickins pgoff_t index = start; 53731560180SMinchan Kim unsigned long ret; 53831560180SMinchan Kim unsigned long count = 0; 5391da177e4SLinus Torvalds int i; 5401da177e4SLinus Torvalds 54186679820SMel Gorman pagevec_init(&pvec); 5420cd6144aSJohannes Weiner while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, 5430cd6144aSJohannes Weiner min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, 5440cd6144aSJohannes Weiner indices)) { 5451da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 5461da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 5471da177e4SLinus Torvalds 548b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 5490cd6144aSJohannes Weiner index = indices[i]; 550b85e0effSHugh Dickins if (index > end) 551b85e0effSHugh Dickins break; 552e0f23603SNeilBrown 5533159f943SMatthew Wilcox if (xa_is_value(page)) { 554c6dcf52cSJan Kara invalidate_exceptional_entry(mapping, index, 555c6dcf52cSJan Kara page); 5560cd6144aSJohannes Weiner continue; 5570cd6144aSJohannes Weiner } 5580cd6144aSJohannes Weiner 559b85e0effSHugh Dickins if (!trylock_page(page)) 560b85e0effSHugh Dickins continue; 561fc127da0SKirill A. Shutemov 5625cbc198aSKirill A. Shutemov WARN_ON(page_to_index(page) != index); 563fc127da0SKirill A. Shutemov 564fc127da0SKirill A. Shutemov /* Middle of THP: skip */ 565fc127da0SKirill A. Shutemov if (PageTransTail(page)) { 566fc127da0SKirill A. Shutemov unlock_page(page); 567fc127da0SKirill A. Shutemov continue; 568fc127da0SKirill A. Shutemov } else if (PageTransHuge(page)) { 569fc127da0SKirill A. Shutemov index += HPAGE_PMD_NR - 1; 570fc127da0SKirill A. Shutemov i += HPAGE_PMD_NR - 1; 57176b6f9b7SJan Kara /* 57276b6f9b7SJan Kara * 'end' is in the middle of THP. Don't 57376b6f9b7SJan Kara * invalidate the page as the part outside of 57476b6f9b7SJan Kara * 'end' could be still useful. 57576b6f9b7SJan Kara */ 57676b6f9b7SJan Kara if (index > end) { 57776b6f9b7SJan Kara unlock_page(page); 578fc127da0SKirill A. Shutemov continue; 579fc127da0SKirill A. Shutemov } 580ef18a1caSKirill A. Shutemov 581ef18a1caSKirill A. Shutemov /* Take a pin outside pagevec */ 582ef18a1caSKirill A. Shutemov get_page(page); 583ef18a1caSKirill A. Shutemov 584ef18a1caSKirill A. Shutemov /* 585ef18a1caSKirill A. Shutemov * Drop extra pins before trying to invalidate 586ef18a1caSKirill A. Shutemov * the huge page. 587ef18a1caSKirill A. Shutemov */ 588ef18a1caSKirill A. Shutemov pagevec_remove_exceptionals(&pvec); 589ef18a1caSKirill A. Shutemov pagevec_release(&pvec); 59076b6f9b7SJan Kara } 591fc127da0SKirill A. Shutemov 59231560180SMinchan Kim ret = invalidate_inode_page(page); 5931da177e4SLinus Torvalds unlock_page(page); 59431560180SMinchan Kim /* 59531560180SMinchan Kim * Invalidation is a hint that the page is no longer 59631560180SMinchan Kim * of interest and try to speed up its reclaim. 59731560180SMinchan Kim */ 598eb1d7a65SYafang Shao if (!ret) { 599cc5993bdSMinchan Kim deactivate_file_page(page); 600eb1d7a65SYafang Shao /* It is likely on the pagevec of a remote CPU */ 601eb1d7a65SYafang Shao if (nr_pagevec) 602eb1d7a65SYafang Shao (*nr_pagevec)++; 603eb1d7a65SYafang Shao } 604eb1d7a65SYafang Shao 605ef18a1caSKirill A. Shutemov if (PageTransHuge(page)) 606ef18a1caSKirill A. Shutemov put_page(page); 60731560180SMinchan Kim count += ret; 6081da177e4SLinus Torvalds } 6090cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 6101da177e4SLinus Torvalds pagevec_release(&pvec); 611fc9a07e7SAndrew Morton cond_resched(); 612b85e0effSHugh Dickins index++; 6131da177e4SLinus Torvalds } 61431560180SMinchan Kim return count; 6151da177e4SLinus Torvalds } 616eb1d7a65SYafang Shao 617eb1d7a65SYafang Shao /** 618eb1d7a65SYafang Shao * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode 619eb1d7a65SYafang Shao * @mapping: the address_space which holds the pages to invalidate 620eb1d7a65SYafang Shao * @start: the offset 'from' which to invalidate 621eb1d7a65SYafang Shao * @end: the offset 'to' which to invalidate (inclusive) 622eb1d7a65SYafang Shao * 623eb1d7a65SYafang Shao * This function only removes the unlocked pages, if you want to 624eb1d7a65SYafang Shao * remove all the pages of one inode, you must call truncate_inode_pages. 625eb1d7a65SYafang Shao * 626eb1d7a65SYafang Shao * invalidate_mapping_pages() will not block on IO activity. It will not 627eb1d7a65SYafang Shao * invalidate pages which are dirty, locked, under writeback or mapped into 628eb1d7a65SYafang Shao * pagetables. 629eb1d7a65SYafang Shao * 630eb1d7a65SYafang Shao * Return: the number of the pages that were invalidated 631eb1d7a65SYafang Shao */ 632eb1d7a65SYafang Shao unsigned long invalidate_mapping_pages(struct address_space *mapping, 633eb1d7a65SYafang Shao pgoff_t start, pgoff_t end) 634eb1d7a65SYafang Shao { 635eb1d7a65SYafang Shao return __invalidate_mapping_pages(mapping, start, end, NULL); 636eb1d7a65SYafang Shao } 63754bc4855SAnton Altaparmakov EXPORT_SYMBOL(invalidate_mapping_pages); 6381da177e4SLinus Torvalds 639eb1d7a65SYafang Shao /** 640*649c6dfeSAlex Shi * invalidate_mapping_pagevec - Invalidate all the unlocked pages of one inode 641*649c6dfeSAlex Shi * @mapping: the address_space which holds the pages to invalidate 642*649c6dfeSAlex Shi * @start: the offset 'from' which to invalidate 643*649c6dfeSAlex Shi * @end: the offset 'to' which to invalidate (inclusive) 644*649c6dfeSAlex Shi * @nr_pagevec: invalidate failed page number for caller 645*649c6dfeSAlex Shi * 646*649c6dfeSAlex Shi * This helper is similar with invalidate_mapping_pages, except that it accounts 647*649c6dfeSAlex Shi * for pages that failed to invalidate on a pagevec and count them in 648*649c6dfeSAlex Shi * @nr_pagevec, which will used by the caller. 649eb1d7a65SYafang Shao */ 650eb1d7a65SYafang Shao void invalidate_mapping_pagevec(struct address_space *mapping, 651eb1d7a65SYafang Shao pgoff_t start, pgoff_t end, unsigned long *nr_pagevec) 652eb1d7a65SYafang Shao { 653eb1d7a65SYafang Shao __invalidate_mapping_pages(mapping, start, end, nr_pagevec); 654eb1d7a65SYafang Shao } 655eb1d7a65SYafang Shao 656bd4c8ce4SAndrew Morton /* 657bd4c8ce4SAndrew Morton * This is like invalidate_complete_page(), except it ignores the page's 658bd4c8ce4SAndrew Morton * refcount. We do this because invalidate_inode_pages2() needs stronger 659bd4c8ce4SAndrew Morton * invalidation guarantees, and cannot afford to leave pages behind because 6602706a1b8SAnderson Briglia * shrink_page_list() has a temp ref on them, or because they're transiently 6612706a1b8SAnderson Briglia * sitting in the lru_cache_add() pagevecs. 662bd4c8ce4SAndrew Morton */ 663bd4c8ce4SAndrew Morton static int 664bd4c8ce4SAndrew Morton invalidate_complete_page2(struct address_space *mapping, struct page *page) 665bd4c8ce4SAndrew Morton { 666c4843a75SGreg Thelen unsigned long flags; 667c4843a75SGreg Thelen 668bd4c8ce4SAndrew Morton if (page->mapping != mapping) 669bd4c8ce4SAndrew Morton return 0; 670bd4c8ce4SAndrew Morton 671266cf658SDavid Howells if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) 672bd4c8ce4SAndrew Morton return 0; 673bd4c8ce4SAndrew Morton 674b93b0163SMatthew Wilcox xa_lock_irqsave(&mapping->i_pages, flags); 675bd4c8ce4SAndrew Morton if (PageDirty(page)) 676bd4c8ce4SAndrew Morton goto failed; 677bd4c8ce4SAndrew Morton 678266cf658SDavid Howells BUG_ON(page_has_private(page)); 67962cccb8cSJohannes Weiner __delete_from_page_cache(page, NULL); 680b93b0163SMatthew Wilcox xa_unlock_irqrestore(&mapping->i_pages, flags); 6816072d13cSLinus Torvalds 6826072d13cSLinus Torvalds if (mapping->a_ops->freepage) 6836072d13cSLinus Torvalds mapping->a_ops->freepage(page); 6846072d13cSLinus Torvalds 68509cbfeafSKirill A. Shutemov put_page(page); /* pagecache ref */ 686bd4c8ce4SAndrew Morton return 1; 687bd4c8ce4SAndrew Morton failed: 688b93b0163SMatthew Wilcox xa_unlock_irqrestore(&mapping->i_pages, flags); 689bd4c8ce4SAndrew Morton return 0; 690bd4c8ce4SAndrew Morton } 691bd4c8ce4SAndrew Morton 692e3db7691STrond Myklebust static int do_launder_page(struct address_space *mapping, struct page *page) 693e3db7691STrond Myklebust { 694e3db7691STrond Myklebust if (!PageDirty(page)) 695e3db7691STrond Myklebust return 0; 696e3db7691STrond Myklebust if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) 697e3db7691STrond Myklebust return 0; 698e3db7691STrond Myklebust return mapping->a_ops->launder_page(page); 699e3db7691STrond Myklebust } 700e3db7691STrond Myklebust 7011da177e4SLinus Torvalds /** 7021da177e4SLinus Torvalds * invalidate_inode_pages2_range - remove range of pages from an address_space 70367be2dd1SMartin Waitz * @mapping: the address_space 7041da177e4SLinus Torvalds * @start: the page offset 'from' which to invalidate 7051da177e4SLinus Torvalds * @end: the page offset 'to' which to invalidate (inclusive) 7061da177e4SLinus Torvalds * 7071da177e4SLinus Torvalds * Any pages which are found to be mapped into pagetables are unmapped prior to 7081da177e4SLinus Torvalds * invalidation. 7091da177e4SLinus Torvalds * 710a862f68aSMike Rapoport * Return: -EBUSY if any pages could not be invalidated. 7111da177e4SLinus Torvalds */ 7121da177e4SLinus Torvalds int invalidate_inode_pages2_range(struct address_space *mapping, 7131da177e4SLinus Torvalds pgoff_t start, pgoff_t end) 7141da177e4SLinus Torvalds { 7150cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 7161da177e4SLinus Torvalds struct pagevec pvec; 717b85e0effSHugh Dickins pgoff_t index; 7181da177e4SLinus Torvalds int i; 7191da177e4SLinus Torvalds int ret = 0; 7200dd1334fSHisashi Hifumi int ret2 = 0; 7211da177e4SLinus Torvalds int did_range_unmap = 0; 7221da177e4SLinus Torvalds 72332691f0fSAndrey Ryabinin if (mapping->nrpages == 0 && mapping->nrexceptional == 0) 72434ccb69eSAndrey Ryabinin goto out; 72532691f0fSAndrey Ryabinin 72686679820SMel Gorman pagevec_init(&pvec); 727b85e0effSHugh Dickins index = start; 7280cd6144aSJohannes Weiner while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, 7290cd6144aSJohannes Weiner min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, 7300cd6144aSJohannes Weiner indices)) { 7317b965e08STrond Myklebust for (i = 0; i < pagevec_count(&pvec); i++) { 7321da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 733b85e0effSHugh Dickins 734b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 7350cd6144aSJohannes Weiner index = indices[i]; 736b85e0effSHugh Dickins if (index > end) 737b85e0effSHugh Dickins break; 7381da177e4SLinus Torvalds 7393159f943SMatthew Wilcox if (xa_is_value(page)) { 740c6dcf52cSJan Kara if (!invalidate_exceptional_entry2(mapping, 741c6dcf52cSJan Kara index, page)) 742c6dcf52cSJan Kara ret = -EBUSY; 7430cd6144aSJohannes Weiner continue; 7440cd6144aSJohannes Weiner } 7450cd6144aSJohannes Weiner 7461da177e4SLinus Torvalds lock_page(page); 7475cbc198aSKirill A. Shutemov WARN_ON(page_to_index(page) != index); 7481da177e4SLinus Torvalds if (page->mapping != mapping) { 7491da177e4SLinus Torvalds unlock_page(page); 7501da177e4SLinus Torvalds continue; 7511da177e4SLinus Torvalds } 7521da177e4SLinus Torvalds wait_on_page_writeback(page); 753d00806b1SNick Piggin if (page_mapped(page)) { 7541da177e4SLinus Torvalds if (!did_range_unmap) { 7551da177e4SLinus Torvalds /* 7561da177e4SLinus Torvalds * Zap the rest of the file in one hit. 7571da177e4SLinus Torvalds */ 758977fbdcdSMatthew Wilcox unmap_mapping_pages(mapping, index, 759977fbdcdSMatthew Wilcox (1 + end - index), false); 7601da177e4SLinus Torvalds did_range_unmap = 1; 7611da177e4SLinus Torvalds } else { 7621da177e4SLinus Torvalds /* 7631da177e4SLinus Torvalds * Just zap this page 7641da177e4SLinus Torvalds */ 765977fbdcdSMatthew Wilcox unmap_mapping_pages(mapping, index, 766977fbdcdSMatthew Wilcox 1, false); 7671da177e4SLinus Torvalds } 7681da177e4SLinus Torvalds } 769d00806b1SNick Piggin BUG_ON(page_mapped(page)); 7700dd1334fSHisashi Hifumi ret2 = do_launder_page(mapping, page); 7710dd1334fSHisashi Hifumi if (ret2 == 0) { 7720dd1334fSHisashi Hifumi if (!invalidate_complete_page2(mapping, page)) 7736ccfa806SHisashi Hifumi ret2 = -EBUSY; 7740dd1334fSHisashi Hifumi } 7750dd1334fSHisashi Hifumi if (ret2 < 0) 7760dd1334fSHisashi Hifumi ret = ret2; 7771da177e4SLinus Torvalds unlock_page(page); 7781da177e4SLinus Torvalds } 7790cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 7801da177e4SLinus Torvalds pagevec_release(&pvec); 7811da177e4SLinus Torvalds cond_resched(); 782b85e0effSHugh Dickins index++; 7831da177e4SLinus Torvalds } 784cd656375SJan Kara /* 78569b6c131SMatthew Wilcox * For DAX we invalidate page tables after invalidating page cache. We 786cd656375SJan Kara * could invalidate page tables while invalidating each entry however 787cd656375SJan Kara * that would be expensive. And doing range unmapping before doesn't 78869b6c131SMatthew Wilcox * work as we have no cheap way to find whether page cache entry didn't 789cd656375SJan Kara * get remapped later. 790cd656375SJan Kara */ 791cd656375SJan Kara if (dax_mapping(mapping)) { 792977fbdcdSMatthew Wilcox unmap_mapping_pages(mapping, start, end - start + 1, false); 793cd656375SJan Kara } 79434ccb69eSAndrey Ryabinin out: 7953167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 7961da177e4SLinus Torvalds return ret; 7971da177e4SLinus Torvalds } 7981da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); 7991da177e4SLinus Torvalds 8001da177e4SLinus Torvalds /** 8011da177e4SLinus Torvalds * invalidate_inode_pages2 - remove all pages from an address_space 80267be2dd1SMartin Waitz * @mapping: the address_space 8031da177e4SLinus Torvalds * 8041da177e4SLinus Torvalds * Any pages which are found to be mapped into pagetables are unmapped prior to 8051da177e4SLinus Torvalds * invalidation. 8061da177e4SLinus Torvalds * 807a862f68aSMike Rapoport * Return: -EBUSY if any pages could not be invalidated. 8081da177e4SLinus Torvalds */ 8091da177e4SLinus Torvalds int invalidate_inode_pages2(struct address_space *mapping) 8101da177e4SLinus Torvalds { 8111da177e4SLinus Torvalds return invalidate_inode_pages2_range(mapping, 0, -1); 8121da177e4SLinus Torvalds } 8131da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(invalidate_inode_pages2); 81425d9e2d1Snpiggin@suse.de 81525d9e2d1Snpiggin@suse.de /** 81625d9e2d1Snpiggin@suse.de * truncate_pagecache - unmap and remove pagecache that has been truncated 81725d9e2d1Snpiggin@suse.de * @inode: inode 8188a549beaSHugh Dickins * @newsize: new file size 81925d9e2d1Snpiggin@suse.de * 82025d9e2d1Snpiggin@suse.de * inode's new i_size must already be written before truncate_pagecache 82125d9e2d1Snpiggin@suse.de * is called. 82225d9e2d1Snpiggin@suse.de * 82325d9e2d1Snpiggin@suse.de * This function should typically be called before the filesystem 82425d9e2d1Snpiggin@suse.de * releases resources associated with the freed range (eg. deallocates 82525d9e2d1Snpiggin@suse.de * blocks). This way, pagecache will always stay logically coherent 82625d9e2d1Snpiggin@suse.de * with on-disk format, and the filesystem would not have to deal with 82725d9e2d1Snpiggin@suse.de * situations such as writepage being called for a page that has already 82825d9e2d1Snpiggin@suse.de * had its underlying blocks deallocated. 82925d9e2d1Snpiggin@suse.de */ 8307caef267SKirill A. Shutemov void truncate_pagecache(struct inode *inode, loff_t newsize) 83125d9e2d1Snpiggin@suse.de { 83225d9e2d1Snpiggin@suse.de struct address_space *mapping = inode->i_mapping; 8338a549beaSHugh Dickins loff_t holebegin = round_up(newsize, PAGE_SIZE); 83425d9e2d1Snpiggin@suse.de 83525d9e2d1Snpiggin@suse.de /* 83625d9e2d1Snpiggin@suse.de * unmap_mapping_range is called twice, first simply for 83725d9e2d1Snpiggin@suse.de * efficiency so that truncate_inode_pages does fewer 83825d9e2d1Snpiggin@suse.de * single-page unmaps. However after this first call, and 83925d9e2d1Snpiggin@suse.de * before truncate_inode_pages finishes, it is possible for 84025d9e2d1Snpiggin@suse.de * private pages to be COWed, which remain after 84125d9e2d1Snpiggin@suse.de * truncate_inode_pages finishes, hence the second 84225d9e2d1Snpiggin@suse.de * unmap_mapping_range call must be made for correctness. 84325d9e2d1Snpiggin@suse.de */ 8448a549beaSHugh Dickins unmap_mapping_range(mapping, holebegin, 0, 1); 8458a549beaSHugh Dickins truncate_inode_pages(mapping, newsize); 8468a549beaSHugh Dickins unmap_mapping_range(mapping, holebegin, 0, 1); 84725d9e2d1Snpiggin@suse.de } 84825d9e2d1Snpiggin@suse.de EXPORT_SYMBOL(truncate_pagecache); 84925d9e2d1Snpiggin@suse.de 85025d9e2d1Snpiggin@suse.de /** 8512c27c65eSChristoph Hellwig * truncate_setsize - update inode and pagecache for a new file size 8522c27c65eSChristoph Hellwig * @inode: inode 8532c27c65eSChristoph Hellwig * @newsize: new file size 8542c27c65eSChristoph Hellwig * 855382e27daSJan Kara * truncate_setsize updates i_size and performs pagecache truncation (if 856382e27daSJan Kara * necessary) to @newsize. It will be typically be called from the filesystem's 857382e27daSJan Kara * setattr function when ATTR_SIZE is passed in. 8582c27c65eSChristoph Hellwig * 85977783d06SJan Kara * Must be called with a lock serializing truncates and writes (generally 86077783d06SJan Kara * i_mutex but e.g. xfs uses a different lock) and before all filesystem 86177783d06SJan Kara * specific block truncation has been performed. 8622c27c65eSChristoph Hellwig */ 8632c27c65eSChristoph Hellwig void truncate_setsize(struct inode *inode, loff_t newsize) 8642c27c65eSChristoph Hellwig { 86590a80202SJan Kara loff_t oldsize = inode->i_size; 86690a80202SJan Kara 8672c27c65eSChristoph Hellwig i_size_write(inode, newsize); 86890a80202SJan Kara if (newsize > oldsize) 86990a80202SJan Kara pagecache_isize_extended(inode, oldsize, newsize); 8707caef267SKirill A. Shutemov truncate_pagecache(inode, newsize); 8712c27c65eSChristoph Hellwig } 8722c27c65eSChristoph Hellwig EXPORT_SYMBOL(truncate_setsize); 8732c27c65eSChristoph Hellwig 8742c27c65eSChristoph Hellwig /** 87590a80202SJan Kara * pagecache_isize_extended - update pagecache after extension of i_size 87690a80202SJan Kara * @inode: inode for which i_size was extended 87790a80202SJan Kara * @from: original inode size 87890a80202SJan Kara * @to: new inode size 87990a80202SJan Kara * 88090a80202SJan Kara * Handle extension of inode size either caused by extending truncate or by 88190a80202SJan Kara * write starting after current i_size. We mark the page straddling current 88290a80202SJan Kara * i_size RO so that page_mkwrite() is called on the nearest write access to 88390a80202SJan Kara * the page. This way filesystem can be sure that page_mkwrite() is called on 88490a80202SJan Kara * the page before user writes to the page via mmap after the i_size has been 88590a80202SJan Kara * changed. 88690a80202SJan Kara * 88790a80202SJan Kara * The function must be called after i_size is updated so that page fault 88890a80202SJan Kara * coming after we unlock the page will already see the new i_size. 88990a80202SJan Kara * The function must be called while we still hold i_mutex - this not only 89090a80202SJan Kara * makes sure i_size is stable but also that userspace cannot observe new 89190a80202SJan Kara * i_size value before we are prepared to store mmap writes at new inode size. 89290a80202SJan Kara */ 89390a80202SJan Kara void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) 89490a80202SJan Kara { 89593407472SFabian Frederick int bsize = i_blocksize(inode); 89690a80202SJan Kara loff_t rounded_from; 89790a80202SJan Kara struct page *page; 89890a80202SJan Kara pgoff_t index; 89990a80202SJan Kara 90090a80202SJan Kara WARN_ON(to > inode->i_size); 90190a80202SJan Kara 90209cbfeafSKirill A. Shutemov if (from >= to || bsize == PAGE_SIZE) 90390a80202SJan Kara return; 90490a80202SJan Kara /* Page straddling @from will not have any hole block created? */ 90590a80202SJan Kara rounded_from = round_up(from, bsize); 90609cbfeafSKirill A. Shutemov if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1))) 90790a80202SJan Kara return; 90890a80202SJan Kara 90909cbfeafSKirill A. Shutemov index = from >> PAGE_SHIFT; 91090a80202SJan Kara page = find_lock_page(inode->i_mapping, index); 91190a80202SJan Kara /* Page not cached? Nothing to do */ 91290a80202SJan Kara if (!page) 91390a80202SJan Kara return; 91490a80202SJan Kara /* 91590a80202SJan Kara * See clear_page_dirty_for_io() for details why set_page_dirty() 91690a80202SJan Kara * is needed. 91790a80202SJan Kara */ 91890a80202SJan Kara if (page_mkclean(page)) 91990a80202SJan Kara set_page_dirty(page); 92090a80202SJan Kara unlock_page(page); 92109cbfeafSKirill A. Shutemov put_page(page); 92290a80202SJan Kara } 92390a80202SJan Kara EXPORT_SYMBOL(pagecache_isize_extended); 92490a80202SJan Kara 92590a80202SJan Kara /** 926623e3db9SHugh Dickins * truncate_pagecache_range - unmap and remove pagecache that is hole-punched 927623e3db9SHugh Dickins * @inode: inode 928623e3db9SHugh Dickins * @lstart: offset of beginning of hole 929623e3db9SHugh Dickins * @lend: offset of last byte of hole 930623e3db9SHugh Dickins * 931623e3db9SHugh Dickins * This function should typically be called before the filesystem 932623e3db9SHugh Dickins * releases resources associated with the freed range (eg. deallocates 933623e3db9SHugh Dickins * blocks). This way, pagecache will always stay logically coherent 934623e3db9SHugh Dickins * with on-disk format, and the filesystem would not have to deal with 935623e3db9SHugh Dickins * situations such as writepage being called for a page that has already 936623e3db9SHugh Dickins * had its underlying blocks deallocated. 937623e3db9SHugh Dickins */ 938623e3db9SHugh Dickins void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) 939623e3db9SHugh Dickins { 940623e3db9SHugh Dickins struct address_space *mapping = inode->i_mapping; 941623e3db9SHugh Dickins loff_t unmap_start = round_up(lstart, PAGE_SIZE); 942623e3db9SHugh Dickins loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; 943623e3db9SHugh Dickins /* 944623e3db9SHugh Dickins * This rounding is currently just for example: unmap_mapping_range 945623e3db9SHugh Dickins * expands its hole outwards, whereas we want it to contract the hole 946623e3db9SHugh Dickins * inwards. However, existing callers of truncate_pagecache_range are 9475a720394SLukas Czerner * doing their own page rounding first. Note that unmap_mapping_range 9485a720394SLukas Czerner * allows holelen 0 for all, and we allow lend -1 for end of file. 949623e3db9SHugh Dickins */ 950623e3db9SHugh Dickins 951623e3db9SHugh Dickins /* 952623e3db9SHugh Dickins * Unlike in truncate_pagecache, unmap_mapping_range is called only 953623e3db9SHugh Dickins * once (before truncating pagecache), and without "even_cows" flag: 954623e3db9SHugh Dickins * hole-punching should not remove private COWed pages from the hole. 955623e3db9SHugh Dickins */ 956623e3db9SHugh Dickins if ((u64)unmap_end > (u64)unmap_start) 957623e3db9SHugh Dickins unmap_mapping_range(mapping, unmap_start, 958623e3db9SHugh Dickins 1 + unmap_end - unmap_start, 0); 959623e3db9SHugh Dickins truncate_inode_pages_range(mapping, lstart, lend); 960623e3db9SHugh Dickins } 961623e3db9SHugh Dickins EXPORT_SYMBOL(truncate_pagecache_range); 962