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/compiler.h> 14#include <linux/fs.h> 15#include <linux/uaccess.h> 16#include <linux/aio.h> 17#include <linux/capability.h> 18#include <linux/kernel_stat.h> 19#include <linux/gfp.h> 20#include <linux/mm.h> 21#include <linux/swap.h> 22#include <linux/mman.h> 23#include <linux/pagemap.h> 24#include <linux/file.h> 25#include <linux/uio.h> 26#include <linux/hash.h> 27#include <linux/writeback.h> 28#include <linux/backing-dev.h> 29#include <linux/pagevec.h> 30#include <linux/blkdev.h> 31#include <linux/security.h> 32#include <linux/syscalls.h> 33#include <linux/cpuset.h> 34#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */ 35#include <linux/memcontrol.h> 36#include <linux/mm_inline.h> /* for page_is_file_cache() */ 37#include <linux/cleancache.h> 38#include "internal.h" 39 40/* 41 * FIXME: remove all knowledge of the buffer layer from the core VM 42 */ 43#include <linux/buffer_head.h> /* for try_to_free_buffers */ 44 45#include <asm/mman.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_mutex (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_mutex (truncate->unmap_mapping_range) 69 * 70 * ->mmap_sem 71 * ->i_mmap_mutex 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_wb_list_lock 85 * sb_lock (fs/fs-writeback.c) 86 * ->mapping->tree_lock (__sync_single_inode) 87 * 88 * ->i_mmap_mutex 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_wb_list_lock (page_remove_rmap->set_page_dirty) 103 * ->inode->i_lock (page_remove_rmap->set_page_dirty) 104 * inode_wb_list_lock (zap_pte_range->set_page_dirty) 105 * ->inode->i_lock (zap_pte_range->set_page_dirty) 106 * ->private_lock (zap_pte_range->__set_page_dirty_buffers) 107 * 108 * (code doesn't rely on that order, so you could switch it around) 109 * ->tasklist_lock (memory_failure, collect_procs_ao) 110 * ->i_mmap_mutex 111 */ 112 113/* 114 * Delete a page from the page cache and free it. Caller has to make 115 * sure the page is locked and that nobody else uses it - or that usage 116 * is safe. The caller must hold the mapping's tree_lock. 117 */ 118void __delete_from_page_cache(struct page *page) 119{ 120 struct address_space *mapping = page->mapping; 121 122 /* 123 * if we're uptodate, flush out into the cleancache, otherwise 124 * invalidate any existing cleancache entries. We can't leave 125 * stale data around in the cleancache once our page is gone 126 */ 127 if (PageUptodate(page) && PageMappedToDisk(page)) 128 cleancache_put_page(page); 129 else 130 cleancache_flush_page(mapping, page); 131 132 radix_tree_delete(&mapping->page_tree, page->index); 133 page->mapping = NULL; 134 mapping->nrpages--; 135 __dec_zone_page_state(page, NR_FILE_PAGES); 136 if (PageSwapBacked(page)) 137 __dec_zone_page_state(page, NR_SHMEM); 138 BUG_ON(page_mapped(page)); 139 140 /* 141 * Some filesystems seem to re-dirty the page even after 142 * the VM has canceled the dirty bit (eg ext3 journaling). 143 * 144 * Fix it up by doing a final dirty accounting check after 145 * having removed the page entirely. 146 */ 147 if (PageDirty(page) && mapping_cap_account_dirty(mapping)) { 148 dec_zone_page_state(page, NR_FILE_DIRTY); 149 dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE); 150 } 151} 152 153/** 154 * delete_from_page_cache - delete page from page cache 155 * @page: the page which the kernel is trying to remove from page cache 156 * 157 * This must be called only on pages that have been verified to be in the page 158 * cache and locked. It will never put the page into the free list, the caller 159 * has a reference on the page. 160 */ 161void delete_from_page_cache(struct page *page) 162{ 163 struct address_space *mapping = page->mapping; 164 void (*freepage)(struct page *); 165 166 BUG_ON(!PageLocked(page)); 167 168 freepage = mapping->a_ops->freepage; 169 spin_lock_irq(&mapping->tree_lock); 170 __delete_from_page_cache(page); 171 spin_unlock_irq(&mapping->tree_lock); 172 mem_cgroup_uncharge_cache_page(page); 173 174 if (freepage) 175 freepage(page); 176 page_cache_release(page); 177} 178EXPORT_SYMBOL(delete_from_page_cache); 179 180static int sleep_on_page(void *word) 181{ 182 io_schedule(); 183 return 0; 184} 185 186static int sleep_on_page_killable(void *word) 187{ 188 sleep_on_page(word); 189 return fatal_signal_pending(current) ? -EINTR : 0; 190} 191 192/** 193 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range 194 * @mapping: address space structure to write 195 * @start: offset in bytes where the range starts 196 * @end: offset in bytes where the range ends (inclusive) 197 * @sync_mode: enable synchronous operation 198 * 199 * Start writeback against all of a mapping's dirty pages that lie 200 * within the byte offsets <start, end> inclusive. 201 * 202 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as 203 * opposed to a regular memory cleansing writeback. The difference between 204 * these two operations is that if a dirty page/buffer is encountered, it must 205 * be waited upon, and not just skipped over. 206 */ 207int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start, 208 loff_t end, int sync_mode) 209{ 210 int ret; 211 struct writeback_control wbc = { 212 .sync_mode = sync_mode, 213 .nr_to_write = LONG_MAX, 214 .range_start = start, 215 .range_end = end, 216 }; 217 218 if (!mapping_cap_writeback_dirty(mapping)) 219 return 0; 220 221 ret = do_writepages(mapping, &wbc); 222 return ret; 223} 224 225static inline int __filemap_fdatawrite(struct address_space *mapping, 226 int sync_mode) 227{ 228 return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode); 229} 230 231int filemap_fdatawrite(struct address_space *mapping) 232{ 233 return __filemap_fdatawrite(mapping, WB_SYNC_ALL); 234} 235EXPORT_SYMBOL(filemap_fdatawrite); 236 237int filemap_fdatawrite_range(struct address_space *mapping, loff_t start, 238 loff_t end) 239{ 240 return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL); 241} 242EXPORT_SYMBOL(filemap_fdatawrite_range); 243 244/** 245 * filemap_flush - mostly a non-blocking flush 246 * @mapping: target address_space 247 * 248 * This is a mostly non-blocking flush. Not suitable for data-integrity 249 * purposes - I/O may not be started against all dirty pages. 250 */ 251int filemap_flush(struct address_space *mapping) 252{ 253 return __filemap_fdatawrite(mapping, WB_SYNC_NONE); 254} 255EXPORT_SYMBOL(filemap_flush); 256 257/** 258 * filemap_fdatawait_range - wait for writeback to complete 259 * @mapping: address space structure to wait for 260 * @start_byte: offset in bytes where the range starts 261 * @end_byte: offset in bytes where the range ends (inclusive) 262 * 263 * Walk the list of under-writeback pages of the given address space 264 * in the given range and wait for all of them. 265 */ 266int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte, 267 loff_t end_byte) 268{ 269 pgoff_t index = start_byte >> PAGE_CACHE_SHIFT; 270 pgoff_t end = end_byte >> PAGE_CACHE_SHIFT; 271 struct pagevec pvec; 272 int nr_pages; 273 int ret = 0; 274 275 if (end_byte < start_byte) 276 return 0; 277 278 pagevec_init(&pvec, 0); 279 while ((index <= end) && 280 (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, 281 PAGECACHE_TAG_WRITEBACK, 282 min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) { 283 unsigned i; 284 285 for (i = 0; i < nr_pages; i++) { 286 struct page *page = pvec.pages[i]; 287 288 /* until radix tree lookup accepts end_index */ 289 if (page->index > end) 290 continue; 291 292 wait_on_page_writeback(page); 293 if (TestClearPageError(page)) 294 ret = -EIO; 295 } 296 pagevec_release(&pvec); 297 cond_resched(); 298 } 299 300 /* Check for outstanding write errors */ 301 if (test_and_clear_bit(AS_ENOSPC, &mapping->flags)) 302 ret = -ENOSPC; 303 if (test_and_clear_bit(AS_EIO, &mapping->flags)) 304 ret = -EIO; 305 306 return ret; 307} 308EXPORT_SYMBOL(filemap_fdatawait_range); 309 310/** 311 * filemap_fdatawait - wait for all under-writeback pages to complete 312 * @mapping: address space structure to wait for 313 * 314 * Walk the list of under-writeback pages of the given address space 315 * and wait for all of them. 316 */ 317int filemap_fdatawait(struct address_space *mapping) 318{ 319 loff_t i_size = i_size_read(mapping->host); 320 321 if (i_size == 0) 322 return 0; 323 324 return filemap_fdatawait_range(mapping, 0, i_size - 1); 325} 326EXPORT_SYMBOL(filemap_fdatawait); 327 328int filemap_write_and_wait(struct address_space *mapping) 329{ 330 int err = 0; 331 332 if (mapping->nrpages) { 333 err = filemap_fdatawrite(mapping); 334 /* 335 * Even if the above returned error, the pages may be 336 * written partially (e.g. -ENOSPC), so we wait for it. 337 * But the -EIO is special case, it may indicate the worst 338 * thing (e.g. bug) happened, so we avoid waiting for it. 339 */ 340 if (err != -EIO) { 341 int err2 = filemap_fdatawait(mapping); 342 if (!err) 343 err = err2; 344 } 345 } 346 return err; 347} 348EXPORT_SYMBOL(filemap_write_and_wait); 349 350/** 351 * filemap_write_and_wait_range - write out & wait on a file range 352 * @mapping: the address_space for the pages 353 * @lstart: offset in bytes where the range starts 354 * @lend: offset in bytes where the range ends (inclusive) 355 * 356 * Write out and wait upon file offsets lstart->lend, inclusive. 357 * 358 * Note that `lend' is inclusive (describes the last byte to be written) so 359 * that this function can be used to write to the very end-of-file (end = -1). 360 */ 361int filemap_write_and_wait_range(struct address_space *mapping, 362 loff_t lstart, loff_t lend) 363{ 364 int err = 0; 365 366 if (mapping->nrpages) { 367 err = __filemap_fdatawrite_range(mapping, lstart, lend, 368 WB_SYNC_ALL); 369 /* See comment of filemap_write_and_wait() */ 370 if (err != -EIO) { 371 int err2 = filemap_fdatawait_range(mapping, 372 lstart, lend); 373 if (!err) 374 err = err2; 375 } 376 } 377 return err; 378} 379EXPORT_SYMBOL(filemap_write_and_wait_range); 380 381/** 382 * replace_page_cache_page - replace a pagecache page with a new one 383 * @old: page to be replaced 384 * @new: page to replace with 385 * @gfp_mask: allocation mode 386 * 387 * This function replaces a page in the pagecache with a new one. On 388 * success it acquires the pagecache reference for the new page and 389 * drops it for the old page. Both the old and new pages must be 390 * locked. This function does not add the new page to the LRU, the 391 * caller must do that. 392 * 393 * The remove + add is atomic. The only way this function can fail is 394 * memory allocation failure. 395 */ 396int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask) 397{ 398 int error; 399 400 VM_BUG_ON(!PageLocked(old)); 401 VM_BUG_ON(!PageLocked(new)); 402 VM_BUG_ON(new->mapping); 403 404 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); 405 if (!error) { 406 struct address_space *mapping = old->mapping; 407 void (*freepage)(struct page *); 408 409 pgoff_t offset = old->index; 410 freepage = mapping->a_ops->freepage; 411 412 page_cache_get(new); 413 new->mapping = mapping; 414 new->index = offset; 415 416 spin_lock_irq(&mapping->tree_lock); 417 __delete_from_page_cache(old); 418 error = radix_tree_insert(&mapping->page_tree, offset, new); 419 BUG_ON(error); 420 mapping->nrpages++; 421 __inc_zone_page_state(new, NR_FILE_PAGES); 422 if (PageSwapBacked(new)) 423 __inc_zone_page_state(new, NR_SHMEM); 424 spin_unlock_irq(&mapping->tree_lock); 425 /* mem_cgroup codes must not be called under tree_lock */ 426 mem_cgroup_replace_page_cache(old, new); 427 radix_tree_preload_end(); 428 if (freepage) 429 freepage(old); 430 page_cache_release(old); 431 } 432 433 return error; 434} 435EXPORT_SYMBOL_GPL(replace_page_cache_page); 436 437/** 438 * add_to_page_cache_locked - add a locked page to the pagecache 439 * @page: page to add 440 * @mapping: the page's address_space 441 * @offset: page index 442 * @gfp_mask: page allocation mode 443 * 444 * This function is used to add a page to the pagecache. It must be locked. 445 * This function does not add the page to the LRU. The caller must do that. 446 */ 447int add_to_page_cache_locked(struct page *page, struct address_space *mapping, 448 pgoff_t offset, gfp_t gfp_mask) 449{ 450 int error; 451 452 VM_BUG_ON(!PageLocked(page)); 453 454 error = mem_cgroup_cache_charge(page, current->mm, 455 gfp_mask & GFP_RECLAIM_MASK); 456 if (error) 457 goto out; 458 459 error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM); 460 if (error == 0) { 461 page_cache_get(page); 462 page->mapping = mapping; 463 page->index = offset; 464 465 spin_lock_irq(&mapping->tree_lock); 466 error = radix_tree_insert(&mapping->page_tree, offset, page); 467 if (likely(!error)) { 468 mapping->nrpages++; 469 __inc_zone_page_state(page, NR_FILE_PAGES); 470 if (PageSwapBacked(page)) 471 __inc_zone_page_state(page, NR_SHMEM); 472 spin_unlock_irq(&mapping->tree_lock); 473 } else { 474 page->mapping = NULL; 475 spin_unlock_irq(&mapping->tree_lock); 476 mem_cgroup_uncharge_cache_page(page); 477 page_cache_release(page); 478 } 479 radix_tree_preload_end(); 480 } else 481 mem_cgroup_uncharge_cache_page(page); 482out: 483 return error; 484} 485EXPORT_SYMBOL(add_to_page_cache_locked); 486 487int add_to_page_cache_lru(struct page *page, struct address_space *mapping, 488 pgoff_t offset, gfp_t gfp_mask) 489{ 490 int ret; 491 492 /* 493 * Splice_read and readahead add shmem/tmpfs pages into the page cache 494 * before shmem_readpage has a chance to mark them as SwapBacked: they 495 * need to go on the anon lru below, and mem_cgroup_cache_charge 496 * (called in add_to_page_cache) needs to know where they're going too. 497 */ 498 if (mapping_cap_swap_backed(mapping)) 499 SetPageSwapBacked(page); 500 501 ret = add_to_page_cache(page, mapping, offset, gfp_mask); 502 if (ret == 0) { 503 if (page_is_file_cache(page)) 504 lru_cache_add_file(page); 505 else 506 lru_cache_add_anon(page); 507 } 508 return ret; 509} 510EXPORT_SYMBOL_GPL(add_to_page_cache_lru); 511 512#ifdef CONFIG_NUMA 513struct page *__page_cache_alloc(gfp_t gfp) 514{ 515 int n; 516 struct page *page; 517 518 if (cpuset_do_page_mem_spread()) { 519 get_mems_allowed(); 520 n = cpuset_mem_spread_node(); 521 page = alloc_pages_exact_node(n, gfp, 0); 522 put_mems_allowed(); 523 return page; 524 } 525 return alloc_pages(gfp, 0); 526} 527EXPORT_SYMBOL(__page_cache_alloc); 528#endif 529 530/* 531 * In order to wait for pages to become available there must be 532 * waitqueues associated with pages. By using a hash table of 533 * waitqueues where the bucket discipline is to maintain all 534 * waiters on the same queue and wake all when any of the pages 535 * become available, and for the woken contexts to check to be 536 * sure the appropriate page became available, this saves space 537 * at a cost of "thundering herd" phenomena during rare hash 538 * collisions. 539 */ 540static wait_queue_head_t *page_waitqueue(struct page *page) 541{ 542 const struct zone *zone = page_zone(page); 543 544 return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)]; 545} 546 547static inline void wake_up_page(struct page *page, int bit) 548{ 549 __wake_up_bit(page_waitqueue(page), &page->flags, bit); 550} 551 552void wait_on_page_bit(struct page *page, int bit_nr) 553{ 554 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); 555 556 if (test_bit(bit_nr, &page->flags)) 557 __wait_on_bit(page_waitqueue(page), &wait, sleep_on_page, 558 TASK_UNINTERRUPTIBLE); 559} 560EXPORT_SYMBOL(wait_on_page_bit); 561 562int wait_on_page_bit_killable(struct page *page, int bit_nr) 563{ 564 DEFINE_WAIT_BIT(wait, &page->flags, bit_nr); 565 566 if (!test_bit(bit_nr, &page->flags)) 567 return 0; 568 569 return __wait_on_bit(page_waitqueue(page), &wait, 570 sleep_on_page_killable, TASK_KILLABLE); 571} 572 573/** 574 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue 575 * @page: Page defining the wait queue of interest 576 * @waiter: Waiter to add to the queue 577 * 578 * Add an arbitrary @waiter to the wait queue for the nominated @page. 579 */ 580void add_page_wait_queue(struct page *page, wait_queue_t *waiter) 581{ 582 wait_queue_head_t *q = page_waitqueue(page); 583 unsigned long flags; 584 585 spin_lock_irqsave(&q->lock, flags); 586 __add_wait_queue(q, waiter); 587 spin_unlock_irqrestore(&q->lock, flags); 588} 589EXPORT_SYMBOL_GPL(add_page_wait_queue); 590 591/** 592 * unlock_page - unlock a locked page 593 * @page: the page 594 * 595 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked(). 596 * Also wakes sleepers in wait_on_page_writeback() because the wakeup 597 * mechananism between PageLocked pages and PageWriteback pages is shared. 598 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep. 599 * 600 * The mb is necessary to enforce ordering between the clear_bit and the read 601 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()). 602 */ 603void unlock_page(struct page *page) 604{ 605 VM_BUG_ON(!PageLocked(page)); 606 clear_bit_unlock(PG_locked, &page->flags); 607 smp_mb__after_clear_bit(); 608 wake_up_page(page, PG_locked); 609} 610EXPORT_SYMBOL(unlock_page); 611 612/** 613 * end_page_writeback - end writeback against a page 614 * @page: the page 615 */ 616void end_page_writeback(struct page *page) 617{ 618 if (TestClearPageReclaim(page)) 619 rotate_reclaimable_page(page); 620 621 if (!test_clear_page_writeback(page)) 622 BUG(); 623 624 smp_mb__after_clear_bit(); 625 wake_up_page(page, PG_writeback); 626} 627EXPORT_SYMBOL(end_page_writeback); 628 629/** 630 * __lock_page - get a lock on the page, assuming we need to sleep to get it 631 * @page: the page to lock 632 */ 633void __lock_page(struct page *page) 634{ 635 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); 636 637 __wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page, 638 TASK_UNINTERRUPTIBLE); 639} 640EXPORT_SYMBOL(__lock_page); 641 642int __lock_page_killable(struct page *page) 643{ 644 DEFINE_WAIT_BIT(wait, &page->flags, PG_locked); 645 646 return __wait_on_bit_lock(page_waitqueue(page), &wait, 647 sleep_on_page_killable, TASK_KILLABLE); 648} 649EXPORT_SYMBOL_GPL(__lock_page_killable); 650 651int __lock_page_or_retry(struct page *page, struct mm_struct *mm, 652 unsigned int flags) 653{ 654 if (flags & FAULT_FLAG_ALLOW_RETRY) { 655 /* 656 * CAUTION! In this case, mmap_sem is not released 657 * even though return 0. 658 */ 659 if (flags & FAULT_FLAG_RETRY_NOWAIT) 660 return 0; 661 662 up_read(&mm->mmap_sem); 663 if (flags & FAULT_FLAG_KILLABLE) 664 wait_on_page_locked_killable(page); 665 else 666 wait_on_page_locked(page); 667 return 0; 668 } else { 669 if (flags & FAULT_FLAG_KILLABLE) { 670 int ret; 671 672 ret = __lock_page_killable(page); 673 if (ret) { 674 up_read(&mm->mmap_sem); 675 return 0; 676 } 677 } else 678 __lock_page(page); 679 return 1; 680 } 681} 682 683/** 684 * find_get_page - find and get a page reference 685 * @mapping: the address_space to search 686 * @offset: the page index 687 * 688 * Is there a pagecache struct page at the given (mapping, offset) tuple? 689 * If yes, increment its refcount and return it; if no, return NULL. 690 */ 691struct page *find_get_page(struct address_space *mapping, pgoff_t offset) 692{ 693 void **pagep; 694 struct page *page; 695 696 rcu_read_lock(); 697repeat: 698 page = NULL; 699 pagep = radix_tree_lookup_slot(&mapping->page_tree, offset); 700 if (pagep) { 701 page = radix_tree_deref_slot(pagep); 702 if (unlikely(!page)) 703 goto out; 704 if (radix_tree_deref_retry(page)) 705 goto repeat; 706 707 if (!page_cache_get_speculative(page)) 708 goto repeat; 709 710 /* 711 * Has the page moved? 712 * This is part of the lockless pagecache protocol. See 713 * include/linux/pagemap.h for details. 714 */ 715 if (unlikely(page != *pagep)) { 716 page_cache_release(page); 717 goto repeat; 718 } 719 } 720out: 721 rcu_read_unlock(); 722 723 return page; 724} 725EXPORT_SYMBOL(find_get_page); 726 727/** 728 * find_lock_page - locate, pin and lock a pagecache page 729 * @mapping: the address_space to search 730 * @offset: the page index 731 * 732 * Locates the desired pagecache page, locks it, increments its reference 733 * count and returns its address. 734 * 735 * Returns zero if the page was not present. find_lock_page() may sleep. 736 */ 737struct page *find_lock_page(struct address_space *mapping, pgoff_t offset) 738{ 739 struct page *page; 740 741repeat: 742 page = find_get_page(mapping, offset); 743 if (page) { 744 lock_page(page); 745 /* Has the page been truncated? */ 746 if (unlikely(page->mapping != mapping)) { 747 unlock_page(page); 748 page_cache_release(page); 749 goto repeat; 750 } 751 VM_BUG_ON(page->index != offset); 752 } 753 return page; 754} 755EXPORT_SYMBOL(find_lock_page); 756 757/** 758 * find_or_create_page - locate or add a pagecache page 759 * @mapping: the page's address_space 760 * @index: the page's index into the mapping 761 * @gfp_mask: page allocation mode 762 * 763 * Locates a page in the pagecache. If the page is not present, a new page 764 * is allocated using @gfp_mask and is added to the pagecache and to the VM's 765 * LRU list. The returned page is locked and has its reference count 766 * incremented. 767 * 768 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic 769 * allocation! 770 * 771 * find_or_create_page() returns the desired page's address, or zero on 772 * memory exhaustion. 773 */ 774struct page *find_or_create_page(struct address_space *mapping, 775 pgoff_t index, gfp_t gfp_mask) 776{ 777 struct page *page; 778 int err; 779repeat: 780 page = find_lock_page(mapping, index); 781 if (!page) { 782 page = __page_cache_alloc(gfp_mask); 783 if (!page) 784 return NULL; 785 /* 786 * We want a regular kernel memory (not highmem or DMA etc) 787 * allocation for the radix tree nodes, but we need to honour 788 * the context-specific requirements the caller has asked for. 789 * GFP_RECLAIM_MASK collects those requirements. 790 */ 791 err = add_to_page_cache_lru(page, mapping, index, 792 (gfp_mask & GFP_RECLAIM_MASK)); 793 if (unlikely(err)) { 794 page_cache_release(page); 795 page = NULL; 796 if (err == -EEXIST) 797 goto repeat; 798 } 799 } 800 return page; 801} 802EXPORT_SYMBOL(find_or_create_page); 803 804/** 805 * find_get_pages - gang pagecache lookup 806 * @mapping: The address_space to search 807 * @start: The starting page index 808 * @nr_pages: The maximum number of pages 809 * @pages: Where the resulting pages are placed 810 * 811 * find_get_pages() will search for and return a group of up to 812 * @nr_pages pages in the mapping. The pages are placed at @pages. 813 * find_get_pages() takes a reference against the returned pages. 814 * 815 * The search returns a group of mapping-contiguous pages with ascending 816 * indexes. There may be holes in the indices due to not-present pages. 817 * 818 * find_get_pages() returns the number of pages which were found. 819 */ 820unsigned find_get_pages(struct address_space *mapping, pgoff_t start, 821 unsigned int nr_pages, struct page **pages) 822{ 823 unsigned int i; 824 unsigned int ret; 825 unsigned int nr_found; 826 827 rcu_read_lock(); 828restart: 829 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree, 830 (void ***)pages, start, nr_pages); 831 ret = 0; 832 for (i = 0; i < nr_found; i++) { 833 struct page *page; 834repeat: 835 page = radix_tree_deref_slot((void **)pages[i]); 836 if (unlikely(!page)) 837 continue; 838 839 /* 840 * This can only trigger when the entry at index 0 moves out 841 * of or back to the root: none yet gotten, safe to restart. 842 */ 843 if (radix_tree_deref_retry(page)) { 844 WARN_ON(start | i); 845 goto restart; 846 } 847 848 if (!page_cache_get_speculative(page)) 849 goto repeat; 850 851 /* Has the page moved? */ 852 if (unlikely(page != *((void **)pages[i]))) { 853 page_cache_release(page); 854 goto repeat; 855 } 856 857 pages[ret] = page; 858 ret++; 859 } 860 861 /* 862 * If all entries were removed before we could secure them, 863 * try again, because callers stop trying once 0 is returned. 864 */ 865 if (unlikely(!ret && nr_found)) 866 goto restart; 867 rcu_read_unlock(); 868 return ret; 869} 870 871/** 872 * find_get_pages_contig - gang contiguous pagecache lookup 873 * @mapping: The address_space to search 874 * @index: The starting page index 875 * @nr_pages: The maximum number of pages 876 * @pages: Where the resulting pages are placed 877 * 878 * find_get_pages_contig() works exactly like find_get_pages(), except 879 * that the returned number of pages are guaranteed to be contiguous. 880 * 881 * find_get_pages_contig() returns the number of pages which were found. 882 */ 883unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index, 884 unsigned int nr_pages, struct page **pages) 885{ 886 unsigned int i; 887 unsigned int ret; 888 unsigned int nr_found; 889 890 rcu_read_lock(); 891restart: 892 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree, 893 (void ***)pages, index, nr_pages); 894 ret = 0; 895 for (i = 0; i < nr_found; i++) { 896 struct page *page; 897repeat: 898 page = radix_tree_deref_slot((void **)pages[i]); 899 if (unlikely(!page)) 900 continue; 901 902 /* 903 * This can only trigger when the entry at index 0 moves out 904 * of or back to the root: none yet gotten, safe to restart. 905 */ 906 if (radix_tree_deref_retry(page)) 907 goto restart; 908 909 if (!page_cache_get_speculative(page)) 910 goto repeat; 911 912 /* Has the page moved? */ 913 if (unlikely(page != *((void **)pages[i]))) { 914 page_cache_release(page); 915 goto repeat; 916 } 917 918 /* 919 * must check mapping and index after taking the ref. 920 * otherwise we can get both false positives and false 921 * negatives, which is just confusing to the caller. 922 */ 923 if (page->mapping == NULL || page->index != index) { 924 page_cache_release(page); 925 break; 926 } 927 928 pages[ret] = page; 929 ret++; 930 index++; 931 } 932 rcu_read_unlock(); 933 return ret; 934} 935EXPORT_SYMBOL(find_get_pages_contig); 936 937/** 938 * find_get_pages_tag - find and return pages that match @tag 939 * @mapping: the address_space to search 940 * @index: the starting page index 941 * @tag: the tag index 942 * @nr_pages: the maximum number of pages 943 * @pages: where the resulting pages are placed 944 * 945 * Like find_get_pages, except we only return pages which are tagged with 946 * @tag. We update @index to index the next page for the traversal. 947 */ 948unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index, 949 int tag, unsigned int nr_pages, struct page **pages) 950{ 951 unsigned int i; 952 unsigned int ret; 953 unsigned int nr_found; 954 955 rcu_read_lock(); 956restart: 957 nr_found = radix_tree_gang_lookup_tag_slot(&mapping->page_tree, 958 (void ***)pages, *index, nr_pages, tag); 959 ret = 0; 960 for (i = 0; i < nr_found; i++) { 961 struct page *page; 962repeat: 963 page = radix_tree_deref_slot((void **)pages[i]); 964 if (unlikely(!page)) 965 continue; 966 967 /* 968 * This can only trigger when the entry at index 0 moves out 969 * of or back to the root: none yet gotten, safe to restart. 970 */ 971 if (radix_tree_deref_retry(page)) 972 goto restart; 973 974 if (!page_cache_get_speculative(page)) 975 goto repeat; 976 977 /* Has the page moved? */ 978 if (unlikely(page != *((void **)pages[i]))) { 979 page_cache_release(page); 980 goto repeat; 981 } 982 983 pages[ret] = page; 984 ret++; 985 } 986 987 /* 988 * If all entries were removed before we could secure them, 989 * try again, because callers stop trying once 0 is returned. 990 */ 991 if (unlikely(!ret && nr_found)) 992 goto restart; 993 rcu_read_unlock(); 994 995 if (ret) 996 *index = pages[ret - 1]->index + 1; 997 998 return ret; 999} 1000EXPORT_SYMBOL(find_get_pages_tag); 1001 1002/** 1003 * grab_cache_page_nowait - returns locked page at given index in given cache 1004 * @mapping: target address_space 1005 * @index: the page index 1006 * 1007 * Same as grab_cache_page(), but do not wait if the page is unavailable. 1008 * This is intended for speculative data generators, where the data can 1009 * be regenerated if the page couldn't be grabbed. This routine should 1010 * be safe to call while holding the lock for another page. 1011 * 1012 * Clear __GFP_FS when allocating the page to avoid recursion into the fs 1013 * and deadlock against the caller's locked page. 1014 */ 1015struct page * 1016grab_cache_page_nowait(struct address_space *mapping, pgoff_t index) 1017{ 1018 struct page *page = find_get_page(mapping, index); 1019 1020 if (page) { 1021 if (trylock_page(page)) 1022 return page; 1023 page_cache_release(page); 1024 return NULL; 1025 } 1026 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS); 1027 if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) { 1028 page_cache_release(page); 1029 page = NULL; 1030 } 1031 return page; 1032} 1033EXPORT_SYMBOL(grab_cache_page_nowait); 1034 1035/* 1036 * CD/DVDs are error prone. When a medium error occurs, the driver may fail 1037 * a _large_ part of the i/o request. Imagine the worst scenario: 1038 * 1039 * ---R__________________________________________B__________ 1040 * ^ reading here ^ bad block(assume 4k) 1041 * 1042 * read(R) => miss => readahead(R...B) => media error => frustrating retries 1043 * => failing the whole request => read(R) => read(R+1) => 1044 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) => 1045 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) => 1046 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ...... 1047 * 1048 * It is going insane. Fix it by quickly scaling down the readahead size. 1049 */ 1050static void shrink_readahead_size_eio(struct file *filp, 1051 struct file_ra_state *ra) 1052{ 1053 ra->ra_pages /= 4; 1054} 1055 1056/** 1057 * do_generic_file_read - generic file read routine 1058 * @filp: the file to read 1059 * @ppos: current file position 1060 * @desc: read_descriptor 1061 * @actor: read method 1062 * 1063 * This is a generic file read routine, and uses the 1064 * mapping->a_ops->readpage() function for the actual low-level stuff. 1065 * 1066 * This is really ugly. But the goto's actually try to clarify some 1067 * of the logic when it comes to error handling etc. 1068 */ 1069static void do_generic_file_read(struct file *filp, loff_t *ppos, 1070 read_descriptor_t *desc, read_actor_t actor) 1071{ 1072 struct address_space *mapping = filp->f_mapping; 1073 struct inode *inode = mapping->host; 1074 struct file_ra_state *ra = &filp->f_ra; 1075 pgoff_t index; 1076 pgoff_t last_index; 1077 pgoff_t prev_index; 1078 unsigned long offset; /* offset into pagecache page */ 1079 unsigned int prev_offset; 1080 int error; 1081 1082 index = *ppos >> PAGE_CACHE_SHIFT; 1083 prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT; 1084 prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1); 1085 last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT; 1086 offset = *ppos & ~PAGE_CACHE_MASK; 1087 1088 for (;;) { 1089 struct page *page; 1090 pgoff_t end_index; 1091 loff_t isize; 1092 unsigned long nr, ret; 1093 1094 cond_resched(); 1095find_page: 1096 page = find_get_page(mapping, index); 1097 if (!page) { 1098 page_cache_sync_readahead(mapping, 1099 ra, filp, 1100 index, last_index - index); 1101 page = find_get_page(mapping, index); 1102 if (unlikely(page == NULL)) 1103 goto no_cached_page; 1104 } 1105 if (PageReadahead(page)) { 1106 page_cache_async_readahead(mapping, 1107 ra, filp, page, 1108 index, last_index - index); 1109 } 1110 if (!PageUptodate(page)) { 1111 if (inode->i_blkbits == PAGE_CACHE_SHIFT || 1112 !mapping->a_ops->is_partially_uptodate) 1113 goto page_not_up_to_date; 1114 if (!trylock_page(page)) 1115 goto page_not_up_to_date; 1116 /* Did it get truncated before we got the lock? */ 1117 if (!page->mapping) 1118 goto page_not_up_to_date_locked; 1119 if (!mapping->a_ops->is_partially_uptodate(page, 1120 desc, offset)) 1121 goto page_not_up_to_date_locked; 1122 unlock_page(page); 1123 } 1124page_ok: 1125 /* 1126 * i_size must be checked after we know the page is Uptodate. 1127 * 1128 * Checking i_size after the check allows us to calculate 1129 * the correct value for "nr", which means the zero-filled 1130 * part of the page is not copied back to userspace (unless 1131 * another truncate extends the file - this is desired though). 1132 */ 1133 1134 isize = i_size_read(inode); 1135 end_index = (isize - 1) >> PAGE_CACHE_SHIFT; 1136 if (unlikely(!isize || index > end_index)) { 1137 page_cache_release(page); 1138 goto out; 1139 } 1140 1141 /* nr is the maximum number of bytes to copy from this page */ 1142 nr = PAGE_CACHE_SIZE; 1143 if (index == end_index) { 1144 nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; 1145 if (nr <= offset) { 1146 page_cache_release(page); 1147 goto out; 1148 } 1149 } 1150 nr = nr - offset; 1151 1152 /* If users can be writing to this page using arbitrary 1153 * virtual addresses, take care about potential aliasing 1154 * before reading the page on the kernel side. 1155 */ 1156 if (mapping_writably_mapped(mapping)) 1157 flush_dcache_page(page); 1158 1159 /* 1160 * When a sequential read accesses a page several times, 1161 * only mark it as accessed the first time. 1162 */ 1163 if (prev_index != index || offset != prev_offset) 1164 mark_page_accessed(page); 1165 prev_index = index; 1166 1167 /* 1168 * Ok, we have the page, and it's up-to-date, so 1169 * now we can copy it to user space... 1170 * 1171 * The actor routine returns how many bytes were actually used.. 1172 * NOTE! This may not be the same as how much of a user buffer 1173 * we filled up (we may be padding etc), so we can only update 1174 * "pos" here (the actor routine has to update the user buffer 1175 * pointers and the remaining count). 1176 */ 1177 ret = actor(desc, page, offset, nr); 1178 offset += ret; 1179 index += offset >> PAGE_CACHE_SHIFT; 1180 offset &= ~PAGE_CACHE_MASK; 1181 prev_offset = offset; 1182 1183 page_cache_release(page); 1184 if (ret == nr && desc->count) 1185 continue; 1186 goto out; 1187 1188page_not_up_to_date: 1189 /* Get exclusive access to the page ... */ 1190 error = lock_page_killable(page); 1191 if (unlikely(error)) 1192 goto readpage_error; 1193 1194page_not_up_to_date_locked: 1195 /* Did it get truncated before we got the lock? */ 1196 if (!page->mapping) { 1197 unlock_page(page); 1198 page_cache_release(page); 1199 continue; 1200 } 1201 1202 /* Did somebody else fill it already? */ 1203 if (PageUptodate(page)) { 1204 unlock_page(page); 1205 goto page_ok; 1206 } 1207 1208readpage: 1209 /* 1210 * A previous I/O error may have been due to temporary 1211 * failures, eg. multipath errors. 1212 * PG_error will be set again if readpage fails. 1213 */ 1214 ClearPageError(page); 1215 /* Start the actual read. The read will unlock the page. */ 1216 error = mapping->a_ops->readpage(filp, page); 1217 1218 if (unlikely(error)) { 1219 if (error == AOP_TRUNCATED_PAGE) { 1220 page_cache_release(page); 1221 goto find_page; 1222 } 1223 goto readpage_error; 1224 } 1225 1226 if (!PageUptodate(page)) { 1227 error = lock_page_killable(page); 1228 if (unlikely(error)) 1229 goto readpage_error; 1230 if (!PageUptodate(page)) { 1231 if (page->mapping == NULL) { 1232 /* 1233 * invalidate_mapping_pages got it 1234 */ 1235 unlock_page(page); 1236 page_cache_release(page); 1237 goto find_page; 1238 } 1239 unlock_page(page); 1240 shrink_readahead_size_eio(filp, ra); 1241 error = -EIO; 1242 goto readpage_error; 1243 } 1244 unlock_page(page); 1245 } 1246 1247 goto page_ok; 1248 1249readpage_error: 1250 /* UHHUH! A synchronous read error occurred. Report it */ 1251 desc->error = error; 1252 page_cache_release(page); 1253 goto out; 1254 1255no_cached_page: 1256 /* 1257 * Ok, it wasn't cached, so we need to create a new 1258 * page.. 1259 */ 1260 page = page_cache_alloc_cold(mapping); 1261 if (!page) { 1262 desc->error = -ENOMEM; 1263 goto out; 1264 } 1265 error = add_to_page_cache_lru(page, mapping, 1266 index, GFP_KERNEL); 1267 if (error) { 1268 page_cache_release(page); 1269 if (error == -EEXIST) 1270 goto find_page; 1271 desc->error = error; 1272 goto out; 1273 } 1274 goto readpage; 1275 } 1276 1277out: 1278 ra->prev_pos = prev_index; 1279 ra->prev_pos <<= PAGE_CACHE_SHIFT; 1280 ra->prev_pos |= prev_offset; 1281 1282 *ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset; 1283 file_accessed(filp); 1284} 1285 1286int file_read_actor(read_descriptor_t *desc, struct page *page, 1287 unsigned long offset, unsigned long size) 1288{ 1289 char *kaddr; 1290 unsigned long left, count = desc->count; 1291 1292 if (size > count) 1293 size = count; 1294 1295 /* 1296 * Faults on the destination of a read are common, so do it before 1297 * taking the kmap. 1298 */ 1299 if (!fault_in_pages_writeable(desc->arg.buf, size)) { 1300 kaddr = kmap_atomic(page, KM_USER0); 1301 left = __copy_to_user_inatomic(desc->arg.buf, 1302 kaddr + offset, size); 1303 kunmap_atomic(kaddr, KM_USER0); 1304 if (left == 0) 1305 goto success; 1306 } 1307 1308 /* Do it the slow way */ 1309 kaddr = kmap(page); 1310 left = __copy_to_user(desc->arg.buf, kaddr + offset, size); 1311 kunmap(page); 1312 1313 if (left) { 1314 size -= left; 1315 desc->error = -EFAULT; 1316 } 1317success: 1318 desc->count = count - size; 1319 desc->written += size; 1320 desc->arg.buf += size; 1321 return size; 1322} 1323 1324/* 1325 * Performs necessary checks before doing a write 1326 * @iov: io vector request 1327 * @nr_segs: number of segments in the iovec 1328 * @count: number of bytes to write 1329 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE 1330 * 1331 * Adjust number of segments and amount of bytes to write (nr_segs should be 1332 * properly initialized first). Returns appropriate error code that caller 1333 * should return or zero in case that write should be allowed. 1334 */ 1335int generic_segment_checks(const struct iovec *iov, 1336 unsigned long *nr_segs, size_t *count, int access_flags) 1337{ 1338 unsigned long seg; 1339 size_t cnt = 0; 1340 for (seg = 0; seg < *nr_segs; seg++) { 1341 const struct iovec *iv = &iov[seg]; 1342 1343 /* 1344 * If any segment has a negative length, or the cumulative 1345 * length ever wraps negative then return -EINVAL. 1346 */ 1347 cnt += iv->iov_len; 1348 if (unlikely((ssize_t)(cnt|iv->iov_len) < 0)) 1349 return -EINVAL; 1350 if (access_ok(access_flags, iv->iov_base, iv->iov_len)) 1351 continue; 1352 if (seg == 0) 1353 return -EFAULT; 1354 *nr_segs = seg; 1355 cnt -= iv->iov_len; /* This segment is no good */ 1356 break; 1357 } 1358 *count = cnt; 1359 return 0; 1360} 1361EXPORT_SYMBOL(generic_segment_checks); 1362 1363/** 1364 * generic_file_aio_read - generic filesystem read routine 1365 * @iocb: kernel I/O control block 1366 * @iov: io vector request 1367 * @nr_segs: number of segments in the iovec 1368 * @pos: current file position 1369 * 1370 * This is the "read()" routine for all filesystems 1371 * that can use the page cache directly. 1372 */ 1373ssize_t 1374generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov, 1375 unsigned long nr_segs, loff_t pos) 1376{ 1377 struct file *filp = iocb->ki_filp; 1378 ssize_t retval; 1379 unsigned long seg = 0; 1380 size_t count; 1381 loff_t *ppos = &iocb->ki_pos; 1382 1383 count = 0; 1384 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); 1385 if (retval) 1386 return retval; 1387 1388 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ 1389 if (filp->f_flags & O_DIRECT) { 1390 loff_t size; 1391 struct address_space *mapping; 1392 struct inode *inode; 1393 1394 mapping = filp->f_mapping; 1395 inode = mapping->host; 1396 if (!count) 1397 goto out; /* skip atime */ 1398 size = i_size_read(inode); 1399 if (pos < size) { 1400 retval = filemap_write_and_wait_range(mapping, pos, 1401 pos + iov_length(iov, nr_segs) - 1); 1402 if (!retval) { 1403 struct blk_plug plug; 1404 1405 blk_start_plug(&plug); 1406 retval = mapping->a_ops->direct_IO(READ, iocb, 1407 iov, pos, nr_segs); 1408 blk_finish_plug(&plug); 1409 } 1410 if (retval > 0) { 1411 *ppos = pos + retval; 1412 count -= retval; 1413 } 1414 1415 /* 1416 * Btrfs can have a short DIO read if we encounter 1417 * compressed extents, so if there was an error, or if 1418 * we've already read everything we wanted to, or if 1419 * there was a short read because we hit EOF, go ahead 1420 * and return. Otherwise fallthrough to buffered io for 1421 * the rest of the read. 1422 */ 1423 if (retval < 0 || !count || *ppos >= size) { 1424 file_accessed(filp); 1425 goto out; 1426 } 1427 } 1428 } 1429 1430 count = retval; 1431 for (seg = 0; seg < nr_segs; seg++) { 1432 read_descriptor_t desc; 1433 loff_t offset = 0; 1434 1435 /* 1436 * If we did a short DIO read we need to skip the section of the 1437 * iov that we've already read data into. 1438 */ 1439 if (count) { 1440 if (count > iov[seg].iov_len) { 1441 count -= iov[seg].iov_len; 1442 continue; 1443 } 1444 offset = count; 1445 count = 0; 1446 } 1447 1448 desc.written = 0; 1449 desc.arg.buf = iov[seg].iov_base + offset; 1450 desc.count = iov[seg].iov_len - offset; 1451 if (desc.count == 0) 1452 continue; 1453 desc.error = 0; 1454 do_generic_file_read(filp, ppos, &desc, file_read_actor); 1455 retval += desc.written; 1456 if (desc.error) { 1457 retval = retval ?: desc.error; 1458 break; 1459 } 1460 if (desc.count > 0) 1461 break; 1462 } 1463out: 1464 return retval; 1465} 1466EXPORT_SYMBOL(generic_file_aio_read); 1467 1468static ssize_t 1469do_readahead(struct address_space *mapping, struct file *filp, 1470 pgoff_t index, unsigned long nr) 1471{ 1472 if (!mapping || !mapping->a_ops || !mapping->a_ops->readpage) 1473 return -EINVAL; 1474 1475 force_page_cache_readahead(mapping, filp, index, nr); 1476 return 0; 1477} 1478 1479SYSCALL_DEFINE(readahead)(int fd, loff_t offset, size_t count) 1480{ 1481 ssize_t ret; 1482 struct file *file; 1483 1484 ret = -EBADF; 1485 file = fget(fd); 1486 if (file) { 1487 if (file->f_mode & FMODE_READ) { 1488 struct address_space *mapping = file->f_mapping; 1489 pgoff_t start = offset >> PAGE_CACHE_SHIFT; 1490 pgoff_t end = (offset + count - 1) >> PAGE_CACHE_SHIFT; 1491 unsigned long len = end - start + 1; 1492 ret = do_readahead(mapping, file, start, len); 1493 } 1494 fput(file); 1495 } 1496 return ret; 1497} 1498#ifdef CONFIG_HAVE_SYSCALL_WRAPPERS 1499asmlinkage long SyS_readahead(long fd, loff_t offset, long count) 1500{ 1501 return SYSC_readahead((int) fd, offset, (size_t) count); 1502} 1503SYSCALL_ALIAS(sys_readahead, SyS_readahead); 1504#endif 1505 1506#ifdef CONFIG_MMU 1507/** 1508 * page_cache_read - adds requested page to the page cache if not already there 1509 * @file: file to read 1510 * @offset: page index 1511 * 1512 * This adds the requested page to the page cache if it isn't already there, 1513 * and schedules an I/O to read in its contents from disk. 1514 */ 1515static int page_cache_read(struct file *file, pgoff_t offset) 1516{ 1517 struct address_space *mapping = file->f_mapping; 1518 struct page *page; 1519 int ret; 1520 1521 do { 1522 page = page_cache_alloc_cold(mapping); 1523 if (!page) 1524 return -ENOMEM; 1525 1526 ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL); 1527 if (ret == 0) 1528 ret = mapping->a_ops->readpage(file, page); 1529 else if (ret == -EEXIST) 1530 ret = 0; /* losing race to add is OK */ 1531 1532 page_cache_release(page); 1533 1534 } while (ret == AOP_TRUNCATED_PAGE); 1535 1536 return ret; 1537} 1538 1539#define MMAP_LOTSAMISS (100) 1540 1541/* 1542 * Synchronous readahead happens when we don't even find 1543 * a page in the page cache at all. 1544 */ 1545static void do_sync_mmap_readahead(struct vm_area_struct *vma, 1546 struct file_ra_state *ra, 1547 struct file *file, 1548 pgoff_t offset) 1549{ 1550 unsigned long ra_pages; 1551 struct address_space *mapping = file->f_mapping; 1552 1553 /* If we don't want any read-ahead, don't bother */ 1554 if (VM_RandomReadHint(vma)) 1555 return; 1556 if (!ra->ra_pages) 1557 return; 1558 1559 if (VM_SequentialReadHint(vma)) { 1560 page_cache_sync_readahead(mapping, ra, file, offset, 1561 ra->ra_pages); 1562 return; 1563 } 1564 1565 /* Avoid banging the cache line if not needed */ 1566 if (ra->mmap_miss < MMAP_LOTSAMISS * 10) 1567 ra->mmap_miss++; 1568 1569 /* 1570 * Do we miss much more than hit in this file? If so, 1571 * stop bothering with read-ahead. It will only hurt. 1572 */ 1573 if (ra->mmap_miss > MMAP_LOTSAMISS) 1574 return; 1575 1576 /* 1577 * mmap read-around 1578 */ 1579 ra_pages = max_sane_readahead(ra->ra_pages); 1580 ra->start = max_t(long, 0, offset - ra_pages / 2); 1581 ra->size = ra_pages; 1582 ra->async_size = ra_pages / 4; 1583 ra_submit(ra, mapping, file); 1584} 1585 1586/* 1587 * Asynchronous readahead happens when we find the page and PG_readahead, 1588 * so we want to possibly extend the readahead further.. 1589 */ 1590static void do_async_mmap_readahead(struct vm_area_struct *vma, 1591 struct file_ra_state *ra, 1592 struct file *file, 1593 struct page *page, 1594 pgoff_t offset) 1595{ 1596 struct address_space *mapping = file->f_mapping; 1597 1598 /* If we don't want any read-ahead, don't bother */ 1599 if (VM_RandomReadHint(vma)) 1600 return; 1601 if (ra->mmap_miss > 0) 1602 ra->mmap_miss--; 1603 if (PageReadahead(page)) 1604 page_cache_async_readahead(mapping, ra, file, 1605 page, offset, ra->ra_pages); 1606} 1607 1608/** 1609 * filemap_fault - read in file data for page fault handling 1610 * @vma: vma in which the fault was taken 1611 * @vmf: struct vm_fault containing details of the fault 1612 * 1613 * filemap_fault() is invoked via the vma operations vector for a 1614 * mapped memory region to read in file data during a page fault. 1615 * 1616 * The goto's are kind of ugly, but this streamlines the normal case of having 1617 * it in the page cache, and handles the special cases reasonably without 1618 * having a lot of duplicated code. 1619 */ 1620int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1621{ 1622 int error; 1623 struct file *file = vma->vm_file; 1624 struct address_space *mapping = file->f_mapping; 1625 struct file_ra_state *ra = &file->f_ra; 1626 struct inode *inode = mapping->host; 1627 pgoff_t offset = vmf->pgoff; 1628 struct page *page; 1629 pgoff_t size; 1630 int ret = 0; 1631 1632 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 1633 if (offset >= size) 1634 return VM_FAULT_SIGBUS; 1635 1636 /* 1637 * Do we have something in the page cache already? 1638 */ 1639 page = find_get_page(mapping, offset); 1640 if (likely(page)) { 1641 /* 1642 * We found the page, so try async readahead before 1643 * waiting for the lock. 1644 */ 1645 do_async_mmap_readahead(vma, ra, file, page, offset); 1646 } else { 1647 /* No page in the page cache at all */ 1648 do_sync_mmap_readahead(vma, ra, file, offset); 1649 count_vm_event(PGMAJFAULT); 1650 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); 1651 ret = VM_FAULT_MAJOR; 1652retry_find: 1653 page = find_get_page(mapping, offset); 1654 if (!page) 1655 goto no_cached_page; 1656 } 1657 1658 if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) { 1659 page_cache_release(page); 1660 return ret | VM_FAULT_RETRY; 1661 } 1662 1663 /* Did it get truncated? */ 1664 if (unlikely(page->mapping != mapping)) { 1665 unlock_page(page); 1666 put_page(page); 1667 goto retry_find; 1668 } 1669 VM_BUG_ON(page->index != offset); 1670 1671 /* 1672 * We have a locked page in the page cache, now we need to check 1673 * that it's up-to-date. If not, it is going to be due to an error. 1674 */ 1675 if (unlikely(!PageUptodate(page))) 1676 goto page_not_uptodate; 1677 1678 /* 1679 * Found the page and have a reference on it. 1680 * We must recheck i_size under page lock. 1681 */ 1682 size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 1683 if (unlikely(offset >= size)) { 1684 unlock_page(page); 1685 page_cache_release(page); 1686 return VM_FAULT_SIGBUS; 1687 } 1688 1689 vmf->page = page; 1690 return ret | VM_FAULT_LOCKED; 1691 1692no_cached_page: 1693 /* 1694 * We're only likely to ever get here if MADV_RANDOM is in 1695 * effect. 1696 */ 1697 error = page_cache_read(file, offset); 1698 1699 /* 1700 * The page we want has now been added to the page cache. 1701 * In the unlikely event that someone removed it in the 1702 * meantime, we'll just come back here and read it again. 1703 */ 1704 if (error >= 0) 1705 goto retry_find; 1706 1707 /* 1708 * An error return from page_cache_read can result if the 1709 * system is low on memory, or a problem occurs while trying 1710 * to schedule I/O. 1711 */ 1712 if (error == -ENOMEM) 1713 return VM_FAULT_OOM; 1714 return VM_FAULT_SIGBUS; 1715 1716page_not_uptodate: 1717 /* 1718 * Umm, take care of errors if the page isn't up-to-date. 1719 * Try to re-read it _once_. We do this synchronously, 1720 * because there really aren't any performance issues here 1721 * and we need to check for errors. 1722 */ 1723 ClearPageError(page); 1724 error = mapping->a_ops->readpage(file, page); 1725 if (!error) { 1726 wait_on_page_locked(page); 1727 if (!PageUptodate(page)) 1728 error = -EIO; 1729 } 1730 page_cache_release(page); 1731 1732 if (!error || error == AOP_TRUNCATED_PAGE) 1733 goto retry_find; 1734 1735 /* Things didn't work out. Return zero to tell the mm layer so. */ 1736 shrink_readahead_size_eio(file, ra); 1737 return VM_FAULT_SIGBUS; 1738} 1739EXPORT_SYMBOL(filemap_fault); 1740 1741const struct vm_operations_struct generic_file_vm_ops = { 1742 .fault = filemap_fault, 1743}; 1744 1745/* This is used for a general mmap of a disk file */ 1746 1747int generic_file_mmap(struct file * file, struct vm_area_struct * vma) 1748{ 1749 struct address_space *mapping = file->f_mapping; 1750 1751 if (!mapping->a_ops->readpage) 1752 return -ENOEXEC; 1753 file_accessed(file); 1754 vma->vm_ops = &generic_file_vm_ops; 1755 vma->vm_flags |= VM_CAN_NONLINEAR; 1756 return 0; 1757} 1758 1759/* 1760 * This is for filesystems which do not implement ->writepage. 1761 */ 1762int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma) 1763{ 1764 if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE)) 1765 return -EINVAL; 1766 return generic_file_mmap(file, vma); 1767} 1768#else 1769int generic_file_mmap(struct file * file, struct vm_area_struct * vma) 1770{ 1771 return -ENOSYS; 1772} 1773int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma) 1774{ 1775 return -ENOSYS; 1776} 1777#endif /* CONFIG_MMU */ 1778 1779EXPORT_SYMBOL(generic_file_mmap); 1780EXPORT_SYMBOL(generic_file_readonly_mmap); 1781 1782static struct page *__read_cache_page(struct address_space *mapping, 1783 pgoff_t index, 1784 int (*filler)(void *,struct page*), 1785 void *data, 1786 gfp_t gfp) 1787{ 1788 struct page *page; 1789 int err; 1790repeat: 1791 page = find_get_page(mapping, index); 1792 if (!page) { 1793 page = __page_cache_alloc(gfp | __GFP_COLD); 1794 if (!page) 1795 return ERR_PTR(-ENOMEM); 1796 err = add_to_page_cache_lru(page, mapping, index, gfp); 1797 if (unlikely(err)) { 1798 page_cache_release(page); 1799 if (err == -EEXIST) 1800 goto repeat; 1801 /* Presumably ENOMEM for radix tree node */ 1802 return ERR_PTR(err); 1803 } 1804 err = filler(data, page); 1805 if (err < 0) { 1806 page_cache_release(page); 1807 page = ERR_PTR(err); 1808 } 1809 } 1810 return page; 1811} 1812 1813static struct page *do_read_cache_page(struct address_space *mapping, 1814 pgoff_t index, 1815 int (*filler)(void *,struct page*), 1816 void *data, 1817 gfp_t gfp) 1818 1819{ 1820 struct page *page; 1821 int err; 1822 1823retry: 1824 page = __read_cache_page(mapping, index, filler, data, gfp); 1825 if (IS_ERR(page)) 1826 return page; 1827 if (PageUptodate(page)) 1828 goto out; 1829 1830 lock_page(page); 1831 if (!page->mapping) { 1832 unlock_page(page); 1833 page_cache_release(page); 1834 goto retry; 1835 } 1836 if (PageUptodate(page)) { 1837 unlock_page(page); 1838 goto out; 1839 } 1840 err = filler(data, page); 1841 if (err < 0) { 1842 page_cache_release(page); 1843 return ERR_PTR(err); 1844 } 1845out: 1846 mark_page_accessed(page); 1847 return page; 1848} 1849 1850/** 1851 * read_cache_page_async - read into page cache, fill it if needed 1852 * @mapping: the page's address_space 1853 * @index: the page index 1854 * @filler: function to perform the read 1855 * @data: destination for read data 1856 * 1857 * Same as read_cache_page, but don't wait for page to become unlocked 1858 * after submitting it to the filler. 1859 * 1860 * Read into the page cache. If a page already exists, and PageUptodate() is 1861 * not set, try to fill the page but don't wait for it to become unlocked. 1862 * 1863 * If the page does not get brought uptodate, return -EIO. 1864 */ 1865struct page *read_cache_page_async(struct address_space *mapping, 1866 pgoff_t index, 1867 int (*filler)(void *,struct page*), 1868 void *data) 1869{ 1870 return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping)); 1871} 1872EXPORT_SYMBOL(read_cache_page_async); 1873 1874static struct page *wait_on_page_read(struct page *page) 1875{ 1876 if (!IS_ERR(page)) { 1877 wait_on_page_locked(page); 1878 if (!PageUptodate(page)) { 1879 page_cache_release(page); 1880 page = ERR_PTR(-EIO); 1881 } 1882 } 1883 return page; 1884} 1885 1886/** 1887 * read_cache_page_gfp - read into page cache, using specified page allocation flags. 1888 * @mapping: the page's address_space 1889 * @index: the page index 1890 * @gfp: the page allocator flags to use if allocating 1891 * 1892 * This is the same as "read_mapping_page(mapping, index, NULL)", but with 1893 * any new page allocations done using the specified allocation flags. 1894 * 1895 * If the page does not get brought uptodate, return -EIO. 1896 */ 1897struct page *read_cache_page_gfp(struct address_space *mapping, 1898 pgoff_t index, 1899 gfp_t gfp) 1900{ 1901 filler_t *filler = (filler_t *)mapping->a_ops->readpage; 1902 1903 return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp)); 1904} 1905EXPORT_SYMBOL(read_cache_page_gfp); 1906 1907/** 1908 * read_cache_page - read into page cache, fill it if needed 1909 * @mapping: the page's address_space 1910 * @index: the page index 1911 * @filler: function to perform the read 1912 * @data: destination for read data 1913 * 1914 * Read into the page cache. If a page already exists, and PageUptodate() is 1915 * not set, try to fill the page then wait for it to become unlocked. 1916 * 1917 * If the page does not get brought uptodate, return -EIO. 1918 */ 1919struct page *read_cache_page(struct address_space *mapping, 1920 pgoff_t index, 1921 int (*filler)(void *,struct page*), 1922 void *data) 1923{ 1924 return wait_on_page_read(read_cache_page_async(mapping, index, filler, data)); 1925} 1926EXPORT_SYMBOL(read_cache_page); 1927 1928/* 1929 * The logic we want is 1930 * 1931 * if suid or (sgid and xgrp) 1932 * remove privs 1933 */ 1934int should_remove_suid(struct dentry *dentry) 1935{ 1936 mode_t mode = dentry->d_inode->i_mode; 1937 int kill = 0; 1938 1939 /* suid always must be killed */ 1940 if (unlikely(mode & S_ISUID)) 1941 kill = ATTR_KILL_SUID; 1942 1943 /* 1944 * sgid without any exec bits is just a mandatory locking mark; leave 1945 * it alone. If some exec bits are set, it's a real sgid; kill it. 1946 */ 1947 if (unlikely((mode & S_ISGID) && (mode & S_IXGRP))) 1948 kill |= ATTR_KILL_SGID; 1949 1950 if (unlikely(kill && !capable(CAP_FSETID) && S_ISREG(mode))) 1951 return kill; 1952 1953 return 0; 1954} 1955EXPORT_SYMBOL(should_remove_suid); 1956 1957static int __remove_suid(struct dentry *dentry, int kill) 1958{ 1959 struct iattr newattrs; 1960 1961 newattrs.ia_valid = ATTR_FORCE | kill; 1962 return notify_change(dentry, &newattrs); 1963} 1964 1965int file_remove_suid(struct file *file) 1966{ 1967 struct dentry *dentry = file->f_path.dentry; 1968 struct inode *inode = dentry->d_inode; 1969 int killsuid; 1970 int killpriv; 1971 int error = 0; 1972 1973 /* Fast path for nothing security related */ 1974 if (IS_NOSEC(inode)) 1975 return 0; 1976 1977 killsuid = should_remove_suid(dentry); 1978 killpriv = security_inode_need_killpriv(dentry); 1979 1980 if (killpriv < 0) 1981 return killpriv; 1982 if (killpriv) 1983 error = security_inode_killpriv(dentry); 1984 if (!error && killsuid) 1985 error = __remove_suid(dentry, killsuid); 1986 if (!error && (inode->i_sb->s_flags & MS_NOSEC)) 1987 inode->i_flags |= S_NOSEC; 1988 1989 return error; 1990} 1991EXPORT_SYMBOL(file_remove_suid); 1992 1993static size_t __iovec_copy_from_user_inatomic(char *vaddr, 1994 const struct iovec *iov, size_t base, size_t bytes) 1995{ 1996 size_t copied = 0, left = 0; 1997 1998 while (bytes) { 1999 char __user *buf = iov->iov_base + base; 2000 int copy = min(bytes, iov->iov_len - base); 2001 2002 base = 0; 2003 left = __copy_from_user_inatomic(vaddr, buf, copy); 2004 copied += copy; 2005 bytes -= copy; 2006 vaddr += copy; 2007 iov++; 2008 2009 if (unlikely(left)) 2010 break; 2011 } 2012 return copied - left; 2013} 2014 2015/* 2016 * Copy as much as we can into the page and return the number of bytes which 2017 * were successfully copied. If a fault is encountered then return the number of 2018 * bytes which were copied. 2019 */ 2020size_t iov_iter_copy_from_user_atomic(struct page *page, 2021 struct iov_iter *i, unsigned long offset, size_t bytes) 2022{ 2023 char *kaddr; 2024 size_t copied; 2025 2026 BUG_ON(!in_atomic()); 2027 kaddr = kmap_atomic(page, KM_USER0); 2028 if (likely(i->nr_segs == 1)) { 2029 int left; 2030 char __user *buf = i->iov->iov_base + i->iov_offset; 2031 left = __copy_from_user_inatomic(kaddr + offset, buf, bytes); 2032 copied = bytes - left; 2033 } else { 2034 copied = __iovec_copy_from_user_inatomic(kaddr + offset, 2035 i->iov, i->iov_offset, bytes); 2036 } 2037 kunmap_atomic(kaddr, KM_USER0); 2038 2039 return copied; 2040} 2041EXPORT_SYMBOL(iov_iter_copy_from_user_atomic); 2042 2043/* 2044 * This has the same sideeffects and return value as 2045 * iov_iter_copy_from_user_atomic(). 2046 * The difference is that it attempts to resolve faults. 2047 * Page must not be locked. 2048 */ 2049size_t iov_iter_copy_from_user(struct page *page, 2050 struct iov_iter *i, unsigned long offset, size_t bytes) 2051{ 2052 char *kaddr; 2053 size_t copied; 2054 2055 kaddr = kmap(page); 2056 if (likely(i->nr_segs == 1)) { 2057 int left; 2058 char __user *buf = i->iov->iov_base + i->iov_offset; 2059 left = __copy_from_user(kaddr + offset, buf, bytes); 2060 copied = bytes - left; 2061 } else { 2062 copied = __iovec_copy_from_user_inatomic(kaddr + offset, 2063 i->iov, i->iov_offset, bytes); 2064 } 2065 kunmap(page); 2066 return copied; 2067} 2068EXPORT_SYMBOL(iov_iter_copy_from_user); 2069 2070void iov_iter_advance(struct iov_iter *i, size_t bytes) 2071{ 2072 BUG_ON(i->count < bytes); 2073 2074 if (likely(i->nr_segs == 1)) { 2075 i->iov_offset += bytes; 2076 i->count -= bytes; 2077 } else { 2078 const struct iovec *iov = i->iov; 2079 size_t base = i->iov_offset; 2080 2081 /* 2082 * The !iov->iov_len check ensures we skip over unlikely 2083 * zero-length segments (without overruning the iovec). 2084 */ 2085 while (bytes || unlikely(i->count && !iov->iov_len)) { 2086 int copy; 2087 2088 copy = min(bytes, iov->iov_len - base); 2089 BUG_ON(!i->count || i->count < copy); 2090 i->count -= copy; 2091 bytes -= copy; 2092 base += copy; 2093 if (iov->iov_len == base) { 2094 iov++; 2095 base = 0; 2096 } 2097 } 2098 i->iov = iov; 2099 i->iov_offset = base; 2100 } 2101} 2102EXPORT_SYMBOL(iov_iter_advance); 2103 2104/* 2105 * Fault in the first iovec of the given iov_iter, to a maximum length 2106 * of bytes. Returns 0 on success, or non-zero if the memory could not be 2107 * accessed (ie. because it is an invalid address). 2108 * 2109 * writev-intensive code may want this to prefault several iovecs -- that 2110 * would be possible (callers must not rely on the fact that _only_ the 2111 * first iovec will be faulted with the current implementation). 2112 */ 2113int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes) 2114{ 2115 char __user *buf = i->iov->iov_base + i->iov_offset; 2116 bytes = min(bytes, i->iov->iov_len - i->iov_offset); 2117 return fault_in_pages_readable(buf, bytes); 2118} 2119EXPORT_SYMBOL(iov_iter_fault_in_readable); 2120 2121/* 2122 * Return the count of just the current iov_iter segment. 2123 */ 2124size_t iov_iter_single_seg_count(struct iov_iter *i) 2125{ 2126 const struct iovec *iov = i->iov; 2127 if (i->nr_segs == 1) 2128 return i->count; 2129 else 2130 return min(i->count, iov->iov_len - i->iov_offset); 2131} 2132EXPORT_SYMBOL(iov_iter_single_seg_count); 2133 2134/* 2135 * Performs necessary checks before doing a write 2136 * 2137 * Can adjust writing position or amount of bytes to write. 2138 * Returns appropriate error code that caller should return or 2139 * zero in case that write should be allowed. 2140 */ 2141inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk) 2142{ 2143 struct inode *inode = file->f_mapping->host; 2144 unsigned long limit = rlimit(RLIMIT_FSIZE); 2145 2146 if (unlikely(*pos < 0)) 2147 return -EINVAL; 2148 2149 if (!isblk) { 2150 /* FIXME: this is for backwards compatibility with 2.4 */ 2151 if (file->f_flags & O_APPEND) 2152 *pos = i_size_read(inode); 2153 2154 if (limit != RLIM_INFINITY) { 2155 if (*pos >= limit) { 2156 send_sig(SIGXFSZ, current, 0); 2157 return -EFBIG; 2158 } 2159 if (*count > limit - (typeof(limit))*pos) { 2160 *count = limit - (typeof(limit))*pos; 2161 } 2162 } 2163 } 2164 2165 /* 2166 * LFS rule 2167 */ 2168 if (unlikely(*pos + *count > MAX_NON_LFS && 2169 !(file->f_flags & O_LARGEFILE))) { 2170 if (*pos >= MAX_NON_LFS) { 2171 return -EFBIG; 2172 } 2173 if (*count > MAX_NON_LFS - (unsigned long)*pos) { 2174 *count = MAX_NON_LFS - (unsigned long)*pos; 2175 } 2176 } 2177 2178 /* 2179 * Are we about to exceed the fs block limit ? 2180 * 2181 * If we have written data it becomes a short write. If we have 2182 * exceeded without writing data we send a signal and return EFBIG. 2183 * Linus frestrict idea will clean these up nicely.. 2184 */ 2185 if (likely(!isblk)) { 2186 if (unlikely(*pos >= inode->i_sb->s_maxbytes)) { 2187 if (*count || *pos > inode->i_sb->s_maxbytes) { 2188 return -EFBIG; 2189 } 2190 /* zero-length writes at ->s_maxbytes are OK */ 2191 } 2192 2193 if (unlikely(*pos + *count > inode->i_sb->s_maxbytes)) 2194 *count = inode->i_sb->s_maxbytes - *pos; 2195 } else { 2196#ifdef CONFIG_BLOCK 2197 loff_t isize; 2198 if (bdev_read_only(I_BDEV(inode))) 2199 return -EPERM; 2200 isize = i_size_read(inode); 2201 if (*pos >= isize) { 2202 if (*count || *pos > isize) 2203 return -ENOSPC; 2204 } 2205 2206 if (*pos + *count > isize) 2207 *count = isize - *pos; 2208#else 2209 return -EPERM; 2210#endif 2211 } 2212 return 0; 2213} 2214EXPORT_SYMBOL(generic_write_checks); 2215 2216int pagecache_write_begin(struct file *file, struct address_space *mapping, 2217 loff_t pos, unsigned len, unsigned flags, 2218 struct page **pagep, void **fsdata) 2219{ 2220 const struct address_space_operations *aops = mapping->a_ops; 2221 2222 return aops->write_begin(file, mapping, pos, len, flags, 2223 pagep, fsdata); 2224} 2225EXPORT_SYMBOL(pagecache_write_begin); 2226 2227int pagecache_write_end(struct file *file, struct address_space *mapping, 2228 loff_t pos, unsigned len, unsigned copied, 2229 struct page *page, void *fsdata) 2230{ 2231 const struct address_space_operations *aops = mapping->a_ops; 2232 2233 mark_page_accessed(page); 2234 return aops->write_end(file, mapping, pos, len, copied, page, fsdata); 2235} 2236EXPORT_SYMBOL(pagecache_write_end); 2237 2238ssize_t 2239generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov, 2240 unsigned long *nr_segs, loff_t pos, loff_t *ppos, 2241 size_t count, size_t ocount) 2242{ 2243 struct file *file = iocb->ki_filp; 2244 struct address_space *mapping = file->f_mapping; 2245 struct inode *inode = mapping->host; 2246 ssize_t written; 2247 size_t write_len; 2248 pgoff_t end; 2249 2250 if (count != ocount) 2251 *nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count); 2252 2253 write_len = iov_length(iov, *nr_segs); 2254 end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT; 2255 2256 written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1); 2257 if (written) 2258 goto out; 2259 2260 /* 2261 * After a write we want buffered reads to be sure to go to disk to get 2262 * the new data. We invalidate clean cached page from the region we're 2263 * about to write. We do this *before* the write so that we can return 2264 * without clobbering -EIOCBQUEUED from ->direct_IO(). 2265 */ 2266 if (mapping->nrpages) { 2267 written = invalidate_inode_pages2_range(mapping, 2268 pos >> PAGE_CACHE_SHIFT, end); 2269 /* 2270 * If a page can not be invalidated, return 0 to fall back 2271 * to buffered write. 2272 */ 2273 if (written) { 2274 if (written == -EBUSY) 2275 return 0; 2276 goto out; 2277 } 2278 } 2279 2280 written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs); 2281 2282 /* 2283 * Finally, try again to invalidate clean pages which might have been 2284 * cached by non-direct readahead, or faulted in by get_user_pages() 2285 * if the source of the write was an mmap'ed region of the file 2286 * we're writing. Either one is a pretty crazy thing to do, 2287 * so we don't support it 100%. If this invalidation 2288 * fails, tough, the write still worked... 2289 */ 2290 if (mapping->nrpages) { 2291 invalidate_inode_pages2_range(mapping, 2292 pos >> PAGE_CACHE_SHIFT, end); 2293 } 2294 2295 if (written > 0) { 2296 pos += written; 2297 if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) { 2298 i_size_write(inode, pos); 2299 mark_inode_dirty(inode); 2300 } 2301 *ppos = pos; 2302 } 2303out: 2304 return written; 2305} 2306EXPORT_SYMBOL(generic_file_direct_write); 2307 2308/* 2309 * Find or create a page at the given pagecache position. Return the locked 2310 * page. This function is specifically for buffered writes. 2311 */ 2312struct page *grab_cache_page_write_begin(struct address_space *mapping, 2313 pgoff_t index, unsigned flags) 2314{ 2315 int status; 2316 struct page *page; 2317 gfp_t gfp_notmask = 0; 2318 if (flags & AOP_FLAG_NOFS) 2319 gfp_notmask = __GFP_FS; 2320repeat: 2321 page = find_lock_page(mapping, index); 2322 if (page) 2323 goto found; 2324 2325 page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~gfp_notmask); 2326 if (!page) 2327 return NULL; 2328 status = add_to_page_cache_lru(page, mapping, index, 2329 GFP_KERNEL & ~gfp_notmask); 2330 if (unlikely(status)) { 2331 page_cache_release(page); 2332 if (status == -EEXIST) 2333 goto repeat; 2334 return NULL; 2335 } 2336found: 2337 wait_on_page_writeback(page); 2338 return page; 2339} 2340EXPORT_SYMBOL(grab_cache_page_write_begin); 2341 2342static ssize_t generic_perform_write(struct file *file, 2343 struct iov_iter *i, loff_t pos) 2344{ 2345 struct address_space *mapping = file->f_mapping; 2346 const struct address_space_operations *a_ops = mapping->a_ops; 2347 long status = 0; 2348 ssize_t written = 0; 2349 unsigned int flags = 0; 2350 2351 /* 2352 * Copies from kernel address space cannot fail (NFSD is a big user). 2353 */ 2354 if (segment_eq(get_fs(), KERNEL_DS)) 2355 flags |= AOP_FLAG_UNINTERRUPTIBLE; 2356 2357 do { 2358 struct page *page; 2359 unsigned long offset; /* Offset into pagecache page */ 2360 unsigned long bytes; /* Bytes to write to page */ 2361 size_t copied; /* Bytes copied from user */ 2362 void *fsdata; 2363 2364 offset = (pos & (PAGE_CACHE_SIZE - 1)); 2365 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, 2366 iov_iter_count(i)); 2367 2368again: 2369 2370 /* 2371 * Bring in the user page that we will copy from _first_. 2372 * Otherwise there's a nasty deadlock on copying from the 2373 * same page as we're writing to, without it being marked 2374 * up-to-date. 2375 * 2376 * Not only is this an optimisation, but it is also required 2377 * to check that the address is actually valid, when atomic 2378 * usercopies are used, below. 2379 */ 2380 if (unlikely(iov_iter_fault_in_readable(i, bytes))) { 2381 status = -EFAULT; 2382 break; 2383 } 2384 2385 status = a_ops->write_begin(file, mapping, pos, bytes, flags, 2386 &page, &fsdata); 2387 if (unlikely(status)) 2388 break; 2389 2390 if (mapping_writably_mapped(mapping)) 2391 flush_dcache_page(page); 2392 2393 pagefault_disable(); 2394 copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes); 2395 pagefault_enable(); 2396 flush_dcache_page(page); 2397 2398 mark_page_accessed(page); 2399 status = a_ops->write_end(file, mapping, pos, bytes, copied, 2400 page, fsdata); 2401 if (unlikely(status < 0)) 2402 break; 2403 copied = status; 2404 2405 cond_resched(); 2406 2407 iov_iter_advance(i, copied); 2408 if (unlikely(copied == 0)) { 2409 /* 2410 * If we were unable to copy any data at all, we must 2411 * fall back to a single segment length write. 2412 * 2413 * If we didn't fallback here, we could livelock 2414 * because not all segments in the iov can be copied at 2415 * once without a pagefault. 2416 */ 2417 bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset, 2418 iov_iter_single_seg_count(i)); 2419 goto again; 2420 } 2421 pos += copied; 2422 written += copied; 2423 2424 balance_dirty_pages_ratelimited(mapping); 2425 2426 } while (iov_iter_count(i)); 2427 2428 return written ? written : status; 2429} 2430 2431ssize_t 2432generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov, 2433 unsigned long nr_segs, loff_t pos, loff_t *ppos, 2434 size_t count, ssize_t written) 2435{ 2436 struct file *file = iocb->ki_filp; 2437 ssize_t status; 2438 struct iov_iter i; 2439 2440 iov_iter_init(&i, iov, nr_segs, count, written); 2441 status = generic_perform_write(file, &i, pos); 2442 2443 if (likely(status >= 0)) { 2444 written += status; 2445 *ppos = pos + status; 2446 } 2447 2448 return written ? written : status; 2449} 2450EXPORT_SYMBOL(generic_file_buffered_write); 2451 2452/** 2453 * __generic_file_aio_write - write data to a file 2454 * @iocb: IO state structure (file, offset, etc.) 2455 * @iov: vector with data to write 2456 * @nr_segs: number of segments in the vector 2457 * @ppos: position where to write 2458 * 2459 * This function does all the work needed for actually writing data to a 2460 * file. It does all basic checks, removes SUID from the file, updates 2461 * modification times and calls proper subroutines depending on whether we 2462 * do direct IO or a standard buffered write. 2463 * 2464 * It expects i_mutex to be grabbed unless we work on a block device or similar 2465 * object which does not need locking at all. 2466 * 2467 * This function does *not* take care of syncing data in case of O_SYNC write. 2468 * A caller has to handle it. This is mainly due to the fact that we want to 2469 * avoid syncing under i_mutex. 2470 */ 2471ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, 2472 unsigned long nr_segs, loff_t *ppos) 2473{ 2474 struct file *file = iocb->ki_filp; 2475 struct address_space * mapping = file->f_mapping; 2476 size_t ocount; /* original count */ 2477 size_t count; /* after file limit checks */ 2478 struct inode *inode = mapping->host; 2479 loff_t pos; 2480 ssize_t written; 2481 ssize_t err; 2482 2483 ocount = 0; 2484 err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ); 2485 if (err) 2486 return err; 2487 2488 count = ocount; 2489 pos = *ppos; 2490 2491 vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE); 2492 2493 /* We can write back this queue in page reclaim */ 2494 current->backing_dev_info = mapping->backing_dev_info; 2495 written = 0; 2496 2497 err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode)); 2498 if (err) 2499 goto out; 2500 2501 if (count == 0) 2502 goto out; 2503 2504 err = file_remove_suid(file); 2505 if (err) 2506 goto out; 2507 2508 file_update_time(file); 2509 2510 /* coalesce the iovecs and go direct-to-BIO for O_DIRECT */ 2511 if (unlikely(file->f_flags & O_DIRECT)) { 2512 loff_t endbyte; 2513 ssize_t written_buffered; 2514 2515 written = generic_file_direct_write(iocb, iov, &nr_segs, pos, 2516 ppos, count, ocount); 2517 if (written < 0 || written == count) 2518 goto out; 2519 /* 2520 * direct-io write to a hole: fall through to buffered I/O 2521 * for completing the rest of the request. 2522 */ 2523 pos += written; 2524 count -= written; 2525 written_buffered = generic_file_buffered_write(iocb, iov, 2526 nr_segs, pos, ppos, count, 2527 written); 2528 /* 2529 * If generic_file_buffered_write() retuned a synchronous error 2530 * then we want to return the number of bytes which were 2531 * direct-written, or the error code if that was zero. Note 2532 * that this differs from normal direct-io semantics, which 2533 * will return -EFOO even if some bytes were written. 2534 */ 2535 if (written_buffered < 0) { 2536 err = written_buffered; 2537 goto out; 2538 } 2539 2540 /* 2541 * We need to ensure that the page cache pages are written to 2542 * disk and invalidated to preserve the expected O_DIRECT 2543 * semantics. 2544 */ 2545 endbyte = pos + written_buffered - written - 1; 2546 err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte); 2547 if (err == 0) { 2548 written = written_buffered; 2549 invalidate_mapping_pages(mapping, 2550 pos >> PAGE_CACHE_SHIFT, 2551 endbyte >> PAGE_CACHE_SHIFT); 2552 } else { 2553 /* 2554 * We don't know how much we wrote, so just return 2555 * the number of bytes which were direct-written 2556 */ 2557 } 2558 } else { 2559 written = generic_file_buffered_write(iocb, iov, nr_segs, 2560 pos, ppos, count, written); 2561 } 2562out: 2563 current->backing_dev_info = NULL; 2564 return written ? written : err; 2565} 2566EXPORT_SYMBOL(__generic_file_aio_write); 2567 2568/** 2569 * generic_file_aio_write - write data to a file 2570 * @iocb: IO state structure 2571 * @iov: vector with data to write 2572 * @nr_segs: number of segments in the vector 2573 * @pos: position in file where to write 2574 * 2575 * This is a wrapper around __generic_file_aio_write() to be used by most 2576 * filesystems. It takes care of syncing the file in case of O_SYNC file 2577 * and acquires i_mutex as needed. 2578 */ 2579ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov, 2580 unsigned long nr_segs, loff_t pos) 2581{ 2582 struct file *file = iocb->ki_filp; 2583 struct inode *inode = file->f_mapping->host; 2584 struct blk_plug plug; 2585 ssize_t ret; 2586 2587 BUG_ON(iocb->ki_pos != pos); 2588 2589 mutex_lock(&inode->i_mutex); 2590 blk_start_plug(&plug); 2591 ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos); 2592 mutex_unlock(&inode->i_mutex); 2593 2594 if (ret > 0 || ret == -EIOCBQUEUED) { 2595 ssize_t err; 2596 2597 err = generic_write_sync(file, pos, ret); 2598 if (err < 0 && ret > 0) 2599 ret = err; 2600 } 2601 blk_finish_plug(&plug); 2602 return ret; 2603} 2604EXPORT_SYMBOL(generic_file_aio_write); 2605 2606/** 2607 * try_to_release_page() - release old fs-specific metadata on a page 2608 * 2609 * @page: the page which the kernel is trying to free 2610 * @gfp_mask: memory allocation flags (and I/O mode) 2611 * 2612 * The address_space is to try to release any data against the page 2613 * (presumably at page->private). If the release was successful, return `1'. 2614 * Otherwise return zero. 2615 * 2616 * This may also be called if PG_fscache is set on a page, indicating that the 2617 * page is known to the local caching routines. 2618 * 2619 * The @gfp_mask argument specifies whether I/O may be performed to release 2620 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS). 2621 * 2622 */ 2623int try_to_release_page(struct page *page, gfp_t gfp_mask) 2624{ 2625 struct address_space * const mapping = page->mapping; 2626 2627 BUG_ON(!PageLocked(page)); 2628 if (PageWriteback(page)) 2629 return 0; 2630 2631 if (mapping && mapping->a_ops->releasepage) 2632 return mapping->a_ops->releasepage(page, gfp_mask); 2633 return try_to_free_buffers(page); 2634} 2635 2636EXPORT_SYMBOL(try_to_release_page); 2637