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  * When looking at page->index outside the page lock we need to be careful to
 203  * copy it into a local to avoid races (it could change at any time).
 204  *
 205  * We pass down the cache-hot hint to the page freeing code.  Even if the
 206  * mapping is large, it is probably the case that the final pages are the most
 207  * recently touched, and freeing happens in ascending file offset order.
 208  */
 209 void truncate_inode_pages_range(struct address_space *mapping,
 210                                 loff_t lstart, loff_t lend)
 211 {
 212         const pgoff_t start = (lstart + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
 213         pgoff_t end;
 214         const unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
 215         struct pagevec pvec;
 216         pgoff_t next;
 217         int i;
 218
 219         cleancache_flush_inode(mapping);
 220         if (mapping->nrpages == 0)
 221                 return;
 222
 223         BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
 224         end = (lend >> PAGE_CACHE_SHIFT);
 225
 226         pagevec_init(&pvec, 0);
 227         next = start;
 228         while (next <= end &&
 229                pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 230                 mem_cgroup_uncharge_start();
 231                 for (i = 0; i < pagevec_count(&pvec); i++) {
 232                         struct page *page = pvec.pages[i];
 233                         pgoff_t page_index = page->index;
 234
 235                         if (page_index > end) {
 236                                 next = page_index;
 237                                 break;
 238                         }
 239
 240                         if (page_index > next)
 241                                 next = page_index;
 242                         next++;
 243                         if (!trylock_page(page))
 244                                 continue;
 245                         if (PageWriteback(page)) {
 246                                 unlock_page(page);
 247                                 continue;
 248                         }
 249                         truncate_inode_page(mapping, page);
 250                         unlock_page(page);
 251                 }
 252                 pagevec_release(&pvec);
 253                 mem_cgroup_uncharge_end();
 254                 cond_resched();
 255         }
 256
 257         if (partial) {
 258                 struct page *page = find_lock_page(mapping, start - 1);
 259                 if (page) {
 260                         wait_on_page_writeback(page);
 261                         truncate_partial_page(page, partial);
 262                         unlock_page(page);
 263                         page_cache_release(page);
 264                 }
 265         }
 266
 267         next = start;
 268         for ( ; ; ) {
 269                 cond_resched();
 270                 if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 271                         if (next == start)
 272                                 break;
 273                         next = start;
 274                         continue;
 275                 }
 276                 if (pvec.pages[0]->index > end) {
 277                         pagevec_release(&pvec);
 278                         break;
 279                 }
 280                 mem_cgroup_uncharge_start();
 281                 for (i = 0; i < pagevec_count(&pvec); i++) {
 282                         struct page *page = pvec.pages[i];
 283
 284                         if (page->index > end)
 285                                 break;
 286                         lock_page(page);
 287                         wait_on_page_writeback(page);
 288                         truncate_inode_page(mapping, page);
 289                         if (page->index > next)
 290                                 next = page->index;
 291                         next++;
 292                         unlock_page(page);
 293                 }
 294                 pagevec_release(&pvec);
 295                 mem_cgroup_uncharge_end();
 296         }
 297         cleancache_flush_inode(mapping);
 298 }
 299 EXPORT_SYMBOL(truncate_inode_pages_range);
 300
 301 /**
 302  * truncate_inode_pages - truncate *all* the pages from an offset
 303  * @mapping: mapping to truncate
 304  * @lstart: offset from which to truncate
 305  *
 306  * Called under (and serialised by) inode->i_mutex.
 307  *
 308  * Note: When this function returns, there can be a page in the process of
 309  * deletion (inside __delete_from_page_cache()) in the specified range.  Thus
 310  * mapping->nrpages can be non-zero when this function returns even after
 311  * truncation of the whole mapping.
 312  */
 313 void truncate_inode_pages(struct address_space *mapping, loff_t lstart)
 314 {
 315         truncate_inode_pages_range(mapping, lstart, (loff_t)-1);
 316 }
 317 EXPORT_SYMBOL(truncate_inode_pages);
 318
 319 /**
 320  * invalidate_mapping_pages - Invalidate all the unlocked pages of one inode
 321  * @mapping: the address_space which holds the pages to invalidate
 322  * @start: the offset 'from' which to invalidate
 323  * @end: the offset 'to' which to invalidate (inclusive)
 324  *
 325  * This function only removes the unlocked pages, if you want to
 326  * remove all the pages of one inode, you must call truncate_inode_pages.
 327  *
 328  * invalidate_mapping_pages() will not block on IO activity. It will not
 329  * invalidate pages which are dirty, locked, under writeback or mapped into
 330  * pagetables.
 331  */
 332 unsigned long invalidate_mapping_pages(struct address_space *mapping,
 333                 pgoff_t start, pgoff_t end)
 334 {
 335         struct pagevec pvec;
 336         pgoff_t next = start;
 337         unsigned long ret;
 338         unsigned long count = 0;
 339         int i;
 340
 341         pagevec_init(&pvec, 0);
 342         while (next <= end &&
 343                         pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) {
 344                 mem_cgroup_uncharge_start();
 345                 for (i = 0; i < pagevec_count(&pvec); i++) {
 346                         struct page *page = pvec.pages[i];
 347                         pgoff_t index;
 348                         int lock_failed;
 349
 350                         lock_failed = !trylock_page(page);
 351
 352                         /*
 353                          * We really shouldn't be looking at the ->index of an
 354                          * unlocked page.  But we're not allowed to lock these
 355                          * pages.  So we rely upon nobody altering the ->index
 356                          * of this (pinned-by-us) page.
 357                          */
 358                         index = page->index;
 359                         if (index > next)
 360                                 next = index;
 361                         next++;
 362                         if (lock_failed)
 363                                 continue;
 364
 365                         ret = invalidate_inode_page(page);
 366                         unlock_page(page);
 367                         /*
 368                          * Invalidation is a hint that the page is no longer
 369                          * of interest and try to speed up its reclaim.
 370                          */
 371                         if (!ret)
 372                                 deactivate_page(page);
 373                         count += ret;
 374                         if (next > end)
 375                                 break;
 376                 }
 377                 pagevec_release(&pvec);
 378                 mem_cgroup_uncharge_end();
 379                 cond_resched();
 380         }
 381         return count;
 382 }
 383 EXPORT_SYMBOL(invalidate_mapping_pages);
 384
 385 /*
 386  * This is like invalidate_complete_page(), except it ignores the page's
 387  * refcount.  We do this because invalidate_inode_pages2() needs stronger
 388  * invalidation guarantees, and cannot afford to leave pages behind because
 389  * shrink_page_list() has a temp ref on them, or because they're transiently
 390  * sitting in the lru_cache_add() pagevecs.
 391  */
 392 static int
 393 invalidate_complete_page2(struct address_space *mapping, struct page *page)
 394 {
 395         if (page->mapping != mapping)
 396                 return 0;
 397
 398         if (page_has_private(page) && !try_to_release_page(page, GFP_KERNEL))
 399                 return 0;
 400
 401         spin_lock_irq(&mapping->tree_lock);
 402         if (PageDirty(page))
 403                 goto failed;
 404
 405         clear_page_mlock(page);
 406         BUG_ON(page_has_private(page));
 407         __delete_from_page_cache(page);
 408         spin_unlock_irq(&mapping->tree_lock);
 409         mem_cgroup_uncharge_cache_page(page);
 410
 411         if (mapping->a_ops->freepage)
 412                 mapping->a_ops->freepage(page);
 413
 414         page_cache_release(page);       /* pagecache ref */
 415         return 1;
 416 failed:
 417         spin_unlock_irq(&mapping->tree_lock);
 418         return 0;
 419 }
 420
 421 static int do_launder_page(struct address_space *mapping, struct page *page)
 422 {
 423         if (!PageDirty(page))
 424                 return 0;
 425         if (page->mapping != mapping || mapping->a_ops->launder_page == NULL)
 426                 return 0;
 427         return mapping->a_ops->launder_page(page);
 428 }
 429
 430 /**
 431  * invalidate_inode_pages2_range - remove range of pages from an address_space
 432  * @mapping: the address_space
 433  * @start: the page offset 'from' which to invalidate
 434  * @end: the page offset 'to' which to invalidate (inclusive)
 435  *
 436  * Any pages which are found to be mapped into pagetables are unmapped prior to
 437  * invalidation.
 438  *
 439  * Returns -EBUSY if any pages could not be invalidated.
 440  */
 441 int invalidate_inode_pages2_range(struct address_space *mapping,
 442                                   pgoff_t start, pgoff_t end)
 443 {
 444         struct pagevec pvec;
 445         pgoff_t next;
 446         int i;
 447         int ret = 0;
 448         int ret2 = 0;
 449         int did_range_unmap = 0;
 450         int wrapped = 0;
 451
 452         cleancache_flush_inode(mapping);
 453         pagevec_init(&pvec, 0);
 454         next = start;
 455         while (next <= end && !wrapped &&
 456                 pagevec_lookup(&pvec, mapping, next,
 457                         min(end - next, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
 458                 mem_cgroup_uncharge_start();
 459                 for (i = 0; i < pagevec_count(&pvec); i++) {
 460                         struct page *page = pvec.pages[i];
 461                         pgoff_t page_index;
 462
 463                         lock_page(page);
 464                         if (page->mapping != mapping) {
 465                                 unlock_page(page);
 466                                 continue;
 467                         }
 468                         page_index = page->index;
 469                         next = page_index + 1;
 470                         if (next == 0)
 471                                 wrapped = 1;
 472                         if (page_index > end) {
 473                                 unlock_page(page);
 474                                 break;
 475                         }
 476                         wait_on_page_writeback(page);
 477                         if (page_mapped(page)) {
 478                                 if (!did_range_unmap) {
 479                                         /*
 480                                          * Zap the rest of the file in one hit.
 481                                          */
 482                                         unmap_mapping_range(mapping,
 483                                            (loff_t)page_index<<PAGE_CACHE_SHIFT,
 484                                            (loff_t)(end - page_index + 1)
 485                                                         << PAGE_CACHE_SHIFT,
 486                                             0);
 487                                         did_range_unmap = 1;
 488                                 } else {
 489                                         /*
 490                                          * Just zap this page
 491                                          */
 492                                         unmap_mapping_range(mapping,
 493                                           (loff_t)page_index<<PAGE_CACHE_SHIFT,
 494                                           PAGE_CACHE_SIZE, 0);
 495                                 }
 496                         }
 497                         BUG_ON(page_mapped(page));
 498                         ret2 = do_launder_page(mapping, page);
 499                         if (ret2 == 0) {
 500                                 if (!invalidate_complete_page2(mapping, page))
 501                                         ret2 = -EBUSY;
 502                         }
 503                         if (ret2 < 0)
 504                                 ret = ret2;
 505                         unlock_page(page);
 506                 }
 507                 pagevec_release(&pvec);
 508                 mem_cgroup_uncharge_end();
 509                 cond_resched();
 510         }
 511         cleancache_flush_inode(mapping);
 512         return ret;
 513 }
 514 EXPORT_SYMBOL_GPL(invalidate_inode_pages2_range);
 515
 516 /**
 517  * invalidate_inode_pages2 - remove all pages from an address_space
 518  * @mapping: the address_space
 519  *
 520  * Any pages which are found to be mapped into pagetables are unmapped prior to
 521  * invalidation.
 522  *
 523  * Returns -EBUSY if any pages could not be invalidated.
 524  */
 525 int invalidate_inode_pages2(struct address_space *mapping)
 526 {
 527         return invalidate_inode_pages2_range(mapping, 0, -1);
 528 }
 529 EXPORT_SYMBOL_GPL(invalidate_inode_pages2);
 530
 531 /**
 532  * truncate_pagecache - unmap and remove pagecache that has been truncated
 533  * @inode: inode
 534  * @old: old file offset
 535  * @new: new file offset
 536  *
 537  * inode's new i_size must already be written before truncate_pagecache
 538  * is called.
 539  *
 540  * This function should typically be called before the filesystem
 541  * releases resources associated with the freed range (eg. deallocates
 542  * blocks). This way, pagecache will always stay logically coherent
 543  * with on-disk format, and the filesystem would not have to deal with
 544  * situations such as writepage being called for a page that has already
 545  * had its underlying blocks deallocated.
 546  */
 547 void truncate_pagecache(struct inode *inode, loff_t old, loff_t new)
 548 {
 549         struct address_space *mapping = inode->i_mapping;
 550
 551         /*
 552          * unmap_mapping_range is called twice, first simply for
 553          * efficiency so that truncate_inode_pages does fewer
 554          * single-page unmaps.  However after this first call, and
 555          * before truncate_inode_pages finishes, it is possible for
 556          * private pages to be COWed, which remain after
 557          * truncate_inode_pages finishes, hence the second
 558          * unmap_mapping_range call must be made for correctness.
 559          */
 560         unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1);
 561         truncate_inode_pages(mapping, new);
 562         unmap_mapping_range(mapping, new + PAGE_SIZE - 1, 0, 1);
 563 }
 564 EXPORT_SYMBOL(truncate_pagecache);
 565
 566 /**
 567  * truncate_setsize - update inode and pagecache for a new file size
 568  * @inode: inode
 569  * @newsize: new file size
 570  *
 571  * truncate_setsize updates i_size and performs pagecache truncation (if
 572  * necessary) to @newsize. It will be typically be called from the filesystem's
 573  * setattr function when ATTR_SIZE is passed in.
 574  *
 575  * Must be called with inode_mutex held and before all filesystem specific
 576  * block truncation has been performed.
 577  */
 578 void truncate_setsize(struct inode *inode, loff_t newsize)
 579 {
 580         loff_t oldsize;
 581
 582         oldsize = inode->i_size;
 583         i_size_write(inode, newsize);
 584
 585         truncate_pagecache(inode, oldsize, newsize);
 586 }
 587 EXPORT_SYMBOL(truncate_setsize);
 588
 589 /**
 590  * vmtruncate - unmap mappings "freed" by truncate() syscall
 591  * @inode: inode of the file used
 592  * @offset: file offset to start truncating
 593  *
 594  * This function is deprecated and truncate_setsize or truncate_pagecache
 595  * should be used instead, together with filesystem specific block truncation.
 596  */
 597 int vmtruncate(struct inode *inode, loff_t offset)
 598 {
 599         int error;
 600
 601         error = inode_newsize_ok(inode, offset);
 602         if (error)
 603                 return error;
 604
 605         truncate_setsize(inode, offset);
 606         if (inode->i_op->truncate)
 607                 inode->i_op->truncate(inode);
 608         return 0;
 609 }
 610 EXPORT_SYMBOL(vmtruncate);
 611
 612 int vmtruncate_range(struct inode *inode, loff_t offset, loff_t end)
 613 {
 614         struct address_space *mapping = inode->i_mapping;
 615
 616         /*
 617          * If the underlying filesystem is not going to provide
 618          * a way to truncate a range of blocks (punch a hole) -
 619          * we should return failure right now.
 620          */
 621         if (!inode->i_op->truncate_range)
 622                 return -ENOSYS;
 623
 624         mutex_lock(&inode->i_mutex);
 625         down_write(&inode->i_alloc_sem);
 626         unmap_mapping_range(mapping, offset, (end - offset), 1);
 627         inode->i_op->truncate_range(inode, offset, end);
 628         /* unmap again to remove racily COWed private pages */
 629         unmap_mapping_range(mapping, offset, (end - offset), 1);
 630         up_write(&inode->i_alloc_sem);
 631         mutex_unlock(&inode->i_mutex);
 632
 633         return 0;
 634 }