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> 223a4f8a0bSHugh Dickins #include <linux/shmem_fs.h> 2390a80202SJan Kara #include <linux/rmap.h> 24ba470de4SRik van Riel #include "internal.h" 251da177e4SLinus Torvalds 26f2187599SMel Gorman /* 27f2187599SMel Gorman * Regular page slots are stabilized by the page lock even without the tree 28f2187599SMel Gorman * itself locked. These unlocked entries need verification under the tree 29f2187599SMel Gorman * lock. 30f2187599SMel Gorman */ 31f2187599SMel Gorman static inline void __clear_shadow_entry(struct address_space *mapping, 32f2187599SMel Gorman pgoff_t index, void *entry) 330cd6144aSJohannes Weiner { 3469b6c131SMatthew Wilcox XA_STATE(xas, &mapping->i_pages, index); 35449dd698SJohannes Weiner 3669b6c131SMatthew Wilcox xas_set_update(&xas, workingset_update_node); 3769b6c131SMatthew Wilcox if (xas_load(&xas) != entry) 38f2187599SMel Gorman return; 3969b6c131SMatthew Wilcox xas_store(&xas, NULL); 40f2187599SMel Gorman } 41f2187599SMel Gorman 42f2187599SMel Gorman static void clear_shadow_entry(struct address_space *mapping, pgoff_t index, 43f2187599SMel Gorman void *entry) 44f2187599SMel Gorman { 4551b8c1feSJohannes Weiner spin_lock(&mapping->host->i_lock); 46b93b0163SMatthew Wilcox xa_lock_irq(&mapping->i_pages); 47f2187599SMel Gorman __clear_shadow_entry(mapping, index, entry); 48b93b0163SMatthew Wilcox xa_unlock_irq(&mapping->i_pages); 4951b8c1feSJohannes Weiner if (mapping_shrinkable(mapping)) 5051b8c1feSJohannes Weiner inode_add_lru(mapping->host); 5151b8c1feSJohannes Weiner spin_unlock(&mapping->host->i_lock); 520cd6144aSJohannes Weiner } 531da177e4SLinus Torvalds 54c6dcf52cSJan Kara /* 55f2187599SMel Gorman * Unconditionally remove exceptional entries. Usually called from truncate 5651dcbdacSMatthew Wilcox (Oracle) * path. Note that the folio_batch may be altered by this function by removing 571613fac9SMatthew Wilcox (Oracle) * exceptional entries similar to what folio_batch_remove_exceptionals() does. 58c6dcf52cSJan Kara */ 5951dcbdacSMatthew Wilcox (Oracle) static void truncate_folio_batch_exceptionals(struct address_space *mapping, 6051dcbdacSMatthew Wilcox (Oracle) struct folio_batch *fbatch, pgoff_t *indices) 61c6dcf52cSJan Kara { 62f2187599SMel Gorman int i, j; 6331d270fdSMatthew Wilcox (Oracle) bool dax; 64f2187599SMel Gorman 65c6dcf52cSJan Kara /* Handled by shmem itself */ 66c6dcf52cSJan Kara if (shmem_mapping(mapping)) 67c6dcf52cSJan Kara return; 68c6dcf52cSJan Kara 6951dcbdacSMatthew Wilcox (Oracle) for (j = 0; j < folio_batch_count(fbatch); j++) 7051dcbdacSMatthew Wilcox (Oracle) if (xa_is_value(fbatch->folios[j])) 71f2187599SMel Gorman break; 72f2187599SMel Gorman 7351dcbdacSMatthew Wilcox (Oracle) if (j == folio_batch_count(fbatch)) 74c6dcf52cSJan Kara return; 75f2187599SMel Gorman 76f2187599SMel Gorman dax = dax_mapping(mapping); 7751b8c1feSJohannes Weiner if (!dax) { 7851b8c1feSJohannes Weiner spin_lock(&mapping->host->i_lock); 79b93b0163SMatthew Wilcox xa_lock_irq(&mapping->i_pages); 8051b8c1feSJohannes Weiner } 81f2187599SMel Gorman 8251dcbdacSMatthew Wilcox (Oracle) for (i = j; i < folio_batch_count(fbatch); i++) { 8351dcbdacSMatthew Wilcox (Oracle) struct folio *folio = fbatch->folios[i]; 84f2187599SMel Gorman pgoff_t index = indices[i]; 85f2187599SMel Gorman 8651dcbdacSMatthew Wilcox (Oracle) if (!xa_is_value(folio)) { 8751dcbdacSMatthew Wilcox (Oracle) fbatch->folios[j++] = folio; 88f2187599SMel Gorman continue; 89c6dcf52cSJan Kara } 90f2187599SMel Gorman 91f2187599SMel Gorman if (unlikely(dax)) { 92f2187599SMel Gorman dax_delete_mapping_entry(mapping, index); 93f2187599SMel Gorman continue; 94f2187599SMel Gorman } 95f2187599SMel Gorman 9651dcbdacSMatthew Wilcox (Oracle) __clear_shadow_entry(mapping, index, folio); 97f2187599SMel Gorman } 98f2187599SMel Gorman 9951b8c1feSJohannes Weiner if (!dax) { 100b93b0163SMatthew Wilcox xa_unlock_irq(&mapping->i_pages); 10151b8c1feSJohannes Weiner if (mapping_shrinkable(mapping)) 10251b8c1feSJohannes Weiner inode_add_lru(mapping->host); 10351b8c1feSJohannes Weiner spin_unlock(&mapping->host->i_lock); 10451b8c1feSJohannes Weiner } 10551dcbdacSMatthew Wilcox (Oracle) fbatch->nr = j; 1060e499ed3SMatthew Wilcox (Oracle) } 1070e499ed3SMatthew Wilcox (Oracle) 108c6dcf52cSJan Kara /* 109c6dcf52cSJan Kara * Invalidate exceptional entry if easily possible. This handles exceptional 1104636e70bSRoss Zwisler * entries for invalidate_inode_pages(). 111c6dcf52cSJan Kara */ 112c6dcf52cSJan Kara static int invalidate_exceptional_entry(struct address_space *mapping, 113c6dcf52cSJan Kara pgoff_t index, void *entry) 114c6dcf52cSJan Kara { 1154636e70bSRoss Zwisler /* Handled by shmem itself, or for DAX we do nothing. */ 1164636e70bSRoss Zwisler if (shmem_mapping(mapping) || dax_mapping(mapping)) 117c6dcf52cSJan Kara return 1; 118c6dcf52cSJan Kara clear_shadow_entry(mapping, index, entry); 119c6dcf52cSJan Kara return 1; 120c6dcf52cSJan Kara } 121c6dcf52cSJan Kara 122c6dcf52cSJan Kara /* 123c6dcf52cSJan Kara * Invalidate exceptional entry if clean. This handles exceptional entries for 124c6dcf52cSJan Kara * invalidate_inode_pages2() so for DAX it evicts only clean entries. 125c6dcf52cSJan Kara */ 126c6dcf52cSJan Kara static int invalidate_exceptional_entry2(struct address_space *mapping, 127c6dcf52cSJan Kara pgoff_t index, void *entry) 128c6dcf52cSJan Kara { 129c6dcf52cSJan Kara /* Handled by shmem itself */ 130c6dcf52cSJan Kara if (shmem_mapping(mapping)) 131c6dcf52cSJan Kara return 1; 132c6dcf52cSJan Kara if (dax_mapping(mapping)) 133c6dcf52cSJan Kara return dax_invalidate_mapping_entry_sync(mapping, index); 134c6dcf52cSJan Kara clear_shadow_entry(mapping, index, entry); 135c6dcf52cSJan Kara return 1; 136c6dcf52cSJan Kara } 137c6dcf52cSJan Kara 138cf9a2ae8SDavid Howells /** 1395ad6b2bdSMatthew Wilcox (Oracle) * folio_invalidate - Invalidate part or all of a folio. 1405ad6b2bdSMatthew Wilcox (Oracle) * @folio: The folio which is affected. 141d47992f8SLukas Czerner * @offset: start of the range to invalidate 142d47992f8SLukas Czerner * @length: length of the range to invalidate 143cf9a2ae8SDavid Howells * 1445ad6b2bdSMatthew Wilcox (Oracle) * folio_invalidate() is called when all or part of the folio has become 145cf9a2ae8SDavid Howells * invalidated by a truncate operation. 146cf9a2ae8SDavid Howells * 1475ad6b2bdSMatthew Wilcox (Oracle) * folio_invalidate() does not have to release all buffers, but it must 148cf9a2ae8SDavid Howells * ensure that no dirty buffer is left outside @offset and that no I/O 149cf9a2ae8SDavid Howells * is underway against any of the blocks which are outside the truncation 150cf9a2ae8SDavid Howells * point. Because the caller is about to free (and possibly reuse) those 151cf9a2ae8SDavid Howells * blocks on-disk. 152cf9a2ae8SDavid Howells */ 1535ad6b2bdSMatthew Wilcox (Oracle) void folio_invalidate(struct folio *folio, size_t offset, size_t length) 154cf9a2ae8SDavid Howells { 155128d1f82SMatthew Wilcox (Oracle) const struct address_space_operations *aops = folio->mapping->a_ops; 156d47992f8SLukas Czerner 157f50015a5SMatthew Wilcox (Oracle) if (aops->invalidate_folio) 158128d1f82SMatthew Wilcox (Oracle) aops->invalidate_folio(folio, offset, length); 159cf9a2ae8SDavid Howells } 1605ad6b2bdSMatthew Wilcox (Oracle) EXPORT_SYMBOL_GPL(folio_invalidate); 161cf9a2ae8SDavid Howells 162ecdfc978SLinus Torvalds /* 1631da177e4SLinus Torvalds * If truncate cannot remove the fs-private metadata from the page, the page 16462e1c553SShaohua Li * becomes orphaned. It will be left on the LRU and may even be mapped into 16554cb8821SNick Piggin * user pagetables if we're racing with filemap_fault(). 1661da177e4SLinus Torvalds * 167fc3a5ac5SMatthew Wilcox (Oracle) * We need to bail out if page->mapping is no longer equal to the original 1681da177e4SLinus Torvalds * mapping. This happens a) when the VM reclaimed the page while we waited on 169fc0ecff6SAndrew Morton * its lock, b) when a concurrent invalidate_mapping_pages got there first and 1701da177e4SLinus Torvalds * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. 1711da177e4SLinus Torvalds */ 172efe99bbaSMatthew Wilcox (Oracle) static void truncate_cleanup_folio(struct folio *folio) 1731da177e4SLinus Torvalds { 174efe99bbaSMatthew Wilcox (Oracle) if (folio_mapped(folio)) 1753506659eSMatthew Wilcox (Oracle) unmap_mapping_folio(folio); 1761da177e4SLinus Torvalds 177efe99bbaSMatthew Wilcox (Oracle) if (folio_has_private(folio)) 1785ad6b2bdSMatthew Wilcox (Oracle) folio_invalidate(folio, 0, folio_size(folio)); 1791da177e4SLinus Torvalds 180b9ea2515SKonstantin Khlebnikov /* 181b9ea2515SKonstantin Khlebnikov * Some filesystems seem to re-dirty the page even after 182b9ea2515SKonstantin Khlebnikov * the VM has canceled the dirty bit (eg ext3 journaling). 183b9ea2515SKonstantin Khlebnikov * Hence dirty accounting check is placed after invalidation. 184b9ea2515SKonstantin Khlebnikov */ 185efe99bbaSMatthew Wilcox (Oracle) folio_cancel_dirty(folio); 186efe99bbaSMatthew Wilcox (Oracle) folio_clear_mappedtodisk(folio); 1871da177e4SLinus Torvalds } 1881da177e4SLinus Torvalds 1891e84a3d9SMatthew Wilcox (Oracle) int truncate_inode_folio(struct address_space *mapping, struct folio *folio) 190750b4987SNick Piggin { 1911e84a3d9SMatthew Wilcox (Oracle) if (folio->mapping != mapping) 1929f4e41f4SJan Kara return -EIO; 1939f4e41f4SJan Kara 194efe99bbaSMatthew Wilcox (Oracle) truncate_cleanup_folio(folio); 195efe99bbaSMatthew Wilcox (Oracle) filemap_remove_folio(folio); 1969f4e41f4SJan Kara return 0; 197750b4987SNick Piggin } 198750b4987SNick Piggin 19983f78668SWu Fengguang /* 200b9a8a419SMatthew Wilcox (Oracle) * Handle partial folios. The folio may be entirely within the 201b9a8a419SMatthew Wilcox (Oracle) * range if a split has raced with us. If not, we zero the part of the 202b9a8a419SMatthew Wilcox (Oracle) * folio that's within the [start, end] range, and then split the folio if 203b9a8a419SMatthew Wilcox (Oracle) * it's large. split_page_range() will discard pages which now lie beyond 204b9a8a419SMatthew Wilcox (Oracle) * i_size, and we rely on the caller to discard pages which lie within a 205b9a8a419SMatthew Wilcox (Oracle) * newly created hole. 206b9a8a419SMatthew Wilcox (Oracle) * 207b9a8a419SMatthew Wilcox (Oracle) * Returns false if splitting failed so the caller can avoid 208b9a8a419SMatthew Wilcox (Oracle) * discarding the entire folio which is stubbornly unsplit. 209b9a8a419SMatthew Wilcox (Oracle) */ 210b9a8a419SMatthew Wilcox (Oracle) bool truncate_inode_partial_folio(struct folio *folio, loff_t start, loff_t end) 211b9a8a419SMatthew Wilcox (Oracle) { 212b9a8a419SMatthew Wilcox (Oracle) loff_t pos = folio_pos(folio); 213b9a8a419SMatthew Wilcox (Oracle) unsigned int offset, length; 214b9a8a419SMatthew Wilcox (Oracle) 215b9a8a419SMatthew Wilcox (Oracle) if (pos < start) 216b9a8a419SMatthew Wilcox (Oracle) offset = start - pos; 217b9a8a419SMatthew Wilcox (Oracle) else 218b9a8a419SMatthew Wilcox (Oracle) offset = 0; 219b9a8a419SMatthew Wilcox (Oracle) length = folio_size(folio); 220b9a8a419SMatthew Wilcox (Oracle) if (pos + length <= (u64)end) 221b9a8a419SMatthew Wilcox (Oracle) length = length - offset; 222b9a8a419SMatthew Wilcox (Oracle) else 223b9a8a419SMatthew Wilcox (Oracle) length = end + 1 - pos - offset; 224b9a8a419SMatthew Wilcox (Oracle) 225b9a8a419SMatthew Wilcox (Oracle) folio_wait_writeback(folio); 226b9a8a419SMatthew Wilcox (Oracle) if (length == folio_size(folio)) { 227b9a8a419SMatthew Wilcox (Oracle) truncate_inode_folio(folio->mapping, folio); 228b9a8a419SMatthew Wilcox (Oracle) return true; 229b9a8a419SMatthew Wilcox (Oracle) } 230b9a8a419SMatthew Wilcox (Oracle) 231b9a8a419SMatthew Wilcox (Oracle) /* 232b9a8a419SMatthew Wilcox (Oracle) * We may be zeroing pages we're about to discard, but it avoids 233b9a8a419SMatthew Wilcox (Oracle) * doing a complex calculation here, and then doing the zeroing 234b9a8a419SMatthew Wilcox (Oracle) * anyway if the page split fails. 235b9a8a419SMatthew Wilcox (Oracle) */ 236b9a8a419SMatthew Wilcox (Oracle) folio_zero_range(folio, offset, length); 237b9a8a419SMatthew Wilcox (Oracle) 238b9a8a419SMatthew Wilcox (Oracle) if (folio_has_private(folio)) 2395ad6b2bdSMatthew Wilcox (Oracle) folio_invalidate(folio, offset, length); 240b9a8a419SMatthew Wilcox (Oracle) if (!folio_test_large(folio)) 241b9a8a419SMatthew Wilcox (Oracle) return true; 242d788f5b3SMatthew Wilcox (Oracle) if (split_folio(folio) == 0) 243b9a8a419SMatthew Wilcox (Oracle) return true; 244b9a8a419SMatthew Wilcox (Oracle) if (folio_test_dirty(folio)) 245b9a8a419SMatthew Wilcox (Oracle) return false; 246b9a8a419SMatthew Wilcox (Oracle) truncate_inode_folio(folio->mapping, folio); 247b9a8a419SMatthew Wilcox (Oracle) return true; 248b9a8a419SMatthew Wilcox (Oracle) } 249b9a8a419SMatthew Wilcox (Oracle) 250b9a8a419SMatthew Wilcox (Oracle) /* 25125718736SAndi Kleen * Used to get rid of pages on hardware memory corruption. 25225718736SAndi Kleen */ 25325718736SAndi Kleen int generic_error_remove_page(struct address_space *mapping, struct page *page) 25425718736SAndi Kleen { 2551e84a3d9SMatthew Wilcox (Oracle) VM_BUG_ON_PAGE(PageTail(page), page); 2561e84a3d9SMatthew Wilcox (Oracle) 25725718736SAndi Kleen if (!mapping) 25825718736SAndi Kleen return -EINVAL; 25925718736SAndi Kleen /* 26025718736SAndi Kleen * Only punch for normal data pages for now. 26125718736SAndi Kleen * Handling other types like directories would need more auditing. 26225718736SAndi Kleen */ 26325718736SAndi Kleen if (!S_ISREG(mapping->host->i_mode)) 26425718736SAndi Kleen return -EIO; 2651e84a3d9SMatthew Wilcox (Oracle) return truncate_inode_folio(mapping, page_folio(page)); 26625718736SAndi Kleen } 26725718736SAndi Kleen EXPORT_SYMBOL(generic_error_remove_page); 26825718736SAndi Kleen 2691e12cbb9SMatthew Wilcox (Oracle) /** 2701e12cbb9SMatthew Wilcox (Oracle) * mapping_evict_folio() - Remove an unused folio from the page-cache. 2711e12cbb9SMatthew Wilcox (Oracle) * @mapping: The mapping this folio belongs to. 2721e12cbb9SMatthew Wilcox (Oracle) * @folio: The folio to remove. 2731e12cbb9SMatthew Wilcox (Oracle) * 2741e12cbb9SMatthew Wilcox (Oracle) * Safely remove one folio from the page cache. 2751e12cbb9SMatthew Wilcox (Oracle) * It only drops clean, unused folios. 2761e12cbb9SMatthew Wilcox (Oracle) * 2771e12cbb9SMatthew Wilcox (Oracle) * Context: Folio must be locked. 2781e12cbb9SMatthew Wilcox (Oracle) * Return: The number of pages successfully removed. 2791e12cbb9SMatthew Wilcox (Oracle) */ 2801e12cbb9SMatthew Wilcox (Oracle) long mapping_evict_folio(struct address_space *mapping, struct folio *folio) 28183f78668SWu Fengguang { 2821e12cbb9SMatthew Wilcox (Oracle) /* The page may have been truncated before it was locked */ 2831e12cbb9SMatthew Wilcox (Oracle) if (!mapping) 2841e12cbb9SMatthew Wilcox (Oracle) return 0; 28544184813SMatthew Wilcox (Oracle) if (folio_test_dirty(folio) || folio_test_writeback(folio)) 28683f78668SWu Fengguang return 0; 287e41c81d0SMatthew Wilcox (Oracle) /* The refcount will be elevated if any page in the folio is mapped */ 288e41c81d0SMatthew Wilcox (Oracle) if (folio_ref_count(folio) > 289e41c81d0SMatthew Wilcox (Oracle) folio_nr_pages(folio) + folio_has_private(folio) + 1) 29083f78668SWu Fengguang return 0; 2910201ebf2SDavid Howells if (!filemap_release_folio(folio, 0)) 2921b8ddbeeSMatthew Wilcox (Oracle) return 0; 2931b8ddbeeSMatthew Wilcox (Oracle) 2945100da38SMatthew Wilcox (Oracle) return remove_mapping(mapping, folio); 29583f78668SWu Fengguang } 29683f78668SWu Fengguang 297d6c75dc2SMatthew Wilcox (Oracle) /** 29873c1e204SLiu Bo * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets 2991da177e4SLinus Torvalds * @mapping: mapping to truncate 3001da177e4SLinus Torvalds * @lstart: offset from which to truncate 3015a720394SLukas Czerner * @lend: offset to which to truncate (inclusive) 3021da177e4SLinus Torvalds * 303d7339071SHans Reiser * Truncate the page cache, removing the pages that are between 3045a720394SLukas Czerner * specified offsets (and zeroing out partial pages 3055a720394SLukas Czerner * if lstart or lend + 1 is not page aligned). 3061da177e4SLinus Torvalds * 3071da177e4SLinus Torvalds * Truncate takes two passes - the first pass is nonblocking. It will not 3081da177e4SLinus Torvalds * block on page locks and it will not block on writeback. The second pass 3091da177e4SLinus Torvalds * will wait. This is to prevent as much IO as possible in the affected region. 3101da177e4SLinus Torvalds * The first pass will remove most pages, so the search cost of the second pass 3111da177e4SLinus Torvalds * is low. 3121da177e4SLinus Torvalds * 3131da177e4SLinus Torvalds * We pass down the cache-hot hint to the page freeing code. Even if the 3141da177e4SLinus Torvalds * mapping is large, it is probably the case that the final pages are the most 3151da177e4SLinus Torvalds * recently touched, and freeing happens in ascending file offset order. 3165a720394SLukas Czerner * 317f50015a5SMatthew Wilcox (Oracle) * Note that since ->invalidate_folio() accepts range to invalidate 3185a720394SLukas Czerner * truncate_inode_pages_range is able to handle cases where lend + 1 is not 3195a720394SLukas Czerner * page aligned properly. 3201da177e4SLinus Torvalds */ 321d7339071SHans Reiser void truncate_inode_pages_range(struct address_space *mapping, 322d7339071SHans Reiser loff_t lstart, loff_t lend) 3231da177e4SLinus Torvalds { 3245a720394SLukas Czerner pgoff_t start; /* inclusive */ 3255a720394SLukas Czerner pgoff_t end; /* exclusive */ 3260e499ed3SMatthew Wilcox (Oracle) struct folio_batch fbatch; 3270cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 328b85e0effSHugh Dickins pgoff_t index; 3291da177e4SLinus Torvalds int i; 330b9a8a419SMatthew Wilcox (Oracle) struct folio *folio; 331b9a8a419SMatthew Wilcox (Oracle) bool same_folio; 3321da177e4SLinus Torvalds 3337716506aSMatthew Wilcox (Oracle) if (mapping_empty(mapping)) 3340a4ee518SChristoph Hellwig return; 3351da177e4SLinus Torvalds 3365a720394SLukas Czerner /* 3375a720394SLukas Czerner * 'start' and 'end' always covers the range of pages to be fully 3385a720394SLukas Czerner * truncated. Partial pages are covered with 'partial_start' at the 3395a720394SLukas Czerner * start of the range and 'partial_end' at the end of the range. 3405a720394SLukas Czerner * Note that 'end' is exclusive while 'lend' is inclusive. 3415a720394SLukas Czerner */ 34209cbfeafSKirill A. Shutemov start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; 3435a720394SLukas Czerner if (lend == -1) 3445a720394SLukas Czerner /* 3455a720394SLukas Czerner * lend == -1 indicates end-of-file so we have to set 'end' 3465a720394SLukas Czerner * to the highest possible pgoff_t and since the type is 3475a720394SLukas Czerner * unsigned we're using -1. 3485a720394SLukas Czerner */ 3495a720394SLukas Czerner end = -1; 3505a720394SLukas Czerner else 35109cbfeafSKirill A. Shutemov end = (lend + 1) >> PAGE_SHIFT; 352d7339071SHans Reiser 35351dcbdacSMatthew Wilcox (Oracle) folio_batch_init(&fbatch); 354b85e0effSHugh Dickins index = start; 3553392ca12SVishal Moola (Oracle) while (index < end && find_lock_entries(mapping, &index, end - 1, 35651dcbdacSMatthew Wilcox (Oracle) &fbatch, indices)) { 35751dcbdacSMatthew Wilcox (Oracle) truncate_folio_batch_exceptionals(mapping, &fbatch, indices); 35851dcbdacSMatthew Wilcox (Oracle) for (i = 0; i < folio_batch_count(&fbatch); i++) 35951dcbdacSMatthew Wilcox (Oracle) truncate_cleanup_folio(fbatch.folios[i]); 36051dcbdacSMatthew Wilcox (Oracle) delete_from_page_cache_batch(mapping, &fbatch); 36151dcbdacSMatthew Wilcox (Oracle) for (i = 0; i < folio_batch_count(&fbatch); i++) 36251dcbdacSMatthew Wilcox (Oracle) folio_unlock(fbatch.folios[i]); 36351dcbdacSMatthew Wilcox (Oracle) folio_batch_release(&fbatch); 3641da177e4SLinus Torvalds cond_resched(); 3651da177e4SLinus Torvalds } 3665c211ba2SMatthew Wilcox (Oracle) 367b9a8a419SMatthew Wilcox (Oracle) same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT); 368b9a8a419SMatthew Wilcox (Oracle) folio = __filemap_get_folio(mapping, lstart >> PAGE_SHIFT, FGP_LOCK, 0); 36966dabbb6SChristoph Hellwig if (!IS_ERR(folio)) { 370b9a8a419SMatthew Wilcox (Oracle) same_folio = lend < folio_pos(folio) + folio_size(folio); 371b9a8a419SMatthew Wilcox (Oracle) if (!truncate_inode_partial_folio(folio, lstart, lend)) { 37287b11f86SSidhartha Kumar start = folio_next_index(folio); 373b9a8a419SMatthew Wilcox (Oracle) if (same_folio) 374b9a8a419SMatthew Wilcox (Oracle) end = folio->index; 3755a720394SLukas Czerner } 376b9a8a419SMatthew Wilcox (Oracle) folio_unlock(folio); 377b9a8a419SMatthew Wilcox (Oracle) folio_put(folio); 378b9a8a419SMatthew Wilcox (Oracle) folio = NULL; 3791da177e4SLinus Torvalds } 380b9a8a419SMatthew Wilcox (Oracle) 38166dabbb6SChristoph Hellwig if (!same_folio) { 382b9a8a419SMatthew Wilcox (Oracle) folio = __filemap_get_folio(mapping, lend >> PAGE_SHIFT, 383b9a8a419SMatthew Wilcox (Oracle) FGP_LOCK, 0); 38466dabbb6SChristoph Hellwig if (!IS_ERR(folio)) { 385b9a8a419SMatthew Wilcox (Oracle) if (!truncate_inode_partial_folio(folio, lstart, lend)) 386b9a8a419SMatthew Wilcox (Oracle) end = folio->index; 387b9a8a419SMatthew Wilcox (Oracle) folio_unlock(folio); 388b9a8a419SMatthew Wilcox (Oracle) folio_put(folio); 3891da177e4SLinus Torvalds } 39066dabbb6SChristoph Hellwig } 3911da177e4SLinus Torvalds 392b85e0effSHugh Dickins index = start; 393b9a8a419SMatthew Wilcox (Oracle) while (index < end) { 3941da177e4SLinus Torvalds cond_resched(); 3959fb6beeaSVishal Moola (Oracle) if (!find_get_entries(mapping, &index, end - 1, &fbatch, 39638cefeb3SMatthew Wilcox (Oracle) indices)) { 397792ceaefSHugh Dickins /* If all gone from start onwards, we're done */ 398b85e0effSHugh Dickins if (index == start) 3991da177e4SLinus Torvalds break; 400792ceaefSHugh Dickins /* Otherwise restart to make sure all gone */ 401b85e0effSHugh Dickins index = start; 4021da177e4SLinus Torvalds continue; 4031da177e4SLinus Torvalds } 404f2187599SMel Gorman 4050e499ed3SMatthew Wilcox (Oracle) for (i = 0; i < folio_batch_count(&fbatch); i++) { 4060e499ed3SMatthew Wilcox (Oracle) struct folio *folio = fbatch.folios[i]; 4071da177e4SLinus Torvalds 408b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 409b85e0effSHugh Dickins 4100e499ed3SMatthew Wilcox (Oracle) if (xa_is_value(folio)) 4110cd6144aSJohannes Weiner continue; 4120cd6144aSJohannes Weiner 4131e84a3d9SMatthew Wilcox (Oracle) folio_lock(folio); 4149fb6beeaSVishal Moola (Oracle) VM_BUG_ON_FOLIO(!folio_contains(folio, indices[i]), folio); 4151e84a3d9SMatthew Wilcox (Oracle) folio_wait_writeback(folio); 4161e84a3d9SMatthew Wilcox (Oracle) truncate_inode_folio(mapping, folio); 4171e84a3d9SMatthew Wilcox (Oracle) folio_unlock(folio); 4181da177e4SLinus Torvalds } 4190e499ed3SMatthew Wilcox (Oracle) truncate_folio_batch_exceptionals(mapping, &fbatch, indices); 4200e499ed3SMatthew Wilcox (Oracle) folio_batch_release(&fbatch); 4211da177e4SLinus Torvalds } 4221da177e4SLinus Torvalds } 423d7339071SHans Reiser EXPORT_SYMBOL(truncate_inode_pages_range); 4241da177e4SLinus Torvalds 425d7339071SHans Reiser /** 426d7339071SHans Reiser * truncate_inode_pages - truncate *all* the pages from an offset 427d7339071SHans Reiser * @mapping: mapping to truncate 428d7339071SHans Reiser * @lstart: offset from which to truncate 429d7339071SHans Reiser * 430730633f0SJan Kara * Called under (and serialised by) inode->i_rwsem and 431730633f0SJan Kara * mapping->invalidate_lock. 43208142579SJan Kara * 43308142579SJan Kara * Note: When this function returns, there can be a page in the process of 4346ffcd825SMatthew Wilcox (Oracle) * deletion (inside __filemap_remove_folio()) in the specified range. Thus 43508142579SJan Kara * mapping->nrpages can be non-zero when this function returns even after 43608142579SJan Kara * truncation of the whole mapping. 437d7339071SHans Reiser */ 438d7339071SHans Reiser void truncate_inode_pages(struct address_space *mapping, loff_t lstart) 439d7339071SHans Reiser { 440d7339071SHans Reiser truncate_inode_pages_range(mapping, lstart, (loff_t)-1); 441d7339071SHans Reiser } 4421da177e4SLinus Torvalds EXPORT_SYMBOL(truncate_inode_pages); 4431da177e4SLinus Torvalds 44428697355SMike Waychison /** 44591b0abe3SJohannes Weiner * truncate_inode_pages_final - truncate *all* pages before inode dies 44691b0abe3SJohannes Weiner * @mapping: mapping to truncate 44791b0abe3SJohannes Weiner * 4489608703eSJan Kara * Called under (and serialized by) inode->i_rwsem. 44991b0abe3SJohannes Weiner * 45091b0abe3SJohannes Weiner * Filesystems have to use this in the .evict_inode path to inform the 45191b0abe3SJohannes Weiner * VM that this is the final truncate and the inode is going away. 45291b0abe3SJohannes Weiner */ 45391b0abe3SJohannes Weiner void truncate_inode_pages_final(struct address_space *mapping) 45491b0abe3SJohannes Weiner { 45591b0abe3SJohannes Weiner /* 45691b0abe3SJohannes Weiner * Page reclaim can not participate in regular inode lifetime 45791b0abe3SJohannes Weiner * management (can't call iput()) and thus can race with the 45891b0abe3SJohannes Weiner * inode teardown. Tell it when the address space is exiting, 45991b0abe3SJohannes Weiner * so that it does not install eviction information after the 46091b0abe3SJohannes Weiner * final truncate has begun. 46191b0abe3SJohannes Weiner */ 46291b0abe3SJohannes Weiner mapping_set_exiting(mapping); 46391b0abe3SJohannes Weiner 4647716506aSMatthew Wilcox (Oracle) if (!mapping_empty(mapping)) { 46591b0abe3SJohannes Weiner /* 46691b0abe3SJohannes Weiner * As truncation uses a lockless tree lookup, cycle 46791b0abe3SJohannes Weiner * the tree lock to make sure any ongoing tree 46891b0abe3SJohannes Weiner * modification that does not see AS_EXITING is 46991b0abe3SJohannes Weiner * completed before starting the final truncate. 47091b0abe3SJohannes Weiner */ 471b93b0163SMatthew Wilcox xa_lock_irq(&mapping->i_pages); 472b93b0163SMatthew Wilcox xa_unlock_irq(&mapping->i_pages); 47391b0abe3SJohannes Weiner } 4746ff38bd4SPavel Tikhomirov 4756ff38bd4SPavel Tikhomirov truncate_inode_pages(mapping, 0); 47691b0abe3SJohannes Weiner } 47791b0abe3SJohannes Weiner EXPORT_SYMBOL(truncate_inode_pages_final); 47891b0abe3SJohannes Weiner 479c56109ddSMatthew Wilcox (Oracle) /** 4801a0fc811SMatthew Wilcox (Oracle) * mapping_try_invalidate - Invalidate all the evictable folios of one inode 4811a0fc811SMatthew Wilcox (Oracle) * @mapping: the address_space which holds the folios to invalidate 482c56109ddSMatthew Wilcox (Oracle) * @start: the offset 'from' which to invalidate 483c56109ddSMatthew Wilcox (Oracle) * @end: the offset 'to' which to invalidate (inclusive) 4841a0fc811SMatthew Wilcox (Oracle) * @nr_failed: How many folio invalidations failed 485c56109ddSMatthew Wilcox (Oracle) * 4861a0fc811SMatthew Wilcox (Oracle) * This function is similar to invalidate_mapping_pages(), except that it 4871a0fc811SMatthew Wilcox (Oracle) * returns the number of folios which could not be evicted in @nr_failed. 488c56109ddSMatthew Wilcox (Oracle) */ 4891a0fc811SMatthew Wilcox (Oracle) unsigned long mapping_try_invalidate(struct address_space *mapping, 4901a0fc811SMatthew Wilcox (Oracle) pgoff_t start, pgoff_t end, unsigned long *nr_failed) 4911da177e4SLinus Torvalds { 4920cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 49351dcbdacSMatthew Wilcox (Oracle) struct folio_batch fbatch; 494b85e0effSHugh Dickins pgoff_t index = start; 49531560180SMinchan Kim unsigned long ret; 49631560180SMinchan Kim unsigned long count = 0; 4971da177e4SLinus Torvalds int i; 4981da177e4SLinus Torvalds 49951dcbdacSMatthew Wilcox (Oracle) folio_batch_init(&fbatch); 5003392ca12SVishal Moola (Oracle) while (find_lock_entries(mapping, &index, end, &fbatch, indices)) { 50151dcbdacSMatthew Wilcox (Oracle) for (i = 0; i < folio_batch_count(&fbatch); i++) { 502b4545f46SMatthew Wilcox (Oracle) struct folio *folio = fbatch.folios[i]; 5031da177e4SLinus Torvalds 504b4545f46SMatthew Wilcox (Oracle) /* We rely upon deletion not changing folio->index */ 505e0f23603SNeilBrown 506b4545f46SMatthew Wilcox (Oracle) if (xa_is_value(folio)) { 5077ae12c80SJohannes Weiner count += invalidate_exceptional_entry(mapping, 5083392ca12SVishal Moola (Oracle) indices[i], folio); 5090cd6144aSJohannes Weiner continue; 5100cd6144aSJohannes Weiner } 511fc127da0SKirill A. Shutemov 512b4545f46SMatthew Wilcox (Oracle) ret = mapping_evict_folio(mapping, folio); 513b4545f46SMatthew Wilcox (Oracle) folio_unlock(folio); 51431560180SMinchan Kim /* 515b4545f46SMatthew Wilcox (Oracle) * Invalidation is a hint that the folio is no longer 51631560180SMinchan Kim * of interest and try to speed up its reclaim. 51731560180SMinchan Kim */ 518eb1d7a65SYafang Shao if (!ret) { 519261b6840SMatthew Wilcox (Oracle) deactivate_file_folio(folio); 5201a0fc811SMatthew Wilcox (Oracle) /* Likely in the lru cache of a remote CPU */ 5211a0fc811SMatthew Wilcox (Oracle) if (nr_failed) 5221a0fc811SMatthew Wilcox (Oracle) (*nr_failed)++; 523eb1d7a65SYafang Shao } 52431560180SMinchan Kim count += ret; 5251da177e4SLinus Torvalds } 52651dcbdacSMatthew Wilcox (Oracle) folio_batch_remove_exceptionals(&fbatch); 52751dcbdacSMatthew Wilcox (Oracle) folio_batch_release(&fbatch); 528fc9a07e7SAndrew Morton cond_resched(); 5291da177e4SLinus Torvalds } 53031560180SMinchan Kim return count; 5311da177e4SLinus Torvalds } 532eb1d7a65SYafang Shao 533eb1d7a65SYafang Shao /** 5347ae12c80SJohannes Weiner * invalidate_mapping_pages - Invalidate all clean, unlocked cache of one inode 5357ae12c80SJohannes Weiner * @mapping: the address_space which holds the cache to invalidate 536eb1d7a65SYafang Shao * @start: the offset 'from' which to invalidate 537eb1d7a65SYafang Shao * @end: the offset 'to' which to invalidate (inclusive) 538eb1d7a65SYafang Shao * 5397ae12c80SJohannes Weiner * This function removes pages that are clean, unmapped and unlocked, 5407ae12c80SJohannes Weiner * as well as shadow entries. It will not block on IO activity. 541eb1d7a65SYafang Shao * 5427ae12c80SJohannes Weiner * If you want to remove all the pages of one inode, regardless of 5437ae12c80SJohannes Weiner * their use and writeback state, use truncate_inode_pages(). 544eb1d7a65SYafang Shao * 5451a0fc811SMatthew Wilcox (Oracle) * Return: The number of indices that had their contents invalidated 546eb1d7a65SYafang Shao */ 547eb1d7a65SYafang Shao unsigned long invalidate_mapping_pages(struct address_space *mapping, 548eb1d7a65SYafang Shao pgoff_t start, pgoff_t end) 549eb1d7a65SYafang Shao { 5501a0fc811SMatthew Wilcox (Oracle) return mapping_try_invalidate(mapping, start, end, NULL); 551eb1d7a65SYafang Shao } 55254bc4855SAnton Altaparmakov EXPORT_SYMBOL(invalidate_mapping_pages); 5531da177e4SLinus Torvalds 554bd4c8ce4SAndrew Morton /* 555*2033c98cSMatthew Wilcox (Oracle) * This is like mapping_evict_folio(), except it ignores the folio's 556bd4c8ce4SAndrew Morton * refcount. We do this because invalidate_inode_pages2() needs stronger 557*2033c98cSMatthew Wilcox (Oracle) * invalidation guarantees, and cannot afford to leave folios behind because 5582706a1b8SAnderson Briglia * shrink_page_list() has a temp ref on them, or because they're transiently 5591fec6890SMatthew Wilcox (Oracle) * sitting in the folio_add_lru() caches. 560bd4c8ce4SAndrew Morton */ 56178f42660SMatthew Wilcox (Oracle) static int invalidate_complete_folio2(struct address_space *mapping, 56278f42660SMatthew Wilcox (Oracle) struct folio *folio) 563bd4c8ce4SAndrew Morton { 56478f42660SMatthew Wilcox (Oracle) if (folio->mapping != mapping) 565bd4c8ce4SAndrew Morton return 0; 566bd4c8ce4SAndrew Morton 5670201ebf2SDavid Howells if (!filemap_release_folio(folio, GFP_KERNEL)) 568bd4c8ce4SAndrew Morton return 0; 569bd4c8ce4SAndrew Morton 57051b8c1feSJohannes Weiner spin_lock(&mapping->host->i_lock); 57130472509SJohannes Weiner xa_lock_irq(&mapping->i_pages); 57278f42660SMatthew Wilcox (Oracle) if (folio_test_dirty(folio)) 573bd4c8ce4SAndrew Morton goto failed; 574bd4c8ce4SAndrew Morton 57578f42660SMatthew Wilcox (Oracle) BUG_ON(folio_has_private(folio)); 57678f42660SMatthew Wilcox (Oracle) __filemap_remove_folio(folio, NULL); 57730472509SJohannes Weiner xa_unlock_irq(&mapping->i_pages); 57851b8c1feSJohannes Weiner if (mapping_shrinkable(mapping)) 57951b8c1feSJohannes Weiner inode_add_lru(mapping->host); 58051b8c1feSJohannes Weiner spin_unlock(&mapping->host->i_lock); 5816072d13cSLinus Torvalds 58278f42660SMatthew Wilcox (Oracle) filemap_free_folio(mapping, folio); 583bd4c8ce4SAndrew Morton return 1; 584bd4c8ce4SAndrew Morton failed: 58530472509SJohannes Weiner xa_unlock_irq(&mapping->i_pages); 58651b8c1feSJohannes Weiner spin_unlock(&mapping->host->i_lock); 587bd4c8ce4SAndrew Morton return 0; 588bd4c8ce4SAndrew Morton } 589bd4c8ce4SAndrew Morton 590affa80e8SMatthew Wilcox (Oracle) static int folio_launder(struct address_space *mapping, struct folio *folio) 591e3db7691STrond Myklebust { 592f6357c3aSMatthew Wilcox (Oracle) if (!folio_test_dirty(folio)) 593e3db7691STrond Myklebust return 0; 594affa80e8SMatthew Wilcox (Oracle) if (folio->mapping != mapping || mapping->a_ops->launder_folio == NULL) 595e3db7691STrond Myklebust return 0; 596affa80e8SMatthew Wilcox (Oracle) return mapping->a_ops->launder_folio(folio); 597e3db7691STrond Myklebust } 598e3db7691STrond Myklebust 5991da177e4SLinus Torvalds /** 6001da177e4SLinus Torvalds * invalidate_inode_pages2_range - remove range of pages from an address_space 60167be2dd1SMartin Waitz * @mapping: the address_space 6021da177e4SLinus Torvalds * @start: the page offset 'from' which to invalidate 6031da177e4SLinus Torvalds * @end: the page offset 'to' which to invalidate (inclusive) 6041da177e4SLinus Torvalds * 6051da177e4SLinus Torvalds * Any pages which are found to be mapped into pagetables are unmapped prior to 6061da177e4SLinus Torvalds * invalidation. 6071da177e4SLinus Torvalds * 608a862f68aSMike Rapoport * Return: -EBUSY if any pages could not be invalidated. 6091da177e4SLinus Torvalds */ 6101da177e4SLinus Torvalds int invalidate_inode_pages2_range(struct address_space *mapping, 6111da177e4SLinus Torvalds pgoff_t start, pgoff_t end) 6121da177e4SLinus Torvalds { 6130cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 6140e499ed3SMatthew Wilcox (Oracle) struct folio_batch fbatch; 615b85e0effSHugh Dickins pgoff_t index; 6161da177e4SLinus Torvalds int i; 6171da177e4SLinus Torvalds int ret = 0; 6180dd1334fSHisashi Hifumi int ret2 = 0; 6191da177e4SLinus Torvalds int did_range_unmap = 0; 6201da177e4SLinus Torvalds 6217716506aSMatthew Wilcox (Oracle) if (mapping_empty(mapping)) 6220a4ee518SChristoph Hellwig return 0; 62332691f0fSAndrey Ryabinin 6240e499ed3SMatthew Wilcox (Oracle) folio_batch_init(&fbatch); 625b85e0effSHugh Dickins index = start; 6269fb6beeaSVishal Moola (Oracle) while (find_get_entries(mapping, &index, end, &fbatch, indices)) { 6270e499ed3SMatthew Wilcox (Oracle) for (i = 0; i < folio_batch_count(&fbatch); i++) { 6280e499ed3SMatthew Wilcox (Oracle) struct folio *folio = fbatch.folios[i]; 629b85e0effSHugh Dickins 630fae9bc4aSMatthew Wilcox (Oracle) /* We rely upon deletion not changing folio->index */ 6311da177e4SLinus Torvalds 6320e499ed3SMatthew Wilcox (Oracle) if (xa_is_value(folio)) { 633c6dcf52cSJan Kara if (!invalidate_exceptional_entry2(mapping, 6349fb6beeaSVishal Moola (Oracle) indices[i], folio)) 635c6dcf52cSJan Kara ret = -EBUSY; 6360cd6144aSJohannes Weiner continue; 6370cd6144aSJohannes Weiner } 6380cd6144aSJohannes Weiner 639fae9bc4aSMatthew Wilcox (Oracle) if (!did_range_unmap && folio_mapped(folio)) { 64022061a1fSHugh Dickins /* 641fae9bc4aSMatthew Wilcox (Oracle) * If folio is mapped, before taking its lock, 64222061a1fSHugh Dickins * zap the rest of the file in one hit. 64322061a1fSHugh Dickins */ 6449fb6beeaSVishal Moola (Oracle) unmap_mapping_pages(mapping, indices[i], 6459fb6beeaSVishal Moola (Oracle) (1 + end - indices[i]), false); 64622061a1fSHugh Dickins did_range_unmap = 1; 64722061a1fSHugh Dickins } 64822061a1fSHugh Dickins 649fae9bc4aSMatthew Wilcox (Oracle) folio_lock(folio); 650aa5b9178SHugh Dickins if (unlikely(folio->mapping != mapping)) { 651fae9bc4aSMatthew Wilcox (Oracle) folio_unlock(folio); 6521da177e4SLinus Torvalds continue; 6531da177e4SLinus Torvalds } 654aa5b9178SHugh Dickins VM_BUG_ON_FOLIO(!folio_contains(folio, indices[i]), folio); 655fae9bc4aSMatthew Wilcox (Oracle) folio_wait_writeback(folio); 65622061a1fSHugh Dickins 657fae9bc4aSMatthew Wilcox (Oracle) if (folio_mapped(folio)) 658fae9bc4aSMatthew Wilcox (Oracle) unmap_mapping_folio(folio); 659fae9bc4aSMatthew Wilcox (Oracle) BUG_ON(folio_mapped(folio)); 66022061a1fSHugh Dickins 661affa80e8SMatthew Wilcox (Oracle) ret2 = folio_launder(mapping, folio); 6620dd1334fSHisashi Hifumi if (ret2 == 0) { 66378f42660SMatthew Wilcox (Oracle) if (!invalidate_complete_folio2(mapping, folio)) 6646ccfa806SHisashi Hifumi ret2 = -EBUSY; 6650dd1334fSHisashi Hifumi } 6660dd1334fSHisashi Hifumi if (ret2 < 0) 6670dd1334fSHisashi Hifumi ret = ret2; 668fae9bc4aSMatthew Wilcox (Oracle) folio_unlock(folio); 6691da177e4SLinus Torvalds } 6700e499ed3SMatthew Wilcox (Oracle) folio_batch_remove_exceptionals(&fbatch); 6710e499ed3SMatthew Wilcox (Oracle) folio_batch_release(&fbatch); 6721da177e4SLinus Torvalds cond_resched(); 6731da177e4SLinus Torvalds } 674cd656375SJan Kara /* 67569b6c131SMatthew Wilcox * For DAX we invalidate page tables after invalidating page cache. We 676cd656375SJan Kara * could invalidate page tables while invalidating each entry however 677cd656375SJan Kara * that would be expensive. And doing range unmapping before doesn't 67869b6c131SMatthew Wilcox * work as we have no cheap way to find whether page cache entry didn't 679cd656375SJan Kara * get remapped later. 680cd656375SJan Kara */ 681cd656375SJan Kara if (dax_mapping(mapping)) { 682977fbdcdSMatthew Wilcox unmap_mapping_pages(mapping, start, end - start + 1, false); 683cd656375SJan Kara } 6841da177e4SLinus Torvalds return ret; 6851da177e4SLinus Torvalds } 6861da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); 6871da177e4SLinus Torvalds 6881da177e4SLinus Torvalds /** 6891da177e4SLinus Torvalds * invalidate_inode_pages2 - remove all pages from an address_space 69067be2dd1SMartin Waitz * @mapping: the address_space 6911da177e4SLinus Torvalds * 6921da177e4SLinus Torvalds * Any pages which are found to be mapped into pagetables are unmapped prior to 6931da177e4SLinus Torvalds * invalidation. 6941da177e4SLinus Torvalds * 695a862f68aSMike Rapoport * Return: -EBUSY if any pages could not be invalidated. 6961da177e4SLinus Torvalds */ 6971da177e4SLinus Torvalds int invalidate_inode_pages2(struct address_space *mapping) 6981da177e4SLinus Torvalds { 6991da177e4SLinus Torvalds return invalidate_inode_pages2_range(mapping, 0, -1); 7001da177e4SLinus Torvalds } 7011da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(invalidate_inode_pages2); 70225d9e2d1Snpiggin@suse.de 70325d9e2d1Snpiggin@suse.de /** 70425d9e2d1Snpiggin@suse.de * truncate_pagecache - unmap and remove pagecache that has been truncated 70525d9e2d1Snpiggin@suse.de * @inode: inode 7068a549beaSHugh Dickins * @newsize: new file size 70725d9e2d1Snpiggin@suse.de * 70825d9e2d1Snpiggin@suse.de * inode's new i_size must already be written before truncate_pagecache 70925d9e2d1Snpiggin@suse.de * is called. 71025d9e2d1Snpiggin@suse.de * 71125d9e2d1Snpiggin@suse.de * This function should typically be called before the filesystem 71225d9e2d1Snpiggin@suse.de * releases resources associated with the freed range (eg. deallocates 71325d9e2d1Snpiggin@suse.de * blocks). This way, pagecache will always stay logically coherent 71425d9e2d1Snpiggin@suse.de * with on-disk format, and the filesystem would not have to deal with 71525d9e2d1Snpiggin@suse.de * situations such as writepage being called for a page that has already 71625d9e2d1Snpiggin@suse.de * had its underlying blocks deallocated. 71725d9e2d1Snpiggin@suse.de */ 7187caef267SKirill A. Shutemov void truncate_pagecache(struct inode *inode, loff_t newsize) 71925d9e2d1Snpiggin@suse.de { 72025d9e2d1Snpiggin@suse.de struct address_space *mapping = inode->i_mapping; 7218a549beaSHugh Dickins loff_t holebegin = round_up(newsize, PAGE_SIZE); 72225d9e2d1Snpiggin@suse.de 72325d9e2d1Snpiggin@suse.de /* 72425d9e2d1Snpiggin@suse.de * unmap_mapping_range is called twice, first simply for 72525d9e2d1Snpiggin@suse.de * efficiency so that truncate_inode_pages does fewer 72625d9e2d1Snpiggin@suse.de * single-page unmaps. However after this first call, and 72725d9e2d1Snpiggin@suse.de * before truncate_inode_pages finishes, it is possible for 72825d9e2d1Snpiggin@suse.de * private pages to be COWed, which remain after 72925d9e2d1Snpiggin@suse.de * truncate_inode_pages finishes, hence the second 73025d9e2d1Snpiggin@suse.de * unmap_mapping_range call must be made for correctness. 73125d9e2d1Snpiggin@suse.de */ 7328a549beaSHugh Dickins unmap_mapping_range(mapping, holebegin, 0, 1); 7338a549beaSHugh Dickins truncate_inode_pages(mapping, newsize); 7348a549beaSHugh Dickins unmap_mapping_range(mapping, holebegin, 0, 1); 73525d9e2d1Snpiggin@suse.de } 73625d9e2d1Snpiggin@suse.de EXPORT_SYMBOL(truncate_pagecache); 73725d9e2d1Snpiggin@suse.de 73825d9e2d1Snpiggin@suse.de /** 7392c27c65eSChristoph Hellwig * truncate_setsize - update inode and pagecache for a new file size 7402c27c65eSChristoph Hellwig * @inode: inode 7412c27c65eSChristoph Hellwig * @newsize: new file size 7422c27c65eSChristoph Hellwig * 743382e27daSJan Kara * truncate_setsize updates i_size and performs pagecache truncation (if 744382e27daSJan Kara * necessary) to @newsize. It will be typically be called from the filesystem's 745382e27daSJan Kara * setattr function when ATTR_SIZE is passed in. 7462c27c65eSChristoph Hellwig * 74777783d06SJan Kara * Must be called with a lock serializing truncates and writes (generally 7489608703eSJan Kara * i_rwsem but e.g. xfs uses a different lock) and before all filesystem 74977783d06SJan Kara * specific block truncation has been performed. 7502c27c65eSChristoph Hellwig */ 7512c27c65eSChristoph Hellwig void truncate_setsize(struct inode *inode, loff_t newsize) 7522c27c65eSChristoph Hellwig { 75390a80202SJan Kara loff_t oldsize = inode->i_size; 75490a80202SJan Kara 7552c27c65eSChristoph Hellwig i_size_write(inode, newsize); 75690a80202SJan Kara if (newsize > oldsize) 75790a80202SJan Kara pagecache_isize_extended(inode, oldsize, newsize); 7587caef267SKirill A. Shutemov truncate_pagecache(inode, newsize); 7592c27c65eSChristoph Hellwig } 7602c27c65eSChristoph Hellwig EXPORT_SYMBOL(truncate_setsize); 7612c27c65eSChristoph Hellwig 7622c27c65eSChristoph Hellwig /** 76390a80202SJan Kara * pagecache_isize_extended - update pagecache after extension of i_size 76490a80202SJan Kara * @inode: inode for which i_size was extended 76590a80202SJan Kara * @from: original inode size 76690a80202SJan Kara * @to: new inode size 76790a80202SJan Kara * 76890a80202SJan Kara * Handle extension of inode size either caused by extending truncate or by 76990a80202SJan Kara * write starting after current i_size. We mark the page straddling current 77090a80202SJan Kara * i_size RO so that page_mkwrite() is called on the nearest write access to 77190a80202SJan Kara * the page. This way filesystem can be sure that page_mkwrite() is called on 77290a80202SJan Kara * the page before user writes to the page via mmap after the i_size has been 77390a80202SJan Kara * changed. 77490a80202SJan Kara * 77590a80202SJan Kara * The function must be called after i_size is updated so that page fault 77690a80202SJan Kara * coming after we unlock the page will already see the new i_size. 7779608703eSJan Kara * The function must be called while we still hold i_rwsem - this not only 77890a80202SJan Kara * makes sure i_size is stable but also that userspace cannot observe new 77990a80202SJan Kara * i_size value before we are prepared to store mmap writes at new inode size. 78090a80202SJan Kara */ 78190a80202SJan Kara void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) 78290a80202SJan Kara { 78393407472SFabian Frederick int bsize = i_blocksize(inode); 78490a80202SJan Kara loff_t rounded_from; 78590a80202SJan Kara struct page *page; 78690a80202SJan Kara pgoff_t index; 78790a80202SJan Kara 78890a80202SJan Kara WARN_ON(to > inode->i_size); 78990a80202SJan Kara 79009cbfeafSKirill A. Shutemov if (from >= to || bsize == PAGE_SIZE) 79190a80202SJan Kara return; 79290a80202SJan Kara /* Page straddling @from will not have any hole block created? */ 79390a80202SJan Kara rounded_from = round_up(from, bsize); 79409cbfeafSKirill A. Shutemov if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1))) 79590a80202SJan Kara return; 79690a80202SJan Kara 79709cbfeafSKirill A. Shutemov index = from >> PAGE_SHIFT; 79890a80202SJan Kara page = find_lock_page(inode->i_mapping, index); 79990a80202SJan Kara /* Page not cached? Nothing to do */ 80090a80202SJan Kara if (!page) 80190a80202SJan Kara return; 80290a80202SJan Kara /* 80390a80202SJan Kara * See clear_page_dirty_for_io() for details why set_page_dirty() 80490a80202SJan Kara * is needed. 80590a80202SJan Kara */ 80690a80202SJan Kara if (page_mkclean(page)) 80790a80202SJan Kara set_page_dirty(page); 80890a80202SJan Kara unlock_page(page); 80909cbfeafSKirill A. Shutemov put_page(page); 81090a80202SJan Kara } 81190a80202SJan Kara EXPORT_SYMBOL(pagecache_isize_extended); 81290a80202SJan Kara 81390a80202SJan Kara /** 814623e3db9SHugh Dickins * truncate_pagecache_range - unmap and remove pagecache that is hole-punched 815623e3db9SHugh Dickins * @inode: inode 816623e3db9SHugh Dickins * @lstart: offset of beginning of hole 817623e3db9SHugh Dickins * @lend: offset of last byte of hole 818623e3db9SHugh Dickins * 819623e3db9SHugh Dickins * This function should typically be called before the filesystem 820623e3db9SHugh Dickins * releases resources associated with the freed range (eg. deallocates 821623e3db9SHugh Dickins * blocks). This way, pagecache will always stay logically coherent 822623e3db9SHugh Dickins * with on-disk format, and the filesystem would not have to deal with 823623e3db9SHugh Dickins * situations such as writepage being called for a page that has already 824623e3db9SHugh Dickins * had its underlying blocks deallocated. 825623e3db9SHugh Dickins */ 826623e3db9SHugh Dickins void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) 827623e3db9SHugh Dickins { 828623e3db9SHugh Dickins struct address_space *mapping = inode->i_mapping; 829623e3db9SHugh Dickins loff_t unmap_start = round_up(lstart, PAGE_SIZE); 830623e3db9SHugh Dickins loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; 831623e3db9SHugh Dickins /* 832623e3db9SHugh Dickins * This rounding is currently just for example: unmap_mapping_range 833623e3db9SHugh Dickins * expands its hole outwards, whereas we want it to contract the hole 834623e3db9SHugh Dickins * inwards. However, existing callers of truncate_pagecache_range are 8355a720394SLukas Czerner * doing their own page rounding first. Note that unmap_mapping_range 8365a720394SLukas Czerner * allows holelen 0 for all, and we allow lend -1 for end of file. 837623e3db9SHugh Dickins */ 838623e3db9SHugh Dickins 839623e3db9SHugh Dickins /* 840623e3db9SHugh Dickins * Unlike in truncate_pagecache, unmap_mapping_range is called only 841623e3db9SHugh Dickins * once (before truncating pagecache), and without "even_cows" flag: 842623e3db9SHugh Dickins * hole-punching should not remove private COWed pages from the hole. 843623e3db9SHugh Dickins */ 844623e3db9SHugh Dickins if ((u64)unmap_end > (u64)unmap_start) 845623e3db9SHugh Dickins unmap_mapping_range(mapping, unmap_start, 846623e3db9SHugh Dickins 1 + unmap_end - unmap_start, 0); 847623e3db9SHugh Dickins truncate_inode_pages_range(mapping, lstart, lend); 848623e3db9SHugh Dickins } 849623e3db9SHugh Dickins EXPORT_SYMBOL(truncate_pagecache_range); 850