filemap.c revision 6fe6900e1e5b6fa9e5c59aa5061f244fe3f467e2
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/pagevec.h> 29#include <linux/blkdev.h> 30#include <linux/security.h> 31#include <linux/syscalls.h> 32#include <linux/cpuset.h> 33#include "filemap.h" 34#include "internal.h" 35 36/* 37 * FIXME: remove all knowledge of the buffer layer from the core VM 38 */ 39#include <linux/buffer_head.h> /* for generic_osync_inode */ 40 41#include <asm/mman.h> 42 43static ssize_t 44generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, 45 loff_t offset, unsigned long nr_segs); 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 a write_lock on 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} 124 125void remove_from_page_cache(struct page *page) 126{ 127 struct address_space *mapping = page->mapping; 128 129 BUG_ON(!PageLocked(page)); 130 131 write_lock_irq(&mapping->tree_lock); 132 __remove_from_page_cache(page); 133 write_unlock_irq(&mapping->tree_lock); 134} 135 136static int sync_page(void *word) 137{ 138 struct address_space *mapping; 139 struct page *page; 140 141 page = container_of((unsigned long *)word, struct page, flags); 142 143 /* 144 * page_mapping() is being called without PG_locked held. 145 * Some knowledge of the state and use of the page is used to 146 * reduce the requirements down to a memory barrier. 147 * The danger here is of a stale page_mapping() return value 148 * indicating a struct address_space different from the one it's 149 * associated with when it is associated with one. 150 * After smp_mb(), it's either the correct page_mapping() for 151 * the page, or an old page_mapping() and the page's own 152 * page_mapping() has gone NULL. 153 * The ->sync_page() address_space operation must tolerate 154 * page_mapping() going NULL. By an amazing coincidence, 155 * this comes about because none of the users of the page 156 * in the ->sync_page() methods make essential use of the 157 * page_mapping(), merely passing the page down to the backing 158 * device's unplug functions when it's non-NULL, which in turn 159 * ignore it for all cases but swap, where only page_private(page) is 160 * of interest. When page_mapping() does go NULL, the entire 161 * call stack gracefully ignores the page and returns. 162 * -- wli 163 */ 164 smp_mb(); 165 mapping = page_mapping(page); 166 if (mapping && mapping->a_ops && mapping->a_ops->sync_page) 167 mapping->a_ops->sync_page(page); 168 io_schedule(); 169 return 0; 170} 171 172/** 173 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range 174 * @mapping: address space structure to write 175 * @start: offset in bytes where the range starts 176 * @end: offset in bytes where the range ends (inclusive) 177 * @sync_mode: enable synchronous operation 178 * 179 * Start writeback against all of a mapping's dirty pages that lie 180 * within the byte offsets <start, end> inclusive. 181 * 182 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as 183 * opposed to a regular memory cleansing writeback. The difference between 184 * these two operations is that if a dirty page/buffer is encountered, it must 185 * be waited upon, and not just skipped over. 186 */ 187int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, 188 loff_t end, int sync_mode) 189{ 190 int ret; 191 struct writeback_control wbc = { 192 .sync_mode = sync_mode, 193 .nr_to_write = mapping->nrpages * 2, 194 .range_start = start, 195 .range_end = end, 196 }; 197 198 if (!mapping_cap_writeback_dirty(mapping)) 199 return 0; 200 201 ret = do_writepages(mapping, &wbc); 202 return ret; 203} 204 205static inline int __filemap_fdatawrite(struct address_space *mapping, 206 int sync_mode) 207{ 208 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); 209} 210 211int filemap_fdatawrite(struct address_space *mapping) 212{ 213 return __filemap_fdatawrite(mapping, WB_SYNC_ALL); 214} 215EXPORT_SYMBOL(filemap_fdatawrite); 216 217static int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, 218 loff_t end) 219{ 220 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); 221} 222 223/** 224 * filemap_flush - mostly a non-blocking flush 225 * @mapping: target address_space 226 * 227 * This is a mostly non-blocking flush. Not suitable for data-integrity 228 * purposes - I/O may not be started against all dirty pages. 229 */ 230int filemap_flush(struct address_space *mapping) 231{ 232 return __filemap_fdatawrite(mapping, WB_SYNC_NONE); 233} 234EXPORT_SYMBOL(filemap_flush); 235 236/** 237 * wait_on_page_writeback_range - wait for writeback to complete 238 * @mapping: target address_space 239 * @start: beginning page index 240 * @end: ending page index 241 * 242 * Wait for writeback to complete against pages indexed by start->end 243 * inclusive 244 */ 245int wait_on_page_writeback_range(struct address_space *mapping, 246 pgoff_t start, pgoff_t end) 247{ 248 struct pagevec pvec; 249 int nr_pages; 250 int ret = 0; 251 pgoff_t index; 252 253 if (end < start) 254 return 0; 255 256 pagevec_init(&pvec, 0); 257 index = start; 258 while ((index <= end) && 259 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, 260 PAGECACHE_TAG_WRITEBACK, 261 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) { 262 unsigned i; 263 264 for (i = 0; i < nr_pages; i++) { 265 struct page *page = pvec.pages[i]; 266 267 /* until radix tree lookup accepts end_index */ 268 if (page->index > end) 269 continue; 270 271 wait_on_page_writeback(page); 272 if (PageError(page)) 273 ret = -EIO; 274 } 275 pagevec_release(&pvec); 276 cond_resched(); 277 } 278 279 /* Check for outstanding write errors */ 280 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags)) 281 ret = -ENOSPC; 282 if (test_and_clear_bit(AS_EIO, &mapping->flags)) 283 ret = -EIO; 284 285 return ret; 286} 287 288/** 289 * sync_page_range - write and wait on all pages in the passed range 290 * @inode: target inode 291 * @mapping: target address_space 292 * @pos: beginning offset in pages to write 293 * @count: number of bytes to write 294 * 295 * Write and wait upon all the pages in the passed range. This is a "data 296 * integrity" operation. It waits upon in-flight writeout before starting and 297 * waiting upon new writeout. If there was an IO error, return it. 298 * 299 * We need to re-take i_mutex during the generic_osync_inode list walk because 300 * it is otherwise livelockable. 301 */ 302int sync_page_range(struct inode *inode, struct address_space *mapping, 303 loff_t pos, loff_t count) 304{ 305 pgoff_t start = pos >> PAGE_CACHE_SHIFT; 306 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT; 307 int ret; 308 309 if (!mapping_cap_writeback_dirty(mapping) || !count) 310 return 0; 311 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1); 312 if (ret == 0) { 313 mutex_lock(&inode->i_mutex); 314 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA); 315 mutex_unlock(&inode->i_mutex); 316 } 317 if (ret == 0) 318 ret = wait_on_page_writeback_range(mapping, start, end); 319 return ret; 320} 321EXPORT_SYMBOL(sync_page_range); 322 323/** 324 * sync_page_range_nolock 325 * @inode: target inode 326 * @mapping: target address_space 327 * @pos: beginning offset in pages to write 328 * @count: number of bytes to write 329 * 330 * Note: Holding i_mutex across sync_page_range_nolock() is not a good idea 331 * as it forces O_SYNC writers to different parts of the same file 332 * to be serialised right until io completion. 333 */ 334int sync_page_range_nolock(struct inode *inode, struct address_space *mapping, 335 loff_t pos, loff_t count) 336{ 337 pgoff_t start = pos >> PAGE_CACHE_SHIFT; 338 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT; 339 int ret; 340 341 if (!mapping_cap_writeback_dirty(mapping) || !count) 342 return 0; 343 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1); 344 if (ret == 0) 345 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA); 346 if (ret == 0) 347 ret = wait_on_page_writeback_range(mapping, start, end); 348 return ret; 349} 350EXPORT_SYMBOL(sync_page_range_nolock); 351 352/** 353 * filemap_fdatawait - wait for all under-writeback pages to complete 354 * @mapping: address space structure to wait for 355 * 356 * Walk the list of under-writeback pages of the given address space 357 * and wait for all of them. 358 */ 359int filemap_fdatawait(struct address_space *mapping) 360{ 361 loff_t i_size = i_size_read(mapping->host); 362 363 if (i_size == 0) 364 return 0; 365 366 return wait_on_page_writeback_range(mapping, 0, 367 (i_size - 1) >> PAGE_CACHE_SHIFT); 368} 369EXPORT_SYMBOL(filemap_fdatawait); 370 371int filemap_write_and_wait(struct address_space *mapping) 372{ 373 int err = 0; 374 375 if (mapping->nrpages) { 376 err = filemap_fdatawrite(mapping); 377 /* 378 * Even if the above returned error, the pages may be 379 * written partially (e.g. -ENOSPC), so we wait for it. 380 * But the -EIO is special case, it may indicate the worst 381 * thing (e.g. bug) happened, so we avoid waiting for it. 382 */ 383 if (err != -EIO) { 384 int err2 = filemap_fdatawait(mapping); 385 if (!err) 386 err = err2; 387 } 388 } 389 return err; 390} 391EXPORT_SYMBOL(filemap_write_and_wait); 392 393/** 394 * filemap_write_and_wait_range - write out & wait on a file range 395 * @mapping: the address_space for the pages 396 * @lstart: offset in bytes where the range starts 397 * @lend: offset in bytes where the range ends (inclusive) 398 * 399 * Write out and wait upon file offsets lstart->lend, inclusive. 400 * 401 * Note that `lend' is inclusive (describes the last byte to be written) so 402 * that this function can be used to write to the very end-of-file (end = -1). 403 */ 404int filemap_write_and_wait_range(struct address_space *mapping, 405 loff_t lstart, loff_t lend) 406{ 407 int err = 0; 408 409 if (mapping->nrpages) { 410 err = __filemap_fdatawrite_range(mapping, lstart, lend, 411 WB_SYNC_ALL); 412 /* See comment of filemap_write_and_wait() */ 413 if (err != -EIO) { 414 int err2 = wait_on_page_writeback_range(mapping, 415 lstart >> PAGE_CACHE_SHIFT, 416 lend >> PAGE_CACHE_SHIFT); 417 if (!err) 418 err = err2; 419 } 420 } 421 return err; 422} 423 424/** 425 * add_to_page_cache - add newly allocated pagecache pages 426 * @page: page to add 427 * @mapping: the page's address_space 428 * @offset: page index 429 * @gfp_mask: page allocation mode 430 * 431 * This function is used to add newly allocated pagecache pages; 432 * the page is new, so we can just run SetPageLocked() against it. 433 * The other page state flags were set by rmqueue(). 434 * 435 * This function does not add the page to the LRU. The caller must do that. 436 */ 437int add_to_page_cache(struct page *page, struct address_space *mapping, 438 pgoff_t offset, gfp_t gfp_mask) 439{ 440 int error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); 441 442 if (error == 0) { 443 write_lock_irq(&mapping->tree_lock); 444 error = radix_tree_insert(&mapping->page_tree, offset, page); 445 if (!error) { 446 page_cache_get(page); 447 SetPageLocked(page); 448 page->mapping = mapping; 449 page->index = offset; 450 mapping->nrpages++; 451 __inc_zone_page_state(page, NR_FILE_PAGES); 452 } 453 write_unlock_irq(&mapping->tree_lock); 454 radix_tree_preload_end(); 455 } 456 return error; 457} 458EXPORT_SYMBOL(add_to_page_cache); 459 460int add_to_page_cache_lru(struct page *page, struct address_space *mapping, 461 pgoff_t offset, gfp_t gfp_mask) 462{ 463 int ret = add_to_page_cache(page, mapping, offset, gfp_mask); 464 if (ret == 0) 465 lru_cache_add(page); 466 return ret; 467} 468 469#ifdef CONFIG_NUMA 470struct page *__page_cache_alloc(gfp_t gfp) 471{ 472 if (cpuset_do_page_mem_spread()) { 473 int n = cpuset_mem_spread_node(); 474 return alloc_pages_node(n, gfp, 0); 475 } 476 return alloc_pages(gfp, 0); 477} 478EXPORT_SYMBOL(__page_cache_alloc); 479#endif 480 481static int __sleep_on_page_lock(void *word) 482{ 483 io_schedule(); 484 return 0; 485} 486 487/* 488 * In order to wait for pages to become available there must be 489 * waitqueues associated with pages. By using a hash table of 490 * waitqueues where the bucket discipline is to maintain all 491 * waiters on the same queue and wake all when any of the pages 492 * become available, and for the woken contexts to check to be 493 * sure the appropriate page became available, this saves space 494 * at a cost of "thundering herd" phenomena during rare hash 495 * collisions. 496 */ 497static wait_queue_head_t *page_waitqueue(struct page *page) 498{ 499 const struct zone *zone = page_zone(page); 500 501 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; 502} 503 504static inline void wake_up_page(struct page *page, int bit) 505{ 506 __wake_up_bit(page_waitqueue(page), &page->flags, bit); 507} 508 509void fastcall wait_on_page_bit(struct page *page, int bit_nr) 510{ 511 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); 512 513 if (test_bit(bit_nr, &page->flags)) 514 __wait_on_bit(page_waitqueue(page), &wait, sync_page, 515 TASK_UNINTERRUPTIBLE); 516} 517EXPORT_SYMBOL(wait_on_page_bit); 518 519/** 520 * unlock_page - unlock a locked page 521 * @page: the page 522 * 523 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). 524 * Also wakes sleepers in wait_on_page_writeback() because the wakeup 525 * mechananism between PageLocked pages and PageWriteback pages is shared. 526 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. 527 * 528 * The first mb is necessary to safely close the critical section opened by the 529 * TestSetPageLocked(), the second mb is necessary to enforce ordering between 530 * the clear_bit and the read of the waitqueue (to avoid SMP races with a 531 * parallel wait_on_page_locked()). 532 */ 533void fastcall unlock_page(struct page *page) 534{ 535 smp_mb__before_clear_bit(); 536 if (!TestClearPageLocked(page)) 537 BUG(); 538 smp_mb__after_clear_bit(); 539 wake_up_page(page, PG_locked); 540} 541EXPORT_SYMBOL(unlock_page); 542 543/** 544 * end_page_writeback - end writeback against a page 545 * @page: the page 546 */ 547void end_page_writeback(struct page *page) 548{ 549 if (!TestClearPageReclaim(page) || rotate_reclaimable_page(page)) { 550 if (!test_clear_page_writeback(page)) 551 BUG(); 552 } 553 smp_mb__after_clear_bit(); 554 wake_up_page(page, PG_writeback); 555} 556EXPORT_SYMBOL(end_page_writeback); 557 558/** 559 * __lock_page - get a lock on the page, assuming we need to sleep to get it 560 * @page: the page to lock 561 * 562 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some 563 * random driver's requestfn sets TASK_RUNNING, we could busywait. However 564 * chances are that on the second loop, the block layer's plug list is empty, 565 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE. 566 */ 567void fastcall __lock_page(struct page *page) 568{ 569 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); 570 571 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page, 572 TASK_UNINTERRUPTIBLE); 573} 574EXPORT_SYMBOL(__lock_page); 575 576/* 577 * Variant of lock_page that does not require the caller to hold a reference 578 * on the page's mapping. 579 */ 580void fastcall __lock_page_nosync(struct page *page) 581{ 582 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); 583 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock, 584 TASK_UNINTERRUPTIBLE); 585} 586 587/** 588 * find_get_page - find and get a page reference 589 * @mapping: the address_space to search 590 * @offset: the page index 591 * 592 * Is there a pagecache struct page at the given (mapping, offset) tuple? 593 * If yes, increment its refcount and return it; if no, return NULL. 594 */ 595struct page * find_get_page(struct address_space *mapping, unsigned long offset) 596{ 597 struct page *page; 598 599 read_lock_irq(&mapping->tree_lock); 600 page = radix_tree_lookup(&mapping->page_tree, offset); 601 if (page) 602 page_cache_get(page); 603 read_unlock_irq(&mapping->tree_lock); 604 return page; 605} 606EXPORT_SYMBOL(find_get_page); 607 608/** 609 * find_lock_page - locate, pin and lock a pagecache page 610 * @mapping: the address_space to search 611 * @offset: the page index 612 * 613 * Locates the desired pagecache page, locks it, increments its reference 614 * count and returns its address. 615 * 616 * Returns zero if the page was not present. find_lock_page() may sleep. 617 */ 618struct page *find_lock_page(struct address_space *mapping, 619 unsigned long offset) 620{ 621 struct page *page; 622 623 read_lock_irq(&mapping->tree_lock); 624repeat: 625 page = radix_tree_lookup(&mapping->page_tree, offset); 626 if (page) { 627 page_cache_get(page); 628 if (TestSetPageLocked(page)) { 629 read_unlock_irq(&mapping->tree_lock); 630 __lock_page(page); 631 read_lock_irq(&mapping->tree_lock); 632 633 /* Has the page been truncated while we slept? */ 634 if (unlikely(page->mapping != mapping || 635 page->index != offset)) { 636 unlock_page(page); 637 page_cache_release(page); 638 goto repeat; 639 } 640 } 641 } 642 read_unlock_irq(&mapping->tree_lock); 643 return page; 644} 645EXPORT_SYMBOL(find_lock_page); 646 647/** 648 * find_or_create_page - locate or add a pagecache page 649 * @mapping: the page's address_space 650 * @index: the page's index into the mapping 651 * @gfp_mask: page allocation mode 652 * 653 * Locates a page in the pagecache. If the page is not present, a new page 654 * is allocated using @gfp_mask and is added to the pagecache and to the VM's 655 * LRU list. The returned page is locked and has its reference count 656 * incremented. 657 * 658 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic 659 * allocation! 660 * 661 * find_or_create_page() returns the desired page's address, or zero on 662 * memory exhaustion. 663 */ 664struct page *find_or_create_page(struct address_space *mapping, 665 unsigned long index, gfp_t gfp_mask) 666{ 667 struct page *page, *cached_page = NULL; 668 int err; 669repeat: 670 page = find_lock_page(mapping, index); 671 if (!page) { 672 if (!cached_page) { 673 cached_page = alloc_page(gfp_mask); 674 if (!cached_page) 675 return NULL; 676 } 677 err = add_to_page_cache_lru(cached_page, mapping, 678 index, gfp_mask); 679 if (!err) { 680 page = cached_page; 681 cached_page = NULL; 682 } else if (err == -EEXIST) 683 goto repeat; 684 } 685 if (cached_page) 686 page_cache_release(cached_page); 687 return page; 688} 689EXPORT_SYMBOL(find_or_create_page); 690 691/** 692 * find_get_pages - gang pagecache lookup 693 * @mapping: The address_space to search 694 * @start: The starting page index 695 * @nr_pages: The maximum number of pages 696 * @pages: Where the resulting pages are placed 697 * 698 * find_get_pages() will search for and return a group of up to 699 * @nr_pages pages in the mapping. The pages are placed at @pages. 700 * find_get_pages() takes a reference against the returned pages. 701 * 702 * The search returns a group of mapping-contiguous pages with ascending 703 * indexes. There may be holes in the indices due to not-present pages. 704 * 705 * find_get_pages() returns the number of pages which were found. 706 */ 707unsigned find_get_pages(struct address_space *mapping, pgoff_t start, 708 unsigned int nr_pages, struct page **pages) 709{ 710 unsigned int i; 711 unsigned int ret; 712 713 read_lock_irq(&mapping->tree_lock); 714 ret = radix_tree_gang_lookup(&mapping->page_tree, 715 (void **)pages, start, nr_pages); 716 for (i = 0; i < ret; i++) 717 page_cache_get(pages[i]); 718 read_unlock_irq(&mapping->tree_lock); 719 return ret; 720} 721 722/** 723 * find_get_pages_contig - gang contiguous pagecache lookup 724 * @mapping: The address_space to search 725 * @index: The starting page index 726 * @nr_pages: The maximum number of pages 727 * @pages: Where the resulting pages are placed 728 * 729 * find_get_pages_contig() works exactly like find_get_pages(), except 730 * that the returned number of pages are guaranteed to be contiguous. 731 * 732 * find_get_pages_contig() returns the number of pages which were found. 733 */ 734unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, 735 unsigned int nr_pages, struct page **pages) 736{ 737 unsigned int i; 738 unsigned int ret; 739 740 read_lock_irq(&mapping->tree_lock); 741 ret = radix_tree_gang_lookup(&mapping->page_tree, 742 (void **)pages, index, nr_pages); 743 for (i = 0; i < ret; i++) { 744 if (pages[i]->mapping == NULL || pages[i]->index != index) 745 break; 746 747 page_cache_get(pages[i]); 748 index++; 749 } 750 read_unlock_irq(&mapping->tree_lock); 751 return i; 752} 753 754/** 755 * find_get_pages_tag - find and return pages that match @tag 756 * @mapping: the address_space to search 757 * @index: the starting page index 758 * @tag: the tag index 759 * @nr_pages: the maximum number of pages 760 * @pages: where the resulting pages are placed 761 * 762 * Like find_get_pages, except we only return pages which are tagged with 763 * @tag. We update @index to index the next page for the traversal. 764 */ 765unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, 766 int tag, unsigned int nr_pages, struct page **pages) 767{ 768 unsigned int i; 769 unsigned int ret; 770 771 read_lock_irq(&mapping->tree_lock); 772 ret = radix_tree_gang_lookup_tag(&mapping->page_tree, 773 (void **)pages, *index, nr_pages, tag); 774 for (i = 0; i < ret; i++) 775 page_cache_get(pages[i]); 776 if (ret) 777 *index = pages[ret - 1]->index + 1; 778 read_unlock_irq(&mapping->tree_lock); 779 return ret; 780} 781 782/** 783 * grab_cache_page_nowait - returns locked page at given index in given cache 784 * @mapping: target address_space 785 * @index: the page index 786 * 787 * Same as grab_cache_page(), but do not wait if the page is unavailable. 788 * This is intended for speculative data generators, where the data can 789 * be regenerated if the page couldn't be grabbed. This routine should 790 * be safe to call while holding the lock for another page. 791 * 792 * Clear __GFP_FS when allocating the page to avoid recursion into the fs 793 * and deadlock against the caller's locked page. 794 */ 795struct page * 796grab_cache_page_nowait(struct address_space *mapping, unsigned long index) 797{ 798 struct page *page = find_get_page(mapping, index); 799 800 if (page) { 801 if (!TestSetPageLocked(page)) 802 return page; 803 page_cache_release(page); 804 return NULL; 805 } 806 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS); 807 if (page && add_to_page_cache_lru(page, mapping, index, GFP_KERNEL)) { 808 page_cache_release(page); 809 page = NULL; 810 } 811 return page; 812} 813EXPORT_SYMBOL(grab_cache_page_nowait); 814 815/* 816 * CD/DVDs are error prone. When a medium error occurs, the driver may fail 817 * a _large_ part of the i/o request. Imagine the worst scenario: 818 * 819 * ---R__________________________________________B__________ 820 * ^ reading here ^ bad block(assume 4k) 821 * 822 * read(R) => miss => readahead(R...B) => media error => frustrating retries 823 * => failing the whole request => read(R) => read(R+1) => 824 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => 825 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => 826 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... 827 * 828 * It is going insane. Fix it by quickly scaling down the readahead size. 829 */ 830static void shrink_readahead_size_eio(struct file *filp, 831 struct file_ra_state *ra) 832{ 833 if (!ra->ra_pages) 834 return; 835 836 ra->ra_pages /= 4; 837} 838 839/** 840 * do_generic_mapping_read - generic file read routine 841 * @mapping: address_space to be read 842 * @_ra: file's readahead state 843 * @filp: the file to read 844 * @ppos: current file position 845 * @desc: read_descriptor 846 * @actor: read method 847 * 848 * This is a generic file read routine, and uses the 849 * mapping->a_ops->readpage() function for the actual low-level stuff. 850 * 851 * This is really ugly. But the goto's actually try to clarify some 852 * of the logic when it comes to error handling etc. 853 * 854 * Note the struct file* is only passed for the use of readpage. 855 * It may be NULL. 856 */ 857void do_generic_mapping_read(struct address_space *mapping, 858 struct file_ra_state *_ra, 859 struct file *filp, 860 loff_t *ppos, 861 read_descriptor_t *desc, 862 read_actor_t actor) 863{ 864 struct inode *inode = mapping->host; 865 unsigned long index; 866 unsigned long end_index; 867 unsigned long offset; 868 unsigned long last_index; 869 unsigned long next_index; 870 unsigned long prev_index; 871 loff_t isize; 872 struct page *cached_page; 873 int error; 874 struct file_ra_state ra = *_ra; 875 876 cached_page = NULL; 877 index = *ppos >> PAGE_CACHE_SHIFT; 878 next_index = index; 879 prev_index = ra.prev_page; 880 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; 881 offset = *ppos & ~PAGE_CACHE_MASK; 882 883 isize = i_size_read(inode); 884 if (!isize) 885 goto out; 886 887 end_index = (isize - 1) >> PAGE_CACHE_SHIFT; 888 for (;;) { 889 struct page *page; 890 unsigned long nr, ret; 891 892 /* nr is the maximum number of bytes to copy from this page */ 893 nr = PAGE_CACHE_SIZE; 894 if (index >= end_index) { 895 if (index > end_index) 896 goto out; 897 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; 898 if (nr <= offset) { 899 goto out; 900 } 901 } 902 nr = nr - offset; 903 904 cond_resched(); 905 if (index == next_index) 906 next_index = page_cache_readahead(mapping, &ra, filp, 907 index, last_index - index); 908 909find_page: 910 page = find_get_page(mapping, index); 911 if (unlikely(page == NULL)) { 912 handle_ra_miss(mapping, &ra, index); 913 goto no_cached_page; 914 } 915 if (!PageUptodate(page)) 916 goto page_not_up_to_date; 917page_ok: 918 919 /* If users can be writing to this page using arbitrary 920 * virtual addresses, take care about potential aliasing 921 * before reading the page on the kernel side. 922 */ 923 if (mapping_writably_mapped(mapping)) 924 flush_dcache_page(page); 925 926 /* 927 * When (part of) the same page is read multiple times 928 * in succession, only mark it as accessed the first time. 929 */ 930 if (prev_index != index) 931 mark_page_accessed(page); 932 prev_index = index; 933 934 /* 935 * Ok, we have the page, and it's up-to-date, so 936 * now we can copy it to user space... 937 * 938 * The actor routine returns how many bytes were actually used.. 939 * NOTE! This may not be the same as how much of a user buffer 940 * we filled up (we may be padding etc), so we can only update 941 * "pos" here (the actor routine has to update the user buffer 942 * pointers and the remaining count). 943 */ 944 ret = actor(desc, page, offset, nr); 945 offset += ret; 946 index += offset >> PAGE_CACHE_SHIFT; 947 offset &= ~PAGE_CACHE_MASK; 948 949 page_cache_release(page); 950 if (ret == nr && desc->count) 951 continue; 952 goto out; 953 954page_not_up_to_date: 955 /* Get exclusive access to the page ... */ 956 lock_page(page); 957 958 /* Did it get truncated before we got the lock? */ 959 if (!page->mapping) { 960 unlock_page(page); 961 page_cache_release(page); 962 continue; 963 } 964 965 /* Did somebody else fill it already? */ 966 if (PageUptodate(page)) { 967 unlock_page(page); 968 goto page_ok; 969 } 970 971readpage: 972 /* Start the actual read. The read will unlock the page. */ 973 error = mapping->a_ops->readpage(filp, page); 974 975 if (unlikely(error)) { 976 if (error == AOP_TRUNCATED_PAGE) { 977 page_cache_release(page); 978 goto find_page; 979 } 980 goto readpage_error; 981 } 982 983 if (!PageUptodate(page)) { 984 lock_page(page); 985 if (!PageUptodate(page)) { 986 if (page->mapping == NULL) { 987 /* 988 * invalidate_inode_pages got it 989 */ 990 unlock_page(page); 991 page_cache_release(page); 992 goto find_page; 993 } 994 unlock_page(page); 995 error = -EIO; 996 shrink_readahead_size_eio(filp, &ra); 997 goto readpage_error; 998 } 999 unlock_page(page); 1000 } 1001 1002 /* 1003 * i_size must be checked after we have done ->readpage. 1004 * 1005 * Checking i_size after the readpage allows us to calculate 1006 * the correct value for "nr", which means the zero-filled 1007 * part of the page is not copied back to userspace (unless 1008 * another truncate extends the file - this is desired though). 1009 */ 1010 isize = i_size_read(inode); 1011 end_index = (isize - 1) >> PAGE_CACHE_SHIFT; 1012 if (unlikely(!isize || index > end_index)) { 1013 page_cache_release(page); 1014 goto out; 1015 } 1016 1017 /* nr is the maximum number of bytes to copy from this page */ 1018 nr = PAGE_CACHE_SIZE; 1019 if (index == end_index) { 1020 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; 1021 if (nr <= offset) { 1022 page_cache_release(page); 1023 goto out; 1024 } 1025 } 1026 nr = nr - offset; 1027 goto page_ok; 1028 1029readpage_error: 1030 /* UHHUH! A synchronous read error occurred. Report it */ 1031 desc->error = error; 1032 page_cache_release(page); 1033 goto out; 1034 1035no_cached_page: 1036 /* 1037 * Ok, it wasn't cached, so we need to create a new 1038 * page.. 1039 */ 1040 if (!cached_page) { 1041 cached_page = page_cache_alloc_cold(mapping); 1042 if (!cached_page) { 1043 desc->error = -ENOMEM; 1044 goto out; 1045 } 1046 } 1047 error = add_to_page_cache_lru(cached_page, mapping, 1048 index, GFP_KERNEL); 1049 if (error) { 1050 if (error == -EEXIST) 1051 goto find_page; 1052 desc->error = error; 1053 goto out; 1054 } 1055 page = cached_page; 1056 cached_page = NULL; 1057 goto readpage; 1058 } 1059 1060out: 1061 *_ra = ra; 1062 1063 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; 1064 if (cached_page) 1065 page_cache_release(cached_page); 1066 if (filp) 1067 file_accessed(filp); 1068} 1069EXPORT_SYMBOL(do_generic_mapping_read); 1070 1071int file_read_actor(read_descriptor_t *desc, struct page *page, 1072 unsigned long offset, unsigned long size) 1073{ 1074 char *kaddr; 1075 unsigned long left, count = desc->count; 1076 1077 if (size > count) 1078 size = count; 1079 1080 /* 1081 * Faults on the destination of a read are common, so do it before 1082 * taking the kmap. 1083 */ 1084 if (!fault_in_pages_writeable(desc->arg.buf, size)) { 1085 kaddr = kmap_atomic(page, KM_USER0); 1086 left = __copy_to_user_inatomic(desc->arg.buf, 1087 kaddr + offset, size); 1088 kunmap_atomic(kaddr, KM_USER0); 1089 if (left == 0) 1090 goto success; 1091 } 1092 1093 /* Do it the slow way */ 1094 kaddr = kmap(page); 1095 left = __copy_to_user(desc->arg.buf, kaddr + offset, size); 1096 kunmap(page); 1097 1098 if (left) { 1099 size -= left; 1100 desc->error = -EFAULT; 1101 } 1102success: 1103 desc->count = count - size; 1104 desc->written += size; 1105 desc->arg.buf += size; 1106 return size; 1107} 1108 1109/** 1110 * generic_file_aio_read - generic filesystem read routine 1111 * @iocb: kernel I/O control block 1112 * @iov: io vector request 1113 * @nr_segs: number of segments in the iovec 1114 * @pos: current file position 1115 * 1116 * This is the "read()" routine for all filesystems 1117 * that can use the page cache directly. 1118 */ 1119ssize_t 1120generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov, 1121 unsigned long nr_segs, loff_t pos) 1122{ 1123 struct file *filp = iocb->ki_filp; 1124 ssize_t retval; 1125 unsigned long seg; 1126 size_t count; 1127 loff_t *ppos = &iocb->ki_pos; 1128 1129 count = 0; 1130 for (seg = 0; seg < nr_segs; seg++) { 1131 const struct iovec *iv = &iov[seg]; 1132 1133 /* 1134 * If any segment has a negative length, or the cumulative 1135 * length ever wraps negative then return -EINVAL. 1136 */ 1137 count += iv->iov_len; 1138 if (unlikely((ssize_t)(count|iv->iov_len) < 0)) 1139 return -EINVAL; 1140 if (access_ok(VERIFY_WRITE, iv->iov_base, iv->iov_len)) 1141 continue; 1142 if (seg == 0) 1143 return -EFAULT; 1144 nr_segs = seg; 1145 count -= iv->iov_len; /* This segment is no good */ 1146 break; 1147 } 1148 1149 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ 1150 if (filp->f_flags & O_DIRECT) { 1151 loff_t size; 1152 struct address_space *mapping; 1153 struct inode *inode; 1154 1155 mapping = filp->f_mapping; 1156 inode = mapping->host; 1157 retval = 0; 1158 if (!count) 1159 goto out; /* skip atime */ 1160 size = i_size_read(inode); 1161 if (pos < size) { 1162 retval = generic_file_direct_IO(READ, iocb, 1163 iov, pos, nr_segs); 1164 if (retval > 0) 1165 *ppos = pos + retval; 1166 } 1167 if (likely(retval != 0)) { 1168 file_accessed(filp); 1169 goto out; 1170 } 1171 } 1172 1173 retval = 0; 1174 if (count) { 1175 for (seg = 0; seg < nr_segs; seg++) { 1176 read_descriptor_t desc; 1177 1178 desc.written = 0; 1179 desc.arg.buf = iov[seg].iov_base; 1180 desc.count = iov[seg].iov_len; 1181 if (desc.count == 0) 1182 continue; 1183 desc.error = 0; 1184 do_generic_file_read(filp,ppos,&desc,file_read_actor); 1185 retval += desc.written; 1186 if (desc.error) { 1187 retval = retval ?: desc.error; 1188 break; 1189 } 1190 } 1191 } 1192out: 1193 return retval; 1194} 1195EXPORT_SYMBOL(generic_file_aio_read); 1196 1197int file_send_actor(read_descriptor_t * desc, struct page *page, unsigned long offset, unsigned long size) 1198{ 1199 ssize_t written; 1200 unsigned long count = desc->count; 1201 struct file *file = desc->arg.data; 1202 1203 if (size > count) 1204 size = count; 1205 1206 written = file->f_op->sendpage(file, page, offset, 1207 size, &file->f_pos, size<count); 1208 if (written < 0) { 1209 desc->error = written; 1210 written = 0; 1211 } 1212 desc->count = count - written; 1213 desc->written += written; 1214 return written; 1215} 1216 1217ssize_t generic_file_sendfile(struct file *in_file, loff_t *ppos, 1218 size_t count, read_actor_t actor, void *target) 1219{ 1220 read_descriptor_t desc; 1221 1222 if (!count) 1223 return 0; 1224 1225 desc.written = 0; 1226 desc.count = count; 1227 desc.arg.data = target; 1228 desc.error = 0; 1229 1230 do_generic_file_read(in_file, ppos, &desc, actor); 1231 if (desc.written) 1232 return desc.written; 1233 return desc.error; 1234} 1235EXPORT_SYMBOL(generic_file_sendfile); 1236 1237static ssize_t 1238do_readahead(struct address_space *mapping, struct file *filp, 1239 unsigned long index, unsigned long nr) 1240{ 1241 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage) 1242 return -EINVAL; 1243 1244 force_page_cache_readahead(mapping, filp, index, 1245 max_sane_readahead(nr)); 1246 return 0; 1247} 1248 1249asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count) 1250{ 1251 ssize_t ret; 1252 struct file *file; 1253 1254 ret = -EBADF; 1255 file = fget(fd); 1256 if (file) { 1257 if (file->f_mode & FMODE_READ) { 1258 struct address_space *mapping = file->f_mapping; 1259 unsigned long start = offset >> PAGE_CACHE_SHIFT; 1260 unsigned long end = (offset + count - 1) >> PAGE_CACHE_SHIFT; 1261 unsigned long len = end - start + 1; 1262 ret = do_readahead(mapping, file, start, len); 1263 } 1264 fput(file); 1265 } 1266 return ret; 1267} 1268 1269#ifdef CONFIG_MMU 1270static int FASTCALL(page_cache_read(struct file * file, unsigned long offset)); 1271/** 1272 * page_cache_read - adds requested page to the page cache if not already there 1273 * @file: file to read 1274 * @offset: page index 1275 * 1276 * This adds the requested page to the page cache if it isn't already there, 1277 * and schedules an I/O to read in its contents from disk. 1278 */ 1279static int fastcall page_cache_read(struct file * file, unsigned long offset) 1280{ 1281 struct address_space *mapping = file->f_mapping; 1282 struct page *page; 1283 int ret; 1284 1285 do { 1286 page = page_cache_alloc_cold(mapping); 1287 if (!page) 1288 return -ENOMEM; 1289 1290 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL); 1291 if (ret == 0) 1292 ret = mapping->a_ops->readpage(file, page); 1293 else if (ret == -EEXIST) 1294 ret = 0; /* losing race to add is OK */ 1295 1296 page_cache_release(page); 1297 1298 } while (ret == AOP_TRUNCATED_PAGE); 1299 1300 return ret; 1301} 1302 1303#define MMAP_LOTSAMISS (100) 1304 1305/** 1306 * filemap_nopage - read in file data for page fault handling 1307 * @area: the applicable vm_area 1308 * @address: target address to read in 1309 * @type: returned with VM_FAULT_{MINOR,MAJOR} if not %NULL 1310 * 1311 * filemap_nopage() is invoked via the vma operations vector for a 1312 * mapped memory region to read in file data during a page fault. 1313 * 1314 * The goto's are kind of ugly, but this streamlines the normal case of having 1315 * it in the page cache, and handles the special cases reasonably without 1316 * having a lot of duplicated code. 1317 */ 1318struct page *filemap_nopage(struct vm_area_struct *area, 1319 unsigned long address, int *type) 1320{ 1321 int error; 1322 struct file *file = area->vm_file; 1323 struct address_space *mapping = file->f_mapping; 1324 struct file_ra_state *ra = &file->f_ra; 1325 struct inode *inode = mapping->host; 1326 struct page *page; 1327 unsigned long size, pgoff; 1328 int did_readaround = 0, majmin = VM_FAULT_MINOR; 1329 1330 pgoff = ((address-area->vm_start) >> PAGE_CACHE_SHIFT) + area->vm_pgoff; 1331 1332retry_all: 1333 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 1334 if (pgoff >= size) 1335 goto outside_data_content; 1336 1337 /* If we don't want any read-ahead, don't bother */ 1338 if (VM_RandomReadHint(area)) 1339 goto no_cached_page; 1340 1341 /* 1342 * The readahead code wants to be told about each and every page 1343 * so it can build and shrink its windows appropriately 1344 * 1345 * For sequential accesses, we use the generic readahead logic. 1346 */ 1347 if (VM_SequentialReadHint(area)) 1348 page_cache_readahead(mapping, ra, file, pgoff, 1); 1349 1350 /* 1351 * Do we have something in the page cache already? 1352 */ 1353retry_find: 1354 page = find_get_page(mapping, pgoff); 1355 if (!page) { 1356 unsigned long ra_pages; 1357 1358 if (VM_SequentialReadHint(area)) { 1359 handle_ra_miss(mapping, ra, pgoff); 1360 goto no_cached_page; 1361 } 1362 ra->mmap_miss++; 1363 1364 /* 1365 * Do we miss much more than hit in this file? If so, 1366 * stop bothering with read-ahead. It will only hurt. 1367 */ 1368 if (ra->mmap_miss > ra->mmap_hit + MMAP_LOTSAMISS) 1369 goto no_cached_page; 1370 1371 /* 1372 * To keep the pgmajfault counter straight, we need to 1373 * check did_readaround, as this is an inner loop. 1374 */ 1375 if (!did_readaround) { 1376 majmin = VM_FAULT_MAJOR; 1377 count_vm_event(PGMAJFAULT); 1378 } 1379 did_readaround = 1; 1380 ra_pages = max_sane_readahead(file->f_ra.ra_pages); 1381 if (ra_pages) { 1382 pgoff_t start = 0; 1383 1384 if (pgoff > ra_pages / 2) 1385 start = pgoff - ra_pages / 2; 1386 do_page_cache_readahead(mapping, file, start, ra_pages); 1387 } 1388 page = find_get_page(mapping, pgoff); 1389 if (!page) 1390 goto no_cached_page; 1391 } 1392 1393 if (!did_readaround) 1394 ra->mmap_hit++; 1395 1396 /* 1397 * Ok, found a page in the page cache, now we need to check 1398 * that it's up-to-date. 1399 */ 1400 if (!PageUptodate(page)) 1401 goto page_not_uptodate; 1402 1403success: 1404 /* 1405 * Found the page and have a reference on it. 1406 */ 1407 mark_page_accessed(page); 1408 if (type) 1409 *type = majmin; 1410 return page; 1411 1412outside_data_content: 1413 /* 1414 * An external ptracer can access pages that normally aren't 1415 * accessible.. 1416 */ 1417 if (area->vm_mm == current->mm) 1418 return NOPAGE_SIGBUS; 1419 /* Fall through to the non-read-ahead case */ 1420no_cached_page: 1421 /* 1422 * We're only likely to ever get here if MADV_RANDOM is in 1423 * effect. 1424 */ 1425 error = page_cache_read(file, pgoff); 1426 1427 /* 1428 * The page we want has now been added to the page cache. 1429 * In the unlikely event that someone removed it in the 1430 * meantime, we'll just come back here and read it again. 1431 */ 1432 if (error >= 0) 1433 goto retry_find; 1434 1435 /* 1436 * An error return from page_cache_read can result if the 1437 * system is low on memory, or a problem occurs while trying 1438 * to schedule I/O. 1439 */ 1440 if (error == -ENOMEM) 1441 return NOPAGE_OOM; 1442 return NOPAGE_SIGBUS; 1443 1444page_not_uptodate: 1445 if (!did_readaround) { 1446 majmin = VM_FAULT_MAJOR; 1447 count_vm_event(PGMAJFAULT); 1448 } 1449 lock_page(page); 1450 1451 /* Did it get unhashed while we waited for it? */ 1452 if (!page->mapping) { 1453 unlock_page(page); 1454 page_cache_release(page); 1455 goto retry_all; 1456 } 1457 1458 /* Did somebody else get it up-to-date? */ 1459 if (PageUptodate(page)) { 1460 unlock_page(page); 1461 goto success; 1462 } 1463 1464 error = mapping->a_ops->readpage(file, page); 1465 if (!error) { 1466 wait_on_page_locked(page); 1467 if (PageUptodate(page)) 1468 goto success; 1469 } else if (error == AOP_TRUNCATED_PAGE) { 1470 page_cache_release(page); 1471 goto retry_find; 1472 } 1473 1474 /* 1475 * Umm, take care of errors if the page isn't up-to-date. 1476 * Try to re-read it _once_. We do this synchronously, 1477 * because there really aren't any performance issues here 1478 * and we need to check for errors. 1479 */ 1480 lock_page(page); 1481 1482 /* Somebody truncated the page on us? */ 1483 if (!page->mapping) { 1484 unlock_page(page); 1485 page_cache_release(page); 1486 goto retry_all; 1487 } 1488 1489 /* Somebody else successfully read it in? */ 1490 if (PageUptodate(page)) { 1491 unlock_page(page); 1492 goto success; 1493 } 1494 ClearPageError(page); 1495 error = mapping->a_ops->readpage(file, page); 1496 if (!error) { 1497 wait_on_page_locked(page); 1498 if (PageUptodate(page)) 1499 goto success; 1500 } else if (error == AOP_TRUNCATED_PAGE) { 1501 page_cache_release(page); 1502 goto retry_find; 1503 } 1504 1505 /* 1506 * Things didn't work out. Return zero to tell the 1507 * mm layer so, possibly freeing the page cache page first. 1508 */ 1509 shrink_readahead_size_eio(file, ra); 1510 page_cache_release(page); 1511 return NOPAGE_SIGBUS; 1512} 1513EXPORT_SYMBOL(filemap_nopage); 1514 1515static struct page * filemap_getpage(struct file *file, unsigned long pgoff, 1516 int nonblock) 1517{ 1518 struct address_space *mapping = file->f_mapping; 1519 struct page *page; 1520 int error; 1521 1522 /* 1523 * Do we have something in the page cache already? 1524 */ 1525retry_find: 1526 page = find_get_page(mapping, pgoff); 1527 if (!page) { 1528 if (nonblock) 1529 return NULL; 1530 goto no_cached_page; 1531 } 1532 1533 /* 1534 * Ok, found a page in the page cache, now we need to check 1535 * that it's up-to-date. 1536 */ 1537 if (!PageUptodate(page)) { 1538 if (nonblock) { 1539 page_cache_release(page); 1540 return NULL; 1541 } 1542 goto page_not_uptodate; 1543 } 1544 1545success: 1546 /* 1547 * Found the page and have a reference on it. 1548 */ 1549 mark_page_accessed(page); 1550 return page; 1551 1552no_cached_page: 1553 error = page_cache_read(file, pgoff); 1554 1555 /* 1556 * The page we want has now been added to the page cache. 1557 * In the unlikely event that someone removed it in the 1558 * meantime, we'll just come back here and read it again. 1559 */ 1560 if (error >= 0) 1561 goto retry_find; 1562 1563 /* 1564 * An error return from page_cache_read can result if the 1565 * system is low on memory, or a problem occurs while trying 1566 * to schedule I/O. 1567 */ 1568 return NULL; 1569 1570page_not_uptodate: 1571 lock_page(page); 1572 1573 /* Did it get truncated while we waited for it? */ 1574 if (!page->mapping) { 1575 unlock_page(page); 1576 goto err; 1577 } 1578 1579 /* Did somebody else get it up-to-date? */ 1580 if (PageUptodate(page)) { 1581 unlock_page(page); 1582 goto success; 1583 } 1584 1585 error = mapping->a_ops->readpage(file, page); 1586 if (!error) { 1587 wait_on_page_locked(page); 1588 if (PageUptodate(page)) 1589 goto success; 1590 } else if (error == AOP_TRUNCATED_PAGE) { 1591 page_cache_release(page); 1592 goto retry_find; 1593 } 1594 1595 /* 1596 * Umm, take care of errors if the page isn't up-to-date. 1597 * Try to re-read it _once_. We do this synchronously, 1598 * because there really aren't any performance issues here 1599 * and we need to check for errors. 1600 */ 1601 lock_page(page); 1602 1603 /* Somebody truncated the page on us? */ 1604 if (!page->mapping) { 1605 unlock_page(page); 1606 goto err; 1607 } 1608 /* Somebody else successfully read it in? */ 1609 if (PageUptodate(page)) { 1610 unlock_page(page); 1611 goto success; 1612 } 1613 1614 ClearPageError(page); 1615 error = mapping->a_ops->readpage(file, page); 1616 if (!error) { 1617 wait_on_page_locked(page); 1618 if (PageUptodate(page)) 1619 goto success; 1620 } else if (error == AOP_TRUNCATED_PAGE) { 1621 page_cache_release(page); 1622 goto retry_find; 1623 } 1624 1625 /* 1626 * Things didn't work out. Return zero to tell the 1627 * mm layer so, possibly freeing the page cache page first. 1628 */ 1629err: 1630 page_cache_release(page); 1631 1632 return NULL; 1633} 1634 1635int filemap_populate(struct vm_area_struct *vma, unsigned long addr, 1636 unsigned long len, pgprot_t prot, unsigned long pgoff, 1637 int nonblock) 1638{ 1639 struct file *file = vma->vm_file; 1640 struct address_space *mapping = file->f_mapping; 1641 struct inode *inode = mapping->host; 1642 unsigned long size; 1643 struct mm_struct *mm = vma->vm_mm; 1644 struct page *page; 1645 int err; 1646 1647 if (!nonblock) 1648 force_page_cache_readahead(mapping, vma->vm_file, 1649 pgoff, len >> PAGE_CACHE_SHIFT); 1650 1651repeat: 1652 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 1653 if (pgoff + (len >> PAGE_CACHE_SHIFT) > size) 1654 return -EINVAL; 1655 1656 page = filemap_getpage(file, pgoff, nonblock); 1657 1658 /* XXX: This is wrong, a filesystem I/O error may have happened. Fix that as 1659 * done in shmem_populate calling shmem_getpage */ 1660 if (!page && !nonblock) 1661 return -ENOMEM; 1662 1663 if (page) { 1664 err = install_page(mm, vma, addr, page, prot); 1665 if (err) { 1666 page_cache_release(page); 1667 return err; 1668 } 1669 } else if (vma->vm_flags & VM_NONLINEAR) { 1670 /* No page was found just because we can't read it in now (being 1671 * here implies nonblock != 0), but the page may exist, so set 1672 * the PTE to fault it in later. */ 1673 err = install_file_pte(mm, vma, addr, pgoff, prot); 1674 if (err) 1675 return err; 1676 } 1677 1678 len -= PAGE_SIZE; 1679 addr += PAGE_SIZE; 1680 pgoff++; 1681 if (len) 1682 goto repeat; 1683 1684 return 0; 1685} 1686EXPORT_SYMBOL(filemap_populate); 1687 1688struct vm_operations_struct generic_file_vm_ops = { 1689 .nopage = filemap_nopage, 1690 .populate = filemap_populate, 1691}; 1692 1693/* This is used for a general mmap of a disk file */ 1694 1695int generic_file_mmap(struct file * file, struct vm_area_struct * vma) 1696{ 1697 struct address_space *mapping = file->f_mapping; 1698 1699 if (!mapping->a_ops->readpage) 1700 return -ENOEXEC; 1701 file_accessed(file); 1702 vma->vm_ops = &generic_file_vm_ops; 1703 return 0; 1704} 1705 1706/* 1707 * This is for filesystems which do not implement ->writepage. 1708 */ 1709int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) 1710{ 1711 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) 1712 return -EINVAL; 1713 return generic_file_mmap(file, vma); 1714} 1715#else 1716int generic_file_mmap(struct file * file, struct vm_area_struct * vma) 1717{ 1718 return -ENOSYS; 1719} 1720int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) 1721{ 1722 return -ENOSYS; 1723} 1724#endif /* CONFIG_MMU */ 1725 1726EXPORT_SYMBOL(generic_file_mmap); 1727EXPORT_SYMBOL(generic_file_readonly_mmap); 1728 1729static struct page *__read_cache_page(struct address_space *mapping, 1730 unsigned long index, 1731 int (*filler)(void *,struct page*), 1732 void *data) 1733{ 1734 struct page *page, *cached_page = NULL; 1735 int err; 1736repeat: 1737 page = find_get_page(mapping, index); 1738 if (!page) { 1739 if (!cached_page) { 1740 cached_page = page_cache_alloc_cold(mapping); 1741 if (!cached_page) 1742 return ERR_PTR(-ENOMEM); 1743 } 1744 err = add_to_page_cache_lru(cached_page, mapping, 1745 index, GFP_KERNEL); 1746 if (err == -EEXIST) 1747 goto repeat; 1748 if (err < 0) { 1749 /* Presumably ENOMEM for radix tree node */ 1750 page_cache_release(cached_page); 1751 return ERR_PTR(err); 1752 } 1753 page = cached_page; 1754 cached_page = NULL; 1755 err = filler(data, page); 1756 if (err < 0) { 1757 page_cache_release(page); 1758 page = ERR_PTR(err); 1759 } 1760 } 1761 if (cached_page) 1762 page_cache_release(cached_page); 1763 return page; 1764} 1765 1766/* 1767 * Same as read_cache_page, but don't wait for page to become unlocked 1768 * after submitting it to the filler. 1769 */ 1770struct page *read_cache_page_async(struct address_space *mapping, 1771 unsigned long index, 1772 int (*filler)(void *,struct page*), 1773 void *data) 1774{ 1775 struct page *page; 1776 int err; 1777 1778retry: 1779 page = __read_cache_page(mapping, index, filler, data); 1780 if (IS_ERR(page)) 1781 goto out; 1782 mark_page_accessed(page); 1783 if (PageUptodate(page)) 1784 goto out; 1785 1786 lock_page(page); 1787 if (!page->mapping) { 1788 unlock_page(page); 1789 page_cache_release(page); 1790 goto retry; 1791 } 1792 if (PageUptodate(page)) { 1793 unlock_page(page); 1794 goto out; 1795 } 1796 err = filler(data, page); 1797 if (err < 0) { 1798 page_cache_release(page); 1799 page = ERR_PTR(err); 1800 } 1801 out: 1802 mark_page_accessed(page); 1803 return page; 1804} 1805EXPORT_SYMBOL(read_cache_page_async); 1806 1807/** 1808 * read_cache_page - read into page cache, fill it if needed 1809 * @mapping: the page's address_space 1810 * @index: the page index 1811 * @filler: function to perform the read 1812 * @data: destination for read data 1813 * 1814 * Read into the page cache. If a page already exists, and PageUptodate() is 1815 * not set, try to fill the page then wait for it to become unlocked. 1816 * 1817 * If the page does not get brought uptodate, return -EIO. 1818 */ 1819struct page *read_cache_page(struct address_space *mapping, 1820 unsigned long index, 1821 int (*filler)(void *,struct page*), 1822 void *data) 1823{ 1824 struct page *page; 1825 1826 page = read_cache_page_async(mapping, index, filler, data); 1827 if (IS_ERR(page)) 1828 goto out; 1829 wait_on_page_locked(page); 1830 if (!PageUptodate(page)) { 1831 page_cache_release(page); 1832 page = ERR_PTR(-EIO); 1833 } 1834 out: 1835 return page; 1836} 1837EXPORT_SYMBOL(read_cache_page); 1838 1839/* 1840 * If the page was newly created, increment its refcount and add it to the 1841 * caller's lru-buffering pagevec. This function is specifically for 1842 * generic_file_write(). 1843 */ 1844static inline struct page * 1845__grab_cache_page(struct address_space *mapping, unsigned long index, 1846 struct page **cached_page, struct pagevec *lru_pvec) 1847{ 1848 int err; 1849 struct page *page; 1850repeat: 1851 page = find_lock_page(mapping, index); 1852 if (!page) { 1853 if (!*cached_page) { 1854 *cached_page = page_cache_alloc(mapping); 1855 if (!*cached_page) 1856 return NULL; 1857 } 1858 err = add_to_page_cache(*cached_page, mapping, 1859 index, GFP_KERNEL); 1860 if (err == -EEXIST) 1861 goto repeat; 1862 if (err == 0) { 1863 page = *cached_page; 1864 page_cache_get(page); 1865 if (!pagevec_add(lru_pvec, page)) 1866 __pagevec_lru_add(lru_pvec); 1867 *cached_page = NULL; 1868 } 1869 } 1870 return page; 1871} 1872 1873/* 1874 * The logic we want is 1875 * 1876 * if suid or (sgid and xgrp) 1877 * remove privs 1878 */ 1879int should_remove_suid(struct dentry *dentry) 1880{ 1881 mode_t mode = dentry->d_inode->i_mode; 1882 int kill = 0; 1883 1884 /* suid always must be killed */ 1885 if (unlikely(mode & S_ISUID)) 1886 kill = ATTR_KILL_SUID; 1887 1888 /* 1889 * sgid without any exec bits is just a mandatory locking mark; leave 1890 * it alone. If some exec bits are set, it's a real sgid; kill it. 1891 */ 1892 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 1893 kill |= ATTR_KILL_SGID; 1894 1895 if (unlikely(kill && !capable(CAP_FSETID))) 1896 return kill; 1897 1898 return 0; 1899} 1900EXPORT_SYMBOL(should_remove_suid); 1901 1902int __remove_suid(struct dentry *dentry, int kill) 1903{ 1904 struct iattr newattrs; 1905 1906 newattrs.ia_valid = ATTR_FORCE | kill; 1907 return notify_change(dentry, &newattrs); 1908} 1909 1910int remove_suid(struct dentry *dentry) 1911{ 1912 int kill = should_remove_suid(dentry); 1913 1914 if (unlikely(kill)) 1915 return __remove_suid(dentry, kill); 1916 1917 return 0; 1918} 1919EXPORT_SYMBOL(remove_suid); 1920 1921size_t 1922__filemap_copy_from_user_iovec_inatomic(char *vaddr, 1923 const struct iovec *iov, size_t base, size_t bytes) 1924{ 1925 size_t copied = 0, left = 0; 1926 1927 while (bytes) { 1928 char __user *buf = iov->iov_base + base; 1929 int copy = min(bytes, iov->iov_len - base); 1930 1931 base = 0; 1932 left = __copy_from_user_inatomic_nocache(vaddr, buf, copy); 1933 copied += copy; 1934 bytes -= copy; 1935 vaddr += copy; 1936 iov++; 1937 1938 if (unlikely(left)) 1939 break; 1940 } 1941 return copied - left; 1942} 1943 1944/* 1945 * Performs necessary checks before doing a write 1946 * 1947 * Can adjust writing position or amount of bytes to write. 1948 * Returns appropriate error code that caller should return or 1949 * zero in case that write should be allowed. 1950 */ 1951inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk) 1952{ 1953 struct inode *inode = file->f_mapping->host; 1954 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; 1955 1956 if (unlikely(*pos < 0)) 1957 return -EINVAL; 1958 1959 if (!isblk) { 1960 /* FIXME: this is for backwards compatibility with 2.4 */ 1961 if (file->f_flags & O_APPEND) 1962 *pos = i_size_read(inode); 1963 1964 if (limit != RLIM_INFINITY) { 1965 if (*pos >= limit) { 1966 send_sig(SIGXFSZ, current, 0); 1967 return -EFBIG; 1968 } 1969 if (*count > limit - (typeof(limit))*pos) { 1970 *count = limit - (typeof(limit))*pos; 1971 } 1972 } 1973 } 1974 1975 /* 1976 * LFS rule 1977 */ 1978 if (unlikely(*pos + *count > MAX_NON_LFS && 1979 !(file->f_flags & O_LARGEFILE))) { 1980 if (*pos >= MAX_NON_LFS) { 1981 send_sig(SIGXFSZ, current, 0); 1982 return -EFBIG; 1983 } 1984 if (*count > MAX_NON_LFS - (unsigned long)*pos) { 1985 *count = MAX_NON_LFS - (unsigned long)*pos; 1986 } 1987 } 1988 1989 /* 1990 * Are we about to exceed the fs block limit ? 1991 * 1992 * If we have written data it becomes a short write. If we have 1993 * exceeded without writing data we send a signal and return EFBIG. 1994 * Linus frestrict idea will clean these up nicely.. 1995 */ 1996 if (likely(!isblk)) { 1997 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) { 1998 if (*count || *pos > inode->i_sb->s_maxbytes) { 1999 send_sig(SIGXFSZ, current, 0); 2000 return -EFBIG; 2001 } 2002 /* zero-length writes at ->s_maxbytes are OK */ 2003 } 2004 2005 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes)) 2006 *count = inode->i_sb->s_maxbytes - *pos; 2007 } else { 2008#ifdef CONFIG_BLOCK 2009 loff_t isize; 2010 if (bdev_read_only(I_BDEV(inode))) 2011 return -EPERM; 2012 isize = i_size_read(inode); 2013 if (*pos >= isize) { 2014 if (*count || *pos > isize) 2015 return -ENOSPC; 2016 } 2017 2018 if (*pos + *count > isize) 2019 *count = isize - *pos; 2020#else 2021 return -EPERM; 2022#endif 2023 } 2024 return 0; 2025} 2026EXPORT_SYMBOL(generic_write_checks); 2027 2028ssize_t 2029generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov, 2030 unsigned long *nr_segs, loff_t pos, loff_t *ppos, 2031 size_t count, size_t ocount) 2032{ 2033 struct file *file = iocb->ki_filp; 2034 struct address_space *mapping = file->f_mapping; 2035 struct inode *inode = mapping->host; 2036 ssize_t written; 2037 2038 if (count != ocount) 2039 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count); 2040 2041 written = generic_file_direct_IO(WRITE, iocb, iov, pos, *nr_segs); 2042 if (written > 0) { 2043 loff_t end = pos + written; 2044 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { 2045 i_size_write(inode, end); 2046 mark_inode_dirty(inode); 2047 } 2048 *ppos = end; 2049 } 2050 2051 /* 2052 * Sync the fs metadata but not the minor inode changes and 2053 * of course not the data as we did direct DMA for the IO. 2054 * i_mutex is held, which protects generic_osync_inode() from 2055 * livelocking. AIO O_DIRECT ops attempt to sync metadata here. 2056 */ 2057 if ((written >= 0 || written == -EIOCBQUEUED) && 2058 ((file->f_flags & O_SYNC) || IS_SYNC(inode))) { 2059 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA); 2060 if (err < 0) 2061 written = err; 2062 } 2063 return written; 2064} 2065EXPORT_SYMBOL(generic_file_direct_write); 2066 2067ssize_t 2068generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov, 2069 unsigned long nr_segs, loff_t pos, loff_t *ppos, 2070 size_t count, ssize_t written) 2071{ 2072 struct file *file = iocb->ki_filp; 2073 struct address_space * mapping = file->f_mapping; 2074 const struct address_space_operations *a_ops = mapping->a_ops; 2075 struct inode *inode = mapping->host; 2076 long status = 0; 2077 struct page *page; 2078 struct page *cached_page = NULL; 2079 size_t bytes; 2080 struct pagevec lru_pvec; 2081 const struct iovec *cur_iov = iov; /* current iovec */ 2082 size_t iov_base = 0; /* offset in the current iovec */ 2083 char __user *buf; 2084 2085 pagevec_init(&lru_pvec, 0); 2086 2087 /* 2088 * handle partial DIO write. Adjust cur_iov if needed. 2089 */ 2090 if (likely(nr_segs == 1)) 2091 buf = iov->iov_base + written; 2092 else { 2093 filemap_set_next_iovec(&cur_iov, &iov_base, written); 2094 buf = cur_iov->iov_base + iov_base; 2095 } 2096 2097 do { 2098 unsigned long index; 2099 unsigned long offset; 2100 size_t copied; 2101 2102 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */ 2103 index = pos >> PAGE_CACHE_SHIFT; 2104 bytes = PAGE_CACHE_SIZE - offset; 2105 2106 /* Limit the size of the copy to the caller's write size */ 2107 bytes = min(bytes, count); 2108 2109 /* We only need to worry about prefaulting when writes are from 2110 * user-space. NFSd uses vfs_writev with several non-aligned 2111 * segments in the vector, and limiting to one segment a time is 2112 * a noticeable performance for re-write 2113 */ 2114 if (!segment_eq(get_fs(), KERNEL_DS)) { 2115 /* 2116 * Limit the size of the copy to that of the current 2117 * segment, because fault_in_pages_readable() doesn't 2118 * know how to walk segments. 2119 */ 2120 bytes = min(bytes, cur_iov->iov_len - iov_base); 2121 2122 /* 2123 * Bring in the user page that we will copy from 2124 * _first_. Otherwise there's a nasty deadlock on 2125 * copying from the same page as we're writing to, 2126 * without it being marked up-to-date. 2127 */ 2128 fault_in_pages_readable(buf, bytes); 2129 } 2130 page = __grab_cache_page(mapping,index,&cached_page,&lru_pvec); 2131 if (!page) { 2132 status = -ENOMEM; 2133 break; 2134 } 2135 2136 if (unlikely(bytes == 0)) { 2137 status = 0; 2138 copied = 0; 2139 goto zero_length_segment; 2140 } 2141 2142 status = a_ops->prepare_write(file, page, offset, offset+bytes); 2143 if (unlikely(status)) { 2144 loff_t isize = i_size_read(inode); 2145 2146 if (status != AOP_TRUNCATED_PAGE) 2147 unlock_page(page); 2148 page_cache_release(page); 2149 if (status == AOP_TRUNCATED_PAGE) 2150 continue; 2151 /* 2152 * prepare_write() may have instantiated a few blocks 2153 * outside i_size. Trim these off again. 2154 */ 2155 if (pos + bytes > isize) 2156 vmtruncate(inode, isize); 2157 break; 2158 } 2159 if (likely(nr_segs == 1)) 2160 copied = filemap_copy_from_user(page, offset, 2161 buf, bytes); 2162 else 2163 copied = filemap_copy_from_user_iovec(page, offset, 2164 cur_iov, iov_base, bytes); 2165 flush_dcache_page(page); 2166 status = a_ops->commit_write(file, page, offset, offset+bytes); 2167 if (status == AOP_TRUNCATED_PAGE) { 2168 page_cache_release(page); 2169 continue; 2170 } 2171zero_length_segment: 2172 if (likely(copied >= 0)) { 2173 if (!status) 2174 status = copied; 2175 2176 if (status >= 0) { 2177 written += status; 2178 count -= status; 2179 pos += status; 2180 buf += status; 2181 if (unlikely(nr_segs > 1)) { 2182 filemap_set_next_iovec(&cur_iov, 2183 &iov_base, status); 2184 if (count) 2185 buf = cur_iov->iov_base + 2186 iov_base; 2187 } else { 2188 iov_base += status; 2189 } 2190 } 2191 } 2192 if (unlikely(copied != bytes)) 2193 if (status >= 0) 2194 status = -EFAULT; 2195 unlock_page(page); 2196 mark_page_accessed(page); 2197 page_cache_release(page); 2198 if (status < 0) 2199 break; 2200 balance_dirty_pages_ratelimited(mapping); 2201 cond_resched(); 2202 } while (count); 2203 *ppos = pos; 2204 2205 if (cached_page) 2206 page_cache_release(cached_page); 2207 2208 /* 2209 * For now, when the user asks for O_SYNC, we'll actually give O_DSYNC 2210 */ 2211 if (likely(status >= 0)) { 2212 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) { 2213 if (!a_ops->writepage || !is_sync_kiocb(iocb)) 2214 status = generic_osync_inode(inode, mapping, 2215 OSYNC_METADATA|OSYNC_DATA); 2216 } 2217 } 2218 2219 /* 2220 * If we get here for O_DIRECT writes then we must have fallen through 2221 * to buffered writes (block instantiation inside i_size). So we sync 2222 * the file data here, to try to honour O_DIRECT expectations. 2223 */ 2224 if (unlikely(file->f_flags & O_DIRECT) && written) 2225 status = filemap_write_and_wait(mapping); 2226 2227 pagevec_lru_add(&lru_pvec); 2228 return written ? written : status; 2229} 2230EXPORT_SYMBOL(generic_file_buffered_write); 2231 2232static ssize_t 2233__generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov, 2234 unsigned long nr_segs, loff_t *ppos) 2235{ 2236 struct file *file = iocb->ki_filp; 2237 struct address_space * mapping = file->f_mapping; 2238 size_t ocount; /* original count */ 2239 size_t count; /* after file limit checks */ 2240 struct inode *inode = mapping->host; 2241 unsigned long seg; 2242 loff_t pos; 2243 ssize_t written; 2244 ssize_t err; 2245 2246 ocount = 0; 2247 for (seg = 0; seg < nr_segs; seg++) { 2248 const struct iovec *iv = &iov[seg]; 2249 2250 /* 2251 * If any segment has a negative length, or the cumulative 2252 * length ever wraps negative then return -EINVAL. 2253 */ 2254 ocount += iv->iov_len; 2255 if (unlikely((ssize_t)(ocount|iv->iov_len) < 0)) 2256 return -EINVAL; 2257 if (access_ok(VERIFY_READ, iv->iov_base, iv->iov_len)) 2258 continue; 2259 if (seg == 0) 2260 return -EFAULT; 2261 nr_segs = seg; 2262 ocount -= iv->iov_len; /* This segment is no good */ 2263 break; 2264 } 2265 2266 count = ocount; 2267 pos = *ppos; 2268 2269 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); 2270 2271 /* We can write back this queue in page reclaim */ 2272 current->backing_dev_info = mapping->backing_dev_info; 2273 written = 0; 2274 2275 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); 2276 if (err) 2277 goto out; 2278 2279 if (count == 0) 2280 goto out; 2281 2282 err = remove_suid(file->f_path.dentry); 2283 if (err) 2284 goto out; 2285 2286 file_update_time(file); 2287 2288 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ 2289 if (unlikely(file->f_flags & O_DIRECT)) { 2290 loff_t endbyte; 2291 ssize_t written_buffered; 2292 2293 written = generic_file_direct_write(iocb, iov, &nr_segs, pos, 2294 ppos, count, ocount); 2295 if (written < 0 || written == count) 2296 goto out; 2297 /* 2298 * direct-io write to a hole: fall through to buffered I/O 2299 * for completing the rest of the request. 2300 */ 2301 pos += written; 2302 count -= written; 2303 written_buffered = generic_file_buffered_write(iocb, iov, 2304 nr_segs, pos, ppos, count, 2305 written); 2306 /* 2307 * If generic_file_buffered_write() retuned a synchronous error 2308 * then we want to return the number of bytes which were 2309 * direct-written, or the error code if that was zero. Note 2310 * that this differs from normal direct-io semantics, which 2311 * will return -EFOO even if some bytes were written. 2312 */ 2313 if (written_buffered < 0) { 2314 err = written_buffered; 2315 goto out; 2316 } 2317 2318 /* 2319 * We need to ensure that the page cache pages are written to 2320 * disk and invalidated to preserve the expected O_DIRECT 2321 * semantics. 2322 */ 2323 endbyte = pos + written_buffered - written - 1; 2324 err = do_sync_file_range(file, pos, endbyte, 2325 SYNC_FILE_RANGE_WAIT_BEFORE| 2326 SYNC_FILE_RANGE_WRITE| 2327 SYNC_FILE_RANGE_WAIT_AFTER); 2328 if (err == 0) { 2329 written = written_buffered; 2330 invalidate_mapping_pages(mapping, 2331 pos >> PAGE_CACHE_SHIFT, 2332 endbyte >> PAGE_CACHE_SHIFT); 2333 } else { 2334 /* 2335 * We don't know how much we wrote, so just return 2336 * the number of bytes which were direct-written 2337 */ 2338 } 2339 } else { 2340 written = generic_file_buffered_write(iocb, iov, nr_segs, 2341 pos, ppos, count, written); 2342 } 2343out: 2344 current->backing_dev_info = NULL; 2345 return written ? written : err; 2346} 2347 2348ssize_t generic_file_aio_write_nolock(struct kiocb *iocb, 2349 const struct iovec *iov, unsigned long nr_segs, loff_t pos) 2350{ 2351 struct file *file = iocb->ki_filp; 2352 struct address_space *mapping = file->f_mapping; 2353 struct inode *inode = mapping->host; 2354 ssize_t ret; 2355 2356 BUG_ON(iocb->ki_pos != pos); 2357 2358 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs, 2359 &iocb->ki_pos); 2360 2361 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) { 2362 ssize_t err; 2363 2364 err = sync_page_range_nolock(inode, mapping, pos, ret); 2365 if (err < 0) 2366 ret = err; 2367 } 2368 return ret; 2369} 2370EXPORT_SYMBOL(generic_file_aio_write_nolock); 2371 2372ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, 2373 unsigned long nr_segs, loff_t pos) 2374{ 2375 struct file *file = iocb->ki_filp; 2376 struct address_space *mapping = file->f_mapping; 2377 struct inode *inode = mapping->host; 2378 ssize_t ret; 2379 2380 BUG_ON(iocb->ki_pos != pos); 2381 2382 mutex_lock(&inode->i_mutex); 2383 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs, 2384 &iocb->ki_pos); 2385 mutex_unlock(&inode->i_mutex); 2386 2387 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) { 2388 ssize_t err; 2389 2390 err = sync_page_range(inode, mapping, pos, ret); 2391 if (err < 0) 2392 ret = err; 2393 } 2394 return ret; 2395} 2396EXPORT_SYMBOL(generic_file_aio_write); 2397 2398/* 2399 * Called under i_mutex for writes to S_ISREG files. Returns -EIO if something 2400 * went wrong during pagecache shootdown. 2401 */ 2402static ssize_t 2403generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, 2404 loff_t offset, unsigned long nr_segs) 2405{ 2406 struct file *file = iocb->ki_filp; 2407 struct address_space *mapping = file->f_mapping; 2408 ssize_t retval; 2409 size_t write_len; 2410 pgoff_t end = 0; /* silence gcc */ 2411 2412 /* 2413 * If it's a write, unmap all mmappings of the file up-front. This 2414 * will cause any pte dirty bits to be propagated into the pageframes 2415 * for the subsequent filemap_write_and_wait(). 2416 */ 2417 if (rw == WRITE) { 2418 write_len = iov_length(iov, nr_segs); 2419 end = (offset + write_len - 1) >> PAGE_CACHE_SHIFT; 2420 if (mapping_mapped(mapping)) 2421 unmap_mapping_range(mapping, offset, write_len, 0); 2422 } 2423 2424 retval = filemap_write_and_wait(mapping); 2425 if (retval) 2426 goto out; 2427 2428 /* 2429 * After a write we want buffered reads to be sure to go to disk to get 2430 * the new data. We invalidate clean cached page from the region we're 2431 * about to write. We do this *before* the write so that we can return 2432 * -EIO without clobbering -EIOCBQUEUED from ->direct_IO(). 2433 */ 2434 if (rw == WRITE && mapping->nrpages) { 2435 retval = invalidate_inode_pages2_range(mapping, 2436 offset >> PAGE_CACHE_SHIFT, end); 2437 if (retval) 2438 goto out; 2439 } 2440 2441 retval = mapping->a_ops->direct_IO(rw, iocb, iov, offset, nr_segs); 2442 if (retval) 2443 goto out; 2444 2445 /* 2446 * Finally, try again to invalidate clean pages which might have been 2447 * faulted in by get_user_pages() if the source of the write was an 2448 * mmap()ed region of the file we're writing. That's a pretty crazy 2449 * thing to do, so we don't support it 100%. If this invalidation 2450 * fails and we have -EIOCBQUEUED we ignore the failure. 2451 */ 2452 if (rw == WRITE && mapping->nrpages) { 2453 int err = invalidate_inode_pages2_range(mapping, 2454 offset >> PAGE_CACHE_SHIFT, end); 2455 if (err && retval >= 0) 2456 retval = err; 2457 } 2458out: 2459 return retval; 2460} 2461 2462/** 2463 * try_to_release_page() - release old fs-specific metadata on a page 2464 * 2465 * @page: the page which the kernel is trying to free 2466 * @gfp_mask: memory allocation flags (and I/O mode) 2467 * 2468 * The address_space is to try to release any data against the page 2469 * (presumably at page->private). If the release was successful, return `1'. 2470 * Otherwise return zero. 2471 * 2472 * The @gfp_mask argument specifies whether I/O may be performed to release 2473 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT). 2474 * 2475 * NOTE: @gfp_mask may go away, and this function may become non-blocking. 2476 */ 2477int try_to_release_page(struct page *page, gfp_t gfp_mask) 2478{ 2479 struct address_space * const mapping = page->mapping; 2480 2481 BUG_ON(!PageLocked(page)); 2482 if (PageWriteback(page)) 2483 return 0; 2484 2485 if (mapping && mapping->a_ops->releasepage) 2486 return mapping->a_ops->releasepage(page, gfp_mask); 2487 return try_to_free_buffers(page); 2488} 2489 2490EXPORT_SYMBOL(try_to_release_page); 2491