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 * 6*e1f8e874SFrancois 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> 121da177e4SLinus Torvalds #include <linux/mm.h> 130fd0e6b0SNick Piggin #include <linux/swap.h> 141da177e4SLinus Torvalds #include <linux/module.h> 151da177e4SLinus Torvalds #include <linux/pagemap.h> 1601f2705dSNate Diller #include <linux/highmem.h> 171da177e4SLinus Torvalds #include <linux/pagevec.h> 18e08748ceSAndrew Morton #include <linux/task_io_accounting_ops.h> 191da177e4SLinus Torvalds #include <linux/buffer_head.h> /* grr. try_to_release_page, 20aaa4059bSJan Kara do_invalidatepage */ 211da177e4SLinus Torvalds 221da177e4SLinus Torvalds 23cf9a2ae8SDavid Howells /** 2428bc44d7SFengguang Wu * do_invalidatepage - invalidate part or all of a page 25cf9a2ae8SDavid Howells * @page: the page which is affected 26cf9a2ae8SDavid Howells * @offset: the index of the truncation point 27cf9a2ae8SDavid Howells * 28cf9a2ae8SDavid Howells * do_invalidatepage() is called when all or part of the page has become 29cf9a2ae8SDavid Howells * invalidated by a truncate operation. 30cf9a2ae8SDavid Howells * 31cf9a2ae8SDavid Howells * do_invalidatepage() does not have to release all buffers, but it must 32cf9a2ae8SDavid Howells * ensure that no dirty buffer is left outside @offset and that no I/O 33cf9a2ae8SDavid Howells * is underway against any of the blocks which are outside the truncation 34cf9a2ae8SDavid Howells * point. Because the caller is about to free (and possibly reuse) those 35cf9a2ae8SDavid Howells * blocks on-disk. 36cf9a2ae8SDavid Howells */ 37cf9a2ae8SDavid Howells void do_invalidatepage(struct page *page, unsigned long offset) 38cf9a2ae8SDavid Howells { 39cf9a2ae8SDavid Howells void (*invalidatepage)(struct page *, unsigned long); 40cf9a2ae8SDavid Howells invalidatepage = page->mapping->a_ops->invalidatepage; 419361401eSDavid Howells #ifdef CONFIG_BLOCK 42cf9a2ae8SDavid Howells if (!invalidatepage) 43cf9a2ae8SDavid Howells invalidatepage = block_invalidatepage; 449361401eSDavid Howells #endif 45cf9a2ae8SDavid Howells if (invalidatepage) 46cf9a2ae8SDavid Howells (*invalidatepage)(page, offset); 47cf9a2ae8SDavid Howells } 48cf9a2ae8SDavid Howells 491da177e4SLinus Torvalds static inline void truncate_partial_page(struct page *page, unsigned partial) 501da177e4SLinus Torvalds { 51eebd2aa3SChristoph Lameter zero_user_segment(page, partial, PAGE_CACHE_SIZE); 521da177e4SLinus Torvalds if (PagePrivate(page)) 531da177e4SLinus Torvalds do_invalidatepage(page, partial); 541da177e4SLinus Torvalds } 551da177e4SLinus Torvalds 56ecdfc978SLinus Torvalds /* 57ecdfc978SLinus Torvalds * This cancels just the dirty bit on the kernel page itself, it 58ecdfc978SLinus Torvalds * does NOT actually remove dirty bits on any mmap's that may be 59ecdfc978SLinus Torvalds * around. It also leaves the page tagged dirty, so any sync 60ecdfc978SLinus Torvalds * activity will still find it on the dirty lists, and in particular, 61ecdfc978SLinus Torvalds * clear_page_dirty_for_io() will still look at the dirty bits in 62ecdfc978SLinus Torvalds * the VM. 63ecdfc978SLinus Torvalds * 64ecdfc978SLinus Torvalds * Doing this should *normally* only ever be done when a page 65ecdfc978SLinus Torvalds * is truncated, and is not actually mapped anywhere at all. However, 66ecdfc978SLinus Torvalds * fs/buffer.c does this when it notices that somebody has cleaned 67ecdfc978SLinus Torvalds * out all the buffers on a page without actually doing it through 68ecdfc978SLinus Torvalds * the VM. Can you say "ext3 is horribly ugly"? Tought you could. 69ecdfc978SLinus Torvalds */ 70fba2591bSLinus Torvalds void cancel_dirty_page(struct page *page, unsigned int account_size) 71fba2591bSLinus Torvalds { 728368e328SLinus Torvalds if (TestClearPageDirty(page)) { 738368e328SLinus Torvalds struct address_space *mapping = page->mapping; 748368e328SLinus Torvalds if (mapping && mapping_cap_account_dirty(mapping)) { 753e67c098SAndrew Morton dec_zone_page_state(page, NR_FILE_DIRTY); 76c9e51e41SPeter Zijlstra dec_bdi_stat(mapping->backing_dev_info, 77c9e51e41SPeter Zijlstra BDI_RECLAIMABLE); 788368e328SLinus Torvalds if (account_size) 79fba2591bSLinus Torvalds task_io_account_cancelled_write(account_size); 80fba2591bSLinus Torvalds } 813e67c098SAndrew Morton } 828368e328SLinus Torvalds } 838368e328SLinus Torvalds EXPORT_SYMBOL(cancel_dirty_page); 84fba2591bSLinus Torvalds 851da177e4SLinus Torvalds /* 861da177e4SLinus Torvalds * If truncate cannot remove the fs-private metadata from the page, the page 8762e1c553SShaohua Li * becomes orphaned. It will be left on the LRU and may even be mapped into 8854cb8821SNick Piggin * user pagetables if we're racing with filemap_fault(). 891da177e4SLinus Torvalds * 901da177e4SLinus Torvalds * We need to bale out if page->mapping is no longer equal to the original 911da177e4SLinus Torvalds * mapping. This happens a) when the VM reclaimed the page while we waited on 92fc0ecff6SAndrew Morton * its lock, b) when a concurrent invalidate_mapping_pages got there first and 931da177e4SLinus Torvalds * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. 941da177e4SLinus Torvalds */ 951da177e4SLinus Torvalds static void 961da177e4SLinus Torvalds truncate_complete_page(struct address_space *mapping, struct page *page) 971da177e4SLinus Torvalds { 981da177e4SLinus Torvalds if (page->mapping != mapping) 991da177e4SLinus Torvalds return; 1001da177e4SLinus Torvalds 1011da177e4SLinus Torvalds if (PagePrivate(page)) 1021da177e4SLinus Torvalds do_invalidatepage(page, 0); 1031da177e4SLinus Torvalds 104a2b34564SBjorn Steinbrink cancel_dirty_page(page, PAGE_CACHE_SIZE); 105a2b34564SBjorn Steinbrink 106787d2214SNick Piggin remove_from_page_cache(page); 1071da177e4SLinus Torvalds ClearPageMappedToDisk(page); 1081da177e4SLinus Torvalds page_cache_release(page); /* pagecache ref */ 1091da177e4SLinus Torvalds } 1101da177e4SLinus Torvalds 1111da177e4SLinus Torvalds /* 112fc0ecff6SAndrew Morton * This is for invalidate_mapping_pages(). That function can be called at 1131da177e4SLinus Torvalds * any time, and is not supposed to throw away dirty pages. But pages can 1140fd0e6b0SNick Piggin * be marked dirty at any time too, so use remove_mapping which safely 1150fd0e6b0SNick Piggin * discards clean, unused pages. 1161da177e4SLinus Torvalds * 1171da177e4SLinus Torvalds * Returns non-zero if the page was successfully invalidated. 1181da177e4SLinus Torvalds */ 1191da177e4SLinus Torvalds static int 1201da177e4SLinus Torvalds invalidate_complete_page(struct address_space *mapping, struct page *page) 1211da177e4SLinus Torvalds { 1220fd0e6b0SNick Piggin int ret; 1230fd0e6b0SNick Piggin 1241da177e4SLinus Torvalds if (page->mapping != mapping) 1251da177e4SLinus Torvalds return 0; 1261da177e4SLinus Torvalds 1271da177e4SLinus Torvalds if (PagePrivate(page) && !try_to_release_page(page, 0)) 1281da177e4SLinus Torvalds return 0; 1291da177e4SLinus Torvalds 1300fd0e6b0SNick Piggin ret = remove_mapping(mapping, page); 1310fd0e6b0SNick Piggin 1320fd0e6b0SNick Piggin return ret; 1331da177e4SLinus Torvalds } 1341da177e4SLinus Torvalds 1351da177e4SLinus Torvalds /** 1360643245fSRandy Dunlap * truncate_inode_pages - truncate range of pages specified by start & end byte offsets 1371da177e4SLinus Torvalds * @mapping: mapping to truncate 1381da177e4SLinus Torvalds * @lstart: offset from which to truncate 139d7339071SHans Reiser * @lend: offset to which to truncate 1401da177e4SLinus Torvalds * 141d7339071SHans Reiser * Truncate the page cache, removing the pages that are between 142d7339071SHans Reiser * specified offsets (and zeroing out partial page 143d7339071SHans Reiser * (if lstart is not page aligned)). 1441da177e4SLinus Torvalds * 1451da177e4SLinus Torvalds * Truncate takes two passes - the first pass is nonblocking. It will not 1461da177e4SLinus Torvalds * block on page locks and it will not block on writeback. The second pass 1471da177e4SLinus Torvalds * will wait. This is to prevent as much IO as possible in the affected region. 1481da177e4SLinus Torvalds * The first pass will remove most pages, so the search cost of the second pass 1491da177e4SLinus Torvalds * is low. 1501da177e4SLinus Torvalds * 1511da177e4SLinus Torvalds * When looking at page->index outside the page lock we need to be careful to 1521da177e4SLinus Torvalds * copy it into a local to avoid races (it could change at any time). 1531da177e4SLinus Torvalds * 1541da177e4SLinus Torvalds * We pass down the cache-hot hint to the page freeing code. Even if the 1551da177e4SLinus Torvalds * mapping is large, it is probably the case that the final pages are the most 1561da177e4SLinus Torvalds * recently touched, and freeing happens in ascending file offset order. 1571da177e4SLinus Torvalds */ 158d7339071SHans Reiser void truncate_inode_pages_range(struct address_space *mapping, 159d7339071SHans Reiser loff_t lstart, loff_t lend) 1601da177e4SLinus Torvalds { 1611da177e4SLinus Torvalds const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; 162d7339071SHans Reiser pgoff_t end; 1631da177e4SLinus Torvalds const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1); 1641da177e4SLinus Torvalds struct pagevec pvec; 1651da177e4SLinus Torvalds pgoff_t next; 1661da177e4SLinus Torvalds int i; 1671da177e4SLinus Torvalds 1681da177e4SLinus Torvalds if (mapping->nrpages == 0) 1691da177e4SLinus Torvalds return; 1701da177e4SLinus Torvalds 171d7339071SHans Reiser BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1)); 172d7339071SHans Reiser end = (lend >> PAGE_CACHE_SHIFT); 173d7339071SHans Reiser 1741da177e4SLinus Torvalds pagevec_init(&pvec, 0); 1751da177e4SLinus Torvalds next = start; 176d7339071SHans Reiser while (next <= end && 177d7339071SHans Reiser pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { 1781da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 1791da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 1801da177e4SLinus Torvalds pgoff_t page_index = page->index; 1811da177e4SLinus Torvalds 182d7339071SHans Reiser if (page_index > end) { 183d7339071SHans Reiser next = page_index; 184d7339071SHans Reiser break; 185d7339071SHans Reiser } 186d7339071SHans Reiser 1871da177e4SLinus Torvalds if (page_index > next) 1881da177e4SLinus Torvalds next = page_index; 1891da177e4SLinus Torvalds next++; 190529ae9aaSNick Piggin if (!trylock_page(page)) 1911da177e4SLinus Torvalds continue; 1921da177e4SLinus Torvalds if (PageWriteback(page)) { 1931da177e4SLinus Torvalds unlock_page(page); 1941da177e4SLinus Torvalds continue; 1951da177e4SLinus Torvalds } 196d00806b1SNick Piggin if (page_mapped(page)) { 197d00806b1SNick Piggin unmap_mapping_range(mapping, 198d00806b1SNick Piggin (loff_t)page_index<<PAGE_CACHE_SHIFT, 199d00806b1SNick Piggin PAGE_CACHE_SIZE, 0); 200d00806b1SNick Piggin } 2011da177e4SLinus Torvalds truncate_complete_page(mapping, page); 2021da177e4SLinus Torvalds unlock_page(page); 2031da177e4SLinus Torvalds } 2041da177e4SLinus Torvalds pagevec_release(&pvec); 2051da177e4SLinus Torvalds cond_resched(); 2061da177e4SLinus Torvalds } 2071da177e4SLinus Torvalds 2081da177e4SLinus Torvalds if (partial) { 2091da177e4SLinus Torvalds struct page *page = find_lock_page(mapping, start - 1); 2101da177e4SLinus Torvalds if (page) { 2111da177e4SLinus Torvalds wait_on_page_writeback(page); 2121da177e4SLinus Torvalds truncate_partial_page(page, partial); 2131da177e4SLinus Torvalds unlock_page(page); 2141da177e4SLinus Torvalds page_cache_release(page); 2151da177e4SLinus Torvalds } 2161da177e4SLinus Torvalds } 2171da177e4SLinus Torvalds 2181da177e4SLinus Torvalds next = start; 2191da177e4SLinus Torvalds for ( ; ; ) { 2201da177e4SLinus Torvalds cond_resched(); 2211da177e4SLinus Torvalds if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { 2221da177e4SLinus Torvalds if (next == start) 2231da177e4SLinus Torvalds break; 2241da177e4SLinus Torvalds next = start; 2251da177e4SLinus Torvalds continue; 2261da177e4SLinus Torvalds } 227d7339071SHans Reiser if (pvec.pages[0]->index > end) { 228d7339071SHans Reiser pagevec_release(&pvec); 229d7339071SHans Reiser break; 230d7339071SHans Reiser } 2311da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 2321da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 2331da177e4SLinus Torvalds 234d7339071SHans Reiser if (page->index > end) 235d7339071SHans Reiser break; 2361da177e4SLinus Torvalds lock_page(page); 2371da177e4SLinus Torvalds wait_on_page_writeback(page); 238d00806b1SNick Piggin if (page_mapped(page)) { 239d00806b1SNick Piggin unmap_mapping_range(mapping, 240d00806b1SNick Piggin (loff_t)page->index<<PAGE_CACHE_SHIFT, 241d00806b1SNick Piggin PAGE_CACHE_SIZE, 0); 242d00806b1SNick Piggin } 2431da177e4SLinus Torvalds if (page->index > next) 2441da177e4SLinus Torvalds next = page->index; 2451da177e4SLinus Torvalds next++; 2461da177e4SLinus Torvalds truncate_complete_page(mapping, page); 2471da177e4SLinus Torvalds unlock_page(page); 2481da177e4SLinus Torvalds } 2491da177e4SLinus Torvalds pagevec_release(&pvec); 2501da177e4SLinus Torvalds } 2511da177e4SLinus Torvalds } 252d7339071SHans Reiser EXPORT_SYMBOL(truncate_inode_pages_range); 2531da177e4SLinus Torvalds 254d7339071SHans Reiser /** 255d7339071SHans Reiser * truncate_inode_pages - truncate *all* the pages from an offset 256d7339071SHans Reiser * @mapping: mapping to truncate 257d7339071SHans Reiser * @lstart: offset from which to truncate 258d7339071SHans Reiser * 2591b1dcc1bSJes Sorensen * Called under (and serialised by) inode->i_mutex. 260d7339071SHans Reiser */ 261d7339071SHans Reiser void truncate_inode_pages(struct address_space *mapping, loff_t lstart) 262d7339071SHans Reiser { 263d7339071SHans Reiser truncate_inode_pages_range(mapping, lstart, (loff_t)-1); 264d7339071SHans Reiser } 2651da177e4SLinus Torvalds EXPORT_SYMBOL(truncate_inode_pages); 2661da177e4SLinus Torvalds 267fc9a07e7SAndrew Morton unsigned long __invalidate_mapping_pages(struct address_space *mapping, 268fc9a07e7SAndrew Morton pgoff_t start, pgoff_t end, bool be_atomic) 2691da177e4SLinus Torvalds { 2701da177e4SLinus Torvalds struct pagevec pvec; 2711da177e4SLinus Torvalds pgoff_t next = start; 2721da177e4SLinus Torvalds unsigned long ret = 0; 2731da177e4SLinus Torvalds int i; 2741da177e4SLinus Torvalds 2751da177e4SLinus Torvalds pagevec_init(&pvec, 0); 2761da177e4SLinus Torvalds while (next <= end && 2771da177e4SLinus Torvalds pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { 2781da177e4SLinus Torvalds for (i = 0; i < pagevec_count(&pvec); i++) { 2791da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 280e0f23603SNeilBrown pgoff_t index; 281e0f23603SNeilBrown int lock_failed; 2821da177e4SLinus Torvalds 283529ae9aaSNick Piggin lock_failed = !trylock_page(page); 284e0f23603SNeilBrown 285e0f23603SNeilBrown /* 286e0f23603SNeilBrown * We really shouldn't be looking at the ->index of an 287e0f23603SNeilBrown * unlocked page. But we're not allowed to lock these 288e0f23603SNeilBrown * pages. So we rely upon nobody altering the ->index 289e0f23603SNeilBrown * of this (pinned-by-us) page. 290e0f23603SNeilBrown */ 291e0f23603SNeilBrown index = page->index; 292e0f23603SNeilBrown if (index > next) 293e0f23603SNeilBrown next = index; 2941da177e4SLinus Torvalds next++; 295e0f23603SNeilBrown if (lock_failed) 2961da177e4SLinus Torvalds continue; 297e0f23603SNeilBrown 2981da177e4SLinus Torvalds if (PageDirty(page) || PageWriteback(page)) 2991da177e4SLinus Torvalds goto unlock; 3001da177e4SLinus Torvalds if (page_mapped(page)) 3011da177e4SLinus Torvalds goto unlock; 3021da177e4SLinus Torvalds ret += invalidate_complete_page(mapping, page); 3031da177e4SLinus Torvalds unlock: 3041da177e4SLinus Torvalds unlock_page(page); 3051da177e4SLinus Torvalds if (next > end) 3061da177e4SLinus Torvalds break; 3071da177e4SLinus Torvalds } 3081da177e4SLinus Torvalds pagevec_release(&pvec); 309fc9a07e7SAndrew Morton if (likely(!be_atomic)) 310fc9a07e7SAndrew Morton cond_resched(); 3111da177e4SLinus Torvalds } 3121da177e4SLinus Torvalds return ret; 3131da177e4SLinus Torvalds } 314fc9a07e7SAndrew Morton 315fc9a07e7SAndrew Morton /** 316fc9a07e7SAndrew Morton * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode 317fc9a07e7SAndrew Morton * @mapping: the address_space which holds the pages to invalidate 318fc9a07e7SAndrew Morton * @start: the offset 'from' which to invalidate 319fc9a07e7SAndrew Morton * @end: the offset 'to' which to invalidate (inclusive) 320fc9a07e7SAndrew Morton * 321fc9a07e7SAndrew Morton * This function only removes the unlocked pages, if you want to 322fc9a07e7SAndrew Morton * remove all the pages of one inode, you must call truncate_inode_pages. 323fc9a07e7SAndrew Morton * 324fc9a07e7SAndrew Morton * invalidate_mapping_pages() will not block on IO activity. It will not 325fc9a07e7SAndrew Morton * invalidate pages which are dirty, locked, under writeback or mapped into 326fc9a07e7SAndrew Morton * pagetables. 327fc9a07e7SAndrew Morton */ 328fc9a07e7SAndrew Morton unsigned long invalidate_mapping_pages(struct address_space *mapping, 329fc9a07e7SAndrew Morton pgoff_t start, pgoff_t end) 330fc9a07e7SAndrew Morton { 331fc9a07e7SAndrew Morton return __invalidate_mapping_pages(mapping, start, end, false); 332fc9a07e7SAndrew Morton } 33354bc4855SAnton Altaparmakov EXPORT_SYMBOL(invalidate_mapping_pages); 3341da177e4SLinus Torvalds 335bd4c8ce4SAndrew Morton /* 336bd4c8ce4SAndrew Morton * This is like invalidate_complete_page(), except it ignores the page's 337bd4c8ce4SAndrew Morton * refcount. We do this because invalidate_inode_pages2() needs stronger 338bd4c8ce4SAndrew Morton * invalidation guarantees, and cannot afford to leave pages behind because 3392706a1b8SAnderson Briglia * shrink_page_list() has a temp ref on them, or because they're transiently 3402706a1b8SAnderson Briglia * sitting in the lru_cache_add() pagevecs. 341bd4c8ce4SAndrew Morton */ 342bd4c8ce4SAndrew Morton static int 343bd4c8ce4SAndrew Morton invalidate_complete_page2(struct address_space *mapping, struct page *page) 344bd4c8ce4SAndrew Morton { 345bd4c8ce4SAndrew Morton if (page->mapping != mapping) 346bd4c8ce4SAndrew Morton return 0; 347bd4c8ce4SAndrew Morton 348887ed2f3STrond Myklebust if (PagePrivate(page) && !try_to_release_page(page, GFP_KERNEL)) 349bd4c8ce4SAndrew Morton return 0; 350bd4c8ce4SAndrew Morton 35119fd6231SNick Piggin spin_lock_irq(&mapping->tree_lock); 352bd4c8ce4SAndrew Morton if (PageDirty(page)) 353bd4c8ce4SAndrew Morton goto failed; 354bd4c8ce4SAndrew Morton 355bd4c8ce4SAndrew Morton BUG_ON(PagePrivate(page)); 356bd4c8ce4SAndrew Morton __remove_from_page_cache(page); 35719fd6231SNick Piggin spin_unlock_irq(&mapping->tree_lock); 358bd4c8ce4SAndrew Morton page_cache_release(page); /* pagecache ref */ 359bd4c8ce4SAndrew Morton return 1; 360bd4c8ce4SAndrew Morton failed: 36119fd6231SNick Piggin spin_unlock_irq(&mapping->tree_lock); 362bd4c8ce4SAndrew Morton return 0; 363bd4c8ce4SAndrew Morton } 364bd4c8ce4SAndrew Morton 365e3db7691STrond Myklebust static int do_launder_page(struct address_space *mapping, struct page *page) 366e3db7691STrond Myklebust { 367e3db7691STrond Myklebust if (!PageDirty(page)) 368e3db7691STrond Myklebust return 0; 369e3db7691STrond Myklebust if (page->mapping != mapping || mapping->a_ops->launder_page == NULL) 370e3db7691STrond Myklebust return 0; 371e3db7691STrond Myklebust return mapping->a_ops->launder_page(page); 372e3db7691STrond Myklebust } 373e3db7691STrond Myklebust 3741da177e4SLinus Torvalds /** 3751da177e4SLinus Torvalds * invalidate_inode_pages2_range - remove range of pages from an address_space 37667be2dd1SMartin Waitz * @mapping: the address_space 3771da177e4SLinus Torvalds * @start: the page offset 'from' which to invalidate 3781da177e4SLinus Torvalds * @end: the page offset 'to' which to invalidate (inclusive) 3791da177e4SLinus Torvalds * 3801da177e4SLinus Torvalds * Any pages which are found to be mapped into pagetables are unmapped prior to 3811da177e4SLinus Torvalds * invalidation. 3821da177e4SLinus Torvalds * 3836ccfa806SHisashi Hifumi * Returns -EBUSY if any pages could not be invalidated. 3841da177e4SLinus Torvalds */ 3851da177e4SLinus Torvalds int invalidate_inode_pages2_range(struct address_space *mapping, 3861da177e4SLinus Torvalds pgoff_t start, pgoff_t end) 3871da177e4SLinus Torvalds { 3881da177e4SLinus Torvalds struct pagevec pvec; 3891da177e4SLinus Torvalds pgoff_t next; 3901da177e4SLinus Torvalds int i; 3911da177e4SLinus Torvalds int ret = 0; 3920dd1334fSHisashi Hifumi int ret2 = 0; 3931da177e4SLinus Torvalds int did_range_unmap = 0; 3941da177e4SLinus Torvalds int wrapped = 0; 3951da177e4SLinus Torvalds 3961da177e4SLinus Torvalds pagevec_init(&pvec, 0); 3971da177e4SLinus Torvalds next = start; 3987b965e08STrond Myklebust while (next <= end && !wrapped && 3991da177e4SLinus Torvalds pagevec_lookup(&pvec, mapping, next, 4001da177e4SLinus Torvalds min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) { 4017b965e08STrond Myklebust for (i = 0; i < pagevec_count(&pvec); i++) { 4021da177e4SLinus Torvalds struct page *page = pvec.pages[i]; 4031da177e4SLinus Torvalds pgoff_t page_index; 4041da177e4SLinus Torvalds 4051da177e4SLinus Torvalds lock_page(page); 4061da177e4SLinus Torvalds if (page->mapping != mapping) { 4071da177e4SLinus Torvalds unlock_page(page); 4081da177e4SLinus Torvalds continue; 4091da177e4SLinus Torvalds } 4101da177e4SLinus Torvalds page_index = page->index; 4111da177e4SLinus Torvalds next = page_index + 1; 4121da177e4SLinus Torvalds if (next == 0) 4131da177e4SLinus Torvalds wrapped = 1; 4141da177e4SLinus Torvalds if (page_index > end) { 4151da177e4SLinus Torvalds unlock_page(page); 4161da177e4SLinus Torvalds break; 4171da177e4SLinus Torvalds } 4181da177e4SLinus Torvalds wait_on_page_writeback(page); 419d00806b1SNick Piggin if (page_mapped(page)) { 4201da177e4SLinus Torvalds if (!did_range_unmap) { 4211da177e4SLinus Torvalds /* 4221da177e4SLinus Torvalds * Zap the rest of the file in one hit. 4231da177e4SLinus Torvalds */ 4241da177e4SLinus Torvalds unmap_mapping_range(mapping, 425479ef592SOleg Drokin (loff_t)page_index<<PAGE_CACHE_SHIFT, 426479ef592SOleg Drokin (loff_t)(end - page_index + 1) 4271da177e4SLinus Torvalds << PAGE_CACHE_SHIFT, 4281da177e4SLinus Torvalds 0); 4291da177e4SLinus Torvalds did_range_unmap = 1; 4301da177e4SLinus Torvalds } else { 4311da177e4SLinus Torvalds /* 4321da177e4SLinus Torvalds * Just zap this page 4331da177e4SLinus Torvalds */ 4341da177e4SLinus Torvalds unmap_mapping_range(mapping, 435479ef592SOleg Drokin (loff_t)page_index<<PAGE_CACHE_SHIFT, 4361da177e4SLinus Torvalds PAGE_CACHE_SIZE, 0); 4371da177e4SLinus Torvalds } 4381da177e4SLinus Torvalds } 439d00806b1SNick Piggin BUG_ON(page_mapped(page)); 4400dd1334fSHisashi Hifumi ret2 = do_launder_page(mapping, page); 4410dd1334fSHisashi Hifumi if (ret2 == 0) { 4420dd1334fSHisashi Hifumi if (!invalidate_complete_page2(mapping, page)) 4436ccfa806SHisashi Hifumi ret2 = -EBUSY; 4440dd1334fSHisashi Hifumi } 4450dd1334fSHisashi Hifumi if (ret2 < 0) 4460dd1334fSHisashi Hifumi ret = ret2; 4471da177e4SLinus Torvalds unlock_page(page); 4481da177e4SLinus Torvalds } 4491da177e4SLinus Torvalds pagevec_release(&pvec); 4501da177e4SLinus Torvalds cond_resched(); 4511da177e4SLinus Torvalds } 4521da177e4SLinus Torvalds return ret; 4531da177e4SLinus Torvalds } 4541da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); 4551da177e4SLinus Torvalds 4561da177e4SLinus Torvalds /** 4571da177e4SLinus Torvalds * invalidate_inode_pages2 - remove all pages from an address_space 45867be2dd1SMartin Waitz * @mapping: the address_space 4591da177e4SLinus Torvalds * 4601da177e4SLinus Torvalds * Any pages which are found to be mapped into pagetables are unmapped prior to 4611da177e4SLinus Torvalds * invalidation. 4621da177e4SLinus Torvalds * 4631da177e4SLinus Torvalds * Returns -EIO if any pages could not be invalidated. 4641da177e4SLinus Torvalds */ 4651da177e4SLinus Torvalds int invalidate_inode_pages2(struct address_space *mapping) 4661da177e4SLinus Torvalds { 4671da177e4SLinus Torvalds return invalidate_inode_pages2_range(mapping, 0, -1); 4681da177e4SLinus Torvalds } 4691da177e4SLinus Torvalds EXPORT_SYMBOL_GPL(invalidate_inode_pages2); 470