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#include <linux/buffer_head.h> /* for generic_osync_inode */ 37 38#include <asm/mman.h> 39 40static ssize_t 41generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, 42 loff_t offset, unsigned long nr_segs); 43 44/* 45 * Shared mappings implemented 30.11.1994. It's not fully working yet, 46 * though. 47 * 48 * Shared mappings now work. 15.8.1995 Bruno. 49 * 50 * finished 'unifying' the page and buffer cache and SMP-threaded the 51 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com> 52 * 53 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de> 54 */ 55 56/* 57 * Lock ordering: 58 * 59 * ->i_mmap_lock (vmtruncate) 60 * ->private_lock (__free_pte->__set_page_dirty_buffers) 61 * ->swap_lock (exclusive_swap_page, others) 62 * ->mapping->tree_lock 63 * 64 * ->i_mutex 65 * ->i_mmap_lock (truncate->unmap_mapping_range) 66 * 67 * ->mmap_sem 68 * ->i_mmap_lock 69 * ->page_table_lock or pte_lock (various, mainly in memory.c) 70 * ->mapping->tree_lock (arch-dependent flush_dcache_mmap_lock) 71 * 72 * ->mmap_sem 73 * ->lock_page (access_process_vm) 74 * 75 * ->i_mutex (generic_file_buffered_write) 76 * ->mmap_sem (fault_in_pages_readable->do_page_fault) 77 * 78 * ->i_mutex 79 * ->i_alloc_sem (various) 80 * 81 * ->inode_lock 82 * ->sb_lock (fs/fs-writeback.c) 83 * ->mapping->tree_lock (__sync_single_inode) 84 * 85 * ->i_mmap_lock 86 * ->anon_vma.lock (vma_adjust) 87 * 88 * ->anon_vma.lock 89 * ->page_table_lock or pte_lock (anon_vma_prepare and various) 90 * 91 * ->page_table_lock or pte_lock 92 * ->swap_lock (try_to_unmap_one) 93 * ->private_lock (try_to_unmap_one) 94 * ->tree_lock (try_to_unmap_one) 95 * ->zone.lru_lock (follow_page->mark_page_accessed) 96 * ->zone.lru_lock (check_pte_range->isolate_lru_page) 97 * ->private_lock (page_remove_rmap->set_page_dirty) 98 * ->tree_lock (page_remove_rmap->set_page_dirty) 99 * ->inode_lock (page_remove_rmap->set_page_dirty) 100 * ->inode_lock (zap_pte_range->set_page_dirty) 101 * ->private_lock (zap_pte_range->__set_page_dirty_buffers) 102 * 103 * ->task->proc_lock 104 * ->dcache_lock (proc_pid_lookup) 105 */ 106 107/* 108 * Remove a page from the page cache and free it. Caller has to make 109 * sure the page is locked and that nobody else uses it - or that usage 110 * is safe. The caller must hold a write_lock on the mapping's tree_lock. 111 */ 112void __remove_from_page_cache(struct page *page) 113{ 114 struct address_space *mapping = page->mapping; 115 116 radix_tree_delete(&mapping->page_tree, page->index); 117 page->mapping = NULL; 118 mapping->nrpages--; 119 __dec_zone_page_state(page, NR_FILE_PAGES); 120} 121 122void remove_from_page_cache(struct page *page) 123{ 124 struct address_space *mapping = page->mapping; 125 126 BUG_ON(!PageLocked(page)); 127 128 write_lock_irq(&mapping->tree_lock); 129 __remove_from_page_cache(page); 130 write_unlock_irq(&mapping->tree_lock); 131} 132 133static int sync_page(void *word) 134{ 135 struct address_space *mapping; 136 struct page *page; 137 138 page = container_of((unsigned long *)word, struct page, flags); 139 140 /* 141 * page_mapping() is being called without PG_locked held. 142 * Some knowledge of the state and use of the page is used to 143 * reduce the requirements down to a memory barrier. 144 * The danger here is of a stale page_mapping() return value 145 * indicating a struct address_space different from the one it's 146 * associated with when it is associated with one. 147 * After smp_mb(), it's either the correct page_mapping() for 148 * the page, or an old page_mapping() and the page's own 149 * page_mapping() has gone NULL. 150 * The ->sync_page() address_space operation must tolerate 151 * page_mapping() going NULL. By an amazing coincidence, 152 * this comes about because none of the users of the page 153 * in the ->sync_page() methods make essential use of the 154 * page_mapping(), merely passing the page down to the backing 155 * device's unplug functions when it's non-NULL, which in turn 156 * ignore it for all cases but swap, where only page_private(page) is 157 * of interest. When page_mapping() does go NULL, the entire 158 * call stack gracefully ignores the page and returns. 159 * -- wli 160 */ 161 smp_mb(); 162 mapping = page_mapping(page); 163 if (mapping && mapping->a_ops && mapping->a_ops->sync_page) 164 mapping->a_ops->sync_page(page); 165 io_schedule(); 166 return 0; 167} 168 169/** 170 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range 171 * @mapping: address space structure to write 172 * @start: offset in bytes where the range starts 173 * @end: offset in bytes where the range ends (inclusive) 174 * @sync_mode: enable synchronous operation 175 * 176 * Start writeback against all of a mapping's dirty pages that lie 177 * within the byte offsets <start, end> inclusive. 178 * 179 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as 180 * opposed to a regular memory cleansing writeback. The difference between 181 * these two operations is that if a dirty page/buffer is encountered, it must 182 * be waited upon, and not just skipped over. 183 */ 184int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, 185 loff_t end, int sync_mode) 186{ 187 int ret; 188 struct writeback_control wbc = { 189 .sync_mode = sync_mode, 190 .nr_to_write = mapping->nrpages * 2, 191 .range_start = start, 192 .range_end = end, 193 }; 194 195 if (!mapping_cap_writeback_dirty(mapping)) 196 return 0; 197 198 ret = do_writepages(mapping, &wbc); 199 return ret; 200} 201 202static inline int __filemap_fdatawrite(struct address_space *mapping, 203 int sync_mode) 204{ 205 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); 206} 207 208int filemap_fdatawrite(struct address_space *mapping) 209{ 210 return __filemap_fdatawrite(mapping, WB_SYNC_ALL); 211} 212EXPORT_SYMBOL(filemap_fdatawrite); 213 214static int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, 215 loff_t end) 216{ 217 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); 218} 219 220/** 221 * filemap_flush - mostly a non-blocking flush 222 * @mapping: target address_space 223 * 224 * This is a mostly non-blocking flush. Not suitable for data-integrity 225 * purposes - I/O may not be started against all dirty pages. 226 */ 227int filemap_flush(struct address_space *mapping) 228{ 229 return __filemap_fdatawrite(mapping, WB_SYNC_NONE); 230} 231EXPORT_SYMBOL(filemap_flush); 232 233/** 234 * wait_on_page_writeback_range - wait for writeback to complete 235 * @mapping: target address_space 236 * @start: beginning page index 237 * @end: ending page index 238 * 239 * Wait for writeback to complete against pages indexed by start->end 240 * inclusive 241 */ 242int wait_on_page_writeback_range(struct address_space *mapping, 243 pgoff_t start, pgoff_t end) 244{ 245 struct pagevec pvec; 246 int nr_pages; 247 int ret = 0; 248 pgoff_t index; 249 250 if (end < start) 251 return 0; 252 253 pagevec_init(&pvec, 0); 254 index = start; 255 while ((index <= end) && 256 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, 257 PAGECACHE_TAG_WRITEBACK, 258 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) { 259 unsigned i; 260 261 for (i = 0; i < nr_pages; i++) { 262 struct page *page = pvec.pages[i]; 263 264 /* until radix tree lookup accepts end_index */ 265 if (page->index > end) 266 continue; 267 268 wait_on_page_writeback(page); 269 if (PageError(page)) 270 ret = -EIO; 271 } 272 pagevec_release(&pvec); 273 cond_resched(); 274 } 275 276 /* Check for outstanding write errors */ 277 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags)) 278 ret = -ENOSPC; 279 if (test_and_clear_bit(AS_EIO, &mapping->flags)) 280 ret = -EIO; 281 282 return ret; 283} 284 285/** 286 * sync_page_range - write and wait on all pages in the passed range 287 * @inode: target inode 288 * @mapping: target address_space 289 * @pos: beginning offset in pages to write 290 * @count: number of bytes to write 291 * 292 * Write and wait upon all the pages in the passed range. This is a "data 293 * integrity" operation. It waits upon in-flight writeout before starting and 294 * waiting upon new writeout. If there was an IO error, return it. 295 * 296 * We need to re-take i_mutex during the generic_osync_inode list walk because 297 * it is otherwise livelockable. 298 */ 299int sync_page_range(struct inode *inode, struct address_space *mapping, 300 loff_t pos, loff_t count) 301{ 302 pgoff_t start = pos >> PAGE_CACHE_SHIFT; 303 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT; 304 int ret; 305 306 if (!mapping_cap_writeback_dirty(mapping) || !count) 307 return 0; 308 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1); 309 if (ret == 0) { 310 mutex_lock(&inode->i_mutex); 311 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA); 312 mutex_unlock(&inode->i_mutex); 313 } 314 if (ret == 0) 315 ret = wait_on_page_writeback_range(mapping, start, end); 316 return ret; 317} 318EXPORT_SYMBOL(sync_page_range); 319 320/** 321 * sync_page_range_nolock 322 * @inode: target inode 323 * @mapping: target address_space 324 * @pos: beginning offset in pages to write 325 * @count: number of bytes to write 326 * 327 * Note: Holding i_mutex across sync_page_range_nolock() is not a good idea 328 * as it forces O_SYNC writers to different parts of the same file 329 * to be serialised right until io completion. 330 */ 331int sync_page_range_nolock(struct inode *inode, struct address_space *mapping, 332 loff_t pos, loff_t count) 333{ 334 pgoff_t start = pos >> PAGE_CACHE_SHIFT; 335 pgoff_t end = (pos + count - 1) >> PAGE_CACHE_SHIFT; 336 int ret; 337 338 if (!mapping_cap_writeback_dirty(mapping) || !count) 339 return 0; 340 ret = filemap_fdatawrite_range(mapping, pos, pos + count - 1); 341 if (ret == 0) 342 ret = generic_osync_inode(inode, mapping, OSYNC_METADATA); 343 if (ret == 0) 344 ret = wait_on_page_writeback_range(mapping, start, end); 345 return ret; 346} 347EXPORT_SYMBOL(sync_page_range_nolock); 348 349/** 350 * filemap_fdatawait - wait for all under-writeback pages to complete 351 * @mapping: address space structure to wait for 352 * 353 * Walk the list of under-writeback pages of the given address space 354 * and wait for all of them. 355 */ 356int filemap_fdatawait(struct address_space *mapping) 357{ 358 loff_t i_size = i_size_read(mapping->host); 359 360 if (i_size == 0) 361 return 0; 362 363 return wait_on_page_writeback_range(mapping, 0, 364 (i_size - 1) >> PAGE_CACHE_SHIFT); 365} 366EXPORT_SYMBOL(filemap_fdatawait); 367 368int filemap_write_and_wait(struct address_space *mapping) 369{ 370 int err = 0; 371 372 if (mapping->nrpages) { 373 err = filemap_fdatawrite(mapping); 374 /* 375 * Even if the above returned error, the pages may be 376 * written partially (e.g. -ENOSPC), so we wait for it. 377 * But the -EIO is special case, it may indicate the worst 378 * thing (e.g. bug) happened, so we avoid waiting for it. 379 */ 380 if (err != -EIO) { 381 int err2 = filemap_fdatawait(mapping); 382 if (!err) 383 err = err2; 384 } 385 } 386 return err; 387} 388EXPORT_SYMBOL(filemap_write_and_wait); 389 390/** 391 * filemap_write_and_wait_range - write out & wait on a file range 392 * @mapping: the address_space for the pages 393 * @lstart: offset in bytes where the range starts 394 * @lend: offset in bytes where the range ends (inclusive) 395 * 396 * Write out and wait upon file offsets lstart->lend, inclusive. 397 * 398 * Note that `lend' is inclusive (describes the last byte to be written) so 399 * that this function can be used to write to the very end-of-file (end = -1). 400 */ 401int filemap_write_and_wait_range(struct address_space *mapping, 402 loff_t lstart, loff_t lend) 403{ 404 int err = 0; 405 406 if (mapping->nrpages) { 407 err = __filemap_fdatawrite_range(mapping, lstart, lend, 408 WB_SYNC_ALL); 409 /* See comment of filemap_write_and_wait() */ 410 if (err != -EIO) { 411 int err2 = wait_on_page_writeback_range(mapping, 412 lstart >> PAGE_CACHE_SHIFT, 413 lend >> PAGE_CACHE_SHIFT); 414 if (!err) 415 err = err2; 416 } 417 } 418 return err; 419} 420 421/** 422 * add_to_page_cache - add newly allocated pagecache pages 423 * @page: page to add 424 * @mapping: the page's address_space 425 * @offset: page index 426 * @gfp_mask: page allocation mode 427 * 428 * This function is used to add newly allocated pagecache pages; 429 * the page is new, so we can just run SetPageLocked() against it. 430 * The other page state flags were set by rmqueue(). 431 * 432 * This function does not add the page to the LRU. The caller must do that. 433 */ 434int add_to_page_cache(struct page *page, struct address_space *mapping, 435 pgoff_t offset, gfp_t gfp_mask) 436{ 437 int error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); 438 439 if (error == 0) { 440 write_lock_irq(&mapping->tree_lock); 441 error = radix_tree_insert(&mapping->page_tree, offset, page); 442 if (!error) { 443 page_cache_get(page); 444 SetPageLocked(page); 445 page->mapping = mapping; 446 page->index = offset; 447 mapping->nrpages++; 448 __inc_zone_page_state(page, NR_FILE_PAGES); 449 } 450 write_unlock_irq(&mapping->tree_lock); 451 radix_tree_preload_end(); 452 } 453 return error; 454} 455EXPORT_SYMBOL(add_to_page_cache); 456 457int add_to_page_cache_lru(struct page *page, struct address_space *mapping, 458 pgoff_t offset, gfp_t gfp_mask) 459{ 460 int ret = add_to_page_cache(page, mapping, offset, gfp_mask); 461 if (ret == 0) 462 lru_cache_add(page); 463 return ret; 464} 465 466#ifdef CONFIG_NUMA 467struct page *__page_cache_alloc(gfp_t gfp) 468{ 469 if (cpuset_do_page_mem_spread()) { 470 int n = cpuset_mem_spread_node(); 471 return alloc_pages_node(n, gfp, 0); 472 } 473 return alloc_pages(gfp, 0); 474} 475EXPORT_SYMBOL(__page_cache_alloc); 476#endif 477 478static int __sleep_on_page_lock(void *word) 479{ 480 io_schedule(); 481 return 0; 482} 483 484/* 485 * In order to wait for pages to become available there must be 486 * waitqueues associated with pages. By using a hash table of 487 * waitqueues where the bucket discipline is to maintain all 488 * waiters on the same queue and wake all when any of the pages 489 * become available, and for the woken contexts to check to be 490 * sure the appropriate page became available, this saves space 491 * at a cost of "thundering herd" phenomena during rare hash 492 * collisions. 493 */ 494static wait_queue_head_t *page_waitqueue(struct page *page) 495{ 496 const struct zone *zone = page_zone(page); 497 498 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; 499} 500 501static inline void wake_up_page(struct page *page, int bit) 502{ 503 __wake_up_bit(page_waitqueue(page), &page->flags, bit); 504} 505 506void fastcall wait_on_page_bit(struct page *page, int bit_nr) 507{ 508 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); 509 510 if (test_bit(bit_nr, &page->flags)) 511 __wait_on_bit(page_waitqueue(page), &wait, sync_page, 512 TASK_UNINTERRUPTIBLE); 513} 514EXPORT_SYMBOL(wait_on_page_bit); 515 516/** 517 * unlock_page - unlock a locked page 518 * @page: the page 519 * 520 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). 521 * Also wakes sleepers in wait_on_page_writeback() because the wakeup 522 * mechananism between PageLocked pages and PageWriteback pages is shared. 523 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. 524 * 525 * The first mb is necessary to safely close the critical section opened by the 526 * TestSetPageLocked(), the second mb is necessary to enforce ordering between 527 * the clear_bit and the read of the waitqueue (to avoid SMP races with a 528 * parallel wait_on_page_locked()). 529 */ 530void fastcall unlock_page(struct page *page) 531{ 532 smp_mb__before_clear_bit(); 533 if (!TestClearPageLocked(page)) 534 BUG(); 535 smp_mb__after_clear_bit(); 536 wake_up_page(page, PG_locked); 537} 538EXPORT_SYMBOL(unlock_page); 539 540/** 541 * end_page_writeback - end writeback against a page 542 * @page: the page 543 */ 544void end_page_writeback(struct page *page) 545{ 546 if (!TestClearPageReclaim(page) || rotate_reclaimable_page(page)) { 547 if (!test_clear_page_writeback(page)) 548 BUG(); 549 } 550 smp_mb__after_clear_bit(); 551 wake_up_page(page, PG_writeback); 552} 553EXPORT_SYMBOL(end_page_writeback); 554 555/** 556 * __lock_page - get a lock on the page, assuming we need to sleep to get it 557 * @page: the page to lock 558 * 559 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some 560 * random driver's requestfn sets TASK_RUNNING, we could busywait. However 561 * chances are that on the second loop, the block layer's plug list is empty, 562 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE. 563 */ 564void fastcall __lock_page(struct page *page) 565{ 566 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); 567 568 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page, 569 TASK_UNINTERRUPTIBLE); 570} 571EXPORT_SYMBOL(__lock_page); 572 573/* 574 * Variant of lock_page that does not require the caller to hold a reference 575 * on the page's mapping. 576 */ 577void fastcall __lock_page_nosync(struct page *page) 578{ 579 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); 580 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock, 581 TASK_UNINTERRUPTIBLE); 582} 583 584/** 585 * find_get_page - find and get a page reference 586 * @mapping: the address_space to search 587 * @offset: the page index 588 * 589 * Is there a pagecache struct page at the given (mapping, offset) tuple? 590 * If yes, increment its refcount and return it; if no, return NULL. 591 */ 592struct page * find_get_page(struct address_space *mapping, unsigned long offset) 593{ 594 struct page *page; 595 596 read_lock_irq(&mapping->tree_lock); 597 page = radix_tree_lookup(&mapping->page_tree, offset); 598 if (page) 599 page_cache_get(page); 600 read_unlock_irq(&mapping->tree_lock); 601 return page; 602} 603EXPORT_SYMBOL(find_get_page); 604 605/** 606 * find_lock_page - locate, pin and lock a pagecache page 607 * @mapping: the address_space to search 608 * @offset: the page index 609 * 610 * Locates the desired pagecache page, locks it, increments its reference 611 * count and returns its address. 612 * 613 * Returns zero if the page was not present. find_lock_page() may sleep. 614 */ 615struct page *find_lock_page(struct address_space *mapping, 616 unsigned long offset) 617{ 618 struct page *page; 619 620 read_lock_irq(&mapping->tree_lock); 621repeat: 622 page = radix_tree_lookup(&mapping->page_tree, offset); 623 if (page) { 624 page_cache_get(page); 625 if (TestSetPageLocked(page)) { 626 read_unlock_irq(&mapping->tree_lock); 627 __lock_page(page); 628 read_lock_irq(&mapping->tree_lock); 629 630 /* Has the page been truncated while we slept? */ 631 if (unlikely(page->mapping != mapping || 632 page->index != offset)) { 633 unlock_page(page); 634 page_cache_release(page); 635 goto repeat; 636 } 637 } 638 } 639 read_unlock_irq(&mapping->tree_lock); 640 return page; 641} 642EXPORT_SYMBOL(find_lock_page); 643 644/** 645 * find_or_create_page - locate or add a pagecache page 646 * @mapping: the page's address_space 647 * @index: the page's index into the mapping 648 * @gfp_mask: page allocation mode 649 * 650 * Locates a page in the pagecache. If the page is not present, a new page 651 * is allocated using @gfp_mask and is added to the pagecache and to the VM's 652 * LRU list. The returned page is locked and has its reference count 653 * incremented. 654 * 655 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic 656 * allocation! 657 * 658 * find_or_create_page() returns the desired page's address, or zero on 659 * memory exhaustion. 660 */ 661struct page *find_or_create_page(struct address_space *mapping, 662 unsigned long index, gfp_t gfp_mask) 663{ 664 struct page *page, *cached_page = NULL; 665 int err; 666repeat: 667 page = find_lock_page(mapping, index); 668 if (!page) { 669 if (!cached_page) { 670 cached_page = 671 __page_cache_alloc(gfp_mask); 672 if (!cached_page) 673 return NULL; 674 } 675 err = add_to_page_cache_lru(cached_page, mapping, 676 index, gfp_mask); 677 if (!err) { 678 page = cached_page; 679 cached_page = NULL; 680 } else if (err == -EEXIST) 681 goto repeat; 682 } 683 if (cached_page) 684 page_cache_release(cached_page); 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, 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 unsigned int prev_offset; 872 loff_t isize; 873 struct page *cached_page; 874 int error; 875 struct file_ra_state ra = *_ra; 876 877 cached_page = NULL; 878 index = *ppos >> PAGE_CACHE_SHIFT; 879 next_index = index; 880 prev_index = ra.prev_index; 881 prev_offset = ra.prev_offset; 882 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; 883 offset = *ppos & ~PAGE_CACHE_MASK; 884 885 isize = i_size_read(inode); 886 if (!isize) 887 goto out; 888 889 end_index = (isize - 1) >> PAGE_CACHE_SHIFT; 890 for (;;) { 891 struct page *page; 892 unsigned long nr, ret; 893 894 /* nr is the maximum number of bytes to copy from this page */ 895 nr = PAGE_CACHE_SIZE; 896 if (index >= end_index) { 897 if (index > end_index) 898 goto out; 899 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; 900 if (nr <= offset) { 901 goto out; 902 } 903 } 904 nr = nr - offset; 905 906 cond_resched(); 907 if (index == next_index) 908 next_index = page_cache_readahead(mapping, &ra, filp, 909 index, last_index - index); 910 911find_page: 912 page = find_get_page(mapping, index); 913 if (unlikely(page == NULL)) { 914 handle_ra_miss(mapping, &ra, index); 915 goto no_cached_page; 916 } 917 if (!PageUptodate(page)) 918 goto page_not_up_to_date; 919page_ok: 920 921 /* If users can be writing to this page using arbitrary 922 * virtual addresses, take care about potential aliasing 923 * before reading the page on the kernel side. 924 */ 925 if (mapping_writably_mapped(mapping)) 926 flush_dcache_page(page); 927 928 /* 929 * When a sequential read accesses a page several times, 930 * only mark it as accessed the first time. 931 */ 932 if (prev_index != index || offset != prev_offset) 933 mark_page_accessed(page); 934 prev_index = index; 935 936 /* 937 * Ok, we have the page, and it's up-to-date, so 938 * now we can copy it to user space... 939 * 940 * The actor routine returns how many bytes were actually used.. 941 * NOTE! This may not be the same as how much of a user buffer 942 * we filled up (we may be padding etc), so we can only update 943 * "pos" here (the actor routine has to update the user buffer 944 * pointers and the remaining count). 945 */ 946 ret = actor(desc, page, offset, nr); 947 offset += ret; 948 index += offset >> PAGE_CACHE_SHIFT; 949 offset &= ~PAGE_CACHE_MASK; 950 prev_offset = offset; 951 ra.prev_offset = offset; 952 953 page_cache_release(page); 954 if (ret == nr && desc->count) 955 continue; 956 goto out; 957 958page_not_up_to_date: 959 /* Get exclusive access to the page ... */ 960 lock_page(page); 961 962 /* Did it get truncated before we got the lock? */ 963 if (!page->mapping) { 964 unlock_page(page); 965 page_cache_release(page); 966 continue; 967 } 968 969 /* Did somebody else fill it already? */ 970 if (PageUptodate(page)) { 971 unlock_page(page); 972 goto page_ok; 973 } 974 975readpage: 976 /* Start the actual read. The read will unlock the page. */ 977 error = mapping->a_ops->readpage(filp, page); 978 979 if (unlikely(error)) { 980 if (error == AOP_TRUNCATED_PAGE) { 981 page_cache_release(page); 982 goto find_page; 983 } 984 goto readpage_error; 985 } 986 987 if (!PageUptodate(page)) { 988 lock_page(page); 989 if (!PageUptodate(page)) { 990 if (page->mapping == NULL) { 991 /* 992 * invalidate_inode_pages got it 993 */ 994 unlock_page(page); 995 page_cache_release(page); 996 goto find_page; 997 } 998 unlock_page(page); 999 error = -EIO; 1000 shrink_readahead_size_eio(filp, &ra); 1001 goto readpage_error; 1002 } 1003 unlock_page(page); 1004 } 1005 1006 /* 1007 * i_size must be checked after we have done ->readpage. 1008 * 1009 * Checking i_size after the readpage allows us to calculate 1010 * the correct value for "nr", which means the zero-filled 1011 * part of the page is not copied back to userspace (unless 1012 * another truncate extends the file - this is desired though). 1013 */ 1014 isize = i_size_read(inode); 1015 end_index = (isize - 1) >> PAGE_CACHE_SHIFT; 1016 if (unlikely(!isize || index > end_index)) { 1017 page_cache_release(page); 1018 goto out; 1019 } 1020 1021 /* nr is the maximum number of bytes to copy from this page */ 1022 nr = PAGE_CACHE_SIZE; 1023 if (index == end_index) { 1024 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; 1025 if (nr <= offset) { 1026 page_cache_release(page); 1027 goto out; 1028 } 1029 } 1030 nr = nr - offset; 1031 goto page_ok; 1032 1033readpage_error: 1034 /* UHHUH! A synchronous read error occurred. Report it */ 1035 desc->error = error; 1036 page_cache_release(page); 1037 goto out; 1038 1039no_cached_page: 1040 /* 1041 * Ok, it wasn't cached, so we need to create a new 1042 * page.. 1043 */ 1044 if (!cached_page) { 1045 cached_page = page_cache_alloc_cold(mapping); 1046 if (!cached_page) { 1047 desc->error = -ENOMEM; 1048 goto out; 1049 } 1050 } 1051 error = add_to_page_cache_lru(cached_page, mapping, 1052 index, GFP_KERNEL); 1053 if (error) { 1054 if (error == -EEXIST) 1055 goto find_page; 1056 desc->error = error; 1057 goto out; 1058 } 1059 page = cached_page; 1060 cached_page = NULL; 1061 goto readpage; 1062 } 1063 1064out: 1065 *_ra = ra; 1066 1067 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; 1068 if (cached_page) 1069 page_cache_release(cached_page); 1070 if (filp) 1071 file_accessed(filp); 1072} 1073EXPORT_SYMBOL(do_generic_mapping_read); 1074 1075int file_read_actor(read_descriptor_t *desc, struct page *page, 1076 unsigned long offset, unsigned long size) 1077{ 1078 char *kaddr; 1079 unsigned long left, count = desc->count; 1080 1081 if (size > count) 1082 size = count; 1083 1084 /* 1085 * Faults on the destination of a read are common, so do it before 1086 * taking the kmap. 1087 */ 1088 if (!fault_in_pages_writeable(desc->arg.buf, size)) { 1089 kaddr = kmap_atomic(page, KM_USER0); 1090 left = __copy_to_user_inatomic(desc->arg.buf, 1091 kaddr + offset, size); 1092 kunmap_atomic(kaddr, KM_USER0); 1093 if (left == 0) 1094 goto success; 1095 } 1096 1097 /* Do it the slow way */ 1098 kaddr = kmap(page); 1099 left = __copy_to_user(desc->arg.buf, kaddr + offset, size); 1100 kunmap(page); 1101 1102 if (left) { 1103 size -= left; 1104 desc->error = -EFAULT; 1105 } 1106success: 1107 desc->count = count - size; 1108 desc->written += size; 1109 desc->arg.buf += size; 1110 return size; 1111} 1112 1113/* 1114 * Performs necessary checks before doing a write 1115 * @iov: io vector request 1116 * @nr_segs: number of segments in the iovec 1117 * @count: number of bytes to write 1118 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE 1119 * 1120 * Adjust number of segments and amount of bytes to write (nr_segs should be 1121 * properly initialized first). Returns appropriate error code that caller 1122 * should return or zero in case that write should be allowed. 1123 */ 1124int generic_segment_checks(const struct iovec *iov, 1125 unsigned long *nr_segs, size_t *count, int access_flags) 1126{ 1127 unsigned long seg; 1128 size_t cnt = 0; 1129 for (seg = 0; seg < *nr_segs; seg++) { 1130 const struct iovec *iv = &iov[seg]; 1131 1132 /* 1133 * If any segment has a negative length, or the cumulative 1134 * length ever wraps negative then return -EINVAL. 1135 */ 1136 cnt += iv->iov_len; 1137 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0)) 1138 return -EINVAL; 1139 if (access_ok(access_flags, iv->iov_base, iv->iov_len)) 1140 continue; 1141 if (seg == 0) 1142 return -EFAULT; 1143 *nr_segs = seg; 1144 cnt -= iv->iov_len; /* This segment is no good */ 1145 break; 1146 } 1147 *count = cnt; 1148 return 0; 1149} 1150EXPORT_SYMBOL(generic_segment_checks); 1151 1152/** 1153 * generic_file_aio_read - generic filesystem read routine 1154 * @iocb: kernel I/O control block 1155 * @iov: io vector request 1156 * @nr_segs: number of segments in the iovec 1157 * @pos: current file position 1158 * 1159 * This is the "read()" routine for all filesystems 1160 * that can use the page cache directly. 1161 */ 1162ssize_t 1163generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov, 1164 unsigned long nr_segs, loff_t pos) 1165{ 1166 struct file *filp = iocb->ki_filp; 1167 ssize_t retval; 1168 unsigned long seg; 1169 size_t count; 1170 loff_t *ppos = &iocb->ki_pos; 1171 1172 count = 0; 1173 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); 1174 if (retval) 1175 return retval; 1176 1177 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ 1178 if (filp->f_flags & O_DIRECT) { 1179 loff_t size; 1180 struct address_space *mapping; 1181 struct inode *inode; 1182 1183 mapping = filp->f_mapping; 1184 inode = mapping->host; 1185 retval = 0; 1186 if (!count) 1187 goto out; /* skip atime */ 1188 size = i_size_read(inode); 1189 if (pos < size) { 1190 retval = generic_file_direct_IO(READ, iocb, 1191 iov, pos, nr_segs); 1192 if (retval > 0) 1193 *ppos = pos + retval; 1194 } 1195 if (likely(retval != 0)) { 1196 file_accessed(filp); 1197 goto out; 1198 } 1199 } 1200 1201 retval = 0; 1202 if (count) { 1203 for (seg = 0; seg < nr_segs; seg++) { 1204 read_descriptor_t desc; 1205 1206 desc.written = 0; 1207 desc.arg.buf = iov[seg].iov_base; 1208 desc.count = iov[seg].iov_len; 1209 if (desc.count == 0) 1210 continue; 1211 desc.error = 0; 1212 do_generic_file_read(filp,ppos,&desc,file_read_actor); 1213 retval += desc.written; 1214 if (desc.error) { 1215 retval = retval ?: desc.error; 1216 break; 1217 } 1218 } 1219 } 1220out: 1221 return retval; 1222} 1223EXPORT_SYMBOL(generic_file_aio_read); 1224 1225int file_send_actor(read_descriptor_t * desc, struct page *page, unsigned long offset, unsigned long size) 1226{ 1227 ssize_t written; 1228 unsigned long count = desc->count; 1229 struct file *file = desc->arg.data; 1230 1231 if (size > count) 1232 size = count; 1233 1234 written = file->f_op->sendpage(file, page, offset, 1235 size, &file->f_pos, size<count); 1236 if (written < 0) { 1237 desc->error = written; 1238 written = 0; 1239 } 1240 desc->count = count - written; 1241 desc->written += written; 1242 return written; 1243} 1244 1245ssize_t generic_file_sendfile(struct file *in_file, loff_t *ppos, 1246 size_t count, read_actor_t actor, void *target) 1247{ 1248 read_descriptor_t desc; 1249 1250 if (!count) 1251 return 0; 1252 1253 desc.written = 0; 1254 desc.count = count; 1255 desc.arg.data = target; 1256 desc.error = 0; 1257 1258 do_generic_file_read(in_file, ppos, &desc, actor); 1259 if (desc.written) 1260 return desc.written; 1261 return desc.error; 1262} 1263EXPORT_SYMBOL(generic_file_sendfile); 1264 1265static ssize_t 1266do_readahead(struct address_space *mapping, struct file *filp, 1267 unsigned long index, unsigned long nr) 1268{ 1269 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage) 1270 return -EINVAL; 1271 1272 force_page_cache_readahead(mapping, filp, index, 1273 max_sane_readahead(nr)); 1274 return 0; 1275} 1276 1277asmlinkage ssize_t sys_readahead(int fd, loff_t offset, size_t count) 1278{ 1279 ssize_t ret; 1280 struct file *file; 1281 1282 ret = -EBADF; 1283 file = fget(fd); 1284 if (file) { 1285 if (file->f_mode & FMODE_READ) { 1286 struct address_space *mapping = file->f_mapping; 1287 unsigned long start = offset >> PAGE_CACHE_SHIFT; 1288 unsigned long end = (offset + count - 1) >> PAGE_CACHE_SHIFT; 1289 unsigned long len = end - start + 1; 1290 ret = do_readahead(mapping, file, start, len); 1291 } 1292 fput(file); 1293 } 1294 return ret; 1295} 1296 1297#ifdef CONFIG_MMU 1298static int FASTCALL(page_cache_read(struct file * file, unsigned long offset)); 1299/** 1300 * page_cache_read - adds requested page to the page cache if not already there 1301 * @file: file to read 1302 * @offset: page index 1303 * 1304 * This adds the requested page to the page cache if it isn't already there, 1305 * and schedules an I/O to read in its contents from disk. 1306 */ 1307static int fastcall page_cache_read(struct file * file, unsigned long offset) 1308{ 1309 struct address_space *mapping = file->f_mapping; 1310 struct page *page; 1311 int ret; 1312 1313 do { 1314 page = page_cache_alloc_cold(mapping); 1315 if (!page) 1316 return -ENOMEM; 1317 1318 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL); 1319 if (ret == 0) 1320 ret = mapping->a_ops->readpage(file, page); 1321 else if (ret == -EEXIST) 1322 ret = 0; /* losing race to add is OK */ 1323 1324 page_cache_release(page); 1325 1326 } while (ret == AOP_TRUNCATED_PAGE); 1327 1328 return ret; 1329} 1330 1331#define MMAP_LOTSAMISS (100) 1332 1333/** 1334 * filemap_nopage - read in file data for page fault handling 1335 * @area: the applicable vm_area 1336 * @address: target address to read in 1337 * @type: returned with VM_FAULT_{MINOR,MAJOR} if not %NULL 1338 * 1339 * filemap_nopage() is invoked via the vma operations vector for a 1340 * mapped memory region to read in file data during a page fault. 1341 * 1342 * The goto's are kind of ugly, but this streamlines the normal case of having 1343 * it in the page cache, and handles the special cases reasonably without 1344 * having a lot of duplicated code. 1345 */ 1346struct page *filemap_nopage(struct vm_area_struct *area, 1347 unsigned long address, int *type) 1348{ 1349 int error; 1350 struct file *file = area->vm_file; 1351 struct address_space *mapping = file->f_mapping; 1352 struct file_ra_state *ra = &file->f_ra; 1353 struct inode *inode = mapping->host; 1354 struct page *page; 1355 unsigned long size, pgoff; 1356 int did_readaround = 0, majmin = VM_FAULT_MINOR; 1357 1358 pgoff = ((address-area->vm_start) >> PAGE_CACHE_SHIFT) + area->vm_pgoff; 1359 1360retry_all: 1361 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 1362 if (pgoff >= size) 1363 goto outside_data_content; 1364 1365 /* If we don't want any read-ahead, don't bother */ 1366 if (VM_RandomReadHint(area)) 1367 goto no_cached_page; 1368 1369 /* 1370 * The readahead code wants to be told about each and every page 1371 * so it can build and shrink its windows appropriately 1372 * 1373 * For sequential accesses, we use the generic readahead logic. 1374 */ 1375 if (VM_SequentialReadHint(area)) 1376 page_cache_readahead(mapping, ra, file, pgoff, 1); 1377 1378 /* 1379 * Do we have something in the page cache already? 1380 */ 1381retry_find: 1382 page = find_get_page(mapping, pgoff); 1383 if (!page) { 1384 unsigned long ra_pages; 1385 1386 if (VM_SequentialReadHint(area)) { 1387 handle_ra_miss(mapping, ra, pgoff); 1388 goto no_cached_page; 1389 } 1390 ra->mmap_miss++; 1391 1392 /* 1393 * Do we miss much more than hit in this file? If so, 1394 * stop bothering with read-ahead. It will only hurt. 1395 */ 1396 if (ra->mmap_miss > ra->mmap_hit + MMAP_LOTSAMISS) 1397 goto no_cached_page; 1398 1399 /* 1400 * To keep the pgmajfault counter straight, we need to 1401 * check did_readaround, as this is an inner loop. 1402 */ 1403 if (!did_readaround) { 1404 majmin = VM_FAULT_MAJOR; 1405 count_vm_event(PGMAJFAULT); 1406 } 1407 did_readaround = 1; 1408 ra_pages = max_sane_readahead(file->f_ra.ra_pages); 1409 if (ra_pages) { 1410 pgoff_t start = 0; 1411 1412 if (pgoff > ra_pages / 2) 1413 start = pgoff - ra_pages / 2; 1414 do_page_cache_readahead(mapping, file, start, ra_pages); 1415 } 1416 page = find_get_page(mapping, pgoff); 1417 if (!page) 1418 goto no_cached_page; 1419 } 1420 1421 if (!did_readaround) 1422 ra->mmap_hit++; 1423 1424 /* 1425 * Ok, found a page in the page cache, now we need to check 1426 * that it's up-to-date. 1427 */ 1428 if (!PageUptodate(page)) 1429 goto page_not_uptodate; 1430 1431success: 1432 /* 1433 * Found the page and have a reference on it. 1434 */ 1435 mark_page_accessed(page); 1436 if (type) 1437 *type = majmin; 1438 return page; 1439 1440outside_data_content: 1441 /* 1442 * An external ptracer can access pages that normally aren't 1443 * accessible.. 1444 */ 1445 if (area->vm_mm == current->mm) 1446 return NOPAGE_SIGBUS; 1447 /* Fall through to the non-read-ahead case */ 1448no_cached_page: 1449 /* 1450 * We're only likely to ever get here if MADV_RANDOM is in 1451 * effect. 1452 */ 1453 error = page_cache_read(file, pgoff); 1454 1455 /* 1456 * The page we want has now been added to the page cache. 1457 * In the unlikely event that someone removed it in the 1458 * meantime, we'll just come back here and read it again. 1459 */ 1460 if (error >= 0) 1461 goto retry_find; 1462 1463 /* 1464 * An error return from page_cache_read can result if the 1465 * system is low on memory, or a problem occurs while trying 1466 * to schedule I/O. 1467 */ 1468 if (error == -ENOMEM) 1469 return NOPAGE_OOM; 1470 return NOPAGE_SIGBUS; 1471 1472page_not_uptodate: 1473 if (!did_readaround) { 1474 majmin = VM_FAULT_MAJOR; 1475 count_vm_event(PGMAJFAULT); 1476 } 1477 1478 /* 1479 * Umm, take care of errors if the page isn't up-to-date. 1480 * Try to re-read it _once_. We do this synchronously, 1481 * because there really aren't any performance issues here 1482 * and we need to check for errors. 1483 */ 1484 lock_page(page); 1485 1486 /* Somebody truncated the page on us? */ 1487 if (!page->mapping) { 1488 unlock_page(page); 1489 page_cache_release(page); 1490 goto retry_all; 1491 } 1492 1493 /* Somebody else successfully read it in? */ 1494 if (PageUptodate(page)) { 1495 unlock_page(page); 1496 goto success; 1497 } 1498 ClearPageError(page); 1499 error = mapping->a_ops->readpage(file, page); 1500 if (!error) { 1501 wait_on_page_locked(page); 1502 if (PageUptodate(page)) 1503 goto success; 1504 } else if (error == AOP_TRUNCATED_PAGE) { 1505 page_cache_release(page); 1506 goto retry_find; 1507 } 1508 1509 /* 1510 * Things didn't work out. Return zero to tell the 1511 * mm layer so, possibly freeing the page cache page first. 1512 */ 1513 shrink_readahead_size_eio(file, ra); 1514 page_cache_release(page); 1515 return NOPAGE_SIGBUS; 1516} 1517EXPORT_SYMBOL(filemap_nopage); 1518 1519static struct page * filemap_getpage(struct file *file, unsigned long pgoff, 1520 int nonblock) 1521{ 1522 struct address_space *mapping = file->f_mapping; 1523 struct page *page; 1524 int error; 1525 1526 /* 1527 * Do we have something in the page cache already? 1528 */ 1529retry_find: 1530 page = find_get_page(mapping, pgoff); 1531 if (!page) { 1532 if (nonblock) 1533 return NULL; 1534 goto no_cached_page; 1535 } 1536 1537 /* 1538 * Ok, found a page in the page cache, now we need to check 1539 * that it's up-to-date. 1540 */ 1541 if (!PageUptodate(page)) { 1542 if (nonblock) { 1543 page_cache_release(page); 1544 return NULL; 1545 } 1546 goto page_not_uptodate; 1547 } 1548 1549success: 1550 /* 1551 * Found the page and have a reference on it. 1552 */ 1553 mark_page_accessed(page); 1554 return page; 1555 1556no_cached_page: 1557 error = page_cache_read(file, pgoff); 1558 1559 /* 1560 * The page we want has now been added to the page cache. 1561 * In the unlikely event that someone removed it in the 1562 * meantime, we'll just come back here and read it again. 1563 */ 1564 if (error >= 0) 1565 goto retry_find; 1566 1567 /* 1568 * An error return from page_cache_read can result if the 1569 * system is low on memory, or a problem occurs while trying 1570 * to schedule I/O. 1571 */ 1572 return NULL; 1573 1574page_not_uptodate: 1575 lock_page(page); 1576 1577 /* Did it get truncated while we waited for it? */ 1578 if (!page->mapping) { 1579 unlock_page(page); 1580 goto err; 1581 } 1582 1583 /* Did somebody else get it up-to-date? */ 1584 if (PageUptodate(page)) { 1585 unlock_page(page); 1586 goto success; 1587 } 1588 1589 error = mapping->a_ops->readpage(file, page); 1590 if (!error) { 1591 wait_on_page_locked(page); 1592 if (PageUptodate(page)) 1593 goto success; 1594 } else if (error == AOP_TRUNCATED_PAGE) { 1595 page_cache_release(page); 1596 goto retry_find; 1597 } 1598 1599 /* 1600 * Umm, take care of errors if the page isn't up-to-date. 1601 * Try to re-read it _once_. We do this synchronously, 1602 * because there really aren't any performance issues here 1603 * and we need to check for errors. 1604 */ 1605 lock_page(page); 1606 1607 /* Somebody truncated the page on us? */ 1608 if (!page->mapping) { 1609 unlock_page(page); 1610 goto err; 1611 } 1612 /* Somebody else successfully read it in? */ 1613 if (PageUptodate(page)) { 1614 unlock_page(page); 1615 goto success; 1616 } 1617 1618 ClearPageError(page); 1619 error = mapping->a_ops->readpage(file, page); 1620 if (!error) { 1621 wait_on_page_locked(page); 1622 if (PageUptodate(page)) 1623 goto success; 1624 } else if (error == AOP_TRUNCATED_PAGE) { 1625 page_cache_release(page); 1626 goto retry_find; 1627 } 1628 1629 /* 1630 * Things didn't work out. Return zero to tell the 1631 * mm layer so, possibly freeing the page cache page first. 1632 */ 1633err: 1634 page_cache_release(page); 1635 1636 return NULL; 1637} 1638 1639int filemap_populate(struct vm_area_struct *vma, unsigned long addr, 1640 unsigned long len, pgprot_t prot, unsigned long pgoff, 1641 int nonblock) 1642{ 1643 struct file *file = vma->vm_file; 1644 struct address_space *mapping = file->f_mapping; 1645 struct inode *inode = mapping->host; 1646 unsigned long size; 1647 struct mm_struct *mm = vma->vm_mm; 1648 struct page *page; 1649 int err; 1650 1651 if (!nonblock) 1652 force_page_cache_readahead(mapping, vma->vm_file, 1653 pgoff, len >> PAGE_CACHE_SHIFT); 1654 1655repeat: 1656 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 1657 if (pgoff + (len >> PAGE_CACHE_SHIFT) > size) 1658 return -EINVAL; 1659 1660 page = filemap_getpage(file, pgoff, nonblock); 1661 1662 if (!page && !nonblock) 1663 return -ENOMEM; 1664 1665 if (page) { 1666 err = install_page(mm, vma, addr, page, prot); 1667 if (err) { 1668 page_cache_release(page); 1669 return err; 1670 } 1671 } else if (vma->vm_flags & VM_NONLINEAR) { 1672 /* No page was found just because we can't read it in now (being 1673 * here implies nonblock != 0), but the page may exist, so set 1674 * the PTE to fault it in later. */ 1675 err = install_file_pte(mm, vma, addr, pgoff, prot); 1676 if (err) 1677 return err; 1678 } 1679 1680 len -= PAGE_SIZE; 1681 addr += PAGE_SIZE; 1682 pgoff++; 1683 if (len) 1684 goto repeat; 1685 1686 return 0; 1687} 1688EXPORT_SYMBOL(filemap_populate); 1689 1690struct vm_operations_struct generic_file_vm_ops = { 1691 .nopage = filemap_nopage, 1692 .populate = filemap_populate, 1693}; 1694 1695/* This is used for a general mmap of a disk file */ 1696 1697int generic_file_mmap(struct file * file, struct vm_area_struct * vma) 1698{ 1699 struct address_space *mapping = file->f_mapping; 1700 1701 if (!mapping->a_ops->readpage) 1702 return -ENOEXEC; 1703 file_accessed(file); 1704 vma->vm_ops = &generic_file_vm_ops; 1705 return 0; 1706} 1707 1708/* 1709 * This is for filesystems which do not implement ->writepage. 1710 */ 1711int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) 1712{ 1713 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) 1714 return -EINVAL; 1715 return generic_file_mmap(file, vma); 1716} 1717#else 1718int generic_file_mmap(struct file * file, struct vm_area_struct * vma) 1719{ 1720 return -ENOSYS; 1721} 1722int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) 1723{ 1724 return -ENOSYS; 1725} 1726#endif /* CONFIG_MMU */ 1727 1728EXPORT_SYMBOL(generic_file_mmap); 1729EXPORT_SYMBOL(generic_file_readonly_mmap); 1730 1731static struct page *__read_cache_page(struct address_space *mapping, 1732 unsigned long index, 1733 int (*filler)(void *,struct page*), 1734 void *data) 1735{ 1736 struct page *page, *cached_page = NULL; 1737 int err; 1738repeat: 1739 page = find_get_page(mapping, index); 1740 if (!page) { 1741 if (!cached_page) { 1742 cached_page = page_cache_alloc_cold(mapping); 1743 if (!cached_page) 1744 return ERR_PTR(-ENOMEM); 1745 } 1746 err = add_to_page_cache_lru(cached_page, mapping, 1747 index, GFP_KERNEL); 1748 if (err == -EEXIST) 1749 goto repeat; 1750 if (err < 0) { 1751 /* Presumably ENOMEM for radix tree node */ 1752 page_cache_release(cached_page); 1753 return ERR_PTR(err); 1754 } 1755 page = cached_page; 1756 cached_page = NULL; 1757 err = filler(data, page); 1758 if (err < 0) { 1759 page_cache_release(page); 1760 page = ERR_PTR(err); 1761 } 1762 } 1763 if (cached_page) 1764 page_cache_release(cached_page); 1765 return page; 1766} 1767 1768/* 1769 * Same as read_cache_page, but don't wait for page to become unlocked 1770 * after submitting it to the filler. 1771 */ 1772struct page *read_cache_page_async(struct address_space *mapping, 1773 unsigned long index, 1774 int (*filler)(void *,struct page*), 1775 void *data) 1776{ 1777 struct page *page; 1778 int err; 1779 1780retry: 1781 page = __read_cache_page(mapping, index, filler, data); 1782 if (IS_ERR(page)) 1783 return page; 1784 if (PageUptodate(page)) 1785 goto out; 1786 1787 lock_page(page); 1788 if (!page->mapping) { 1789 unlock_page(page); 1790 page_cache_release(page); 1791 goto retry; 1792 } 1793 if (PageUptodate(page)) { 1794 unlock_page(page); 1795 goto out; 1796 } 1797 err = filler(data, page); 1798 if (err < 0) { 1799 page_cache_release(page); 1800 return ERR_PTR(err); 1801 } 1802out: 1803 mark_page_accessed(page); 1804 return page; 1805} 1806EXPORT_SYMBOL(read_cache_page_async); 1807 1808/** 1809 * read_cache_page - read into page cache, fill it if needed 1810 * @mapping: the page's address_space 1811 * @index: the page index 1812 * @filler: function to perform the read 1813 * @data: destination for read data 1814 * 1815 * Read into the page cache. If a page already exists, and PageUptodate() is 1816 * not set, try to fill the page then wait for it to become unlocked. 1817 * 1818 * If the page does not get brought uptodate, return -EIO. 1819 */ 1820struct page *read_cache_page(struct address_space *mapping, 1821 unsigned long index, 1822 int (*filler)(void *,struct page*), 1823 void *data) 1824{ 1825 struct page *page; 1826 1827 page = read_cache_page_async(mapping, index, filler, data); 1828 if (IS_ERR(page)) 1829 goto out; 1830 wait_on_page_locked(page); 1831 if (!PageUptodate(page)) { 1832 page_cache_release(page); 1833 page = ERR_PTR(-EIO); 1834 } 1835 out: 1836 return page; 1837} 1838EXPORT_SYMBOL(read_cache_page); 1839 1840/* 1841 * If the page was newly created, increment its refcount and add it to the 1842 * caller's lru-buffering pagevec. This function is specifically for 1843 * generic_file_write(). 1844 */ 1845static inline struct page * 1846__grab_cache_page(struct address_space *mapping, unsigned long index, 1847 struct page **cached_page, struct pagevec *lru_pvec) 1848{ 1849 int err; 1850 struct page *page; 1851repeat: 1852 page = find_lock_page(mapping, index); 1853 if (!page) { 1854 if (!*cached_page) { 1855 *cached_page = page_cache_alloc(mapping); 1856 if (!*cached_page) 1857 return NULL; 1858 } 1859 err = add_to_page_cache(*cached_page, mapping, 1860 index, GFP_KERNEL); 1861 if (err == -EEXIST) 1862 goto repeat; 1863 if (err == 0) { 1864 page = *cached_page; 1865 page_cache_get(page); 1866 if (!pagevec_add(lru_pvec, page)) 1867 __pagevec_lru_add(lru_pvec); 1868 *cached_page = NULL; 1869 } 1870 } 1871 return page; 1872} 1873 1874/* 1875 * The logic we want is 1876 * 1877 * if suid or (sgid and xgrp) 1878 * remove privs 1879 */ 1880int should_remove_suid(struct dentry *dentry) 1881{ 1882 mode_t mode = dentry->d_inode->i_mode; 1883 int kill = 0; 1884 1885 /* suid always must be killed */ 1886 if (unlikely(mode & S_ISUID)) 1887 kill = ATTR_KILL_SUID; 1888 1889 /* 1890 * sgid without any exec bits is just a mandatory locking mark; leave 1891 * it alone. If some exec bits are set, it's a real sgid; kill it. 1892 */ 1893 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 1894 kill |= ATTR_KILL_SGID; 1895 1896 if (unlikely(kill && !capable(CAP_FSETID))) 1897 return kill; 1898 1899 return 0; 1900} 1901EXPORT_SYMBOL(should_remove_suid); 1902 1903int __remove_suid(struct dentry *dentry, int kill) 1904{ 1905 struct iattr newattrs; 1906 1907 newattrs.ia_valid = ATTR_FORCE | kill; 1908 return notify_change(dentry, &newattrs); 1909} 1910 1911int remove_suid(struct dentry *dentry) 1912{ 1913 int kill = should_remove_suid(dentry); 1914 1915 if (unlikely(kill)) 1916 return __remove_suid(dentry, kill); 1917 1918 return 0; 1919} 1920EXPORT_SYMBOL(remove_suid); 1921 1922size_t 1923__filemap_copy_from_user_iovec_inatomic(char *vaddr, 1924 const struct iovec *iov, size_t base, size_t bytes) 1925{ 1926 size_t copied = 0, left = 0; 1927 1928 while (bytes) { 1929 char __user *buf = iov->iov_base + base; 1930 int copy = min(bytes, iov->iov_len - base); 1931 1932 base = 0; 1933 left = __copy_from_user_inatomic_nocache(vaddr, buf, copy); 1934 copied += copy; 1935 bytes -= copy; 1936 vaddr += copy; 1937 iov++; 1938 1939 if (unlikely(left)) 1940 break; 1941 } 1942 return copied - left; 1943} 1944 1945/* 1946 * Performs necessary checks before doing a write 1947 * 1948 * Can adjust writing position or amount of bytes to write. 1949 * Returns appropriate error code that caller should return or 1950 * zero in case that write should be allowed. 1951 */ 1952inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk) 1953{ 1954 struct inode *inode = file->f_mapping->host; 1955 unsigned long limit = current->signal->rlim[RLIMIT_FSIZE].rlim_cur; 1956 1957 if (unlikely(*pos < 0)) 1958 return -EINVAL; 1959 1960 if (!isblk) { 1961 if (file->f_flags & O_APPEND) 1962 *pos = i_size_read(inode); 1963 /* Foxconn modified start pling 12/04/2009 */ 1964 /* Remove large file limitation */ 1965#if (!defined SAMBA_ENABLE) 1966 if (limit != RLIM_INFINITY) { 1967 if (*pos >= limit) { 1968 send_sig(SIGXFSZ, current, 0); 1969 return -EFBIG; 1970 } 1971 if (*count > limit - (typeof(limit))*pos) { 1972 *count = limit - (typeof(limit))*pos; 1973 } 1974 } 1975#endif 1976 /* Foxconn modified end pling 12/04/2009 */ 1977 } 1978 1979 /* Foxconn modified start pling 12/04/2009 */ 1980 /* Ignore LFS rule to support large files */ 1981#if (!defined SAMBA_ENABLE) 1982 /* 1983 * LFS rule 1984 */ 1985 if (unlikely(*pos + *count > MAX_NON_LFS && 1986 !(file->f_flags & O_LARGEFILE))) { 1987 if (*pos >= MAX_NON_LFS) { 1988 send_sig(SIGXFSZ, current, 0); 1989 return -EFBIG; 1990 } 1991 if (*count > MAX_NON_LFS - (unsigned long)*pos) { 1992 *count = MAX_NON_LFS - (unsigned long)*pos; 1993 } 1994 } 1995#endif 1996 /* Foxconn modified end pling 12/04/2009 */ 1997 1998 /* 1999 * Are we about to exceed the fs block limit ? 2000 * 2001 * If we have written data it becomes a short write. If we have 2002 * exceeded without writing data we send a signal and return EFBIG. 2003 * Linus frestrict idea will clean these up nicely.. 2004 */ 2005 if (likely(!isblk)) { 2006 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) { 2007 if (*count || *pos > inode->i_sb->s_maxbytes) { 2008 send_sig(SIGXFSZ, current, 0); 2009 return -EFBIG; 2010 } 2011 /* zero-length writes at ->s_maxbytes are OK */ 2012 } 2013 2014 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes)) 2015 *count = inode->i_sb->s_maxbytes - *pos; 2016 } else { 2017#ifdef CONFIG_BLOCK 2018 loff_t isize; 2019 if (bdev_read_only(I_BDEV(inode))) 2020 return -EPERM; 2021 isize = i_size_read(inode); 2022 if (*pos >= isize) { 2023 if (*count || *pos > isize) 2024 return -ENOSPC; 2025 } 2026 2027 if (*pos + *count > isize) 2028 *count = isize - *pos; 2029#else 2030 return -EPERM; 2031#endif 2032 } 2033 return 0; 2034} 2035EXPORT_SYMBOL(generic_write_checks); 2036 2037ssize_t 2038generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov, 2039 unsigned long *nr_segs, loff_t pos, loff_t *ppos, 2040 size_t count, size_t ocount) 2041{ 2042 struct file *file = iocb->ki_filp; 2043 struct address_space *mapping = file->f_mapping; 2044 struct inode *inode = mapping->host; 2045 ssize_t written; 2046 2047 if (count != ocount) 2048 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count); 2049 2050 written = generic_file_direct_IO(WRITE, iocb, iov, pos, *nr_segs); 2051 if (written > 0) { 2052 loff_t end = pos + written; 2053 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { 2054 i_size_write(inode, end); 2055 mark_inode_dirty(inode); 2056 } 2057 *ppos = end; 2058 } 2059 2060 /* 2061 * Sync the fs metadata but not the minor inode changes and 2062 * of course not the data as we did direct DMA for the IO. 2063 * i_mutex is held, which protects generic_osync_inode() from 2064 * livelocking. AIO O_DIRECT ops attempt to sync metadata here. 2065 */ 2066 if ((written >= 0 || written == -EIOCBQUEUED) && 2067 ((file->f_flags & O_SYNC) || IS_SYNC(inode))) { 2068 int err = generic_osync_inode(inode, mapping, OSYNC_METADATA); 2069 if (err < 0) 2070 written = err; 2071 } 2072 return written; 2073} 2074EXPORT_SYMBOL(generic_file_direct_write); 2075 2076ssize_t 2077generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov, 2078 unsigned long nr_segs, loff_t pos, loff_t *ppos, 2079 size_t count, ssize_t written) 2080{ 2081 struct file *file = iocb->ki_filp; 2082 struct address_space * mapping = file->f_mapping; 2083 const struct address_space_operations *a_ops = mapping->a_ops; 2084 struct inode *inode = mapping->host; 2085 long status = 0; 2086 struct page *page; 2087 struct page *cached_page = NULL; 2088 size_t bytes; 2089 struct pagevec lru_pvec; 2090 const struct iovec *cur_iov = iov; /* current iovec */ 2091 size_t iov_base = 0; /* offset in the current iovec */ 2092 char __user *buf; 2093 2094 pagevec_init(&lru_pvec, 0); 2095 2096 /* 2097 * handle partial DIO write. Adjust cur_iov if needed. 2098 */ 2099 if (likely(nr_segs == 1)) 2100 buf = iov->iov_base + written; 2101 else { 2102 filemap_set_next_iovec(&cur_iov, &iov_base, written); 2103 buf = cur_iov->iov_base + iov_base; 2104 } 2105 2106 do { 2107 unsigned long index; 2108 unsigned long offset; 2109 size_t copied; 2110 2111 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */ 2112 index = pos >> PAGE_CACHE_SHIFT; 2113 bytes = PAGE_CACHE_SIZE - offset; 2114 2115 /* Limit the size of the copy to the caller's write size */ 2116 bytes = min(bytes, count); 2117 2118 /* We only need to worry about prefaulting when writes are from 2119 * user-space. NFSd uses vfs_writev with several non-aligned 2120 * segments in the vector, and limiting to one segment a time is 2121 * a noticeable performance for re-write 2122 */ 2123 if (!segment_eq(get_fs(), KERNEL_DS)) { 2124 /* 2125 * Limit the size of the copy to that of the current 2126 * segment, because fault_in_pages_readable() doesn't 2127 * know how to walk segments. 2128 */ 2129 bytes = min(bytes, cur_iov->iov_len - iov_base); 2130 2131 /* 2132 * Bring in the user page that we will copy from 2133 * _first_. Otherwise there's a nasty deadlock on 2134 * copying from the same page as we're writing to, 2135 * without it being marked up-to-date. 2136 */ 2137 fault_in_pages_readable(buf, bytes); 2138 } 2139 page = __grab_cache_page(mapping,index,&cached_page,&lru_pvec); 2140 if (!page) { 2141 status = -ENOMEM; 2142 break; 2143 } 2144 2145 if (unlikely(bytes == 0)) { 2146 status = 0; 2147 copied = 0; 2148 goto zero_length_segment; 2149 } 2150 2151 status = a_ops->prepare_write(file, page, offset, offset+bytes); 2152 if (unlikely(status)) { 2153 loff_t isize = i_size_read(inode); 2154 2155 if (status != AOP_TRUNCATED_PAGE) 2156 unlock_page(page); 2157 page_cache_release(page); 2158 if (status == AOP_TRUNCATED_PAGE) 2159 continue; 2160 /* 2161 * prepare_write() may have instantiated a few blocks 2162 * outside i_size. Trim these off again. 2163 */ 2164 if (pos + bytes > isize) 2165 vmtruncate(inode, isize); 2166 break; 2167 } 2168 if (likely(nr_segs == 1)) 2169 copied = filemap_copy_from_user(page, offset, 2170 buf, bytes); 2171 else 2172 copied = filemap_copy_from_user_iovec(page, offset, 2173 cur_iov, iov_base, bytes); 2174 flush_dcache_page(page); 2175 status = a_ops->commit_write(file, page, offset, offset+bytes); 2176 if (status == AOP_TRUNCATED_PAGE) { 2177 page_cache_release(page); 2178 continue; 2179 } 2180zero_length_segment: 2181 if (likely(copied >= 0)) { 2182 if (!status) 2183 status = copied; 2184 2185 if (status >= 0) { 2186 written += status; 2187 count -= status; 2188 pos += status; 2189 buf += status; 2190 if (unlikely(nr_segs > 1)) { 2191 filemap_set_next_iovec(&cur_iov, 2192 &iov_base, status); 2193 if (count) 2194 buf = cur_iov->iov_base + 2195 iov_base; 2196 } else { 2197 iov_base += status; 2198 } 2199 } 2200 } 2201 if (unlikely(copied != bytes)) 2202 if (status >= 0) 2203 status = -EFAULT; 2204 unlock_page(page); 2205 mark_page_accessed(page); 2206 page_cache_release(page); 2207 if (status < 0) 2208 break; 2209 balance_dirty_pages_ratelimited(mapping); 2210 cond_resched(); 2211 } while (count); 2212 *ppos = pos; 2213 2214 if (cached_page) 2215 page_cache_release(cached_page); 2216 2217 /* 2218 * For now, when the user asks for O_SYNC, we'll actually give O_DSYNC 2219 */ 2220 if (likely(status >= 0)) { 2221 if (unlikely((file->f_flags & O_SYNC) || IS_SYNC(inode))) { 2222 if (!a_ops->writepage || !is_sync_kiocb(iocb)) 2223 status = generic_osync_inode(inode, mapping, 2224 OSYNC_METADATA|OSYNC_DATA); 2225 } 2226 } 2227 2228 /* 2229 * If we get here for O_DIRECT writes then we must have fallen through 2230 * to buffered writes (block instantiation inside i_size). So we sync 2231 * the file data here, to try to honour O_DIRECT expectations. 2232 */ 2233 if (unlikely(file->f_flags & O_DIRECT) && written) 2234 status = filemap_write_and_wait(mapping); 2235 2236 pagevec_lru_add(&lru_pvec); 2237 return written ? written : status; 2238} 2239EXPORT_SYMBOL(generic_file_buffered_write); 2240 2241static ssize_t 2242__generic_file_aio_write_nolock(struct kiocb *iocb, const struct iovec *iov, 2243 unsigned long nr_segs, loff_t *ppos) 2244{ 2245 struct file *file = iocb->ki_filp; 2246 struct address_space * mapping = file->f_mapping; 2247 size_t ocount; /* original count */ 2248 size_t count; /* after file limit checks */ 2249 struct inode *inode = mapping->host; 2250 loff_t pos; 2251 ssize_t written; 2252 ssize_t err; 2253 2254 ocount = 0; 2255 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); 2256 if (err) 2257 return err; 2258 2259 count = ocount; 2260 pos = *ppos; 2261 2262 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); 2263 2264 /* We can write back this queue in page reclaim */ 2265 current->backing_dev_info = mapping->backing_dev_info; 2266 written = 0; 2267 2268 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); 2269 if (err) 2270 goto out; 2271 2272 if (count == 0) 2273 goto out; 2274 2275 err = remove_suid(file->f_path.dentry); 2276 if (err) 2277 goto out; 2278 2279 file_update_time(file); 2280 2281 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ 2282 if (unlikely(file->f_flags & O_DIRECT)) { 2283 loff_t endbyte; 2284 ssize_t written_buffered; 2285 2286 written = generic_file_direct_write(iocb, iov, &nr_segs, pos, 2287 ppos, count, ocount); 2288 if (written < 0 || written == count) 2289 goto out; 2290 /* 2291 * direct-io write to a hole: fall through to buffered I/O 2292 * for completing the rest of the request. 2293 */ 2294 pos += written; 2295 count -= written; 2296 written_buffered = generic_file_buffered_write(iocb, iov, 2297 nr_segs, pos, ppos, count, 2298 written); 2299 /* 2300 * If generic_file_buffered_write() retuned a synchronous error 2301 * then we want to return the number of bytes which were 2302 * direct-written, or the error code if that was zero. Note 2303 * that this differs from normal direct-io semantics, which 2304 * will return -EFOO even if some bytes were written. 2305 */ 2306 if (written_buffered < 0) { 2307 err = written_buffered; 2308 goto out; 2309 } 2310 2311 /* 2312 * We need to ensure that the page cache pages are written to 2313 * disk and invalidated to preserve the expected O_DIRECT 2314 * semantics. 2315 */ 2316 endbyte = pos + written_buffered - written - 1; 2317 err = do_sync_mapping_range(file->f_mapping, pos, endbyte, 2318 SYNC_FILE_RANGE_WAIT_BEFORE| 2319 SYNC_FILE_RANGE_WRITE| 2320 SYNC_FILE_RANGE_WAIT_AFTER); 2321 if (err == 0) { 2322 written = written_buffered; 2323 invalidate_mapping_pages(mapping, 2324 pos >> PAGE_CACHE_SHIFT, 2325 endbyte >> PAGE_CACHE_SHIFT); 2326 } else { 2327 /* 2328 * We don't know how much we wrote, so just return 2329 * the number of bytes which were direct-written 2330 */ 2331 } 2332 } else { 2333 written = generic_file_buffered_write(iocb, iov, nr_segs, 2334 pos, ppos, count, written); 2335 } 2336out: 2337 current->backing_dev_info = NULL; 2338 return written ? written : err; 2339} 2340 2341ssize_t generic_file_aio_write_nolock(struct kiocb *iocb, 2342 const struct iovec *iov, unsigned long nr_segs, loff_t pos) 2343{ 2344 struct file *file = iocb->ki_filp; 2345 struct address_space *mapping = file->f_mapping; 2346 struct inode *inode = mapping->host; 2347 ssize_t ret; 2348 2349 BUG_ON(iocb->ki_pos != pos); 2350 2351 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs, 2352 &iocb->ki_pos); 2353 2354 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) { 2355 ssize_t err; 2356 2357 err = sync_page_range_nolock(inode, mapping, pos, ret); 2358 if (err < 0) 2359 ret = err; 2360 } 2361 return ret; 2362} 2363EXPORT_SYMBOL(generic_file_aio_write_nolock); 2364 2365ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, 2366 unsigned long nr_segs, loff_t pos) 2367{ 2368 struct file *file = iocb->ki_filp; 2369 struct address_space *mapping = file->f_mapping; 2370 struct inode *inode = mapping->host; 2371 ssize_t ret; 2372 2373 BUG_ON(iocb->ki_pos != pos); 2374 2375 mutex_lock(&inode->i_mutex); 2376 ret = __generic_file_aio_write_nolock(iocb, iov, nr_segs, 2377 &iocb->ki_pos); 2378 mutex_unlock(&inode->i_mutex); 2379 2380 if (ret > 0 && ((file->f_flags & O_SYNC) || IS_SYNC(inode))) { 2381 ssize_t err; 2382 2383 err = sync_page_range(inode, mapping, pos, ret); 2384 if (err < 0) 2385 ret = err; 2386 } 2387 return ret; 2388} 2389EXPORT_SYMBOL(generic_file_aio_write); 2390 2391/* 2392 * Called under i_mutex for writes to S_ISREG files. Returns -EIO if something 2393 * went wrong during pagecache shootdown. 2394 */ 2395static ssize_t 2396generic_file_direct_IO(int rw, struct kiocb *iocb, const struct iovec *iov, 2397 loff_t offset, unsigned long nr_segs) 2398{ 2399 struct file *file = iocb->ki_filp; 2400 struct address_space *mapping = file->f_mapping; 2401 ssize_t retval; 2402 size_t write_len; 2403 pgoff_t end = 0; /* silence gcc */ 2404 2405 /* 2406 * If it's a write, unmap all mmappings of the file up-front. This 2407 * will cause any pte dirty bits to be propagated into the pageframes 2408 * for the subsequent filemap_write_and_wait(). 2409 */ 2410 if (rw == WRITE) { 2411 write_len = iov_length(iov, nr_segs); 2412 end = (offset + write_len - 1) >> PAGE_CACHE_SHIFT; 2413 if (mapping_mapped(mapping)) 2414 unmap_mapping_range(mapping, offset, write_len, 0); 2415 } 2416 2417 retval = filemap_write_and_wait(mapping); 2418 if (retval) 2419 goto out; 2420 2421 /* 2422 * After a write we want buffered reads to be sure to go to disk to get 2423 * the new data. We invalidate clean cached page from the region we're 2424 * about to write. We do this *before* the write so that we can return 2425 * -EIO without clobbering -EIOCBQUEUED from ->direct_IO(). 2426 */ 2427 if (rw == WRITE && mapping->nrpages) { 2428 retval = invalidate_inode_pages2_range(mapping, 2429 offset >> PAGE_CACHE_SHIFT, end); 2430 if (retval) 2431 goto out; 2432 } 2433 2434 retval = mapping->a_ops->direct_IO(rw, iocb, iov, offset, nr_segs); 2435 if (retval) 2436 goto out; 2437 2438 /* 2439 * Finally, try again to invalidate clean pages which might have been 2440 * faulted in by get_user_pages() if the source of the write was an 2441 * mmap()ed region of the file we're writing. That's a pretty crazy 2442 * thing to do, so we don't support it 100%. If this invalidation 2443 * fails and we have -EIOCBQUEUED we ignore the failure. 2444 */ 2445 if (rw == WRITE && mapping->nrpages) { 2446 int err = invalidate_inode_pages2_range(mapping, 2447 offset >> PAGE_CACHE_SHIFT, end); 2448 if (err && retval >= 0) 2449 retval = err; 2450 } 2451out: 2452 return retval; 2453} 2454 2455/** 2456 * try_to_release_page() - release old fs-specific metadata on a page 2457 * 2458 * @page: the page which the kernel is trying to free 2459 * @gfp_mask: memory allocation flags (and I/O mode) 2460 * 2461 * The address_space is to try to release any data against the page 2462 * (presumably at page->private). If the release was successful, return `1'. 2463 * Otherwise return zero. 2464 * 2465 * The @gfp_mask argument specifies whether I/O may be performed to release 2466 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT). 2467 * 2468 * NOTE: @gfp_mask may go away, and this function may become non-blocking. 2469 */ 2470int try_to_release_page(struct page *page, gfp_t gfp_mask) 2471{ 2472 struct address_space * const mapping = page->mapping; 2473 2474 BUG_ON(!PageLocked(page)); 2475 if (PageWriteback(page)) 2476 return 0; 2477 2478 if (mapping && mapping->a_ops->releasepage) 2479 return mapping->a_ops->releasepage(page, gfp_mask); 2480 return try_to_free_buffers(page); 2481} 2482 2483EXPORT_SYMBOL(try_to_release_page); 2484