Linux Audio

Check our new training course

Loading...
v4.17
 
   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/dax.h>
  15#include <linux/fs.h>
  16#include <linux/sched/signal.h>
  17#include <linux/uaccess.h>
  18#include <linux/capability.h>
  19#include <linux/kernel_stat.h>
  20#include <linux/gfp.h>
  21#include <linux/mm.h>
  22#include <linux/swap.h>
  23#include <linux/mman.h>
  24#include <linux/pagemap.h>
  25#include <linux/file.h>
  26#include <linux/uio.h>
 
  27#include <linux/hash.h>
  28#include <linux/writeback.h>
  29#include <linux/backing-dev.h>
  30#include <linux/pagevec.h>
  31#include <linux/blkdev.h>
  32#include <linux/security.h>
  33#include <linux/cpuset.h>
  34#include <linux/hugetlb.h>
  35#include <linux/memcontrol.h>
  36#include <linux/cleancache.h>
  37#include <linux/shmem_fs.h>
  38#include <linux/rmap.h>
 
 
 
  39#include "internal.h"
  40
  41#define CREATE_TRACE_POINTS
  42#include <trace/events/filemap.h>
  43
  44/*
  45 * FIXME: remove all knowledge of the buffer layer from the core VM
  46 */
  47#include <linux/buffer_head.h> /* for try_to_free_buffers */
  48
  49#include <asm/mman.h>
  50
  51/*
  52 * Shared mappings implemented 30.11.1994. It's not fully working yet,
  53 * though.
  54 *
  55 * Shared mappings now work. 15.8.1995  Bruno.
  56 *
  57 * finished 'unifying' the page and buffer cache and SMP-threaded the
  58 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
  59 *
  60 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
  61 */
  62
  63/*
  64 * Lock ordering:
  65 *
  66 *  ->i_mmap_rwsem		(truncate_pagecache)
  67 *    ->private_lock		(__free_pte->__set_page_dirty_buffers)
  68 *      ->swap_lock		(exclusive_swap_page, others)
  69 *        ->i_pages lock
  70 *
  71 *  ->i_mutex
  72 *    ->i_mmap_rwsem		(truncate->unmap_mapping_range)
  73 *
  74 *  ->mmap_sem
  75 *    ->i_mmap_rwsem
  76 *      ->page_table_lock or pte_lock	(various, mainly in memory.c)
  77 *        ->i_pages lock	(arch-dependent flush_dcache_mmap_lock)
  78 *
  79 *  ->mmap_sem
  80 *    ->lock_page		(access_process_vm)
  81 *
  82 *  ->i_mutex			(generic_perform_write)
  83 *    ->mmap_sem		(fault_in_pages_readable->do_page_fault)
  84 *
  85 *  bdi->wb.list_lock
  86 *    sb_lock			(fs/fs-writeback.c)
  87 *    ->i_pages lock		(__sync_single_inode)
  88 *
  89 *  ->i_mmap_rwsem
  90 *    ->anon_vma.lock		(vma_adjust)
  91 *
  92 *  ->anon_vma.lock
  93 *    ->page_table_lock or pte_lock	(anon_vma_prepare and various)
  94 *
  95 *  ->page_table_lock or pte_lock
  96 *    ->swap_lock		(try_to_unmap_one)
  97 *    ->private_lock		(try_to_unmap_one)
  98 *    ->i_pages lock		(try_to_unmap_one)
  99 *    ->zone_lru_lock(zone)	(follow_page->mark_page_accessed)
 100 *    ->zone_lru_lock(zone)	(check_pte_range->isolate_lru_page)
 101 *    ->private_lock		(page_remove_rmap->set_page_dirty)
 102 *    ->i_pages lock		(page_remove_rmap->set_page_dirty)
 103 *    bdi.wb->list_lock		(page_remove_rmap->set_page_dirty)
 104 *    ->inode->i_lock		(page_remove_rmap->set_page_dirty)
 105 *    ->memcg->move_lock	(page_remove_rmap->lock_page_memcg)
 106 *    bdi.wb->list_lock		(zap_pte_range->set_page_dirty)
 107 *    ->inode->i_lock		(zap_pte_range->set_page_dirty)
 108 *    ->private_lock		(zap_pte_range->__set_page_dirty_buffers)
 109 *
 110 * ->i_mmap_rwsem
 111 *   ->tasklist_lock            (memory_failure, collect_procs_ao)
 112 */
 113
 114static int page_cache_tree_insert(struct address_space *mapping,
 115				  struct page *page, void **shadowp)
 116{
 117	struct radix_tree_node *node;
 118	void **slot;
 119	int error;
 120
 121	error = __radix_tree_create(&mapping->i_pages, page->index, 0,
 122				    &node, &slot);
 123	if (error)
 124		return error;
 125	if (*slot) {
 126		void *p;
 127
 128		p = radix_tree_deref_slot_protected(slot,
 129						    &mapping->i_pages.xa_lock);
 130		if (!radix_tree_exceptional_entry(p))
 131			return -EEXIST;
 132
 133		mapping->nrexceptional--;
 134		if (shadowp)
 135			*shadowp = p;
 136	}
 137	__radix_tree_replace(&mapping->i_pages, node, slot, page,
 138			     workingset_lookup_update(mapping));
 139	mapping->nrpages++;
 140	return 0;
 141}
 142
 143static void page_cache_tree_delete(struct address_space *mapping,
 144				   struct page *page, void *shadow)
 145{
 146	int i, nr;
 
 
 
 147
 148	/* hugetlb pages are represented by one entry in the radix tree */
 149	nr = PageHuge(page) ? 1 : hpage_nr_pages(page);
 
 
 
 150
 151	VM_BUG_ON_PAGE(!PageLocked(page), page);
 152	VM_BUG_ON_PAGE(PageTail(page), page);
 153	VM_BUG_ON_PAGE(nr != 1 && shadow, page);
 154
 155	for (i = 0; i < nr; i++) {
 156		struct radix_tree_node *node;
 157		void **slot;
 158
 159		__radix_tree_lookup(&mapping->i_pages, page->index + i,
 160				    &node, &slot);
 161
 162		VM_BUG_ON_PAGE(!node && nr != 1, page);
 163
 164		radix_tree_clear_tags(&mapping->i_pages, node, slot);
 165		__radix_tree_replace(&mapping->i_pages, node, slot, shadow,
 166				workingset_lookup_update(mapping));
 167	}
 168
 169	page->mapping = NULL;
 170	/* Leave page->index set: truncation lookup relies upon it */
 171
 172	if (shadow) {
 173		mapping->nrexceptional += nr;
 174		/*
 175		 * Make sure the nrexceptional update is committed before
 176		 * the nrpages update so that final truncate racing
 177		 * with reclaim does not see both counters 0 at the
 178		 * same time and miss a shadow entry.
 179		 */
 180		smp_wmb();
 181	}
 182	mapping->nrpages -= nr;
 183}
 184
 185static void unaccount_page_cache_page(struct address_space *mapping,
 186				      struct page *page)
 187{
 188	int nr;
 189
 190	/*
 191	 * if we're uptodate, flush out into the cleancache, otherwise
 192	 * invalidate any existing cleancache entries.  We can't leave
 193	 * stale data around in the cleancache once our page is gone
 194	 */
 195	if (PageUptodate(page) && PageMappedToDisk(page))
 196		cleancache_put_page(page);
 197	else
 198		cleancache_invalidate_page(mapping, page);
 199
 200	VM_BUG_ON_PAGE(PageTail(page), page);
 201	VM_BUG_ON_PAGE(page_mapped(page), page);
 202	if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
 203		int mapcount;
 204
 205		pr_alert("BUG: Bad page cache in process %s  pfn:%05lx\n",
 206			 current->comm, page_to_pfn(page));
 207		dump_page(page, "still mapped when deleted");
 208		dump_stack();
 209		add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
 210
 211		mapcount = page_mapcount(page);
 212		if (mapping_exiting(mapping) &&
 213		    page_count(page) >= mapcount + 2) {
 214			/*
 215			 * All vmas have already been torn down, so it's
 216			 * a good bet that actually the page is unmapped,
 217			 * and we'd prefer not to leak it: if we're wrong,
 218			 * some other bad page check should catch it later.
 219			 */
 220			page_mapcount_reset(page);
 221			page_ref_sub(page, mapcount);
 222		}
 223	}
 224
 225	/* hugetlb pages do not participate in page cache accounting. */
 226	if (PageHuge(page))
 227		return;
 228
 229	nr = hpage_nr_pages(page);
 230
 231	__mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
 232	if (PageSwapBacked(page)) {
 233		__mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
 234		if (PageTransHuge(page))
 235			__dec_node_page_state(page, NR_SHMEM_THPS);
 236	} else {
 237		VM_BUG_ON_PAGE(PageTransHuge(page), page);
 
 238	}
 239
 240	/*
 241	 * At this point page must be either written or cleaned by
 242	 * truncate.  Dirty page here signals a bug and loss of
 243	 * unwritten data.
 244	 *
 245	 * This fixes dirty accounting after removing the page entirely
 246	 * but leaves PageDirty set: it has no effect for truncated
 247	 * page and anyway will be cleared before returning page into
 248	 * buddy allocator.
 249	 */
 250	if (WARN_ON_ONCE(PageDirty(page)))
 251		account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
 252}
 253
 254/*
 255 * Delete a page from the page cache and free it. Caller has to make
 256 * sure the page is locked and that nobody else uses it - or that usage
 257 * is safe.  The caller must hold the i_pages lock.
 258 */
 259void __delete_from_page_cache(struct page *page, void *shadow)
 260{
 261	struct address_space *mapping = page->mapping;
 262
 263	trace_mm_filemap_delete_from_page_cache(page);
 264
 265	unaccount_page_cache_page(mapping, page);
 266	page_cache_tree_delete(mapping, page, shadow);
 267}
 268
 269static void page_cache_free_page(struct address_space *mapping,
 270				struct page *page)
 271{
 272	void (*freepage)(struct page *);
 273
 274	freepage = mapping->a_ops->freepage;
 275	if (freepage)
 276		freepage(page);
 277
 278	if (PageTransHuge(page) && !PageHuge(page)) {
 279		page_ref_sub(page, HPAGE_PMD_NR);
 280		VM_BUG_ON_PAGE(page_count(page) <= 0, page);
 281	} else {
 282		put_page(page);
 283	}
 284}
 285
 286/**
 287 * delete_from_page_cache - delete page from page cache
 288 * @page: the page which the kernel is trying to remove from page cache
 289 *
 290 * This must be called only on pages that have been verified to be in the page
 291 * cache and locked.  It will never put the page into the free list, the caller
 292 * has a reference on the page.
 293 */
 294void delete_from_page_cache(struct page *page)
 295{
 296	struct address_space *mapping = page_mapping(page);
 297	unsigned long flags;
 298
 299	BUG_ON(!PageLocked(page));
 300	xa_lock_irqsave(&mapping->i_pages, flags);
 301	__delete_from_page_cache(page, NULL);
 302	xa_unlock_irqrestore(&mapping->i_pages, flags);
 303
 304	page_cache_free_page(mapping, page);
 305}
 306EXPORT_SYMBOL(delete_from_page_cache);
 307
 308/*
 309 * page_cache_tree_delete_batch - delete several pages from page cache
 310 * @mapping: the mapping to which pages belong
 311 * @pvec: pagevec with pages to delete
 312 *
 313 * The function walks over mapping->i_pages and removes pages passed in @pvec
 314 * from the mapping. The function expects @pvec to be sorted by page index.
 
 315 * It tolerates holes in @pvec (mapping entries at those indices are not
 316 * modified). The function expects only THP head pages to be present in the
 317 * @pvec and takes care to delete all corresponding tail pages from the
 318 * mapping as well.
 319 *
 320 * The function expects the i_pages lock to be held.
 321 */
 322static void
 323page_cache_tree_delete_batch(struct address_space *mapping,
 324			     struct pagevec *pvec)
 325{
 326	struct radix_tree_iter iter;
 327	void **slot;
 328	int total_pages = 0;
 329	int i = 0, tail_pages = 0;
 330	struct page *page;
 331	pgoff_t start;
 332
 333	start = pvec->pages[0]->index;
 334	radix_tree_for_each_slot(slot, &mapping->i_pages, &iter, start) {
 335		if (i >= pagevec_count(pvec) && !tail_pages)
 336			break;
 337		page = radix_tree_deref_slot_protected(slot,
 338						       &mapping->i_pages.xa_lock);
 339		if (radix_tree_exceptional_entry(page))
 340			continue;
 341		if (!tail_pages) {
 342			/*
 343			 * Some page got inserted in our range? Skip it. We
 344			 * have our pages locked so they are protected from
 345			 * being removed.
 346			 */
 347			if (page != pvec->pages[i])
 348				continue;
 349			WARN_ON_ONCE(!PageLocked(page));
 350			if (PageTransHuge(page) && !PageHuge(page))
 351				tail_pages = HPAGE_PMD_NR - 1;
 
 
 
 
 
 352			page->mapping = NULL;
 353			/*
 354			 * Leave page->index set: truncation lookup relies
 355			 * upon it
 356			 */
 
 
 
 
 357			i++;
 358		} else {
 359			tail_pages--;
 360		}
 361		radix_tree_clear_tags(&mapping->i_pages, iter.node, slot);
 362		__radix_tree_replace(&mapping->i_pages, iter.node, slot, NULL,
 363				workingset_lookup_update(mapping));
 364		total_pages++;
 365	}
 366	mapping->nrpages -= total_pages;
 367}
 368
 369void delete_from_page_cache_batch(struct address_space *mapping,
 370				  struct pagevec *pvec)
 371{
 372	int i;
 373	unsigned long flags;
 374
 375	if (!pagevec_count(pvec))
 376		return;
 377
 378	xa_lock_irqsave(&mapping->i_pages, flags);
 379	for (i = 0; i < pagevec_count(pvec); i++) {
 380		trace_mm_filemap_delete_from_page_cache(pvec->pages[i]);
 381
 382		unaccount_page_cache_page(mapping, pvec->pages[i]);
 383	}
 384	page_cache_tree_delete_batch(mapping, pvec);
 385	xa_unlock_irqrestore(&mapping->i_pages, flags);
 386
 387	for (i = 0; i < pagevec_count(pvec); i++)
 388		page_cache_free_page(mapping, pvec->pages[i]);
 389}
 390
 391int filemap_check_errors(struct address_space *mapping)
 392{
 393	int ret = 0;
 394	/* Check for outstanding write errors */
 395	if (test_bit(AS_ENOSPC, &mapping->flags) &&
 396	    test_and_clear_bit(AS_ENOSPC, &mapping->flags))
 397		ret = -ENOSPC;
 398	if (test_bit(AS_EIO, &mapping->flags) &&
 399	    test_and_clear_bit(AS_EIO, &mapping->flags))
 400		ret = -EIO;
 401	return ret;
 402}
 403EXPORT_SYMBOL(filemap_check_errors);
 404
 405static int filemap_check_and_keep_errors(struct address_space *mapping)
 406{
 407	/* Check for outstanding write errors */
 408	if (test_bit(AS_EIO, &mapping->flags))
 409		return -EIO;
 410	if (test_bit(AS_ENOSPC, &mapping->flags))
 411		return -ENOSPC;
 412	return 0;
 413}
 414
 415/**
 416 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
 417 * @mapping:	address space structure to write
 418 * @start:	offset in bytes where the range starts
 419 * @end:	offset in bytes where the range ends (inclusive)
 420 * @sync_mode:	enable synchronous operation
 421 *
 422 * Start writeback against all of a mapping's dirty pages that lie
 423 * within the byte offsets <start, end> inclusive.
 424 *
 425 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
 426 * opposed to a regular memory cleansing writeback.  The difference between
 427 * these two operations is that if a dirty page/buffer is encountered, it must
 428 * be waited upon, and not just skipped over.
 
 
 429 */
 430int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
 431				loff_t end, int sync_mode)
 432{
 433	int ret;
 434	struct writeback_control wbc = {
 435		.sync_mode = sync_mode,
 436		.nr_to_write = LONG_MAX,
 437		.range_start = start,
 438		.range_end = end,
 439	};
 440
 441	if (!mapping_cap_writeback_dirty(mapping))
 
 442		return 0;
 443
 444	wbc_attach_fdatawrite_inode(&wbc, mapping->host);
 445	ret = do_writepages(mapping, &wbc);
 446	wbc_detach_inode(&wbc);
 447	return ret;
 448}
 449
 450static inline int __filemap_fdatawrite(struct address_space *mapping,
 451	int sync_mode)
 452{
 453	return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
 454}
 455
 456int filemap_fdatawrite(struct address_space *mapping)
 457{
 458	return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
 459}
 460EXPORT_SYMBOL(filemap_fdatawrite);
 461
 462int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
 463				loff_t end)
 464{
 465	return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
 466}
 467EXPORT_SYMBOL(filemap_fdatawrite_range);
 468
 469/**
 470 * filemap_flush - mostly a non-blocking flush
 471 * @mapping:	target address_space
 472 *
 473 * This is a mostly non-blocking flush.  Not suitable for data-integrity
 474 * purposes - I/O may not be started against all dirty pages.
 
 
 475 */
 476int filemap_flush(struct address_space *mapping)
 477{
 478	return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
 479}
 480EXPORT_SYMBOL(filemap_flush);
 481
 482/**
 483 * filemap_range_has_page - check if a page exists in range.
 484 * @mapping:           address space within which to check
 485 * @start_byte:        offset in bytes where the range starts
 486 * @end_byte:          offset in bytes where the range ends (inclusive)
 487 *
 488 * Find at least one page in the range supplied, usually used to check if
 489 * direct writing in this range will trigger a writeback.
 
 
 
 490 */
 491bool filemap_range_has_page(struct address_space *mapping,
 492			   loff_t start_byte, loff_t end_byte)
 493{
 494	pgoff_t index = start_byte >> PAGE_SHIFT;
 495	pgoff_t end = end_byte >> PAGE_SHIFT;
 496	struct page *page;
 
 
 497
 498	if (end_byte < start_byte)
 499		return false;
 500
 501	if (mapping->nrpages == 0)
 502		return false;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 503
 504	if (!find_get_pages_range(mapping, &index, end, 1, &page))
 505		return false;
 506	put_page(page);
 507	return true;
 508}
 509EXPORT_SYMBOL(filemap_range_has_page);
 510
 511static void __filemap_fdatawait_range(struct address_space *mapping,
 512				     loff_t start_byte, loff_t end_byte)
 513{
 514	pgoff_t index = start_byte >> PAGE_SHIFT;
 515	pgoff_t end = end_byte >> PAGE_SHIFT;
 516	struct pagevec pvec;
 517	int nr_pages;
 518
 519	if (end_byte < start_byte)
 520		return;
 521
 522	pagevec_init(&pvec);
 523	while (index <= end) {
 524		unsigned i;
 525
 526		nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
 527				end, PAGECACHE_TAG_WRITEBACK);
 528		if (!nr_pages)
 529			break;
 530
 531		for (i = 0; i < nr_pages; i++) {
 532			struct page *page = pvec.pages[i];
 533
 534			wait_on_page_writeback(page);
 535			ClearPageError(page);
 536		}
 537		pagevec_release(&pvec);
 538		cond_resched();
 539	}
 540}
 541
 542/**
 543 * filemap_fdatawait_range - wait for writeback to complete
 544 * @mapping:		address space structure to wait for
 545 * @start_byte:		offset in bytes where the range starts
 546 * @end_byte:		offset in bytes where the range ends (inclusive)
 547 *
 548 * Walk the list of under-writeback pages of the given address space
 549 * in the given range and wait for all of them.  Check error status of
 550 * the address space and return it.
 551 *
 552 * Since the error status of the address space is cleared by this function,
 553 * callers are responsible for checking the return value and handling and/or
 554 * reporting the error.
 
 
 555 */
 556int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
 557			    loff_t end_byte)
 558{
 559	__filemap_fdatawait_range(mapping, start_byte, end_byte);
 560	return filemap_check_errors(mapping);
 561}
 562EXPORT_SYMBOL(filemap_fdatawait_range);
 563
 564/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 565 * file_fdatawait_range - wait for writeback to complete
 566 * @file:		file pointing to address space structure to wait for
 567 * @start_byte:		offset in bytes where the range starts
 568 * @end_byte:		offset in bytes where the range ends (inclusive)
 569 *
 570 * Walk the list of under-writeback pages of the address space that file
 571 * refers to, in the given range and wait for all of them.  Check error
 572 * status of the address space vs. the file->f_wb_err cursor and return it.
 573 *
 574 * Since the error status of the file is advanced by this function,
 575 * callers are responsible for checking the return value and handling and/or
 576 * reporting the error.
 
 
 577 */
 578int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
 579{
 580	struct address_space *mapping = file->f_mapping;
 581
 582	__filemap_fdatawait_range(mapping, start_byte, end_byte);
 583	return file_check_and_advance_wb_err(file);
 584}
 585EXPORT_SYMBOL(file_fdatawait_range);
 586
 587/**
 588 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
 589 * @mapping: address space structure to wait for
 590 *
 591 * Walk the list of under-writeback pages of the given address space
 592 * and wait for all of them.  Unlike filemap_fdatawait(), this function
 593 * does not clear error status of the address space.
 594 *
 595 * Use this function if callers don't handle errors themselves.  Expected
 596 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
 597 * fsfreeze(8)
 
 
 598 */
 599int filemap_fdatawait_keep_errors(struct address_space *mapping)
 600{
 601	__filemap_fdatawait_range(mapping, 0, LLONG_MAX);
 602	return filemap_check_and_keep_errors(mapping);
 603}
 604EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
 605
 
 606static bool mapping_needs_writeback(struct address_space *mapping)
 607{
 608	return (!dax_mapping(mapping) && mapping->nrpages) ||
 609	    (dax_mapping(mapping) && mapping->nrexceptional);
 
 
 610}
 611
 612int filemap_write_and_wait(struct address_space *mapping)
 613{
 614	int err = 0;
 615
 616	if (mapping_needs_writeback(mapping)) {
 617		err = filemap_fdatawrite(mapping);
 618		/*
 619		 * Even if the above returned error, the pages may be
 620		 * written partially (e.g. -ENOSPC), so we wait for it.
 621		 * But the -EIO is special case, it may indicate the worst
 622		 * thing (e.g. bug) happened, so we avoid waiting for it.
 623		 */
 624		if (err != -EIO) {
 625			int err2 = filemap_fdatawait(mapping);
 626			if (!err)
 627				err = err2;
 628		} else {
 629			/* Clear any previously stored errors */
 630			filemap_check_errors(mapping);
 631		}
 632	} else {
 633		err = filemap_check_errors(mapping);
 634	}
 635	return err;
 636}
 637EXPORT_SYMBOL(filemap_write_and_wait);
 638
 639/**
 640 * filemap_write_and_wait_range - write out & wait on a file range
 641 * @mapping:	the address_space for the pages
 642 * @lstart:	offset in bytes where the range starts
 643 * @lend:	offset in bytes where the range ends (inclusive)
 644 *
 645 * Write out and wait upon file offsets lstart->lend, inclusive.
 646 *
 647 * Note that @lend is inclusive (describes the last byte to be written) so
 648 * that this function can be used to write to the very end-of-file (end = -1).
 
 
 649 */
 650int filemap_write_and_wait_range(struct address_space *mapping,
 651				 loff_t lstart, loff_t lend)
 652{
 653	int err = 0;
 654
 655	if (mapping_needs_writeback(mapping)) {
 656		err = __filemap_fdatawrite_range(mapping, lstart, lend,
 657						 WB_SYNC_ALL);
 658		/* See comment of filemap_write_and_wait() */
 659		if (err != -EIO) {
 660			int err2 = filemap_fdatawait_range(mapping,
 661						lstart, lend);
 662			if (!err)
 663				err = err2;
 664		} else {
 665			/* Clear any previously stored errors */
 666			filemap_check_errors(mapping);
 667		}
 668	} else {
 669		err = filemap_check_errors(mapping);
 670	}
 671	return err;
 672}
 673EXPORT_SYMBOL(filemap_write_and_wait_range);
 674
 675void __filemap_set_wb_err(struct address_space *mapping, int err)
 676{
 677	errseq_t eseq = errseq_set(&mapping->wb_err, err);
 678
 679	trace_filemap_set_wb_err(mapping, eseq);
 680}
 681EXPORT_SYMBOL(__filemap_set_wb_err);
 682
 683/**
 684 * file_check_and_advance_wb_err - report wb error (if any) that was previously
 685 * 				   and advance wb_err to current one
 686 * @file: struct file on which the error is being reported
 687 *
 688 * When userland calls fsync (or something like nfsd does the equivalent), we
 689 * want to report any writeback errors that occurred since the last fsync (or
 690 * since the file was opened if there haven't been any).
 691 *
 692 * Grab the wb_err from the mapping. If it matches what we have in the file,
 693 * then just quickly return 0. The file is all caught up.
 694 *
 695 * If it doesn't match, then take the mapping value, set the "seen" flag in
 696 * it and try to swap it into place. If it works, or another task beat us
 697 * to it with the new value, then update the f_wb_err and return the error
 698 * portion. The error at this point must be reported via proper channels
 699 * (a'la fsync, or NFS COMMIT operation, etc.).
 700 *
 701 * While we handle mapping->wb_err with atomic operations, the f_wb_err
 702 * value is protected by the f_lock since we must ensure that it reflects
 703 * the latest value swapped in for this file descriptor.
 
 
 704 */
 705int file_check_and_advance_wb_err(struct file *file)
 706{
 707	int err = 0;
 708	errseq_t old = READ_ONCE(file->f_wb_err);
 709	struct address_space *mapping = file->f_mapping;
 710
 711	/* Locklessly handle the common case where nothing has changed */
 712	if (errseq_check(&mapping->wb_err, old)) {
 713		/* Something changed, must use slow path */
 714		spin_lock(&file->f_lock);
 715		old = file->f_wb_err;
 716		err = errseq_check_and_advance(&mapping->wb_err,
 717						&file->f_wb_err);
 718		trace_file_check_and_advance_wb_err(file, old);
 719		spin_unlock(&file->f_lock);
 720	}
 721
 722	/*
 723	 * We're mostly using this function as a drop in replacement for
 724	 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
 725	 * that the legacy code would have had on these flags.
 726	 */
 727	clear_bit(AS_EIO, &mapping->flags);
 728	clear_bit(AS_ENOSPC, &mapping->flags);
 729	return err;
 730}
 731EXPORT_SYMBOL(file_check_and_advance_wb_err);
 732
 733/**
 734 * file_write_and_wait_range - write out & wait on a file range
 735 * @file:	file pointing to address_space with pages
 736 * @lstart:	offset in bytes where the range starts
 737 * @lend:	offset in bytes where the range ends (inclusive)
 738 *
 739 * Write out and wait upon file offsets lstart->lend, inclusive.
 740 *
 741 * Note that @lend is inclusive (describes the last byte to be written) so
 742 * that this function can be used to write to the very end-of-file (end = -1).
 743 *
 744 * After writing out and waiting on the data, we check and advance the
 745 * f_wb_err cursor to the latest value, and return any errors detected there.
 
 
 746 */
 747int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
 748{
 749	int err = 0, err2;
 750	struct address_space *mapping = file->f_mapping;
 751
 752	if (mapping_needs_writeback(mapping)) {
 753		err = __filemap_fdatawrite_range(mapping, lstart, lend,
 754						 WB_SYNC_ALL);
 755		/* See comment of filemap_write_and_wait() */
 756		if (err != -EIO)
 757			__filemap_fdatawait_range(mapping, lstart, lend);
 758	}
 759	err2 = file_check_and_advance_wb_err(file);
 760	if (!err)
 761		err = err2;
 762	return err;
 763}
 764EXPORT_SYMBOL(file_write_and_wait_range);
 765
 766/**
 767 * replace_page_cache_page - replace a pagecache page with a new one
 768 * @old:	page to be replaced
 769 * @new:	page to replace with
 770 * @gfp_mask:	allocation mode
 771 *
 772 * This function replaces a page in the pagecache with a new one.  On
 773 * success it acquires the pagecache reference for the new page and
 774 * drops it for the old page.  Both the old and new pages must be
 775 * locked.  This function does not add the new page to the LRU, the
 776 * caller must do that.
 777 *
 778 * The remove + add is atomic.  The only way this function can fail is
 779 * memory allocation failure.
 
 780 */
 781int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
 782{
 783	int error;
 
 
 
 
 784
 785	VM_BUG_ON_PAGE(!PageLocked(old), old);
 786	VM_BUG_ON_PAGE(!PageLocked(new), new);
 787	VM_BUG_ON_PAGE(new->mapping, new);
 788
 789	error = radix_tree_preload(gfp_mask & GFP_RECLAIM_MASK);
 790	if (!error) {
 791		struct address_space *mapping = old->mapping;
 792		void (*freepage)(struct page *);
 793		unsigned long flags;
 794
 795		pgoff_t offset = old->index;
 796		freepage = mapping->a_ops->freepage;
 797
 798		get_page(new);
 799		new->mapping = mapping;
 800		new->index = offset;
 801
 802		xa_lock_irqsave(&mapping->i_pages, flags);
 803		__delete_from_page_cache(old, NULL);
 804		error = page_cache_tree_insert(mapping, new, NULL);
 805		BUG_ON(error);
 806
 807		/*
 808		 * hugetlb pages do not participate in page cache accounting.
 809		 */
 810		if (!PageHuge(new))
 811			__inc_node_page_state(new, NR_FILE_PAGES);
 812		if (PageSwapBacked(new))
 813			__inc_node_page_state(new, NR_SHMEM);
 814		xa_unlock_irqrestore(&mapping->i_pages, flags);
 815		mem_cgroup_migrate(old, new);
 816		radix_tree_preload_end();
 817		if (freepage)
 818			freepage(old);
 819		put_page(old);
 820	}
 821
 822	return error;
 823}
 824EXPORT_SYMBOL_GPL(replace_page_cache_page);
 825
 826static int __add_to_page_cache_locked(struct page *page,
 827				      struct address_space *mapping,
 828				      pgoff_t offset, gfp_t gfp_mask,
 829				      void **shadowp)
 830{
 
 831	int huge = PageHuge(page);
 832	struct mem_cgroup *memcg;
 833	int error;
 
 834
 835	VM_BUG_ON_PAGE(!PageLocked(page), page);
 836	VM_BUG_ON_PAGE(PageSwapBacked(page), page);
 
 837
 838	if (!huge) {
 839		error = mem_cgroup_try_charge(page, current->mm,
 840					      gfp_mask, &memcg, false);
 841		if (error)
 842			return error;
 843	}
 844
 845	error = radix_tree_maybe_preload(gfp_mask & GFP_RECLAIM_MASK);
 846	if (error) {
 847		if (!huge)
 848			mem_cgroup_cancel_charge(page, memcg, false);
 849		return error;
 850	}
 851
 852	get_page(page);
 853	page->mapping = mapping;
 854	page->index = offset;
 855
 856	xa_lock_irq(&mapping->i_pages);
 857	error = page_cache_tree_insert(mapping, page, shadowp);
 858	radix_tree_preload_end();
 859	if (unlikely(error))
 860		goto err_insert;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 861
 862	/* hugetlb pages do not participate in page cache accounting. */
 863	if (!huge)
 864		__inc_node_page_state(page, NR_FILE_PAGES);
 865	xa_unlock_irq(&mapping->i_pages);
 866	if (!huge)
 867		mem_cgroup_commit_charge(page, memcg, false, false);
 868	trace_mm_filemap_add_to_page_cache(page);
 869	return 0;
 870err_insert:
 871	page->mapping = NULL;
 872	/* Leave page->index set: truncation relies upon it */
 873	xa_unlock_irq(&mapping->i_pages);
 874	if (!huge)
 875		mem_cgroup_cancel_charge(page, memcg, false);
 876	put_page(page);
 877	return error;
 878}
 
 879
 880/**
 881 * add_to_page_cache_locked - add a locked page to the pagecache
 882 * @page:	page to add
 883 * @mapping:	the page's address_space
 884 * @offset:	page index
 885 * @gfp_mask:	page allocation mode
 886 *
 887 * This function is used to add a page to the pagecache. It must be locked.
 888 * This function does not add the page to the LRU.  The caller must do that.
 
 
 889 */
 890int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
 891		pgoff_t offset, gfp_t gfp_mask)
 892{
 893	return __add_to_page_cache_locked(page, mapping, offset,
 894					  gfp_mask, NULL);
 895}
 896EXPORT_SYMBOL(add_to_page_cache_locked);
 897
 898int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
 899				pgoff_t offset, gfp_t gfp_mask)
 900{
 901	void *shadow = NULL;
 902	int ret;
 903
 904	__SetPageLocked(page);
 905	ret = __add_to_page_cache_locked(page, mapping, offset,
 906					 gfp_mask, &shadow);
 907	if (unlikely(ret))
 908		__ClearPageLocked(page);
 909	else {
 910		/*
 911		 * The page might have been evicted from cache only
 912		 * recently, in which case it should be activated like
 913		 * any other repeatedly accessed page.
 914		 * The exception is pages getting rewritten; evicting other
 915		 * data from the working set, only to cache data that will
 916		 * get overwritten with something else, is a waste of memory.
 917		 */
 918		if (!(gfp_mask & __GFP_WRITE) &&
 919		    shadow && workingset_refault(shadow)) {
 920			SetPageActive(page);
 921			workingset_activation(page);
 922		} else
 923			ClearPageActive(page);
 924		lru_cache_add(page);
 925	}
 926	return ret;
 927}
 928EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
 929
 930#ifdef CONFIG_NUMA
 931struct page *__page_cache_alloc(gfp_t gfp)
 932{
 933	int n;
 934	struct page *page;
 935
 936	if (cpuset_do_page_mem_spread()) {
 937		unsigned int cpuset_mems_cookie;
 938		do {
 939			cpuset_mems_cookie = read_mems_allowed_begin();
 940			n = cpuset_mem_spread_node();
 941			page = __alloc_pages_node(n, gfp, 0);
 942		} while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
 943
 944		return page;
 945	}
 946	return alloc_pages(gfp, 0);
 947}
 948EXPORT_SYMBOL(__page_cache_alloc);
 949#endif
 950
 951/*
 952 * In order to wait for pages to become available there must be
 953 * waitqueues associated with pages. By using a hash table of
 954 * waitqueues where the bucket discipline is to maintain all
 955 * waiters on the same queue and wake all when any of the pages
 956 * become available, and for the woken contexts to check to be
 957 * sure the appropriate page became available, this saves space
 958 * at a cost of "thundering herd" phenomena during rare hash
 959 * collisions.
 960 */
 961#define PAGE_WAIT_TABLE_BITS 8
 962#define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
 963static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
 964
 965static wait_queue_head_t *page_waitqueue(struct page *page)
 966{
 967	return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
 968}
 969
 970void __init pagecache_init(void)
 971{
 972	int i;
 973
 974	for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
 975		init_waitqueue_head(&page_wait_table[i]);
 976
 977	page_writeback_init();
 978}
 979
 980/* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */
 981struct wait_page_key {
 982	struct page *page;
 983	int bit_nr;
 984	int page_match;
 985};
 986
 987struct wait_page_queue {
 988	struct page *page;
 989	int bit_nr;
 990	wait_queue_entry_t wait;
 991};
 992
 993static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
 994{
 995	struct wait_page_key *key = arg;
 996	struct wait_page_queue *wait_page
 997		= container_of(wait, struct wait_page_queue, wait);
 998
 999	if (wait_page->page != key->page)
1000	       return 0;
1001	key->page_match = 1;
1002
1003	if (wait_page->bit_nr != key->bit_nr)
1004		return 0;
1005
1006	/* Stop walking if it's locked */
 
 
 
 
 
 
 
1007	if (test_bit(key->bit_nr, &key->page->flags))
1008		return -1;
1009
1010	return autoremove_wake_function(wait, mode, sync, key);
1011}
1012
1013static void wake_up_page_bit(struct page *page, int bit_nr)
1014{
1015	wait_queue_head_t *q = page_waitqueue(page);
1016	struct wait_page_key key;
1017	unsigned long flags;
1018	wait_queue_entry_t bookmark;
1019
1020	key.page = page;
1021	key.bit_nr = bit_nr;
1022	key.page_match = 0;
1023
1024	bookmark.flags = 0;
1025	bookmark.private = NULL;
1026	bookmark.func = NULL;
1027	INIT_LIST_HEAD(&bookmark.entry);
1028
1029	spin_lock_irqsave(&q->lock, flags);
1030	__wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1031
1032	while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1033		/*
1034		 * Take a breather from holding the lock,
1035		 * allow pages that finish wake up asynchronously
1036		 * to acquire the lock and remove themselves
1037		 * from wait queue
1038		 */
1039		spin_unlock_irqrestore(&q->lock, flags);
1040		cpu_relax();
1041		spin_lock_irqsave(&q->lock, flags);
1042		__wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1043	}
1044
1045	/*
1046	 * It is possible for other pages to have collided on the waitqueue
1047	 * hash, so in that case check for a page match. That prevents a long-
1048	 * term waiter
1049	 *
1050	 * It is still possible to miss a case here, when we woke page waiters
1051	 * and removed them from the waitqueue, but there are still other
1052	 * page waiters.
1053	 */
1054	if (!waitqueue_active(q) || !key.page_match) {
1055		ClearPageWaiters(page);
1056		/*
1057		 * It's possible to miss clearing Waiters here, when we woke
1058		 * our page waiters, but the hashed waitqueue has waiters for
1059		 * other pages on it.
1060		 *
1061		 * That's okay, it's a rare case. The next waker will clear it.
1062		 */
1063	}
1064	spin_unlock_irqrestore(&q->lock, flags);
1065}
1066
1067static void wake_up_page(struct page *page, int bit)
1068{
1069	if (!PageWaiters(page))
1070		return;
1071	wake_up_page_bit(page, bit);
1072}
1073
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1074static inline int wait_on_page_bit_common(wait_queue_head_t *q,
1075		struct page *page, int bit_nr, int state, bool lock)
1076{
1077	struct wait_page_queue wait_page;
1078	wait_queue_entry_t *wait = &wait_page.wait;
 
 
 
 
1079	int ret = 0;
1080
 
 
 
 
 
 
 
 
 
 
1081	init_wait(wait);
1082	wait->flags = lock ? WQ_FLAG_EXCLUSIVE : 0;
1083	wait->func = wake_page_function;
1084	wait_page.page = page;
1085	wait_page.bit_nr = bit_nr;
1086
1087	for (;;) {
1088		spin_lock_irq(&q->lock);
1089
1090		if (likely(list_empty(&wait->entry))) {
1091			__add_wait_queue_entry_tail(q, wait);
1092			SetPageWaiters(page);
1093		}
1094
1095		set_current_state(state);
1096
1097		spin_unlock_irq(&q->lock);
1098
1099		if (likely(test_bit(bit_nr, &page->flags))) {
 
 
 
 
1100			io_schedule();
1101		}
1102
1103		if (lock) {
1104			if (!test_and_set_bit_lock(bit_nr, &page->flags))
1105				break;
1106		} else {
1107			if (!test_bit(bit_nr, &page->flags))
1108				break;
1109		}
1110
1111		if (unlikely(signal_pending_state(state, current))) {
1112			ret = -EINTR;
1113			break;
1114		}
 
 
 
 
 
 
 
 
 
 
 
1115	}
1116
1117	finish_wait(q, wait);
1118
 
 
 
 
 
 
1119	/*
1120	 * A signal could leave PageWaiters set. Clearing it here if
1121	 * !waitqueue_active would be possible (by open-coding finish_wait),
1122	 * but still fail to catch it in the case of wait hash collision. We
1123	 * already can fail to clear wait hash collision cases, so don't
1124	 * bother with signals either.
1125	 */
1126
1127	return ret;
1128}
1129
1130void wait_on_page_bit(struct page *page, int bit_nr)
1131{
1132	wait_queue_head_t *q = page_waitqueue(page);
1133	wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, false);
1134}
1135EXPORT_SYMBOL(wait_on_page_bit);
1136
1137int wait_on_page_bit_killable(struct page *page, int bit_nr)
1138{
1139	wait_queue_head_t *q = page_waitqueue(page);
1140	return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, false);
1141}
1142EXPORT_SYMBOL(wait_on_page_bit_killable);
1143
1144/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1145 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1146 * @page: Page defining the wait queue of interest
1147 * @waiter: Waiter to add to the queue
1148 *
1149 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1150 */
1151void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
1152{
1153	wait_queue_head_t *q = page_waitqueue(page);
1154	unsigned long flags;
1155
1156	spin_lock_irqsave(&q->lock, flags);
1157	__add_wait_queue_entry_tail(q, waiter);
1158	SetPageWaiters(page);
1159	spin_unlock_irqrestore(&q->lock, flags);
1160}
1161EXPORT_SYMBOL_GPL(add_page_wait_queue);
1162
1163#ifndef clear_bit_unlock_is_negative_byte
1164
1165/*
1166 * PG_waiters is the high bit in the same byte as PG_lock.
1167 *
1168 * On x86 (and on many other architectures), we can clear PG_lock and
1169 * test the sign bit at the same time. But if the architecture does
1170 * not support that special operation, we just do this all by hand
1171 * instead.
1172 *
1173 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1174 * being cleared, but a memory barrier should be unneccssary since it is
1175 * in the same byte as PG_locked.
1176 */
1177static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1178{
1179	clear_bit_unlock(nr, mem);
1180	/* smp_mb__after_atomic(); */
1181	return test_bit(PG_waiters, mem);
1182}
1183
1184#endif
1185
1186/**
1187 * unlock_page - unlock a locked page
1188 * @page: the page
1189 *
1190 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
1191 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1192 * mechanism between PageLocked pages and PageWriteback pages is shared.
1193 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1194 *
1195 * Note that this depends on PG_waiters being the sign bit in the byte
1196 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1197 * clear the PG_locked bit and test PG_waiters at the same time fairly
1198 * portably (architectures that do LL/SC can test any bit, while x86 can
1199 * test the sign bit).
1200 */
1201void unlock_page(struct page *page)
1202{
1203	BUILD_BUG_ON(PG_waiters != 7);
1204	page = compound_head(page);
1205	VM_BUG_ON_PAGE(!PageLocked(page), page);
1206	if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
1207		wake_up_page_bit(page, PG_locked);
1208}
1209EXPORT_SYMBOL(unlock_page);
1210
1211/**
1212 * end_page_writeback - end writeback against a page
1213 * @page: the page
1214 */
1215void end_page_writeback(struct page *page)
1216{
1217	/*
1218	 * TestClearPageReclaim could be used here but it is an atomic
1219	 * operation and overkill in this particular case. Failing to
1220	 * shuffle a page marked for immediate reclaim is too mild to
1221	 * justify taking an atomic operation penalty at the end of
1222	 * ever page writeback.
1223	 */
1224	if (PageReclaim(page)) {
1225		ClearPageReclaim(page);
1226		rotate_reclaimable_page(page);
1227	}
1228
1229	if (!test_clear_page_writeback(page))
1230		BUG();
1231
1232	smp_mb__after_atomic();
1233	wake_up_page(page, PG_writeback);
1234}
1235EXPORT_SYMBOL(end_page_writeback);
1236
1237/*
1238 * After completing I/O on a page, call this routine to update the page
1239 * flags appropriately
1240 */
1241void page_endio(struct page *page, bool is_write, int err)
1242{
1243	if (!is_write) {
1244		if (!err) {
1245			SetPageUptodate(page);
1246		} else {
1247			ClearPageUptodate(page);
1248			SetPageError(page);
1249		}
1250		unlock_page(page);
1251	} else {
1252		if (err) {
1253			struct address_space *mapping;
1254
1255			SetPageError(page);
1256			mapping = page_mapping(page);
1257			if (mapping)
1258				mapping_set_error(mapping, err);
1259		}
1260		end_page_writeback(page);
1261	}
1262}
1263EXPORT_SYMBOL_GPL(page_endio);
1264
1265/**
1266 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1267 * @__page: the page to lock
1268 */
1269void __lock_page(struct page *__page)
1270{
1271	struct page *page = compound_head(__page);
1272	wait_queue_head_t *q = page_waitqueue(page);
1273	wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, true);
 
1274}
1275EXPORT_SYMBOL(__lock_page);
1276
1277int __lock_page_killable(struct page *__page)
1278{
1279	struct page *page = compound_head(__page);
1280	wait_queue_head_t *q = page_waitqueue(page);
1281	return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE, true);
 
1282}
1283EXPORT_SYMBOL_GPL(__lock_page_killable);
1284
1285/*
1286 * Return values:
1287 * 1 - page is locked; mmap_sem is still held.
1288 * 0 - page is not locked.
1289 *     mmap_sem has been released (up_read()), unless flags had both
1290 *     FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1291 *     which case mmap_sem is still held.
1292 *
1293 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1294 * with the page locked and the mmap_sem unperturbed.
1295 */
1296int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1297			 unsigned int flags)
1298{
1299	if (flags & FAULT_FLAG_ALLOW_RETRY) {
1300		/*
1301		 * CAUTION! In this case, mmap_sem is not released
1302		 * even though return 0.
1303		 */
1304		if (flags & FAULT_FLAG_RETRY_NOWAIT)
1305			return 0;
1306
1307		up_read(&mm->mmap_sem);
1308		if (flags & FAULT_FLAG_KILLABLE)
1309			wait_on_page_locked_killable(page);
1310		else
1311			wait_on_page_locked(page);
1312		return 0;
1313	} else {
1314		if (flags & FAULT_FLAG_KILLABLE) {
1315			int ret;
1316
1317			ret = __lock_page_killable(page);
1318			if (ret) {
1319				up_read(&mm->mmap_sem);
1320				return 0;
1321			}
1322		} else
1323			__lock_page(page);
1324		return 1;
1325	}
1326}
1327
1328/**
1329 * page_cache_next_hole - find the next hole (not-present entry)
1330 * @mapping: mapping
1331 * @index: index
1332 * @max_scan: maximum range to search
1333 *
1334 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
1335 * lowest indexed hole.
1336 *
1337 * Returns: the index of the hole if found, otherwise returns an index
1338 * outside of the set specified (in which case 'return - index >=
1339 * max_scan' will be true). In rare cases of index wrap-around, 0 will
1340 * be returned.
1341 *
1342 * page_cache_next_hole may be called under rcu_read_lock. However,
1343 * like radix_tree_gang_lookup, this will not atomically search a
1344 * snapshot of the tree at a single point in time. For example, if a
1345 * hole is created at index 5, then subsequently a hole is created at
1346 * index 10, page_cache_next_hole covering both indexes may return 10
1347 * if called under rcu_read_lock.
1348 */
1349pgoff_t page_cache_next_hole(struct address_space *mapping,
1350			     pgoff_t index, unsigned long max_scan)
1351{
1352	unsigned long i;
1353
1354	for (i = 0; i < max_scan; i++) {
1355		struct page *page;
1356
1357		page = radix_tree_lookup(&mapping->i_pages, index);
1358		if (!page || radix_tree_exceptional_entry(page))
1359			break;
1360		index++;
1361		if (index == 0)
1362			break;
1363	}
1364
1365	return index;
1366}
1367EXPORT_SYMBOL(page_cache_next_hole);
1368
1369/**
1370 * page_cache_prev_hole - find the prev hole (not-present entry)
1371 * @mapping: mapping
1372 * @index: index
1373 * @max_scan: maximum range to search
1374 *
1375 * Search backwards in the range [max(index-max_scan+1, 0), index] for
1376 * the first hole.
1377 *
1378 * Returns: the index of the hole if found, otherwise returns an index
1379 * outside of the set specified (in which case 'index - return >=
1380 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
1381 * will be returned.
 
1382 *
1383 * page_cache_prev_hole may be called under rcu_read_lock. However,
1384 * like radix_tree_gang_lookup, this will not atomically search a
1385 * snapshot of the tree at a single point in time. For example, if a
1386 * hole is created at index 10, then subsequently a hole is created at
1387 * index 5, page_cache_prev_hole covering both indexes may return 5 if
1388 * called under rcu_read_lock.
1389 */
1390pgoff_t page_cache_prev_hole(struct address_space *mapping,
1391			     pgoff_t index, unsigned long max_scan)
1392{
1393	unsigned long i;
1394
1395	for (i = 0; i < max_scan; i++) {
1396		struct page *page;
1397
1398		page = radix_tree_lookup(&mapping->i_pages, index);
1399		if (!page || radix_tree_exceptional_entry(page))
 
1400			break;
1401		index--;
1402		if (index == ULONG_MAX)
1403			break;
1404	}
1405
1406	return index;
1407}
1408EXPORT_SYMBOL(page_cache_prev_hole);
1409
1410/**
1411 * find_get_entry - find and get a page cache entry
1412 * @mapping: the address_space to search
1413 * @offset: the page cache index
1414 *
1415 * Looks up the page cache slot at @mapping & @offset.  If there is a
1416 * page cache page, it is returned with an increased refcount.
1417 *
1418 * If the slot holds a shadow entry of a previously evicted page, or a
1419 * swap entry from shmem/tmpfs, it is returned.
1420 *
1421 * Otherwise, %NULL is returned.
1422 */
1423struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1424{
1425	void **pagep;
1426	struct page *head, *page;
1427
1428	rcu_read_lock();
1429repeat:
1430	page = NULL;
1431	pagep = radix_tree_lookup_slot(&mapping->i_pages, offset);
1432	if (pagep) {
1433		page = radix_tree_deref_slot(pagep);
1434		if (unlikely(!page))
1435			goto out;
1436		if (radix_tree_exception(page)) {
1437			if (radix_tree_deref_retry(page))
1438				goto repeat;
1439			/*
1440			 * A shadow entry of a recently evicted page,
1441			 * or a swap entry from shmem/tmpfs.  Return
1442			 * it without attempting to raise page count.
1443			 */
1444			goto out;
1445		}
1446
1447		head = compound_head(page);
1448		if (!page_cache_get_speculative(head))
1449			goto repeat;
1450
1451		/* The page was split under us? */
1452		if (compound_head(page) != head) {
1453			put_page(head);
1454			goto repeat;
1455		}
1456
1457		/*
1458		 * Has the page moved?
1459		 * This is part of the lockless pagecache protocol. See
1460		 * include/linux/pagemap.h for details.
1461		 */
1462		if (unlikely(page != *pagep)) {
1463			put_page(head);
1464			goto repeat;
1465		}
1466	}
 
1467out:
1468	rcu_read_unlock();
1469
1470	return page;
1471}
1472EXPORT_SYMBOL(find_get_entry);
1473
1474/**
1475 * find_lock_entry - locate, pin and lock a page cache entry
1476 * @mapping: the address_space to search
1477 * @offset: the page cache index
1478 *
1479 * Looks up the page cache slot at @mapping & @offset.  If there is a
1480 * page cache page, it is returned locked and with an increased
1481 * refcount.
1482 *
1483 * If the slot holds a shadow entry of a previously evicted page, or a
1484 * swap entry from shmem/tmpfs, it is returned.
1485 *
1486 * Otherwise, %NULL is returned.
1487 *
1488 * find_lock_entry() may sleep.
 
 
1489 */
1490struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1491{
1492	struct page *page;
1493
1494repeat:
1495	page = find_get_entry(mapping, offset);
1496	if (page && !radix_tree_exception(page)) {
1497		lock_page(page);
1498		/* Has the page been truncated? */
1499		if (unlikely(page_mapping(page) != mapping)) {
1500			unlock_page(page);
1501			put_page(page);
1502			goto repeat;
1503		}
1504		VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
1505	}
1506	return page;
1507}
1508EXPORT_SYMBOL(find_lock_entry);
1509
1510/**
1511 * pagecache_get_page - find and get a page reference
1512 * @mapping: the address_space to search
1513 * @offset: the page index
1514 * @fgp_flags: PCG flags
1515 * @gfp_mask: gfp mask to use for the page cache data page allocation
1516 *
1517 * Looks up the page cache slot at @mapping & @offset.
1518 *
1519 * PCG flags modify how the page is returned.
1520 *
1521 * @fgp_flags can be:
1522 *
1523 * - FGP_ACCESSED: the page will be marked accessed
1524 * - FGP_LOCK: Page is return locked
1525 * - FGP_CREAT: If page is not present then a new page is allocated using
1526 *   @gfp_mask and added to the page cache and the VM's LRU
1527 *   list. The page is returned locked and with an increased
1528 *   refcount. Otherwise, NULL is returned.
 
 
 
1529 *
1530 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1531 * if the GFP flags specified for FGP_CREAT are atomic.
1532 *
1533 * If there is a page cache page, it is returned with an increased refcount.
 
 
1534 */
1535struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1536	int fgp_flags, gfp_t gfp_mask)
1537{
1538	struct page *page;
1539
1540repeat:
1541	page = find_get_entry(mapping, offset);
1542	if (radix_tree_exceptional_entry(page))
1543		page = NULL;
1544	if (!page)
1545		goto no_page;
1546
1547	if (fgp_flags & FGP_LOCK) {
1548		if (fgp_flags & FGP_NOWAIT) {
1549			if (!trylock_page(page)) {
1550				put_page(page);
1551				return NULL;
1552			}
1553		} else {
1554			lock_page(page);
1555		}
1556
1557		/* Has the page been truncated? */
1558		if (unlikely(page->mapping != mapping)) {
1559			unlock_page(page);
1560			put_page(page);
1561			goto repeat;
1562		}
1563		VM_BUG_ON_PAGE(page->index != offset, page);
1564	}
1565
1566	if (page && (fgp_flags & FGP_ACCESSED))
1567		mark_page_accessed(page);
1568
1569no_page:
1570	if (!page && (fgp_flags & FGP_CREAT)) {
1571		int err;
1572		if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1573			gfp_mask |= __GFP_WRITE;
1574		if (fgp_flags & FGP_NOFS)
1575			gfp_mask &= ~__GFP_FS;
1576
1577		page = __page_cache_alloc(gfp_mask);
1578		if (!page)
1579			return NULL;
1580
1581		if (WARN_ON_ONCE(!(fgp_flags & FGP_LOCK)))
1582			fgp_flags |= FGP_LOCK;
1583
1584		/* Init accessed so avoid atomic mark_page_accessed later */
1585		if (fgp_flags & FGP_ACCESSED)
1586			__SetPageReferenced(page);
1587
1588		err = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
1589		if (unlikely(err)) {
1590			put_page(page);
1591			page = NULL;
1592			if (err == -EEXIST)
1593				goto repeat;
1594		}
 
 
 
 
 
 
 
1595	}
1596
1597	return page;
1598}
1599EXPORT_SYMBOL(pagecache_get_page);
1600
1601/**
1602 * find_get_entries - gang pagecache lookup
1603 * @mapping:	The address_space to search
1604 * @start:	The starting page cache index
1605 * @nr_entries:	The maximum number of entries
1606 * @entries:	Where the resulting entries are placed
1607 * @indices:	The cache indices corresponding to the entries in @entries
1608 *
1609 * find_get_entries() will search for and return a group of up to
1610 * @nr_entries entries in the mapping.  The entries are placed at
1611 * @entries.  find_get_entries() takes a reference against any actual
1612 * pages it returns.
1613 *
1614 * The search returns a group of mapping-contiguous page cache entries
1615 * with ascending indexes.  There may be holes in the indices due to
1616 * not-present pages.
1617 *
1618 * Any shadow entries of evicted pages, or swap entries from
1619 * shmem/tmpfs, are included in the returned array.
1620 *
1621 * find_get_entries() returns the number of pages and shadow entries
1622 * which were found.
1623 */
1624unsigned find_get_entries(struct address_space *mapping,
1625			  pgoff_t start, unsigned int nr_entries,
1626			  struct page **entries, pgoff_t *indices)
1627{
1628	void **slot;
 
1629	unsigned int ret = 0;
1630	struct radix_tree_iter iter;
1631
1632	if (!nr_entries)
1633		return 0;
1634
1635	rcu_read_lock();
1636	radix_tree_for_each_slot(slot, &mapping->i_pages, &iter, start) {
1637		struct page *head, *page;
1638repeat:
1639		page = radix_tree_deref_slot(slot);
1640		if (unlikely(!page))
1641			continue;
1642		if (radix_tree_exception(page)) {
1643			if (radix_tree_deref_retry(page)) {
1644				slot = radix_tree_iter_retry(&iter);
1645				continue;
1646			}
1647			/*
1648			 * A shadow entry of a recently evicted page, a swap
1649			 * entry from shmem/tmpfs or a DAX entry.  Return it
1650			 * without attempting to raise page count.
1651			 */
1652			goto export;
1653		}
1654
1655		head = compound_head(page);
1656		if (!page_cache_get_speculative(head))
1657			goto repeat;
1658
1659		/* The page was split under us? */
1660		if (compound_head(page) != head) {
1661			put_page(head);
1662			goto repeat;
1663		}
1664
1665		/* Has the page moved? */
1666		if (unlikely(page != *slot)) {
1667			put_page(head);
1668			goto repeat;
1669		}
1670export:
1671		indices[ret] = iter.index;
1672		entries[ret] = page;
1673		if (++ret == nr_entries)
1674			break;
 
 
 
 
 
1675	}
1676	rcu_read_unlock();
1677	return ret;
1678}
1679
1680/**
1681 * find_get_pages_range - gang pagecache lookup
1682 * @mapping:	The address_space to search
1683 * @start:	The starting page index
1684 * @end:	The final page index (inclusive)
1685 * @nr_pages:	The maximum number of pages
1686 * @pages:	Where the resulting pages are placed
1687 *
1688 * find_get_pages_range() will search for and return a group of up to @nr_pages
1689 * pages in the mapping starting at index @start and up to index @end
1690 * (inclusive).  The pages are placed at @pages.  find_get_pages_range() takes
1691 * a reference against the returned pages.
1692 *
1693 * The search returns a group of mapping-contiguous pages with ascending
1694 * indexes.  There may be holes in the indices due to not-present pages.
1695 * We also update @start to index the next page for the traversal.
1696 *
1697 * find_get_pages_range() returns the number of pages which were found. If this
1698 * number is smaller than @nr_pages, the end of specified range has been
1699 * reached.
1700 */
1701unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
1702			      pgoff_t end, unsigned int nr_pages,
1703			      struct page **pages)
1704{
1705	struct radix_tree_iter iter;
1706	void **slot;
1707	unsigned ret = 0;
1708
1709	if (unlikely(!nr_pages))
1710		return 0;
1711
1712	rcu_read_lock();
1713	radix_tree_for_each_slot(slot, &mapping->i_pages, &iter, *start) {
1714		struct page *head, *page;
1715
1716		if (iter.index > end)
1717			break;
1718repeat:
1719		page = radix_tree_deref_slot(slot);
1720		if (unlikely(!page))
1721			continue;
1722
1723		if (radix_tree_exception(page)) {
1724			if (radix_tree_deref_retry(page)) {
1725				slot = radix_tree_iter_retry(&iter);
1726				continue;
1727			}
1728			/*
1729			 * A shadow entry of a recently evicted page,
1730			 * or a swap entry from shmem/tmpfs.  Skip
1731			 * over it.
1732			 */
1733			continue;
1734		}
1735
1736		head = compound_head(page);
1737		if (!page_cache_get_speculative(head))
1738			goto repeat;
1739
1740		/* The page was split under us? */
1741		if (compound_head(page) != head) {
1742			put_page(head);
1743			goto repeat;
1744		}
1745
1746		/* Has the page moved? */
1747		if (unlikely(page != *slot)) {
1748			put_page(head);
1749			goto repeat;
1750		}
1751
1752		pages[ret] = page;
1753		if (++ret == nr_pages) {
1754			*start = pages[ret - 1]->index + 1;
1755			goto out;
1756		}
 
 
 
 
 
1757	}
1758
1759	/*
1760	 * We come here when there is no page beyond @end. We take care to not
1761	 * overflow the index @start as it confuses some of the callers. This
1762	 * breaks the iteration when there is page at index -1 but that is
1763	 * already broken anyway.
1764	 */
1765	if (end == (pgoff_t)-1)
1766		*start = (pgoff_t)-1;
1767	else
1768		*start = end + 1;
1769out:
1770	rcu_read_unlock();
1771
1772	return ret;
1773}
1774
1775/**
1776 * find_get_pages_contig - gang contiguous pagecache lookup
1777 * @mapping:	The address_space to search
1778 * @index:	The starting page index
1779 * @nr_pages:	The maximum number of pages
1780 * @pages:	Where the resulting pages are placed
1781 *
1782 * find_get_pages_contig() works exactly like find_get_pages(), except
1783 * that the returned number of pages are guaranteed to be contiguous.
1784 *
1785 * find_get_pages_contig() returns the number of pages which were found.
1786 */
1787unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1788			       unsigned int nr_pages, struct page **pages)
1789{
1790	struct radix_tree_iter iter;
1791	void **slot;
1792	unsigned int ret = 0;
1793
1794	if (unlikely(!nr_pages))
1795		return 0;
1796
1797	rcu_read_lock();
1798	radix_tree_for_each_contig(slot, &mapping->i_pages, &iter, index) {
1799		struct page *head, *page;
1800repeat:
1801		page = radix_tree_deref_slot(slot);
1802		/* The hole, there no reason to continue */
1803		if (unlikely(!page))
1804			break;
1805
1806		if (radix_tree_exception(page)) {
1807			if (radix_tree_deref_retry(page)) {
1808				slot = radix_tree_iter_retry(&iter);
1809				continue;
1810			}
1811			/*
1812			 * A shadow entry of a recently evicted page,
1813			 * or a swap entry from shmem/tmpfs.  Stop
1814			 * looking for contiguous pages.
1815			 */
1816			break;
1817		}
1818
1819		head = compound_head(page);
1820		if (!page_cache_get_speculative(head))
1821			goto repeat;
1822
1823		/* The page was split under us? */
1824		if (compound_head(page) != head) {
1825			put_page(head);
1826			goto repeat;
1827		}
1828
1829		/* Has the page moved? */
1830		if (unlikely(page != *slot)) {
1831			put_page(head);
1832			goto repeat;
1833		}
1834
1835		/*
1836		 * must check mapping and index after taking the ref.
1837		 * otherwise we can get both false positives and false
1838		 * negatives, which is just confusing to the caller.
1839		 */
1840		if (page->mapping == NULL || page_to_pgoff(page) != iter.index) {
1841			put_page(page);
1842			break;
1843		}
1844
1845		pages[ret] = page;
 
 
 
 
 
 
 
1846		if (++ret == nr_pages)
1847			break;
 
 
 
 
 
1848	}
1849	rcu_read_unlock();
1850	return ret;
1851}
1852EXPORT_SYMBOL(find_get_pages_contig);
1853
1854/**
1855 * find_get_pages_range_tag - find and return pages in given range matching @tag
1856 * @mapping:	the address_space to search
1857 * @index:	the starting page index
1858 * @end:	The final page index (inclusive)
1859 * @tag:	the tag index
1860 * @nr_pages:	the maximum number of pages
1861 * @pages:	where the resulting pages are placed
1862 *
1863 * Like find_get_pages, except we only return pages which are tagged with
1864 * @tag.   We update @index to index the next page for the traversal.
 
 
1865 */
1866unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
1867			pgoff_t end, int tag, unsigned int nr_pages,
1868			struct page **pages)
1869{
1870	struct radix_tree_iter iter;
1871	void **slot;
1872	unsigned ret = 0;
1873
1874	if (unlikely(!nr_pages))
1875		return 0;
1876
1877	rcu_read_lock();
1878	radix_tree_for_each_tagged(slot, &mapping->i_pages, &iter, *index, tag) {
1879		struct page *head, *page;
1880
1881		if (iter.index > end)
1882			break;
1883repeat:
1884		page = radix_tree_deref_slot(slot);
1885		if (unlikely(!page))
1886			continue;
1887
1888		if (radix_tree_exception(page)) {
1889			if (radix_tree_deref_retry(page)) {
1890				slot = radix_tree_iter_retry(&iter);
1891				continue;
1892			}
1893			/*
1894			 * A shadow entry of a recently evicted page.
1895			 *
1896			 * Those entries should never be tagged, but
1897			 * this tree walk is lockless and the tags are
1898			 * looked up in bulk, one radix tree node at a
1899			 * time, so there is a sizable window for page
1900			 * reclaim to evict a page we saw tagged.
1901			 *
1902			 * Skip over it.
1903			 */
1904			continue;
1905		}
1906
1907		head = compound_head(page);
1908		if (!page_cache_get_speculative(head))
1909			goto repeat;
1910
1911		/* The page was split under us? */
1912		if (compound_head(page) != head) {
1913			put_page(head);
1914			goto repeat;
1915		}
1916
1917		/* Has the page moved? */
1918		if (unlikely(page != *slot)) {
1919			put_page(head);
1920			goto repeat;
1921		}
1922
1923		pages[ret] = page;
1924		if (++ret == nr_pages) {
1925			*index = pages[ret - 1]->index + 1;
1926			goto out;
1927		}
 
 
 
 
 
1928	}
1929
1930	/*
1931	 * We come here when we got at @end. We take care to not overflow the
1932	 * index @index as it confuses some of the callers. This breaks the
1933	 * iteration when there is page at index -1 but that is already broken
1934	 * anyway.
1935	 */
1936	if (end == (pgoff_t)-1)
1937		*index = (pgoff_t)-1;
1938	else
1939		*index = end + 1;
1940out:
1941	rcu_read_unlock();
1942
1943	return ret;
1944}
1945EXPORT_SYMBOL(find_get_pages_range_tag);
1946
1947/**
1948 * find_get_entries_tag - find and return entries that match @tag
1949 * @mapping:	the address_space to search
1950 * @start:	the starting page cache index
1951 * @tag:	the tag index
1952 * @nr_entries:	the maximum number of entries
1953 * @entries:	where the resulting entries are placed
1954 * @indices:	the cache indices corresponding to the entries in @entries
1955 *
1956 * Like find_get_entries, except we only return entries which are tagged with
1957 * @tag.
1958 */
1959unsigned find_get_entries_tag(struct address_space *mapping, pgoff_t start,
1960			int tag, unsigned int nr_entries,
1961			struct page **entries, pgoff_t *indices)
1962{
1963	void **slot;
1964	unsigned int ret = 0;
1965	struct radix_tree_iter iter;
1966
1967	if (!nr_entries)
1968		return 0;
1969
1970	rcu_read_lock();
1971	radix_tree_for_each_tagged(slot, &mapping->i_pages, &iter, start, tag) {
1972		struct page *head, *page;
1973repeat:
1974		page = radix_tree_deref_slot(slot);
1975		if (unlikely(!page))
1976			continue;
1977		if (radix_tree_exception(page)) {
1978			if (radix_tree_deref_retry(page)) {
1979				slot = radix_tree_iter_retry(&iter);
1980				continue;
1981			}
1982
1983			/*
1984			 * A shadow entry of a recently evicted page, a swap
1985			 * entry from shmem/tmpfs or a DAX entry.  Return it
1986			 * without attempting to raise page count.
1987			 */
1988			goto export;
1989		}
1990
1991		head = compound_head(page);
1992		if (!page_cache_get_speculative(head))
1993			goto repeat;
1994
1995		/* The page was split under us? */
1996		if (compound_head(page) != head) {
1997			put_page(head);
1998			goto repeat;
1999		}
2000
2001		/* Has the page moved? */
2002		if (unlikely(page != *slot)) {
2003			put_page(head);
2004			goto repeat;
2005		}
2006export:
2007		indices[ret] = iter.index;
2008		entries[ret] = page;
2009		if (++ret == nr_entries)
2010			break;
2011	}
2012	rcu_read_unlock();
2013	return ret;
2014}
2015EXPORT_SYMBOL(find_get_entries_tag);
2016
2017/*
2018 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2019 * a _large_ part of the i/o request. Imagine the worst scenario:
2020 *
2021 *      ---R__________________________________________B__________
2022 *         ^ reading here                             ^ bad block(assume 4k)
2023 *
2024 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2025 * => failing the whole request => read(R) => read(R+1) =>
2026 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2027 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2028 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2029 *
2030 * It is going insane. Fix it by quickly scaling down the readahead size.
2031 */
2032static void shrink_readahead_size_eio(struct file *filp,
2033					struct file_ra_state *ra)
2034{
2035	ra->ra_pages /= 4;
2036}
2037
2038/**
2039 * generic_file_buffered_read - generic file read routine
2040 * @iocb:	the iocb to read
2041 * @iter:	data destination
2042 * @written:	already copied
2043 *
2044 * This is a generic file read routine, and uses the
2045 * mapping->a_ops->readpage() function for the actual low-level stuff.
2046 *
2047 * This is really ugly. But the goto's actually try to clarify some
2048 * of the logic when it comes to error handling etc.
 
 
 
 
2049 */
2050static ssize_t generic_file_buffered_read(struct kiocb *iocb,
2051		struct iov_iter *iter, ssize_t written)
2052{
2053	struct file *filp = iocb->ki_filp;
2054	struct address_space *mapping = filp->f_mapping;
2055	struct inode *inode = mapping->host;
2056	struct file_ra_state *ra = &filp->f_ra;
2057	loff_t *ppos = &iocb->ki_pos;
2058	pgoff_t index;
2059	pgoff_t last_index;
2060	pgoff_t prev_index;
2061	unsigned long offset;      /* offset into pagecache page */
2062	unsigned int prev_offset;
2063	int error = 0;
2064
2065	if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
2066		return 0;
2067	iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2068
2069	index = *ppos >> PAGE_SHIFT;
2070	prev_index = ra->prev_pos >> PAGE_SHIFT;
2071	prev_offset = ra->prev_pos & (PAGE_SIZE-1);
2072	last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
2073	offset = *ppos & ~PAGE_MASK;
2074
2075	for (;;) {
2076		struct page *page;
2077		pgoff_t end_index;
2078		loff_t isize;
2079		unsigned long nr, ret;
2080
2081		cond_resched();
2082find_page:
2083		if (fatal_signal_pending(current)) {
2084			error = -EINTR;
2085			goto out;
2086		}
2087
2088		page = find_get_page(mapping, index);
2089		if (!page) {
2090			if (iocb->ki_flags & IOCB_NOWAIT)
2091				goto would_block;
2092			page_cache_sync_readahead(mapping,
2093					ra, filp,
2094					index, last_index - index);
2095			page = find_get_page(mapping, index);
2096			if (unlikely(page == NULL))
2097				goto no_cached_page;
2098		}
2099		if (PageReadahead(page)) {
2100			page_cache_async_readahead(mapping,
2101					ra, filp, page,
2102					index, last_index - index);
2103		}
2104		if (!PageUptodate(page)) {
2105			if (iocb->ki_flags & IOCB_NOWAIT) {
2106				put_page(page);
2107				goto would_block;
2108			}
2109
2110			/*
2111			 * See comment in do_read_cache_page on why
2112			 * wait_on_page_locked is used to avoid unnecessarily
2113			 * serialisations and why it's safe.
2114			 */
2115			error = wait_on_page_locked_killable(page);
2116			if (unlikely(error))
2117				goto readpage_error;
2118			if (PageUptodate(page))
2119				goto page_ok;
2120
2121			if (inode->i_blkbits == PAGE_SHIFT ||
2122					!mapping->a_ops->is_partially_uptodate)
2123				goto page_not_up_to_date;
2124			/* pipes can't handle partially uptodate pages */
2125			if (unlikely(iter->type & ITER_PIPE))
2126				goto page_not_up_to_date;
2127			if (!trylock_page(page))
2128				goto page_not_up_to_date;
2129			/* Did it get truncated before we got the lock? */
2130			if (!page->mapping)
2131				goto page_not_up_to_date_locked;
2132			if (!mapping->a_ops->is_partially_uptodate(page,
2133							offset, iter->count))
2134				goto page_not_up_to_date_locked;
2135			unlock_page(page);
2136		}
2137page_ok:
2138		/*
2139		 * i_size must be checked after we know the page is Uptodate.
2140		 *
2141		 * Checking i_size after the check allows us to calculate
2142		 * the correct value for "nr", which means the zero-filled
2143		 * part of the page is not copied back to userspace (unless
2144		 * another truncate extends the file - this is desired though).
2145		 */
2146
2147		isize = i_size_read(inode);
2148		end_index = (isize - 1) >> PAGE_SHIFT;
2149		if (unlikely(!isize || index > end_index)) {
2150			put_page(page);
2151			goto out;
2152		}
2153
2154		/* nr is the maximum number of bytes to copy from this page */
2155		nr = PAGE_SIZE;
2156		if (index == end_index) {
2157			nr = ((isize - 1) & ~PAGE_MASK) + 1;
2158			if (nr <= offset) {
2159				put_page(page);
2160				goto out;
2161			}
2162		}
2163		nr = nr - offset;
2164
2165		/* If users can be writing to this page using arbitrary
2166		 * virtual addresses, take care about potential aliasing
2167		 * before reading the page on the kernel side.
2168		 */
2169		if (mapping_writably_mapped(mapping))
2170			flush_dcache_page(page);
2171
2172		/*
2173		 * When a sequential read accesses a page several times,
2174		 * only mark it as accessed the first time.
2175		 */
2176		if (prev_index != index || offset != prev_offset)
2177			mark_page_accessed(page);
2178		prev_index = index;
2179
2180		/*
2181		 * Ok, we have the page, and it's up-to-date, so
2182		 * now we can copy it to user space...
2183		 */
2184
2185		ret = copy_page_to_iter(page, offset, nr, iter);
2186		offset += ret;
2187		index += offset >> PAGE_SHIFT;
2188		offset &= ~PAGE_MASK;
2189		prev_offset = offset;
2190
2191		put_page(page);
2192		written += ret;
2193		if (!iov_iter_count(iter))
2194			goto out;
2195		if (ret < nr) {
2196			error = -EFAULT;
2197			goto out;
2198		}
2199		continue;
2200
2201page_not_up_to_date:
2202		/* Get exclusive access to the page ... */
2203		error = lock_page_killable(page);
2204		if (unlikely(error))
2205			goto readpage_error;
2206
2207page_not_up_to_date_locked:
2208		/* Did it get truncated before we got the lock? */
2209		if (!page->mapping) {
2210			unlock_page(page);
2211			put_page(page);
2212			continue;
2213		}
2214
2215		/* Did somebody else fill it already? */
2216		if (PageUptodate(page)) {
2217			unlock_page(page);
2218			goto page_ok;
2219		}
2220
2221readpage:
2222		/*
2223		 * A previous I/O error may have been due to temporary
2224		 * failures, eg. multipath errors.
2225		 * PG_error will be set again if readpage fails.
2226		 */
2227		ClearPageError(page);
2228		/* Start the actual read. The read will unlock the page. */
2229		error = mapping->a_ops->readpage(filp, page);
2230
2231		if (unlikely(error)) {
2232			if (error == AOP_TRUNCATED_PAGE) {
2233				put_page(page);
2234				error = 0;
2235				goto find_page;
2236			}
2237			goto readpage_error;
2238		}
2239
2240		if (!PageUptodate(page)) {
2241			error = lock_page_killable(page);
2242			if (unlikely(error))
2243				goto readpage_error;
2244			if (!PageUptodate(page)) {
2245				if (page->mapping == NULL) {
2246					/*
2247					 * invalidate_mapping_pages got it
2248					 */
2249					unlock_page(page);
2250					put_page(page);
2251					goto find_page;
2252				}
2253				unlock_page(page);
2254				shrink_readahead_size_eio(filp, ra);
2255				error = -EIO;
2256				goto readpage_error;
2257			}
2258			unlock_page(page);
2259		}
2260
2261		goto page_ok;
2262
2263readpage_error:
2264		/* UHHUH! A synchronous read error occurred. Report it */
2265		put_page(page);
2266		goto out;
2267
2268no_cached_page:
2269		/*
2270		 * Ok, it wasn't cached, so we need to create a new
2271		 * page..
2272		 */
2273		page = page_cache_alloc(mapping);
2274		if (!page) {
2275			error = -ENOMEM;
2276			goto out;
2277		}
2278		error = add_to_page_cache_lru(page, mapping, index,
2279				mapping_gfp_constraint(mapping, GFP_KERNEL));
2280		if (error) {
2281			put_page(page);
2282			if (error == -EEXIST) {
2283				error = 0;
2284				goto find_page;
2285			}
2286			goto out;
2287		}
2288		goto readpage;
2289	}
2290
2291would_block:
2292	error = -EAGAIN;
2293out:
2294	ra->prev_pos = prev_index;
2295	ra->prev_pos <<= PAGE_SHIFT;
2296	ra->prev_pos |= prev_offset;
2297
2298	*ppos = ((loff_t)index << PAGE_SHIFT) + offset;
2299	file_accessed(filp);
2300	return written ? written : error;
2301}
2302
2303/**
2304 * generic_file_read_iter - generic filesystem read routine
2305 * @iocb:	kernel I/O control block
2306 * @iter:	destination for the data read
2307 *
2308 * This is the "read_iter()" routine for all filesystems
2309 * that can use the page cache directly.
 
 
 
2310 */
2311ssize_t
2312generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2313{
2314	size_t count = iov_iter_count(iter);
2315	ssize_t retval = 0;
2316
2317	if (!count)
2318		goto out; /* skip atime */
2319
2320	if (iocb->ki_flags & IOCB_DIRECT) {
2321		struct file *file = iocb->ki_filp;
2322		struct address_space *mapping = file->f_mapping;
2323		struct inode *inode = mapping->host;
2324		loff_t size;
2325
2326		size = i_size_read(inode);
2327		if (iocb->ki_flags & IOCB_NOWAIT) {
2328			if (filemap_range_has_page(mapping, iocb->ki_pos,
2329						   iocb->ki_pos + count - 1))
2330				return -EAGAIN;
2331		} else {
2332			retval = filemap_write_and_wait_range(mapping,
2333						iocb->ki_pos,
2334					        iocb->ki_pos + count - 1);
2335			if (retval < 0)
2336				goto out;
2337		}
2338
2339		file_accessed(file);
2340
2341		retval = mapping->a_ops->direct_IO(iocb, iter);
2342		if (retval >= 0) {
2343			iocb->ki_pos += retval;
2344			count -= retval;
2345		}
2346		iov_iter_revert(iter, count - iov_iter_count(iter));
2347
2348		/*
2349		 * Btrfs can have a short DIO read if we encounter
2350		 * compressed extents, so if there was an error, or if
2351		 * we've already read everything we wanted to, or if
2352		 * there was a short read because we hit EOF, go ahead
2353		 * and return.  Otherwise fallthrough to buffered io for
2354		 * the rest of the read.  Buffered reads will not work for
2355		 * DAX files, so don't bother trying.
2356		 */
2357		if (retval < 0 || !count || iocb->ki_pos >= size ||
2358		    IS_DAX(inode))
2359			goto out;
2360	}
2361
2362	retval = generic_file_buffered_read(iocb, iter, retval);
2363out:
2364	return retval;
2365}
2366EXPORT_SYMBOL(generic_file_read_iter);
2367
2368#ifdef CONFIG_MMU
2369/**
2370 * page_cache_read - adds requested page to the page cache if not already there
2371 * @file:	file to read
2372 * @offset:	page index
2373 * @gfp_mask:	memory allocation flags
2374 *
2375 * This adds the requested page to the page cache if it isn't already there,
2376 * and schedules an I/O to read in its contents from disk.
2377 */
2378static int page_cache_read(struct file *file, pgoff_t offset, gfp_t gfp_mask)
2379{
2380	struct address_space *mapping = file->f_mapping;
2381	struct page *page;
2382	int ret;
2383
2384	do {
2385		page = __page_cache_alloc(gfp_mask);
2386		if (!page)
2387			return -ENOMEM;
2388
2389		ret = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
2390		if (ret == 0)
2391			ret = mapping->a_ops->readpage(file, page);
2392		else if (ret == -EEXIST)
2393			ret = 0; /* losing race to add is OK */
 
 
 
 
 
 
 
2394
2395		put_page(page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2396
2397	} while (ret == AOP_TRUNCATED_PAGE);
 
 
 
 
 
 
2398
2399	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2400}
2401
2402#define MMAP_LOTSAMISS  (100)
2403
2404/*
2405 * Synchronous readahead happens when we don't even find
2406 * a page in the page cache at all.
 
 
 
2407 */
2408static void do_sync_mmap_readahead(struct vm_area_struct *vma,
2409				   struct file_ra_state *ra,
2410				   struct file *file,
2411				   pgoff_t offset)
2412{
 
 
2413	struct address_space *mapping = file->f_mapping;
 
 
2414
2415	/* If we don't want any read-ahead, don't bother */
2416	if (vma->vm_flags & VM_RAND_READ)
2417		return;
2418	if (!ra->ra_pages)
2419		return;
2420
2421	if (vma->vm_flags & VM_SEQ_READ) {
 
2422		page_cache_sync_readahead(mapping, ra, file, offset,
2423					  ra->ra_pages);
2424		return;
2425	}
2426
2427	/* Avoid banging the cache line if not needed */
2428	if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
2429		ra->mmap_miss++;
2430
2431	/*
2432	 * Do we miss much more than hit in this file? If so,
2433	 * stop bothering with read-ahead. It will only hurt.
2434	 */
2435	if (ra->mmap_miss > MMAP_LOTSAMISS)
2436		return;
2437
2438	/*
2439	 * mmap read-around
2440	 */
 
2441	ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2442	ra->size = ra->ra_pages;
2443	ra->async_size = ra->ra_pages / 4;
2444	ra_submit(ra, mapping, file);
 
2445}
2446
2447/*
2448 * Asynchronous readahead happens when we find the page and PG_readahead,
2449 * so we want to possibly extend the readahead further..
 
2450 */
2451static void do_async_mmap_readahead(struct vm_area_struct *vma,
2452				    struct file_ra_state *ra,
2453				    struct file *file,
2454				    struct page *page,
2455				    pgoff_t offset)
2456{
 
 
2457	struct address_space *mapping = file->f_mapping;
 
 
2458
2459	/* If we don't want any read-ahead, don't bother */
2460	if (vma->vm_flags & VM_RAND_READ)
2461		return;
2462	if (ra->mmap_miss > 0)
2463		ra->mmap_miss--;
2464	if (PageReadahead(page))
 
2465		page_cache_async_readahead(mapping, ra, file,
2466					   page, offset, ra->ra_pages);
 
 
2467}
2468
2469/**
2470 * filemap_fault - read in file data for page fault handling
2471 * @vmf:	struct vm_fault containing details of the fault
2472 *
2473 * filemap_fault() is invoked via the vma operations vector for a
2474 * mapped memory region to read in file data during a page fault.
2475 *
2476 * The goto's are kind of ugly, but this streamlines the normal case of having
2477 * it in the page cache, and handles the special cases reasonably without
2478 * having a lot of duplicated code.
2479 *
2480 * vma->vm_mm->mmap_sem must be held on entry.
2481 *
2482 * If our return value has VM_FAULT_RETRY set, it's because
2483 * lock_page_or_retry() returned 0.
2484 * The mmap_sem has usually been released in this case.
2485 * See __lock_page_or_retry() for the exception.
2486 *
2487 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2488 * has not been released.
2489 *
2490 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
 
 
2491 */
2492int filemap_fault(struct vm_fault *vmf)
2493{
2494	int error;
2495	struct file *file = vmf->vma->vm_file;
 
2496	struct address_space *mapping = file->f_mapping;
2497	struct file_ra_state *ra = &file->f_ra;
2498	struct inode *inode = mapping->host;
2499	pgoff_t offset = vmf->pgoff;
2500	pgoff_t max_off;
2501	struct page *page;
2502	int ret = 0;
2503
2504	max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2505	if (unlikely(offset >= max_off))
2506		return VM_FAULT_SIGBUS;
2507
2508	/*
2509	 * Do we have something in the page cache already?
2510	 */
2511	page = find_get_page(mapping, offset);
2512	if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2513		/*
2514		 * We found the page, so try async readahead before
2515		 * waiting for the lock.
2516		 */
2517		do_async_mmap_readahead(vmf->vma, ra, file, page, offset);
2518	} else if (!page) {
2519		/* No page in the page cache at all */
2520		do_sync_mmap_readahead(vmf->vma, ra, file, offset);
2521		count_vm_event(PGMAJFAULT);
2522		count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
2523		ret = VM_FAULT_MAJOR;
 
2524retry_find:
2525		page = find_get_page(mapping, offset);
2526		if (!page)
2527			goto no_cached_page;
 
 
 
 
 
2528	}
2529
2530	if (!lock_page_or_retry(page, vmf->vma->vm_mm, vmf->flags)) {
2531		put_page(page);
2532		return ret | VM_FAULT_RETRY;
2533	}
2534
2535	/* Did it get truncated? */
2536	if (unlikely(page->mapping != mapping)) {
2537		unlock_page(page);
2538		put_page(page);
2539		goto retry_find;
2540	}
2541	VM_BUG_ON_PAGE(page->index != offset, page);
2542
2543	/*
2544	 * We have a locked page in the page cache, now we need to check
2545	 * that it's up-to-date. If not, it is going to be due to an error.
2546	 */
2547	if (unlikely(!PageUptodate(page)))
2548		goto page_not_uptodate;
2549
2550	/*
 
 
 
 
 
 
 
 
 
 
2551	 * Found the page and have a reference on it.
2552	 * We must recheck i_size under page lock.
2553	 */
2554	max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2555	if (unlikely(offset >= max_off)) {
2556		unlock_page(page);
2557		put_page(page);
2558		return VM_FAULT_SIGBUS;
2559	}
2560
2561	vmf->page = page;
2562	return ret | VM_FAULT_LOCKED;
2563
2564no_cached_page:
2565	/*
2566	 * We're only likely to ever get here if MADV_RANDOM is in
2567	 * effect.
2568	 */
2569	error = page_cache_read(file, offset, vmf->gfp_mask);
2570
2571	/*
2572	 * The page we want has now been added to the page cache.
2573	 * In the unlikely event that someone removed it in the
2574	 * meantime, we'll just come back here and read it again.
2575	 */
2576	if (error >= 0)
2577		goto retry_find;
2578
2579	/*
2580	 * An error return from page_cache_read can result if the
2581	 * system is low on memory, or a problem occurs while trying
2582	 * to schedule I/O.
2583	 */
2584	if (error == -ENOMEM)
2585		return VM_FAULT_OOM;
2586	return VM_FAULT_SIGBUS;
2587
2588page_not_uptodate:
2589	/*
2590	 * Umm, take care of errors if the page isn't up-to-date.
2591	 * Try to re-read it _once_. We do this synchronously,
2592	 * because there really aren't any performance issues here
2593	 * and we need to check for errors.
2594	 */
2595	ClearPageError(page);
 
2596	error = mapping->a_ops->readpage(file, page);
2597	if (!error) {
2598		wait_on_page_locked(page);
2599		if (!PageUptodate(page))
2600			error = -EIO;
2601	}
 
 
2602	put_page(page);
2603
2604	if (!error || error == AOP_TRUNCATED_PAGE)
2605		goto retry_find;
2606
2607	/* Things didn't work out. Return zero to tell the mm layer so. */
2608	shrink_readahead_size_eio(file, ra);
2609	return VM_FAULT_SIGBUS;
 
 
 
 
 
 
 
 
 
 
 
 
2610}
2611EXPORT_SYMBOL(filemap_fault);
2612
2613void filemap_map_pages(struct vm_fault *vmf,
2614		pgoff_t start_pgoff, pgoff_t end_pgoff)
2615{
2616	struct radix_tree_iter iter;
2617	void **slot;
2618	struct file *file = vmf->vma->vm_file;
2619	struct address_space *mapping = file->f_mapping;
2620	pgoff_t last_pgoff = start_pgoff;
2621	unsigned long max_idx;
2622	struct page *head, *page;
 
2623
2624	rcu_read_lock();
2625	radix_tree_for_each_slot(slot, &mapping->i_pages, &iter, start_pgoff) {
2626		if (iter.index > end_pgoff)
2627			break;
2628repeat:
2629		page = radix_tree_deref_slot(slot);
2630		if (unlikely(!page))
2631			goto next;
2632		if (radix_tree_exception(page)) {
2633			if (radix_tree_deref_retry(page)) {
2634				slot = radix_tree_iter_retry(&iter);
2635				continue;
2636			}
2637			goto next;
2638		}
2639
2640		head = compound_head(page);
2641		if (!page_cache_get_speculative(head))
2642			goto repeat;
2643
2644		/* The page was split under us? */
2645		if (compound_head(page) != head) {
2646			put_page(head);
2647			goto repeat;
2648		}
2649
2650		/* Has the page moved? */
2651		if (unlikely(page != *slot)) {
2652			put_page(head);
2653			goto repeat;
2654		}
2655
2656		if (!PageUptodate(page) ||
2657				PageReadahead(page) ||
2658				PageHWPoison(page))
2659			goto skip;
2660		if (!trylock_page(page))
2661			goto skip;
2662
2663		if (page->mapping != mapping || !PageUptodate(page))
2664			goto unlock;
2665
2666		max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
2667		if (page->index >= max_idx)
2668			goto unlock;
2669
2670		if (file->f_ra.mmap_miss > 0)
2671			file->f_ra.mmap_miss--;
2672
2673		vmf->address += (iter.index - last_pgoff) << PAGE_SHIFT;
2674		if (vmf->pte)
2675			vmf->pte += iter.index - last_pgoff;
2676		last_pgoff = iter.index;
2677		if (alloc_set_pte(vmf, NULL, page))
2678			goto unlock;
2679		unlock_page(page);
2680		goto next;
2681unlock:
2682		unlock_page(page);
2683skip:
2684		put_page(page);
2685next:
2686		/* Huge page is mapped? No need to proceed. */
2687		if (pmd_trans_huge(*vmf->pmd))
2688			break;
2689		if (iter.index == end_pgoff)
2690			break;
2691	}
2692	rcu_read_unlock();
2693}
2694EXPORT_SYMBOL(filemap_map_pages);
2695
2696int filemap_page_mkwrite(struct vm_fault *vmf)
2697{
2698	struct page *page = vmf->page;
2699	struct inode *inode = file_inode(vmf->vma->vm_file);
2700	int ret = VM_FAULT_LOCKED;
2701
2702	sb_start_pagefault(inode->i_sb);
2703	file_update_time(vmf->vma->vm_file);
2704	lock_page(page);
2705	if (page->mapping != inode->i_mapping) {
2706		unlock_page(page);
2707		ret = VM_FAULT_NOPAGE;
2708		goto out;
2709	}
2710	/*
2711	 * We mark the page dirty already here so that when freeze is in
2712	 * progress, we are guaranteed that writeback during freezing will
2713	 * see the dirty page and writeprotect it again.
2714	 */
2715	set_page_dirty(page);
2716	wait_for_stable_page(page);
2717out:
2718	sb_end_pagefault(inode->i_sb);
2719	return ret;
2720}
2721
2722const struct vm_operations_struct generic_file_vm_ops = {
2723	.fault		= filemap_fault,
2724	.map_pages	= filemap_map_pages,
2725	.page_mkwrite	= filemap_page_mkwrite,
2726};
2727
2728/* This is used for a general mmap of a disk file */
2729
2730int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2731{
2732	struct address_space *mapping = file->f_mapping;
2733
2734	if (!mapping->a_ops->readpage)
2735		return -ENOEXEC;
2736	file_accessed(file);
2737	vma->vm_ops = &generic_file_vm_ops;
2738	return 0;
2739}
2740
2741/*
2742 * This is for filesystems which do not implement ->writepage.
2743 */
2744int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2745{
2746	if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2747		return -EINVAL;
2748	return generic_file_mmap(file, vma);
2749}
2750#else
2751int filemap_page_mkwrite(struct vm_fault *vmf)
2752{
2753	return -ENOSYS;
2754}
2755int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2756{
2757	return -ENOSYS;
2758}
2759int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2760{
2761	return -ENOSYS;
2762}
2763#endif /* CONFIG_MMU */
2764
2765EXPORT_SYMBOL(filemap_page_mkwrite);
2766EXPORT_SYMBOL(generic_file_mmap);
2767EXPORT_SYMBOL(generic_file_readonly_mmap);
2768
2769static struct page *wait_on_page_read(struct page *page)
2770{
2771	if (!IS_ERR(page)) {
2772		wait_on_page_locked(page);
2773		if (!PageUptodate(page)) {
2774			put_page(page);
2775			page = ERR_PTR(-EIO);
2776		}
2777	}
2778	return page;
2779}
2780
2781static struct page *do_read_cache_page(struct address_space *mapping,
2782				pgoff_t index,
2783				int (*filler)(void *, struct page *),
2784				void *data,
2785				gfp_t gfp)
2786{
2787	struct page *page;
2788	int err;
2789repeat:
2790	page = find_get_page(mapping, index);
2791	if (!page) {
2792		page = __page_cache_alloc(gfp);
2793		if (!page)
2794			return ERR_PTR(-ENOMEM);
2795		err = add_to_page_cache_lru(page, mapping, index, gfp);
2796		if (unlikely(err)) {
2797			put_page(page);
2798			if (err == -EEXIST)
2799				goto repeat;
2800			/* Presumably ENOMEM for radix tree node */
2801			return ERR_PTR(err);
2802		}
2803
2804filler:
2805		err = filler(data, page);
 
 
 
 
2806		if (err < 0) {
2807			put_page(page);
2808			return ERR_PTR(err);
2809		}
2810
2811		page = wait_on_page_read(page);
2812		if (IS_ERR(page))
2813			return page;
2814		goto out;
2815	}
2816	if (PageUptodate(page))
2817		goto out;
2818
2819	/*
2820	 * Page is not up to date and may be locked due one of the following
2821	 * case a: Page is being filled and the page lock is held
2822	 * case b: Read/write error clearing the page uptodate status
2823	 * case c: Truncation in progress (page locked)
2824	 * case d: Reclaim in progress
2825	 *
2826	 * Case a, the page will be up to date when the page is unlocked.
2827	 *    There is no need to serialise on the page lock here as the page
2828	 *    is pinned so the lock gives no additional protection. Even if the
2829	 *    the page is truncated, the data is still valid if PageUptodate as
2830	 *    it's a race vs truncate race.
2831	 * Case b, the page will not be up to date
2832	 * Case c, the page may be truncated but in itself, the data may still
2833	 *    be valid after IO completes as it's a read vs truncate race. The
2834	 *    operation must restart if the page is not uptodate on unlock but
2835	 *    otherwise serialising on page lock to stabilise the mapping gives
2836	 *    no additional guarantees to the caller as the page lock is
2837	 *    released before return.
2838	 * Case d, similar to truncation. If reclaim holds the page lock, it
2839	 *    will be a race with remove_mapping that determines if the mapping
2840	 *    is valid on unlock but otherwise the data is valid and there is
2841	 *    no need to serialise with page lock.
2842	 *
2843	 * As the page lock gives no additional guarantee, we optimistically
2844	 * wait on the page to be unlocked and check if it's up to date and
2845	 * use the page if it is. Otherwise, the page lock is required to
2846	 * distinguish between the different cases. The motivation is that we
2847	 * avoid spurious serialisations and wakeups when multiple processes
2848	 * wait on the same page for IO to complete.
2849	 */
2850	wait_on_page_locked(page);
2851	if (PageUptodate(page))
2852		goto out;
2853
2854	/* Distinguish between all the cases under the safety of the lock */
2855	lock_page(page);
2856
2857	/* Case c or d, restart the operation */
2858	if (!page->mapping) {
2859		unlock_page(page);
2860		put_page(page);
2861		goto repeat;
2862	}
2863
2864	/* Someone else locked and filled the page in a very small window */
2865	if (PageUptodate(page)) {
2866		unlock_page(page);
2867		goto out;
2868	}
2869	goto filler;
2870
2871out:
2872	mark_page_accessed(page);
2873	return page;
2874}
2875
2876/**
2877 * read_cache_page - read into page cache, fill it if needed
2878 * @mapping:	the page's address_space
2879 * @index:	the page index
2880 * @filler:	function to perform the read
2881 * @data:	first arg to filler(data, page) function, often left as NULL
2882 *
2883 * Read into the page cache. If a page already exists, and PageUptodate() is
2884 * not set, try to fill the page and wait for it to become unlocked.
2885 *
2886 * If the page does not get brought uptodate, return -EIO.
 
 
2887 */
2888struct page *read_cache_page(struct address_space *mapping,
2889				pgoff_t index,
2890				int (*filler)(void *, struct page *),
2891				void *data)
2892{
2893	return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
 
2894}
2895EXPORT_SYMBOL(read_cache_page);
2896
2897/**
2898 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2899 * @mapping:	the page's address_space
2900 * @index:	the page index
2901 * @gfp:	the page allocator flags to use if allocating
2902 *
2903 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2904 * any new page allocations done using the specified allocation flags.
2905 *
2906 * If the page does not get brought uptodate, return -EIO.
 
 
2907 */
2908struct page *read_cache_page_gfp(struct address_space *mapping,
2909				pgoff_t index,
2910				gfp_t gfp)
2911{
2912	filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2913
2914	return do_read_cache_page(mapping, index, filler, NULL, gfp);
2915}
2916EXPORT_SYMBOL(read_cache_page_gfp);
2917
2918/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2919 * Performs necessary checks before doing a write
2920 *
2921 * Can adjust writing position or amount of bytes to write.
2922 * Returns appropriate error code that caller should return or
2923 * zero in case that write should be allowed.
2924 */
2925inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
2926{
2927	struct file *file = iocb->ki_filp;
2928	struct inode *inode = file->f_mapping->host;
2929	unsigned long limit = rlimit(RLIMIT_FSIZE);
2930	loff_t pos;
 
 
 
2931
2932	if (!iov_iter_count(from))
2933		return 0;
2934
2935	/* FIXME: this is for backwards compatibility with 2.4 */
2936	if (iocb->ki_flags & IOCB_APPEND)
2937		iocb->ki_pos = i_size_read(inode);
2938
2939	pos = iocb->ki_pos;
2940
2941	if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT))
2942		return -EINVAL;
2943
2944	if (limit != RLIM_INFINITY) {
2945		if (iocb->ki_pos >= limit) {
2946			send_sig(SIGXFSZ, current, 0);
2947			return -EFBIG;
2948		}
2949		iov_iter_truncate(from, limit - (unsigned long)pos);
2950	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2951
2952	/*
2953	 * LFS rule
 
 
 
 
2954	 */
2955	if (unlikely(pos + iov_iter_count(from) > MAX_NON_LFS &&
2956				!(file->f_flags & O_LARGEFILE))) {
2957		if (pos >= MAX_NON_LFS)
2958			return -EFBIG;
2959		iov_iter_truncate(from, MAX_NON_LFS - (unsigned long)pos);
 
2960	}
2961
 
 
 
 
 
 
2962	/*
2963	 * Are we about to exceed the fs block limit ?
2964	 *
2965	 * If we have written data it becomes a short write.  If we have
2966	 * exceeded without writing data we send a signal and return EFBIG.
2967	 * Linus frestrict idea will clean these up nicely..
2968	 */
2969	if (unlikely(pos >= inode->i_sb->s_maxbytes))
2970		return -EFBIG;
2971
2972	iov_iter_truncate(from, inode->i_sb->s_maxbytes - pos);
2973	return iov_iter_count(from);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2974}
2975EXPORT_SYMBOL(generic_write_checks);
2976
2977int pagecache_write_begin(struct file *file, struct address_space *mapping,
2978				loff_t pos, unsigned len, unsigned flags,
2979				struct page **pagep, void **fsdata)
2980{
2981	const struct address_space_operations *aops = mapping->a_ops;
2982
2983	return aops->write_begin(file, mapping, pos, len, flags,
2984							pagep, fsdata);
2985}
2986EXPORT_SYMBOL(pagecache_write_begin);
2987
2988int pagecache_write_end(struct file *file, struct address_space *mapping,
2989				loff_t pos, unsigned len, unsigned copied,
2990				struct page *page, void *fsdata)
2991{
2992	const struct address_space_operations *aops = mapping->a_ops;
2993
2994	return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2995}
2996EXPORT_SYMBOL(pagecache_write_end);
2997
2998ssize_t
2999generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3000{
3001	struct file	*file = iocb->ki_filp;
3002	struct address_space *mapping = file->f_mapping;
3003	struct inode	*inode = mapping->host;
3004	loff_t		pos = iocb->ki_pos;
3005	ssize_t		written;
3006	size_t		write_len;
3007	pgoff_t		end;
3008
3009	write_len = iov_iter_count(from);
3010	end = (pos + write_len - 1) >> PAGE_SHIFT;
3011
3012	if (iocb->ki_flags & IOCB_NOWAIT) {
3013		/* If there are pages to writeback, return */
3014		if (filemap_range_has_page(inode->i_mapping, pos,
3015					   pos + iov_iter_count(from)))
3016			return -EAGAIN;
3017	} else {
3018		written = filemap_write_and_wait_range(mapping, pos,
3019							pos + write_len - 1);
3020		if (written)
3021			goto out;
3022	}
3023
3024	/*
3025	 * After a write we want buffered reads to be sure to go to disk to get
3026	 * the new data.  We invalidate clean cached page from the region we're
3027	 * about to write.  We do this *before* the write so that we can return
3028	 * without clobbering -EIOCBQUEUED from ->direct_IO().
3029	 */
3030	written = invalidate_inode_pages2_range(mapping,
3031					pos >> PAGE_SHIFT, end);
3032	/*
3033	 * If a page can not be invalidated, return 0 to fall back
3034	 * to buffered write.
3035	 */
3036	if (written) {
3037		if (written == -EBUSY)
3038			return 0;
3039		goto out;
3040	}
3041
3042	written = mapping->a_ops->direct_IO(iocb, from);
3043
3044	/*
3045	 * Finally, try again to invalidate clean pages which might have been
3046	 * cached by non-direct readahead, or faulted in by get_user_pages()
3047	 * if the source of the write was an mmap'ed region of the file
3048	 * we're writing.  Either one is a pretty crazy thing to do,
3049	 * so we don't support it 100%.  If this invalidation
3050	 * fails, tough, the write still worked...
3051	 *
3052	 * Most of the time we do not need this since dio_complete() will do
3053	 * the invalidation for us. However there are some file systems that
3054	 * do not end up with dio_complete() being called, so let's not break
3055	 * them by removing it completely
3056	 */
3057	if (mapping->nrpages)
3058		invalidate_inode_pages2_range(mapping,
3059					pos >> PAGE_SHIFT, end);
3060
3061	if (written > 0) {
3062		pos += written;
3063		write_len -= written;
3064		if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3065			i_size_write(inode, pos);
3066			mark_inode_dirty(inode);
3067		}
3068		iocb->ki_pos = pos;
3069	}
3070	iov_iter_revert(from, write_len - iov_iter_count(from));
3071out:
3072	return written;
3073}
3074EXPORT_SYMBOL(generic_file_direct_write);
3075
3076/*
3077 * Find or create a page at the given pagecache position. Return the locked
3078 * page. This function is specifically for buffered writes.
3079 */
3080struct page *grab_cache_page_write_begin(struct address_space *mapping,
3081					pgoff_t index, unsigned flags)
3082{
3083	struct page *page;
3084	int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
3085
3086	if (flags & AOP_FLAG_NOFS)
3087		fgp_flags |= FGP_NOFS;
3088
3089	page = pagecache_get_page(mapping, index, fgp_flags,
3090			mapping_gfp_mask(mapping));
3091	if (page)
3092		wait_for_stable_page(page);
3093
3094	return page;
3095}
3096EXPORT_SYMBOL(grab_cache_page_write_begin);
3097
3098ssize_t generic_perform_write(struct file *file,
3099				struct iov_iter *i, loff_t pos)
3100{
3101	struct address_space *mapping = file->f_mapping;
3102	const struct address_space_operations *a_ops = mapping->a_ops;
3103	long status = 0;
3104	ssize_t written = 0;
3105	unsigned int flags = 0;
3106
3107	do {
3108		struct page *page;
3109		unsigned long offset;	/* Offset into pagecache page */
3110		unsigned long bytes;	/* Bytes to write to page */
3111		size_t copied;		/* Bytes copied from user */
3112		void *fsdata;
3113
3114		offset = (pos & (PAGE_SIZE - 1));
3115		bytes = min_t(unsigned long, PAGE_SIZE - offset,
3116						iov_iter_count(i));
3117
3118again:
3119		/*
3120		 * Bring in the user page that we will copy from _first_.
3121		 * Otherwise there's a nasty deadlock on copying from the
3122		 * same page as we're writing to, without it being marked
3123		 * up-to-date.
3124		 *
3125		 * Not only is this an optimisation, but it is also required
3126		 * to check that the address is actually valid, when atomic
3127		 * usercopies are used, below.
3128		 */
3129		if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
3130			status = -EFAULT;
3131			break;
3132		}
3133
3134		if (fatal_signal_pending(current)) {
3135			status = -EINTR;
3136			break;
3137		}
3138
3139		status = a_ops->write_begin(file, mapping, pos, bytes, flags,
3140						&page, &fsdata);
3141		if (unlikely(status < 0))
3142			break;
3143
3144		if (mapping_writably_mapped(mapping))
3145			flush_dcache_page(page);
3146
3147		copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
3148		flush_dcache_page(page);
3149
3150		status = a_ops->write_end(file, mapping, pos, bytes, copied,
3151						page, fsdata);
3152		if (unlikely(status < 0))
3153			break;
3154		copied = status;
3155
3156		cond_resched();
3157
3158		iov_iter_advance(i, copied);
3159		if (unlikely(copied == 0)) {
3160			/*
3161			 * If we were unable to copy any data at all, we must
3162			 * fall back to a single segment length write.
3163			 *
3164			 * If we didn't fallback here, we could livelock
3165			 * because not all segments in the iov can be copied at
3166			 * once without a pagefault.
3167			 */
3168			bytes = min_t(unsigned long, PAGE_SIZE - offset,
3169						iov_iter_single_seg_count(i));
3170			goto again;
3171		}
3172		pos += copied;
3173		written += copied;
3174
3175		balance_dirty_pages_ratelimited(mapping);
3176	} while (iov_iter_count(i));
3177
3178	return written ? written : status;
3179}
3180EXPORT_SYMBOL(generic_perform_write);
3181
3182/**
3183 * __generic_file_write_iter - write data to a file
3184 * @iocb:	IO state structure (file, offset, etc.)
3185 * @from:	iov_iter with data to write
3186 *
3187 * This function does all the work needed for actually writing data to a
3188 * file. It does all basic checks, removes SUID from the file, updates
3189 * modification times and calls proper subroutines depending on whether we
3190 * do direct IO or a standard buffered write.
3191 *
3192 * It expects i_mutex to be grabbed unless we work on a block device or similar
3193 * object which does not need locking at all.
3194 *
3195 * This function does *not* take care of syncing data in case of O_SYNC write.
3196 * A caller has to handle it. This is mainly due to the fact that we want to
3197 * avoid syncing under i_mutex.
 
 
 
 
3198 */
3199ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3200{
3201	struct file *file = iocb->ki_filp;
3202	struct address_space * mapping = file->f_mapping;
3203	struct inode 	*inode = mapping->host;
3204	ssize_t		written = 0;
3205	ssize_t		err;
3206	ssize_t		status;
3207
3208	/* We can write back this queue in page reclaim */
3209	current->backing_dev_info = inode_to_bdi(inode);
3210	err = file_remove_privs(file);
3211	if (err)
3212		goto out;
3213
3214	err = file_update_time(file);
3215	if (err)
3216		goto out;
3217
3218	if (iocb->ki_flags & IOCB_DIRECT) {
3219		loff_t pos, endbyte;
3220
3221		written = generic_file_direct_write(iocb, from);
3222		/*
3223		 * If the write stopped short of completing, fall back to
3224		 * buffered writes.  Some filesystems do this for writes to
3225		 * holes, for example.  For DAX files, a buffered write will
3226		 * not succeed (even if it did, DAX does not handle dirty
3227		 * page-cache pages correctly).
3228		 */
3229		if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3230			goto out;
3231
3232		status = generic_perform_write(file, from, pos = iocb->ki_pos);
3233		/*
3234		 * If generic_perform_write() returned a synchronous error
3235		 * then we want to return the number of bytes which were
3236		 * direct-written, or the error code if that was zero.  Note
3237		 * that this differs from normal direct-io semantics, which
3238		 * will return -EFOO even if some bytes were written.
3239		 */
3240		if (unlikely(status < 0)) {
3241			err = status;
3242			goto out;
3243		}
3244		/*
3245		 * We need to ensure that the page cache pages are written to
3246		 * disk and invalidated to preserve the expected O_DIRECT
3247		 * semantics.
3248		 */
3249		endbyte = pos + status - 1;
3250		err = filemap_write_and_wait_range(mapping, pos, endbyte);
3251		if (err == 0) {
3252			iocb->ki_pos = endbyte + 1;
3253			written += status;
3254			invalidate_mapping_pages(mapping,
3255						 pos >> PAGE_SHIFT,
3256						 endbyte >> PAGE_SHIFT);
3257		} else {
3258			/*
3259			 * We don't know how much we wrote, so just return
3260			 * the number of bytes which were direct-written
3261			 */
3262		}
3263	} else {
3264		written = generic_perform_write(file, from, iocb->ki_pos);
3265		if (likely(written > 0))
3266			iocb->ki_pos += written;
3267	}
3268out:
3269	current->backing_dev_info = NULL;
3270	return written ? written : err;
3271}
3272EXPORT_SYMBOL(__generic_file_write_iter);
3273
3274/**
3275 * generic_file_write_iter - write data to a file
3276 * @iocb:	IO state structure
3277 * @from:	iov_iter with data to write
3278 *
3279 * This is a wrapper around __generic_file_write_iter() to be used by most
3280 * filesystems. It takes care of syncing the file in case of O_SYNC file
3281 * and acquires i_mutex as needed.
 
 
 
 
3282 */
3283ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3284{
3285	struct file *file = iocb->ki_filp;
3286	struct inode *inode = file->f_mapping->host;
3287	ssize_t ret;
3288
3289	inode_lock(inode);
3290	ret = generic_write_checks(iocb, from);
3291	if (ret > 0)
3292		ret = __generic_file_write_iter(iocb, from);
3293	inode_unlock(inode);
3294
3295	if (ret > 0)
3296		ret = generic_write_sync(iocb, ret);
3297	return ret;
3298}
3299EXPORT_SYMBOL(generic_file_write_iter);
3300
3301/**
3302 * try_to_release_page() - release old fs-specific metadata on a page
3303 *
3304 * @page: the page which the kernel is trying to free
3305 * @gfp_mask: memory allocation flags (and I/O mode)
3306 *
3307 * The address_space is to try to release any data against the page
3308 * (presumably at page->private).  If the release was successful, return '1'.
3309 * Otherwise return zero.
3310 *
3311 * This may also be called if PG_fscache is set on a page, indicating that the
3312 * page is known to the local caching routines.
3313 *
3314 * The @gfp_mask argument specifies whether I/O may be performed to release
3315 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3316 *
 
3317 */
3318int try_to_release_page(struct page *page, gfp_t gfp_mask)
3319{
3320	struct address_space * const mapping = page->mapping;
3321
3322	BUG_ON(!PageLocked(page));
3323	if (PageWriteback(page))
3324		return 0;
3325
3326	if (mapping && mapping->a_ops->releasepage)
3327		return mapping->a_ops->releasepage(page, gfp_mask);
3328	return try_to_free_buffers(page);
3329}
3330
3331EXPORT_SYMBOL(try_to_release_page);
v5.4
   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 "internal.h"
  45
  46#define CREATE_TRACE_POINTS
  47#include <trace/events/filemap.h>
  48
  49/*
  50 * FIXME: remove all knowledge of the buffer layer from the core VM
  51 */
  52#include <linux/buffer_head.h> /* for try_to_free_buffers */
  53
  54#include <asm/mman.h>
  55
  56/*
  57 * Shared mappings implemented 30.11.1994. It's not fully working yet,
  58 * though.
  59 *
  60 * Shared mappings now work. 15.8.1995  Bruno.
  61 *
  62 * finished 'unifying' the page and buffer cache and SMP-threaded the
  63 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
  64 *
  65 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
  66 */
  67
  68/*
  69 * Lock ordering:
  70 *
  71 *  ->i_mmap_rwsem		(truncate_pagecache)
  72 *    ->private_lock		(__free_pte->__set_page_dirty_buffers)
  73 *      ->swap_lock		(exclusive_swap_page, others)
  74 *        ->i_pages lock
  75 *
  76 *  ->i_mutex
  77 *    ->i_mmap_rwsem		(truncate->unmap_mapping_range)
  78 *
  79 *  ->mmap_sem
  80 *    ->i_mmap_rwsem
  81 *      ->page_table_lock or pte_lock	(various, mainly in memory.c)
  82 *        ->i_pages lock	(arch-dependent flush_dcache_mmap_lock)
  83 *
  84 *  ->mmap_sem
  85 *    ->lock_page		(access_process_vm)
  86 *
  87 *  ->i_mutex			(generic_perform_write)
  88 *    ->mmap_sem		(fault_in_pages_readable->do_page_fault)
  89 *
  90 *  bdi->wb.list_lock
  91 *    sb_lock			(fs/fs-writeback.c)
  92 *    ->i_pages lock		(__sync_single_inode)
  93 *
  94 *  ->i_mmap_rwsem
  95 *    ->anon_vma.lock		(vma_adjust)
  96 *
  97 *  ->anon_vma.lock
  98 *    ->page_table_lock or pte_lock	(anon_vma_prepare and various)
  99 *
 100 *  ->page_table_lock or pte_lock
 101 *    ->swap_lock		(try_to_unmap_one)
 102 *    ->private_lock		(try_to_unmap_one)
 103 *    ->i_pages lock		(try_to_unmap_one)
 104 *    ->pgdat->lru_lock		(follow_page->mark_page_accessed)
 105 *    ->pgdat->lru_lock		(check_pte_range->isolate_lru_page)
 106 *    ->private_lock		(page_remove_rmap->set_page_dirty)
 107 *    ->i_pages lock		(page_remove_rmap->set_page_dirty)
 108 *    bdi.wb->list_lock		(page_remove_rmap->set_page_dirty)
 109 *    ->inode->i_lock		(page_remove_rmap->set_page_dirty)
 110 *    ->memcg->move_lock	(page_remove_rmap->lock_page_memcg)
 111 *    bdi.wb->list_lock		(zap_pte_range->set_page_dirty)
 112 *    ->inode->i_lock		(zap_pte_range->set_page_dirty)
 113 *    ->private_lock		(zap_pte_range->__set_page_dirty_buffers)
 114 *
 115 * ->i_mmap_rwsem
 116 *   ->tasklist_lock            (memory_failure, collect_procs_ao)
 117 */
 118
 119static void page_cache_delete(struct address_space *mapping,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 120				   struct page *page, void *shadow)
 121{
 122	XA_STATE(xas, &mapping->i_pages, page->index);
 123	unsigned int nr = 1;
 124
 125	mapping_set_update(&xas, mapping);
 126
 127	/* hugetlb pages are represented by a single entry in the xarray */
 128	if (!PageHuge(page)) {
 129		xas_set_order(&xas, page->index, compound_order(page));
 130		nr = compound_nr(page);
 131	}
 132
 133	VM_BUG_ON_PAGE(!PageLocked(page), page);
 134	VM_BUG_ON_PAGE(PageTail(page), page);
 135	VM_BUG_ON_PAGE(nr != 1 && shadow, page);
 136
 137	xas_store(&xas, shadow);
 138	xas_init_marks(&xas);
 
 
 
 
 
 
 
 
 
 
 
 139
 140	page->mapping = NULL;
 141	/* Leave page->index set: truncation lookup relies upon it */
 142
 143	if (shadow) {
 144		mapping->nrexceptional += nr;
 145		/*
 146		 * Make sure the nrexceptional update is committed before
 147		 * the nrpages update so that final truncate racing
 148		 * with reclaim does not see both counters 0 at the
 149		 * same time and miss a shadow entry.
 150		 */
 151		smp_wmb();
 152	}
 153	mapping->nrpages -= nr;
 154}
 155
 156static void unaccount_page_cache_page(struct address_space *mapping,
 157				      struct page *page)
 158{
 159	int nr;
 160
 161	/*
 162	 * if we're uptodate, flush out into the cleancache, otherwise
 163	 * invalidate any existing cleancache entries.  We can't leave
 164	 * stale data around in the cleancache once our page is gone
 165	 */
 166	if (PageUptodate(page) && PageMappedToDisk(page))
 167		cleancache_put_page(page);
 168	else
 169		cleancache_invalidate_page(mapping, page);
 170
 171	VM_BUG_ON_PAGE(PageTail(page), page);
 172	VM_BUG_ON_PAGE(page_mapped(page), page);
 173	if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
 174		int mapcount;
 175
 176		pr_alert("BUG: Bad page cache in process %s  pfn:%05lx\n",
 177			 current->comm, page_to_pfn(page));
 178		dump_page(page, "still mapped when deleted");
 179		dump_stack();
 180		add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
 181
 182		mapcount = page_mapcount(page);
 183		if (mapping_exiting(mapping) &&
 184		    page_count(page) >= mapcount + 2) {
 185			/*
 186			 * All vmas have already been torn down, so it's
 187			 * a good bet that actually the page is unmapped,
 188			 * and we'd prefer not to leak it: if we're wrong,
 189			 * some other bad page check should catch it later.
 190			 */
 191			page_mapcount_reset(page);
 192			page_ref_sub(page, mapcount);
 193		}
 194	}
 195
 196	/* hugetlb pages do not participate in page cache accounting. */
 197	if (PageHuge(page))
 198		return;
 199
 200	nr = hpage_nr_pages(page);
 201
 202	__mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, -nr);
 203	if (PageSwapBacked(page)) {
 204		__mod_node_page_state(page_pgdat(page), NR_SHMEM, -nr);
 205		if (PageTransHuge(page))
 206			__dec_node_page_state(page, NR_SHMEM_THPS);
 207	} else if (PageTransHuge(page)) {
 208		__dec_node_page_state(page, NR_FILE_THPS);
 209		filemap_nr_thps_dec(mapping);
 210	}
 211
 212	/*
 213	 * At this point page must be either written or cleaned by
 214	 * truncate.  Dirty page here signals a bug and loss of
 215	 * unwritten data.
 216	 *
 217	 * This fixes dirty accounting after removing the page entirely
 218	 * but leaves PageDirty set: it has no effect for truncated
 219	 * page and anyway will be cleared before returning page into
 220	 * buddy allocator.
 221	 */
 222	if (WARN_ON_ONCE(PageDirty(page)))
 223		account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
 224}
 225
 226/*
 227 * Delete a page from the page cache and free it. Caller has to make
 228 * sure the page is locked and that nobody else uses it - or that usage
 229 * is safe.  The caller must hold the i_pages lock.
 230 */
 231void __delete_from_page_cache(struct page *page, void *shadow)
 232{
 233	struct address_space *mapping = page->mapping;
 234
 235	trace_mm_filemap_delete_from_page_cache(page);
 236
 237	unaccount_page_cache_page(mapping, page);
 238	page_cache_delete(mapping, page, shadow);
 239}
 240
 241static void page_cache_free_page(struct address_space *mapping,
 242				struct page *page)
 243{
 244	void (*freepage)(struct page *);
 245
 246	freepage = mapping->a_ops->freepage;
 247	if (freepage)
 248		freepage(page);
 249
 250	if (PageTransHuge(page) && !PageHuge(page)) {
 251		page_ref_sub(page, HPAGE_PMD_NR);
 252		VM_BUG_ON_PAGE(page_count(page) <= 0, page);
 253	} else {
 254		put_page(page);
 255	}
 256}
 257
 258/**
 259 * delete_from_page_cache - delete page from page cache
 260 * @page: the page which the kernel is trying to remove from page cache
 261 *
 262 * This must be called only on pages that have been verified to be in the page
 263 * cache and locked.  It will never put the page into the free list, the caller
 264 * has a reference on the page.
 265 */
 266void delete_from_page_cache(struct page *page)
 267{
 268	struct address_space *mapping = page_mapping(page);
 269	unsigned long flags;
 270
 271	BUG_ON(!PageLocked(page));
 272	xa_lock_irqsave(&mapping->i_pages, flags);
 273	__delete_from_page_cache(page, NULL);
 274	xa_unlock_irqrestore(&mapping->i_pages, flags);
 275
 276	page_cache_free_page(mapping, page);
 277}
 278EXPORT_SYMBOL(delete_from_page_cache);
 279
 280/*
 281 * page_cache_delete_batch - delete several pages from page cache
 282 * @mapping: the mapping to which pages belong
 283 * @pvec: pagevec with pages to delete
 284 *
 285 * The function walks over mapping->i_pages and removes pages passed in @pvec
 286 * from the mapping. The function expects @pvec to be sorted by page index
 287 * and is optimised for it to be dense.
 288 * It tolerates holes in @pvec (mapping entries at those indices are not
 289 * modified). The function expects only THP head pages to be present in the
 290 * @pvec.
 
 291 *
 292 * The function expects the i_pages lock to be held.
 293 */
 294static void page_cache_delete_batch(struct address_space *mapping,
 
 295			     struct pagevec *pvec)
 296{
 297	XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index);
 
 298	int total_pages = 0;
 299	int i = 0;
 300	struct page *page;
 
 301
 302	mapping_set_update(&xas, mapping);
 303	xas_for_each(&xas, page, ULONG_MAX) {
 304		if (i >= pagevec_count(pvec))
 305			break;
 306
 307		/* A swap/dax/shadow entry got inserted? Skip it. */
 308		if (xa_is_value(page))
 309			continue;
 310		/*
 311		 * A page got inserted in our range? Skip it. We have our
 312		 * pages locked so they are protected from being removed.
 313		 * If we see a page whose index is higher than ours, it
 314		 * means our page has been removed, which shouldn't be
 315		 * possible because we're holding the PageLock.
 316		 */
 317		if (page != pvec->pages[i]) {
 318			VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index,
 319					page);
 320			continue;
 321		}
 322
 323		WARN_ON_ONCE(!PageLocked(page));
 324
 325		if (page->index == xas.xa_index)
 326			page->mapping = NULL;
 327		/* Leave page->index set: truncation lookup relies on it */
 328
 329		/*
 330		 * Move to the next page in the vector if this is a regular
 331		 * page or the index is of the last sub-page of this compound
 332		 * page.
 333		 */
 334		if (page->index + compound_nr(page) - 1 == xas.xa_index)
 335			i++;
 336		xas_store(&xas, NULL);
 
 
 
 
 
 337		total_pages++;
 338	}
 339	mapping->nrpages -= total_pages;
 340}
 341
 342void delete_from_page_cache_batch(struct address_space *mapping,
 343				  struct pagevec *pvec)
 344{
 345	int i;
 346	unsigned long flags;
 347
 348	if (!pagevec_count(pvec))
 349		return;
 350
 351	xa_lock_irqsave(&mapping->i_pages, flags);
 352	for (i = 0; i < pagevec_count(pvec); i++) {
 353		trace_mm_filemap_delete_from_page_cache(pvec->pages[i]);
 354
 355		unaccount_page_cache_page(mapping, pvec->pages[i]);
 356	}
 357	page_cache_delete_batch(mapping, pvec);
 358	xa_unlock_irqrestore(&mapping->i_pages, flags);
 359
 360	for (i = 0; i < pagevec_count(pvec); i++)
 361		page_cache_free_page(mapping, pvec->pages[i]);
 362}
 363
 364int filemap_check_errors(struct address_space *mapping)
 365{
 366	int ret = 0;
 367	/* Check for outstanding write errors */
 368	if (test_bit(AS_ENOSPC, &mapping->flags) &&
 369	    test_and_clear_bit(AS_ENOSPC, &mapping->flags))
 370		ret = -ENOSPC;
 371	if (test_bit(AS_EIO, &mapping->flags) &&
 372	    test_and_clear_bit(AS_EIO, &mapping->flags))
 373		ret = -EIO;
 374	return ret;
 375}
 376EXPORT_SYMBOL(filemap_check_errors);
 377
 378static int filemap_check_and_keep_errors(struct address_space *mapping)
 379{
 380	/* Check for outstanding write errors */
 381	if (test_bit(AS_EIO, &mapping->flags))
 382		return -EIO;
 383	if (test_bit(AS_ENOSPC, &mapping->flags))
 384		return -ENOSPC;
 385	return 0;
 386}
 387
 388/**
 389 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
 390 * @mapping:	address space structure to write
 391 * @start:	offset in bytes where the range starts
 392 * @end:	offset in bytes where the range ends (inclusive)
 393 * @sync_mode:	enable synchronous operation
 394 *
 395 * Start writeback against all of a mapping's dirty pages that lie
 396 * within the byte offsets <start, end> inclusive.
 397 *
 398 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
 399 * opposed to a regular memory cleansing writeback.  The difference between
 400 * these two operations is that if a dirty page/buffer is encountered, it must
 401 * be waited upon, and not just skipped over.
 402 *
 403 * Return: %0 on success, negative error code otherwise.
 404 */
 405int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
 406				loff_t end, int sync_mode)
 407{
 408	int ret;
 409	struct writeback_control wbc = {
 410		.sync_mode = sync_mode,
 411		.nr_to_write = LONG_MAX,
 412		.range_start = start,
 413		.range_end = end,
 414	};
 415
 416	if (!mapping_cap_writeback_dirty(mapping) ||
 417	    !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
 418		return 0;
 419
 420	wbc_attach_fdatawrite_inode(&wbc, mapping->host);
 421	ret = do_writepages(mapping, &wbc);
 422	wbc_detach_inode(&wbc);
 423	return ret;
 424}
 425
 426static inline int __filemap_fdatawrite(struct address_space *mapping,
 427	int sync_mode)
 428{
 429	return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
 430}
 431
 432int filemap_fdatawrite(struct address_space *mapping)
 433{
 434	return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
 435}
 436EXPORT_SYMBOL(filemap_fdatawrite);
 437
 438int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
 439				loff_t end)
 440{
 441	return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
 442}
 443EXPORT_SYMBOL(filemap_fdatawrite_range);
 444
 445/**
 446 * filemap_flush - mostly a non-blocking flush
 447 * @mapping:	target address_space
 448 *
 449 * This is a mostly non-blocking flush.  Not suitable for data-integrity
 450 * purposes - I/O may not be started against all dirty pages.
 451 *
 452 * Return: %0 on success, negative error code otherwise.
 453 */
 454int filemap_flush(struct address_space *mapping)
 455{
 456	return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
 457}
 458EXPORT_SYMBOL(filemap_flush);
 459
 460/**
 461 * filemap_range_has_page - check if a page exists in range.
 462 * @mapping:           address space within which to check
 463 * @start_byte:        offset in bytes where the range starts
 464 * @end_byte:          offset in bytes where the range ends (inclusive)
 465 *
 466 * Find at least one page in the range supplied, usually used to check if
 467 * direct writing in this range will trigger a writeback.
 468 *
 469 * Return: %true if at least one page exists in the specified range,
 470 * %false otherwise.
 471 */
 472bool filemap_range_has_page(struct address_space *mapping,
 473			   loff_t start_byte, loff_t end_byte)
 474{
 
 
 475	struct page *page;
 476	XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
 477	pgoff_t max = end_byte >> PAGE_SHIFT;
 478
 479	if (end_byte < start_byte)
 480		return false;
 481
 482	rcu_read_lock();
 483	for (;;) {
 484		page = xas_find(&xas, max);
 485		if (xas_retry(&xas, page))
 486			continue;
 487		/* Shadow entries don't count */
 488		if (xa_is_value(page))
 489			continue;
 490		/*
 491		 * We don't need to try to pin this page; we're about to
 492		 * release the RCU lock anyway.  It is enough to know that
 493		 * there was a page here recently.
 494		 */
 495		break;
 496	}
 497	rcu_read_unlock();
 498
 499	return page != NULL;
 
 
 
 500}
 501EXPORT_SYMBOL(filemap_range_has_page);
 502
 503static void __filemap_fdatawait_range(struct address_space *mapping,
 504				     loff_t start_byte, loff_t end_byte)
 505{
 506	pgoff_t index = start_byte >> PAGE_SHIFT;
 507	pgoff_t end = end_byte >> PAGE_SHIFT;
 508	struct pagevec pvec;
 509	int nr_pages;
 510
 511	if (end_byte < start_byte)
 512		return;
 513
 514	pagevec_init(&pvec);
 515	while (index <= end) {
 516		unsigned i;
 517
 518		nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
 519				end, PAGECACHE_TAG_WRITEBACK);
 520		if (!nr_pages)
 521			break;
 522
 523		for (i = 0; i < nr_pages; i++) {
 524			struct page *page = pvec.pages[i];
 525
 526			wait_on_page_writeback(page);
 527			ClearPageError(page);
 528		}
 529		pagevec_release(&pvec);
 530		cond_resched();
 531	}
 532}
 533
 534/**
 535 * filemap_fdatawait_range - wait for writeback to complete
 536 * @mapping:		address space structure to wait for
 537 * @start_byte:		offset in bytes where the range starts
 538 * @end_byte:		offset in bytes where the range ends (inclusive)
 539 *
 540 * Walk the list of under-writeback pages of the given address space
 541 * in the given range and wait for all of them.  Check error status of
 542 * the address space and return it.
 543 *
 544 * Since the error status of the address space is cleared by this function,
 545 * callers are responsible for checking the return value and handling and/or
 546 * reporting the error.
 547 *
 548 * Return: error status of the address space.
 549 */
 550int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
 551			    loff_t end_byte)
 552{
 553	__filemap_fdatawait_range(mapping, start_byte, end_byte);
 554	return filemap_check_errors(mapping);
 555}
 556EXPORT_SYMBOL(filemap_fdatawait_range);
 557
 558/**
 559 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
 560 * @mapping:		address space structure to wait for
 561 * @start_byte:		offset in bytes where the range starts
 562 * @end_byte:		offset in bytes where the range ends (inclusive)
 563 *
 564 * Walk the list of under-writeback pages of the given address space in the
 565 * given range and wait for all of them.  Unlike filemap_fdatawait_range(),
 566 * this function does not clear error status of the address space.
 567 *
 568 * Use this function if callers don't handle errors themselves.  Expected
 569 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
 570 * fsfreeze(8)
 571 */
 572int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
 573		loff_t start_byte, loff_t end_byte)
 574{
 575	__filemap_fdatawait_range(mapping, start_byte, end_byte);
 576	return filemap_check_and_keep_errors(mapping);
 577}
 578EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
 579
 580/**
 581 * file_fdatawait_range - wait for writeback to complete
 582 * @file:		file pointing to address space structure to wait for
 583 * @start_byte:		offset in bytes where the range starts
 584 * @end_byte:		offset in bytes where the range ends (inclusive)
 585 *
 586 * Walk the list of under-writeback pages of the address space that file
 587 * refers to, in the given range and wait for all of them.  Check error
 588 * status of the address space vs. the file->f_wb_err cursor and return it.
 589 *
 590 * Since the error status of the file is advanced by this function,
 591 * callers are responsible for checking the return value and handling and/or
 592 * reporting the error.
 593 *
 594 * Return: error status of the address space vs. the file->f_wb_err cursor.
 595 */
 596int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
 597{
 598	struct address_space *mapping = file->f_mapping;
 599
 600	__filemap_fdatawait_range(mapping, start_byte, end_byte);
 601	return file_check_and_advance_wb_err(file);
 602}
 603EXPORT_SYMBOL(file_fdatawait_range);
 604
 605/**
 606 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
 607 * @mapping: address space structure to wait for
 608 *
 609 * Walk the list of under-writeback pages of the given address space
 610 * and wait for all of them.  Unlike filemap_fdatawait(), this function
 611 * does not clear error status of the address space.
 612 *
 613 * Use this function if callers don't handle errors themselves.  Expected
 614 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
 615 * fsfreeze(8)
 616 *
 617 * Return: error status of the address space.
 618 */
 619int filemap_fdatawait_keep_errors(struct address_space *mapping)
 620{
 621	__filemap_fdatawait_range(mapping, 0, LLONG_MAX);
 622	return filemap_check_and_keep_errors(mapping);
 623}
 624EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
 625
 626/* Returns true if writeback might be needed or already in progress. */
 627static bool mapping_needs_writeback(struct address_space *mapping)
 628{
 629	if (dax_mapping(mapping))
 630		return mapping->nrexceptional;
 631
 632	return mapping->nrpages;
 633}
 634
 635int filemap_write_and_wait(struct address_space *mapping)
 636{
 637	int err = 0;
 638
 639	if (mapping_needs_writeback(mapping)) {
 640		err = filemap_fdatawrite(mapping);
 641		/*
 642		 * Even if the above returned error, the pages may be
 643		 * written partially (e.g. -ENOSPC), so we wait for it.
 644		 * But the -EIO is special case, it may indicate the worst
 645		 * thing (e.g. bug) happened, so we avoid waiting for it.
 646		 */
 647		if (err != -EIO) {
 648			int err2 = filemap_fdatawait(mapping);
 649			if (!err)
 650				err = err2;
 651		} else {
 652			/* Clear any previously stored errors */
 653			filemap_check_errors(mapping);
 654		}
 655	} else {
 656		err = filemap_check_errors(mapping);
 657	}
 658	return err;
 659}
 660EXPORT_SYMBOL(filemap_write_and_wait);
 661
 662/**
 663 * filemap_write_and_wait_range - write out & wait on a file range
 664 * @mapping:	the address_space for the pages
 665 * @lstart:	offset in bytes where the range starts
 666 * @lend:	offset in bytes where the range ends (inclusive)
 667 *
 668 * Write out and wait upon file offsets lstart->lend, inclusive.
 669 *
 670 * Note that @lend is inclusive (describes the last byte to be written) so
 671 * that this function can be used to write to the very end-of-file (end = -1).
 672 *
 673 * Return: error status of the address space.
 674 */
 675int filemap_write_and_wait_range(struct address_space *mapping,
 676				 loff_t lstart, loff_t lend)
 677{
 678	int err = 0;
 679
 680	if (mapping_needs_writeback(mapping)) {
 681		err = __filemap_fdatawrite_range(mapping, lstart, lend,
 682						 WB_SYNC_ALL);
 683		/* See comment of filemap_write_and_wait() */
 684		if (err != -EIO) {
 685			int err2 = filemap_fdatawait_range(mapping,
 686						lstart, lend);
 687			if (!err)
 688				err = err2;
 689		} else {
 690			/* Clear any previously stored errors */
 691			filemap_check_errors(mapping);
 692		}
 693	} else {
 694		err = filemap_check_errors(mapping);
 695	}
 696	return err;
 697}
 698EXPORT_SYMBOL(filemap_write_and_wait_range);
 699
 700void __filemap_set_wb_err(struct address_space *mapping, int err)
 701{
 702	errseq_t eseq = errseq_set(&mapping->wb_err, err);
 703
 704	trace_filemap_set_wb_err(mapping, eseq);
 705}
 706EXPORT_SYMBOL(__filemap_set_wb_err);
 707
 708/**
 709 * file_check_and_advance_wb_err - report wb error (if any) that was previously
 710 * 				   and advance wb_err to current one
 711 * @file: struct file on which the error is being reported
 712 *
 713 * When userland calls fsync (or something like nfsd does the equivalent), we
 714 * want to report any writeback errors that occurred since the last fsync (or
 715 * since the file was opened if there haven't been any).
 716 *
 717 * Grab the wb_err from the mapping. If it matches what we have in the file,
 718 * then just quickly return 0. The file is all caught up.
 719 *
 720 * If it doesn't match, then take the mapping value, set the "seen" flag in
 721 * it and try to swap it into place. If it works, or another task beat us
 722 * to it with the new value, then update the f_wb_err and return the error
 723 * portion. The error at this point must be reported via proper channels
 724 * (a'la fsync, or NFS COMMIT operation, etc.).
 725 *
 726 * While we handle mapping->wb_err with atomic operations, the f_wb_err
 727 * value is protected by the f_lock since we must ensure that it reflects
 728 * the latest value swapped in for this file descriptor.
 729 *
 730 * Return: %0 on success, negative error code otherwise.
 731 */
 732int file_check_and_advance_wb_err(struct file *file)
 733{
 734	int err = 0;
 735	errseq_t old = READ_ONCE(file->f_wb_err);
 736	struct address_space *mapping = file->f_mapping;
 737
 738	/* Locklessly handle the common case where nothing has changed */
 739	if (errseq_check(&mapping->wb_err, old)) {
 740		/* Something changed, must use slow path */
 741		spin_lock(&file->f_lock);
 742		old = file->f_wb_err;
 743		err = errseq_check_and_advance(&mapping->wb_err,
 744						&file->f_wb_err);
 745		trace_file_check_and_advance_wb_err(file, old);
 746		spin_unlock(&file->f_lock);
 747	}
 748
 749	/*
 750	 * We're mostly using this function as a drop in replacement for
 751	 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
 752	 * that the legacy code would have had on these flags.
 753	 */
 754	clear_bit(AS_EIO, &mapping->flags);
 755	clear_bit(AS_ENOSPC, &mapping->flags);
 756	return err;
 757}
 758EXPORT_SYMBOL(file_check_and_advance_wb_err);
 759
 760/**
 761 * file_write_and_wait_range - write out & wait on a file range
 762 * @file:	file pointing to address_space with pages
 763 * @lstart:	offset in bytes where the range starts
 764 * @lend:	offset in bytes where the range ends (inclusive)
 765 *
 766 * Write out and wait upon file offsets lstart->lend, inclusive.
 767 *
 768 * Note that @lend is inclusive (describes the last byte to be written) so
 769 * that this function can be used to write to the very end-of-file (end = -1).
 770 *
 771 * After writing out and waiting on the data, we check and advance the
 772 * f_wb_err cursor to the latest value, and return any errors detected there.
 773 *
 774 * Return: %0 on success, negative error code otherwise.
 775 */
 776int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
 777{
 778	int err = 0, err2;
 779	struct address_space *mapping = file->f_mapping;
 780
 781	if (mapping_needs_writeback(mapping)) {
 782		err = __filemap_fdatawrite_range(mapping, lstart, lend,
 783						 WB_SYNC_ALL);
 784		/* See comment of filemap_write_and_wait() */
 785		if (err != -EIO)
 786			__filemap_fdatawait_range(mapping, lstart, lend);
 787	}
 788	err2 = file_check_and_advance_wb_err(file);
 789	if (!err)
 790		err = err2;
 791	return err;
 792}
 793EXPORT_SYMBOL(file_write_and_wait_range);
 794
 795/**
 796 * replace_page_cache_page - replace a pagecache page with a new one
 797 * @old:	page to be replaced
 798 * @new:	page to replace with
 799 * @gfp_mask:	allocation mode
 800 *
 801 * This function replaces a page in the pagecache with a new one.  On
 802 * success it acquires the pagecache reference for the new page and
 803 * drops it for the old page.  Both the old and new pages must be
 804 * locked.  This function does not add the new page to the LRU, the
 805 * caller must do that.
 806 *
 807 * The remove + add is atomic.  This function cannot fail.
 808 *
 809 * Return: %0
 810 */
 811int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
 812{
 813	struct address_space *mapping = old->mapping;
 814	void (*freepage)(struct page *) = mapping->a_ops->freepage;
 815	pgoff_t offset = old->index;
 816	XA_STATE(xas, &mapping->i_pages, offset);
 817	unsigned long flags;
 818
 819	VM_BUG_ON_PAGE(!PageLocked(old), old);
 820	VM_BUG_ON_PAGE(!PageLocked(new), new);
 821	VM_BUG_ON_PAGE(new->mapping, new);
 822
 823	get_page(new);
 824	new->mapping = mapping;
 825	new->index = offset;
 826
 827	xas_lock_irqsave(&xas, flags);
 828	xas_store(&xas, new);
 829
 830	old->mapping = NULL;
 831	/* hugetlb pages do not participate in page cache accounting. */
 832	if (!PageHuge(old))
 833		__dec_node_page_state(new, NR_FILE_PAGES);
 834	if (!PageHuge(new))
 835		__inc_node_page_state(new, NR_FILE_PAGES);
 836	if (PageSwapBacked(old))
 837		__dec_node_page_state(new, NR_SHMEM);
 838	if (PageSwapBacked(new))
 839		__inc_node_page_state(new, NR_SHMEM);
 840	xas_unlock_irqrestore(&xas, flags);
 841	mem_cgroup_migrate(old, new);
 842	if (freepage)
 843		freepage(old);
 844	put_page(old);
 
 
 
 
 
 
 
 
 
 
 845
 846	return 0;
 847}
 848EXPORT_SYMBOL_GPL(replace_page_cache_page);
 849
 850static int __add_to_page_cache_locked(struct page *page,
 851				      struct address_space *mapping,
 852				      pgoff_t offset, gfp_t gfp_mask,
 853				      void **shadowp)
 854{
 855	XA_STATE(xas, &mapping->i_pages, offset);
 856	int huge = PageHuge(page);
 857	struct mem_cgroup *memcg;
 858	int error;
 859	void *old;
 860
 861	VM_BUG_ON_PAGE(!PageLocked(page), page);
 862	VM_BUG_ON_PAGE(PageSwapBacked(page), page);
 863	mapping_set_update(&xas, mapping);
 864
 865	if (!huge) {
 866		error = mem_cgroup_try_charge(page, current->mm,
 867					      gfp_mask, &memcg, false);
 868		if (error)
 869			return error;
 870	}
 871
 
 
 
 
 
 
 
 872	get_page(page);
 873	page->mapping = mapping;
 874	page->index = offset;
 875
 876	do {
 877		xas_lock_irq(&xas);
 878		old = xas_load(&xas);
 879		if (old && !xa_is_value(old))
 880			xas_set_err(&xas, -EEXIST);
 881		xas_store(&xas, page);
 882		if (xas_error(&xas))
 883			goto unlock;
 884
 885		if (xa_is_value(old)) {
 886			mapping->nrexceptional--;
 887			if (shadowp)
 888				*shadowp = old;
 889		}
 890		mapping->nrpages++;
 891
 892		/* hugetlb pages do not participate in page cache accounting */
 893		if (!huge)
 894			__inc_node_page_state(page, NR_FILE_PAGES);
 895unlock:
 896		xas_unlock_irq(&xas);
 897	} while (xas_nomem(&xas, gfp_mask & GFP_RECLAIM_MASK));
 898
 899	if (xas_error(&xas))
 900		goto error;
 901
 
 
 
 
 902	if (!huge)
 903		mem_cgroup_commit_charge(page, memcg, false, false);
 904	trace_mm_filemap_add_to_page_cache(page);
 905	return 0;
 906error:
 907	page->mapping = NULL;
 908	/* Leave page->index set: truncation relies upon it */
 
 909	if (!huge)
 910		mem_cgroup_cancel_charge(page, memcg, false);
 911	put_page(page);
 912	return xas_error(&xas);
 913}
 914ALLOW_ERROR_INJECTION(__add_to_page_cache_locked, ERRNO);
 915
 916/**
 917 * add_to_page_cache_locked - add a locked page to the pagecache
 918 * @page:	page to add
 919 * @mapping:	the page's address_space
 920 * @offset:	page index
 921 * @gfp_mask:	page allocation mode
 922 *
 923 * This function is used to add a page to the pagecache. It must be locked.
 924 * This function does not add the page to the LRU.  The caller must do that.
 925 *
 926 * Return: %0 on success, negative error code otherwise.
 927 */
 928int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
 929		pgoff_t offset, gfp_t gfp_mask)
 930{
 931	return __add_to_page_cache_locked(page, mapping, offset,
 932					  gfp_mask, NULL);
 933}
 934EXPORT_SYMBOL(add_to_page_cache_locked);
 935
 936int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
 937				pgoff_t offset, gfp_t gfp_mask)
 938{
 939	void *shadow = NULL;
 940	int ret;
 941
 942	__SetPageLocked(page);
 943	ret = __add_to_page_cache_locked(page, mapping, offset,
 944					 gfp_mask, &shadow);
 945	if (unlikely(ret))
 946		__ClearPageLocked(page);
 947	else {
 948		/*
 949		 * The page might have been evicted from cache only
 950		 * recently, in which case it should be activated like
 951		 * any other repeatedly accessed page.
 952		 * The exception is pages getting rewritten; evicting other
 953		 * data from the working set, only to cache data that will
 954		 * get overwritten with something else, is a waste of memory.
 955		 */
 956		WARN_ON_ONCE(PageActive(page));
 957		if (!(gfp_mask & __GFP_WRITE) && shadow)
 958			workingset_refault(page, shadow);
 
 
 
 959		lru_cache_add(page);
 960	}
 961	return ret;
 962}
 963EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
 964
 965#ifdef CONFIG_NUMA
 966struct page *__page_cache_alloc(gfp_t gfp)
 967{
 968	int n;
 969	struct page *page;
 970
 971	if (cpuset_do_page_mem_spread()) {
 972		unsigned int cpuset_mems_cookie;
 973		do {
 974			cpuset_mems_cookie = read_mems_allowed_begin();
 975			n = cpuset_mem_spread_node();
 976			page = __alloc_pages_node(n, gfp, 0);
 977		} while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
 978
 979		return page;
 980	}
 981	return alloc_pages(gfp, 0);
 982}
 983EXPORT_SYMBOL(__page_cache_alloc);
 984#endif
 985
 986/*
 987 * In order to wait for pages to become available there must be
 988 * waitqueues associated with pages. By using a hash table of
 989 * waitqueues where the bucket discipline is to maintain all
 990 * waiters on the same queue and wake all when any of the pages
 991 * become available, and for the woken contexts to check to be
 992 * sure the appropriate page became available, this saves space
 993 * at a cost of "thundering herd" phenomena during rare hash
 994 * collisions.
 995 */
 996#define PAGE_WAIT_TABLE_BITS 8
 997#define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
 998static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
 999
1000static wait_queue_head_t *page_waitqueue(struct page *page)
1001{
1002	return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
1003}
1004
1005void __init pagecache_init(void)
1006{
1007	int i;
1008
1009	for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1010		init_waitqueue_head(&page_wait_table[i]);
1011
1012	page_writeback_init();
1013}
1014
1015/* This has the same layout as wait_bit_key - see fs/cachefiles/rdwr.c */
1016struct wait_page_key {
1017	struct page *page;
1018	int bit_nr;
1019	int page_match;
1020};
1021
1022struct wait_page_queue {
1023	struct page *page;
1024	int bit_nr;
1025	wait_queue_entry_t wait;
1026};
1027
1028static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1029{
1030	struct wait_page_key *key = arg;
1031	struct wait_page_queue *wait_page
1032		= container_of(wait, struct wait_page_queue, wait);
1033
1034	if (wait_page->page != key->page)
1035	       return 0;
1036	key->page_match = 1;
1037
1038	if (wait_page->bit_nr != key->bit_nr)
1039		return 0;
1040
1041	/*
1042	 * Stop walking if it's locked.
1043	 * Is this safe if put_and_wait_on_page_locked() is in use?
1044	 * Yes: the waker must hold a reference to this page, and if PG_locked
1045	 * has now already been set by another task, that task must also hold
1046	 * a reference to the *same usage* of this page; so there is no need
1047	 * to walk on to wake even the put_and_wait_on_page_locked() callers.
1048	 */
1049	if (test_bit(key->bit_nr, &key->page->flags))
1050		return -1;
1051
1052	return autoremove_wake_function(wait, mode, sync, key);
1053}
1054
1055static void wake_up_page_bit(struct page *page, int bit_nr)
1056{
1057	wait_queue_head_t *q = page_waitqueue(page);
1058	struct wait_page_key key;
1059	unsigned long flags;
1060	wait_queue_entry_t bookmark;
1061
1062	key.page = page;
1063	key.bit_nr = bit_nr;
1064	key.page_match = 0;
1065
1066	bookmark.flags = 0;
1067	bookmark.private = NULL;
1068	bookmark.func = NULL;
1069	INIT_LIST_HEAD(&bookmark.entry);
1070
1071	spin_lock_irqsave(&q->lock, flags);
1072	__wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1073
1074	while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1075		/*
1076		 * Take a breather from holding the lock,
1077		 * allow pages that finish wake up asynchronously
1078		 * to acquire the lock and remove themselves
1079		 * from wait queue
1080		 */
1081		spin_unlock_irqrestore(&q->lock, flags);
1082		cpu_relax();
1083		spin_lock_irqsave(&q->lock, flags);
1084		__wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1085	}
1086
1087	/*
1088	 * It is possible for other pages to have collided on the waitqueue
1089	 * hash, so in that case check for a page match. That prevents a long-
1090	 * term waiter
1091	 *
1092	 * It is still possible to miss a case here, when we woke page waiters
1093	 * and removed them from the waitqueue, but there are still other
1094	 * page waiters.
1095	 */
1096	if (!waitqueue_active(q) || !key.page_match) {
1097		ClearPageWaiters(page);
1098		/*
1099		 * It's possible to miss clearing Waiters here, when we woke
1100		 * our page waiters, but the hashed waitqueue has waiters for
1101		 * other pages on it.
1102		 *
1103		 * That's okay, it's a rare case. The next waker will clear it.
1104		 */
1105	}
1106	spin_unlock_irqrestore(&q->lock, flags);
1107}
1108
1109static void wake_up_page(struct page *page, int bit)
1110{
1111	if (!PageWaiters(page))
1112		return;
1113	wake_up_page_bit(page, bit);
1114}
1115
1116/*
1117 * A choice of three behaviors for wait_on_page_bit_common():
1118 */
1119enum behavior {
1120	EXCLUSIVE,	/* Hold ref to page and take the bit when woken, like
1121			 * __lock_page() waiting on then setting PG_locked.
1122			 */
1123	SHARED,		/* Hold ref to page and check the bit when woken, like
1124			 * wait_on_page_writeback() waiting on PG_writeback.
1125			 */
1126	DROP,		/* Drop ref to page before wait, no check when woken,
1127			 * like put_and_wait_on_page_locked() on PG_locked.
1128			 */
1129};
1130
1131static inline int wait_on_page_bit_common(wait_queue_head_t *q,
1132	struct page *page, int bit_nr, int state, enum behavior behavior)
1133{
1134	struct wait_page_queue wait_page;
1135	wait_queue_entry_t *wait = &wait_page.wait;
1136	bool bit_is_set;
1137	bool thrashing = false;
1138	bool delayacct = false;
1139	unsigned long pflags;
1140	int ret = 0;
1141
1142	if (bit_nr == PG_locked &&
1143	    !PageUptodate(page) && PageWorkingset(page)) {
1144		if (!PageSwapBacked(page)) {
1145			delayacct_thrashing_start();
1146			delayacct = true;
1147		}
1148		psi_memstall_enter(&pflags);
1149		thrashing = true;
1150	}
1151
1152	init_wait(wait);
1153	wait->flags = behavior == EXCLUSIVE ? WQ_FLAG_EXCLUSIVE : 0;
1154	wait->func = wake_page_function;
1155	wait_page.page = page;
1156	wait_page.bit_nr = bit_nr;
1157
1158	for (;;) {
1159		spin_lock_irq(&q->lock);
1160
1161		if (likely(list_empty(&wait->entry))) {
1162			__add_wait_queue_entry_tail(q, wait);
1163			SetPageWaiters(page);
1164		}
1165
1166		set_current_state(state);
1167
1168		spin_unlock_irq(&q->lock);
1169
1170		bit_is_set = test_bit(bit_nr, &page->flags);
1171		if (behavior == DROP)
1172			put_page(page);
1173
1174		if (likely(bit_is_set))
1175			io_schedule();
 
1176
1177		if (behavior == EXCLUSIVE) {
1178			if (!test_and_set_bit_lock(bit_nr, &page->flags))
1179				break;
1180		} else if (behavior == SHARED) {
1181			if (!test_bit(bit_nr, &page->flags))
1182				break;
1183		}
1184
1185		if (signal_pending_state(state, current)) {
1186			ret = -EINTR;
1187			break;
1188		}
1189
1190		if (behavior == DROP) {
1191			/*
1192			 * We can no longer safely access page->flags:
1193			 * even if CONFIG_MEMORY_HOTREMOVE is not enabled,
1194			 * there is a risk of waiting forever on a page reused
1195			 * for something that keeps it locked indefinitely.
1196			 * But best check for -EINTR above before breaking.
1197			 */
1198			break;
1199		}
1200	}
1201
1202	finish_wait(q, wait);
1203
1204	if (thrashing) {
1205		if (delayacct)
1206			delayacct_thrashing_end();
1207		psi_memstall_leave(&pflags);
1208	}
1209
1210	/*
1211	 * A signal could leave PageWaiters set. Clearing it here if
1212	 * !waitqueue_active would be possible (by open-coding finish_wait),
1213	 * but still fail to catch it in the case of wait hash collision. We
1214	 * already can fail to clear wait hash collision cases, so don't
1215	 * bother with signals either.
1216	 */
1217
1218	return ret;
1219}
1220
1221void wait_on_page_bit(struct page *page, int bit_nr)
1222{
1223	wait_queue_head_t *q = page_waitqueue(page);
1224	wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1225}
1226EXPORT_SYMBOL(wait_on_page_bit);
1227
1228int wait_on_page_bit_killable(struct page *page, int bit_nr)
1229{
1230	wait_queue_head_t *q = page_waitqueue(page);
1231	return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED);
1232}
1233EXPORT_SYMBOL(wait_on_page_bit_killable);
1234
1235/**
1236 * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1237 * @page: The page to wait for.
1238 *
1239 * The caller should hold a reference on @page.  They expect the page to
1240 * become unlocked relatively soon, but do not wish to hold up migration
1241 * (for example) by holding the reference while waiting for the page to
1242 * come unlocked.  After this function returns, the caller should not
1243 * dereference @page.
1244 */
1245void put_and_wait_on_page_locked(struct page *page)
1246{
1247	wait_queue_head_t *q;
1248
1249	page = compound_head(page);
1250	q = page_waitqueue(page);
1251	wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, DROP);
1252}
1253
1254/**
1255 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1256 * @page: Page defining the wait queue of interest
1257 * @waiter: Waiter to add to the queue
1258 *
1259 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1260 */
1261void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
1262{
1263	wait_queue_head_t *q = page_waitqueue(page);
1264	unsigned long flags;
1265
1266	spin_lock_irqsave(&q->lock, flags);
1267	__add_wait_queue_entry_tail(q, waiter);
1268	SetPageWaiters(page);
1269	spin_unlock_irqrestore(&q->lock, flags);
1270}
1271EXPORT_SYMBOL_GPL(add_page_wait_queue);
1272
1273#ifndef clear_bit_unlock_is_negative_byte
1274
1275/*
1276 * PG_waiters is the high bit in the same byte as PG_lock.
1277 *
1278 * On x86 (and on many other architectures), we can clear PG_lock and
1279 * test the sign bit at the same time. But if the architecture does
1280 * not support that special operation, we just do this all by hand
1281 * instead.
1282 *
1283 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1284 * being cleared, but a memory barrier should be unneccssary since it is
1285 * in the same byte as PG_locked.
1286 */
1287static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1288{
1289	clear_bit_unlock(nr, mem);
1290	/* smp_mb__after_atomic(); */
1291	return test_bit(PG_waiters, mem);
1292}
1293
1294#endif
1295
1296/**
1297 * unlock_page - unlock a locked page
1298 * @page: the page
1299 *
1300 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
1301 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1302 * mechanism between PageLocked pages and PageWriteback pages is shared.
1303 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1304 *
1305 * Note that this depends on PG_waiters being the sign bit in the byte
1306 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1307 * clear the PG_locked bit and test PG_waiters at the same time fairly
1308 * portably (architectures that do LL/SC can test any bit, while x86 can
1309 * test the sign bit).
1310 */
1311void unlock_page(struct page *page)
1312{
1313	BUILD_BUG_ON(PG_waiters != 7);
1314	page = compound_head(page);
1315	VM_BUG_ON_PAGE(!PageLocked(page), page);
1316	if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
1317		wake_up_page_bit(page, PG_locked);
1318}
1319EXPORT_SYMBOL(unlock_page);
1320
1321/**
1322 * end_page_writeback - end writeback against a page
1323 * @page: the page
1324 */
1325void end_page_writeback(struct page *page)
1326{
1327	/*
1328	 * TestClearPageReclaim could be used here but it is an atomic
1329	 * operation and overkill in this particular case. Failing to
1330	 * shuffle a page marked for immediate reclaim is too mild to
1331	 * justify taking an atomic operation penalty at the end of
1332	 * ever page writeback.
1333	 */
1334	if (PageReclaim(page)) {
1335		ClearPageReclaim(page);
1336		rotate_reclaimable_page(page);
1337	}
1338
1339	if (!test_clear_page_writeback(page))
1340		BUG();
1341
1342	smp_mb__after_atomic();
1343	wake_up_page(page, PG_writeback);
1344}
1345EXPORT_SYMBOL(end_page_writeback);
1346
1347/*
1348 * After completing I/O on a page, call this routine to update the page
1349 * flags appropriately
1350 */
1351void page_endio(struct page *page, bool is_write, int err)
1352{
1353	if (!is_write) {
1354		if (!err) {
1355			SetPageUptodate(page);
1356		} else {
1357			ClearPageUptodate(page);
1358			SetPageError(page);
1359		}
1360		unlock_page(page);
1361	} else {
1362		if (err) {
1363			struct address_space *mapping;
1364
1365			SetPageError(page);
1366			mapping = page_mapping(page);
1367			if (mapping)
1368				mapping_set_error(mapping, err);
1369		}
1370		end_page_writeback(page);
1371	}
1372}
1373EXPORT_SYMBOL_GPL(page_endio);
1374
1375/**
1376 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1377 * @__page: the page to lock
1378 */
1379void __lock_page(struct page *__page)
1380{
1381	struct page *page = compound_head(__page);
1382	wait_queue_head_t *q = page_waitqueue(page);
1383	wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE,
1384				EXCLUSIVE);
1385}
1386EXPORT_SYMBOL(__lock_page);
1387
1388int __lock_page_killable(struct page *__page)
1389{
1390	struct page *page = compound_head(__page);
1391	wait_queue_head_t *q = page_waitqueue(page);
1392	return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE,
1393					EXCLUSIVE);
1394}
1395EXPORT_SYMBOL_GPL(__lock_page_killable);
1396
1397/*
1398 * Return values:
1399 * 1 - page is locked; mmap_sem is still held.
1400 * 0 - page is not locked.
1401 *     mmap_sem has been released (up_read()), unless flags had both
1402 *     FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1403 *     which case mmap_sem is still held.
1404 *
1405 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1406 * with the page locked and the mmap_sem unperturbed.
1407 */
1408int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1409			 unsigned int flags)
1410{
1411	if (flags & FAULT_FLAG_ALLOW_RETRY) {
1412		/*
1413		 * CAUTION! In this case, mmap_sem is not released
1414		 * even though return 0.
1415		 */
1416		if (flags & FAULT_FLAG_RETRY_NOWAIT)
1417			return 0;
1418
1419		up_read(&mm->mmap_sem);
1420		if (flags & FAULT_FLAG_KILLABLE)
1421			wait_on_page_locked_killable(page);
1422		else
1423			wait_on_page_locked(page);
1424		return 0;
1425	} else {
1426		if (flags & FAULT_FLAG_KILLABLE) {
1427			int ret;
1428
1429			ret = __lock_page_killable(page);
1430			if (ret) {
1431				up_read(&mm->mmap_sem);
1432				return 0;
1433			}
1434		} else
1435			__lock_page(page);
1436		return 1;
1437	}
1438}
1439
1440/**
1441 * page_cache_next_miss() - Find the next gap in the page cache.
1442 * @mapping: Mapping.
1443 * @index: Index.
1444 * @max_scan: Maximum range to search.
1445 *
1446 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1447 * gap with the lowest index.
1448 *
1449 * This function may be called under the rcu_read_lock.  However, this will
1450 * not atomically search a snapshot of the cache at a single point in time.
1451 * For example, if a gap is created at index 5, then subsequently a gap is
1452 * created at index 10, page_cache_next_miss covering both indices may
1453 * return 10 if called under the rcu_read_lock.
1454 *
1455 * Return: The index of the gap if found, otherwise an index outside the
1456 * range specified (in which case 'return - index >= max_scan' will be true).
1457 * In the rare case of index wrap-around, 0 will be returned.
 
 
1458 */
1459pgoff_t page_cache_next_miss(struct address_space *mapping,
1460			     pgoff_t index, unsigned long max_scan)
1461{
1462	XA_STATE(xas, &mapping->i_pages, index);
1463
1464	while (max_scan--) {
1465		void *entry = xas_next(&xas);
1466		if (!entry || xa_is_value(entry))
 
 
1467			break;
1468		if (xas.xa_index == 0)
 
1469			break;
1470	}
1471
1472	return xas.xa_index;
1473}
1474EXPORT_SYMBOL(page_cache_next_miss);
1475
1476/**
1477 * page_cache_prev_miss() - Find the previous gap in the page cache.
1478 * @mapping: Mapping.
1479 * @index: Index.
1480 * @max_scan: Maximum range to search.
1481 *
1482 * Search the range [max(index - max_scan + 1, 0), index] for the
1483 * gap with the highest index.
1484 *
1485 * This function may be called under the rcu_read_lock.  However, this will
1486 * not atomically search a snapshot of the cache at a single point in time.
1487 * For example, if a gap is created at index 10, then subsequently a gap is
1488 * created at index 5, page_cache_prev_miss() covering both indices may
1489 * return 5 if called under the rcu_read_lock.
1490 *
1491 * Return: The index of the gap if found, otherwise an index outside the
1492 * range specified (in which case 'index - return >= max_scan' will be true).
1493 * In the rare case of wrap-around, ULONG_MAX will be returned.
 
 
 
1494 */
1495pgoff_t page_cache_prev_miss(struct address_space *mapping,
1496			     pgoff_t index, unsigned long max_scan)
1497{
1498	XA_STATE(xas, &mapping->i_pages, index);
 
 
 
1499
1500	while (max_scan--) {
1501		void *entry = xas_prev(&xas);
1502		if (!entry || xa_is_value(entry))
1503			break;
1504		if (xas.xa_index == ULONG_MAX)
 
1505			break;
1506	}
1507
1508	return xas.xa_index;
1509}
1510EXPORT_SYMBOL(page_cache_prev_miss);
1511
1512/**
1513 * find_get_entry - find and get a page cache entry
1514 * @mapping: the address_space to search
1515 * @offset: the page cache index
1516 *
1517 * Looks up the page cache slot at @mapping & @offset.  If there is a
1518 * page cache page, it is returned with an increased refcount.
1519 *
1520 * If the slot holds a shadow entry of a previously evicted page, or a
1521 * swap entry from shmem/tmpfs, it is returned.
1522 *
1523 * Return: the found page or shadow entry, %NULL if nothing is found.
1524 */
1525struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1526{
1527	XA_STATE(xas, &mapping->i_pages, offset);
1528	struct page *page;
1529
1530	rcu_read_lock();
1531repeat:
1532	xas_reset(&xas);
1533	page = xas_load(&xas);
1534	if (xas_retry(&xas, page))
1535		goto repeat;
1536	/*
1537	 * A shadow entry of a recently evicted page, or a swap entry from
1538	 * shmem/tmpfs.  Return it without attempting to raise page count.
1539	 */
1540	if (!page || xa_is_value(page))
1541		goto out;
 
 
 
 
 
 
 
 
 
 
1542
1543	if (!page_cache_get_speculative(page))
1544		goto repeat;
 
 
 
1545
1546	/*
1547	 * Has the page moved or been split?
1548	 * This is part of the lockless pagecache protocol. See
1549	 * include/linux/pagemap.h for details.
1550	 */
1551	if (unlikely(page != xas_reload(&xas))) {
1552		put_page(page);
1553		goto repeat;
 
1554	}
1555	page = find_subpage(page, offset);
1556out:
1557	rcu_read_unlock();
1558
1559	return page;
1560}
1561EXPORT_SYMBOL(find_get_entry);
1562
1563/**
1564 * find_lock_entry - locate, pin and lock a page cache entry
1565 * @mapping: the address_space to search
1566 * @offset: the page cache index
1567 *
1568 * Looks up the page cache slot at @mapping & @offset.  If there is a
1569 * page cache page, it is returned locked and with an increased
1570 * refcount.
1571 *
1572 * If the slot holds a shadow entry of a previously evicted page, or a
1573 * swap entry from shmem/tmpfs, it is returned.
1574 *
 
 
1575 * find_lock_entry() may sleep.
1576 *
1577 * Return: the found page or shadow entry, %NULL if nothing is found.
1578 */
1579struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1580{
1581	struct page *page;
1582
1583repeat:
1584	page = find_get_entry(mapping, offset);
1585	if (page && !xa_is_value(page)) {
1586		lock_page(page);
1587		/* Has the page been truncated? */
1588		if (unlikely(page_mapping(page) != mapping)) {
1589			unlock_page(page);
1590			put_page(page);
1591			goto repeat;
1592		}
1593		VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
1594	}
1595	return page;
1596}
1597EXPORT_SYMBOL(find_lock_entry);
1598
1599/**
1600 * pagecache_get_page - find and get a page reference
1601 * @mapping: the address_space to search
1602 * @offset: the page index
1603 * @fgp_flags: PCG flags
1604 * @gfp_mask: gfp mask to use for the page cache data page allocation
1605 *
1606 * Looks up the page cache slot at @mapping & @offset.
1607 *
1608 * PCG flags modify how the page is returned.
1609 *
1610 * @fgp_flags can be:
1611 *
1612 * - FGP_ACCESSED: the page will be marked accessed
1613 * - FGP_LOCK: Page is return locked
1614 * - FGP_CREAT: If page is not present then a new page is allocated using
1615 *   @gfp_mask and added to the page cache and the VM's LRU
1616 *   list. The page is returned locked and with an increased
1617 *   refcount.
1618 * - FGP_FOR_MMAP: Similar to FGP_CREAT, only we want to allow the caller to do
1619 *   its own locking dance if the page is already in cache, or unlock the page
1620 *   before returning if we had to add the page to pagecache.
1621 *
1622 * If FGP_LOCK or FGP_CREAT are specified then the function may sleep even
1623 * if the GFP flags specified for FGP_CREAT are atomic.
1624 *
1625 * If there is a page cache page, it is returned with an increased refcount.
1626 *
1627 * Return: the found page or %NULL otherwise.
1628 */
1629struct page *pagecache_get_page(struct address_space *mapping, pgoff_t offset,
1630	int fgp_flags, gfp_t gfp_mask)
1631{
1632	struct page *page;
1633
1634repeat:
1635	page = find_get_entry(mapping, offset);
1636	if (xa_is_value(page))
1637		page = NULL;
1638	if (!page)
1639		goto no_page;
1640
1641	if (fgp_flags & FGP_LOCK) {
1642		if (fgp_flags & FGP_NOWAIT) {
1643			if (!trylock_page(page)) {
1644				put_page(page);
1645				return NULL;
1646			}
1647		} else {
1648			lock_page(page);
1649		}
1650
1651		/* Has the page been truncated? */
1652		if (unlikely(compound_head(page)->mapping != mapping)) {
1653			unlock_page(page);
1654			put_page(page);
1655			goto repeat;
1656		}
1657		VM_BUG_ON_PAGE(page->index != offset, page);
1658	}
1659
1660	if (fgp_flags & FGP_ACCESSED)
1661		mark_page_accessed(page);
1662
1663no_page:
1664	if (!page && (fgp_flags & FGP_CREAT)) {
1665		int err;
1666		if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1667			gfp_mask |= __GFP_WRITE;
1668		if (fgp_flags & FGP_NOFS)
1669			gfp_mask &= ~__GFP_FS;
1670
1671		page = __page_cache_alloc(gfp_mask);
1672		if (!page)
1673			return NULL;
1674
1675		if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1676			fgp_flags |= FGP_LOCK;
1677
1678		/* Init accessed so avoid atomic mark_page_accessed later */
1679		if (fgp_flags & FGP_ACCESSED)
1680			__SetPageReferenced(page);
1681
1682		err = add_to_page_cache_lru(page, mapping, offset, gfp_mask);
1683		if (unlikely(err)) {
1684			put_page(page);
1685			page = NULL;
1686			if (err == -EEXIST)
1687				goto repeat;
1688		}
1689
1690		/*
1691		 * add_to_page_cache_lru locks the page, and for mmap we expect
1692		 * an unlocked page.
1693		 */
1694		if (page && (fgp_flags & FGP_FOR_MMAP))
1695			unlock_page(page);
1696	}
1697
1698	return page;
1699}
1700EXPORT_SYMBOL(pagecache_get_page);
1701
1702/**
1703 * find_get_entries - gang pagecache lookup
1704 * @mapping:	The address_space to search
1705 * @start:	The starting page cache index
1706 * @nr_entries:	The maximum number of entries
1707 * @entries:	Where the resulting entries are placed
1708 * @indices:	The cache indices corresponding to the entries in @entries
1709 *
1710 * find_get_entries() will search for and return a group of up to
1711 * @nr_entries entries in the mapping.  The entries are placed at
1712 * @entries.  find_get_entries() takes a reference against any actual
1713 * pages it returns.
1714 *
1715 * The search returns a group of mapping-contiguous page cache entries
1716 * with ascending indexes.  There may be holes in the indices due to
1717 * not-present pages.
1718 *
1719 * Any shadow entries of evicted pages, or swap entries from
1720 * shmem/tmpfs, are included in the returned array.
1721 *
1722 * Return: the number of pages and shadow entries which were found.
 
1723 */
1724unsigned find_get_entries(struct address_space *mapping,
1725			  pgoff_t start, unsigned int nr_entries,
1726			  struct page **entries, pgoff_t *indices)
1727{
1728	XA_STATE(xas, &mapping->i_pages, start);
1729	struct page *page;
1730	unsigned int ret = 0;
 
1731
1732	if (!nr_entries)
1733		return 0;
1734
1735	rcu_read_lock();
1736	xas_for_each(&xas, page, ULONG_MAX) {
1737		if (xas_retry(&xas, page))
 
 
 
1738			continue;
1739		/*
1740		 * A shadow entry of a recently evicted page, a swap
1741		 * entry from shmem/tmpfs or a DAX entry.  Return it
1742		 * without attempting to raise page count.
1743		 */
1744		if (xa_is_value(page))
 
 
 
 
1745			goto export;
 
1746
1747		if (!page_cache_get_speculative(page))
1748			goto retry;
 
1749
1750		/* Has the page moved or been split? */
1751		if (unlikely(page != xas_reload(&xas)))
1752			goto put_page;
1753		page = find_subpage(page, xas.xa_index);
 
1754
 
 
 
 
 
1755export:
1756		indices[ret] = xas.xa_index;
1757		entries[ret] = page;
1758		if (++ret == nr_entries)
1759			break;
1760		continue;
1761put_page:
1762		put_page(page);
1763retry:
1764		xas_reset(&xas);
1765	}
1766	rcu_read_unlock();
1767	return ret;
1768}
1769
1770/**
1771 * find_get_pages_range - gang pagecache lookup
1772 * @mapping:	The address_space to search
1773 * @start:	The starting page index
1774 * @end:	The final page index (inclusive)
1775 * @nr_pages:	The maximum number of pages
1776 * @pages:	Where the resulting pages are placed
1777 *
1778 * find_get_pages_range() will search for and return a group of up to @nr_pages
1779 * pages in the mapping starting at index @start and up to index @end
1780 * (inclusive).  The pages are placed at @pages.  find_get_pages_range() takes
1781 * a reference against the returned pages.
1782 *
1783 * The search returns a group of mapping-contiguous pages with ascending
1784 * indexes.  There may be holes in the indices due to not-present pages.
1785 * We also update @start to index the next page for the traversal.
1786 *
1787 * Return: the number of pages which were found. If this number is
1788 * smaller than @nr_pages, the end of specified range has been
1789 * reached.
1790 */
1791unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
1792			      pgoff_t end, unsigned int nr_pages,
1793			      struct page **pages)
1794{
1795	XA_STATE(xas, &mapping->i_pages, *start);
1796	struct page *page;
1797	unsigned ret = 0;
1798
1799	if (unlikely(!nr_pages))
1800		return 0;
1801
1802	rcu_read_lock();
1803	xas_for_each(&xas, page, end) {
1804		if (xas_retry(&xas, page))
 
 
 
 
 
 
1805			continue;
1806		/* Skip over shadow, swap and DAX entries */
1807		if (xa_is_value(page))
 
 
 
 
 
 
 
 
 
1808			continue;
 
1809
1810		if (!page_cache_get_speculative(page))
1811			goto retry;
 
 
 
 
 
 
 
1812
1813		/* Has the page moved or been split? */
1814		if (unlikely(page != xas_reload(&xas)))
1815			goto put_page;
 
 
1816
1817		pages[ret] = find_subpage(page, xas.xa_index);
1818		if (++ret == nr_pages) {
1819			*start = xas.xa_index + 1;
1820			goto out;
1821		}
1822		continue;
1823put_page:
1824		put_page(page);
1825retry:
1826		xas_reset(&xas);
1827	}
1828
1829	/*
1830	 * We come here when there is no page beyond @end. We take care to not
1831	 * overflow the index @start as it confuses some of the callers. This
1832	 * breaks the iteration when there is a page at index -1 but that is
1833	 * already broken anyway.
1834	 */
1835	if (end == (pgoff_t)-1)
1836		*start = (pgoff_t)-1;
1837	else
1838		*start = end + 1;
1839out:
1840	rcu_read_unlock();
1841
1842	return ret;
1843}
1844
1845/**
1846 * find_get_pages_contig - gang contiguous pagecache lookup
1847 * @mapping:	The address_space to search
1848 * @index:	The starting page index
1849 * @nr_pages:	The maximum number of pages
1850 * @pages:	Where the resulting pages are placed
1851 *
1852 * find_get_pages_contig() works exactly like find_get_pages(), except
1853 * that the returned number of pages are guaranteed to be contiguous.
1854 *
1855 * Return: the number of pages which were found.
1856 */
1857unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1858			       unsigned int nr_pages, struct page **pages)
1859{
1860	XA_STATE(xas, &mapping->i_pages, index);
1861	struct page *page;
1862	unsigned int ret = 0;
1863
1864	if (unlikely(!nr_pages))
1865		return 0;
1866
1867	rcu_read_lock();
1868	for (page = xas_load(&xas); page; page = xas_next(&xas)) {
1869		if (xas_retry(&xas, page))
1870			continue;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1871		/*
1872		 * If the entry has been swapped out, we can stop looking.
1873		 * No current caller is looking for DAX entries.
 
1874		 */
1875		if (xa_is_value(page))
 
1876			break;
 
1877
1878		if (!page_cache_get_speculative(page))
1879			goto retry;
1880
1881		/* Has the page moved or been split? */
1882		if (unlikely(page != xas_reload(&xas)))
1883			goto put_page;
1884
1885		pages[ret] = find_subpage(page, xas.xa_index);
1886		if (++ret == nr_pages)
1887			break;
1888		continue;
1889put_page:
1890		put_page(page);
1891retry:
1892		xas_reset(&xas);
1893	}
1894	rcu_read_unlock();
1895	return ret;
1896}
1897EXPORT_SYMBOL(find_get_pages_contig);
1898
1899/**
1900 * find_get_pages_range_tag - find and return pages in given range matching @tag
1901 * @mapping:	the address_space to search
1902 * @index:	the starting page index
1903 * @end:	The final page index (inclusive)
1904 * @tag:	the tag index
1905 * @nr_pages:	the maximum number of pages
1906 * @pages:	where the resulting pages are placed
1907 *
1908 * Like find_get_pages, except we only return pages which are tagged with
1909 * @tag.   We update @index to index the next page for the traversal.
1910 *
1911 * Return: the number of pages which were found.
1912 */
1913unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
1914			pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
1915			struct page **pages)
1916{
1917	XA_STATE(xas, &mapping->i_pages, *index);
1918	struct page *page;
1919	unsigned ret = 0;
1920
1921	if (unlikely(!nr_pages))
1922		return 0;
1923
1924	rcu_read_lock();
1925	xas_for_each_marked(&xas, page, end, tag) {
1926		if (xas_retry(&xas, page))
 
 
 
 
 
 
1927			continue;
1928		/*
1929		 * Shadow entries should never be tagged, but this iteration
1930		 * is lockless so there is a window for page reclaim to evict
1931		 * a page we saw tagged.  Skip over it.
1932		 */
1933		if (xa_is_value(page))
 
 
 
 
 
 
 
 
 
 
 
1934			continue;
 
1935
1936		if (!page_cache_get_speculative(page))
1937			goto retry;
 
1938
1939		/* Has the page moved or been split? */
1940		if (unlikely(page != xas_reload(&xas)))
1941			goto put_page;
 
 
 
 
 
 
 
 
1942
1943		pages[ret] = find_subpage(page, xas.xa_index);
1944		if (++ret == nr_pages) {
1945			*index = xas.xa_index + 1;
1946			goto out;
1947		}
1948		continue;
1949put_page:
1950		put_page(page);
1951retry:
1952		xas_reset(&xas);
1953	}
1954
1955	/*
1956	 * We come here when we got to @end. We take care to not overflow the
1957	 * index @index as it confuses some of the callers. This breaks the
1958	 * iteration when there is a page at index -1 but that is already
1959	 * broken anyway.
1960	 */
1961	if (end == (pgoff_t)-1)
1962		*index = (pgoff_t)-1;
1963	else
1964		*index = end + 1;
1965out:
1966	rcu_read_unlock();
1967
1968	return ret;
1969}
1970EXPORT_SYMBOL(find_get_pages_range_tag);
1971
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1972/*
1973 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1974 * a _large_ part of the i/o request. Imagine the worst scenario:
1975 *
1976 *      ---R__________________________________________B__________
1977 *         ^ reading here                             ^ bad block(assume 4k)
1978 *
1979 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1980 * => failing the whole request => read(R) => read(R+1) =>
1981 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1982 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1983 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1984 *
1985 * It is going insane. Fix it by quickly scaling down the readahead size.
1986 */
1987static void shrink_readahead_size_eio(struct file *filp,
1988					struct file_ra_state *ra)
1989{
1990	ra->ra_pages /= 4;
1991}
1992
1993/**
1994 * generic_file_buffered_read - generic file read routine
1995 * @iocb:	the iocb to read
1996 * @iter:	data destination
1997 * @written:	already copied
1998 *
1999 * This is a generic file read routine, and uses the
2000 * mapping->a_ops->readpage() function for the actual low-level stuff.
2001 *
2002 * This is really ugly. But the goto's actually try to clarify some
2003 * of the logic when it comes to error handling etc.
2004 *
2005 * Return:
2006 * * total number of bytes copied, including those the were already @written
2007 * * negative error code if nothing was copied
2008 */
2009static ssize_t generic_file_buffered_read(struct kiocb *iocb,
2010		struct iov_iter *iter, ssize_t written)
2011{
2012	struct file *filp = iocb->ki_filp;
2013	struct address_space *mapping = filp->f_mapping;
2014	struct inode *inode = mapping->host;
2015	struct file_ra_state *ra = &filp->f_ra;
2016	loff_t *ppos = &iocb->ki_pos;
2017	pgoff_t index;
2018	pgoff_t last_index;
2019	pgoff_t prev_index;
2020	unsigned long offset;      /* offset into pagecache page */
2021	unsigned int prev_offset;
2022	int error = 0;
2023
2024	if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
2025		return 0;
2026	iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2027
2028	index = *ppos >> PAGE_SHIFT;
2029	prev_index = ra->prev_pos >> PAGE_SHIFT;
2030	prev_offset = ra->prev_pos & (PAGE_SIZE-1);
2031	last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
2032	offset = *ppos & ~PAGE_MASK;
2033
2034	for (;;) {
2035		struct page *page;
2036		pgoff_t end_index;
2037		loff_t isize;
2038		unsigned long nr, ret;
2039
2040		cond_resched();
2041find_page:
2042		if (fatal_signal_pending(current)) {
2043			error = -EINTR;
2044			goto out;
2045		}
2046
2047		page = find_get_page(mapping, index);
2048		if (!page) {
2049			if (iocb->ki_flags & IOCB_NOWAIT)
2050				goto would_block;
2051			page_cache_sync_readahead(mapping,
2052					ra, filp,
2053					index, last_index - index);
2054			page = find_get_page(mapping, index);
2055			if (unlikely(page == NULL))
2056				goto no_cached_page;
2057		}
2058		if (PageReadahead(page)) {
2059			page_cache_async_readahead(mapping,
2060					ra, filp, page,
2061					index, last_index - index);
2062		}
2063		if (!PageUptodate(page)) {
2064			if (iocb->ki_flags & IOCB_NOWAIT) {
2065				put_page(page);
2066				goto would_block;
2067			}
2068
2069			/*
2070			 * See comment in do_read_cache_page on why
2071			 * wait_on_page_locked is used to avoid unnecessarily
2072			 * serialisations and why it's safe.
2073			 */
2074			error = wait_on_page_locked_killable(page);
2075			if (unlikely(error))
2076				goto readpage_error;
2077			if (PageUptodate(page))
2078				goto page_ok;
2079
2080			if (inode->i_blkbits == PAGE_SHIFT ||
2081					!mapping->a_ops->is_partially_uptodate)
2082				goto page_not_up_to_date;
2083			/* pipes can't handle partially uptodate pages */
2084			if (unlikely(iov_iter_is_pipe(iter)))
2085				goto page_not_up_to_date;
2086			if (!trylock_page(page))
2087				goto page_not_up_to_date;
2088			/* Did it get truncated before we got the lock? */
2089			if (!page->mapping)
2090				goto page_not_up_to_date_locked;
2091			if (!mapping->a_ops->is_partially_uptodate(page,
2092							offset, iter->count))
2093				goto page_not_up_to_date_locked;
2094			unlock_page(page);
2095		}
2096page_ok:
2097		/*
2098		 * i_size must be checked after we know the page is Uptodate.
2099		 *
2100		 * Checking i_size after the check allows us to calculate
2101		 * the correct value for "nr", which means the zero-filled
2102		 * part of the page is not copied back to userspace (unless
2103		 * another truncate extends the file - this is desired though).
2104		 */
2105
2106		isize = i_size_read(inode);
2107		end_index = (isize - 1) >> PAGE_SHIFT;
2108		if (unlikely(!isize || index > end_index)) {
2109			put_page(page);
2110			goto out;
2111		}
2112
2113		/* nr is the maximum number of bytes to copy from this page */
2114		nr = PAGE_SIZE;
2115		if (index == end_index) {
2116			nr = ((isize - 1) & ~PAGE_MASK) + 1;
2117			if (nr <= offset) {
2118				put_page(page);
2119				goto out;
2120			}
2121		}
2122		nr = nr - offset;
2123
2124		/* If users can be writing to this page using arbitrary
2125		 * virtual addresses, take care about potential aliasing
2126		 * before reading the page on the kernel side.
2127		 */
2128		if (mapping_writably_mapped(mapping))
2129			flush_dcache_page(page);
2130
2131		/*
2132		 * When a sequential read accesses a page several times,
2133		 * only mark it as accessed the first time.
2134		 */
2135		if (prev_index != index || offset != prev_offset)
2136			mark_page_accessed(page);
2137		prev_index = index;
2138
2139		/*
2140		 * Ok, we have the page, and it's up-to-date, so
2141		 * now we can copy it to user space...
2142		 */
2143
2144		ret = copy_page_to_iter(page, offset, nr, iter);
2145		offset += ret;
2146		index += offset >> PAGE_SHIFT;
2147		offset &= ~PAGE_MASK;
2148		prev_offset = offset;
2149
2150		put_page(page);
2151		written += ret;
2152		if (!iov_iter_count(iter))
2153			goto out;
2154		if (ret < nr) {
2155			error = -EFAULT;
2156			goto out;
2157		}
2158		continue;
2159
2160page_not_up_to_date:
2161		/* Get exclusive access to the page ... */
2162		error = lock_page_killable(page);
2163		if (unlikely(error))
2164			goto readpage_error;
2165
2166page_not_up_to_date_locked:
2167		/* Did it get truncated before we got the lock? */
2168		if (!page->mapping) {
2169			unlock_page(page);
2170			put_page(page);
2171			continue;
2172		}
2173
2174		/* Did somebody else fill it already? */
2175		if (PageUptodate(page)) {
2176			unlock_page(page);
2177			goto page_ok;
2178		}
2179
2180readpage:
2181		/*
2182		 * A previous I/O error may have been due to temporary
2183		 * failures, eg. multipath errors.
2184		 * PG_error will be set again if readpage fails.
2185		 */
2186		ClearPageError(page);
2187		/* Start the actual read. The read will unlock the page. */
2188		error = mapping->a_ops->readpage(filp, page);
2189
2190		if (unlikely(error)) {
2191			if (error == AOP_TRUNCATED_PAGE) {
2192				put_page(page);
2193				error = 0;
2194				goto find_page;
2195			}
2196			goto readpage_error;
2197		}
2198
2199		if (!PageUptodate(page)) {
2200			error = lock_page_killable(page);
2201			if (unlikely(error))
2202				goto readpage_error;
2203			if (!PageUptodate(page)) {
2204				if (page->mapping == NULL) {
2205					/*
2206					 * invalidate_mapping_pages got it
2207					 */
2208					unlock_page(page);
2209					put_page(page);
2210					goto find_page;
2211				}
2212				unlock_page(page);
2213				shrink_readahead_size_eio(filp, ra);
2214				error = -EIO;
2215				goto readpage_error;
2216			}
2217			unlock_page(page);
2218		}
2219
2220		goto page_ok;
2221
2222readpage_error:
2223		/* UHHUH! A synchronous read error occurred. Report it */
2224		put_page(page);
2225		goto out;
2226
2227no_cached_page:
2228		/*
2229		 * Ok, it wasn't cached, so we need to create a new
2230		 * page..
2231		 */
2232		page = page_cache_alloc(mapping);
2233		if (!page) {
2234			error = -ENOMEM;
2235			goto out;
2236		}
2237		error = add_to_page_cache_lru(page, mapping, index,
2238				mapping_gfp_constraint(mapping, GFP_KERNEL));
2239		if (error) {
2240			put_page(page);
2241			if (error == -EEXIST) {
2242				error = 0;
2243				goto find_page;
2244			}
2245			goto out;
2246		}
2247		goto readpage;
2248	}
2249
2250would_block:
2251	error = -EAGAIN;
2252out:
2253	ra->prev_pos = prev_index;
2254	ra->prev_pos <<= PAGE_SHIFT;
2255	ra->prev_pos |= prev_offset;
2256
2257	*ppos = ((loff_t)index << PAGE_SHIFT) + offset;
2258	file_accessed(filp);
2259	return written ? written : error;
2260}
2261
2262/**
2263 * generic_file_read_iter - generic filesystem read routine
2264 * @iocb:	kernel I/O control block
2265 * @iter:	destination for the data read
2266 *
2267 * This is the "read_iter()" routine for all filesystems
2268 * that can use the page cache directly.
2269 * Return:
2270 * * number of bytes copied, even for partial reads
2271 * * negative error code if nothing was read
2272 */
2273ssize_t
2274generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2275{
2276	size_t count = iov_iter_count(iter);
2277	ssize_t retval = 0;
2278
2279	if (!count)
2280		goto out; /* skip atime */
2281
2282	if (iocb->ki_flags & IOCB_DIRECT) {
2283		struct file *file = iocb->ki_filp;
2284		struct address_space *mapping = file->f_mapping;
2285		struct inode *inode = mapping->host;
2286		loff_t size;
2287
2288		size = i_size_read(inode);
2289		if (iocb->ki_flags & IOCB_NOWAIT) {
2290			if (filemap_range_has_page(mapping, iocb->ki_pos,
2291						   iocb->ki_pos + count - 1))
2292				return -EAGAIN;
2293		} else {
2294			retval = filemap_write_and_wait_range(mapping,
2295						iocb->ki_pos,
2296					        iocb->ki_pos + count - 1);
2297			if (retval < 0)
2298				goto out;
2299		}
2300
2301		file_accessed(file);
2302
2303		retval = mapping->a_ops->direct_IO(iocb, iter);
2304		if (retval >= 0) {
2305			iocb->ki_pos += retval;
2306			count -= retval;
2307		}
2308		iov_iter_revert(iter, count - iov_iter_count(iter));
2309
2310		/*
2311		 * Btrfs can have a short DIO read if we encounter
2312		 * compressed extents, so if there was an error, or if
2313		 * we've already read everything we wanted to, or if
2314		 * there was a short read because we hit EOF, go ahead
2315		 * and return.  Otherwise fallthrough to buffered io for
2316		 * the rest of the read.  Buffered reads will not work for
2317		 * DAX files, so don't bother trying.
2318		 */
2319		if (retval < 0 || !count || iocb->ki_pos >= size ||
2320		    IS_DAX(inode))
2321			goto out;
2322	}
2323
2324	retval = generic_file_buffered_read(iocb, iter, retval);
2325out:
2326	return retval;
2327}
2328EXPORT_SYMBOL(generic_file_read_iter);
2329
2330#ifdef CONFIG_MMU
2331#define MMAP_LOTSAMISS  (100)
2332static struct file *maybe_unlock_mmap_for_io(struct vm_fault *vmf,
2333					     struct file *fpin)
 
 
 
 
 
 
 
2334{
2335	int flags = vmf->flags;
 
 
2336
2337	if (fpin)
2338		return fpin;
 
 
2339
2340	/*
2341	 * FAULT_FLAG_RETRY_NOWAIT means we don't want to wait on page locks or
2342	 * anything, so we only pin the file and drop the mmap_sem if only
2343	 * FAULT_FLAG_ALLOW_RETRY is set.
2344	 */
2345	if ((flags & (FAULT_FLAG_ALLOW_RETRY | FAULT_FLAG_RETRY_NOWAIT)) ==
2346	    FAULT_FLAG_ALLOW_RETRY) {
2347		fpin = get_file(vmf->vma->vm_file);
2348		up_read(&vmf->vma->vm_mm->mmap_sem);
2349	}
2350	return fpin;
2351}
2352
2353/*
2354 * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_sem
2355 * @vmf - the vm_fault for this fault.
2356 * @page - the page to lock.
2357 * @fpin - the pointer to the file we may pin (or is already pinned).
2358 *
2359 * This works similar to lock_page_or_retry in that it can drop the mmap_sem.
2360 * It differs in that it actually returns the page locked if it returns 1 and 0
2361 * if it couldn't lock the page.  If we did have to drop the mmap_sem then fpin
2362 * will point to the pinned file and needs to be fput()'ed at a later point.
2363 */
2364static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page,
2365				     struct file **fpin)
2366{
2367	if (trylock_page(page))
2368		return 1;
2369
2370	/*
2371	 * NOTE! This will make us return with VM_FAULT_RETRY, but with
2372	 * the mmap_sem still held. That's how FAULT_FLAG_RETRY_NOWAIT
2373	 * is supposed to work. We have way too many special cases..
2374	 */
2375	if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
2376		return 0;
2377
2378	*fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
2379	if (vmf->flags & FAULT_FLAG_KILLABLE) {
2380		if (__lock_page_killable(page)) {
2381			/*
2382			 * We didn't have the right flags to drop the mmap_sem,
2383			 * but all fault_handlers only check for fatal signals
2384			 * if we return VM_FAULT_RETRY, so we need to drop the
2385			 * mmap_sem here and return 0 if we don't have a fpin.
2386			 */
2387			if (*fpin == NULL)
2388				up_read(&vmf->vma->vm_mm->mmap_sem);
2389			return 0;
2390		}
2391	} else
2392		__lock_page(page);
2393	return 1;
2394}
2395
 
2396
2397/*
2398 * Synchronous readahead happens when we don't even find a page in the page
2399 * cache at all.  We don't want to perform IO under the mmap sem, so if we have
2400 * to drop the mmap sem we return the file that was pinned in order for us to do
2401 * that.  If we didn't pin a file then we return NULL.  The file that is
2402 * returned needs to be fput()'ed when we're done with it.
2403 */
2404static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
 
 
 
2405{
2406	struct file *file = vmf->vma->vm_file;
2407	struct file_ra_state *ra = &file->f_ra;
2408	struct address_space *mapping = file->f_mapping;
2409	struct file *fpin = NULL;
2410	pgoff_t offset = vmf->pgoff;
2411
2412	/* If we don't want any read-ahead, don't bother */
2413	if (vmf->vma->vm_flags & VM_RAND_READ)
2414		return fpin;
2415	if (!ra->ra_pages)
2416		return fpin;
2417
2418	if (vmf->vma->vm_flags & VM_SEQ_READ) {
2419		fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2420		page_cache_sync_readahead(mapping, ra, file, offset,
2421					  ra->ra_pages);
2422		return fpin;
2423	}
2424
2425	/* Avoid banging the cache line if not needed */
2426	if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
2427		ra->mmap_miss++;
2428
2429	/*
2430	 * Do we miss much more than hit in this file? If so,
2431	 * stop bothering with read-ahead. It will only hurt.
2432	 */
2433	if (ra->mmap_miss > MMAP_LOTSAMISS)
2434		return fpin;
2435
2436	/*
2437	 * mmap read-around
2438	 */
2439	fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2440	ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2441	ra->size = ra->ra_pages;
2442	ra->async_size = ra->ra_pages / 4;
2443	ra_submit(ra, mapping, file);
2444	return fpin;
2445}
2446
2447/*
2448 * Asynchronous readahead happens when we find the page and PG_readahead,
2449 * so we want to possibly extend the readahead further.  We return the file that
2450 * was pinned if we have to drop the mmap_sem in order to do IO.
2451 */
2452static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
2453					    struct page *page)
 
 
 
2454{
2455	struct file *file = vmf->vma->vm_file;
2456	struct file_ra_state *ra = &file->f_ra;
2457	struct address_space *mapping = file->f_mapping;
2458	struct file *fpin = NULL;
2459	pgoff_t offset = vmf->pgoff;
2460
2461	/* If we don't want any read-ahead, don't bother */
2462	if (vmf->vma->vm_flags & VM_RAND_READ)
2463		return fpin;
2464	if (ra->mmap_miss > 0)
2465		ra->mmap_miss--;
2466	if (PageReadahead(page)) {
2467		fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2468		page_cache_async_readahead(mapping, ra, file,
2469					   page, offset, ra->ra_pages);
2470	}
2471	return fpin;
2472}
2473
2474/**
2475 * filemap_fault - read in file data for page fault handling
2476 * @vmf:	struct vm_fault containing details of the fault
2477 *
2478 * filemap_fault() is invoked via the vma operations vector for a
2479 * mapped memory region to read in file data during a page fault.
2480 *
2481 * The goto's are kind of ugly, but this streamlines the normal case of having
2482 * it in the page cache, and handles the special cases reasonably without
2483 * having a lot of duplicated code.
2484 *
2485 * vma->vm_mm->mmap_sem must be held on entry.
2486 *
2487 * If our return value has VM_FAULT_RETRY set, it's because the mmap_sem
2488 * may be dropped before doing I/O or by lock_page_maybe_drop_mmap().
 
 
2489 *
2490 * If our return value does not have VM_FAULT_RETRY set, the mmap_sem
2491 * has not been released.
2492 *
2493 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2494 *
2495 * Return: bitwise-OR of %VM_FAULT_ codes.
2496 */
2497vm_fault_t filemap_fault(struct vm_fault *vmf)
2498{
2499	int error;
2500	struct file *file = vmf->vma->vm_file;
2501	struct file *fpin = NULL;
2502	struct address_space *mapping = file->f_mapping;
2503	struct file_ra_state *ra = &file->f_ra;
2504	struct inode *inode = mapping->host;
2505	pgoff_t offset = vmf->pgoff;
2506	pgoff_t max_off;
2507	struct page *page;
2508	vm_fault_t ret = 0;
2509
2510	max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2511	if (unlikely(offset >= max_off))
2512		return VM_FAULT_SIGBUS;
2513
2514	/*
2515	 * Do we have something in the page cache already?
2516	 */
2517	page = find_get_page(mapping, offset);
2518	if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2519		/*
2520		 * We found the page, so try async readahead before
2521		 * waiting for the lock.
2522		 */
2523		fpin = do_async_mmap_readahead(vmf, page);
2524	} else if (!page) {
2525		/* No page in the page cache at all */
 
2526		count_vm_event(PGMAJFAULT);
2527		count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
2528		ret = VM_FAULT_MAJOR;
2529		fpin = do_sync_mmap_readahead(vmf);
2530retry_find:
2531		page = pagecache_get_page(mapping, offset,
2532					  FGP_CREAT|FGP_FOR_MMAP,
2533					  vmf->gfp_mask);
2534		if (!page) {
2535			if (fpin)
2536				goto out_retry;
2537			return vmf_error(-ENOMEM);
2538		}
2539	}
2540
2541	if (!lock_page_maybe_drop_mmap(vmf, page, &fpin))
2542		goto out_retry;
 
 
2543
2544	/* Did it get truncated? */
2545	if (unlikely(compound_head(page)->mapping != mapping)) {
2546		unlock_page(page);
2547		put_page(page);
2548		goto retry_find;
2549	}
2550	VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
2551
2552	/*
2553	 * We have a locked page in the page cache, now we need to check
2554	 * that it's up-to-date. If not, it is going to be due to an error.
2555	 */
2556	if (unlikely(!PageUptodate(page)))
2557		goto page_not_uptodate;
2558
2559	/*
2560	 * We've made it this far and we had to drop our mmap_sem, now is the
2561	 * time to return to the upper layer and have it re-find the vma and
2562	 * redo the fault.
2563	 */
2564	if (fpin) {
2565		unlock_page(page);
2566		goto out_retry;
2567	}
2568
2569	/*
2570	 * Found the page and have a reference on it.
2571	 * We must recheck i_size under page lock.
2572	 */
2573	max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2574	if (unlikely(offset >= max_off)) {
2575		unlock_page(page);
2576		put_page(page);
2577		return VM_FAULT_SIGBUS;
2578	}
2579
2580	vmf->page = page;
2581	return ret | VM_FAULT_LOCKED;
2582
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2583page_not_uptodate:
2584	/*
2585	 * Umm, take care of errors if the page isn't up-to-date.
2586	 * Try to re-read it _once_. We do this synchronously,
2587	 * because there really aren't any performance issues here
2588	 * and we need to check for errors.
2589	 */
2590	ClearPageError(page);
2591	fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2592	error = mapping->a_ops->readpage(file, page);
2593	if (!error) {
2594		wait_on_page_locked(page);
2595		if (!PageUptodate(page))
2596			error = -EIO;
2597	}
2598	if (fpin)
2599		goto out_retry;
2600	put_page(page);
2601
2602	if (!error || error == AOP_TRUNCATED_PAGE)
2603		goto retry_find;
2604
2605	/* Things didn't work out. Return zero to tell the mm layer so. */
2606	shrink_readahead_size_eio(file, ra);
2607	return VM_FAULT_SIGBUS;
2608
2609out_retry:
2610	/*
2611	 * We dropped the mmap_sem, we need to return to the fault handler to
2612	 * re-find the vma and come back and find our hopefully still populated
2613	 * page.
2614	 */
2615	if (page)
2616		put_page(page);
2617	if (fpin)
2618		fput(fpin);
2619	return ret | VM_FAULT_RETRY;
2620}
2621EXPORT_SYMBOL(filemap_fault);
2622
2623void filemap_map_pages(struct vm_fault *vmf,
2624		pgoff_t start_pgoff, pgoff_t end_pgoff)
2625{
 
 
2626	struct file *file = vmf->vma->vm_file;
2627	struct address_space *mapping = file->f_mapping;
2628	pgoff_t last_pgoff = start_pgoff;
2629	unsigned long max_idx;
2630	XA_STATE(xas, &mapping->i_pages, start_pgoff);
2631	struct page *page;
2632
2633	rcu_read_lock();
2634	xas_for_each(&xas, page, end_pgoff) {
2635		if (xas_retry(&xas, page))
2636			continue;
2637		if (xa_is_value(page))
 
 
 
 
 
 
 
 
2638			goto next;
 
2639
2640		/*
2641		 * Check for a locked page first, as a speculative
2642		 * reference may adversely influence page migration.
2643		 */
2644		if (PageLocked(page))
2645			goto next;
2646		if (!page_cache_get_speculative(page))
2647			goto next;
 
2648
2649		/* Has the page moved or been split? */
2650		if (unlikely(page != xas_reload(&xas)))
2651			goto skip;
2652		page = find_subpage(page, xas.xa_index);
 
2653
2654		if (!PageUptodate(page) ||
2655				PageReadahead(page) ||
2656				PageHWPoison(page))
2657			goto skip;
2658		if (!trylock_page(page))
2659			goto skip;
2660
2661		if (page->mapping != mapping || !PageUptodate(page))
2662			goto unlock;
2663
2664		max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
2665		if (page->index >= max_idx)
2666			goto unlock;
2667
2668		if (file->f_ra.mmap_miss > 0)
2669			file->f_ra.mmap_miss--;
2670
2671		vmf->address += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
2672		if (vmf->pte)
2673			vmf->pte += xas.xa_index - last_pgoff;
2674		last_pgoff = xas.xa_index;
2675		if (alloc_set_pte(vmf, NULL, page))
2676			goto unlock;
2677		unlock_page(page);
2678		goto next;
2679unlock:
2680		unlock_page(page);
2681skip:
2682		put_page(page);
2683next:
2684		/* Huge page is mapped? No need to proceed. */
2685		if (pmd_trans_huge(*vmf->pmd))
2686			break;
 
 
2687	}
2688	rcu_read_unlock();
2689}
2690EXPORT_SYMBOL(filemap_map_pages);
2691
2692vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
2693{
2694	struct page *page = vmf->page;
2695	struct inode *inode = file_inode(vmf->vma->vm_file);
2696	vm_fault_t ret = VM_FAULT_LOCKED;
2697
2698	sb_start_pagefault(inode->i_sb);
2699	file_update_time(vmf->vma->vm_file);
2700	lock_page(page);
2701	if (page->mapping != inode->i_mapping) {
2702		unlock_page(page);
2703		ret = VM_FAULT_NOPAGE;
2704		goto out;
2705	}
2706	/*
2707	 * We mark the page dirty already here so that when freeze is in
2708	 * progress, we are guaranteed that writeback during freezing will
2709	 * see the dirty page and writeprotect it again.
2710	 */
2711	set_page_dirty(page);
2712	wait_for_stable_page(page);
2713out:
2714	sb_end_pagefault(inode->i_sb);
2715	return ret;
2716}
2717
2718const struct vm_operations_struct generic_file_vm_ops = {
2719	.fault		= filemap_fault,
2720	.map_pages	= filemap_map_pages,
2721	.page_mkwrite	= filemap_page_mkwrite,
2722};
2723
2724/* This is used for a general mmap of a disk file */
2725
2726int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2727{
2728	struct address_space *mapping = file->f_mapping;
2729
2730	if (!mapping->a_ops->readpage)
2731		return -ENOEXEC;
2732	file_accessed(file);
2733	vma->vm_ops = &generic_file_vm_ops;
2734	return 0;
2735}
2736
2737/*
2738 * This is for filesystems which do not implement ->writepage.
2739 */
2740int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2741{
2742	if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2743		return -EINVAL;
2744	return generic_file_mmap(file, vma);
2745}
2746#else
2747vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
2748{
2749	return VM_FAULT_SIGBUS;
2750}
2751int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2752{
2753	return -ENOSYS;
2754}
2755int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2756{
2757	return -ENOSYS;
2758}
2759#endif /* CONFIG_MMU */
2760
2761EXPORT_SYMBOL(filemap_page_mkwrite);
2762EXPORT_SYMBOL(generic_file_mmap);
2763EXPORT_SYMBOL(generic_file_readonly_mmap);
2764
2765static struct page *wait_on_page_read(struct page *page)
2766{
2767	if (!IS_ERR(page)) {
2768		wait_on_page_locked(page);
2769		if (!PageUptodate(page)) {
2770			put_page(page);
2771			page = ERR_PTR(-EIO);
2772		}
2773	}
2774	return page;
2775}
2776
2777static struct page *do_read_cache_page(struct address_space *mapping,
2778				pgoff_t index,
2779				int (*filler)(void *, struct page *),
2780				void *data,
2781				gfp_t gfp)
2782{
2783	struct page *page;
2784	int err;
2785repeat:
2786	page = find_get_page(mapping, index);
2787	if (!page) {
2788		page = __page_cache_alloc(gfp);
2789		if (!page)
2790			return ERR_PTR(-ENOMEM);
2791		err = add_to_page_cache_lru(page, mapping, index, gfp);
2792		if (unlikely(err)) {
2793			put_page(page);
2794			if (err == -EEXIST)
2795				goto repeat;
2796			/* Presumably ENOMEM for xarray node */
2797			return ERR_PTR(err);
2798		}
2799
2800filler:
2801		if (filler)
2802			err = filler(data, page);
2803		else
2804			err = mapping->a_ops->readpage(data, page);
2805
2806		if (err < 0) {
2807			put_page(page);
2808			return ERR_PTR(err);
2809		}
2810
2811		page = wait_on_page_read(page);
2812		if (IS_ERR(page))
2813			return page;
2814		goto out;
2815	}
2816	if (PageUptodate(page))
2817		goto out;
2818
2819	/*
2820	 * Page is not up to date and may be locked due one of the following
2821	 * case a: Page is being filled and the page lock is held
2822	 * case b: Read/write error clearing the page uptodate status
2823	 * case c: Truncation in progress (page locked)
2824	 * case d: Reclaim in progress
2825	 *
2826	 * Case a, the page will be up to date when the page is unlocked.
2827	 *    There is no need to serialise on the page lock here as the page
2828	 *    is pinned so the lock gives no additional protection. Even if the
2829	 *    the page is truncated, the data is still valid if PageUptodate as
2830	 *    it's a race vs truncate race.
2831	 * Case b, the page will not be up to date
2832	 * Case c, the page may be truncated but in itself, the data may still
2833	 *    be valid after IO completes as it's a read vs truncate race. The
2834	 *    operation must restart if the page is not uptodate on unlock but
2835	 *    otherwise serialising on page lock to stabilise the mapping gives
2836	 *    no additional guarantees to the caller as the page lock is
2837	 *    released before return.
2838	 * Case d, similar to truncation. If reclaim holds the page lock, it
2839	 *    will be a race with remove_mapping that determines if the mapping
2840	 *    is valid on unlock but otherwise the data is valid and there is
2841	 *    no need to serialise with page lock.
2842	 *
2843	 * As the page lock gives no additional guarantee, we optimistically
2844	 * wait on the page to be unlocked and check if it's up to date and
2845	 * use the page if it is. Otherwise, the page lock is required to
2846	 * distinguish between the different cases. The motivation is that we
2847	 * avoid spurious serialisations and wakeups when multiple processes
2848	 * wait on the same page for IO to complete.
2849	 */
2850	wait_on_page_locked(page);
2851	if (PageUptodate(page))
2852		goto out;
2853
2854	/* Distinguish between all the cases under the safety of the lock */
2855	lock_page(page);
2856
2857	/* Case c or d, restart the operation */
2858	if (!page->mapping) {
2859		unlock_page(page);
2860		put_page(page);
2861		goto repeat;
2862	}
2863
2864	/* Someone else locked and filled the page in a very small window */
2865	if (PageUptodate(page)) {
2866		unlock_page(page);
2867		goto out;
2868	}
2869	goto filler;
2870
2871out:
2872	mark_page_accessed(page);
2873	return page;
2874}
2875
2876/**
2877 * read_cache_page - read into page cache, fill it if needed
2878 * @mapping:	the page's address_space
2879 * @index:	the page index
2880 * @filler:	function to perform the read
2881 * @data:	first arg to filler(data, page) function, often left as NULL
2882 *
2883 * Read into the page cache. If a page already exists, and PageUptodate() is
2884 * not set, try to fill the page and wait for it to become unlocked.
2885 *
2886 * If the page does not get brought uptodate, return -EIO.
2887 *
2888 * Return: up to date page on success, ERR_PTR() on failure.
2889 */
2890struct page *read_cache_page(struct address_space *mapping,
2891				pgoff_t index,
2892				int (*filler)(void *, struct page *),
2893				void *data)
2894{
2895	return do_read_cache_page(mapping, index, filler, data,
2896			mapping_gfp_mask(mapping));
2897}
2898EXPORT_SYMBOL(read_cache_page);
2899
2900/**
2901 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2902 * @mapping:	the page's address_space
2903 * @index:	the page index
2904 * @gfp:	the page allocator flags to use if allocating
2905 *
2906 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2907 * any new page allocations done using the specified allocation flags.
2908 *
2909 * If the page does not get brought uptodate, return -EIO.
2910 *
2911 * Return: up to date page on success, ERR_PTR() on failure.
2912 */
2913struct page *read_cache_page_gfp(struct address_space *mapping,
2914				pgoff_t index,
2915				gfp_t gfp)
2916{
2917	return do_read_cache_page(mapping, index, NULL, NULL, gfp);
 
 
2918}
2919EXPORT_SYMBOL(read_cache_page_gfp);
2920
2921/*
2922 * Don't operate on ranges the page cache doesn't support, and don't exceed the
2923 * LFS limits.  If pos is under the limit it becomes a short access.  If it
2924 * exceeds the limit we return -EFBIG.
2925 */
2926static int generic_write_check_limits(struct file *file, loff_t pos,
2927				      loff_t *count)
2928{
2929	struct inode *inode = file->f_mapping->host;
2930	loff_t max_size = inode->i_sb->s_maxbytes;
2931	loff_t limit = rlimit(RLIMIT_FSIZE);
2932
2933	if (limit != RLIM_INFINITY) {
2934		if (pos >= limit) {
2935			send_sig(SIGXFSZ, current, 0);
2936			return -EFBIG;
2937		}
2938		*count = min(*count, limit - pos);
2939	}
2940
2941	if (!(file->f_flags & O_LARGEFILE))
2942		max_size = MAX_NON_LFS;
2943
2944	if (unlikely(pos >= max_size))
2945		return -EFBIG;
2946
2947	*count = min(*count, max_size - pos);
2948
2949	return 0;
2950}
2951
2952/*
2953 * Performs necessary checks before doing a write
2954 *
2955 * Can adjust writing position or amount of bytes to write.
2956 * Returns appropriate error code that caller should return or
2957 * zero in case that write should be allowed.
2958 */
2959inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
2960{
2961	struct file *file = iocb->ki_filp;
2962	struct inode *inode = file->f_mapping->host;
2963	loff_t count;
2964	int ret;
2965
2966	if (IS_SWAPFILE(inode))
2967		return -ETXTBSY;
2968
2969	if (!iov_iter_count(from))
2970		return 0;
2971
2972	/* FIXME: this is for backwards compatibility with 2.4 */
2973	if (iocb->ki_flags & IOCB_APPEND)
2974		iocb->ki_pos = i_size_read(inode);
2975
 
 
2976	if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT))
2977		return -EINVAL;
2978
2979	count = iov_iter_count(from);
2980	ret = generic_write_check_limits(file, iocb->ki_pos, &count);
2981	if (ret)
2982		return ret;
2983
2984	iov_iter_truncate(from, count);
2985	return iov_iter_count(from);
2986}
2987EXPORT_SYMBOL(generic_write_checks);
2988
2989/*
2990 * Performs necessary checks before doing a clone.
2991 *
2992 * Can adjust amount of bytes to clone via @req_count argument.
2993 * Returns appropriate error code that caller should return or
2994 * zero in case the clone should be allowed.
2995 */
2996int generic_remap_checks(struct file *file_in, loff_t pos_in,
2997			 struct file *file_out, loff_t pos_out,
2998			 loff_t *req_count, unsigned int remap_flags)
2999{
3000	struct inode *inode_in = file_in->f_mapping->host;
3001	struct inode *inode_out = file_out->f_mapping->host;
3002	uint64_t count = *req_count;
3003	uint64_t bcount;
3004	loff_t size_in, size_out;
3005	loff_t bs = inode_out->i_sb->s_blocksize;
3006	int ret;
3007
3008	/* The start of both ranges must be aligned to an fs block. */
3009	if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs))
3010		return -EINVAL;
3011
3012	/* Ensure offsets don't wrap. */
3013	if (pos_in + count < pos_in || pos_out + count < pos_out)
3014		return -EINVAL;
3015
3016	size_in = i_size_read(inode_in);
3017	size_out = i_size_read(inode_out);
3018
3019	/* Dedupe requires both ranges to be within EOF. */
3020	if ((remap_flags & REMAP_FILE_DEDUP) &&
3021	    (pos_in >= size_in || pos_in + count > size_in ||
3022	     pos_out >= size_out || pos_out + count > size_out))
3023		return -EINVAL;
3024
3025	/* Ensure the infile range is within the infile. */
3026	if (pos_in >= size_in)
3027		return -EINVAL;
3028	count = min(count, size_in - (uint64_t)pos_in);
3029
3030	ret = generic_write_check_limits(file_out, pos_out, &count);
3031	if (ret)
3032		return ret;
3033
3034	/*
3035	 * If the user wanted us to link to the infile's EOF, round up to the
3036	 * next block boundary for this check.
3037	 *
3038	 * Otherwise, make sure the count is also block-aligned, having
3039	 * already confirmed the starting offsets' block alignment.
3040	 */
3041	if (pos_in + count == size_in) {
3042		bcount = ALIGN(size_in, bs) - pos_in;
3043	} else {
3044		if (!IS_ALIGNED(count, bs))
3045			count = ALIGN_DOWN(count, bs);
3046		bcount = count;
3047	}
3048
3049	/* Don't allow overlapped cloning within the same file. */
3050	if (inode_in == inode_out &&
3051	    pos_out + bcount > pos_in &&
3052	    pos_out < pos_in + bcount)
3053		return -EINVAL;
3054
3055	/*
3056	 * We shortened the request but the caller can't deal with that, so
3057	 * bounce the request back to userspace.
 
 
 
3058	 */
3059	if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN))
3060		return -EINVAL;
3061
3062	*req_count = count;
3063	return 0;
3064}
3065
3066
3067/*
3068 * Performs common checks before doing a file copy/clone
3069 * from @file_in to @file_out.
3070 */
3071int generic_file_rw_checks(struct file *file_in, struct file *file_out)
3072{
3073	struct inode *inode_in = file_inode(file_in);
3074	struct inode *inode_out = file_inode(file_out);
3075
3076	/* Don't copy dirs, pipes, sockets... */
3077	if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
3078		return -EISDIR;
3079	if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
3080		return -EINVAL;
3081
3082	if (!(file_in->f_mode & FMODE_READ) ||
3083	    !(file_out->f_mode & FMODE_WRITE) ||
3084	    (file_out->f_flags & O_APPEND))
3085		return -EBADF;
3086
3087	return 0;
3088}
3089
3090/*
3091 * Performs necessary checks before doing a file copy
3092 *
3093 * Can adjust amount of bytes to copy via @req_count argument.
3094 * Returns appropriate error code that caller should return or
3095 * zero in case the copy should be allowed.
3096 */
3097int generic_copy_file_checks(struct file *file_in, loff_t pos_in,
3098			     struct file *file_out, loff_t pos_out,
3099			     size_t *req_count, unsigned int flags)
3100{
3101	struct inode *inode_in = file_inode(file_in);
3102	struct inode *inode_out = file_inode(file_out);
3103	uint64_t count = *req_count;
3104	loff_t size_in;
3105	int ret;
3106
3107	ret = generic_file_rw_checks(file_in, file_out);
3108	if (ret)
3109		return ret;
3110
3111	/* Don't touch certain kinds of inodes */
3112	if (IS_IMMUTABLE(inode_out))
3113		return -EPERM;
3114
3115	if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
3116		return -ETXTBSY;
3117
3118	/* Ensure offsets don't wrap. */
3119	if (pos_in + count < pos_in || pos_out + count < pos_out)
3120		return -EOVERFLOW;
3121
3122	/* Shorten the copy to EOF */
3123	size_in = i_size_read(inode_in);
3124	if (pos_in >= size_in)
3125		count = 0;
3126	else
3127		count = min(count, size_in - (uint64_t)pos_in);
3128
3129	ret = generic_write_check_limits(file_out, pos_out, &count);
3130	if (ret)
3131		return ret;
3132
3133	/* Don't allow overlapped copying within the same file. */
3134	if (inode_in == inode_out &&
3135	    pos_out + count > pos_in &&
3136	    pos_out < pos_in + count)
3137		return -EINVAL;
3138
3139	*req_count = count;
3140	return 0;
3141}
 
3142
3143int pagecache_write_begin(struct file *file, struct address_space *mapping,
3144				loff_t pos, unsigned len, unsigned flags,
3145				struct page **pagep, void **fsdata)
3146{
3147	const struct address_space_operations *aops = mapping->a_ops;
3148
3149	return aops->write_begin(file, mapping, pos, len, flags,
3150							pagep, fsdata);
3151}
3152EXPORT_SYMBOL(pagecache_write_begin);
3153
3154int pagecache_write_end(struct file *file, struct address_space *mapping,
3155				loff_t pos, unsigned len, unsigned copied,
3156				struct page *page, void *fsdata)
3157{
3158	const struct address_space_operations *aops = mapping->a_ops;
3159
3160	return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
3161}
3162EXPORT_SYMBOL(pagecache_write_end);
3163
3164ssize_t
3165generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3166{
3167	struct file	*file = iocb->ki_filp;
3168	struct address_space *mapping = file->f_mapping;
3169	struct inode	*inode = mapping->host;
3170	loff_t		pos = iocb->ki_pos;
3171	ssize_t		written;
3172	size_t		write_len;
3173	pgoff_t		end;
3174
3175	write_len = iov_iter_count(from);
3176	end = (pos + write_len - 1) >> PAGE_SHIFT;
3177
3178	if (iocb->ki_flags & IOCB_NOWAIT) {
3179		/* If there are pages to writeback, return */
3180		if (filemap_range_has_page(inode->i_mapping, pos,
3181					   pos + write_len - 1))
3182			return -EAGAIN;
3183	} else {
3184		written = filemap_write_and_wait_range(mapping, pos,
3185							pos + write_len - 1);
3186		if (written)
3187			goto out;
3188	}
3189
3190	/*
3191	 * After a write we want buffered reads to be sure to go to disk to get
3192	 * the new data.  We invalidate clean cached page from the region we're
3193	 * about to write.  We do this *before* the write so that we can return
3194	 * without clobbering -EIOCBQUEUED from ->direct_IO().
3195	 */
3196	written = invalidate_inode_pages2_range(mapping,
3197					pos >> PAGE_SHIFT, end);
3198	/*
3199	 * If a page can not be invalidated, return 0 to fall back
3200	 * to buffered write.
3201	 */
3202	if (written) {
3203		if (written == -EBUSY)
3204			return 0;
3205		goto out;
3206	}
3207
3208	written = mapping->a_ops->direct_IO(iocb, from);
3209
3210	/*
3211	 * Finally, try again to invalidate clean pages which might have been
3212	 * cached by non-direct readahead, or faulted in by get_user_pages()
3213	 * if the source of the write was an mmap'ed region of the file
3214	 * we're writing.  Either one is a pretty crazy thing to do,
3215	 * so we don't support it 100%.  If this invalidation
3216	 * fails, tough, the write still worked...
3217	 *
3218	 * Most of the time we do not need this since dio_complete() will do
3219	 * the invalidation for us. However there are some file systems that
3220	 * do not end up with dio_complete() being called, so let's not break
3221	 * them by removing it completely
3222	 */
3223	if (mapping->nrpages)
3224		invalidate_inode_pages2_range(mapping,
3225					pos >> PAGE_SHIFT, end);
3226
3227	if (written > 0) {
3228		pos += written;
3229		write_len -= written;
3230		if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3231			i_size_write(inode, pos);
3232			mark_inode_dirty(inode);
3233		}
3234		iocb->ki_pos = pos;
3235	}
3236	iov_iter_revert(from, write_len - iov_iter_count(from));
3237out:
3238	return written;
3239}
3240EXPORT_SYMBOL(generic_file_direct_write);
3241
3242/*
3243 * Find or create a page at the given pagecache position. Return the locked
3244 * page. This function is specifically for buffered writes.
3245 */
3246struct page *grab_cache_page_write_begin(struct address_space *mapping,
3247					pgoff_t index, unsigned flags)
3248{
3249	struct page *page;
3250	int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
3251
3252	if (flags & AOP_FLAG_NOFS)
3253		fgp_flags |= FGP_NOFS;
3254
3255	page = pagecache_get_page(mapping, index, fgp_flags,
3256			mapping_gfp_mask(mapping));
3257	if (page)
3258		wait_for_stable_page(page);
3259
3260	return page;
3261}
3262EXPORT_SYMBOL(grab_cache_page_write_begin);
3263
3264ssize_t generic_perform_write(struct file *file,
3265				struct iov_iter *i, loff_t pos)
3266{
3267	struct address_space *mapping = file->f_mapping;
3268	const struct address_space_operations *a_ops = mapping->a_ops;
3269	long status = 0;
3270	ssize_t written = 0;
3271	unsigned int flags = 0;
3272
3273	do {
3274		struct page *page;
3275		unsigned long offset;	/* Offset into pagecache page */
3276		unsigned long bytes;	/* Bytes to write to page */
3277		size_t copied;		/* Bytes copied from user */
3278		void *fsdata;
3279
3280		offset = (pos & (PAGE_SIZE - 1));
3281		bytes = min_t(unsigned long, PAGE_SIZE - offset,
3282						iov_iter_count(i));
3283
3284again:
3285		/*
3286		 * Bring in the user page that we will copy from _first_.
3287		 * Otherwise there's a nasty deadlock on copying from the
3288		 * same page as we're writing to, without it being marked
3289		 * up-to-date.
3290		 *
3291		 * Not only is this an optimisation, but it is also required
3292		 * to check that the address is actually valid, when atomic
3293		 * usercopies are used, below.
3294		 */
3295		if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
3296			status = -EFAULT;
3297			break;
3298		}
3299
3300		if (fatal_signal_pending(current)) {
3301			status = -EINTR;
3302			break;
3303		}
3304
3305		status = a_ops->write_begin(file, mapping, pos, bytes, flags,
3306						&page, &fsdata);
3307		if (unlikely(status < 0))
3308			break;
3309
3310		if (mapping_writably_mapped(mapping))
3311			flush_dcache_page(page);
3312
3313		copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
3314		flush_dcache_page(page);
3315
3316		status = a_ops->write_end(file, mapping, pos, bytes, copied,
3317						page, fsdata);
3318		if (unlikely(status < 0))
3319			break;
3320		copied = status;
3321
3322		cond_resched();
3323
3324		iov_iter_advance(i, copied);
3325		if (unlikely(copied == 0)) {
3326			/*
3327			 * If we were unable to copy any data at all, we must
3328			 * fall back to a single segment length write.
3329			 *
3330			 * If we didn't fallback here, we could livelock
3331			 * because not all segments in the iov can be copied at
3332			 * once without a pagefault.
3333			 */
3334			bytes = min_t(unsigned long, PAGE_SIZE - offset,
3335						iov_iter_single_seg_count(i));
3336			goto again;
3337		}
3338		pos += copied;
3339		written += copied;
3340
3341		balance_dirty_pages_ratelimited(mapping);
3342	} while (iov_iter_count(i));
3343
3344	return written ? written : status;
3345}
3346EXPORT_SYMBOL(generic_perform_write);
3347
3348/**
3349 * __generic_file_write_iter - write data to a file
3350 * @iocb:	IO state structure (file, offset, etc.)
3351 * @from:	iov_iter with data to write
3352 *
3353 * This function does all the work needed for actually writing data to a
3354 * file. It does all basic checks, removes SUID from the file, updates
3355 * modification times and calls proper subroutines depending on whether we
3356 * do direct IO or a standard buffered write.
3357 *
3358 * It expects i_mutex to be grabbed unless we work on a block device or similar
3359 * object which does not need locking at all.
3360 *
3361 * This function does *not* take care of syncing data in case of O_SYNC write.
3362 * A caller has to handle it. This is mainly due to the fact that we want to
3363 * avoid syncing under i_mutex.
3364 *
3365 * Return:
3366 * * number of bytes written, even for truncated writes
3367 * * negative error code if no data has been written at all
3368 */
3369ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3370{
3371	struct file *file = iocb->ki_filp;
3372	struct address_space * mapping = file->f_mapping;
3373	struct inode 	*inode = mapping->host;
3374	ssize_t		written = 0;
3375	ssize_t		err;
3376	ssize_t		status;
3377
3378	/* We can write back this queue in page reclaim */
3379	current->backing_dev_info = inode_to_bdi(inode);
3380	err = file_remove_privs(file);
3381	if (err)
3382		goto out;
3383
3384	err = file_update_time(file);
3385	if (err)
3386		goto out;
3387
3388	if (iocb->ki_flags & IOCB_DIRECT) {
3389		loff_t pos, endbyte;
3390
3391		written = generic_file_direct_write(iocb, from);
3392		/*
3393		 * If the write stopped short of completing, fall back to
3394		 * buffered writes.  Some filesystems do this for writes to
3395		 * holes, for example.  For DAX files, a buffered write will
3396		 * not succeed (even if it did, DAX does not handle dirty
3397		 * page-cache pages correctly).
3398		 */
3399		if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3400			goto out;
3401
3402		status = generic_perform_write(file, from, pos = iocb->ki_pos);
3403		/*
3404		 * If generic_perform_write() returned a synchronous error
3405		 * then we want to return the number of bytes which were
3406		 * direct-written, or the error code if that was zero.  Note
3407		 * that this differs from normal direct-io semantics, which
3408		 * will return -EFOO even if some bytes were written.
3409		 */
3410		if (unlikely(status < 0)) {
3411			err = status;
3412			goto out;
3413		}
3414		/*
3415		 * We need to ensure that the page cache pages are written to
3416		 * disk and invalidated to preserve the expected O_DIRECT
3417		 * semantics.
3418		 */
3419		endbyte = pos + status - 1;
3420		err = filemap_write_and_wait_range(mapping, pos, endbyte);
3421		if (err == 0) {
3422			iocb->ki_pos = endbyte + 1;
3423			written += status;
3424			invalidate_mapping_pages(mapping,
3425						 pos >> PAGE_SHIFT,
3426						 endbyte >> PAGE_SHIFT);
3427		} else {
3428			/*
3429			 * We don't know how much we wrote, so just return
3430			 * the number of bytes which were direct-written
3431			 */
3432		}
3433	} else {
3434		written = generic_perform_write(file, from, iocb->ki_pos);
3435		if (likely(written > 0))
3436			iocb->ki_pos += written;
3437	}
3438out:
3439	current->backing_dev_info = NULL;
3440	return written ? written : err;
3441}
3442EXPORT_SYMBOL(__generic_file_write_iter);
3443
3444/**
3445 * generic_file_write_iter - write data to a file
3446 * @iocb:	IO state structure
3447 * @from:	iov_iter with data to write
3448 *
3449 * This is a wrapper around __generic_file_write_iter() to be used by most
3450 * filesystems. It takes care of syncing the file in case of O_SYNC file
3451 * and acquires i_mutex as needed.
3452 * Return:
3453 * * negative error code if no data has been written at all of
3454 *   vfs_fsync_range() failed for a synchronous write
3455 * * number of bytes written, even for truncated writes
3456 */
3457ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3458{
3459	struct file *file = iocb->ki_filp;
3460	struct inode *inode = file->f_mapping->host;
3461	ssize_t ret;
3462
3463	inode_lock(inode);
3464	ret = generic_write_checks(iocb, from);
3465	if (ret > 0)
3466		ret = __generic_file_write_iter(iocb, from);
3467	inode_unlock(inode);
3468
3469	if (ret > 0)
3470		ret = generic_write_sync(iocb, ret);
3471	return ret;
3472}
3473EXPORT_SYMBOL(generic_file_write_iter);
3474
3475/**
3476 * try_to_release_page() - release old fs-specific metadata on a page
3477 *
3478 * @page: the page which the kernel is trying to free
3479 * @gfp_mask: memory allocation flags (and I/O mode)
3480 *
3481 * The address_space is to try to release any data against the page
3482 * (presumably at page->private).
 
3483 *
3484 * This may also be called if PG_fscache is set on a page, indicating that the
3485 * page is known to the local caching routines.
3486 *
3487 * The @gfp_mask argument specifies whether I/O may be performed to release
3488 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
3489 *
3490 * Return: %1 if the release was successful, otherwise return zero.
3491 */
3492int try_to_release_page(struct page *page, gfp_t gfp_mask)
3493{
3494	struct address_space * const mapping = page->mapping;
3495
3496	BUG_ON(!PageLocked(page));
3497	if (PageWriteback(page))
3498		return 0;
3499
3500	if (mapping && mapping->a_ops->releasepage)
3501		return mapping->a_ops->releasepage(page, gfp_mask);
3502	return try_to_free_buffers(page);
3503}
3504
3505EXPORT_SYMBOL(try_to_release_page);