xref: /linux/mm/truncate.c (revision c4843a7593a9df3ff5b1806084cefdfa81dd7c79)
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>
2390a80202SJan 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 /*
961da177e4SLinus Torvalds  * If truncate cannot remove the fs-private metadata from the page, the page
9762e1c553SShaohua Li  * becomes orphaned.  It will be left on the LRU and may even be mapped into
9854cb8821SNick Piggin  * user pagetables if we're racing with filemap_fault().
991da177e4SLinus Torvalds  *
1001da177e4SLinus Torvalds  * We need to bale out if page->mapping is no longer equal to the original
1011da177e4SLinus Torvalds  * mapping.  This happens a) when the VM reclaimed the page while we waited on
102fc0ecff6SAndrew Morton  * its lock, b) when a concurrent invalidate_mapping_pages got there first and
1031da177e4SLinus Torvalds  * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
1041da177e4SLinus Torvalds  */
105750b4987SNick Piggin static int
1061da177e4SLinus Torvalds truncate_complete_page(struct address_space *mapping, struct page *page)
1071da177e4SLinus Torvalds {
1081da177e4SLinus Torvalds 	if (page->mapping != mapping)
109750b4987SNick Piggin 		return -EIO;
1101da177e4SLinus Torvalds 
111266cf658SDavid Howells 	if (page_has_private(page))
112d47992f8SLukas Czerner 		do_invalidatepage(page, 0, PAGE_CACHE_SIZE);
1131da177e4SLinus Torvalds 
114b9ea2515SKonstantin Khlebnikov 	/*
115b9ea2515SKonstantin Khlebnikov 	 * Some filesystems seem to re-dirty the page even after
116b9ea2515SKonstantin Khlebnikov 	 * the VM has canceled the dirty bit (eg ext3 journaling).
117b9ea2515SKonstantin Khlebnikov 	 * Hence dirty accounting check is placed after invalidation.
118b9ea2515SKonstantin Khlebnikov 	 */
11911f81becSTejun Heo 	cancel_dirty_page(page);
1201da177e4SLinus Torvalds 	ClearPageMappedToDisk(page);
1215adc7b51SMinchan Kim 	delete_from_page_cache(page);
122750b4987SNick Piggin 	return 0;
1231da177e4SLinus Torvalds }
1241da177e4SLinus Torvalds 
1251da177e4SLinus Torvalds /*
126fc0ecff6SAndrew Morton  * This is for invalidate_mapping_pages().  That function can be called at
1271da177e4SLinus Torvalds  * any time, and is not supposed to throw away dirty pages.  But pages can
1280fd0e6b0SNick Piggin  * be marked dirty at any time too, so use remove_mapping which safely
1290fd0e6b0SNick Piggin  * discards clean, unused pages.
1301da177e4SLinus Torvalds  *
1311da177e4SLinus Torvalds  * Returns non-zero if the page was successfully invalidated.
1321da177e4SLinus Torvalds  */
1331da177e4SLinus Torvalds static int
1341da177e4SLinus Torvalds invalidate_complete_page(struct address_space *mapping, struct page *page)
1351da177e4SLinus Torvalds {
1360fd0e6b0SNick Piggin 	int ret;
1370fd0e6b0SNick Piggin 
1381da177e4SLinus Torvalds 	if (page->mapping != mapping)
1391da177e4SLinus Torvalds 		return 0;
1401da177e4SLinus Torvalds 
141266cf658SDavid Howells 	if (page_has_private(page) && !try_to_release_page(page, 0))
1421da177e4SLinus Torvalds 		return 0;
1431da177e4SLinus Torvalds 
1440fd0e6b0SNick Piggin 	ret = remove_mapping(mapping, page);
1450fd0e6b0SNick Piggin 
1460fd0e6b0SNick Piggin 	return ret;
1471da177e4SLinus Torvalds }
1481da177e4SLinus Torvalds 
149750b4987SNick Piggin int truncate_inode_page(struct address_space *mapping, struct page *page)
150750b4987SNick Piggin {
151750b4987SNick Piggin 	if (page_mapped(page)) {
152750b4987SNick Piggin 		unmap_mapping_range(mapping,
153750b4987SNick Piggin 				   (loff_t)page->index << PAGE_CACHE_SHIFT,
154750b4987SNick Piggin 				   PAGE_CACHE_SIZE, 0);
155750b4987SNick Piggin 	}
156750b4987SNick Piggin 	return truncate_complete_page(mapping, page);
157750b4987SNick Piggin }
158750b4987SNick Piggin 
15983f78668SWu Fengguang /*
16025718736SAndi Kleen  * Used to get rid of pages on hardware memory corruption.
16125718736SAndi Kleen  */
16225718736SAndi Kleen int generic_error_remove_page(struct address_space *mapping, struct page *page)
16325718736SAndi Kleen {
16425718736SAndi Kleen 	if (!mapping)
16525718736SAndi Kleen 		return -EINVAL;
16625718736SAndi Kleen 	/*
16725718736SAndi Kleen 	 * Only punch for normal data pages for now.
16825718736SAndi Kleen 	 * Handling other types like directories would need more auditing.
16925718736SAndi Kleen 	 */
17025718736SAndi Kleen 	if (!S_ISREG(mapping->host->i_mode))
17125718736SAndi Kleen 		return -EIO;
17225718736SAndi Kleen 	return truncate_inode_page(mapping, page);
17325718736SAndi Kleen }
17425718736SAndi Kleen EXPORT_SYMBOL(generic_error_remove_page);
17525718736SAndi Kleen 
17625718736SAndi Kleen /*
17783f78668SWu Fengguang  * Safely invalidate one page from its pagecache mapping.
17883f78668SWu Fengguang  * It only drops clean, unused pages. The page must be locked.
17983f78668SWu Fengguang  *
18083f78668SWu Fengguang  * Returns 1 if the page is successfully invalidated, otherwise 0.
18183f78668SWu Fengguang  */
18283f78668SWu Fengguang int invalidate_inode_page(struct page *page)
18383f78668SWu Fengguang {
18483f78668SWu Fengguang 	struct address_space *mapping = page_mapping(page);
18583f78668SWu Fengguang 	if (!mapping)
18683f78668SWu Fengguang 		return 0;
18783f78668SWu Fengguang 	if (PageDirty(page) || PageWriteback(page))
18883f78668SWu Fengguang 		return 0;
18983f78668SWu Fengguang 	if (page_mapped(page))
19083f78668SWu Fengguang 		return 0;
19183f78668SWu Fengguang 	return invalidate_complete_page(mapping, page);
19283f78668SWu Fengguang }
19383f78668SWu Fengguang 
1941da177e4SLinus Torvalds /**
19573c1e204SLiu Bo  * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
1961da177e4SLinus Torvalds  * @mapping: mapping to truncate
1971da177e4SLinus Torvalds  * @lstart: offset from which to truncate
1985a720394SLukas Czerner  * @lend: offset to which to truncate (inclusive)
1991da177e4SLinus Torvalds  *
200d7339071SHans Reiser  * Truncate the page cache, removing the pages that are between
2015a720394SLukas Czerner  * specified offsets (and zeroing out partial pages
2025a720394SLukas Czerner  * if lstart or lend + 1 is not page aligned).
2031da177e4SLinus Torvalds  *
2041da177e4SLinus Torvalds  * Truncate takes two passes - the first pass is nonblocking.  It will not
2051da177e4SLinus Torvalds  * block on page locks and it will not block on writeback.  The second pass
2061da177e4SLinus Torvalds  * will wait.  This is to prevent as much IO as possible in the affected region.
2071da177e4SLinus Torvalds  * The first pass will remove most pages, so the search cost of the second pass
2081da177e4SLinus Torvalds  * is low.
2091da177e4SLinus Torvalds  *
2101da177e4SLinus Torvalds  * We pass down the cache-hot hint to the page freeing code.  Even if the
2111da177e4SLinus Torvalds  * mapping is large, it is probably the case that the final pages are the most
2121da177e4SLinus Torvalds  * recently touched, and freeing happens in ascending file offset order.
2135a720394SLukas Czerner  *
2145a720394SLukas Czerner  * Note that since ->invalidatepage() accepts range to invalidate
2155a720394SLukas Czerner  * truncate_inode_pages_range is able to handle cases where lend + 1 is not
2165a720394SLukas Czerner  * page aligned properly.
2171da177e4SLinus Torvalds  */
218d7339071SHans Reiser void truncate_inode_pages_range(struct address_space *mapping,
219d7339071SHans Reiser 				loff_t lstart, loff_t lend)
2201da177e4SLinus Torvalds {
2215a720394SLukas Czerner 	pgoff_t		start;		/* inclusive */
2225a720394SLukas Czerner 	pgoff_t		end;		/* exclusive */
2235a720394SLukas Czerner 	unsigned int	partial_start;	/* inclusive */
2245a720394SLukas Czerner 	unsigned int	partial_end;	/* exclusive */
2251da177e4SLinus Torvalds 	struct pagevec	pvec;
2260cd6144aSJohannes Weiner 	pgoff_t		indices[PAGEVEC_SIZE];
227b85e0effSHugh Dickins 	pgoff_t		index;
2281da177e4SLinus Torvalds 	int		i;
2291da177e4SLinus Torvalds 
2303167760fSDan Magenheimer 	cleancache_invalidate_inode(mapping);
23191b0abe3SJohannes Weiner 	if (mapping->nrpages == 0 && mapping->nrshadows == 0)
2321da177e4SLinus Torvalds 		return;
2331da177e4SLinus Torvalds 
2345a720394SLukas Czerner 	/* Offsets within partial pages */
2355a720394SLukas Czerner 	partial_start = lstart & (PAGE_CACHE_SIZE - 1);
2365a720394SLukas Czerner 	partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
2375a720394SLukas Czerner 
2385a720394SLukas Czerner 	/*
2395a720394SLukas Czerner 	 * 'start' and 'end' always covers the range of pages to be fully
2405a720394SLukas Czerner 	 * truncated. Partial pages are covered with 'partial_start' at the
2415a720394SLukas Czerner 	 * start of the range and 'partial_end' at the end of the range.
2425a720394SLukas Czerner 	 * Note that 'end' is exclusive while 'lend' is inclusive.
2435a720394SLukas Czerner 	 */
2445a720394SLukas Czerner 	start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
2455a720394SLukas Czerner 	if (lend == -1)
2465a720394SLukas Czerner 		/*
2475a720394SLukas Czerner 		 * lend == -1 indicates end-of-file so we have to set 'end'
2485a720394SLukas Czerner 		 * to the highest possible pgoff_t and since the type is
2495a720394SLukas Czerner 		 * unsigned we're using -1.
2505a720394SLukas Czerner 		 */
2515a720394SLukas Czerner 		end = -1;
2525a720394SLukas Czerner 	else
2535a720394SLukas Czerner 		end = (lend + 1) >> PAGE_CACHE_SHIFT;
254d7339071SHans Reiser 
2551da177e4SLinus Torvalds 	pagevec_init(&pvec, 0);
256b85e0effSHugh Dickins 	index = start;
2570cd6144aSJohannes Weiner 	while (index < end && pagevec_lookup_entries(&pvec, mapping, index,
2580cd6144aSJohannes Weiner 			min(end - index, (pgoff_t)PAGEVEC_SIZE),
2590cd6144aSJohannes Weiner 			indices)) {
2601da177e4SLinus Torvalds 		for (i = 0; i < pagevec_count(&pvec); i++) {
2611da177e4SLinus Torvalds 			struct page *page = pvec.pages[i];
2621da177e4SLinus Torvalds 
263b85e0effSHugh Dickins 			/* We rely upon deletion not changing page->index */
2640cd6144aSJohannes Weiner 			index = indices[i];
2655a720394SLukas Czerner 			if (index >= end)
266d7339071SHans Reiser 				break;
267d7339071SHans Reiser 
2680cd6144aSJohannes Weiner 			if (radix_tree_exceptional_entry(page)) {
2690cd6144aSJohannes Weiner 				clear_exceptional_entry(mapping, index, page);
2700cd6144aSJohannes Weiner 				continue;
2710cd6144aSJohannes Weiner 			}
2720cd6144aSJohannes Weiner 
273529ae9aaSNick Piggin 			if (!trylock_page(page))
2741da177e4SLinus Torvalds 				continue;
275b85e0effSHugh Dickins 			WARN_ON(page->index != index);
2761da177e4SLinus Torvalds 			if (PageWriteback(page)) {
2771da177e4SLinus Torvalds 				unlock_page(page);
2781da177e4SLinus Torvalds 				continue;
2791da177e4SLinus Torvalds 			}
280750b4987SNick Piggin 			truncate_inode_page(mapping, page);
2811da177e4SLinus Torvalds 			unlock_page(page);
2821da177e4SLinus Torvalds 		}
2830cd6144aSJohannes Weiner 		pagevec_remove_exceptionals(&pvec);
2841da177e4SLinus Torvalds 		pagevec_release(&pvec);
2851da177e4SLinus Torvalds 		cond_resched();
286b85e0effSHugh Dickins 		index++;
2871da177e4SLinus Torvalds 	}
2881da177e4SLinus Torvalds 
2895a720394SLukas Czerner 	if (partial_start) {
2901da177e4SLinus Torvalds 		struct page *page = find_lock_page(mapping, start - 1);
2911da177e4SLinus Torvalds 		if (page) {
2925a720394SLukas Czerner 			unsigned int top = PAGE_CACHE_SIZE;
2935a720394SLukas Czerner 			if (start > end) {
2945a720394SLukas Czerner 				/* Truncation within a single page */
2955a720394SLukas Czerner 				top = partial_end;
2965a720394SLukas Czerner 				partial_end = 0;
2975a720394SLukas Czerner 			}
2981da177e4SLinus Torvalds 			wait_on_page_writeback(page);
2995a720394SLukas Czerner 			zero_user_segment(page, partial_start, top);
3005a720394SLukas Czerner 			cleancache_invalidate_page(mapping, page);
3015a720394SLukas Czerner 			if (page_has_private(page))
3025a720394SLukas Czerner 				do_invalidatepage(page, partial_start,
3035a720394SLukas Czerner 						  top - partial_start);
3041da177e4SLinus Torvalds 			unlock_page(page);
3051da177e4SLinus Torvalds 			page_cache_release(page);
3061da177e4SLinus Torvalds 		}
3071da177e4SLinus Torvalds 	}
3085a720394SLukas Czerner 	if (partial_end) {
3095a720394SLukas Czerner 		struct page *page = find_lock_page(mapping, end);
3105a720394SLukas Czerner 		if (page) {
3115a720394SLukas Czerner 			wait_on_page_writeback(page);
3125a720394SLukas Czerner 			zero_user_segment(page, 0, partial_end);
3135a720394SLukas Czerner 			cleancache_invalidate_page(mapping, page);
3145a720394SLukas Czerner 			if (page_has_private(page))
3155a720394SLukas Czerner 				do_invalidatepage(page, 0,
3165a720394SLukas Czerner 						  partial_end);
3175a720394SLukas Czerner 			unlock_page(page);
3185a720394SLukas Czerner 			page_cache_release(page);
3195a720394SLukas Czerner 		}
3205a720394SLukas Czerner 	}
3215a720394SLukas Czerner 	/*
3225a720394SLukas Czerner 	 * If the truncation happened within a single page no pages
3235a720394SLukas Czerner 	 * will be released, just zeroed, so we can bail out now.
3245a720394SLukas Czerner 	 */
3255a720394SLukas Czerner 	if (start >= end)
3265a720394SLukas Czerner 		return;
3271da177e4SLinus Torvalds 
328b85e0effSHugh Dickins 	index = start;
3291da177e4SLinus Torvalds 	for ( ; ; ) {
3301da177e4SLinus Torvalds 		cond_resched();
3310cd6144aSJohannes Weiner 		if (!pagevec_lookup_entries(&pvec, mapping, index,
332792ceaefSHugh Dickins 			min(end - index, (pgoff_t)PAGEVEC_SIZE), indices)) {
333792ceaefSHugh Dickins 			/* If all gone from start onwards, we're done */
334b85e0effSHugh Dickins 			if (index == start)
3351da177e4SLinus Torvalds 				break;
336792ceaefSHugh Dickins 			/* Otherwise restart to make sure all gone */
337b85e0effSHugh Dickins 			index = start;
3381da177e4SLinus Torvalds 			continue;
3391da177e4SLinus Torvalds 		}
3400cd6144aSJohannes Weiner 		if (index == start && indices[0] >= end) {
341792ceaefSHugh Dickins 			/* All gone out of hole to be punched, we're done */
3420cd6144aSJohannes Weiner 			pagevec_remove_exceptionals(&pvec);
343d7339071SHans Reiser 			pagevec_release(&pvec);
344d7339071SHans Reiser 			break;
345d7339071SHans Reiser 		}
3461da177e4SLinus Torvalds 		for (i = 0; i < pagevec_count(&pvec); i++) {
3471da177e4SLinus Torvalds 			struct page *page = pvec.pages[i];
3481da177e4SLinus Torvalds 
349b85e0effSHugh Dickins 			/* We rely upon deletion not changing page->index */
3500cd6144aSJohannes Weiner 			index = indices[i];
351792ceaefSHugh Dickins 			if (index >= end) {
352792ceaefSHugh Dickins 				/* Restart punch to make sure all gone */
353792ceaefSHugh Dickins 				index = start - 1;
354d7339071SHans Reiser 				break;
355792ceaefSHugh Dickins 			}
356b85e0effSHugh Dickins 
3570cd6144aSJohannes Weiner 			if (radix_tree_exceptional_entry(page)) {
3580cd6144aSJohannes Weiner 				clear_exceptional_entry(mapping, index, page);
3590cd6144aSJohannes Weiner 				continue;
3600cd6144aSJohannes Weiner 			}
3610cd6144aSJohannes Weiner 
3621da177e4SLinus Torvalds 			lock_page(page);
363b85e0effSHugh Dickins 			WARN_ON(page->index != index);
3641da177e4SLinus Torvalds 			wait_on_page_writeback(page);
365750b4987SNick Piggin 			truncate_inode_page(mapping, page);
3661da177e4SLinus Torvalds 			unlock_page(page);
3671da177e4SLinus Torvalds 		}
3680cd6144aSJohannes Weiner 		pagevec_remove_exceptionals(&pvec);
3691da177e4SLinus Torvalds 		pagevec_release(&pvec);
370b85e0effSHugh Dickins 		index++;
3711da177e4SLinus Torvalds 	}
3723167760fSDan Magenheimer 	cleancache_invalidate_inode(mapping);
3731da177e4SLinus Torvalds }
374d7339071SHans Reiser EXPORT_SYMBOL(truncate_inode_pages_range);
3751da177e4SLinus Torvalds 
376d7339071SHans Reiser /**
377d7339071SHans Reiser  * truncate_inode_pages - truncate *all* the pages from an offset
378d7339071SHans Reiser  * @mapping: mapping to truncate
379d7339071SHans Reiser  * @lstart: offset from which to truncate
380d7339071SHans Reiser  *
3811b1dcc1bSJes Sorensen  * Called under (and serialised by) inode->i_mutex.
38208142579SJan Kara  *
38308142579SJan Kara  * Note: When this function returns, there can be a page in the process of
38408142579SJan Kara  * deletion (inside __delete_from_page_cache()) in the specified range.  Thus
38508142579SJan Kara  * mapping->nrpages can be non-zero when this function returns even after
38608142579SJan Kara  * truncation of the whole mapping.
387d7339071SHans Reiser  */
388d7339071SHans Reiser void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
389d7339071SHans Reiser {
390d7339071SHans Reiser 	truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
391d7339071SHans Reiser }
3921da177e4SLinus Torvalds EXPORT_SYMBOL(truncate_inode_pages);
3931da177e4SLinus Torvalds 
39428697355SMike Waychison /**
39591b0abe3SJohannes Weiner  * truncate_inode_pages_final - truncate *all* pages before inode dies
39691b0abe3SJohannes Weiner  * @mapping: mapping to truncate
39791b0abe3SJohannes Weiner  *
39891b0abe3SJohannes Weiner  * Called under (and serialized by) inode->i_mutex.
39991b0abe3SJohannes Weiner  *
40091b0abe3SJohannes Weiner  * Filesystems have to use this in the .evict_inode path to inform the
40191b0abe3SJohannes Weiner  * VM that this is the final truncate and the inode is going away.
40291b0abe3SJohannes Weiner  */
40391b0abe3SJohannes Weiner void truncate_inode_pages_final(struct address_space *mapping)
40491b0abe3SJohannes Weiner {
40591b0abe3SJohannes Weiner 	unsigned long nrshadows;
40691b0abe3SJohannes Weiner 	unsigned long nrpages;
40791b0abe3SJohannes Weiner 
40891b0abe3SJohannes Weiner 	/*
40991b0abe3SJohannes Weiner 	 * Page reclaim can not participate in regular inode lifetime
41091b0abe3SJohannes Weiner 	 * management (can't call iput()) and thus can race with the
41191b0abe3SJohannes Weiner 	 * inode teardown.  Tell it when the address space is exiting,
41291b0abe3SJohannes Weiner 	 * so that it does not install eviction information after the
41391b0abe3SJohannes Weiner 	 * final truncate has begun.
41491b0abe3SJohannes Weiner 	 */
41591b0abe3SJohannes Weiner 	mapping_set_exiting(mapping);
41691b0abe3SJohannes Weiner 
41791b0abe3SJohannes Weiner 	/*
41891b0abe3SJohannes Weiner 	 * When reclaim installs eviction entries, it increases
41991b0abe3SJohannes Weiner 	 * nrshadows first, then decreases nrpages.  Make sure we see
42091b0abe3SJohannes Weiner 	 * this in the right order or we might miss an entry.
42191b0abe3SJohannes Weiner 	 */
42291b0abe3SJohannes Weiner 	nrpages = mapping->nrpages;
42391b0abe3SJohannes Weiner 	smp_rmb();
42491b0abe3SJohannes Weiner 	nrshadows = mapping->nrshadows;
42591b0abe3SJohannes Weiner 
42691b0abe3SJohannes Weiner 	if (nrpages || nrshadows) {
42791b0abe3SJohannes Weiner 		/*
42891b0abe3SJohannes Weiner 		 * As truncation uses a lockless tree lookup, cycle
42991b0abe3SJohannes Weiner 		 * the tree lock to make sure any ongoing tree
43091b0abe3SJohannes Weiner 		 * modification that does not see AS_EXITING is
43191b0abe3SJohannes Weiner 		 * completed before starting the final truncate.
43291b0abe3SJohannes Weiner 		 */
43391b0abe3SJohannes Weiner 		spin_lock_irq(&mapping->tree_lock);
43491b0abe3SJohannes Weiner 		spin_unlock_irq(&mapping->tree_lock);
43591b0abe3SJohannes Weiner 
43691b0abe3SJohannes Weiner 		truncate_inode_pages(mapping, 0);
43791b0abe3SJohannes Weiner 	}
43891b0abe3SJohannes Weiner }
43991b0abe3SJohannes Weiner EXPORT_SYMBOL(truncate_inode_pages_final);
44091b0abe3SJohannes Weiner 
44191b0abe3SJohannes Weiner /**
44228697355SMike Waychison  * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
44328697355SMike Waychison  * @mapping: the address_space which holds the pages to invalidate
44428697355SMike Waychison  * @start: the offset 'from' which to invalidate
44528697355SMike Waychison  * @end: the offset 'to' which to invalidate (inclusive)
44628697355SMike Waychison  *
44728697355SMike Waychison  * This function only removes the unlocked pages, if you want to
44828697355SMike Waychison  * remove all the pages of one inode, you must call truncate_inode_pages.
44928697355SMike Waychison  *
45028697355SMike Waychison  * invalidate_mapping_pages() will not block on IO activity. It will not
45128697355SMike Waychison  * invalidate pages which are dirty, locked, under writeback or mapped into
45228697355SMike Waychison  * pagetables.
45328697355SMike Waychison  */
45428697355SMike Waychison unsigned long invalidate_mapping_pages(struct address_space *mapping,
45528697355SMike Waychison 		pgoff_t start, pgoff_t end)
4561da177e4SLinus Torvalds {
4570cd6144aSJohannes Weiner 	pgoff_t indices[PAGEVEC_SIZE];
4581da177e4SLinus Torvalds 	struct pagevec pvec;
459b85e0effSHugh Dickins 	pgoff_t index = start;
46031560180SMinchan Kim 	unsigned long ret;
46131560180SMinchan Kim 	unsigned long count = 0;
4621da177e4SLinus Torvalds 	int i;
4631da177e4SLinus Torvalds 
4641da177e4SLinus Torvalds 	pagevec_init(&pvec, 0);
4650cd6144aSJohannes Weiner 	while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
4660cd6144aSJohannes Weiner 			min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
4670cd6144aSJohannes Weiner 			indices)) {
4681da177e4SLinus Torvalds 		for (i = 0; i < pagevec_count(&pvec); i++) {
4691da177e4SLinus Torvalds 			struct page *page = pvec.pages[i];
4701da177e4SLinus Torvalds 
471b85e0effSHugh Dickins 			/* We rely upon deletion not changing page->index */
4720cd6144aSJohannes Weiner 			index = indices[i];
473b85e0effSHugh Dickins 			if (index > end)
474b85e0effSHugh Dickins 				break;
475e0f23603SNeilBrown 
4760cd6144aSJohannes Weiner 			if (radix_tree_exceptional_entry(page)) {
4770cd6144aSJohannes Weiner 				clear_exceptional_entry(mapping, index, page);
4780cd6144aSJohannes Weiner 				continue;
4790cd6144aSJohannes Weiner 			}
4800cd6144aSJohannes Weiner 
481b85e0effSHugh Dickins 			if (!trylock_page(page))
482b85e0effSHugh Dickins 				continue;
483b85e0effSHugh Dickins 			WARN_ON(page->index != index);
48431560180SMinchan Kim 			ret = invalidate_inode_page(page);
4851da177e4SLinus Torvalds 			unlock_page(page);
48631560180SMinchan Kim 			/*
48731560180SMinchan Kim 			 * Invalidation is a hint that the page is no longer
48831560180SMinchan Kim 			 * of interest and try to speed up its reclaim.
48931560180SMinchan Kim 			 */
49031560180SMinchan Kim 			if (!ret)
491cc5993bdSMinchan Kim 				deactivate_file_page(page);
49231560180SMinchan Kim 			count += ret;
4931da177e4SLinus Torvalds 		}
4940cd6144aSJohannes Weiner 		pagevec_remove_exceptionals(&pvec);
4951da177e4SLinus Torvalds 		pagevec_release(&pvec);
496fc9a07e7SAndrew Morton 		cond_resched();
497b85e0effSHugh Dickins 		index++;
4981da177e4SLinus Torvalds 	}
49931560180SMinchan Kim 	return count;
5001da177e4SLinus Torvalds }
50154bc4855SAnton Altaparmakov EXPORT_SYMBOL(invalidate_mapping_pages);
5021da177e4SLinus Torvalds 
503bd4c8ce4SAndrew Morton /*
504bd4c8ce4SAndrew Morton  * This is like invalidate_complete_page(), except it ignores the page's
505bd4c8ce4SAndrew Morton  * refcount.  We do this because invalidate_inode_pages2() needs stronger
506bd4c8ce4SAndrew Morton  * invalidation guarantees, and cannot afford to leave pages behind because
5072706a1b8SAnderson Briglia  * shrink_page_list() has a temp ref on them, or because they're transiently
5082706a1b8SAnderson Briglia  * sitting in the lru_cache_add() pagevecs.
509bd4c8ce4SAndrew Morton  */
510bd4c8ce4SAndrew Morton static int
511bd4c8ce4SAndrew Morton invalidate_complete_page2(struct address_space *mapping, struct page *page)
512bd4c8ce4SAndrew Morton {
513*c4843a75SGreg Thelen 	struct mem_cgroup *memcg;
514*c4843a75SGreg Thelen 	unsigned long flags;
515*c4843a75SGreg Thelen 
516bd4c8ce4SAndrew Morton 	if (page->mapping != mapping)
517bd4c8ce4SAndrew Morton 		return 0;
518bd4c8ce4SAndrew Morton 
519266cf658SDavid Howells 	if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
520bd4c8ce4SAndrew Morton 		return 0;
521bd4c8ce4SAndrew Morton 
522*c4843a75SGreg Thelen 	memcg = mem_cgroup_begin_page_stat(page);
523*c4843a75SGreg Thelen 	spin_lock_irqsave(&mapping->tree_lock, flags);
524bd4c8ce4SAndrew Morton 	if (PageDirty(page))
525bd4c8ce4SAndrew Morton 		goto failed;
526bd4c8ce4SAndrew Morton 
527266cf658SDavid Howells 	BUG_ON(page_has_private(page));
528*c4843a75SGreg Thelen 	__delete_from_page_cache(page, NULL, memcg);
529*c4843a75SGreg Thelen 	spin_unlock_irqrestore(&mapping->tree_lock, flags);
530*c4843a75SGreg Thelen 	mem_cgroup_end_page_stat(memcg);
5316072d13cSLinus Torvalds 
5326072d13cSLinus Torvalds 	if (mapping->a_ops->freepage)
5336072d13cSLinus Torvalds 		mapping->a_ops->freepage(page);
5346072d13cSLinus Torvalds 
535bd4c8ce4SAndrew Morton 	page_cache_release(page);	/* pagecache ref */
536bd4c8ce4SAndrew Morton 	return 1;
537bd4c8ce4SAndrew Morton failed:
538*c4843a75SGreg Thelen 	spin_unlock_irqrestore(&mapping->tree_lock, flags);
539*c4843a75SGreg Thelen 	mem_cgroup_end_page_stat(memcg);
540bd4c8ce4SAndrew Morton 	return 0;
541bd4c8ce4SAndrew Morton }
542bd4c8ce4SAndrew Morton 
543e3db7691STrond Myklebust static int do_launder_page(struct address_space *mapping, struct page *page)
544e3db7691STrond Myklebust {
545e3db7691STrond Myklebust 	if (!PageDirty(page))
546e3db7691STrond Myklebust 		return 0;
547e3db7691STrond Myklebust 	if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
548e3db7691STrond Myklebust 		return 0;
549e3db7691STrond Myklebust 	return mapping->a_ops->launder_page(page);
550e3db7691STrond Myklebust }
551e3db7691STrond Myklebust 
5521da177e4SLinus Torvalds /**
5531da177e4SLinus Torvalds  * invalidate_inode_pages2_range - remove range of pages from an address_space
55467be2dd1SMartin Waitz  * @mapping: the address_space
5551da177e4SLinus Torvalds  * @start: the page offset 'from' which to invalidate
5561da177e4SLinus Torvalds  * @end: the page offset 'to' which to invalidate (inclusive)
5571da177e4SLinus Torvalds  *
5581da177e4SLinus Torvalds  * Any pages which are found to be mapped into pagetables are unmapped prior to
5591da177e4SLinus Torvalds  * invalidation.
5601da177e4SLinus Torvalds  *
5616ccfa806SHisashi Hifumi  * Returns -EBUSY if any pages could not be invalidated.
5621da177e4SLinus Torvalds  */
5631da177e4SLinus Torvalds int invalidate_inode_pages2_range(struct address_space *mapping,
5641da177e4SLinus Torvalds 				  pgoff_t start, pgoff_t end)
5651da177e4SLinus Torvalds {
5660cd6144aSJohannes Weiner 	pgoff_t indices[PAGEVEC_SIZE];
5671da177e4SLinus Torvalds 	struct pagevec pvec;
568b85e0effSHugh Dickins 	pgoff_t index;
5691da177e4SLinus Torvalds 	int i;
5701da177e4SLinus Torvalds 	int ret = 0;
5710dd1334fSHisashi Hifumi 	int ret2 = 0;
5721da177e4SLinus Torvalds 	int did_range_unmap = 0;
5731da177e4SLinus Torvalds 
5743167760fSDan Magenheimer 	cleancache_invalidate_inode(mapping);
5751da177e4SLinus Torvalds 	pagevec_init(&pvec, 0);
576b85e0effSHugh Dickins 	index = start;
5770cd6144aSJohannes Weiner 	while (index <= end && pagevec_lookup_entries(&pvec, mapping, index,
5780cd6144aSJohannes Weiner 			min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
5790cd6144aSJohannes Weiner 			indices)) {
5807b965e08STrond Myklebust 		for (i = 0; i < pagevec_count(&pvec); i++) {
5811da177e4SLinus Torvalds 			struct page *page = pvec.pages[i];
582b85e0effSHugh Dickins 
583b85e0effSHugh Dickins 			/* We rely upon deletion not changing page->index */
5840cd6144aSJohannes Weiner 			index = indices[i];
585b85e0effSHugh Dickins 			if (index > end)
586b85e0effSHugh Dickins 				break;
5871da177e4SLinus Torvalds 
5880cd6144aSJohannes Weiner 			if (radix_tree_exceptional_entry(page)) {
5890cd6144aSJohannes Weiner 				clear_exceptional_entry(mapping, index, page);
5900cd6144aSJohannes Weiner 				continue;
5910cd6144aSJohannes Weiner 			}
5920cd6144aSJohannes Weiner 
5931da177e4SLinus Torvalds 			lock_page(page);
594b85e0effSHugh Dickins 			WARN_ON(page->index != index);
5951da177e4SLinus Torvalds 			if (page->mapping != mapping) {
5961da177e4SLinus Torvalds 				unlock_page(page);
5971da177e4SLinus Torvalds 				continue;
5981da177e4SLinus Torvalds 			}
5991da177e4SLinus Torvalds 			wait_on_page_writeback(page);
600d00806b1SNick Piggin 			if (page_mapped(page)) {
6011da177e4SLinus Torvalds 				if (!did_range_unmap) {
6021da177e4SLinus Torvalds 					/*
6031da177e4SLinus Torvalds 					 * Zap the rest of the file in one hit.
6041da177e4SLinus Torvalds 					 */
6051da177e4SLinus Torvalds 					unmap_mapping_range(mapping,
606b85e0effSHugh Dickins 					   (loff_t)index << PAGE_CACHE_SHIFT,
607b85e0effSHugh Dickins 					   (loff_t)(1 + end - index)
6081da177e4SLinus Torvalds 							 << PAGE_CACHE_SHIFT,
6091da177e4SLinus Torvalds 					    0);
6101da177e4SLinus Torvalds 					did_range_unmap = 1;
6111da177e4SLinus Torvalds 				} else {
6121da177e4SLinus Torvalds 					/*
6131da177e4SLinus Torvalds 					 * Just zap this page
6141da177e4SLinus Torvalds 					 */
6151da177e4SLinus Torvalds 					unmap_mapping_range(mapping,
616b85e0effSHugh Dickins 					   (loff_t)index << PAGE_CACHE_SHIFT,
6171da177e4SLinus Torvalds 					   PAGE_CACHE_SIZE, 0);
6181da177e4SLinus Torvalds 				}
6191da177e4SLinus Torvalds 			}
620d00806b1SNick Piggin 			BUG_ON(page_mapped(page));
6210dd1334fSHisashi Hifumi 			ret2 = do_launder_page(mapping, page);
6220dd1334fSHisashi Hifumi 			if (ret2 == 0) {
6230dd1334fSHisashi Hifumi 				if (!invalidate_complete_page2(mapping, page))
6246ccfa806SHisashi Hifumi 					ret2 = -EBUSY;
6250dd1334fSHisashi Hifumi 			}
6260dd1334fSHisashi Hifumi 			if (ret2 < 0)
6270dd1334fSHisashi Hifumi 				ret = ret2;
6281da177e4SLinus Torvalds 			unlock_page(page);
6291da177e4SLinus Torvalds 		}
6300cd6144aSJohannes Weiner 		pagevec_remove_exceptionals(&pvec);
6311da177e4SLinus Torvalds 		pagevec_release(&pvec);
6321da177e4SLinus Torvalds 		cond_resched();
633b85e0effSHugh Dickins 		index++;
6341da177e4SLinus Torvalds 	}
6353167760fSDan Magenheimer 	cleancache_invalidate_inode(mapping);
6361da177e4SLinus Torvalds 	return ret;
6371da177e4SLinus Torvalds }
6381da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
6391da177e4SLinus Torvalds 
6401da177e4SLinus Torvalds /**
6411da177e4SLinus Torvalds  * invalidate_inode_pages2 - remove all pages from an address_space
64267be2dd1SMartin Waitz  * @mapping: the address_space
6431da177e4SLinus Torvalds  *
6441da177e4SLinus Torvalds  * Any pages which are found to be mapped into pagetables are unmapped prior to
6451da177e4SLinus Torvalds  * invalidation.
6461da177e4SLinus Torvalds  *
647e9de25ddSPeng Tao  * Returns -EBUSY if any pages could not be invalidated.
6481da177e4SLinus Torvalds  */
6491da177e4SLinus Torvalds int invalidate_inode_pages2(struct address_space *mapping)
6501da177e4SLinus Torvalds {
6511da177e4SLinus Torvalds 	return invalidate_inode_pages2_range(mapping, 0, -1);
6521da177e4SLinus Torvalds }
6531da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
65425d9e2d1Snpiggin@suse.de 
65525d9e2d1Snpiggin@suse.de /**
65625d9e2d1Snpiggin@suse.de  * truncate_pagecache - unmap and remove pagecache that has been truncated
65725d9e2d1Snpiggin@suse.de  * @inode: inode
6588a549beaSHugh Dickins  * @newsize: new file size
65925d9e2d1Snpiggin@suse.de  *
66025d9e2d1Snpiggin@suse.de  * inode's new i_size must already be written before truncate_pagecache
66125d9e2d1Snpiggin@suse.de  * is called.
66225d9e2d1Snpiggin@suse.de  *
66325d9e2d1Snpiggin@suse.de  * This function should typically be called before the filesystem
66425d9e2d1Snpiggin@suse.de  * releases resources associated with the freed range (eg. deallocates
66525d9e2d1Snpiggin@suse.de  * blocks). This way, pagecache will always stay logically coherent
66625d9e2d1Snpiggin@suse.de  * with on-disk format, and the filesystem would not have to deal with
66725d9e2d1Snpiggin@suse.de  * situations such as writepage being called for a page that has already
66825d9e2d1Snpiggin@suse.de  * had its underlying blocks deallocated.
66925d9e2d1Snpiggin@suse.de  */
6707caef267SKirill A. Shutemov void truncate_pagecache(struct inode *inode, loff_t newsize)
67125d9e2d1Snpiggin@suse.de {
67225d9e2d1Snpiggin@suse.de 	struct address_space *mapping = inode->i_mapping;
6738a549beaSHugh Dickins 	loff_t holebegin = round_up(newsize, PAGE_SIZE);
67425d9e2d1Snpiggin@suse.de 
67525d9e2d1Snpiggin@suse.de 	/*
67625d9e2d1Snpiggin@suse.de 	 * unmap_mapping_range is called twice, first simply for
67725d9e2d1Snpiggin@suse.de 	 * efficiency so that truncate_inode_pages does fewer
67825d9e2d1Snpiggin@suse.de 	 * single-page unmaps.  However after this first call, and
67925d9e2d1Snpiggin@suse.de 	 * before truncate_inode_pages finishes, it is possible for
68025d9e2d1Snpiggin@suse.de 	 * private pages to be COWed, which remain after
68125d9e2d1Snpiggin@suse.de 	 * truncate_inode_pages finishes, hence the second
68225d9e2d1Snpiggin@suse.de 	 * unmap_mapping_range call must be made for correctness.
68325d9e2d1Snpiggin@suse.de 	 */
6848a549beaSHugh Dickins 	unmap_mapping_range(mapping, holebegin, 0, 1);
6858a549beaSHugh Dickins 	truncate_inode_pages(mapping, newsize);
6868a549beaSHugh Dickins 	unmap_mapping_range(mapping, holebegin, 0, 1);
68725d9e2d1Snpiggin@suse.de }
68825d9e2d1Snpiggin@suse.de EXPORT_SYMBOL(truncate_pagecache);
68925d9e2d1Snpiggin@suse.de 
69025d9e2d1Snpiggin@suse.de /**
6912c27c65eSChristoph Hellwig  * truncate_setsize - update inode and pagecache for a new file size
6922c27c65eSChristoph Hellwig  * @inode: inode
6932c27c65eSChristoph Hellwig  * @newsize: new file size
6942c27c65eSChristoph Hellwig  *
695382e27daSJan Kara  * truncate_setsize updates i_size and performs pagecache truncation (if
696382e27daSJan Kara  * necessary) to @newsize. It will be typically be called from the filesystem's
697382e27daSJan Kara  * setattr function when ATTR_SIZE is passed in.
6982c27c65eSChristoph Hellwig  *
69977783d06SJan Kara  * Must be called with a lock serializing truncates and writes (generally
70077783d06SJan Kara  * i_mutex but e.g. xfs uses a different lock) and before all filesystem
70177783d06SJan Kara  * specific block truncation has been performed.
7022c27c65eSChristoph Hellwig  */
7032c27c65eSChristoph Hellwig void truncate_setsize(struct inode *inode, loff_t newsize)
7042c27c65eSChristoph Hellwig {
70590a80202SJan Kara 	loff_t oldsize = inode->i_size;
70690a80202SJan Kara 
7072c27c65eSChristoph Hellwig 	i_size_write(inode, newsize);
70890a80202SJan Kara 	if (newsize > oldsize)
70990a80202SJan Kara 		pagecache_isize_extended(inode, oldsize, newsize);
7107caef267SKirill A. Shutemov 	truncate_pagecache(inode, newsize);
7112c27c65eSChristoph Hellwig }
7122c27c65eSChristoph Hellwig EXPORT_SYMBOL(truncate_setsize);
7132c27c65eSChristoph Hellwig 
7142c27c65eSChristoph Hellwig /**
71590a80202SJan Kara  * pagecache_isize_extended - update pagecache after extension of i_size
71690a80202SJan Kara  * @inode:	inode for which i_size was extended
71790a80202SJan Kara  * @from:	original inode size
71890a80202SJan Kara  * @to:		new inode size
71990a80202SJan Kara  *
72090a80202SJan Kara  * Handle extension of inode size either caused by extending truncate or by
72190a80202SJan Kara  * write starting after current i_size. We mark the page straddling current
72290a80202SJan Kara  * i_size RO so that page_mkwrite() is called on the nearest write access to
72390a80202SJan Kara  * the page.  This way filesystem can be sure that page_mkwrite() is called on
72490a80202SJan Kara  * the page before user writes to the page via mmap after the i_size has been
72590a80202SJan Kara  * changed.
72690a80202SJan Kara  *
72790a80202SJan Kara  * The function must be called after i_size is updated so that page fault
72890a80202SJan Kara  * coming after we unlock the page will already see the new i_size.
72990a80202SJan Kara  * The function must be called while we still hold i_mutex - this not only
73090a80202SJan Kara  * makes sure i_size is stable but also that userspace cannot observe new
73190a80202SJan Kara  * i_size value before we are prepared to store mmap writes at new inode size.
73290a80202SJan Kara  */
73390a80202SJan Kara void pagecache_isize_extended(struct inode *inode, loff_t from, loff_t to)
73490a80202SJan Kara {
73590a80202SJan Kara 	int bsize = 1 << inode->i_blkbits;
73690a80202SJan Kara 	loff_t rounded_from;
73790a80202SJan Kara 	struct page *page;
73890a80202SJan Kara 	pgoff_t index;
73990a80202SJan Kara 
74090a80202SJan Kara 	WARN_ON(to > inode->i_size);
74190a80202SJan Kara 
74290a80202SJan Kara 	if (from >= to || bsize == PAGE_CACHE_SIZE)
74390a80202SJan Kara 		return;
74490a80202SJan Kara 	/* Page straddling @from will not have any hole block created? */
74590a80202SJan Kara 	rounded_from = round_up(from, bsize);
74690a80202SJan Kara 	if (to <= rounded_from || !(rounded_from & (PAGE_CACHE_SIZE - 1)))
74790a80202SJan Kara 		return;
74890a80202SJan Kara 
74990a80202SJan Kara 	index = from >> PAGE_CACHE_SHIFT;
75090a80202SJan Kara 	page = find_lock_page(inode->i_mapping, index);
75190a80202SJan Kara 	/* Page not cached? Nothing to do */
75290a80202SJan Kara 	if (!page)
75390a80202SJan Kara 		return;
75490a80202SJan Kara 	/*
75590a80202SJan Kara 	 * See clear_page_dirty_for_io() for details why set_page_dirty()
75690a80202SJan Kara 	 * is needed.
75790a80202SJan Kara 	 */
75890a80202SJan Kara 	if (page_mkclean(page))
75990a80202SJan Kara 		set_page_dirty(page);
76090a80202SJan Kara 	unlock_page(page);
76190a80202SJan Kara 	page_cache_release(page);
76290a80202SJan Kara }
76390a80202SJan Kara EXPORT_SYMBOL(pagecache_isize_extended);
76490a80202SJan Kara 
76590a80202SJan Kara /**
766623e3db9SHugh Dickins  * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
767623e3db9SHugh Dickins  * @inode: inode
768623e3db9SHugh Dickins  * @lstart: offset of beginning of hole
769623e3db9SHugh Dickins  * @lend: offset of last byte of hole
770623e3db9SHugh Dickins  *
771623e3db9SHugh Dickins  * This function should typically be called before the filesystem
772623e3db9SHugh Dickins  * releases resources associated with the freed range (eg. deallocates
773623e3db9SHugh Dickins  * blocks). This way, pagecache will always stay logically coherent
774623e3db9SHugh Dickins  * with on-disk format, and the filesystem would not have to deal with
775623e3db9SHugh Dickins  * situations such as writepage being called for a page that has already
776623e3db9SHugh Dickins  * had its underlying blocks deallocated.
777623e3db9SHugh Dickins  */
778623e3db9SHugh Dickins void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
779623e3db9SHugh Dickins {
780623e3db9SHugh Dickins 	struct address_space *mapping = inode->i_mapping;
781623e3db9SHugh Dickins 	loff_t unmap_start = round_up(lstart, PAGE_SIZE);
782623e3db9SHugh Dickins 	loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
783623e3db9SHugh Dickins 	/*
784623e3db9SHugh Dickins 	 * This rounding is currently just for example: unmap_mapping_range
785623e3db9SHugh Dickins 	 * expands its hole outwards, whereas we want it to contract the hole
786623e3db9SHugh Dickins 	 * inwards.  However, existing callers of truncate_pagecache_range are
7875a720394SLukas Czerner 	 * doing their own page rounding first.  Note that unmap_mapping_range
7885a720394SLukas Czerner 	 * allows holelen 0 for all, and we allow lend -1 for end of file.
789623e3db9SHugh Dickins 	 */
790623e3db9SHugh Dickins 
791623e3db9SHugh Dickins 	/*
792623e3db9SHugh Dickins 	 * Unlike in truncate_pagecache, unmap_mapping_range is called only
793623e3db9SHugh Dickins 	 * once (before truncating pagecache), and without "even_cows" flag:
794623e3db9SHugh Dickins 	 * hole-punching should not remove private COWed pages from the hole.
795623e3db9SHugh Dickins 	 */
796623e3db9SHugh Dickins 	if ((u64)unmap_end > (u64)unmap_start)
797623e3db9SHugh Dickins 		unmap_mapping_range(mapping, unmap_start,
798623e3db9SHugh Dickins 				    1 + unmap_end - unmap_start, 0);
799623e3db9SHugh Dickins 	truncate_inode_pages_range(mapping, lstart, lend);
800623e3db9SHugh Dickins }
801623e3db9SHugh Dickins EXPORT_SYMBOL(truncate_pagecache_range);
802