11da177e4SLinus Torvalds /* 21da177e4SLinus Torvalds * mm/truncate.c - code for taking down pages from address_spaces 31da177e4SLinus Torvalds * 41da177e4SLinus Torvalds * Copyright (C) 2002, Linus Torvalds 51da177e4SLinus Torvalds * 6e1f8e874SFrancois Cami * 10Sep2002 Andrew Morton 71da177e4SLinus Torvalds * Initial version. 81da177e4SLinus Torvalds */ 91da177e4SLinus Torvalds 101da177e4SLinus Torvalds #include <linux/kernel.h> 114af3c9ccSAlexey Dobriyan #include <linux/backing-dev.h> 125a0e3ad6STejun Heo #include <linux/gfp.h> 131da177e4SLinus Torvalds #include <linux/mm.h> 140fd0e6b0SNick Piggin #include <linux/swap.h> 15b95f1b31SPaul Gortmaker #include <linux/export.h> 161da177e4SLinus Torvalds #include <linux/pagemap.h> 1701f2705dSNate Diller #include <linux/highmem.h> 181da177e4SLinus Torvalds #include <linux/pagevec.h> 19e08748ceSAndrew Morton #include <linux/task_io_accounting_ops.h> 201da177e4SLinus Torvalds #include <linux/buffer_head.h> /* grr. try_to_release_page, 21aaa4059bSJan Kara do_invalidatepage */ 22c515e1fdSDan Magenheimer #include <linux/cleancache.h> 23*90a80202SJan Kara #include <linux/rmap.h> 24ba470de4SRik van Riel #include "internal.h" 251da177e4SLinus Torvalds 260cd6144aSJohannes Weiner static void clear_exceptional_entry(struct address_space *mapping, 270cd6144aSJohannes Weiner pgoff_t index, void *entry) 280cd6144aSJohannes Weiner { 29449dd698SJohannes Weiner struct radix_tree_node *node; 30449dd698SJohannes Weiner void **slot; 31449dd698SJohannes Weiner 320cd6144aSJohannes Weiner /* Handled by shmem itself */ 330cd6144aSJohannes Weiner if (shmem_mapping(mapping)) 340cd6144aSJohannes Weiner return; 350cd6144aSJohannes Weiner 360cd6144aSJohannes Weiner spin_lock_irq(&mapping->tree_lock); 370cd6144aSJohannes Weiner /* 380cd6144aSJohannes Weiner * Regular page slots are stabilized by the page lock even 390cd6144aSJohannes Weiner * without the tree itself locked. These unlocked entries 400cd6144aSJohannes Weiner * need verification under the tree lock. 410cd6144aSJohannes Weiner */ 42449dd698SJohannes Weiner if (!__radix_tree_lookup(&mapping->page_tree, index, &node, &slot)) 43449dd698SJohannes Weiner goto unlock; 44449dd698SJohannes Weiner if (*slot != entry) 45449dd698SJohannes Weiner goto unlock; 46449dd698SJohannes Weiner radix_tree_replace_slot(slot, NULL); 4791b0abe3SJohannes Weiner mapping->nrshadows--; 48449dd698SJohannes Weiner if (!node) 49449dd698SJohannes Weiner goto unlock; 50449dd698SJohannes Weiner workingset_node_shadows_dec(node); 51449dd698SJohannes Weiner /* 52449dd698SJohannes Weiner * Don't track node without shadow entries. 53449dd698SJohannes Weiner * 54449dd698SJohannes Weiner * Avoid acquiring the list_lru lock if already untracked. 55449dd698SJohannes Weiner * The list_empty() test is safe as node->private_list is 56449dd698SJohannes Weiner * protected by mapping->tree_lock. 57449dd698SJohannes Weiner */ 58449dd698SJohannes Weiner if (!workingset_node_shadows(node) && 59449dd698SJohannes Weiner !list_empty(&node->private_list)) 60449dd698SJohannes Weiner list_lru_del(&workingset_shadow_nodes, &node->private_list); 61449dd698SJohannes Weiner __radix_tree_delete_node(&mapping->page_tree, node); 62449dd698SJohannes Weiner unlock: 630cd6144aSJohannes Weiner spin_unlock_irq(&mapping->tree_lock); 640cd6144aSJohannes Weiner } 651da177e4SLinus Torvalds 66cf9a2ae8SDavid Howells /** 6728bc44d7SFengguang Wu * do_invalidatepage - invalidate part or all of a page 68cf9a2ae8SDavid Howells * @page: the page which is affected 69d47992f8SLukas Czerner * @offset: start of the range to invalidate 70d47992f8SLukas Czerner * @length: length of the range to invalidate 71cf9a2ae8SDavid Howells * 72cf9a2ae8SDavid Howells * do_invalidatepage() is called when all or part of the page has become 73cf9a2ae8SDavid Howells * invalidated by a truncate operation. 74cf9a2ae8SDavid Howells * 75cf9a2ae8SDavid Howells * do_invalidatepage() does not have to release all buffers, but it must 76cf9a2ae8SDavid Howells * ensure that no dirty buffer is left outside @offset and that no I/O 77cf9a2ae8SDavid Howells * is underway against any of the blocks which are outside the truncation 78cf9a2ae8SDavid Howells * point. Because the caller is about to free (and possibly reuse) those 79cf9a2ae8SDavid Howells * blocks on-disk. 80cf9a2ae8SDavid Howells */ 81d47992f8SLukas Czerner void do_invalidatepage(struct page *page, unsigned int offset, 82d47992f8SLukas Czerner unsigned int length) 83cf9a2ae8SDavid Howells { 84d47992f8SLukas Czerner void (*invalidatepage)(struct page *, unsigned int, unsigned int); 85d47992f8SLukas Czerner 86cf9a2ae8SDavid Howells invalidatepage = page->mapping->a_ops->invalidatepage; 879361401eSDavid Howells #ifdef CONFIG_BLOCK 88cf9a2ae8SDavid Howells if (!invalidatepage) 89cf9a2ae8SDavid Howells invalidatepage = block_invalidatepage; 909361401eSDavid Howells #endif 91cf9a2ae8SDavid Howells if (invalidatepage) 92d47992f8SLukas Czerner (*invalidatepage)(page, offset, length); 93cf9a2ae8SDavid Howells } 94cf9a2ae8SDavid Howells 95ecdfc978SLinus Torvalds /* 96ecdfc978SLinus Torvalds * This cancels just the dirty bit on the kernel page itself, it 97ecdfc978SLinus Torvalds * does NOT actually remove dirty bits on any mmap's that may be 98ecdfc978SLinus Torvalds * around. It also leaves the page tagged dirty, so any sync 99ecdfc978SLinus Torvalds * activity will still find it on the dirty lists, and in particular, 100ecdfc978SLinus Torvalds * clear_page_dirty_for_io() will still look at the dirty bits in 101ecdfc978SLinus Torvalds * the VM. 102ecdfc978SLinus Torvalds * 103ecdfc978SLinus Torvalds * Doing this should *normally* only ever be done when a page 104ecdfc978SLinus Torvalds * is truncated, and is not actually mapped anywhere at all. However, 105ecdfc978SLinus Torvalds * fs/buffer.c does this when it notices that somebody has cleaned 106ecdfc978SLinus Torvalds * out all the buffers on a page without actually doing it through 107ecdfc978SLinus Torvalds * the VM. Can you say "ext3 is horribly ugly"? Tought you could. 108ecdfc978SLinus Torvalds */ 109fba2591bSLinus Torvalds void cancel_dirty_page(struct page *page, unsigned int account_size) 110fba2591bSLinus Torvalds { 1118368e328SLinus Torvalds if (TestClearPageDirty(page)) { 1128368e328SLinus Torvalds struct address_space *mapping = page->mapping; 1138368e328SLinus Torvalds if (mapping && mapping_cap_account_dirty(mapping)) { 1143e67c098SAndrew Morton dec_zone_page_state(page, NR_FILE_DIRTY); 115c9e51e41SPeter Zijlstra dec_bdi_stat(mapping->backing_dev_info, 116c9e51e41SPeter Zijlstra BDI_RECLAIMABLE); 1178368e328SLinus Torvalds if (account_size) 118fba2591bSLinus Torvalds task_io_account_cancelled_write(account_size); 119fba2591bSLinus Torvalds } 1203e67c098SAndrew Morton } 1218368e328SLinus Torvalds } 1228368e328SLinus Torvalds EXPORT_SYMBOL(cancel_dirty_page); 123fba2591bSLinus Torvalds 1241da177e4SLinus Torvalds /* 1251da177e4SLinus Torvalds * If truncate cannot remove the fs-private metadata from the page, the page 12662e1c553SShaohua Li * becomes orphaned. It will be left on the LRU and may even be mapped into 12754cb8821SNick Piggin * user pagetables if we're racing with filemap_fault(). 1281da177e4SLinus Torvalds * 1291da177e4SLinus Torvalds * We need to bale out if page->mapping is no longer equal to the original 1301da177e4SLinus Torvalds * mapping. This happens a) when the VM reclaimed the page while we waited on 131fc0ecff6SAndrew Morton * its lock, b) when a concurrent invalidate_mapping_pages got there first and 1321da177e4SLinus Torvalds * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. 1331da177e4SLinus Torvalds */ 134750b4987SNick Piggin static int 1351da177e4SLinus Torvalds truncate_complete_page(struct address_space *mapping, struct page *page) 1361da177e4SLinus Torvalds { 1371da177e4SLinus Torvalds if (page->mapping != mapping) 138750b4987SNick Piggin return -EIO; 1391da177e4SLinus Torvalds 140266cf658SDavid Howells if (page_has_private(page)) 141d47992f8SLukas Czerner do_invalidatepage(page, 0, PAGE_CACHE_SIZE); 1421da177e4SLinus Torvalds 143a2b34564SBjorn Steinbrink cancel_dirty_page(page, PAGE_CACHE_SIZE); 144a2b34564SBjorn Steinbrink 1451da177e4SLinus Torvalds ClearPageMappedToDisk(page); 1465adc7b51SMinchan Kim delete_from_page_cache(page); 147750b4987SNick Piggin return 0; 1481da177e4SLinus Torvalds } 1491da177e4SLinus Torvalds 1501da177e4SLinus Torvalds /* 151fc0ecff6SAndrew Morton * This is for invalidate_mapping_pages(). That function can be called at 1521da177e4SLinus Torvalds * any time, and is not supposed to throw away dirty pages. But pages can 1530fd0e6b0SNick Piggin * be marked dirty at any time too, so use remove_mapping which safely 1540fd0e6b0SNick Piggin * discards clean, unused pages. 1551da177e4SLinus Torvalds * 1561da177e4SLinus Torvalds * Returns non-zero if the page was successfully invalidated. 1571da177e4SLinus Torvalds */ 1581da177e4SLinus Torvalds static int 1591da177e4SLinus Torvalds invalidate_complete_page(struct address_space *mapping, struct page *page) 1601da177e4SLinus Torvalds { 1610fd0e6b0SNick Piggin int ret; 1620fd0e6b0SNick Piggin 1631da177e4SLinus Torvalds if (page->mapping != mapping) 1641da177e4SLinus Torvalds return 0; 1651da177e4SLinus Torvalds 166266cf658SDavid Howells if (page_has_private(page) && !try_to_release_page(page, 0)) 1671da177e4SLinus Torvalds return 0; 1681da177e4SLinus Torvalds 1690fd0e6b0SNick Piggin ret = remove_mapping(mapping, page); 1700fd0e6b0SNick Piggin 1710fd0e6b0SNick Piggin return ret; 1721da177e4SLinus Torvalds } 1731da177e4SLinus Torvalds 174750b4987SNick Piggin int truncate_inode_page(struct address_space *mapping, struct page *page) 175750b4987SNick Piggin { 176750b4987SNick Piggin if (page_mapped(page)) { 177750b4987SNick Piggin unmap_mapping_range(mapping, 178750b4987SNick Piggin (loff_t)page->index << PAGE_CACHE_SHIFT, 179750b4987SNick Piggin PAGE_CACHE_SIZE, 0); 180750b4987SNick Piggin } 181750b4987SNick Piggin return truncate_complete_page(mapping, page); 182750b4987SNick Piggin } 183750b4987SNick Piggin 18483f78668SWu Fengguang /* 18525718736SAndi Kleen * Used to get rid of pages on hardware memory corruption. 18625718736SAndi Kleen */ 18725718736SAndi Kleen int generic_error_remove_page(struct address_space *mapping, struct page *page) 18825718736SAndi Kleen { 18925718736SAndi Kleen if (!mapping) 19025718736SAndi Kleen return -EINVAL; 19125718736SAndi Kleen /* 19225718736SAndi Kleen * Only punch for normal data pages for now. 19325718736SAndi Kleen * Handling other types like directories would need more auditing. 19425718736SAndi Kleen */ 19525718736SAndi Kleen if (!S_ISREG(mapping->host->i_mode)) 19625718736SAndi Kleen return -EIO; 19725718736SAndi Kleen return truncate_inode_page(mapping, page); 19825718736SAndi Kleen } 19925718736SAndi Kleen EXPORT_SYMBOL(generic_error_remove_page); 20025718736SAndi Kleen 20125718736SAndi Kleen /* 20283f78668SWu Fengguang * Safely invalidate one page from its pagecache mapping. 20383f78668SWu Fengguang * It only drops clean, unused pages. The page must be locked. 20483f78668SWu Fengguang * 20583f78668SWu Fengguang * Returns 1 if the page is successfully invalidated, otherwise 0. 20683f78668SWu Fengguang */ 20783f78668SWu Fengguang int invalidate_inode_page(struct page *page) 20883f78668SWu Fengguang { 20983f78668SWu Fengguang struct address_space *mapping = page_mapping(page); 21083f78668SWu Fengguang if (!mapping) 21183f78668SWu Fengguang return 0; 21283f78668SWu Fengguang if (PageDirty(page) || PageWriteback(page)) 21383f78668SWu Fengguang return 0; 21483f78668SWu Fengguang if (page_mapped(page)) 21583f78668SWu Fengguang return 0; 21683f78668SWu Fengguang return invalidate_complete_page(mapping, page); 21783f78668SWu Fengguang } 21883f78668SWu Fengguang 2191da177e4SLinus Torvalds /** 22073c1e204SLiu Bo * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets 2211da177e4SLinus Torvalds * @mapping: mapping to truncate 2221da177e4SLinus Torvalds * @lstart: offset from which to truncate 2235a720394SLukas Czerner * @lend: offset to which to truncate (inclusive) 2241da177e4SLinus Torvalds * 225d7339071SHans Reiser * Truncate the page cache, removing the pages that are between 2265a720394SLukas Czerner * specified offsets (and zeroing out partial pages 2275a720394SLukas Czerner * if lstart or lend + 1 is not page aligned). 2281da177e4SLinus Torvalds * 2291da177e4SLinus Torvalds * Truncate takes two passes - the first pass is nonblocking. It will not 2301da177e4SLinus Torvalds * block on page locks and it will not block on writeback. The second pass 2311da177e4SLinus Torvalds * will wait. This is to prevent as much IO as possible in the affected region. 2321da177e4SLinus Torvalds * The first pass will remove most pages, so the search cost of the second pass 2331da177e4SLinus Torvalds * is low. 2341da177e4SLinus Torvalds * 2351da177e4SLinus Torvalds * We pass down the cache-hot hint to the page freeing code. Even if the 2361da177e4SLinus Torvalds * mapping is large, it is probably the case that the final pages are the most 2371da177e4SLinus Torvalds * recently touched, and freeing happens in ascending file offset order. 2385a720394SLukas Czerner * 2395a720394SLukas Czerner * Note that since ->invalidatepage() accepts range to invalidate 2405a720394SLukas Czerner * truncate_inode_pages_range is able to handle cases where lend + 1 is not 2415a720394SLukas Czerner * page aligned properly. 2421da177e4SLinus Torvalds */ 243d7339071SHans Reiser void truncate_inode_pages_range(struct address_space *mapping, 244d7339071SHans Reiser loff_t lstart, loff_t lend) 2451da177e4SLinus Torvalds { 2465a720394SLukas Czerner pgoff_t start; /* inclusive */ 2475a720394SLukas Czerner pgoff_t end; /* exclusive */ 2485a720394SLukas Czerner unsigned int partial_start; /* inclusive */ 2495a720394SLukas Czerner unsigned int partial_end; /* exclusive */ 2501da177e4SLinus Torvalds struct pagevec pvec; 2510cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 252b85e0effSHugh Dickins pgoff_t index; 2531da177e4SLinus Torvalds int i; 2541da177e4SLinus Torvalds 2553167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 25691b0abe3SJohannes Weiner if (mapping->nrpages == 0 && mapping->nrshadows == 0) 2571da177e4SLinus Torvalds return; 2581da177e4SLinus Torvalds 2595a720394SLukas Czerner /* Offsets within partial pages */ 2605a720394SLukas Czerner partial_start = lstart & (PAGE_CACHE_SIZE - 1); 2615a720394SLukas Czerner partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1); 2625a720394SLukas Czerner 2635a720394SLukas Czerner /* 2645a720394SLukas Czerner * 'start' and 'end' always covers the range of pages to be fully 2655a720394SLukas Czerner * truncated. Partial pages are covered with 'partial_start' at the 2665a720394SLukas Czerner * start of the range and 'partial_end' at the end of the range. 2675a720394SLukas Czerner * Note that 'end' is exclusive while 'lend' is inclusive. 2685a720394SLukas Czerner */ 2695a720394SLukas Czerner start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 2705a720394SLukas Czerner if (lend == -1) 2715a720394SLukas Czerner /* 2725a720394SLukas Czerner * lend == -1 indicates end-of-file so we have to set 'end' 2735a720394SLukas Czerner * to the highest possible pgoff_t and since the type is 2745a720394SLukas Czerner * unsigned we're using -1. 2755a720394SLukas Czerner */ 2765a720394SLukas Czerner end = -1; 2775a720394SLukas Czerner else 2785a720394SLukas Czerner end = (lend + 1) >> PAGE_CACHE_SHIFT; 279d7339071SHans Reiser 2801da177e4SLinus Torvalds pagevec_init(&pvec, 0); 281b85e0effSHugh Dickins index = start; 2820cd6144aSJohannes Weiner while (index < end && pagevec_lookup_entries(&pvec, mapping, index, 2830cd6144aSJohannes Weiner min(end - index, (pgoff_t)PAGEVEC_SIZE), 2840cd6144aSJohannes Weiner indices)) { 2851da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 2861da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 2871da177e4SLinus Torvalds 288b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 2890cd6144aSJohannes Weiner index = indices[i]; 2905a720394SLukas Czerner if (index >= end) 291d7339071SHans Reiser break; 292d7339071SHans Reiser 2930cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 2940cd6144aSJohannes Weiner clear_exceptional_entry(mapping, index, page); 2950cd6144aSJohannes Weiner continue; 2960cd6144aSJohannes Weiner } 2970cd6144aSJohannes Weiner 298529ae9aaSNick Piggin if (!trylock_page(page)) 2991da177e4SLinus Torvalds continue; 300b85e0effSHugh Dickins WARN_ON(page->index != index); 3011da177e4SLinus Torvalds if (PageWriteback(page)) { 3021da177e4SLinus Torvalds unlock_page(page); 3031da177e4SLinus Torvalds continue; 3041da177e4SLinus Torvalds } 305750b4987SNick Piggin truncate_inode_page(mapping, page); 3061da177e4SLinus Torvalds unlock_page(page); 3071da177e4SLinus Torvalds } 3080cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 3091da177e4SLinus Torvalds pagevec_release(&pvec); 3101da177e4SLinus Torvalds cond_resched(); 311b85e0effSHugh Dickins index++; 3121da177e4SLinus Torvalds } 3131da177e4SLinus Torvalds 3145a720394SLukas Czerner if (partial_start) { 3151da177e4SLinus Torvalds struct page *page = find_lock_page(mapping, start - 1); 3161da177e4SLinus Torvalds if (page) { 3175a720394SLukas Czerner unsigned int top = PAGE_CACHE_SIZE; 3185a720394SLukas Czerner if (start > end) { 3195a720394SLukas Czerner /* Truncation within a single page */ 3205a720394SLukas Czerner top = partial_end; 3215a720394SLukas Czerner partial_end = 0; 3225a720394SLukas Czerner } 3231da177e4SLinus Torvalds wait_on_page_writeback(page); 3245a720394SLukas Czerner zero_user_segment(page, partial_start, top); 3255a720394SLukas Czerner cleancache_invalidate_page(mapping, page); 3265a720394SLukas Czerner if (page_has_private(page)) 3275a720394SLukas Czerner do_invalidatepage(page, partial_start, 3285a720394SLukas Czerner top - partial_start); 3291da177e4SLinus Torvalds unlock_page(page); 3301da177e4SLinus Torvalds page_cache_release(page); 3311da177e4SLinus Torvalds } 3321da177e4SLinus Torvalds } 3335a720394SLukas Czerner if (partial_end) { 3345a720394SLukas Czerner struct page *page = find_lock_page(mapping, end); 3355a720394SLukas Czerner if (page) { 3365a720394SLukas Czerner wait_on_page_writeback(page); 3375a720394SLukas Czerner zero_user_segment(page, 0, partial_end); 3385a720394SLukas Czerner cleancache_invalidate_page(mapping, page); 3395a720394SLukas Czerner if (page_has_private(page)) 3405a720394SLukas Czerner do_invalidatepage(page, 0, 3415a720394SLukas Czerner partial_end); 3425a720394SLukas Czerner unlock_page(page); 3435a720394SLukas Czerner page_cache_release(page); 3445a720394SLukas Czerner } 3455a720394SLukas Czerner } 3465a720394SLukas Czerner /* 3475a720394SLukas Czerner * If the truncation happened within a single page no pages 3485a720394SLukas Czerner * will be released, just zeroed, so we can bail out now. 3495a720394SLukas Czerner */ 3505a720394SLukas Czerner if (start >= end) 3515a720394SLukas Czerner return; 3521da177e4SLinus Torvalds 353b85e0effSHugh Dickins index = start; 3541da177e4SLinus Torvalds for ( ; ; ) { 3551da177e4SLinus Torvalds cond_resched(); 3560cd6144aSJohannes Weiner if (!pagevec_lookup_entries(&pvec, mapping, index, 357792ceaefSHugh Dickins min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) { 358792ceaefSHugh Dickins /* If all gone from start onwards, we're done */ 359b85e0effSHugh Dickins if (index == start) 3601da177e4SLinus Torvalds break; 361792ceaefSHugh Dickins /* Otherwise restart to make sure all gone */ 362b85e0effSHugh Dickins index = start; 3631da177e4SLinus Torvalds continue; 3641da177e4SLinus Torvalds } 3650cd6144aSJohannes Weiner if (index == start && indices[0] >= end) { 366792ceaefSHugh Dickins /* All gone out of hole to be punched, we're done */ 3670cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 368d7339071SHans Reiser pagevec_release(&pvec); 369d7339071SHans Reiser break; 370d7339071SHans Reiser } 3711da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 3721da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 3731da177e4SLinus Torvalds 374b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 3750cd6144aSJohannes Weiner index = indices[i]; 376792ceaefSHugh Dickins if (index >= end) { 377792ceaefSHugh Dickins /* Restart punch to make sure all gone */ 378792ceaefSHugh Dickins index = start - 1; 379d7339071SHans Reiser break; 380792ceaefSHugh Dickins } 381b85e0effSHugh Dickins 3820cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 3830cd6144aSJohannes Weiner clear_exceptional_entry(mapping, index, page); 3840cd6144aSJohannes Weiner continue; 3850cd6144aSJohannes Weiner } 3860cd6144aSJohannes Weiner 3871da177e4SLinus Torvalds lock_page(page); 388b85e0effSHugh Dickins WARN_ON(page->index != index); 3891da177e4SLinus Torvalds wait_on_page_writeback(page); 390750b4987SNick Piggin truncate_inode_page(mapping, page); 3911da177e4SLinus Torvalds unlock_page(page); 3921da177e4SLinus Torvalds } 3930cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 3941da177e4SLinus Torvalds pagevec_release(&pvec); 395b85e0effSHugh Dickins index++; 3961da177e4SLinus Torvalds } 3973167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 3981da177e4SLinus Torvalds } 399d7339071SHans Reiser EXPORT_SYMBOL(truncate_inode_pages_range); 4001da177e4SLinus Torvalds 401d7339071SHans Reiser /** 402d7339071SHans Reiser * truncate_inode_pages - truncate *all* the pages from an offset 403d7339071SHans Reiser * @mapping: mapping to truncate 404d7339071SHans Reiser * @lstart: offset from which to truncate 405d7339071SHans Reiser * 4061b1dcc1bSJes Sorensen * Called under (and serialised by) inode->i_mutex. 40708142579SJan Kara * 40808142579SJan Kara * Note: When this function returns, there can be a page in the process of 40908142579SJan Kara * deletion (inside __delete_from_page_cache()) in the specified range. Thus 41008142579SJan Kara * mapping->nrpages can be non-zero when this function returns even after 41108142579SJan Kara * truncation of the whole mapping. 412d7339071SHans Reiser */ 413d7339071SHans Reiser void truncate_inode_pages(struct address_space *mapping, loff_t lstart) 414d7339071SHans Reiser { 415d7339071SHans Reiser truncate_inode_pages_range(mapping, lstart, (loff_t)-1); 416d7339071SHans Reiser } 4171da177e4SLinus Torvalds EXPORT_SYMBOL(truncate_inode_pages); 4181da177e4SLinus Torvalds 41928697355SMike Waychison /** 42091b0abe3SJohannes Weiner * truncate_inode_pages_final - truncate *all* pages before inode dies 42191b0abe3SJohannes Weiner * @mapping: mapping to truncate 42291b0abe3SJohannes Weiner * 42391b0abe3SJohannes Weiner * Called under (and serialized by) inode->i_mutex. 42491b0abe3SJohannes Weiner * 42591b0abe3SJohannes Weiner * Filesystems have to use this in the .evict_inode path to inform the 42691b0abe3SJohannes Weiner * VM that this is the final truncate and the inode is going away. 42791b0abe3SJohannes Weiner */ 42891b0abe3SJohannes Weiner void truncate_inode_pages_final(struct address_space *mapping) 42991b0abe3SJohannes Weiner { 43091b0abe3SJohannes Weiner unsigned long nrshadows; 43191b0abe3SJohannes Weiner unsigned long nrpages; 43291b0abe3SJohannes Weiner 43391b0abe3SJohannes Weiner /* 43491b0abe3SJohannes Weiner * Page reclaim can not participate in regular inode lifetime 43591b0abe3SJohannes Weiner * management (can't call iput()) and thus can race with the 43691b0abe3SJohannes Weiner * inode teardown. Tell it when the address space is exiting, 43791b0abe3SJohannes Weiner * so that it does not install eviction information after the 43891b0abe3SJohannes Weiner * final truncate has begun. 43991b0abe3SJohannes Weiner */ 44091b0abe3SJohannes Weiner mapping_set_exiting(mapping); 44191b0abe3SJohannes Weiner 44291b0abe3SJohannes Weiner /* 44391b0abe3SJohannes Weiner * When reclaim installs eviction entries, it increases 44491b0abe3SJohannes Weiner * nrshadows first, then decreases nrpages. Make sure we see 44591b0abe3SJohannes Weiner * this in the right order or we might miss an entry. 44691b0abe3SJohannes Weiner */ 44791b0abe3SJohannes Weiner nrpages = mapping->nrpages; 44891b0abe3SJohannes Weiner smp_rmb(); 44991b0abe3SJohannes Weiner nrshadows = mapping->nrshadows; 45091b0abe3SJohannes Weiner 45191b0abe3SJohannes Weiner if (nrpages || nrshadows) { 45291b0abe3SJohannes Weiner /* 45391b0abe3SJohannes Weiner * As truncation uses a lockless tree lookup, cycle 45491b0abe3SJohannes Weiner * the tree lock to make sure any ongoing tree 45591b0abe3SJohannes Weiner * modification that does not see AS_EXITING is 45691b0abe3SJohannes Weiner * completed before starting the final truncate. 45791b0abe3SJohannes Weiner */ 45891b0abe3SJohannes Weiner spin_lock_irq(&mapping->tree_lock); 45991b0abe3SJohannes Weiner spin_unlock_irq(&mapping->tree_lock); 46091b0abe3SJohannes Weiner 46191b0abe3SJohannes Weiner truncate_inode_pages(mapping, 0); 46291b0abe3SJohannes Weiner } 46391b0abe3SJohannes Weiner } 46491b0abe3SJohannes Weiner EXPORT_SYMBOL(truncate_inode_pages_final); 46591b0abe3SJohannes Weiner 46691b0abe3SJohannes Weiner /** 46728697355SMike Waychison * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode 46828697355SMike Waychison * @mapping: the address_space which holds the pages to invalidate 46928697355SMike Waychison * @start: the offset 'from' which to invalidate 47028697355SMike Waychison * @end: the offset 'to' which to invalidate (inclusive) 47128697355SMike Waychison * 47228697355SMike Waychison * This function only removes the unlocked pages, if you want to 47328697355SMike Waychison * remove all the pages of one inode, you must call truncate_inode_pages. 47428697355SMike Waychison * 47528697355SMike Waychison * invalidate_mapping_pages() will not block on IO activity. It will not 47628697355SMike Waychison * invalidate pages which are dirty, locked, under writeback or mapped into 47728697355SMike Waychison * pagetables. 47828697355SMike Waychison */ 47928697355SMike Waychison unsigned long invalidate_mapping_pages(struct address_space *mapping, 48028697355SMike Waychison pgoff_t start, pgoff_t end) 4811da177e4SLinus Torvalds { 4820cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 4831da177e4SLinus Torvalds struct pagevec pvec; 484b85e0effSHugh Dickins pgoff_t index = start; 48531560180SMinchan Kim unsigned long ret; 48631560180SMinchan Kim unsigned long count = 0; 4871da177e4SLinus Torvalds int i; 4881da177e4SLinus Torvalds 4891da177e4SLinus Torvalds pagevec_init(&pvec, 0); 4900cd6144aSJohannes Weiner while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, 4910cd6144aSJohannes Weiner min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, 4920cd6144aSJohannes Weiner indices)) { 4931da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 4941da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 4951da177e4SLinus Torvalds 496b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 4970cd6144aSJohannes Weiner index = indices[i]; 498b85e0effSHugh Dickins if (index > end) 499b85e0effSHugh Dickins break; 500e0f23603SNeilBrown 5010cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 5020cd6144aSJohannes Weiner clear_exceptional_entry(mapping, index, page); 5030cd6144aSJohannes Weiner continue; 5040cd6144aSJohannes Weiner } 5050cd6144aSJohannes Weiner 506b85e0effSHugh Dickins if (!trylock_page(page)) 507b85e0effSHugh Dickins continue; 508b85e0effSHugh Dickins WARN_ON(page->index != index); 50931560180SMinchan Kim ret = invalidate_inode_page(page); 5101da177e4SLinus Torvalds unlock_page(page); 51131560180SMinchan Kim /* 51231560180SMinchan Kim * Invalidation is a hint that the page is no longer 51331560180SMinchan Kim * of interest and try to speed up its reclaim. 51431560180SMinchan Kim */ 51531560180SMinchan Kim if (!ret) 51631560180SMinchan Kim deactivate_page(page); 51731560180SMinchan Kim count += ret; 5181da177e4SLinus Torvalds } 5190cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 5201da177e4SLinus Torvalds pagevec_release(&pvec); 521fc9a07e7SAndrew Morton cond_resched(); 522b85e0effSHugh Dickins index++; 5231da177e4SLinus Torvalds } 52431560180SMinchan Kim return count; 5251da177e4SLinus Torvalds } 52654bc4855SAnton Altaparmakov EXPORT_SYMBOL(invalidate_mapping_pages); 5271da177e4SLinus Torvalds 528bd4c8ce4SAndrew Morton /* 529bd4c8ce4SAndrew Morton * This is like invalidate_complete_page(), except it ignores the page's 530bd4c8ce4SAndrew Morton * refcount. We do this because invalidate_inode_pages2() needs stronger 531bd4c8ce4SAndrew Morton * invalidation guarantees, and cannot afford to leave pages behind because 5322706a1b8SAnderson Briglia * shrink_page_list() has a temp ref on them, or because they're transiently 5332706a1b8SAnderson Briglia * sitting in the lru_cache_add() pagevecs. 534bd4c8ce4SAndrew Morton */ 535bd4c8ce4SAndrew Morton static int 536bd4c8ce4SAndrew Morton invalidate_complete_page2(struct address_space *mapping, struct page *page) 537bd4c8ce4SAndrew Morton { 538bd4c8ce4SAndrew Morton if (page->mapping != mapping) 539bd4c8ce4SAndrew Morton return 0; 540bd4c8ce4SAndrew Morton 541266cf658SDavid Howells if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL)) 542bd4c8ce4SAndrew Morton return 0; 543bd4c8ce4SAndrew Morton 54419fd6231SNick Piggin spin_lock_irq(&mapping->tree_lock); 545bd4c8ce4SAndrew Morton if (PageDirty(page)) 546bd4c8ce4SAndrew Morton goto failed; 547bd4c8ce4SAndrew Morton 548266cf658SDavid Howells BUG_ON(page_has_private(page)); 54991b0abe3SJohannes Weiner __delete_from_page_cache(page, NULL); 55019fd6231SNick Piggin spin_unlock_irq(&mapping->tree_lock); 5516072d13cSLinus Torvalds 5526072d13cSLinus Torvalds if (mapping->a_ops->freepage) 5536072d13cSLinus Torvalds mapping->a_ops->freepage(page); 5546072d13cSLinus Torvalds 555bd4c8ce4SAndrew Morton page_cache_release(page); /* pagecache ref */ 556bd4c8ce4SAndrew Morton return 1; 557bd4c8ce4SAndrew Morton failed: 55819fd6231SNick Piggin spin_unlock_irq(&mapping->tree_lock); 559bd4c8ce4SAndrew Morton return 0; 560bd4c8ce4SAndrew Morton } 561bd4c8ce4SAndrew Morton 562e3db7691STrond Myklebust static int do_launder_page(struct address_space *mapping, struct page *page) 563e3db7691STrond Myklebust { 564e3db7691STrond Myklebust if (!PageDirty(page)) 565e3db7691STrond Myklebust return 0; 566e3db7691STrond Myklebust if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) 567e3db7691STrond Myklebust return 0; 568e3db7691STrond Myklebust return mapping->a_ops->launder_page(page); 569e3db7691STrond Myklebust } 570e3db7691STrond Myklebust 5711da177e4SLinus Torvalds /** 5721da177e4SLinus Torvalds * invalidate_inode_pages2_range - remove range of pages from an address_space 57367be2dd1SMartin Waitz * @mapping: the address_space 5741da177e4SLinus Torvalds * @start: the page offset 'from' which to invalidate 5751da177e4SLinus Torvalds * @end: the page offset 'to' which to invalidate (inclusive) 5761da177e4SLinus Torvalds * 5771da177e4SLinus Torvalds * Any pages which are found to be mapped into pagetables are unmapped prior to 5781da177e4SLinus Torvalds * invalidation. 5791da177e4SLinus Torvalds * 5806ccfa806SHisashi Hifumi * Returns -EBUSY if any pages could not be invalidated. 5811da177e4SLinus Torvalds */ 5821da177e4SLinus Torvalds int invalidate_inode_pages2_range(struct address_space *mapping, 5831da177e4SLinus Torvalds pgoff_t start, pgoff_t end) 5841da177e4SLinus Torvalds { 5850cd6144aSJohannes Weiner pgoff_t indices[PAGEVEC_SIZE]; 5861da177e4SLinus Torvalds struct pagevec pvec; 587b85e0effSHugh Dickins pgoff_t index; 5881da177e4SLinus Torvalds int i; 5891da177e4SLinus Torvalds int ret = 0; 5900dd1334fSHisashi Hifumi int ret2 = 0; 5911da177e4SLinus Torvalds int did_range_unmap = 0; 5921da177e4SLinus Torvalds 5933167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 5941da177e4SLinus Torvalds pagevec_init(&pvec, 0); 595b85e0effSHugh Dickins index = start; 5960cd6144aSJohannes Weiner while (index <= end && pagevec_lookup_entries(&pvec, mapping, index, 5970cd6144aSJohannes Weiner min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1, 5980cd6144aSJohannes Weiner indices)) { 5997b965e08STrond Myklebust for (i = 0; i < pagevec_count(&pvec); i++) { 6001da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 601b85e0effSHugh Dickins 602b85e0effSHugh Dickins /* We rely upon deletion not changing page->index */ 6030cd6144aSJohannes Weiner index = indices[i]; 604b85e0effSHugh Dickins if (index > end) 605b85e0effSHugh Dickins break; 6061da177e4SLinus Torvalds 6070cd6144aSJohannes Weiner if (radix_tree_exceptional_entry(page)) { 6080cd6144aSJohannes Weiner clear_exceptional_entry(mapping, index, page); 6090cd6144aSJohannes Weiner continue; 6100cd6144aSJohannes Weiner } 6110cd6144aSJohannes Weiner 6121da177e4SLinus Torvalds lock_page(page); 613b85e0effSHugh Dickins WARN_ON(page->index != index); 6141da177e4SLinus Torvalds if (page->mapping != mapping) { 6151da177e4SLinus Torvalds unlock_page(page); 6161da177e4SLinus Torvalds continue; 6171da177e4SLinus Torvalds } 6181da177e4SLinus Torvalds wait_on_page_writeback(page); 619d00806b1SNick Piggin if (page_mapped(page)) { 6201da177e4SLinus Torvalds if (!did_range_unmap) { 6211da177e4SLinus Torvalds /* 6221da177e4SLinus Torvalds * Zap the rest of the file in one hit. 6231da177e4SLinus Torvalds */ 6241da177e4SLinus Torvalds unmap_mapping_range(mapping, 625b85e0effSHugh Dickins (loff_t)index << PAGE_CACHE_SHIFT, 626b85e0effSHugh Dickins (loff_t)(1 + end - index) 6271da177e4SLinus Torvalds << PAGE_CACHE_SHIFT, 6281da177e4SLinus Torvalds 0); 6291da177e4SLinus Torvalds did_range_unmap = 1; 6301da177e4SLinus Torvalds } else { 6311da177e4SLinus Torvalds /* 6321da177e4SLinus Torvalds * Just zap this page 6331da177e4SLinus Torvalds */ 6341da177e4SLinus Torvalds unmap_mapping_range(mapping, 635b85e0effSHugh Dickins (loff_t)index << PAGE_CACHE_SHIFT, 6361da177e4SLinus Torvalds PAGE_CACHE_SIZE, 0); 6371da177e4SLinus Torvalds } 6381da177e4SLinus Torvalds } 639d00806b1SNick Piggin BUG_ON(page_mapped(page)); 6400dd1334fSHisashi Hifumi ret2 = do_launder_page(mapping, page); 6410dd1334fSHisashi Hifumi if (ret2 == 0) { 6420dd1334fSHisashi Hifumi if (!invalidate_complete_page2(mapping, page)) 6436ccfa806SHisashi Hifumi ret2 = -EBUSY; 6440dd1334fSHisashi Hifumi } 6450dd1334fSHisashi Hifumi if (ret2 < 0) 6460dd1334fSHisashi Hifumi ret = ret2; 6471da177e4SLinus Torvalds unlock_page(page); 6481da177e4SLinus Torvalds } 6490cd6144aSJohannes Weiner pagevec_remove_exceptionals(&pvec); 6501da177e4SLinus Torvalds pagevec_release(&pvec); 6511da177e4SLinus Torvalds cond_resched(); 652b85e0effSHugh Dickins index++; 6531da177e4SLinus Torvalds } 6543167760fSDan Magenheimer cleancache_invalidate_inode(mapping); 6551da177e4SLinus Torvalds return ret; 6561da177e4SLinus Torvalds } 6571da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); 6581da177e4SLinus Torvalds 6591da177e4SLinus Torvalds /** 6601da177e4SLinus Torvalds * invalidate_inode_pages2 - remove all pages from an address_space 66167be2dd1SMartin Waitz * @mapping: the address_space 6621da177e4SLinus Torvalds * 6631da177e4SLinus Torvalds * Any pages which are found to be mapped into pagetables are unmapped prior to 6641da177e4SLinus Torvalds * invalidation. 6651da177e4SLinus Torvalds * 666e9de25ddSPeng Tao * Returns -EBUSY if any pages could not be invalidated. 6671da177e4SLinus Torvalds */ 6681da177e4SLinus Torvalds int invalidate_inode_pages2(struct address_space *mapping) 6691da177e4SLinus Torvalds { 6701da177e4SLinus Torvalds return invalidate_inode_pages2_range(mapping, 0, -1); 6711da177e4SLinus Torvalds } 6721da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(invalidate_inode_pages2); 67325d9e2d1Snpiggin@suse.de 67425d9e2d1Snpiggin@suse.de /** 67525d9e2d1Snpiggin@suse.de * truncate_pagecache - unmap and remove pagecache that has been truncated 67625d9e2d1Snpiggin@suse.de * @inode: inode 6778a549beaSHugh Dickins * @newsize: new file size 67825d9e2d1Snpiggin@suse.de * 67925d9e2d1Snpiggin@suse.de * inode's new i_size must already be written before truncate_pagecache 68025d9e2d1Snpiggin@suse.de * is called. 68125d9e2d1Snpiggin@suse.de * 68225d9e2d1Snpiggin@suse.de * This function should typically be called before the filesystem 68325d9e2d1Snpiggin@suse.de * releases resources associated with the freed range (eg. deallocates 68425d9e2d1Snpiggin@suse.de * blocks). This way, pagecache will always stay logically coherent 68525d9e2d1Snpiggin@suse.de * with on-disk format, and the filesystem would not have to deal with 68625d9e2d1Snpiggin@suse.de * situations such as writepage being called for a page that has already 68725d9e2d1Snpiggin@suse.de * had its underlying blocks deallocated. 68825d9e2d1Snpiggin@suse.de */ 6897caef267SKirill A. Shutemov void truncate_pagecache(struct inode *inode, loff_t newsize) 69025d9e2d1Snpiggin@suse.de { 69125d9e2d1Snpiggin@suse.de struct address_space *mapping = inode->i_mapping; 6928a549beaSHugh Dickins loff_t holebegin = round_up(newsize, PAGE_SIZE); 69325d9e2d1Snpiggin@suse.de 69425d9e2d1Snpiggin@suse.de /* 69525d9e2d1Snpiggin@suse.de * unmap_mapping_range is called twice, first simply for 69625d9e2d1Snpiggin@suse.de * efficiency so that truncate_inode_pages does fewer 69725d9e2d1Snpiggin@suse.de * single-page unmaps. However after this first call, and 69825d9e2d1Snpiggin@suse.de * before truncate_inode_pages finishes, it is possible for 69925d9e2d1Snpiggin@suse.de * private pages to be COWed, which remain after 70025d9e2d1Snpiggin@suse.de * truncate_inode_pages finishes, hence the second 70125d9e2d1Snpiggin@suse.de * unmap_mapping_range call must be made for correctness. 70225d9e2d1Snpiggin@suse.de */ 7038a549beaSHugh Dickins unmap_mapping_range(mapping, holebegin, 0, 1); 7048a549beaSHugh Dickins truncate_inode_pages(mapping, newsize); 7058a549beaSHugh Dickins unmap_mapping_range(mapping, holebegin, 0, 1); 70625d9e2d1Snpiggin@suse.de } 70725d9e2d1Snpiggin@suse.de EXPORT_SYMBOL(truncate_pagecache); 70825d9e2d1Snpiggin@suse.de 70925d9e2d1Snpiggin@suse.de /** 7102c27c65eSChristoph Hellwig * truncate_setsize - update inode and pagecache for a new file size 7112c27c65eSChristoph Hellwig * @inode: inode 7122c27c65eSChristoph Hellwig * @newsize: new file size 7132c27c65eSChristoph Hellwig * 714382e27daSJan Kara * truncate_setsize updates i_size and performs pagecache truncation (if 715382e27daSJan Kara * necessary) to @newsize. It will be typically be called from the filesystem's 716382e27daSJan Kara * setattr function when ATTR_SIZE is passed in. 7172c27c65eSChristoph Hellwig * 718382e27daSJan Kara * Must be called with inode_mutex held and before all filesystem specific 719382e27daSJan Kara * block truncation has been performed. 7202c27c65eSChristoph Hellwig */ 7212c27c65eSChristoph Hellwig void truncate_setsize(struct inode *inode, loff_t newsize) 7222c27c65eSChristoph Hellwig { 723*90a80202SJan Kara loff_t oldsize = inode->i_size; 724*90a80202SJan Kara 7252c27c65eSChristoph Hellwig i_size_write(inode, newsize); 726*90a80202SJan Kara if (newsize > oldsize) 727*90a80202SJan Kara pagecache_isize_extended(inode, oldsize, newsize); 7287caef267SKirill A. Shutemov truncate_pagecache(inode, newsize); 7292c27c65eSChristoph Hellwig } 7302c27c65eSChristoph Hellwig EXPORT_SYMBOL(truncate_setsize); 7312c27c65eSChristoph Hellwig 7322c27c65eSChristoph Hellwig /** 733*90a80202SJan Kara * pagecache_isize_extended - update pagecache after extension of i_size 734*90a80202SJan Kara * @inode: inode for which i_size was extended 735*90a80202SJan Kara * @from: original inode size 736*90a80202SJan Kara * @to: new inode size 737*90a80202SJan Kara * 738*90a80202SJan Kara * Handle extension of inode size either caused by extending truncate or by 739*90a80202SJan Kara * write starting after current i_size. We mark the page straddling current 740*90a80202SJan Kara * i_size RO so that page_mkwrite() is called on the nearest write access to 741*90a80202SJan Kara * the page. This way filesystem can be sure that page_mkwrite() is called on 742*90a80202SJan Kara * the page before user writes to the page via mmap after the i_size has been 743*90a80202SJan Kara * changed. 744*90a80202SJan Kara * 745*90a80202SJan Kara * The function must be called after i_size is updated so that page fault 746*90a80202SJan Kara * coming after we unlock the page will already see the new i_size. 747*90a80202SJan Kara * The function must be called while we still hold i_mutex - this not only 748*90a80202SJan Kara * makes sure i_size is stable but also that userspace cannot observe new 749*90a80202SJan Kara * i_size value before we are prepared to store mmap writes at new inode size. 750*90a80202SJan Kara */ 751*90a80202SJan Kara void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to) 752*90a80202SJan Kara { 753*90a80202SJan Kara int bsize = 1 << inode->i_blkbits; 754*90a80202SJan Kara loff_t rounded_from; 755*90a80202SJan Kara struct page *page; 756*90a80202SJan Kara pgoff_t index; 757*90a80202SJan Kara 758*90a80202SJan Kara WARN_ON(!mutex_is_locked(&inode->i_mutex)); 759*90a80202SJan Kara WARN_ON(to > inode->i_size); 760*90a80202SJan Kara 761*90a80202SJan Kara if (from >= to || bsize == PAGE_CACHE_SIZE) 762*90a80202SJan Kara return; 763*90a80202SJan Kara /* Page straddling @from will not have any hole block created? */ 764*90a80202SJan Kara rounded_from = round_up(from, bsize); 765*90a80202SJan Kara if (to <= rounded_from || !(rounded_from & (PAGE_CACHE_SIZE - 1))) 766*90a80202SJan Kara return; 767*90a80202SJan Kara 768*90a80202SJan Kara index = from >> PAGE_CACHE_SHIFT; 769*90a80202SJan Kara page = find_lock_page(inode->i_mapping, index); 770*90a80202SJan Kara /* Page not cached? Nothing to do */ 771*90a80202SJan Kara if (!page) 772*90a80202SJan Kara return; 773*90a80202SJan Kara /* 774*90a80202SJan Kara * See clear_page_dirty_for_io() for details why set_page_dirty() 775*90a80202SJan Kara * is needed. 776*90a80202SJan Kara */ 777*90a80202SJan Kara if (page_mkclean(page)) 778*90a80202SJan Kara set_page_dirty(page); 779*90a80202SJan Kara unlock_page(page); 780*90a80202SJan Kara page_cache_release(page); 781*90a80202SJan Kara } 782*90a80202SJan Kara EXPORT_SYMBOL(pagecache_isize_extended); 783*90a80202SJan Kara 784*90a80202SJan Kara /** 785623e3db9SHugh Dickins * truncate_pagecache_range - unmap and remove pagecache that is hole-punched 786623e3db9SHugh Dickins * @inode: inode 787623e3db9SHugh Dickins * @lstart: offset of beginning of hole 788623e3db9SHugh Dickins * @lend: offset of last byte of hole 789623e3db9SHugh Dickins * 790623e3db9SHugh Dickins * This function should typically be called before the filesystem 791623e3db9SHugh Dickins * releases resources associated with the freed range (eg. deallocates 792623e3db9SHugh Dickins * blocks). This way, pagecache will always stay logically coherent 793623e3db9SHugh Dickins * with on-disk format, and the filesystem would not have to deal with 794623e3db9SHugh Dickins * situations such as writepage being called for a page that has already 795623e3db9SHugh Dickins * had its underlying blocks deallocated. 796623e3db9SHugh Dickins */ 797623e3db9SHugh Dickins void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend) 798623e3db9SHugh Dickins { 799623e3db9SHugh Dickins struct address_space *mapping = inode->i_mapping; 800623e3db9SHugh Dickins loff_t unmap_start = round_up(lstart, PAGE_SIZE); 801623e3db9SHugh Dickins loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1; 802623e3db9SHugh Dickins /* 803623e3db9SHugh Dickins * This rounding is currently just for example: unmap_mapping_range 804623e3db9SHugh Dickins * expands its hole outwards, whereas we want it to contract the hole 805623e3db9SHugh Dickins * inwards. However, existing callers of truncate_pagecache_range are 8065a720394SLukas Czerner * doing their own page rounding first. Note that unmap_mapping_range 8075a720394SLukas Czerner * allows holelen 0 for all, and we allow lend -1 for end of file. 808623e3db9SHugh Dickins */ 809623e3db9SHugh Dickins 810623e3db9SHugh Dickins /* 811623e3db9SHugh Dickins * Unlike in truncate_pagecache, unmap_mapping_range is called only 812623e3db9SHugh Dickins * once (before truncating pagecache), and without "even_cows" flag: 813623e3db9SHugh Dickins * hole-punching should not remove private COWed pages from the hole. 814623e3db9SHugh Dickins */ 815623e3db9SHugh Dickins if ((u64)unmap_end > (u64)unmap_start) 816623e3db9SHugh Dickins unmap_mapping_range(mapping, unmap_start, 817623e3db9SHugh Dickins 1 + unmap_end - unmap_start, 0); 818623e3db9SHugh Dickins truncate_inode_pages_range(mapping, lstart, lend); 819623e3db9SHugh Dickins } 820623e3db9SHugh Dickins EXPORT_SYMBOL(truncate_pagecache_range); 821