truncate.c revision 887ed2f3aecde2ff24e06666932dc5f144745044
1/* 2 * mm/truncate.c - code for taking down pages from address_spaces 3 * 4 * Copyright (C) 2002, Linus Torvalds 5 * 6 * 10Sep2002 akpm@zip.com.au 7 * Initial version. 8 */ 9 10#include <linux/kernel.h> 11#include <linux/mm.h> 12#include <linux/swap.h> 13#include <linux/module.h> 14#include <linux/pagemap.h> 15#include <linux/pagevec.h> 16#include <linux/buffer_head.h> /* grr. try_to_release_page, 17 do_invalidatepage */ 18 19 20/** 21 * do_invalidatepage - invalidate part of all of a page 22 * @page: the page which is affected 23 * @offset: the index of the truncation point 24 * 25 * do_invalidatepage() is called when all or part of the page has become 26 * invalidated by a truncate operation. 27 * 28 * do_invalidatepage() does not have to release all buffers, but it must 29 * ensure that no dirty buffer is left outside @offset and that no I/O 30 * is underway against any of the blocks which are outside the truncation 31 * point. Because the caller is about to free (and possibly reuse) those 32 * blocks on-disk. 33 */ 34void do_invalidatepage(struct page *page, unsigned long offset) 35{ 36 void (*invalidatepage)(struct page *, unsigned long); 37 invalidatepage = page->mapping->a_ops->invalidatepage; 38#ifdef CONFIG_BLOCK 39 if (!invalidatepage) 40 invalidatepage = block_invalidatepage; 41#endif 42 if (invalidatepage) 43 (*invalidatepage)(page, offset); 44} 45 46static inline void truncate_partial_page(struct page *page, unsigned partial) 47{ 48 memclear_highpage_flush(page, partial, PAGE_CACHE_SIZE-partial); 49 if (PagePrivate(page)) 50 do_invalidatepage(page, partial); 51} 52 53/* 54 * If truncate cannot remove the fs-private metadata from the page, the page 55 * becomes anonymous. It will be left on the LRU and may even be mapped into 56 * user pagetables if we're racing with filemap_nopage(). 57 * 58 * We need to bale out if page->mapping is no longer equal to the original 59 * mapping. This happens a) when the VM reclaimed the page while we waited on 60 * its lock, b) when a concurrent invalidate_inode_pages got there first and 61 * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space. 62 */ 63static void 64truncate_complete_page(struct address_space *mapping, struct page *page) 65{ 66 if (page->mapping != mapping) 67 return; 68 69 if (PagePrivate(page)) 70 do_invalidatepage(page, 0); 71 72 clear_page_dirty(page); 73 ClearPageUptodate(page); 74 ClearPageMappedToDisk(page); 75 remove_from_page_cache(page); 76 page_cache_release(page); /* pagecache ref */ 77} 78 79/* 80 * This is for invalidate_inode_pages(). That function can be called at 81 * any time, and is not supposed to throw away dirty pages. But pages can 82 * be marked dirty at any time too, so use remove_mapping which safely 83 * discards clean, unused pages. 84 * 85 * Returns non-zero if the page was successfully invalidated. 86 */ 87static int 88invalidate_complete_page(struct address_space *mapping, struct page *page) 89{ 90 int ret; 91 92 if (page->mapping != mapping) 93 return 0; 94 95 if (PagePrivate(page) && !try_to_release_page(page, 0)) 96 return 0; 97 98 ret = remove_mapping(mapping, page); 99 ClearPageUptodate(page); 100 101 return ret; 102} 103 104/** 105 * truncate_inode_pages - truncate range of pages specified by start and 106 * end byte offsets 107 * @mapping: mapping to truncate 108 * @lstart: offset from which to truncate 109 * @lend: offset to which to truncate 110 * 111 * Truncate the page cache, removing the pages that are between 112 * specified offsets (and zeroing out partial page 113 * (if lstart is not page aligned)). 114 * 115 * Truncate takes two passes - the first pass is nonblocking. It will not 116 * block on page locks and it will not block on writeback. The second pass 117 * will wait. This is to prevent as much IO as possible in the affected region. 118 * The first pass will remove most pages, so the search cost of the second pass 119 * is low. 120 * 121 * When looking at page->index outside the page lock we need to be careful to 122 * copy it into a local to avoid races (it could change at any time). 123 * 124 * We pass down the cache-hot hint to the page freeing code. Even if the 125 * mapping is large, it is probably the case that the final pages are the most 126 * recently touched, and freeing happens in ascending file offset order. 127 */ 128void truncate_inode_pages_range(struct address_space *mapping, 129 loff_t lstart, loff_t lend) 130{ 131 const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; 132 pgoff_t end; 133 const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1); 134 struct pagevec pvec; 135 pgoff_t next; 136 int i; 137 138 if (mapping->nrpages == 0) 139 return; 140 141 BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1)); 142 end = (lend >> PAGE_CACHE_SHIFT); 143 144 pagevec_init(&pvec, 0); 145 next = start; 146 while (next <= end && 147 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { 148 for (i = 0; i < pagevec_count(&pvec); i++) { 149 struct page *page = pvec.pages[i]; 150 pgoff_t page_index = page->index; 151 152 if (page_index > end) { 153 next = page_index; 154 break; 155 } 156 157 if (page_index > next) 158 next = page_index; 159 next++; 160 if (TestSetPageLocked(page)) 161 continue; 162 if (PageWriteback(page)) { 163 unlock_page(page); 164 continue; 165 } 166 truncate_complete_page(mapping, page); 167 unlock_page(page); 168 } 169 pagevec_release(&pvec); 170 cond_resched(); 171 } 172 173 if (partial) { 174 struct page *page = find_lock_page(mapping, start - 1); 175 if (page) { 176 wait_on_page_writeback(page); 177 truncate_partial_page(page, partial); 178 unlock_page(page); 179 page_cache_release(page); 180 } 181 } 182 183 next = start; 184 for ( ; ; ) { 185 cond_resched(); 186 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { 187 if (next == start) 188 break; 189 next = start; 190 continue; 191 } 192 if (pvec.pages[0]->index > end) { 193 pagevec_release(&pvec); 194 break; 195 } 196 for (i = 0; i < pagevec_count(&pvec); i++) { 197 struct page *page = pvec.pages[i]; 198 199 if (page->index > end) 200 break; 201 lock_page(page); 202 wait_on_page_writeback(page); 203 if (page->index > next) 204 next = page->index; 205 next++; 206 truncate_complete_page(mapping, page); 207 unlock_page(page); 208 } 209 pagevec_release(&pvec); 210 } 211} 212EXPORT_SYMBOL(truncate_inode_pages_range); 213 214/** 215 * truncate_inode_pages - truncate *all* the pages from an offset 216 * @mapping: mapping to truncate 217 * @lstart: offset from which to truncate 218 * 219 * Called under (and serialised by) inode->i_mutex. 220 */ 221void truncate_inode_pages(struct address_space *mapping, loff_t lstart) 222{ 223 truncate_inode_pages_range(mapping, lstart, (loff_t)-1); 224} 225EXPORT_SYMBOL(truncate_inode_pages); 226 227/** 228 * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode 229 * @mapping: the address_space which holds the pages to invalidate 230 * @start: the offset 'from' which to invalidate 231 * @end: the offset 'to' which to invalidate (inclusive) 232 * 233 * This function only removes the unlocked pages, if you want to 234 * remove all the pages of one inode, you must call truncate_inode_pages. 235 * 236 * invalidate_mapping_pages() will not block on IO activity. It will not 237 * invalidate pages which are dirty, locked, under writeback or mapped into 238 * pagetables. 239 */ 240unsigned long invalidate_mapping_pages(struct address_space *mapping, 241 pgoff_t start, pgoff_t end) 242{ 243 struct pagevec pvec; 244 pgoff_t next = start; 245 unsigned long ret = 0; 246 int i; 247 248 pagevec_init(&pvec, 0); 249 while (next <= end && 250 pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { 251 for (i = 0; i < pagevec_count(&pvec); i++) { 252 struct page *page = pvec.pages[i]; 253 pgoff_t index; 254 int lock_failed; 255 256 lock_failed = TestSetPageLocked(page); 257 258 /* 259 * We really shouldn't be looking at the ->index of an 260 * unlocked page. But we're not allowed to lock these 261 * pages. So we rely upon nobody altering the ->index 262 * of this (pinned-by-us) page. 263 */ 264 index = page->index; 265 if (index > next) 266 next = index; 267 next++; 268 if (lock_failed) 269 continue; 270 271 if (PageDirty(page) || PageWriteback(page)) 272 goto unlock; 273 if (page_mapped(page)) 274 goto unlock; 275 ret += invalidate_complete_page(mapping, page); 276unlock: 277 unlock_page(page); 278 if (next > end) 279 break; 280 } 281 pagevec_release(&pvec); 282 } 283 return ret; 284} 285 286unsigned long invalidate_inode_pages(struct address_space *mapping) 287{ 288 return invalidate_mapping_pages(mapping, 0, ~0UL); 289} 290EXPORT_SYMBOL(invalidate_inode_pages); 291 292/* 293 * This is like invalidate_complete_page(), except it ignores the page's 294 * refcount. We do this because invalidate_inode_pages2() needs stronger 295 * invalidation guarantees, and cannot afford to leave pages behind because 296 * shrink_list() has a temp ref on them, or because they're transiently sitting 297 * in the lru_cache_add() pagevecs. 298 */ 299static int 300invalidate_complete_page2(struct address_space *mapping, struct page *page) 301{ 302 if (page->mapping != mapping) 303 return 0; 304 305 if (PagePrivate(page) && !try_to_release_page(page, GFP_KERNEL)) 306 return 0; 307 308 write_lock_irq(&mapping->tree_lock); 309 if (PageDirty(page)) 310 goto failed; 311 312 BUG_ON(PagePrivate(page)); 313 __remove_from_page_cache(page); 314 write_unlock_irq(&mapping->tree_lock); 315 ClearPageUptodate(page); 316 page_cache_release(page); /* pagecache ref */ 317 return 1; 318failed: 319 write_unlock_irq(&mapping->tree_lock); 320 return 0; 321} 322 323/** 324 * invalidate_inode_pages2_range - remove range of pages from an address_space 325 * @mapping: the address_space 326 * @start: the page offset 'from' which to invalidate 327 * @end: the page offset 'to' which to invalidate (inclusive) 328 * 329 * Any pages which are found to be mapped into pagetables are unmapped prior to 330 * invalidation. 331 * 332 * Returns -EIO if any pages could not be invalidated. 333 */ 334int invalidate_inode_pages2_range(struct address_space *mapping, 335 pgoff_t start, pgoff_t end) 336{ 337 struct pagevec pvec; 338 pgoff_t next; 339 int i; 340 int ret = 0; 341 int did_range_unmap = 0; 342 int wrapped = 0; 343 344 pagevec_init(&pvec, 0); 345 next = start; 346 while (next <= end && !ret && !wrapped && 347 pagevec_lookup(&pvec, mapping, next, 348 min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) { 349 for (i = 0; !ret && i < pagevec_count(&pvec); i++) { 350 struct page *page = pvec.pages[i]; 351 pgoff_t page_index; 352 int was_dirty; 353 354 lock_page(page); 355 if (page->mapping != mapping) { 356 unlock_page(page); 357 continue; 358 } 359 page_index = page->index; 360 next = page_index + 1; 361 if (next == 0) 362 wrapped = 1; 363 if (page_index > end) { 364 unlock_page(page); 365 break; 366 } 367 wait_on_page_writeback(page); 368 while (page_mapped(page)) { 369 if (!did_range_unmap) { 370 /* 371 * Zap the rest of the file in one hit. 372 */ 373 unmap_mapping_range(mapping, 374 (loff_t)page_index<<PAGE_CACHE_SHIFT, 375 (loff_t)(end - page_index + 1) 376 << PAGE_CACHE_SHIFT, 377 0); 378 did_range_unmap = 1; 379 } else { 380 /* 381 * Just zap this page 382 */ 383 unmap_mapping_range(mapping, 384 (loff_t)page_index<<PAGE_CACHE_SHIFT, 385 PAGE_CACHE_SIZE, 0); 386 } 387 } 388 was_dirty = test_clear_page_dirty(page); 389 if (!invalidate_complete_page2(mapping, page)) { 390 if (was_dirty) 391 set_page_dirty(page); 392 ret = -EIO; 393 } 394 unlock_page(page); 395 } 396 pagevec_release(&pvec); 397 cond_resched(); 398 } 399 WARN_ON_ONCE(ret); 400 return ret; 401} 402EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range); 403 404/** 405 * invalidate_inode_pages2 - remove all pages from an address_space 406 * @mapping: the address_space 407 * 408 * Any pages which are found to be mapped into pagetables are unmapped prior to 409 * invalidation. 410 * 411 * Returns -EIO if any pages could not be invalidated. 412 */ 413int invalidate_inode_pages2(struct address_space *mapping) 414{ 415 return invalidate_inode_pages2_range(mapping, 0, -1); 416} 417EXPORT_SYMBOL_GPL(invalidate_inode_pages2); 418