filemap.c revision 6673e0c3fbeaed2cd08e2fd4a4aa97382d6fedb0
1/* 2 * linux/mm/filemap.c 3 * 4 * Copyright (C) 1994-1999 Linus Torvalds 5 */ 6 7/* 8 * This file handles the generic file mmap semantics used by 9 * most "normal" filesystems (but you don't /have/ to use this: 10 * the NFS filesystem used to do this differently, for example) 11 */ 12#include <linux/module.h> 13#include <linux/slab.h> 14#include <linux/compiler.h> 15#include <linux/fs.h> 16#include <linux/uaccess.h> 17#include <linux/aio.h> 18#include <linux/capability.h> 19#include <linux/kernel_stat.h> 20#include <linux/mm.h> 21#include <linux/swap.h> 22#include <linux/mman.h> 23#include <linux/pagemap.h> 24#include <linux/file.h> 25#include <linux/uio.h> 26#include <linux/hash.h> 27#include <linux/writeback.h> 28#include <linux/backing-dev.h> 29#include <linux/pagevec.h> 30#include <linux/blkdev.h> 31#include <linux/security.h> 32#include <linux/syscalls.h> 33#include <linux/cpuset.h> 34#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */ 35#include <linux/memcontrol.h> 36#include <linux/mm_inline.h> /* for page_is_file_cache() */ 37#include "internal.h" 38 39/* 40 * FIXME: remove all knowledge of the buffer layer from the core VM 41 */ 42#include <linux/buffer_head.h> /* for generic_osync_inode */ 43 44#include <asm/mman.h> 45 46 47/* 48 * Shared mappings implemented 30.11.1994. It's not fully working yet, 49 * though. 50 * 51 * Shared mappings now work. 15.8.1995 Bruno. 52 * 53 * finished 'unifying' the page and buffer cache and SMP-threaded the 54 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> 55 * 56 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> 57 */ 58 59/* 60 * Lock ordering: 61 * 62 * ->i_mmap_lock (vmtruncate) 63 * ->private_lock (__free_pte->__set_page_dirty_buffers) 64 * ->swap_lock (exclusive_swap_page, others) 65 * ->mapping->tree_lock 66 * 67 * ->i_mutex 68 * ->i_mmap_lock (truncate->unmap_mapping_range) 69 * 70 * ->mmap_sem 71 * ->i_mmap_lock 72 * ->page_table_lock or pte_lock (various, mainly in memory.c) 73 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock) 74 * 75 * ->mmap_sem 76 * ->lock_page (access_process_vm) 77 * 78 * ->i_mutex (generic_file_buffered_write) 79 * ->mmap_sem (fault_in_pages_readable->do_page_fault) 80 * 81 * ->i_mutex 82 * ->i_alloc_sem (various) 83 * 84 * ->inode_lock 85 * ->sb_lock (fs/fs-writeback.c) 86 * ->mapping->tree_lock (__sync_single_inode) 87 * 88 * ->i_mmap_lock 89 * ->anon_vma.lock (vma_adjust) 90 * 91 * ->anon_vma.lock 92 * ->page_table_lock or pte_lock (anon_vma_prepare and various) 93 * 94 * ->page_table_lock or pte_lock 95 * ->swap_lock (try_to_unmap_one) 96 * ->private_lock (try_to_unmap_one) 97 * ->tree_lock (try_to_unmap_one) 98 * ->zone.lru_lock (follow_page->mark_page_accessed) 99 * ->zone.lru_lock (check_pte_range->isolate_lru_page) 100 * ->private_lock (page_remove_rmap->set_page_dirty) 101 * ->tree_lock (page_remove_rmap->set_page_dirty) 102 * ->inode_lock (page_remove_rmap->set_page_dirty) 103 * ->inode_lock (zap_pte_range->set_page_dirty) 104 * ->private_lock (zap_pte_range->__set_page_dirty_buffers) 105 * 106 * ->task->proc_lock 107 * ->dcache_lock (proc_pid_lookup) 108 */ 109 110/* 111 * Remove a page from the page cache and free it. Caller has to make 112 * sure the page is locked and that nobody else uses it - or that usage 113 * is safe. The caller must hold the mapping's tree_lock. 114 */ 115void __remove_from_page_cache(struct page *page) 116{ 117 struct address_space *mapping = page->mapping; 118 119 radix_tree_delete(&mapping->page_tree, page->index); 120 page->mapping = NULL; 121 mapping->nrpages--; 122 __dec_zone_page_state(page, NR_FILE_PAGES); 123 BUG_ON(page_mapped(page)); 124 mem_cgroup_uncharge_cache_page(page); 125 126 /* 127 * Some filesystems seem to re-dirty the page even after 128 * the VM has canceled the dirty bit (eg ext3 journaling). 129 * 130 * Fix it up by doing a final dirty accounting check after 131 * having removed the page entirely. 132 */ 133 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) { 134 dec_zone_page_state(page, NR_FILE_DIRTY); 135 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); 136 } 137} 138 139void remove_from_page_cache(struct page *page) 140{ 141 struct address_space *mapping = page->mapping; 142 143 BUG_ON(!PageLocked(page)); 144 145 spin_lock_irq(&mapping->tree_lock); 146 __remove_from_page_cache(page); 147 spin_unlock_irq(&mapping->tree_lock); 148} 149 150static int sync_page(void *word) 151{ 152 struct address_space *mapping; 153 struct page *page; 154 155 page = container_of((unsigned long *)word, struct page, flags); 156 157 /* 158 * page_mapping() is being called without PG_locked held. 159 * Some knowledge of the state and use of the page is used to 160 * reduce the requirements down to a memory barrier. 161 * The danger here is of a stale page_mapping() return value 162 * indicating a struct address_space different from the one it's 163 * associated with when it is associated with one. 164 * After smp_mb(), it's either the correct page_mapping() for 165 * the page, or an old page_mapping() and the page's own 166 * page_mapping() has gone NULL. 167 * The ->sync_page() address_space operation must tolerate 168 * page_mapping() going NULL. By an amazing coincidence, 169 * this comes about because none of the users of the page 170 * in the ->sync_page() methods make essential use of the 171 * page_mapping(), merely passing the page down to the backing 172 * device's unplug functions when it's non-NULL, which in turn 173 * ignore it for all cases but swap, where only page_private(page) is 174 * of interest. When page_mapping() does go NULL, the entire 175 * call stack gracefully ignores the page and returns. 176 * -- wli 177 */ 178 smp_mb(); 179 mapping = page_mapping(page); 180 if (mapping && mapping->a_ops && mapping->a_ops->sync_page) 181 mapping->a_ops->sync_page(page); 182 io_schedule(); 183 return 0; 184} 185 186static int sync_page_killable(void *word) 187{ 188 sync_page(word); 189 return fatal_signal_pending(current) ? -EINTR : 0; 190} 191 192/** 193 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range 194 * @mapping: address space structure to write 195 * @start: offset in bytes where the range starts 196 * @end: offset in bytes where the range ends (inclusive) 197 * @sync_mode: enable synchronous operation 198 * 199 * Start writeback against all of a mapping's dirty pages that lie 200 * within the byte offsets <start, end> inclusive. 201 * 202 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as 203 * opposed to a regular memory cleansing writeback. The difference between 204 * these two operations is that if a dirty page/buffer is encountered, it must 205 * be waited upon, and not just skipped over. 206 */ 207int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, 208 loff_t end, int sync_mode) 209{ 210 int ret; 211 struct writeback_control wbc = { 212 .sync_mode = sync_mode, 213 .nr_to_write = LONG_MAX, 214 .range_start = start, 215 .range_end = end, 216 }; 217 218 if (!mapping_cap_writeback_dirty(mapping)) 219 return 0; 220 221 ret = do_writepages(mapping, &wbc); 222 return ret; 223} 224 225static inline int __filemap_fdatawrite(struct address_space *mapping, 226 int sync_mode) 227{ 228 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); 229} 230 231int filemap_fdatawrite(struct address_space *mapping) 232{ 233 return __filemap_fdatawrite(mapping, WB_SYNC_ALL); 234} 235EXPORT_SYMBOL(filemap_fdatawrite); 236 237int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, 238 loff_t end) 239{ 240 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); 241} 242EXPORT_SYMBOL(filemap_fdatawrite_range); 243 244/** 245 * filemap_flush - mostly a non-blocking flush 246 * @mapping: target address_space 247 * 248 * This is a mostly non-blocking flush. Not suitable for data-integrity 249 * purposes - I/O may not be started against all dirty pages. 250 */ 251int filemap_flush(struct address_space *mapping) 252{ 253 return __filemap_fdatawrite(mapping, WB_SYNC_NONE); 254} 255EXPORT_SYMBOL(filemap_flush); 256 257/** 258 * wait_on_page_writeback_range - wait for writeback to complete 259 * @mapping: target address_space 260 * @start: beginning page index 261 * @end: ending page index 262 * 263 * Wait for writeback to complete against pages indexed by start->end 264 * inclusive 265 */ 266int wait_on_page_writeback_range(struct address_space *mapping, 267 pgoff_t start, pgoff_t end) 268{ 269 struct pagevec pvec; 270 int nr_pages; 271 int ret = 0; 272 pgoff_t index; 273 274 if (end < start) 275 return 0; 276 277 pagevec_init(&pvec, 0); 278 index = start; 279 while ((index <= end) && 280 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, 281 PAGECACHE_TAG_WRITEBACK, 282 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) { 283 unsigned i; 284 285 for (i = 0; i < nr_pages; i++) { 286 struct page *page = pvec.pages[i]; 287 288 /* until radix tree lookup accepts end_index */ 289 if (page->index > end) 290 continue; 291 292 wait_on_page_writeback(page); 293 if (PageError(page)) 294 ret = -EIO; 295 } 296 pagevec_release(&pvec); 297 cond_resched(); 298 } 299 300 /* Check for outstanding write errors */ 301 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags)) 302 ret = -ENOSPC; 303 if (test_and_clear_bit(AS_EIO, &mapping->flags)) 304 ret = -EIO; 305 306 return ret; 307} 308 309/** 310 * sync_page_range - write and wait on all pages in the passed range 311 * @inode: target inode 312 * @mapping: target address_space 313 * @pos: beginning offset in pages to write 314 * @count: number of bytes to write 315 * 316 * Write and wait upon all the pages in the passed range. This is a "data 317 * integrity" operation. It waits upon in-flight writeout before starting and 318 * waiting upon new writeout. If there was an IO error, return it. 319 * 320 * We need to re-take i_mutex during the generic_osync_inode list walk because 321 * it is otherwise livelockable. 322 */ 323int sync_page_range(struct inode *inode, struct address_space *mapping, 324 loff_t pos, loff_t count) 325{ 326 pgoff_t start = pos >> PAGE_CACHE_SHIFT; 327 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT; 328 int ret; 329 330 if (!mapping_cap_writeback_dirty(mapping) || !count) 331 return 0; 332 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1); 333 if (ret == 0) { 334 mutex_lock(&inode->i_mutex); 335 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA); 336 mutex_unlock(&inode->i_mutex); 337 } 338 if (ret == 0) 339 ret = wait_on_page_writeback_range(mapping, start, end); 340 return ret; 341} 342EXPORT_SYMBOL(sync_page_range); 343 344/** 345 * sync_page_range_nolock - write & wait on all pages in the passed range without locking 346 * @inode: target inode 347 * @mapping: target address_space 348 * @pos: beginning offset in pages to write 349 * @count: number of bytes to write 350 * 351 * Note: Holding i_mutex across sync_page_range_nolock() is not a good idea 352 * as it forces O_SYNC writers to different parts of the same file 353 * to be serialised right until io completion. 354 */ 355int sync_page_range_nolock(struct inode *inode, struct address_space *mapping, 356 loff_t pos, loff_t count) 357{ 358 pgoff_t start = pos >> PAGE_CACHE_SHIFT; 359 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT; 360 int ret; 361 362 if (!mapping_cap_writeback_dirty(mapping) || !count) 363 return 0; 364 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1); 365 if (ret == 0) 366 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA); 367 if (ret == 0) 368 ret = wait_on_page_writeback_range(mapping, start, end); 369 return ret; 370} 371EXPORT_SYMBOL(sync_page_range_nolock); 372 373/** 374 * filemap_fdatawait - wait for all under-writeback pages to complete 375 * @mapping: address space structure to wait for 376 * 377 * Walk the list of under-writeback pages of the given address space 378 * and wait for all of them. 379 */ 380int filemap_fdatawait(struct address_space *mapping) 381{ 382 loff_t i_size = i_size_read(mapping->host); 383 384 if (i_size == 0) 385 return 0; 386 387 return wait_on_page_writeback_range(mapping, 0, 388 (i_size - 1) >> PAGE_CACHE_SHIFT); 389} 390EXPORT_SYMBOL(filemap_fdatawait); 391 392int filemap_write_and_wait(struct address_space *mapping) 393{ 394 int err = 0; 395 396 if (mapping->nrpages) { 397 err = filemap_fdatawrite(mapping); 398 /* 399 * Even if the above returned error, the pages may be 400 * written partially (e.g. -ENOSPC), so we wait for it. 401 * But the -EIO is special case, it may indicate the worst 402 * thing (e.g. bug) happened, so we avoid waiting for it. 403 */ 404 if (err != -EIO) { 405 int err2 = filemap_fdatawait(mapping); 406 if (!err) 407 err = err2; 408 } 409 } 410 return err; 411} 412EXPORT_SYMBOL(filemap_write_and_wait); 413 414/** 415 * filemap_write_and_wait_range - write out & wait on a file range 416 * @mapping: the address_space for the pages 417 * @lstart: offset in bytes where the range starts 418 * @lend: offset in bytes where the range ends (inclusive) 419 * 420 * Write out and wait upon file offsets lstart->lend, inclusive. 421 * 422 * Note that `lend' is inclusive (describes the last byte to be written) so 423 * that this function can be used to write to the very end-of-file (end = -1). 424 */ 425int filemap_write_and_wait_range(struct address_space *mapping, 426 loff_t lstart, loff_t lend) 427{ 428 int err = 0; 429 430 if (mapping->nrpages) { 431 err = __filemap_fdatawrite_range(mapping, lstart, lend, 432 WB_SYNC_ALL); 433 /* See comment of filemap_write_and_wait() */ 434 if (err != -EIO) { 435 int err2 = wait_on_page_writeback_range(mapping, 436 lstart >> PAGE_CACHE_SHIFT, 437 lend >> PAGE_CACHE_SHIFT); 438 if (!err) 439 err = err2; 440 } 441 } 442 return err; 443} 444 445/** 446 * add_to_page_cache_locked - add a locked page to the pagecache 447 * @page: page to add 448 * @mapping: the page's address_space 449 * @offset: page index 450 * @gfp_mask: page allocation mode 451 * 452 * This function is used to add a page to the pagecache. It must be locked. 453 * This function does not add the page to the LRU. The caller must do that. 454 */ 455int add_to_page_cache_locked(struct page *page, struct address_space *mapping, 456 pgoff_t offset, gfp_t gfp_mask) 457{ 458 int error; 459 460 VM_BUG_ON(!PageLocked(page)); 461 462 error = mem_cgroup_cache_charge(page, current->mm, 463 gfp_mask & GFP_RECLAIM_MASK); 464 if (error) 465 goto out; 466 467 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); 468 if (error == 0) { 469 page_cache_get(page); 470 page->mapping = mapping; 471 page->index = offset; 472 473 spin_lock_irq(&mapping->tree_lock); 474 error = radix_tree_insert(&mapping->page_tree, offset, page); 475 if (likely(!error)) { 476 mapping->nrpages++; 477 __inc_zone_page_state(page, NR_FILE_PAGES); 478 } else { 479 page->mapping = NULL; 480 mem_cgroup_uncharge_cache_page(page); 481 page_cache_release(page); 482 } 483 484 spin_unlock_irq(&mapping->tree_lock); 485 radix_tree_preload_end(); 486 } else 487 mem_cgroup_uncharge_cache_page(page); 488out: 489 return error; 490} 491EXPORT_SYMBOL(add_to_page_cache_locked); 492 493int add_to_page_cache_lru(struct page *page, struct address_space *mapping, 494 pgoff_t offset, gfp_t gfp_mask) 495{ 496 int ret; 497 498 /* 499 * Splice_read and readahead add shmem/tmpfs pages into the page cache 500 * before shmem_readpage has a chance to mark them as SwapBacked: they 501 * need to go on the active_anon lru below, and mem_cgroup_cache_charge 502 * (called in add_to_page_cache) needs to know where they're going too. 503 */ 504 if (mapping_cap_swap_backed(mapping)) 505 SetPageSwapBacked(page); 506 507 ret = add_to_page_cache(page, mapping, offset, gfp_mask); 508 if (ret == 0) { 509 if (page_is_file_cache(page)) 510 lru_cache_add_file(page); 511 else 512 lru_cache_add_active_anon(page); 513 } 514 return ret; 515} 516 517#ifdef CONFIG_NUMA 518struct page *__page_cache_alloc(gfp_t gfp) 519{ 520 if (cpuset_do_page_mem_spread()) { 521 int n = cpuset_mem_spread_node(); 522 return alloc_pages_node(n, gfp, 0); 523 } 524 return alloc_pages(gfp, 0); 525} 526EXPORT_SYMBOL(__page_cache_alloc); 527#endif 528 529static int __sleep_on_page_lock(void *word) 530{ 531 io_schedule(); 532 return 0; 533} 534 535/* 536 * In order to wait for pages to become available there must be 537 * waitqueues associated with pages. By using a hash table of 538 * waitqueues where the bucket discipline is to maintain all 539 * waiters on the same queue and wake all when any of the pages 540 * become available, and for the woken contexts to check to be 541 * sure the appropriate page became available, this saves space 542 * at a cost of "thundering herd" phenomena during rare hash 543 * collisions. 544 */ 545static wait_queue_head_t *page_waitqueue(struct page *page) 546{ 547 const struct zone *zone = page_zone(page); 548 549 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; 550} 551 552static inline void wake_up_page(struct page *page, int bit) 553{ 554 __wake_up_bit(page_waitqueue(page), &page->flags, bit); 555} 556 557void wait_on_page_bit(struct page *page, int bit_nr) 558{ 559 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); 560 561 if (test_bit(bit_nr, &page->flags)) 562 __wait_on_bit(page_waitqueue(page), &wait, sync_page, 563 TASK_UNINTERRUPTIBLE); 564} 565EXPORT_SYMBOL(wait_on_page_bit); 566 567/** 568 * unlock_page - unlock a locked page 569 * @page: the page 570 * 571 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). 572 * Also wakes sleepers in wait_on_page_writeback() because the wakeup 573 * mechananism between PageLocked pages and PageWriteback pages is shared. 574 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. 575 * 576 * The mb is necessary to enforce ordering between the clear_bit and the read 577 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()). 578 */ 579void unlock_page(struct page *page) 580{ 581 VM_BUG_ON(!PageLocked(page)); 582 clear_bit_unlock(PG_locked, &page->flags); 583 smp_mb__after_clear_bit(); 584 wake_up_page(page, PG_locked); 585} 586EXPORT_SYMBOL(unlock_page); 587 588/** 589 * end_page_writeback - end writeback against a page 590 * @page: the page 591 */ 592void end_page_writeback(struct page *page) 593{ 594 if (TestClearPageReclaim(page)) 595 rotate_reclaimable_page(page); 596 597 if (!test_clear_page_writeback(page)) 598 BUG(); 599 600 smp_mb__after_clear_bit(); 601 wake_up_page(page, PG_writeback); 602} 603EXPORT_SYMBOL(end_page_writeback); 604 605/** 606 * __lock_page - get a lock on the page, assuming we need to sleep to get it 607 * @page: the page to lock 608 * 609 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some 610 * random driver's requestfn sets TASK_RUNNING, we could busywait. However 611 * chances are that on the second loop, the block layer's plug list is empty, 612 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE. 613 */ 614void __lock_page(struct page *page) 615{ 616 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); 617 618 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page, 619 TASK_UNINTERRUPTIBLE); 620} 621EXPORT_SYMBOL(__lock_page); 622 623int __lock_page_killable(struct page *page) 624{ 625 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); 626 627 return __wait_on_bit_lock(page_waitqueue(page), &wait, 628 sync_page_killable, TASK_KILLABLE); 629} 630 631/** 632 * __lock_page_nosync - get a lock on the page, without calling sync_page() 633 * @page: the page to lock 634 * 635 * Variant of lock_page that does not require the caller to hold a reference 636 * on the page's mapping. 637 */ 638void __lock_page_nosync(struct page *page) 639{ 640 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); 641 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock, 642 TASK_UNINTERRUPTIBLE); 643} 644 645/** 646 * find_get_page - find and get a page reference 647 * @mapping: the address_space to search 648 * @offset: the page index 649 * 650 * Is there a pagecache struct page at the given (mapping, offset) tuple? 651 * If yes, increment its refcount and return it; if no, return NULL. 652 */ 653struct page *find_get_page(struct address_space *mapping, pgoff_t offset) 654{ 655 void **pagep; 656 struct page *page; 657 658 rcu_read_lock(); 659repeat: 660 page = NULL; 661 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); 662 if (pagep) { 663 page = radix_tree_deref_slot(pagep); 664 if (unlikely(!page || page == RADIX_TREE_RETRY)) 665 goto repeat; 666 667 if (!page_cache_get_speculative(page)) 668 goto repeat; 669 670 /* 671 * Has the page moved? 672 * This is part of the lockless pagecache protocol. See 673 * include/linux/pagemap.h for details. 674 */ 675 if (unlikely(page != *pagep)) { 676 page_cache_release(page); 677 goto repeat; 678 } 679 } 680 rcu_read_unlock(); 681 682 return page; 683} 684EXPORT_SYMBOL(find_get_page); 685 686/** 687 * find_lock_page - locate, pin and lock a pagecache page 688 * @mapping: the address_space to search 689 * @offset: the page index 690 * 691 * Locates the desired pagecache page, locks it, increments its reference 692 * count and returns its address. 693 * 694 * Returns zero if the page was not present. find_lock_page() may sleep. 695 */ 696struct page *find_lock_page(struct address_space *mapping, pgoff_t offset) 697{ 698 struct page *page; 699 700repeat: 701 page = find_get_page(mapping, offset); 702 if (page) { 703 lock_page(page); 704 /* Has the page been truncated? */ 705 if (unlikely(page->mapping != mapping)) { 706 unlock_page(page); 707 page_cache_release(page); 708 goto repeat; 709 } 710 VM_BUG_ON(page->index != offset); 711 } 712 return page; 713} 714EXPORT_SYMBOL(find_lock_page); 715 716/** 717 * find_or_create_page - locate or add a pagecache page 718 * @mapping: the page's address_space 719 * @index: the page's index into the mapping 720 * @gfp_mask: page allocation mode 721 * 722 * Locates a page in the pagecache. If the page is not present, a new page 723 * is allocated using @gfp_mask and is added to the pagecache and to the VM's 724 * LRU list. The returned page is locked and has its reference count 725 * incremented. 726 * 727 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic 728 * allocation! 729 * 730 * find_or_create_page() returns the desired page's address, or zero on 731 * memory exhaustion. 732 */ 733struct page *find_or_create_page(struct address_space *mapping, 734 pgoff_t index, gfp_t gfp_mask) 735{ 736 struct page *page; 737 int err; 738repeat: 739 page = find_lock_page(mapping, index); 740 if (!page) { 741 page = __page_cache_alloc(gfp_mask); 742 if (!page) 743 return NULL; 744 /* 745 * We want a regular kernel memory (not highmem or DMA etc) 746 * allocation for the radix tree nodes, but we need to honour 747 * the context-specific requirements the caller has asked for. 748 * GFP_RECLAIM_MASK collects those requirements. 749 */ 750 err = add_to_page_cache_lru(page, mapping, index, 751 (gfp_mask & GFP_RECLAIM_MASK)); 752 if (unlikely(err)) { 753 page_cache_release(page); 754 page = NULL; 755 if (err == -EEXIST) 756 goto repeat; 757 } 758 } 759 return page; 760} 761EXPORT_SYMBOL(find_or_create_page); 762 763/** 764 * find_get_pages - gang pagecache lookup 765 * @mapping: The address_space to search 766 * @start: The starting page index 767 * @nr_pages: The maximum number of pages 768 * @pages: Where the resulting pages are placed 769 * 770 * find_get_pages() will search for and return a group of up to 771 * @nr_pages pages in the mapping. The pages are placed at @pages. 772 * find_get_pages() takes a reference against the returned pages. 773 * 774 * The search returns a group of mapping-contiguous pages with ascending 775 * indexes. There may be holes in the indices due to not-present pages. 776 * 777 * find_get_pages() returns the number of pages which were found. 778 */ 779unsigned find_get_pages(struct address_space *mapping, pgoff_t start, 780 unsigned int nr_pages, struct page **pages) 781{ 782 unsigned int i; 783 unsigned int ret; 784 unsigned int nr_found; 785 786 rcu_read_lock(); 787restart: 788 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree, 789 (void ***)pages, start, nr_pages); 790 ret = 0; 791 for (i = 0; i < nr_found; i++) { 792 struct page *page; 793repeat: 794 page = radix_tree_deref_slot((void **)pages[i]); 795 if (unlikely(!page)) 796 continue; 797 /* 798 * this can only trigger if nr_found == 1, making livelock 799 * a non issue. 800 */ 801 if (unlikely(page == RADIX_TREE_RETRY)) 802 goto restart; 803 804 if (!page_cache_get_speculative(page)) 805 goto repeat; 806 807 /* Has the page moved? */ 808 if (unlikely(page != *((void **)pages[i]))) { 809 page_cache_release(page); 810 goto repeat; 811 } 812 813 pages[ret] = page; 814 ret++; 815 } 816 rcu_read_unlock(); 817 return ret; 818} 819 820/** 821 * find_get_pages_contig - gang contiguous pagecache lookup 822 * @mapping: The address_space to search 823 * @index: The starting page index 824 * @nr_pages: The maximum number of pages 825 * @pages: Where the resulting pages are placed 826 * 827 * find_get_pages_contig() works exactly like find_get_pages(), except 828 * that the returned number of pages are guaranteed to be contiguous. 829 * 830 * find_get_pages_contig() returns the number of pages which were found. 831 */ 832unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, 833 unsigned int nr_pages, struct page **pages) 834{ 835 unsigned int i; 836 unsigned int ret; 837 unsigned int nr_found; 838 839 rcu_read_lock(); 840restart: 841 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree, 842 (void ***)pages, index, nr_pages); 843 ret = 0; 844 for (i = 0; i < nr_found; i++) { 845 struct page *page; 846repeat: 847 page = radix_tree_deref_slot((void **)pages[i]); 848 if (unlikely(!page)) 849 continue; 850 /* 851 * this can only trigger if nr_found == 1, making livelock 852 * a non issue. 853 */ 854 if (unlikely(page == RADIX_TREE_RETRY)) 855 goto restart; 856 857 if (page->mapping == NULL || page->index != index) 858 break; 859 860 if (!page_cache_get_speculative(page)) 861 goto repeat; 862 863 /* Has the page moved? */ 864 if (unlikely(page != *((void **)pages[i]))) { 865 page_cache_release(page); 866 goto repeat; 867 } 868 869 pages[ret] = page; 870 ret++; 871 index++; 872 } 873 rcu_read_unlock(); 874 return ret; 875} 876EXPORT_SYMBOL(find_get_pages_contig); 877 878/** 879 * find_get_pages_tag - find and return pages that match @tag 880 * @mapping: the address_space to search 881 * @index: the starting page index 882 * @tag: the tag index 883 * @nr_pages: the maximum number of pages 884 * @pages: where the resulting pages are placed 885 * 886 * Like find_get_pages, except we only return pages which are tagged with 887 * @tag. We update @index to index the next page for the traversal. 888 */ 889unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, 890 int tag, unsigned int nr_pages, struct page **pages) 891{ 892 unsigned int i; 893 unsigned int ret; 894 unsigned int nr_found; 895 896 rcu_read_lock(); 897restart: 898 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree, 899 (void ***)pages, *index, nr_pages, tag); 900 ret = 0; 901 for (i = 0; i < nr_found; i++) { 902 struct page *page; 903repeat: 904 page = radix_tree_deref_slot((void **)pages[i]); 905 if (unlikely(!page)) 906 continue; 907 /* 908 * this can only trigger if nr_found == 1, making livelock 909 * a non issue. 910 */ 911 if (unlikely(page == RADIX_TREE_RETRY)) 912 goto restart; 913 914 if (!page_cache_get_speculative(page)) 915 goto repeat; 916 917 /* Has the page moved? */ 918 if (unlikely(page != *((void **)pages[i]))) { 919 page_cache_release(page); 920 goto repeat; 921 } 922 923 pages[ret] = page; 924 ret++; 925 } 926 rcu_read_unlock(); 927 928 if (ret) 929 *index = pages[ret - 1]->index + 1; 930 931 return ret; 932} 933EXPORT_SYMBOL(find_get_pages_tag); 934 935/** 936 * grab_cache_page_nowait - returns locked page at given index in given cache 937 * @mapping: target address_space 938 * @index: the page index 939 * 940 * Same as grab_cache_page(), but do not wait if the page is unavailable. 941 * This is intended for speculative data generators, where the data can 942 * be regenerated if the page couldn't be grabbed. This routine should 943 * be safe to call while holding the lock for another page. 944 * 945 * Clear __GFP_FS when allocating the page to avoid recursion into the fs 946 * and deadlock against the caller's locked page. 947 */ 948struct page * 949grab_cache_page_nowait(struct address_space *mapping, pgoff_t index) 950{ 951 struct page *page = find_get_page(mapping, index); 952 953 if (page) { 954 if (trylock_page(page)) 955 return page; 956 page_cache_release(page); 957 return NULL; 958 } 959 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS); 960 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) { 961 page_cache_release(page); 962 page = NULL; 963 } 964 return page; 965} 966EXPORT_SYMBOL(grab_cache_page_nowait); 967 968/* 969 * CD/DVDs are error prone. When a medium error occurs, the driver may fail 970 * a _large_ part of the i/o request. Imagine the worst scenario: 971 * 972 * ---R__________________________________________B__________ 973 * ^ reading here ^ bad block(assume 4k) 974 * 975 * read(R) => miss => readahead(R...B) => media error => frustrating retries 976 * => failing the whole request => read(R) => read(R+1) => 977 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => 978 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => 979 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... 980 * 981 * It is going insane. Fix it by quickly scaling down the readahead size. 982 */ 983static void shrink_readahead_size_eio(struct file *filp, 984 struct file_ra_state *ra) 985{ 986 if (!ra->ra_pages) 987 return; 988 989 ra->ra_pages /= 4; 990} 991 992/** 993 * do_generic_file_read - generic file read routine 994 * @filp: the file to read 995 * @ppos: current file position 996 * @desc: read_descriptor 997 * @actor: read method 998 * 999 * This is a generic file read routine, and uses the 1000 * mapping->a_ops->readpage() function for the actual low-level stuff. 1001 * 1002 * This is really ugly. But the goto's actually try to clarify some 1003 * of the logic when it comes to error handling etc. 1004 */ 1005static void do_generic_file_read(struct file *filp, loff_t *ppos, 1006 read_descriptor_t *desc, read_actor_t actor) 1007{ 1008 struct address_space *mapping = filp->f_mapping; 1009 struct inode *inode = mapping->host; 1010 struct file_ra_state *ra = &filp->f_ra; 1011 pgoff_t index; 1012 pgoff_t last_index; 1013 pgoff_t prev_index; 1014 unsigned long offset; /* offset into pagecache page */ 1015 unsigned int prev_offset; 1016 int error; 1017 1018 index = *ppos >> PAGE_CACHE_SHIFT; 1019 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT; 1020 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1); 1021 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; 1022 offset = *ppos & ~PAGE_CACHE_MASK; 1023 1024 for (;;) { 1025 struct page *page; 1026 pgoff_t end_index; 1027 loff_t isize; 1028 unsigned long nr, ret; 1029 1030 cond_resched(); 1031find_page: 1032 page = find_get_page(mapping, index); 1033 if (!page) { 1034 page_cache_sync_readahead(mapping, 1035 ra, filp, 1036 index, last_index - index); 1037 page = find_get_page(mapping, index); 1038 if (unlikely(page == NULL)) 1039 goto no_cached_page; 1040 } 1041 if (PageReadahead(page)) { 1042 page_cache_async_readahead(mapping, 1043 ra, filp, page, 1044 index, last_index - index); 1045 } 1046 if (!PageUptodate(page)) { 1047 if (inode->i_blkbits == PAGE_CACHE_SHIFT || 1048 !mapping->a_ops->is_partially_uptodate) 1049 goto page_not_up_to_date; 1050 if (!trylock_page(page)) 1051 goto page_not_up_to_date; 1052 if (!mapping->a_ops->is_partially_uptodate(page, 1053 desc, offset)) 1054 goto page_not_up_to_date_locked; 1055 unlock_page(page); 1056 } 1057page_ok: 1058 /* 1059 * i_size must be checked after we know the page is Uptodate. 1060 * 1061 * Checking i_size after the check allows us to calculate 1062 * the correct value for "nr", which means the zero-filled 1063 * part of the page is not copied back to userspace (unless 1064 * another truncate extends the file - this is desired though). 1065 */ 1066 1067 isize = i_size_read(inode); 1068 end_index = (isize - 1) >> PAGE_CACHE_SHIFT; 1069 if (unlikely(!isize || index > end_index)) { 1070 page_cache_release(page); 1071 goto out; 1072 } 1073 1074 /* nr is the maximum number of bytes to copy from this page */ 1075 nr = PAGE_CACHE_SIZE; 1076 if (index == end_index) { 1077 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; 1078 if (nr <= offset) { 1079 page_cache_release(page); 1080 goto out; 1081 } 1082 } 1083 nr = nr - offset; 1084 1085 /* If users can be writing to this page using arbitrary 1086 * virtual addresses, take care about potential aliasing 1087 * before reading the page on the kernel side. 1088 */ 1089 if (mapping_writably_mapped(mapping)) 1090 flush_dcache_page(page); 1091 1092 /* 1093 * When a sequential read accesses a page several times, 1094 * only mark it as accessed the first time. 1095 */ 1096 if (prev_index != index || offset != prev_offset) 1097 mark_page_accessed(page); 1098 prev_index = index; 1099 1100 /* 1101 * Ok, we have the page, and it's up-to-date, so 1102 * now we can copy it to user space... 1103 * 1104 * The actor routine returns how many bytes were actually used.. 1105 * NOTE! This may not be the same as how much of a user buffer 1106 * we filled up (we may be padding etc), so we can only update 1107 * "pos" here (the actor routine has to update the user buffer 1108 * pointers and the remaining count). 1109 */ 1110 ret = actor(desc, page, offset, nr); 1111 offset += ret; 1112 index += offset >> PAGE_CACHE_SHIFT; 1113 offset &= ~PAGE_CACHE_MASK; 1114 prev_offset = offset; 1115 1116 page_cache_release(page); 1117 if (ret == nr && desc->count) 1118 continue; 1119 goto out; 1120 1121page_not_up_to_date: 1122 /* Get exclusive access to the page ... */ 1123 error = lock_page_killable(page); 1124 if (unlikely(error)) 1125 goto readpage_error; 1126 1127page_not_up_to_date_locked: 1128 /* Did it get truncated before we got the lock? */ 1129 if (!page->mapping) { 1130 unlock_page(page); 1131 page_cache_release(page); 1132 continue; 1133 } 1134 1135 /* Did somebody else fill it already? */ 1136 if (PageUptodate(page)) { 1137 unlock_page(page); 1138 goto page_ok; 1139 } 1140 1141readpage: 1142 /* Start the actual read. The read will unlock the page. */ 1143 error = mapping->a_ops->readpage(filp, page); 1144 1145 if (unlikely(error)) { 1146 if (error == AOP_TRUNCATED_PAGE) { 1147 page_cache_release(page); 1148 goto find_page; 1149 } 1150 goto readpage_error; 1151 } 1152 1153 if (!PageUptodate(page)) { 1154 error = lock_page_killable(page); 1155 if (unlikely(error)) 1156 goto readpage_error; 1157 if (!PageUptodate(page)) { 1158 if (page->mapping == NULL) { 1159 /* 1160 * invalidate_inode_pages got it 1161 */ 1162 unlock_page(page); 1163 page_cache_release(page); 1164 goto find_page; 1165 } 1166 unlock_page(page); 1167 shrink_readahead_size_eio(filp, ra); 1168 error = -EIO; 1169 goto readpage_error; 1170 } 1171 unlock_page(page); 1172 } 1173 1174 goto page_ok; 1175 1176readpage_error: 1177 /* UHHUH! A synchronous read error occurred. Report it */ 1178 desc->error = error; 1179 page_cache_release(page); 1180 goto out; 1181 1182no_cached_page: 1183 /* 1184 * Ok, it wasn't cached, so we need to create a new 1185 * page.. 1186 */ 1187 page = page_cache_alloc_cold(mapping); 1188 if (!page) { 1189 desc->error = -ENOMEM; 1190 goto out; 1191 } 1192 error = add_to_page_cache_lru(page, mapping, 1193 index, GFP_KERNEL); 1194 if (error) { 1195 page_cache_release(page); 1196 if (error == -EEXIST) 1197 goto find_page; 1198 desc->error = error; 1199 goto out; 1200 } 1201 goto readpage; 1202 } 1203 1204out: 1205 ra->prev_pos = prev_index; 1206 ra->prev_pos <<= PAGE_CACHE_SHIFT; 1207 ra->prev_pos |= prev_offset; 1208 1209 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset; 1210 file_accessed(filp); 1211} 1212 1213int file_read_actor(read_descriptor_t *desc, struct page *page, 1214 unsigned long offset, unsigned long size) 1215{ 1216 char *kaddr; 1217 unsigned long left, count = desc->count; 1218 1219 if (size > count) 1220 size = count; 1221 1222 /* 1223 * Faults on the destination of a read are common, so do it before 1224 * taking the kmap. 1225 */ 1226 if (!fault_in_pages_writeable(desc->arg.buf, size)) { 1227 kaddr = kmap_atomic(page, KM_USER0); 1228 left = __copy_to_user_inatomic(desc->arg.buf, 1229 kaddr + offset, size); 1230 kunmap_atomic(kaddr, KM_USER0); 1231 if (left == 0) 1232 goto success; 1233 } 1234 1235 /* Do it the slow way */ 1236 kaddr = kmap(page); 1237 left = __copy_to_user(desc->arg.buf, kaddr + offset, size); 1238 kunmap(page); 1239 1240 if (left) { 1241 size -= left; 1242 desc->error = -EFAULT; 1243 } 1244success: 1245 desc->count = count - size; 1246 desc->written += size; 1247 desc->arg.buf += size; 1248 return size; 1249} 1250 1251/* 1252 * Performs necessary checks before doing a write 1253 * @iov: io vector request 1254 * @nr_segs: number of segments in the iovec 1255 * @count: number of bytes to write 1256 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE 1257 * 1258 * Adjust number of segments and amount of bytes to write (nr_segs should be 1259 * properly initialized first). Returns appropriate error code that caller 1260 * should return or zero in case that write should be allowed. 1261 */ 1262int generic_segment_checks(const struct iovec *iov, 1263 unsigned long *nr_segs, size_t *count, int access_flags) 1264{ 1265 unsigned long seg; 1266 size_t cnt = 0; 1267 for (seg = 0; seg < *nr_segs; seg++) { 1268 const struct iovec *iv = &iov[seg]; 1269 1270 /* 1271 * If any segment has a negative length, or the cumulative 1272 * length ever wraps negative then return -EINVAL. 1273 */ 1274 cnt += iv->iov_len; 1275 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0)) 1276 return -EINVAL; 1277 if (access_ok(access_flags, iv->iov_base, iv->iov_len)) 1278 continue; 1279 if (seg == 0) 1280 return -EFAULT; 1281 *nr_segs = seg; 1282 cnt -= iv->iov_len; /* This segment is no good */ 1283 break; 1284 } 1285 *count = cnt; 1286 return 0; 1287} 1288EXPORT_SYMBOL(generic_segment_checks); 1289 1290/** 1291 * generic_file_aio_read - generic filesystem read routine 1292 * @iocb: kernel I/O control block 1293 * @iov: io vector request 1294 * @nr_segs: number of segments in the iovec 1295 * @pos: current file position 1296 * 1297 * This is the "read()" routine for all filesystems 1298 * that can use the page cache directly. 1299 */ 1300ssize_t 1301generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov, 1302 unsigned long nr_segs, loff_t pos) 1303{ 1304 struct file *filp = iocb->ki_filp; 1305 ssize_t retval; 1306 unsigned long seg; 1307 size_t count; 1308 loff_t *ppos = &iocb->ki_pos; 1309 1310 count = 0; 1311 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); 1312 if (retval) 1313 return retval; 1314 1315 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ 1316 if (filp->f_flags & O_DIRECT) { 1317 loff_t size; 1318 struct address_space *mapping; 1319 struct inode *inode; 1320 1321 mapping = filp->f_mapping; 1322 inode = mapping->host; 1323 if (!count) 1324 goto out; /* skip atime */ 1325 size = i_size_read(inode); 1326 if (pos < size) { 1327 retval = filemap_write_and_wait_range(mapping, pos, 1328 pos + iov_length(iov, nr_segs) - 1); 1329 if (!retval) { 1330 retval = mapping->a_ops->direct_IO(READ, iocb, 1331 iov, pos, nr_segs); 1332 } 1333 if (retval > 0) 1334 *ppos = pos + retval; 1335 if (retval) { 1336 file_accessed(filp); 1337 goto out; 1338 } 1339 } 1340 } 1341 1342 for (seg = 0; seg < nr_segs; seg++) { 1343 read_descriptor_t desc; 1344 1345 desc.written = 0; 1346 desc.arg.buf = iov[seg].iov_base; 1347 desc.count = iov[seg].iov_len; 1348 if (desc.count == 0) 1349 continue; 1350 desc.error = 0; 1351 do_generic_file_read(filp, ppos, &desc, file_read_actor); 1352 retval += desc.written; 1353 if (desc.error) { 1354 retval = retval ?: desc.error; 1355 break; 1356 } 1357 if (desc.count > 0) 1358 break; 1359 } 1360out: 1361 return retval; 1362} 1363EXPORT_SYMBOL(generic_file_aio_read); 1364 1365static ssize_t 1366do_readahead(struct address_space *mapping, struct file *filp, 1367 pgoff_t index, unsigned long nr) 1368{ 1369 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage) 1370 return -EINVAL; 1371 1372 force_page_cache_readahead(mapping, filp, index, 1373 max_sane_readahead(nr)); 1374 return 0; 1375} 1376 1377SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count) 1378{ 1379 ssize_t ret; 1380 struct file *file; 1381 1382 ret = -EBADF; 1383 file = fget(fd); 1384 if (file) { 1385 if (file->f_mode & FMODE_READ) { 1386 struct address_space *mapping = file->f_mapping; 1387 pgoff_t start = offset >> PAGE_CACHE_SHIFT; 1388 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT; 1389 unsigned long len = end - start + 1; 1390 ret = do_readahead(mapping, file, start, len); 1391 } 1392 fput(file); 1393 } 1394 return ret; 1395} 1396#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS 1397asmlinkage long SyS_readahead(long fd, loff_t offset, long count) 1398{ 1399 return SYSC_readahead((int) fd, offset, (size_t) count); 1400} 1401SYSCALL_ALIAS(sys_readahead, SyS_readahead); 1402#endif 1403 1404#ifdef CONFIG_MMU 1405/** 1406 * page_cache_read - adds requested page to the page cache if not already there 1407 * @file: file to read 1408 * @offset: page index 1409 * 1410 * This adds the requested page to the page cache if it isn't already there, 1411 * and schedules an I/O to read in its contents from disk. 1412 */ 1413static int page_cache_read(struct file *file, pgoff_t offset) 1414{ 1415 struct address_space *mapping = file->f_mapping; 1416 struct page *page; 1417 int ret; 1418 1419 do { 1420 page = page_cache_alloc_cold(mapping); 1421 if (!page) 1422 return -ENOMEM; 1423 1424 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL); 1425 if (ret == 0) 1426 ret = mapping->a_ops->readpage(file, page); 1427 else if (ret == -EEXIST) 1428 ret = 0; /* losing race to add is OK */ 1429 1430 page_cache_release(page); 1431 1432 } while (ret == AOP_TRUNCATED_PAGE); 1433 1434 return ret; 1435} 1436 1437#define MMAP_LOTSAMISS (100) 1438 1439/** 1440 * filemap_fault - read in file data for page fault handling 1441 * @vma: vma in which the fault was taken 1442 * @vmf: struct vm_fault containing details of the fault 1443 * 1444 * filemap_fault() is invoked via the vma operations vector for a 1445 * mapped memory region to read in file data during a page fault. 1446 * 1447 * The goto's are kind of ugly, but this streamlines the normal case of having 1448 * it in the page cache, and handles the special cases reasonably without 1449 * having a lot of duplicated code. 1450 */ 1451int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1452{ 1453 int error; 1454 struct file *file = vma->vm_file; 1455 struct address_space *mapping = file->f_mapping; 1456 struct file_ra_state *ra = &file->f_ra; 1457 struct inode *inode = mapping->host; 1458 struct page *page; 1459 pgoff_t size; 1460 int did_readaround = 0; 1461 int ret = 0; 1462 1463 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 1464 if (vmf->pgoff >= size) 1465 return VM_FAULT_SIGBUS; 1466 1467 /* If we don't want any read-ahead, don't bother */ 1468 if (VM_RandomReadHint(vma)) 1469 goto no_cached_page; 1470 1471 /* 1472 * Do we have something in the page cache already? 1473 */ 1474retry_find: 1475 page = find_lock_page(mapping, vmf->pgoff); 1476 /* 1477 * For sequential accesses, we use the generic readahead logic. 1478 */ 1479 if (VM_SequentialReadHint(vma)) { 1480 if (!page) { 1481 page_cache_sync_readahead(mapping, ra, file, 1482 vmf->pgoff, 1); 1483 page = find_lock_page(mapping, vmf->pgoff); 1484 if (!page) 1485 goto no_cached_page; 1486 } 1487 if (PageReadahead(page)) { 1488 page_cache_async_readahead(mapping, ra, file, page, 1489 vmf->pgoff, 1); 1490 } 1491 } 1492 1493 if (!page) { 1494 unsigned long ra_pages; 1495 1496 ra->mmap_miss++; 1497 1498 /* 1499 * Do we miss much more than hit in this file? If so, 1500 * stop bothering with read-ahead. It will only hurt. 1501 */ 1502 if (ra->mmap_miss > MMAP_LOTSAMISS) 1503 goto no_cached_page; 1504 1505 /* 1506 * To keep the pgmajfault counter straight, we need to 1507 * check did_readaround, as this is an inner loop. 1508 */ 1509 if (!did_readaround) { 1510 ret = VM_FAULT_MAJOR; 1511 count_vm_event(PGMAJFAULT); 1512 } 1513 did_readaround = 1; 1514 ra_pages = max_sane_readahead(file->f_ra.ra_pages); 1515 if (ra_pages) { 1516 pgoff_t start = 0; 1517 1518 if (vmf->pgoff > ra_pages / 2) 1519 start = vmf->pgoff - ra_pages / 2; 1520 do_page_cache_readahead(mapping, file, start, ra_pages); 1521 } 1522 page = find_lock_page(mapping, vmf->pgoff); 1523 if (!page) 1524 goto no_cached_page; 1525 } 1526 1527 if (!did_readaround) 1528 ra->mmap_miss--; 1529 1530 /* 1531 * We have a locked page in the page cache, now we need to check 1532 * that it's up-to-date. If not, it is going to be due to an error. 1533 */ 1534 if (unlikely(!PageUptodate(page))) 1535 goto page_not_uptodate; 1536 1537 /* Must recheck i_size under page lock */ 1538 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 1539 if (unlikely(vmf->pgoff >= size)) { 1540 unlock_page(page); 1541 page_cache_release(page); 1542 return VM_FAULT_SIGBUS; 1543 } 1544 1545 /* 1546 * Found the page and have a reference on it. 1547 */ 1548 ra->prev_pos = (loff_t)page->index << PAGE_CACHE_SHIFT; 1549 vmf->page = page; 1550 return ret | VM_FAULT_LOCKED; 1551 1552no_cached_page: 1553 /* 1554 * We're only likely to ever get here if MADV_RANDOM is in 1555 * effect. 1556 */ 1557 error = page_cache_read(file, vmf->pgoff); 1558 1559 /* 1560 * The page we want has now been added to the page cache. 1561 * In the unlikely event that someone removed it in the 1562 * meantime, we'll just come back here and read it again. 1563 */ 1564 if (error >= 0) 1565 goto retry_find; 1566 1567 /* 1568 * An error return from page_cache_read can result if the 1569 * system is low on memory, or a problem occurs while trying 1570 * to schedule I/O. 1571 */ 1572 if (error == -ENOMEM) 1573 return VM_FAULT_OOM; 1574 return VM_FAULT_SIGBUS; 1575 1576page_not_uptodate: 1577 /* IO error path */ 1578 if (!did_readaround) { 1579 ret = VM_FAULT_MAJOR; 1580 count_vm_event(PGMAJFAULT); 1581 } 1582 1583 /* 1584 * Umm, take care of errors if the page isn't up-to-date. 1585 * Try to re-read it _once_. We do this synchronously, 1586 * because there really aren't any performance issues here 1587 * and we need to check for errors. 1588 */ 1589 ClearPageError(page); 1590 error = mapping->a_ops->readpage(file, page); 1591 if (!error) { 1592 wait_on_page_locked(page); 1593 if (!PageUptodate(page)) 1594 error = -EIO; 1595 } 1596 page_cache_release(page); 1597 1598 if (!error || error == AOP_TRUNCATED_PAGE) 1599 goto retry_find; 1600 1601 /* Things didn't work out. Return zero to tell the mm layer so. */ 1602 shrink_readahead_size_eio(file, ra); 1603 return VM_FAULT_SIGBUS; 1604} 1605EXPORT_SYMBOL(filemap_fault); 1606 1607struct vm_operations_struct generic_file_vm_ops = { 1608 .fault = filemap_fault, 1609}; 1610 1611/* This is used for a general mmap of a disk file */ 1612 1613int generic_file_mmap(struct file * file, struct vm_area_struct * vma) 1614{ 1615 struct address_space *mapping = file->f_mapping; 1616 1617 if (!mapping->a_ops->readpage) 1618 return -ENOEXEC; 1619 file_accessed(file); 1620 vma->vm_ops = &generic_file_vm_ops; 1621 vma->vm_flags |= VM_CAN_NONLINEAR; 1622 return 0; 1623} 1624 1625/* 1626 * This is for filesystems which do not implement ->writepage. 1627 */ 1628int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) 1629{ 1630 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) 1631 return -EINVAL; 1632 return generic_file_mmap(file, vma); 1633} 1634#else 1635int generic_file_mmap(struct file * file, struct vm_area_struct * vma) 1636{ 1637 return -ENOSYS; 1638} 1639int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) 1640{ 1641 return -ENOSYS; 1642} 1643#endif /* CONFIG_MMU */ 1644 1645EXPORT_SYMBOL(generic_file_mmap); 1646EXPORT_SYMBOL(generic_file_readonly_mmap); 1647 1648static struct page *__read_cache_page(struct address_space *mapping, 1649 pgoff_t index, 1650 int (*filler)(void *,struct page*), 1651 void *data) 1652{ 1653 struct page *page; 1654 int err; 1655repeat: 1656 page = find_get_page(mapping, index); 1657 if (!page) { 1658 page = page_cache_alloc_cold(mapping); 1659 if (!page) 1660 return ERR_PTR(-ENOMEM); 1661 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL); 1662 if (unlikely(err)) { 1663 page_cache_release(page); 1664 if (err == -EEXIST) 1665 goto repeat; 1666 /* Presumably ENOMEM for radix tree node */ 1667 return ERR_PTR(err); 1668 } 1669 err = filler(data, page); 1670 if (err < 0) { 1671 page_cache_release(page); 1672 page = ERR_PTR(err); 1673 } 1674 } 1675 return page; 1676} 1677 1678/** 1679 * read_cache_page_async - read into page cache, fill it if needed 1680 * @mapping: the page's address_space 1681 * @index: the page index 1682 * @filler: function to perform the read 1683 * @data: destination for read data 1684 * 1685 * Same as read_cache_page, but don't wait for page to become unlocked 1686 * after submitting it to the filler. 1687 * 1688 * Read into the page cache. If a page already exists, and PageUptodate() is 1689 * not set, try to fill the page but don't wait for it to become unlocked. 1690 * 1691 * If the page does not get brought uptodate, return -EIO. 1692 */ 1693struct page *read_cache_page_async(struct address_space *mapping, 1694 pgoff_t index, 1695 int (*filler)(void *,struct page*), 1696 void *data) 1697{ 1698 struct page *page; 1699 int err; 1700 1701retry: 1702 page = __read_cache_page(mapping, index, filler, data); 1703 if (IS_ERR(page)) 1704 return page; 1705 if (PageUptodate(page)) 1706 goto out; 1707 1708 lock_page(page); 1709 if (!page->mapping) { 1710 unlock_page(page); 1711 page_cache_release(page); 1712 goto retry; 1713 } 1714 if (PageUptodate(page)) { 1715 unlock_page(page); 1716 goto out; 1717 } 1718 err = filler(data, page); 1719 if (err < 0) { 1720 page_cache_release(page); 1721 return ERR_PTR(err); 1722 } 1723out: 1724 mark_page_accessed(page); 1725 return page; 1726} 1727EXPORT_SYMBOL(read_cache_page_async); 1728 1729/** 1730 * read_cache_page - read into page cache, fill it if needed 1731 * @mapping: the page's address_space 1732 * @index: the page index 1733 * @filler: function to perform the read 1734 * @data: destination for read data 1735 * 1736 * Read into the page cache. If a page already exists, and PageUptodate() is 1737 * not set, try to fill the page then wait for it to become unlocked. 1738 * 1739 * If the page does not get brought uptodate, return -EIO. 1740 */ 1741struct page *read_cache_page(struct address_space *mapping, 1742 pgoff_t index, 1743 int (*filler)(void *,struct page*), 1744 void *data) 1745{ 1746 struct page *page; 1747 1748 page = read_cache_page_async(mapping, index, filler, data); 1749 if (IS_ERR(page)) 1750 goto out; 1751 wait_on_page_locked(page); 1752 if (!PageUptodate(page)) { 1753 page_cache_release(page); 1754 page = ERR_PTR(-EIO); 1755 } 1756 out: 1757 return page; 1758} 1759EXPORT_SYMBOL(read_cache_page); 1760 1761/* 1762 * The logic we want is 1763 * 1764 * if suid or (sgid and xgrp) 1765 * remove privs 1766 */ 1767int should_remove_suid(struct dentry *dentry) 1768{ 1769 mode_t mode = dentry->d_inode->i_mode; 1770 int kill = 0; 1771 1772 /* suid always must be killed */ 1773 if (unlikely(mode & S_ISUID)) 1774 kill = ATTR_KILL_SUID; 1775 1776 /* 1777 * sgid without any exec bits is just a mandatory locking mark; leave 1778 * it alone. If some exec bits are set, it's a real sgid; kill it. 1779 */ 1780 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 1781 kill |= ATTR_KILL_SGID; 1782 1783 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode))) 1784 return kill; 1785 1786 return 0; 1787} 1788EXPORT_SYMBOL(should_remove_suid); 1789 1790static int __remove_suid(struct dentry *dentry, int kill) 1791{ 1792 struct iattr newattrs; 1793 1794 newattrs.ia_valid = ATTR_FORCE | kill; 1795 return notify_change(dentry, &newattrs); 1796} 1797 1798int file_remove_suid(struct file *file) 1799{ 1800 struct dentry *dentry = file->f_path.dentry; 1801 int killsuid = should_remove_suid(dentry); 1802 int killpriv = security_inode_need_killpriv(dentry); 1803 int error = 0; 1804 1805 if (killpriv < 0) 1806 return killpriv; 1807 if (killpriv) 1808 error = security_inode_killpriv(dentry); 1809 if (!error && killsuid) 1810 error = __remove_suid(dentry, killsuid); 1811 1812 return error; 1813} 1814EXPORT_SYMBOL(file_remove_suid); 1815 1816static size_t __iovec_copy_from_user_inatomic(char *vaddr, 1817 const struct iovec *iov, size_t base, size_t bytes) 1818{ 1819 size_t copied = 0, left = 0; 1820 1821 while (bytes) { 1822 char __user *buf = iov->iov_base + base; 1823 int copy = min(bytes, iov->iov_len - base); 1824 1825 base = 0; 1826 left = __copy_from_user_inatomic_nocache(vaddr, buf, copy); 1827 copied += copy; 1828 bytes -= copy; 1829 vaddr += copy; 1830 iov++; 1831 1832 if (unlikely(left)) 1833 break; 1834 } 1835 return copied - left; 1836} 1837 1838/* 1839 * Copy as much as we can into the page and return the number of bytes which 1840 * were sucessfully copied. If a fault is encountered then return the number of 1841 * bytes which were copied. 1842 */ 1843size_t iov_iter_copy_from_user_atomic(struct page *page, 1844 struct iov_iter *i, unsigned long offset, size_t bytes) 1845{ 1846 char *kaddr; 1847 size_t copied; 1848 1849 BUG_ON(!in_atomic()); 1850 kaddr = kmap_atomic(page, KM_USER0); 1851 if (likely(i->nr_segs == 1)) { 1852 int left; 1853 char __user *buf = i->iov->iov_base + i->iov_offset; 1854 left = __copy_from_user_inatomic_nocache(kaddr + offset, 1855 buf, bytes); 1856 copied = bytes - left; 1857 } else { 1858 copied = __iovec_copy_from_user_inatomic(kaddr + offset, 1859 i->iov, i->iov_offset, bytes); 1860 } 1861 kunmap_atomic(kaddr, KM_USER0); 1862 1863 return copied; 1864} 1865EXPORT_SYMBOL(iov_iter_copy_from_user_atomic); 1866 1867/* 1868 * This has the same sideeffects and return value as 1869 * iov_iter_copy_from_user_atomic(). 1870 * The difference is that it attempts to resolve faults. 1871 * Page must not be locked. 1872 */ 1873size_t iov_iter_copy_from_user(struct page *page, 1874 struct iov_iter *i, unsigned long offset, size_t bytes) 1875{ 1876 char *kaddr; 1877 size_t copied; 1878 1879 kaddr = kmap(page); 1880 if (likely(i->nr_segs == 1)) { 1881 int left; 1882 char __user *buf = i->iov->iov_base + i->iov_offset; 1883 left = __copy_from_user_nocache(kaddr + offset, buf, bytes); 1884 copied = bytes - left; 1885 } else { 1886 copied = __iovec_copy_from_user_inatomic(kaddr + offset, 1887 i->iov, i->iov_offset, bytes); 1888 } 1889 kunmap(page); 1890 return copied; 1891} 1892EXPORT_SYMBOL(iov_iter_copy_from_user); 1893 1894void iov_iter_advance(struct iov_iter *i, size_t bytes) 1895{ 1896 BUG_ON(i->count < bytes); 1897 1898 if (likely(i->nr_segs == 1)) { 1899 i->iov_offset += bytes; 1900 i->count -= bytes; 1901 } else { 1902 const struct iovec *iov = i->iov; 1903 size_t base = i->iov_offset; 1904 1905 /* 1906 * The !iov->iov_len check ensures we skip over unlikely 1907 * zero-length segments (without overruning the iovec). 1908 */ 1909 while (bytes || unlikely(i->count && !iov->iov_len)) { 1910 int copy; 1911 1912 copy = min(bytes, iov->iov_len - base); 1913 BUG_ON(!i->count || i->count < copy); 1914 i->count -= copy; 1915 bytes -= copy; 1916 base += copy; 1917 if (iov->iov_len == base) { 1918 iov++; 1919 base = 0; 1920 } 1921 } 1922 i->iov = iov; 1923 i->iov_offset = base; 1924 } 1925} 1926EXPORT_SYMBOL(iov_iter_advance); 1927 1928/* 1929 * Fault in the first iovec of the given iov_iter, to a maximum length 1930 * of bytes. Returns 0 on success, or non-zero if the memory could not be 1931 * accessed (ie. because it is an invalid address). 1932 * 1933 * writev-intensive code may want this to prefault several iovecs -- that 1934 * would be possible (callers must not rely on the fact that _only_ the 1935 * first iovec will be faulted with the current implementation). 1936 */ 1937int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes) 1938{ 1939 char __user *buf = i->iov->iov_base + i->iov_offset; 1940 bytes = min(bytes, i->iov->iov_len - i->iov_offset); 1941 return fault_in_pages_readable(buf, bytes); 1942} 1943EXPORT_SYMBOL(iov_iter_fault_in_readable); 1944 1945/* 1946 * Return the count of just the current iov_iter segment. 1947 */ 1948size_t iov_iter_single_seg_count(struct iov_iter *i) 1949{ 1950 const struct iovec *iov = i->iov; 1951 if (i->nr_segs == 1) 1952 return i->count; 1953 else 1954 return min(i->count, iov->iov_len - i->iov_offset); 1955} 1956EXPORT_SYMBOL(iov_iter_single_seg_count); 1957 1958/* 1959 * Performs necessary checks before doing a write 1960 * 1961 * Can adjust writing position or amount of bytes to write. 1962 * Returns appropriate error code that caller should return or 1963 * zero in case that write should be allowed. 1964 */ 1965inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk) 1966{ 1967 struct inode *inode = file->f_mapping->host; 1968 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; 1969 1970 if (unlikely(*pos < 0)) 1971 return -EINVAL; 1972 1973 if (!isblk) { 1974 /* FIXME: this is for backwards compatibility with 2.4 */ 1975 if (file->f_flags & O_APPEND) 1976 *pos = i_size_read(inode); 1977 1978 if (limit != RLIM_INFINITY) { 1979 if (*pos >= limit) { 1980 send_sig(SIGXFSZ, current, 0); 1981 return -EFBIG; 1982 } 1983 if (*count > limit - (typeof(limit))*pos) { 1984 *count = limit - (typeof(limit))*pos; 1985 } 1986 } 1987 } 1988 1989 /* 1990 * LFS rule 1991 */ 1992 if (unlikely(*pos + *count > MAX_NON_LFS && 1993 !(file->f_flags & O_LARGEFILE))) { 1994 if (*pos >= MAX_NON_LFS) { 1995 return -EFBIG; 1996 } 1997 if (*count > MAX_NON_LFS - (unsigned long)*pos) { 1998 *count = MAX_NON_LFS - (unsigned long)*pos; 1999 } 2000 } 2001 2002 /* 2003 * Are we about to exceed the fs block limit ? 2004 * 2005 * If we have written data it becomes a short write. If we have 2006 * exceeded without writing data we send a signal and return EFBIG. 2007 * Linus frestrict idea will clean these up nicely.. 2008 */ 2009 if (likely(!isblk)) { 2010 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) { 2011 if (*count || *pos > inode->i_sb->s_maxbytes) { 2012 return -EFBIG; 2013 } 2014 /* zero-length writes at ->s_maxbytes are OK */ 2015 } 2016 2017 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes)) 2018 *count = inode->i_sb->s_maxbytes - *pos; 2019 } else { 2020#ifdef CONFIG_BLOCK 2021 loff_t isize; 2022 if (bdev_read_only(I_BDEV(inode))) 2023 return -EPERM; 2024 isize = i_size_read(inode); 2025 if (*pos >= isize) { 2026 if (*count || *pos > isize) 2027 return -ENOSPC; 2028 } 2029 2030 if (*pos + *count > isize) 2031 *count = isize - *pos; 2032#else 2033 return -EPERM; 2034#endif 2035 } 2036 return 0; 2037} 2038EXPORT_SYMBOL(generic_write_checks); 2039 2040int pagecache_write_begin(struct file *file, struct address_space *mapping, 2041 loff_t pos, unsigned len, unsigned flags, 2042 struct page **pagep, void **fsdata) 2043{ 2044 const struct address_space_operations *aops = mapping->a_ops; 2045 2046 return aops->write_begin(file, mapping, pos, len, flags, 2047 pagep, fsdata); 2048} 2049EXPORT_SYMBOL(pagecache_write_begin); 2050 2051int pagecache_write_end(struct file *file, struct address_space *mapping, 2052 loff_t pos, unsigned len, unsigned copied, 2053 struct page *page, void *fsdata) 2054{ 2055 const struct address_space_operations *aops = mapping->a_ops; 2056 2057 mark_page_accessed(page); 2058 return aops->write_end(file, mapping, pos, len, copied, page, fsdata); 2059} 2060EXPORT_SYMBOL(pagecache_write_end); 2061 2062ssize_t 2063generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov, 2064 unsigned long *nr_segs, loff_t pos, loff_t *ppos, 2065 size_t count, size_t ocount) 2066{ 2067 struct file *file = iocb->ki_filp; 2068 struct address_space *mapping = file->f_mapping; 2069 struct inode *inode = mapping->host; 2070 ssize_t written; 2071 size_t write_len; 2072 pgoff_t end; 2073 2074 if (count != ocount) 2075 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count); 2076 2077 write_len = iov_length(iov, *nr_segs); 2078 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT; 2079 2080 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1); 2081 if (written) 2082 goto out; 2083 2084 /* 2085 * After a write we want buffered reads to be sure to go to disk to get 2086 * the new data. We invalidate clean cached page from the region we're 2087 * about to write. We do this *before* the write so that we can return 2088 * without clobbering -EIOCBQUEUED from ->direct_IO(). 2089 */ 2090 if (mapping->nrpages) { 2091 written = invalidate_inode_pages2_range(mapping, 2092 pos >> PAGE_CACHE_SHIFT, end); 2093 /* 2094 * If a page can not be invalidated, return 0 to fall back 2095 * to buffered write. 2096 */ 2097 if (written) { 2098 if (written == -EBUSY) 2099 return 0; 2100 goto out; 2101 } 2102 } 2103 2104 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs); 2105 2106 /* 2107 * Finally, try again to invalidate clean pages which might have been 2108 * cached by non-direct readahead, or faulted in by get_user_pages() 2109 * if the source of the write was an mmap'ed region of the file 2110 * we're writing. Either one is a pretty crazy thing to do, 2111 * so we don't support it 100%. If this invalidation 2112 * fails, tough, the write still worked... 2113 */ 2114 if (mapping->nrpages) { 2115 invalidate_inode_pages2_range(mapping, 2116 pos >> PAGE_CACHE_SHIFT, end); 2117 } 2118 2119 if (written > 0) { 2120 loff_t end = pos + written; 2121 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { 2122 i_size_write(inode, end); 2123 mark_inode_dirty(inode); 2124 } 2125 *ppos = end; 2126 } 2127 2128 /* 2129 * Sync the fs metadata but not the minor inode changes and 2130 * of course not the data as we did direct DMA for the IO. 2131 * i_mutex is held, which protects generic_osync_inode() from 2132 * livelocking. AIO O_DIRECT ops attempt to sync metadata here. 2133 */ 2134out: 2135 if ((written >= 0 || written == -EIOCBQUEUED) && 2136 ((file->f_flags & O_SYNC) || IS_SYNC(inode))) { 2137 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA); 2138 if (err < 0) 2139 written = err; 2140 } 2141 return written; 2142} 2143EXPORT_SYMBOL(generic_file_direct_write); 2144 2145/* 2146 * Find or create a page at the given pagecache position. Return the locked 2147 * page. This function is specifically for buffered writes. 2148 */ 2149struct page *grab_cache_page_write_begin(struct address_space *mapping, 2150 pgoff_t index, unsigned flags) 2151{ 2152 int status; 2153 struct page *page; 2154 gfp_t gfp_notmask = 0; 2155 if (flags & AOP_FLAG_NOFS) 2156 gfp_notmask = __GFP_FS; 2157repeat: 2158 page = find_lock_page(mapping, index); 2159 if (likely(page)) 2160 return page; 2161 2162 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask); 2163 if (!page) 2164 return NULL; 2165 status = add_to_page_cache_lru(page, mapping, index, 2166 GFP_KERNEL & ~gfp_notmask); 2167 if (unlikely(status)) { 2168 page_cache_release(page); 2169 if (status == -EEXIST) 2170 goto repeat; 2171 return NULL; 2172 } 2173 return page; 2174} 2175EXPORT_SYMBOL(grab_cache_page_write_begin); 2176 2177static ssize_t generic_perform_write(struct file *file, 2178 struct iov_iter *i, loff_t pos) 2179{ 2180 struct address_space *mapping = file->f_mapping; 2181 const struct address_space_operations *a_ops = mapping->a_ops; 2182 long status = 0; 2183 ssize_t written = 0; 2184 unsigned int flags = 0; 2185 2186 /* 2187 * Copies from kernel address space cannot fail (NFSD is a big user). 2188 */ 2189 if (segment_eq(get_fs(), KERNEL_DS)) 2190 flags |= AOP_FLAG_UNINTERRUPTIBLE; 2191 2192 do { 2193 struct page *page; 2194 pgoff_t index; /* Pagecache index for current page */ 2195 unsigned long offset; /* Offset into pagecache page */ 2196 unsigned long bytes; /* Bytes to write to page */ 2197 size_t copied; /* Bytes copied from user */ 2198 void *fsdata; 2199 2200 offset = (pos & (PAGE_CACHE_SIZE - 1)); 2201 index = pos >> PAGE_CACHE_SHIFT; 2202 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, 2203 iov_iter_count(i)); 2204 2205again: 2206 2207 /* 2208 * Bring in the user page that we will copy from _first_. 2209 * Otherwise there's a nasty deadlock on copying from the 2210 * same page as we're writing to, without it being marked 2211 * up-to-date. 2212 * 2213 * Not only is this an optimisation, but it is also required 2214 * to check that the address is actually valid, when atomic 2215 * usercopies are used, below. 2216 */ 2217 if (unlikely(iov_iter_fault_in_readable(i, bytes))) { 2218 status = -EFAULT; 2219 break; 2220 } 2221 2222 status = a_ops->write_begin(file, mapping, pos, bytes, flags, 2223 &page, &fsdata); 2224 if (unlikely(status)) 2225 break; 2226 2227 pagefault_disable(); 2228 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); 2229 pagefault_enable(); 2230 flush_dcache_page(page); 2231 2232 status = a_ops->write_end(file, mapping, pos, bytes, copied, 2233 page, fsdata); 2234 if (unlikely(status < 0)) 2235 break; 2236 copied = status; 2237 2238 cond_resched(); 2239 2240 iov_iter_advance(i, copied); 2241 if (unlikely(copied == 0)) { 2242 /* 2243 * If we were unable to copy any data at all, we must 2244 * fall back to a single segment length write. 2245 * 2246 * If we didn't fallback here, we could livelock 2247 * because not all segments in the iov can be copied at 2248 * once without a pagefault. 2249 */ 2250 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, 2251 iov_iter_single_seg_count(i)); 2252 goto again; 2253 } 2254 pos += copied; 2255 written += copied; 2256 2257 balance_dirty_pages_ratelimited(mapping); 2258 2259 } while (iov_iter_count(i)); 2260 2261 return written ? written : status; 2262} 2263 2264ssize_t 2265generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov, 2266 unsigned long nr_segs, loff_t pos, loff_t *ppos, 2267 size_t count, ssize_t written) 2268{ 2269 struct file *file = iocb->ki_filp; 2270 struct address_space *mapping = file->f_mapping; 2271 const struct address_space_operations *a_ops = mapping->a_ops; 2272 struct inode *inode = mapping->host; 2273 ssize_t status; 2274 struct iov_iter i; 2275 2276 iov_iter_init(&i, iov, nr_segs, count, written); 2277 status = generic_perform_write(file, &i, pos); 2278 2279 if (likely(status >= 0)) { 2280 written += status; 2281 *ppos = pos + status; 2282 2283 /* 2284 * For now, when the user asks for O_SYNC, we'll actually give 2285 * O_DSYNC 2286 */ 2287 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) { 2288 if (!a_ops->writepage || !is_sync_kiocb(iocb)) 2289 status = generic_osync_inode(inode, mapping, 2290 OSYNC_METADATA|OSYNC_DATA); 2291 } 2292 } 2293 2294 /* 2295 * If we get here for O_DIRECT writes then we must have fallen through 2296 * to buffered writes (block instantiation inside i_size). So we sync 2297 * the file data here, to try to honour O_DIRECT expectations. 2298 */ 2299 if (unlikely(file->f_flags & O_DIRECT) && written) 2300 status = filemap_write_and_wait_range(mapping, 2301 pos, pos + written - 1); 2302 2303 return written ? written : status; 2304} 2305EXPORT_SYMBOL(generic_file_buffered_write); 2306 2307static ssize_t 2308__generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov, 2309 unsigned long nr_segs, loff_t *ppos) 2310{ 2311 struct file *file = iocb->ki_filp; 2312 struct address_space * mapping = file->f_mapping; 2313 size_t ocount; /* original count */ 2314 size_t count; /* after file limit checks */ 2315 struct inode *inode = mapping->host; 2316 loff_t pos; 2317 ssize_t written; 2318 ssize_t err; 2319 2320 ocount = 0; 2321 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); 2322 if (err) 2323 return err; 2324 2325 count = ocount; 2326 pos = *ppos; 2327 2328 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); 2329 2330 /* We can write back this queue in page reclaim */ 2331 current->backing_dev_info = mapping->backing_dev_info; 2332 written = 0; 2333 2334 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); 2335 if (err) 2336 goto out; 2337 2338 if (count == 0) 2339 goto out; 2340 2341 err = file_remove_suid(file); 2342 if (err) 2343 goto out; 2344 2345 file_update_time(file); 2346 2347 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ 2348 if (unlikely(file->f_flags & O_DIRECT)) { 2349 loff_t endbyte; 2350 ssize_t written_buffered; 2351 2352 written = generic_file_direct_write(iocb, iov, &nr_segs, pos, 2353 ppos, count, ocount); 2354 if (written < 0 || written == count) 2355 goto out; 2356 /* 2357 * direct-io write to a hole: fall through to buffered I/O 2358 * for completing the rest of the request. 2359 */ 2360 pos += written; 2361 count -= written; 2362 written_buffered = generic_file_buffered_write(iocb, iov, 2363 nr_segs, pos, ppos, count, 2364 written); 2365 /* 2366 * If generic_file_buffered_write() retuned a synchronous error 2367 * then we want to return the number of bytes which were 2368 * direct-written, or the error code if that was zero. Note 2369 * that this differs from normal direct-io semantics, which 2370 * will return -EFOO even if some bytes were written. 2371 */ 2372 if (written_buffered < 0) { 2373 err = written_buffered; 2374 goto out; 2375 } 2376 2377 /* 2378 * We need to ensure that the page cache pages are written to 2379 * disk and invalidated to preserve the expected O_DIRECT 2380 * semantics. 2381 */ 2382 endbyte = pos + written_buffered - written - 1; 2383 err = do_sync_mapping_range(file->f_mapping, pos, endbyte, 2384 SYNC_FILE_RANGE_WAIT_BEFORE| 2385 SYNC_FILE_RANGE_WRITE| 2386 SYNC_FILE_RANGE_WAIT_AFTER); 2387 if (err == 0) { 2388 written = written_buffered; 2389 invalidate_mapping_pages(mapping, 2390 pos >> PAGE_CACHE_SHIFT, 2391 endbyte >> PAGE_CACHE_SHIFT); 2392 } else { 2393 /* 2394 * We don't know how much we wrote, so just return 2395 * the number of bytes which were direct-written 2396 */ 2397 } 2398 } else { 2399 written = generic_file_buffered_write(iocb, iov, nr_segs, 2400 pos, ppos, count, written); 2401 } 2402out: 2403 current->backing_dev_info = NULL; 2404 return written ? written : err; 2405} 2406 2407ssize_t generic_file_aio_write_nolock(struct kiocb *iocb, 2408 const struct iovec *iov, unsigned long nr_segs, loff_t pos) 2409{ 2410 struct file *file = iocb->ki_filp; 2411 struct address_space *mapping = file->f_mapping; 2412 struct inode *inode = mapping->host; 2413 ssize_t ret; 2414 2415 BUG_ON(iocb->ki_pos != pos); 2416 2417 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs, 2418 &iocb->ki_pos); 2419 2420 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) { 2421 ssize_t err; 2422 2423 err = sync_page_range_nolock(inode, mapping, pos, ret); 2424 if (err < 0) 2425 ret = err; 2426 } 2427 return ret; 2428} 2429EXPORT_SYMBOL(generic_file_aio_write_nolock); 2430 2431ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, 2432 unsigned long nr_segs, loff_t pos) 2433{ 2434 struct file *file = iocb->ki_filp; 2435 struct address_space *mapping = file->f_mapping; 2436 struct inode *inode = mapping->host; 2437 ssize_t ret; 2438 2439 BUG_ON(iocb->ki_pos != pos); 2440 2441 mutex_lock(&inode->i_mutex); 2442 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs, 2443 &iocb->ki_pos); 2444 mutex_unlock(&inode->i_mutex); 2445 2446 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) { 2447 ssize_t err; 2448 2449 err = sync_page_range(inode, mapping, pos, ret); 2450 if (err < 0) 2451 ret = err; 2452 } 2453 return ret; 2454} 2455EXPORT_SYMBOL(generic_file_aio_write); 2456 2457/** 2458 * try_to_release_page() - release old fs-specific metadata on a page 2459 * 2460 * @page: the page which the kernel is trying to free 2461 * @gfp_mask: memory allocation flags (and I/O mode) 2462 * 2463 * The address_space is to try to release any data against the page 2464 * (presumably at page->private). If the release was successful, return `1'. 2465 * Otherwise return zero. 2466 * 2467 * The @gfp_mask argument specifies whether I/O may be performed to release 2468 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS). 2469 * 2470 */ 2471int try_to_release_page(struct page *page, gfp_t gfp_mask) 2472{ 2473 struct address_space * const mapping = page->mapping; 2474 2475 BUG_ON(!PageLocked(page)); 2476 if (PageWriteback(page)) 2477 return 0; 2478 2479 if (mapping && mapping->a_ops->releasepage) 2480 return mapping->a_ops->releasepage(page, gfp_mask); 2481 return try_to_free_buffers(page); 2482} 2483 2484EXPORT_SYMBOL(try_to_release_page); 2485