Linux Audio

Check our new training course

Loading...
v5.9
   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 "internal.h"
  46
  47#define CREATE_TRACE_POINTS
  48#include <trace/events/filemap.h>
  49
  50/*
  51 * FIXME: remove all knowledge of the buffer layer from the core VM
  52 */
  53#include <linux/buffer_head.h> /* for try_to_free_buffers */
  54
  55#include <asm/mman.h>
  56
  57/*
  58 * Shared mappings implemented 30.11.1994. It's not fully working yet,
  59 * though.
  60 *
  61 * Shared mappings now work. 15.8.1995  Bruno.
  62 *
  63 * finished 'unifying' the page and buffer cache and SMP-threaded the
  64 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
  65 *
  66 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
  67 */
  68
  69/*
  70 * Lock ordering:
  71 *
  72 *  ->i_mmap_rwsem		(truncate_pagecache)
  73 *    ->private_lock		(__free_pte->__set_page_dirty_buffers)
  74 *      ->swap_lock		(exclusive_swap_page, others)
  75 *        ->i_pages lock
  76 *
  77 *  ->i_mutex
  78 *    ->i_mmap_rwsem		(truncate->unmap_mapping_range)
  79 *
  80 *  ->mmap_lock
  81 *    ->i_mmap_rwsem
  82 *      ->page_table_lock or pte_lock	(various, mainly in memory.c)
  83 *        ->i_pages lock	(arch-dependent flush_dcache_mmap_lock)
  84 *
  85 *  ->mmap_lock
  86 *    ->lock_page		(access_process_vm)
  87 *
  88 *  ->i_mutex			(generic_perform_write)
  89 *    ->mmap_lock		(fault_in_pages_readable->do_page_fault)
  90 *
  91 *  bdi->wb.list_lock
  92 *    sb_lock			(fs/fs-writeback.c)
  93 *    ->i_pages lock		(__sync_single_inode)
  94 *
  95 *  ->i_mmap_rwsem
  96 *    ->anon_vma.lock		(vma_adjust)
  97 *
  98 *  ->anon_vma.lock
  99 *    ->page_table_lock or pte_lock	(anon_vma_prepare and various)
 100 *
 101 *  ->page_table_lock or pte_lock
 102 *    ->swap_lock		(try_to_unmap_one)
 103 *    ->private_lock		(try_to_unmap_one)
 104 *    ->i_pages lock		(try_to_unmap_one)
 105 *    ->pgdat->lru_lock		(follow_page->mark_page_accessed)
 106 *    ->pgdat->lru_lock		(check_pte_range->isolate_lru_page)
 107 *    ->private_lock		(page_remove_rmap->set_page_dirty)
 108 *    ->i_pages lock		(page_remove_rmap->set_page_dirty)
 109 *    bdi.wb->list_lock		(page_remove_rmap->set_page_dirty)
 110 *    ->inode->i_lock		(page_remove_rmap->set_page_dirty)
 111 *    ->memcg->move_lock	(page_remove_rmap->lock_page_memcg)
 112 *    bdi.wb->list_lock		(zap_pte_range->set_page_dirty)
 113 *    ->inode->i_lock		(zap_pte_range->set_page_dirty)
 114 *    ->private_lock		(zap_pte_range->__set_page_dirty_buffers)
 115 *
 116 * ->i_mmap_rwsem
 117 *   ->tasklist_lock            (memory_failure, collect_procs_ao)
 118 */
 119
 120static void page_cache_delete(struct address_space *mapping,
 121				   struct page *page, void *shadow)
 122{
 123	XA_STATE(xas, &mapping->i_pages, page->index);
 124	unsigned int nr = 1;
 
 
 
 125
 126	mapping_set_update(&xas, mapping);
 127
 128	/* hugetlb pages are represented by a single entry in the xarray */
 129	if (!PageHuge(page)) {
 130		xas_set_order(&xas, page->index, compound_order(page));
 131		nr = compound_nr(page);
 132	}
 133
 134	VM_BUG_ON_PAGE(!PageLocked(page), page);
 135	VM_BUG_ON_PAGE(PageTail(page), page);
 136	VM_BUG_ON_PAGE(nr != 1 && shadow, page);
 137
 138	xas_store(&xas, shadow);
 139	xas_init_marks(&xas);
 140
 141	page->mapping = NULL;
 142	/* Leave page->index set: truncation lookup relies upon it */
 143
 144	if (shadow) {
 145		mapping->nrexceptional += nr;
 146		/*
 147		 * Make sure the nrexceptional update is committed before
 148		 * the nrpages update so that final truncate racing
 149		 * with reclaim does not see both counters 0 at the
 150		 * same time and miss a shadow entry.
 151		 */
 152		smp_wmb();
 153	}
 154	mapping->nrpages -= nr;
 155}
 156
 157static void unaccount_page_cache_page(struct address_space *mapping,
 158				      struct page *page)
 159{
 160	int nr;
 161
 162	/*
 163	 * if we're uptodate, flush out into the cleancache, otherwise
 164	 * invalidate any existing cleancache entries.  We can't leave
 165	 * stale data around in the cleancache once our page is gone
 166	 */
 167	if (PageUptodate(page) && PageMappedToDisk(page))
 168		cleancache_put_page(page);
 169	else
 170		cleancache_invalidate_page(mapping, page);
 171
 172	VM_BUG_ON_PAGE(PageTail(page), page);
 173	VM_BUG_ON_PAGE(page_mapped(page), page);
 174	if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
 175		int mapcount;
 176
 177		pr_alert("BUG: Bad page cache in process %s  pfn:%05lx\n",
 178			 current->comm, page_to_pfn(page));
 179		dump_page(page, "still mapped when deleted");
 180		dump_stack();
 181		add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
 182
 183		mapcount = page_mapcount(page);
 184		if (mapping_exiting(mapping) &&
 185		    page_count(page) >= mapcount + 2) {
 186			/*
 187			 * All vmas have already been torn down, so it's
 188			 * a good bet that actually the page is unmapped,
 189			 * and we'd prefer not to leak it: if we're wrong,
 190			 * some other bad page check should catch it later.
 191			 */
 192			page_mapcount_reset(page);
 193			page_ref_sub(page, mapcount);
 194		}
 195	}
 196
 197	/* hugetlb pages do not participate in page cache accounting. */
 198	if (PageHuge(page))
 199		return;
 
 200
 201	nr = thp_nr_pages(page);
 202
 203	__mod_lruvec_page_state(page, NR_FILE_PAGES, -nr);
 204	if (PageSwapBacked(page)) {
 205		__mod_lruvec_page_state(page, NR_SHMEM, -nr);
 206		if (PageTransHuge(page))
 207			__dec_node_page_state(page, NR_SHMEM_THPS);
 208	} else if (PageTransHuge(page)) {
 209		__dec_node_page_state(page, NR_FILE_THPS);
 210		filemap_nr_thps_dec(mapping);
 211	}
 212
 
 
 
 
 
 
 
 
 
 213	/*
 214	 * At this point page must be either written or cleaned by
 215	 * truncate.  Dirty page here signals a bug and loss of
 216	 * unwritten data.
 217	 *
 218	 * This fixes dirty accounting after removing the page entirely
 219	 * but leaves PageDirty set: it has no effect for truncated
 220	 * page and anyway will be cleared before returning page into
 221	 * buddy allocator.
 222	 */
 223	if (WARN_ON_ONCE(PageDirty(page)))
 224		account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
 
 
 225}
 226
 227/*
 228 * Delete a page from the page cache and free it. Caller has to make
 229 * sure the page is locked and that nobody else uses it - or that usage
 230 * is safe.  The caller must hold the i_pages lock.
 231 */
 232void __delete_from_page_cache(struct page *page, void *shadow)
 233{
 234	struct address_space *mapping = page->mapping;
 235
 236	trace_mm_filemap_delete_from_page_cache(page);
 
 
 
 
 
 
 
 
 
 237
 238	unaccount_page_cache_page(mapping, page);
 239	page_cache_delete(mapping, page, shadow);
 240}
 241
 242static void page_cache_free_page(struct address_space *mapping,
 243				struct page *page)
 244{
 245	void (*freepage)(struct page *);
 246
 247	freepage = mapping->a_ops->freepage;
 248	if (freepage)
 249		freepage(page);
 
 250
 251	if (PageTransHuge(page) && !PageHuge(page)) {
 252		page_ref_sub(page, HPAGE_PMD_NR);
 253		VM_BUG_ON_PAGE(page_count(page) <= 0, page);
 254	} else {
 255		put_page(page);
 
 
 
 
 
 256	}
 257}
 258
 259/**
 260 * delete_from_page_cache - delete page from page cache
 261 * @page: the page which the kernel is trying to remove from page cache
 262 *
 263 * This must be called only on pages that have been verified to be in the page
 264 * cache and locked.  It will never put the page into the free list, the caller
 265 * has a reference on the page.
 266 */
 267void delete_from_page_cache(struct page *page)
 268{
 269	struct address_space *mapping = page_mapping(page);
 270	unsigned long flags;
 271
 272	BUG_ON(!PageLocked(page));
 273	xa_lock_irqsave(&mapping->i_pages, flags);
 
 
 274	__delete_from_page_cache(page, NULL);
 275	xa_unlock_irqrestore(&mapping->i_pages, flags);
 
 276
 277	page_cache_free_page(mapping, page);
 
 
 278}
 279EXPORT_SYMBOL(delete_from_page_cache);
 280
 281/*
 282 * page_cache_delete_batch - delete several pages from page cache
 283 * @mapping: the mapping to which pages belong
 284 * @pvec: pagevec with pages to delete
 285 *
 286 * The function walks over mapping->i_pages and removes pages passed in @pvec
 287 * from the mapping. The function expects @pvec to be sorted by page index
 288 * and is optimised for it to be dense.
 289 * It tolerates holes in @pvec (mapping entries at those indices are not
 290 * modified). The function expects only THP head pages to be present in the
 291 * @pvec.
 292 *
 293 * The function expects the i_pages lock to be held.
 294 */
 295static void page_cache_delete_batch(struct address_space *mapping,
 296			     struct pagevec *pvec)
 297{
 298	XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index);
 299	int total_pages = 0;
 300	int i = 0;
 301	struct page *page;
 302
 303	mapping_set_update(&xas, mapping);
 304	xas_for_each(&xas, page, ULONG_MAX) {
 305		if (i >= pagevec_count(pvec))
 306			break;
 307
 308		/* A swap/dax/shadow entry got inserted? Skip it. */
 309		if (xa_is_value(page))
 310			continue;
 311		/*
 312		 * A page got inserted in our range? Skip it. We have our
 313		 * pages locked so they are protected from being removed.
 314		 * If we see a page whose index is higher than ours, it
 315		 * means our page has been removed, which shouldn't be
 316		 * possible because we're holding the PageLock.
 317		 */
 318		if (page != pvec->pages[i]) {
 319			VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index,
 320					page);
 321			continue;
 322		}
 323
 324		WARN_ON_ONCE(!PageLocked(page));
 325
 326		if (page->index == xas.xa_index)
 327			page->mapping = NULL;
 328		/* Leave page->index set: truncation lookup relies on it */
 329
 330		/*
 331		 * Move to the next page in the vector if this is a regular
 332		 * page or the index is of the last sub-page of this compound
 333		 * page.
 334		 */
 335		if (page->index + compound_nr(page) - 1 == xas.xa_index)
 336			i++;
 337		xas_store(&xas, NULL);
 338		total_pages++;
 339	}
 340	mapping->nrpages -= total_pages;
 341}
 342
 343void delete_from_page_cache_batch(struct address_space *mapping,
 344				  struct pagevec *pvec)
 345{
 346	int i;
 347	unsigned long flags;
 348
 349	if (!pagevec_count(pvec))
 350		return;
 351
 352	xa_lock_irqsave(&mapping->i_pages, flags);
 353	for (i = 0; i < pagevec_count(pvec); i++) {
 354		trace_mm_filemap_delete_from_page_cache(pvec->pages[i]);
 355
 356		unaccount_page_cache_page(mapping, pvec->pages[i]);
 357	}
 358	page_cache_delete_batch(mapping, pvec);
 359	xa_unlock_irqrestore(&mapping->i_pages, flags);
 360
 361	for (i = 0; i < pagevec_count(pvec); i++)
 362		page_cache_free_page(mapping, pvec->pages[i]);
 363}
 364
 365int filemap_check_errors(struct address_space *mapping)
 366{
 367	int ret = 0;
 368	/* Check for outstanding write errors */
 369	if (test_bit(AS_ENOSPC, &mapping->flags) &&
 370	    test_and_clear_bit(AS_ENOSPC, &mapping->flags))
 371		ret = -ENOSPC;
 372	if (test_bit(AS_EIO, &mapping->flags) &&
 373	    test_and_clear_bit(AS_EIO, &mapping->flags))
 374		ret = -EIO;
 375	return ret;
 376}
 377EXPORT_SYMBOL(filemap_check_errors);
 378
 379static int filemap_check_and_keep_errors(struct address_space *mapping)
 380{
 381	/* Check for outstanding write errors */
 382	if (test_bit(AS_EIO, &mapping->flags))
 383		return -EIO;
 384	if (test_bit(AS_ENOSPC, &mapping->flags))
 385		return -ENOSPC;
 386	return 0;
 387}
 388
 389/**
 390 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
 391 * @mapping:	address space structure to write
 392 * @start:	offset in bytes where the range starts
 393 * @end:	offset in bytes where the range ends (inclusive)
 394 * @sync_mode:	enable synchronous operation
 395 *
 396 * Start writeback against all of a mapping's dirty pages that lie
 397 * within the byte offsets <start, end> inclusive.
 398 *
 399 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
 400 * opposed to a regular memory cleansing writeback.  The difference between
 401 * these two operations is that if a dirty page/buffer is encountered, it must
 402 * be waited upon, and not just skipped over.
 403 *
 404 * Return: %0 on success, negative error code otherwise.
 405 */
 406int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
 407				loff_t end, int sync_mode)
 408{
 409	int ret;
 410	struct writeback_control wbc = {
 411		.sync_mode = sync_mode,
 412		.nr_to_write = LONG_MAX,
 413		.range_start = start,
 414		.range_end = end,
 415	};
 416
 417	if (!mapping_cap_writeback_dirty(mapping) ||
 418	    !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
 419		return 0;
 420
 421	wbc_attach_fdatawrite_inode(&wbc, mapping->host);
 422	ret = do_writepages(mapping, &wbc);
 423	wbc_detach_inode(&wbc);
 424	return ret;
 425}
 426
 427static inline int __filemap_fdatawrite(struct address_space *mapping,
 428	int sync_mode)
 429{
 430	return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
 431}
 432
 433int filemap_fdatawrite(struct address_space *mapping)
 434{
 435	return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
 436}
 437EXPORT_SYMBOL(filemap_fdatawrite);
 438
 439int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
 440				loff_t end)
 441{
 442	return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
 443}
 444EXPORT_SYMBOL(filemap_fdatawrite_range);
 445
 446/**
 447 * filemap_flush - mostly a non-blocking flush
 448 * @mapping:	target address_space
 449 *
 450 * This is a mostly non-blocking flush.  Not suitable for data-integrity
 451 * purposes - I/O may not be started against all dirty pages.
 452 *
 453 * Return: %0 on success, negative error code otherwise.
 454 */
 455int filemap_flush(struct address_space *mapping)
 456{
 457	return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
 458}
 459EXPORT_SYMBOL(filemap_flush);
 460
 461/**
 462 * filemap_range_has_page - check if a page exists in range.
 463 * @mapping:           address space within which to check
 464 * @start_byte:        offset in bytes where the range starts
 465 * @end_byte:          offset in bytes where the range ends (inclusive)
 466 *
 467 * Find at least one page in the range supplied, usually used to check if
 468 * direct writing in this range will trigger a writeback.
 469 *
 470 * Return: %true if at least one page exists in the specified range,
 471 * %false otherwise.
 472 */
 473bool filemap_range_has_page(struct address_space *mapping,
 474			   loff_t start_byte, loff_t end_byte)
 475{
 476	struct page *page;
 477	XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
 478	pgoff_t max = end_byte >> PAGE_SHIFT;
 479
 480	if (end_byte < start_byte)
 481		return false;
 482
 483	rcu_read_lock();
 484	for (;;) {
 485		page = xas_find(&xas, max);
 486		if (xas_retry(&xas, page))
 487			continue;
 488		/* Shadow entries don't count */
 489		if (xa_is_value(page))
 490			continue;
 491		/*
 492		 * We don't need to try to pin this page; we're about to
 493		 * release the RCU lock anyway.  It is enough to know that
 494		 * there was a page here recently.
 495		 */
 496		break;
 497	}
 498	rcu_read_unlock();
 499
 500	return page != NULL;
 501}
 502EXPORT_SYMBOL(filemap_range_has_page);
 503
 504static void __filemap_fdatawait_range(struct address_space *mapping,
 505				     loff_t start_byte, loff_t end_byte)
 506{
 507	pgoff_t index = start_byte >> PAGE_SHIFT;
 508	pgoff_t end = end_byte >> PAGE_SHIFT;
 509	struct pagevec pvec;
 510	int nr_pages;
 
 511
 512	if (end_byte < start_byte)
 513		return;
 514
 515	pagevec_init(&pvec);
 516	while (index <= end) {
 
 
 
 517		unsigned i;
 518
 519		nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
 520				end, PAGECACHE_TAG_WRITEBACK);
 521		if (!nr_pages)
 522			break;
 523
 524		for (i = 0; i < nr_pages; i++) {
 525			struct page *page = pvec.pages[i];
 526
 
 
 
 
 527			wait_on_page_writeback(page);
 528			ClearPageError(page);
 
 529		}
 530		pagevec_release(&pvec);
 531		cond_resched();
 532	}
 533}
 
 
 
 534
 535/**
 536 * filemap_fdatawait_range - wait for writeback to complete
 537 * @mapping:		address space structure to wait for
 538 * @start_byte:		offset in bytes where the range starts
 539 * @end_byte:		offset in bytes where the range ends (inclusive)
 540 *
 541 * Walk the list of under-writeback pages of the given address space
 542 * in the given range and wait for all of them.  Check error status of
 543 * the address space and return it.
 544 *
 545 * Since the error status of the address space is cleared by this function,
 546 * callers are responsible for checking the return value and handling and/or
 547 * reporting the error.
 548 *
 549 * Return: error status of the address space.
 550 */
 551int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
 552			    loff_t end_byte)
 553{
 554	__filemap_fdatawait_range(mapping, start_byte, end_byte);
 555	return filemap_check_errors(mapping);
 556}
 557EXPORT_SYMBOL(filemap_fdatawait_range);
 558
 559/**
 560 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
 561 * @mapping:		address space structure to wait for
 562 * @start_byte:		offset in bytes where the range starts
 563 * @end_byte:		offset in bytes where the range ends (inclusive)
 564 *
 565 * Walk the list of under-writeback pages of the given address space in the
 566 * given range and wait for all of them.  Unlike filemap_fdatawait_range(),
 567 * this function does not clear error status of the address space.
 568 *
 569 * Use this function if callers don't handle errors themselves.  Expected
 570 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
 571 * fsfreeze(8)
 572 */
 573int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
 574		loff_t start_byte, loff_t end_byte)
 575{
 576	__filemap_fdatawait_range(mapping, start_byte, end_byte);
 577	return filemap_check_and_keep_errors(mapping);
 578}
 579EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
 580
 581/**
 582 * file_fdatawait_range - wait for writeback to complete
 583 * @file:		file pointing to address space structure to wait for
 584 * @start_byte:		offset in bytes where the range starts
 585 * @end_byte:		offset in bytes where the range ends (inclusive)
 586 *
 587 * Walk the list of under-writeback pages of the address space that file
 588 * refers to, in the given range and wait for all of them.  Check error
 589 * status of the address space vs. the file->f_wb_err cursor and return it.
 590 *
 591 * Since the error status of the file is advanced by this function,
 592 * callers are responsible for checking the return value and handling and/or
 593 * reporting the error.
 594 *
 595 * Return: error status of the address space vs. the file->f_wb_err cursor.
 596 */
 597int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
 598{
 599	struct address_space *mapping = file->f_mapping;
 600
 601	__filemap_fdatawait_range(mapping, start_byte, end_byte);
 602	return file_check_and_advance_wb_err(file);
 603}
 604EXPORT_SYMBOL(file_fdatawait_range);
 605
 606/**
 607 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
 608 * @mapping: address space structure to wait for
 609 *
 610 * Walk the list of under-writeback pages of the given address space
 611 * and wait for all of them.  Unlike filemap_fdatawait(), this function
 612 * does not clear error status of the address space.
 613 *
 614 * Use this function if callers don't handle errors themselves.  Expected
 615 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
 616 * fsfreeze(8)
 617 *
 618 * Return: error status of the address space.
 619 */
 620int filemap_fdatawait_keep_errors(struct address_space *mapping)
 621{
 622	__filemap_fdatawait_range(mapping, 0, LLONG_MAX);
 623	return filemap_check_and_keep_errors(mapping);
 624}
 625EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
 626
 627/* Returns true if writeback might be needed or already in progress. */
 628static bool mapping_needs_writeback(struct address_space *mapping)
 629{
 630	if (dax_mapping(mapping))
 631		return mapping->nrexceptional;
 632
 633	return mapping->nrpages;
 634}
 
 635
 636/**
 637 * filemap_write_and_wait_range - write out & wait on a file range
 638 * @mapping:	the address_space for the pages
 639 * @lstart:	offset in bytes where the range starts
 640 * @lend:	offset in bytes where the range ends (inclusive)
 641 *
 642 * Write out and wait upon file offsets lstart->lend, inclusive.
 643 *
 644 * Note that @lend is inclusive (describes the last byte to be written) so
 645 * that this function can be used to write to the very end-of-file (end = -1).
 646 *
 647 * Return: error status of the address space.
 648 */
 649int filemap_write_and_wait_range(struct address_space *mapping,
 650				 loff_t lstart, loff_t lend)
 651{
 652	int err = 0;
 653
 654	if (mapping_needs_writeback(mapping)) {
 655		err = __filemap_fdatawrite_range(mapping, lstart, lend,
 656						 WB_SYNC_ALL);
 657		/*
 658		 * Even if the above returned error, the pages may be
 659		 * written partially (e.g. -ENOSPC), so we wait for it.
 660		 * But the -EIO is special case, it may indicate the worst
 661		 * thing (e.g. bug) happened, so we avoid waiting for it.
 662		 */
 663		if (err != -EIO) {
 664			int err2 = filemap_fdatawait_range(mapping,
 665						lstart, lend);
 666			if (!err)
 667				err = err2;
 668		} else {
 669			/* Clear any previously stored errors */
 670			filemap_check_errors(mapping);
 671		}
 672	} else {
 673		err = filemap_check_errors(mapping);
 674	}
 675	return err;
 676}
 677EXPORT_SYMBOL(filemap_write_and_wait_range);
 678
 679void __filemap_set_wb_err(struct address_space *mapping, int err)
 680{
 681	errseq_t eseq = errseq_set(&mapping->wb_err, err);
 682
 683	trace_filemap_set_wb_err(mapping, eseq);
 684}
 685EXPORT_SYMBOL(__filemap_set_wb_err);
 686
 687/**
 688 * file_check_and_advance_wb_err - report wb error (if any) that was previously
 689 * 				   and advance wb_err to current one
 690 * @file: struct file on which the error is being reported
 691 *
 692 * When userland calls fsync (or something like nfsd does the equivalent), we
 693 * want to report any writeback errors that occurred since the last fsync (or
 694 * since the file was opened if there haven't been any).
 695 *
 696 * Grab the wb_err from the mapping. If it matches what we have in the file,
 697 * then just quickly return 0. The file is all caught up.
 698 *
 699 * If it doesn't match, then take the mapping value, set the "seen" flag in
 700 * it and try to swap it into place. If it works, or another task beat us
 701 * to it with the new value, then update the f_wb_err and return the error
 702 * portion. The error at this point must be reported via proper channels
 703 * (a'la fsync, or NFS COMMIT operation, etc.).
 704 *
 705 * While we handle mapping->wb_err with atomic operations, the f_wb_err
 706 * value is protected by the f_lock since we must ensure that it reflects
 707 * the latest value swapped in for this file descriptor.
 708 *
 709 * Return: %0 on success, negative error code otherwise.
 710 */
 711int file_check_and_advance_wb_err(struct file *file)
 712{
 713	int err = 0;
 714	errseq_t old = READ_ONCE(file->f_wb_err);
 715	struct address_space *mapping = file->f_mapping;
 716
 717	/* Locklessly handle the common case where nothing has changed */
 718	if (errseq_check(&mapping->wb_err, old)) {
 719		/* Something changed, must use slow path */
 720		spin_lock(&file->f_lock);
 721		old = file->f_wb_err;
 722		err = errseq_check_and_advance(&mapping->wb_err,
 723						&file->f_wb_err);
 724		trace_file_check_and_advance_wb_err(file, old);
 725		spin_unlock(&file->f_lock);
 726	}
 727
 728	/*
 729	 * We're mostly using this function as a drop in replacement for
 730	 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
 731	 * that the legacy code would have had on these flags.
 732	 */
 733	clear_bit(AS_EIO, &mapping->flags);
 734	clear_bit(AS_ENOSPC, &mapping->flags);
 735	return err;
 736}
 737EXPORT_SYMBOL(file_check_and_advance_wb_err);
 738
 739/**
 740 * file_write_and_wait_range - write out & wait on a file range
 741 * @file:	file pointing to address_space with pages
 742 * @lstart:	offset in bytes where the range starts
 743 * @lend:	offset in bytes where the range ends (inclusive)
 744 *
 745 * Write out and wait upon file offsets lstart->lend, inclusive.
 746 *
 747 * Note that @lend is inclusive (describes the last byte to be written) so
 748 * that this function can be used to write to the very end-of-file (end = -1).
 749 *
 750 * After writing out and waiting on the data, we check and advance the
 751 * f_wb_err cursor to the latest value, and return any errors detected there.
 752 *
 753 * Return: %0 on success, negative error code otherwise.
 754 */
 755int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
 
 756{
 757	int err = 0, err2;
 758	struct address_space *mapping = file->f_mapping;
 759
 760	if (mapping_needs_writeback(mapping)) {
 761		err = __filemap_fdatawrite_range(mapping, lstart, lend,
 762						 WB_SYNC_ALL);
 763		/* See comment of filemap_write_and_wait() */
 764		if (err != -EIO)
 765			__filemap_fdatawait_range(mapping, lstart, lend);
 
 
 
 
 
 
 766	}
 767	err2 = file_check_and_advance_wb_err(file);
 768	if (!err)
 769		err = err2;
 770	return err;
 771}
 772EXPORT_SYMBOL(file_write_and_wait_range);
 773
 774/**
 775 * replace_page_cache_page - replace a pagecache page with a new one
 776 * @old:	page to be replaced
 777 * @new:	page to replace with
 778 * @gfp_mask:	allocation mode
 779 *
 780 * This function replaces a page in the pagecache with a new one.  On
 781 * success it acquires the pagecache reference for the new page and
 782 * drops it for the old page.  Both the old and new pages must be
 783 * locked.  This function does not add the new page to the LRU, the
 784 * caller must do that.
 785 *
 786 * The remove + add is atomic.  This function cannot fail.
 787 *
 788 * Return: %0
 789 */
 790int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
 791{
 792	struct address_space *mapping = old->mapping;
 793	void (*freepage)(struct page *) = mapping->a_ops->freepage;
 794	pgoff_t offset = old->index;
 795	XA_STATE(xas, &mapping->i_pages, offset);
 796	unsigned long flags;
 797
 798	VM_BUG_ON_PAGE(!PageLocked(old), old);
 799	VM_BUG_ON_PAGE(!PageLocked(new), new);
 800	VM_BUG_ON_PAGE(new->mapping, new);
 801
 802	get_page(new);
 803	new->mapping = mapping;
 804	new->index = offset;
 805
 806	mem_cgroup_migrate(old, new);
 807
 808	xas_lock_irqsave(&xas, flags);
 809	xas_store(&xas, new);
 810
 811	old->mapping = NULL;
 812	/* hugetlb pages do not participate in page cache accounting. */
 813	if (!PageHuge(old))
 814		__dec_lruvec_page_state(old, NR_FILE_PAGES);
 815	if (!PageHuge(new))
 816		__inc_lruvec_page_state(new, NR_FILE_PAGES);
 817	if (PageSwapBacked(old))
 818		__dec_lruvec_page_state(old, NR_SHMEM);
 819	if (PageSwapBacked(new))
 820		__inc_lruvec_page_state(new, NR_SHMEM);
 821	xas_unlock_irqrestore(&xas, flags);
 822	if (freepage)
 823		freepage(old);
 824	put_page(old);
 825
 826	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 827}
 828EXPORT_SYMBOL_GPL(replace_page_cache_page);
 829
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 830static int __add_to_page_cache_locked(struct page *page,
 831				      struct address_space *mapping,
 832				      pgoff_t offset, gfp_t gfp_mask,
 833				      void **shadowp)
 834{
 835	XA_STATE(xas, &mapping->i_pages, offset);
 836	int huge = PageHuge(page);
 837	int error;
 838	void *old;
 839
 840	VM_BUG_ON_PAGE(!PageLocked(page), page);
 841	VM_BUG_ON_PAGE(PageSwapBacked(page), page);
 842	mapping_set_update(&xas, mapping);
 843
 844	get_page(page);
 845	page->mapping = mapping;
 846	page->index = offset;
 847
 848	if (!huge) {
 849		error = mem_cgroup_charge(page, current->mm, gfp_mask);
 850		if (error)
 851			goto error;
 
 
 
 
 
 852	}
 853
 854	do {
 855		xas_lock_irq(&xas);
 856		old = xas_load(&xas);
 857		if (old && !xa_is_value(old))
 858			xas_set_err(&xas, -EEXIST);
 859		xas_store(&xas, page);
 860		if (xas_error(&xas))
 861			goto unlock;
 862
 863		if (xa_is_value(old)) {
 864			mapping->nrexceptional--;
 865			if (shadowp)
 866				*shadowp = old;
 867		}
 868		mapping->nrpages++;
 869
 870		/* hugetlb pages do not participate in page cache accounting */
 871		if (!huge)
 872			__inc_lruvec_page_state(page, NR_FILE_PAGES);
 873unlock:
 874		xas_unlock_irq(&xas);
 875	} while (xas_nomem(&xas, gfp_mask & GFP_RECLAIM_MASK));
 876
 877	if (xas_error(&xas)) {
 878		error = xas_error(&xas);
 879		goto error;
 880	}
 881
 
 
 
 
 
 
 
 882	trace_mm_filemap_add_to_page_cache(page);
 883	return 0;
 884error:
 885	page->mapping = NULL;
 886	/* Leave page->index set: truncation relies upon it */
 887	put_page(page);
 
 
 888	return error;
 889}
 890ALLOW_ERROR_INJECTION(__add_to_page_cache_locked, ERRNO);
 891
 892/**
 893 * add_to_page_cache_locked - add a locked page to the pagecache
 894 * @page:	page to add
 895 * @mapping:	the page's address_space
 896 * @offset:	page index
 897 * @gfp_mask:	page allocation mode
 898 *
 899 * This function is used to add a page to the pagecache. It must be locked.
 900 * This function does not add the page to the LRU.  The caller must do that.
 901 *
 902 * Return: %0 on success, negative error code otherwise.
 903 */
 904int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
 905		pgoff_t offset, gfp_t gfp_mask)
 906{
 907	return __add_to_page_cache_locked(page, mapping, offset,
 908					  gfp_mask, NULL);
 909}
 910EXPORT_SYMBOL(add_to_page_cache_locked);
 911
 912int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
 913				pgoff_t offset, gfp_t gfp_mask)
 914{
 915	void *shadow = NULL;
 916	int ret;
 917
 918	__SetPageLocked(page);
 919	ret = __add_to_page_cache_locked(page, mapping, offset,
 920					 gfp_mask, &shadow);
 921	if (unlikely(ret))
 922		__ClearPageLocked(page);
 923	else {
 924		/*
 925		 * The page might have been evicted from cache only
 926		 * recently, in which case it should be activated like
 927		 * any other repeatedly accessed page.
 928		 * The exception is pages getting rewritten; evicting other
 929		 * data from the working set, only to cache data that will
 930		 * get overwritten with something else, is a waste of memory.
 931		 */
 932		WARN_ON_ONCE(PageActive(page));
 933		if (!(gfp_mask & __GFP_WRITE) && shadow)
 934			workingset_refault(page, shadow);
 935		lru_cache_add(page);
 936	}
 937	return ret;
 938}
 939EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
 940
 941#ifdef CONFIG_NUMA
 942struct page *__page_cache_alloc(gfp_t gfp)
 943{
 944	int n;
 945	struct page *page;
 946
 947	if (cpuset_do_page_mem_spread()) {
 948		unsigned int cpuset_mems_cookie;
 949		do {
 950			cpuset_mems_cookie = read_mems_allowed_begin();
 951			n = cpuset_mem_spread_node();
 952			page = __alloc_pages_node(n, gfp, 0);
 953		} while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
 954
 955		return page;
 956	}
 957	return alloc_pages(gfp, 0);
 958}
 959EXPORT_SYMBOL(__page_cache_alloc);
 960#endif
 961
 962/*
 963 * In order to wait for pages to become available there must be
 964 * waitqueues associated with pages. By using a hash table of
 965 * waitqueues where the bucket discipline is to maintain all
 966 * waiters on the same queue and wake all when any of the pages
 967 * become available, and for the woken contexts to check to be
 968 * sure the appropriate page became available, this saves space
 969 * at a cost of "thundering herd" phenomena during rare hash
 970 * collisions.
 971 */
 972#define PAGE_WAIT_TABLE_BITS 8
 973#define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
 974static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
 975
 976static wait_queue_head_t *page_waitqueue(struct page *page)
 977{
 978	return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
 979}
 980
 981void __init pagecache_init(void)
 982{
 983	int i;
 984
 985	for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
 986		init_waitqueue_head(&page_wait_table[i]);
 987
 988	page_writeback_init();
 989}
 990
 991/*
 992 * The page wait code treats the "wait->flags" somewhat unusually, because
 993 * we have multiple different kinds of waits, not just the usual "exclusive"
 994 * one.
 995 *
 996 * We have:
 997 *
 998 *  (a) no special bits set:
 999 *
1000 *	We're just waiting for the bit to be released, and when a waker
1001 *	calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1002 *	and remove it from the wait queue.
1003 *
1004 *	Simple and straightforward.
1005 *
1006 *  (b) WQ_FLAG_EXCLUSIVE:
1007 *
1008 *	The waiter is waiting to get the lock, and only one waiter should
1009 *	be woken up to avoid any thundering herd behavior. We'll set the
1010 *	WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1011 *
1012 *	This is the traditional exclusive wait.
1013 *
1014 *  (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1015 *
1016 *	The waiter is waiting to get the bit, and additionally wants the
1017 *	lock to be transferred to it for fair lock behavior. If the lock
1018 *	cannot be taken, we stop walking the wait queue without waking
1019 *	the waiter.
1020 *
1021 *	This is the "fair lock handoff" case, and in addition to setting
1022 *	WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1023 *	that it now has the lock.
1024 */
1025static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1026{
1027	unsigned int flags;
1028	struct wait_page_key *key = arg;
1029	struct wait_page_queue *wait_page
1030		= container_of(wait, struct wait_page_queue, wait);
1031
1032	if (!wake_page_match(wait_page, key))
1033		return 0;
1034
1035	/*
1036	 * If it's a lock handoff wait, we get the bit for it, and
1037	 * stop walking (and do not wake it up) if we can't.
1038	 */
1039	flags = wait->flags;
1040	if (flags & WQ_FLAG_EXCLUSIVE) {
1041		if (test_bit(key->bit_nr, &key->page->flags))
1042			return -1;
1043		if (flags & WQ_FLAG_CUSTOM) {
1044			if (test_and_set_bit(key->bit_nr, &key->page->flags))
1045				return -1;
1046			flags |= WQ_FLAG_DONE;
1047		}
1048	}
1049
1050	/*
1051	 * We are holding the wait-queue lock, but the waiter that
1052	 * is waiting for this will be checking the flags without
1053	 * any locking.
1054	 *
1055	 * So update the flags atomically, and wake up the waiter
1056	 * afterwards to avoid any races. This store-release pairs
1057	 * with the load-acquire in wait_on_page_bit_common().
1058	 */
1059	smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1060	wake_up_state(wait->private, mode);
1061
1062	/*
1063	 * Ok, we have successfully done what we're waiting for,
1064	 * and we can unconditionally remove the wait entry.
1065	 *
1066	 * Note that this pairs with the "finish_wait()" in the
1067	 * waiter, and has to be the absolute last thing we do.
1068	 * After this list_del_init(&wait->entry) the wait entry
1069	 * might be de-allocated and the process might even have
1070	 * exited.
1071	 */
1072	list_del_init_careful(&wait->entry);
1073	return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1074}
1075
1076static void wake_up_page_bit(struct page *page, int bit_nr)
1077{
1078	wait_queue_head_t *q = page_waitqueue(page);
1079	struct wait_page_key key;
1080	unsigned long flags;
1081	wait_queue_entry_t bookmark;
1082
1083	key.page = page;
1084	key.bit_nr = bit_nr;
1085	key.page_match = 0;
1086
1087	bookmark.flags = 0;
1088	bookmark.private = NULL;
1089	bookmark.func = NULL;
1090	INIT_LIST_HEAD(&bookmark.entry);
1091
1092	spin_lock_irqsave(&q->lock, flags);
1093	__wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1094
1095	while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1096		/*
1097		 * Take a breather from holding the lock,
1098		 * allow pages that finish wake up asynchronously
1099		 * to acquire the lock and remove themselves
1100		 * from wait queue
1101		 */
1102		spin_unlock_irqrestore(&q->lock, flags);
1103		cpu_relax();
1104		spin_lock_irqsave(&q->lock, flags);
1105		__wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1106	}
1107
1108	/*
1109	 * It is possible for other pages to have collided on the waitqueue
1110	 * hash, so in that case check for a page match. That prevents a long-
1111	 * term waiter
1112	 *
1113	 * It is still possible to miss a case here, when we woke page waiters
1114	 * and removed them from the waitqueue, but there are still other
1115	 * page waiters.
1116	 */
1117	if (!waitqueue_active(q) || !key.page_match) {
1118		ClearPageWaiters(page);
1119		/*
1120		 * It's possible to miss clearing Waiters here, when we woke
1121		 * our page waiters, but the hashed waitqueue has waiters for
1122		 * other pages on it.
1123		 *
1124		 * That's okay, it's a rare case. The next waker will clear it.
1125		 */
1126	}
1127	spin_unlock_irqrestore(&q->lock, flags);
1128}
1129
1130static void wake_up_page(struct page *page, int bit)
1131{
1132	if (!PageWaiters(page))
1133		return;
1134	wake_up_page_bit(page, bit);
1135}
1136
1137/*
1138 * A choice of three behaviors for wait_on_page_bit_common():
1139 */
1140enum behavior {
1141	EXCLUSIVE,	/* Hold ref to page and take the bit when woken, like
1142			 * __lock_page() waiting on then setting PG_locked.
1143			 */
1144	SHARED,		/* Hold ref to page and check the bit when woken, like
1145			 * wait_on_page_writeback() waiting on PG_writeback.
1146			 */
1147	DROP,		/* Drop ref to page before wait, no check when woken,
1148			 * like put_and_wait_on_page_locked() on PG_locked.
1149			 */
1150};
1151
1152/*
1153 * Attempt to check (or get) the page bit, and mark us done
1154 * if successful.
1155 */
1156static inline bool trylock_page_bit_common(struct page *page, int bit_nr,
1157					struct wait_queue_entry *wait)
1158{
1159	if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1160		if (test_and_set_bit(bit_nr, &page->flags))
1161			return false;
1162	} else if (test_bit(bit_nr, &page->flags))
1163		return false;
1164
1165	wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1166	return true;
1167}
1168
1169/* How many times do we accept lock stealing from under a waiter? */
1170int sysctl_page_lock_unfairness = 5;
1171
1172static inline int wait_on_page_bit_common(wait_queue_head_t *q,
1173	struct page *page, int bit_nr, int state, enum behavior behavior)
1174{
1175	int unfairness = sysctl_page_lock_unfairness;
1176	struct wait_page_queue wait_page;
1177	wait_queue_entry_t *wait = &wait_page.wait;
1178	bool thrashing = false;
1179	bool delayacct = false;
1180	unsigned long pflags;
1181
1182	if (bit_nr == PG_locked &&
1183	    !PageUptodate(page) && PageWorkingset(page)) {
1184		if (!PageSwapBacked(page)) {
1185			delayacct_thrashing_start();
1186			delayacct = true;
1187		}
1188		psi_memstall_enter(&pflags);
1189		thrashing = true;
1190	}
1191
1192	init_wait(wait);
1193	wait->func = wake_page_function;
1194	wait_page.page = page;
1195	wait_page.bit_nr = bit_nr;
1196
1197repeat:
1198	wait->flags = 0;
1199	if (behavior == EXCLUSIVE) {
1200		wait->flags = WQ_FLAG_EXCLUSIVE;
1201		if (--unfairness < 0)
1202			wait->flags |= WQ_FLAG_CUSTOM;
1203	}
1204
1205	/*
1206	 * Do one last check whether we can get the
1207	 * page bit synchronously.
1208	 *
1209	 * Do the SetPageWaiters() marking before that
1210	 * to let any waker we _just_ missed know they
1211	 * need to wake us up (otherwise they'll never
1212	 * even go to the slow case that looks at the
1213	 * page queue), and add ourselves to the wait
1214	 * queue if we need to sleep.
1215	 *
1216	 * This part needs to be done under the queue
1217	 * lock to avoid races.
1218	 */
1219	spin_lock_irq(&q->lock);
1220	SetPageWaiters(page);
1221	if (!trylock_page_bit_common(page, bit_nr, wait))
1222		__add_wait_queue_entry_tail(q, wait);
1223	spin_unlock_irq(&q->lock);
1224
1225	/*
1226	 * From now on, all the logic will be based on
1227	 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1228	 * see whether the page bit testing has already
1229	 * been done by the wake function.
1230	 *
1231	 * We can drop our reference to the page.
1232	 */
1233	if (behavior == DROP)
1234		put_page(page);
1235
1236	/*
1237	 * Note that until the "finish_wait()", or until
1238	 * we see the WQ_FLAG_WOKEN flag, we need to
1239	 * be very careful with the 'wait->flags', because
1240	 * we may race with a waker that sets them.
1241	 */
1242	for (;;) {
1243		unsigned int flags;
1244
1245		set_current_state(state);
1246
1247		/* Loop until we've been woken or interrupted */
1248		flags = smp_load_acquire(&wait->flags);
1249		if (!(flags & WQ_FLAG_WOKEN)) {
1250			if (signal_pending_state(state, current))
1251				break;
1252
1253			io_schedule();
1254			continue;
1255		}
1256
1257		/* If we were non-exclusive, we're done */
1258		if (behavior != EXCLUSIVE)
1259			break;
1260
1261		/* If the waker got the lock for us, we're done */
1262		if (flags & WQ_FLAG_DONE)
1263			break;
1264
1265		/*
1266		 * Otherwise, if we're getting the lock, we need to
1267		 * try to get it ourselves.
1268		 *
1269		 * And if that fails, we'll have to retry this all.
1270		 */
1271		if (unlikely(test_and_set_bit(bit_nr, &page->flags)))
1272			goto repeat;
1273
1274		wait->flags |= WQ_FLAG_DONE;
1275		break;
1276	}
1277
1278	/*
1279	 * If a signal happened, this 'finish_wait()' may remove the last
1280	 * waiter from the wait-queues, but the PageWaiters bit will remain
1281	 * set. That's ok. The next wakeup will take care of it, and trying
1282	 * to do it here would be difficult and prone to races.
1283	 */
1284	finish_wait(q, wait);
1285
1286	if (thrashing) {
1287		if (delayacct)
1288			delayacct_thrashing_end();
1289		psi_memstall_leave(&pflags);
1290	}
1291
1292	/*
1293	 * NOTE! The wait->flags weren't stable until we've done the
1294	 * 'finish_wait()', and we could have exited the loop above due
1295	 * to a signal, and had a wakeup event happen after the signal
1296	 * test but before the 'finish_wait()'.
1297	 *
1298	 * So only after the finish_wait() can we reliably determine
1299	 * if we got woken up or not, so we can now figure out the final
1300	 * return value based on that state without races.
1301	 *
1302	 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1303	 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1304	 */
1305	if (behavior == EXCLUSIVE)
1306		return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1307
1308	return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1309}
1310
1311void wait_on_page_bit(struct page *page, int bit_nr)
1312{
1313	wait_queue_head_t *q = page_waitqueue(page);
1314	wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
 
 
 
1315}
1316EXPORT_SYMBOL(wait_on_page_bit);
1317
1318int wait_on_page_bit_killable(struct page *page, int bit_nr)
1319{
1320	wait_queue_head_t *q = page_waitqueue(page);
1321	return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED);
1322}
1323EXPORT_SYMBOL(wait_on_page_bit_killable);
1324
1325static int __wait_on_page_locked_async(struct page *page,
1326				       struct wait_page_queue *wait, bool set)
1327{
1328	struct wait_queue_head *q = page_waitqueue(page);
1329	int ret = 0;
1330
1331	wait->page = page;
1332	wait->bit_nr = PG_locked;
1333
1334	spin_lock_irq(&q->lock);
1335	__add_wait_queue_entry_tail(q, &wait->wait);
1336	SetPageWaiters(page);
1337	if (set)
1338		ret = !trylock_page(page);
1339	else
1340		ret = PageLocked(page);
1341	/*
1342	 * If we were succesful now, we know we're still on the
1343	 * waitqueue as we're still under the lock. This means it's
1344	 * safe to remove and return success, we know the callback
1345	 * isn't going to trigger.
1346	 */
1347	if (!ret)
1348		__remove_wait_queue(q, &wait->wait);
1349	else
1350		ret = -EIOCBQUEUED;
1351	spin_unlock_irq(&q->lock);
1352	return ret;
1353}
1354
1355static int wait_on_page_locked_async(struct page *page,
1356				     struct wait_page_queue *wait)
1357{
1358	if (!PageLocked(page))
1359		return 0;
1360	return __wait_on_page_locked_async(compound_head(page), wait, false);
1361}
1362
1363/**
1364 * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1365 * @page: The page to wait for.
1366 *
1367 * The caller should hold a reference on @page.  They expect the page to
1368 * become unlocked relatively soon, but do not wish to hold up migration
1369 * (for example) by holding the reference while waiting for the page to
1370 * come unlocked.  After this function returns, the caller should not
1371 * dereference @page.
1372 */
1373void put_and_wait_on_page_locked(struct page *page)
1374{
1375	wait_queue_head_t *q;
1376
1377	page = compound_head(page);
1378	q = page_waitqueue(page);
1379	wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, DROP);
1380}
1381
1382/**
1383 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1384 * @page: Page defining the wait queue of interest
1385 * @waiter: Waiter to add to the queue
1386 *
1387 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1388 */
1389void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
1390{
1391	wait_queue_head_t *q = page_waitqueue(page);
1392	unsigned long flags;
1393
1394	spin_lock_irqsave(&q->lock, flags);
1395	__add_wait_queue_entry_tail(q, waiter);
1396	SetPageWaiters(page);
1397	spin_unlock_irqrestore(&q->lock, flags);
1398}
1399EXPORT_SYMBOL_GPL(add_page_wait_queue);
1400
1401#ifndef clear_bit_unlock_is_negative_byte
1402
1403/*
1404 * PG_waiters is the high bit in the same byte as PG_lock.
1405 *
1406 * On x86 (and on many other architectures), we can clear PG_lock and
1407 * test the sign bit at the same time. But if the architecture does
1408 * not support that special operation, we just do this all by hand
1409 * instead.
1410 *
1411 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1412 * being cleared, but a memory barrier should be unnecessary since it is
1413 * in the same byte as PG_locked.
1414 */
1415static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1416{
1417	clear_bit_unlock(nr, mem);
1418	/* smp_mb__after_atomic(); */
1419	return test_bit(PG_waiters, mem);
1420}
1421
1422#endif
1423
1424/**
1425 * unlock_page - unlock a locked page
1426 * @page: the page
1427 *
1428 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
1429 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1430 * mechanism between PageLocked pages and PageWriteback pages is shared.
1431 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1432 *
1433 * Note that this depends on PG_waiters being the sign bit in the byte
1434 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1435 * clear the PG_locked bit and test PG_waiters at the same time fairly
1436 * portably (architectures that do LL/SC can test any bit, while x86 can
1437 * test the sign bit).
1438 */
1439void unlock_page(struct page *page)
1440{
1441	BUILD_BUG_ON(PG_waiters != 7);
1442	page = compound_head(page);
1443	VM_BUG_ON_PAGE(!PageLocked(page), page);
1444	if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
1445		wake_up_page_bit(page, PG_locked);
 
1446}
1447EXPORT_SYMBOL(unlock_page);
1448
1449/**
1450 * end_page_writeback - end writeback against a page
1451 * @page: the page
1452 */
1453void end_page_writeback(struct page *page)
1454{
1455	/*
1456	 * TestClearPageReclaim could be used here but it is an atomic
1457	 * operation and overkill in this particular case. Failing to
1458	 * shuffle a page marked for immediate reclaim is too mild to
1459	 * justify taking an atomic operation penalty at the end of
1460	 * ever page writeback.
1461	 */
1462	if (PageReclaim(page)) {
1463		ClearPageReclaim(page);
1464		rotate_reclaimable_page(page);
1465	}
1466
1467	if (!test_clear_page_writeback(page))
1468		BUG();
1469
1470	smp_mb__after_atomic();
1471	wake_up_page(page, PG_writeback);
1472}
1473EXPORT_SYMBOL(end_page_writeback);
1474
1475/*
1476 * After completing I/O on a page, call this routine to update the page
1477 * flags appropriately
1478 */
1479void page_endio(struct page *page, bool is_write, int err)
1480{
1481	if (!is_write) {
1482		if (!err) {
1483			SetPageUptodate(page);
1484		} else {
1485			ClearPageUptodate(page);
1486			SetPageError(page);
1487		}
1488		unlock_page(page);
1489	} else {
1490		if (err) {
1491			struct address_space *mapping;
1492
1493			SetPageError(page);
1494			mapping = page_mapping(page);
1495			if (mapping)
1496				mapping_set_error(mapping, err);
1497		}
1498		end_page_writeback(page);
1499	}
1500}
1501EXPORT_SYMBOL_GPL(page_endio);
1502
1503/**
1504 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1505 * @__page: the page to lock
1506 */
1507void __lock_page(struct page *__page)
1508{
1509	struct page *page = compound_head(__page);
1510	wait_queue_head_t *q = page_waitqueue(page);
1511	wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE,
1512				EXCLUSIVE);
1513}
1514EXPORT_SYMBOL(__lock_page);
1515
1516int __lock_page_killable(struct page *__page)
1517{
1518	struct page *page = compound_head(__page);
1519	wait_queue_head_t *q = page_waitqueue(page);
1520	return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE,
1521					EXCLUSIVE);
1522}
1523EXPORT_SYMBOL_GPL(__lock_page_killable);
1524
1525int __lock_page_async(struct page *page, struct wait_page_queue *wait)
1526{
1527	return __wait_on_page_locked_async(page, wait, true);
1528}
 
1529
1530/*
1531 * Return values:
1532 * 1 - page is locked; mmap_lock is still held.
1533 * 0 - page is not locked.
1534 *     mmap_lock has been released (mmap_read_unlock(), unless flags had both
1535 *     FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1536 *     which case mmap_lock is still held.
1537 *
1538 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1539 * with the page locked and the mmap_lock unperturbed.
1540 */
1541int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1542			 unsigned int flags)
1543{
1544	if (fault_flag_allow_retry_first(flags)) {
1545		/*
1546		 * CAUTION! In this case, mmap_lock is not released
1547		 * even though return 0.
1548		 */
1549		if (flags & FAULT_FLAG_RETRY_NOWAIT)
1550			return 0;
1551
1552		mmap_read_unlock(mm);
1553		if (flags & FAULT_FLAG_KILLABLE)
1554			wait_on_page_locked_killable(page);
1555		else
1556			wait_on_page_locked(page);
1557		return 0;
1558	} else {
1559		if (flags & FAULT_FLAG_KILLABLE) {
1560			int ret;
1561
1562			ret = __lock_page_killable(page);
1563			if (ret) {
1564				mmap_read_unlock(mm);
1565				return 0;
1566			}
1567		} else
1568			__lock_page(page);
1569		return 1;
1570	}
1571}
1572
1573/**
1574 * page_cache_next_miss() - Find the next gap in the page cache.
1575 * @mapping: Mapping.
1576 * @index: Index.
1577 * @max_scan: Maximum range to search.
1578 *
1579 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1580 * gap with the lowest index.
1581 *
1582 * This function may be called under the rcu_read_lock.  However, this will
1583 * not atomically search a snapshot of the cache at a single point in time.
1584 * For example, if a gap is created at index 5, then subsequently a gap is
1585 * created at index 10, page_cache_next_miss covering both indices may
1586 * return 10 if called under the rcu_read_lock.
1587 *
1588 * Return: The index of the gap if found, otherwise an index outside the
1589 * range specified (in which case 'return - index >= max_scan' will be true).
1590 * In the rare case of index wrap-around, 0 will be returned.
 
 
1591 */
1592pgoff_t page_cache_next_miss(struct address_space *mapping,
1593			     pgoff_t index, unsigned long max_scan)
1594{
1595	XA_STATE(xas, &mapping->i_pages, index);
1596
1597	while (max_scan--) {
1598		void *entry = xas_next(&xas);
1599		if (!entry || xa_is_value(entry))
 
 
1600			break;
1601		if (xas.xa_index == 0)
 
1602			break;
1603	}
1604
1605	return xas.xa_index;
1606}
1607EXPORT_SYMBOL(page_cache_next_miss);
1608
1609/**
1610 * page_cache_prev_miss() - Find the previous gap in the page cache.
1611 * @mapping: Mapping.
1612 * @index: Index.
1613 * @max_scan: Maximum range to search.
1614 *
1615 * Search the range [max(index - max_scan + 1, 0), index] for the
1616 * gap with the highest index.
1617 *
1618 * This function may be called under the rcu_read_lock.  However, this will
1619 * not atomically search a snapshot of the cache at a single point in time.
1620 * For example, if a gap is created at index 10, then subsequently a gap is
1621 * created at index 5, page_cache_prev_miss() covering both indices may
1622 * return 5 if called under the rcu_read_lock.
1623 *
1624 * Return: The index of the gap if found, otherwise an index outside the
1625 * range specified (in which case 'index - return >= max_scan' will be true).
1626 * In the rare case of wrap-around, ULONG_MAX will be returned.
 
 
 
1627 */
1628pgoff_t page_cache_prev_miss(struct address_space *mapping,
1629			     pgoff_t index, unsigned long max_scan)
1630{
1631	XA_STATE(xas, &mapping->i_pages, index);
 
 
 
1632
1633	while (max_scan--) {
1634		void *entry = xas_prev(&xas);
1635		if (!entry || xa_is_value(entry))
1636			break;
1637		if (xas.xa_index == ULONG_MAX)
 
1638			break;
1639	}
1640
1641	return xas.xa_index;
1642}
1643EXPORT_SYMBOL(page_cache_prev_miss);
1644
1645/**
1646 * find_get_entry - find and get a page cache entry
1647 * @mapping: the address_space to search
1648 * @offset: the page cache index
1649 *
1650 * Looks up the page cache slot at @mapping & @offset.  If there is a
1651 * page cache page, it is returned with an increased refcount.
1652 *
1653 * If the slot holds a shadow entry of a previously evicted page, or a
1654 * swap entry from shmem/tmpfs, it is returned.
1655 *
1656 * Return: the found page or shadow entry, %NULL if nothing is found.
1657 */
1658struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1659{
1660	XA_STATE(xas, &mapping->i_pages, offset);
1661	struct page *page;
1662
1663	rcu_read_lock();
1664repeat:
1665	xas_reset(&xas);
1666	page = xas_load(&xas);
1667	if (xas_retry(&xas, page))
1668		goto repeat;
1669	/*
1670	 * A shadow entry of a recently evicted page, or a swap entry from
1671	 * shmem/tmpfs.  Return it without attempting to raise page count.
1672	 */
1673	if (!page || xa_is_value(page))
1674		goto out;
1675
1676	if (!page_cache_get_speculative(page))
1677		goto repeat;
 
 
 
 
 
1678
1679	/*
1680	 * Has the page moved or been split?
1681	 * This is part of the lockless pagecache protocol. See
1682	 * include/linux/pagemap.h for details.
1683	 */
1684	if (unlikely(page != xas_reload(&xas))) {
1685		put_page(page);
1686		goto repeat;
 
1687	}
1688	page = find_subpage(page, offset);
1689out:
1690	rcu_read_unlock();
1691
1692	return page;
1693}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1694
1695/**
1696 * find_lock_entry - locate, pin and lock a page cache entry
1697 * @mapping: the address_space to search
1698 * @offset: the page cache index
1699 *
1700 * Looks up the page cache slot at @mapping & @offset.  If there is a
1701 * page cache page, it is returned locked and with an increased
1702 * refcount.
1703 *
1704 * If the slot holds a shadow entry of a previously evicted page, or a
1705 * swap entry from shmem/tmpfs, it is returned.
1706 *
1707 * find_lock_entry() may sleep.
1708 *
1709 * Return: the found page or shadow entry, %NULL if nothing is found.
1710 */
1711struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1712{
1713	struct page *page;
1714
1715repeat:
1716	page = find_get_entry(mapping, offset);
1717	if (page && !xa_is_value(page)) {
1718		lock_page(page);
1719		/* Has the page been truncated? */
1720		if (unlikely(page_mapping(page) != mapping)) {
1721			unlock_page(page);
1722			put_page(page);
1723			goto repeat;
1724		}
1725		VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
1726	}
1727	return page;
1728}
1729EXPORT_SYMBOL(find_lock_entry);
1730
1731/**
1732 * pagecache_get_page - Find and get a reference to a page.
1733 * @mapping: The address_space to search.
1734 * @index: The page index.
1735 * @fgp_flags: %FGP flags modify how the page is returned.
1736 * @gfp_mask: Memory allocation flags to use if %FGP_CREAT is specified.
1737 *
1738 * Looks up the page cache entry at @mapping & @index.
1739 *
1740 * @fgp_flags can be zero or more of these flags:
1741 *
1742 * * %FGP_ACCESSED - The page will be marked accessed.
1743 * * %FGP_LOCK - The page is returned locked.
1744 * * %FGP_CREAT - If no page is present then a new page is allocated using
1745 *   @gfp_mask and added to the page cache and the VM's LRU list.
1746 *   The page is returned locked and with an increased refcount.
1747 * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
1748 *   page is already in cache.  If the page was allocated, unlock it before
1749 *   returning so the caller can do the same dance.
1750 * * %FGP_WRITE - The page will be written
1751 * * %FGP_NOFS - __GFP_FS will get cleared in gfp mask
1752 * * %FGP_NOWAIT - Don't get blocked by page lock
1753 *
1754 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1755 * if the %GFP flags specified for %FGP_CREAT are atomic.
1756 *
1757 * If there is a page cache page, it is returned with an increased refcount.
 
 
1758 *
1759 * Return: The found page or %NULL otherwise.
 
 
1760 */
1761struct page *pagecache_get_page(struct address_space *mapping, pgoff_t index,
1762		int fgp_flags, gfp_t gfp_mask)
1763{
1764	struct page *page;
1765
1766repeat:
1767	page = find_get_entry(mapping, index);
1768	if (xa_is_value(page))
1769		page = NULL;
1770	if (!page)
1771		goto no_page;
1772
1773	if (fgp_flags & FGP_LOCK) {
1774		if (fgp_flags & FGP_NOWAIT) {
1775			if (!trylock_page(page)) {
1776				put_page(page);
1777				return NULL;
1778			}
1779		} else {
1780			lock_page(page);
1781		}
1782
1783		/* Has the page been truncated? */
1784		if (unlikely(compound_head(page)->mapping != mapping)) {
1785			unlock_page(page);
1786			put_page(page);
1787			goto repeat;
1788		}
1789		VM_BUG_ON_PAGE(page->index != index, page);
1790	}
1791
1792	if (fgp_flags & FGP_ACCESSED)
1793		mark_page_accessed(page);
1794	else if (fgp_flags & FGP_WRITE) {
1795		/* Clear idle flag for buffer write */
1796		if (page_is_idle(page))
1797			clear_page_idle(page);
1798	}
1799
1800no_page:
1801	if (!page && (fgp_flags & FGP_CREAT)) {
1802		int err;
1803		if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1804			gfp_mask |= __GFP_WRITE;
1805		if (fgp_flags & FGP_NOFS)
1806			gfp_mask &= ~__GFP_FS;
1807
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1808		page = __page_cache_alloc(gfp_mask);
1809		if (!page)
1810			return NULL;
1811
1812		if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1813			fgp_flags |= FGP_LOCK;
1814
1815		/* Init accessed so avoid atomic mark_page_accessed later */
1816		if (fgp_flags & FGP_ACCESSED)
1817			__SetPageReferenced(page);
1818
1819		err = add_to_page_cache_lru(page, mapping, index, gfp_mask);
1820		if (unlikely(err)) {
1821			put_page(page);
1822			page = NULL;
1823			if (err == -EEXIST)
1824				goto repeat;
1825		}
1826
1827		/*
1828		 * add_to_page_cache_lru locks the page, and for mmap we expect
1829		 * an unlocked page.
1830		 */
1831		if (page && (fgp_flags & FGP_FOR_MMAP))
1832			unlock_page(page);
1833	}
1834
1835	return page;
1836}
1837EXPORT_SYMBOL(pagecache_get_page);
1838
1839/**
1840 * find_get_entries - gang pagecache lookup
1841 * @mapping:	The address_space to search
1842 * @start:	The starting page cache index
1843 * @nr_entries:	The maximum number of entries
1844 * @entries:	Where the resulting entries are placed
1845 * @indices:	The cache indices corresponding to the entries in @entries
1846 *
1847 * find_get_entries() will search for and return a group of up to
1848 * @nr_entries entries in the mapping.  The entries are placed at
1849 * @entries.  find_get_entries() takes a reference against any actual
1850 * pages it returns.
1851 *
1852 * The search returns a group of mapping-contiguous page cache entries
1853 * with ascending indexes.  There may be holes in the indices due to
1854 * not-present pages.
1855 *
1856 * Any shadow entries of evicted pages, or swap entries from
1857 * shmem/tmpfs, are included in the returned array.
1858 *
1859 * If it finds a Transparent Huge Page, head or tail, find_get_entries()
1860 * stops at that page: the caller is likely to have a better way to handle
1861 * the compound page as a whole, and then skip its extent, than repeatedly
1862 * calling find_get_entries() to return all its tails.
1863 *
1864 * Return: the number of pages and shadow entries which were found.
1865 */
1866unsigned find_get_entries(struct address_space *mapping,
1867			  pgoff_t start, unsigned int nr_entries,
1868			  struct page **entries, pgoff_t *indices)
1869{
1870	XA_STATE(xas, &mapping->i_pages, start);
1871	struct page *page;
1872	unsigned int ret = 0;
 
1873
1874	if (!nr_entries)
1875		return 0;
1876
1877	rcu_read_lock();
1878	xas_for_each(&xas, page, ULONG_MAX) {
1879		if (xas_retry(&xas, page))
 
 
 
 
1880			continue;
1881		/*
1882		 * A shadow entry of a recently evicted page, a swap
1883		 * entry from shmem/tmpfs or a DAX entry.  Return it
1884		 * without attempting to raise page count.
1885		 */
1886		if (xa_is_value(page))
 
 
1887			goto export;
1888
1889		if (!page_cache_get_speculative(page))
1890			goto retry;
1891
1892		/* Has the page moved or been split? */
1893		if (unlikely(page != xas_reload(&xas)))
1894			goto put_page;
1895
1896		/*
1897		 * Terminate early on finding a THP, to allow the caller to
1898		 * handle it all at once; but continue if this is hugetlbfs.
1899		 */
1900		if (PageTransHuge(page) && !PageHuge(page)) {
1901			page = find_subpage(page, xas.xa_index);
1902			nr_entries = ret + 1;
1903		}
1904export:
1905		indices[ret] = xas.xa_index;
1906		entries[ret] = page;
1907		if (++ret == nr_entries)
1908			break;
1909		continue;
1910put_page:
1911		put_page(page);
1912retry:
1913		xas_reset(&xas);
1914	}
1915	rcu_read_unlock();
1916	return ret;
1917}
1918
1919/**
1920 * find_get_pages_range - gang pagecache lookup
1921 * @mapping:	The address_space to search
1922 * @start:	The starting page index
1923 * @end:	The final page index (inclusive)
1924 * @nr_pages:	The maximum number of pages
1925 * @pages:	Where the resulting pages are placed
1926 *
1927 * find_get_pages_range() will search for and return a group of up to @nr_pages
1928 * pages in the mapping starting at index @start and up to index @end
1929 * (inclusive).  The pages are placed at @pages.  find_get_pages_range() takes
1930 * a reference against the returned pages.
1931 *
1932 * The search returns a group of mapping-contiguous pages with ascending
1933 * indexes.  There may be holes in the indices due to not-present pages.
1934 * We also update @start to index the next page for the traversal.
1935 *
1936 * Return: the number of pages which were found. If this number is
1937 * smaller than @nr_pages, the end of specified range has been
1938 * reached.
1939 */
1940unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
1941			      pgoff_t end, unsigned int nr_pages,
1942			      struct page **pages)
1943{
1944	XA_STATE(xas, &mapping->i_pages, *start);
1945	struct page *page;
1946	unsigned ret = 0;
1947
1948	if (unlikely(!nr_pages))
1949		return 0;
1950
1951	rcu_read_lock();
1952	xas_for_each(&xas, page, end) {
1953		if (xas_retry(&xas, page))
 
 
 
 
1954			continue;
1955		/* Skip over shadow, swap and DAX entries */
1956		if (xa_is_value(page))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1957			continue;
 
1958
1959		if (!page_cache_get_speculative(page))
1960			goto retry;
1961
1962		/* Has the page moved or been split? */
1963		if (unlikely(page != xas_reload(&xas)))
1964			goto put_page;
1965
1966		pages[ret] = find_subpage(page, xas.xa_index);
1967		if (++ret == nr_pages) {
1968			*start = xas.xa_index + 1;
1969			goto out;
1970		}
1971		continue;
1972put_page:
1973		put_page(page);
1974retry:
1975		xas_reset(&xas);
1976	}
1977
1978	/*
1979	 * We come here when there is no page beyond @end. We take care to not
1980	 * overflow the index @start as it confuses some of the callers. This
1981	 * breaks the iteration when there is a page at index -1 but that is
1982	 * already broken anyway.
1983	 */
1984	if (end == (pgoff_t)-1)
1985		*start = (pgoff_t)-1;
1986	else
1987		*start = end + 1;
1988out:
1989	rcu_read_unlock();
1990
1991	return ret;
1992}
1993
1994/**
1995 * find_get_pages_contig - gang contiguous pagecache lookup
1996 * @mapping:	The address_space to search
1997 * @index:	The starting page index
1998 * @nr_pages:	The maximum number of pages
1999 * @pages:	Where the resulting pages are placed
2000 *
2001 * find_get_pages_contig() works exactly like find_get_pages(), except
2002 * that the returned number of pages are guaranteed to be contiguous.
2003 *
2004 * Return: the number of pages which were found.
2005 */
2006unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
2007			       unsigned int nr_pages, struct page **pages)
2008{
2009	XA_STATE(xas, &mapping->i_pages, index);
2010	struct page *page;
2011	unsigned int ret = 0;
2012
2013	if (unlikely(!nr_pages))
2014		return 0;
2015
2016	rcu_read_lock();
2017	for (page = xas_load(&xas); page; page = xas_next(&xas)) {
2018		if (xas_retry(&xas, page))
2019			continue;
2020		/*
2021		 * If the entry has been swapped out, we can stop looking.
2022		 * No current caller is looking for DAX entries.
2023		 */
2024		if (xa_is_value(page))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2025			break;
 
2026
2027		if (!page_cache_get_speculative(page))
2028			goto retry;
2029
2030		/* Has the page moved or been split? */
2031		if (unlikely(page != xas_reload(&xas)))
2032			goto put_page;
 
 
2033
2034		pages[ret] = find_subpage(page, xas.xa_index);
 
 
 
 
 
 
 
 
 
 
2035		if (++ret == nr_pages)
2036			break;
2037		continue;
2038put_page:
2039		put_page(page);
2040retry:
2041		xas_reset(&xas);
2042	}
2043	rcu_read_unlock();
2044	return ret;
2045}
2046EXPORT_SYMBOL(find_get_pages_contig);
2047
2048/**
2049 * find_get_pages_range_tag - find and return pages in given range matching @tag
2050 * @mapping:	the address_space to search
2051 * @index:	the starting page index
2052 * @end:	The final page index (inclusive)
2053 * @tag:	the tag index
2054 * @nr_pages:	the maximum number of pages
2055 * @pages:	where the resulting pages are placed
2056 *
2057 * Like find_get_pages, except we only return pages which are tagged with
2058 * @tag.   We update @index to index the next page for the traversal.
2059 *
2060 * Return: the number of pages which were found.
2061 */
2062unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
2063			pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
2064			struct page **pages)
2065{
2066	XA_STATE(xas, &mapping->i_pages, *index);
2067	struct page *page;
2068	unsigned ret = 0;
2069
2070	if (unlikely(!nr_pages))
2071		return 0;
2072
2073	rcu_read_lock();
2074	xas_for_each_marked(&xas, page, end, tag) {
2075		if (xas_retry(&xas, page))
 
 
 
 
 
2076			continue;
2077		/*
2078		 * Shadow entries should never be tagged, but this iteration
2079		 * is lockless so there is a window for page reclaim to evict
2080		 * a page we saw tagged.  Skip over it.
2081		 */
2082		if (xa_is_value(page))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2083			continue;
 
2084
2085		if (!page_cache_get_speculative(page))
2086			goto retry;
2087
2088		/* Has the page moved or been split? */
2089		if (unlikely(page != xas_reload(&xas)))
2090			goto put_page;
2091
2092		pages[ret] = find_subpage(page, xas.xa_index);
2093		if (++ret == nr_pages) {
2094			*index = xas.xa_index + 1;
2095			goto out;
2096		}
2097		continue;
2098put_page:
2099		put_page(page);
2100retry:
2101		xas_reset(&xas);
2102	}
2103
2104	/*
2105	 * We come here when we got to @end. We take care to not overflow the
2106	 * index @index as it confuses some of the callers. This breaks the
2107	 * iteration when there is a page at index -1 but that is already
2108	 * broken anyway.
2109	 */
2110	if (end == (pgoff_t)-1)
2111		*index = (pgoff_t)-1;
2112	else
2113		*index = end + 1;
2114out:
2115	rcu_read_unlock();
2116
 
 
 
2117	return ret;
2118}
2119EXPORT_SYMBOL(find_get_pages_range_tag);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2120
2121/*
2122 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2123 * a _large_ part of the i/o request. Imagine the worst scenario:
2124 *
2125 *      ---R__________________________________________B__________
2126 *         ^ reading here                             ^ bad block(assume 4k)
2127 *
2128 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2129 * => failing the whole request => read(R) => read(R+1) =>
2130 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2131 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2132 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2133 *
2134 * It is going insane. Fix it by quickly scaling down the readahead size.
2135 */
2136static void shrink_readahead_size_eio(struct file_ra_state *ra)
 
2137{
2138	ra->ra_pages /= 4;
2139}
2140
2141/**
2142 * generic_file_buffered_read - generic file read routine
2143 * @iocb:	the iocb to read
 
2144 * @iter:	data destination
2145 * @written:	already copied
2146 *
2147 * This is a generic file read routine, and uses the
2148 * mapping->a_ops->readpage() function for the actual low-level stuff.
2149 *
2150 * This is really ugly. But the goto's actually try to clarify some
2151 * of the logic when it comes to error handling etc.
2152 *
2153 * Return:
2154 * * total number of bytes copied, including those the were already @written
2155 * * negative error code if nothing was copied
2156 */
2157ssize_t generic_file_buffered_read(struct kiocb *iocb,
2158		struct iov_iter *iter, ssize_t written)
2159{
2160	struct file *filp = iocb->ki_filp;
2161	struct address_space *mapping = filp->f_mapping;
2162	struct inode *inode = mapping->host;
2163	struct file_ra_state *ra = &filp->f_ra;
2164	loff_t *ppos = &iocb->ki_pos;
2165	pgoff_t index;
2166	pgoff_t last_index;
2167	pgoff_t prev_index;
2168	unsigned long offset;      /* offset into pagecache page */
2169	unsigned int prev_offset;
2170	int error = 0;
2171
2172	if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
2173		return 0;
2174	iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2175
2176	index = *ppos >> PAGE_SHIFT;
2177	prev_index = ra->prev_pos >> PAGE_SHIFT;
2178	prev_offset = ra->prev_pos & (PAGE_SIZE-1);
2179	last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
2180	offset = *ppos & ~PAGE_MASK;
2181
2182	for (;;) {
2183		struct page *page;
2184		pgoff_t end_index;
2185		loff_t isize;
2186		unsigned long nr, ret;
2187
2188		cond_resched();
2189find_page:
2190		if (fatal_signal_pending(current)) {
2191			error = -EINTR;
2192			goto out;
2193		}
2194
2195		page = find_get_page(mapping, index);
2196		if (!page) {
2197			if (iocb->ki_flags & IOCB_NOIO)
2198				goto would_block;
2199			page_cache_sync_readahead(mapping,
2200					ra, filp,
2201					index, last_index - index);
2202			page = find_get_page(mapping, index);
2203			if (unlikely(page == NULL))
2204				goto no_cached_page;
2205		}
2206		if (PageReadahead(page)) {
2207			if (iocb->ki_flags & IOCB_NOIO) {
2208				put_page(page);
2209				goto out;
2210			}
2211			page_cache_async_readahead(mapping,
2212					ra, filp, page,
2213					index, last_index - index);
2214		}
2215		if (!PageUptodate(page)) {
2216			/*
2217			 * See comment in do_read_cache_page on why
2218			 * wait_on_page_locked is used to avoid unnecessarily
2219			 * serialisations and why it's safe.
2220			 */
2221			if (iocb->ki_flags & IOCB_WAITQ) {
2222				if (written) {
2223					put_page(page);
2224					goto out;
2225				}
2226				error = wait_on_page_locked_async(page,
2227								iocb->ki_waitq);
2228			} else {
2229				if (iocb->ki_flags & IOCB_NOWAIT) {
2230					put_page(page);
2231					goto would_block;
2232				}
2233				error = wait_on_page_locked_killable(page);
2234			}
2235			if (unlikely(error))
2236				goto readpage_error;
2237			if (PageUptodate(page))
2238				goto page_ok;
2239
2240			if (inode->i_blkbits == PAGE_SHIFT ||
2241					!mapping->a_ops->is_partially_uptodate)
2242				goto page_not_up_to_date;
2243			/* pipes can't handle partially uptodate pages */
2244			if (unlikely(iov_iter_is_pipe(iter)))
2245				goto page_not_up_to_date;
2246			if (!trylock_page(page))
2247				goto page_not_up_to_date;
2248			/* Did it get truncated before we got the lock? */
2249			if (!page->mapping)
2250				goto page_not_up_to_date_locked;
2251			if (!mapping->a_ops->is_partially_uptodate(page,
2252							offset, iter->count))
2253				goto page_not_up_to_date_locked;
2254			unlock_page(page);
2255		}
2256page_ok:
2257		/*
2258		 * i_size must be checked after we know the page is Uptodate.
2259		 *
2260		 * Checking i_size after the check allows us to calculate
2261		 * the correct value for "nr", which means the zero-filled
2262		 * part of the page is not copied back to userspace (unless
2263		 * another truncate extends the file - this is desired though).
2264		 */
2265
2266		isize = i_size_read(inode);
2267		end_index = (isize - 1) >> PAGE_SHIFT;
2268		if (unlikely(!isize || index > end_index)) {
2269			put_page(page);
2270			goto out;
2271		}
2272
2273		/* nr is the maximum number of bytes to copy from this page */
2274		nr = PAGE_SIZE;
2275		if (index == end_index) {
2276			nr = ((isize - 1) & ~PAGE_MASK) + 1;
2277			if (nr <= offset) {
2278				put_page(page);
2279				goto out;
2280			}
2281		}
2282		nr = nr - offset;
2283
2284		/* If users can be writing to this page using arbitrary
2285		 * virtual addresses, take care about potential aliasing
2286		 * before reading the page on the kernel side.
2287		 */
2288		if (mapping_writably_mapped(mapping))
2289			flush_dcache_page(page);
2290
2291		/*
2292		 * When a sequential read accesses a page several times,
2293		 * only mark it as accessed the first time.
2294		 */
2295		if (prev_index != index || offset != prev_offset)
2296			mark_page_accessed(page);
2297		prev_index = index;
2298
2299		/*
2300		 * Ok, we have the page, and it's up-to-date, so
2301		 * now we can copy it to user space...
2302		 */
2303
2304		ret = copy_page_to_iter(page, offset, nr, iter);
2305		offset += ret;
2306		index += offset >> PAGE_SHIFT;
2307		offset &= ~PAGE_MASK;
2308		prev_offset = offset;
2309
2310		put_page(page);
2311		written += ret;
2312		if (!iov_iter_count(iter))
2313			goto out;
2314		if (ret < nr) {
2315			error = -EFAULT;
2316			goto out;
2317		}
2318		continue;
2319
2320page_not_up_to_date:
2321		/* Get exclusive access to the page ... */
2322		if (iocb->ki_flags & IOCB_WAITQ)
2323			error = lock_page_async(page, iocb->ki_waitq);
2324		else
2325			error = lock_page_killable(page);
2326		if (unlikely(error))
2327			goto readpage_error;
2328
2329page_not_up_to_date_locked:
2330		/* Did it get truncated before we got the lock? */
2331		if (!page->mapping) {
2332			unlock_page(page);
2333			put_page(page);
2334			continue;
2335		}
2336
2337		/* Did somebody else fill it already? */
2338		if (PageUptodate(page)) {
2339			unlock_page(page);
2340			goto page_ok;
2341		}
2342
2343readpage:
2344		if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT)) {
2345			unlock_page(page);
2346			put_page(page);
2347			goto would_block;
2348		}
2349		/*
2350		 * A previous I/O error may have been due to temporary
2351		 * failures, eg. multipath errors.
2352		 * PG_error will be set again if readpage fails.
2353		 */
2354		ClearPageError(page);
2355		/* Start the actual read. The read will unlock the page. */
2356		error = mapping->a_ops->readpage(filp, page);
2357
2358		if (unlikely(error)) {
2359			if (error == AOP_TRUNCATED_PAGE) {
2360				put_page(page);
2361				error = 0;
2362				goto find_page;
2363			}
2364			goto readpage_error;
2365		}
2366
2367		if (!PageUptodate(page)) {
2368			if (iocb->ki_flags & IOCB_WAITQ)
2369				error = lock_page_async(page, iocb->ki_waitq);
2370			else
2371				error = lock_page_killable(page);
2372
2373			if (unlikely(error))
2374				goto readpage_error;
2375			if (!PageUptodate(page)) {
2376				if (page->mapping == NULL) {
2377					/*
2378					 * invalidate_mapping_pages got it
2379					 */
2380					unlock_page(page);
2381					put_page(page);
2382					goto find_page;
2383				}
2384				unlock_page(page);
2385				shrink_readahead_size_eio(ra);
2386				error = -EIO;
2387				goto readpage_error;
2388			}
2389			unlock_page(page);
2390		}
2391
2392		goto page_ok;
2393
2394readpage_error:
2395		/* UHHUH! A synchronous read error occurred. Report it */
2396		put_page(page);
2397		goto out;
2398
2399no_cached_page:
2400		/*
2401		 * Ok, it wasn't cached, so we need to create a new
2402		 * page..
2403		 */
2404		page = page_cache_alloc(mapping);
2405		if (!page) {
2406			error = -ENOMEM;
2407			goto out;
2408		}
2409		error = add_to_page_cache_lru(page, mapping, index,
2410				mapping_gfp_constraint(mapping, GFP_KERNEL));
2411		if (error) {
2412			put_page(page);
2413			if (error == -EEXIST) {
2414				error = 0;
2415				goto find_page;
2416			}
2417			goto out;
2418		}
2419		goto readpage;
2420	}
2421
2422would_block:
2423	error = -EAGAIN;
2424out:
2425	ra->prev_pos = prev_index;
2426	ra->prev_pos <<= PAGE_SHIFT;
2427	ra->prev_pos |= prev_offset;
2428
2429	*ppos = ((loff_t)index << PAGE_SHIFT) + offset;
2430	file_accessed(filp);
2431	return written ? written : error;
2432}
2433EXPORT_SYMBOL_GPL(generic_file_buffered_read);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2434
2435/**
2436 * generic_file_read_iter - generic filesystem read routine
2437 * @iocb:	kernel I/O control block
2438 * @iter:	destination for the data read
 
 
2439 *
2440 * This is the "read_iter()" routine for all filesystems
2441 * that can use the page cache directly.
2442 *
2443 * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2444 * be returned when no data can be read without waiting for I/O requests
2445 * to complete; it doesn't prevent readahead.
2446 *
2447 * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2448 * requests shall be made for the read or for readahead.  When no data
2449 * can be read, -EAGAIN shall be returned.  When readahead would be
2450 * triggered, a partial, possibly empty read shall be returned.
2451 *
2452 * Return:
2453 * * number of bytes copied, even for partial reads
2454 * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2455 */
2456ssize_t
2457generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
 
2458{
2459	size_t count = iov_iter_count(iter);
2460	ssize_t retval = 0;
 
 
 
2461
2462	if (!count)
2463		goto out; /* skip atime */
 
 
 
2464
2465	if (iocb->ki_flags & IOCB_DIRECT) {
2466		struct file *file = iocb->ki_filp;
2467		struct address_space *mapping = file->f_mapping;
2468		struct inode *inode = mapping->host;
2469		loff_t size;
 
 
2470
 
 
 
 
2471		size = i_size_read(inode);
2472		if (iocb->ki_flags & IOCB_NOWAIT) {
2473			if (filemap_range_has_page(mapping, iocb->ki_pos,
2474						   iocb->ki_pos + count - 1))
2475				return -EAGAIN;
2476		} else {
2477			retval = filemap_write_and_wait_range(mapping,
2478						iocb->ki_pos,
2479					        iocb->ki_pos + count - 1);
2480			if (retval < 0)
2481				goto out;
2482		}
2483
2484		file_accessed(file);
2485
2486		retval = mapping->a_ops->direct_IO(iocb, iter);
2487		if (retval >= 0) {
2488			iocb->ki_pos += retval;
2489			count -= retval;
 
 
 
 
 
2490		}
2491		iov_iter_revert(iter, count - iov_iter_count(iter));
2492
2493		/*
2494		 * Btrfs can have a short DIO read if we encounter
2495		 * compressed extents, so if there was an error, or if
2496		 * we've already read everything we wanted to, or if
2497		 * there was a short read because we hit EOF, go ahead
2498		 * and return.  Otherwise fallthrough to buffered io for
2499		 * the rest of the read.  Buffered reads will not work for
2500		 * DAX files, so don't bother trying.
2501		 */
2502		if (retval < 0 || !count || iocb->ki_pos >= size ||
2503		    IS_DAX(inode))
2504			goto out;
 
2505	}
2506
2507	retval = generic_file_buffered_read(iocb, iter, retval);
2508out:
2509	return retval;
2510}
2511EXPORT_SYMBOL(generic_file_read_iter);
2512
2513#ifdef CONFIG_MMU
2514#define MMAP_LOTSAMISS  (100)
2515/*
2516 * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
2517 * @vmf - the vm_fault for this fault.
2518 * @page - the page to lock.
2519 * @fpin - the pointer to the file we may pin (or is already pinned).
2520 *
2521 * This works similar to lock_page_or_retry in that it can drop the mmap_lock.
2522 * It differs in that it actually returns the page locked if it returns 1 and 0
2523 * if it couldn't lock the page.  If we did have to drop the mmap_lock then fpin
2524 * will point to the pinned file and needs to be fput()'ed at a later point.
2525 */
2526static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page,
2527				     struct file **fpin)
2528{
2529	if (trylock_page(page))
2530		return 1;
 
2531
2532	/*
2533	 * NOTE! This will make us return with VM_FAULT_RETRY, but with
2534	 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
2535	 * is supposed to work. We have way too many special cases..
2536	 */
2537	if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
2538		return 0;
2539
2540	*fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
2541	if (vmf->flags & FAULT_FLAG_KILLABLE) {
2542		if (__lock_page_killable(page)) {
2543			/*
2544			 * We didn't have the right flags to drop the mmap_lock,
2545			 * but all fault_handlers only check for fatal signals
2546			 * if we return VM_FAULT_RETRY, so we need to drop the
2547			 * mmap_lock here and return 0 if we don't have a fpin.
2548			 */
2549			if (*fpin == NULL)
2550				mmap_read_unlock(vmf->vma->vm_mm);
2551			return 0;
2552		}
2553	} else
2554		__lock_page(page);
2555	return 1;
2556}
2557
 
2558
2559/*
2560 * Synchronous readahead happens when we don't even find a page in the page
2561 * cache at all.  We don't want to perform IO under the mmap sem, so if we have
2562 * to drop the mmap sem we return the file that was pinned in order for us to do
2563 * that.  If we didn't pin a file then we return NULL.  The file that is
2564 * returned needs to be fput()'ed when we're done with it.
2565 */
2566static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
 
 
 
2567{
2568	struct file *file = vmf->vma->vm_file;
2569	struct file_ra_state *ra = &file->f_ra;
2570	struct address_space *mapping = file->f_mapping;
2571	struct file *fpin = NULL;
2572	pgoff_t offset = vmf->pgoff;
2573	unsigned int mmap_miss;
2574
2575	/* If we don't want any read-ahead, don't bother */
2576	if (vmf->vma->vm_flags & VM_RAND_READ)
2577		return fpin;
2578	if (!ra->ra_pages)
2579		return fpin;
2580
2581	if (vmf->vma->vm_flags & VM_SEQ_READ) {
2582		fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2583		page_cache_sync_readahead(mapping, ra, file, offset,
2584					  ra->ra_pages);
2585		return fpin;
2586	}
2587
2588	/* Avoid banging the cache line if not needed */
2589	mmap_miss = READ_ONCE(ra->mmap_miss);
2590	if (mmap_miss < MMAP_LOTSAMISS * 10)
2591		WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
2592
2593	/*
2594	 * Do we miss much more than hit in this file? If so,
2595	 * stop bothering with read-ahead. It will only hurt.
2596	 */
2597	if (mmap_miss > MMAP_LOTSAMISS)
2598		return fpin;
2599
2600	/*
2601	 * mmap read-around
2602	 */
2603	fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2604	ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2605	ra->size = ra->ra_pages;
2606	ra->async_size = ra->ra_pages / 4;
2607	ra_submit(ra, mapping, file);
2608	return fpin;
2609}
2610
2611/*
2612 * Asynchronous readahead happens when we find the page and PG_readahead,
2613 * so we want to possibly extend the readahead further.  We return the file that
2614 * was pinned if we have to drop the mmap_lock in order to do IO.
2615 */
2616static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
2617					    struct page *page)
 
 
 
2618{
2619	struct file *file = vmf->vma->vm_file;
2620	struct file_ra_state *ra = &file->f_ra;
2621	struct address_space *mapping = file->f_mapping;
2622	struct file *fpin = NULL;
2623	unsigned int mmap_miss;
2624	pgoff_t offset = vmf->pgoff;
2625
2626	/* If we don't want any read-ahead, don't bother */
2627	if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
2628		return fpin;
2629	mmap_miss = READ_ONCE(ra->mmap_miss);
2630	if (mmap_miss)
2631		WRITE_ONCE(ra->mmap_miss, --mmap_miss);
2632	if (PageReadahead(page)) {
2633		fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2634		page_cache_async_readahead(mapping, ra, file,
2635					   page, offset, ra->ra_pages);
2636	}
2637	return fpin;
2638}
2639
2640/**
2641 * filemap_fault - read in file data for page fault handling
 
2642 * @vmf:	struct vm_fault containing details of the fault
2643 *
2644 * filemap_fault() is invoked via the vma operations vector for a
2645 * mapped memory region to read in file data during a page fault.
2646 *
2647 * The goto's are kind of ugly, but this streamlines the normal case of having
2648 * it in the page cache, and handles the special cases reasonably without
2649 * having a lot of duplicated code.
2650 *
2651 * vma->vm_mm->mmap_lock must be held on entry.
2652 *
2653 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
2654 * may be dropped before doing I/O or by lock_page_maybe_drop_mmap().
2655 *
2656 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
2657 * has not been released.
2658 *
2659 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2660 *
2661 * Return: bitwise-OR of %VM_FAULT_ codes.
2662 */
2663vm_fault_t filemap_fault(struct vm_fault *vmf)
2664{
2665	int error;
2666	struct file *file = vmf->vma->vm_file;
2667	struct file *fpin = NULL;
2668	struct address_space *mapping = file->f_mapping;
2669	struct file_ra_state *ra = &file->f_ra;
2670	struct inode *inode = mapping->host;
2671	pgoff_t offset = vmf->pgoff;
2672	pgoff_t max_off;
2673	struct page *page;
2674	vm_fault_t ret = 0;
 
2675
2676	max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2677	if (unlikely(offset >= max_off))
2678		return VM_FAULT_SIGBUS;
2679
2680	/*
2681	 * Do we have something in the page cache already?
2682	 */
2683	page = find_get_page(mapping, offset);
2684	if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2685		/*
2686		 * We found the page, so try async readahead before
2687		 * waiting for the lock.
2688		 */
2689		fpin = do_async_mmap_readahead(vmf, page);
2690	} else if (!page) {
2691		/* No page in the page cache at all */
 
2692		count_vm_event(PGMAJFAULT);
2693		count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
2694		ret = VM_FAULT_MAJOR;
2695		fpin = do_sync_mmap_readahead(vmf);
2696retry_find:
2697		page = pagecache_get_page(mapping, offset,
2698					  FGP_CREAT|FGP_FOR_MMAP,
2699					  vmf->gfp_mask);
2700		if (!page) {
2701			if (fpin)
2702				goto out_retry;
2703			return VM_FAULT_OOM;
2704		}
2705	}
2706
2707	if (!lock_page_maybe_drop_mmap(vmf, page, &fpin))
2708		goto out_retry;
 
 
2709
2710	/* Did it get truncated? */
2711	if (unlikely(compound_head(page)->mapping != mapping)) {
2712		unlock_page(page);
2713		put_page(page);
2714		goto retry_find;
2715	}
2716	VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
2717
2718	/*
2719	 * We have a locked page in the page cache, now we need to check
2720	 * that it's up-to-date. If not, it is going to be due to an error.
2721	 */
2722	if (unlikely(!PageUptodate(page)))
2723		goto page_not_uptodate;
2724
2725	/*
2726	 * We've made it this far and we had to drop our mmap_lock, now is the
2727	 * time to return to the upper layer and have it re-find the vma and
2728	 * redo the fault.
2729	 */
2730	if (fpin) {
2731		unlock_page(page);
2732		goto out_retry;
2733	}
2734
2735	/*
2736	 * Found the page and have a reference on it.
2737	 * We must recheck i_size under page lock.
2738	 */
2739	max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2740	if (unlikely(offset >= max_off)) {
2741		unlock_page(page);
2742		put_page(page);
2743		return VM_FAULT_SIGBUS;
2744	}
2745
2746	vmf->page = page;
2747	return ret | VM_FAULT_LOCKED;
2748
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2749page_not_uptodate:
2750	/*
2751	 * Umm, take care of errors if the page isn't up-to-date.
2752	 * Try to re-read it _once_. We do this synchronously,
2753	 * because there really aren't any performance issues here
2754	 * and we need to check for errors.
2755	 */
2756	ClearPageError(page);
2757	fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2758	error = mapping->a_ops->readpage(file, page);
2759	if (!error) {
2760		wait_on_page_locked(page);
2761		if (!PageUptodate(page))
2762			error = -EIO;
2763	}
2764	if (fpin)
2765		goto out_retry;
2766	put_page(page);
2767
2768	if (!error || error == AOP_TRUNCATED_PAGE)
2769		goto retry_find;
2770
2771	shrink_readahead_size_eio(ra);
 
2772	return VM_FAULT_SIGBUS;
2773
2774out_retry:
2775	/*
2776	 * We dropped the mmap_lock, we need to return to the fault handler to
2777	 * re-find the vma and come back and find our hopefully still populated
2778	 * page.
2779	 */
2780	if (page)
2781		put_page(page);
2782	if (fpin)
2783		fput(fpin);
2784	return ret | VM_FAULT_RETRY;
2785}
2786EXPORT_SYMBOL(filemap_fault);
2787
2788void filemap_map_pages(struct vm_fault *vmf,
2789		pgoff_t start_pgoff, pgoff_t end_pgoff)
2790{
2791	struct file *file = vmf->vma->vm_file;
 
 
2792	struct address_space *mapping = file->f_mapping;
2793	pgoff_t last_pgoff = start_pgoff;
2794	unsigned long max_idx;
2795	XA_STATE(xas, &mapping->i_pages, start_pgoff);
2796	struct page *page;
2797	unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss);
 
 
2798
2799	rcu_read_lock();
2800	xas_for_each(&xas, page, end_pgoff) {
2801		if (xas_retry(&xas, page))
2802			continue;
2803		if (xa_is_value(page))
 
 
2804			goto next;
 
 
 
 
 
 
2805
2806		/*
2807		 * Check for a locked page first, as a speculative
2808		 * reference may adversely influence page migration.
2809		 */
2810		if (PageLocked(page))
2811			goto next;
2812		if (!page_cache_get_speculative(page))
2813			goto next;
2814
2815		/* Has the page moved or been split? */
2816		if (unlikely(page != xas_reload(&xas)))
2817			goto skip;
2818		page = find_subpage(page, xas.xa_index);
 
2819
2820		if (!PageUptodate(page) ||
2821				PageReadahead(page) ||
2822				PageHWPoison(page))
2823			goto skip;
2824		if (!trylock_page(page))
2825			goto skip;
2826
2827		if (page->mapping != mapping || !PageUptodate(page))
2828			goto unlock;
2829
2830		max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
2831		if (page->index >= max_idx)
2832			goto unlock;
2833
2834		if (mmap_miss > 0)
2835			mmap_miss--;
2836
2837		vmf->address += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
2838		if (vmf->pte)
2839			vmf->pte += xas.xa_index - last_pgoff;
2840		last_pgoff = xas.xa_index;
2841		if (alloc_set_pte(vmf, page))
2842			goto unlock;
 
 
 
 
 
2843		unlock_page(page);
2844		goto next;
2845unlock:
2846		unlock_page(page);
2847skip:
2848		put_page(page);
2849next:
2850		/* Huge page is mapped? No need to proceed. */
2851		if (pmd_trans_huge(*vmf->pmd))
2852			break;
2853	}
2854	rcu_read_unlock();
2855	WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss);
2856}
2857EXPORT_SYMBOL(filemap_map_pages);
2858
2859vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
2860{
2861	struct page *page = vmf->page;
2862	struct inode *inode = file_inode(vmf->vma->vm_file);
2863	vm_fault_t ret = VM_FAULT_LOCKED;
2864
2865	sb_start_pagefault(inode->i_sb);
2866	file_update_time(vmf->vma->vm_file);
2867	lock_page(page);
2868	if (page->mapping != inode->i_mapping) {
2869		unlock_page(page);
2870		ret = VM_FAULT_NOPAGE;
2871		goto out;
2872	}
2873	/*
2874	 * We mark the page dirty already here so that when freeze is in
2875	 * progress, we are guaranteed that writeback during freezing will
2876	 * see the dirty page and writeprotect it again.
2877	 */
2878	set_page_dirty(page);
2879	wait_for_stable_page(page);
2880out:
2881	sb_end_pagefault(inode->i_sb);
2882	return ret;
2883}
 
2884
2885const struct vm_operations_struct generic_file_vm_ops = {
2886	.fault		= filemap_fault,
2887	.map_pages	= filemap_map_pages,
2888	.page_mkwrite	= filemap_page_mkwrite,
 
2889};
2890
2891/* This is used for a general mmap of a disk file */
2892
2893int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2894{
2895	struct address_space *mapping = file->f_mapping;
2896
2897	if (!mapping->a_ops->readpage)
2898		return -ENOEXEC;
2899	file_accessed(file);
2900	vma->vm_ops = &generic_file_vm_ops;
2901	return 0;
2902}
2903
2904/*
2905 * This is for filesystems which do not implement ->writepage.
2906 */
2907int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2908{
2909	if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2910		return -EINVAL;
2911	return generic_file_mmap(file, vma);
2912}
2913#else
2914vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
2915{
2916	return VM_FAULT_SIGBUS;
2917}
2918int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2919{
2920	return -ENOSYS;
2921}
2922int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2923{
2924	return -ENOSYS;
2925}
2926#endif /* CONFIG_MMU */
2927
2928EXPORT_SYMBOL(filemap_page_mkwrite);
2929EXPORT_SYMBOL(generic_file_mmap);
2930EXPORT_SYMBOL(generic_file_readonly_mmap);
2931
2932static struct page *wait_on_page_read(struct page *page)
2933{
2934	if (!IS_ERR(page)) {
2935		wait_on_page_locked(page);
2936		if (!PageUptodate(page)) {
2937			put_page(page);
2938			page = ERR_PTR(-EIO);
2939		}
2940	}
2941	return page;
2942}
2943
2944static struct page *do_read_cache_page(struct address_space *mapping,
2945				pgoff_t index,
2946				int (*filler)(void *, struct page *),
2947				void *data,
2948				gfp_t gfp)
2949{
2950	struct page *page;
2951	int err;
2952repeat:
2953	page = find_get_page(mapping, index);
2954	if (!page) {
2955		page = __page_cache_alloc(gfp);
2956		if (!page)
2957			return ERR_PTR(-ENOMEM);
2958		err = add_to_page_cache_lru(page, mapping, index, gfp);
2959		if (unlikely(err)) {
2960			put_page(page);
2961			if (err == -EEXIST)
2962				goto repeat;
2963			/* Presumably ENOMEM for xarray node */
2964			return ERR_PTR(err);
2965		}
2966
2967filler:
2968		if (filler)
2969			err = filler(data, page);
2970		else
2971			err = mapping->a_ops->readpage(data, page);
2972
2973		if (err < 0) {
2974			put_page(page);
2975			return ERR_PTR(err);
 
 
2976		}
2977
2978		page = wait_on_page_read(page);
2979		if (IS_ERR(page))
2980			return page;
2981		goto out;
2982	}
2983	if (PageUptodate(page))
2984		goto out;
2985
2986	/*
2987	 * Page is not up to date and may be locked due one of the following
2988	 * case a: Page is being filled and the page lock is held
2989	 * case b: Read/write error clearing the page uptodate status
2990	 * case c: Truncation in progress (page locked)
2991	 * case d: Reclaim in progress
2992	 *
2993	 * Case a, the page will be up to date when the page is unlocked.
2994	 *    There is no need to serialise on the page lock here as the page
2995	 *    is pinned so the lock gives no additional protection. Even if the
2996	 *    page is truncated, the data is still valid if PageUptodate as
2997	 *    it's a race vs truncate race.
2998	 * Case b, the page will not be up to date
2999	 * Case c, the page may be truncated but in itself, the data may still
3000	 *    be valid after IO completes as it's a read vs truncate race. The
3001	 *    operation must restart if the page is not uptodate on unlock but
3002	 *    otherwise serialising on page lock to stabilise the mapping gives
3003	 *    no additional guarantees to the caller as the page lock is
3004	 *    released before return.
3005	 * Case d, similar to truncation. If reclaim holds the page lock, it
3006	 *    will be a race with remove_mapping that determines if the mapping
3007	 *    is valid on unlock but otherwise the data is valid and there is
3008	 *    no need to serialise with page lock.
3009	 *
3010	 * As the page lock gives no additional guarantee, we optimistically
3011	 * wait on the page to be unlocked and check if it's up to date and
3012	 * use the page if it is. Otherwise, the page lock is required to
3013	 * distinguish between the different cases. The motivation is that we
3014	 * avoid spurious serialisations and wakeups when multiple processes
3015	 * wait on the same page for IO to complete.
3016	 */
3017	wait_on_page_locked(page);
3018	if (PageUptodate(page))
3019		goto out;
3020
3021	/* Distinguish between all the cases under the safety of the lock */
3022	lock_page(page);
3023
3024	/* Case c or d, restart the operation */
3025	if (!page->mapping) {
3026		unlock_page(page);
3027		put_page(page);
3028		goto repeat;
3029	}
3030
3031	/* Someone else locked and filled the page in a very small window */
3032	if (PageUptodate(page)) {
3033		unlock_page(page);
3034		goto out;
3035	}
3036
3037	/*
3038	 * A previous I/O error may have been due to temporary
3039	 * failures.
3040	 * Clear page error before actual read, PG_error will be
3041	 * set again if read page fails.
3042	 */
3043	ClearPageError(page);
3044	goto filler;
3045
3046out:
3047	mark_page_accessed(page);
3048	return page;
3049}
3050
3051/**
3052 * read_cache_page - read into page cache, fill it if needed
3053 * @mapping:	the page's address_space
3054 * @index:	the page index
3055 * @filler:	function to perform the read
3056 * @data:	first arg to filler(data, page) function, often left as NULL
3057 *
3058 * Read into the page cache. If a page already exists, and PageUptodate() is
3059 * not set, try to fill the page and wait for it to become unlocked.
3060 *
3061 * If the page does not get brought uptodate, return -EIO.
3062 *
3063 * Return: up to date page on success, ERR_PTR() on failure.
3064 */
3065struct page *read_cache_page(struct address_space *mapping,
3066				pgoff_t index,
3067				int (*filler)(void *, struct page *),
3068				void *data)
3069{
3070	return do_read_cache_page(mapping, index, filler, data,
3071			mapping_gfp_mask(mapping));
3072}
3073EXPORT_SYMBOL(read_cache_page);
3074
3075/**
3076 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3077 * @mapping:	the page's address_space
3078 * @index:	the page index
3079 * @gfp:	the page allocator flags to use if allocating
3080 *
3081 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3082 * any new page allocations done using the specified allocation flags.
3083 *
3084 * If the page does not get brought uptodate, return -EIO.
3085 *
3086 * Return: up to date page on success, ERR_PTR() on failure.
3087 */
3088struct page *read_cache_page_gfp(struct address_space *mapping,
3089				pgoff_t index,
3090				gfp_t gfp)
3091{
3092	return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3093}
3094EXPORT_SYMBOL(read_cache_page_gfp);
3095
3096/*
3097 * Don't operate on ranges the page cache doesn't support, and don't exceed the
3098 * LFS limits.  If pos is under the limit it becomes a short access.  If it
3099 * exceeds the limit we return -EFBIG.
3100 */
3101static int generic_write_check_limits(struct file *file, loff_t pos,
3102				      loff_t *count)
3103{
3104	struct inode *inode = file->f_mapping->host;
3105	loff_t max_size = inode->i_sb->s_maxbytes;
3106	loff_t limit = rlimit(RLIMIT_FSIZE);
3107
3108	if (limit != RLIM_INFINITY) {
3109		if (pos >= limit) {
3110			send_sig(SIGXFSZ, current, 0);
3111			return -EFBIG;
3112		}
3113		*count = min(*count, limit - pos);
3114	}
3115
3116	if (!(file->f_flags & O_LARGEFILE))
3117		max_size = MAX_NON_LFS;
3118
3119	if (unlikely(pos >= max_size))
3120		return -EFBIG;
3121
3122	*count = min(*count, max_size - pos);
3123
3124	return 0;
3125}
 
3126
3127/*
3128 * Performs necessary checks before doing a write
3129 *
3130 * Can adjust writing position or amount of bytes to write.
3131 * Returns appropriate error code that caller should return or
3132 * zero in case that write should be allowed.
3133 */
3134inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
3135{
3136	struct file *file = iocb->ki_filp;
3137	struct inode *inode = file->f_mapping->host;
3138	loff_t count;
3139	int ret;
3140
3141	if (IS_SWAPFILE(inode))
3142		return -ETXTBSY;
3143
3144	if (!iov_iter_count(from))
3145		return 0;
3146
3147	/* FIXME: this is for backwards compatibility with 2.4 */
3148	if (iocb->ki_flags & IOCB_APPEND)
3149		iocb->ki_pos = i_size_read(inode);
3150
3151	if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT))
3152		return -EINVAL;
3153
3154	count = iov_iter_count(from);
3155	ret = generic_write_check_limits(file, iocb->ki_pos, &count);
3156	if (ret)
3157		return ret;
3158
3159	iov_iter_truncate(from, count);
3160	return iov_iter_count(from);
3161}
3162EXPORT_SYMBOL(generic_write_checks);
3163
3164/*
3165 * Performs necessary checks before doing a clone.
3166 *
3167 * Can adjust amount of bytes to clone via @req_count argument.
3168 * Returns appropriate error code that caller should return or
3169 * zero in case the clone should be allowed.
3170 */
3171int generic_remap_checks(struct file *file_in, loff_t pos_in,
3172			 struct file *file_out, loff_t pos_out,
3173			 loff_t *req_count, unsigned int remap_flags)
3174{
3175	struct inode *inode_in = file_in->f_mapping->host;
3176	struct inode *inode_out = file_out->f_mapping->host;
3177	uint64_t count = *req_count;
3178	uint64_t bcount;
3179	loff_t size_in, size_out;
3180	loff_t bs = inode_out->i_sb->s_blocksize;
3181	int ret;
3182
3183	/* The start of both ranges must be aligned to an fs block. */
3184	if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs))
3185		return -EINVAL;
3186
3187	/* Ensure offsets don't wrap. */
3188	if (pos_in + count < pos_in || pos_out + count < pos_out)
3189		return -EINVAL;
3190
3191	size_in = i_size_read(inode_in);
3192	size_out = i_size_read(inode_out);
3193
3194	/* Dedupe requires both ranges to be within EOF. */
3195	if ((remap_flags & REMAP_FILE_DEDUP) &&
3196	    (pos_in >= size_in || pos_in + count > size_in ||
3197	     pos_out >= size_out || pos_out + count > size_out))
3198		return -EINVAL;
3199
3200	/* Ensure the infile range is within the infile. */
3201	if (pos_in >= size_in)
3202		return -EINVAL;
3203	count = min(count, size_in - (uint64_t)pos_in);
3204
3205	ret = generic_write_check_limits(file_out, pos_out, &count);
3206	if (ret)
3207		return ret;
3208
3209	/*
3210	 * If the user wanted us to link to the infile's EOF, round up to the
3211	 * next block boundary for this check.
3212	 *
3213	 * Otherwise, make sure the count is also block-aligned, having
3214	 * already confirmed the starting offsets' block alignment.
3215	 */
3216	if (pos_in + count == size_in) {
3217		bcount = ALIGN(size_in, bs) - pos_in;
3218	} else {
3219		if (!IS_ALIGNED(count, bs))
3220			count = ALIGN_DOWN(count, bs);
3221		bcount = count;
 
 
3222	}
3223
3224	/* Don't allow overlapped cloning within the same file. */
3225	if (inode_in == inode_out &&
3226	    pos_out + bcount > pos_in &&
3227	    pos_out < pos_in + bcount)
3228		return -EINVAL;
3229
3230	/*
3231	 * We shortened the request but the caller can't deal with that, so
3232	 * bounce the request back to userspace.
3233	 */
3234	if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN))
3235		return -EINVAL;
3236
3237	*req_count = count;
3238	return 0;
3239}
3240
3241
3242/*
3243 * Performs common checks before doing a file copy/clone
3244 * from @file_in to @file_out.
3245 */
3246int generic_file_rw_checks(struct file *file_in, struct file *file_out)
3247{
3248	struct inode *inode_in = file_inode(file_in);
3249	struct inode *inode_out = file_inode(file_out);
3250
3251	/* Don't copy dirs, pipes, sockets... */
3252	if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
3253		return -EISDIR;
3254	if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
3255		return -EINVAL;
3256
3257	if (!(file_in->f_mode & FMODE_READ) ||
3258	    !(file_out->f_mode & FMODE_WRITE) ||
3259	    (file_out->f_flags & O_APPEND))
3260		return -EBADF;
3261
3262	return 0;
3263}
3264
3265/*
3266 * Performs necessary checks before doing a file copy
3267 *
3268 * Can adjust amount of bytes to copy via @req_count argument.
3269 * Returns appropriate error code that caller should return or
3270 * zero in case the copy should be allowed.
3271 */
3272int generic_copy_file_checks(struct file *file_in, loff_t pos_in,
3273			     struct file *file_out, loff_t pos_out,
3274			     size_t *req_count, unsigned int flags)
3275{
3276	struct inode *inode_in = file_inode(file_in);
3277	struct inode *inode_out = file_inode(file_out);
3278	uint64_t count = *req_count;
3279	loff_t size_in;
3280	int ret;
3281
3282	ret = generic_file_rw_checks(file_in, file_out);
3283	if (ret)
3284		return ret;
 
 
 
 
 
 
 
 
 
3285
3286	/* Don't touch certain kinds of inodes */
3287	if (IS_IMMUTABLE(inode_out))
 
3288		return -EPERM;
3289
3290	if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
3291		return -ETXTBSY;
3292
3293	/* Ensure offsets don't wrap. */
3294	if (pos_in + count < pos_in || pos_out + count < pos_out)
3295		return -EOVERFLOW;
3296
3297	/* Shorten the copy to EOF */
3298	size_in = i_size_read(inode_in);
3299	if (pos_in >= size_in)
3300		count = 0;
3301	else
3302		count = min(count, size_in - (uint64_t)pos_in);
3303
3304	ret = generic_write_check_limits(file_out, pos_out, &count);
3305	if (ret)
3306		return ret;
3307
3308	/* Don't allow overlapped copying within the same file. */
3309	if (inode_in == inode_out &&
3310	    pos_out + count > pos_in &&
3311	    pos_out < pos_in + count)
3312		return -EINVAL;
3313
3314	*req_count = count;
3315	return 0;
3316}
 
3317
3318int pagecache_write_begin(struct file *file, struct address_space *mapping,
3319				loff_t pos, unsigned len, unsigned flags,
3320				struct page **pagep, void **fsdata)
3321{
3322	const struct address_space_operations *aops = mapping->a_ops;
3323
3324	return aops->write_begin(file, mapping, pos, len, flags,
3325							pagep, fsdata);
3326}
3327EXPORT_SYMBOL(pagecache_write_begin);
3328
3329int pagecache_write_end(struct file *file, struct address_space *mapping,
3330				loff_t pos, unsigned len, unsigned copied,
3331				struct page *page, void *fsdata)
3332{
3333	const struct address_space_operations *aops = mapping->a_ops;
3334
 
3335	return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
3336}
3337EXPORT_SYMBOL(pagecache_write_end);
3338
3339/*
3340 * Warn about a page cache invalidation failure during a direct I/O write.
3341 */
3342void dio_warn_stale_pagecache(struct file *filp)
3343{
3344	static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3345	char pathname[128];
3346	struct inode *inode = file_inode(filp);
3347	char *path;
3348
3349	errseq_set(&inode->i_mapping->wb_err, -EIO);
3350	if (__ratelimit(&_rs)) {
3351		path = file_path(filp, pathname, sizeof(pathname));
3352		if (IS_ERR(path))
3353			path = "(unknown)";
3354		pr_crit("Page cache invalidation failure on direct I/O.  Possible data corruption due to collision with buffered I/O!\n");
3355		pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3356			current->comm);
3357	}
3358}
3359
3360ssize_t
3361generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
 
 
3362{
3363	struct file	*file = iocb->ki_filp;
3364	struct address_space *mapping = file->f_mapping;
3365	struct inode	*inode = mapping->host;
3366	loff_t		pos = iocb->ki_pos;
3367	ssize_t		written;
3368	size_t		write_len;
3369	pgoff_t		end;
3370
3371	write_len = iov_iter_count(from);
3372	end = (pos + write_len - 1) >> PAGE_SHIFT;
3373
3374	if (iocb->ki_flags & IOCB_NOWAIT) {
3375		/* If there are pages to writeback, return */
3376		if (filemap_range_has_page(inode->i_mapping, pos,
3377					   pos + write_len - 1))
3378			return -EAGAIN;
3379	} else {
3380		written = filemap_write_and_wait_range(mapping, pos,
3381							pos + write_len - 1);
3382		if (written)
3383			goto out;
3384	}
3385
3386	/*
3387	 * After a write we want buffered reads to be sure to go to disk to get
3388	 * the new data.  We invalidate clean cached page from the region we're
3389	 * about to write.  We do this *before* the write so that we can return
3390	 * without clobbering -EIOCBQUEUED from ->direct_IO().
3391	 */
3392	written = invalidate_inode_pages2_range(mapping,
3393					pos >> PAGE_SHIFT, end);
3394	/*
3395	 * If a page can not be invalidated, return 0 to fall back
3396	 * to buffered write.
3397	 */
3398	if (written) {
3399		if (written == -EBUSY)
3400			return 0;
3401		goto out;
 
 
3402	}
3403
3404	written = mapping->a_ops->direct_IO(iocb, from);
3405
3406	/*
3407	 * Finally, try again to invalidate clean pages which might have been
3408	 * cached by non-direct readahead, or faulted in by get_user_pages()
3409	 * if the source of the write was an mmap'ed region of the file
3410	 * we're writing.  Either one is a pretty crazy thing to do,
3411	 * so we don't support it 100%.  If this invalidation
3412	 * fails, tough, the write still worked...
3413	 *
3414	 * Most of the time we do not need this since dio_complete() will do
3415	 * the invalidation for us. However there are some file systems that
3416	 * do not end up with dio_complete() being called, so let's not break
3417	 * them by removing it completely.
3418	 *
3419	 * Noticeable example is a blkdev_direct_IO().
3420	 *
3421	 * Skip invalidation for async writes or if mapping has no pages.
3422	 */
3423	if (written > 0 && mapping->nrpages &&
3424	    invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end))
3425		dio_warn_stale_pagecache(file);
 
3426
3427	if (written > 0) {
3428		pos += written;
3429		write_len -= written;
3430		if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3431			i_size_write(inode, pos);
3432			mark_inode_dirty(inode);
3433		}
3434		iocb->ki_pos = pos;
3435	}
3436	iov_iter_revert(from, write_len - iov_iter_count(from));
3437out:
3438	return written;
3439}
3440EXPORT_SYMBOL(generic_file_direct_write);
3441
3442/*
3443 * Find or create a page at the given pagecache position. Return the locked
3444 * page. This function is specifically for buffered writes.
3445 */
3446struct page *grab_cache_page_write_begin(struct address_space *mapping,
3447					pgoff_t index, unsigned flags)
3448{
 
 
3449	struct page *page;
3450	int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
3451
 
 
 
3452	if (flags & AOP_FLAG_NOFS)
3453		fgp_flags |= FGP_NOFS;
3454
3455	page = pagecache_get_page(mapping, index, fgp_flags,
3456			mapping_gfp_mask(mapping));
3457	if (page)
3458		wait_for_stable_page(page);
3459
 
 
 
 
 
 
 
 
 
 
 
 
 
3460	return page;
3461}
3462EXPORT_SYMBOL(grab_cache_page_write_begin);
3463
3464ssize_t generic_perform_write(struct file *file,
3465				struct iov_iter *i, loff_t pos)
3466{
3467	struct address_space *mapping = file->f_mapping;
3468	const struct address_space_operations *a_ops = mapping->a_ops;
3469	long status = 0;
3470	ssize_t written = 0;
3471	unsigned int flags = 0;
3472
 
 
 
 
 
 
3473	do {
3474		struct page *page;
3475		unsigned long offset;	/* Offset into pagecache page */
3476		unsigned long bytes;	/* Bytes to write to page */
3477		size_t copied;		/* Bytes copied from user */
3478		void *fsdata;
3479
3480		offset = (pos & (PAGE_SIZE - 1));
3481		bytes = min_t(unsigned long, PAGE_SIZE - offset,
3482						iov_iter_count(i));
3483
3484again:
3485		/*
3486		 * Bring in the user page that we will copy from _first_.
3487		 * Otherwise there's a nasty deadlock on copying from the
3488		 * same page as we're writing to, without it being marked
3489		 * up-to-date.
3490		 *
3491		 * Not only is this an optimisation, but it is also required
3492		 * to check that the address is actually valid, when atomic
3493		 * usercopies are used, below.
3494		 */
3495		if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
3496			status = -EFAULT;
3497			break;
3498		}
3499
3500		if (fatal_signal_pending(current)) {
3501			status = -EINTR;
3502			break;
3503		}
3504
3505		status = a_ops->write_begin(file, mapping, pos, bytes, flags,
3506						&page, &fsdata);
3507		if (unlikely(status < 0))
3508			break;
3509
3510		if (mapping_writably_mapped(mapping))
3511			flush_dcache_page(page);
3512
3513		copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
3514		flush_dcache_page(page);
3515
 
3516		status = a_ops->write_end(file, mapping, pos, bytes, copied,
3517						page, fsdata);
3518		if (unlikely(status < 0))
3519			break;
3520		copied = status;
3521
3522		cond_resched();
3523
3524		iov_iter_advance(i, copied);
3525		if (unlikely(copied == 0)) {
3526			/*
3527			 * If we were unable to copy any data at all, we must
3528			 * fall back to a single segment length write.
3529			 *
3530			 * If we didn't fallback here, we could livelock
3531			 * because not all segments in the iov can be copied at
3532			 * once without a pagefault.
3533			 */
3534			bytes = min_t(unsigned long, PAGE_SIZE - offset,
3535						iov_iter_single_seg_count(i));
3536			goto again;
3537		}
3538		pos += copied;
3539		written += copied;
3540
3541		balance_dirty_pages_ratelimited(mapping);
 
 
 
 
3542	} while (iov_iter_count(i));
3543
3544	return written ? written : status;
3545}
3546EXPORT_SYMBOL(generic_perform_write);
3547
3548/**
3549 * __generic_file_write_iter - write data to a file
3550 * @iocb:	IO state structure (file, offset, etc.)
3551 * @from:	iov_iter with data to write
 
3552 *
3553 * This function does all the work needed for actually writing data to a
3554 * file. It does all basic checks, removes SUID from the file, updates
3555 * modification times and calls proper subroutines depending on whether we
3556 * do direct IO or a standard buffered write.
3557 *
3558 * It expects i_mutex to be grabbed unless we work on a block device or similar
3559 * object which does not need locking at all.
3560 *
3561 * This function does *not* take care of syncing data in case of O_SYNC write.
3562 * A caller has to handle it. This is mainly due to the fact that we want to
3563 * avoid syncing under i_mutex.
3564 *
3565 * Return:
3566 * * number of bytes written, even for truncated writes
3567 * * negative error code if no data has been written at all
3568 */
3569ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 
3570{
3571	struct file *file = iocb->ki_filp;
3572	struct address_space * mapping = file->f_mapping;
 
 
3573	struct inode 	*inode = mapping->host;
 
3574	ssize_t		written = 0;
3575	ssize_t		err;
3576	ssize_t		status;
 
 
 
 
 
 
 
 
3577
3578	/* We can write back this queue in page reclaim */
3579	current->backing_dev_info = inode_to_bdi(inode);
3580	err = file_remove_privs(file);
 
 
 
 
 
 
 
3581	if (err)
3582		goto out;
3583
3584	err = file_update_time(file);
3585	if (err)
3586		goto out;
3587
3588	if (iocb->ki_flags & IOCB_DIRECT) {
3589		loff_t pos, endbyte;
 
 
 
 
 
 
 
 
 
3590
3591		written = generic_file_direct_write(iocb, from);
3592		/*
3593		 * If the write stopped short of completing, fall back to
3594		 * buffered writes.  Some filesystems do this for writes to
3595		 * holes, for example.  For DAX files, a buffered write will
3596		 * not succeed (even if it did, DAX does not handle dirty
3597		 * page-cache pages correctly).
3598		 */
3599		if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3600			goto out;
3601
3602		status = generic_perform_write(file, from, pos = iocb->ki_pos);
3603		/*
3604		 * If generic_perform_write() returned a synchronous error
3605		 * then we want to return the number of bytes which were
3606		 * direct-written, or the error code if that was zero.  Note
3607		 * that this differs from normal direct-io semantics, which
3608		 * will return -EFOO even if some bytes were written.
3609		 */
3610		if (unlikely(status < 0)) {
3611			err = status;
3612			goto out;
3613		}
 
3614		/*
3615		 * We need to ensure that the page cache pages are written to
3616		 * disk and invalidated to preserve the expected O_DIRECT
3617		 * semantics.
3618		 */
3619		endbyte = pos + status - 1;
3620		err = filemap_write_and_wait_range(mapping, pos, endbyte);
3621		if (err == 0) {
3622			iocb->ki_pos = endbyte + 1;
3623			written += status;
3624			invalidate_mapping_pages(mapping,
3625						 pos >> PAGE_SHIFT,
3626						 endbyte >> PAGE_SHIFT);
3627		} else {
3628			/*
3629			 * We don't know how much we wrote, so just return
3630			 * the number of bytes which were direct-written
3631			 */
3632		}
3633	} else {
3634		written = generic_perform_write(file, from, iocb->ki_pos);
3635		if (likely(written > 0))
3636			iocb->ki_pos += written;
3637	}
3638out:
3639	current->backing_dev_info = NULL;
3640	return written ? written : err;
3641}
3642EXPORT_SYMBOL(__generic_file_write_iter);
3643
3644/**
3645 * generic_file_write_iter - write data to a file
3646 * @iocb:	IO state structure
3647 * @from:	iov_iter with data to write
 
 
3648 *
3649 * This is a wrapper around __generic_file_write_iter() to be used by most
3650 * filesystems. It takes care of syncing the file in case of O_SYNC file
3651 * and acquires i_mutex as needed.
3652 * Return:
3653 * * negative error code if no data has been written at all of
3654 *   vfs_fsync_range() failed for a synchronous write
3655 * * number of bytes written, even for truncated writes
3656 */
3657ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
 
3658{
3659	struct file *file = iocb->ki_filp;
3660	struct inode *inode = file->f_mapping->host;
3661	ssize_t ret;
3662
3663	inode_lock(inode);
3664	ret = generic_write_checks(iocb, from);
3665	if (ret > 0)
3666		ret = __generic_file_write_iter(iocb, from);
3667	inode_unlock(inode);
3668
3669	if (ret > 0)
3670		ret = generic_write_sync(iocb, ret);
 
 
 
 
 
 
 
 
 
3671	return ret;
3672}
3673EXPORT_SYMBOL(generic_file_write_iter);
3674
3675/**
3676 * try_to_release_page() - release old fs-specific metadata on a page
3677 *
3678 * @page: the page which the kernel is trying to free
3679 * @gfp_mask: memory allocation flags (and I/O mode)
3680 *
3681 * The address_space is to try to release any data against the page
3682 * (presumably at page->private).
 
3683 *
3684 * This may also be called if PG_fscache is set on a page, indicating that the
3685 * page is known to the local caching routines.
3686 *
3687 * The @gfp_mask argument specifies whether I/O may be performed to release
3688 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3689 *
3690 * Return: %1 if the release was successful, otherwise return zero.
3691 */
3692int try_to_release_page(struct page *page, gfp_t gfp_mask)
3693{
3694	struct address_space * const mapping = page->mapping;
3695
3696	BUG_ON(!PageLocked(page));
3697	if (PageWriteback(page))
3698		return 0;
3699
3700	if (mapping && mapping->a_ops->releasepage)
3701		return mapping->a_ops->releasepage(page, gfp_mask);
3702	return try_to_free_buffers(page);
3703}
3704
3705EXPORT_SYMBOL(try_to_release_page);
v3.15
 
   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 <linux/rmap.h>
 
 
 
 
  37#include "internal.h"
  38
  39#define CREATE_TRACE_POINTS
  40#include <trace/events/filemap.h>
  41
  42/*
  43 * FIXME: remove all knowledge of the buffer layer from the core VM
  44 */
  45#include <linux/buffer_head.h> /* for try_to_free_buffers */
  46
  47#include <asm/mman.h>
  48
  49/*
  50 * Shared mappings implemented 30.11.1994. It's not fully working yet,
  51 * though.
  52 *
  53 * Shared mappings now work. 15.8.1995  Bruno.
  54 *
  55 * finished 'unifying' the page and buffer cache and SMP-threaded the
  56 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
  57 *
  58 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
  59 */
  60
  61/*
  62 * Lock ordering:
  63 *
  64 *  ->i_mmap_mutex		(truncate_pagecache)
  65 *    ->private_lock		(__free_pte->__set_page_dirty_buffers)
  66 *      ->swap_lock		(exclusive_swap_page, others)
  67 *        ->mapping->tree_lock
  68 *
  69 *  ->i_mutex
  70 *    ->i_mmap_mutex		(truncate->unmap_mapping_range)
  71 *
  72 *  ->mmap_sem
  73 *    ->i_mmap_mutex
  74 *      ->page_table_lock or pte_lock	(various, mainly in memory.c)
  75 *        ->mapping->tree_lock	(arch-dependent flush_dcache_mmap_lock)
  76 *
  77 *  ->mmap_sem
  78 *    ->lock_page		(access_process_vm)
  79 *
  80 *  ->i_mutex			(generic_perform_write)
  81 *    ->mmap_sem		(fault_in_pages_readable->do_page_fault)
  82 *
  83 *  bdi->wb.list_lock
  84 *    sb_lock			(fs/fs-writeback.c)
  85 *    ->mapping->tree_lock	(__sync_single_inode)
  86 *
  87 *  ->i_mmap_mutex
  88 *    ->anon_vma.lock		(vma_adjust)
  89 *
  90 *  ->anon_vma.lock
  91 *    ->page_table_lock or pte_lock	(anon_vma_prepare and various)
  92 *
  93 *  ->page_table_lock or pte_lock
  94 *    ->swap_lock		(try_to_unmap_one)
  95 *    ->private_lock		(try_to_unmap_one)
  96 *    ->tree_lock		(try_to_unmap_one)
  97 *    ->zone.lru_lock		(follow_page->mark_page_accessed)
  98 *    ->zone.lru_lock		(check_pte_range->isolate_lru_page)
  99 *    ->private_lock		(page_remove_rmap->set_page_dirty)
 100 *    ->tree_lock		(page_remove_rmap->set_page_dirty)
 101 *    bdi.wb->list_lock		(page_remove_rmap->set_page_dirty)
 102 *    ->inode->i_lock		(page_remove_rmap->set_page_dirty)
 
 103 *    bdi.wb->list_lock		(zap_pte_range->set_page_dirty)
 104 *    ->inode->i_lock		(zap_pte_range->set_page_dirty)
 105 *    ->private_lock		(zap_pte_range->__set_page_dirty_buffers)
 106 *
 107 * ->i_mmap_mutex
 108 *   ->tasklist_lock            (memory_failure, collect_procs_ao)
 109 */
 110
 111static void page_cache_tree_delete(struct address_space *mapping,
 112				   struct page *page, void *shadow)
 113{
 114	struct radix_tree_node *node;
 115	unsigned long index;
 116	unsigned int offset;
 117	unsigned int tag;
 118	void **slot;
 119
 120	VM_BUG_ON(!PageLocked(page));
 121
 122	__radix_tree_lookup(&mapping->page_tree, page->index, &node, &slot);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 123
 124	if (shadow) {
 125		mapping->nrshadows++;
 126		/*
 127		 * Make sure the nrshadows update is committed before
 128		 * the nrpages update so that final truncate racing
 129		 * with reclaim does not see both counters 0 at the
 130		 * same time and miss a shadow entry.
 131		 */
 132		smp_wmb();
 133	}
 134	mapping->nrpages--;
 
 135
 136	if (!node) {
 137		/* Clear direct pointer tags in root node */
 138		mapping->page_tree.gfp_mask &= __GFP_BITS_MASK;
 139		radix_tree_replace_slot(slot, shadow);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 140		return;
 141	}
 142
 143	/* Clear tree tags for the removed page */
 144	index = page->index;
 145	offset = index & RADIX_TREE_MAP_MASK;
 146	for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
 147		if (test_bit(offset, node->tags[tag]))
 148			radix_tree_tag_clear(&mapping->page_tree, index, tag);
 
 
 
 
 149	}
 150
 151	/* Delete page, swap shadow entry */
 152	radix_tree_replace_slot(slot, shadow);
 153	workingset_node_pages_dec(node);
 154	if (shadow)
 155		workingset_node_shadows_inc(node);
 156	else
 157		if (__radix_tree_delete_node(&mapping->page_tree, node))
 158			return;
 159
 160	/*
 161	 * Track node that only contains shadow entries.
 
 
 162	 *
 163	 * Avoid acquiring the list_lru lock if already tracked.  The
 164	 * list_empty() test is safe as node->private_list is
 165	 * protected by mapping->tree_lock.
 166	 */
 167	if (!workingset_node_pages(node) &&
 168	    list_empty(&node->private_list)) {
 169		node->private_data = mapping;
 170		list_lru_add(&workingset_shadow_nodes, &node->private_list);
 171	}
 172}
 173
 174/*
 175 * Delete a page from the page cache and free it. Caller has to make
 176 * sure the page is locked and that nobody else uses it - or that usage
 177 * is safe.  The caller must hold the mapping's tree_lock.
 178 */
 179void __delete_from_page_cache(struct page *page, void *shadow)
 180{
 181	struct address_space *mapping = page->mapping;
 182
 183	trace_mm_filemap_delete_from_page_cache(page);
 184	/*
 185	 * if we're uptodate, flush out into the cleancache, otherwise
 186	 * invalidate any existing cleancache entries.  We can't leave
 187	 * stale data around in the cleancache once our page is gone
 188	 */
 189	if (PageUptodate(page) && PageMappedToDisk(page))
 190		cleancache_put_page(page);
 191	else
 192		cleancache_invalidate_page(mapping, page);
 193
 194	page_cache_tree_delete(mapping, page, shadow);
 
 
 195
 196	page->mapping = NULL;
 197	/* Leave page->index set: truncation lookup relies upon it */
 
 
 198
 199	__dec_zone_page_state(page, NR_FILE_PAGES);
 200	if (PageSwapBacked(page))
 201		__dec_zone_page_state(page, NR_SHMEM);
 202	BUG_ON(page_mapped(page));
 203
 204	/*
 205	 * Some filesystems seem to re-dirty the page even after
 206	 * the VM has canceled the dirty bit (eg ext3 journaling).
 207	 *
 208	 * Fix it up by doing a final dirty accounting check after
 209	 * having removed the page entirely.
 210	 */
 211	if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
 212		dec_zone_page_state(page, NR_FILE_DIRTY);
 213		dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
 214	}
 215}
 216
 217/**
 218 * delete_from_page_cache - delete page from page cache
 219 * @page: the page which the kernel is trying to remove from page cache
 220 *
 221 * This must be called only on pages that have been verified to be in the page
 222 * cache and locked.  It will never put the page into the free list, the caller
 223 * has a reference on the page.
 224 */
 225void delete_from_page_cache(struct page *page)
 226{
 227	struct address_space *mapping = page->mapping;
 228	void (*freepage)(struct page *);
 229
 230	BUG_ON(!PageLocked(page));
 231
 232	freepage = mapping->a_ops->freepage;
 233	spin_lock_irq(&mapping->tree_lock);
 234	__delete_from_page_cache(page, NULL);
 235	spin_unlock_irq(&mapping->tree_lock);
 236	mem_cgroup_uncharge_cache_page(page);
 237
 238	if (freepage)
 239		freepage(page);
 240	page_cache_release(page);
 241}
 242EXPORT_SYMBOL(delete_from_page_cache);
 243
 244static int sleep_on_page(void *word)
 245{
 246	io_schedule();
 247	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 248}
 249
 250static int sleep_on_page_killable(void *word)
 
 251{
 252	sleep_on_page(word);
 253	return fatal_signal_pending(current) ? -EINTR : 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 254}
 255
 256static int filemap_check_errors(struct address_space *mapping)
 257{
 258	int ret = 0;
 259	/* Check for outstanding write errors */
 260	if (test_bit(AS_ENOSPC, &mapping->flags) &&
 261	    test_and_clear_bit(AS_ENOSPC, &mapping->flags))
 262		ret = -ENOSPC;
 263	if (test_bit(AS_EIO, &mapping->flags) &&
 264	    test_and_clear_bit(AS_EIO, &mapping->flags))
 265		ret = -EIO;
 266	return ret;
 267}
 
 
 
 
 
 
 
 
 
 
 
 268
 269/**
 270 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
 271 * @mapping:	address space structure to write
 272 * @start:	offset in bytes where the range starts
 273 * @end:	offset in bytes where the range ends (inclusive)
 274 * @sync_mode:	enable synchronous operation
 275 *
 276 * Start writeback against all of a mapping's dirty pages that lie
 277 * within the byte offsets <start, end> inclusive.
 278 *
 279 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
 280 * opposed to a regular memory cleansing writeback.  The difference between
 281 * these two operations is that if a dirty page/buffer is encountered, it must
 282 * be waited upon, and not just skipped over.
 
 
 283 */
 284int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
 285				loff_t end, int sync_mode)
 286{
 287	int ret;
 288	struct writeback_control wbc = {
 289		.sync_mode = sync_mode,
 290		.nr_to_write = LONG_MAX,
 291		.range_start = start,
 292		.range_end = end,
 293	};
 294
 295	if (!mapping_cap_writeback_dirty(mapping))
 
 296		return 0;
 297
 
 298	ret = do_writepages(mapping, &wbc);
 
 299	return ret;
 300}
 301
 302static inline int __filemap_fdatawrite(struct address_space *mapping,
 303	int sync_mode)
 304{
 305	return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
 306}
 307
 308int filemap_fdatawrite(struct address_space *mapping)
 309{
 310	return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
 311}
 312EXPORT_SYMBOL(filemap_fdatawrite);
 313
 314int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
 315				loff_t end)
 316{
 317	return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
 318}
 319EXPORT_SYMBOL(filemap_fdatawrite_range);
 320
 321/**
 322 * filemap_flush - mostly a non-blocking flush
 323 * @mapping:	target address_space
 324 *
 325 * This is a mostly non-blocking flush.  Not suitable for data-integrity
 326 * purposes - I/O may not be started against all dirty pages.
 
 
 327 */
 328int filemap_flush(struct address_space *mapping)
 329{
 330	return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
 331}
 332EXPORT_SYMBOL(filemap_flush);
 333
 334/**
 335 * filemap_fdatawait_range - wait for writeback to complete
 336 * @mapping:		address space structure to wait for
 337 * @start_byte:		offset in bytes where the range starts
 338 * @end_byte:		offset in bytes where the range ends (inclusive)
 
 
 
 339 *
 340 * Walk the list of under-writeback pages of the given address space
 341 * in the given range and wait for all of them.
 342 */
 343int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
 344			    loff_t end_byte)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 345{
 346	pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
 347	pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
 348	struct pagevec pvec;
 349	int nr_pages;
 350	int ret2, ret = 0;
 351
 352	if (end_byte < start_byte)
 353		goto out;
 354
 355	pagevec_init(&pvec, 0);
 356	while ((index <= end) &&
 357			(nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 358			PAGECACHE_TAG_WRITEBACK,
 359			min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
 360		unsigned i;
 361
 
 
 
 
 
 362		for (i = 0; i < nr_pages; i++) {
 363			struct page *page = pvec.pages[i];
 364
 365			/* until radix tree lookup accepts end_index */
 366			if (page->index > end)
 367				continue;
 368
 369			wait_on_page_writeback(page);
 370			if (TestClearPageError(page))
 371				ret = -EIO;
 372		}
 373		pagevec_release(&pvec);
 374		cond_resched();
 375	}
 376out:
 377	ret2 = filemap_check_errors(mapping);
 378	if (!ret)
 379		ret = ret2;
 380
 381	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 382}
 383EXPORT_SYMBOL(filemap_fdatawait_range);
 384
 385/**
 386 * filemap_fdatawait - wait for all under-writeback pages to complete
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 387 * @mapping: address space structure to wait for
 388 *
 389 * Walk the list of under-writeback pages of the given address space
 390 * and wait for all of them.
 
 
 
 
 
 
 
 391 */
 392int filemap_fdatawait(struct address_space *mapping)
 393{
 394	loff_t i_size = i_size_read(mapping->host);
 
 
 
 395
 396	if (i_size == 0)
 397		return 0;
 
 
 
 398
 399	return filemap_fdatawait_range(mapping, 0, i_size - 1);
 400}
 401EXPORT_SYMBOL(filemap_fdatawait);
 402
 403int filemap_write_and_wait(struct address_space *mapping)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 404{
 405	int err = 0;
 406
 407	if (mapping->nrpages) {
 408		err = filemap_fdatawrite(mapping);
 
 409		/*
 410		 * Even if the above returned error, the pages may be
 411		 * written partially (e.g. -ENOSPC), so we wait for it.
 412		 * But the -EIO is special case, it may indicate the worst
 413		 * thing (e.g. bug) happened, so we avoid waiting for it.
 414		 */
 415		if (err != -EIO) {
 416			int err2 = filemap_fdatawait(mapping);
 
 417			if (!err)
 418				err = err2;
 
 
 
 419		}
 420	} else {
 421		err = filemap_check_errors(mapping);
 422	}
 423	return err;
 424}
 425EXPORT_SYMBOL(filemap_write_and_wait);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 426
 427/**
 428 * filemap_write_and_wait_range - write out & wait on a file range
 429 * @mapping:	the address_space for the pages
 430 * @lstart:	offset in bytes where the range starts
 431 * @lend:	offset in bytes where the range ends (inclusive)
 432 *
 433 * Write out and wait upon file offsets lstart->lend, inclusive.
 434 *
 435 * Note that `lend' is inclusive (describes the last byte to be written) so
 436 * that this function can be used to write to the very end-of-file (end = -1).
 
 
 
 
 
 437 */
 438int filemap_write_and_wait_range(struct address_space *mapping,
 439				 loff_t lstart, loff_t lend)
 440{
 441	int err = 0;
 
 442
 443	if (mapping->nrpages) {
 444		err = __filemap_fdatawrite_range(mapping, lstart, lend,
 445						 WB_SYNC_ALL);
 446		/* See comment of filemap_write_and_wait() */
 447		if (err != -EIO) {
 448			int err2 = filemap_fdatawait_range(mapping,
 449						lstart, lend);
 450			if (!err)
 451				err = err2;
 452		}
 453	} else {
 454		err = filemap_check_errors(mapping);
 455	}
 
 
 
 456	return err;
 457}
 458EXPORT_SYMBOL(filemap_write_and_wait_range);
 459
 460/**
 461 * replace_page_cache_page - replace a pagecache page with a new one
 462 * @old:	page to be replaced
 463 * @new:	page to replace with
 464 * @gfp_mask:	allocation mode
 465 *
 466 * This function replaces a page in the pagecache with a new one.  On
 467 * success it acquires the pagecache reference for the new page and
 468 * drops it for the old page.  Both the old and new pages must be
 469 * locked.  This function does not add the new page to the LRU, the
 470 * caller must do that.
 471 *
 472 * The remove + add is atomic.  The only way this function can fail is
 473 * memory allocation failure.
 
 474 */
 475int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
 476{
 477	int error;
 
 
 
 
 478
 479	VM_BUG_ON_PAGE(!PageLocked(old), old);
 480	VM_BUG_ON_PAGE(!PageLocked(new), new);
 481	VM_BUG_ON_PAGE(new->mapping, new);
 482
 483	error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
 484	if (!error) {
 485		struct address_space *mapping = old->mapping;
 486		void (*freepage)(struct page *);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 487
 488		pgoff_t offset = old->index;
 489		freepage = mapping->a_ops->freepage;
 490
 491		page_cache_get(new);
 492		new->mapping = mapping;
 493		new->index = offset;
 494
 495		spin_lock_irq(&mapping->tree_lock);
 496		__delete_from_page_cache(old, NULL);
 497		error = radix_tree_insert(&mapping->page_tree, offset, new);
 498		BUG_ON(error);
 499		mapping->nrpages++;
 500		__inc_zone_page_state(new, NR_FILE_PAGES);
 501		if (PageSwapBacked(new))
 502			__inc_zone_page_state(new, NR_SHMEM);
 503		spin_unlock_irq(&mapping->tree_lock);
 504		/* mem_cgroup codes must not be called under tree_lock */
 505		mem_cgroup_replace_page_cache(old, new);
 506		radix_tree_preload_end();
 507		if (freepage)
 508			freepage(old);
 509		page_cache_release(old);
 510	}
 511
 512	return error;
 513}
 514EXPORT_SYMBOL_GPL(replace_page_cache_page);
 515
 516static int page_cache_tree_insert(struct address_space *mapping,
 517				  struct page *page, void **shadowp)
 518{
 519	struct radix_tree_node *node;
 520	void **slot;
 521	int error;
 522
 523	error = __radix_tree_create(&mapping->page_tree, page->index,
 524				    &node, &slot);
 525	if (error)
 526		return error;
 527	if (*slot) {
 528		void *p;
 529
 530		p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
 531		if (!radix_tree_exceptional_entry(p))
 532			return -EEXIST;
 533		if (shadowp)
 534			*shadowp = p;
 535		mapping->nrshadows--;
 536		if (node)
 537			workingset_node_shadows_dec(node);
 538	}
 539	radix_tree_replace_slot(slot, page);
 540	mapping->nrpages++;
 541	if (node) {
 542		workingset_node_pages_inc(node);
 543		/*
 544		 * Don't track node that contains actual pages.
 545		 *
 546		 * Avoid acquiring the list_lru lock if already
 547		 * untracked.  The list_empty() test is safe as
 548		 * node->private_list is protected by
 549		 * mapping->tree_lock.
 550		 */
 551		if (!list_empty(&node->private_list))
 552			list_lru_del(&workingset_shadow_nodes,
 553				     &node->private_list);
 554	}
 555	return 0;
 556}
 557
 558static int __add_to_page_cache_locked(struct page *page,
 559				      struct address_space *mapping,
 560				      pgoff_t offset, gfp_t gfp_mask,
 561				      void **shadowp)
 562{
 
 
 563	int error;
 
 564
 565	VM_BUG_ON_PAGE(!PageLocked(page), page);
 566	VM_BUG_ON_PAGE(PageSwapBacked(page), page);
 
 
 
 
 
 567
 568	error = mem_cgroup_charge_file(page, current->mm,
 569					gfp_mask & GFP_RECLAIM_MASK);
 570	if (error)
 571		return error;
 572
 573	error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
 574	if (error) {
 575		mem_cgroup_uncharge_cache_page(page);
 576		return error;
 577	}
 578
 579	page_cache_get(page);
 580	page->mapping = mapping;
 581	page->index = offset;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 582
 583	spin_lock_irq(&mapping->tree_lock);
 584	error = page_cache_tree_insert(mapping, page, shadowp);
 585	radix_tree_preload_end();
 586	if (unlikely(error))
 587		goto err_insert;
 588	__inc_zone_page_state(page, NR_FILE_PAGES);
 589	spin_unlock_irq(&mapping->tree_lock);
 590	trace_mm_filemap_add_to_page_cache(page);
 591	return 0;
 592err_insert:
 593	page->mapping = NULL;
 594	/* Leave page->index set: truncation relies upon it */
 595	spin_unlock_irq(&mapping->tree_lock);
 596	mem_cgroup_uncharge_cache_page(page);
 597	page_cache_release(page);
 598	return error;
 599}
 
 600
 601/**
 602 * add_to_page_cache_locked - add a locked page to the pagecache
 603 * @page:	page to add
 604 * @mapping:	the page's address_space
 605 * @offset:	page index
 606 * @gfp_mask:	page allocation mode
 607 *
 608 * This function is used to add a page to the pagecache. It must be locked.
 609 * This function does not add the page to the LRU.  The caller must do that.
 
 
 610 */
 611int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
 612		pgoff_t offset, gfp_t gfp_mask)
 613{
 614	return __add_to_page_cache_locked(page, mapping, offset,
 615					  gfp_mask, NULL);
 616}
 617EXPORT_SYMBOL(add_to_page_cache_locked);
 618
 619int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
 620				pgoff_t offset, gfp_t gfp_mask)
 621{
 622	void *shadow = NULL;
 623	int ret;
 624
 625	__set_page_locked(page);
 626	ret = __add_to_page_cache_locked(page, mapping, offset,
 627					 gfp_mask, &shadow);
 628	if (unlikely(ret))
 629		__clear_page_locked(page);
 630	else {
 631		/*
 632		 * The page might have been evicted from cache only
 633		 * recently, in which case it should be activated like
 634		 * any other repeatedly accessed page.
 635		 */
 636		if (shadow && workingset_refault(shadow)) {
 637			SetPageActive(page);
 638			workingset_activation(page);
 639		} else
 640			ClearPageActive(page);
 
 641		lru_cache_add(page);
 642	}
 643	return ret;
 644}
 645EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
 646
 647#ifdef CONFIG_NUMA
 648struct page *__page_cache_alloc(gfp_t gfp)
 649{
 650	int n;
 651	struct page *page;
 652
 653	if (cpuset_do_page_mem_spread()) {
 654		unsigned int cpuset_mems_cookie;
 655		do {
 656			cpuset_mems_cookie = read_mems_allowed_begin();
 657			n = cpuset_mem_spread_node();
 658			page = alloc_pages_exact_node(n, gfp, 0);
 659		} while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
 660
 661		return page;
 662	}
 663	return alloc_pages(gfp, 0);
 664}
 665EXPORT_SYMBOL(__page_cache_alloc);
 666#endif
 667
 668/*
 669 * In order to wait for pages to become available there must be
 670 * waitqueues associated with pages. By using a hash table of
 671 * waitqueues where the bucket discipline is to maintain all
 672 * waiters on the same queue and wake all when any of the pages
 673 * become available, and for the woken contexts to check to be
 674 * sure the appropriate page became available, this saves space
 675 * at a cost of "thundering herd" phenomena during rare hash
 676 * collisions.
 677 */
 
 
 
 
 678static wait_queue_head_t *page_waitqueue(struct page *page)
 679{
 680	const struct zone *zone = page_zone(page);
 
 
 
 
 
 681
 682	return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
 
 
 
 683}
 684
 685static inline void wake_up_page(struct page *page, int bit)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 686{
 687	__wake_up_bit(page_waitqueue(page), &page->flags, bit);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 688}
 689
 690void wait_on_page_bit(struct page *page, int bit_nr)
 691{
 692	DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
 693
 694	if (test_bit(bit_nr, &page->flags))
 695		__wait_on_bit(page_waitqueue(page), &wait, sleep_on_page,
 696							TASK_UNINTERRUPTIBLE);
 697}
 698EXPORT_SYMBOL(wait_on_page_bit);
 699
 700int wait_on_page_bit_killable(struct page *page, int bit_nr)
 701{
 702	DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 703
 704	if (!test_bit(bit_nr, &page->flags))
 
 
 
 705		return 0;
 
 
 706
 707	return __wait_on_bit(page_waitqueue(page), &wait,
 708			     sleep_on_page_killable, TASK_KILLABLE);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 709}
 710
 711/**
 712 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
 713 * @page: Page defining the wait queue of interest
 714 * @waiter: Waiter to add to the queue
 715 *
 716 * Add an arbitrary @waiter to the wait queue for the nominated @page.
 717 */
 718void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
 719{
 720	wait_queue_head_t *q = page_waitqueue(page);
 721	unsigned long flags;
 722
 723	spin_lock_irqsave(&q->lock, flags);
 724	__add_wait_queue(q, waiter);
 
 725	spin_unlock_irqrestore(&q->lock, flags);
 726}
 727EXPORT_SYMBOL_GPL(add_page_wait_queue);
 728
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 729/**
 730 * unlock_page - unlock a locked page
 731 * @page: the page
 732 *
 733 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
 734 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
 735 * mechananism between PageLocked pages and PageWriteback pages is shared.
 736 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
 737 *
 738 * The mb is necessary to enforce ordering between the clear_bit and the read
 739 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
 
 
 
 740 */
 741void unlock_page(struct page *page)
 742{
 
 
 743	VM_BUG_ON_PAGE(!PageLocked(page), page);
 744	clear_bit_unlock(PG_locked, &page->flags);
 745	smp_mb__after_clear_bit();
 746	wake_up_page(page, PG_locked);
 747}
 748EXPORT_SYMBOL(unlock_page);
 749
 750/**
 751 * end_page_writeback - end writeback against a page
 752 * @page: the page
 753 */
 754void end_page_writeback(struct page *page)
 755{
 756	if (TestClearPageReclaim(page))
 
 
 
 
 
 
 
 
 757		rotate_reclaimable_page(page);
 
 758
 759	if (!test_clear_page_writeback(page))
 760		BUG();
 761
 762	smp_mb__after_clear_bit();
 763	wake_up_page(page, PG_writeback);
 764}
 765EXPORT_SYMBOL(end_page_writeback);
 766
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 767/**
 768 * __lock_page - get a lock on the page, assuming we need to sleep to get it
 769 * @page: the page to lock
 770 */
 771void __lock_page(struct page *page)
 772{
 773	DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
 774
 775	__wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page,
 776							TASK_UNINTERRUPTIBLE);
 777}
 778EXPORT_SYMBOL(__lock_page);
 779
 780int __lock_page_killable(struct page *page)
 781{
 782	DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
 
 
 
 
 
 783
 784	return __wait_on_bit_lock(page_waitqueue(page), &wait,
 785					sleep_on_page_killable, TASK_KILLABLE);
 
 786}
 787EXPORT_SYMBOL_GPL(__lock_page_killable);
 788
 
 
 
 
 
 
 
 
 
 
 
 789int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
 790			 unsigned int flags)
 791{
 792	if (flags & FAULT_FLAG_ALLOW_RETRY) {
 793		/*
 794		 * CAUTION! In this case, mmap_sem is not released
 795		 * even though return 0.
 796		 */
 797		if (flags & FAULT_FLAG_RETRY_NOWAIT)
 798			return 0;
 799
 800		up_read(&mm->mmap_sem);
 801		if (flags & FAULT_FLAG_KILLABLE)
 802			wait_on_page_locked_killable(page);
 803		else
 804			wait_on_page_locked(page);
 805		return 0;
 806	} else {
 807		if (flags & FAULT_FLAG_KILLABLE) {
 808			int ret;
 809
 810			ret = __lock_page_killable(page);
 811			if (ret) {
 812				up_read(&mm->mmap_sem);
 813				return 0;
 814			}
 815		} else
 816			__lock_page(page);
 817		return 1;
 818	}
 819}
 820
 821/**
 822 * page_cache_next_hole - find the next hole (not-present entry)
 823 * @mapping: mapping
 824 * @index: index
 825 * @max_scan: maximum range to search
 826 *
 827 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
 828 * lowest indexed hole.
 829 *
 830 * Returns: the index of the hole if found, otherwise returns an index
 831 * outside of the set specified (in which case 'return - index >=
 832 * max_scan' will be true). In rare cases of index wrap-around, 0 will
 833 * be returned.
 834 *
 835 * page_cache_next_hole may be called under rcu_read_lock. However,
 836 * like radix_tree_gang_lookup, this will not atomically search a
 837 * snapshot of the tree at a single point in time. For example, if a
 838 * hole is created at index 5, then subsequently a hole is created at
 839 * index 10, page_cache_next_hole covering both indexes may return 10
 840 * if called under rcu_read_lock.
 841 */
 842pgoff_t page_cache_next_hole(struct address_space *mapping,
 843			     pgoff_t index, unsigned long max_scan)
 844{
 845	unsigned long i;
 846
 847	for (i = 0; i < max_scan; i++) {
 848		struct page *page;
 849
 850		page = radix_tree_lookup(&mapping->page_tree, index);
 851		if (!page || radix_tree_exceptional_entry(page))
 852			break;
 853		index++;
 854		if (index == 0)
 855			break;
 856	}
 857
 858	return index;
 859}
 860EXPORT_SYMBOL(page_cache_next_hole);
 861
 862/**
 863 * page_cache_prev_hole - find the prev hole (not-present entry)
 864 * @mapping: mapping
 865 * @index: index
 866 * @max_scan: maximum range to search
 867 *
 868 * Search backwards in the range [max(index-max_scan+1, 0), index] for
 869 * the first hole.
 870 *
 871 * Returns: the index of the hole if found, otherwise returns an index
 872 * outside of the set specified (in which case 'index - return >=
 873 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
 874 * will be returned.
 
 875 *
 876 * page_cache_prev_hole may be called under rcu_read_lock. However,
 877 * like radix_tree_gang_lookup, this will not atomically search a
 878 * snapshot of the tree at a single point in time. For example, if a
 879 * hole is created at index 10, then subsequently a hole is created at
 880 * index 5, page_cache_prev_hole covering both indexes may return 5 if
 881 * called under rcu_read_lock.
 882 */
 883pgoff_t page_cache_prev_hole(struct address_space *mapping,
 884			     pgoff_t index, unsigned long max_scan)
 885{
 886	unsigned long i;
 887
 888	for (i = 0; i < max_scan; i++) {
 889		struct page *page;
 890
 891		page = radix_tree_lookup(&mapping->page_tree, index);
 892		if (!page || radix_tree_exceptional_entry(page))
 
 893			break;
 894		index--;
 895		if (index == ULONG_MAX)
 896			break;
 897	}
 898
 899	return index;
 900}
 901EXPORT_SYMBOL(page_cache_prev_hole);
 902
 903/**
 904 * find_get_entry - find and get a page cache entry
 905 * @mapping: the address_space to search
 906 * @offset: the page cache index
 907 *
 908 * Looks up the page cache slot at @mapping & @offset.  If there is a
 909 * page cache page, it is returned with an increased refcount.
 910 *
 911 * If the slot holds a shadow entry of a previously evicted page, or a
 912 * swap entry from shmem/tmpfs, it is returned.
 913 *
 914 * Otherwise, %NULL is returned.
 915 */
 916struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
 917{
 918	void **pagep;
 919	struct page *page;
 920
 921	rcu_read_lock();
 922repeat:
 923	page = NULL;
 924	pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
 925	if (pagep) {
 926		page = radix_tree_deref_slot(pagep);
 927		if (unlikely(!page))
 928			goto out;
 929		if (radix_tree_exception(page)) {
 930			if (radix_tree_deref_retry(page))
 931				goto repeat;
 932			/*
 933			 * A shadow entry of a recently evicted page,
 934			 * or a swap entry from shmem/tmpfs.  Return
 935			 * it without attempting to raise page count.
 936			 */
 937			goto out;
 938		}
 939		if (!page_cache_get_speculative(page))
 940			goto repeat;
 941
 942		/*
 943		 * Has the page moved?
 944		 * This is part of the lockless pagecache protocol. See
 945		 * include/linux/pagemap.h for details.
 946		 */
 947		if (unlikely(page != *pagep)) {
 948			page_cache_release(page);
 949			goto repeat;
 950		}
 951	}
 
 952out:
 953	rcu_read_unlock();
 954
 955	return page;
 956}
 957EXPORT_SYMBOL(find_get_entry);
 958
 959/**
 960 * find_get_page - find and get a page reference
 961 * @mapping: the address_space to search
 962 * @offset: the page index
 963 *
 964 * Looks up the page cache slot at @mapping & @offset.  If there is a
 965 * page cache page, it is returned with an increased refcount.
 966 *
 967 * Otherwise, %NULL is returned.
 968 */
 969struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
 970{
 971	struct page *page = find_get_entry(mapping, offset);
 972
 973	if (radix_tree_exceptional_entry(page))
 974		page = NULL;
 975	return page;
 976}
 977EXPORT_SYMBOL(find_get_page);
 978
 979/**
 980 * find_lock_entry - locate, pin and lock a page cache entry
 981 * @mapping: the address_space to search
 982 * @offset: the page cache index
 983 *
 984 * Looks up the page cache slot at @mapping & @offset.  If there is a
 985 * page cache page, it is returned locked and with an increased
 986 * refcount.
 987 *
 988 * If the slot holds a shadow entry of a previously evicted page, or a
 989 * swap entry from shmem/tmpfs, it is returned.
 990 *
 991 * Otherwise, %NULL is returned.
 992 *
 993 * find_lock_entry() may sleep.
 994 */
 995struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
 996{
 997	struct page *page;
 998
 999repeat:
1000	page = find_get_entry(mapping, offset);
1001	if (page && !radix_tree_exception(page)) {
1002		lock_page(page);
1003		/* Has the page been truncated? */
1004		if (unlikely(page->mapping != mapping)) {
1005			unlock_page(page);
1006			page_cache_release(page);
1007			goto repeat;
1008		}
1009		VM_BUG_ON_PAGE(page->index != offset, page);
1010	}
1011	return page;
1012}
1013EXPORT_SYMBOL(find_lock_entry);
1014
1015/**
1016 * find_lock_page - locate, pin and lock a pagecache page
1017 * @mapping: the address_space to search
1018 * @offset: the page index
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1019 *
1020 * Looks up the page cache slot at @mapping & @offset.  If there is a
1021 * page cache page, it is returned locked and with an increased
1022 * refcount.
1023 *
1024 * Otherwise, %NULL is returned.
1025 *
1026 * find_lock_page() may sleep.
1027 */
1028struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
 
1029{
1030	struct page *page = find_lock_entry(mapping, offset);
1031
1032	if (radix_tree_exceptional_entry(page))
 
 
1033		page = NULL;
1034	return page;
1035}
1036EXPORT_SYMBOL(find_lock_page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1037
1038/**
1039 * find_or_create_page - locate or add a pagecache page
1040 * @mapping: the page's address_space
1041 * @index: the page's index into the mapping
1042 * @gfp_mask: page allocation mode
1043 *
1044 * Looks up the page cache slot at @mapping & @offset.  If there is a
1045 * page cache page, it is returned locked and with an increased
1046 * refcount.
1047 *
1048 * If the page is not present, a new page is allocated using @gfp_mask
1049 * and added to the page cache and the VM's LRU list.  The page is
1050 * returned locked and with an increased refcount.
1051 *
1052 * On memory exhaustion, %NULL is returned.
1053 *
1054 * find_or_create_page() may sleep, even if @gfp_flags specifies an
1055 * atomic allocation!
1056 */
1057struct page *find_or_create_page(struct address_space *mapping,
1058		pgoff_t index, gfp_t gfp_mask)
1059{
1060	struct page *page;
1061	int err;
1062repeat:
1063	page = find_lock_page(mapping, index);
1064	if (!page) {
1065		page = __page_cache_alloc(gfp_mask);
1066		if (!page)
1067			return NULL;
1068		/*
1069		 * We want a regular kernel memory (not highmem or DMA etc)
1070		 * allocation for the radix tree nodes, but we need to honour
1071		 * the context-specific requirements the caller has asked for.
1072		 * GFP_RECLAIM_MASK collects those requirements.
1073		 */
1074		err = add_to_page_cache_lru(page, mapping, index,
1075			(gfp_mask & GFP_RECLAIM_MASK));
 
1076		if (unlikely(err)) {
1077			page_cache_release(page);
1078			page = NULL;
1079			if (err == -EEXIST)
1080				goto repeat;
1081		}
 
 
 
 
 
 
 
1082	}
 
1083	return page;
1084}
1085EXPORT_SYMBOL(find_or_create_page);
1086
1087/**
1088 * find_get_entries - gang pagecache lookup
1089 * @mapping:	The address_space to search
1090 * @start:	The starting page cache index
1091 * @nr_entries:	The maximum number of entries
1092 * @entries:	Where the resulting entries are placed
1093 * @indices:	The cache indices corresponding to the entries in @entries
1094 *
1095 * find_get_entries() will search for and return a group of up to
1096 * @nr_entries entries in the mapping.  The entries are placed at
1097 * @entries.  find_get_entries() takes a reference against any actual
1098 * pages it returns.
1099 *
1100 * The search returns a group of mapping-contiguous page cache entries
1101 * with ascending indexes.  There may be holes in the indices due to
1102 * not-present pages.
1103 *
1104 * Any shadow entries of evicted pages, or swap entries from
1105 * shmem/tmpfs, are included in the returned array.
1106 *
1107 * find_get_entries() returns the number of pages and shadow entries
1108 * which were found.
 
 
 
 
1109 */
1110unsigned find_get_entries(struct address_space *mapping,
1111			  pgoff_t start, unsigned int nr_entries,
1112			  struct page **entries, pgoff_t *indices)
1113{
1114	void **slot;
 
1115	unsigned int ret = 0;
1116	struct radix_tree_iter iter;
1117
1118	if (!nr_entries)
1119		return 0;
1120
1121	rcu_read_lock();
1122restart:
1123	radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1124		struct page *page;
1125repeat:
1126		page = radix_tree_deref_slot(slot);
1127		if (unlikely(!page))
1128			continue;
1129		if (radix_tree_exception(page)) {
1130			if (radix_tree_deref_retry(page))
1131				goto restart;
1132			/*
1133			 * A shadow entry of a recently evicted page,
1134			 * or a swap entry from shmem/tmpfs.  Return
1135			 * it without attempting to raise page count.
1136			 */
1137			goto export;
1138		}
1139		if (!page_cache_get_speculative(page))
1140			goto repeat;
 
 
 
 
1141
1142		/* Has the page moved? */
1143		if (unlikely(page != *slot)) {
1144			page_cache_release(page);
1145			goto repeat;
 
 
 
1146		}
1147export:
1148		indices[ret] = iter.index;
1149		entries[ret] = page;
1150		if (++ret == nr_entries)
1151			break;
 
 
 
 
 
1152	}
1153	rcu_read_unlock();
1154	return ret;
1155}
1156
1157/**
1158 * find_get_pages - gang pagecache lookup
1159 * @mapping:	The address_space to search
1160 * @start:	The starting page index
 
1161 * @nr_pages:	The maximum number of pages
1162 * @pages:	Where the resulting pages are placed
1163 *
1164 * find_get_pages() will search for and return a group of up to
1165 * @nr_pages pages in the mapping.  The pages are placed at @pages.
1166 * find_get_pages() takes a reference against the returned pages.
 
1167 *
1168 * The search returns a group of mapping-contiguous pages with ascending
1169 * indexes.  There may be holes in the indices due to not-present pages.
 
1170 *
1171 * find_get_pages() returns the number of pages which were found.
 
 
1172 */
1173unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1174			    unsigned int nr_pages, struct page **pages)
 
1175{
1176	struct radix_tree_iter iter;
1177	void **slot;
1178	unsigned ret = 0;
1179
1180	if (unlikely(!nr_pages))
1181		return 0;
1182
1183	rcu_read_lock();
1184restart:
1185	radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1186		struct page *page;
1187repeat:
1188		page = radix_tree_deref_slot(slot);
1189		if (unlikely(!page))
1190			continue;
1191
1192		if (radix_tree_exception(page)) {
1193			if (radix_tree_deref_retry(page)) {
1194				/*
1195				 * Transient condition which can only trigger
1196				 * when entry at index 0 moves out of or back
1197				 * to root: none yet gotten, safe to restart.
1198				 */
1199				WARN_ON(iter.index);
1200				goto restart;
1201			}
1202			/*
1203			 * A shadow entry of a recently evicted page,
1204			 * or a swap entry from shmem/tmpfs.  Skip
1205			 * over it.
1206			 */
1207			continue;
1208		}
1209
1210		if (!page_cache_get_speculative(page))
1211			goto repeat;
1212
1213		/* Has the page moved? */
1214		if (unlikely(page != *slot)) {
1215			page_cache_release(page);
1216			goto repeat;
 
 
 
 
1217		}
1218
1219		pages[ret] = page;
1220		if (++ret == nr_pages)
1221			break;
 
1222	}
1223
 
 
 
 
 
 
 
 
 
 
 
1224	rcu_read_unlock();
 
1225	return ret;
1226}
1227
1228/**
1229 * find_get_pages_contig - gang contiguous pagecache lookup
1230 * @mapping:	The address_space to search
1231 * @index:	The starting page index
1232 * @nr_pages:	The maximum number of pages
1233 * @pages:	Where the resulting pages are placed
1234 *
1235 * find_get_pages_contig() works exactly like find_get_pages(), except
1236 * that the returned number of pages are guaranteed to be contiguous.
1237 *
1238 * find_get_pages_contig() returns the number of pages which were found.
1239 */
1240unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1241			       unsigned int nr_pages, struct page **pages)
1242{
1243	struct radix_tree_iter iter;
1244	void **slot;
1245	unsigned int ret = 0;
1246
1247	if (unlikely(!nr_pages))
1248		return 0;
1249
1250	rcu_read_lock();
1251restart:
1252	radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
1253		struct page *page;
1254repeat:
1255		page = radix_tree_deref_slot(slot);
1256		/* The hole, there no reason to continue */
1257		if (unlikely(!page))
1258			break;
1259
1260		if (radix_tree_exception(page)) {
1261			if (radix_tree_deref_retry(page)) {
1262				/*
1263				 * Transient condition which can only trigger
1264				 * when entry at index 0 moves out of or back
1265				 * to root: none yet gotten, safe to restart.
1266				 */
1267				goto restart;
1268			}
1269			/*
1270			 * A shadow entry of a recently evicted page,
1271			 * or a swap entry from shmem/tmpfs.  Stop
1272			 * looking for contiguous pages.
1273			 */
1274			break;
1275		}
1276
1277		if (!page_cache_get_speculative(page))
1278			goto repeat;
1279
1280		/* Has the page moved? */
1281		if (unlikely(page != *slot)) {
1282			page_cache_release(page);
1283			goto repeat;
1284		}
1285
1286		/*
1287		 * must check mapping and index after taking the ref.
1288		 * otherwise we can get both false positives and false
1289		 * negatives, which is just confusing to the caller.
1290		 */
1291		if (page->mapping == NULL || page->index != iter.index) {
1292			page_cache_release(page);
1293			break;
1294		}
1295
1296		pages[ret] = page;
1297		if (++ret == nr_pages)
1298			break;
 
 
 
 
 
1299	}
1300	rcu_read_unlock();
1301	return ret;
1302}
1303EXPORT_SYMBOL(find_get_pages_contig);
1304
1305/**
1306 * find_get_pages_tag - find and return pages that match @tag
1307 * @mapping:	the address_space to search
1308 * @index:	the starting page index
 
1309 * @tag:	the tag index
1310 * @nr_pages:	the maximum number of pages
1311 * @pages:	where the resulting pages are placed
1312 *
1313 * Like find_get_pages, except we only return pages which are tagged with
1314 * @tag.   We update @index to index the next page for the traversal.
 
 
1315 */
1316unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1317			int tag, unsigned int nr_pages, struct page **pages)
 
1318{
1319	struct radix_tree_iter iter;
1320	void **slot;
1321	unsigned ret = 0;
1322
1323	if (unlikely(!nr_pages))
1324		return 0;
1325
1326	rcu_read_lock();
1327restart:
1328	radix_tree_for_each_tagged(slot, &mapping->page_tree,
1329				   &iter, *index, tag) {
1330		struct page *page;
1331repeat:
1332		page = radix_tree_deref_slot(slot);
1333		if (unlikely(!page))
1334			continue;
1335
1336		if (radix_tree_exception(page)) {
1337			if (radix_tree_deref_retry(page)) {
1338				/*
1339				 * Transient condition which can only trigger
1340				 * when entry at index 0 moves out of or back
1341				 * to root: none yet gotten, safe to restart.
1342				 */
1343				goto restart;
1344			}
1345			/*
1346			 * A shadow entry of a recently evicted page.
1347			 *
1348			 * Those entries should never be tagged, but
1349			 * this tree walk is lockless and the tags are
1350			 * looked up in bulk, one radix tree node at a
1351			 * time, so there is a sizable window for page
1352			 * reclaim to evict a page we saw tagged.
1353			 *
1354			 * Skip over it.
1355			 */
1356			continue;
1357		}
1358
1359		if (!page_cache_get_speculative(page))
1360			goto repeat;
1361
1362		/* Has the page moved? */
1363		if (unlikely(page != *slot)) {
1364			page_cache_release(page);
1365			goto repeat;
 
 
 
 
1366		}
1367
1368		pages[ret] = page;
1369		if (++ret == nr_pages)
1370			break;
 
1371	}
1372
 
 
 
 
 
 
 
 
 
 
 
1373	rcu_read_unlock();
1374
1375	if (ret)
1376		*index = pages[ret - 1]->index + 1;
1377
1378	return ret;
1379}
1380EXPORT_SYMBOL(find_get_pages_tag);
1381
1382/**
1383 * grab_cache_page_nowait - returns locked page at given index in given cache
1384 * @mapping: target address_space
1385 * @index: the page index
1386 *
1387 * Same as grab_cache_page(), but do not wait if the page is unavailable.
1388 * This is intended for speculative data generators, where the data can
1389 * be regenerated if the page couldn't be grabbed.  This routine should
1390 * be safe to call while holding the lock for another page.
1391 *
1392 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
1393 * and deadlock against the caller's locked page.
1394 */
1395struct page *
1396grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
1397{
1398	struct page *page = find_get_page(mapping, index);
1399
1400	if (page) {
1401		if (trylock_page(page))
1402			return page;
1403		page_cache_release(page);
1404		return NULL;
1405	}
1406	page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
1407	if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
1408		page_cache_release(page);
1409		page = NULL;
1410	}
1411	return page;
1412}
1413EXPORT_SYMBOL(grab_cache_page_nowait);
1414
1415/*
1416 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1417 * a _large_ part of the i/o request. Imagine the worst scenario:
1418 *
1419 *      ---R__________________________________________B__________
1420 *         ^ reading here                             ^ bad block(assume 4k)
1421 *
1422 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1423 * => failing the whole request => read(R) => read(R+1) =>
1424 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1425 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1426 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1427 *
1428 * It is going insane. Fix it by quickly scaling down the readahead size.
1429 */
1430static void shrink_readahead_size_eio(struct file *filp,
1431					struct file_ra_state *ra)
1432{
1433	ra->ra_pages /= 4;
1434}
1435
1436/**
1437 * do_generic_file_read - generic file read routine
1438 * @filp:	the file to read
1439 * @ppos:	current file position
1440 * @iter:	data destination
1441 * @written:	already copied
1442 *
1443 * This is a generic file read routine, and uses the
1444 * mapping->a_ops->readpage() function for the actual low-level stuff.
1445 *
1446 * This is really ugly. But the goto's actually try to clarify some
1447 * of the logic when it comes to error handling etc.
 
 
 
 
1448 */
1449static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1450		struct iov_iter *iter, ssize_t written)
1451{
 
1452	struct address_space *mapping = filp->f_mapping;
1453	struct inode *inode = mapping->host;
1454	struct file_ra_state *ra = &filp->f_ra;
 
1455	pgoff_t index;
1456	pgoff_t last_index;
1457	pgoff_t prev_index;
1458	unsigned long offset;      /* offset into pagecache page */
1459	unsigned int prev_offset;
1460	int error = 0;
1461
1462	index = *ppos >> PAGE_CACHE_SHIFT;
1463	prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1464	prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1465	last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1466	offset = *ppos & ~PAGE_CACHE_MASK;
 
 
 
 
1467
1468	for (;;) {
1469		struct page *page;
1470		pgoff_t end_index;
1471		loff_t isize;
1472		unsigned long nr, ret;
1473
1474		cond_resched();
1475find_page:
 
 
 
 
 
1476		page = find_get_page(mapping, index);
1477		if (!page) {
 
 
1478			page_cache_sync_readahead(mapping,
1479					ra, filp,
1480					index, last_index - index);
1481			page = find_get_page(mapping, index);
1482			if (unlikely(page == NULL))
1483				goto no_cached_page;
1484		}
1485		if (PageReadahead(page)) {
 
 
 
 
1486			page_cache_async_readahead(mapping,
1487					ra, filp, page,
1488					index, last_index - index);
1489		}
1490		if (!PageUptodate(page)) {
1491			if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1492					!mapping->a_ops->is_partially_uptodate)
1493				goto page_not_up_to_date;
 
 
 
1494			if (!trylock_page(page))
1495				goto page_not_up_to_date;
1496			/* Did it get truncated before we got the lock? */
1497			if (!page->mapping)
1498				goto page_not_up_to_date_locked;
1499			if (!mapping->a_ops->is_partially_uptodate(page,
1500							offset, iter->count))
1501				goto page_not_up_to_date_locked;
1502			unlock_page(page);
1503		}
1504page_ok:
1505		/*
1506		 * i_size must be checked after we know the page is Uptodate.
1507		 *
1508		 * Checking i_size after the check allows us to calculate
1509		 * the correct value for "nr", which means the zero-filled
1510		 * part of the page is not copied back to userspace (unless
1511		 * another truncate extends the file - this is desired though).
1512		 */
1513
1514		isize = i_size_read(inode);
1515		end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1516		if (unlikely(!isize || index > end_index)) {
1517			page_cache_release(page);
1518			goto out;
1519		}
1520
1521		/* nr is the maximum number of bytes to copy from this page */
1522		nr = PAGE_CACHE_SIZE;
1523		if (index == end_index) {
1524			nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1525			if (nr <= offset) {
1526				page_cache_release(page);
1527				goto out;
1528			}
1529		}
1530		nr = nr - offset;
1531
1532		/* If users can be writing to this page using arbitrary
1533		 * virtual addresses, take care about potential aliasing
1534		 * before reading the page on the kernel side.
1535		 */
1536		if (mapping_writably_mapped(mapping))
1537			flush_dcache_page(page);
1538
1539		/*
1540		 * When a sequential read accesses a page several times,
1541		 * only mark it as accessed the first time.
1542		 */
1543		if (prev_index != index || offset != prev_offset)
1544			mark_page_accessed(page);
1545		prev_index = index;
1546
1547		/*
1548		 * Ok, we have the page, and it's up-to-date, so
1549		 * now we can copy it to user space...
1550		 */
1551
1552		ret = copy_page_to_iter(page, offset, nr, iter);
1553		offset += ret;
1554		index += offset >> PAGE_CACHE_SHIFT;
1555		offset &= ~PAGE_CACHE_MASK;
1556		prev_offset = offset;
1557
1558		page_cache_release(page);
1559		written += ret;
1560		if (!iov_iter_count(iter))
1561			goto out;
1562		if (ret < nr) {
1563			error = -EFAULT;
1564			goto out;
1565		}
1566		continue;
1567
1568page_not_up_to_date:
1569		/* Get exclusive access to the page ... */
1570		error = lock_page_killable(page);
 
 
 
1571		if (unlikely(error))
1572			goto readpage_error;
1573
1574page_not_up_to_date_locked:
1575		/* Did it get truncated before we got the lock? */
1576		if (!page->mapping) {
1577			unlock_page(page);
1578			page_cache_release(page);
1579			continue;
1580		}
1581
1582		/* Did somebody else fill it already? */
1583		if (PageUptodate(page)) {
1584			unlock_page(page);
1585			goto page_ok;
1586		}
1587
1588readpage:
 
 
 
 
 
1589		/*
1590		 * A previous I/O error may have been due to temporary
1591		 * failures, eg. multipath errors.
1592		 * PG_error will be set again if readpage fails.
1593		 */
1594		ClearPageError(page);
1595		/* Start the actual read. The read will unlock the page. */
1596		error = mapping->a_ops->readpage(filp, page);
1597
1598		if (unlikely(error)) {
1599			if (error == AOP_TRUNCATED_PAGE) {
1600				page_cache_release(page);
1601				error = 0;
1602				goto find_page;
1603			}
1604			goto readpage_error;
1605		}
1606
1607		if (!PageUptodate(page)) {
1608			error = lock_page_killable(page);
 
 
 
 
1609			if (unlikely(error))
1610				goto readpage_error;
1611			if (!PageUptodate(page)) {
1612				if (page->mapping == NULL) {
1613					/*
1614					 * invalidate_mapping_pages got it
1615					 */
1616					unlock_page(page);
1617					page_cache_release(page);
1618					goto find_page;
1619				}
1620				unlock_page(page);
1621				shrink_readahead_size_eio(filp, ra);
1622				error = -EIO;
1623				goto readpage_error;
1624			}
1625			unlock_page(page);
1626		}
1627
1628		goto page_ok;
1629
1630readpage_error:
1631		/* UHHUH! A synchronous read error occurred. Report it */
1632		page_cache_release(page);
1633		goto out;
1634
1635no_cached_page:
1636		/*
1637		 * Ok, it wasn't cached, so we need to create a new
1638		 * page..
1639		 */
1640		page = page_cache_alloc_cold(mapping);
1641		if (!page) {
1642			error = -ENOMEM;
1643			goto out;
1644		}
1645		error = add_to_page_cache_lru(page, mapping,
1646						index, GFP_KERNEL);
1647		if (error) {
1648			page_cache_release(page);
1649			if (error == -EEXIST) {
1650				error = 0;
1651				goto find_page;
1652			}
1653			goto out;
1654		}
1655		goto readpage;
1656	}
1657
 
 
1658out:
1659	ra->prev_pos = prev_index;
1660	ra->prev_pos <<= PAGE_CACHE_SHIFT;
1661	ra->prev_pos |= prev_offset;
1662
1663	*ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1664	file_accessed(filp);
1665	return written ? written : error;
1666}
1667
1668/*
1669 * Performs necessary checks before doing a write
1670 * @iov:	io vector request
1671 * @nr_segs:	number of segments in the iovec
1672 * @count:	number of bytes to write
1673 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1674 *
1675 * Adjust number of segments and amount of bytes to write (nr_segs should be
1676 * properly initialized first). Returns appropriate error code that caller
1677 * should return or zero in case that write should be allowed.
1678 */
1679int generic_segment_checks(const struct iovec *iov,
1680			unsigned long *nr_segs, size_t *count, int access_flags)
1681{
1682	unsigned long   seg;
1683	size_t cnt = 0;
1684	for (seg = 0; seg < *nr_segs; seg++) {
1685		const struct iovec *iv = &iov[seg];
1686
1687		/*
1688		 * If any segment has a negative length, or the cumulative
1689		 * length ever wraps negative then return -EINVAL.
1690		 */
1691		cnt += iv->iov_len;
1692		if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1693			return -EINVAL;
1694		if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1695			continue;
1696		if (seg == 0)
1697			return -EFAULT;
1698		*nr_segs = seg;
1699		cnt -= iv->iov_len;	/* This segment is no good */
1700		break;
1701	}
1702	*count = cnt;
1703	return 0;
1704}
1705EXPORT_SYMBOL(generic_segment_checks);
1706
1707/**
1708 * generic_file_aio_read - generic filesystem read routine
1709 * @iocb:	kernel I/O control block
1710 * @iov:	io vector request
1711 * @nr_segs:	number of segments in the iovec
1712 * @pos:	current file position
1713 *
1714 * This is the "read()" routine for all filesystems
1715 * that can use the page cache directly.
 
 
 
 
 
 
 
 
 
 
 
 
 
1716 */
1717ssize_t
1718generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1719		unsigned long nr_segs, loff_t pos)
1720{
1721	struct file *filp = iocb->ki_filp;
1722	ssize_t retval;
1723	size_t count;
1724	loff_t *ppos = &iocb->ki_pos;
1725	struct iov_iter i;
1726
1727	count = 0;
1728	retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1729	if (retval)
1730		return retval;
1731	iov_iter_init(&i, iov, nr_segs, count, 0);
1732
1733	/* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1734	if (filp->f_flags & O_DIRECT) {
 
 
1735		loff_t size;
1736		struct address_space *mapping;
1737		struct inode *inode;
1738
1739		mapping = filp->f_mapping;
1740		inode = mapping->host;
1741		if (!count)
1742			goto out; /* skip atime */
1743		size = i_size_read(inode);
1744		retval = filemap_write_and_wait_range(mapping, pos,
1745					pos + iov_length(iov, nr_segs) - 1);
1746		if (!retval) {
1747			retval = mapping->a_ops->direct_IO(READ, iocb,
1748							   iov, pos, nr_segs);
 
 
 
 
 
1749		}
1750		if (retval > 0) {
1751			*ppos = pos + retval;
 
 
 
 
1752			count -= retval;
1753			/*
1754			 * If we did a short DIO read we need to skip the
1755			 * section of the iov that we've already read data into.
1756			 */
1757			iov_iter_advance(&i, retval);
1758		}
 
1759
1760		/*
1761		 * Btrfs can have a short DIO read if we encounter
1762		 * compressed extents, so if there was an error, or if
1763		 * we've already read everything we wanted to, or if
1764		 * there was a short read because we hit EOF, go ahead
1765		 * and return.  Otherwise fallthrough to buffered io for
1766		 * the rest of the read.
 
1767		 */
1768		if (retval < 0 || !count || *ppos >= size) {
1769			file_accessed(filp);
1770			goto out;
1771		}
1772	}
1773
1774	retval = do_generic_file_read(filp, ppos, &i, retval);
1775out:
1776	return retval;
1777}
1778EXPORT_SYMBOL(generic_file_aio_read);
1779
1780#ifdef CONFIG_MMU
1781/**
1782 * page_cache_read - adds requested page to the page cache if not already there
1783 * @file:	file to read
1784 * @offset:	page index
1785 *
1786 * This adds the requested page to the page cache if it isn't already there,
1787 * and schedules an I/O to read in its contents from disk.
 
 
 
 
1788 */
1789static int page_cache_read(struct file *file, pgoff_t offset)
 
1790{
1791	struct address_space *mapping = file->f_mapping;
1792	struct page *page; 
1793	int ret;
1794
1795	do {
1796		page = page_cache_alloc_cold(mapping);
1797		if (!page)
1798			return -ENOMEM;
 
 
 
1799
1800		ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1801		if (ret == 0)
1802			ret = mapping->a_ops->readpage(file, page);
1803		else if (ret == -EEXIST)
1804			ret = 0; /* losing race to add is OK */
1805
1806		page_cache_release(page);
1807
1808	} while (ret == AOP_TRUNCATED_PAGE);
1809		
1810	return ret;
 
 
 
 
 
1811}
1812
1813#define MMAP_LOTSAMISS  (100)
1814
1815/*
1816 * Synchronous readahead happens when we don't even find
1817 * a page in the page cache at all.
 
 
 
1818 */
1819static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1820				   struct file_ra_state *ra,
1821				   struct file *file,
1822				   pgoff_t offset)
1823{
1824	unsigned long ra_pages;
 
1825	struct address_space *mapping = file->f_mapping;
 
 
 
1826
1827	/* If we don't want any read-ahead, don't bother */
1828	if (vma->vm_flags & VM_RAND_READ)
1829		return;
1830	if (!ra->ra_pages)
1831		return;
1832
1833	if (vma->vm_flags & VM_SEQ_READ) {
 
1834		page_cache_sync_readahead(mapping, ra, file, offset,
1835					  ra->ra_pages);
1836		return;
1837	}
1838
1839	/* Avoid banging the cache line if not needed */
1840	if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
1841		ra->mmap_miss++;
 
1842
1843	/*
1844	 * Do we miss much more than hit in this file? If so,
1845	 * stop bothering with read-ahead. It will only hurt.
1846	 */
1847	if (ra->mmap_miss > MMAP_LOTSAMISS)
1848		return;
1849
1850	/*
1851	 * mmap read-around
1852	 */
1853	ra_pages = max_sane_readahead(ra->ra_pages);
1854	ra->start = max_t(long, 0, offset - ra_pages / 2);
1855	ra->size = ra_pages;
1856	ra->async_size = ra_pages / 4;
1857	ra_submit(ra, mapping, file);
 
1858}
1859
1860/*
1861 * Asynchronous readahead happens when we find the page and PG_readahead,
1862 * so we want to possibly extend the readahead further..
 
1863 */
1864static void do_async_mmap_readahead(struct vm_area_struct *vma,
1865				    struct file_ra_state *ra,
1866				    struct file *file,
1867				    struct page *page,
1868				    pgoff_t offset)
1869{
 
 
1870	struct address_space *mapping = file->f_mapping;
 
 
 
1871
1872	/* If we don't want any read-ahead, don't bother */
1873	if (vma->vm_flags & VM_RAND_READ)
1874		return;
1875	if (ra->mmap_miss > 0)
1876		ra->mmap_miss--;
1877	if (PageReadahead(page))
 
 
1878		page_cache_async_readahead(mapping, ra, file,
1879					   page, offset, ra->ra_pages);
 
 
1880}
1881
1882/**
1883 * filemap_fault - read in file data for page fault handling
1884 * @vma:	vma in which the fault was taken
1885 * @vmf:	struct vm_fault containing details of the fault
1886 *
1887 * filemap_fault() is invoked via the vma operations vector for a
1888 * mapped memory region to read in file data during a page fault.
1889 *
1890 * The goto's are kind of ugly, but this streamlines the normal case of having
1891 * it in the page cache, and handles the special cases reasonably without
1892 * having a lot of duplicated code.
 
 
 
 
 
 
 
 
 
 
 
 
1893 */
1894int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1895{
1896	int error;
1897	struct file *file = vma->vm_file;
 
1898	struct address_space *mapping = file->f_mapping;
1899	struct file_ra_state *ra = &file->f_ra;
1900	struct inode *inode = mapping->host;
1901	pgoff_t offset = vmf->pgoff;
 
1902	struct page *page;
1903	loff_t size;
1904	int ret = 0;
1905
1906	size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1907	if (offset >= size >> PAGE_CACHE_SHIFT)
1908		return VM_FAULT_SIGBUS;
1909
1910	/*
1911	 * Do we have something in the page cache already?
1912	 */
1913	page = find_get_page(mapping, offset);
1914	if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
1915		/*
1916		 * We found the page, so try async readahead before
1917		 * waiting for the lock.
1918		 */
1919		do_async_mmap_readahead(vma, ra, file, page, offset);
1920	} else if (!page) {
1921		/* No page in the page cache at all */
1922		do_sync_mmap_readahead(vma, ra, file, offset);
1923		count_vm_event(PGMAJFAULT);
1924		mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1925		ret = VM_FAULT_MAJOR;
 
1926retry_find:
1927		page = find_get_page(mapping, offset);
1928		if (!page)
1929			goto no_cached_page;
 
 
 
 
 
1930	}
1931
1932	if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1933		page_cache_release(page);
1934		return ret | VM_FAULT_RETRY;
1935	}
1936
1937	/* Did it get truncated? */
1938	if (unlikely(page->mapping != mapping)) {
1939		unlock_page(page);
1940		put_page(page);
1941		goto retry_find;
1942	}
1943	VM_BUG_ON_PAGE(page->index != offset, page);
1944
1945	/*
1946	 * We have a locked page in the page cache, now we need to check
1947	 * that it's up-to-date. If not, it is going to be due to an error.
1948	 */
1949	if (unlikely(!PageUptodate(page)))
1950		goto page_not_uptodate;
1951
1952	/*
 
 
 
 
 
 
 
 
 
 
1953	 * Found the page and have a reference on it.
1954	 * We must recheck i_size under page lock.
1955	 */
1956	size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1957	if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) {
1958		unlock_page(page);
1959		page_cache_release(page);
1960		return VM_FAULT_SIGBUS;
1961	}
1962
1963	vmf->page = page;
1964	return ret | VM_FAULT_LOCKED;
1965
1966no_cached_page:
1967	/*
1968	 * We're only likely to ever get here if MADV_RANDOM is in
1969	 * effect.
1970	 */
1971	error = page_cache_read(file, offset);
1972
1973	/*
1974	 * The page we want has now been added to the page cache.
1975	 * In the unlikely event that someone removed it in the
1976	 * meantime, we'll just come back here and read it again.
1977	 */
1978	if (error >= 0)
1979		goto retry_find;
1980
1981	/*
1982	 * An error return from page_cache_read can result if the
1983	 * system is low on memory, or a problem occurs while trying
1984	 * to schedule I/O.
1985	 */
1986	if (error == -ENOMEM)
1987		return VM_FAULT_OOM;
1988	return VM_FAULT_SIGBUS;
1989
1990page_not_uptodate:
1991	/*
1992	 * Umm, take care of errors if the page isn't up-to-date.
1993	 * Try to re-read it _once_. We do this synchronously,
1994	 * because there really aren't any performance issues here
1995	 * and we need to check for errors.
1996	 */
1997	ClearPageError(page);
 
1998	error = mapping->a_ops->readpage(file, page);
1999	if (!error) {
2000		wait_on_page_locked(page);
2001		if (!PageUptodate(page))
2002			error = -EIO;
2003	}
2004	page_cache_release(page);
 
 
2005
2006	if (!error || error == AOP_TRUNCATED_PAGE)
2007		goto retry_find;
2008
2009	/* Things didn't work out. Return zero to tell the mm layer so. */
2010	shrink_readahead_size_eio(file, ra);
2011	return VM_FAULT_SIGBUS;
 
 
 
 
 
 
 
 
 
 
 
 
2012}
2013EXPORT_SYMBOL(filemap_fault);
2014
2015void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
 
2016{
2017	struct radix_tree_iter iter;
2018	void **slot;
2019	struct file *file = vma->vm_file;
2020	struct address_space *mapping = file->f_mapping;
2021	loff_t size;
 
 
2022	struct page *page;
2023	unsigned long address = (unsigned long) vmf->virtual_address;
2024	unsigned long addr;
2025	pte_t *pte;
2026
2027	rcu_read_lock();
2028	radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) {
2029		if (iter.index > vmf->max_pgoff)
2030			break;
2031repeat:
2032		page = radix_tree_deref_slot(slot);
2033		if (unlikely(!page))
2034			goto next;
2035		if (radix_tree_exception(page)) {
2036			if (radix_tree_deref_retry(page))
2037				break;
2038			else
2039				goto next;
2040		}
2041
 
 
 
 
 
 
2042		if (!page_cache_get_speculative(page))
2043			goto repeat;
2044
2045		/* Has the page moved? */
2046		if (unlikely(page != *slot)) {
2047			page_cache_release(page);
2048			goto repeat;
2049		}
2050
2051		if (!PageUptodate(page) ||
2052				PageReadahead(page) ||
2053				PageHWPoison(page))
2054			goto skip;
2055		if (!trylock_page(page))
2056			goto skip;
2057
2058		if (page->mapping != mapping || !PageUptodate(page))
2059			goto unlock;
2060
2061		size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE);
2062		if (page->index >= size >> PAGE_CACHE_SHIFT)
2063			goto unlock;
2064
2065		pte = vmf->pte + page->index - vmf->pgoff;
2066		if (!pte_none(*pte))
 
 
 
 
 
 
2067			goto unlock;
2068
2069		if (file->f_ra.mmap_miss > 0)
2070			file->f_ra.mmap_miss--;
2071		addr = address + (page->index - vmf->pgoff) * PAGE_SIZE;
2072		do_set_pte(vma, addr, page, pte, false, false);
2073		unlock_page(page);
2074		goto next;
2075unlock:
2076		unlock_page(page);
2077skip:
2078		page_cache_release(page);
2079next:
2080		if (iter.index == vmf->max_pgoff)
 
2081			break;
2082	}
2083	rcu_read_unlock();
 
2084}
2085EXPORT_SYMBOL(filemap_map_pages);
2086
2087int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2088{
2089	struct page *page = vmf->page;
2090	struct inode *inode = file_inode(vma->vm_file);
2091	int ret = VM_FAULT_LOCKED;
2092
2093	sb_start_pagefault(inode->i_sb);
2094	file_update_time(vma->vm_file);
2095	lock_page(page);
2096	if (page->mapping != inode->i_mapping) {
2097		unlock_page(page);
2098		ret = VM_FAULT_NOPAGE;
2099		goto out;
2100	}
2101	/*
2102	 * We mark the page dirty already here so that when freeze is in
2103	 * progress, we are guaranteed that writeback during freezing will
2104	 * see the dirty page and writeprotect it again.
2105	 */
2106	set_page_dirty(page);
2107	wait_for_stable_page(page);
2108out:
2109	sb_end_pagefault(inode->i_sb);
2110	return ret;
2111}
2112EXPORT_SYMBOL(filemap_page_mkwrite);
2113
2114const struct vm_operations_struct generic_file_vm_ops = {
2115	.fault		= filemap_fault,
2116	.map_pages	= filemap_map_pages,
2117	.page_mkwrite	= filemap_page_mkwrite,
2118	.remap_pages	= generic_file_remap_pages,
2119};
2120
2121/* This is used for a general mmap of a disk file */
2122
2123int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2124{
2125	struct address_space *mapping = file->f_mapping;
2126
2127	if (!mapping->a_ops->readpage)
2128		return -ENOEXEC;
2129	file_accessed(file);
2130	vma->vm_ops = &generic_file_vm_ops;
2131	return 0;
2132}
2133
2134/*
2135 * This is for filesystems which do not implement ->writepage.
2136 */
2137int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2138{
2139	if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2140		return -EINVAL;
2141	return generic_file_mmap(file, vma);
2142}
2143#else
 
 
 
 
2144int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2145{
2146	return -ENOSYS;
2147}
2148int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2149{
2150	return -ENOSYS;
2151}
2152#endif /* CONFIG_MMU */
2153
 
2154EXPORT_SYMBOL(generic_file_mmap);
2155EXPORT_SYMBOL(generic_file_readonly_mmap);
2156
2157static struct page *wait_on_page_read(struct page *page)
2158{
2159	if (!IS_ERR(page)) {
2160		wait_on_page_locked(page);
2161		if (!PageUptodate(page)) {
2162			page_cache_release(page);
2163			page = ERR_PTR(-EIO);
2164		}
2165	}
2166	return page;
2167}
2168
2169static struct page *__read_cache_page(struct address_space *mapping,
2170				pgoff_t index,
2171				int (*filler)(void *, struct page *),
2172				void *data,
2173				gfp_t gfp)
2174{
2175	struct page *page;
2176	int err;
2177repeat:
2178	page = find_get_page(mapping, index);
2179	if (!page) {
2180		page = __page_cache_alloc(gfp | __GFP_COLD);
2181		if (!page)
2182			return ERR_PTR(-ENOMEM);
2183		err = add_to_page_cache_lru(page, mapping, index, gfp);
2184		if (unlikely(err)) {
2185			page_cache_release(page);
2186			if (err == -EEXIST)
2187				goto repeat;
2188			/* Presumably ENOMEM for radix tree node */
2189			return ERR_PTR(err);
2190		}
2191		err = filler(data, page);
 
 
 
 
 
 
2192		if (err < 0) {
2193			page_cache_release(page);
2194			page = ERR_PTR(err);
2195		} else {
2196			page = wait_on_page_read(page);
2197		}
 
 
 
 
 
2198	}
2199	return page;
2200}
2201
2202static struct page *do_read_cache_page(struct address_space *mapping,
2203				pgoff_t index,
2204				int (*filler)(void *, struct page *),
2205				void *data,
2206				gfp_t gfp)
2207
2208{
2209	struct page *page;
2210	int err;
2211
2212retry:
2213	page = __read_cache_page(mapping, index, filler, data, gfp);
2214	if (IS_ERR(page))
2215		return page;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2216	if (PageUptodate(page))
2217		goto out;
2218
 
2219	lock_page(page);
 
 
2220	if (!page->mapping) {
2221		unlock_page(page);
2222		page_cache_release(page);
2223		goto retry;
2224	}
 
 
2225	if (PageUptodate(page)) {
2226		unlock_page(page);
2227		goto out;
2228	}
2229	err = filler(data, page);
2230	if (err < 0) {
2231		page_cache_release(page);
2232		return ERR_PTR(err);
2233	} else {
2234		page = wait_on_page_read(page);
2235		if (IS_ERR(page))
2236			return page;
2237	}
 
2238out:
2239	mark_page_accessed(page);
2240	return page;
2241}
2242
2243/**
2244 * read_cache_page - read into page cache, fill it if needed
2245 * @mapping:	the page's address_space
2246 * @index:	the page index
2247 * @filler:	function to perform the read
2248 * @data:	first arg to filler(data, page) function, often left as NULL
2249 *
2250 * Read into the page cache. If a page already exists, and PageUptodate() is
2251 * not set, try to fill the page and wait for it to become unlocked.
2252 *
2253 * If the page does not get brought uptodate, return -EIO.
 
 
2254 */
2255struct page *read_cache_page(struct address_space *mapping,
2256				pgoff_t index,
2257				int (*filler)(void *, struct page *),
2258				void *data)
2259{
2260	return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
 
2261}
2262EXPORT_SYMBOL(read_cache_page);
2263
2264/**
2265 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2266 * @mapping:	the page's address_space
2267 * @index:	the page index
2268 * @gfp:	the page allocator flags to use if allocating
2269 *
2270 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2271 * any new page allocations done using the specified allocation flags.
2272 *
2273 * If the page does not get brought uptodate, return -EIO.
 
 
2274 */
2275struct page *read_cache_page_gfp(struct address_space *mapping,
2276				pgoff_t index,
2277				gfp_t gfp)
2278{
2279	filler_t *filler = (filler_t *)mapping->a_ops->readpage;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2280
2281	return do_read_cache_page(mapping, index, filler, NULL, gfp);
 
 
 
 
 
2282}
2283EXPORT_SYMBOL(read_cache_page_gfp);
2284
2285/*
2286 * Performs necessary checks before doing a write
2287 *
2288 * Can adjust writing position or amount of bytes to write.
2289 * Returns appropriate error code that caller should return or
2290 * zero in case that write should be allowed.
2291 */
2292inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2293{
 
2294	struct inode *inode = file->f_mapping->host;
2295	unsigned long limit = rlimit(RLIMIT_FSIZE);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2296
2297        if (unlikely(*pos < 0))
2298                return -EINVAL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2299
2300	if (!isblk) {
2301		/* FIXME: this is for backwards compatibility with 2.4 */
2302		if (file->f_flags & O_APPEND)
2303                        *pos = i_size_read(inode);
2304
2305		if (limit != RLIM_INFINITY) {
2306			if (*pos >= limit) {
2307				send_sig(SIGXFSZ, current, 0);
2308				return -EFBIG;
2309			}
2310			if (*count > limit - (typeof(limit))*pos) {
2311				*count = limit - (typeof(limit))*pos;
2312			}
2313		}
2314	}
 
 
2315
2316	/*
2317	 * LFS rule
 
 
 
 
2318	 */
2319	if (unlikely(*pos + *count > MAX_NON_LFS &&
2320				!(file->f_flags & O_LARGEFILE))) {
2321		if (*pos >= MAX_NON_LFS) {
2322			return -EFBIG;
2323		}
2324		if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2325			*count = MAX_NON_LFS - (unsigned long)*pos;
2326		}
2327	}
2328
 
 
 
 
 
 
2329	/*
2330	 * Are we about to exceed the fs block limit ?
2331	 *
2332	 * If we have written data it becomes a short write.  If we have
2333	 * exceeded without writing data we send a signal and return EFBIG.
2334	 * Linus frestrict idea will clean these up nicely..
2335	 */
2336	if (likely(!isblk)) {
2337		if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2338			if (*count || *pos > inode->i_sb->s_maxbytes) {
2339				return -EFBIG;
2340			}
2341			/* zero-length writes at ->s_maxbytes are OK */
2342		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2343
2344		if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2345			*count = inode->i_sb->s_maxbytes - *pos;
2346	} else {
2347#ifdef CONFIG_BLOCK
2348		loff_t isize;
2349		if (bdev_read_only(I_BDEV(inode)))
2350			return -EPERM;
2351		isize = i_size_read(inode);
2352		if (*pos >= isize) {
2353			if (*count || *pos > isize)
2354				return -ENOSPC;
2355		}
2356
2357		if (*pos + *count > isize)
2358			*count = isize - *pos;
2359#else
2360		return -EPERM;
2361#endif
2362	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2363	return 0;
2364}
2365EXPORT_SYMBOL(generic_write_checks);
2366
2367int pagecache_write_begin(struct file *file, struct address_space *mapping,
2368				loff_t pos, unsigned len, unsigned flags,
2369				struct page **pagep, void **fsdata)
2370{
2371	const struct address_space_operations *aops = mapping->a_ops;
2372
2373	return aops->write_begin(file, mapping, pos, len, flags,
2374							pagep, fsdata);
2375}
2376EXPORT_SYMBOL(pagecache_write_begin);
2377
2378int pagecache_write_end(struct file *file, struct address_space *mapping,
2379				loff_t pos, unsigned len, unsigned copied,
2380				struct page *page, void *fsdata)
2381{
2382	const struct address_space_operations *aops = mapping->a_ops;
2383
2384	mark_page_accessed(page);
2385	return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2386}
2387EXPORT_SYMBOL(pagecache_write_end);
2388
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2389ssize_t
2390generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2391		unsigned long *nr_segs, loff_t pos,
2392		size_t count, size_t ocount)
2393{
2394	struct file	*file = iocb->ki_filp;
2395	struct address_space *mapping = file->f_mapping;
2396	struct inode	*inode = mapping->host;
 
2397	ssize_t		written;
2398	size_t		write_len;
2399	pgoff_t		end;
2400
2401	if (count != ocount)
2402		*nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2403
2404	write_len = iov_length(iov, *nr_segs);
2405	end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2406
2407	written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2408	if (written)
2409		goto out;
 
 
 
 
 
2410
2411	/*
2412	 * After a write we want buffered reads to be sure to go to disk to get
2413	 * the new data.  We invalidate clean cached page from the region we're
2414	 * about to write.  We do this *before* the write so that we can return
2415	 * without clobbering -EIOCBQUEUED from ->direct_IO().
2416	 */
2417	if (mapping->nrpages) {
2418		written = invalidate_inode_pages2_range(mapping,
2419					pos >> PAGE_CACHE_SHIFT, end);
2420		/*
2421		 * If a page can not be invalidated, return 0 to fall back
2422		 * to buffered write.
2423		 */
2424		if (written) {
2425			if (written == -EBUSY)
2426				return 0;
2427			goto out;
2428		}
2429	}
2430
2431	written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2432
2433	/*
2434	 * Finally, try again to invalidate clean pages which might have been
2435	 * cached by non-direct readahead, or faulted in by get_user_pages()
2436	 * if the source of the write was an mmap'ed region of the file
2437	 * we're writing.  Either one is a pretty crazy thing to do,
2438	 * so we don't support it 100%.  If this invalidation
2439	 * fails, tough, the write still worked...
 
 
 
 
 
 
 
 
 
2440	 */
2441	if (mapping->nrpages) {
2442		invalidate_inode_pages2_range(mapping,
2443					      pos >> PAGE_CACHE_SHIFT, end);
2444	}
2445
2446	if (written > 0) {
2447		pos += written;
 
2448		if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2449			i_size_write(inode, pos);
2450			mark_inode_dirty(inode);
2451		}
2452		iocb->ki_pos = pos;
2453	}
 
2454out:
2455	return written;
2456}
2457EXPORT_SYMBOL(generic_file_direct_write);
2458
2459/*
2460 * Find or create a page at the given pagecache position. Return the locked
2461 * page. This function is specifically for buffered writes.
2462 */
2463struct page *grab_cache_page_write_begin(struct address_space *mapping,
2464					pgoff_t index, unsigned flags)
2465{
2466	int status;
2467	gfp_t gfp_mask;
2468	struct page *page;
2469	gfp_t gfp_notmask = 0;
2470
2471	gfp_mask = mapping_gfp_mask(mapping);
2472	if (mapping_cap_account_dirty(mapping))
2473		gfp_mask |= __GFP_WRITE;
2474	if (flags & AOP_FLAG_NOFS)
2475		gfp_notmask = __GFP_FS;
2476repeat:
2477	page = find_lock_page(mapping, index);
 
2478	if (page)
2479		goto found;
2480
2481	page = __page_cache_alloc(gfp_mask & ~gfp_notmask);
2482	if (!page)
2483		return NULL;
2484	status = add_to_page_cache_lru(page, mapping, index,
2485						GFP_KERNEL & ~gfp_notmask);
2486	if (unlikely(status)) {
2487		page_cache_release(page);
2488		if (status == -EEXIST)
2489			goto repeat;
2490		return NULL;
2491	}
2492found:
2493	wait_for_stable_page(page);
2494	return page;
2495}
2496EXPORT_SYMBOL(grab_cache_page_write_begin);
2497
2498ssize_t generic_perform_write(struct file *file,
2499				struct iov_iter *i, loff_t pos)
2500{
2501	struct address_space *mapping = file->f_mapping;
2502	const struct address_space_operations *a_ops = mapping->a_ops;
2503	long status = 0;
2504	ssize_t written = 0;
2505	unsigned int flags = 0;
2506
2507	/*
2508	 * Copies from kernel address space cannot fail (NFSD is a big user).
2509	 */
2510	if (segment_eq(get_fs(), KERNEL_DS))
2511		flags |= AOP_FLAG_UNINTERRUPTIBLE;
2512
2513	do {
2514		struct page *page;
2515		unsigned long offset;	/* Offset into pagecache page */
2516		unsigned long bytes;	/* Bytes to write to page */
2517		size_t copied;		/* Bytes copied from user */
2518		void *fsdata;
2519
2520		offset = (pos & (PAGE_CACHE_SIZE - 1));
2521		bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2522						iov_iter_count(i));
2523
2524again:
2525		/*
2526		 * Bring in the user page that we will copy from _first_.
2527		 * Otherwise there's a nasty deadlock on copying from the
2528		 * same page as we're writing to, without it being marked
2529		 * up-to-date.
2530		 *
2531		 * Not only is this an optimisation, but it is also required
2532		 * to check that the address is actually valid, when atomic
2533		 * usercopies are used, below.
2534		 */
2535		if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2536			status = -EFAULT;
2537			break;
2538		}
2539
 
 
 
 
 
2540		status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2541						&page, &fsdata);
2542		if (unlikely(status))
2543			break;
2544
2545		if (mapping_writably_mapped(mapping))
2546			flush_dcache_page(page);
2547
2548		copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2549		flush_dcache_page(page);
2550
2551		mark_page_accessed(page);
2552		status = a_ops->write_end(file, mapping, pos, bytes, copied,
2553						page, fsdata);
2554		if (unlikely(status < 0))
2555			break;
2556		copied = status;
2557
2558		cond_resched();
2559
2560		iov_iter_advance(i, copied);
2561		if (unlikely(copied == 0)) {
2562			/*
2563			 * If we were unable to copy any data at all, we must
2564			 * fall back to a single segment length write.
2565			 *
2566			 * If we didn't fallback here, we could livelock
2567			 * because not all segments in the iov can be copied at
2568			 * once without a pagefault.
2569			 */
2570			bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2571						iov_iter_single_seg_count(i));
2572			goto again;
2573		}
2574		pos += copied;
2575		written += copied;
2576
2577		balance_dirty_pages_ratelimited(mapping);
2578		if (fatal_signal_pending(current)) {
2579			status = -EINTR;
2580			break;
2581		}
2582	} while (iov_iter_count(i));
2583
2584	return written ? written : status;
2585}
2586EXPORT_SYMBOL(generic_perform_write);
2587
2588/**
2589 * __generic_file_aio_write - write data to a file
2590 * @iocb:	IO state structure (file, offset, etc.)
2591 * @iov:	vector with data to write
2592 * @nr_segs:	number of segments in the vector
2593 *
2594 * This function does all the work needed for actually writing data to a
2595 * file. It does all basic checks, removes SUID from the file, updates
2596 * modification times and calls proper subroutines depending on whether we
2597 * do direct IO or a standard buffered write.
2598 *
2599 * It expects i_mutex to be grabbed unless we work on a block device or similar
2600 * object which does not need locking at all.
2601 *
2602 * This function does *not* take care of syncing data in case of O_SYNC write.
2603 * A caller has to handle it. This is mainly due to the fact that we want to
2604 * avoid syncing under i_mutex.
 
 
 
 
2605 */
2606ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2607				 unsigned long nr_segs)
2608{
2609	struct file *file = iocb->ki_filp;
2610	struct address_space * mapping = file->f_mapping;
2611	size_t ocount;		/* original count */
2612	size_t count;		/* after file limit checks */
2613	struct inode 	*inode = mapping->host;
2614	loff_t		pos = iocb->ki_pos;
2615	ssize_t		written = 0;
2616	ssize_t		err;
2617	ssize_t		status;
2618	struct iov_iter from;
2619
2620	ocount = 0;
2621	err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2622	if (err)
2623		return err;
2624
2625	count = ocount;
2626
2627	/* We can write back this queue in page reclaim */
2628	current->backing_dev_info = mapping->backing_dev_info;
2629	err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2630	if (err)
2631		goto out;
2632
2633	if (count == 0)
2634		goto out;
2635
2636	err = file_remove_suid(file);
2637	if (err)
2638		goto out;
2639
2640	err = file_update_time(file);
2641	if (err)
2642		goto out;
2643
2644	iov_iter_init(&from, iov, nr_segs, count, 0);
2645
2646	/* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2647	if (unlikely(file->f_flags & O_DIRECT)) {
2648		loff_t endbyte;
2649
2650		written = generic_file_direct_write(iocb, iov, &from.nr_segs, pos,
2651							count, ocount);
2652		if (written < 0 || written == count)
2653			goto out;
2654		iov_iter_advance(&from, written);
2655
 
2656		/*
2657		 * direct-io write to a hole: fall through to buffered I/O
2658		 * for completing the rest of the request.
 
 
 
2659		 */
2660		pos += written;
2661		count -= written;
2662
2663		status = generic_perform_write(file, &from, pos);
2664		/*
2665		 * If generic_perform_write() returned a synchronous error
2666		 * then we want to return the number of bytes which were
2667		 * direct-written, or the error code if that was zero.  Note
2668		 * that this differs from normal direct-io semantics, which
2669		 * will return -EFOO even if some bytes were written.
2670		 */
2671		if (unlikely(status < 0) && !written) {
2672			err = status;
2673			goto out;
2674		}
2675		iocb->ki_pos = pos + status;
2676		/*
2677		 * We need to ensure that the page cache pages are written to
2678		 * disk and invalidated to preserve the expected O_DIRECT
2679		 * semantics.
2680		 */
2681		endbyte = pos + status - 1;
2682		err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
2683		if (err == 0) {
 
2684			written += status;
2685			invalidate_mapping_pages(mapping,
2686						 pos >> PAGE_CACHE_SHIFT,
2687						 endbyte >> PAGE_CACHE_SHIFT);
2688		} else {
2689			/*
2690			 * We don't know how much we wrote, so just return
2691			 * the number of bytes which were direct-written
2692			 */
2693		}
2694	} else {
2695		written = generic_perform_write(file, &from, pos);
2696		if (likely(written >= 0))
2697			iocb->ki_pos = pos + written;
2698	}
2699out:
2700	current->backing_dev_info = NULL;
2701	return written ? written : err;
2702}
2703EXPORT_SYMBOL(__generic_file_aio_write);
2704
2705/**
2706 * generic_file_aio_write - write data to a file
2707 * @iocb:	IO state structure
2708 * @iov:	vector with data to write
2709 * @nr_segs:	number of segments in the vector
2710 * @pos:	position in file where to write
2711 *
2712 * This is a wrapper around __generic_file_aio_write() to be used by most
2713 * filesystems. It takes care of syncing the file in case of O_SYNC file
2714 * and acquires i_mutex as needed.
 
 
 
 
2715 */
2716ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2717		unsigned long nr_segs, loff_t pos)
2718{
2719	struct file *file = iocb->ki_filp;
2720	struct inode *inode = file->f_mapping->host;
2721	ssize_t ret;
2722
2723	BUG_ON(iocb->ki_pos != pos);
 
 
 
 
2724
2725	mutex_lock(&inode->i_mutex);
2726	ret = __generic_file_aio_write(iocb, iov, nr_segs);
2727	mutex_unlock(&inode->i_mutex);
2728
2729	if (ret > 0) {
2730		ssize_t err;
2731
2732		err = generic_write_sync(file, iocb->ki_pos - ret, ret);
2733		if (err < 0)
2734			ret = err;
2735	}
2736	return ret;
2737}
2738EXPORT_SYMBOL(generic_file_aio_write);
2739
2740/**
2741 * try_to_release_page() - release old fs-specific metadata on a page
2742 *
2743 * @page: the page which the kernel is trying to free
2744 * @gfp_mask: memory allocation flags (and I/O mode)
2745 *
2746 * The address_space is to try to release any data against the page
2747 * (presumably at page->private).  If the release was successful, return `1'.
2748 * Otherwise return zero.
2749 *
2750 * This may also be called if PG_fscache is set on a page, indicating that the
2751 * page is known to the local caching routines.
2752 *
2753 * The @gfp_mask argument specifies whether I/O may be performed to release
2754 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2755 *
 
2756 */
2757int try_to_release_page(struct page *page, gfp_t gfp_mask)
2758{
2759	struct address_space * const mapping = page->mapping;
2760
2761	BUG_ON(!PageLocked(page));
2762	if (PageWriteback(page))
2763		return 0;
2764
2765	if (mapping && mapping->a_ops->releasepage)
2766		return mapping->a_ops->releasepage(page, gfp_mask);
2767	return try_to_free_buffers(page);
2768}
2769
2770EXPORT_SYMBOL(try_to_release_page);