1/* Modified by Broadcom Corp. Portions Copyright (c) Broadcom Corp, 2012. */ 2/* 3 * linux/mm/filemap.c 4 * 5 * Copyright (C) 1994-1999 Linus Torvalds 6 */ 7 8/* 9 * This file handles the generic file mmap semantics used by 10 * most "normal" filesystems (but you don't /have/ to use this: 11 * the NFS filesystem used to do this differently, for example) 12 */ 13#include <linux/module.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/gfp.h> 21#include <linux/mm.h> 22#include <linux/swap.h> 23#include <linux/mman.h> 24#include <linux/pagemap.h> 25#include <linux/file.h> 26#include <linux/uio.h> 27#include <linux/hash.h> 28#include <linux/writeback.h> 29#include <linux/backing-dev.h> 30#include <linux/pagevec.h> 31#include <linux/blkdev.h> 32#include <linux/security.h> 33#include <linux/syscalls.h> 34#include <linux/cpuset.h> 35#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */ 36#include <linux/memcontrol.h> 37#include <linux/mm_inline.h> /* for page_is_file_cache() */ 38#include "internal.h" 39 40#include <linux/buffer_head.h> /* for try_to_free_buffers */ 41 42#include <asm/mman.h> 43 44#include <typedefs.h> 45#include <bcmdefs.h> 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 (truncate_pagecache) 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 * (code doesn't rely on that order, so you could switch it around) 110 * ->tasklist_lock (memory_failure, collect_procs_ao) 111 * ->i_mmap_lock 112 */ 113 114/* 115 * Remove a page from the page cache and free it. Caller has to make 116 * sure the page is locked and that nobody else uses it - or that usage 117 * is safe. The caller must hold the mapping's tree_lock. 118 */ 119void __remove_from_page_cache(struct page *page) 120{ 121 struct address_space *mapping = page->mapping; 122 123 radix_tree_delete(&mapping->page_tree, page->index); 124 page->mapping = NULL; 125 mapping->nrpages--; 126 __dec_zone_page_state(page, NR_FILE_PAGES); 127 if (PageSwapBacked(page)) 128 __dec_zone_page_state(page, NR_SHMEM); 129 BUG_ON(page_mapped(page)); 130 131 /* 132 * Some filesystems seem to re-dirty the page even after 133 * the VM has canceled the dirty bit (eg ext3 journaling). 134 * 135 * Fix it up by doing a final dirty accounting check after 136 * having removed the page entirely. 137 */ 138 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) { 139 dec_zone_page_state(page, NR_FILE_DIRTY); 140 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); 141 } 142} 143 144void remove_from_page_cache(struct page *page) 145{ 146 struct address_space *mapping = page->mapping; 147 148 BUG_ON(!PageLocked(page)); 149 150 spin_lock_irq(&mapping->tree_lock); 151 __remove_from_page_cache(page); 152 spin_unlock_irq(&mapping->tree_lock); 153 mem_cgroup_uncharge_cache_page(page); 154} 155EXPORT_SYMBOL(remove_from_page_cache); 156 157static int sync_page(void *word) 158{ 159 struct address_space *mapping; 160 struct page *page; 161 162 page = container_of((unsigned long *)word, struct page, flags); 163 164 /* 165 * page_mapping() is being called without PG_locked held. 166 * Some knowledge of the state and use of the page is used to 167 * reduce the requirements down to a memory barrier. 168 * The danger here is of a stale page_mapping() return value 169 * indicating a struct address_space different from the one it's 170 * associated with when it is associated with one. 171 * After smp_mb(), it's either the correct page_mapping() for 172 * the page, or an old page_mapping() and the page's own 173 * page_mapping() has gone NULL. 174 * The ->sync_page() address_space operation must tolerate 175 * page_mapping() going NULL. By an amazing coincidence, 176 * this comes about because none of the users of the page 177 * in the ->sync_page() methods make essential use of the 178 * page_mapping(), merely passing the page down to the backing 179 * device's unplug functions when it's non-NULL, which in turn 180 * ignore it for all cases but swap, where only page_private(page) is 181 * of interest. When page_mapping() does go NULL, the entire 182 * call stack gracefully ignores the page and returns. 183 * -- wli 184 */ 185 smp_mb(); 186 mapping = page_mapping(page); 187 if (mapping && mapping->a_ops && mapping->a_ops->sync_page) 188 mapping->a_ops->sync_page(page); 189 io_schedule(); 190 return 0; 191} 192 193static int sync_page_killable(void *word) 194{ 195 sync_page(word); 196 return fatal_signal_pending(current) ? -EINTR : 0; 197} 198 199/** 200 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range 201 * @mapping: address space structure to write 202 * @start: offset in bytes where the range starts 203 * @end: offset in bytes where the range ends (inclusive) 204 * @sync_mode: enable synchronous operation 205 * 206 * Start writeback against all of a mapping's dirty pages that lie 207 * within the byte offsets <start, end> inclusive. 208 * 209 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as 210 * opposed to a regular memory cleansing writeback. The difference between 211 * these two operations is that if a dirty page/buffer is encountered, it must 212 * be waited upon, and not just skipped over. 213 */ 214int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, 215 loff_t end, int sync_mode) 216{ 217 int ret; 218 struct writeback_control wbc = { 219 .sync_mode = sync_mode, 220 .nr_to_write = LONG_MAX, 221 .range_start = start, 222 .range_end = end, 223 }; 224 225 if (!mapping_cap_writeback_dirty(mapping)) 226 return 0; 227 228 ret = do_writepages(mapping, &wbc); 229 return ret; 230} 231 232static inline int __filemap_fdatawrite(struct address_space *mapping, 233 int sync_mode) 234{ 235 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); 236} 237 238int filemap_fdatawrite(struct address_space *mapping) 239{ 240 return __filemap_fdatawrite(mapping, WB_SYNC_ALL); 241} 242EXPORT_SYMBOL(filemap_fdatawrite); 243 244int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, 245 loff_t end) 246{ 247 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); 248} 249EXPORT_SYMBOL(filemap_fdatawrite_range); 250 251/** 252 * filemap_flush - mostly a non-blocking flush 253 * @mapping: target address_space 254 * 255 * This is a mostly non-blocking flush. Not suitable for data-integrity 256 * purposes - I/O may not be started against all dirty pages. 257 */ 258int filemap_flush(struct address_space *mapping) 259{ 260 return __filemap_fdatawrite(mapping, WB_SYNC_NONE); 261} 262EXPORT_SYMBOL(filemap_flush); 263 264/** 265 * filemap_fdatawait_range - wait for writeback to complete 266 * @mapping: address space structure to wait for 267 * @start_byte: offset in bytes where the range starts 268 * @end_byte: offset in bytes where the range ends (inclusive) 269 * 270 * Walk the list of under-writeback pages of the given address space 271 * in the given range and wait for all of them. 272 */ 273int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, 274 loff_t end_byte) 275{ 276 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT; 277 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT; 278 struct pagevec pvec; 279 int nr_pages; 280 int ret = 0; 281 282 if (end_byte < start_byte) 283 return 0; 284 285 pagevec_init(&pvec, 0); 286 while ((index <= end) && 287 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, 288 PAGECACHE_TAG_WRITEBACK, 289 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) { 290 unsigned i; 291 292 for (i = 0; i < nr_pages; i++) { 293 struct page *page = pvec.pages[i]; 294 295 /* until radix tree lookup accepts end_index */ 296 if (page->index > end) 297 continue; 298 299 wait_on_page_writeback(page); 300 if (PageError(page)) 301 ret = -EIO; 302 } 303 pagevec_release(&pvec); 304 cond_resched(); 305 } 306 307 /* Check for outstanding write errors */ 308 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags)) 309 ret = -ENOSPC; 310 if (test_and_clear_bit(AS_EIO, &mapping->flags)) 311 ret = -EIO; 312 313 return ret; 314} 315EXPORT_SYMBOL(filemap_fdatawait_range); 316 317/** 318 * filemap_fdatawait - wait for all under-writeback pages to complete 319 * @mapping: address space structure to wait for 320 * 321 * Walk the list of under-writeback pages of the given address space 322 * and wait for all of them. 323 */ 324int filemap_fdatawait(struct address_space *mapping) 325{ 326 loff_t i_size = i_size_read(mapping->host); 327 328 if (i_size == 0) 329 return 0; 330 331 return filemap_fdatawait_range(mapping, 0, i_size - 1); 332} 333EXPORT_SYMBOL(filemap_fdatawait); 334 335int filemap_write_and_wait(struct address_space *mapping) 336{ 337 int err = 0; 338 339 if (mapping->nrpages) { 340 err = filemap_fdatawrite(mapping); 341 /* 342 * Even if the above returned error, the pages may be 343 * written partially (e.g. -ENOSPC), so we wait for it. 344 * But the -EIO is special case, it may indicate the worst 345 * thing (e.g. bug) happened, so we avoid waiting for it. 346 */ 347 if (err != -EIO) { 348 int err2 = filemap_fdatawait(mapping); 349 if (!err) 350 err = err2; 351 } 352 } 353 return err; 354} 355EXPORT_SYMBOL(filemap_write_and_wait); 356 357/** 358 * filemap_write_and_wait_range - write out & wait on a file range 359 * @mapping: the address_space for the pages 360 * @lstart: offset in bytes where the range starts 361 * @lend: offset in bytes where the range ends (inclusive) 362 * 363 * Write out and wait upon file offsets lstart->lend, inclusive. 364 * 365 * Note that `lend' is inclusive (describes the last byte to be written) so 366 * that this function can be used to write to the very end-of-file (end = -1). 367 */ 368int filemap_write_and_wait_range(struct address_space *mapping, 369 loff_t lstart, loff_t lend) 370{ 371 int err = 0; 372 373 if (mapping->nrpages) { 374 err = __filemap_fdatawrite_range(mapping, lstart, lend, 375 WB_SYNC_ALL); 376 /* See comment of filemap_write_and_wait() */ 377 if (err != -EIO) { 378 int err2 = filemap_fdatawait_range(mapping, 379 lstart, lend); 380 if (!err) 381 err = err2; 382 } 383 } 384 return err; 385} 386EXPORT_SYMBOL(filemap_write_and_wait_range); 387 388/** 389 * add_to_page_cache_locked - add a locked page to the pagecache 390 * @page: page to add 391 * @mapping: the page's address_space 392 * @offset: page index 393 * @gfp_mask: page allocation mode 394 * 395 * This function is used to add a page to the pagecache. It must be locked. 396 * This function does not add the page to the LRU. The caller must do that. 397 */ 398int add_to_page_cache_locked(struct page *page, struct address_space *mapping, 399 pgoff_t offset, gfp_t gfp_mask) 400{ 401 int error; 402 403 VM_BUG_ON(!PageLocked(page)); 404 405 error = mem_cgroup_cache_charge(page, current->mm, 406 gfp_mask & GFP_RECLAIM_MASK); 407 if (error) 408 goto out; 409 410 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); 411 if (error == 0) { 412 page_cache_get(page); 413 page->mapping = mapping; 414 page->index = offset; 415 416 spin_lock_irq(&mapping->tree_lock); 417 error = radix_tree_insert(&mapping->page_tree, offset, page); 418 if (likely(!error)) { 419 mapping->nrpages++; 420 __inc_zone_page_state(page, NR_FILE_PAGES); 421 if (PageSwapBacked(page)) 422 __inc_zone_page_state(page, NR_SHMEM); 423 spin_unlock_irq(&mapping->tree_lock); 424 } else { 425 page->mapping = NULL; 426 spin_unlock_irq(&mapping->tree_lock); 427 mem_cgroup_uncharge_cache_page(page); 428 page_cache_release(page); 429 } 430 radix_tree_preload_end(); 431 } else 432 mem_cgroup_uncharge_cache_page(page); 433out: 434 return error; 435} 436EXPORT_SYMBOL(add_to_page_cache_locked); 437 438int add_to_page_cache_lru(struct page *page, struct address_space *mapping, 439 pgoff_t offset, gfp_t gfp_mask) 440{ 441 int ret; 442 443 /* 444 * Splice_read and readahead add shmem/tmpfs pages into the page cache 445 * before shmem_readpage has a chance to mark them as SwapBacked: they 446 * need to go on the anon lru below, and mem_cgroup_cache_charge 447 * (called in add_to_page_cache) needs to know where they're going too. 448 */ 449 if (mapping_cap_swap_backed(mapping)) 450 SetPageSwapBacked(page); 451 452 ret = add_to_page_cache(page, mapping, offset, gfp_mask); 453 if (ret == 0) { 454 if (page_is_file_cache(page)) 455 lru_cache_add_file(page); 456 else 457 lru_cache_add_anon(page); 458 } 459 return ret; 460} 461EXPORT_SYMBOL_GPL(add_to_page_cache_lru); 462 463#ifdef CONFIG_NUMA 464struct page *__page_cache_alloc(gfp_t gfp) 465{ 466 int n; 467 struct page *page; 468 469 if (cpuset_do_page_mem_spread()) { 470 get_mems_allowed(); 471 n = cpuset_mem_spread_node(); 472 page = alloc_pages_exact_node(n, gfp, 0); 473 put_mems_allowed(); 474 return page; 475 } 476 return alloc_pages(gfp, 0); 477} 478EXPORT_SYMBOL(__page_cache_alloc); 479#endif 480 481static int __sleep_on_page_lock(void *word) 482{ 483 io_schedule(); 484 return 0; 485} 486 487/* 488 * In order to wait for pages to become available there must be 489 * waitqueues associated with pages. By using a hash table of 490 * waitqueues where the bucket discipline is to maintain all 491 * waiters on the same queue and wake all when any of the pages 492 * become available, and for the woken contexts to check to be 493 * sure the appropriate page became available, this saves space 494 * at a cost of "thundering herd" phenomena during rare hash 495 * collisions. 496 */ 497static wait_queue_head_t *page_waitqueue(struct page *page) 498{ 499 const struct zone *zone = page_zone(page); 500 501 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; 502} 503 504static inline void wake_up_page(struct page *page, int bit) 505{ 506 __wake_up_bit(page_waitqueue(page), &page->flags, bit); 507} 508 509void wait_on_page_bit(struct page *page, int bit_nr) 510{ 511 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); 512 513 if (test_bit(bit_nr, &page->flags)) 514 __wait_on_bit(page_waitqueue(page), &wait, sync_page, 515 TASK_UNINTERRUPTIBLE); 516} 517EXPORT_SYMBOL(wait_on_page_bit); 518 519/** 520 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue 521 * @page: Page defining the wait queue of interest 522 * @waiter: Waiter to add to the queue 523 * 524 * Add an arbitrary @waiter to the wait queue for the nominated @page. 525 */ 526void add_page_wait_queue(struct page *page, wait_queue_t *waiter) 527{ 528 wait_queue_head_t *q = page_waitqueue(page); 529 unsigned long flags; 530 531 spin_lock_irqsave(&q->lock, flags); 532 __add_wait_queue(q, waiter); 533 spin_unlock_irqrestore(&q->lock, flags); 534} 535EXPORT_SYMBOL_GPL(add_page_wait_queue); 536 537/** 538 * unlock_page - unlock a locked page 539 * @page: the page 540 * 541 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). 542 * Also wakes sleepers in wait_on_page_writeback() because the wakeup 543 * mechananism between PageLocked pages and PageWriteback pages is shared. 544 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. 545 * 546 * The mb is necessary to enforce ordering between the clear_bit and the read 547 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()). 548 */ 549void BCMFASTPATH_HOST unlock_page(struct page *page) 550{ 551 VM_BUG_ON(!PageLocked(page)); 552 clear_bit_unlock(PG_locked, &page->flags); 553 smp_mb__after_clear_bit(); 554 wake_up_page(page, PG_locked); 555} 556EXPORT_SYMBOL(unlock_page); 557 558/** 559 * end_page_writeback - end writeback against a page 560 * @page: the page 561 */ 562void end_page_writeback(struct page *page) 563{ 564 if (TestClearPageReclaim(page)) 565 rotate_reclaimable_page(page); 566 567 if (!test_clear_page_writeback(page)) 568 BUG(); 569 570 smp_mb__after_clear_bit(); 571 wake_up_page(page, PG_writeback); 572} 573EXPORT_SYMBOL(end_page_writeback); 574 575/** 576 * __lock_page - get a lock on the page, assuming we need to sleep to get it 577 * @page: the page to lock 578 * 579 * Ugly. Running sync_page() in state TASK_UNINTERRUPTIBLE is scary. If some 580 * random driver's requestfn sets TASK_RUNNING, we could busywait. However 581 * chances are that on the second loop, the block layer's plug list is empty, 582 * so sync_page() will then return in state TASK_UNINTERRUPTIBLE. 583 */ 584void __lock_page(struct page *page) 585{ 586 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); 587 588 __wait_on_bit_lock(page_waitqueue(page), &wait, sync_page, 589 TASK_UNINTERRUPTIBLE); 590} 591EXPORT_SYMBOL(__lock_page); 592 593int __lock_page_killable(struct page *page) 594{ 595 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); 596 597 return __wait_on_bit_lock(page_waitqueue(page), &wait, 598 sync_page_killable, TASK_KILLABLE); 599} 600EXPORT_SYMBOL_GPL(__lock_page_killable); 601 602/** 603 * __lock_page_nosync - get a lock on the page, without calling sync_page() 604 * @page: the page to lock 605 * 606 * Variant of lock_page that does not require the caller to hold a reference 607 * on the page's mapping. 608 */ 609void __lock_page_nosync(struct page *page) 610{ 611 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); 612 __wait_on_bit_lock(page_waitqueue(page), &wait, __sleep_on_page_lock, 613 TASK_UNINTERRUPTIBLE); 614} 615 616/** 617 * find_get_page - find and get a page reference 618 * @mapping: the address_space to search 619 * @offset: the page index 620 * 621 * Is there a pagecache struct page at the given (mapping, offset) tuple? 622 * If yes, increment its refcount and return it; if no, return NULL. 623 */ 624struct page *find_get_page(struct address_space *mapping, pgoff_t offset) 625{ 626 void **pagep; 627 struct page *page; 628 629 rcu_read_lock(); 630repeat: 631 page = NULL; 632 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); 633 if (pagep) { 634 page = radix_tree_deref_slot(pagep); 635 if (unlikely(!page)) 636 goto out; 637 if (radix_tree_deref_retry(page)) 638 goto repeat; 639 640 if (!page_cache_get_speculative(page)) 641 goto repeat; 642 643 /* 644 * Has the page moved? 645 * This is part of the lockless pagecache protocol. See 646 * include/linux/pagemap.h for details. 647 */ 648 if (unlikely(page != *pagep)) { 649 page_cache_release(page); 650 goto repeat; 651 } 652 } 653out: 654 rcu_read_unlock(); 655 656 return page; 657} 658EXPORT_SYMBOL(find_get_page); 659 660/** 661 * find_lock_page - locate, pin and lock a pagecache page 662 * @mapping: the address_space to search 663 * @offset: the page index 664 * 665 * Locates the desired pagecache page, locks it, increments its reference 666 * count and returns its address. 667 * 668 * Returns zero if the page was not present. find_lock_page() may sleep. 669 */ 670struct page *find_lock_page(struct address_space *mapping, pgoff_t offset) 671{ 672 struct page *page; 673 674repeat: 675 page = find_get_page(mapping, offset); 676 if (page) { 677 lock_page(page); 678 /* Has the page been truncated? */ 679 if (unlikely(page->mapping != mapping)) { 680 unlock_page(page); 681 page_cache_release(page); 682 goto repeat; 683 } 684 VM_BUG_ON(page->index != offset); 685 } 686 return page; 687} 688EXPORT_SYMBOL(find_lock_page); 689 690/** 691 * find_or_create_page - locate or add a pagecache page 692 * @mapping: the page's address_space 693 * @index: the page's index into the mapping 694 * @gfp_mask: page allocation mode 695 * 696 * Locates a page in the pagecache. If the page is not present, a new page 697 * is allocated using @gfp_mask and is added to the pagecache and to the VM's 698 * LRU list. The returned page is locked and has its reference count 699 * incremented. 700 * 701 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic 702 * allocation! 703 * 704 * find_or_create_page() returns the desired page's address, or zero on 705 * memory exhaustion. 706 */ 707struct page *find_or_create_page(struct address_space *mapping, 708 pgoff_t index, gfp_t gfp_mask) 709{ 710 struct page *page; 711 int err; 712repeat: 713 page = find_lock_page(mapping, index); 714 if (!page) { 715 page = __page_cache_alloc(gfp_mask); 716 if (!page) 717 return NULL; 718 /* 719 * We want a regular kernel memory (not highmem or DMA etc) 720 * allocation for the radix tree nodes, but we need to honour 721 * the context-specific requirements the caller has asked for. 722 * GFP_RECLAIM_MASK collects those requirements. 723 */ 724 err = add_to_page_cache_lru(page, mapping, index, 725 (gfp_mask & GFP_RECLAIM_MASK)); 726 if (unlikely(err)) { 727 page_cache_release(page); 728 page = NULL; 729 if (err == -EEXIST) 730 goto repeat; 731 } 732 } 733 return page; 734} 735EXPORT_SYMBOL(find_or_create_page); 736 737/** 738 * find_get_pages - gang pagecache lookup 739 * @mapping: The address_space to search 740 * @start: The starting page index 741 * @nr_pages: The maximum number of pages 742 * @pages: Where the resulting pages are placed 743 * 744 * find_get_pages() will search for and return a group of up to 745 * @nr_pages pages in the mapping. The pages are placed at @pages. 746 * find_get_pages() takes a reference against the returned pages. 747 * 748 * The search returns a group of mapping-contiguous pages with ascending 749 * indexes. There may be holes in the indices due to not-present pages. 750 * 751 * find_get_pages() returns the number of pages which were found. 752 */ 753unsigned find_get_pages(struct address_space *mapping, pgoff_t start, 754 unsigned int nr_pages, struct page **pages) 755{ 756 unsigned int i; 757 unsigned int ret; 758 unsigned int nr_found; 759 760 rcu_read_lock(); 761restart: 762 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree, 763 (void ***)pages, start, nr_pages); 764 ret = 0; 765 for (i = 0; i < nr_found; i++) { 766 struct page *page; 767repeat: 768 page = radix_tree_deref_slot((void **)pages[i]); 769 if (unlikely(!page)) 770 continue; 771 if (radix_tree_deref_retry(page)) { 772 if (ret) 773 start = pages[ret-1]->index; 774 goto restart; 775 } 776 777 if (!page_cache_get_speculative(page)) 778 goto repeat; 779 780 /* Has the page moved? */ 781 if (unlikely(page != *((void **)pages[i]))) { 782 page_cache_release(page); 783 goto repeat; 784 } 785 786 pages[ret] = page; 787 ret++; 788 } 789 rcu_read_unlock(); 790 return ret; 791} 792 793/** 794 * find_get_pages_contig - gang contiguous pagecache lookup 795 * @mapping: The address_space to search 796 * @index: The starting page index 797 * @nr_pages: The maximum number of pages 798 * @pages: Where the resulting pages are placed 799 * 800 * find_get_pages_contig() works exactly like find_get_pages(), except 801 * that the returned number of pages are guaranteed to be contiguous. 802 * 803 * find_get_pages_contig() returns the number of pages which were found. 804 */ 805unsigned BCMFASTPATH_HOST find_get_pages_contig(struct address_space *mapping, pgoff_t index, 806 unsigned int nr_pages, struct page **pages) 807{ 808 unsigned int i; 809 unsigned int ret; 810 unsigned int nr_found; 811 812 rcu_read_lock(); 813restart: 814 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree, 815 (void ***)pages, index, nr_pages); 816 ret = 0; 817 for (i = 0; i < nr_found; i++) { 818 struct page *page; 819repeat: 820 page = radix_tree_deref_slot((void **)pages[i]); 821 if (unlikely(!page)) 822 continue; 823 if (radix_tree_deref_retry(page)) 824 goto restart; 825 826 if (page->mapping == NULL || page->index != index) 827 break; 828 829 if (!page_cache_get_speculative(page)) 830 goto repeat; 831 832 /* Has the page moved? */ 833 if (unlikely(page != *((void **)pages[i]))) { 834 page_cache_release(page); 835 goto repeat; 836 } 837 838 pages[ret] = page; 839 ret++; 840 index++; 841 } 842 rcu_read_unlock(); 843 return ret; 844} 845EXPORT_SYMBOL(find_get_pages_contig); 846 847/** 848 * find_get_pages_tag - find and return pages that match @tag 849 * @mapping: the address_space to search 850 * @index: the starting page index 851 * @tag: the tag index 852 * @nr_pages: the maximum number of pages 853 * @pages: where the resulting pages are placed 854 * 855 * Like find_get_pages, except we only return pages which are tagged with 856 * @tag. We update @index to index the next page for the traversal. 857 */ 858unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, 859 int tag, unsigned int nr_pages, struct page **pages) 860{ 861 unsigned int i; 862 unsigned int ret; 863 unsigned int nr_found; 864 865 rcu_read_lock(); 866restart: 867 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree, 868 (void ***)pages, *index, nr_pages, tag); 869 ret = 0; 870 for (i = 0; i < nr_found; i++) { 871 struct page *page; 872repeat: 873 page = radix_tree_deref_slot((void **)pages[i]); 874 if (unlikely(!page)) 875 continue; 876 if (radix_tree_deref_retry(page)) 877 goto restart; 878 879 if (!page_cache_get_speculative(page)) 880 goto repeat; 881 882 /* Has the page moved? */ 883 if (unlikely(page != *((void **)pages[i]))) { 884 page_cache_release(page); 885 goto repeat; 886 } 887 888 pages[ret] = page; 889 ret++; 890 } 891 rcu_read_unlock(); 892 893 if (ret) 894 *index = pages[ret - 1]->index + 1; 895 896 return ret; 897} 898EXPORT_SYMBOL(find_get_pages_tag); 899 900/** 901 * grab_cache_page_nowait - returns locked page at given index in given cache 902 * @mapping: target address_space 903 * @index: the page index 904 * 905 * Same as grab_cache_page(), but do not wait if the page is unavailable. 906 * This is intended for speculative data generators, where the data can 907 * be regenerated if the page couldn't be grabbed. This routine should 908 * be safe to call while holding the lock for another page. 909 * 910 * Clear __GFP_FS when allocating the page to avoid recursion into the fs 911 * and deadlock against the caller's locked page. 912 */ 913struct page * 914grab_cache_page_nowait(struct address_space *mapping, pgoff_t index) 915{ 916 struct page *page = find_get_page(mapping, index); 917 918 if (page) { 919 if (trylock_page(page)) 920 return page; 921 page_cache_release(page); 922 return NULL; 923 } 924 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS); 925 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) { 926 page_cache_release(page); 927 page = NULL; 928 } 929 return page; 930} 931EXPORT_SYMBOL(grab_cache_page_nowait); 932 933/* 934 * CD/DVDs are error prone. When a medium error occurs, the driver may fail 935 * a _large_ part of the i/o request. Imagine the worst scenario: 936 * 937 * ---R__________________________________________B__________ 938 * ^ reading here ^ bad block(assume 4k) 939 * 940 * read(R) => miss => readahead(R...B) => media error => frustrating retries 941 * => failing the whole request => read(R) => read(R+1) => 942 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => 943 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => 944 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... 945 * 946 * It is going insane. Fix it by quickly scaling down the readahead size. 947 */ 948static void shrink_readahead_size_eio(struct file *filp, 949 struct file_ra_state *ra) 950{ 951 ra->ra_pages /= 4; 952} 953 954/** 955 * do_generic_file_read - generic file read routine 956 * @filp: the file to read 957 * @ppos: current file position 958 * @desc: read_descriptor 959 * @actor: read method 960 * 961 * This is a generic file read routine, and uses the 962 * mapping->a_ops->readpage() function for the actual low-level stuff. 963 * 964 * This is really ugly. But the goto's actually try to clarify some 965 * of the logic when it comes to error handling etc. 966 */ 967static void do_generic_file_read(struct file *filp, loff_t *ppos, 968 read_descriptor_t *desc, read_actor_t actor) 969{ 970 struct address_space *mapping = filp->f_mapping; 971 struct inode *inode = mapping->host; 972 struct file_ra_state *ra = &filp->f_ra; 973 pgoff_t index; 974 pgoff_t last_index; 975 pgoff_t prev_index; 976 unsigned long offset; /* offset into pagecache page */ 977 unsigned int prev_offset; 978 int error; 979 980 index = *ppos >> PAGE_CACHE_SHIFT; 981 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT; 982 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1); 983 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; 984 offset = *ppos & ~PAGE_CACHE_MASK; 985 986 for (;;) { 987 struct page *page; 988 pgoff_t end_index; 989 loff_t isize; 990 unsigned long nr, ret; 991 992 cond_resched(); 993find_page: 994 page = find_get_page(mapping, index); 995 if (!page) { 996 page_cache_sync_readahead(mapping, 997 ra, filp, 998 index, last_index - index); 999 page = find_get_page(mapping, index); 1000 if (unlikely(page == NULL)) 1001 goto no_cached_page; 1002 } 1003 if (PageReadahead(page)) { 1004 page_cache_async_readahead(mapping, 1005 ra, filp, page, 1006 index, last_index - index); 1007 } 1008 if (!PageUptodate(page)) { 1009 if (inode->i_blkbits == PAGE_CACHE_SHIFT || 1010 !mapping->a_ops->is_partially_uptodate) 1011 goto page_not_up_to_date; 1012 if (!trylock_page(page)) 1013 goto page_not_up_to_date; 1014 /* Did it get truncated before we got the lock? */ 1015 if (!page->mapping) 1016 goto page_not_up_to_date_locked; 1017 if (!mapping->a_ops->is_partially_uptodate(page, 1018 desc, offset)) 1019 goto page_not_up_to_date_locked; 1020 unlock_page(page); 1021 } 1022page_ok: 1023 /* 1024 * i_size must be checked after we know the page is Uptodate. 1025 * 1026 * Checking i_size after the check allows us to calculate 1027 * the correct value for "nr", which means the zero-filled 1028 * part of the page is not copied back to userspace (unless 1029 * another truncate extends the file - this is desired though). 1030 */ 1031 1032 isize = i_size_read(inode); 1033 end_index = (isize - 1) >> PAGE_CACHE_SHIFT; 1034 if (unlikely(!isize || index > end_index)) { 1035 page_cache_release(page); 1036 goto out; 1037 } 1038 1039 /* nr is the maximum number of bytes to copy from this page */ 1040 nr = PAGE_CACHE_SIZE; 1041 if (index == end_index) { 1042 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; 1043 if (nr <= offset) { 1044 page_cache_release(page); 1045 goto out; 1046 } 1047 } 1048 nr = nr - offset; 1049 1050 /* If users can be writing to this page using arbitrary 1051 * virtual addresses, take care about potential aliasing 1052 * before reading the page on the kernel side. 1053 */ 1054 if (mapping_writably_mapped(mapping)) 1055 flush_dcache_page(page); 1056 1057 /* 1058 * When a sequential read accesses a page several times, 1059 * only mark it as accessed the first time. 1060 */ 1061 if (prev_index != index || offset != prev_offset) 1062 mark_page_accessed(page); 1063 prev_index = index; 1064 1065 /* 1066 * Ok, we have the page, and it's up-to-date, so 1067 * now we can copy it to user space... 1068 * 1069 * The actor routine returns how many bytes were actually used.. 1070 * NOTE! This may not be the same as how much of a user buffer 1071 * we filled up (we may be padding etc), so we can only update 1072 * "pos" here (the actor routine has to update the user buffer 1073 * pointers and the remaining count). 1074 */ 1075 ret = actor(desc, page, offset, nr); 1076 offset += ret; 1077 index += offset >> PAGE_CACHE_SHIFT; 1078 offset &= ~PAGE_CACHE_MASK; 1079 prev_offset = offset; 1080 1081 page_cache_release(page); 1082 if (ret == nr && desc->count) 1083 continue; 1084 goto out; 1085 1086page_not_up_to_date: 1087 /* Get exclusive access to the page ... */ 1088 error = lock_page_killable(page); 1089 if (unlikely(error)) 1090 goto readpage_error; 1091 1092page_not_up_to_date_locked: 1093 /* Did it get truncated before we got the lock? */ 1094 if (!page->mapping) { 1095 unlock_page(page); 1096 page_cache_release(page); 1097 continue; 1098 } 1099 1100 /* Did somebody else fill it already? */ 1101 if (PageUptodate(page)) { 1102 unlock_page(page); 1103 goto page_ok; 1104 } 1105 1106readpage: 1107 /* 1108 * A previous I/O error may have been due to temporary 1109 * failures, eg. multipath errors. 1110 * PG_error will be set again if readpage fails. 1111 */ 1112 ClearPageError(page); 1113 /* Start the actual read. The read will unlock the page. */ 1114 error = mapping->a_ops->readpage(filp, page); 1115 1116 if (unlikely(error)) { 1117 if (error == AOP_TRUNCATED_PAGE) { 1118 page_cache_release(page); 1119 goto find_page; 1120 } 1121 goto readpage_error; 1122 } 1123 1124 if (!PageUptodate(page)) { 1125 error = lock_page_killable(page); 1126 if (unlikely(error)) 1127 goto readpage_error; 1128 if (!PageUptodate(page)) { 1129 if (page->mapping == NULL) { 1130 /* 1131 * invalidate_mapping_pages got it 1132 */ 1133 unlock_page(page); 1134 page_cache_release(page); 1135 goto find_page; 1136 } 1137 unlock_page(page); 1138 shrink_readahead_size_eio(filp, ra); 1139 error = -EIO; 1140 goto readpage_error; 1141 } 1142 unlock_page(page); 1143 } 1144 1145 goto page_ok; 1146 1147readpage_error: 1148 /* UHHUH! A synchronous read error occurred. Report it */ 1149 desc->error = error; 1150 page_cache_release(page); 1151 goto out; 1152 1153no_cached_page: 1154 /* 1155 * Ok, it wasn't cached, so we need to create a new 1156 * page.. 1157 */ 1158 page = page_cache_alloc_cold(mapping); 1159 if (!page) { 1160 desc->error = -ENOMEM; 1161 goto out; 1162 } 1163 error = add_to_page_cache_lru(page, mapping, 1164 index, GFP_KERNEL); 1165 if (error) { 1166 page_cache_release(page); 1167 if (error == -EEXIST) 1168 goto find_page; 1169 desc->error = error; 1170 goto out; 1171 } 1172 goto readpage; 1173 } 1174 1175out: 1176 ra->prev_pos = prev_index; 1177 ra->prev_pos <<= PAGE_CACHE_SHIFT; 1178 ra->prev_pos |= prev_offset; 1179 1180 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset; 1181 file_accessed(filp); 1182} 1183 1184int file_read_actor(read_descriptor_t *desc, struct page *page, 1185 unsigned long offset, unsigned long size) 1186{ 1187 char *kaddr; 1188 unsigned long left, count = desc->count; 1189 1190 if (size > count) 1191 size = count; 1192 1193 /* 1194 * Faults on the destination of a read are common, so do it before 1195 * taking the kmap. 1196 */ 1197 if (!fault_in_pages_writeable(desc->arg.buf, size)) { 1198 kaddr = kmap_atomic(page, KM_USER0); 1199 left = __copy_to_user_inatomic(desc->arg.buf, 1200 kaddr + offset, size); 1201 kunmap_atomic(kaddr, KM_USER0); 1202 if (left == 0) 1203 goto success; 1204 } 1205 1206 /* Do it the slow way */ 1207 kaddr = kmap(page); 1208 left = __copy_to_user(desc->arg.buf, kaddr + offset, size); 1209 kunmap(page); 1210 1211 if (left) { 1212 size -= left; 1213 desc->error = -EFAULT; 1214 } 1215success: 1216 desc->count = count - size; 1217 desc->written += size; 1218 desc->arg.buf += size; 1219 return size; 1220} 1221 1222/* 1223 * Performs necessary checks before doing a write 1224 * @iov: io vector request 1225 * @nr_segs: number of segments in the iovec 1226 * @count: number of bytes to write 1227 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE 1228 * 1229 * Adjust number of segments and amount of bytes to write (nr_segs should be 1230 * properly initialized first). Returns appropriate error code that caller 1231 * should return or zero in case that write should be allowed. 1232 */ 1233int generic_segment_checks(const struct iovec *iov, 1234 unsigned long *nr_segs, size_t *count, int access_flags) 1235{ 1236 unsigned long seg; 1237 size_t cnt = 0; 1238 for (seg = 0; seg < *nr_segs; seg++) { 1239 const struct iovec *iv = &iov[seg]; 1240 1241 /* 1242 * If any segment has a negative length, or the cumulative 1243 * length ever wraps negative then return -EINVAL. 1244 */ 1245 cnt += iv->iov_len; 1246 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0)) 1247 return -EINVAL; 1248 if (access_ok(access_flags, iv->iov_base, iv->iov_len)) 1249 continue; 1250 if (seg == 0) 1251 return -EFAULT; 1252 *nr_segs = seg; 1253 cnt -= iv->iov_len; /* This segment is no good */ 1254 break; 1255 } 1256 *count = cnt; 1257 return 0; 1258} 1259EXPORT_SYMBOL(generic_segment_checks); 1260 1261/** 1262 * generic_file_aio_read - generic filesystem read routine 1263 * @iocb: kernel I/O control block 1264 * @iov: io vector request 1265 * @nr_segs: number of segments in the iovec 1266 * @pos: current file position 1267 * 1268 * This is the "read()" routine for all filesystems 1269 * that can use the page cache directly. 1270 */ 1271ssize_t 1272generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov, 1273 unsigned long nr_segs, loff_t pos) 1274{ 1275 struct file *filp = iocb->ki_filp; 1276 ssize_t retval; 1277 unsigned long seg = 0; 1278 size_t count; 1279 loff_t *ppos = &iocb->ki_pos; 1280 1281 count = 0; 1282 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); 1283 if (retval) 1284 return retval; 1285 1286 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ 1287 if (filp->f_flags & O_DIRECT) { 1288 loff_t size; 1289 struct address_space *mapping; 1290 struct inode *inode; 1291 1292 mapping = filp->f_mapping; 1293 inode = mapping->host; 1294 if (!count) 1295 goto out; /* skip atime */ 1296 size = i_size_read(inode); 1297 if (pos < size) { 1298 retval = filemap_write_and_wait_range(mapping, pos, 1299 pos + iov_length(iov, nr_segs) - 1); 1300 if (!retval) { 1301 retval = mapping->a_ops->direct_IO(READ, iocb, 1302 iov, pos, nr_segs); 1303 } 1304 if (retval > 0) { 1305 *ppos = pos + retval; 1306 count -= retval; 1307 } 1308 1309 /* 1310 * Btrfs can have a short DIO read if we encounter 1311 * compressed extents, so if there was an error, or if 1312 * we've already read everything we wanted to, or if 1313 * there was a short read because we hit EOF, go ahead 1314 * and return. Otherwise fallthrough to buffered io for 1315 * the rest of the read. 1316 */ 1317 if (retval < 0 || !count || *ppos >= size) { 1318 file_accessed(filp); 1319 goto out; 1320 } 1321 } 1322 } 1323 1324 count = retval; 1325 for (seg = 0; seg < nr_segs; seg++) { 1326 read_descriptor_t desc; 1327 loff_t offset = 0; 1328 1329 /* 1330 * If we did a short DIO read we need to skip the section of the 1331 * iov that we've already read data into. 1332 */ 1333 if (count) { 1334 if (count > iov[seg].iov_len) { 1335 count -= iov[seg].iov_len; 1336 continue; 1337 } 1338 offset = count; 1339 count = 0; 1340 } 1341 1342 desc.written = 0; 1343 desc.arg.buf = iov[seg].iov_base + offset; 1344 desc.count = iov[seg].iov_len - offset; 1345 if (desc.count == 0) 1346 continue; 1347 desc.error = 0; 1348 do_generic_file_read(filp, ppos, &desc, file_read_actor); 1349 retval += desc.written; 1350 if (desc.error) { 1351 retval = retval ?: desc.error; 1352 break; 1353 } 1354 if (desc.count > 0) 1355 break; 1356 } 1357out: 1358 return retval; 1359} 1360EXPORT_SYMBOL(generic_file_aio_read); 1361 1362static ssize_t 1363do_readahead(struct address_space *mapping, struct file *filp, 1364 pgoff_t index, unsigned long nr) 1365{ 1366 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage) 1367 return -EINVAL; 1368 1369 force_page_cache_readahead(mapping, filp, index, nr); 1370 return 0; 1371} 1372 1373SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count) 1374{ 1375 ssize_t ret; 1376 struct file *file; 1377 1378 ret = -EBADF; 1379 file = fget(fd); 1380 if (file) { 1381 if (file->f_mode & FMODE_READ) { 1382 struct address_space *mapping = file->f_mapping; 1383 pgoff_t start = offset >> PAGE_CACHE_SHIFT; 1384 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT; 1385 unsigned long len = end - start + 1; 1386 ret = do_readahead(mapping, file, start, len); 1387 } 1388 fput(file); 1389 } 1390 return ret; 1391} 1392#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS 1393asmlinkage long SyS_readahead(long fd, loff_t offset, long count) 1394{ 1395 return SYSC_readahead((int) fd, offset, (size_t) count); 1396} 1397SYSCALL_ALIAS(sys_readahead, SyS_readahead); 1398#endif 1399 1400#ifdef CONFIG_MMU 1401/** 1402 * page_cache_read - adds requested page to the page cache if not already there 1403 * @file: file to read 1404 * @offset: page index 1405 * 1406 * This adds the requested page to the page cache if it isn't already there, 1407 * and schedules an I/O to read in its contents from disk. 1408 */ 1409static int page_cache_read(struct file *file, pgoff_t offset) 1410{ 1411 struct address_space *mapping = file->f_mapping; 1412 struct page *page; 1413 int ret; 1414 1415 do { 1416 page = page_cache_alloc_cold(mapping); 1417 if (!page) 1418 return -ENOMEM; 1419 1420 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL); 1421 if (ret == 0) 1422 ret = mapping->a_ops->readpage(file, page); 1423 else if (ret == -EEXIST) 1424 ret = 0; /* losing race to add is OK */ 1425 1426 page_cache_release(page); 1427 1428 } while (ret == AOP_TRUNCATED_PAGE); 1429 1430 return ret; 1431} 1432 1433#define MMAP_LOTSAMISS (100) 1434 1435/* 1436 * Synchronous readahead happens when we don't even find 1437 * a page in the page cache at all. 1438 */ 1439static void do_sync_mmap_readahead(struct vm_area_struct *vma, 1440 struct file_ra_state *ra, 1441 struct file *file, 1442 pgoff_t offset) 1443{ 1444 unsigned long ra_pages; 1445 struct address_space *mapping = file->f_mapping; 1446 1447 /* If we don't want any read-ahead, don't bother */ 1448 if (VM_RandomReadHint(vma)) 1449 return; 1450 1451 if (VM_SequentialReadHint(vma) || 1452 offset - 1 == (ra->prev_pos >> PAGE_CACHE_SHIFT)) { 1453 page_cache_sync_readahead(mapping, ra, file, offset, 1454 ra->ra_pages); 1455 return; 1456 } 1457 1458 if (ra->mmap_miss < INT_MAX) 1459 ra->mmap_miss++; 1460 1461 /* 1462 * Do we miss much more than hit in this file? If so, 1463 * stop bothering with read-ahead. It will only hurt. 1464 */ 1465 if (ra->mmap_miss > MMAP_LOTSAMISS) 1466 return; 1467 1468 /* 1469 * mmap read-around 1470 */ 1471 ra_pages = max_sane_readahead(ra->ra_pages); 1472 if (ra_pages) { 1473 ra->start = max_t(long, 0, offset - ra_pages/2); 1474 ra->size = ra_pages; 1475 ra->async_size = 0; 1476 ra_submit(ra, mapping, file); 1477 } 1478} 1479 1480/* 1481 * Asynchronous readahead happens when we find the page and PG_readahead, 1482 * so we want to possibly extend the readahead further.. 1483 */ 1484static void do_async_mmap_readahead(struct vm_area_struct *vma, 1485 struct file_ra_state *ra, 1486 struct file *file, 1487 struct page *page, 1488 pgoff_t offset) 1489{ 1490 struct address_space *mapping = file->f_mapping; 1491 1492 /* If we don't want any read-ahead, don't bother */ 1493 if (VM_RandomReadHint(vma)) 1494 return; 1495 if (ra->mmap_miss > 0) 1496 ra->mmap_miss--; 1497 if (PageReadahead(page)) 1498 page_cache_async_readahead(mapping, ra, file, 1499 page, offset, ra->ra_pages); 1500} 1501 1502/** 1503 * filemap_fault - read in file data for page fault handling 1504 * @vma: vma in which the fault was taken 1505 * @vmf: struct vm_fault containing details of the fault 1506 * 1507 * filemap_fault() is invoked via the vma operations vector for a 1508 * mapped memory region to read in file data during a page fault. 1509 * 1510 * The goto's are kind of ugly, but this streamlines the normal case of having 1511 * it in the page cache, and handles the special cases reasonably without 1512 * having a lot of duplicated code. 1513 */ 1514int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1515{ 1516 int error; 1517 struct file *file = vma->vm_file; 1518 struct address_space *mapping = file->f_mapping; 1519 struct file_ra_state *ra = &file->f_ra; 1520 struct inode *inode = mapping->host; 1521 pgoff_t offset = vmf->pgoff; 1522 struct page *page; 1523 pgoff_t size; 1524 int ret = 0; 1525 1526 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 1527 if (offset >= size) 1528 return VM_FAULT_SIGBUS; 1529 1530 /* 1531 * Do we have something in the page cache already? 1532 */ 1533 page = find_get_page(mapping, offset); 1534 if (likely(page)) { 1535 /* 1536 * We found the page, so try async readahead before 1537 * waiting for the lock. 1538 */ 1539 do_async_mmap_readahead(vma, ra, file, page, offset); 1540 lock_page(page); 1541 1542 /* Did it get truncated? */ 1543 if (unlikely(page->mapping != mapping)) { 1544 unlock_page(page); 1545 put_page(page); 1546 goto no_cached_page; 1547 } 1548 } else { 1549 /* No page in the page cache at all */ 1550 do_sync_mmap_readahead(vma, ra, file, offset); 1551 count_vm_event(PGMAJFAULT); 1552 ret = VM_FAULT_MAJOR; 1553retry_find: 1554 page = find_lock_page(mapping, offset); 1555 if (!page) 1556 goto no_cached_page; 1557 } 1558 1559 /* 1560 * We have a locked page in the page cache, now we need to check 1561 * that it's up-to-date. If not, it is going to be due to an error. 1562 */ 1563 if (unlikely(!PageUptodate(page))) 1564 goto page_not_uptodate; 1565 1566 /* 1567 * Found the page and have a reference on it. 1568 * We must recheck i_size under page lock. 1569 */ 1570 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 1571 if (unlikely(offset >= size)) { 1572 unlock_page(page); 1573 page_cache_release(page); 1574 return VM_FAULT_SIGBUS; 1575 } 1576 1577 ra->prev_pos = (loff_t)offset << PAGE_CACHE_SHIFT; 1578 vmf->page = page; 1579 return ret | VM_FAULT_LOCKED; 1580 1581no_cached_page: 1582 /* 1583 * We're only likely to ever get here if MADV_RANDOM is in 1584 * effect. 1585 */ 1586 error = page_cache_read(file, offset); 1587 1588 /* 1589 * The page we want has now been added to the page cache. 1590 * In the unlikely event that someone removed it in the 1591 * meantime, we'll just come back here and read it again. 1592 */ 1593 if (error >= 0) 1594 goto retry_find; 1595 1596 /* 1597 * An error return from page_cache_read can result if the 1598 * system is low on memory, or a problem occurs while trying 1599 * to schedule I/O. 1600 */ 1601 if (error == -ENOMEM) 1602 return VM_FAULT_OOM; 1603 return VM_FAULT_SIGBUS; 1604 1605page_not_uptodate: 1606 /* 1607 * Umm, take care of errors if the page isn't up-to-date. 1608 * Try to re-read it _once_. We do this synchronously, 1609 * because there really aren't any performance issues here 1610 * and we need to check for errors. 1611 */ 1612 ClearPageError(page); 1613 error = mapping->a_ops->readpage(file, page); 1614 if (!error) { 1615 wait_on_page_locked(page); 1616 if (!PageUptodate(page)) 1617 error = -EIO; 1618 } 1619 page_cache_release(page); 1620 1621 if (!error || error == AOP_TRUNCATED_PAGE) 1622 goto retry_find; 1623 1624 /* Things didn't work out. Return zero to tell the mm layer so. */ 1625 shrink_readahead_size_eio(file, ra); 1626 return VM_FAULT_SIGBUS; 1627} 1628EXPORT_SYMBOL(filemap_fault); 1629 1630const struct vm_operations_struct generic_file_vm_ops = { 1631 .fault = filemap_fault, 1632}; 1633 1634/* This is used for a general mmap of a disk file */ 1635 1636int generic_file_mmap(struct file * file, struct vm_area_struct * vma) 1637{ 1638 struct address_space *mapping = file->f_mapping; 1639 1640 if (!mapping->a_ops->readpage) 1641 return -ENOEXEC; 1642 file_accessed(file); 1643 vma->vm_ops = &generic_file_vm_ops; 1644 vma->vm_flags |= VM_CAN_NONLINEAR; 1645 return 0; 1646} 1647 1648/* 1649 * This is for filesystems which do not implement ->writepage. 1650 */ 1651int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) 1652{ 1653 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) 1654 return -EINVAL; 1655 return generic_file_mmap(file, vma); 1656} 1657#else 1658int generic_file_mmap(struct file * file, struct vm_area_struct * vma) 1659{ 1660 return -ENOSYS; 1661} 1662int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) 1663{ 1664 return -ENOSYS; 1665} 1666#endif /* CONFIG_MMU */ 1667 1668EXPORT_SYMBOL(generic_file_mmap); 1669EXPORT_SYMBOL(generic_file_readonly_mmap); 1670 1671static struct page *__read_cache_page(struct address_space *mapping, 1672 pgoff_t index, 1673 int (*filler)(void *,struct page*), 1674 void *data, 1675 gfp_t gfp) 1676{ 1677 struct page *page; 1678 int err; 1679repeat: 1680 page = find_get_page(mapping, index); 1681 if (!page) { 1682 page = __page_cache_alloc(gfp | __GFP_COLD); 1683 if (!page) 1684 return ERR_PTR(-ENOMEM); 1685 err = add_to_page_cache_lru(page, mapping, index, GFP_KERNEL); 1686 if (unlikely(err)) { 1687 page_cache_release(page); 1688 if (err == -EEXIST) 1689 goto repeat; 1690 /* Presumably ENOMEM for radix tree node */ 1691 return ERR_PTR(err); 1692 } 1693 err = filler(data, page); 1694 if (err < 0) { 1695 page_cache_release(page); 1696 page = ERR_PTR(err); 1697 } 1698 } 1699 return page; 1700} 1701 1702static struct page *do_read_cache_page(struct address_space *mapping, 1703 pgoff_t index, 1704 int (*filler)(void *,struct page*), 1705 void *data, 1706 gfp_t gfp) 1707 1708{ 1709 struct page *page; 1710 int err; 1711 1712retry: 1713 page = __read_cache_page(mapping, index, filler, data, gfp); 1714 if (IS_ERR(page)) 1715 return page; 1716 if (PageUptodate(page)) 1717 goto out; 1718 1719 lock_page(page); 1720 if (!page->mapping) { 1721 unlock_page(page); 1722 page_cache_release(page); 1723 goto retry; 1724 } 1725 if (PageUptodate(page)) { 1726 unlock_page(page); 1727 goto out; 1728 } 1729 err = filler(data, page); 1730 if (err < 0) { 1731 page_cache_release(page); 1732 return ERR_PTR(err); 1733 } 1734out: 1735 mark_page_accessed(page); 1736 return page; 1737} 1738 1739/** 1740 * read_cache_page_async - read into page cache, fill it if needed 1741 * @mapping: the page's address_space 1742 * @index: the page index 1743 * @filler: function to perform the read 1744 * @data: destination for read data 1745 * 1746 * Same as read_cache_page, but don't wait for page to become unlocked 1747 * after submitting it to the filler. 1748 * 1749 * Read into the page cache. If a page already exists, and PageUptodate() is 1750 * not set, try to fill the page but don't wait for it to become unlocked. 1751 * 1752 * If the page does not get brought uptodate, return -EIO. 1753 */ 1754struct page *read_cache_page_async(struct address_space *mapping, 1755 pgoff_t index, 1756 int (*filler)(void *,struct page*), 1757 void *data) 1758{ 1759 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); 1760} 1761EXPORT_SYMBOL(read_cache_page_async); 1762 1763static struct page *wait_on_page_read(struct page *page) 1764{ 1765 if (!IS_ERR(page)) { 1766 wait_on_page_locked(page); 1767 if (!PageUptodate(page)) { 1768 page_cache_release(page); 1769 page = ERR_PTR(-EIO); 1770 } 1771 } 1772 return page; 1773} 1774 1775/** 1776 * read_cache_page_gfp - read into page cache, using specified page allocation flags. 1777 * @mapping: the page's address_space 1778 * @index: the page index 1779 * @gfp: the page allocator flags to use if allocating 1780 * 1781 * This is the same as "read_mapping_page(mapping, index, NULL)", but with 1782 * any new page allocations done using the specified allocation flags. Note 1783 * that the Radix tree operations will still use GFP_KERNEL, so you can't 1784 * expect to do this atomically or anything like that - but you can pass in 1785 * other page requirements. 1786 * 1787 * If the page does not get brought uptodate, return -EIO. 1788 */ 1789struct page *read_cache_page_gfp(struct address_space *mapping, 1790 pgoff_t index, 1791 gfp_t gfp) 1792{ 1793 filler_t *filler = (filler_t *)mapping->a_ops->readpage; 1794 1795 return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp)); 1796} 1797EXPORT_SYMBOL(read_cache_page_gfp); 1798 1799/** 1800 * read_cache_page - read into page cache, fill it if needed 1801 * @mapping: the page's address_space 1802 * @index: the page index 1803 * @filler: function to perform the read 1804 * @data: destination for read data 1805 * 1806 * Read into the page cache. If a page already exists, and PageUptodate() is 1807 * not set, try to fill the page then wait for it to become unlocked. 1808 * 1809 * If the page does not get brought uptodate, return -EIO. 1810 */ 1811struct page *read_cache_page(struct address_space *mapping, 1812 pgoff_t index, 1813 int (*filler)(void *,struct page*), 1814 void *data) 1815{ 1816 return wait_on_page_read(read_cache_page_async(mapping, index, filler, data)); 1817} 1818EXPORT_SYMBOL(read_cache_page); 1819 1820/* 1821 * The logic we want is 1822 * 1823 * if suid or (sgid and xgrp) 1824 * remove privs 1825 */ 1826int should_remove_suid(struct dentry *dentry) 1827{ 1828 mode_t mode = dentry->d_inode->i_mode; 1829 int kill = 0; 1830 1831 /* suid always must be killed */ 1832 if (unlikely(mode & S_ISUID)) 1833 kill = ATTR_KILL_SUID; 1834 1835 /* 1836 * sgid without any exec bits is just a mandatory locking mark; leave 1837 * it alone. If some exec bits are set, it's a real sgid; kill it. 1838 */ 1839 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 1840 kill |= ATTR_KILL_SGID; 1841 1842 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode))) 1843 return kill; 1844 1845 return 0; 1846} 1847EXPORT_SYMBOL(should_remove_suid); 1848 1849static int __remove_suid(struct dentry *dentry, int kill) 1850{ 1851 struct iattr newattrs; 1852 1853 newattrs.ia_valid = ATTR_FORCE | kill; 1854 return notify_change(dentry, &newattrs); 1855} 1856 1857int file_remove_suid(struct file *file) 1858{ 1859 struct dentry *dentry = file->f_path.dentry; 1860 int killsuid = should_remove_suid(dentry); 1861 int killpriv = security_inode_need_killpriv(dentry); 1862 int error = 0; 1863 1864 if (killpriv < 0) 1865 return killpriv; 1866 if (killpriv) 1867 error = security_inode_killpriv(dentry); 1868 if (!error && killsuid) 1869 error = __remove_suid(dentry, killsuid); 1870 1871 return error; 1872} 1873EXPORT_SYMBOL(file_remove_suid); 1874 1875static size_t __iovec_copy_from_user_inatomic(char *vaddr, 1876 const struct iovec *iov, size_t base, size_t bytes) 1877{ 1878 size_t copied = 0, left = 0; 1879 1880 while (bytes) { 1881 char __user *buf = iov->iov_base + base; 1882 int copy = min(bytes, iov->iov_len - base); 1883 1884 base = 0; 1885 left = __copy_from_user_inatomic(vaddr, buf, copy); 1886 copied += copy; 1887 bytes -= copy; 1888 vaddr += copy; 1889 iov++; 1890 1891 if (unlikely(left)) 1892 break; 1893 } 1894 return copied - left; 1895} 1896 1897/* 1898 * Copy as much as we can into the page and return the number of bytes which 1899 * were successfully copied. If a fault is encountered then return the number of 1900 * bytes which were copied. 1901 */ 1902size_t iov_iter_copy_from_user_atomic(struct page *page, 1903 struct iov_iter *i, unsigned long offset, size_t bytes) 1904{ 1905 char *kaddr; 1906 size_t copied; 1907 1908 BUG_ON(!in_atomic()); 1909 kaddr = kmap_atomic(page, KM_USER0); 1910 if (likely(i->nr_segs == 1)) { 1911 int left; 1912 char __user *buf = i->iov->iov_base + i->iov_offset; 1913 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes); 1914 copied = bytes - left; 1915 } else { 1916 copied = __iovec_copy_from_user_inatomic(kaddr + offset, 1917 i->iov, i->iov_offset, bytes); 1918 } 1919 kunmap_atomic(kaddr, KM_USER0); 1920 1921 return copied; 1922} 1923EXPORT_SYMBOL(iov_iter_copy_from_user_atomic); 1924 1925/* 1926 * This has the same sideeffects and return value as 1927 * iov_iter_copy_from_user_atomic(). 1928 * The difference is that it attempts to resolve faults. 1929 * Page must not be locked. 1930 */ 1931size_t iov_iter_copy_from_user(struct page *page, 1932 struct iov_iter *i, unsigned long offset, size_t bytes) 1933{ 1934 char *kaddr; 1935 size_t copied; 1936 1937 kaddr = kmap(page); 1938 if (likely(i->nr_segs == 1)) { 1939 int left; 1940 char __user *buf = i->iov->iov_base + i->iov_offset; 1941 left = __copy_from_user(kaddr + offset, buf, bytes); 1942 copied = bytes - left; 1943 } else { 1944 copied = __iovec_copy_from_user_inatomic(kaddr + offset, 1945 i->iov, i->iov_offset, bytes); 1946 } 1947 kunmap(page); 1948 return copied; 1949} 1950EXPORT_SYMBOL(iov_iter_copy_from_user); 1951 1952void iov_iter_advance(struct iov_iter *i, size_t bytes) 1953{ 1954 BUG_ON(i->count < bytes); 1955 1956 if (likely(i->nr_segs == 1)) { 1957 i->iov_offset += bytes; 1958 i->count -= bytes; 1959 } else { 1960 const struct iovec *iov = i->iov; 1961 size_t base = i->iov_offset; 1962 1963 /* 1964 * The !iov->iov_len check ensures we skip over unlikely 1965 * zero-length segments (without overruning the iovec). 1966 */ 1967 while (bytes || unlikely(i->count && !iov->iov_len)) { 1968 int copy; 1969 1970 copy = min(bytes, iov->iov_len - base); 1971 BUG_ON(!i->count || i->count < copy); 1972 i->count -= copy; 1973 bytes -= copy; 1974 base += copy; 1975 if (iov->iov_len == base) { 1976 iov++; 1977 base = 0; 1978 } 1979 } 1980 i->iov = iov; 1981 i->iov_offset = base; 1982 } 1983} 1984EXPORT_SYMBOL(iov_iter_advance); 1985 1986/* 1987 * Fault in the first iovec of the given iov_iter, to a maximum length 1988 * of bytes. Returns 0 on success, or non-zero if the memory could not be 1989 * accessed (ie. because it is an invalid address). 1990 * 1991 * writev-intensive code may want this to prefault several iovecs -- that 1992 * would be possible (callers must not rely on the fact that _only_ the 1993 * first iovec will be faulted with the current implementation). 1994 */ 1995int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes) 1996{ 1997 char __user *buf = i->iov->iov_base + i->iov_offset; 1998 bytes = min(bytes, i->iov->iov_len - i->iov_offset); 1999 return fault_in_pages_readable(buf, bytes); 2000} 2001EXPORT_SYMBOL(iov_iter_fault_in_readable); 2002 2003/* 2004 * Return the count of just the current iov_iter segment. 2005 */ 2006size_t iov_iter_single_seg_count(struct iov_iter *i) 2007{ 2008 const struct iovec *iov = i->iov; 2009 if (i->nr_segs == 1) 2010 return i->count; 2011 else 2012 return min(i->count, iov->iov_len - i->iov_offset); 2013} 2014EXPORT_SYMBOL(iov_iter_single_seg_count); 2015 2016/* 2017 * Performs necessary checks before doing a write 2018 * 2019 * Can adjust writing position or amount of bytes to write. 2020 * Returns appropriate error code that caller should return or 2021 * zero in case that write should be allowed. 2022 */ 2023inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk) 2024{ 2025 struct inode *inode = file->f_mapping->host; 2026 unsigned long limit = rlimit(RLIMIT_FSIZE); 2027 2028 if (unlikely(*pos < 0)) 2029 return -EINVAL; 2030 2031 if (!isblk) { 2032 if (file->f_flags & O_APPEND) 2033 *pos = i_size_read(inode); 2034 /* Foxconn modified start pling 12/04/2009 */ 2035 /* Remove large file limitation */ 2036#if (!defined SAMBA_ENABLE) 2037 if (limit != RLIM_INFINITY) { 2038 if (*pos >= limit) { 2039 send_sig(SIGXFSZ, current, 0); 2040 return -EFBIG; 2041 } 2042 if (*count > limit - (typeof(limit))*pos) { 2043 *count = limit - (typeof(limit))*pos; 2044 } 2045 } 2046#endif 2047 /* Foxconn modified end pling 12/04/2009 */ 2048 } 2049 2050 /* Foxconn modified start pling 12/04/2009 */ 2051 /* Ignore LFS rule to support large files */ 2052#if (!defined SAMBA_ENABLE) 2053 /* 2054 * LFS rule 2055 */ 2056 if (unlikely(*pos + *count > MAX_NON_LFS && 2057 !(file->f_flags & O_LARGEFILE))) { 2058 if (*pos >= MAX_NON_LFS) { 2059 return -EFBIG; 2060 } 2061 if (*count > MAX_NON_LFS - (unsigned long)*pos) { 2062 *count = MAX_NON_LFS - (unsigned long)*pos; 2063 } 2064 } 2065#endif 2066 /* Foxconn modified end pling 12/04/2009 */ 2067 2068 /* 2069 * Are we about to exceed the fs block limit ? 2070 * 2071 * If we have written data it becomes a short write. If we have 2072 * exceeded without writing data we send a signal and return EFBIG. 2073 * Linus frestrict idea will clean these up nicely.. 2074 */ 2075 if (likely(!isblk)) { 2076 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) { 2077 if (*count || *pos > inode->i_sb->s_maxbytes) { 2078 return -EFBIG; 2079 } 2080 /* zero-length writes at ->s_maxbytes are OK */ 2081 } 2082 2083 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes)) 2084 *count = inode->i_sb->s_maxbytes - *pos; 2085 } else { 2086#ifdef CONFIG_BLOCK 2087 loff_t isize; 2088 if (bdev_read_only(I_BDEV(inode))) 2089 return -EPERM; 2090 isize = i_size_read(inode); 2091 if (*pos >= isize) { 2092 if (*count || *pos > isize) 2093 return -ENOSPC; 2094 } 2095 2096 if (*pos + *count > isize) 2097 *count = isize - *pos; 2098#else 2099 return -EPERM; 2100#endif 2101 } 2102 return 0; 2103} 2104EXPORT_SYMBOL(generic_write_checks); 2105 2106int pagecache_write_begin(struct file *file, struct address_space *mapping, 2107 loff_t pos, unsigned len, unsigned flags, 2108 struct page **pagep, void **fsdata) 2109{ 2110 const struct address_space_operations *aops = mapping->a_ops; 2111 2112 return aops->write_begin(file, mapping, pos, len, flags, 2113 pagep, fsdata); 2114} 2115EXPORT_SYMBOL(pagecache_write_begin); 2116 2117int pagecache_write_end(struct file *file, struct address_space *mapping, 2118 loff_t pos, unsigned len, unsigned copied, 2119 struct page *page, void *fsdata) 2120{ 2121 const struct address_space_operations *aops = mapping->a_ops; 2122 2123 mark_page_accessed(page); 2124 return aops->write_end(file, mapping, pos, len, copied, page, fsdata); 2125} 2126EXPORT_SYMBOL(pagecache_write_end); 2127 2128ssize_t 2129generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov, 2130 unsigned long *nr_segs, loff_t pos, loff_t *ppos, 2131 size_t count, size_t ocount) 2132{ 2133 struct file *file = iocb->ki_filp; 2134 struct address_space *mapping = file->f_mapping; 2135 struct inode *inode = mapping->host; 2136 ssize_t written; 2137 size_t write_len; 2138 pgoff_t end; 2139 2140 if (count != ocount) 2141 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count); 2142 2143 write_len = iov_length(iov, *nr_segs); 2144 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT; 2145 2146 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1); 2147 if (written) 2148 goto out; 2149 2150 /* 2151 * After a write we want buffered reads to be sure to go to disk to get 2152 * the new data. We invalidate clean cached page from the region we're 2153 * about to write. We do this *before* the write so that we can return 2154 * without clobbering -EIOCBQUEUED from ->direct_IO(). 2155 */ 2156 if (mapping->nrpages) { 2157 written = invalidate_inode_pages2_range(mapping, 2158 pos >> PAGE_CACHE_SHIFT, end); 2159 /* 2160 * If a page can not be invalidated, return 0 to fall back 2161 * to buffered write. 2162 */ 2163 if (written) { 2164 if (written == -EBUSY) 2165 return 0; 2166 goto out; 2167 } 2168 } 2169 2170 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs); 2171 2172 /* 2173 * Finally, try again to invalidate clean pages which might have been 2174 * cached by non-direct readahead, or faulted in by get_user_pages() 2175 * if the source of the write was an mmap'ed region of the file 2176 * we're writing. Either one is a pretty crazy thing to do, 2177 * so we don't support it 100%. If this invalidation 2178 * fails, tough, the write still worked... 2179 */ 2180 if (mapping->nrpages) { 2181 invalidate_inode_pages2_range(mapping, 2182 pos >> PAGE_CACHE_SHIFT, end); 2183 } 2184 2185 if (written > 0) { 2186 loff_t end = pos + written; 2187 if (end > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { 2188 i_size_write(inode, end); 2189 mark_inode_dirty(inode); 2190 } 2191 *ppos = end; 2192 } 2193out: 2194 return written; 2195} 2196EXPORT_SYMBOL(generic_file_direct_write); 2197 2198/* 2199 * Find or create a page at the given pagecache position. Return the locked 2200 * page. This function is specifically for buffered writes. 2201 */ 2202struct page *grab_cache_page_write_begin(struct address_space *mapping, 2203 pgoff_t index, unsigned flags) 2204{ 2205 int status; 2206 struct page *page; 2207 gfp_t gfp_notmask = 0; 2208 if (flags & AOP_FLAG_NOFS) 2209 gfp_notmask = __GFP_FS; 2210repeat: 2211 page = find_lock_page(mapping, index); 2212 if (likely(page)) 2213 return page; 2214 2215 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask); 2216 if (!page) 2217 return NULL; 2218 status = add_to_page_cache_lru(page, mapping, index, 2219 GFP_KERNEL & ~gfp_notmask); 2220 if (unlikely(status)) { 2221 page_cache_release(page); 2222 if (status == -EEXIST) 2223 goto repeat; 2224 return NULL; 2225 } 2226 return page; 2227} 2228EXPORT_SYMBOL(grab_cache_page_write_begin); 2229 2230static ssize_t generic_perform_write(struct file *file, 2231 struct iov_iter *i, loff_t pos) 2232{ 2233 struct address_space *mapping = file->f_mapping; 2234 const struct address_space_operations *a_ops = mapping->a_ops; 2235 long status = 0; 2236 ssize_t written = 0; 2237 unsigned int flags = 0; 2238 2239 /* 2240 * Copies from kernel address space cannot fail (NFSD is a big user). 2241 */ 2242 if (segment_eq(get_fs(), KERNEL_DS)) 2243 flags |= AOP_FLAG_UNINTERRUPTIBLE; 2244 2245 do { 2246 struct page *page; 2247 unsigned long offset; /* Offset into pagecache page */ 2248 unsigned long bytes; /* Bytes to write to page */ 2249 size_t copied; /* Bytes copied from user */ 2250 void *fsdata; 2251 2252 offset = (pos & (PAGE_CACHE_SIZE - 1)); 2253 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, 2254 iov_iter_count(i)); 2255 2256again: 2257 2258 /* 2259 * Bring in the user page that we will copy from _first_. 2260 * Otherwise there's a nasty deadlock on copying from the 2261 * same page as we're writing to, without it being marked 2262 * up-to-date. 2263 * 2264 * Not only is this an optimisation, but it is also required 2265 * to check that the address is actually valid, when atomic 2266 * usercopies are used, below. 2267 */ 2268 if (unlikely(iov_iter_fault_in_readable(i, bytes))) { 2269 status = -EFAULT; 2270 break; 2271 } 2272 2273 status = a_ops->write_begin(file, mapping, pos, bytes, flags, 2274 &page, &fsdata); 2275 if (unlikely(status)) 2276 break; 2277 2278 if (mapping_writably_mapped(mapping)) 2279 flush_dcache_page(page); 2280 2281 pagefault_disable(); 2282 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); 2283 pagefault_enable(); 2284 flush_dcache_page(page); 2285 2286 mark_page_accessed(page); 2287 status = a_ops->write_end(file, mapping, pos, bytes, copied, 2288 page, fsdata); 2289 if (unlikely(status < 0)) 2290 break; 2291 copied = status; 2292 2293 cond_resched(); 2294 2295 iov_iter_advance(i, copied); 2296 if (unlikely(copied == 0)) { 2297 /* 2298 * If we were unable to copy any data at all, we must 2299 * fall back to a single segment length write. 2300 * 2301 * If we didn't fallback here, we could livelock 2302 * because not all segments in the iov can be copied at 2303 * once without a pagefault. 2304 */ 2305 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, 2306 iov_iter_single_seg_count(i)); 2307 goto again; 2308 } 2309 pos += copied; 2310 written += copied; 2311 2312 balance_dirty_pages_ratelimited(mapping); 2313 2314 } while (iov_iter_count(i)); 2315 2316 return written ? written : status; 2317} 2318 2319ssize_t 2320generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov, 2321 unsigned long nr_segs, loff_t pos, loff_t *ppos, 2322 size_t count, ssize_t written) 2323{ 2324 struct file *file = iocb->ki_filp; 2325 ssize_t status; 2326 struct iov_iter i; 2327 2328 iov_iter_init(&i, iov, nr_segs, count, written); 2329 status = generic_perform_write(file, &i, pos); 2330 2331 if (likely(status >= 0)) { 2332 written += status; 2333 *ppos = pos + status; 2334 } 2335 2336 return written ? written : status; 2337} 2338EXPORT_SYMBOL(generic_file_buffered_write); 2339 2340/** 2341 * __generic_file_aio_write - write data to a file 2342 * @iocb: IO state structure (file, offset, etc.) 2343 * @iov: vector with data to write 2344 * @nr_segs: number of segments in the vector 2345 * @ppos: position where to write 2346 * 2347 * This function does all the work needed for actually writing data to a 2348 * file. It does all basic checks, removes SUID from the file, updates 2349 * modification times and calls proper subroutines depending on whether we 2350 * do direct IO or a standard buffered write. 2351 * 2352 * It expects i_mutex to be grabbed unless we work on a block device or similar 2353 * object which does not need locking at all. 2354 * 2355 * This function does *not* take care of syncing data in case of O_SYNC write. 2356 * A caller has to handle it. This is mainly due to the fact that we want to 2357 * avoid syncing under i_mutex. 2358 */ 2359ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, 2360 unsigned long nr_segs, loff_t *ppos) 2361{ 2362 struct file *file = iocb->ki_filp; 2363 struct address_space * mapping = file->f_mapping; 2364 size_t ocount; /* original count */ 2365 size_t count; /* after file limit checks */ 2366 struct inode *inode = mapping->host; 2367 loff_t pos; 2368 ssize_t written; 2369 ssize_t err; 2370 2371 ocount = 0; 2372 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); 2373 if (err) 2374 return err; 2375 2376 count = ocount; 2377 pos = *ppos; 2378 2379 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); 2380 2381 /* We can write back this queue in page reclaim */ 2382 current->backing_dev_info = mapping->backing_dev_info; 2383 written = 0; 2384 2385 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); 2386 if (err) 2387 goto out; 2388 2389 if (count == 0) 2390 goto out; 2391 2392 err = file_remove_suid(file); 2393 if (err) 2394 goto out; 2395 2396 file_update_time(file); 2397 2398 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ 2399 if (unlikely(file->f_flags & O_DIRECT)) { 2400 loff_t endbyte; 2401 ssize_t written_buffered; 2402 2403 written = generic_file_direct_write(iocb, iov, &nr_segs, pos, 2404 ppos, count, ocount); 2405 if (written < 0 || written == count) 2406 goto out; 2407 /* 2408 * direct-io write to a hole: fall through to buffered I/O 2409 * for completing the rest of the request. 2410 */ 2411 pos += written; 2412 count -= written; 2413 written_buffered = generic_file_buffered_write(iocb, iov, 2414 nr_segs, pos, ppos, count, 2415 written); 2416 /* 2417 * If generic_file_buffered_write() retuned a synchronous error 2418 * then we want to return the number of bytes which were 2419 * direct-written, or the error code if that was zero. Note 2420 * that this differs from normal direct-io semantics, which 2421 * will return -EFOO even if some bytes were written. 2422 */ 2423 if (written_buffered < 0) { 2424 err = written_buffered; 2425 goto out; 2426 } 2427 2428 /* 2429 * We need to ensure that the page cache pages are written to 2430 * disk and invalidated to preserve the expected O_DIRECT 2431 * semantics. 2432 */ 2433 endbyte = pos + written_buffered - written - 1; 2434 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte); 2435 if (err == 0) { 2436 written = written_buffered; 2437 invalidate_mapping_pages(mapping, 2438 pos >> PAGE_CACHE_SHIFT, 2439 endbyte >> PAGE_CACHE_SHIFT); 2440 } else { 2441 /* 2442 * We don't know how much we wrote, so just return 2443 * the number of bytes which were direct-written 2444 */ 2445 } 2446 } else { 2447 written = generic_file_buffered_write(iocb, iov, nr_segs, 2448 pos, ppos, count, written); 2449 } 2450out: 2451 current->backing_dev_info = NULL; 2452 return written ? written : err; 2453} 2454EXPORT_SYMBOL(__generic_file_aio_write); 2455 2456/** 2457 * generic_file_aio_write - write data to a file 2458 * @iocb: IO state structure 2459 * @iov: vector with data to write 2460 * @nr_segs: number of segments in the vector 2461 * @pos: position in file where to write 2462 * 2463 * This is a wrapper around __generic_file_aio_write() to be used by most 2464 * filesystems. It takes care of syncing the file in case of O_SYNC file 2465 * and acquires i_mutex as needed. 2466 */ 2467ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, 2468 unsigned long nr_segs, loff_t pos) 2469{ 2470 struct file *file = iocb->ki_filp; 2471 struct inode *inode = file->f_mapping->host; 2472 ssize_t ret; 2473 2474 BUG_ON(iocb->ki_pos != pos); 2475 2476 mutex_lock(&inode->i_mutex); 2477 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos); 2478 mutex_unlock(&inode->i_mutex); 2479 2480 if (ret > 0 || ret == -EIOCBQUEUED) { 2481 ssize_t err; 2482 2483 err = generic_write_sync(file, pos, ret); 2484 if (err < 0 && ret > 0) 2485 ret = err; 2486 } 2487 return ret; 2488} 2489EXPORT_SYMBOL(generic_file_aio_write); 2490 2491/** 2492 * try_to_release_page() - release old fs-specific metadata on a page 2493 * 2494 * @page: the page which the kernel is trying to free 2495 * @gfp_mask: memory allocation flags (and I/O mode) 2496 * 2497 * The address_space is to try to release any data against the page 2498 * (presumably at page->private). If the release was successful, return `1'. 2499 * Otherwise return zero. 2500 * 2501 * This may also be called if PG_fscache is set on a page, indicating that the 2502 * page is known to the local caching routines. 2503 * 2504 * The @gfp_mask argument specifies whether I/O may be performed to release 2505 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS). 2506 * 2507 */ 2508int try_to_release_page(struct page *page, gfp_t gfp_mask) 2509{ 2510 struct address_space * const mapping = page->mapping; 2511 2512 BUG_ON(!PageLocked(page)); 2513 if (PageWriteback(page)) 2514 return 0; 2515 2516 if (mapping && mapping->a_ops->releasepage) 2517 return mapping->a_ops->releasepage(page, gfp_mask); 2518 return try_to_free_buffers(page); 2519} 2520 2521EXPORT_SYMBOL(try_to_release_page); 2522