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> 22f50015a5SMatthew Wilcox (Oracle) #include <linux/buffer_head.h> /* grr. try_to_release_page */ 233a4f8a0bSHugh Dickins #include <linux/shmem_fs.h> 2490a80202SJan Kara #include <linux/rmap.h> 25ba470de4SRik van Riel #include "internal.h" 261da177e4SLinus Torvalds 27f2187599SMel Gorman /* 28f2187599SMel Gorman * Regular page slots are stabilized by the page lock even without the tree 29f2187599SMel Gorman * itself locked. These unlocked entries need verification under the tree 30f2187599SMel Gorman * lock. 31f2187599SMel Gorman */ 32f2187599SMel Gorman static inline void __clear_shadow_entry(struct address_space *mapping, 33f2187599SMel Gorman pgoff_t index, void *entry) 340cd6144aSJohannes Weiner { 3569b6c131SMatthew Wilcox XA_STATE(xas, &mapping->i_pages, index); 36449dd698SJohannes Weiner 3769b6c131SMatthew Wilcox xas_set_update(&xas, workingset_update_node); 3869b6c131SMatthew Wilcox if (xas_load(&xas) != entry) 39f2187599SMel Gorman return; 4069b6c131SMatthew Wilcox xas_store(&xas, NULL); 41f2187599SMel Gorman } 42f2187599SMel Gorman 43f2187599SMel Gorman static void clear_shadow_entry(struct address_space *mapping, pgoff_t index, 44f2187599SMel Gorman void *entry) 45f2187599SMel Gorman { 4651b8c1feSJohannes Weiner spin_lock(&mapping->host->i_lock); 47b93b0163SMatthew Wilcox xa_lock_irq(&mapping->i_pages); 48f2187599SMel Gorman __clear_shadow_entry(mapping, index, entry); 49b93b0163SMatthew Wilcox xa_unlock_irq(&mapping->i_pages); 5051b8c1feSJohannes Weiner if (mapping_shrinkable(mapping)) 5151b8c1feSJohannes Weiner inode_add_lru(mapping->host); 5251b8c1feSJohannes Weiner spin_unlock(&mapping->host->i_lock); 530cd6144aSJohannes Weiner } 541da177e4SLinus Torvalds 55c6dcf52cSJan Kara /* 56f2187599SMel Gorman * Unconditionally remove exceptional entries. Usually called from truncate 5751dcbdacSMatthew Wilcox (Oracle) * path. Note that the folio_batch may be altered by this function by removing 581613fac9SMatthew Wilcox (Oracle) * exceptional entries similar to what folio_batch_remove_exceptionals() does. 59c6dcf52cSJan Kara */ 6051dcbdacSMatthew Wilcox (Oracle) static void truncate_folio_batch_exceptionals(struct address_space *mapping, 6151dcbdacSMatthew Wilcox (Oracle) struct folio_batch *fbatch, pgoff_t *indices) 62c6dcf52cSJan Kara { 63f2187599SMel Gorman int i, j; 6431d270fdSMatthew Wilcox (Oracle) bool dax; 65f2187599SMel Gorman 66c6dcf52cSJan Kara /* Handled by shmem itself */ 67c6dcf52cSJan Kara if (shmem_mapping(mapping)) 68c6dcf52cSJan Kara return; 69c6dcf52cSJan Kara 7051dcbdacSMatthew Wilcox (Oracle) for (j = 0; j < folio_batch_count(fbatch); j++) 7151dcbdacSMatthew Wilcox (Oracle) if (xa_is_value(fbatch->folios[j])) 72f2187599SMel Gorman break; 73f2187599SMel Gorman 7451dcbdacSMatthew Wilcox (Oracle) if (j == folio_batch_count(fbatch)) 75c6dcf52cSJan Kara return; 76f2187599SMel Gorman 77f2187599SMel Gorman dax = dax_mapping(mapping); 7851b8c1feSJohannes Weiner if (!dax) { 7951b8c1feSJohannes Weiner spin_lock(&mapping->host->i_lock); 80b93b0163SMatthew Wilcox xa_lock_irq(&mapping->i_pages); 8151b8c1feSJohannes Weiner } 82f2187599SMel Gorman 8351dcbdacSMatthew Wilcox (Oracle) for (i = j; i < folio_batch_count(fbatch); i++) { 8451dcbdacSMatthew Wilcox (Oracle) struct folio *folio = fbatch->folios[i]; 85f2187599SMel Gorman pgoff_t index = indices[i]; 86f2187599SMel Gorman 8751dcbdacSMatthew Wilcox (Oracle) if (!xa_is_value(folio)) { 8851dcbdacSMatthew Wilcox (Oracle) fbatch->folios[j++] = folio; 89f2187599SMel Gorman continue; 90c6dcf52cSJan Kara } 91f2187599SMel Gorman 92f2187599SMel Gorman if (unlikely(dax)) { 93f2187599SMel Gorman dax_delete_mapping_entry(mapping, index); 94f2187599SMel Gorman continue; 95f2187599SMel Gorman } 96f2187599SMel Gorman 9751dcbdacSMatthew Wilcox (Oracle) __clear_shadow_entry(mapping, index, folio); 98f2187599SMel Gorman } 99f2187599SMel Gorman 10051b8c1feSJohannes Weiner if (!dax) { 101b93b0163SMatthew Wilcox xa_unlock_irq(&mapping->i_pages); 10251b8c1feSJohannes Weiner if (mapping_shrinkable(mapping)) 10351b8c1feSJohannes Weiner inode_add_lru(mapping->host); 10451b8c1feSJohannes Weiner spin_unlock(&mapping->host->i_lock); 10551b8c1feSJohannes Weiner } 10651dcbdacSMatthew Wilcox (Oracle) fbatch->nr = j; 1070e499ed3SMatthew Wilcox (Oracle) } 1080e499ed3SMatthew Wilcox (Oracle) 109c6dcf52cSJan Kara /* 110c6dcf52cSJan Kara * Invalidate exceptional entry if easily possible. This handles exceptional 1114636e70bSRoss Zwisler * entries for invalidate_inode_pages(). 112c6dcf52cSJan Kara */ 113c6dcf52cSJan Kara static int invalidate_exceptional_entry(struct address_space *mapping, 114c6dcf52cSJan Kara pgoff_t index, void *entry) 115c6dcf52cSJan Kara { 1164636e70bSRoss Zwisler /* Handled by shmem itself, or for DAX we do nothing. */ 1174636e70bSRoss Zwisler if (shmem_mapping(mapping) || dax_mapping(mapping)) 118c6dcf52cSJan Kara return 1; 119c6dcf52cSJan Kara clear_shadow_entry(mapping, index, entry); 120c6dcf52cSJan Kara return 1; 121c6dcf52cSJan Kara } 122c6dcf52cSJan Kara 123c6dcf52cSJan Kara /* 124c6dcf52cSJan Kara * Invalidate exceptional entry if clean. This handles exceptional entries for 125c6dcf52cSJan Kara * invalidate_inode_pages2() so for DAX it evicts only clean entries. 126c6dcf52cSJan Kara */ 127c6dcf52cSJan Kara static int invalidate_exceptional_entry2(struct address_space *mapping, 128c6dcf52cSJan Kara pgoff_t index, void *entry) 129c6dcf52cSJan Kara { 130c6dcf52cSJan Kara /* Handled by shmem itself */ 131c6dcf52cSJan Kara if (shmem_mapping(mapping)) 132c6dcf52cSJan Kara return 1; 133c6dcf52cSJan Kara if (dax_mapping(mapping)) 134c6dcf52cSJan Kara return dax_invalidate_mapping_entry_sync(mapping, index); 135c6dcf52cSJan Kara clear_shadow_entry(mapping, index, entry); 136c6dcf52cSJan Kara return 1; 137c6dcf52cSJan Kara } 138c6dcf52cSJan Kara 139cf9a2ae8SDavid Howells /** 1405ad6b2bdSMatthew Wilcox (Oracle) * folio_invalidate - Invalidate part or all of a folio. 1415ad6b2bdSMatthew Wilcox (Oracle) * @folio: The folio which is affected. 142d47992f8SLukas Czerner * @offset: start of the range to invalidate 143d47992f8SLukas Czerner * @length: length of the range to invalidate 144cf9a2ae8SDavid Howells * 1455ad6b2bdSMatthew Wilcox (Oracle) * folio_invalidate() is called when all or part of the folio has become 146cf9a2ae8SDavid Howells * invalidated by a truncate operation. 147cf9a2ae8SDavid Howells * 1485ad6b2bdSMatthew Wilcox (Oracle) * folio_invalidate() does not have to release all buffers, but it must 149cf9a2ae8SDavid Howells * ensure that no dirty buffer is left outside @offset and that no I/O 150cf9a2ae8SDavid Howells * is underway against any of the blocks which are outside the truncation 151cf9a2ae8SDavid Howells * point. Because the caller is about to free (and possibly reuse) those 152cf9a2ae8SDavid Howells * blocks on-disk. 153cf9a2ae8SDavid Howells */ 1545ad6b2bdSMatthew Wilcox (Oracle) void folio_invalidate(struct folio *folio, size_t offset, size_t length) 155cf9a2ae8SDavid Howells { 156128d1f82SMatthew Wilcox (Oracle) const struct address_space_operations *aops = folio->mapping->a_ops; 157d47992f8SLukas Czerner 158f50015a5SMatthew Wilcox (Oracle) if (aops->invalidate_folio) 159128d1f82SMatthew Wilcox (Oracle) aops->invalidate_folio(folio, offset, length); 160cf9a2ae8SDavid Howells } 1615ad6b2bdSMatthew Wilcox (Oracle) EXPORT_SYMBOL_GPL(folio_invalidate); 162cf9a2ae8SDavid Howells 163ecdfc978SLinus Torvalds /* 1641da177e4SLinus Torvalds * If truncate cannot remove the fs-private metadata from the page, the page 16562e1c553SShaohua Li * becomes orphaned. It will be left on the LRU and may even be mapped into 16654cb8821SNick Piggin * user pagetables if we're racing with filemap_fault(). 1671da177e4SLinus Torvalds * 168fc3a5ac5SMatthew Wilcox (Oracle) * We need to bail out if page->mapping is no longer equal to the original 1691da177e4SLinus Torvalds * mapping. This happens a) when the VM reclaimed the page while we waited on 170fc0ecff6SAndrew Morton * its lock, b) when a concurrent invalidate_mapping_pages got there first and 1711da177e4SLinus Torvalds * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. 1721da177e4SLinus Torvalds */ 173efe99bbaSMatthew Wilcox (Oracle) static void truncate_cleanup_folio(struct folio *folio) 1741da177e4SLinus Torvalds { 175efe99bbaSMatthew Wilcox (Oracle) if (folio_mapped(folio)) 1763506659eSMatthew Wilcox (Oracle) unmap_mapping_folio(folio); 1771da177e4SLinus Torvalds 178efe99bbaSMatthew Wilcox (Oracle) if (folio_has_private(folio)) 1795ad6b2bdSMatthew Wilcox (Oracle) folio_invalidate(folio, 0, folio_size(folio)); 1801da177e4SLinus Torvalds 181b9ea2515SKonstantin Khlebnikov /* 182b9ea2515SKonstantin Khlebnikov * Some filesystems seem to re-dirty the page even after 183b9ea2515SKonstantin Khlebnikov * the VM has canceled the dirty bit (eg ext3 journaling). 184b9ea2515SKonstantin Khlebnikov * Hence dirty accounting check is placed after invalidation. 185b9ea2515SKonstantin Khlebnikov */ 186efe99bbaSMatthew Wilcox (Oracle) folio_cancel_dirty(folio); 187efe99bbaSMatthew Wilcox (Oracle) folio_clear_mappedtodisk(folio); 1881da177e4SLinus Torvalds } 1891da177e4SLinus Torvalds 1901e84a3d9SMatthew Wilcox (Oracle) int truncate_inode_folio(struct address_space *mapping, struct folio *folio) 191750b4987SNick Piggin { 1921e84a3d9SMatthew Wilcox (Oracle) if (folio->mapping != mapping) 1939f4e41f4SJan Kara return -EIO; 1949f4e41f4SJan Kara 195efe99bbaSMatthew Wilcox (Oracle) truncate_cleanup_folio(folio); 196efe99bbaSMatthew Wilcox (Oracle) filemap_remove_folio(folio); 1979f4e41f4SJan Kara return 0; 198750b4987SNick Piggin } 199750b4987SNick Piggin 20083f78668SWu Fengguang /* 201b9a8a419SMatthew Wilcox (Oracle) * Handle partial folios. The folio may be entirely within the 202b9a8a419SMatthew Wilcox (Oracle) * range if a split has raced with us. If not, we zero the part of the 203b9a8a419SMatthew Wilcox (Oracle) * folio that's within the [start, end] range, and then split the folio if 204b9a8a419SMatthew Wilcox (Oracle) * it's large. split_page_range() will discard pages which now lie beyond 205b9a8a419SMatthew Wilcox (Oracle) * i_size, and we rely on the caller to discard pages which lie within a 206b9a8a419SMatthew Wilcox (Oracle) * newly created hole. 207b9a8a419SMatthew Wilcox (Oracle) * 208b9a8a419SMatthew Wilcox (Oracle) * Returns false if splitting failed so the caller can avoid 209b9a8a419SMatthew Wilcox (Oracle) * discarding the entire folio which is stubbornly unsplit. 210b9a8a419SMatthew Wilcox (Oracle) */ 211b9a8a419SMatthew Wilcox (Oracle) bool truncate_inode_partial_folio(struct folio *folio, loff_t start, loff_t end) 212b9a8a419SMatthew Wilcox (Oracle) { 213b9a8a419SMatthew Wilcox (Oracle) loff_t pos = folio_pos(folio); 214b9a8a419SMatthew Wilcox (Oracle) unsigned int offset, length; 215b9a8a419SMatthew Wilcox (Oracle) 216b9a8a419SMatthew Wilcox (Oracle) if (pos < start) 217b9a8a419SMatthew Wilcox (Oracle) offset = start - pos; 218b9a8a419SMatthew Wilcox (Oracle) else 219b9a8a419SMatthew Wilcox (Oracle) offset = 0; 220b9a8a419SMatthew Wilcox (Oracle) length = folio_size(folio); 221b9a8a419SMatthew Wilcox (Oracle) if (pos + length <= (u64)end) 222b9a8a419SMatthew Wilcox (Oracle) length = length - offset; 223b9a8a419SMatthew Wilcox (Oracle) else 224b9a8a419SMatthew Wilcox (Oracle) length = end + 1 - pos - offset; 225b9a8a419SMatthew Wilcox (Oracle) 226b9a8a419SMatthew Wilcox (Oracle) folio_wait_writeback(folio); 227b9a8a419SMatthew Wilcox (Oracle) if (length == folio_size(folio)) { 228b9a8a419SMatthew Wilcox (Oracle) truncate_inode_folio(folio->mapping, folio); 229b9a8a419SMatthew Wilcox (Oracle) return true; 230b9a8a419SMatthew Wilcox (Oracle) } 231b9a8a419SMatthew Wilcox (Oracle) 232b9a8a419SMatthew Wilcox (Oracle) /* 233b9a8a419SMatthew Wilcox (Oracle) * We may be zeroing pages we're about to discard, but it avoids 234b9a8a419SMatthew Wilcox (Oracle) * doing a complex calculation here, and then doing the zeroing 235b9a8a419SMatthew Wilcox (Oracle) * anyway if the page split fails. 236b9a8a419SMatthew Wilcox (Oracle) */ 237b9a8a419SMatthew Wilcox (Oracle) folio_zero_range(folio, offset, length); 238b9a8a419SMatthew Wilcox (Oracle) 239b9a8a419SMatthew Wilcox (Oracle) if (folio_has_private(folio)) 2405ad6b2bdSMatthew Wilcox (Oracle) folio_invalidate(folio, offset, length); 241b9a8a419SMatthew Wilcox (Oracle) if (!folio_test_large(folio)) 242b9a8a419SMatthew Wilcox (Oracle) return true; 243d788f5b3SMatthew Wilcox (Oracle) if (split_folio(folio) == 0) 244b9a8a419SMatthew Wilcox (Oracle) return true; 245b9a8a419SMatthew Wilcox (Oracle) if (folio_test_dirty(folio)) 246b9a8a419SMatthew Wilcox (Oracle) return false; 247b9a8a419SMatthew Wilcox (Oracle) truncate_inode_folio(folio->mapping, folio); 248b9a8a419SMatthew Wilcox (Oracle) return true; 249b9a8a419SMatthew Wilcox (Oracle) } 250b9a8a419SMatthew Wilcox (Oracle) 251b9a8a419SMatthew Wilcox (Oracle) /* 25225718736SAndi Kleen * Used to get rid of pages on hardware memory corruption. 25325718736SAndi Kleen */ 25425718736SAndi Kleen int generic_error_remove_page(struct address_space *mapping, struct page *page) 25525718736SAndi Kleen { 2561e84a3d9SMatthew Wilcox (Oracle) VM_BUG_ON_PAGE(PageTail(page), page); 2571e84a3d9SMatthew Wilcox (Oracle) 25825718736SAndi Kleen if (!mapping) 25925718736SAndi Kleen return -EINVAL; 26025718736SAndi Kleen /* 26125718736SAndi Kleen * Only punch for normal data pages for now. 26225718736SAndi Kleen * Handling other types like directories would need more auditing. 26325718736SAndi Kleen */ 26425718736SAndi Kleen if (!S_ISREG(mapping->host->i_mode)) 26525718736SAndi Kleen return -EIO; 2661e84a3d9SMatthew Wilcox (Oracle) return truncate_inode_folio(mapping, page_folio(page)); 26725718736SAndi Kleen } 26825718736SAndi Kleen EXPORT_SYMBOL(generic_error_remove_page); 26925718736SAndi Kleen 270d6c75dc2SMatthew Wilcox (Oracle) static long mapping_evict_folio(struct address_space *mapping, 271d6c75dc2SMatthew Wilcox (Oracle) struct folio *folio) 27283f78668SWu Fengguang { 27344184813SMatthew Wilcox (Oracle) if (folio_test_dirty(folio) || folio_test_writeback(folio)) 27483f78668SWu Fengguang return 0; 275e41c81d0SMatthew Wilcox (Oracle) /* The refcount will be elevated if any page in the folio is mapped */ 276e41c81d0SMatthew Wilcox (Oracle) if (folio_ref_count(folio) > 277e41c81d0SMatthew Wilcox (Oracle) folio_nr_pages(folio) + folio_has_private(folio) + 1) 27883f78668SWu Fengguang return 0; 27944184813SMatthew Wilcox (Oracle) if (folio_has_private(folio) && !filemap_release_folio(folio, 0)) 2801b8ddbeeSMatthew Wilcox (Oracle) return 0; 2811b8ddbeeSMatthew Wilcox (Oracle) 2825100da38SMatthew Wilcox (Oracle) return remove_mapping(mapping, folio); 28383f78668SWu Fengguang } 28483f78668SWu Fengguang 2851da177e4SLinus Torvalds /** 286d6c75dc2SMatthew Wilcox (Oracle) * invalidate_inode_page() - Remove an unused page from the pagecache. 287d6c75dc2SMatthew Wilcox (Oracle) * @page: The page to remove. 288d6c75dc2SMatthew Wilcox (Oracle) * 289d6c75dc2SMatthew Wilcox (Oracle) * Safely invalidate one page from its pagecache mapping. 290d6c75dc2SMatthew Wilcox (Oracle) * It only drops clean, unused pages. 291d6c75dc2SMatthew Wilcox (Oracle) * 292d6c75dc2SMatthew Wilcox (Oracle) * Context: Page must be locked. 293d6c75dc2SMatthew Wilcox (Oracle) * Return: The number of pages successfully removed. 294d6c75dc2SMatthew Wilcox (Oracle) */ 295d6c75dc2SMatthew Wilcox (Oracle) long invalidate_inode_page(struct page *page) 296d6c75dc2SMatthew Wilcox (Oracle) { 297d6c75dc2SMatthew Wilcox (Oracle) struct folio *folio = page_folio(page); 298d6c75dc2SMatthew Wilcox (Oracle) struct address_space *mapping = folio_mapping(folio); 299d6c75dc2SMatthew Wilcox (Oracle) 300d6c75dc2SMatthew Wilcox (Oracle) /* The page may have been truncated before it was locked */ 301d6c75dc2SMatthew Wilcox (Oracle) if (!mapping) 302d6c75dc2SMatthew Wilcox (Oracle) return 0; 303d6c75dc2SMatthew Wilcox (Oracle) return mapping_evict_folio(mapping, folio); 304d6c75dc2SMatthew Wilcox (Oracle) } 305d6c75dc2SMatthew Wilcox (Oracle) 306d6c75dc2SMatthew Wilcox (Oracle) /** 30773c1e204SLiu Bo * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets 3081da177e4SLinus Torvalds * @mapping: mapping to truncate 3091da177e4SLinus Torvalds * @lstart: offset from which to truncate 3105a720394SLukas Czerner * @lend: offset to which to truncate (inclusive) 3111da177e4SLinus Torvalds * 312d7339071SHans Reiser * Truncate the page cache, removing the pages that are between 3135a720394SLukas Czerner * specified offsets (and zeroing out partial pages 3145a720394SLukas Czerner * if lstart or lend + 1 is not page aligned). 3151da177e4SLinus Torvalds * 3161da177e4SLinus Torvalds * Truncate takes two passes - the first pass is nonblocking. It will not 3171da177e4SLinus Torvalds * block on page locks and it will not block on writeback. The second pass 3181da177e4SLinus Torvalds * will wait. This is to prevent as much IO as possible in the affected region. 3191da177e4SLinus Torvalds * The first pass will remove most pages, so the search cost of the second pass 3201da177e4SLinus Torvalds * is low. 3211da177e4SLinus Torvalds * 3221da177e4SLinus Torvalds * We pass down the cache-hot hint to the page freeing code. Even if the 3231da177e4SLinus Torvalds * mapping is large, it is probably the case that the final pages are the most 3241da177e4SLinus Torvalds * recently touched, and freeing happens in ascending file offset order. 3255a720394SLukas Czerner * 326f50015a5SMatthew Wilcox (Oracle) * Note that since ->invalidate_folio() accepts range to invalidate 3275a720394SLukas Czerner * truncate_inode_pages_range is able to handle cases where lend + 1 is not 3285a720394SLukas Czerner * page aligned properly. 3291da177e4SLinus Torvalds */ 330d7339071SHans Reiser void truncate_inode_pages_range(struct address_space *mapping, 331d7339071SHans Reiser loff_t lstart, loff_t lend) 3321da177e4SLinus Torvalds { 3335a720394SLukas Czerner pgoff_t start; /* inclusive */ 3345a720394SLukas Czerner pgoff_t end; /* exclusive */ 3350e499ed3SMatthew Wilcox (Oracle) struct folio_batch fbatch; 3360cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 337b85e0effSHugh Dickins pgoff_t index; 3381da177e4SLinus Torvalds int i; 339b9a8a419SMatthew Wilcox (Oracle) struct folio *folio; 340b9a8a419SMatthew Wilcox (Oracle) bool same_folio; 3411da177e4SLinus Torvalds 3427716506aSMatthew Wilcox (Oracle) if (mapping_empty(mapping)) 3430a4ee518SChristoph Hellwig return; 3441da177e4SLinus Torvalds 3455a720394SLukas Czerner /* 3465a720394SLukas Czerner * 'start' and 'end' always covers the range of pages to be fully 3475a720394SLukas Czerner * truncated. Partial pages are covered with 'partial_start' at the 3485a720394SLukas Czerner * start of the range and 'partial_end' at the end of the range. 3495a720394SLukas Czerner * Note that 'end' is exclusive while 'lend' is inclusive. 3505a720394SLukas Czerner */ 35109cbfeafSKirill A. Shutemov start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT; 3525a720394SLukas Czerner if (lend == -1) 3535a720394SLukas Czerner /* 3545a720394SLukas Czerner * lend == -1 indicates end-of-file so we have to set 'end' 3555a720394SLukas Czerner * to the highest possible pgoff_t and since the type is 3565a720394SLukas Czerner * unsigned we're using -1. 3575a720394SLukas Czerner */ 3585a720394SLukas Czerner end = -1; 3595a720394SLukas Czerner else 36009cbfeafSKirill A. Shutemov end = (lend + 1) >> PAGE_SHIFT; 361d7339071SHans Reiser 36251dcbdacSMatthew Wilcox (Oracle) folio_batch_init(&fbatch); 363b85e0effSHugh Dickins index = start; 3643392ca12SVishal Moola (Oracle) while (index < end && find_lock_entries(mapping, &index, end - 1, 36551dcbdacSMatthew Wilcox (Oracle) &fbatch, indices)) { 36651dcbdacSMatthew Wilcox (Oracle) truncate_folio_batch_exceptionals(mapping, &fbatch, indices); 36751dcbdacSMatthew Wilcox (Oracle) for (i = 0; i < folio_batch_count(&fbatch); i++) 36851dcbdacSMatthew Wilcox (Oracle) truncate_cleanup_folio(fbatch.folios[i]); 36951dcbdacSMatthew Wilcox (Oracle) delete_from_page_cache_batch(mapping, &fbatch); 37051dcbdacSMatthew Wilcox (Oracle) for (i = 0; i < folio_batch_count(&fbatch); i++) 37151dcbdacSMatthew Wilcox (Oracle) folio_unlock(fbatch.folios[i]); 37251dcbdacSMatthew Wilcox (Oracle) folio_batch_release(&fbatch); 3731da177e4SLinus Torvalds cond_resched(); 3741da177e4SLinus Torvalds } 3755c211ba2SMatthew Wilcox (Oracle) 376b9a8a419SMatthew Wilcox (Oracle) same_folio = (lstart >> PAGE_SHIFT) == (lend >> PAGE_SHIFT); 377b9a8a419SMatthew Wilcox (Oracle) folio = __filemap_get_folio(mapping, lstart >> PAGE_SHIFT, FGP_LOCK, 0); 37866dabbb6SChristoph Hellwig if (!IS_ERR(folio)) { 379b9a8a419SMatthew Wilcox (Oracle) same_folio = lend < folio_pos(folio) + folio_size(folio); 380b9a8a419SMatthew Wilcox (Oracle) if (!truncate_inode_partial_folio(folio, lstart, lend)) { 381*87b11f86SSidhartha Kumar start = folio_next_index(folio); 382b9a8a419SMatthew Wilcox (Oracle) if (same_folio) 383b9a8a419SMatthew Wilcox (Oracle) end = folio->index; 3845a720394SLukas Czerner } 385b9a8a419SMatthew Wilcox (Oracle) folio_unlock(folio); 386b9a8a419SMatthew Wilcox (Oracle) folio_put(folio); 387b9a8a419SMatthew Wilcox (Oracle) folio = NULL; 3881da177e4SLinus Torvalds } 389b9a8a419SMatthew Wilcox (Oracle) 39066dabbb6SChristoph Hellwig if (!same_folio) { 391b9a8a419SMatthew Wilcox (Oracle) folio = __filemap_get_folio(mapping, lend >> PAGE_SHIFT, 392b9a8a419SMatthew Wilcox (Oracle) FGP_LOCK, 0); 39366dabbb6SChristoph Hellwig if (!IS_ERR(folio)) { 394b9a8a419SMatthew Wilcox (Oracle) if (!truncate_inode_partial_folio(folio, lstart, lend)) 395b9a8a419SMatthew Wilcox (Oracle) end = folio->index; 396b9a8a419SMatthew Wilcox (Oracle) folio_unlock(folio); 397b9a8a419SMatthew Wilcox (Oracle) folio_put(folio); 3981da177e4SLinus Torvalds } 39966dabbb6SChristoph Hellwig } 4001da177e4SLinus Torvalds 401b85e0effSHugh Dickins index = start; 402b9a8a419SMatthew Wilcox (Oracle) while (index < end) { 4031da177e4SLinus Torvalds cond_resched(); 4049fb6beeaSVishal Moola (Oracle) if (!find_get_entries(mapping, &index, end - 1, &fbatch, 40538cefeb3SMatthew Wilcox (Oracle) indices)) { 406792ceaefSHugh Dickins /* If all gone from start onwards, we're done */ 407b85e0effSHugh Dickins if (index == start) 4081da177e4SLinus Torvalds break; 409792ceaefSHugh Dickins /* Otherwise restart to make sure all gone */ 410b85e0effSHugh Dickins index = start; 4111da177e4SLinus Torvalds continue; 4121da177e4SLinus Torvalds } 413f2187599SMel Gorman 4140e499ed3SMatthew Wilcox (Oracle) for (i = 0; i < folio_batch_count(&fbatch); i++) { 4150e499ed3SMatthew Wilcox (Oracle) struct folio *folio = fbatch.folios[i]; 4161da177e4SLinus Torvalds 417b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 418b85e0effSHugh Dickins 4190e499ed3SMatthew Wilcox (Oracle) if (xa_is_value(folio)) 4200cd6144aSJohannes Weiner continue; 4210cd6144aSJohannes Weiner 4221e84a3d9SMatthew Wilcox (Oracle) folio_lock(folio); 4239fb6beeaSVishal Moola (Oracle) VM_BUG_ON_FOLIO(!folio_contains(folio, indices[i]), folio); 4241e84a3d9SMatthew Wilcox (Oracle) folio_wait_writeback(folio); 4251e84a3d9SMatthew Wilcox (Oracle) truncate_inode_folio(mapping, folio); 4261e84a3d9SMatthew Wilcox (Oracle) folio_unlock(folio); 4271da177e4SLinus Torvalds } 4280e499ed3SMatthew Wilcox (Oracle) truncate_folio_batch_exceptionals(mapping, &fbatch, indices); 4290e499ed3SMatthew Wilcox (Oracle) folio_batch_release(&fbatch); 4301da177e4SLinus Torvalds } 4311da177e4SLinus Torvalds } 432d7339071SHans Reiser EXPORT_SYMBOL(truncate_inode_pages_range); 4331da177e4SLinus Torvalds 434d7339071SHans Reiser /** 435d7339071SHans Reiser * truncate_inode_pages - truncate *all* the pages from an offset 436d7339071SHans Reiser * @mapping: mapping to truncate 437d7339071SHans Reiser * @lstart: offset from which to truncate 438d7339071SHans Reiser * 439730633f0SJan Kara * Called under (and serialised by) inode->i_rwsem and 440730633f0SJan Kara * mapping->invalidate_lock. 44108142579SJan Kara * 44208142579SJan Kara * Note: When this function returns, there can be a page in the process of 4436ffcd825SMatthew Wilcox (Oracle) * deletion (inside __filemap_remove_folio()) in the specified range. Thus 44408142579SJan Kara * mapping->nrpages can be non-zero when this function returns even after 44508142579SJan Kara * truncation of the whole mapping. 446d7339071SHans Reiser */ 447d7339071SHans Reiser void truncate_inode_pages(struct address_space *mapping, loff_t lstart) 448d7339071SHans Reiser { 449d7339071SHans Reiser truncate_inode_pages_range(mapping, lstart, (loff_t)-1); 450d7339071SHans Reiser } 4511da177e4SLinus Torvalds EXPORT_SYMBOL(truncate_inode_pages); 4521da177e4SLinus Torvalds 45328697355SMike Waychison /** 45491b0abe3SJohannes Weiner * truncate_inode_pages_final - truncate *all* pages before inode dies 45591b0abe3SJohannes Weiner * @mapping: mapping to truncate 45691b0abe3SJohannes Weiner * 4579608703eSJan Kara * Called under (and serialized by) inode->i_rwsem. 45891b0abe3SJohannes Weiner * 45991b0abe3SJohannes Weiner * Filesystems have to use this in the .evict_inode path to inform the 46091b0abe3SJohannes Weiner * VM that this is the final truncate and the inode is going away. 46191b0abe3SJohannes Weiner */ 46291b0abe3SJohannes Weiner void truncate_inode_pages_final(struct address_space *mapping) 46391b0abe3SJohannes Weiner { 46491b0abe3SJohannes Weiner /* 46591b0abe3SJohannes Weiner * Page reclaim can not participate in regular inode lifetime 46691b0abe3SJohannes Weiner * management (can't call iput()) and thus can race with the 46791b0abe3SJohannes Weiner * inode teardown. Tell it when the address space is exiting, 46891b0abe3SJohannes Weiner * so that it does not install eviction information after the 46991b0abe3SJohannes Weiner * final truncate has begun. 47091b0abe3SJohannes Weiner */ 47191b0abe3SJohannes Weiner mapping_set_exiting(mapping); 47291b0abe3SJohannes Weiner 4737716506aSMatthew Wilcox (Oracle) if (!mapping_empty(mapping)) { 47491b0abe3SJohannes Weiner /* 47591b0abe3SJohannes Weiner * As truncation uses a lockless tree lookup, cycle 47691b0abe3SJohannes Weiner * the tree lock to make sure any ongoing tree 47791b0abe3SJohannes Weiner * modification that does not see AS_EXITING is 47891b0abe3SJohannes Weiner * completed before starting the final truncate. 47991b0abe3SJohannes Weiner */ 480b93b0163SMatthew Wilcox xa_lock_irq(&mapping->i_pages); 481b93b0163SMatthew Wilcox xa_unlock_irq(&mapping->i_pages); 48291b0abe3SJohannes Weiner } 4836ff38bd4SPavel Tikhomirov 4846ff38bd4SPavel Tikhomirov truncate_inode_pages(mapping, 0); 48591b0abe3SJohannes Weiner } 48691b0abe3SJohannes Weiner EXPORT_SYMBOL(truncate_inode_pages_final); 48791b0abe3SJohannes Weiner 488c56109ddSMatthew Wilcox (Oracle) /** 4891a0fc811SMatthew Wilcox (Oracle) * mapping_try_invalidate - Invalidate all the evictable folios of one inode 4901a0fc811SMatthew Wilcox (Oracle) * @mapping: the address_space which holds the folios to invalidate 491c56109ddSMatthew Wilcox (Oracle) * @start: the offset 'from' which to invalidate 492c56109ddSMatthew Wilcox (Oracle) * @end: the offset 'to' which to invalidate (inclusive) 4931a0fc811SMatthew Wilcox (Oracle) * @nr_failed: How many folio invalidations failed 494c56109ddSMatthew Wilcox (Oracle) * 4951a0fc811SMatthew Wilcox (Oracle) * This function is similar to invalidate_mapping_pages(), except that it 4961a0fc811SMatthew Wilcox (Oracle) * returns the number of folios which could not be evicted in @nr_failed. 497c56109ddSMatthew Wilcox (Oracle) */ 4981a0fc811SMatthew Wilcox (Oracle) unsigned long mapping_try_invalidate(struct address_space *mapping, 4991a0fc811SMatthew Wilcox (Oracle) pgoff_t start, pgoff_t end, unsigned long *nr_failed) 5001da177e4SLinus Torvalds { 5010cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 50251dcbdacSMatthew Wilcox (Oracle) struct folio_batch fbatch; 503b85e0effSHugh Dickins pgoff_t index = start; 50431560180SMinchan Kim unsigned long ret; 50531560180SMinchan Kim unsigned long count = 0; 5061da177e4SLinus Torvalds int i; 5071da177e4SLinus Torvalds 50851dcbdacSMatthew Wilcox (Oracle) folio_batch_init(&fbatch); 5093392ca12SVishal Moola (Oracle) while (find_lock_entries(mapping, &index, end, &fbatch, indices)) { 51051dcbdacSMatthew Wilcox (Oracle) for (i = 0; i < folio_batch_count(&fbatch); i++) { 511b4545f46SMatthew Wilcox (Oracle) struct folio *folio = fbatch.folios[i]; 5121da177e4SLinus Torvalds 513b4545f46SMatthew Wilcox (Oracle) /* We rely upon deletion not changing folio->index */ 514e0f23603SNeilBrown 515b4545f46SMatthew Wilcox (Oracle) if (xa_is_value(folio)) { 5167ae12c80SJohannes Weiner count += invalidate_exceptional_entry(mapping, 5173392ca12SVishal Moola (Oracle) indices[i], folio); 5180cd6144aSJohannes Weiner continue; 5190cd6144aSJohannes Weiner } 520fc127da0SKirill A. Shutemov 521b4545f46SMatthew Wilcox (Oracle) ret = mapping_evict_folio(mapping, folio); 522b4545f46SMatthew Wilcox (Oracle) folio_unlock(folio); 52331560180SMinchan Kim /* 524b4545f46SMatthew Wilcox (Oracle) * Invalidation is a hint that the folio is no longer 52531560180SMinchan Kim * of interest and try to speed up its reclaim. 52631560180SMinchan Kim */ 527eb1d7a65SYafang Shao if (!ret) { 528261b6840SMatthew Wilcox (Oracle) deactivate_file_folio(folio); 5291a0fc811SMatthew Wilcox (Oracle) /* Likely in the lru cache of a remote CPU */ 5301a0fc811SMatthew Wilcox (Oracle) if (nr_failed) 5311a0fc811SMatthew Wilcox (Oracle) (*nr_failed)++; 532eb1d7a65SYafang Shao } 53331560180SMinchan Kim count += ret; 5341da177e4SLinus Torvalds } 53551dcbdacSMatthew Wilcox (Oracle) folio_batch_remove_exceptionals(&fbatch); 53651dcbdacSMatthew Wilcox (Oracle) folio_batch_release(&fbatch); 537fc9a07e7SAndrew Morton cond_resched(); 5381da177e4SLinus Torvalds } 53931560180SMinchan Kim return count; 5401da177e4SLinus Torvalds } 541eb1d7a65SYafang Shao 542eb1d7a65SYafang Shao /** 5437ae12c80SJohannes Weiner * invalidate_mapping_pages - Invalidate all clean, unlocked cache of one inode 5447ae12c80SJohannes Weiner * @mapping: the address_space which holds the cache to invalidate 545eb1d7a65SYafang Shao * @start: the offset 'from' which to invalidate 546eb1d7a65SYafang Shao * @end: the offset 'to' which to invalidate (inclusive) 547eb1d7a65SYafang Shao * 5487ae12c80SJohannes Weiner * This function removes pages that are clean, unmapped and unlocked, 5497ae12c80SJohannes Weiner * as well as shadow entries. It will not block on IO activity. 550eb1d7a65SYafang Shao * 5517ae12c80SJohannes Weiner * If you want to remove all the pages of one inode, regardless of 5527ae12c80SJohannes Weiner * their use and writeback state, use truncate_inode_pages(). 553eb1d7a65SYafang Shao * 5541a0fc811SMatthew Wilcox (Oracle) * Return: The number of indices that had their contents invalidated 555eb1d7a65SYafang Shao */ 556eb1d7a65SYafang Shao unsigned long invalidate_mapping_pages(struct address_space *mapping, 557eb1d7a65SYafang Shao pgoff_t start, pgoff_t end) 558eb1d7a65SYafang Shao { 5591a0fc811SMatthew Wilcox (Oracle) return mapping_try_invalidate(mapping, start, end, NULL); 560eb1d7a65SYafang Shao } 56154bc4855SAnton Altaparmakov EXPORT_SYMBOL(invalidate_mapping_pages); 5621da177e4SLinus Torvalds 563bd4c8ce4SAndrew Morton /* 5641b8ddbeeSMatthew Wilcox (Oracle) * This is like invalidate_inode_page(), except it ignores the page's 565bd4c8ce4SAndrew Morton * refcount. We do this because invalidate_inode_pages2() needs stronger 566bd4c8ce4SAndrew Morton * invalidation guarantees, and cannot afford to leave pages behind because 5672706a1b8SAnderson Briglia * shrink_page_list() has a temp ref on them, or because they're transiently 5681fec6890SMatthew Wilcox (Oracle) * sitting in the folio_add_lru() caches. 569bd4c8ce4SAndrew Morton */ 57078f42660SMatthew Wilcox (Oracle) static int invalidate_complete_folio2(struct address_space *mapping, 57178f42660SMatthew Wilcox (Oracle) struct folio *folio) 572bd4c8ce4SAndrew Morton { 57378f42660SMatthew Wilcox (Oracle) if (folio->mapping != mapping) 574bd4c8ce4SAndrew Morton return 0; 575bd4c8ce4SAndrew Morton 57678f42660SMatthew Wilcox (Oracle) if (folio_has_private(folio) && 57778f42660SMatthew Wilcox (Oracle) !filemap_release_folio(folio, GFP_KERNEL)) 578bd4c8ce4SAndrew Morton return 0; 579bd4c8ce4SAndrew Morton 58051b8c1feSJohannes Weiner spin_lock(&mapping->host->i_lock); 58130472509SJohannes Weiner xa_lock_irq(&mapping->i_pages); 58278f42660SMatthew Wilcox (Oracle) if (folio_test_dirty(folio)) 583bd4c8ce4SAndrew Morton goto failed; 584bd4c8ce4SAndrew Morton 58578f42660SMatthew Wilcox (Oracle) BUG_ON(folio_has_private(folio)); 58678f42660SMatthew Wilcox (Oracle) __filemap_remove_folio(folio, NULL); 58730472509SJohannes Weiner xa_unlock_irq(&mapping->i_pages); 58851b8c1feSJohannes Weiner if (mapping_shrinkable(mapping)) 58951b8c1feSJohannes Weiner inode_add_lru(mapping->host); 59051b8c1feSJohannes Weiner spin_unlock(&mapping->host->i_lock); 5916072d13cSLinus Torvalds 59278f42660SMatthew Wilcox (Oracle) filemap_free_folio(mapping, folio); 593bd4c8ce4SAndrew Morton return 1; 594bd4c8ce4SAndrew Morton failed: 59530472509SJohannes Weiner xa_unlock_irq(&mapping->i_pages); 59651b8c1feSJohannes Weiner spin_unlock(&mapping->host->i_lock); 597bd4c8ce4SAndrew Morton return 0; 598bd4c8ce4SAndrew Morton } 599bd4c8ce4SAndrew Morton 600affa80e8SMatthew Wilcox (Oracle) static int folio_launder(struct address_space *mapping, struct folio *folio) 601e3db7691STrond Myklebust { 602f6357c3aSMatthew Wilcox (Oracle) if (!folio_test_dirty(folio)) 603e3db7691STrond Myklebust return 0; 604affa80e8SMatthew Wilcox (Oracle) if (folio->mapping != mapping || mapping->a_ops->launder_folio == NULL) 605e3db7691STrond Myklebust return 0; 606affa80e8SMatthew Wilcox (Oracle) return mapping->a_ops->launder_folio(folio); 607e3db7691STrond Myklebust } 608e3db7691STrond Myklebust 6091da177e4SLinus Torvalds /** 6101da177e4SLinus Torvalds * invalidate_inode_pages2_range - remove range of pages from an address_space 61167be2dd1SMartin Waitz * @mapping: the address_space 6121da177e4SLinus Torvalds * @start: the page offset 'from' which to invalidate 6131da177e4SLinus Torvalds * @end: the page offset 'to' which to invalidate (inclusive) 6141da177e4SLinus Torvalds * 6151da177e4SLinus Torvalds * Any pages which are found to be mapped into pagetables are unmapped prior to 6161da177e4SLinus Torvalds * invalidation. 6171da177e4SLinus Torvalds * 618a862f68aSMike Rapoport * Return: -EBUSY if any pages could not be invalidated. 6191da177e4SLinus Torvalds */ 6201da177e4SLinus Torvalds int invalidate_inode_pages2_range(struct address_space *mapping, 6211da177e4SLinus Torvalds pgoff_t start, pgoff_t end) 6221da177e4SLinus Torvalds { 6230cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 6240e499ed3SMatthew Wilcox (Oracle) struct folio_batch fbatch; 625b85e0effSHugh Dickins pgoff_t index; 6261da177e4SLinus Torvalds int i; 6271da177e4SLinus Torvalds int ret = 0; 6280dd1334fSHisashi Hifumi int ret2 = 0; 6291da177e4SLinus Torvalds int did_range_unmap = 0; 6301da177e4SLinus Torvalds 6317716506aSMatthew Wilcox (Oracle) if (mapping_empty(mapping)) 6320a4ee518SChristoph Hellwig return 0; 63332691f0fSAndrey Ryabinin 6340e499ed3SMatthew Wilcox (Oracle) folio_batch_init(&fbatch); 635b85e0effSHugh Dickins index = start; 6369fb6beeaSVishal Moola (Oracle) while (find_get_entries(mapping, &index, end, &fbatch, indices)) { 6370e499ed3SMatthew Wilcox (Oracle) for (i = 0; i < folio_batch_count(&fbatch); i++) { 6380e499ed3SMatthew Wilcox (Oracle) struct folio *folio = fbatch.folios[i]; 639b85e0effSHugh Dickins 640fae9bc4aSMatthew Wilcox (Oracle) /* We rely upon deletion not changing folio->index */ 6411da177e4SLinus Torvalds 6420e499ed3SMatthew Wilcox (Oracle) if (xa_is_value(folio)) { 643c6dcf52cSJan Kara if (!invalidate_exceptional_entry2(mapping, 6449fb6beeaSVishal Moola (Oracle) indices[i], folio)) 645c6dcf52cSJan Kara ret = -EBUSY; 6460cd6144aSJohannes Weiner continue; 6470cd6144aSJohannes Weiner } 6480cd6144aSJohannes Weiner 649fae9bc4aSMatthew Wilcox (Oracle) if (!did_range_unmap && folio_mapped(folio)) { 65022061a1fSHugh Dickins /* 651fae9bc4aSMatthew Wilcox (Oracle) * If folio is mapped, before taking its lock, 65222061a1fSHugh Dickins * zap the rest of the file in one hit. 65322061a1fSHugh Dickins */ 6549fb6beeaSVishal Moola (Oracle) unmap_mapping_pages(mapping, indices[i], 6559fb6beeaSVishal Moola (Oracle) (1 + end - indices[i]), false); 65622061a1fSHugh Dickins did_range_unmap = 1; 65722061a1fSHugh Dickins } 65822061a1fSHugh Dickins 659fae9bc4aSMatthew Wilcox (Oracle) folio_lock(folio); 6609fb6beeaSVishal Moola (Oracle) VM_BUG_ON_FOLIO(!folio_contains(folio, indices[i]), folio); 661fae9bc4aSMatthew Wilcox (Oracle) if (folio->mapping != mapping) { 662fae9bc4aSMatthew Wilcox (Oracle) folio_unlock(folio); 6631da177e4SLinus Torvalds continue; 6641da177e4SLinus Torvalds } 665fae9bc4aSMatthew Wilcox (Oracle) folio_wait_writeback(folio); 66622061a1fSHugh Dickins 667fae9bc4aSMatthew Wilcox (Oracle) if (folio_mapped(folio)) 668fae9bc4aSMatthew Wilcox (Oracle) unmap_mapping_folio(folio); 669fae9bc4aSMatthew Wilcox (Oracle) BUG_ON(folio_mapped(folio)); 67022061a1fSHugh Dickins 671affa80e8SMatthew Wilcox (Oracle) ret2 = folio_launder(mapping, folio); 6720dd1334fSHisashi Hifumi if (ret2 == 0) { 67378f42660SMatthew Wilcox (Oracle) if (!invalidate_complete_folio2(mapping, folio)) 6746ccfa806SHisashi Hifumi ret2 = -EBUSY; 6750dd1334fSHisashi Hifumi } 6760dd1334fSHisashi Hifumi if (ret2 < 0) 6770dd1334fSHisashi Hifumi ret = ret2; 678fae9bc4aSMatthew Wilcox (Oracle) folio_unlock(folio); 6791da177e4SLinus Torvalds } 6800e499ed3SMatthew Wilcox (Oracle) folio_batch_remove_exceptionals(&fbatch); 6810e499ed3SMatthew Wilcox (Oracle) folio_batch_release(&fbatch); 6821da177e4SLinus Torvalds cond_resched(); 6831da177e4SLinus Torvalds } 684cd656375SJan Kara /* 68569b6c131SMatthew Wilcox * For DAX we invalidate page tables after invalidating page cache. We 686cd656375SJan Kara * could invalidate page tables while invalidating each entry however 687cd656375SJan Kara * that would be expensive. And doing range unmapping before doesn't 68869b6c131SMatthew Wilcox * work as we have no cheap way to find whether page cache entry didn't 689cd656375SJan Kara * get remapped later. 690cd656375SJan Kara */ 691cd656375SJan Kara if (dax_mapping(mapping)) { 692977fbdcdSMatthew Wilcox unmap_mapping_pages(mapping, start, end - start + 1, false); 693cd656375SJan Kara } 6941da177e4SLinus Torvalds return ret; 6951da177e4SLinus Torvalds } 6961da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); 6971da177e4SLinus Torvalds 6981da177e4SLinus Torvalds /** 6991da177e4SLinus Torvalds * invalidate_inode_pages2 - remove all pages from an address_space 70067be2dd1SMartin Waitz * @mapping: the address_space 7011da177e4SLinus Torvalds * 7021da177e4SLinus Torvalds * Any pages which are found to be mapped into pagetables are unmapped prior to 7031da177e4SLinus Torvalds * invalidation. 7041da177e4SLinus Torvalds * 705a862f68aSMike Rapoport * Return: -EBUSY if any pages could not be invalidated. 7061da177e4SLinus Torvalds */ 7071da177e4SLinus Torvalds int invalidate_inode_pages2(struct address_space *mapping) 7081da177e4SLinus Torvalds { 7091da177e4SLinus Torvalds return invalidate_inode_pages2_range(mapping, 0, -1); 7101da177e4SLinus Torvalds } 7111da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(invalidate_inode_pages2); 71225d9e2d1Snpiggin@suse.de 71325d9e2d1Snpiggin@suse.de /** 71425d9e2d1Snpiggin@suse.de * truncate_pagecache - unmap and remove pagecache that has been truncated 71525d9e2d1Snpiggin@suse.de * @inode: inode 7168a549beaSHugh Dickins * @newsize: new file size 71725d9e2d1Snpiggin@suse.de * 71825d9e2d1Snpiggin@suse.de * inode's new i_size must already be written before truncate_pagecache 71925d9e2d1Snpiggin@suse.de * is called. 72025d9e2d1Snpiggin@suse.de * 72125d9e2d1Snpiggin@suse.de * This function should typically be called before the filesystem 72225d9e2d1Snpiggin@suse.de * releases resources associated with the freed range (eg. deallocates 72325d9e2d1Snpiggin@suse.de * blocks). This way, pagecache will always stay logically coherent 72425d9e2d1Snpiggin@suse.de * with on-disk format, and the filesystem would not have to deal with 72525d9e2d1Snpiggin@suse.de * situations such as writepage being called for a page that has already 72625d9e2d1Snpiggin@suse.de * had its underlying blocks deallocated. 72725d9e2d1Snpiggin@suse.de */ 7287caef267SKirill A. Shutemov void truncate_pagecache(struct inode *inode, loff_t newsize) 72925d9e2d1Snpiggin@suse.de { 73025d9e2d1Snpiggin@suse.de struct address_space *mapping = inode->i_mapping; 7318a549beaSHugh Dickins loff_t holebegin = round_up(newsize, PAGE_SIZE); 73225d9e2d1Snpiggin@suse.de 73325d9e2d1Snpiggin@suse.de /* 73425d9e2d1Snpiggin@suse.de * unmap_mapping_range is called twice, first simply for 73525d9e2d1Snpiggin@suse.de * efficiency so that truncate_inode_pages does fewer 73625d9e2d1Snpiggin@suse.de * single-page unmaps. However after this first call, and 73725d9e2d1Snpiggin@suse.de * before truncate_inode_pages finishes, it is possible for 73825d9e2d1Snpiggin@suse.de * private pages to be COWed, which remain after 73925d9e2d1Snpiggin@suse.de * truncate_inode_pages finishes, hence the second 74025d9e2d1Snpiggin@suse.de * unmap_mapping_range call must be made for correctness. 74125d9e2d1Snpiggin@suse.de */ 7428a549beaSHugh Dickins unmap_mapping_range(mapping, holebegin, 0, 1); 7438a549beaSHugh Dickins truncate_inode_pages(mapping, newsize); 7448a549beaSHugh Dickins unmap_mapping_range(mapping, holebegin, 0, 1); 74525d9e2d1Snpiggin@suse.de } 74625d9e2d1Snpiggin@suse.de EXPORT_SYMBOL(truncate_pagecache); 74725d9e2d1Snpiggin@suse.de 74825d9e2d1Snpiggin@suse.de /** 7492c27c65eSChristoph Hellwig * truncate_setsize - update inode and pagecache for a new file size 7502c27c65eSChristoph Hellwig * @inode: inode 7512c27c65eSChristoph Hellwig * @newsize: new file size 7522c27c65eSChristoph Hellwig * 753382e27daSJan Kara * truncate_setsize updates i_size and performs pagecache truncation (if 754382e27daSJan Kara * necessary) to @newsize. It will be typically be called from the filesystem's 755382e27daSJan Kara * setattr function when ATTR_SIZE is passed in. 7562c27c65eSChristoph Hellwig * 75777783d06SJan Kara * Must be called with a lock serializing truncates and writes (generally 7589608703eSJan Kara * i_rwsem but e.g. xfs uses a different lock) and before all filesystem 75977783d06SJan Kara * specific block truncation has been performed. 7602c27c65eSChristoph Hellwig */ 7612c27c65eSChristoph Hellwig void truncate_setsize(struct inode *inode, loff_t newsize) 7622c27c65eSChristoph Hellwig { 76390a80202SJan Kara loff_t oldsize = inode->i_size; 76490a80202SJan Kara 7652c27c65eSChristoph Hellwig i_size_write(inode, newsize); 76690a80202SJan Kara if (newsize > oldsize) 76790a80202SJan Kara pagecache_isize_extended(inode, oldsize, newsize); 7687caef267SKirill A. Shutemov truncate_pagecache(inode, newsize); 7692c27c65eSChristoph Hellwig } 7702c27c65eSChristoph Hellwig EXPORT_SYMBOL(truncate_setsize); 7712c27c65eSChristoph Hellwig 7722c27c65eSChristoph Hellwig /** 77390a80202SJan Kara * pagecache_isize_extended - update pagecache after extension of i_size 77490a80202SJan Kara * @inode: inode for which i_size was extended 77590a80202SJan Kara * @from: original inode size 77690a80202SJan Kara * @to: new inode size 77790a80202SJan Kara * 77890a80202SJan Kara * Handle extension of inode size either caused by extending truncate or by 77990a80202SJan Kara * write starting after current i_size. We mark the page straddling current 78090a80202SJan Kara * i_size RO so that page_mkwrite() is called on the nearest write access to 78190a80202SJan Kara * the page. This way filesystem can be sure that page_mkwrite() is called on 78290a80202SJan Kara * the page before user writes to the page via mmap after the i_size has been 78390a80202SJan Kara * changed. 78490a80202SJan Kara * 78590a80202SJan Kara * The function must be called after i_size is updated so that page fault 78690a80202SJan Kara * coming after we unlock the page will already see the new i_size. 7879608703eSJan Kara * The function must be called while we still hold i_rwsem - this not only 78890a80202SJan Kara * makes sure i_size is stable but also that userspace cannot observe new 78990a80202SJan Kara * i_size value before we are prepared to store mmap writes at new inode size. 79090a80202SJan Kara */ 79190a80202SJan Kara void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) 79290a80202SJan Kara { 79393407472SFabian Frederick int bsize = i_blocksize(inode); 79490a80202SJan Kara loff_t rounded_from; 79590a80202SJan Kara struct page *page; 79690a80202SJan Kara pgoff_t index; 79790a80202SJan Kara 79890a80202SJan Kara WARN_ON(to > inode->i_size); 79990a80202SJan Kara 80009cbfeafSKirill A. Shutemov if (from >= to || bsize == PAGE_SIZE) 80190a80202SJan Kara return; 80290a80202SJan Kara /* Page straddling @from will not have any hole block created? */ 80390a80202SJan Kara rounded_from = round_up(from, bsize); 80409cbfeafSKirill A. Shutemov if (to <= rounded_from || !(rounded_from & (PAGE_SIZE - 1))) 80590a80202SJan Kara return; 80690a80202SJan Kara 80709cbfeafSKirill A. Shutemov index = from >> PAGE_SHIFT; 80890a80202SJan Kara page = find_lock_page(inode->i_mapping, index); 80990a80202SJan Kara /* Page not cached? Nothing to do */ 81090a80202SJan Kara if (!page) 81190a80202SJan Kara return; 81290a80202SJan Kara /* 81390a80202SJan Kara * See clear_page_dirty_for_io() for details why set_page_dirty() 81490a80202SJan Kara * is needed. 81590a80202SJan Kara */ 81690a80202SJan Kara if (page_mkclean(page)) 81790a80202SJan Kara set_page_dirty(page); 81890a80202SJan Kara unlock_page(page); 81909cbfeafSKirill A. Shutemov put_page(page); 82090a80202SJan Kara } 82190a80202SJan Kara EXPORT_SYMBOL(pagecache_isize_extended); 82290a80202SJan Kara 82390a80202SJan Kara /** 824623e3db9SHugh Dickins * truncate_pagecache_range - unmap and remove pagecache that is hole-punched 825623e3db9SHugh Dickins * @inode: inode 826623e3db9SHugh Dickins * @lstart: offset of beginning of hole 827623e3db9SHugh Dickins * @lend: offset of last byte of hole 828623e3db9SHugh Dickins * 829623e3db9SHugh Dickins * This function should typically be called before the filesystem 830623e3db9SHugh Dickins * releases resources associated with the freed range (eg. deallocates 831623e3db9SHugh Dickins * blocks). This way, pagecache will always stay logically coherent 832623e3db9SHugh Dickins * with on-disk format, and the filesystem would not have to deal with 833623e3db9SHugh Dickins * situations such as writepage being called for a page that has already 834623e3db9SHugh Dickins * had its underlying blocks deallocated. 835623e3db9SHugh Dickins */ 836623e3db9SHugh Dickins void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) 837623e3db9SHugh Dickins { 838623e3db9SHugh Dickins struct address_space *mapping = inode->i_mapping; 839623e3db9SHugh Dickins loff_t unmap_start = round_up(lstart, PAGE_SIZE); 840623e3db9SHugh Dickins loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; 841623e3db9SHugh Dickins /* 842623e3db9SHugh Dickins * This rounding is currently just for example: unmap_mapping_range 843623e3db9SHugh Dickins * expands its hole outwards, whereas we want it to contract the hole 844623e3db9SHugh Dickins * inwards. However, existing callers of truncate_pagecache_range are 8455a720394SLukas Czerner * doing their own page rounding first. Note that unmap_mapping_range 8465a720394SLukas Czerner * allows holelen 0 for all, and we allow lend -1 for end of file. 847623e3db9SHugh Dickins */ 848623e3db9SHugh Dickins 849623e3db9SHugh Dickins /* 850623e3db9SHugh Dickins * Unlike in truncate_pagecache, unmap_mapping_range is called only 851623e3db9SHugh Dickins * once (before truncating pagecache), and without "even_cows" flag: 852623e3db9SHugh Dickins * hole-punching should not remove private COWed pages from the hole. 853623e3db9SHugh Dickins */ 854623e3db9SHugh Dickins if ((u64)unmap_end > (u64)unmap_start) 855623e3db9SHugh Dickins unmap_mapping_range(mapping, unmap_start, 856623e3db9SHugh Dickins 1 + unmap_end - unmap_start, 0); 857623e3db9SHugh Dickins truncate_inode_pages_range(mapping, lstart, lend); 858623e3db9SHugh Dickins } 859623e3db9SHugh Dickins EXPORT_SYMBOL(truncate_pagecache_range); 860