Linux Audio

Check our new training course

Loading...
v5.14.15
   1// SPDX-License-Identifier: GPL-2.0-only
   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/export.h>
  14#include <linux/compiler.h>
  15#include <linux/dax.h>
  16#include <linux/fs.h>
  17#include <linux/sched/signal.h>
  18#include <linux/uaccess.h>
 
  19#include <linux/capability.h>
  20#include <linux/kernel_stat.h>
  21#include <linux/gfp.h>
  22#include <linux/mm.h>
  23#include <linux/swap.h>
  24#include <linux/mman.h>
  25#include <linux/pagemap.h>
  26#include <linux/file.h>
  27#include <linux/uio.h>
  28#include <linux/error-injection.h>
  29#include <linux/hash.h>
  30#include <linux/writeback.h>
  31#include <linux/backing-dev.h>
  32#include <linux/pagevec.h>
  33#include <linux/blkdev.h>
  34#include <linux/security.h>
  35#include <linux/cpuset.h>
  36#include <linux/hugetlb.h>
  37#include <linux/memcontrol.h>
  38#include <linux/cleancache.h>
  39#include <linux/shmem_fs.h>
  40#include <linux/rmap.h>
  41#include <linux/delayacct.h>
  42#include <linux/psi.h>
  43#include <linux/ramfs.h>
  44#include <linux/page_idle.h>
  45#include <asm/pgalloc.h>
  46#include <asm/tlbflush.h>
  47#include "internal.h"
  48
  49#define CREATE_TRACE_POINTS
  50#include <trace/events/filemap.h>
  51
  52/*
  53 * FIXME: remove all knowledge of the buffer layer from the core VM
  54 */
  55#include <linux/buffer_head.h> /* for try_to_free_buffers */
  56
  57#include <asm/mman.h>
  58
  59/*
  60 * Shared mappings implemented 30.11.1994. It's not fully working yet,
  61 * though.
  62 *
  63 * Shared mappings now work. 15.8.1995  Bruno.
  64 *
  65 * finished 'unifying' the page and buffer cache and SMP-threaded the
  66 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
  67 *
  68 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
  69 */
  70
  71/*
  72 * Lock ordering:
  73 *
  74 *  ->i_mmap_rwsem		(truncate_pagecache)
  75 *    ->private_lock		(__free_pte->__set_page_dirty_buffers)
  76 *      ->swap_lock		(exclusive_swap_page, others)
  77 *        ->i_pages lock
  78 *
  79 *  ->i_mutex
  80 *    ->i_mmap_rwsem		(truncate->unmap_mapping_range)
  81 *
  82 *  ->mmap_lock
  83 *    ->i_mmap_rwsem
  84 *      ->page_table_lock or pte_lock	(various, mainly in memory.c)
  85 *        ->i_pages lock	(arch-dependent flush_dcache_mmap_lock)
  86 *
  87 *  ->mmap_lock
  88 *    ->lock_page		(access_process_vm)
  89 *
  90 *  ->i_mutex			(generic_perform_write)
  91 *    ->mmap_lock		(fault_in_pages_readable->do_page_fault)
  92 *
  93 *  bdi->wb.list_lock
  94 *    sb_lock			(fs/fs-writeback.c)
  95 *    ->i_pages lock		(__sync_single_inode)
  96 *
  97 *  ->i_mmap_rwsem
  98 *    ->anon_vma.lock		(vma_adjust)
  99 *
 100 *  ->anon_vma.lock
 101 *    ->page_table_lock or pte_lock	(anon_vma_prepare and various)
 102 *
 103 *  ->page_table_lock or pte_lock
 104 *    ->swap_lock		(try_to_unmap_one)
 105 *    ->private_lock		(try_to_unmap_one)
 106 *    ->i_pages lock		(try_to_unmap_one)
 107 *    ->lruvec->lru_lock	(follow_page->mark_page_accessed)
 108 *    ->lruvec->lru_lock	(check_pte_range->isolate_lru_page)
 109 *    ->private_lock		(page_remove_rmap->set_page_dirty)
 110 *    ->i_pages lock		(page_remove_rmap->set_page_dirty)
 111 *    bdi.wb->list_lock		(page_remove_rmap->set_page_dirty)
 112 *    ->inode->i_lock		(page_remove_rmap->set_page_dirty)
 113 *    ->memcg->move_lock	(page_remove_rmap->lock_page_memcg)
 114 *    bdi.wb->list_lock		(zap_pte_range->set_page_dirty)
 115 *    ->inode->i_lock		(zap_pte_range->set_page_dirty)
 116 *    ->private_lock		(zap_pte_range->__set_page_dirty_buffers)
 117 *
 118 * ->i_mmap_rwsem
 119 *   ->tasklist_lock            (memory_failure, collect_procs_ao)
 120 */
 121
 122static void page_cache_delete(struct address_space *mapping,
 123				   struct page *page, void *shadow)
 124{
 125	XA_STATE(xas, &mapping->i_pages, page->index);
 126	unsigned int nr = 1;
 127
 128	mapping_set_update(&xas, mapping);
 129
 130	/* hugetlb pages are represented by a single entry in the xarray */
 131	if (!PageHuge(page)) {
 132		xas_set_order(&xas, page->index, compound_order(page));
 133		nr = compound_nr(page);
 134	}
 135
 136	VM_BUG_ON_PAGE(!PageLocked(page), page);
 137	VM_BUG_ON_PAGE(PageTail(page), page);
 138	VM_BUG_ON_PAGE(nr != 1 && shadow, page);
 139
 140	xas_store(&xas, shadow);
 141	xas_init_marks(&xas);
 142
 143	page->mapping = NULL;
 144	/* Leave page->index set: truncation lookup relies upon it */
 145	mapping->nrpages -= nr;
 146}
 147
 148static void unaccount_page_cache_page(struct address_space *mapping,
 149				      struct page *page)
 150{
 151	int nr;
 152
 153	/*
 154	 * if we're uptodate, flush out into the cleancache, otherwise
 155	 * invalidate any existing cleancache entries.  We can't leave
 156	 * stale data around in the cleancache once our page is gone
 157	 */
 158	if (PageUptodate(page) && PageMappedToDisk(page))
 159		cleancache_put_page(page);
 160	else
 161		cleancache_invalidate_page(mapping, page);
 162
 163	VM_BUG_ON_PAGE(PageTail(page), page);
 164	VM_BUG_ON_PAGE(page_mapped(page), page);
 165	if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
 166		int mapcount;
 167
 168		pr_alert("BUG: Bad page cache in process %s  pfn:%05lx\n",
 169			 current->comm, page_to_pfn(page));
 170		dump_page(page, "still mapped when deleted");
 171		dump_stack();
 172		add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
 173
 174		mapcount = page_mapcount(page);
 175		if (mapping_exiting(mapping) &&
 176		    page_count(page) >= mapcount + 2) {
 177			/*
 178			 * All vmas have already been torn down, so it's
 179			 * a good bet that actually the page is unmapped,
 180			 * and we'd prefer not to leak it: if we're wrong,
 181			 * some other bad page check should catch it later.
 182			 */
 183			page_mapcount_reset(page);
 184			page_ref_sub(page, mapcount);
 185		}
 186	}
 187
 188	/* hugetlb pages do not participate in page cache accounting. */
 189	if (PageHuge(page))
 190		return;
 191
 192	nr = thp_nr_pages(page);
 193
 194	__mod_lruvec_page_state(page, NR_FILE_PAGES, -nr);
 195	if (PageSwapBacked(page)) {
 196		__mod_lruvec_page_state(page, NR_SHMEM, -nr);
 197		if (PageTransHuge(page))
 198			__mod_lruvec_page_state(page, NR_SHMEM_THPS, -nr);
 199	} else if (PageTransHuge(page)) {
 200		__mod_lruvec_page_state(page, NR_FILE_THPS, -nr);
 201		filemap_nr_thps_dec(mapping);
 202	}
 203
 204	/*
 205	 * At this point page must be either written or cleaned by
 206	 * truncate.  Dirty page here signals a bug and loss of
 207	 * unwritten data.
 208	 *
 209	 * This fixes dirty accounting after removing the page entirely
 210	 * but leaves PageDirty set: it has no effect for truncated
 211	 * page and anyway will be cleared before returning page into
 212	 * buddy allocator.
 213	 */
 214	if (WARN_ON_ONCE(PageDirty(page)))
 215		account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
 216}
 217
 218/*
 219 * Delete a page from the page cache and free it. Caller has to make
 220 * sure the page is locked and that nobody else uses it - or that usage
 221 * is safe.  The caller must hold the i_pages lock.
 222 */
 223void __delete_from_page_cache(struct page *page, void *shadow)
 224{
 225	struct address_space *mapping = page->mapping;
 226
 227	trace_mm_filemap_delete_from_page_cache(page);
 228
 229	unaccount_page_cache_page(mapping, page);
 230	page_cache_delete(mapping, page, shadow);
 231}
 232
 233static void page_cache_free_page(struct address_space *mapping,
 234				struct page *page)
 235{
 236	void (*freepage)(struct page *);
 237
 238	freepage = mapping->a_ops->freepage;
 239	if (freepage)
 240		freepage(page);
 241
 242	if (PageTransHuge(page) && !PageHuge(page)) {
 243		page_ref_sub(page, thp_nr_pages(page));
 244		VM_BUG_ON_PAGE(page_count(page) <= 0, page);
 245	} else {
 246		put_page(page);
 247	}
 248}
 249
 250/**
 251 * delete_from_page_cache - delete page from page cache
 252 * @page: the page which the kernel is trying to remove from page cache
 253 *
 254 * This must be called only on pages that have been verified to be in the page
 255 * cache and locked.  It will never put the page into the free list, the caller
 256 * has a reference on the page.
 257 */
 258void delete_from_page_cache(struct page *page)
 259{
 260	struct address_space *mapping = page_mapping(page);
 261	unsigned long flags;
 262
 263	BUG_ON(!PageLocked(page));
 264	xa_lock_irqsave(&mapping->i_pages, flags);
 265	__delete_from_page_cache(page, NULL);
 266	xa_unlock_irqrestore(&mapping->i_pages, flags);
 267
 268	page_cache_free_page(mapping, page);
 269}
 270EXPORT_SYMBOL(delete_from_page_cache);
 271
 272/*
 273 * page_cache_delete_batch - delete several pages from page cache
 274 * @mapping: the mapping to which pages belong
 275 * @pvec: pagevec with pages to delete
 276 *
 277 * The function walks over mapping->i_pages and removes pages passed in @pvec
 278 * from the mapping. The function expects @pvec to be sorted by page index
 279 * and is optimised for it to be dense.
 280 * It tolerates holes in @pvec (mapping entries at those indices are not
 281 * modified). The function expects only THP head pages to be present in the
 282 * @pvec.
 283 *
 284 * The function expects the i_pages lock to be held.
 285 */
 286static void page_cache_delete_batch(struct address_space *mapping,
 287			     struct pagevec *pvec)
 288{
 289	XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index);
 290	int total_pages = 0;
 291	int i = 0;
 292	struct page *page;
 293
 294	mapping_set_update(&xas, mapping);
 295	xas_for_each(&xas, page, ULONG_MAX) {
 296		if (i >= pagevec_count(pvec))
 297			break;
 298
 299		/* A swap/dax/shadow entry got inserted? Skip it. */
 300		if (xa_is_value(page))
 301			continue;
 302		/*
 303		 * A page got inserted in our range? Skip it. We have our
 304		 * pages locked so they are protected from being removed.
 305		 * If we see a page whose index is higher than ours, it
 306		 * means our page has been removed, which shouldn't be
 307		 * possible because we're holding the PageLock.
 308		 */
 309		if (page != pvec->pages[i]) {
 310			VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index,
 311					page);
 312			continue;
 313		}
 314
 315		WARN_ON_ONCE(!PageLocked(page));
 316
 317		if (page->index == xas.xa_index)
 318			page->mapping = NULL;
 319		/* Leave page->index set: truncation lookup relies on it */
 320
 321		/*
 322		 * Move to the next page in the vector if this is a regular
 323		 * page or the index is of the last sub-page of this compound
 324		 * page.
 325		 */
 326		if (page->index + compound_nr(page) - 1 == xas.xa_index)
 327			i++;
 328		xas_store(&xas, NULL);
 329		total_pages++;
 330	}
 331	mapping->nrpages -= total_pages;
 332}
 333
 334void delete_from_page_cache_batch(struct address_space *mapping,
 335				  struct pagevec *pvec)
 336{
 337	int i;
 338	unsigned long flags;
 339
 340	if (!pagevec_count(pvec))
 341		return;
 342
 343	xa_lock_irqsave(&mapping->i_pages, flags);
 344	for (i = 0; i < pagevec_count(pvec); i++) {
 345		trace_mm_filemap_delete_from_page_cache(pvec->pages[i]);
 346
 347		unaccount_page_cache_page(mapping, pvec->pages[i]);
 348	}
 349	page_cache_delete_batch(mapping, pvec);
 350	xa_unlock_irqrestore(&mapping->i_pages, flags);
 351
 352	for (i = 0; i < pagevec_count(pvec); i++)
 353		page_cache_free_page(mapping, pvec->pages[i]);
 
 354}
 
 355
 356int filemap_check_errors(struct address_space *mapping)
 357{
 358	int ret = 0;
 359	/* Check for outstanding write errors */
 360	if (test_bit(AS_ENOSPC, &mapping->flags) &&
 361	    test_and_clear_bit(AS_ENOSPC, &mapping->flags))
 362		ret = -ENOSPC;
 363	if (test_bit(AS_EIO, &mapping->flags) &&
 364	    test_and_clear_bit(AS_EIO, &mapping->flags))
 365		ret = -EIO;
 366	return ret;
 367}
 368EXPORT_SYMBOL(filemap_check_errors);
 369
 370static int filemap_check_and_keep_errors(struct address_space *mapping)
 371{
 372	/* Check for outstanding write errors */
 373	if (test_bit(AS_EIO, &mapping->flags))
 374		return -EIO;
 375	if (test_bit(AS_ENOSPC, &mapping->flags))
 376		return -ENOSPC;
 377	return 0;
 378}
 379
 380/**
 381 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
 382 * @mapping:	address space structure to write
 383 * @start:	offset in bytes where the range starts
 384 * @end:	offset in bytes where the range ends (inclusive)
 385 * @sync_mode:	enable synchronous operation
 386 *
 387 * Start writeback against all of a mapping's dirty pages that lie
 388 * within the byte offsets <start, end> inclusive.
 389 *
 390 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
 391 * opposed to a regular memory cleansing writeback.  The difference between
 392 * these two operations is that if a dirty page/buffer is encountered, it must
 393 * be waited upon, and not just skipped over.
 394 *
 395 * Return: %0 on success, negative error code otherwise.
 396 */
 397int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
 398				loff_t end, int sync_mode)
 399{
 400	int ret;
 401	struct writeback_control wbc = {
 402		.sync_mode = sync_mode,
 403		.nr_to_write = LONG_MAX,
 404		.range_start = start,
 405		.range_end = end,
 406	};
 407
 408	if (!mapping_can_writeback(mapping) ||
 409	    !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
 410		return 0;
 411
 412	wbc_attach_fdatawrite_inode(&wbc, mapping->host);
 413	ret = do_writepages(mapping, &wbc);
 414	wbc_detach_inode(&wbc);
 415	return ret;
 416}
 417
 418static inline int __filemap_fdatawrite(struct address_space *mapping,
 419	int sync_mode)
 420{
 421	return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
 422}
 423
 424int filemap_fdatawrite(struct address_space *mapping)
 425{
 426	return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
 427}
 428EXPORT_SYMBOL(filemap_fdatawrite);
 429
 430int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
 431				loff_t end)
 432{
 433	return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
 434}
 435EXPORT_SYMBOL(filemap_fdatawrite_range);
 436
 437/**
 438 * filemap_flush - mostly a non-blocking flush
 439 * @mapping:	target address_space
 440 *
 441 * This is a mostly non-blocking flush.  Not suitable for data-integrity
 442 * purposes - I/O may not be started against all dirty pages.
 443 *
 444 * Return: %0 on success, negative error code otherwise.
 445 */
 446int filemap_flush(struct address_space *mapping)
 447{
 448	return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
 449}
 450EXPORT_SYMBOL(filemap_flush);
 451
 452/**
 453 * filemap_range_has_page - check if a page exists in range.
 454 * @mapping:           address space within which to check
 455 * @start_byte:        offset in bytes where the range starts
 456 * @end_byte:          offset in bytes where the range ends (inclusive)
 457 *
 458 * Find at least one page in the range supplied, usually used to check if
 459 * direct writing in this range will trigger a writeback.
 460 *
 461 * Return: %true if at least one page exists in the specified range,
 462 * %false otherwise.
 463 */
 464bool filemap_range_has_page(struct address_space *mapping,
 465			   loff_t start_byte, loff_t end_byte)
 466{
 467	struct page *page;
 468	XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
 469	pgoff_t max = end_byte >> PAGE_SHIFT;
 470
 471	if (end_byte < start_byte)
 472		return false;
 473
 474	rcu_read_lock();
 475	for (;;) {
 476		page = xas_find(&xas, max);
 477		if (xas_retry(&xas, page))
 478			continue;
 479		/* Shadow entries don't count */
 480		if (xa_is_value(page))
 481			continue;
 482		/*
 483		 * We don't need to try to pin this page; we're about to
 484		 * release the RCU lock anyway.  It is enough to know that
 485		 * there was a page here recently.
 486		 */
 487		break;
 488	}
 489	rcu_read_unlock();
 490
 491	return page != NULL;
 492}
 493EXPORT_SYMBOL(filemap_range_has_page);
 494
 495static void __filemap_fdatawait_range(struct address_space *mapping,
 496				     loff_t start_byte, loff_t end_byte)
 497{
 498	pgoff_t index = start_byte >> PAGE_SHIFT;
 499	pgoff_t end = end_byte >> PAGE_SHIFT;
 500	struct pagevec pvec;
 501	int nr_pages;
 
 502
 503	if (end_byte < start_byte)
 504		return;
 505
 506	pagevec_init(&pvec);
 507	while (index <= end) {
 
 
 
 508		unsigned i;
 509
 510		nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
 511				end, PAGECACHE_TAG_WRITEBACK);
 512		if (!nr_pages)
 513			break;
 514
 515		for (i = 0; i < nr_pages; i++) {
 516			struct page *page = pvec.pages[i];
 517
 
 
 
 
 518			wait_on_page_writeback(page);
 519			ClearPageError(page);
 
 520		}
 521		pagevec_release(&pvec);
 522		cond_resched();
 523	}
 524}
 525
 526/**
 527 * filemap_fdatawait_range - wait for writeback to complete
 528 * @mapping:		address space structure to wait for
 529 * @start_byte:		offset in bytes where the range starts
 530 * @end_byte:		offset in bytes where the range ends (inclusive)
 531 *
 532 * Walk the list of under-writeback pages of the given address space
 533 * in the given range and wait for all of them.  Check error status of
 534 * the address space and return it.
 535 *
 536 * Since the error status of the address space is cleared by this function,
 537 * callers are responsible for checking the return value and handling and/or
 538 * reporting the error.
 539 *
 540 * Return: error status of the address space.
 541 */
 542int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
 543			    loff_t end_byte)
 544{
 545	__filemap_fdatawait_range(mapping, start_byte, end_byte);
 546	return filemap_check_errors(mapping);
 547}
 548EXPORT_SYMBOL(filemap_fdatawait_range);
 549
 550/**
 551 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
 552 * @mapping:		address space structure to wait for
 553 * @start_byte:		offset in bytes where the range starts
 554 * @end_byte:		offset in bytes where the range ends (inclusive)
 555 *
 556 * Walk the list of under-writeback pages of the given address space in the
 557 * given range and wait for all of them.  Unlike filemap_fdatawait_range(),
 558 * this function does not clear error status of the address space.
 559 *
 560 * Use this function if callers don't handle errors themselves.  Expected
 561 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
 562 * fsfreeze(8)
 563 */
 564int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
 565		loff_t start_byte, loff_t end_byte)
 566{
 567	__filemap_fdatawait_range(mapping, start_byte, end_byte);
 568	return filemap_check_and_keep_errors(mapping);
 569}
 570EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
 571
 572/**
 573 * file_fdatawait_range - wait for writeback to complete
 574 * @file:		file pointing to address space structure to wait for
 575 * @start_byte:		offset in bytes where the range starts
 576 * @end_byte:		offset in bytes where the range ends (inclusive)
 577 *
 578 * Walk the list of under-writeback pages of the address space that file
 579 * refers to, in the given range and wait for all of them.  Check error
 580 * status of the address space vs. the file->f_wb_err cursor and return it.
 581 *
 582 * Since the error status of the file is advanced by this function,
 583 * callers are responsible for checking the return value and handling and/or
 584 * reporting the error.
 585 *
 586 * Return: error status of the address space vs. the file->f_wb_err cursor.
 587 */
 588int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
 589{
 590	struct address_space *mapping = file->f_mapping;
 591
 592	__filemap_fdatawait_range(mapping, start_byte, end_byte);
 593	return file_check_and_advance_wb_err(file);
 594}
 595EXPORT_SYMBOL(file_fdatawait_range);
 596
 597/**
 598 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
 599 * @mapping: address space structure to wait for
 600 *
 601 * Walk the list of under-writeback pages of the given address space
 602 * and wait for all of them.  Unlike filemap_fdatawait(), this function
 603 * does not clear error status of the address space.
 604 *
 605 * Use this function if callers don't handle errors themselves.  Expected
 606 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
 607 * fsfreeze(8)
 608 *
 609 * Return: error status of the address space.
 610 */
 611int filemap_fdatawait_keep_errors(struct address_space *mapping)
 612{
 613	__filemap_fdatawait_range(mapping, 0, LLONG_MAX);
 614	return filemap_check_and_keep_errors(mapping);
 615}
 616EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
 617
 618/* Returns true if writeback might be needed or already in progress. */
 619static bool mapping_needs_writeback(struct address_space *mapping)
 620{
 621	return mapping->nrpages;
 622}
 623
 624/**
 625 * filemap_range_needs_writeback - check if range potentially needs writeback
 626 * @mapping:           address space within which to check
 627 * @start_byte:        offset in bytes where the range starts
 628 * @end_byte:          offset in bytes where the range ends (inclusive)
 629 *
 630 * Find at least one page in the range supplied, usually used to check if
 631 * direct writing in this range will trigger a writeback. Used by O_DIRECT
 632 * read/write with IOCB_NOWAIT, to see if the caller needs to do
 633 * filemap_write_and_wait_range() before proceeding.
 634 *
 635 * Return: %true if the caller should do filemap_write_and_wait_range() before
 636 * doing O_DIRECT to a page in this range, %false otherwise.
 637 */
 638bool filemap_range_needs_writeback(struct address_space *mapping,
 639				   loff_t start_byte, loff_t end_byte)
 640{
 641	XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
 642	pgoff_t max = end_byte >> PAGE_SHIFT;
 643	struct page *page;
 644
 645	if (!mapping_needs_writeback(mapping))
 646		return false;
 647	if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY) &&
 648	    !mapping_tagged(mapping, PAGECACHE_TAG_WRITEBACK))
 649		return false;
 650	if (end_byte < start_byte)
 651		return false;
 652
 653	rcu_read_lock();
 654	xas_for_each(&xas, page, max) {
 655		if (xas_retry(&xas, page))
 656			continue;
 657		if (xa_is_value(page))
 658			continue;
 659		if (PageDirty(page) || PageLocked(page) || PageWriteback(page))
 660			break;
 661	}
 662	rcu_read_unlock();
 663	return page != NULL;
 664}
 665EXPORT_SYMBOL_GPL(filemap_range_needs_writeback);
 666
 667/**
 668 * filemap_write_and_wait_range - write out & wait on a file range
 669 * @mapping:	the address_space for the pages
 670 * @lstart:	offset in bytes where the range starts
 671 * @lend:	offset in bytes where the range ends (inclusive)
 672 *
 673 * Write out and wait upon file offsets lstart->lend, inclusive.
 674 *
 675 * Note that @lend is inclusive (describes the last byte to be written) so
 676 * that this function can be used to write to the very end-of-file (end = -1).
 677 *
 678 * Return: error status of the address space.
 679 */
 680int filemap_write_and_wait_range(struct address_space *mapping,
 681				 loff_t lstart, loff_t lend)
 682{
 683	int err = 0;
 684
 685	if (mapping_needs_writeback(mapping)) {
 686		err = __filemap_fdatawrite_range(mapping, lstart, lend,
 687						 WB_SYNC_ALL);
 688		/*
 689		 * Even if the above returned error, the pages may be
 690		 * written partially (e.g. -ENOSPC), so we wait for it.
 691		 * But the -EIO is special case, it may indicate the worst
 692		 * thing (e.g. bug) happened, so we avoid waiting for it.
 693		 */
 694		if (err != -EIO) {
 695			int err2 = filemap_fdatawait_range(mapping,
 696						lstart, lend);
 697			if (!err)
 698				err = err2;
 699		} else {
 700			/* Clear any previously stored errors */
 701			filemap_check_errors(mapping);
 702		}
 703	} else {
 704		err = filemap_check_errors(mapping);
 705	}
 706	return err;
 707}
 708EXPORT_SYMBOL(filemap_write_and_wait_range);
 709
 710void __filemap_set_wb_err(struct address_space *mapping, int err)
 711{
 712	errseq_t eseq = errseq_set(&mapping->wb_err, err);
 713
 714	trace_filemap_set_wb_err(mapping, eseq);
 715}
 716EXPORT_SYMBOL(__filemap_set_wb_err);
 717
 718/**
 719 * file_check_and_advance_wb_err - report wb error (if any) that was previously
 720 * 				   and advance wb_err to current one
 721 * @file: struct file on which the error is being reported
 722 *
 723 * When userland calls fsync (or something like nfsd does the equivalent), we
 724 * want to report any writeback errors that occurred since the last fsync (or
 725 * since the file was opened if there haven't been any).
 726 *
 727 * Grab the wb_err from the mapping. If it matches what we have in the file,
 728 * then just quickly return 0. The file is all caught up.
 729 *
 730 * If it doesn't match, then take the mapping value, set the "seen" flag in
 731 * it and try to swap it into place. If it works, or another task beat us
 732 * to it with the new value, then update the f_wb_err and return the error
 733 * portion. The error at this point must be reported via proper channels
 734 * (a'la fsync, or NFS COMMIT operation, etc.).
 735 *
 736 * While we handle mapping->wb_err with atomic operations, the f_wb_err
 737 * value is protected by the f_lock since we must ensure that it reflects
 738 * the latest value swapped in for this file descriptor.
 739 *
 740 * Return: %0 on success, negative error code otherwise.
 741 */
 742int file_check_and_advance_wb_err(struct file *file)
 743{
 744	int err = 0;
 745	errseq_t old = READ_ONCE(file->f_wb_err);
 746	struct address_space *mapping = file->f_mapping;
 747
 748	/* Locklessly handle the common case where nothing has changed */
 749	if (errseq_check(&mapping->wb_err, old)) {
 750		/* Something changed, must use slow path */
 751		spin_lock(&file->f_lock);
 752		old = file->f_wb_err;
 753		err = errseq_check_and_advance(&mapping->wb_err,
 754						&file->f_wb_err);
 755		trace_file_check_and_advance_wb_err(file, old);
 756		spin_unlock(&file->f_lock);
 757	}
 758
 759	/*
 760	 * We're mostly using this function as a drop in replacement for
 761	 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
 762	 * that the legacy code would have had on these flags.
 763	 */
 764	clear_bit(AS_EIO, &mapping->flags);
 765	clear_bit(AS_ENOSPC, &mapping->flags);
 766	return err;
 767}
 768EXPORT_SYMBOL(file_check_and_advance_wb_err);
 769
 770/**
 771 * file_write_and_wait_range - write out & wait on a file range
 772 * @file:	file pointing to address_space with pages
 773 * @lstart:	offset in bytes where the range starts
 774 * @lend:	offset in bytes where the range ends (inclusive)
 775 *
 776 * Write out and wait upon file offsets lstart->lend, inclusive.
 777 *
 778 * Note that @lend is inclusive (describes the last byte to be written) so
 779 * that this function can be used to write to the very end-of-file (end = -1).
 780 *
 781 * After writing out and waiting on the data, we check and advance the
 782 * f_wb_err cursor to the latest value, and return any errors detected there.
 783 *
 784 * Return: %0 on success, negative error code otherwise.
 785 */
 786int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
 
 787{
 788	int err = 0, err2;
 789	struct address_space *mapping = file->f_mapping;
 790
 791	if (mapping_needs_writeback(mapping)) {
 792		err = __filemap_fdatawrite_range(mapping, lstart, lend,
 793						 WB_SYNC_ALL);
 794		/* See comment of filemap_write_and_wait() */
 795		if (err != -EIO)
 796			__filemap_fdatawait_range(mapping, lstart, lend);
 
 
 
 
 797	}
 798	err2 = file_check_and_advance_wb_err(file);
 799	if (!err)
 800		err = err2;
 801	return err;
 802}
 803EXPORT_SYMBOL(file_write_and_wait_range);
 804
 805/**
 806 * replace_page_cache_page - replace a pagecache page with a new one
 807 * @old:	page to be replaced
 808 * @new:	page to replace with
 
 809 *
 810 * This function replaces a page in the pagecache with a new one.  On
 811 * success it acquires the pagecache reference for the new page and
 812 * drops it for the old page.  Both the old and new pages must be
 813 * locked.  This function does not add the new page to the LRU, the
 814 * caller must do that.
 815 *
 816 * The remove + add is atomic.  This function cannot fail.
 
 817 */
 818void replace_page_cache_page(struct page *old, struct page *new)
 819{
 820	struct address_space *mapping = old->mapping;
 821	void (*freepage)(struct page *) = mapping->a_ops->freepage;
 822	pgoff_t offset = old->index;
 823	XA_STATE(xas, &mapping->i_pages, offset);
 824	unsigned long flags;
 825
 826	VM_BUG_ON_PAGE(!PageLocked(old), old);
 827	VM_BUG_ON_PAGE(!PageLocked(new), new);
 828	VM_BUG_ON_PAGE(new->mapping, new);
 829
 830	get_page(new);
 831	new->mapping = mapping;
 832	new->index = offset;
 833
 834	mem_cgroup_migrate(old, new);
 835
 836	xas_lock_irqsave(&xas, flags);
 837	xas_store(&xas, new);
 838
 839	old->mapping = NULL;
 840	/* hugetlb pages do not participate in page cache accounting. */
 841	if (!PageHuge(old))
 842		__dec_lruvec_page_state(old, NR_FILE_PAGES);
 843	if (!PageHuge(new))
 844		__inc_lruvec_page_state(new, NR_FILE_PAGES);
 845	if (PageSwapBacked(old))
 846		__dec_lruvec_page_state(old, NR_SHMEM);
 847	if (PageSwapBacked(new))
 848		__inc_lruvec_page_state(new, NR_SHMEM);
 849	xas_unlock_irqrestore(&xas, flags);
 850	if (freepage)
 851		freepage(old);
 852	put_page(old);
 853}
 854EXPORT_SYMBOL_GPL(replace_page_cache_page);
 855
 856noinline int __add_to_page_cache_locked(struct page *page,
 857					struct address_space *mapping,
 858					pgoff_t offset, gfp_t gfp,
 859					void **shadowp)
 860{
 861	XA_STATE(xas, &mapping->i_pages, offset);
 862	int huge = PageHuge(page);
 863	int error;
 864	bool charged = false;
 865
 866	VM_BUG_ON_PAGE(!PageLocked(page), page);
 867	VM_BUG_ON_PAGE(PageSwapBacked(page), page);
 868	mapping_set_update(&xas, mapping);
 869
 870	get_page(page);
 871	page->mapping = mapping;
 872	page->index = offset;
 873
 874	if (!huge) {
 875		error = mem_cgroup_charge(page, NULL, gfp);
 876		if (error)
 877			goto error;
 878		charged = true;
 879	}
 880
 881	gfp &= GFP_RECLAIM_MASK;
 882
 883	do {
 884		unsigned int order = xa_get_order(xas.xa, xas.xa_index);
 885		void *entry, *old = NULL;
 886
 887		if (order > thp_order(page))
 888			xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index),
 889					order, gfp);
 890		xas_lock_irq(&xas);
 891		xas_for_each_conflict(&xas, entry) {
 892			old = entry;
 893			if (!xa_is_value(entry)) {
 894				xas_set_err(&xas, -EEXIST);
 895				goto unlock;
 896			}
 897		}
 898
 899		if (old) {
 900			if (shadowp)
 901				*shadowp = old;
 902			/* entry may have been split before we acquired lock */
 903			order = xa_get_order(xas.xa, xas.xa_index);
 904			if (order > thp_order(page)) {
 905				xas_split(&xas, old, order);
 906				xas_reset(&xas);
 907			}
 908		}
 909
 910		xas_store(&xas, page);
 911		if (xas_error(&xas))
 912			goto unlock;
 913
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 914		mapping->nrpages++;
 915
 916		/* hugetlb pages do not participate in page cache accounting */
 917		if (!huge)
 918			__inc_lruvec_page_state(page, NR_FILE_PAGES);
 919unlock:
 920		xas_unlock_irq(&xas);
 921	} while (xas_nomem(&xas, gfp));
 922
 923	if (xas_error(&xas)) {
 924		error = xas_error(&xas);
 925		if (charged)
 926			mem_cgroup_uncharge(page);
 927		goto error;
 928	}
 929
 930	trace_mm_filemap_add_to_page_cache(page);
 931	return 0;
 932error:
 933	page->mapping = NULL;
 934	/* Leave page->index set: truncation relies upon it */
 935	put_page(page);
 936	return error;
 937}
 938ALLOW_ERROR_INJECTION(__add_to_page_cache_locked, ERRNO);
 939
 940/**
 941 * add_to_page_cache_locked - add a locked page to the pagecache
 942 * @page:	page to add
 943 * @mapping:	the page's address_space
 944 * @offset:	page index
 945 * @gfp_mask:	page allocation mode
 946 *
 947 * This function is used to add a page to the pagecache. It must be locked.
 948 * This function does not add the page to the LRU.  The caller must do that.
 949 *
 950 * Return: %0 on success, negative error code otherwise.
 951 */
 952int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
 953		pgoff_t offset, gfp_t gfp_mask)
 954{
 955	return __add_to_page_cache_locked(page, mapping, offset,
 956					  gfp_mask, NULL);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 957}
 958EXPORT_SYMBOL(add_to_page_cache_locked);
 959
 960int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
 961				pgoff_t offset, gfp_t gfp_mask)
 962{
 963	void *shadow = NULL;
 964	int ret;
 965
 966	__SetPageLocked(page);
 967	ret = __add_to_page_cache_locked(page, mapping, offset,
 968					 gfp_mask, &shadow);
 969	if (unlikely(ret))
 970		__ClearPageLocked(page);
 971	else {
 972		/*
 973		 * The page might have been evicted from cache only
 974		 * recently, in which case it should be activated like
 975		 * any other repeatedly accessed page.
 976		 * The exception is pages getting rewritten; evicting other
 977		 * data from the working set, only to cache data that will
 978		 * get overwritten with something else, is a waste of memory.
 979		 */
 980		WARN_ON_ONCE(PageActive(page));
 981		if (!(gfp_mask & __GFP_WRITE) && shadow)
 982			workingset_refault(page, shadow);
 983		lru_cache_add(page);
 984	}
 985	return ret;
 986}
 987EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
 988
 989#ifdef CONFIG_NUMA
 990struct page *__page_cache_alloc(gfp_t gfp)
 991{
 992	int n;
 993	struct page *page;
 994
 995	if (cpuset_do_page_mem_spread()) {
 996		unsigned int cpuset_mems_cookie;
 997		do {
 998			cpuset_mems_cookie = read_mems_allowed_begin();
 999			n = cpuset_mem_spread_node();
1000			page = __alloc_pages_node(n, gfp, 0);
1001		} while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
1002
1003		return page;
1004	}
1005	return alloc_pages(gfp, 0);
1006}
1007EXPORT_SYMBOL(__page_cache_alloc);
1008#endif
1009
1010/*
1011 * In order to wait for pages to become available there must be
1012 * waitqueues associated with pages. By using a hash table of
1013 * waitqueues where the bucket discipline is to maintain all
1014 * waiters on the same queue and wake all when any of the pages
1015 * become available, and for the woken contexts to check to be
1016 * sure the appropriate page became available, this saves space
1017 * at a cost of "thundering herd" phenomena during rare hash
1018 * collisions.
1019 */
1020#define PAGE_WAIT_TABLE_BITS 8
1021#define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1022static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
1023
1024static wait_queue_head_t *page_waitqueue(struct page *page)
1025{
1026	return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
1027}
1028
1029void __init pagecache_init(void)
1030{
1031	int i;
1032
1033	for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1034		init_waitqueue_head(&page_wait_table[i]);
1035
1036	page_writeback_init();
1037}
1038
1039/*
1040 * The page wait code treats the "wait->flags" somewhat unusually, because
1041 * we have multiple different kinds of waits, not just the usual "exclusive"
1042 * one.
1043 *
1044 * We have:
1045 *
1046 *  (a) no special bits set:
1047 *
1048 *	We're just waiting for the bit to be released, and when a waker
1049 *	calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1050 *	and remove it from the wait queue.
1051 *
1052 *	Simple and straightforward.
1053 *
1054 *  (b) WQ_FLAG_EXCLUSIVE:
1055 *
1056 *	The waiter is waiting to get the lock, and only one waiter should
1057 *	be woken up to avoid any thundering herd behavior. We'll set the
1058 *	WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1059 *
1060 *	This is the traditional exclusive wait.
1061 *
1062 *  (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1063 *
1064 *	The waiter is waiting to get the bit, and additionally wants the
1065 *	lock to be transferred to it for fair lock behavior. If the lock
1066 *	cannot be taken, we stop walking the wait queue without waking
1067 *	the waiter.
1068 *
1069 *	This is the "fair lock handoff" case, and in addition to setting
1070 *	WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1071 *	that it now has the lock.
1072 */
1073static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1074{
1075	unsigned int flags;
1076	struct wait_page_key *key = arg;
1077	struct wait_page_queue *wait_page
1078		= container_of(wait, struct wait_page_queue, wait);
1079
1080	if (!wake_page_match(wait_page, key))
1081		return 0;
1082
1083	/*
1084	 * If it's a lock handoff wait, we get the bit for it, and
1085	 * stop walking (and do not wake it up) if we can't.
1086	 */
1087	flags = wait->flags;
1088	if (flags & WQ_FLAG_EXCLUSIVE) {
1089		if (test_bit(key->bit_nr, &key->page->flags))
1090			return -1;
1091		if (flags & WQ_FLAG_CUSTOM) {
1092			if (test_and_set_bit(key->bit_nr, &key->page->flags))
1093				return -1;
1094			flags |= WQ_FLAG_DONE;
1095		}
1096	}
1097
1098	/*
1099	 * We are holding the wait-queue lock, but the waiter that
1100	 * is waiting for this will be checking the flags without
1101	 * any locking.
1102	 *
1103	 * So update the flags atomically, and wake up the waiter
1104	 * afterwards to avoid any races. This store-release pairs
1105	 * with the load-acquire in wait_on_page_bit_common().
1106	 */
1107	smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1108	wake_up_state(wait->private, mode);
1109
1110	/*
1111	 * Ok, we have successfully done what we're waiting for,
1112	 * and we can unconditionally remove the wait entry.
1113	 *
1114	 * Note that this pairs with the "finish_wait()" in the
1115	 * waiter, and has to be the absolute last thing we do.
1116	 * After this list_del_init(&wait->entry) the wait entry
1117	 * might be de-allocated and the process might even have
1118	 * exited.
1119	 */
1120	list_del_init_careful(&wait->entry);
1121	return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1122}
1123
1124static void wake_up_page_bit(struct page *page, int bit_nr)
1125{
1126	wait_queue_head_t *q = page_waitqueue(page);
1127	struct wait_page_key key;
1128	unsigned long flags;
1129	wait_queue_entry_t bookmark;
1130
1131	key.page = page;
1132	key.bit_nr = bit_nr;
1133	key.page_match = 0;
1134
1135	bookmark.flags = 0;
1136	bookmark.private = NULL;
1137	bookmark.func = NULL;
1138	INIT_LIST_HEAD(&bookmark.entry);
1139
1140	spin_lock_irqsave(&q->lock, flags);
1141	__wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1142
1143	while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1144		/*
1145		 * Take a breather from holding the lock,
1146		 * allow pages that finish wake up asynchronously
1147		 * to acquire the lock and remove themselves
1148		 * from wait queue
1149		 */
1150		spin_unlock_irqrestore(&q->lock, flags);
1151		cpu_relax();
1152		spin_lock_irqsave(&q->lock, flags);
1153		__wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1154	}
1155
1156	/*
1157	 * It is possible for other pages to have collided on the waitqueue
1158	 * hash, so in that case check for a page match. That prevents a long-
1159	 * term waiter
1160	 *
1161	 * It is still possible to miss a case here, when we woke page waiters
1162	 * and removed them from the waitqueue, but there are still other
1163	 * page waiters.
1164	 */
1165	if (!waitqueue_active(q) || !key.page_match) {
1166		ClearPageWaiters(page);
1167		/*
1168		 * It's possible to miss clearing Waiters here, when we woke
1169		 * our page waiters, but the hashed waitqueue has waiters for
1170		 * other pages on it.
1171		 *
1172		 * That's okay, it's a rare case. The next waker will clear it.
1173		 */
1174	}
1175	spin_unlock_irqrestore(&q->lock, flags);
1176}
1177
1178static void wake_up_page(struct page *page, int bit)
1179{
1180	if (!PageWaiters(page))
1181		return;
1182	wake_up_page_bit(page, bit);
1183}
1184
1185/*
1186 * A choice of three behaviors for wait_on_page_bit_common():
1187 */
1188enum behavior {
1189	EXCLUSIVE,	/* Hold ref to page and take the bit when woken, like
1190			 * __lock_page() waiting on then setting PG_locked.
1191			 */
1192	SHARED,		/* Hold ref to page and check the bit when woken, like
1193			 * wait_on_page_writeback() waiting on PG_writeback.
1194			 */
1195	DROP,		/* Drop ref to page before wait, no check when woken,
1196			 * like put_and_wait_on_page_locked() on PG_locked.
1197			 */
1198};
1199
1200/*
1201 * Attempt to check (or get) the page bit, and mark us done
1202 * if successful.
1203 */
1204static inline bool trylock_page_bit_common(struct page *page, int bit_nr,
1205					struct wait_queue_entry *wait)
1206{
1207	if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1208		if (test_and_set_bit(bit_nr, &page->flags))
1209			return false;
1210	} else if (test_bit(bit_nr, &page->flags))
1211		return false;
1212
1213	wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1214	return true;
1215}
1216
1217/* How many times do we accept lock stealing from under a waiter? */
1218int sysctl_page_lock_unfairness = 5;
1219
1220static inline int wait_on_page_bit_common(wait_queue_head_t *q,
1221	struct page *page, int bit_nr, int state, enum behavior behavior)
1222{
1223	int unfairness = sysctl_page_lock_unfairness;
1224	struct wait_page_queue wait_page;
1225	wait_queue_entry_t *wait = &wait_page.wait;
1226	bool thrashing = false;
1227	bool delayacct = false;
1228	unsigned long pflags;
1229
1230	if (bit_nr == PG_locked &&
1231	    !PageUptodate(page) && PageWorkingset(page)) {
1232		if (!PageSwapBacked(page)) {
1233			delayacct_thrashing_start();
1234			delayacct = true;
1235		}
1236		psi_memstall_enter(&pflags);
1237		thrashing = true;
1238	}
1239
1240	init_wait(wait);
1241	wait->func = wake_page_function;
1242	wait_page.page = page;
1243	wait_page.bit_nr = bit_nr;
1244
1245repeat:
1246	wait->flags = 0;
1247	if (behavior == EXCLUSIVE) {
1248		wait->flags = WQ_FLAG_EXCLUSIVE;
1249		if (--unfairness < 0)
1250			wait->flags |= WQ_FLAG_CUSTOM;
1251	}
1252
1253	/*
1254	 * Do one last check whether we can get the
1255	 * page bit synchronously.
1256	 *
1257	 * Do the SetPageWaiters() marking before that
1258	 * to let any waker we _just_ missed know they
1259	 * need to wake us up (otherwise they'll never
1260	 * even go to the slow case that looks at the
1261	 * page queue), and add ourselves to the wait
1262	 * queue if we need to sleep.
1263	 *
1264	 * This part needs to be done under the queue
1265	 * lock to avoid races.
1266	 */
1267	spin_lock_irq(&q->lock);
1268	SetPageWaiters(page);
1269	if (!trylock_page_bit_common(page, bit_nr, wait))
1270		__add_wait_queue_entry_tail(q, wait);
1271	spin_unlock_irq(&q->lock);
1272
1273	/*
1274	 * From now on, all the logic will be based on
1275	 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1276	 * see whether the page bit testing has already
1277	 * been done by the wake function.
1278	 *
1279	 * We can drop our reference to the page.
1280	 */
1281	if (behavior == DROP)
1282		put_page(page);
1283
1284	/*
1285	 * Note that until the "finish_wait()", or until
1286	 * we see the WQ_FLAG_WOKEN flag, we need to
1287	 * be very careful with the 'wait->flags', because
1288	 * we may race with a waker that sets them.
1289	 */
1290	for (;;) {
1291		unsigned int flags;
1292
1293		set_current_state(state);
1294
1295		/* Loop until we've been woken or interrupted */
1296		flags = smp_load_acquire(&wait->flags);
1297		if (!(flags & WQ_FLAG_WOKEN)) {
1298			if (signal_pending_state(state, current))
1299				break;
1300
1301			io_schedule();
1302			continue;
1303		}
1304
1305		/* If we were non-exclusive, we're done */
1306		if (behavior != EXCLUSIVE)
1307			break;
1308
1309		/* If the waker got the lock for us, we're done */
1310		if (flags & WQ_FLAG_DONE)
1311			break;
1312
1313		/*
1314		 * Otherwise, if we're getting the lock, we need to
1315		 * try to get it ourselves.
1316		 *
1317		 * And if that fails, we'll have to retry this all.
1318		 */
1319		if (unlikely(test_and_set_bit(bit_nr, &page->flags)))
1320			goto repeat;
1321
1322		wait->flags |= WQ_FLAG_DONE;
1323		break;
1324	}
1325
1326	/*
1327	 * If a signal happened, this 'finish_wait()' may remove the last
1328	 * waiter from the wait-queues, but the PageWaiters bit will remain
1329	 * set. That's ok. The next wakeup will take care of it, and trying
1330	 * to do it here would be difficult and prone to races.
1331	 */
1332	finish_wait(q, wait);
1333
1334	if (thrashing) {
1335		if (delayacct)
1336			delayacct_thrashing_end();
1337		psi_memstall_leave(&pflags);
1338	}
1339
1340	/*
1341	 * NOTE! The wait->flags weren't stable until we've done the
1342	 * 'finish_wait()', and we could have exited the loop above due
1343	 * to a signal, and had a wakeup event happen after the signal
1344	 * test but before the 'finish_wait()'.
1345	 *
1346	 * So only after the finish_wait() can we reliably determine
1347	 * if we got woken up or not, so we can now figure out the final
1348	 * return value based on that state without races.
1349	 *
1350	 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1351	 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1352	 */
1353	if (behavior == EXCLUSIVE)
1354		return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1355
1356	return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1357}
1358
1359void wait_on_page_bit(struct page *page, int bit_nr)
1360{
1361	wait_queue_head_t *q = page_waitqueue(page);
1362	wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
 
 
 
1363}
1364EXPORT_SYMBOL(wait_on_page_bit);
1365
1366int wait_on_page_bit_killable(struct page *page, int bit_nr)
1367{
1368	wait_queue_head_t *q = page_waitqueue(page);
1369	return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED);
1370}
1371EXPORT_SYMBOL(wait_on_page_bit_killable);
1372
1373/**
1374 * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1375 * @page: The page to wait for.
1376 * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1377 *
1378 * The caller should hold a reference on @page.  They expect the page to
1379 * become unlocked relatively soon, but do not wish to hold up migration
1380 * (for example) by holding the reference while waiting for the page to
1381 * come unlocked.  After this function returns, the caller should not
1382 * dereference @page.
1383 *
1384 * Return: 0 if the page was unlocked or -EINTR if interrupted by a signal.
1385 */
1386int put_and_wait_on_page_locked(struct page *page, int state)
1387{
1388	wait_queue_head_t *q;
1389
1390	page = compound_head(page);
1391	q = page_waitqueue(page);
1392	return wait_on_page_bit_common(q, page, PG_locked, state, DROP);
1393}
1394
1395/**
1396 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1397 * @page: Page defining the wait queue of interest
1398 * @waiter: Waiter to add to the queue
1399 *
1400 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1401 */
1402void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
1403{
1404	wait_queue_head_t *q = page_waitqueue(page);
1405	unsigned long flags;
1406
1407	spin_lock_irqsave(&q->lock, flags);
1408	__add_wait_queue_entry_tail(q, waiter);
1409	SetPageWaiters(page);
1410	spin_unlock_irqrestore(&q->lock, flags);
1411}
1412EXPORT_SYMBOL_GPL(add_page_wait_queue);
1413
1414#ifndef clear_bit_unlock_is_negative_byte
1415
1416/*
1417 * PG_waiters is the high bit in the same byte as PG_lock.
1418 *
1419 * On x86 (and on many other architectures), we can clear PG_lock and
1420 * test the sign bit at the same time. But if the architecture does
1421 * not support that special operation, we just do this all by hand
1422 * instead.
1423 *
1424 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1425 * being cleared, but a memory barrier should be unnecessary since it is
1426 * in the same byte as PG_locked.
1427 */
1428static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1429{
1430	clear_bit_unlock(nr, mem);
1431	/* smp_mb__after_atomic(); */
1432	return test_bit(PG_waiters, mem);
1433}
1434
1435#endif
1436
1437/**
1438 * unlock_page - unlock a locked page
1439 * @page: the page
1440 *
1441 * Unlocks the page and wakes up sleepers in wait_on_page_locked().
1442 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1443 * mechanism between PageLocked pages and PageWriteback pages is shared.
1444 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1445 *
1446 * Note that this depends on PG_waiters being the sign bit in the byte
1447 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1448 * clear the PG_locked bit and test PG_waiters at the same time fairly
1449 * portably (architectures that do LL/SC can test any bit, while x86 can
1450 * test the sign bit).
1451 */
1452void unlock_page(struct page *page)
1453{
1454	BUILD_BUG_ON(PG_waiters != 7);
1455	page = compound_head(page);
1456	VM_BUG_ON_PAGE(!PageLocked(page), page);
1457	if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
1458		wake_up_page_bit(page, PG_locked);
1459}
1460EXPORT_SYMBOL(unlock_page);
1461
1462/**
1463 * end_page_private_2 - Clear PG_private_2 and release any waiters
1464 * @page: The page
1465 *
1466 * Clear the PG_private_2 bit on a page and wake up any sleepers waiting for
1467 * this.  The page ref held for PG_private_2 being set is released.
1468 *
1469 * This is, for example, used when a netfs page is being written to a local
1470 * disk cache, thereby allowing writes to the cache for the same page to be
1471 * serialised.
1472 */
1473void end_page_private_2(struct page *page)
1474{
1475	page = compound_head(page);
1476	VM_BUG_ON_PAGE(!PagePrivate2(page), page);
1477	clear_bit_unlock(PG_private_2, &page->flags);
1478	wake_up_page_bit(page, PG_private_2);
1479	put_page(page);
1480}
1481EXPORT_SYMBOL(end_page_private_2);
1482
1483/**
1484 * wait_on_page_private_2 - Wait for PG_private_2 to be cleared on a page
1485 * @page: The page to wait on
1486 *
1487 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a page.
1488 */
1489void wait_on_page_private_2(struct page *page)
1490{
1491	page = compound_head(page);
1492	while (PagePrivate2(page))
1493		wait_on_page_bit(page, PG_private_2);
1494}
1495EXPORT_SYMBOL(wait_on_page_private_2);
1496
1497/**
1498 * wait_on_page_private_2_killable - Wait for PG_private_2 to be cleared on a page
1499 * @page: The page to wait on
1500 *
1501 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a page or until a
1502 * fatal signal is received by the calling task.
1503 *
1504 * Return:
1505 * - 0 if successful.
1506 * - -EINTR if a fatal signal was encountered.
1507 */
1508int wait_on_page_private_2_killable(struct page *page)
1509{
1510	int ret = 0;
1511
1512	page = compound_head(page);
1513	while (PagePrivate2(page)) {
1514		ret = wait_on_page_bit_killable(page, PG_private_2);
1515		if (ret < 0)
1516			break;
1517	}
1518
1519	return ret;
1520}
1521EXPORT_SYMBOL(wait_on_page_private_2_killable);
1522
1523/**
1524 * end_page_writeback - end writeback against a page
1525 * @page: the page
1526 */
1527void end_page_writeback(struct page *page)
1528{
1529	/*
1530	 * TestClearPageReclaim could be used here but it is an atomic
1531	 * operation and overkill in this particular case. Failing to
1532	 * shuffle a page marked for immediate reclaim is too mild to
1533	 * justify taking an atomic operation penalty at the end of
1534	 * ever page writeback.
1535	 */
1536	if (PageReclaim(page)) {
1537		ClearPageReclaim(page);
1538		rotate_reclaimable_page(page);
1539	}
1540
1541	/*
1542	 * Writeback does not hold a page reference of its own, relying
1543	 * on truncation to wait for the clearing of PG_writeback.
1544	 * But here we must make sure that the page is not freed and
1545	 * reused before the wake_up_page().
1546	 */
1547	get_page(page);
1548	if (!test_clear_page_writeback(page))
1549		BUG();
1550
1551	smp_mb__after_atomic();
1552	wake_up_page(page, PG_writeback);
1553	put_page(page);
1554}
1555EXPORT_SYMBOL(end_page_writeback);
1556
1557/*
1558 * After completing I/O on a page, call this routine to update the page
1559 * flags appropriately
1560 */
1561void page_endio(struct page *page, bool is_write, int err)
1562{
1563	if (!is_write) {
1564		if (!err) {
1565			SetPageUptodate(page);
1566		} else {
1567			ClearPageUptodate(page);
1568			SetPageError(page);
1569		}
1570		unlock_page(page);
1571	} else {
1572		if (err) {
1573			struct address_space *mapping;
1574
1575			SetPageError(page);
1576			mapping = page_mapping(page);
1577			if (mapping)
1578				mapping_set_error(mapping, err);
1579		}
1580		end_page_writeback(page);
1581	}
1582}
1583EXPORT_SYMBOL_GPL(page_endio);
1584
1585/**
1586 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1587 * @__page: the page to lock
1588 */
1589void __lock_page(struct page *__page)
1590{
1591	struct page *page = compound_head(__page);
1592	wait_queue_head_t *q = page_waitqueue(page);
1593	wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE,
1594				EXCLUSIVE);
1595}
1596EXPORT_SYMBOL(__lock_page);
1597
1598int __lock_page_killable(struct page *__page)
1599{
1600	struct page *page = compound_head(__page);
1601	wait_queue_head_t *q = page_waitqueue(page);
1602	return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE,
1603					EXCLUSIVE);
1604}
1605EXPORT_SYMBOL_GPL(__lock_page_killable);
1606
1607int __lock_page_async(struct page *page, struct wait_page_queue *wait)
1608{
1609	struct wait_queue_head *q = page_waitqueue(page);
1610	int ret = 0;
1611
1612	wait->page = page;
1613	wait->bit_nr = PG_locked;
1614
1615	spin_lock_irq(&q->lock);
1616	__add_wait_queue_entry_tail(q, &wait->wait);
1617	SetPageWaiters(page);
1618	ret = !trylock_page(page);
1619	/*
1620	 * If we were successful now, we know we're still on the
1621	 * waitqueue as we're still under the lock. This means it's
1622	 * safe to remove and return success, we know the callback
1623	 * isn't going to trigger.
1624	 */
1625	if (!ret)
1626		__remove_wait_queue(q, &wait->wait);
1627	else
1628		ret = -EIOCBQUEUED;
1629	spin_unlock_irq(&q->lock);
1630	return ret;
1631}
 
1632
1633/*
1634 * Return values:
1635 * 1 - page is locked; mmap_lock is still held.
1636 * 0 - page is not locked.
1637 *     mmap_lock has been released (mmap_read_unlock(), unless flags had both
1638 *     FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1639 *     which case mmap_lock is still held.
1640 *
1641 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1642 * with the page locked and the mmap_lock unperturbed.
1643 */
1644int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1645			 unsigned int flags)
1646{
1647	if (fault_flag_allow_retry_first(flags)) {
1648		/*
1649		 * CAUTION! In this case, mmap_lock is not released
1650		 * even though return 0.
1651		 */
1652		if (flags & FAULT_FLAG_RETRY_NOWAIT)
1653			return 0;
1654
1655		mmap_read_unlock(mm);
1656		if (flags & FAULT_FLAG_KILLABLE)
1657			wait_on_page_locked_killable(page);
1658		else
1659			wait_on_page_locked(page);
1660		return 0;
1661	}
1662	if (flags & FAULT_FLAG_KILLABLE) {
1663		int ret;
1664
1665		ret = __lock_page_killable(page);
1666		if (ret) {
1667			mmap_read_unlock(mm);
1668			return 0;
1669		}
1670	} else {
1671		__lock_page(page);
1672	}
1673	return 1;
1674
1675}
1676
1677/**
1678 * page_cache_next_miss() - Find the next gap in the page cache.
1679 * @mapping: Mapping.
1680 * @index: Index.
1681 * @max_scan: Maximum range to search.
1682 *
1683 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1684 * gap with the lowest index.
1685 *
1686 * This function may be called under the rcu_read_lock.  However, this will
1687 * not atomically search a snapshot of the cache at a single point in time.
1688 * For example, if a gap is created at index 5, then subsequently a gap is
1689 * created at index 10, page_cache_next_miss covering both indices may
1690 * return 10 if called under the rcu_read_lock.
1691 *
1692 * Return: The index of the gap if found, otherwise an index outside the
1693 * range specified (in which case 'return - index >= max_scan' will be true).
1694 * In the rare case of index wrap-around, 0 will be returned.
1695 */
1696pgoff_t page_cache_next_miss(struct address_space *mapping,
1697			     pgoff_t index, unsigned long max_scan)
1698{
1699	XA_STATE(xas, &mapping->i_pages, index);
1700
1701	while (max_scan--) {
1702		void *entry = xas_next(&xas);
1703		if (!entry || xa_is_value(entry))
1704			break;
1705		if (xas.xa_index == 0)
1706			break;
1707	}
1708
1709	return xas.xa_index;
1710}
1711EXPORT_SYMBOL(page_cache_next_miss);
1712
1713/**
1714 * page_cache_prev_miss() - Find the previous gap in the page cache.
1715 * @mapping: Mapping.
1716 * @index: Index.
1717 * @max_scan: Maximum range to search.
1718 *
1719 * Search the range [max(index - max_scan + 1, 0), index] for the
1720 * gap with the highest index.
1721 *
1722 * This function may be called under the rcu_read_lock.  However, this will
1723 * not atomically search a snapshot of the cache at a single point in time.
1724 * For example, if a gap is created at index 10, then subsequently a gap is
1725 * created at index 5, page_cache_prev_miss() covering both indices may
1726 * return 5 if called under the rcu_read_lock.
1727 *
1728 * Return: The index of the gap if found, otherwise an index outside the
1729 * range specified (in which case 'index - return >= max_scan' will be true).
1730 * In the rare case of wrap-around, ULONG_MAX will be returned.
1731 */
1732pgoff_t page_cache_prev_miss(struct address_space *mapping,
1733			     pgoff_t index, unsigned long max_scan)
1734{
1735	XA_STATE(xas, &mapping->i_pages, index);
1736
1737	while (max_scan--) {
1738		void *entry = xas_prev(&xas);
1739		if (!entry || xa_is_value(entry))
1740			break;
1741		if (xas.xa_index == ULONG_MAX)
1742			break;
1743	}
1744
1745	return xas.xa_index;
1746}
1747EXPORT_SYMBOL(page_cache_prev_miss);
1748
1749/*
1750 * mapping_get_entry - Get a page cache entry.
1751 * @mapping: the address_space to search
1752 * @index: The page cache index.
1753 *
1754 * Looks up the page cache slot at @mapping & @index.  If there is a
1755 * page cache page, the head page is returned with an increased refcount.
1756 *
1757 * If the slot holds a shadow entry of a previously evicted page, or a
1758 * swap entry from shmem/tmpfs, it is returned.
1759 *
1760 * Return: The head page or shadow entry, %NULL if nothing is found.
1761 */
1762static struct page *mapping_get_entry(struct address_space *mapping,
1763		pgoff_t index)
1764{
1765	XA_STATE(xas, &mapping->i_pages, index);
1766	struct page *page;
1767
1768	rcu_read_lock();
1769repeat:
1770	xas_reset(&xas);
1771	page = xas_load(&xas);
1772	if (xas_retry(&xas, page))
1773		goto repeat;
1774	/*
1775	 * A shadow entry of a recently evicted page, or a swap entry from
1776	 * shmem/tmpfs.  Return it without attempting to raise page count.
1777	 */
1778	if (!page || xa_is_value(page))
1779		goto out;
1780
1781	if (!page_cache_get_speculative(page))
1782		goto repeat;
 
 
 
 
 
1783
1784	/*
1785	 * Has the page moved or been split?
1786	 * This is part of the lockless pagecache protocol. See
1787	 * include/linux/pagemap.h for details.
1788	 */
1789	if (unlikely(page != xas_reload(&xas))) {
1790		put_page(page);
1791		goto repeat;
 
1792	}
1793out:
1794	rcu_read_unlock();
1795
1796	return page;
1797}
 
1798
1799/**
1800 * pagecache_get_page - Find and get a reference to a page.
1801 * @mapping: The address_space to search.
1802 * @index: The page index.
1803 * @fgp_flags: %FGP flags modify how the page is returned.
1804 * @gfp_mask: Memory allocation flags to use if %FGP_CREAT is specified.
1805 *
1806 * Looks up the page cache entry at @mapping & @index.
1807 *
1808 * @fgp_flags can be zero or more of these flags:
1809 *
1810 * * %FGP_ACCESSED - The page will be marked accessed.
1811 * * %FGP_LOCK - The page is returned locked.
1812 * * %FGP_HEAD - If the page is present and a THP, return the head page
1813 *   rather than the exact page specified by the index.
1814 * * %FGP_ENTRY - If there is a shadow / swap / DAX entry, return it
1815 *   instead of allocating a new page to replace it.
1816 * * %FGP_CREAT - If no page is present then a new page is allocated using
1817 *   @gfp_mask and added to the page cache and the VM's LRU list.
1818 *   The page is returned locked and with an increased refcount.
1819 * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
1820 *   page is already in cache.  If the page was allocated, unlock it before
1821 *   returning so the caller can do the same dance.
1822 * * %FGP_WRITE - The page will be written
1823 * * %FGP_NOFS - __GFP_FS will get cleared in gfp mask
1824 * * %FGP_NOWAIT - Don't get blocked by page lock
1825 *
1826 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1827 * if the %GFP flags specified for %FGP_CREAT are atomic.
1828 *
1829 * If there is a page cache page, it is returned with an increased refcount.
 
1830 *
1831 * Return: The found page or %NULL otherwise.
1832 */
1833struct page *pagecache_get_page(struct address_space *mapping, pgoff_t index,
1834		int fgp_flags, gfp_t gfp_mask)
1835{
1836	struct page *page;
1837
1838repeat:
1839	page = mapping_get_entry(mapping, index);
1840	if (xa_is_value(page)) {
1841		if (fgp_flags & FGP_ENTRY)
1842			return page;
1843		page = NULL;
1844	}
1845	if (!page)
1846		goto no_page;
1847
1848	if (fgp_flags & FGP_LOCK) {
1849		if (fgp_flags & FGP_NOWAIT) {
1850			if (!trylock_page(page)) {
1851				put_page(page);
1852				return NULL;
1853			}
1854		} else {
1855			lock_page(page);
1856		}
1857
1858		/* Has the page been truncated? */
1859		if (unlikely(page->mapping != mapping)) {
1860			unlock_page(page);
1861			put_page(page);
1862			goto repeat;
1863		}
1864		VM_BUG_ON_PAGE(!thp_contains(page, index), page);
1865	}
1866
1867	if (fgp_flags & FGP_ACCESSED)
1868		mark_page_accessed(page);
1869	else if (fgp_flags & FGP_WRITE) {
1870		/* Clear idle flag for buffer write */
1871		if (page_is_idle(page))
1872			clear_page_idle(page);
1873	}
1874	if (!(fgp_flags & FGP_HEAD))
1875		page = find_subpage(page, index);
1876
1877no_page:
1878	if (!page && (fgp_flags & FGP_CREAT)) {
1879		int err;
1880		if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
1881			gfp_mask |= __GFP_WRITE;
1882		if (fgp_flags & FGP_NOFS)
1883			gfp_mask &= ~__GFP_FS;
1884
1885		page = __page_cache_alloc(gfp_mask);
1886		if (!page)
1887			return NULL;
1888
1889		if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1890			fgp_flags |= FGP_LOCK;
1891
1892		/* Init accessed so avoid atomic mark_page_accessed later */
1893		if (fgp_flags & FGP_ACCESSED)
1894			__SetPageReferenced(page);
1895
1896		err = add_to_page_cache_lru(page, mapping, index, gfp_mask);
1897		if (unlikely(err)) {
1898			put_page(page);
1899			page = NULL;
1900			if (err == -EEXIST)
1901				goto repeat;
1902		}
1903
1904		/*
1905		 * add_to_page_cache_lru locks the page, and for mmap we expect
1906		 * an unlocked page.
1907		 */
1908		if (page && (fgp_flags & FGP_FOR_MMAP))
1909			unlock_page(page);
1910	}
1911
1912	return page;
1913}
1914EXPORT_SYMBOL(pagecache_get_page);
1915
1916static inline struct page *find_get_entry(struct xa_state *xas, pgoff_t max,
1917		xa_mark_t mark)
1918{
1919	struct page *page;
1920
1921retry:
1922	if (mark == XA_PRESENT)
1923		page = xas_find(xas, max);
1924	else
1925		page = xas_find_marked(xas, max, mark);
1926
1927	if (xas_retry(xas, page))
1928		goto retry;
1929	/*
1930	 * A shadow entry of a recently evicted page, a swap
1931	 * entry from shmem/tmpfs or a DAX entry.  Return it
1932	 * without attempting to raise page count.
1933	 */
1934	if (!page || xa_is_value(page))
1935		return page;
1936
1937	if (!page_cache_get_speculative(page))
1938		goto reset;
1939
1940	/* Has the page moved or been split? */
1941	if (unlikely(page != xas_reload(xas))) {
1942		put_page(page);
1943		goto reset;
1944	}
1945
1946	return page;
1947reset:
1948	xas_reset(xas);
1949	goto retry;
1950}
 
1951
1952/**
1953 * find_get_entries - gang pagecache lookup
1954 * @mapping:	The address_space to search
1955 * @start:	The starting page cache index
1956 * @end:	The final page index (inclusive).
1957 * @pvec:	Where the resulting entries are placed.
1958 * @indices:	The cache indices corresponding to the entries in @entries
1959 *
1960 * find_get_entries() will search for and return a batch of entries in
1961 * the mapping.  The entries are placed in @pvec.  find_get_entries()
1962 * takes a reference on any actual pages it returns.
1963 *
1964 * The search returns a group of mapping-contiguous page cache entries
1965 * with ascending indexes.  There may be holes in the indices due to
1966 * not-present pages.
 
1967 *
1968 * Any shadow entries of evicted pages, or swap entries from
1969 * shmem/tmpfs, are included in the returned array.
1970 *
1971 * If it finds a Transparent Huge Page, head or tail, find_get_entries()
1972 * stops at that page: the caller is likely to have a better way to handle
1973 * the compound page as a whole, and then skip its extent, than repeatedly
1974 * calling find_get_entries() to return all its tails.
1975 *
1976 * Return: the number of pages and shadow entries which were found.
1977 */
1978unsigned find_get_entries(struct address_space *mapping, pgoff_t start,
1979		pgoff_t end, struct pagevec *pvec, pgoff_t *indices)
1980{
1981	XA_STATE(xas, &mapping->i_pages, start);
1982	struct page *page;
1983	unsigned int ret = 0;
1984	unsigned nr_entries = PAGEVEC_SIZE;
1985
1986	rcu_read_lock();
1987	while ((page = find_get_entry(&xas, end, XA_PRESENT))) {
 
 
1988		/*
1989		 * Terminate early on finding a THP, to allow the caller to
1990		 * handle it all at once; but continue if this is hugetlbfs.
 
 
1991		 */
1992		if (!xa_is_value(page) && PageTransHuge(page) &&
1993				!PageHuge(page)) {
1994			page = find_subpage(page, xas.xa_index);
1995			nr_entries = ret + 1;
1996		}
1997
1998		indices[ret] = xas.xa_index;
1999		pvec->pages[ret] = page;
2000		if (++ret == nr_entries)
2001			break;
2002	}
2003	rcu_read_unlock();
2004
2005	pvec->nr = ret;
2006	return ret;
2007}
2008
2009/**
2010 * find_lock_entries - Find a batch of pagecache entries.
2011 * @mapping:	The address_space to search.
2012 * @start:	The starting page cache index.
2013 * @end:	The final page index (inclusive).
2014 * @pvec:	Where the resulting entries are placed.
2015 * @indices:	The cache indices of the entries in @pvec.
2016 *
2017 * find_lock_entries() will return a batch of entries from @mapping.
2018 * Swap, shadow and DAX entries are included.  Pages are returned
2019 * locked and with an incremented refcount.  Pages which are locked by
2020 * somebody else or under writeback are skipped.  Only the head page of
2021 * a THP is returned.  Pages which are partially outside the range are
2022 * not returned.
2023 *
2024 * The entries have ascending indexes.  The indices may not be consecutive
2025 * due to not-present entries, THP pages, pages which could not be locked
2026 * or pages under writeback.
2027 *
2028 * Return: The number of entries which were found.
2029 */
2030unsigned find_lock_entries(struct address_space *mapping, pgoff_t start,
2031		pgoff_t end, struct pagevec *pvec, pgoff_t *indices)
2032{
2033	XA_STATE(xas, &mapping->i_pages, start);
2034	struct page *page;
2035
2036	rcu_read_lock();
2037	while ((page = find_get_entry(&xas, end, XA_PRESENT))) {
2038		if (!xa_is_value(page)) {
2039			if (page->index < start)
2040				goto put;
2041			VM_BUG_ON_PAGE(page->index != xas.xa_index, page);
2042			if (page->index + thp_nr_pages(page) - 1 > end)
2043				goto put;
2044			if (!trylock_page(page))
2045				goto put;
2046			if (page->mapping != mapping || PageWriteback(page))
2047				goto unlock;
2048			VM_BUG_ON_PAGE(!thp_contains(page, xas.xa_index),
2049					page);
2050		}
2051		indices[pvec->nr] = xas.xa_index;
2052		if (!pagevec_add(pvec, page))
2053			break;
2054		goto next;
2055unlock:
2056		unlock_page(page);
2057put:
2058		put_page(page);
2059next:
2060		if (!xa_is_value(page) && PageTransHuge(page)) {
2061			unsigned int nr_pages = thp_nr_pages(page);
2062
2063			/* Final THP may cross MAX_LFS_FILESIZE on 32-bit */
2064			xas_set(&xas, page->index + nr_pages);
2065			if (xas.xa_index < nr_pages)
2066				break;
2067		}
2068	}
2069	rcu_read_unlock();
2070
2071	return pagevec_count(pvec);
2072}
 
2073
2074/**
2075 * find_get_pages_range - gang pagecache lookup
2076 * @mapping:	The address_space to search
2077 * @start:	The starting page index
2078 * @end:	The final page index (inclusive)
2079 * @nr_pages:	The maximum number of pages
2080 * @pages:	Where the resulting pages are placed
2081 *
2082 * find_get_pages_range() will search for and return a group of up to @nr_pages
2083 * pages in the mapping starting at index @start and up to index @end
2084 * (inclusive).  The pages are placed at @pages.  find_get_pages_range() takes
2085 * a reference against the returned pages.
2086 *
2087 * The search returns a group of mapping-contiguous pages with ascending
2088 * indexes.  There may be holes in the indices due to not-present pages.
2089 * We also update @start to index the next page for the traversal.
2090 *
2091 * Return: the number of pages which were found. If this number is
2092 * smaller than @nr_pages, the end of specified range has been
2093 * reached.
2094 */
2095unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
2096			      pgoff_t end, unsigned int nr_pages,
2097			      struct page **pages)
2098{
2099	XA_STATE(xas, &mapping->i_pages, *start);
2100	struct page *page;
2101	unsigned ret = 0;
2102
2103	if (unlikely(!nr_pages))
2104		return 0;
2105
2106	rcu_read_lock();
2107	while ((page = find_get_entry(&xas, end, XA_PRESENT))) {
2108		/* Skip over shadow, swap and DAX entries */
2109		if (xa_is_value(page))
 
 
 
2110			continue;
2111
2112		pages[ret] = find_subpage(page, xas.xa_index);
2113		if (++ret == nr_pages) {
2114			*start = xas.xa_index + 1;
2115			goto out;
 
 
 
 
 
 
 
 
 
 
 
 
2116		}
 
 
 
 
 
 
 
 
 
 
 
 
 
2117	}
2118
2119	/*
2120	 * We come here when there is no page beyond @end. We take care to not
2121	 * overflow the index @start as it confuses some of the callers. This
2122	 * breaks the iteration when there is a page at index -1 but that is
2123	 * already broken anyway.
2124	 */
2125	if (end == (pgoff_t)-1)
2126		*start = (pgoff_t)-1;
2127	else
2128		*start = end + 1;
2129out:
2130	rcu_read_unlock();
2131
2132	return ret;
2133}
2134
2135/**
2136 * find_get_pages_contig - gang contiguous pagecache lookup
2137 * @mapping:	The address_space to search
2138 * @index:	The starting page index
2139 * @nr_pages:	The maximum number of pages
2140 * @pages:	Where the resulting pages are placed
2141 *
2142 * find_get_pages_contig() works exactly like find_get_pages(), except
2143 * that the returned number of pages are guaranteed to be contiguous.
2144 *
2145 * Return: the number of pages which were found.
2146 */
2147unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
2148			       unsigned int nr_pages, struct page **pages)
2149{
2150	XA_STATE(xas, &mapping->i_pages, index);
2151	struct page *page;
2152	unsigned int ret = 0;
2153
2154	if (unlikely(!nr_pages))
2155		return 0;
2156
2157	rcu_read_lock();
2158	for (page = xas_load(&xas); page; page = xas_next(&xas)) {
2159		if (xas_retry(&xas, page))
2160			continue;
2161		/*
2162		 * If the entry has been swapped out, we can stop looking.
2163		 * No current caller is looking for DAX entries.
2164		 */
2165		if (xa_is_value(page))
2166			break;
2167
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2168		if (!page_cache_get_speculative(page))
2169			goto retry;
2170
2171		/* Has the page moved or been split? */
2172		if (unlikely(page != xas_reload(&xas)))
2173			goto put_page;
 
 
2174
2175		pages[ret] = find_subpage(page, xas.xa_index);
 
 
 
 
 
 
 
 
 
 
2176		if (++ret == nr_pages)
2177			break;
2178		continue;
2179put_page:
2180		put_page(page);
2181retry:
2182		xas_reset(&xas);
2183	}
2184	rcu_read_unlock();
2185	return ret;
2186}
2187EXPORT_SYMBOL(find_get_pages_contig);
2188
2189/**
2190 * find_get_pages_range_tag - Find and return head pages matching @tag.
2191 * @mapping:	the address_space to search
2192 * @index:	the starting page index
2193 * @end:	The final page index (inclusive)
2194 * @tag:	the tag index
2195 * @nr_pages:	the maximum number of pages
2196 * @pages:	where the resulting pages are placed
2197 *
2198 * Like find_get_pages(), except we only return head pages which are tagged
2199 * with @tag.  @index is updated to the index immediately after the last
2200 * page we return, ready for the next iteration.
2201 *
2202 * Return: the number of pages which were found.
2203 */
2204unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
2205			pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
2206			struct page **pages)
2207{
2208	XA_STATE(xas, &mapping->i_pages, *index);
2209	struct page *page;
2210	unsigned ret = 0;
2211
2212	if (unlikely(!nr_pages))
2213		return 0;
2214
2215	rcu_read_lock();
2216	while ((page = find_get_entry(&xas, end, tag))) {
2217		/*
2218		 * Shadow entries should never be tagged, but this iteration
2219		 * is lockless so there is a window for page reclaim to evict
2220		 * a page we saw tagged.  Skip over it.
2221		 */
2222		if (xa_is_value(page))
2223			continue;
2224
2225		pages[ret] = page;
2226		if (++ret == nr_pages) {
2227			*index = page->index + thp_nr_pages(page);
2228			goto out;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2229		}
 
 
 
 
2230	}
2231
2232	/*
2233	 * We come here when we got to @end. We take care to not overflow the
2234	 * index @index as it confuses some of the callers. This breaks the
2235	 * iteration when there is a page at index -1 but that is already
2236	 * broken anyway.
2237	 */
2238	if (end == (pgoff_t)-1)
2239		*index = (pgoff_t)-1;
2240	else
2241		*index = end + 1;
2242out:
2243	rcu_read_unlock();
2244
 
 
 
2245	return ret;
2246}
2247EXPORT_SYMBOL(find_get_pages_range_tag);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2248
2249/*
2250 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2251 * a _large_ part of the i/o request. Imagine the worst scenario:
2252 *
2253 *      ---R__________________________________________B__________
2254 *         ^ reading here                             ^ bad block(assume 4k)
2255 *
2256 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2257 * => failing the whole request => read(R) => read(R+1) =>
2258 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2259 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2260 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2261 *
2262 * It is going insane. Fix it by quickly scaling down the readahead size.
2263 */
2264static void shrink_readahead_size_eio(struct file_ra_state *ra)
 
2265{
2266	ra->ra_pages /= 4;
2267}
2268
2269/*
2270 * filemap_get_read_batch - Get a batch of pages for read
 
 
 
 
 
 
 
2271 *
2272 * Get a batch of pages which represent a contiguous range of bytes
2273 * in the file.  No tail pages will be returned.  If @index is in the
2274 * middle of a THP, the entire THP will be returned.  The last page in
2275 * the batch may have Readahead set or be not Uptodate so that the
2276 * caller can take the appropriate action.
2277 */
2278static void filemap_get_read_batch(struct address_space *mapping,
2279		pgoff_t index, pgoff_t max, struct pagevec *pvec)
2280{
2281	XA_STATE(xas, &mapping->i_pages, index);
2282	struct page *head;
 
 
 
 
 
 
 
2283
2284	rcu_read_lock();
2285	for (head = xas_load(&xas); head; head = xas_next(&xas)) {
2286		if (xas_retry(&xas, head))
2287			continue;
2288		if (xas.xa_index > max || xa_is_value(head))
2289			break;
2290		if (!page_cache_get_speculative(head))
2291			goto retry;
2292
2293		/* Has the page moved or been split? */
2294		if (unlikely(head != xas_reload(&xas)))
2295			goto put_page;
 
 
2296
2297		if (!pagevec_add(pvec, head))
2298			break;
2299		if (!PageUptodate(head))
2300			break;
2301		if (PageReadahead(head))
2302			break;
2303		xas.xa_index = head->index + thp_nr_pages(head) - 1;
2304		xas.xa_offset = (xas.xa_index >> xas.xa_shift) & XA_CHUNK_MASK;
2305		continue;
2306put_page:
2307		put_page(head);
2308retry:
2309		xas_reset(&xas);
2310	}
2311	rcu_read_unlock();
2312}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2313
2314static int filemap_read_page(struct file *file, struct address_space *mapping,
2315		struct page *page)
2316{
2317	int error;
 
 
2318
2319	/*
2320	 * A previous I/O error may have been due to temporary failures,
2321	 * eg. multipath errors.  PG_error will be set again if readpage
2322	 * fails.
2323	 */
2324	ClearPageError(page);
2325	/* Start the actual read. The read will unlock the page. */
2326	error = mapping->a_ops->readpage(file, page);
2327	if (error)
2328		return error;
2329
2330	error = wait_on_page_locked_killable(page);
2331	if (error)
2332		return error;
2333	if (PageUptodate(page))
2334		return 0;
2335	shrink_readahead_size_eio(&file->f_ra);
2336	return -EIO;
2337}
2338
2339static bool filemap_range_uptodate(struct address_space *mapping,
2340		loff_t pos, struct iov_iter *iter, struct page *page)
2341{
2342	int count;
 
 
 
2343
2344	if (PageUptodate(page))
2345		return true;
2346	/* pipes can't handle partially uptodate pages */
2347	if (iov_iter_is_pipe(iter))
2348		return false;
2349	if (!mapping->a_ops->is_partially_uptodate)
2350		return false;
2351	if (mapping->host->i_blkbits >= (PAGE_SHIFT + thp_order(page)))
2352		return false;
2353
2354	count = iter->count;
2355	if (page_offset(page) > pos) {
2356		count -= page_offset(page) - pos;
2357		pos = 0;
2358	} else {
2359		pos -= page_offset(page);
2360	}
2361
2362	return mapping->a_ops->is_partially_uptodate(page, pos, count);
2363}
 
 
2364
2365static int filemap_update_page(struct kiocb *iocb,
2366		struct address_space *mapping, struct iov_iter *iter,
2367		struct page *page)
2368{
2369	int error;
 
 
 
 
 
 
 
 
2370
2371	if (!trylock_page(page)) {
2372		if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
2373			return -EAGAIN;
2374		if (!(iocb->ki_flags & IOCB_WAITQ)) {
2375			put_and_wait_on_page_locked(page, TASK_KILLABLE);
2376			return AOP_TRUNCATED_PAGE;
2377		}
2378		error = __lock_page_async(page, iocb->ki_waitq);
2379		if (error)
2380			return error;
2381	}
2382
2383	if (!page->mapping)
2384		goto truncated;
2385
2386	error = 0;
2387	if (filemap_range_uptodate(mapping, iocb->ki_pos, iter, page))
2388		goto unlock;
2389
2390	error = -EAGAIN;
2391	if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ))
2392		goto unlock;
2393
2394	error = filemap_read_page(iocb->ki_filp, mapping, page);
2395	if (error == AOP_TRUNCATED_PAGE)
2396		put_page(page);
2397	return error;
2398truncated:
2399	unlock_page(page);
2400	put_page(page);
2401	return AOP_TRUNCATED_PAGE;
2402unlock:
2403	unlock_page(page);
2404	return error;
2405}
2406
2407static int filemap_create_page(struct file *file,
2408		struct address_space *mapping, pgoff_t index,
2409		struct pagevec *pvec)
2410{
2411	struct page *page;
2412	int error;
 
2413
2414	page = page_cache_alloc(mapping);
2415	if (!page)
2416		return -ENOMEM;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2417
2418	error = add_to_page_cache_lru(page, mapping, index,
2419			mapping_gfp_constraint(mapping, GFP_KERNEL));
2420	if (error == -EEXIST)
2421		error = AOP_TRUNCATED_PAGE;
2422	if (error)
2423		goto error;
2424
2425	error = filemap_read_page(file, mapping, page);
2426	if (error)
2427		goto error;
 
 
2428
2429	pagevec_add(pvec, page);
2430	return 0;
2431error:
2432	put_page(page);
2433	return error;
2434}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2435
2436static int filemap_readahead(struct kiocb *iocb, struct file *file,
2437		struct address_space *mapping, struct page *page,
2438		pgoff_t last_index)
2439{
2440	if (iocb->ki_flags & IOCB_NOIO)
2441		return -EAGAIN;
2442	page_cache_async_readahead(mapping, &file->f_ra, file, page,
2443			page->index, last_index - page->index);
2444	return 0;
2445}
2446
2447static int filemap_get_pages(struct kiocb *iocb, struct iov_iter *iter,
2448		struct pagevec *pvec)
2449{
2450	struct file *filp = iocb->ki_filp;
2451	struct address_space *mapping = filp->f_mapping;
2452	struct file_ra_state *ra = &filp->f_ra;
2453	pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
2454	pgoff_t last_index;
2455	struct page *page;
2456	int err = 0;
2457
2458	last_index = DIV_ROUND_UP(iocb->ki_pos + iter->count, PAGE_SIZE);
2459retry:
2460	if (fatal_signal_pending(current))
2461		return -EINTR;
2462
2463	filemap_get_read_batch(mapping, index, last_index, pvec);
2464	if (!pagevec_count(pvec)) {
2465		if (iocb->ki_flags & IOCB_NOIO)
2466			return -EAGAIN;
2467		page_cache_sync_readahead(mapping, ra, filp, index,
2468				last_index - index);
2469		filemap_get_read_batch(mapping, index, last_index, pvec);
2470	}
2471	if (!pagevec_count(pvec)) {
2472		if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ))
2473			return -EAGAIN;
2474		err = filemap_create_page(filp, mapping,
2475				iocb->ki_pos >> PAGE_SHIFT, pvec);
2476		if (err == AOP_TRUNCATED_PAGE)
2477			goto retry;
2478		return err;
2479	}
2480
2481	page = pvec->pages[pagevec_count(pvec) - 1];
2482	if (PageReadahead(page)) {
2483		err = filemap_readahead(iocb, filp, mapping, page, last_index);
2484		if (err)
2485			goto err;
2486	}
2487	if (!PageUptodate(page)) {
2488		if ((iocb->ki_flags & IOCB_WAITQ) && pagevec_count(pvec) > 1)
2489			iocb->ki_flags |= IOCB_NOWAIT;
2490		err = filemap_update_page(iocb, mapping, iter, page);
2491		if (err)
2492			goto err;
2493	}
2494
2495	return 0;
2496err:
2497	if (err < 0)
2498		put_page(page);
2499	if (likely(--pvec->nr))
2500		return 0;
2501	if (err == AOP_TRUNCATED_PAGE)
2502		goto retry;
2503	return err;
2504}
2505
2506/**
2507 * filemap_read - Read data from the page cache.
2508 * @iocb: The iocb to read.
2509 * @iter: Destination for the data.
2510 * @already_read: Number of bytes already read by the caller.
2511 *
2512 * Copies data from the page cache.  If the data is not currently present,
2513 * uses the readahead and readpage address_space operations to fetch it.
2514 *
2515 * Return: Total number of bytes copied, including those already read by
2516 * the caller.  If an error happens before any bytes are copied, returns
2517 * a negative error number.
2518 */
2519ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
2520		ssize_t already_read)
2521{
2522	struct file *filp = iocb->ki_filp;
2523	struct file_ra_state *ra = &filp->f_ra;
2524	struct address_space *mapping = filp->f_mapping;
2525	struct inode *inode = mapping->host;
2526	struct pagevec pvec;
2527	int i, error = 0;
2528	bool writably_mapped;
2529	loff_t isize, end_offset;
2530
2531	if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
2532		return 0;
2533	if (unlikely(!iov_iter_count(iter)))
2534		return 0;
2535
2536	iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2537	pagevec_init(&pvec);
2538
2539	do {
2540		cond_resched();
2541
2542		/*
2543		 * If we've already successfully copied some data, then we
2544		 * can no longer safely return -EIOCBQUEUED. Hence mark
2545		 * an async read NOWAIT at that point.
2546		 */
2547		if ((iocb->ki_flags & IOCB_WAITQ) && already_read)
2548			iocb->ki_flags |= IOCB_NOWAIT;
2549
2550		error = filemap_get_pages(iocb, iter, &pvec);
2551		if (error < 0)
2552			break;
2553
2554		/*
2555		 * i_size must be checked after we know the pages are Uptodate.
2556		 *
2557		 * Checking i_size after the check allows us to calculate
2558		 * the correct value for "nr", which means the zero-filled
2559		 * part of the page is not copied back to userspace (unless
2560		 * another truncate extends the file - this is desired though).
2561		 */
2562		isize = i_size_read(inode);
2563		if (unlikely(iocb->ki_pos >= isize))
2564			goto put_pages;
2565		end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
2566
2567		/*
2568		 * Once we start copying data, we don't want to be touching any
2569		 * cachelines that might be contended:
2570		 */
2571		writably_mapped = mapping_writably_mapped(mapping);
2572
2573		/*
2574		 * When a sequential read accesses a page several times, only
2575		 * mark it as accessed the first time.
2576		 */
2577		if (iocb->ki_pos >> PAGE_SHIFT !=
2578		    ra->prev_pos >> PAGE_SHIFT)
2579			mark_page_accessed(pvec.pages[0]);
2580
2581		for (i = 0; i < pagevec_count(&pvec); i++) {
2582			struct page *page = pvec.pages[i];
2583			size_t page_size = thp_size(page);
2584			size_t offset = iocb->ki_pos & (page_size - 1);
2585			size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos,
2586					     page_size - offset);
2587			size_t copied;
2588
2589			if (end_offset < page_offset(page))
2590				break;
2591			if (i > 0)
2592				mark_page_accessed(page);
2593			/*
2594			 * If users can be writing to this page using arbitrary
2595			 * virtual addresses, take care about potential aliasing
2596			 * before reading the page on the kernel side.
2597			 */
2598			if (writably_mapped) {
2599				int j;
2600
2601				for (j = 0; j < thp_nr_pages(page); j++)
2602					flush_dcache_page(page + j);
2603			}
2604
2605			copied = copy_page_to_iter(page, offset, bytes, iter);
2606
2607			already_read += copied;
2608			iocb->ki_pos += copied;
2609			ra->prev_pos = iocb->ki_pos;
2610
2611			if (copied < bytes) {
2612				error = -EFAULT;
2613				break;
2614			}
2615		}
2616put_pages:
2617		for (i = 0; i < pagevec_count(&pvec); i++)
2618			put_page(pvec.pages[i]);
2619		pagevec_reinit(&pvec);
2620	} while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
2621
2622	file_accessed(filp);
2623
2624	return already_read ? already_read : error;
2625}
2626EXPORT_SYMBOL_GPL(filemap_read);
2627
2628/**
2629 * generic_file_read_iter - generic filesystem read routine
2630 * @iocb:	kernel I/O control block
2631 * @iter:	destination for the data read
 
 
2632 *
2633 * This is the "read_iter()" routine for all filesystems
2634 * that can use the page cache directly.
2635 *
2636 * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2637 * be returned when no data can be read without waiting for I/O requests
2638 * to complete; it doesn't prevent readahead.
2639 *
2640 * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2641 * requests shall be made for the read or for readahead.  When no data
2642 * can be read, -EAGAIN shall be returned.  When readahead would be
2643 * triggered, a partial, possibly empty read shall be returned.
2644 *
2645 * Return:
2646 * * number of bytes copied, even for partial reads
2647 * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2648 */
2649ssize_t
2650generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
 
2651{
2652	size_t count = iov_iter_count(iter);
2653	ssize_t retval = 0;
2654
2655	if (!count)
2656		return 0; /* skip atime */
 
 
 
 
 
2657
2658	if (iocb->ki_flags & IOCB_DIRECT) {
2659		struct file *file = iocb->ki_filp;
2660		struct address_space *mapping = file->f_mapping;
2661		struct inode *inode = mapping->host;
2662		loff_t size;
 
 
2663
 
 
 
 
2664		size = i_size_read(inode);
2665		if (iocb->ki_flags & IOCB_NOWAIT) {
2666			if (filemap_range_needs_writeback(mapping, iocb->ki_pos,
2667						iocb->ki_pos + count - 1))
2668				return -EAGAIN;
2669		} else {
2670			retval = filemap_write_and_wait_range(mapping,
2671						iocb->ki_pos,
2672					        iocb->ki_pos + count - 1);
2673			if (retval < 0)
2674				return retval;
2675		}
2676
2677		file_accessed(file);
2678
2679		retval = mapping->a_ops->direct_IO(iocb, iter);
2680		if (retval >= 0) {
2681			iocb->ki_pos += retval;
2682			count -= retval;
2683		}
2684		if (retval != -EIOCBQUEUED)
2685			iov_iter_revert(iter, count - iov_iter_count(iter));
2686
2687		/*
2688		 * Btrfs can have a short DIO read if we encounter
2689		 * compressed extents, so if there was an error, or if
2690		 * we've already read everything we wanted to, or if
2691		 * there was a short read because we hit EOF, go ahead
2692		 * and return.  Otherwise fallthrough to buffered io for
2693		 * the rest of the read.  Buffered reads will not work for
2694		 * DAX files, so don't bother trying.
2695		 */
2696		if (retval < 0 || !count || iocb->ki_pos >= size ||
2697		    IS_DAX(inode))
2698			return retval;
 
2699	}
2700
2701	return filemap_read(iocb, iter, retval);
2702}
2703EXPORT_SYMBOL(generic_file_read_iter);
2704
2705static inline loff_t page_seek_hole_data(struct xa_state *xas,
2706		struct address_space *mapping, struct page *page,
2707		loff_t start, loff_t end, bool seek_data)
2708{
2709	const struct address_space_operations *ops = mapping->a_ops;
2710	size_t offset, bsz = i_blocksize(mapping->host);
2711
2712	if (xa_is_value(page) || PageUptodate(page))
2713		return seek_data ? start : end;
2714	if (!ops->is_partially_uptodate)
2715		return seek_data ? end : start;
2716
2717	xas_pause(xas);
2718	rcu_read_unlock();
2719	lock_page(page);
2720	if (unlikely(page->mapping != mapping))
2721		goto unlock;
2722
2723	offset = offset_in_thp(page, start) & ~(bsz - 1);
 
 
 
 
 
 
 
 
 
 
 
2724
2725	do {
2726		if (ops->is_partially_uptodate(page, offset, bsz) == seek_data)
 
 
 
 
 
 
 
 
2727			break;
2728		start = (start + bsz) & ~(bsz - 1);
2729		offset += bsz;
2730	} while (offset < thp_size(page));
2731unlock:
2732	unlock_page(page);
2733	rcu_read_lock();
2734	return start;
2735}
2736
2737static inline
2738unsigned int seek_page_size(struct xa_state *xas, struct page *page)
2739{
2740	if (xa_is_value(page))
2741		return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index);
2742	return thp_size(page);
2743}
 
2744
 
2745/**
2746 * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
2747 * @mapping: Address space to search.
2748 * @start: First byte to consider.
2749 * @end: Limit of search (exclusive).
2750 * @whence: Either SEEK_HOLE or SEEK_DATA.
2751 *
2752 * If the page cache knows which blocks contain holes and which blocks
2753 * contain data, your filesystem can use this function to implement
2754 * SEEK_HOLE and SEEK_DATA.  This is useful for filesystems which are
2755 * entirely memory-based such as tmpfs, and filesystems which support
2756 * unwritten extents.
2757 *
2758 * Return: The requested offset on success, or -ENXIO if @whence specifies
2759 * SEEK_DATA and there is no data after @start.  There is an implicit hole
2760 * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
2761 * and @end contain data.
2762 */
2763loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start,
2764		loff_t end, int whence)
2765{
2766	XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT);
2767	pgoff_t max = (end - 1) >> PAGE_SHIFT;
2768	bool seek_data = (whence == SEEK_DATA);
2769	struct page *page;
2770
2771	if (end <= start)
2772		return -ENXIO;
2773
2774	rcu_read_lock();
2775	while ((page = find_get_entry(&xas, max, XA_PRESENT))) {
2776		loff_t pos = (u64)xas.xa_index << PAGE_SHIFT;
2777		unsigned int seek_size;
2778
2779		if (start < pos) {
2780			if (!seek_data)
2781				goto unlock;
2782			start = pos;
2783		}
2784
2785		seek_size = seek_page_size(&xas, page);
2786		pos = round_up(pos + 1, seek_size);
2787		start = page_seek_hole_data(&xas, mapping, page, start, pos,
2788				seek_data);
2789		if (start < pos)
2790			goto unlock;
2791		if (start >= end)
2792			break;
2793		if (seek_size > PAGE_SIZE)
2794			xas_set(&xas, pos >> PAGE_SHIFT);
2795		if (!xa_is_value(page))
2796			put_page(page);
2797	}
2798	if (seek_data)
2799		start = -ENXIO;
2800unlock:
2801	rcu_read_unlock();
2802	if (page && !xa_is_value(page))
2803		put_page(page);
2804	if (start > end)
2805		return end;
2806	return start;
2807}
2808
2809#ifdef CONFIG_MMU
2810#define MMAP_LOTSAMISS  (100)
2811/*
2812 * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
2813 * @vmf - the vm_fault for this fault.
2814 * @page - the page to lock.
2815 * @fpin - the pointer to the file we may pin (or is already pinned).
2816 *
2817 * This works similar to lock_page_or_retry in that it can drop the mmap_lock.
2818 * It differs in that it actually returns the page locked if it returns 1 and 0
2819 * if it couldn't lock the page.  If we did have to drop the mmap_lock then fpin
2820 * will point to the pinned file and needs to be fput()'ed at a later point.
2821 */
2822static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page,
2823				     struct file **fpin)
2824{
2825	if (trylock_page(page))
2826		return 1;
2827
2828	/*
2829	 * NOTE! This will make us return with VM_FAULT_RETRY, but with
2830	 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
2831	 * is supposed to work. We have way too many special cases..
2832	 */
2833	if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
2834		return 0;
2835
2836	*fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
2837	if (vmf->flags & FAULT_FLAG_KILLABLE) {
2838		if (__lock_page_killable(page)) {
2839			/*
2840			 * We didn't have the right flags to drop the mmap_lock,
2841			 * but all fault_handlers only check for fatal signals
2842			 * if we return VM_FAULT_RETRY, so we need to drop the
2843			 * mmap_lock here and return 0 if we don't have a fpin.
2844			 */
2845			if (*fpin == NULL)
2846				mmap_read_unlock(vmf->vma->vm_mm);
2847			return 0;
2848		}
2849	} else
2850		__lock_page(page);
2851	return 1;
2852}
2853
 
2854
2855/*
2856 * Synchronous readahead happens when we don't even find a page in the page
2857 * cache at all.  We don't want to perform IO under the mmap sem, so if we have
2858 * to drop the mmap sem we return the file that was pinned in order for us to do
2859 * that.  If we didn't pin a file then we return NULL.  The file that is
2860 * returned needs to be fput()'ed when we're done with it.
2861 */
2862static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
 
 
 
2863{
2864	struct file *file = vmf->vma->vm_file;
2865	struct file_ra_state *ra = &file->f_ra;
2866	struct address_space *mapping = file->f_mapping;
2867	DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff);
2868	struct file *fpin = NULL;
2869	unsigned int mmap_miss;
2870
2871	/* If we don't want any read-ahead, don't bother */
2872	if (vmf->vma->vm_flags & VM_RAND_READ)
2873		return fpin;
2874	if (!ra->ra_pages)
2875		return fpin;
2876
2877	if (vmf->vma->vm_flags & VM_SEQ_READ) {
2878		fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2879		page_cache_sync_ra(&ractl, ra->ra_pages);
2880		return fpin;
2881	}
2882
2883	/* Avoid banging the cache line if not needed */
2884	mmap_miss = READ_ONCE(ra->mmap_miss);
2885	if (mmap_miss < MMAP_LOTSAMISS * 10)
2886		WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
2887
2888	/*
2889	 * Do we miss much more than hit in this file? If so,
2890	 * stop bothering with read-ahead. It will only hurt.
2891	 */
2892	if (mmap_miss > MMAP_LOTSAMISS)
2893		return fpin;
2894
2895	/*
2896	 * mmap read-around
2897	 */
2898	fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2899	ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
2900	ra->size = ra->ra_pages;
2901	ra->async_size = ra->ra_pages / 4;
2902	ractl._index = ra->start;
2903	do_page_cache_ra(&ractl, ra->size, ra->async_size);
2904	return fpin;
2905}
2906
2907/*
2908 * Asynchronous readahead happens when we find the page and PG_readahead,
2909 * so we want to possibly extend the readahead further.  We return the file that
2910 * was pinned if we have to drop the mmap_lock in order to do IO.
2911 */
2912static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
2913					    struct page *page)
 
 
 
2914{
2915	struct file *file = vmf->vma->vm_file;
2916	struct file_ra_state *ra = &file->f_ra;
2917	struct address_space *mapping = file->f_mapping;
2918	struct file *fpin = NULL;
2919	unsigned int mmap_miss;
2920	pgoff_t offset = vmf->pgoff;
2921
2922	/* If we don't want any read-ahead, don't bother */
2923	if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
2924		return fpin;
2925	mmap_miss = READ_ONCE(ra->mmap_miss);
2926	if (mmap_miss)
2927		WRITE_ONCE(ra->mmap_miss, --mmap_miss);
2928	if (PageReadahead(page)) {
2929		fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2930		page_cache_async_readahead(mapping, ra, file,
2931					   page, offset, ra->ra_pages);
2932	}
2933	return fpin;
2934}
2935
2936/**
2937 * filemap_fault - read in file data for page fault handling
 
2938 * @vmf:	struct vm_fault containing details of the fault
2939 *
2940 * filemap_fault() is invoked via the vma operations vector for a
2941 * mapped memory region to read in file data during a page fault.
2942 *
2943 * The goto's are kind of ugly, but this streamlines the normal case of having
2944 * it in the page cache, and handles the special cases reasonably without
2945 * having a lot of duplicated code.
2946 *
2947 * vma->vm_mm->mmap_lock must be held on entry.
2948 *
2949 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
2950 * may be dropped before doing I/O or by lock_page_maybe_drop_mmap().
2951 *
2952 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
2953 * has not been released.
2954 *
2955 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2956 *
2957 * Return: bitwise-OR of %VM_FAULT_ codes.
2958 */
2959vm_fault_t filemap_fault(struct vm_fault *vmf)
2960{
2961	int error;
2962	struct file *file = vmf->vma->vm_file;
2963	struct file *fpin = NULL;
2964	struct address_space *mapping = file->f_mapping;
 
2965	struct inode *inode = mapping->host;
2966	pgoff_t offset = vmf->pgoff;
2967	pgoff_t max_off;
2968	struct page *page;
2969	vm_fault_t ret = 0;
 
2970
2971	max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2972	if (unlikely(offset >= max_off))
2973		return VM_FAULT_SIGBUS;
2974
2975	/*
2976	 * Do we have something in the page cache already?
2977	 */
2978	page = find_get_page(mapping, offset);
2979	if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2980		/*
2981		 * We found the page, so try async readahead before
2982		 * waiting for the lock.
2983		 */
2984		fpin = do_async_mmap_readahead(vmf, page);
2985	} else if (!page) {
2986		/* No page in the page cache at all */
 
2987		count_vm_event(PGMAJFAULT);
2988		count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
2989		ret = VM_FAULT_MAJOR;
2990		fpin = do_sync_mmap_readahead(vmf);
2991retry_find:
2992		page = pagecache_get_page(mapping, offset,
2993					  FGP_CREAT|FGP_FOR_MMAP,
2994					  vmf->gfp_mask);
2995		if (!page) {
2996			if (fpin)
2997				goto out_retry;
2998			return VM_FAULT_OOM;
2999		}
3000	}
3001
3002	if (!lock_page_maybe_drop_mmap(vmf, page, &fpin))
3003		goto out_retry;
 
 
3004
3005	/* Did it get truncated? */
3006	if (unlikely(compound_head(page)->mapping != mapping)) {
3007		unlock_page(page);
3008		put_page(page);
3009		goto retry_find;
3010	}
3011	VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
3012
3013	/*
3014	 * We have a locked page in the page cache, now we need to check
3015	 * that it's up-to-date. If not, it is going to be due to an error.
3016	 */
3017	if (unlikely(!PageUptodate(page)))
3018		goto page_not_uptodate;
3019
3020	/*
3021	 * We've made it this far and we had to drop our mmap_lock, now is the
3022	 * time to return to the upper layer and have it re-find the vma and
3023	 * redo the fault.
3024	 */
3025	if (fpin) {
3026		unlock_page(page);
3027		goto out_retry;
3028	}
3029
3030	/*
3031	 * Found the page and have a reference on it.
3032	 * We must recheck i_size under page lock.
3033	 */
3034	max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3035	if (unlikely(offset >= max_off)) {
3036		unlock_page(page);
3037		put_page(page);
3038		return VM_FAULT_SIGBUS;
3039	}
3040
3041	vmf->page = page;
3042	return ret | VM_FAULT_LOCKED;
3043
3044page_not_uptodate:
3045	/*
3046	 * Umm, take care of errors if the page isn't up-to-date.
3047	 * Try to re-read it _once_. We do this synchronously,
3048	 * because there really aren't any performance issues here
3049	 * and we need to check for errors.
3050	 */
3051	fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3052	error = filemap_read_page(file, mapping, page);
3053	if (fpin)
3054		goto out_retry;
3055	put_page(page);
3056
3057	if (!error || error == AOP_TRUNCATED_PAGE)
 
 
 
 
 
3058		goto retry_find;
3059
 
 
 
 
 
 
 
3060	return VM_FAULT_SIGBUS;
3061
3062out_retry:
3063	/*
3064	 * We dropped the mmap_lock, we need to return to the fault handler to
3065	 * re-find the vma and come back and find our hopefully still populated
3066	 * page.
 
3067	 */
3068	if (page)
3069		put_page(page);
3070	if (fpin)
3071		fput(fpin);
3072	return ret | VM_FAULT_RETRY;
3073}
3074EXPORT_SYMBOL(filemap_fault);
3075
3076static bool filemap_map_pmd(struct vm_fault *vmf, struct page *page)
3077{
3078	struct mm_struct *mm = vmf->vma->vm_mm;
3079
3080	/* Huge page is mapped? No need to proceed. */
3081	if (pmd_trans_huge(*vmf->pmd)) {
3082		unlock_page(page);
3083		put_page(page);
3084		return true;
3085	}
3086
3087	if (pmd_none(*vmf->pmd) && PageTransHuge(page)) {
3088	    vm_fault_t ret = do_set_pmd(vmf, page);
3089	    if (!ret) {
3090		    /* The page is mapped successfully, reference consumed. */
3091		    unlock_page(page);
3092		    return true;
3093	    }
3094	}
3095
3096	if (pmd_none(*vmf->pmd)) {
3097		vmf->ptl = pmd_lock(mm, vmf->pmd);
3098		if (likely(pmd_none(*vmf->pmd))) {
3099			mm_inc_nr_ptes(mm);
3100			pmd_populate(mm, vmf->pmd, vmf->prealloc_pte);
3101			vmf->prealloc_pte = NULL;
3102		}
3103		spin_unlock(vmf->ptl);
3104	}
3105
3106	/* See comment in handle_pte_fault() */
3107	if (pmd_devmap_trans_unstable(vmf->pmd)) {
3108		unlock_page(page);
3109		put_page(page);
3110		return true;
3111	}
3112
3113	return false;
3114}
3115
3116static struct page *next_uptodate_page(struct page *page,
3117				       struct address_space *mapping,
3118				       struct xa_state *xas, pgoff_t end_pgoff)
3119{
3120	unsigned long max_idx;
3121
3122	do {
3123		if (!page)
3124			return NULL;
3125		if (xas_retry(xas, page))
3126			continue;
3127		if (xa_is_value(page))
3128			continue;
3129		if (PageLocked(page))
3130			continue;
3131		if (!page_cache_get_speculative(page))
3132			continue;
3133		/* Has the page moved or been split? */
3134		if (unlikely(page != xas_reload(xas)))
3135			goto skip;
3136		if (!PageUptodate(page) || PageReadahead(page))
3137			goto skip;
3138		if (PageHWPoison(page))
3139			goto skip;
3140		if (!trylock_page(page))
3141			goto skip;
3142		if (page->mapping != mapping)
3143			goto unlock;
3144		if (!PageUptodate(page))
3145			goto unlock;
3146		max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3147		if (xas->xa_index >= max_idx)
3148			goto unlock;
3149		return page;
3150unlock:
3151		unlock_page(page);
3152skip:
3153		put_page(page);
3154	} while ((page = xas_next_entry(xas, end_pgoff)) != NULL);
3155
3156	return NULL;
3157}
3158
3159static inline struct page *first_map_page(struct address_space *mapping,
3160					  struct xa_state *xas,
3161					  pgoff_t end_pgoff)
3162{
3163	return next_uptodate_page(xas_find(xas, end_pgoff),
3164				  mapping, xas, end_pgoff);
3165}
3166
3167static inline struct page *next_map_page(struct address_space *mapping,
3168					 struct xa_state *xas,
3169					 pgoff_t end_pgoff)
3170{
3171	return next_uptodate_page(xas_next_entry(xas, end_pgoff),
3172				  mapping, xas, end_pgoff);
3173}
3174
3175vm_fault_t filemap_map_pages(struct vm_fault *vmf,
3176			     pgoff_t start_pgoff, pgoff_t end_pgoff)
3177{
3178	struct vm_area_struct *vma = vmf->vma;
3179	struct file *file = vma->vm_file;
3180	struct address_space *mapping = file->f_mapping;
3181	pgoff_t last_pgoff = start_pgoff;
3182	unsigned long addr;
3183	XA_STATE(xas, &mapping->i_pages, start_pgoff);
3184	struct page *head, *page;
3185	unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss);
3186	vm_fault_t ret = 0;
3187
3188	rcu_read_lock();
3189	head = first_map_page(mapping, &xas, end_pgoff);
3190	if (!head)
3191		goto out;
3192
3193	if (filemap_map_pmd(vmf, head)) {
3194		ret = VM_FAULT_NOPAGE;
3195		goto out;
3196	}
 
3197
3198	addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT);
3199	vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl);
3200	do {
3201		page = find_subpage(head, xas.xa_index);
3202		if (PageHWPoison(page))
3203			goto unlock;
3204
3205		if (mmap_miss > 0)
3206			mmap_miss--;
3207
3208		addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3209		vmf->pte += xas.xa_index - last_pgoff;
3210		last_pgoff = xas.xa_index;
3211
3212		if (!pte_none(*vmf->pte))
3213			goto unlock;
3214
3215		/* We're about to handle the fault */
3216		if (vmf->address == addr)
3217			ret = VM_FAULT_NOPAGE;
3218
3219		do_set_pte(vmf, page, addr);
3220		/* no need to invalidate: a not-present page won't be cached */
3221		update_mmu_cache(vma, addr, vmf->pte);
3222		unlock_page(head);
3223		continue;
3224unlock:
3225		unlock_page(head);
3226		put_page(head);
3227	} while ((head = next_map_page(mapping, &xas, end_pgoff)) != NULL);
3228	pte_unmap_unlock(vmf->pte, vmf->ptl);
3229out:
3230	rcu_read_unlock();
3231	WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss);
3232	return ret;
3233}
3234EXPORT_SYMBOL(filemap_map_pages);
3235
3236vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3237{
3238	struct address_space *mapping = vmf->vma->vm_file->f_mapping;
3239	struct page *page = vmf->page;
3240	vm_fault_t ret = VM_FAULT_LOCKED;
3241
3242	sb_start_pagefault(mapping->host->i_sb);
3243	file_update_time(vmf->vma->vm_file);
3244	lock_page(page);
3245	if (page->mapping != mapping) {
3246		unlock_page(page);
3247		ret = VM_FAULT_NOPAGE;
3248		goto out;
3249	}
3250	/*
3251	 * We mark the page dirty already here so that when freeze is in
3252	 * progress, we are guaranteed that writeback during freezing will
3253	 * see the dirty page and writeprotect it again.
3254	 */
3255	set_page_dirty(page);
3256	wait_for_stable_page(page);
3257out:
3258	sb_end_pagefault(mapping->host->i_sb);
3259	return ret;
3260}
 
3261
3262const struct vm_operations_struct generic_file_vm_ops = {
3263	.fault		= filemap_fault,
3264	.map_pages	= filemap_map_pages,
3265	.page_mkwrite	= filemap_page_mkwrite,
3266};
3267
3268/* This is used for a general mmap of a disk file */
3269
3270int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3271{
3272	struct address_space *mapping = file->f_mapping;
3273
3274	if (!mapping->a_ops->readpage)
3275		return -ENOEXEC;
3276	file_accessed(file);
3277	vma->vm_ops = &generic_file_vm_ops;
 
3278	return 0;
3279}
3280
3281/*
3282 * This is for filesystems which do not implement ->writepage.
3283 */
3284int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3285{
3286	if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
3287		return -EINVAL;
3288	return generic_file_mmap(file, vma);
3289}
3290#else
3291vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3292{
3293	return VM_FAULT_SIGBUS;
3294}
3295int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3296{
3297	return -ENOSYS;
3298}
3299int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3300{
3301	return -ENOSYS;
3302}
3303#endif /* CONFIG_MMU */
3304
3305EXPORT_SYMBOL(filemap_page_mkwrite);
3306EXPORT_SYMBOL(generic_file_mmap);
3307EXPORT_SYMBOL(generic_file_readonly_mmap);
3308
3309static struct page *wait_on_page_read(struct page *page)
3310{
3311	if (!IS_ERR(page)) {
3312		wait_on_page_locked(page);
3313		if (!PageUptodate(page)) {
3314			put_page(page);
3315			page = ERR_PTR(-EIO);
3316		}
3317	}
3318	return page;
3319}
3320
3321static struct page *do_read_cache_page(struct address_space *mapping,
3322				pgoff_t index,
3323				int (*filler)(void *, struct page *),
3324				void *data,
3325				gfp_t gfp)
3326{
3327	struct page *page;
3328	int err;
3329repeat:
3330	page = find_get_page(mapping, index);
3331	if (!page) {
3332		page = __page_cache_alloc(gfp);
3333		if (!page)
3334			return ERR_PTR(-ENOMEM);
3335		err = add_to_page_cache_lru(page, mapping, index, gfp);
3336		if (unlikely(err)) {
3337			put_page(page);
3338			if (err == -EEXIST)
3339				goto repeat;
3340			/* Presumably ENOMEM for xarray node */
3341			return ERR_PTR(err);
3342		}
3343
3344filler:
3345		if (filler)
3346			err = filler(data, page);
3347		else
3348			err = mapping->a_ops->readpage(data, page);
3349
3350		if (err < 0) {
3351			put_page(page);
3352			return ERR_PTR(err);
3353		}
3354
3355		page = wait_on_page_read(page);
3356		if (IS_ERR(page))
3357			return page;
3358		goto out;
3359	}
3360	if (PageUptodate(page))
3361		goto out;
3362
3363	/*
3364	 * Page is not up to date and may be locked due to one of the following
3365	 * case a: Page is being filled and the page lock is held
3366	 * case b: Read/write error clearing the page uptodate status
3367	 * case c: Truncation in progress (page locked)
3368	 * case d: Reclaim in progress
3369	 *
3370	 * Case a, the page will be up to date when the page is unlocked.
3371	 *    There is no need to serialise on the page lock here as the page
3372	 *    is pinned so the lock gives no additional protection. Even if the
3373	 *    page is truncated, the data is still valid if PageUptodate as
3374	 *    it's a race vs truncate race.
3375	 * Case b, the page will not be up to date
3376	 * Case c, the page may be truncated but in itself, the data may still
3377	 *    be valid after IO completes as it's a read vs truncate race. The
3378	 *    operation must restart if the page is not uptodate on unlock but
3379	 *    otherwise serialising on page lock to stabilise the mapping gives
3380	 *    no additional guarantees to the caller as the page lock is
3381	 *    released before return.
3382	 * Case d, similar to truncation. If reclaim holds the page lock, it
3383	 *    will be a race with remove_mapping that determines if the mapping
3384	 *    is valid on unlock but otherwise the data is valid and there is
3385	 *    no need to serialise with page lock.
3386	 *
3387	 * As the page lock gives no additional guarantee, we optimistically
3388	 * wait on the page to be unlocked and check if it's up to date and
3389	 * use the page if it is. Otherwise, the page lock is required to
3390	 * distinguish between the different cases. The motivation is that we
3391	 * avoid spurious serialisations and wakeups when multiple processes
3392	 * wait on the same page for IO to complete.
3393	 */
3394	wait_on_page_locked(page);
3395	if (PageUptodate(page))
3396		goto out;
3397
3398	/* Distinguish between all the cases under the safety of the lock */
3399	lock_page(page);
3400
3401	/* Case c or d, restart the operation */
3402	if (!page->mapping) {
3403		unlock_page(page);
3404		put_page(page);
3405		goto repeat;
3406	}
3407
3408	/* Someone else locked and filled the page in a very small window */
3409	if (PageUptodate(page)) {
3410		unlock_page(page);
3411		goto out;
3412	}
3413
3414	/*
3415	 * A previous I/O error may have been due to temporary
3416	 * failures.
3417	 * Clear page error before actual read, PG_error will be
3418	 * set again if read page fails.
3419	 */
3420	ClearPageError(page);
3421	goto filler;
3422
3423out:
3424	mark_page_accessed(page);
3425	return page;
3426}
3427
3428/**
3429 * read_cache_page - read into page cache, fill it if needed
3430 * @mapping:	the page's address_space
3431 * @index:	the page index
3432 * @filler:	function to perform the read
3433 * @data:	first arg to filler(data, page) function, often left as NULL
3434 *
 
 
 
3435 * Read into the page cache. If a page already exists, and PageUptodate() is
3436 * not set, try to fill the page and wait for it to become unlocked.
3437 *
3438 * If the page does not get brought uptodate, return -EIO.
3439 *
3440 * Return: up to date page on success, ERR_PTR() on failure.
3441 */
3442struct page *read_cache_page(struct address_space *mapping,
3443				pgoff_t index,
3444				int (*filler)(void *, struct page *),
3445				void *data)
3446{
3447	return do_read_cache_page(mapping, index, filler, data,
3448			mapping_gfp_mask(mapping));
 
 
 
 
 
 
 
 
 
 
 
 
3449}
3450EXPORT_SYMBOL(read_cache_page);
3451
3452/**
3453 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3454 * @mapping:	the page's address_space
3455 * @index:	the page index
3456 * @gfp:	the page allocator flags to use if allocating
3457 *
3458 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3459 * any new page allocations done using the specified allocation flags.
3460 *
3461 * If the page does not get brought uptodate, return -EIO.
3462 *
3463 * Return: up to date page on success, ERR_PTR() on failure.
3464 */
3465struct page *read_cache_page_gfp(struct address_space *mapping,
3466				pgoff_t index,
3467				gfp_t gfp)
3468{
3469	return do_read_cache_page(mapping, index, NULL, NULL, gfp);
 
 
3470}
3471EXPORT_SYMBOL(read_cache_page_gfp);
3472
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3473int pagecache_write_begin(struct file *file, struct address_space *mapping,
3474				loff_t pos, unsigned len, unsigned flags,
3475				struct page **pagep, void **fsdata)
3476{
3477	const struct address_space_operations *aops = mapping->a_ops;
3478
3479	return aops->write_begin(file, mapping, pos, len, flags,
3480							pagep, fsdata);
3481}
3482EXPORT_SYMBOL(pagecache_write_begin);
3483
3484int pagecache_write_end(struct file *file, struct address_space *mapping,
3485				loff_t pos, unsigned len, unsigned copied,
3486				struct page *page, void *fsdata)
3487{
3488	const struct address_space_operations *aops = mapping->a_ops;
3489
 
3490	return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
3491}
3492EXPORT_SYMBOL(pagecache_write_end);
3493
3494/*
3495 * Warn about a page cache invalidation failure during a direct I/O write.
3496 */
3497void dio_warn_stale_pagecache(struct file *filp)
3498{
3499	static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3500	char pathname[128];
3501	char *path;
3502
3503	errseq_set(&filp->f_mapping->wb_err, -EIO);
3504	if (__ratelimit(&_rs)) {
3505		path = file_path(filp, pathname, sizeof(pathname));
3506		if (IS_ERR(path))
3507			path = "(unknown)";
3508		pr_crit("Page cache invalidation failure on direct I/O.  Possible data corruption due to collision with buffered I/O!\n");
3509		pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3510			current->comm);
3511	}
3512}
3513
3514ssize_t
3515generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
 
 
3516{
3517	struct file	*file = iocb->ki_filp;
3518	struct address_space *mapping = file->f_mapping;
3519	struct inode	*inode = mapping->host;
3520	loff_t		pos = iocb->ki_pos;
3521	ssize_t		written;
3522	size_t		write_len;
3523	pgoff_t		end;
3524
3525	write_len = iov_iter_count(from);
3526	end = (pos + write_len - 1) >> PAGE_SHIFT;
3527
3528	if (iocb->ki_flags & IOCB_NOWAIT) {
3529		/* If there are pages to writeback, return */
3530		if (filemap_range_has_page(file->f_mapping, pos,
3531					   pos + write_len - 1))
3532			return -EAGAIN;
3533	} else {
3534		written = filemap_write_and_wait_range(mapping, pos,
3535							pos + write_len - 1);
3536		if (written)
3537			goto out;
3538	}
3539
3540	/*
3541	 * After a write we want buffered reads to be sure to go to disk to get
3542	 * the new data.  We invalidate clean cached page from the region we're
3543	 * about to write.  We do this *before* the write so that we can return
3544	 * without clobbering -EIOCBQUEUED from ->direct_IO().
3545	 */
3546	written = invalidate_inode_pages2_range(mapping,
3547					pos >> PAGE_SHIFT, end);
3548	/*
3549	 * If a page can not be invalidated, return 0 to fall back
3550	 * to buffered write.
3551	 */
3552	if (written) {
3553		if (written == -EBUSY)
3554			return 0;
3555		goto out;
 
 
3556	}
3557
3558	written = mapping->a_ops->direct_IO(iocb, from);
3559
3560	/*
3561	 * Finally, try again to invalidate clean pages which might have been
3562	 * cached by non-direct readahead, or faulted in by get_user_pages()
3563	 * if the source of the write was an mmap'ed region of the file
3564	 * we're writing.  Either one is a pretty crazy thing to do,
3565	 * so we don't support it 100%.  If this invalidation
3566	 * fails, tough, the write still worked...
3567	 *
3568	 * Most of the time we do not need this since dio_complete() will do
3569	 * the invalidation for us. However there are some file systems that
3570	 * do not end up with dio_complete() being called, so let's not break
3571	 * them by removing it completely.
3572	 *
3573	 * Noticeable example is a blkdev_direct_IO().
3574	 *
3575	 * Skip invalidation for async writes or if mapping has no pages.
3576	 */
3577	if (written > 0 && mapping->nrpages &&
3578	    invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end))
3579		dio_warn_stale_pagecache(file);
 
3580
3581	if (written > 0) {
3582		pos += written;
3583		write_len -= written;
3584		if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3585			i_size_write(inode, pos);
3586			mark_inode_dirty(inode);
3587		}
3588		iocb->ki_pos = pos;
3589	}
3590	if (written != -EIOCBQUEUED)
3591		iov_iter_revert(from, write_len - iov_iter_count(from));
3592out:
3593	return written;
3594}
3595EXPORT_SYMBOL(generic_file_direct_write);
3596
3597/*
3598 * Find or create a page at the given pagecache position. Return the locked
3599 * page. This function is specifically for buffered writes.
3600 */
3601struct page *grab_cache_page_write_begin(struct address_space *mapping,
3602					pgoff_t index, unsigned flags)
3603{
 
 
3604	struct page *page;
3605	int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
3606
 
 
 
3607	if (flags & AOP_FLAG_NOFS)
3608		fgp_flags |= FGP_NOFS;
3609
3610	page = pagecache_get_page(mapping, index, fgp_flags,
3611			mapping_gfp_mask(mapping));
3612	if (page)
3613		wait_for_stable_page(page);
3614
 
 
 
 
 
 
 
 
 
 
 
 
 
3615	return page;
3616}
3617EXPORT_SYMBOL(grab_cache_page_write_begin);
3618
3619ssize_t generic_perform_write(struct file *file,
3620				struct iov_iter *i, loff_t pos)
3621{
3622	struct address_space *mapping = file->f_mapping;
3623	const struct address_space_operations *a_ops = mapping->a_ops;
3624	long status = 0;
3625	ssize_t written = 0;
3626	unsigned int flags = 0;
3627
 
 
 
 
 
 
3628	do {
3629		struct page *page;
3630		unsigned long offset;	/* Offset into pagecache page */
3631		unsigned long bytes;	/* Bytes to write to page */
3632		size_t copied;		/* Bytes copied from user */
3633		void *fsdata;
3634
3635		offset = (pos & (PAGE_SIZE - 1));
3636		bytes = min_t(unsigned long, PAGE_SIZE - offset,
3637						iov_iter_count(i));
3638
3639again:
3640		/*
3641		 * Bring in the user page that we will copy from _first_.
3642		 * Otherwise there's a nasty deadlock on copying from the
3643		 * same page as we're writing to, without it being marked
3644		 * up-to-date.
 
 
 
 
3645		 */
3646		if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
3647			status = -EFAULT;
3648			break;
3649		}
3650
3651		if (fatal_signal_pending(current)) {
3652			status = -EINTR;
3653			break;
3654		}
3655
3656		status = a_ops->write_begin(file, mapping, pos, bytes, flags,
3657						&page, &fsdata);
3658		if (unlikely(status < 0))
3659			break;
3660
3661		if (mapping_writably_mapped(mapping))
3662			flush_dcache_page(page);
3663
3664		copied = copy_page_from_iter_atomic(page, offset, bytes, i);
 
 
3665		flush_dcache_page(page);
3666
 
3667		status = a_ops->write_end(file, mapping, pos, bytes, copied,
3668						page, fsdata);
3669		if (unlikely(status != copied)) {
3670			iov_iter_revert(i, copied - max(status, 0L));
3671			if (unlikely(status < 0))
3672				break;
3673		}
3674		cond_resched();
3675
3676		if (unlikely(status == 0)) {
 
3677			/*
3678			 * A short copy made ->write_end() reject the
3679			 * thing entirely.  Might be memory poisoning
3680			 * halfway through, might be a race with munmap,
3681			 * might be severe memory pressure.
 
 
3682			 */
3683			if (copied)
3684				bytes = copied;
3685			goto again;
3686		}
3687		pos += status;
3688		written += status;
3689
3690		balance_dirty_pages_ratelimited(mapping);
 
 
 
 
3691	} while (iov_iter_count(i));
3692
3693	return written ? written : status;
3694}
3695EXPORT_SYMBOL(generic_perform_write);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3696
3697/**
3698 * __generic_file_write_iter - write data to a file
3699 * @iocb:	IO state structure (file, offset, etc.)
3700 * @from:	iov_iter with data to write
 
 
3701 *
3702 * This function does all the work needed for actually writing data to a
3703 * file. It does all basic checks, removes SUID from the file, updates
3704 * modification times and calls proper subroutines depending on whether we
3705 * do direct IO or a standard buffered write.
3706 *
3707 * It expects i_mutex to be grabbed unless we work on a block device or similar
3708 * object which does not need locking at all.
3709 *
3710 * This function does *not* take care of syncing data in case of O_SYNC write.
3711 * A caller has to handle it. This is mainly due to the fact that we want to
3712 * avoid syncing under i_mutex.
3713 *
3714 * Return:
3715 * * number of bytes written, even for truncated writes
3716 * * negative error code if no data has been written at all
3717 */
3718ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 
3719{
3720	struct file *file = iocb->ki_filp;
3721	struct address_space *mapping = file->f_mapping;
 
 
3722	struct inode 	*inode = mapping->host;
3723	ssize_t		written = 0;
 
3724	ssize_t		err;
3725	ssize_t		status;
 
 
 
 
 
 
 
 
 
3726
3727	/* We can write back this queue in page reclaim */
3728	current->backing_dev_info = inode_to_bdi(inode);
3729	err = file_remove_privs(file);
 
 
3730	if (err)
3731		goto out;
3732
3733	err = file_update_time(file);
 
 
 
3734	if (err)
3735		goto out;
3736
3737	if (iocb->ki_flags & IOCB_DIRECT) {
3738		loff_t pos, endbyte;
 
3739
3740		written = generic_file_direct_write(iocb, from);
 
 
 
 
 
 
 
 
3741		/*
3742		 * If the write stopped short of completing, fall back to
3743		 * buffered writes.  Some filesystems do this for writes to
3744		 * holes, for example.  For DAX files, a buffered write will
3745		 * not succeed (even if it did, DAX does not handle dirty
3746		 * page-cache pages correctly).
3747		 */
3748		if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3749			goto out;
3750
3751		status = generic_perform_write(file, from, pos = iocb->ki_pos);
 
3752		/*
3753		 * If generic_perform_write() returned a synchronous error
3754		 * then we want to return the number of bytes which were
3755		 * direct-written, or the error code if that was zero.  Note
3756		 * that this differs from normal direct-io semantics, which
3757		 * will return -EFOO even if some bytes were written.
3758		 */
3759		if (unlikely(status < 0)) {
3760			err = status;
3761			goto out;
3762		}
 
3763		/*
3764		 * We need to ensure that the page cache pages are written to
3765		 * disk and invalidated to preserve the expected O_DIRECT
3766		 * semantics.
3767		 */
3768		endbyte = pos + status - 1;
3769		err = filemap_write_and_wait_range(mapping, pos, endbyte);
3770		if (err == 0) {
3771			iocb->ki_pos = endbyte + 1;
3772			written += status;
3773			invalidate_mapping_pages(mapping,
3774						 pos >> PAGE_SHIFT,
3775						 endbyte >> PAGE_SHIFT);
3776		} else {
3777			/*
3778			 * We don't know how much we wrote, so just return
3779			 * the number of bytes which were direct-written
3780			 */
3781		}
3782	} else {
3783		written = generic_perform_write(file, from, iocb->ki_pos);
3784		if (likely(written > 0))
3785			iocb->ki_pos += written;
3786	}
3787out:
3788	current->backing_dev_info = NULL;
3789	return written ? written : err;
3790}
3791EXPORT_SYMBOL(__generic_file_write_iter);
3792
3793/**
3794 * generic_file_write_iter - write data to a file
3795 * @iocb:	IO state structure
3796 * @from:	iov_iter with data to write
 
 
3797 *
3798 * This is a wrapper around __generic_file_write_iter() to be used by most
3799 * filesystems. It takes care of syncing the file in case of O_SYNC file
3800 * and acquires i_mutex as needed.
3801 * Return:
3802 * * negative error code if no data has been written at all of
3803 *   vfs_fsync_range() failed for a synchronous write
3804 * * number of bytes written, even for truncated writes
3805 */
3806ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 
3807{
3808	struct file *file = iocb->ki_filp;
3809	struct inode *inode = file->f_mapping->host;
 
3810	ssize_t ret;
3811
3812	inode_lock(inode);
3813	ret = generic_write_checks(iocb, from);
3814	if (ret > 0)
3815		ret = __generic_file_write_iter(iocb, from);
3816	inode_unlock(inode);
 
3817
3818	if (ret > 0)
3819		ret = generic_write_sync(iocb, ret);
 
 
 
 
 
 
3820	return ret;
3821}
3822EXPORT_SYMBOL(generic_file_write_iter);
3823
3824/**
3825 * try_to_release_page() - release old fs-specific metadata on a page
3826 *
3827 * @page: the page which the kernel is trying to free
3828 * @gfp_mask: memory allocation flags (and I/O mode)
3829 *
3830 * The address_space is to try to release any data against the page
3831 * (presumably at page->private).
 
3832 *
3833 * This may also be called if PG_fscache is set on a page, indicating that the
3834 * page is known to the local caching routines.
3835 *
3836 * The @gfp_mask argument specifies whether I/O may be performed to release
3837 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3838 *
3839 * Return: %1 if the release was successful, otherwise return zero.
3840 */
3841int try_to_release_page(struct page *page, gfp_t gfp_mask)
3842{
3843	struct address_space * const mapping = page->mapping;
3844
3845	BUG_ON(!PageLocked(page));
3846	if (PageWriteback(page))
3847		return 0;
3848
3849	if (mapping && mapping->a_ops->releasepage)
3850		return mapping->a_ops->releasepage(page, gfp_mask);
3851	return try_to_free_buffers(page);
3852}
3853
3854EXPORT_SYMBOL(try_to_release_page);
v3.5.6
 
   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/export.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/cpuset.h>
  33#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
  34#include <linux/memcontrol.h>
  35#include <linux/cleancache.h>
 
 
 
 
 
 
 
 
  36#include "internal.h"
  37
 
 
 
  38/*
  39 * FIXME: remove all knowledge of the buffer layer from the core VM
  40 */
  41#include <linux/buffer_head.h> /* for try_to_free_buffers */
  42
  43#include <asm/mman.h>
  44
  45/*
  46 * Shared mappings implemented 30.11.1994. It's not fully working yet,
  47 * though.
  48 *
  49 * Shared mappings now work. 15.8.1995  Bruno.
  50 *
  51 * finished 'unifying' the page and buffer cache and SMP-threaded the
  52 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
  53 *
  54 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
  55 */
  56
  57/*
  58 * Lock ordering:
  59 *
  60 *  ->i_mmap_mutex		(truncate_pagecache)
  61 *    ->private_lock		(__free_pte->__set_page_dirty_buffers)
  62 *      ->swap_lock		(exclusive_swap_page, others)
  63 *        ->mapping->tree_lock
  64 *
  65 *  ->i_mutex
  66 *    ->i_mmap_mutex		(truncate->unmap_mapping_range)
  67 *
  68 *  ->mmap_sem
  69 *    ->i_mmap_mutex
  70 *      ->page_table_lock or pte_lock	(various, mainly in memory.c)
  71 *        ->mapping->tree_lock	(arch-dependent flush_dcache_mmap_lock)
  72 *
  73 *  ->mmap_sem
  74 *    ->lock_page		(access_process_vm)
  75 *
  76 *  ->i_mutex			(generic_file_buffered_write)
  77 *    ->mmap_sem		(fault_in_pages_readable->do_page_fault)
  78 *
  79 *  bdi->wb.list_lock
  80 *    sb_lock			(fs/fs-writeback.c)
  81 *    ->mapping->tree_lock	(__sync_single_inode)
  82 *
  83 *  ->i_mmap_mutex
  84 *    ->anon_vma.lock		(vma_adjust)
  85 *
  86 *  ->anon_vma.lock
  87 *    ->page_table_lock or pte_lock	(anon_vma_prepare and various)
  88 *
  89 *  ->page_table_lock or pte_lock
  90 *    ->swap_lock		(try_to_unmap_one)
  91 *    ->private_lock		(try_to_unmap_one)
  92 *    ->tree_lock		(try_to_unmap_one)
  93 *    ->zone.lru_lock		(follow_page->mark_page_accessed)
  94 *    ->zone.lru_lock		(check_pte_range->isolate_lru_page)
  95 *    ->private_lock		(page_remove_rmap->set_page_dirty)
  96 *    ->tree_lock		(page_remove_rmap->set_page_dirty)
  97 *    bdi.wb->list_lock		(page_remove_rmap->set_page_dirty)
  98 *    ->inode->i_lock		(page_remove_rmap->set_page_dirty)
 
  99 *    bdi.wb->list_lock		(zap_pte_range->set_page_dirty)
 100 *    ->inode->i_lock		(zap_pte_range->set_page_dirty)
 101 *    ->private_lock		(zap_pte_range->__set_page_dirty_buffers)
 102 *
 103 * ->i_mmap_mutex
 104 *   ->tasklist_lock            (memory_failure, collect_procs_ao)
 105 */
 106
 107/*
 108 * Delete a page from the page cache and free it. Caller has to make
 109 * sure the page is locked and that nobody else uses it - or that usage
 110 * is safe.  The caller must hold the mapping's tree_lock.
 111 */
 112void __delete_from_page_cache(struct page *page)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 113{
 114	struct address_space *mapping = page->mapping;
 115
 116	/*
 117	 * if we're uptodate, flush out into the cleancache, otherwise
 118	 * invalidate any existing cleancache entries.  We can't leave
 119	 * stale data around in the cleancache once our page is gone
 120	 */
 121	if (PageUptodate(page) && PageMappedToDisk(page))
 122		cleancache_put_page(page);
 123	else
 124		cleancache_invalidate_page(mapping, page);
 125
 126	radix_tree_delete(&mapping->page_tree, page->index);
 127	page->mapping = NULL;
 128	/* Leave page->index set: truncation lookup relies upon it */
 129	mapping->nrpages--;
 130	__dec_zone_page_state(page, NR_FILE_PAGES);
 131	if (PageSwapBacked(page))
 132		__dec_zone_page_state(page, NR_SHMEM);
 133	BUG_ON(page_mapped(page));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 134
 135	/*
 136	 * Some filesystems seem to re-dirty the page even after
 137	 * the VM has canceled the dirty bit (eg ext3 journaling).
 
 138	 *
 139	 * Fix it up by doing a final dirty accounting check after
 140	 * having removed the page entirely.
 
 
 141	 */
 142	if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
 143		dec_zone_page_state(page, NR_FILE_DIRTY);
 144		dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 145	}
 146}
 147
 148/**
 149 * delete_from_page_cache - delete page from page cache
 150 * @page: the page which the kernel is trying to remove from page cache
 151 *
 152 * This must be called only on pages that have been verified to be in the page
 153 * cache and locked.  It will never put the page into the free list, the caller
 154 * has a reference on the page.
 155 */
 156void delete_from_page_cache(struct page *page)
 157{
 158	struct address_space *mapping = page->mapping;
 159	void (*freepage)(struct page *);
 160
 161	BUG_ON(!PageLocked(page));
 
 
 
 162
 163	freepage = mapping->a_ops->freepage;
 164	spin_lock_irq(&mapping->tree_lock);
 165	__delete_from_page_cache(page);
 166	spin_unlock_irq(&mapping->tree_lock);
 167	mem_cgroup_uncharge_cache_page(page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 168
 169	if (freepage)
 170		freepage(page);
 171	page_cache_release(page);
 172}
 173EXPORT_SYMBOL(delete_from_page_cache);
 174
 175static int sleep_on_page(void *word)
 176{
 177	io_schedule();
 178	return 0;
 
 
 
 
 
 
 
 179}
 
 180
 181static int sleep_on_page_killable(void *word)
 182{
 183	sleep_on_page(word);
 184	return fatal_signal_pending(current) ? -EINTR : 0;
 
 
 
 
 185}
 186
 187/**
 188 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
 189 * @mapping:	address space structure to write
 190 * @start:	offset in bytes where the range starts
 191 * @end:	offset in bytes where the range ends (inclusive)
 192 * @sync_mode:	enable synchronous operation
 193 *
 194 * Start writeback against all of a mapping's dirty pages that lie
 195 * within the byte offsets <start, end> inclusive.
 196 *
 197 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
 198 * opposed to a regular memory cleansing writeback.  The difference between
 199 * these two operations is that if a dirty page/buffer is encountered, it must
 200 * be waited upon, and not just skipped over.
 
 
 201 */
 202int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
 203				loff_t end, int sync_mode)
 204{
 205	int ret;
 206	struct writeback_control wbc = {
 207		.sync_mode = sync_mode,
 208		.nr_to_write = LONG_MAX,
 209		.range_start = start,
 210		.range_end = end,
 211	};
 212
 213	if (!mapping_cap_writeback_dirty(mapping))
 
 214		return 0;
 215
 
 216	ret = do_writepages(mapping, &wbc);
 
 217	return ret;
 218}
 219
 220static inline int __filemap_fdatawrite(struct address_space *mapping,
 221	int sync_mode)
 222{
 223	return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
 224}
 225
 226int filemap_fdatawrite(struct address_space *mapping)
 227{
 228	return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
 229}
 230EXPORT_SYMBOL(filemap_fdatawrite);
 231
 232int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
 233				loff_t end)
 234{
 235	return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
 236}
 237EXPORT_SYMBOL(filemap_fdatawrite_range);
 238
 239/**
 240 * filemap_flush - mostly a non-blocking flush
 241 * @mapping:	target address_space
 242 *
 243 * This is a mostly non-blocking flush.  Not suitable for data-integrity
 244 * purposes - I/O may not be started against all dirty pages.
 
 
 245 */
 246int filemap_flush(struct address_space *mapping)
 247{
 248	return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
 249}
 250EXPORT_SYMBOL(filemap_flush);
 251
 252/**
 253 * filemap_fdatawait_range - wait for writeback to complete
 254 * @mapping:		address space structure to wait for
 255 * @start_byte:		offset in bytes where the range starts
 256 * @end_byte:		offset in bytes where the range ends (inclusive)
 
 
 
 257 *
 258 * Walk the list of under-writeback pages of the given address space
 259 * in the given range and wait for all of them.
 260 */
 261int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
 262			    loff_t end_byte)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 263{
 264	pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
 265	pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
 266	struct pagevec pvec;
 267	int nr_pages;
 268	int ret = 0;
 269
 270	if (end_byte < start_byte)
 271		return 0;
 272
 273	pagevec_init(&pvec, 0);
 274	while ((index <= end) &&
 275			(nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 276			PAGECACHE_TAG_WRITEBACK,
 277			min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
 278		unsigned i;
 279
 
 
 
 
 
 280		for (i = 0; i < nr_pages; i++) {
 281			struct page *page = pvec.pages[i];
 282
 283			/* until radix tree lookup accepts end_index */
 284			if (page->index > end)
 285				continue;
 286
 287			wait_on_page_writeback(page);
 288			if (TestClearPageError(page))
 289				ret = -EIO;
 290		}
 291		pagevec_release(&pvec);
 292		cond_resched();
 293	}
 
 294
 295	/* Check for outstanding write errors */
 296	if (test_and_clear_bit(AS_ENOSPC, &mapping->flags))
 297		ret = -ENOSPC;
 298	if (test_and_clear_bit(AS_EIO, &mapping->flags))
 299		ret = -EIO;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 300
 301	return ret;
 
 302}
 303EXPORT_SYMBOL(filemap_fdatawait_range);
 304
 305/**
 306 * filemap_fdatawait - wait for all under-writeback pages to complete
 307 * @mapping: address space structure to wait for
 308 *
 309 * Walk the list of under-writeback pages of the given address space
 310 * and wait for all of them.
 
 
 
 
 
 
 
 311 */
 312int filemap_fdatawait(struct address_space *mapping)
 313{
 314	loff_t i_size = i_size_read(mapping->host);
 
 
 
 315
 316	if (i_size == 0)
 317		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 318
 319	return filemap_fdatawait_range(mapping, 0, i_size - 1);
 
 
 
 
 
 
 
 
 
 
 320}
 321EXPORT_SYMBOL(filemap_fdatawait);
 322
 323int filemap_write_and_wait(struct address_space *mapping)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 324{
 325	int err = 0;
 326
 327	if (mapping->nrpages) {
 328		err = filemap_fdatawrite(mapping);
 
 329		/*
 330		 * Even if the above returned error, the pages may be
 331		 * written partially (e.g. -ENOSPC), so we wait for it.
 332		 * But the -EIO is special case, it may indicate the worst
 333		 * thing (e.g. bug) happened, so we avoid waiting for it.
 334		 */
 335		if (err != -EIO) {
 336			int err2 = filemap_fdatawait(mapping);
 
 337			if (!err)
 338				err = err2;
 
 
 
 339		}
 
 
 340	}
 341	return err;
 342}
 343EXPORT_SYMBOL(filemap_write_and_wait);
 
 
 
 
 
 
 
 
 344
 345/**
 346 * filemap_write_and_wait_range - write out & wait on a file range
 347 * @mapping:	the address_space for the pages
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 348 * @lstart:	offset in bytes where the range starts
 349 * @lend:	offset in bytes where the range ends (inclusive)
 350 *
 351 * Write out and wait upon file offsets lstart->lend, inclusive.
 352 *
 353 * Note that `lend' is inclusive (describes the last byte to be written) so
 354 * that this function can be used to write to the very end-of-file (end = -1).
 
 
 
 
 
 355 */
 356int filemap_write_and_wait_range(struct address_space *mapping,
 357				 loff_t lstart, loff_t lend)
 358{
 359	int err = 0;
 
 360
 361	if (mapping->nrpages) {
 362		err = __filemap_fdatawrite_range(mapping, lstart, lend,
 363						 WB_SYNC_ALL);
 364		/* See comment of filemap_write_and_wait() */
 365		if (err != -EIO) {
 366			int err2 = filemap_fdatawait_range(mapping,
 367						lstart, lend);
 368			if (!err)
 369				err = err2;
 370		}
 371	}
 
 
 
 372	return err;
 373}
 374EXPORT_SYMBOL(filemap_write_and_wait_range);
 375
 376/**
 377 * replace_page_cache_page - replace a pagecache page with a new one
 378 * @old:	page to be replaced
 379 * @new:	page to replace with
 380 * @gfp_mask:	allocation mode
 381 *
 382 * This function replaces a page in the pagecache with a new one.  On
 383 * success it acquires the pagecache reference for the new page and
 384 * drops it for the old page.  Both the old and new pages must be
 385 * locked.  This function does not add the new page to the LRU, the
 386 * caller must do that.
 387 *
 388 * The remove + add is atomic.  The only way this function can fail is
 389 * memory allocation failure.
 390 */
 391int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 392{
 
 
 393	int error;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 394
 395	VM_BUG_ON(!PageLocked(old));
 396	VM_BUG_ON(!PageLocked(new));
 397	VM_BUG_ON(new->mapping);
 398
 399	error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
 400	if (!error) {
 401		struct address_space *mapping = old->mapping;
 402		void (*freepage)(struct page *);
 403
 404		pgoff_t offset = old->index;
 405		freepage = mapping->a_ops->freepage;
 406
 407		page_cache_get(new);
 408		new->mapping = mapping;
 409		new->index = offset;
 410
 411		spin_lock_irq(&mapping->tree_lock);
 412		__delete_from_page_cache(old);
 413		error = radix_tree_insert(&mapping->page_tree, offset, new);
 414		BUG_ON(error);
 415		mapping->nrpages++;
 416		__inc_zone_page_state(new, NR_FILE_PAGES);
 417		if (PageSwapBacked(new))
 418			__inc_zone_page_state(new, NR_SHMEM);
 419		spin_unlock_irq(&mapping->tree_lock);
 420		/* mem_cgroup codes must not be called under tree_lock */
 421		mem_cgroup_replace_page_cache(old, new);
 422		radix_tree_preload_end();
 423		if (freepage)
 424			freepage(old);
 425		page_cache_release(old);
 
 
 
 426	}
 427
 
 
 
 
 
 
 428	return error;
 429}
 430EXPORT_SYMBOL_GPL(replace_page_cache_page);
 431
 432/**
 433 * add_to_page_cache_locked - add a locked page to the pagecache
 434 * @page:	page to add
 435 * @mapping:	the page's address_space
 436 * @offset:	page index
 437 * @gfp_mask:	page allocation mode
 438 *
 439 * This function is used to add a page to the pagecache. It must be locked.
 440 * This function does not add the page to the LRU.  The caller must do that.
 
 
 441 */
 442int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
 443		pgoff_t offset, gfp_t gfp_mask)
 444{
 445	int error;
 446
 447	VM_BUG_ON(!PageLocked(page));
 448	VM_BUG_ON(PageSwapBacked(page));
 449
 450	error = mem_cgroup_cache_charge(page, current->mm,
 451					gfp_mask & GFP_RECLAIM_MASK);
 452	if (error)
 453		goto out;
 454
 455	error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
 456	if (error == 0) {
 457		page_cache_get(page);
 458		page->mapping = mapping;
 459		page->index = offset;
 460
 461		spin_lock_irq(&mapping->tree_lock);
 462		error = radix_tree_insert(&mapping->page_tree, offset, page);
 463		if (likely(!error)) {
 464			mapping->nrpages++;
 465			__inc_zone_page_state(page, NR_FILE_PAGES);
 466			spin_unlock_irq(&mapping->tree_lock);
 467		} else {
 468			page->mapping = NULL;
 469			/* Leave page->index set: truncation relies upon it */
 470			spin_unlock_irq(&mapping->tree_lock);
 471			mem_cgroup_uncharge_cache_page(page);
 472			page_cache_release(page);
 473		}
 474		radix_tree_preload_end();
 475	} else
 476		mem_cgroup_uncharge_cache_page(page);
 477out:
 478	return error;
 479}
 480EXPORT_SYMBOL(add_to_page_cache_locked);
 481
 482int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
 483				pgoff_t offset, gfp_t gfp_mask)
 484{
 
 485	int ret;
 486
 487	ret = add_to_page_cache(page, mapping, offset, gfp_mask);
 488	if (ret == 0)
 489		lru_cache_add_file(page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 490	return ret;
 491}
 492EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
 493
 494#ifdef CONFIG_NUMA
 495struct page *__page_cache_alloc(gfp_t gfp)
 496{
 497	int n;
 498	struct page *page;
 499
 500	if (cpuset_do_page_mem_spread()) {
 501		unsigned int cpuset_mems_cookie;
 502		do {
 503			cpuset_mems_cookie = get_mems_allowed();
 504			n = cpuset_mem_spread_node();
 505			page = alloc_pages_exact_node(n, gfp, 0);
 506		} while (!put_mems_allowed(cpuset_mems_cookie) && !page);
 507
 508		return page;
 509	}
 510	return alloc_pages(gfp, 0);
 511}
 512EXPORT_SYMBOL(__page_cache_alloc);
 513#endif
 514
 515/*
 516 * In order to wait for pages to become available there must be
 517 * waitqueues associated with pages. By using a hash table of
 518 * waitqueues where the bucket discipline is to maintain all
 519 * waiters on the same queue and wake all when any of the pages
 520 * become available, and for the woken contexts to check to be
 521 * sure the appropriate page became available, this saves space
 522 * at a cost of "thundering herd" phenomena during rare hash
 523 * collisions.
 524 */
 
 
 
 
 525static wait_queue_head_t *page_waitqueue(struct page *page)
 526{
 527	const struct zone *zone = page_zone(page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 528
 529	return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 530}
 531
 532static inline void wake_up_page(struct page *page, int bit)
 
 
 
 
 533{
 534	__wake_up_bit(page_waitqueue(page), &page->flags, bit);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 535}
 536
 537void wait_on_page_bit(struct page *page, int bit_nr)
 538{
 539	DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
 540
 541	if (test_bit(bit_nr, &page->flags))
 542		__wait_on_bit(page_waitqueue(page), &wait, sleep_on_page,
 543							TASK_UNINTERRUPTIBLE);
 544}
 545EXPORT_SYMBOL(wait_on_page_bit);
 546
 547int wait_on_page_bit_killable(struct page *page, int bit_nr)
 548{
 549	DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
 
 
 
 550
 551	if (!test_bit(bit_nr, &page->flags))
 552		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 553
 554	return __wait_on_bit(page_waitqueue(page), &wait,
 555			     sleep_on_page_killable, TASK_KILLABLE);
 
 556}
 557
 558/**
 559 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
 560 * @page: Page defining the wait queue of interest
 561 * @waiter: Waiter to add to the queue
 562 *
 563 * Add an arbitrary @waiter to the wait queue for the nominated @page.
 564 */
 565void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
 566{
 567	wait_queue_head_t *q = page_waitqueue(page);
 568	unsigned long flags;
 569
 570	spin_lock_irqsave(&q->lock, flags);
 571	__add_wait_queue(q, waiter);
 
 572	spin_unlock_irqrestore(&q->lock, flags);
 573}
 574EXPORT_SYMBOL_GPL(add_page_wait_queue);
 575
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 576/**
 577 * unlock_page - unlock a locked page
 578 * @page: the page
 579 *
 580 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
 581 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
 582 * mechananism between PageLocked pages and PageWriteback pages is shared.
 583 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
 584 *
 585 * The mb is necessary to enforce ordering between the clear_bit and the read
 586 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
 
 
 
 587 */
 588void unlock_page(struct page *page)
 589{
 590	VM_BUG_ON(!PageLocked(page));
 591	clear_bit_unlock(PG_locked, &page->flags);
 592	smp_mb__after_clear_bit();
 593	wake_up_page(page, PG_locked);
 
 594}
 595EXPORT_SYMBOL(unlock_page);
 596
 597/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 598 * end_page_writeback - end writeback against a page
 599 * @page: the page
 600 */
 601void end_page_writeback(struct page *page)
 602{
 603	if (TestClearPageReclaim(page))
 
 
 
 
 
 
 
 
 604		rotate_reclaimable_page(page);
 
 605
 
 
 
 
 
 
 
 606	if (!test_clear_page_writeback(page))
 607		BUG();
 608
 609	smp_mb__after_clear_bit();
 610	wake_up_page(page, PG_writeback);
 
 611}
 612EXPORT_SYMBOL(end_page_writeback);
 613
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 614/**
 615 * __lock_page - get a lock on the page, assuming we need to sleep to get it
 616 * @page: the page to lock
 617 */
 618void __lock_page(struct page *page)
 619{
 620	DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
 
 
 
 
 
 621
 622	__wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page,
 623							TASK_UNINTERRUPTIBLE);
 
 
 
 
 624}
 625EXPORT_SYMBOL(__lock_page);
 626
 627int __lock_page_killable(struct page *page)
 628{
 629	DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
 
 630
 631	return __wait_on_bit_lock(page_waitqueue(page), &wait,
 632					sleep_on_page_killable, TASK_KILLABLE);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 633}
 634EXPORT_SYMBOL_GPL(__lock_page_killable);
 635
 
 
 
 
 
 
 
 
 
 
 
 636int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
 637			 unsigned int flags)
 638{
 639	if (flags & FAULT_FLAG_ALLOW_RETRY) {
 640		/*
 641		 * CAUTION! In this case, mmap_sem is not released
 642		 * even though return 0.
 643		 */
 644		if (flags & FAULT_FLAG_RETRY_NOWAIT)
 645			return 0;
 646
 647		up_read(&mm->mmap_sem);
 648		if (flags & FAULT_FLAG_KILLABLE)
 649			wait_on_page_locked_killable(page);
 650		else
 651			wait_on_page_locked(page);
 652		return 0;
 
 
 
 
 
 
 
 
 
 653	} else {
 654		if (flags & FAULT_FLAG_KILLABLE) {
 655			int ret;
 
 
 
 656
 657			ret = __lock_page_killable(page);
 658			if (ret) {
 659				up_read(&mm->mmap_sem);
 660				return 0;
 661			}
 662		} else
 663			__lock_page(page);
 664		return 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 665	}
 
 
 666}
 
 667
 668/**
 669 * find_get_page - find and get a page reference
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 670 * @mapping: the address_space to search
 671 * @offset: the page index
 672 *
 673 * Is there a pagecache struct page at the given (mapping, offset) tuple?
 674 * If yes, increment its refcount and return it; if no, return NULL.
 
 
 
 
 
 675 */
 676struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
 
 677{
 678	void **pagep;
 679	struct page *page;
 680
 681	rcu_read_lock();
 682repeat:
 683	page = NULL;
 684	pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
 685	if (pagep) {
 686		page = radix_tree_deref_slot(pagep);
 687		if (unlikely(!page))
 688			goto out;
 689		if (radix_tree_exception(page)) {
 690			if (radix_tree_deref_retry(page))
 691				goto repeat;
 692			/*
 693			 * Otherwise, shmem/tmpfs must be storing a swap entry
 694			 * here as an exceptional entry: so return it without
 695			 * attempting to raise page count.
 696			 */
 697			goto out;
 698		}
 699		if (!page_cache_get_speculative(page))
 700			goto repeat;
 701
 702		/*
 703		 * Has the page moved?
 704		 * This is part of the lockless pagecache protocol. See
 705		 * include/linux/pagemap.h for details.
 706		 */
 707		if (unlikely(page != *pagep)) {
 708			page_cache_release(page);
 709			goto repeat;
 710		}
 711	}
 712out:
 713	rcu_read_unlock();
 714
 715	return page;
 716}
 717EXPORT_SYMBOL(find_get_page);
 718
 719/**
 720 * find_lock_page - locate, pin and lock a pagecache page
 721 * @mapping: the address_space to search
 722 * @offset: the page index
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 723 *
 724 * Locates the desired pagecache page, locks it, increments its reference
 725 * count and returns its address.
 726 *
 727 * Returns zero if the page was not present. find_lock_page() may sleep.
 728 */
 729struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
 
 730{
 731	struct page *page;
 732
 733repeat:
 734	page = find_get_page(mapping, offset);
 735	if (page && !radix_tree_exception(page)) {
 736		lock_page(page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 737		/* Has the page been truncated? */
 738		if (unlikely(page->mapping != mapping)) {
 739			unlock_page(page);
 740			page_cache_release(page);
 741			goto repeat;
 742		}
 743		VM_BUG_ON(page->index != offset);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 744	}
 
 745	return page;
 
 
 
 746}
 747EXPORT_SYMBOL(find_lock_page);
 748
 749/**
 750 * find_or_create_page - locate or add a pagecache page
 751 * @mapping: the page's address_space
 752 * @index: the page's index into the mapping
 753 * @gfp_mask: page allocation mode
 
 
 
 
 
 
 754 *
 755 * Locates a page in the pagecache.  If the page is not present, a new page
 756 * is allocated using @gfp_mask and is added to the pagecache and to the VM's
 757 * LRU list.  The returned page is locked and has its reference count
 758 * incremented.
 759 *
 760 * find_or_create_page() may sleep, even if @gfp_flags specifies an atomic
 761 * allocation!
 762 *
 763 * find_or_create_page() returns the desired page's address, or zero on
 764 * memory exhaustion.
 
 
 
 
 765 */
 766struct page *find_or_create_page(struct address_space *mapping,
 767		pgoff_t index, gfp_t gfp_mask)
 768{
 
 769	struct page *page;
 770	int err;
 771repeat:
 772	page = find_lock_page(mapping, index);
 773	if (!page) {
 774		page = __page_cache_alloc(gfp_mask);
 775		if (!page)
 776			return NULL;
 777		/*
 778		 * We want a regular kernel memory (not highmem or DMA etc)
 779		 * allocation for the radix tree nodes, but we need to honour
 780		 * the context-specific requirements the caller has asked for.
 781		 * GFP_RECLAIM_MASK collects those requirements.
 782		 */
 783		err = add_to_page_cache_lru(page, mapping, index,
 784			(gfp_mask & GFP_RECLAIM_MASK));
 785		if (unlikely(err)) {
 786			page_cache_release(page);
 787			page = NULL;
 788			if (err == -EEXIST)
 789				goto repeat;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 790		}
 791	}
 792	return page;
 
 
 793}
 794EXPORT_SYMBOL(find_or_create_page);
 795
 796/**
 797 * find_get_pages - gang pagecache lookup
 798 * @mapping:	The address_space to search
 799 * @start:	The starting page index
 
 800 * @nr_pages:	The maximum number of pages
 801 * @pages:	Where the resulting pages are placed
 802 *
 803 * find_get_pages() will search for and return a group of up to
 804 * @nr_pages pages in the mapping.  The pages are placed at @pages.
 805 * find_get_pages() takes a reference against the returned pages.
 
 806 *
 807 * The search returns a group of mapping-contiguous pages with ascending
 808 * indexes.  There may be holes in the indices due to not-present pages.
 
 809 *
 810 * find_get_pages() returns the number of pages which were found.
 
 
 811 */
 812unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
 813			    unsigned int nr_pages, struct page **pages)
 
 814{
 815	struct radix_tree_iter iter;
 816	void **slot;
 817	unsigned ret = 0;
 818
 819	if (unlikely(!nr_pages))
 820		return 0;
 821
 822	rcu_read_lock();
 823restart:
 824	radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
 825		struct page *page;
 826repeat:
 827		page = radix_tree_deref_slot(slot);
 828		if (unlikely(!page))
 829			continue;
 830
 831		if (radix_tree_exception(page)) {
 832			if (radix_tree_deref_retry(page)) {
 833				/*
 834				 * Transient condition which can only trigger
 835				 * when entry at index 0 moves out of or back
 836				 * to root: none yet gotten, safe to restart.
 837				 */
 838				WARN_ON(iter.index);
 839				goto restart;
 840			}
 841			/*
 842			 * Otherwise, shmem/tmpfs must be storing a swap entry
 843			 * here as an exceptional entry: so skip over it -
 844			 * we only reach this from invalidate_mapping_pages().
 845			 */
 846			continue;
 847		}
 848
 849		if (!page_cache_get_speculative(page))
 850			goto repeat;
 851
 852		/* Has the page moved? */
 853		if (unlikely(page != *slot)) {
 854			page_cache_release(page);
 855			goto repeat;
 856		}
 857
 858		pages[ret] = page;
 859		if (++ret == nr_pages)
 860			break;
 861	}
 862
 
 
 
 
 
 
 
 
 
 
 
 863	rcu_read_unlock();
 
 864	return ret;
 865}
 866
 867/**
 868 * find_get_pages_contig - gang contiguous pagecache lookup
 869 * @mapping:	The address_space to search
 870 * @index:	The starting page index
 871 * @nr_pages:	The maximum number of pages
 872 * @pages:	Where the resulting pages are placed
 873 *
 874 * find_get_pages_contig() works exactly like find_get_pages(), except
 875 * that the returned number of pages are guaranteed to be contiguous.
 876 *
 877 * find_get_pages_contig() returns the number of pages which were found.
 878 */
 879unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
 880			       unsigned int nr_pages, struct page **pages)
 881{
 882	struct radix_tree_iter iter;
 883	void **slot;
 884	unsigned int ret = 0;
 885
 886	if (unlikely(!nr_pages))
 887		return 0;
 888
 889	rcu_read_lock();
 890restart:
 891	radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
 892		struct page *page;
 893repeat:
 894		page = radix_tree_deref_slot(slot);
 895		/* The hole, there no reason to continue */
 896		if (unlikely(!page))
 
 897			break;
 898
 899		if (radix_tree_exception(page)) {
 900			if (radix_tree_deref_retry(page)) {
 901				/*
 902				 * Transient condition which can only trigger
 903				 * when entry at index 0 moves out of or back
 904				 * to root: none yet gotten, safe to restart.
 905				 */
 906				goto restart;
 907			}
 908			/*
 909			 * Otherwise, shmem/tmpfs must be storing a swap entry
 910			 * here as an exceptional entry: so stop looking for
 911			 * contiguous pages.
 912			 */
 913			break;
 914		}
 915
 916		if (!page_cache_get_speculative(page))
 917			goto repeat;
 918
 919		/* Has the page moved? */
 920		if (unlikely(page != *slot)) {
 921			page_cache_release(page);
 922			goto repeat;
 923		}
 924
 925		/*
 926		 * must check mapping and index after taking the ref.
 927		 * otherwise we can get both false positives and false
 928		 * negatives, which is just confusing to the caller.
 929		 */
 930		if (page->mapping == NULL || page->index != iter.index) {
 931			page_cache_release(page);
 932			break;
 933		}
 934
 935		pages[ret] = page;
 936		if (++ret == nr_pages)
 937			break;
 
 
 
 
 
 938	}
 939	rcu_read_unlock();
 940	return ret;
 941}
 942EXPORT_SYMBOL(find_get_pages_contig);
 943
 944/**
 945 * find_get_pages_tag - find and return pages that match @tag
 946 * @mapping:	the address_space to search
 947 * @index:	the starting page index
 
 948 * @tag:	the tag index
 949 * @nr_pages:	the maximum number of pages
 950 * @pages:	where the resulting pages are placed
 951 *
 952 * Like find_get_pages, except we only return pages which are tagged with
 953 * @tag.   We update @index to index the next page for the traversal.
 
 
 
 954 */
 955unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
 956			int tag, unsigned int nr_pages, struct page **pages)
 
 957{
 958	struct radix_tree_iter iter;
 959	void **slot;
 960	unsigned ret = 0;
 961
 962	if (unlikely(!nr_pages))
 963		return 0;
 964
 965	rcu_read_lock();
 966restart:
 967	radix_tree_for_each_tagged(slot, &mapping->page_tree,
 968				   &iter, *index, tag) {
 969		struct page *page;
 970repeat:
 971		page = radix_tree_deref_slot(slot);
 972		if (unlikely(!page))
 973			continue;
 974
 975		if (radix_tree_exception(page)) {
 976			if (radix_tree_deref_retry(page)) {
 977				/*
 978				 * Transient condition which can only trigger
 979				 * when entry at index 0 moves out of or back
 980				 * to root: none yet gotten, safe to restart.
 981				 */
 982				goto restart;
 983			}
 984			/*
 985			 * This function is never used on a shmem/tmpfs
 986			 * mapping, so a swap entry won't be found here.
 987			 */
 988			BUG();
 989		}
 990
 991		if (!page_cache_get_speculative(page))
 992			goto repeat;
 993
 994		/* Has the page moved? */
 995		if (unlikely(page != *slot)) {
 996			page_cache_release(page);
 997			goto repeat;
 998		}
 999
1000		pages[ret] = page;
1001		if (++ret == nr_pages)
1002			break;
1003	}
1004
 
 
 
 
 
 
 
 
 
 
 
1005	rcu_read_unlock();
1006
1007	if (ret)
1008		*index = pages[ret - 1]->index + 1;
1009
1010	return ret;
1011}
1012EXPORT_SYMBOL(find_get_pages_tag);
1013
1014/**
1015 * grab_cache_page_nowait - returns locked page at given index in given cache
1016 * @mapping: target address_space
1017 * @index: the page index
1018 *
1019 * Same as grab_cache_page(), but do not wait if the page is unavailable.
1020 * This is intended for speculative data generators, where the data can
1021 * be regenerated if the page couldn't be grabbed.  This routine should
1022 * be safe to call while holding the lock for another page.
1023 *
1024 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
1025 * and deadlock against the caller's locked page.
1026 */
1027struct page *
1028grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
1029{
1030	struct page *page = find_get_page(mapping, index);
1031
1032	if (page) {
1033		if (trylock_page(page))
1034			return page;
1035		page_cache_release(page);
1036		return NULL;
1037	}
1038	page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
1039	if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
1040		page_cache_release(page);
1041		page = NULL;
1042	}
1043	return page;
1044}
1045EXPORT_SYMBOL(grab_cache_page_nowait);
1046
1047/*
1048 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1049 * a _large_ part of the i/o request. Imagine the worst scenario:
1050 *
1051 *      ---R__________________________________________B__________
1052 *         ^ reading here                             ^ bad block(assume 4k)
1053 *
1054 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1055 * => failing the whole request => read(R) => read(R+1) =>
1056 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1057 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1058 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1059 *
1060 * It is going insane. Fix it by quickly scaling down the readahead size.
1061 */
1062static void shrink_readahead_size_eio(struct file *filp,
1063					struct file_ra_state *ra)
1064{
1065	ra->ra_pages /= 4;
1066}
1067
1068/**
1069 * do_generic_file_read - generic file read routine
1070 * @filp:	the file to read
1071 * @ppos:	current file position
1072 * @desc:	read_descriptor
1073 * @actor:	read method
1074 *
1075 * This is a generic file read routine, and uses the
1076 * mapping->a_ops->readpage() function for the actual low-level stuff.
1077 *
1078 * This is really ugly. But the goto's actually try to clarify some
1079 * of the logic when it comes to error handling etc.
 
 
 
1080 */
1081static void do_generic_file_read(struct file *filp, loff_t *ppos,
1082		read_descriptor_t *desc, read_actor_t actor)
1083{
1084	struct address_space *mapping = filp->f_mapping;
1085	struct inode *inode = mapping->host;
1086	struct file_ra_state *ra = &filp->f_ra;
1087	pgoff_t index;
1088	pgoff_t last_index;
1089	pgoff_t prev_index;
1090	unsigned long offset;      /* offset into pagecache page */
1091	unsigned int prev_offset;
1092	int error;
1093
1094	index = *ppos >> PAGE_CACHE_SHIFT;
1095	prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1096	prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1097	last_index = (*ppos + desc->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1098	offset = *ppos & ~PAGE_CACHE_MASK;
 
 
 
1099
1100	for (;;) {
1101		struct page *page;
1102		pgoff_t end_index;
1103		loff_t isize;
1104		unsigned long nr, ret;
1105
1106		cond_resched();
1107find_page:
1108		page = find_get_page(mapping, index);
1109		if (!page) {
1110			page_cache_sync_readahead(mapping,
1111					ra, filp,
1112					index, last_index - index);
1113			page = find_get_page(mapping, index);
1114			if (unlikely(page == NULL))
1115				goto no_cached_page;
1116		}
1117		if (PageReadahead(page)) {
1118			page_cache_async_readahead(mapping,
1119					ra, filp, page,
1120					index, last_index - index);
1121		}
1122		if (!PageUptodate(page)) {
1123			if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
1124					!mapping->a_ops->is_partially_uptodate)
1125				goto page_not_up_to_date;
1126			if (!trylock_page(page))
1127				goto page_not_up_to_date;
1128			/* Did it get truncated before we got the lock? */
1129			if (!page->mapping)
1130				goto page_not_up_to_date_locked;
1131			if (!mapping->a_ops->is_partially_uptodate(page,
1132								desc, offset))
1133				goto page_not_up_to_date_locked;
1134			unlock_page(page);
1135		}
1136page_ok:
1137		/*
1138		 * i_size must be checked after we know the page is Uptodate.
1139		 *
1140		 * Checking i_size after the check allows us to calculate
1141		 * the correct value for "nr", which means the zero-filled
1142		 * part of the page is not copied back to userspace (unless
1143		 * another truncate extends the file - this is desired though).
1144		 */
1145
1146		isize = i_size_read(inode);
1147		end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1148		if (unlikely(!isize || index > end_index)) {
1149			page_cache_release(page);
1150			goto out;
1151		}
1152
1153		/* nr is the maximum number of bytes to copy from this page */
1154		nr = PAGE_CACHE_SIZE;
1155		if (index == end_index) {
1156			nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1157			if (nr <= offset) {
1158				page_cache_release(page);
1159				goto out;
1160			}
1161		}
1162		nr = nr - offset;
1163
1164		/* If users can be writing to this page using arbitrary
1165		 * virtual addresses, take care about potential aliasing
1166		 * before reading the page on the kernel side.
1167		 */
1168		if (mapping_writably_mapped(mapping))
1169			flush_dcache_page(page);
 
 
1170
1171		/*
1172		 * When a sequential read accesses a page several times,
1173		 * only mark it as accessed the first time.
1174		 */
1175		if (prev_index != index || offset != prev_offset)
1176			mark_page_accessed(page);
1177		prev_index = index;
1178
1179		/*
1180		 * Ok, we have the page, and it's up-to-date, so
1181		 * now we can copy it to user space...
1182		 *
1183		 * The actor routine returns how many bytes were actually used..
1184		 * NOTE! This may not be the same as how much of a user buffer
1185		 * we filled up (we may be padding etc), so we can only update
1186		 * "pos" here (the actor routine has to update the user buffer
1187		 * pointers and the remaining count).
1188		 */
1189		ret = actor(desc, page, offset, nr);
1190		offset += ret;
1191		index += offset >> PAGE_CACHE_SHIFT;
1192		offset &= ~PAGE_CACHE_MASK;
1193		prev_offset = offset;
 
 
1194
1195		page_cache_release(page);
1196		if (ret == nr && desc->count)
1197			continue;
1198		goto out;
1199
1200page_not_up_to_date:
1201		/* Get exclusive access to the page ... */
1202		error = lock_page_killable(page);
1203		if (unlikely(error))
1204			goto readpage_error;
1205
1206page_not_up_to_date_locked:
1207		/* Did it get truncated before we got the lock? */
1208		if (!page->mapping) {
1209			unlock_page(page);
1210			page_cache_release(page);
1211			continue;
1212		}
1213
1214		/* Did somebody else fill it already? */
1215		if (PageUptodate(page)) {
1216			unlock_page(page);
1217			goto page_ok;
1218		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1219
1220readpage:
1221		/*
1222		 * A previous I/O error may have been due to temporary
1223		 * failures, eg. multipath errors.
1224		 * PG_error will be set again if readpage fails.
1225		 */
1226		ClearPageError(page);
1227		/* Start the actual read. The read will unlock the page. */
1228		error = mapping->a_ops->readpage(filp, page);
 
 
 
1229
1230		if (unlikely(error)) {
1231			if (error == AOP_TRUNCATED_PAGE) {
1232				page_cache_release(page);
1233				goto find_page;
1234			}
1235			goto readpage_error;
1236		}
1237
1238		if (!PageUptodate(page)) {
1239			error = lock_page_killable(page);
1240			if (unlikely(error))
1241				goto readpage_error;
1242			if (!PageUptodate(page)) {
1243				if (page->mapping == NULL) {
1244					/*
1245					 * invalidate_mapping_pages got it
1246					 */
1247					unlock_page(page);
1248					page_cache_release(page);
1249					goto find_page;
1250				}
1251				unlock_page(page);
1252				shrink_readahead_size_eio(filp, ra);
1253				error = -EIO;
1254				goto readpage_error;
1255			}
1256			unlock_page(page);
1257		}
1258
1259		goto page_ok;
 
 
 
 
 
1260
1261readpage_error:
1262		/* UHHUH! A synchronous read error occurred. Report it */
1263		desc->error = error;
1264		page_cache_release(page);
1265		goto out;
1266
1267no_cached_page:
1268		/*
1269		 * Ok, it wasn't cached, so we need to create a new
1270		 * page..
1271		 */
1272		page = page_cache_alloc_cold(mapping);
1273		if (!page) {
1274			desc->error = -ENOMEM;
1275			goto out;
1276		}
1277		error = add_to_page_cache_lru(page, mapping,
1278						index, GFP_KERNEL);
1279		if (error) {
1280			page_cache_release(page);
1281			if (error == -EEXIST)
1282				goto find_page;
1283			desc->error = error;
1284			goto out;
1285		}
1286		goto readpage;
1287	}
1288
1289out:
1290	ra->prev_pos = prev_index;
1291	ra->prev_pos <<= PAGE_CACHE_SHIFT;
1292	ra->prev_pos |= prev_offset;
1293
1294	*ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1295	file_accessed(filp);
 
 
1296}
1297
1298int file_read_actor(read_descriptor_t *desc, struct page *page,
1299			unsigned long offset, unsigned long size)
1300{
1301	char *kaddr;
1302	unsigned long left, count = desc->count;
 
 
 
 
 
1303
1304	if (size > count)
1305		size = count;
 
 
1306
1307	/*
1308	 * Faults on the destination of a read are common, so do it before
1309	 * taking the kmap.
1310	 */
1311	if (!fault_in_pages_writeable(desc->arg.buf, size)) {
1312		kaddr = kmap_atomic(page);
1313		left = __copy_to_user_inatomic(desc->arg.buf,
1314						kaddr + offset, size);
1315		kunmap_atomic(kaddr);
1316		if (left == 0)
1317			goto success;
 
 
 
 
 
1318	}
1319
1320	/* Do it the slow way */
1321	kaddr = kmap(page);
1322	left = __copy_to_user(desc->arg.buf, kaddr + offset, size);
1323	kunmap(page);
 
 
 
 
 
 
 
 
 
1324
1325	if (left) {
1326		size -= left;
1327		desc->error = -EFAULT;
1328	}
1329success:
1330	desc->count = count - size;
1331	desc->written += size;
1332	desc->arg.buf += size;
1333	return size;
1334}
1335
1336/*
1337 * Performs necessary checks before doing a write
1338 * @iov:	io vector request
1339 * @nr_segs:	number of segments in the iovec
1340 * @count:	number of bytes to write
1341 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1342 *
1343 * Adjust number of segments and amount of bytes to write (nr_segs should be
1344 * properly initialized first). Returns appropriate error code that caller
1345 * should return or zero in case that write should be allowed.
 
 
1346 */
1347int generic_segment_checks(const struct iovec *iov,
1348			unsigned long *nr_segs, size_t *count, int access_flags)
1349{
1350	unsigned long   seg;
1351	size_t cnt = 0;
1352	for (seg = 0; seg < *nr_segs; seg++) {
1353		const struct iovec *iv = &iov[seg];
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1354
1355		/*
1356		 * If any segment has a negative length, or the cumulative
1357		 * length ever wraps negative then return -EINVAL.
1358		 */
1359		cnt += iv->iov_len;
1360		if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1361			return -EINVAL;
1362		if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1363			continue;
1364		if (seg == 0)
1365			return -EFAULT;
1366		*nr_segs = seg;
1367		cnt -= iv->iov_len;	/* This segment is no good */
1368		break;
1369	}
1370	*count = cnt;
1371	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1372}
1373EXPORT_SYMBOL(generic_segment_checks);
1374
1375/**
1376 * generic_file_aio_read - generic filesystem read routine
1377 * @iocb:	kernel I/O control block
1378 * @iov:	io vector request
1379 * @nr_segs:	number of segments in the iovec
1380 * @pos:	current file position
1381 *
1382 * This is the "read()" routine for all filesystems
1383 * that can use the page cache directly.
 
 
 
 
 
 
 
 
 
 
 
 
 
1384 */
1385ssize_t
1386generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1387		unsigned long nr_segs, loff_t pos)
1388{
1389	struct file *filp = iocb->ki_filp;
1390	ssize_t retval;
1391	unsigned long seg = 0;
1392	size_t count;
1393	loff_t *ppos = &iocb->ki_pos;
1394
1395	count = 0;
1396	retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1397	if (retval)
1398		return retval;
1399
1400	/* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1401	if (filp->f_flags & O_DIRECT) {
 
 
1402		loff_t size;
1403		struct address_space *mapping;
1404		struct inode *inode;
1405
1406		mapping = filp->f_mapping;
1407		inode = mapping->host;
1408		if (!count)
1409			goto out; /* skip atime */
1410		size = i_size_read(inode);
1411		if (pos < size) {
1412			retval = filemap_write_and_wait_range(mapping, pos,
1413					pos + iov_length(iov, nr_segs) - 1);
1414			if (!retval) {
1415				struct blk_plug plug;
1416
1417				blk_start_plug(&plug);
1418				retval = mapping->a_ops->direct_IO(READ, iocb,
1419							iov, pos, nr_segs);
1420				blk_finish_plug(&plug);
1421			}
1422			if (retval > 0) {
1423				*ppos = pos + retval;
1424				count -= retval;
1425			}
 
 
 
 
 
 
1426
1427			/*
1428			 * Btrfs can have a short DIO read if we encounter
1429			 * compressed extents, so if there was an error, or if
1430			 * we've already read everything we wanted to, or if
1431			 * there was a short read because we hit EOF, go ahead
1432			 * and return.  Otherwise fallthrough to buffered io for
1433			 * the rest of the read.
1434			 */
1435			if (retval < 0 || !count || *ppos >= size) {
1436				file_accessed(filp);
1437				goto out;
1438			}
1439		}
1440	}
1441
1442	count = retval;
1443	for (seg = 0; seg < nr_segs; seg++) {
1444		read_descriptor_t desc;
1445		loff_t offset = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1446
1447		/*
1448		 * If we did a short DIO read we need to skip the section of the
1449		 * iov that we've already read data into.
1450		 */
1451		if (count) {
1452			if (count > iov[seg].iov_len) {
1453				count -= iov[seg].iov_len;
1454				continue;
1455			}
1456			offset = count;
1457			count = 0;
1458		}
1459
1460		desc.written = 0;
1461		desc.arg.buf = iov[seg].iov_base + offset;
1462		desc.count = iov[seg].iov_len - offset;
1463		if (desc.count == 0)
1464			continue;
1465		desc.error = 0;
1466		do_generic_file_read(filp, ppos, &desc, file_read_actor);
1467		retval += desc.written;
1468		if (desc.error) {
1469			retval = retval ?: desc.error;
1470			break;
1471		}
1472		if (desc.count > 0)
1473			break;
1474	}
1475out:
1476	return retval;
 
 
 
 
 
 
 
 
 
1477}
1478EXPORT_SYMBOL(generic_file_aio_read);
1479
1480#ifdef CONFIG_MMU
1481/**
1482 * page_cache_read - adds requested page to the page cache if not already there
1483 * @file:	file to read
1484 * @offset:	page index
 
 
 
 
 
 
 
 
1485 *
1486 * This adds the requested page to the page cache if it isn't already there,
1487 * and schedules an I/O to read in its contents from disk.
 
 
1488 */
1489static int page_cache_read(struct file *file, pgoff_t offset)
 
1490{
1491	struct address_space *mapping = file->f_mapping;
1492	struct page *page; 
1493	int ret;
 
 
 
 
1494
1495	do {
1496		page = page_cache_alloc_cold(mapping);
1497		if (!page)
1498			return -ENOMEM;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1499
1500		ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1501		if (ret == 0)
1502			ret = mapping->a_ops->readpage(file, page);
1503		else if (ret == -EEXIST)
1504			ret = 0; /* losing race to add is OK */
 
 
 
 
 
 
 
 
 
 
 
 
 
1505
1506		page_cache_release(page);
 
 
 
 
 
 
1507
1508	} while (ret == AOP_TRUNCATED_PAGE);
1509		
1510	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
1511}
1512
1513#define MMAP_LOTSAMISS  (100)
1514
1515/*
1516 * Synchronous readahead happens when we don't even find
1517 * a page in the page cache at all.
 
 
 
1518 */
1519static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1520				   struct file_ra_state *ra,
1521				   struct file *file,
1522				   pgoff_t offset)
1523{
1524	unsigned long ra_pages;
 
1525	struct address_space *mapping = file->f_mapping;
 
 
 
1526
1527	/* If we don't want any read-ahead, don't bother */
1528	if (VM_RandomReadHint(vma))
1529		return;
1530	if (!ra->ra_pages)
1531		return;
1532
1533	if (VM_SequentialReadHint(vma)) {
1534		page_cache_sync_readahead(mapping, ra, file, offset,
1535					  ra->ra_pages);
1536		return;
1537	}
1538
1539	/* Avoid banging the cache line if not needed */
1540	if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
1541		ra->mmap_miss++;
 
1542
1543	/*
1544	 * Do we miss much more than hit in this file? If so,
1545	 * stop bothering with read-ahead. It will only hurt.
1546	 */
1547	if (ra->mmap_miss > MMAP_LOTSAMISS)
1548		return;
1549
1550	/*
1551	 * mmap read-around
1552	 */
1553	ra_pages = max_sane_readahead(ra->ra_pages);
1554	ra->start = max_t(long, 0, offset - ra_pages / 2);
1555	ra->size = ra_pages;
1556	ra->async_size = ra_pages / 4;
1557	ra_submit(ra, mapping, file);
 
 
1558}
1559
1560/*
1561 * Asynchronous readahead happens when we find the page and PG_readahead,
1562 * so we want to possibly extend the readahead further..
 
1563 */
1564static void do_async_mmap_readahead(struct vm_area_struct *vma,
1565				    struct file_ra_state *ra,
1566				    struct file *file,
1567				    struct page *page,
1568				    pgoff_t offset)
1569{
 
 
1570	struct address_space *mapping = file->f_mapping;
 
 
 
1571
1572	/* If we don't want any read-ahead, don't bother */
1573	if (VM_RandomReadHint(vma))
1574		return;
1575	if (ra->mmap_miss > 0)
1576		ra->mmap_miss--;
1577	if (PageReadahead(page))
 
 
1578		page_cache_async_readahead(mapping, ra, file,
1579					   page, offset, ra->ra_pages);
 
 
1580}
1581
1582/**
1583 * filemap_fault - read in file data for page fault handling
1584 * @vma:	vma in which the fault was taken
1585 * @vmf:	struct vm_fault containing details of the fault
1586 *
1587 * filemap_fault() is invoked via the vma operations vector for a
1588 * mapped memory region to read in file data during a page fault.
1589 *
1590 * The goto's are kind of ugly, but this streamlines the normal case of having
1591 * it in the page cache, and handles the special cases reasonably without
1592 * having a lot of duplicated code.
 
 
 
 
 
 
 
 
 
 
 
 
1593 */
1594int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1595{
1596	int error;
1597	struct file *file = vma->vm_file;
 
1598	struct address_space *mapping = file->f_mapping;
1599	struct file_ra_state *ra = &file->f_ra;
1600	struct inode *inode = mapping->host;
1601	pgoff_t offset = vmf->pgoff;
 
1602	struct page *page;
1603	pgoff_t size;
1604	int ret = 0;
1605
1606	size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1607	if (offset >= size)
1608		return VM_FAULT_SIGBUS;
1609
1610	/*
1611	 * Do we have something in the page cache already?
1612	 */
1613	page = find_get_page(mapping, offset);
1614	if (likely(page)) {
1615		/*
1616		 * We found the page, so try async readahead before
1617		 * waiting for the lock.
1618		 */
1619		do_async_mmap_readahead(vma, ra, file, page, offset);
1620	} else {
1621		/* No page in the page cache at all */
1622		do_sync_mmap_readahead(vma, ra, file, offset);
1623		count_vm_event(PGMAJFAULT);
1624		mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1625		ret = VM_FAULT_MAJOR;
 
1626retry_find:
1627		page = find_get_page(mapping, offset);
1628		if (!page)
1629			goto no_cached_page;
 
 
 
 
 
1630	}
1631
1632	if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1633		page_cache_release(page);
1634		return ret | VM_FAULT_RETRY;
1635	}
1636
1637	/* Did it get truncated? */
1638	if (unlikely(page->mapping != mapping)) {
1639		unlock_page(page);
1640		put_page(page);
1641		goto retry_find;
1642	}
1643	VM_BUG_ON(page->index != offset);
1644
1645	/*
1646	 * We have a locked page in the page cache, now we need to check
1647	 * that it's up-to-date. If not, it is going to be due to an error.
1648	 */
1649	if (unlikely(!PageUptodate(page)))
1650		goto page_not_uptodate;
1651
1652	/*
 
 
 
 
 
 
 
 
 
 
1653	 * Found the page and have a reference on it.
1654	 * We must recheck i_size under page lock.
1655	 */
1656	size = (i_size_read(inode) + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1657	if (unlikely(offset >= size)) {
1658		unlock_page(page);
1659		page_cache_release(page);
1660		return VM_FAULT_SIGBUS;
1661	}
1662
1663	vmf->page = page;
1664	return ret | VM_FAULT_LOCKED;
1665
1666no_cached_page:
1667	/*
1668	 * We're only likely to ever get here if MADV_RANDOM is in
1669	 * effect.
 
 
1670	 */
1671	error = page_cache_read(file, offset);
 
 
 
 
1672
1673	/*
1674	 * The page we want has now been added to the page cache.
1675	 * In the unlikely event that someone removed it in the
1676	 * meantime, we'll just come back here and read it again.
1677	 */
1678	if (error >= 0)
1679		goto retry_find;
1680
1681	/*
1682	 * An error return from page_cache_read can result if the
1683	 * system is low on memory, or a problem occurs while trying
1684	 * to schedule I/O.
1685	 */
1686	if (error == -ENOMEM)
1687		return VM_FAULT_OOM;
1688	return VM_FAULT_SIGBUS;
1689
1690page_not_uptodate:
1691	/*
1692	 * Umm, take care of errors if the page isn't up-to-date.
1693	 * Try to re-read it _once_. We do this synchronously,
1694	 * because there really aren't any performance issues here
1695	 * and we need to check for errors.
1696	 */
1697	ClearPageError(page);
1698	error = mapping->a_ops->readpage(file, page);
1699	if (!error) {
1700		wait_on_page_locked(page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1701		if (!PageUptodate(page))
1702			error = -EIO;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1703	}
1704	page_cache_release(page);
1705
1706	if (!error || error == AOP_TRUNCATED_PAGE)
1707		goto retry_find;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1708
1709	/* Things didn't work out. Return zero to tell the mm layer so. */
1710	shrink_readahead_size_eio(file, ra);
1711	return VM_FAULT_SIGBUS;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1712}
1713EXPORT_SYMBOL(filemap_fault);
1714
1715const struct vm_operations_struct generic_file_vm_ops = {
1716	.fault		= filemap_fault,
 
 
1717};
1718
1719/* This is used for a general mmap of a disk file */
1720
1721int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
1722{
1723	struct address_space *mapping = file->f_mapping;
1724
1725	if (!mapping->a_ops->readpage)
1726		return -ENOEXEC;
1727	file_accessed(file);
1728	vma->vm_ops = &generic_file_vm_ops;
1729	vma->vm_flags |= VM_CAN_NONLINEAR;
1730	return 0;
1731}
1732
1733/*
1734 * This is for filesystems which do not implement ->writepage.
1735 */
1736int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
1737{
1738	if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
1739		return -EINVAL;
1740	return generic_file_mmap(file, vma);
1741}
1742#else
1743int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
 
 
 
 
1744{
1745	return -ENOSYS;
1746}
1747int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
1748{
1749	return -ENOSYS;
1750}
1751#endif /* CONFIG_MMU */
1752
 
1753EXPORT_SYMBOL(generic_file_mmap);
1754EXPORT_SYMBOL(generic_file_readonly_mmap);
1755
1756static struct page *__read_cache_page(struct address_space *mapping,
 
 
 
 
 
 
 
 
 
 
 
 
1757				pgoff_t index,
1758				int (*filler)(void *, struct page *),
1759				void *data,
1760				gfp_t gfp)
1761{
1762	struct page *page;
1763	int err;
1764repeat:
1765	page = find_get_page(mapping, index);
1766	if (!page) {
1767		page = __page_cache_alloc(gfp | __GFP_COLD);
1768		if (!page)
1769			return ERR_PTR(-ENOMEM);
1770		err = add_to_page_cache_lru(page, mapping, index, gfp);
1771		if (unlikely(err)) {
1772			page_cache_release(page);
1773			if (err == -EEXIST)
1774				goto repeat;
1775			/* Presumably ENOMEM for radix tree node */
1776			return ERR_PTR(err);
1777		}
1778		err = filler(data, page);
 
 
 
 
 
 
1779		if (err < 0) {
1780			page_cache_release(page);
1781			page = ERR_PTR(err);
1782		}
 
 
 
 
 
1783	}
1784	return page;
1785}
1786
1787static struct page *do_read_cache_page(struct address_space *mapping,
1788				pgoff_t index,
1789				int (*filler)(void *, struct page *),
1790				void *data,
1791				gfp_t gfp)
1792
1793{
1794	struct page *page;
1795	int err;
1796
1797retry:
1798	page = __read_cache_page(mapping, index, filler, data, gfp);
1799	if (IS_ERR(page))
1800		return page;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1801	if (PageUptodate(page))
1802		goto out;
1803
 
1804	lock_page(page);
 
 
1805	if (!page->mapping) {
1806		unlock_page(page);
1807		page_cache_release(page);
1808		goto retry;
1809	}
 
 
1810	if (PageUptodate(page)) {
1811		unlock_page(page);
1812		goto out;
1813	}
1814	err = filler(data, page);
1815	if (err < 0) {
1816		page_cache_release(page);
1817		return ERR_PTR(err);
1818	}
 
 
 
 
 
1819out:
1820	mark_page_accessed(page);
1821	return page;
1822}
1823
1824/**
1825 * read_cache_page_async - read into page cache, fill it if needed
1826 * @mapping:	the page's address_space
1827 * @index:	the page index
1828 * @filler:	function to perform the read
1829 * @data:	first arg to filler(data, page) function, often left as NULL
1830 *
1831 * Same as read_cache_page, but don't wait for page to become unlocked
1832 * after submitting it to the filler.
1833 *
1834 * Read into the page cache. If a page already exists, and PageUptodate() is
1835 * not set, try to fill the page but don't wait for it to become unlocked.
1836 *
1837 * If the page does not get brought uptodate, return -EIO.
 
 
1838 */
1839struct page *read_cache_page_async(struct address_space *mapping,
1840				pgoff_t index,
1841				int (*filler)(void *, struct page *),
1842				void *data)
1843{
1844	return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
1845}
1846EXPORT_SYMBOL(read_cache_page_async);
1847
1848static struct page *wait_on_page_read(struct page *page)
1849{
1850	if (!IS_ERR(page)) {
1851		wait_on_page_locked(page);
1852		if (!PageUptodate(page)) {
1853			page_cache_release(page);
1854			page = ERR_PTR(-EIO);
1855		}
1856	}
1857	return page;
1858}
 
1859
1860/**
1861 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
1862 * @mapping:	the page's address_space
1863 * @index:	the page index
1864 * @gfp:	the page allocator flags to use if allocating
1865 *
1866 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
1867 * any new page allocations done using the specified allocation flags.
1868 *
1869 * If the page does not get brought uptodate, return -EIO.
 
 
1870 */
1871struct page *read_cache_page_gfp(struct address_space *mapping,
1872				pgoff_t index,
1873				gfp_t gfp)
1874{
1875	filler_t *filler = (filler_t *)mapping->a_ops->readpage;
1876
1877	return wait_on_page_read(do_read_cache_page(mapping, index, filler, NULL, gfp));
1878}
1879EXPORT_SYMBOL(read_cache_page_gfp);
1880
1881/**
1882 * read_cache_page - read into page cache, fill it if needed
1883 * @mapping:	the page's address_space
1884 * @index:	the page index
1885 * @filler:	function to perform the read
1886 * @data:	first arg to filler(data, page) function, often left as NULL
1887 *
1888 * Read into the page cache. If a page already exists, and PageUptodate() is
1889 * not set, try to fill the page then wait for it to become unlocked.
1890 *
1891 * If the page does not get brought uptodate, return -EIO.
1892 */
1893struct page *read_cache_page(struct address_space *mapping,
1894				pgoff_t index,
1895				int (*filler)(void *, struct page *),
1896				void *data)
1897{
1898	return wait_on_page_read(read_cache_page_async(mapping, index, filler, data));
1899}
1900EXPORT_SYMBOL(read_cache_page);
1901
1902static size_t __iovec_copy_from_user_inatomic(char *vaddr,
1903			const struct iovec *iov, size_t base, size_t bytes)
1904{
1905	size_t copied = 0, left = 0;
1906
1907	while (bytes) {
1908		char __user *buf = iov->iov_base + base;
1909		int copy = min(bytes, iov->iov_len - base);
1910
1911		base = 0;
1912		left = __copy_from_user_inatomic(vaddr, buf, copy);
1913		copied += copy;
1914		bytes -= copy;
1915		vaddr += copy;
1916		iov++;
1917
1918		if (unlikely(left))
1919			break;
1920	}
1921	return copied - left;
1922}
1923
1924/*
1925 * Copy as much as we can into the page and return the number of bytes which
1926 * were successfully copied.  If a fault is encountered then return the number of
1927 * bytes which were copied.
1928 */
1929size_t iov_iter_copy_from_user_atomic(struct page *page,
1930		struct iov_iter *i, unsigned long offset, size_t bytes)
1931{
1932	char *kaddr;
1933	size_t copied;
1934
1935	BUG_ON(!in_atomic());
1936	kaddr = kmap_atomic(page);
1937	if (likely(i->nr_segs == 1)) {
1938		int left;
1939		char __user *buf = i->iov->iov_base + i->iov_offset;
1940		left = __copy_from_user_inatomic(kaddr + offset, buf, bytes);
1941		copied = bytes - left;
1942	} else {
1943		copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1944						i->iov, i->iov_offset, bytes);
1945	}
1946	kunmap_atomic(kaddr);
1947
1948	return copied;
1949}
1950EXPORT_SYMBOL(iov_iter_copy_from_user_atomic);
1951
1952/*
1953 * This has the same sideeffects and return value as
1954 * iov_iter_copy_from_user_atomic().
1955 * The difference is that it attempts to resolve faults.
1956 * Page must not be locked.
1957 */
1958size_t iov_iter_copy_from_user(struct page *page,
1959		struct iov_iter *i, unsigned long offset, size_t bytes)
1960{
1961	char *kaddr;
1962	size_t copied;
1963
1964	kaddr = kmap(page);
1965	if (likely(i->nr_segs == 1)) {
1966		int left;
1967		char __user *buf = i->iov->iov_base + i->iov_offset;
1968		left = __copy_from_user(kaddr + offset, buf, bytes);
1969		copied = bytes - left;
1970	} else {
1971		copied = __iovec_copy_from_user_inatomic(kaddr + offset,
1972						i->iov, i->iov_offset, bytes);
1973	}
1974	kunmap(page);
1975	return copied;
1976}
1977EXPORT_SYMBOL(iov_iter_copy_from_user);
1978
1979void iov_iter_advance(struct iov_iter *i, size_t bytes)
1980{
1981	BUG_ON(i->count < bytes);
1982
1983	if (likely(i->nr_segs == 1)) {
1984		i->iov_offset += bytes;
1985		i->count -= bytes;
1986	} else {
1987		const struct iovec *iov = i->iov;
1988		size_t base = i->iov_offset;
1989		unsigned long nr_segs = i->nr_segs;
1990
1991		/*
1992		 * The !iov->iov_len check ensures we skip over unlikely
1993		 * zero-length segments (without overruning the iovec).
1994		 */
1995		while (bytes || unlikely(i->count && !iov->iov_len)) {
1996			int copy;
1997
1998			copy = min(bytes, iov->iov_len - base);
1999			BUG_ON(!i->count || i->count < copy);
2000			i->count -= copy;
2001			bytes -= copy;
2002			base += copy;
2003			if (iov->iov_len == base) {
2004				iov++;
2005				nr_segs--;
2006				base = 0;
2007			}
2008		}
2009		i->iov = iov;
2010		i->iov_offset = base;
2011		i->nr_segs = nr_segs;
2012	}
2013}
2014EXPORT_SYMBOL(iov_iter_advance);
2015
2016/*
2017 * Fault in the first iovec of the given iov_iter, to a maximum length
2018 * of bytes. Returns 0 on success, or non-zero if the memory could not be
2019 * accessed (ie. because it is an invalid address).
2020 *
2021 * writev-intensive code may want this to prefault several iovecs -- that
2022 * would be possible (callers must not rely on the fact that _only_ the
2023 * first iovec will be faulted with the current implementation).
2024 */
2025int iov_iter_fault_in_readable(struct iov_iter *i, size_t bytes)
2026{
2027	char __user *buf = i->iov->iov_base + i->iov_offset;
2028	bytes = min(bytes, i->iov->iov_len - i->iov_offset);
2029	return fault_in_pages_readable(buf, bytes);
2030}
2031EXPORT_SYMBOL(iov_iter_fault_in_readable);
2032
2033/*
2034 * Return the count of just the current iov_iter segment.
2035 */
2036size_t iov_iter_single_seg_count(struct iov_iter *i)
2037{
2038	const struct iovec *iov = i->iov;
2039	if (i->nr_segs == 1)
2040		return i->count;
2041	else
2042		return min(i->count, iov->iov_len - i->iov_offset);
2043}
2044EXPORT_SYMBOL(iov_iter_single_seg_count);
2045
2046/*
2047 * Performs necessary checks before doing a write
2048 *
2049 * Can adjust writing position or amount of bytes to write.
2050 * Returns appropriate error code that caller should return or
2051 * zero in case that write should be allowed.
2052 */
2053inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2054{
2055	struct inode *inode = file->f_mapping->host;
2056	unsigned long limit = rlimit(RLIMIT_FSIZE);
2057
2058        if (unlikely(*pos < 0))
2059                return -EINVAL;
2060
2061	if (!isblk) {
2062		/* FIXME: this is for backwards compatibility with 2.4 */
2063		if (file->f_flags & O_APPEND)
2064                        *pos = i_size_read(inode);
2065
2066		if (limit != RLIM_INFINITY) {
2067			if (*pos >= limit) {
2068				send_sig(SIGXFSZ, current, 0);
2069				return -EFBIG;
2070			}
2071			if (*count > limit - (typeof(limit))*pos) {
2072				*count = limit - (typeof(limit))*pos;
2073			}
2074		}
2075	}
2076
2077	/*
2078	 * LFS rule
2079	 */
2080	if (unlikely(*pos + *count > MAX_NON_LFS &&
2081				!(file->f_flags & O_LARGEFILE))) {
2082		if (*pos >= MAX_NON_LFS) {
2083			return -EFBIG;
2084		}
2085		if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2086			*count = MAX_NON_LFS - (unsigned long)*pos;
2087		}
2088	}
2089
2090	/*
2091	 * Are we about to exceed the fs block limit ?
2092	 *
2093	 * If we have written data it becomes a short write.  If we have
2094	 * exceeded without writing data we send a signal and return EFBIG.
2095	 * Linus frestrict idea will clean these up nicely..
2096	 */
2097	if (likely(!isblk)) {
2098		if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2099			if (*count || *pos > inode->i_sb->s_maxbytes) {
2100				return -EFBIG;
2101			}
2102			/* zero-length writes at ->s_maxbytes are OK */
2103		}
2104
2105		if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2106			*count = inode->i_sb->s_maxbytes - *pos;
2107	} else {
2108#ifdef CONFIG_BLOCK
2109		loff_t isize;
2110		if (bdev_read_only(I_BDEV(inode)))
2111			return -EPERM;
2112		isize = i_size_read(inode);
2113		if (*pos >= isize) {
2114			if (*count || *pos > isize)
2115				return -ENOSPC;
2116		}
2117
2118		if (*pos + *count > isize)
2119			*count = isize - *pos;
2120#else
2121		return -EPERM;
2122#endif
2123	}
2124	return 0;
2125}
2126EXPORT_SYMBOL(generic_write_checks);
2127
2128int pagecache_write_begin(struct file *file, struct address_space *mapping,
2129				loff_t pos, unsigned len, unsigned flags,
2130				struct page **pagep, void **fsdata)
2131{
2132	const struct address_space_operations *aops = mapping->a_ops;
2133
2134	return aops->write_begin(file, mapping, pos, len, flags,
2135							pagep, fsdata);
2136}
2137EXPORT_SYMBOL(pagecache_write_begin);
2138
2139int pagecache_write_end(struct file *file, struct address_space *mapping,
2140				loff_t pos, unsigned len, unsigned copied,
2141				struct page *page, void *fsdata)
2142{
2143	const struct address_space_operations *aops = mapping->a_ops;
2144
2145	mark_page_accessed(page);
2146	return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2147}
2148EXPORT_SYMBOL(pagecache_write_end);
2149
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2150ssize_t
2151generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2152		unsigned long *nr_segs, loff_t pos, loff_t *ppos,
2153		size_t count, size_t ocount)
2154{
2155	struct file	*file = iocb->ki_filp;
2156	struct address_space *mapping = file->f_mapping;
2157	struct inode	*inode = mapping->host;
 
2158	ssize_t		written;
2159	size_t		write_len;
2160	pgoff_t		end;
2161
2162	if (count != ocount)
2163		*nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2164
2165	write_len = iov_length(iov, *nr_segs);
2166	end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2167
2168	written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2169	if (written)
2170		goto out;
 
 
 
 
 
2171
2172	/*
2173	 * After a write we want buffered reads to be sure to go to disk to get
2174	 * the new data.  We invalidate clean cached page from the region we're
2175	 * about to write.  We do this *before* the write so that we can return
2176	 * without clobbering -EIOCBQUEUED from ->direct_IO().
2177	 */
2178	if (mapping->nrpages) {
2179		written = invalidate_inode_pages2_range(mapping,
2180					pos >> PAGE_CACHE_SHIFT, end);
2181		/*
2182		 * If a page can not be invalidated, return 0 to fall back
2183		 * to buffered write.
2184		 */
2185		if (written) {
2186			if (written == -EBUSY)
2187				return 0;
2188			goto out;
2189		}
2190	}
2191
2192	written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2193
2194	/*
2195	 * Finally, try again to invalidate clean pages which might have been
2196	 * cached by non-direct readahead, or faulted in by get_user_pages()
2197	 * if the source of the write was an mmap'ed region of the file
2198	 * we're writing.  Either one is a pretty crazy thing to do,
2199	 * so we don't support it 100%.  If this invalidation
2200	 * fails, tough, the write still worked...
 
 
 
 
 
 
 
 
 
2201	 */
2202	if (mapping->nrpages) {
2203		invalidate_inode_pages2_range(mapping,
2204					      pos >> PAGE_CACHE_SHIFT, end);
2205	}
2206
2207	if (written > 0) {
2208		pos += written;
 
2209		if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2210			i_size_write(inode, pos);
2211			mark_inode_dirty(inode);
2212		}
2213		*ppos = pos;
2214	}
 
 
2215out:
2216	return written;
2217}
2218EXPORT_SYMBOL(generic_file_direct_write);
2219
2220/*
2221 * Find or create a page at the given pagecache position. Return the locked
2222 * page. This function is specifically for buffered writes.
2223 */
2224struct page *grab_cache_page_write_begin(struct address_space *mapping,
2225					pgoff_t index, unsigned flags)
2226{
2227	int status;
2228	gfp_t gfp_mask;
2229	struct page *page;
2230	gfp_t gfp_notmask = 0;
2231
2232	gfp_mask = mapping_gfp_mask(mapping);
2233	if (mapping_cap_account_dirty(mapping))
2234		gfp_mask |= __GFP_WRITE;
2235	if (flags & AOP_FLAG_NOFS)
2236		gfp_notmask = __GFP_FS;
2237repeat:
2238	page = find_lock_page(mapping, index);
 
2239	if (page)
2240		goto found;
2241
2242	page = __page_cache_alloc(gfp_mask & ~gfp_notmask);
2243	if (!page)
2244		return NULL;
2245	status = add_to_page_cache_lru(page, mapping, index,
2246						GFP_KERNEL & ~gfp_notmask);
2247	if (unlikely(status)) {
2248		page_cache_release(page);
2249		if (status == -EEXIST)
2250			goto repeat;
2251		return NULL;
2252	}
2253found:
2254	wait_on_page_writeback(page);
2255	return page;
2256}
2257EXPORT_SYMBOL(grab_cache_page_write_begin);
2258
2259static ssize_t generic_perform_write(struct file *file,
2260				struct iov_iter *i, loff_t pos)
2261{
2262	struct address_space *mapping = file->f_mapping;
2263	const struct address_space_operations *a_ops = mapping->a_ops;
2264	long status = 0;
2265	ssize_t written = 0;
2266	unsigned int flags = 0;
2267
2268	/*
2269	 * Copies from kernel address space cannot fail (NFSD is a big user).
2270	 */
2271	if (segment_eq(get_fs(), KERNEL_DS))
2272		flags |= AOP_FLAG_UNINTERRUPTIBLE;
2273
2274	do {
2275		struct page *page;
2276		unsigned long offset;	/* Offset into pagecache page */
2277		unsigned long bytes;	/* Bytes to write to page */
2278		size_t copied;		/* Bytes copied from user */
2279		void *fsdata;
2280
2281		offset = (pos & (PAGE_CACHE_SIZE - 1));
2282		bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2283						iov_iter_count(i));
2284
2285again:
2286		/*
2287		 * Bring in the user page that we will copy from _first_.
2288		 * Otherwise there's a nasty deadlock on copying from the
2289		 * same page as we're writing to, without it being marked
2290		 * up-to-date.
2291		 *
2292		 * Not only is this an optimisation, but it is also required
2293		 * to check that the address is actually valid, when atomic
2294		 * usercopies are used, below.
2295		 */
2296		if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2297			status = -EFAULT;
2298			break;
2299		}
2300
 
 
 
 
 
2301		status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2302						&page, &fsdata);
2303		if (unlikely(status))
2304			break;
2305
2306		if (mapping_writably_mapped(mapping))
2307			flush_dcache_page(page);
2308
2309		pagefault_disable();
2310		copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2311		pagefault_enable();
2312		flush_dcache_page(page);
2313
2314		mark_page_accessed(page);
2315		status = a_ops->write_end(file, mapping, pos, bytes, copied,
2316						page, fsdata);
2317		if (unlikely(status < 0))
2318			break;
2319		copied = status;
2320
 
2321		cond_resched();
2322
2323		iov_iter_advance(i, copied);
2324		if (unlikely(copied == 0)) {
2325			/*
2326			 * If we were unable to copy any data at all, we must
2327			 * fall back to a single segment length write.
2328			 *
2329			 * If we didn't fallback here, we could livelock
2330			 * because not all segments in the iov can be copied at
2331			 * once without a pagefault.
2332			 */
2333			bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2334						iov_iter_single_seg_count(i));
2335			goto again;
2336		}
2337		pos += copied;
2338		written += copied;
2339
2340		balance_dirty_pages_ratelimited(mapping);
2341		if (fatal_signal_pending(current)) {
2342			status = -EINTR;
2343			break;
2344		}
2345	} while (iov_iter_count(i));
2346
2347	return written ? written : status;
2348}
2349
2350ssize_t
2351generic_file_buffered_write(struct kiocb *iocb, const struct iovec *iov,
2352		unsigned long nr_segs, loff_t pos, loff_t *ppos,
2353		size_t count, ssize_t written)
2354{
2355	struct file *file = iocb->ki_filp;
2356	ssize_t status;
2357	struct iov_iter i;
2358
2359	iov_iter_init(&i, iov, nr_segs, count, written);
2360	status = generic_perform_write(file, &i, pos);
2361
2362	if (likely(status >= 0)) {
2363		written += status;
2364		*ppos = pos + status;
2365  	}
2366	
2367	return written ? written : status;
2368}
2369EXPORT_SYMBOL(generic_file_buffered_write);
2370
2371/**
2372 * __generic_file_aio_write - write data to a file
2373 * @iocb:	IO state structure (file, offset, etc.)
2374 * @iov:	vector with data to write
2375 * @nr_segs:	number of segments in the vector
2376 * @ppos:	position where to write
2377 *
2378 * This function does all the work needed for actually writing data to a
2379 * file. It does all basic checks, removes SUID from the file, updates
2380 * modification times and calls proper subroutines depending on whether we
2381 * do direct IO or a standard buffered write.
2382 *
2383 * It expects i_mutex to be grabbed unless we work on a block device or similar
2384 * object which does not need locking at all.
2385 *
2386 * This function does *not* take care of syncing data in case of O_SYNC write.
2387 * A caller has to handle it. This is mainly due to the fact that we want to
2388 * avoid syncing under i_mutex.
 
 
 
 
2389 */
2390ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2391				 unsigned long nr_segs, loff_t *ppos)
2392{
2393	struct file *file = iocb->ki_filp;
2394	struct address_space * mapping = file->f_mapping;
2395	size_t ocount;		/* original count */
2396	size_t count;		/* after file limit checks */
2397	struct inode 	*inode = mapping->host;
2398	loff_t		pos;
2399	ssize_t		written;
2400	ssize_t		err;
2401
2402	ocount = 0;
2403	err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2404	if (err)
2405		return err;
2406
2407	count = ocount;
2408	pos = *ppos;
2409
2410	vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
2411
2412	/* We can write back this queue in page reclaim */
2413	current->backing_dev_info = mapping->backing_dev_info;
2414	written = 0;
2415
2416	err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2417	if (err)
2418		goto out;
2419
2420	if (count == 0)
2421		goto out;
2422
2423	err = file_remove_suid(file);
2424	if (err)
2425		goto out;
2426
2427	err = file_update_time(file);
2428	if (err)
2429		goto out;
2430
2431	/* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2432	if (unlikely(file->f_flags & O_DIRECT)) {
2433		loff_t endbyte;
2434		ssize_t written_buffered;
2435
2436		written = generic_file_direct_write(iocb, iov, &nr_segs, pos,
2437							ppos, count, ocount);
2438		if (written < 0 || written == count)
2439			goto out;
2440		/*
2441		 * direct-io write to a hole: fall through to buffered I/O
2442		 * for completing the rest of the request.
 
 
 
2443		 */
2444		pos += written;
2445		count -= written;
2446		written_buffered = generic_file_buffered_write(iocb, iov,
2447						nr_segs, pos, ppos, count,
2448						written);
2449		/*
2450		 * If generic_file_buffered_write() retuned a synchronous error
2451		 * then we want to return the number of bytes which were
2452		 * direct-written, or the error code if that was zero.  Note
2453		 * that this differs from normal direct-io semantics, which
2454		 * will return -EFOO even if some bytes were written.
2455		 */
2456		if (written_buffered < 0) {
2457			err = written_buffered;
2458			goto out;
2459		}
2460
2461		/*
2462		 * We need to ensure that the page cache pages are written to
2463		 * disk and invalidated to preserve the expected O_DIRECT
2464		 * semantics.
2465		 */
2466		endbyte = pos + written_buffered - written - 1;
2467		err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
2468		if (err == 0) {
2469			written = written_buffered;
 
2470			invalidate_mapping_pages(mapping,
2471						 pos >> PAGE_CACHE_SHIFT,
2472						 endbyte >> PAGE_CACHE_SHIFT);
2473		} else {
2474			/*
2475			 * We don't know how much we wrote, so just return
2476			 * the number of bytes which were direct-written
2477			 */
2478		}
2479	} else {
2480		written = generic_file_buffered_write(iocb, iov, nr_segs,
2481				pos, ppos, count, written);
 
2482	}
2483out:
2484	current->backing_dev_info = NULL;
2485	return written ? written : err;
2486}
2487EXPORT_SYMBOL(__generic_file_aio_write);
2488
2489/**
2490 * generic_file_aio_write - write data to a file
2491 * @iocb:	IO state structure
2492 * @iov:	vector with data to write
2493 * @nr_segs:	number of segments in the vector
2494 * @pos:	position in file where to write
2495 *
2496 * This is a wrapper around __generic_file_aio_write() to be used by most
2497 * filesystems. It takes care of syncing the file in case of O_SYNC file
2498 * and acquires i_mutex as needed.
 
 
 
 
2499 */
2500ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2501		unsigned long nr_segs, loff_t pos)
2502{
2503	struct file *file = iocb->ki_filp;
2504	struct inode *inode = file->f_mapping->host;
2505	struct blk_plug plug;
2506	ssize_t ret;
2507
2508	BUG_ON(iocb->ki_pos != pos);
2509
2510	mutex_lock(&inode->i_mutex);
2511	blk_start_plug(&plug);
2512	ret = __generic_file_aio_write(iocb, iov, nr_segs, &iocb->ki_pos);
2513	mutex_unlock(&inode->i_mutex);
2514
2515	if (ret > 0 || ret == -EIOCBQUEUED) {
2516		ssize_t err;
2517
2518		err = generic_write_sync(file, pos, ret);
2519		if (err < 0 && ret > 0)
2520			ret = err;
2521	}
2522	blk_finish_plug(&plug);
2523	return ret;
2524}
2525EXPORT_SYMBOL(generic_file_aio_write);
2526
2527/**
2528 * try_to_release_page() - release old fs-specific metadata on a page
2529 *
2530 * @page: the page which the kernel is trying to free
2531 * @gfp_mask: memory allocation flags (and I/O mode)
2532 *
2533 * The address_space is to try to release any data against the page
2534 * (presumably at page->private).  If the release was successful, return `1'.
2535 * Otherwise return zero.
2536 *
2537 * This may also be called if PG_fscache is set on a page, indicating that the
2538 * page is known to the local caching routines.
2539 *
2540 * The @gfp_mask argument specifies whether I/O may be performed to release
2541 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2542 *
 
2543 */
2544int try_to_release_page(struct page *page, gfp_t gfp_mask)
2545{
2546	struct address_space * const mapping = page->mapping;
2547
2548	BUG_ON(!PageLocked(page));
2549	if (PageWriteback(page))
2550		return 0;
2551
2552	if (mapping && mapping->a_ops->releasepage)
2553		return mapping->a_ops->releasepage(page, gfp_mask);
2554	return try_to_free_buffers(page);
2555}
2556
2557EXPORT_SYMBOL(try_to_release_page);