filemap.c revision 08291429cfa6258c4cd95d8833beb40f828b194e
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 BUG_ON(page_mapped(page)); 124} 125 126void remove_from_page_cache(struct page *page) 127{ 128 struct address_space *mapping = page->mapping; 129 130 BUG_ON(!PageLocked(page)); 131 132 write_lock_irq(&mapping->tree_lock); 133 __remove_from_page_cache(page); 134 write_unlock_irq(&mapping->tree_lock); 135} 136 137static int sync_page(void *word) 138{ 139 struct address_space *mapping; 140 struct page *page; 141 142 page = container_of((unsigned long *)word, struct page, flags); 143 144 /* 145 * page_mapping() is being called without PG_locked held. 146 * Some knowledge of the state and use of the page is used to 147 * reduce the requirements down to a memory barrier. 148 * The danger here is of a stale page_mapping() return value 149 * indicating a struct address_space different from the one it's 150 * associated with when it is associated with one. 151 * After smp_mb(), it's either the correct page_mapping() for 152 * the page, or an old page_mapping() and the page's own 153 * page_mapping() has gone NULL. 154 * The ->sync_page() address_space operation must tolerate 155 * page_mapping() going NULL. By an amazing coincidence, 156 * this comes about because none of the users of the page 157 * in the ->sync_page() methods make essential use of the 158 * page_mapping(), merely passing the page down to the backing 159 * device's unplug functions when it's non-NULL, which in turn 160 * ignore it for all cases but swap, where only page_private(page) is 161 * of interest. When page_mapping() does go NULL, the entire 162 * call stack gracefully ignores the page and returns. 163 * -- wli 164 */ 165 smp_mb(); 166 mapping = page_mapping(page); 167 if (mapping && mapping->a_ops && mapping->a_ops->sync_page) 168 mapping->a_ops->sync_page(page); 169 io_schedule(); 170 return 0; 171} 172 173/** 174 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range 175 * @mapping: address space structure to write 176 * @start: offset in bytes where the range starts 177 * @end: offset in bytes where the range ends (inclusive) 178 * @sync_mode: enable synchronous operation 179 * 180 * Start writeback against all of a mapping's dirty pages that lie 181 * within the byte offsets <start, end> inclusive. 182 * 183 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as 184 * opposed to a regular memory cleansing writeback. The difference between 185 * these two operations is that if a dirty page/buffer is encountered, it must 186 * be waited upon, and not just skipped over. 187 */ 188int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, 189 loff_t end, int sync_mode) 190{ 191 int ret; 192 struct writeback_control wbc = { 193 .sync_mode = sync_mode, 194 .nr_to_write = mapping->nrpages * 2, 195 .range_start = start, 196 .range_end = end, 197 }; 198 199 if (!mapping_cap_writeback_dirty(mapping)) 200 return 0; 201 202 ret = do_writepages(mapping, &wbc); 203 return ret; 204} 205 206static inline int __filemap_fdatawrite(struct address_space *mapping, 207 int sync_mode) 208{ 209 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); 210} 211 212int filemap_fdatawrite(struct address_space *mapping) 213{ 214 return __filemap_fdatawrite(mapping, WB_SYNC_ALL); 215} 216EXPORT_SYMBOL(filemap_fdatawrite); 217 218static int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, 219 loff_t end) 220{ 221 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); 222} 223 224/** 225 * filemap_flush - mostly a non-blocking flush 226 * @mapping: target address_space 227 * 228 * This is a mostly non-blocking flush. Not suitable for data-integrity 229 * purposes - I/O may not be started against all dirty pages. 230 */ 231int filemap_flush(struct address_space *mapping) 232{ 233 return __filemap_fdatawrite(mapping, WB_SYNC_NONE); 234} 235EXPORT_SYMBOL(filemap_flush); 236 237/** 238 * wait_on_page_writeback_range - wait for writeback to complete 239 * @mapping: target address_space 240 * @start: beginning page index 241 * @end: ending page index 242 * 243 * Wait for writeback to complete against pages indexed by start->end 244 * inclusive 245 */ 246int wait_on_page_writeback_range(struct address_space *mapping, 247 pgoff_t start, pgoff_t end) 248{ 249 struct pagevec pvec; 250 int nr_pages; 251 int ret = 0; 252 pgoff_t index; 253 254 if (end < start) 255 return 0; 256 257 pagevec_init(&pvec, 0); 258 index = start; 259 while ((index <= end) && 260 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, 261 PAGECACHE_TAG_WRITEBACK, 262 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) { 263 unsigned i; 264 265 for (i = 0; i < nr_pages; i++) { 266 struct page *page = pvec.pages[i]; 267 268 /* until radix tree lookup accepts end_index */ 269 if (page->index > end) 270 continue; 271 272 wait_on_page_writeback(page); 273 if (PageError(page)) 274 ret = -EIO; 275 } 276 pagevec_release(&pvec); 277 cond_resched(); 278 } 279 280 /* Check for outstanding write errors */ 281 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags)) 282 ret = -ENOSPC; 283 if (test_and_clear_bit(AS_EIO, &mapping->flags)) 284 ret = -EIO; 285 286 return ret; 287} 288 289/** 290 * sync_page_range - write and wait on all pages in the passed range 291 * @inode: target inode 292 * @mapping: target address_space 293 * @pos: beginning offset in pages to write 294 * @count: number of bytes to write 295 * 296 * Write and wait upon all the pages in the passed range. This is a "data 297 * integrity" operation. It waits upon in-flight writeout before starting and 298 * waiting upon new writeout. If there was an IO error, return it. 299 * 300 * We need to re-take i_mutex during the generic_osync_inode list walk because 301 * it is otherwise livelockable. 302 */ 303int sync_page_range(struct inode *inode, struct address_space *mapping, 304 loff_t pos, loff_t count) 305{ 306 pgoff_t start = pos >> PAGE_CACHE_SHIFT; 307 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT; 308 int ret; 309 310 if (!mapping_cap_writeback_dirty(mapping) || !count) 311 return 0; 312 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1); 313 if (ret == 0) { 314 mutex_lock(&inode->i_mutex); 315 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA); 316 mutex_unlock(&inode->i_mutex); 317 } 318 if (ret == 0) 319 ret = wait_on_page_writeback_range(mapping, start, end); 320 return ret; 321} 322EXPORT_SYMBOL(sync_page_range); 323 324/** 325 * sync_page_range_nolock 326 * @inode: target inode 327 * @mapping: target address_space 328 * @pos: beginning offset in pages to write 329 * @count: number of bytes to write 330 * 331 * Note: Holding i_mutex across sync_page_range_nolock() is not a good idea 332 * as it forces O_SYNC writers to different parts of the same file 333 * to be serialised right until io completion. 334 */ 335int sync_page_range_nolock(struct inode *inode, struct address_space *mapping, 336 loff_t pos, loff_t count) 337{ 338 pgoff_t start = pos >> PAGE_CACHE_SHIFT; 339 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT; 340 int ret; 341 342 if (!mapping_cap_writeback_dirty(mapping) || !count) 343 return 0; 344 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1); 345 if (ret == 0) 346 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA); 347 if (ret == 0) 348 ret = wait_on_page_writeback_range(mapping, start, end); 349 return ret; 350} 351EXPORT_SYMBOL(sync_page_range_nolock); 352 353/** 354 * filemap_fdatawait - wait for all under-writeback pages to complete 355 * @mapping: address space structure to wait for 356 * 357 * Walk the list of under-writeback pages of the given address space 358 * and wait for all of them. 359 */ 360int filemap_fdatawait(struct address_space *mapping) 361{ 362 loff_t i_size = i_size_read(mapping->host); 363 364 if (i_size == 0) 365 return 0; 366 367 return wait_on_page_writeback_range(mapping, 0, 368 (i_size - 1) >> PAGE_CACHE_SHIFT); 369} 370EXPORT_SYMBOL(filemap_fdatawait); 371 372int filemap_write_and_wait(struct address_space *mapping) 373{ 374 int err = 0; 375 376 if (mapping->nrpages) { 377 err = filemap_fdatawrite(mapping); 378 /* 379 * Even if the above returned error, the pages may be 380 * written partially (e.g. -ENOSPC), so we wait for it. 381 * But the -EIO is special case, it may indicate the worst 382 * thing (e.g. bug) happened, so we avoid waiting for it. 383 */ 384 if (err != -EIO) { 385 int err2 = filemap_fdatawait(mapping); 386 if (!err) 387 err = err2; 388 } 389 } 390 return err; 391} 392EXPORT_SYMBOL(filemap_write_and_wait); 393 394/** 395 * filemap_write_and_wait_range - write out & wait on a file range 396 * @mapping: the address_space for the pages 397 * @lstart: offset in bytes where the range starts 398 * @lend: offset in bytes where the range ends (inclusive) 399 * 400 * Write out and wait upon file offsets lstart->lend, inclusive. 401 * 402 * Note that `lend' is inclusive (describes the last byte to be written) so 403 * that this function can be used to write to the very end-of-file (end = -1). 404 */ 405int filemap_write_and_wait_range(struct address_space *mapping, 406 loff_t lstart, loff_t lend) 407{ 408 int err = 0; 409 410 if (mapping->nrpages) { 411 err = __filemap_fdatawrite_range(mapping, lstart, lend, 412 WB_SYNC_ALL); 413 /* See comment of filemap_write_and_wait() */ 414 if (err != -EIO) { 415 int err2 = wait_on_page_writeback_range(mapping, 416 lstart >> PAGE_CACHE_SHIFT, 417 lend >> PAGE_CACHE_SHIFT); 418 if (!err) 419 err = err2; 420 } 421 } 422 return err; 423} 424 425/** 426 * add_to_page_cache - add newly allocated pagecache pages 427 * @page: page to add 428 * @mapping: the page's address_space 429 * @offset: page index 430 * @gfp_mask: page allocation mode 431 * 432 * This function is used to add newly allocated pagecache pages; 433 * the page is new, so we can just run SetPageLocked() against it. 434 * The other page state flags were set by rmqueue(). 435 * 436 * This function does not add the page to the LRU. The caller must do that. 437 */ 438int add_to_page_cache(struct page *page, struct address_space *mapping, 439 pgoff_t offset, gfp_t gfp_mask) 440{ 441 int error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); 442 443 if (error == 0) { 444 write_lock_irq(&mapping->tree_lock); 445 error = radix_tree_insert(&mapping->page_tree, offset, page); 446 if (!error) { 447 page_cache_get(page); 448 SetPageLocked(page); 449 page->mapping = mapping; 450 page->index = offset; 451 mapping->nrpages++; 452 __inc_zone_page_state(page, NR_FILE_PAGES); 453 } 454 write_unlock_irq(&mapping->tree_lock); 455 radix_tree_preload_end(); 456 } 457 return error; 458} 459EXPORT_SYMBOL(add_to_page_cache); 460 461int add_to_page_cache_lru(struct page *page, struct address_space *mapping, 462 pgoff_t offset, gfp_t gfp_mask) 463{ 464 int ret = add_to_page_cache(page, mapping, offset, gfp_mask); 465 if (ret == 0) 466 lru_cache_add(page); 467 return ret; 468} 469 470#ifdef CONFIG_NUMA 471struct page *__page_cache_alloc(gfp_t gfp) 472{ 473 if (cpuset_do_page_mem_spread()) { 474 int n = cpuset_mem_spread_node(); 475 return alloc_pages_node(n, gfp, 0); 476 } 477 return alloc_pages(gfp, 0); 478} 479EXPORT_SYMBOL(__page_cache_alloc); 480#endif 481 482static int __sleep_on_page_lock(void *word) 483{ 484 io_schedule(); 485 return 0; 486} 487 488/* 489 * In order to wait for pages to become available there must be 490 * waitqueues associated with pages. By using a hash table of 491 * waitqueues where the bucket discipline is to maintain all 492 * waiters on the same queue and wake all when any of the pages 493 * become available, and for the woken contexts to check to be 494 * sure the appropriate page became available, this saves space 495 * at a cost of "thundering herd" phenomena during rare hash 496 * collisions. 497 */ 498static wait_queue_head_t *page_waitqueue(struct page *page) 499{ 500 const struct zone *zone = page_zone(page); 501 502 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; 503} 504 505static inline void wake_up_page(struct page *page, int bit) 506{ 507 __wake_up_bit(page_waitqueue(page), &page->flags, bit); 508} 509 510void fastcall wait_on_page_bit(struct page *page, int bit_nr) 511{ 512 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); 513 514 if (test_bit(bit_nr, &page->flags)) 515 __wait_on_bit(page_waitqueue(page), &wait, sync_page, 516 TASK_UNINTERRUPTIBLE); 517} 518EXPORT_SYMBOL(wait_on_page_bit); 519 520/** 521 * unlock_page - unlock a locked page 522 * @page: the page 523 * 524 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). 525 * Also wakes sleepers in wait_on_page_writeback() because the wakeup 526 * mechananism between PageLocked pages and PageWriteback pages is shared. 527 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. 528 * 529 * The first mb is necessary to safely close the critical section opened by the 530 * TestSetPageLocked(), the second mb is necessary to enforce ordering between 531 * the clear_bit and the read of the waitqueue (to avoid SMP races with a 532 * parallel wait_on_page_locked()). 533 */ 534void fastcall unlock_page(struct page *page) 535{ 536 smp_mb__before_clear_bit(); 537 if (!TestClearPageLocked(page)) 538 BUG(); 539 smp_mb__after_clear_bit(); 540 wake_up_page(page, PG_locked); 541} 542EXPORT_SYMBOL(unlock_page); 543 544/** 545 * end_page_writeback - end writeback against a page 546 * @page: the page 547 */ 548void end_page_writeback(struct page *page) 549{ 550 if (!TestClearPageReclaim(page) || rotate_reclaimable_page(page)) { 551 if (!test_clear_page_writeback(page)) 552 BUG(); 553 } 554 smp_mb__after_clear_bit(); 555 wake_up_page(page, PG_writeback); 556} 557EXPORT_SYMBOL(end_page_writeback); 558 559/** 560 * __lock_page - get a lock on the page, assuming we need to sleep to get it 561 * @page: the page to lock 562 * 563 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some 564 * random driver's requestfn sets TASK_RUNNING, we could busywait. However 565 * chances are that on the second loop, the block layer's plug list is empty, 566 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE. 567 */ 568void fastcall __lock_page(struct page *page) 569{ 570 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); 571 572 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page, 573 TASK_UNINTERRUPTIBLE); 574} 575EXPORT_SYMBOL(__lock_page); 576 577/* 578 * Variant of lock_page that does not require the caller to hold a reference 579 * on the page's mapping. 580 */ 581void fastcall __lock_page_nosync(struct page *page) 582{ 583 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); 584 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock, 585 TASK_UNINTERRUPTIBLE); 586} 587 588/** 589 * find_get_page - find and get a page reference 590 * @mapping: the address_space to search 591 * @offset: the page index 592 * 593 * Is there a pagecache struct page at the given (mapping, offset) tuple? 594 * If yes, increment its refcount and return it; if no, return NULL. 595 */ 596struct page * find_get_page(struct address_space *mapping, pgoff_t offset) 597{ 598 struct page *page; 599 600 read_lock_irq(&mapping->tree_lock); 601 page = radix_tree_lookup(&mapping->page_tree, offset); 602 if (page) 603 page_cache_get(page); 604 read_unlock_irq(&mapping->tree_lock); 605 return page; 606} 607EXPORT_SYMBOL(find_get_page); 608 609/** 610 * find_lock_page - locate, pin and lock a pagecache page 611 * @mapping: the address_space to search 612 * @offset: the page index 613 * 614 * Locates the desired pagecache page, locks it, increments its reference 615 * count and returns its address. 616 * 617 * Returns zero if the page was not present. find_lock_page() may sleep. 618 */ 619struct page *find_lock_page(struct address_space *mapping, 620 pgoff_t offset) 621{ 622 struct page *page; 623 624repeat: 625 read_lock_irq(&mapping->tree_lock); 626 page = radix_tree_lookup(&mapping->page_tree, offset); 627 if (page) { 628 page_cache_get(page); 629 if (TestSetPageLocked(page)) { 630 read_unlock_irq(&mapping->tree_lock); 631 __lock_page(page); 632 633 /* Has the page been truncated while we slept? */ 634 if (unlikely(page->mapping != mapping)) { 635 unlock_page(page); 636 page_cache_release(page); 637 goto repeat; 638 } 639 VM_BUG_ON(page->index != offset); 640 goto out; 641 } 642 } 643 read_unlock_irq(&mapping->tree_lock); 644out: 645 return page; 646} 647EXPORT_SYMBOL(find_lock_page); 648 649/** 650 * find_or_create_page - locate or add a pagecache page 651 * @mapping: the page's address_space 652 * @index: the page's index into the mapping 653 * @gfp_mask: page allocation mode 654 * 655 * Locates a page in the pagecache. If the page is not present, a new page 656 * is allocated using @gfp_mask and is added to the pagecache and to the VM's 657 * LRU list. The returned page is locked and has its reference count 658 * incremented. 659 * 660 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic 661 * allocation! 662 * 663 * find_or_create_page() returns the desired page's address, or zero on 664 * memory exhaustion. 665 */ 666struct page *find_or_create_page(struct address_space *mapping, 667 pgoff_t index, gfp_t gfp_mask) 668{ 669 struct page *page; 670 int err; 671repeat: 672 page = find_lock_page(mapping, index); 673 if (!page) { 674 page = __page_cache_alloc(gfp_mask); 675 if (!page) 676 return NULL; 677 err = add_to_page_cache_lru(page, mapping, index, gfp_mask); 678 if (unlikely(err)) { 679 page_cache_release(page); 680 page = NULL; 681 if (err == -EEXIST) 682 goto repeat; 683 } 684 } 685 return page; 686} 687EXPORT_SYMBOL(find_or_create_page); 688 689/** 690 * find_get_pages - gang pagecache lookup 691 * @mapping: The address_space to search 692 * @start: The starting page index 693 * @nr_pages: The maximum number of pages 694 * @pages: Where the resulting pages are placed 695 * 696 * find_get_pages() will search for and return a group of up to 697 * @nr_pages pages in the mapping. The pages are placed at @pages. 698 * find_get_pages() takes a reference against the returned pages. 699 * 700 * The search returns a group of mapping-contiguous pages with ascending 701 * indexes. There may be holes in the indices due to not-present pages. 702 * 703 * find_get_pages() returns the number of pages which were found. 704 */ 705unsigned find_get_pages(struct address_space *mapping, pgoff_t start, 706 unsigned int nr_pages, struct page **pages) 707{ 708 unsigned int i; 709 unsigned int ret; 710 711 read_lock_irq(&mapping->tree_lock); 712 ret = radix_tree_gang_lookup(&mapping->page_tree, 713 (void **)pages, start, nr_pages); 714 for (i = 0; i < ret; i++) 715 page_cache_get(pages[i]); 716 read_unlock_irq(&mapping->tree_lock); 717 return ret; 718} 719 720/** 721 * find_get_pages_contig - gang contiguous pagecache lookup 722 * @mapping: The address_space to search 723 * @index: The starting page index 724 * @nr_pages: The maximum number of pages 725 * @pages: Where the resulting pages are placed 726 * 727 * find_get_pages_contig() works exactly like find_get_pages(), except 728 * that the returned number of pages are guaranteed to be contiguous. 729 * 730 * find_get_pages_contig() returns the number of pages which were found. 731 */ 732unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, 733 unsigned int nr_pages, struct page **pages) 734{ 735 unsigned int i; 736 unsigned int ret; 737 738 read_lock_irq(&mapping->tree_lock); 739 ret = radix_tree_gang_lookup(&mapping->page_tree, 740 (void **)pages, index, nr_pages); 741 for (i = 0; i < ret; i++) { 742 if (pages[i]->mapping == NULL || pages[i]->index != index) 743 break; 744 745 page_cache_get(pages[i]); 746 index++; 747 } 748 read_unlock_irq(&mapping->tree_lock); 749 return i; 750} 751EXPORT_SYMBOL(find_get_pages_contig); 752 753/** 754 * find_get_pages_tag - find and return pages that match @tag 755 * @mapping: the address_space to search 756 * @index: the starting page index 757 * @tag: the tag index 758 * @nr_pages: the maximum number of pages 759 * @pages: where the resulting pages are placed 760 * 761 * Like find_get_pages, except we only return pages which are tagged with 762 * @tag. We update @index to index the next page for the traversal. 763 */ 764unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, 765 int tag, unsigned int nr_pages, struct page **pages) 766{ 767 unsigned int i; 768 unsigned int ret; 769 770 read_lock_irq(&mapping->tree_lock); 771 ret = radix_tree_gang_lookup_tag(&mapping->page_tree, 772 (void **)pages, *index, nr_pages, tag); 773 for (i = 0; i < ret; i++) 774 page_cache_get(pages[i]); 775 if (ret) 776 *index = pages[ret - 1]->index + 1; 777 read_unlock_irq(&mapping->tree_lock); 778 return ret; 779} 780EXPORT_SYMBOL(find_get_pages_tag); 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, pgoff_t 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 pgoff_t index; 866 pgoff_t last_index; 867 pgoff_t prev_index; 868 unsigned long offset; /* offset into pagecache page */ 869 unsigned int prev_offset; 870 int error; 871 872 index = *ppos >> PAGE_CACHE_SHIFT; 873 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT; 874 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1); 875 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; 876 offset = *ppos & ~PAGE_CACHE_MASK; 877 878 for (;;) { 879 struct page *page; 880 pgoff_t end_index; 881 loff_t isize; 882 unsigned long nr, ret; 883 884 cond_resched(); 885find_page: 886 page = find_get_page(mapping, index); 887 if (!page) { 888 page_cache_sync_readahead(mapping, 889 ra, filp, 890 index, last_index - index); 891 page = find_get_page(mapping, index); 892 if (unlikely(page == NULL)) 893 goto no_cached_page; 894 } 895 if (PageReadahead(page)) { 896 page_cache_async_readahead(mapping, 897 ra, filp, page, 898 index, last_index - index); 899 } 900 if (!PageUptodate(page)) 901 goto page_not_up_to_date; 902page_ok: 903 /* 904 * i_size must be checked after we know the page is Uptodate. 905 * 906 * Checking i_size after the check allows us to calculate 907 * the correct value for "nr", which means the zero-filled 908 * part of the page is not copied back to userspace (unless 909 * another truncate extends the file - this is desired though). 910 */ 911 912 isize = i_size_read(inode); 913 end_index = (isize - 1) >> PAGE_CACHE_SHIFT; 914 if (unlikely(!isize || index > end_index)) { 915 page_cache_release(page); 916 goto out; 917 } 918 919 /* nr is the maximum number of bytes to copy from this page */ 920 nr = PAGE_CACHE_SIZE; 921 if (index == end_index) { 922 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; 923 if (nr <= offset) { 924 page_cache_release(page); 925 goto out; 926 } 927 } 928 nr = nr - offset; 929 930 /* If users can be writing to this page using arbitrary 931 * virtual addresses, take care about potential aliasing 932 * before reading the page on the kernel side. 933 */ 934 if (mapping_writably_mapped(mapping)) 935 flush_dcache_page(page); 936 937 /* 938 * When a sequential read accesses a page several times, 939 * only mark it as accessed the first time. 940 */ 941 if (prev_index != index || offset != prev_offset) 942 mark_page_accessed(page); 943 prev_index = index; 944 945 /* 946 * Ok, we have the page, and it's up-to-date, so 947 * now we can copy it to user space... 948 * 949 * The actor routine returns how many bytes were actually used.. 950 * NOTE! This may not be the same as how much of a user buffer 951 * we filled up (we may be padding etc), so we can only update 952 * "pos" here (the actor routine has to update the user buffer 953 * pointers and the remaining count). 954 */ 955 ret = actor(desc, page, offset, nr); 956 offset += ret; 957 index += offset >> PAGE_CACHE_SHIFT; 958 offset &= ~PAGE_CACHE_MASK; 959 prev_offset = offset; 960 961 page_cache_release(page); 962 if (ret == nr && desc->count) 963 continue; 964 goto out; 965 966page_not_up_to_date: 967 /* Get exclusive access to the page ... */ 968 lock_page(page); 969 970 /* Did it get truncated before we got the lock? */ 971 if (!page->mapping) { 972 unlock_page(page); 973 page_cache_release(page); 974 continue; 975 } 976 977 /* Did somebody else fill it already? */ 978 if (PageUptodate(page)) { 979 unlock_page(page); 980 goto page_ok; 981 } 982 983readpage: 984 /* Start the actual read. The read will unlock the page. */ 985 error = mapping->a_ops->readpage(filp, page); 986 987 if (unlikely(error)) { 988 if (error == AOP_TRUNCATED_PAGE) { 989 page_cache_release(page); 990 goto find_page; 991 } 992 goto readpage_error; 993 } 994 995 if (!PageUptodate(page)) { 996 lock_page(page); 997 if (!PageUptodate(page)) { 998 if (page->mapping == NULL) { 999 /* 1000 * invalidate_inode_pages got it 1001 */ 1002 unlock_page(page); 1003 page_cache_release(page); 1004 goto find_page; 1005 } 1006 unlock_page(page); 1007 error = -EIO; 1008 shrink_readahead_size_eio(filp, ra); 1009 goto readpage_error; 1010 } 1011 unlock_page(page); 1012 } 1013 1014 goto page_ok; 1015 1016readpage_error: 1017 /* UHHUH! A synchronous read error occurred. Report it */ 1018 desc->error = error; 1019 page_cache_release(page); 1020 goto out; 1021 1022no_cached_page: 1023 /* 1024 * Ok, it wasn't cached, so we need to create a new 1025 * page.. 1026 */ 1027 page = page_cache_alloc_cold(mapping); 1028 if (!page) { 1029 desc->error = -ENOMEM; 1030 goto out; 1031 } 1032 error = add_to_page_cache_lru(page, mapping, 1033 index, GFP_KERNEL); 1034 if (error) { 1035 page_cache_release(page); 1036 if (error == -EEXIST) 1037 goto find_page; 1038 desc->error = error; 1039 goto out; 1040 } 1041 goto readpage; 1042 } 1043 1044out: 1045 ra->prev_pos = prev_index; 1046 ra->prev_pos <<= PAGE_CACHE_SHIFT; 1047 ra->prev_pos |= prev_offset; 1048 1049 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset; 1050 if (filp) 1051 file_accessed(filp); 1052} 1053EXPORT_SYMBOL(do_generic_mapping_read); 1054 1055int file_read_actor(read_descriptor_t *desc, struct page *page, 1056 unsigned long offset, unsigned long size) 1057{ 1058 char *kaddr; 1059 unsigned long left, count = desc->count; 1060 1061 if (size > count) 1062 size = count; 1063 1064 /* 1065 * Faults on the destination of a read are common, so do it before 1066 * taking the kmap. 1067 */ 1068 if (!fault_in_pages_writeable(desc->arg.buf, size)) { 1069 kaddr = kmap_atomic(page, KM_USER0); 1070 left = __copy_to_user_inatomic(desc->arg.buf, 1071 kaddr + offset, size); 1072 kunmap_atomic(kaddr, KM_USER0); 1073 if (left == 0) 1074 goto success; 1075 } 1076 1077 /* Do it the slow way */ 1078 kaddr = kmap(page); 1079 left = __copy_to_user(desc->arg.buf, kaddr + offset, size); 1080 kunmap(page); 1081 1082 if (left) { 1083 size -= left; 1084 desc->error = -EFAULT; 1085 } 1086success: 1087 desc->count = count - size; 1088 desc->written += size; 1089 desc->arg.buf += size; 1090 return size; 1091} 1092 1093/* 1094 * Performs necessary checks before doing a write 1095 * @iov: io vector request 1096 * @nr_segs: number of segments in the iovec 1097 * @count: number of bytes to write 1098 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE 1099 * 1100 * Adjust number of segments and amount of bytes to write (nr_segs should be 1101 * properly initialized first). Returns appropriate error code that caller 1102 * should return or zero in case that write should be allowed. 1103 */ 1104int generic_segment_checks(const struct iovec *iov, 1105 unsigned long *nr_segs, size_t *count, int access_flags) 1106{ 1107 unsigned long seg; 1108 size_t cnt = 0; 1109 for (seg = 0; seg < *nr_segs; seg++) { 1110 const struct iovec *iv = &iov[seg]; 1111 1112 /* 1113 * If any segment has a negative length, or the cumulative 1114 * length ever wraps negative then return -EINVAL. 1115 */ 1116 cnt += iv->iov_len; 1117 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0)) 1118 return -EINVAL; 1119 if (access_ok(access_flags, iv->iov_base, iv->iov_len)) 1120 continue; 1121 if (seg == 0) 1122 return -EFAULT; 1123 *nr_segs = seg; 1124 cnt -= iv->iov_len; /* This segment is no good */ 1125 break; 1126 } 1127 *count = cnt; 1128 return 0; 1129} 1130EXPORT_SYMBOL(generic_segment_checks); 1131 1132/** 1133 * generic_file_aio_read - generic filesystem read routine 1134 * @iocb: kernel I/O control block 1135 * @iov: io vector request 1136 * @nr_segs: number of segments in the iovec 1137 * @pos: current file position 1138 * 1139 * This is the "read()" routine for all filesystems 1140 * that can use the page cache directly. 1141 */ 1142ssize_t 1143generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov, 1144 unsigned long nr_segs, loff_t pos) 1145{ 1146 struct file *filp = iocb->ki_filp; 1147 ssize_t retval; 1148 unsigned long seg; 1149 size_t count; 1150 loff_t *ppos = &iocb->ki_pos; 1151 1152 count = 0; 1153 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); 1154 if (retval) 1155 return retval; 1156 1157 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ 1158 if (filp->f_flags & O_DIRECT) { 1159 loff_t size; 1160 struct address_space *mapping; 1161 struct inode *inode; 1162 1163 mapping = filp->f_mapping; 1164 inode = mapping->host; 1165 retval = 0; 1166 if (!count) 1167 goto out; /* skip atime */ 1168 size = i_size_read(inode); 1169 if (pos < size) { 1170 retval = generic_file_direct_IO(READ, iocb, 1171 iov, pos, nr_segs); 1172 if (retval > 0) 1173 *ppos = pos + retval; 1174 } 1175 if (likely(retval != 0)) { 1176 file_accessed(filp); 1177 goto out; 1178 } 1179 } 1180 1181 retval = 0; 1182 if (count) { 1183 for (seg = 0; seg < nr_segs; seg++) { 1184 read_descriptor_t desc; 1185 1186 desc.written = 0; 1187 desc.arg.buf = iov[seg].iov_base; 1188 desc.count = iov[seg].iov_len; 1189 if (desc.count == 0) 1190 continue; 1191 desc.error = 0; 1192 do_generic_file_read(filp,ppos,&desc,file_read_actor); 1193 retval += desc.written; 1194 if (desc.error) { 1195 retval = retval ?: desc.error; 1196 break; 1197 } 1198 if (desc.count > 0) 1199 break; 1200 } 1201 } 1202out: 1203 return retval; 1204} 1205EXPORT_SYMBOL(generic_file_aio_read); 1206 1207static ssize_t 1208do_readahead(struct address_space *mapping, struct file *filp, 1209 pgoff_t index, unsigned long nr) 1210{ 1211 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage) 1212 return -EINVAL; 1213 1214 force_page_cache_readahead(mapping, filp, index, 1215 max_sane_readahead(nr)); 1216 return 0; 1217} 1218 1219asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count) 1220{ 1221 ssize_t ret; 1222 struct file *file; 1223 1224 ret = -EBADF; 1225 file = fget(fd); 1226 if (file) { 1227 if (file->f_mode & FMODE_READ) { 1228 struct address_space *mapping = file->f_mapping; 1229 pgoff_t start = offset >> PAGE_CACHE_SHIFT; 1230 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT; 1231 unsigned long len = end - start + 1; 1232 ret = do_readahead(mapping, file, start, len); 1233 } 1234 fput(file); 1235 } 1236 return ret; 1237} 1238 1239#ifdef CONFIG_MMU 1240/** 1241 * page_cache_read - adds requested page to the page cache if not already there 1242 * @file: file to read 1243 * @offset: page index 1244 * 1245 * This adds the requested page to the page cache if it isn't already there, 1246 * and schedules an I/O to read in its contents from disk. 1247 */ 1248static int fastcall page_cache_read(struct file * file, pgoff_t offset) 1249{ 1250 struct address_space *mapping = file->f_mapping; 1251 struct page *page; 1252 int ret; 1253 1254 do { 1255 page = page_cache_alloc_cold(mapping); 1256 if (!page) 1257 return -ENOMEM; 1258 1259 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL); 1260 if (ret == 0) 1261 ret = mapping->a_ops->readpage(file, page); 1262 else if (ret == -EEXIST) 1263 ret = 0; /* losing race to add is OK */ 1264 1265 page_cache_release(page); 1266 1267 } while (ret == AOP_TRUNCATED_PAGE); 1268 1269 return ret; 1270} 1271 1272#define MMAP_LOTSAMISS (100) 1273 1274/** 1275 * filemap_fault - read in file data for page fault handling 1276 * @vma: vma in which the fault was taken 1277 * @vmf: struct vm_fault containing details of the fault 1278 * 1279 * filemap_fault() is invoked via the vma operations vector for a 1280 * mapped memory region to read in file data during a page fault. 1281 * 1282 * The goto's are kind of ugly, but this streamlines the normal case of having 1283 * it in the page cache, and handles the special cases reasonably without 1284 * having a lot of duplicated code. 1285 */ 1286int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1287{ 1288 int error; 1289 struct file *file = vma->vm_file; 1290 struct address_space *mapping = file->f_mapping; 1291 struct file_ra_state *ra = &file->f_ra; 1292 struct inode *inode = mapping->host; 1293 struct page *page; 1294 unsigned long size; 1295 int did_readaround = 0; 1296 int ret = 0; 1297 1298 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 1299 if (vmf->pgoff >= size) 1300 goto outside_data_content; 1301 1302 /* If we don't want any read-ahead, don't bother */ 1303 if (VM_RandomReadHint(vma)) 1304 goto no_cached_page; 1305 1306 /* 1307 * Do we have something in the page cache already? 1308 */ 1309retry_find: 1310 page = find_lock_page(mapping, vmf->pgoff); 1311 /* 1312 * For sequential accesses, we use the generic readahead logic. 1313 */ 1314 if (VM_SequentialReadHint(vma)) { 1315 if (!page) { 1316 page_cache_sync_readahead(mapping, ra, file, 1317 vmf->pgoff, 1); 1318 page = find_lock_page(mapping, vmf->pgoff); 1319 if (!page) 1320 goto no_cached_page; 1321 } 1322 if (PageReadahead(page)) { 1323 page_cache_async_readahead(mapping, ra, file, page, 1324 vmf->pgoff, 1); 1325 } 1326 } 1327 1328 if (!page) { 1329 unsigned long ra_pages; 1330 1331 ra->mmap_miss++; 1332 1333 /* 1334 * Do we miss much more than hit in this file? If so, 1335 * stop bothering with read-ahead. It will only hurt. 1336 */ 1337 if (ra->mmap_miss > MMAP_LOTSAMISS) 1338 goto no_cached_page; 1339 1340 /* 1341 * To keep the pgmajfault counter straight, we need to 1342 * check did_readaround, as this is an inner loop. 1343 */ 1344 if (!did_readaround) { 1345 ret = VM_FAULT_MAJOR; 1346 count_vm_event(PGMAJFAULT); 1347 } 1348 did_readaround = 1; 1349 ra_pages = max_sane_readahead(file->f_ra.ra_pages); 1350 if (ra_pages) { 1351 pgoff_t start = 0; 1352 1353 if (vmf->pgoff > ra_pages / 2) 1354 start = vmf->pgoff - ra_pages / 2; 1355 do_page_cache_readahead(mapping, file, start, ra_pages); 1356 } 1357 page = find_lock_page(mapping, vmf->pgoff); 1358 if (!page) 1359 goto no_cached_page; 1360 } 1361 1362 if (!did_readaround) 1363 ra->mmap_miss--; 1364 1365 /* 1366 * We have a locked page in the page cache, now we need to check 1367 * that it's up-to-date. If not, it is going to be due to an error. 1368 */ 1369 if (unlikely(!PageUptodate(page))) 1370 goto page_not_uptodate; 1371 1372 /* Must recheck i_size under page lock */ 1373 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 1374 if (unlikely(vmf->pgoff >= size)) { 1375 unlock_page(page); 1376 page_cache_release(page); 1377 goto outside_data_content; 1378 } 1379 1380 /* 1381 * Found the page and have a reference on it. 1382 */ 1383 mark_page_accessed(page); 1384 ra->prev_pos = (loff_t)page->index << PAGE_CACHE_SHIFT; 1385 vmf->page = page; 1386 return ret | VM_FAULT_LOCKED; 1387 1388outside_data_content: 1389 /* 1390 * An external ptracer can access pages that normally aren't 1391 * accessible.. 1392 */ 1393 if (vma->vm_mm == current->mm) 1394 return VM_FAULT_SIGBUS; 1395 1396 /* Fall through to the non-read-ahead case */ 1397no_cached_page: 1398 /* 1399 * We're only likely to ever get here if MADV_RANDOM is in 1400 * effect. 1401 */ 1402 error = page_cache_read(file, vmf->pgoff); 1403 1404 /* 1405 * The page we want has now been added to the page cache. 1406 * In the unlikely event that someone removed it in the 1407 * meantime, we'll just come back here and read it again. 1408 */ 1409 if (error >= 0) 1410 goto retry_find; 1411 1412 /* 1413 * An error return from page_cache_read can result if the 1414 * system is low on memory, or a problem occurs while trying 1415 * to schedule I/O. 1416 */ 1417 if (error == -ENOMEM) 1418 return VM_FAULT_OOM; 1419 return VM_FAULT_SIGBUS; 1420 1421page_not_uptodate: 1422 /* IO error path */ 1423 if (!did_readaround) { 1424 ret = VM_FAULT_MAJOR; 1425 count_vm_event(PGMAJFAULT); 1426 } 1427 1428 /* 1429 * Umm, take care of errors if the page isn't up-to-date. 1430 * Try to re-read it _once_. We do this synchronously, 1431 * because there really aren't any performance issues here 1432 * and we need to check for errors. 1433 */ 1434 ClearPageError(page); 1435 error = mapping->a_ops->readpage(file, page); 1436 page_cache_release(page); 1437 1438 if (!error || error == AOP_TRUNCATED_PAGE) 1439 goto retry_find; 1440 1441 /* Things didn't work out. Return zero to tell the mm layer so. */ 1442 shrink_readahead_size_eio(file, ra); 1443 return VM_FAULT_SIGBUS; 1444} 1445EXPORT_SYMBOL(filemap_fault); 1446 1447struct vm_operations_struct generic_file_vm_ops = { 1448 .fault = filemap_fault, 1449}; 1450 1451/* This is used for a general mmap of a disk file */ 1452 1453int generic_file_mmap(struct file * file, struct vm_area_struct * vma) 1454{ 1455 struct address_space *mapping = file->f_mapping; 1456 1457 if (!mapping->a_ops->readpage) 1458 return -ENOEXEC; 1459 file_accessed(file); 1460 vma->vm_ops = &generic_file_vm_ops; 1461 vma->vm_flags |= VM_CAN_NONLINEAR; 1462 return 0; 1463} 1464 1465/* 1466 * This is for filesystems which do not implement ->writepage. 1467 */ 1468int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) 1469{ 1470 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) 1471 return -EINVAL; 1472 return generic_file_mmap(file, vma); 1473} 1474#else 1475int generic_file_mmap(struct file * file, struct vm_area_struct * vma) 1476{ 1477 return -ENOSYS; 1478} 1479int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) 1480{ 1481 return -ENOSYS; 1482} 1483#endif /* CONFIG_MMU */ 1484 1485EXPORT_SYMBOL(generic_file_mmap); 1486EXPORT_SYMBOL(generic_file_readonly_mmap); 1487 1488static struct page *__read_cache_page(struct address_space *mapping, 1489 pgoff_t index, 1490 int (*filler)(void *,struct page*), 1491 void *data) 1492{ 1493 struct page *page; 1494 int err; 1495repeat: 1496 page = find_get_page(mapping, index); 1497 if (!page) { 1498 page = page_cache_alloc_cold(mapping); 1499 if (!page) 1500 return ERR_PTR(-ENOMEM); 1501 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL); 1502 if (unlikely(err)) { 1503 page_cache_release(page); 1504 if (err == -EEXIST) 1505 goto repeat; 1506 /* Presumably ENOMEM for radix tree node */ 1507 return ERR_PTR(err); 1508 } 1509 err = filler(data, page); 1510 if (err < 0) { 1511 page_cache_release(page); 1512 page = ERR_PTR(err); 1513 } 1514 } 1515 return page; 1516} 1517 1518/* 1519 * Same as read_cache_page, but don't wait for page to become unlocked 1520 * after submitting it to the filler. 1521 */ 1522struct page *read_cache_page_async(struct address_space *mapping, 1523 pgoff_t index, 1524 int (*filler)(void *,struct page*), 1525 void *data) 1526{ 1527 struct page *page; 1528 int err; 1529 1530retry: 1531 page = __read_cache_page(mapping, index, filler, data); 1532 if (IS_ERR(page)) 1533 return page; 1534 if (PageUptodate(page)) 1535 goto out; 1536 1537 lock_page(page); 1538 if (!page->mapping) { 1539 unlock_page(page); 1540 page_cache_release(page); 1541 goto retry; 1542 } 1543 if (PageUptodate(page)) { 1544 unlock_page(page); 1545 goto out; 1546 } 1547 err = filler(data, page); 1548 if (err < 0) { 1549 page_cache_release(page); 1550 return ERR_PTR(err); 1551 } 1552out: 1553 mark_page_accessed(page); 1554 return page; 1555} 1556EXPORT_SYMBOL(read_cache_page_async); 1557 1558/** 1559 * read_cache_page - read into page cache, fill it if needed 1560 * @mapping: the page's address_space 1561 * @index: the page index 1562 * @filler: function to perform the read 1563 * @data: destination for read data 1564 * 1565 * Read into the page cache. If a page already exists, and PageUptodate() is 1566 * not set, try to fill the page then wait for it to become unlocked. 1567 * 1568 * If the page does not get brought uptodate, return -EIO. 1569 */ 1570struct page *read_cache_page(struct address_space *mapping, 1571 pgoff_t index, 1572 int (*filler)(void *,struct page*), 1573 void *data) 1574{ 1575 struct page *page; 1576 1577 page = read_cache_page_async(mapping, index, filler, data); 1578 if (IS_ERR(page)) 1579 goto out; 1580 wait_on_page_locked(page); 1581 if (!PageUptodate(page)) { 1582 page_cache_release(page); 1583 page = ERR_PTR(-EIO); 1584 } 1585 out: 1586 return page; 1587} 1588EXPORT_SYMBOL(read_cache_page); 1589 1590/* 1591 * The logic we want is 1592 * 1593 * if suid or (sgid and xgrp) 1594 * remove privs 1595 */ 1596int should_remove_suid(struct dentry *dentry) 1597{ 1598 mode_t mode = dentry->d_inode->i_mode; 1599 int kill = 0; 1600 1601 /* suid always must be killed */ 1602 if (unlikely(mode & S_ISUID)) 1603 kill = ATTR_KILL_SUID; 1604 1605 /* 1606 * sgid without any exec bits is just a mandatory locking mark; leave 1607 * it alone. If some exec bits are set, it's a real sgid; kill it. 1608 */ 1609 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 1610 kill |= ATTR_KILL_SGID; 1611 1612 if (unlikely(kill && !capable(CAP_FSETID))) 1613 return kill; 1614 1615 return 0; 1616} 1617EXPORT_SYMBOL(should_remove_suid); 1618 1619int __remove_suid(struct dentry *dentry, int kill) 1620{ 1621 struct iattr newattrs; 1622 1623 newattrs.ia_valid = ATTR_FORCE | kill; 1624 return notify_change(dentry, &newattrs); 1625} 1626 1627int remove_suid(struct dentry *dentry) 1628{ 1629 int kill = should_remove_suid(dentry); 1630 1631 if (unlikely(kill)) 1632 return __remove_suid(dentry, kill); 1633 1634 return 0; 1635} 1636EXPORT_SYMBOL(remove_suid); 1637 1638size_t 1639__filemap_copy_from_user_iovec_inatomic(char *vaddr, 1640 const struct iovec *iov, size_t base, size_t bytes) 1641{ 1642 size_t copied = 0, left = 0; 1643 1644 while (bytes) { 1645 char __user *buf = iov->iov_base + base; 1646 int copy = min(bytes, iov->iov_len - base); 1647 1648 base = 0; 1649 left = __copy_from_user_inatomic_nocache(vaddr, buf, copy); 1650 copied += copy; 1651 bytes -= copy; 1652 vaddr += copy; 1653 iov++; 1654 1655 if (unlikely(left)) 1656 break; 1657 } 1658 return copied - left; 1659} 1660 1661/* 1662 * Performs necessary checks before doing a write 1663 * 1664 * Can adjust writing position or amount of bytes to write. 1665 * Returns appropriate error code that caller should return or 1666 * zero in case that write should be allowed. 1667 */ 1668inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk) 1669{ 1670 struct inode *inode = file->f_mapping->host; 1671 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; 1672 1673 if (unlikely(*pos < 0)) 1674 return -EINVAL; 1675 1676 if (!isblk) { 1677 /* FIXME: this is for backwards compatibility with 2.4 */ 1678 if (file->f_flags & O_APPEND) 1679 *pos = i_size_read(inode); 1680 1681 if (limit != RLIM_INFINITY) { 1682 if (*pos >= limit) { 1683 send_sig(SIGXFSZ, current, 0); 1684 return -EFBIG; 1685 } 1686 if (*count > limit - (typeof(limit))*pos) { 1687 *count = limit - (typeof(limit))*pos; 1688 } 1689 } 1690 } 1691 1692 /* 1693 * LFS rule 1694 */ 1695 if (unlikely(*pos + *count > MAX_NON_LFS && 1696 !(file->f_flags & O_LARGEFILE))) { 1697 if (*pos >= MAX_NON_LFS) { 1698 return -EFBIG; 1699 } 1700 if (*count > MAX_NON_LFS - (unsigned long)*pos) { 1701 *count = MAX_NON_LFS - (unsigned long)*pos; 1702 } 1703 } 1704 1705 /* 1706 * Are we about to exceed the fs block limit ? 1707 * 1708 * If we have written data it becomes a short write. If we have 1709 * exceeded without writing data we send a signal and return EFBIG. 1710 * Linus frestrict idea will clean these up nicely.. 1711 */ 1712 if (likely(!isblk)) { 1713 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) { 1714 if (*count || *pos > inode->i_sb->s_maxbytes) { 1715 return -EFBIG; 1716 } 1717 /* zero-length writes at ->s_maxbytes are OK */ 1718 } 1719 1720 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes)) 1721 *count = inode->i_sb->s_maxbytes - *pos; 1722 } else { 1723#ifdef CONFIG_BLOCK 1724 loff_t isize; 1725 if (bdev_read_only(I_BDEV(inode))) 1726 return -EPERM; 1727 isize = i_size_read(inode); 1728 if (*pos >= isize) { 1729 if (*count || *pos > isize) 1730 return -ENOSPC; 1731 } 1732 1733 if (*pos + *count > isize) 1734 *count = isize - *pos; 1735#else 1736 return -EPERM; 1737#endif 1738 } 1739 return 0; 1740} 1741EXPORT_SYMBOL(generic_write_checks); 1742 1743ssize_t 1744generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov, 1745 unsigned long *nr_segs, loff_t pos, loff_t *ppos, 1746 size_t count, size_t ocount) 1747{ 1748 struct file *file = iocb->ki_filp; 1749 struct address_space *mapping = file->f_mapping; 1750 struct inode *inode = mapping->host; 1751 ssize_t written; 1752 1753 if (count != ocount) 1754 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count); 1755 1756 written = generic_file_direct_IO(WRITE, iocb, iov, pos, *nr_segs); 1757 if (written > 0) { 1758 loff_t end = pos + written; 1759 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { 1760 i_size_write(inode, end); 1761 mark_inode_dirty(inode); 1762 } 1763 *ppos = end; 1764 } 1765 1766 /* 1767 * Sync the fs metadata but not the minor inode changes and 1768 * of course not the data as we did direct DMA for the IO. 1769 * i_mutex is held, which protects generic_osync_inode() from 1770 * livelocking. AIO O_DIRECT ops attempt to sync metadata here. 1771 */ 1772 if ((written >= 0 || written == -EIOCBQUEUED) && 1773 ((file->f_flags & O_SYNC) || IS_SYNC(inode))) { 1774 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA); 1775 if (err < 0) 1776 written = err; 1777 } 1778 return written; 1779} 1780EXPORT_SYMBOL(generic_file_direct_write); 1781 1782/* 1783 * Find or create a page at the given pagecache position. Return the locked 1784 * page. This function is specifically for buffered writes. 1785 */ 1786static struct page *__grab_cache_page(struct address_space *mapping, 1787 pgoff_t index) 1788{ 1789 int status; 1790 struct page *page; 1791repeat: 1792 page = find_lock_page(mapping, index); 1793 if (likely(page)) 1794 return page; 1795 1796 page = page_cache_alloc(mapping); 1797 if (!page) 1798 return NULL; 1799 status = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL); 1800 if (unlikely(status)) { 1801 page_cache_release(page); 1802 if (status == -EEXIST) 1803 goto repeat; 1804 return NULL; 1805 } 1806 return page; 1807} 1808 1809ssize_t 1810generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov, 1811 unsigned long nr_segs, loff_t pos, loff_t *ppos, 1812 size_t count, ssize_t written) 1813{ 1814 struct file *file = iocb->ki_filp; 1815 struct address_space *mapping = file->f_mapping; 1816 const struct address_space_operations *a_ops = mapping->a_ops; 1817 struct inode *inode = mapping->host; 1818 long status = 0; 1819 const struct iovec *cur_iov = iov; /* current iovec */ 1820 size_t iov_offset = 0; /* offset in the current iovec */ 1821 char __user *buf; 1822 1823 /* 1824 * handle partial DIO write. Adjust cur_iov if needed. 1825 */ 1826 filemap_set_next_iovec(&cur_iov, nr_segs, &iov_offset, written); 1827 1828 do { 1829 struct page *src_page; 1830 struct page *page; 1831 pgoff_t index; /* Pagecache index for current page */ 1832 unsigned long offset; /* Offset into pagecache page */ 1833 unsigned long seglen; /* Bytes remaining in current iovec */ 1834 unsigned long bytes; /* Bytes to write to page */ 1835 size_t copied; /* Bytes copied from user */ 1836 1837 buf = cur_iov->iov_base + iov_offset; 1838 offset = (pos & (PAGE_CACHE_SIZE - 1)); 1839 index = pos >> PAGE_CACHE_SHIFT; 1840 bytes = PAGE_CACHE_SIZE - offset; 1841 if (bytes > count) 1842 bytes = count; 1843 1844 /* 1845 * a non-NULL src_page indicates that we're doing the 1846 * copy via get_user_pages and kmap. 1847 */ 1848 src_page = NULL; 1849 1850 seglen = cur_iov->iov_len - iov_offset; 1851 if (seglen > bytes) 1852 seglen = bytes; 1853 1854 /* 1855 * Bring in the user page that we will copy from _first_. 1856 * Otherwise there's a nasty deadlock on copying from the 1857 * same page as we're writing to, without it being marked 1858 * up-to-date. 1859 * 1860 * Not only is this an optimisation, but it is also required 1861 * to check that the address is actually valid, when atomic 1862 * usercopies are used, below. 1863 */ 1864 if (unlikely(fault_in_pages_readable(buf, seglen))) { 1865 status = -EFAULT; 1866 break; 1867 } 1868 1869 page = __grab_cache_page(mapping, index); 1870 if (!page) { 1871 status = -ENOMEM; 1872 break; 1873 } 1874 1875 /* 1876 * non-uptodate pages cannot cope with short copies, and we 1877 * cannot take a pagefault with the destination page locked. 1878 * So pin the source page to copy it. 1879 */ 1880 if (!PageUptodate(page)) { 1881 unlock_page(page); 1882 1883 src_page = alloc_page(GFP_KERNEL); 1884 if (!src_page) { 1885 page_cache_release(page); 1886 status = -ENOMEM; 1887 break; 1888 } 1889 1890 /* 1891 * Cannot get_user_pages with a page locked for the 1892 * same reason as we can't take a page fault with a 1893 * page locked (as explained below). 1894 */ 1895 copied = filemap_copy_from_user(src_page, offset, 1896 cur_iov, nr_segs, iov_offset, bytes); 1897 if (unlikely(copied == 0)) { 1898 status = -EFAULT; 1899 page_cache_release(page); 1900 page_cache_release(src_page); 1901 break; 1902 } 1903 bytes = copied; 1904 1905 lock_page(page); 1906 /* 1907 * Can't handle the page going uptodate here, because 1908 * that means we would use non-atomic usercopies, which 1909 * zero out the tail of the page, which can cause 1910 * zeroes to become transiently visible. We could just 1911 * use a non-zeroing copy, but the APIs aren't too 1912 * consistent. 1913 */ 1914 if (unlikely(!page->mapping || PageUptodate(page))) { 1915 unlock_page(page); 1916 page_cache_release(page); 1917 page_cache_release(src_page); 1918 continue; 1919 } 1920 1921 } 1922 1923 status = a_ops->prepare_write(file, page, offset, offset+bytes); 1924 if (unlikely(status)) 1925 goto fs_write_aop_error; 1926 1927 if (!src_page) { 1928 /* 1929 * Must not enter the pagefault handler here, because 1930 * we hold the page lock, so we might recursively 1931 * deadlock on the same lock, or get an ABBA deadlock 1932 * against a different lock, or against the mmap_sem 1933 * (which nests outside the page lock). So increment 1934 * preempt count, and use _atomic usercopies. 1935 * 1936 * The page is uptodate so we are OK to encounter a 1937 * short copy: if unmodified parts of the page are 1938 * marked dirty and written out to disk, it doesn't 1939 * really matter. 1940 */ 1941 pagefault_disable(); 1942 copied = filemap_copy_from_user_atomic(page, offset, 1943 cur_iov, nr_segs, iov_offset, bytes); 1944 pagefault_enable(); 1945 } else { 1946 void *src, *dst; 1947 src = kmap_atomic(src_page, KM_USER0); 1948 dst = kmap_atomic(page, KM_USER1); 1949 memcpy(dst + offset, src + offset, bytes); 1950 kunmap_atomic(dst, KM_USER1); 1951 kunmap_atomic(src, KM_USER0); 1952 copied = bytes; 1953 } 1954 flush_dcache_page(page); 1955 1956 status = a_ops->commit_write(file, page, offset, offset+bytes); 1957 if (unlikely(status < 0 || status == AOP_TRUNCATED_PAGE)) 1958 goto fs_write_aop_error; 1959 if (unlikely(status > 0)) /* filesystem did partial write */ 1960 copied = min_t(size_t, copied, status); 1961 1962 unlock_page(page); 1963 mark_page_accessed(page); 1964 page_cache_release(page); 1965 if (src_page) 1966 page_cache_release(src_page); 1967 1968 written += copied; 1969 count -= copied; 1970 pos += copied; 1971 filemap_set_next_iovec(&cur_iov, nr_segs, &iov_offset, copied); 1972 1973 balance_dirty_pages_ratelimited(mapping); 1974 cond_resched(); 1975 continue; 1976 1977fs_write_aop_error: 1978 if (status != AOP_TRUNCATED_PAGE) 1979 unlock_page(page); 1980 page_cache_release(page); 1981 if (src_page) 1982 page_cache_release(src_page); 1983 1984 /* 1985 * prepare_write() may have instantiated a few blocks 1986 * outside i_size. Trim these off again. Don't need 1987 * i_size_read because we hold i_mutex. 1988 */ 1989 if (pos + bytes > inode->i_size) 1990 vmtruncate(inode, inode->i_size); 1991 if (status == AOP_TRUNCATED_PAGE) 1992 continue; 1993 else 1994 break; 1995 } while (count); 1996 *ppos = pos; 1997 1998 /* 1999 * For now, when the user asks for O_SYNC, we'll actually give O_DSYNC 2000 */ 2001 if (likely(status >= 0)) { 2002 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) { 2003 if (!a_ops->writepage || !is_sync_kiocb(iocb)) 2004 status = generic_osync_inode(inode, mapping, 2005 OSYNC_METADATA|OSYNC_DATA); 2006 } 2007 } 2008 2009 /* 2010 * If we get here for O_DIRECT writes then we must have fallen through 2011 * to buffered writes (block instantiation inside i_size). So we sync 2012 * the file data here, to try to honour O_DIRECT expectations. 2013 */ 2014 if (unlikely(file->f_flags & O_DIRECT) && written) 2015 status = filemap_write_and_wait(mapping); 2016 2017 return written ? written : status; 2018} 2019EXPORT_SYMBOL(generic_file_buffered_write); 2020 2021static ssize_t 2022__generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov, 2023 unsigned long nr_segs, loff_t *ppos) 2024{ 2025 struct file *file = iocb->ki_filp; 2026 struct address_space * mapping = file->f_mapping; 2027 size_t ocount; /* original count */ 2028 size_t count; /* after file limit checks */ 2029 struct inode *inode = mapping->host; 2030 loff_t pos; 2031 ssize_t written; 2032 ssize_t err; 2033 2034 ocount = 0; 2035 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); 2036 if (err) 2037 return err; 2038 2039 count = ocount; 2040 pos = *ppos; 2041 2042 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); 2043 2044 /* We can write back this queue in page reclaim */ 2045 current->backing_dev_info = mapping->backing_dev_info; 2046 written = 0; 2047 2048 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); 2049 if (err) 2050 goto out; 2051 2052 if (count == 0) 2053 goto out; 2054 2055 err = remove_suid(file->f_path.dentry); 2056 if (err) 2057 goto out; 2058 2059 file_update_time(file); 2060 2061 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ 2062 if (unlikely(file->f_flags & O_DIRECT)) { 2063 loff_t endbyte; 2064 ssize_t written_buffered; 2065 2066 written = generic_file_direct_write(iocb, iov, &nr_segs, pos, 2067 ppos, count, ocount); 2068 if (written < 0 || written == count) 2069 goto out; 2070 /* 2071 * direct-io write to a hole: fall through to buffered I/O 2072 * for completing the rest of the request. 2073 */ 2074 pos += written; 2075 count -= written; 2076 written_buffered = generic_file_buffered_write(iocb, iov, 2077 nr_segs, pos, ppos, count, 2078 written); 2079 /* 2080 * If generic_file_buffered_write() retuned a synchronous error 2081 * then we want to return the number of bytes which were 2082 * direct-written, or the error code if that was zero. Note 2083 * that this differs from normal direct-io semantics, which 2084 * will return -EFOO even if some bytes were written. 2085 */ 2086 if (written_buffered < 0) { 2087 err = written_buffered; 2088 goto out; 2089 } 2090 2091 /* 2092 * We need to ensure that the page cache pages are written to 2093 * disk and invalidated to preserve the expected O_DIRECT 2094 * semantics. 2095 */ 2096 endbyte = pos + written_buffered - written - 1; 2097 err = do_sync_mapping_range(file->f_mapping, pos, endbyte, 2098 SYNC_FILE_RANGE_WAIT_BEFORE| 2099 SYNC_FILE_RANGE_WRITE| 2100 SYNC_FILE_RANGE_WAIT_AFTER); 2101 if (err == 0) { 2102 written = written_buffered; 2103 invalidate_mapping_pages(mapping, 2104 pos >> PAGE_CACHE_SHIFT, 2105 endbyte >> PAGE_CACHE_SHIFT); 2106 } else { 2107 /* 2108 * We don't know how much we wrote, so just return 2109 * the number of bytes which were direct-written 2110 */ 2111 } 2112 } else { 2113 written = generic_file_buffered_write(iocb, iov, nr_segs, 2114 pos, ppos, count, written); 2115 } 2116out: 2117 current->backing_dev_info = NULL; 2118 return written ? written : err; 2119} 2120 2121ssize_t generic_file_aio_write_nolock(struct kiocb *iocb, 2122 const struct iovec *iov, unsigned long nr_segs, loff_t pos) 2123{ 2124 struct file *file = iocb->ki_filp; 2125 struct address_space *mapping = file->f_mapping; 2126 struct inode *inode = mapping->host; 2127 ssize_t ret; 2128 2129 BUG_ON(iocb->ki_pos != pos); 2130 2131 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs, 2132 &iocb->ki_pos); 2133 2134 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) { 2135 ssize_t err; 2136 2137 err = sync_page_range_nolock(inode, mapping, pos, ret); 2138 if (err < 0) 2139 ret = err; 2140 } 2141 return ret; 2142} 2143EXPORT_SYMBOL(generic_file_aio_write_nolock); 2144 2145ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, 2146 unsigned long nr_segs, loff_t pos) 2147{ 2148 struct file *file = iocb->ki_filp; 2149 struct address_space *mapping = file->f_mapping; 2150 struct inode *inode = mapping->host; 2151 ssize_t ret; 2152 2153 BUG_ON(iocb->ki_pos != pos); 2154 2155 mutex_lock(&inode->i_mutex); 2156 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs, 2157 &iocb->ki_pos); 2158 mutex_unlock(&inode->i_mutex); 2159 2160 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) { 2161 ssize_t err; 2162 2163 err = sync_page_range(inode, mapping, pos, ret); 2164 if (err < 0) 2165 ret = err; 2166 } 2167 return ret; 2168} 2169EXPORT_SYMBOL(generic_file_aio_write); 2170 2171/* 2172 * Called under i_mutex for writes to S_ISREG files. Returns -EIO if something 2173 * went wrong during pagecache shootdown. 2174 */ 2175static ssize_t 2176generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, 2177 loff_t offset, unsigned long nr_segs) 2178{ 2179 struct file *file = iocb->ki_filp; 2180 struct address_space *mapping = file->f_mapping; 2181 ssize_t retval; 2182 size_t write_len; 2183 pgoff_t end = 0; /* silence gcc */ 2184 2185 /* 2186 * If it's a write, unmap all mmappings of the file up-front. This 2187 * will cause any pte dirty bits to be propagated into the pageframes 2188 * for the subsequent filemap_write_and_wait(). 2189 */ 2190 if (rw == WRITE) { 2191 write_len = iov_length(iov, nr_segs); 2192 end = (offset + write_len - 1) >> PAGE_CACHE_SHIFT; 2193 if (mapping_mapped(mapping)) 2194 unmap_mapping_range(mapping, offset, write_len, 0); 2195 } 2196 2197 retval = filemap_write_and_wait(mapping); 2198 if (retval) 2199 goto out; 2200 2201 /* 2202 * After a write we want buffered reads to be sure to go to disk to get 2203 * the new data. We invalidate clean cached page from the region we're 2204 * about to write. We do this *before* the write so that we can return 2205 * -EIO without clobbering -EIOCBQUEUED from ->direct_IO(). 2206 */ 2207 if (rw == WRITE && mapping->nrpages) { 2208 retval = invalidate_inode_pages2_range(mapping, 2209 offset >> PAGE_CACHE_SHIFT, end); 2210 if (retval) 2211 goto out; 2212 } 2213 2214 retval = mapping->a_ops->direct_IO(rw, iocb, iov, offset, nr_segs); 2215 if (retval) 2216 goto out; 2217 2218 /* 2219 * Finally, try again to invalidate clean pages which might have been 2220 * faulted in by get_user_pages() if the source of the write was an 2221 * mmap()ed region of the file we're writing. That's a pretty crazy 2222 * thing to do, so we don't support it 100%. If this invalidation 2223 * fails and we have -EIOCBQUEUED we ignore the failure. 2224 */ 2225 if (rw == WRITE && mapping->nrpages) { 2226 int err = invalidate_inode_pages2_range(mapping, 2227 offset >> PAGE_CACHE_SHIFT, end); 2228 if (err && retval >= 0) 2229 retval = err; 2230 } 2231out: 2232 return retval; 2233} 2234 2235/** 2236 * try_to_release_page() - release old fs-specific metadata on a page 2237 * 2238 * @page: the page which the kernel is trying to free 2239 * @gfp_mask: memory allocation flags (and I/O mode) 2240 * 2241 * The address_space is to try to release any data against the page 2242 * (presumably at page->private). If the release was successful, return `1'. 2243 * Otherwise return zero. 2244 * 2245 * The @gfp_mask argument specifies whether I/O may be performed to release 2246 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT). 2247 * 2248 * NOTE: @gfp_mask may go away, and this function may become non-blocking. 2249 */ 2250int try_to_release_page(struct page *page, gfp_t gfp_mask) 2251{ 2252 struct address_space * const mapping = page->mapping; 2253 2254 BUG_ON(!PageLocked(page)); 2255 if (PageWriteback(page)) 2256 return 0; 2257 2258 if (mapping && mapping->a_ops->releasepage) 2259 return mapping->a_ops->releasepage(page, gfp_mask); 2260 return try_to_free_buffers(page); 2261} 2262 2263EXPORT_SYMBOL(try_to_release_page); 2264