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