Root/mm/truncate.c

1/*
2 * mm/truncate.c - code for taking down pages from address_spaces
3 *
4 * Copyright (C) 2002, Linus Torvalds
5 *
6 * 10Sep2002 Andrew Morton
7 * Initial version.
8 */
9
10#include <linux/kernel.h>
11#include <linux/backing-dev.h>
12#include <linux/gfp.h>
13#include <linux/mm.h>
14#include <linux/swap.h>
15#include <linux/export.h>
16#include <linux/pagemap.h>
17#include <linux/highmem.h>
18#include <linux/pagevec.h>
19#include <linux/task_io_accounting_ops.h>
20#include <linux/buffer_head.h> /* grr. try_to_release_page,
21                   do_invalidatepage */
22#include <linux/cleancache.h>
23#include "internal.h"
24
25
26/**
27 * do_invalidatepage - invalidate part or all of a page
28 * @page: the page which is affected
29 * @offset: start of the range to invalidate
30 * @length: length of the range to invalidate
31 *
32 * do_invalidatepage() is called when all or part of the page has become
33 * invalidated by a truncate operation.
34 *
35 * do_invalidatepage() does not have to release all buffers, but it must
36 * ensure that no dirty buffer is left outside @offset and that no I/O
37 * is underway against any of the blocks which are outside the truncation
38 * point. Because the caller is about to free (and possibly reuse) those
39 * blocks on-disk.
40 */
41void do_invalidatepage(struct page *page, unsigned int offset,
42               unsigned int length)
43{
44    void (*invalidatepage)(struct page *, unsigned int, unsigned int);
45
46    invalidatepage = page->mapping->a_ops->invalidatepage;
47#ifdef CONFIG_BLOCK
48    if (!invalidatepage)
49        invalidatepage = block_invalidatepage;
50#endif
51    if (invalidatepage)
52        (*invalidatepage)(page, offset, length);
53}
54
55/*
56 * This cancels just the dirty bit on the kernel page itself, it
57 * does NOT actually remove dirty bits on any mmap's that may be
58 * around. It also leaves the page tagged dirty, so any sync
59 * activity will still find it on the dirty lists, and in particular,
60 * clear_page_dirty_for_io() will still look at the dirty bits in
61 * the VM.
62 *
63 * Doing this should *normally* only ever be done when a page
64 * is truncated, and is not actually mapped anywhere at all. However,
65 * fs/buffer.c does this when it notices that somebody has cleaned
66 * out all the buffers on a page without actually doing it through
67 * the VM. Can you say "ext3 is horribly ugly"? Tought you could.
68 */
69void cancel_dirty_page(struct page *page, unsigned int account_size)
70{
71    if (TestClearPageDirty(page)) {
72        struct address_space *mapping = page->mapping;
73        if (mapping && mapping_cap_account_dirty(mapping)) {
74            dec_zone_page_state(page, NR_FILE_DIRTY);
75            dec_bdi_stat(mapping->backing_dev_info,
76                    BDI_RECLAIMABLE);
77            if (account_size)
78                task_io_account_cancelled_write(account_size);
79        }
80    }
81}
82EXPORT_SYMBOL(cancel_dirty_page);
83
84/*
85 * If truncate cannot remove the fs-private metadata from the page, the page
86 * becomes orphaned. It will be left on the LRU and may even be mapped into
87 * user pagetables if we're racing with filemap_fault().
88 *
89 * We need to bale out if page->mapping is no longer equal to the original
90 * mapping. This happens a) when the VM reclaimed the page while we waited on
91 * its lock, b) when a concurrent invalidate_mapping_pages got there first and
92 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
93 */
94static int
95truncate_complete_page(struct address_space *mapping, struct page *page)
96{
97    if (page->mapping != mapping)
98        return -EIO;
99
100    if (page_has_private(page))
101        do_invalidatepage(page, 0, PAGE_CACHE_SIZE);
102
103    cancel_dirty_page(page, PAGE_CACHE_SIZE);
104
105    ClearPageMappedToDisk(page);
106    delete_from_page_cache(page);
107    return 0;
108}
109
110/*
111 * This is for invalidate_mapping_pages(). That function can be called at
112 * any time, and is not supposed to throw away dirty pages. But pages can
113 * be marked dirty at any time too, so use remove_mapping which safely
114 * discards clean, unused pages.
115 *
116 * Returns non-zero if the page was successfully invalidated.
117 */
118static int
119invalidate_complete_page(struct address_space *mapping, struct page *page)
120{
121    int ret;
122
123    if (page->mapping != mapping)
124        return 0;
125
126    if (page_has_private(page) && !try_to_release_page(page, 0))
127        return 0;
128
129    ret = remove_mapping(mapping, page);
130
131    return ret;
132}
133
134int truncate_inode_page(struct address_space *mapping, struct page *page)
135{
136    if (page_mapped(page)) {
137        unmap_mapping_range(mapping,
138                   (loff_t)page->index << PAGE_CACHE_SHIFT,
139                   PAGE_CACHE_SIZE, 0);
140    }
141    return truncate_complete_page(mapping, page);
142}
143
144/*
145 * Used to get rid of pages on hardware memory corruption.
146 */
147int generic_error_remove_page(struct address_space *mapping, struct page *page)
148{
149    if (!mapping)
150        return -EINVAL;
151    /*
152     * Only punch for normal data pages for now.
153     * Handling other types like directories would need more auditing.
154     */
155    if (!S_ISREG(mapping->host->i_mode))
156        return -EIO;
157    return truncate_inode_page(mapping, page);
158}
159EXPORT_SYMBOL(generic_error_remove_page);
160
161/*
162 * Safely invalidate one page from its pagecache mapping.
163 * It only drops clean, unused pages. The page must be locked.
164 *
165 * Returns 1 if the page is successfully invalidated, otherwise 0.
166 */
167int invalidate_inode_page(struct page *page)
168{
169    struct address_space *mapping = page_mapping(page);
170    if (!mapping)
171        return 0;
172    if (PageDirty(page) || PageWriteback(page))
173        return 0;
174    if (page_mapped(page))
175        return 0;
176    return invalidate_complete_page(mapping, page);
177}
178
179/**
180 * truncate_inode_pages_range - truncate range of pages specified by start & end byte offsets
181 * @mapping: mapping to truncate
182 * @lstart: offset from which to truncate
183 * @lend: offset to which to truncate (inclusive)
184 *
185 * Truncate the page cache, removing the pages that are between
186 * specified offsets (and zeroing out partial pages
187 * if lstart or lend + 1 is not page aligned).
188 *
189 * Truncate takes two passes - the first pass is nonblocking. It will not
190 * block on page locks and it will not block on writeback. The second pass
191 * will wait. This is to prevent as much IO as possible in the affected region.
192 * The first pass will remove most pages, so the search cost of the second pass
193 * is low.
194 *
195 * We pass down the cache-hot hint to the page freeing code. Even if the
196 * mapping is large, it is probably the case that the final pages are the most
197 * recently touched, and freeing happens in ascending file offset order.
198 *
199 * Note that since ->invalidatepage() accepts range to invalidate
200 * truncate_inode_pages_range is able to handle cases where lend + 1 is not
201 * page aligned properly.
202 */
203void truncate_inode_pages_range(struct address_space *mapping,
204                loff_t lstart, loff_t lend)
205{
206    pgoff_t start; /* inclusive */
207    pgoff_t end; /* exclusive */
208    unsigned int partial_start; /* inclusive */
209    unsigned int partial_end; /* exclusive */
210    struct pagevec pvec;
211    pgoff_t index;
212    int i;
213
214    cleancache_invalidate_inode(mapping);
215    if (mapping->nrpages == 0)
216        return;
217
218    /* Offsets within partial pages */
219    partial_start = lstart & (PAGE_CACHE_SIZE - 1);
220    partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
221
222    /*
223     * 'start' and 'end' always covers the range of pages to be fully
224     * truncated. Partial pages are covered with 'partial_start' at the
225     * start of the range and 'partial_end' at the end of the range.
226     * Note that 'end' is exclusive while 'lend' is inclusive.
227     */
228    start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
229    if (lend == -1)
230        /*
231         * lend == -1 indicates end-of-file so we have to set 'end'
232         * to the highest possible pgoff_t and since the type is
233         * unsigned we're using -1.
234         */
235        end = -1;
236    else
237        end = (lend + 1) >> PAGE_CACHE_SHIFT;
238
239    pagevec_init(&pvec, 0);
240    index = start;
241    while (index < end && pagevec_lookup(&pvec, mapping, index,
242            min(end - index, (pgoff_t)PAGEVEC_SIZE))) {
243        mem_cgroup_uncharge_start();
244        for (i = 0; i < pagevec_count(&pvec); i++) {
245            struct page *page = pvec.pages[i];
246
247            /* We rely upon deletion not changing page->index */
248            index = page->index;
249            if (index >= end)
250                break;
251
252            if (!trylock_page(page))
253                continue;
254            WARN_ON(page->index != index);
255            if (PageWriteback(page)) {
256                unlock_page(page);
257                continue;
258            }
259            truncate_inode_page(mapping, page);
260            unlock_page(page);
261        }
262        pagevec_release(&pvec);
263        mem_cgroup_uncharge_end();
264        cond_resched();
265        index++;
266    }
267
268    if (partial_start) {
269        struct page *page = find_lock_page(mapping, start - 1);
270        if (page) {
271            unsigned int top = PAGE_CACHE_SIZE;
272            if (start > end) {
273                /* Truncation within a single page */
274                top = partial_end;
275                partial_end = 0;
276            }
277            wait_on_page_writeback(page);
278            zero_user_segment(page, partial_start, top);
279            cleancache_invalidate_page(mapping, page);
280            if (page_has_private(page))
281                do_invalidatepage(page, partial_start,
282                          top - partial_start);
283            unlock_page(page);
284            page_cache_release(page);
285        }
286    }
287    if (partial_end) {
288        struct page *page = find_lock_page(mapping, end);
289        if (page) {
290            wait_on_page_writeback(page);
291            zero_user_segment(page, 0, partial_end);
292            cleancache_invalidate_page(mapping, page);
293            if (page_has_private(page))
294                do_invalidatepage(page, 0,
295                          partial_end);
296            unlock_page(page);
297            page_cache_release(page);
298        }
299    }
300    /*
301     * If the truncation happened within a single page no pages
302     * will be released, just zeroed, so we can bail out now.
303     */
304    if (start >= end)
305        return;
306
307    index = start;
308    for ( ; ; ) {
309        cond_resched();
310        if (!pagevec_lookup(&pvec, mapping, index,
311            min(end - index, (pgoff_t)PAGEVEC_SIZE))) {
312            if (index == start)
313                break;
314            index = start;
315            continue;
316        }
317        if (index == start && pvec.pages[0]->index >= end) {
318            pagevec_release(&pvec);
319            break;
320        }
321        mem_cgroup_uncharge_start();
322        for (i = 0; i < pagevec_count(&pvec); i++) {
323            struct page *page = pvec.pages[i];
324
325            /* We rely upon deletion not changing page->index */
326            index = page->index;
327            if (index >= end)
328                break;
329
330            lock_page(page);
331            WARN_ON(page->index != index);
332            wait_on_page_writeback(page);
333            truncate_inode_page(mapping, page);
334            unlock_page(page);
335        }
336        pagevec_release(&pvec);
337        mem_cgroup_uncharge_end();
338        index++;
339    }
340    cleancache_invalidate_inode(mapping);
341}
342EXPORT_SYMBOL(truncate_inode_pages_range);
343
344/**
345 * truncate_inode_pages - truncate *all* the pages from an offset
346 * @mapping: mapping to truncate
347 * @lstart: offset from which to truncate
348 *
349 * Called under (and serialised by) inode->i_mutex.
350 *
351 * Note: When this function returns, there can be a page in the process of
352 * deletion (inside __delete_from_page_cache()) in the specified range. Thus
353 * mapping->nrpages can be non-zero when this function returns even after
354 * truncation of the whole mapping.
355 */
356void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
357{
358    truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
359}
360EXPORT_SYMBOL(truncate_inode_pages);
361
362/**
363 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
364 * @mapping: the address_space which holds the pages to invalidate
365 * @start: the offset 'from' which to invalidate
366 * @end: the offset 'to' which to invalidate (inclusive)
367 *
368 * This function only removes the unlocked pages, if you want to
369 * remove all the pages of one inode, you must call truncate_inode_pages.
370 *
371 * invalidate_mapping_pages() will not block on IO activity. It will not
372 * invalidate pages which are dirty, locked, under writeback or mapped into
373 * pagetables.
374 */
375unsigned long invalidate_mapping_pages(struct address_space *mapping,
376        pgoff_t start, pgoff_t end)
377{
378    struct pagevec pvec;
379    pgoff_t index = start;
380    unsigned long ret;
381    unsigned long count = 0;
382    int i;
383
384    /*
385     * Note: this function may get called on a shmem/tmpfs mapping:
386     * pagevec_lookup() might then return 0 prematurely (because it
387     * got a gangful of swap entries); but it's hardly worth worrying
388     * about - it can rarely have anything to free from such a mapping
389     * (most pages are dirty), and already skips over any difficulties.
390     */
391
392    pagevec_init(&pvec, 0);
393    while (index <= end && pagevec_lookup(&pvec, mapping, index,
394            min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
395        mem_cgroup_uncharge_start();
396        for (i = 0; i < pagevec_count(&pvec); i++) {
397            struct page *page = pvec.pages[i];
398
399            /* We rely upon deletion not changing page->index */
400            index = page->index;
401            if (index > end)
402                break;
403
404            if (!trylock_page(page))
405                continue;
406            WARN_ON(page->index != index);
407            ret = invalidate_inode_page(page);
408            unlock_page(page);
409            /*
410             * Invalidation is a hint that the page is no longer
411             * of interest and try to speed up its reclaim.
412             */
413            if (!ret)
414                deactivate_page(page);
415            count += ret;
416        }
417        pagevec_release(&pvec);
418        mem_cgroup_uncharge_end();
419        cond_resched();
420        index++;
421    }
422    return count;
423}
424EXPORT_SYMBOL(invalidate_mapping_pages);
425
426/*
427 * This is like invalidate_complete_page(), except it ignores the page's
428 * refcount. We do this because invalidate_inode_pages2() needs stronger
429 * invalidation guarantees, and cannot afford to leave pages behind because
430 * shrink_page_list() has a temp ref on them, or because they're transiently
431 * sitting in the lru_cache_add() pagevecs.
432 */
433static int
434invalidate_complete_page2(struct address_space *mapping, struct page *page)
435{
436    if (page->mapping != mapping)
437        return 0;
438
439    if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
440        return 0;
441
442    spin_lock_irq(&mapping->tree_lock);
443    if (PageDirty(page))
444        goto failed;
445
446    BUG_ON(page_has_private(page));
447    __delete_from_page_cache(page);
448    spin_unlock_irq(&mapping->tree_lock);
449    mem_cgroup_uncharge_cache_page(page);
450
451    if (mapping->a_ops->freepage)
452        mapping->a_ops->freepage(page);
453
454    page_cache_release(page); /* pagecache ref */
455    return 1;
456failed:
457    spin_unlock_irq(&mapping->tree_lock);
458    return 0;
459}
460
461static int do_launder_page(struct address_space *mapping, struct page *page)
462{
463    if (!PageDirty(page))
464        return 0;
465    if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
466        return 0;
467    return mapping->a_ops->launder_page(page);
468}
469
470/**
471 * invalidate_inode_pages2_range - remove range of pages from an address_space
472 * @mapping: the address_space
473 * @start: the page offset 'from' which to invalidate
474 * @end: the page offset 'to' which to invalidate (inclusive)
475 *
476 * Any pages which are found to be mapped into pagetables are unmapped prior to
477 * invalidation.
478 *
479 * Returns -EBUSY if any pages could not be invalidated.
480 */
481int invalidate_inode_pages2_range(struct address_space *mapping,
482                  pgoff_t start, pgoff_t end)
483{
484    struct pagevec pvec;
485    pgoff_t index;
486    int i;
487    int ret = 0;
488    int ret2 = 0;
489    int did_range_unmap = 0;
490
491    cleancache_invalidate_inode(mapping);
492    pagevec_init(&pvec, 0);
493    index = start;
494    while (index <= end && pagevec_lookup(&pvec, mapping, index,
495            min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
496        mem_cgroup_uncharge_start();
497        for (i = 0; i < pagevec_count(&pvec); i++) {
498            struct page *page = pvec.pages[i];
499
500            /* We rely upon deletion not changing page->index */
501            index = page->index;
502            if (index > end)
503                break;
504
505            lock_page(page);
506            WARN_ON(page->index != index);
507            if (page->mapping != mapping) {
508                unlock_page(page);
509                continue;
510            }
511            wait_on_page_writeback(page);
512            if (page_mapped(page)) {
513                if (!did_range_unmap) {
514                    /*
515                     * Zap the rest of the file in one hit.
516                     */
517                    unmap_mapping_range(mapping,
518                       (loff_t)index << PAGE_CACHE_SHIFT,
519                       (loff_t)(1 + end - index)
520                             << PAGE_CACHE_SHIFT,
521                        0);
522                    did_range_unmap = 1;
523                } else {
524                    /*
525                     * Just zap this page
526                     */
527                    unmap_mapping_range(mapping,
528                       (loff_t)index << PAGE_CACHE_SHIFT,
529                       PAGE_CACHE_SIZE, 0);
530                }
531            }
532            BUG_ON(page_mapped(page));
533            ret2 = do_launder_page(mapping, page);
534            if (ret2 == 0) {
535                if (!invalidate_complete_page2(mapping, page))
536                    ret2 = -EBUSY;
537            }
538            if (ret2 < 0)
539                ret = ret2;
540            unlock_page(page);
541        }
542        pagevec_release(&pvec);
543        mem_cgroup_uncharge_end();
544        cond_resched();
545        index++;
546    }
547    cleancache_invalidate_inode(mapping);
548    return ret;
549}
550EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
551
552/**
553 * invalidate_inode_pages2 - remove all pages from an address_space
554 * @mapping: the address_space
555 *
556 * Any pages which are found to be mapped into pagetables are unmapped prior to
557 * invalidation.
558 *
559 * Returns -EBUSY if any pages could not be invalidated.
560 */
561int invalidate_inode_pages2(struct address_space *mapping)
562{
563    return invalidate_inode_pages2_range(mapping, 0, -1);
564}
565EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
566
567/**
568 * truncate_pagecache - unmap and remove pagecache that has been truncated
569 * @inode: inode
570 * @newsize: new file size
571 *
572 * inode's new i_size must already be written before truncate_pagecache
573 * is called.
574 *
575 * This function should typically be called before the filesystem
576 * releases resources associated with the freed range (eg. deallocates
577 * blocks). This way, pagecache will always stay logically coherent
578 * with on-disk format, and the filesystem would not have to deal with
579 * situations such as writepage being called for a page that has already
580 * had its underlying blocks deallocated.
581 */
582void truncate_pagecache(struct inode *inode, loff_t newsize)
583{
584    struct address_space *mapping = inode->i_mapping;
585    loff_t holebegin = round_up(newsize, PAGE_SIZE);
586
587    /*
588     * unmap_mapping_range is called twice, first simply for
589     * efficiency so that truncate_inode_pages does fewer
590     * single-page unmaps. However after this first call, and
591     * before truncate_inode_pages finishes, it is possible for
592     * private pages to be COWed, which remain after
593     * truncate_inode_pages finishes, hence the second
594     * unmap_mapping_range call must be made for correctness.
595     */
596    unmap_mapping_range(mapping, holebegin, 0, 1);
597    truncate_inode_pages(mapping, newsize);
598    unmap_mapping_range(mapping, holebegin, 0, 1);
599}
600EXPORT_SYMBOL(truncate_pagecache);
601
602/**
603 * truncate_setsize - update inode and pagecache for a new file size
604 * @inode: inode
605 * @newsize: new file size
606 *
607 * truncate_setsize updates i_size and performs pagecache truncation (if
608 * necessary) to @newsize. It will be typically be called from the filesystem's
609 * setattr function when ATTR_SIZE is passed in.
610 *
611 * Must be called with inode_mutex held and before all filesystem specific
612 * block truncation has been performed.
613 */
614void truncate_setsize(struct inode *inode, loff_t newsize)
615{
616    i_size_write(inode, newsize);
617    truncate_pagecache(inode, newsize);
618}
619EXPORT_SYMBOL(truncate_setsize);
620
621/**
622 * truncate_pagecache_range - unmap and remove pagecache that is hole-punched
623 * @inode: inode
624 * @lstart: offset of beginning of hole
625 * @lend: offset of last byte of hole
626 *
627 * This function should typically be called before the filesystem
628 * releases resources associated with the freed range (eg. deallocates
629 * blocks). This way, pagecache will always stay logically coherent
630 * with on-disk format, and the filesystem would not have to deal with
631 * situations such as writepage being called for a page that has already
632 * had its underlying blocks deallocated.
633 */
634void truncate_pagecache_range(struct inode *inode, loff_t lstart, loff_t lend)
635{
636    struct address_space *mapping = inode->i_mapping;
637    loff_t unmap_start = round_up(lstart, PAGE_SIZE);
638    loff_t unmap_end = round_down(1 + lend, PAGE_SIZE) - 1;
639    /*
640     * This rounding is currently just for example: unmap_mapping_range
641     * expands its hole outwards, whereas we want it to contract the hole
642     * inwards. However, existing callers of truncate_pagecache_range are
643     * doing their own page rounding first. Note that unmap_mapping_range
644     * allows holelen 0 for all, and we allow lend -1 for end of file.
645     */
646
647    /*
648     * Unlike in truncate_pagecache, unmap_mapping_range is called only
649     * once (before truncating pagecache), and without "even_cows" flag:
650     * hole-punching should not remove private COWed pages from the hole.
651     */
652    if ((u64)unmap_end > (u64)unmap_start)
653        unmap_mapping_range(mapping, unmap_start,
654                    1 + unmap_end - unmap_start, 0);
655    truncate_inode_pages_range(mapping, lstart, lend);
656}
657EXPORT_SYMBOL(truncate_pagecache_range);
658

Archive Download this file



interactive