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    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/task_io_accounting_ops.h>
  17 #include <linux/buffer_head.h>  /* grr. try_to_release_page,
  18                                    do_invalidatepage */
  19
  20
  21 /**
  22  * do_invalidatepage - invalidate part of all of a page
  23  * @page: the page which is affected
  24  * @offset: the index of the truncation point
  25  *
  26  * do_invalidatepage() is called when all or part of the page has become
  27  * invalidated by a truncate operation.
  28  *
  29  * do_invalidatepage() does not have to release all buffers, but it must
  30  * ensure that no dirty buffer is left outside @offset and that no I/O
  31  * is underway against any of the blocks which are outside the truncation
  32  * point.  Because the caller is about to free (and possibly reuse) those
  33  * blocks on-disk.
  34  */
  35 void do_invalidatepage(struct page *page, unsigned long offset)
  36 {
  37         void (*invalidatepage)(struct page *, unsigned long);
  38         invalidatepage = page->mapping->a_ops->invalidatepage;
  39 #ifdef CONFIG_BLOCK
  40         if (!invalidatepage)
  41                 invalidatepage = block_invalidatepage;
  42 #endif
  43         if (invalidatepage)
  44                 (*invalidatepage)(page, offset);
  45 }
  46
  47 static inline void truncate_partial_page(struct page *page, unsigned partial)
  48 {
  49         memclear_highpage_flush(page, partial, PAGE_CACHE_SIZE-partial);
  50         if (PagePrivate(page))
  51                 do_invalidatepage(page, partial);
  52 }
  53
  54 void cancel_dirty_page(struct page *page, unsigned int account_size)
  55 {
  56         /* If we're cancelling the page, it had better not be mapped any more */
  57         if (page_mapped(page)) {
  58                 static unsigned int warncount;
  59
  60                 WARN_ON(++warncount < 5);
  61         }
  62
  63         if (TestClearPageDirty(page)) {
  64                 struct address_space *mapping = page->mapping;
  65                 if (mapping && mapping_cap_account_dirty(mapping)) {
  66                         dec_zone_page_state(page, NR_FILE_DIRTY);
  67                         if (account_size)
  68                                 task_io_account_cancelled_write(account_size);
  69                 }
  70         }
  71 }
  72 EXPORT_SYMBOL(cancel_dirty_page);
  73
  74 /*
  75  * If truncate cannot remove the fs-private metadata from the page, the page
  76  * becomes anonymous.  It will be left on the LRU and may even be mapped into
  77  * user pagetables if we're racing with filemap_nopage().
  78  *
  79  * We need to bale out if page->mapping is no longer equal to the original
  80  * mapping.  This happens a) when the VM reclaimed the page while we waited on
  81  * its lock, b) when a concurrent invalidate_inode_pages got there first and
  82  * c) when tmpfs swizzles a page between a tmpfs inode and swapper_space.
  83  */
  84 static void
  85 truncate_complete_page(struct address_space *mapping, struct page *page)
  86 {
  87         if (page->mapping != mapping)
  88                 return;
  89
  90         cancel_dirty_page(page, PAGE_CACHE_SIZE);
  91
  92         if (PagePrivate(page))
  93                 do_invalidatepage(page, 0);
  94
  95         ClearPageUptodate(page);
  96         ClearPageMappedToDisk(page);
  97         remove_from_page_cache(page);
  98         page_cache_release(page);       /* pagecache ref */
  99 }
 100
 101 /*
 102  * This is for invalidate_inode_pages().  That function can be called at
 103  * any time, and is not supposed to throw away dirty pages.  But pages can
 104  * be marked dirty at any time too, so use remove_mapping which safely
 105  * discards clean, unused pages.
 106  *
 107  * Returns non-zero if the page was successfully invalidated.
 108  */
 109 static int
 110 invalidate_complete_page(struct address_space *mapping, struct page *page)
 111 {
 112         int ret;
 113
 114         if (page->mapping != mapping)
 115                 return 0;
 116
 117         if (PagePrivate(page) && !try_to_release_page(page, 0))
 118                 return 0;
 119
 120         ret = remove_mapping(mapping, page);
 121
 122         return ret;
 123 }
 124
 125 /**
 126  * truncate_inode_pages - truncate range of pages specified by start and
 127  * end byte offsets
 128  * @mapping: mapping to truncate
 129  * @lstart: offset from which to truncate
 130  * @lend: offset to which to truncate
 131  *
 132  * Truncate the page cache, removing the pages that are between
 133  * specified offsets (and zeroing out partial page
 134  * (if lstart is not page aligned)).
 135  *
 136  * Truncate takes two passes - the first pass is nonblocking.  It will not
 137  * block on page locks and it will not block on writeback.  The second pass
 138  * will wait.  This is to prevent as much IO as possible in the affected region.
 139  * The first pass will remove most pages, so the search cost of the second pass
 140  * is low.
 141  *
 142  * When looking at page->index outside the page lock we need to be careful to
 143  * copy it into a local to avoid races (it could change at any time).
 144  *
 145  * We pass down the cache-hot hint to the page freeing code.  Even if the
 146  * mapping is large, it is probably the case that the final pages are the most
 147  * recently touched, and freeing happens in ascending file offset order.
 148  */
 149 void truncate_inode_pages_range(struct address_space *mapping,
 150                                 loff_t lstart, loff_t lend)
 151 {
 152         const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
 153         pgoff_t end;
 154         const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
 155         struct pagevec pvec;
 156         pgoff_t next;
 157         int i;
 158
 159         if (mapping->nrpages == 0)
 160                 return;
 161
 162         BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
 163         end = (lend >> PAGE_CACHE_SHIFT);
 164
 165         pagevec_init(&pvec, 0);
 166         next = start;
 167         while (next <= end &&
 168                pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 169                 for (i = 0; i < pagevec_count(&pvec); i++) {
 170                         struct page *page = pvec.pages[i];
 171                         pgoff_t page_index = page->index;
 172
 173                         if (page_index > end) {
 174                                 next = page_index;
 175                                 break;
 176                         }
 177
 178                         if (page_index > next)
 179                                 next = page_index;
 180                         next++;
 181                         if (TestSetPageLocked(page))
 182                                 continue;
 183                         if (PageWriteback(page)) {
 184                                 unlock_page(page);
 185                                 continue;
 186                         }
 187                         truncate_complete_page(mapping, page);
 188                         unlock_page(page);
 189                 }
 190                 pagevec_release(&pvec);
 191                 cond_resched();
 192         }
 193
 194         if (partial) {
 195                 struct page *page = find_lock_page(mapping, start - 1);
 196                 if (page) {
 197                         wait_on_page_writeback(page);
 198                         truncate_partial_page(page, partial);
 199                         unlock_page(page);
 200                         page_cache_release(page);
 201                 }
 202         }
 203
 204         next = start;
 205         for ( ; ; ) {
 206                 cond_resched();
 207                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 208                         if (next == start)
 209                                 break;
 210                         next = start;
 211                         continue;
 212                 }
 213                 if (pvec.pages[0]->index > end) {
 214                         pagevec_release(&pvec);
 215                         break;
 216                 }
 217                 for (i = 0; i < pagevec_count(&pvec); i++) {
 218                         struct page *page = pvec.pages[i];
 219
 220                         if (page->index > end)
 221                                 break;
 222                         lock_page(page);
 223                         wait_on_page_writeback(page);
 224                         if (page->index > next)
 225                                 next = page->index;
 226                         next++;
 227                         truncate_complete_page(mapping, page);
 228                         unlock_page(page);
 229                 }
 230                 pagevec_release(&pvec);
 231         }
 232 }
 233 EXPORT_SYMBOL(truncate_inode_pages_range);
 234
 235 /**
 236  * truncate_inode_pages - truncate *all* the pages from an offset
 237  * @mapping: mapping to truncate
 238  * @lstart: offset from which to truncate
 239  *
 240  * Called under (and serialised by) inode->i_mutex.
 241  */
 242 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
 243 {
 244         truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
 245 }
 246 EXPORT_SYMBOL(truncate_inode_pages);
 247
 248 /**
 249  * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
 250  * @mapping: the address_space which holds the pages to invalidate
 251  * @start: the offset 'from' which to invalidate
 252  * @end: the offset 'to' which to invalidate (inclusive)
 253  *
 254  * This function only removes the unlocked pages, if you want to
 255  * remove all the pages of one inode, you must call truncate_inode_pages.
 256  *
 257  * invalidate_mapping_pages() will not block on IO activity. It will not
 258  * invalidate pages which are dirty, locked, under writeback or mapped into
 259  * pagetables.
 260  */
 261 unsigned long invalidate_mapping_pages(struct address_space *mapping,
 262                                 pgoff_t start, pgoff_t end)
 263 {
 264         struct pagevec pvec;
 265         pgoff_t next = start;
 266         unsigned long ret = 0;
 267         int i;
 268
 269         pagevec_init(&pvec, 0);
 270         while (next <= end &&
 271                         pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 272                 for (i = 0; i < pagevec_count(&pvec); i++) {
 273                         struct page *page = pvec.pages[i];
 274                         pgoff_t index;
 275                         int lock_failed;
 276
 277                         lock_failed = TestSetPageLocked(page);
 278
 279                         /*
 280                          * We really shouldn't be looking at the ->index of an
 281                          * unlocked page.  But we're not allowed to lock these
 282                          * pages.  So we rely upon nobody altering the ->index
 283                          * of this (pinned-by-us) page.
 284                          */
 285                         index = page->index;
 286                         if (index > next)
 287                                 next = index;
 288                         next++;
 289                         if (lock_failed)
 290                                 continue;
 291
 292                         if (PageDirty(page) || PageWriteback(page))
 293                                 goto unlock;
 294                         if (page_mapped(page))
 295                                 goto unlock;
 296                         ret += invalidate_complete_page(mapping, page);
 297 unlock:
 298                         unlock_page(page);
 299                         if (next > end)
 300                                 break;
 301                 }
 302                 pagevec_release(&pvec);
 303         }
 304         return ret;
 305 }
 306
 307 unsigned long invalidate_inode_pages(struct address_space *mapping)
 308 {
 309         return invalidate_mapping_pages(mapping, 0, ~0UL);
 310 }
 311 EXPORT_SYMBOL(invalidate_inode_pages);
 312
 313 /*
 314  * This is like invalidate_complete_page(), except it ignores the page's
 315  * refcount.  We do this because invalidate_inode_pages2() needs stronger
 316  * invalidation guarantees, and cannot afford to leave pages behind because
 317  * shrink_list() has a temp ref on them, or because they're transiently sitting
 318  * in the lru_cache_add() pagevecs.
 319  */
 320 static int
 321 invalidate_complete_page2(struct address_space *mapping, struct page *page)
 322 {
 323         if (page->mapping != mapping)
 324                 return 0;
 325
 326         if (PagePrivate(page) && !try_to_release_page(page, GFP_KERNEL))
 327                 return 0;
 328
 329         write_lock_irq(&mapping->tree_lock);
 330         if (PageDirty(page))
 331                 goto failed;
 332
 333         BUG_ON(PagePrivate(page));
 334         __remove_from_page_cache(page);
 335         write_unlock_irq(&mapping->tree_lock);
 336         ClearPageUptodate(page);
 337         page_cache_release(page);       /* pagecache ref */
 338         return 1;
 339 failed:
 340         write_unlock_irq(&mapping->tree_lock);
 341         return 0;
 342 }
 343
 344 /**
 345  * invalidate_inode_pages2_range - remove range of pages from an address_space
 346  * @mapping: the address_space
 347  * @start: the page offset 'from' which to invalidate
 348  * @end: the page offset 'to' which to invalidate (inclusive)
 349  *
 350  * Any pages which are found to be mapped into pagetables are unmapped prior to
 351  * invalidation.
 352  *
 353  * Returns -EIO if any pages could not be invalidated.
 354  */
 355 int invalidate_inode_pages2_range(struct address_space *mapping,
 356                                   pgoff_t start, pgoff_t end)
 357 {
 358         struct pagevec pvec;
 359         pgoff_t next;
 360         int i;
 361         int ret = 0;
 362         int did_range_unmap = 0;
 363         int wrapped = 0;
 364
 365         pagevec_init(&pvec, 0);
 366         next = start;
 367         while (next <= end && !ret && !wrapped &&
 368                 pagevec_lookup(&pvec, mapping, next,
 369                         min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
 370                 for (i = 0; !ret && i < pagevec_count(&pvec); i++) {
 371                         struct page *page = pvec.pages[i];
 372                         pgoff_t page_index;
 373
 374                         lock_page(page);
 375                         if (page->mapping != mapping) {
 376                                 unlock_page(page);
 377                                 continue;
 378                         }
 379                         page_index = page->index;
 380                         next = page_index + 1;
 381                         if (next == 0)
 382                                 wrapped = 1;
 383                         if (page_index > end) {
 384                                 unlock_page(page);
 385                                 break;
 386                         }
 387                         wait_on_page_writeback(page);
 388                         while (page_mapped(page)) {
 389                                 if (!did_range_unmap) {
 390                                         /*
 391                                          * Zap the rest of the file in one hit.
 392                                          */
 393                                         unmap_mapping_range(mapping,
 394                                            (loff_t)page_index<<PAGE_CACHE_SHIFT,
 395                                            (loff_t)(end - page_index + 1)
 396                                                         << PAGE_CACHE_SHIFT,
 397                                             0);
 398                                         did_range_unmap = 1;
 399                                 } else {
 400                                         /*
 401                                          * Just zap this page
 402                                          */
 403                                         unmap_mapping_range(mapping,
 404                                           (loff_t)page_index<<PAGE_CACHE_SHIFT,
 405                                           PAGE_CACHE_SIZE, 0);
 406                                 }
 407                         }
 408                         if (!invalidate_complete_page2(mapping, page))
 409                                 ret = -EIO;
 410                         unlock_page(page);
 411                 }
 412                 pagevec_release(&pvec);
 413                 cond_resched();
 414         }
 415         WARN_ON_ONCE(ret);
 416         return ret;
 417 }
 418 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
 419
 420 /**
 421  * invalidate_inode_pages2 - remove all pages from an address_space
 422  * @mapping: the address_space
 423  *
 424  * Any pages which are found to be mapped into pagetables are unmapped prior to
 425  * invalidation.
 426  *
 427  * Returns -EIO if any pages could not be invalidated.
 428  */
 429 int invalidate_inode_pages2(struct address_space *mapping)
 430 {
 431         return invalidate_inode_pages2_range(mapping, 0, -1);
 432 }
 433 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);