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);
v3.15
   1/*
   2 *	linux/mm/filemap.c
   3 *
   4 * Copyright (C) 1994-1999  Linus Torvalds
   5 */
   6
   7/*
   8 * This file handles the generic file mmap semantics used by
   9 * most "normal" filesystems (but you don't /have/ to use this:
  10 * the NFS filesystem used to do this differently, for example)
  11 */
  12#include <linux/export.h>
  13#include <linux/compiler.h>
 
  14#include <linux/fs.h>
 
  15#include <linux/uaccess.h>
  16#include <linux/aio.h>
  17#include <linux/capability.h>
  18#include <linux/kernel_stat.h>
  19#include <linux/gfp.h>
  20#include <linux/mm.h>
  21#include <linux/swap.h>
  22#include <linux/mman.h>
  23#include <linux/pagemap.h>
  24#include <linux/file.h>
  25#include <linux/uio.h>
  26#include <linux/hash.h>
  27#include <linux/writeback.h>
  28#include <linux/backing-dev.h>
  29#include <linux/pagevec.h>
  30#include <linux/blkdev.h>
  31#include <linux/security.h>
  32#include <linux/cpuset.h>
  33#include <linux/hardirq.h> /* for BUG_ON(!in_atomic()) only */
  34#include <linux/memcontrol.h>
  35#include <linux/cleancache.h>
 
  36#include <linux/rmap.h>
  37#include "internal.h"
  38
  39#define CREATE_TRACE_POINTS
  40#include <trace/events/filemap.h>
  41
  42/*
  43 * FIXME: remove all knowledge of the buffer layer from the core VM
  44 */
  45#include <linux/buffer_head.h> /* for try_to_free_buffers */
  46
  47#include <asm/mman.h>
  48
  49/*
  50 * Shared mappings implemented 30.11.1994. It's not fully working yet,
  51 * though.
  52 *
  53 * Shared mappings now work. 15.8.1995  Bruno.
  54 *
  55 * finished 'unifying' the page and buffer cache and SMP-threaded the
  56 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
  57 *
  58 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
  59 */
  60
  61/*
  62 * Lock ordering:
  63 *
  64 *  ->i_mmap_mutex		(truncate_pagecache)
  65 *    ->private_lock		(__free_pte->__set_page_dirty_buffers)
  66 *      ->swap_lock		(exclusive_swap_page, others)
  67 *        ->mapping->tree_lock
  68 *
  69 *  ->i_mutex
  70 *    ->i_mmap_mutex		(truncate->unmap_mapping_range)
  71 *
  72 *  ->mmap_sem
  73 *    ->i_mmap_mutex
  74 *      ->page_table_lock or pte_lock	(various, mainly in memory.c)
  75 *        ->mapping->tree_lock	(arch-dependent flush_dcache_mmap_lock)
  76 *
  77 *  ->mmap_sem
  78 *    ->lock_page		(access_process_vm)
  79 *
  80 *  ->i_mutex			(generic_perform_write)
  81 *    ->mmap_sem		(fault_in_pages_readable->do_page_fault)
  82 *
  83 *  bdi->wb.list_lock
  84 *    sb_lock			(fs/fs-writeback.c)
  85 *    ->mapping->tree_lock	(__sync_single_inode)
  86 *
  87 *  ->i_mmap_mutex
  88 *    ->anon_vma.lock		(vma_adjust)
  89 *
  90 *  ->anon_vma.lock
  91 *    ->page_table_lock or pte_lock	(anon_vma_prepare and various)
  92 *
  93 *  ->page_table_lock or pte_lock
  94 *    ->swap_lock		(try_to_unmap_one)
  95 *    ->private_lock		(try_to_unmap_one)
  96 *    ->tree_lock		(try_to_unmap_one)
  97 *    ->zone.lru_lock		(follow_page->mark_page_accessed)
  98 *    ->zone.lru_lock		(check_pte_range->isolate_lru_page)
  99 *    ->private_lock		(page_remove_rmap->set_page_dirty)
 100 *    ->tree_lock		(page_remove_rmap->set_page_dirty)
 101 *    bdi.wb->list_lock		(page_remove_rmap->set_page_dirty)
 102 *    ->inode->i_lock		(page_remove_rmap->set_page_dirty)
 
 103 *    bdi.wb->list_lock		(zap_pte_range->set_page_dirty)
 104 *    ->inode->i_lock		(zap_pte_range->set_page_dirty)
 105 *    ->private_lock		(zap_pte_range->__set_page_dirty_buffers)
 106 *
 107 * ->i_mmap_mutex
 108 *   ->tasklist_lock            (memory_failure, collect_procs_ao)
 109 */
 110
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 111static void page_cache_tree_delete(struct address_space *mapping,
 112				   struct page *page, void *shadow)
 113{
 114	struct radix_tree_node *node;
 115	unsigned long index;
 116	unsigned int offset;
 117	unsigned int tag;
 118	void **slot;
 
 
 
 
 
 
 
 
 
 
 
 
 119
 120	VM_BUG_ON(!PageLocked(page));
 
 
 
 121
 122	__radix_tree_lookup(&mapping->page_tree, page->index, &node, &slot);
 
 123
 124	if (shadow) {
 125		mapping->nrshadows++;
 126		/*
 127		 * Make sure the nrshadows update is committed before
 128		 * the nrpages update so that final truncate racing
 129		 * with reclaim does not see both counters 0 at the
 130		 * same time and miss a shadow entry.
 131		 */
 132		smp_wmb();
 133	}
 134	mapping->nrpages--;
 
 135
 136	if (!node) {
 137		/* Clear direct pointer tags in root node */
 138		mapping->page_tree.gfp_mask &= __GFP_BITS_MASK;
 139		radix_tree_replace_slot(slot, shadow);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 140		return;
 141	}
 142
 143	/* Clear tree tags for the removed page */
 144	index = page->index;
 145	offset = index & RADIX_TREE_MAP_MASK;
 146	for (tag = 0; tag < RADIX_TREE_MAX_TAGS; tag++) {
 147		if (test_bit(offset, node->tags[tag]))
 148			radix_tree_tag_clear(&mapping->page_tree, index, tag);
 
 
 
 149	}
 150
 151	/* Delete page, swap shadow entry */
 152	radix_tree_replace_slot(slot, shadow);
 153	workingset_node_pages_dec(node);
 154	if (shadow)
 155		workingset_node_shadows_inc(node);
 156	else
 157		if (__radix_tree_delete_node(&mapping->page_tree, node))
 158			return;
 159
 160	/*
 161	 * Track node that only contains shadow entries.
 
 
 162	 *
 163	 * Avoid acquiring the list_lru lock if already tracked.  The
 164	 * list_empty() test is safe as node->private_list is
 165	 * protected by mapping->tree_lock.
 166	 */
 167	if (!workingset_node_pages(node) &&
 168	    list_empty(&node->private_list)) {
 169		node->private_data = mapping;
 170		list_lru_add(&workingset_shadow_nodes, &node->private_list);
 171	}
 172}
 173
 174/*
 175 * Delete a page from the page cache and free it. Caller has to make
 176 * sure the page is locked and that nobody else uses it - or that usage
 177 * is safe.  The caller must hold the mapping's tree_lock.
 178 */
 179void __delete_from_page_cache(struct page *page, void *shadow)
 180{
 181	struct address_space *mapping = page->mapping;
 182
 183	trace_mm_filemap_delete_from_page_cache(page);
 184	/*
 185	 * if we're uptodate, flush out into the cleancache, otherwise
 186	 * invalidate any existing cleancache entries.  We can't leave
 187	 * stale data around in the cleancache once our page is gone
 188	 */
 189	if (PageUptodate(page) && PageMappedToDisk(page))
 190		cleancache_put_page(page);
 191	else
 192		cleancache_invalidate_page(mapping, page);
 193
 
 194	page_cache_tree_delete(mapping, page, shadow);
 
 195
 196	page->mapping = NULL;
 197	/* Leave page->index set: truncation lookup relies upon it */
 
 
 198
 199	__dec_zone_page_state(page, NR_FILE_PAGES);
 200	if (PageSwapBacked(page))
 201		__dec_zone_page_state(page, NR_SHMEM);
 202	BUG_ON(page_mapped(page));
 203
 204	/*
 205	 * Some filesystems seem to re-dirty the page even after
 206	 * the VM has canceled the dirty bit (eg ext3 journaling).
 207	 *
 208	 * Fix it up by doing a final dirty accounting check after
 209	 * having removed the page entirely.
 210	 */
 211	if (PageDirty(page) && mapping_cap_account_dirty(mapping)) {
 212		dec_zone_page_state(page, NR_FILE_DIRTY);
 213		dec_bdi_stat(mapping->backing_dev_info, BDI_RECLAIMABLE);
 214	}
 215}
 216
 217/**
 218 * delete_from_page_cache - delete page from page cache
 219 * @page: the page which the kernel is trying to remove from page cache
 220 *
 221 * This must be called only on pages that have been verified to be in the page
 222 * cache and locked.  It will never put the page into the free list, the caller
 223 * has a reference on the page.
 224 */
 225void delete_from_page_cache(struct page *page)
 226{
 227	struct address_space *mapping = page->mapping;
 228	void (*freepage)(struct page *);
 229
 230	BUG_ON(!PageLocked(page));
 231
 232	freepage = mapping->a_ops->freepage;
 233	spin_lock_irq(&mapping->tree_lock);
 234	__delete_from_page_cache(page, NULL);
 235	spin_unlock_irq(&mapping->tree_lock);
 236	mem_cgroup_uncharge_cache_page(page);
 237
 238	if (freepage)
 239		freepage(page);
 240	page_cache_release(page);
 241}
 242EXPORT_SYMBOL(delete_from_page_cache);
 243
 244static int sleep_on_page(void *word)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 245{
 246	io_schedule();
 247	return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 248}
 249
 250static int sleep_on_page_killable(void *word)
 
 251{
 252	sleep_on_page(word);
 253	return fatal_signal_pending(current) ? -EINTR : 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 254}
 255
 256static int filemap_check_errors(struct address_space *mapping)
 257{
 258	int ret = 0;
 259	/* Check for outstanding write errors */
 260	if (test_bit(AS_ENOSPC, &mapping->flags) &&
 261	    test_and_clear_bit(AS_ENOSPC, &mapping->flags))
 262		ret = -ENOSPC;
 263	if (test_bit(AS_EIO, &mapping->flags) &&
 264	    test_and_clear_bit(AS_EIO, &mapping->flags))
 265		ret = -EIO;
 266	return ret;
 267}
 
 
 
 
 
 
 
 
 
 
 
 268
 269/**
 270 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
 271 * @mapping:	address space structure to write
 272 * @start:	offset in bytes where the range starts
 273 * @end:	offset in bytes where the range ends (inclusive)
 274 * @sync_mode:	enable synchronous operation
 275 *
 276 * Start writeback against all of a mapping's dirty pages that lie
 277 * within the byte offsets <start, end> inclusive.
 278 *
 279 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
 280 * opposed to a regular memory cleansing writeback.  The difference between
 281 * these two operations is that if a dirty page/buffer is encountered, it must
 282 * be waited upon, and not just skipped over.
 283 */
 284int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
 285				loff_t end, int sync_mode)
 286{
 287	int ret;
 288	struct writeback_control wbc = {
 289		.sync_mode = sync_mode,
 290		.nr_to_write = LONG_MAX,
 291		.range_start = start,
 292		.range_end = end,
 293	};
 294
 295	if (!mapping_cap_writeback_dirty(mapping))
 296		return 0;
 297
 
 298	ret = do_writepages(mapping, &wbc);
 
 299	return ret;
 300}
 301
 302static inline int __filemap_fdatawrite(struct address_space *mapping,
 303	int sync_mode)
 304{
 305	return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
 306}
 307
 308int filemap_fdatawrite(struct address_space *mapping)
 309{
 310	return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
 311}
 312EXPORT_SYMBOL(filemap_fdatawrite);
 313
 314int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
 315				loff_t end)
 316{
 317	return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
 318}
 319EXPORT_SYMBOL(filemap_fdatawrite_range);
 320
 321/**
 322 * filemap_flush - mostly a non-blocking flush
 323 * @mapping:	target address_space
 324 *
 325 * This is a mostly non-blocking flush.  Not suitable for data-integrity
 326 * purposes - I/O may not be started against all dirty pages.
 327 */
 328int filemap_flush(struct address_space *mapping)
 329{
 330	return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
 331}
 332EXPORT_SYMBOL(filemap_flush);
 333
 334/**
 335 * filemap_fdatawait_range - wait for writeback to complete
 336 * @mapping:		address space structure to wait for
 337 * @start_byte:		offset in bytes where the range starts
 338 * @end_byte:		offset in bytes where the range ends (inclusive)
 339 *
 340 * Walk the list of under-writeback pages of the given address space
 341 * in the given range and wait for all of them.
 342 */
 343int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
 344			    loff_t end_byte)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 345{
 346	pgoff_t index = start_byte >> PAGE_CACHE_SHIFT;
 347	pgoff_t end = end_byte >> PAGE_CACHE_SHIFT;
 348	struct pagevec pvec;
 349	int nr_pages;
 350	int ret2, ret = 0;
 351
 352	if (end_byte < start_byte)
 353		goto out;
 354
 355	pagevec_init(&pvec, 0);
 356	while ((index <= end) &&
 357			(nr_pages = pagevec_lookup_tag(&pvec, mapping, &index,
 358			PAGECACHE_TAG_WRITEBACK,
 359			min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1)) != 0) {
 360		unsigned i;
 361
 
 
 
 
 
 362		for (i = 0; i < nr_pages; i++) {
 363			struct page *page = pvec.pages[i];
 364
 365			/* until radix tree lookup accepts end_index */
 366			if (page->index > end)
 367				continue;
 368
 369			wait_on_page_writeback(page);
 370			if (TestClearPageError(page))
 371				ret = -EIO;
 372		}
 373		pagevec_release(&pvec);
 374		cond_resched();
 375	}
 376out:
 377	ret2 = filemap_check_errors(mapping);
 378	if (!ret)
 379		ret = ret2;
 380
 381	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 382}
 383EXPORT_SYMBOL(filemap_fdatawait_range);
 384
 385/**
 386 * filemap_fdatawait - wait for all under-writeback pages to complete
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 387 * @mapping: address space structure to wait for
 388 *
 389 * Walk the list of under-writeback pages of the given address space
 390 * and wait for all of them.
 
 
 
 
 
 391 */
 392int filemap_fdatawait(struct address_space *mapping)
 393{
 394	loff_t i_size = i_size_read(mapping->host);
 
 
 
 395
 396	if (i_size == 0)
 397		return 0;
 398
 399	return filemap_fdatawait_range(mapping, 0, i_size - 1);
 400}
 401EXPORT_SYMBOL(filemap_fdatawait);
 402
 403int filemap_write_and_wait(struct address_space *mapping)
 404{
 405	int err = 0;
 406
 407	if (mapping->nrpages) {
 408		err = filemap_fdatawrite(mapping);
 409		/*
 410		 * Even if the above returned error, the pages may be
 411		 * written partially (e.g. -ENOSPC), so we wait for it.
 412		 * But the -EIO is special case, it may indicate the worst
 413		 * thing (e.g. bug) happened, so we avoid waiting for it.
 414		 */
 415		if (err != -EIO) {
 416			int err2 = filemap_fdatawait(mapping);
 417			if (!err)
 418				err = err2;
 
 
 
 419		}
 420	} else {
 421		err = filemap_check_errors(mapping);
 422	}
 423	return err;
 424}
 425EXPORT_SYMBOL(filemap_write_and_wait);
 426
 427/**
 428 * filemap_write_and_wait_range - write out & wait on a file range
 429 * @mapping:	the address_space for the pages
 430 * @lstart:	offset in bytes where the range starts
 431 * @lend:	offset in bytes where the range ends (inclusive)
 432 *
 433 * Write out and wait upon file offsets lstart->lend, inclusive.
 434 *
 435 * Note that `lend' is inclusive (describes the last byte to be written) so
 436 * that this function can be used to write to the very end-of-file (end = -1).
 437 */
 438int filemap_write_and_wait_range(struct address_space *mapping,
 439				 loff_t lstart, loff_t lend)
 440{
 441	int err = 0;
 442
 443	if (mapping->nrpages) {
 444		err = __filemap_fdatawrite_range(mapping, lstart, lend,
 445						 WB_SYNC_ALL);
 446		/* See comment of filemap_write_and_wait() */
 447		if (err != -EIO) {
 448			int err2 = filemap_fdatawait_range(mapping,
 449						lstart, lend);
 450			if (!err)
 451				err = err2;
 
 
 
 452		}
 453	} else {
 454		err = filemap_check_errors(mapping);
 455	}
 456	return err;
 457}
 458EXPORT_SYMBOL(filemap_write_and_wait_range);
 459
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 460/**
 461 * replace_page_cache_page - replace a pagecache page with a new one
 462 * @old:	page to be replaced
 463 * @new:	page to replace with
 464 * @gfp_mask:	allocation mode
 465 *
 466 * This function replaces a page in the pagecache with a new one.  On
 467 * success it acquires the pagecache reference for the new page and
 468 * drops it for the old page.  Both the old and new pages must be
 469 * locked.  This function does not add the new page to the LRU, the
 470 * caller must do that.
 471 *
 472 * The remove + add is atomic.  The only way this function can fail is
 473 * memory allocation failure.
 474 */
 475int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
 476{
 477	int error;
 478
 479	VM_BUG_ON_PAGE(!PageLocked(old), old);
 480	VM_BUG_ON_PAGE(!PageLocked(new), new);
 481	VM_BUG_ON_PAGE(new->mapping, new);
 482
 483	error = radix_tree_preload(gfp_mask & ~__GFP_HIGHMEM);
 484	if (!error) {
 485		struct address_space *mapping = old->mapping;
 486		void (*freepage)(struct page *);
 
 487
 488		pgoff_t offset = old->index;
 489		freepage = mapping->a_ops->freepage;
 490
 491		page_cache_get(new);
 492		new->mapping = mapping;
 493		new->index = offset;
 494
 495		spin_lock_irq(&mapping->tree_lock);
 496		__delete_from_page_cache(old, NULL);
 497		error = radix_tree_insert(&mapping->page_tree, offset, new);
 498		BUG_ON(error);
 499		mapping->nrpages++;
 500		__inc_zone_page_state(new, NR_FILE_PAGES);
 
 
 
 
 501		if (PageSwapBacked(new))
 502			__inc_zone_page_state(new, NR_SHMEM);
 503		spin_unlock_irq(&mapping->tree_lock);
 504		/* mem_cgroup codes must not be called under tree_lock */
 505		mem_cgroup_replace_page_cache(old, new);
 506		radix_tree_preload_end();
 507		if (freepage)
 508			freepage(old);
 509		page_cache_release(old);
 510	}
 511
 512	return error;
 513}
 514EXPORT_SYMBOL_GPL(replace_page_cache_page);
 515
 516static int page_cache_tree_insert(struct address_space *mapping,
 517				  struct page *page, void **shadowp)
 518{
 519	struct radix_tree_node *node;
 520	void **slot;
 521	int error;
 522
 523	error = __radix_tree_create(&mapping->page_tree, page->index,
 524				    &node, &slot);
 525	if (error)
 526		return error;
 527	if (*slot) {
 528		void *p;
 529
 530		p = radix_tree_deref_slot_protected(slot, &mapping->tree_lock);
 531		if (!radix_tree_exceptional_entry(p))
 532			return -EEXIST;
 533		if (shadowp)
 534			*shadowp = p;
 535		mapping->nrshadows--;
 536		if (node)
 537			workingset_node_shadows_dec(node);
 538	}
 539	radix_tree_replace_slot(slot, page);
 540	mapping->nrpages++;
 541	if (node) {
 542		workingset_node_pages_inc(node);
 543		/*
 544		 * Don't track node that contains actual pages.
 545		 *
 546		 * Avoid acquiring the list_lru lock if already
 547		 * untracked.  The list_empty() test is safe as
 548		 * node->private_list is protected by
 549		 * mapping->tree_lock.
 550		 */
 551		if (!list_empty(&node->private_list))
 552			list_lru_del(&workingset_shadow_nodes,
 553				     &node->private_list);
 554	}
 555	return 0;
 556}
 557
 558static int __add_to_page_cache_locked(struct page *page,
 559				      struct address_space *mapping,
 560				      pgoff_t offset, gfp_t gfp_mask,
 561				      void **shadowp)
 562{
 
 
 563	int error;
 564
 565	VM_BUG_ON_PAGE(!PageLocked(page), page);
 566	VM_BUG_ON_PAGE(PageSwapBacked(page), page);
 567
 568	error = mem_cgroup_charge_file(page, current->mm,
 569					gfp_mask & GFP_RECLAIM_MASK);
 570	if (error)
 571		return error;
 
 
 572
 573	error = radix_tree_maybe_preload(gfp_mask & ~__GFP_HIGHMEM);
 574	if (error) {
 575		mem_cgroup_uncharge_cache_page(page);
 
 576		return error;
 577	}
 578
 579	page_cache_get(page);
 580	page->mapping = mapping;
 581	page->index = offset;
 582
 583	spin_lock_irq(&mapping->tree_lock);
 584	error = page_cache_tree_insert(mapping, page, shadowp);
 585	radix_tree_preload_end();
 586	if (unlikely(error))
 587		goto err_insert;
 588	__inc_zone_page_state(page, NR_FILE_PAGES);
 589	spin_unlock_irq(&mapping->tree_lock);
 
 
 
 
 
 590	trace_mm_filemap_add_to_page_cache(page);
 591	return 0;
 592err_insert:
 593	page->mapping = NULL;
 594	/* Leave page->index set: truncation relies upon it */
 595	spin_unlock_irq(&mapping->tree_lock);
 596	mem_cgroup_uncharge_cache_page(page);
 597	page_cache_release(page);
 
 598	return error;
 599}
 600
 601/**
 602 * add_to_page_cache_locked - add a locked page to the pagecache
 603 * @page:	page to add
 604 * @mapping:	the page's address_space
 605 * @offset:	page index
 606 * @gfp_mask:	page allocation mode
 607 *
 608 * This function is used to add a page to the pagecache. It must be locked.
 609 * This function does not add the page to the LRU.  The caller must do that.
 610 */
 611int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
 612		pgoff_t offset, gfp_t gfp_mask)
 613{
 614	return __add_to_page_cache_locked(page, mapping, offset,
 615					  gfp_mask, NULL);
 616}
 617EXPORT_SYMBOL(add_to_page_cache_locked);
 618
 619int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
 620				pgoff_t offset, gfp_t gfp_mask)
 621{
 622	void *shadow = NULL;
 623	int ret;
 624
 625	__set_page_locked(page);
 626	ret = __add_to_page_cache_locked(page, mapping, offset,
 627					 gfp_mask, &shadow);
 628	if (unlikely(ret))
 629		__clear_page_locked(page);
 630	else {
 631		/*
 632		 * The page might have been evicted from cache only
 633		 * recently, in which case it should be activated like
 634		 * any other repeatedly accessed page.
 
 
 
 635		 */
 636		if (shadow && workingset_refault(shadow)) {
 
 637			SetPageActive(page);
 638			workingset_activation(page);
 639		} else
 640			ClearPageActive(page);
 641		lru_cache_add(page);
 642	}
 643	return ret;
 644}
 645EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
 646
 647#ifdef CONFIG_NUMA
 648struct page *__page_cache_alloc(gfp_t gfp)
 649{
 650	int n;
 651	struct page *page;
 652
 653	if (cpuset_do_page_mem_spread()) {
 654		unsigned int cpuset_mems_cookie;
 655		do {
 656			cpuset_mems_cookie = read_mems_allowed_begin();
 657			n = cpuset_mem_spread_node();
 658			page = alloc_pages_exact_node(n, gfp, 0);
 659		} while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
 660
 661		return page;
 662	}
 663	return alloc_pages(gfp, 0);
 664}
 665EXPORT_SYMBOL(__page_cache_alloc);
 666#endif
 667
 668/*
 669 * In order to wait for pages to become available there must be
 670 * waitqueues associated with pages. By using a hash table of
 671 * waitqueues where the bucket discipline is to maintain all
 672 * waiters on the same queue and wake all when any of the pages
 673 * become available, and for the woken contexts to check to be
 674 * sure the appropriate page became available, this saves space
 675 * at a cost of "thundering herd" phenomena during rare hash
 676 * collisions.
 677 */
 
 
 
 
 678static wait_queue_head_t *page_waitqueue(struct page *page)
 679{
 680	const struct zone *zone = page_zone(page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 681
 682	return &zone->wait_table[hash_ptr(page, zone->wait_table_bits)];
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 683}
 684
 685static inline void wake_up_page(struct page *page, int bit)
 686{
 687	__wake_up_bit(page_waitqueue(page), &page->flags, bit);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 688}
 689
 690void wait_on_page_bit(struct page *page, int bit_nr)
 691{
 692	DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
 693
 694	if (test_bit(bit_nr, &page->flags))
 695		__wait_on_bit(page_waitqueue(page), &wait, sleep_on_page,
 696							TASK_UNINTERRUPTIBLE);
 697}
 698EXPORT_SYMBOL(wait_on_page_bit);
 699
 700int wait_on_page_bit_killable(struct page *page, int bit_nr)
 701{
 702	DEFINE_WAIT_BIT(wait, &page->flags, bit_nr);
 703
 704	if (!test_bit(bit_nr, &page->flags))
 705		return 0;
 706
 707	return __wait_on_bit(page_waitqueue(page), &wait,
 708			     sleep_on_page_killable, TASK_KILLABLE);
 709}
 
 710
 711/**
 712 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
 713 * @page: Page defining the wait queue of interest
 714 * @waiter: Waiter to add to the queue
 715 *
 716 * Add an arbitrary @waiter to the wait queue for the nominated @page.
 717 */
 718void add_page_wait_queue(struct page *page, wait_queue_t *waiter)
 719{
 720	wait_queue_head_t *q = page_waitqueue(page);
 721	unsigned long flags;
 722
 723	spin_lock_irqsave(&q->lock, flags);
 724	__add_wait_queue(q, waiter);
 
 725	spin_unlock_irqrestore(&q->lock, flags);
 726}
 727EXPORT_SYMBOL_GPL(add_page_wait_queue);
 728
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 729/**
 730 * unlock_page - unlock a locked page
 731 * @page: the page
 732 *
 733 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
 734 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
 735 * mechananism between PageLocked pages and PageWriteback pages is shared.
 736 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
 737 *
 738 * The mb is necessary to enforce ordering between the clear_bit and the read
 739 * of the waitqueue (to avoid SMP races with a parallel wait_on_page_locked()).
 
 
 
 740 */
 741void unlock_page(struct page *page)
 742{
 
 
 743	VM_BUG_ON_PAGE(!PageLocked(page), page);
 744	clear_bit_unlock(PG_locked, &page->flags);
 745	smp_mb__after_clear_bit();
 746	wake_up_page(page, PG_locked);
 747}
 748EXPORT_SYMBOL(unlock_page);
 749
 750/**
 751 * end_page_writeback - end writeback against a page
 752 * @page: the page
 753 */
 754void end_page_writeback(struct page *page)
 755{
 756	if (TestClearPageReclaim(page))
 
 
 
 
 
 
 
 
 757		rotate_reclaimable_page(page);
 
 758
 759	if (!test_clear_page_writeback(page))
 760		BUG();
 761
 762	smp_mb__after_clear_bit();
 763	wake_up_page(page, PG_writeback);
 764}
 765EXPORT_SYMBOL(end_page_writeback);
 766
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 767/**
 768 * __lock_page - get a lock on the page, assuming we need to sleep to get it
 769 * @page: the page to lock
 770 */
 771void __lock_page(struct page *page)
 772{
 773	DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
 774
 775	__wait_on_bit_lock(page_waitqueue(page), &wait, sleep_on_page,
 776							TASK_UNINTERRUPTIBLE);
 777}
 778EXPORT_SYMBOL(__lock_page);
 779
 780int __lock_page_killable(struct page *page)
 781{
 782	DEFINE_WAIT_BIT(wait, &page->flags, PG_locked);
 783
 784	return __wait_on_bit_lock(page_waitqueue(page), &wait,
 785					sleep_on_page_killable, TASK_KILLABLE);
 786}
 787EXPORT_SYMBOL_GPL(__lock_page_killable);
 788
 
 
 
 
 
 
 
 
 
 
 
 789int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
 790			 unsigned int flags)
 791{
 792	if (flags & FAULT_FLAG_ALLOW_RETRY) {
 793		/*
 794		 * CAUTION! In this case, mmap_sem is not released
 795		 * even though return 0.
 796		 */
 797		if (flags & FAULT_FLAG_RETRY_NOWAIT)
 798			return 0;
 799
 800		up_read(&mm->mmap_sem);
 801		if (flags & FAULT_FLAG_KILLABLE)
 802			wait_on_page_locked_killable(page);
 803		else
 804			wait_on_page_locked(page);
 805		return 0;
 806	} else {
 807		if (flags & FAULT_FLAG_KILLABLE) {
 808			int ret;
 809
 810			ret = __lock_page_killable(page);
 811			if (ret) {
 812				up_read(&mm->mmap_sem);
 813				return 0;
 814			}
 815		} else
 816			__lock_page(page);
 817		return 1;
 818	}
 819}
 820
 821/**
 822 * page_cache_next_hole - find the next hole (not-present entry)
 823 * @mapping: mapping
 824 * @index: index
 825 * @max_scan: maximum range to search
 826 *
 827 * Search the set [index, min(index+max_scan-1, MAX_INDEX)] for the
 828 * lowest indexed hole.
 829 *
 830 * Returns: the index of the hole if found, otherwise returns an index
 831 * outside of the set specified (in which case 'return - index >=
 832 * max_scan' will be true). In rare cases of index wrap-around, 0 will
 833 * be returned.
 834 *
 835 * page_cache_next_hole may be called under rcu_read_lock. However,
 836 * like radix_tree_gang_lookup, this will not atomically search a
 837 * snapshot of the tree at a single point in time. For example, if a
 838 * hole is created at index 5, then subsequently a hole is created at
 839 * index 10, page_cache_next_hole covering both indexes may return 10
 840 * if called under rcu_read_lock.
 841 */
 842pgoff_t page_cache_next_hole(struct address_space *mapping,
 843			     pgoff_t index, unsigned long max_scan)
 844{
 845	unsigned long i;
 846
 847	for (i = 0; i < max_scan; i++) {
 848		struct page *page;
 849
 850		page = radix_tree_lookup(&mapping->page_tree, index);
 851		if (!page || radix_tree_exceptional_entry(page))
 852			break;
 853		index++;
 854		if (index == 0)
 855			break;
 856	}
 857
 858	return index;
 859}
 860EXPORT_SYMBOL(page_cache_next_hole);
 861
 862/**
 863 * page_cache_prev_hole - find the prev hole (not-present entry)
 864 * @mapping: mapping
 865 * @index: index
 866 * @max_scan: maximum range to search
 867 *
 868 * Search backwards in the range [max(index-max_scan+1, 0), index] for
 869 * the first hole.
 870 *
 871 * Returns: the index of the hole if found, otherwise returns an index
 872 * outside of the set specified (in which case 'index - return >=
 873 * max_scan' will be true). In rare cases of wrap-around, ULONG_MAX
 874 * will be returned.
 875 *
 876 * page_cache_prev_hole may be called under rcu_read_lock. However,
 877 * like radix_tree_gang_lookup, this will not atomically search a
 878 * snapshot of the tree at a single point in time. For example, if a
 879 * hole is created at index 10, then subsequently a hole is created at
 880 * index 5, page_cache_prev_hole covering both indexes may return 5 if
 881 * called under rcu_read_lock.
 882 */
 883pgoff_t page_cache_prev_hole(struct address_space *mapping,
 884			     pgoff_t index, unsigned long max_scan)
 885{
 886	unsigned long i;
 887
 888	for (i = 0; i < max_scan; i++) {
 889		struct page *page;
 890
 891		page = radix_tree_lookup(&mapping->page_tree, index);
 892		if (!page || radix_tree_exceptional_entry(page))
 893			break;
 894		index--;
 895		if (index == ULONG_MAX)
 896			break;
 897	}
 898
 899	return index;
 900}
 901EXPORT_SYMBOL(page_cache_prev_hole);
 902
 903/**
 904 * find_get_entry - find and get a page cache entry
 905 * @mapping: the address_space to search
 906 * @offset: the page cache index
 907 *
 908 * Looks up the page cache slot at @mapping & @offset.  If there is a
 909 * page cache page, it is returned with an increased refcount.
 910 *
 911 * If the slot holds a shadow entry of a previously evicted page, or a
 912 * swap entry from shmem/tmpfs, it is returned.
 913 *
 914 * Otherwise, %NULL is returned.
 915 */
 916struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
 917{
 918	void **pagep;
 919	struct page *page;
 920
 921	rcu_read_lock();
 922repeat:
 923	page = NULL;
 924	pagep = radix_tree_lookup_slot(&mapping->page_tree, offset);
 925	if (pagep) {
 926		page = radix_tree_deref_slot(pagep);
 927		if (unlikely(!page))
 928			goto out;
 929		if (radix_tree_exception(page)) {
 930			if (radix_tree_deref_retry(page))
 931				goto repeat;
 932			/*
 933			 * A shadow entry of a recently evicted page,
 934			 * or a swap entry from shmem/tmpfs.  Return
 935			 * it without attempting to raise page count.
 936			 */
 937			goto out;
 938		}
 939		if (!page_cache_get_speculative(page))
 
 
 
 
 
 
 
 940			goto repeat;
 
 941
 942		/*
 943		 * Has the page moved?
 944		 * This is part of the lockless pagecache protocol. See
 945		 * include/linux/pagemap.h for details.
 946		 */
 947		if (unlikely(page != *pagep)) {
 948			page_cache_release(page);
 949			goto repeat;
 950		}
 951	}
 952out:
 953	rcu_read_unlock();
 954
 955	return page;
 956}
 957EXPORT_SYMBOL(find_get_entry);
 958
 959/**
 960 * find_get_page - find and get a page reference
 961 * @mapping: the address_space to search
 962 * @offset: the page index
 963 *
 964 * Looks up the page cache slot at @mapping & @offset.  If there is a
 965 * page cache page, it is returned with an increased refcount.
 966 *
 967 * Otherwise, %NULL is returned.
 968 */
 969struct page *find_get_page(struct address_space *mapping, pgoff_t offset)
 970{
 971	struct page *page = find_get_entry(mapping, offset);
 972
 973	if (radix_tree_exceptional_entry(page))
 974		page = NULL;
 975	return page;
 976}
 977EXPORT_SYMBOL(find_get_page);
 978
 979/**
 980 * find_lock_entry - locate, pin and lock a page cache entry
 981 * @mapping: the address_space to search
 982 * @offset: the page cache index
 983 *
 984 * Looks up the page cache slot at @mapping & @offset.  If there is a
 985 * page cache page, it is returned locked and with an increased
 986 * refcount.
 987 *
 988 * If the slot holds a shadow entry of a previously evicted page, or a
 989 * swap entry from shmem/tmpfs, it is returned.
 990 *
 991 * Otherwise, %NULL is returned.
 992 *
 993 * find_lock_entry() may sleep.
 994 */
 995struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
 996{
 997	struct page *page;
 998
 999repeat:
1000	page = find_get_entry(mapping, offset);
1001	if (page && !radix_tree_exception(page)) {
1002		lock_page(page);
1003		/* Has the page been truncated? */
1004		if (unlikely(page->mapping != mapping)) {
1005			unlock_page(page);
1006			page_cache_release(page);
1007			goto repeat;
1008		}
1009		VM_BUG_ON_PAGE(page->index != offset, page);
1010	}
1011	return page;
1012}
1013EXPORT_SYMBOL(find_lock_entry);
1014
1015/**
1016 * find_lock_page - locate, pin and lock a pagecache page
1017 * @mapping: the address_space to search
1018 * @offset: the page index
 
 
1019 *
1020 * Looks up the page cache slot at @mapping & @offset.  If there is a
1021 * page cache page, it is returned locked and with an increased
1022 * refcount.
1023 *
1024 * Otherwise, %NULL is returned.
1025 *
1026 * find_lock_page() may sleep.
1027 */
1028struct page *find_lock_page(struct address_space *mapping, pgoff_t offset)
1029{
1030	struct page *page = find_lock_entry(mapping, offset);
1031
1032	if (radix_tree_exceptional_entry(page))
1033		page = NULL;
1034	return page;
1035}
1036EXPORT_SYMBOL(find_lock_page);
1037
1038/**
1039 * find_or_create_page - locate or add a pagecache page
1040 * @mapping: the page's address_space
1041 * @index: the page's index into the mapping
1042 * @gfp_mask: page allocation mode
1043 *
1044 * Looks up the page cache slot at @mapping & @offset.  If there is a
1045 * page cache page, it is returned locked and with an increased
1046 * refcount.
1047 *
1048 * If the page is not present, a new page is allocated using @gfp_mask
1049 * and added to the page cache and the VM's LRU list.  The page is
1050 * returned locked and with an increased refcount.
 
 
 
1051 *
1052 * On memory exhaustion, %NULL is returned.
 
1053 *
1054 * find_or_create_page() may sleep, even if @gfp_flags specifies an
1055 * atomic allocation!
1056 */
1057struct page *find_or_create_page(struct address_space *mapping,
1058		pgoff_t index, gfp_t gfp_mask)
1059{
1060	struct page *page;
1061	int err;
1062repeat:
1063	page = find_lock_page(mapping, index);
1064	if (!page) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1065		page = __page_cache_alloc(gfp_mask);
1066		if (!page)
1067			return NULL;
1068		/*
1069		 * We want a regular kernel memory (not highmem or DMA etc)
1070		 * allocation for the radix tree nodes, but we need to honour
1071		 * the context-specific requirements the caller has asked for.
1072		 * GFP_RECLAIM_MASK collects those requirements.
1073		 */
1074		err = add_to_page_cache_lru(page, mapping, index,
1075			(gfp_mask & GFP_RECLAIM_MASK));
 
1076		if (unlikely(err)) {
1077			page_cache_release(page);
1078			page = NULL;
1079			if (err == -EEXIST)
1080				goto repeat;
1081		}
1082	}
 
1083	return page;
1084}
1085EXPORT_SYMBOL(find_or_create_page);
1086
1087/**
1088 * find_get_entries - gang pagecache lookup
1089 * @mapping:	The address_space to search
1090 * @start:	The starting page cache index
1091 * @nr_entries:	The maximum number of entries
1092 * @entries:	Where the resulting entries are placed
1093 * @indices:	The cache indices corresponding to the entries in @entries
1094 *
1095 * find_get_entries() will search for and return a group of up to
1096 * @nr_entries entries in the mapping.  The entries are placed at
1097 * @entries.  find_get_entries() takes a reference against any actual
1098 * pages it returns.
1099 *
1100 * The search returns a group of mapping-contiguous page cache entries
1101 * with ascending indexes.  There may be holes in the indices due to
1102 * not-present pages.
1103 *
1104 * Any shadow entries of evicted pages, or swap entries from
1105 * shmem/tmpfs, are included in the returned array.
1106 *
1107 * find_get_entries() returns the number of pages and shadow entries
1108 * which were found.
1109 */
1110unsigned find_get_entries(struct address_space *mapping,
1111			  pgoff_t start, unsigned int nr_entries,
1112			  struct page **entries, pgoff_t *indices)
1113{
1114	void **slot;
1115	unsigned int ret = 0;
1116	struct radix_tree_iter iter;
1117
1118	if (!nr_entries)
1119		return 0;
1120
1121	rcu_read_lock();
1122restart:
1123	radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1124		struct page *page;
1125repeat:
1126		page = radix_tree_deref_slot(slot);
1127		if (unlikely(!page))
1128			continue;
1129		if (radix_tree_exception(page)) {
1130			if (radix_tree_deref_retry(page))
1131				goto restart;
 
 
1132			/*
1133			 * A shadow entry of a recently evicted page,
1134			 * or a swap entry from shmem/tmpfs.  Return
1135			 * it without attempting to raise page count.
1136			 */
1137			goto export;
1138		}
1139		if (!page_cache_get_speculative(page))
 
 
1140			goto repeat;
1141
 
 
 
 
 
 
1142		/* Has the page moved? */
1143		if (unlikely(page != *slot)) {
1144			page_cache_release(page);
1145			goto repeat;
1146		}
1147export:
1148		indices[ret] = iter.index;
1149		entries[ret] = page;
1150		if (++ret == nr_entries)
1151			break;
1152	}
1153	rcu_read_unlock();
1154	return ret;
1155}
1156
1157/**
1158 * find_get_pages - gang pagecache lookup
1159 * @mapping:	The address_space to search
1160 * @start:	The starting page index
 
1161 * @nr_pages:	The maximum number of pages
1162 * @pages:	Where the resulting pages are placed
1163 *
1164 * find_get_pages() will search for and return a group of up to
1165 * @nr_pages pages in the mapping.  The pages are placed at @pages.
1166 * find_get_pages() takes a reference against the returned pages.
 
1167 *
1168 * The search returns a group of mapping-contiguous pages with ascending
1169 * indexes.  There may be holes in the indices due to not-present pages.
 
1170 *
1171 * find_get_pages() returns the number of pages which were found.
1172 */
1173unsigned find_get_pages(struct address_space *mapping, pgoff_t start,
1174			    unsigned int nr_pages, struct page **pages)
 
 
 
1175{
1176	struct radix_tree_iter iter;
1177	void **slot;
1178	unsigned ret = 0;
1179
1180	if (unlikely(!nr_pages))
1181		return 0;
1182
1183	rcu_read_lock();
1184restart:
1185	radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
1186		struct page *page;
 
 
1187repeat:
1188		page = radix_tree_deref_slot(slot);
1189		if (unlikely(!page))
1190			continue;
1191
1192		if (radix_tree_exception(page)) {
1193			if (radix_tree_deref_retry(page)) {
1194				/*
1195				 * Transient condition which can only trigger
1196				 * when entry at index 0 moves out of or back
1197				 * to root: none yet gotten, safe to restart.
1198				 */
1199				WARN_ON(iter.index);
1200				goto restart;
1201			}
1202			/*
1203			 * A shadow entry of a recently evicted page,
1204			 * or a swap entry from shmem/tmpfs.  Skip
1205			 * over it.
1206			 */
1207			continue;
1208		}
1209
1210		if (!page_cache_get_speculative(page))
 
1211			goto repeat;
1212
 
 
 
 
 
 
1213		/* Has the page moved? */
1214		if (unlikely(page != *slot)) {
1215			page_cache_release(page);
1216			goto repeat;
1217		}
1218
1219		pages[ret] = page;
1220		if (++ret == nr_pages)
1221			break;
 
 
1222	}
1223
 
 
 
 
 
 
 
 
 
 
 
1224	rcu_read_unlock();
 
1225	return ret;
1226}
1227
1228/**
1229 * find_get_pages_contig - gang contiguous pagecache lookup
1230 * @mapping:	The address_space to search
1231 * @index:	The starting page index
1232 * @nr_pages:	The maximum number of pages
1233 * @pages:	Where the resulting pages are placed
1234 *
1235 * find_get_pages_contig() works exactly like find_get_pages(), except
1236 * that the returned number of pages are guaranteed to be contiguous.
1237 *
1238 * find_get_pages_contig() returns the number of pages which were found.
1239 */
1240unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
1241			       unsigned int nr_pages, struct page **pages)
1242{
1243	struct radix_tree_iter iter;
1244	void **slot;
1245	unsigned int ret = 0;
1246
1247	if (unlikely(!nr_pages))
1248		return 0;
1249
1250	rcu_read_lock();
1251restart:
1252	radix_tree_for_each_contig(slot, &mapping->page_tree, &iter, index) {
1253		struct page *page;
1254repeat:
1255		page = radix_tree_deref_slot(slot);
1256		/* The hole, there no reason to continue */
1257		if (unlikely(!page))
1258			break;
1259
1260		if (radix_tree_exception(page)) {
1261			if (radix_tree_deref_retry(page)) {
1262				/*
1263				 * Transient condition which can only trigger
1264				 * when entry at index 0 moves out of or back
1265				 * to root: none yet gotten, safe to restart.
1266				 */
1267				goto restart;
1268			}
1269			/*
1270			 * A shadow entry of a recently evicted page,
1271			 * or a swap entry from shmem/tmpfs.  Stop
1272			 * looking for contiguous pages.
1273			 */
1274			break;
1275		}
1276
1277		if (!page_cache_get_speculative(page))
 
1278			goto repeat;
1279
 
 
 
 
 
 
1280		/* Has the page moved? */
1281		if (unlikely(page != *slot)) {
1282			page_cache_release(page);
1283			goto repeat;
1284		}
1285
1286		/*
1287		 * must check mapping and index after taking the ref.
1288		 * otherwise we can get both false positives and false
1289		 * negatives, which is just confusing to the caller.
1290		 */
1291		if (page->mapping == NULL || page->index != iter.index) {
1292			page_cache_release(page);
1293			break;
1294		}
1295
1296		pages[ret] = page;
1297		if (++ret == nr_pages)
1298			break;
1299	}
1300	rcu_read_unlock();
1301	return ret;
1302}
1303EXPORT_SYMBOL(find_get_pages_contig);
1304
1305/**
1306 * find_get_pages_tag - find and return pages that match @tag
1307 * @mapping:	the address_space to search
1308 * @index:	the starting page index
 
1309 * @tag:	the tag index
1310 * @nr_pages:	the maximum number of pages
1311 * @pages:	where the resulting pages are placed
1312 *
1313 * Like find_get_pages, except we only return pages which are tagged with
1314 * @tag.   We update @index to index the next page for the traversal.
1315 */
1316unsigned find_get_pages_tag(struct address_space *mapping, pgoff_t *index,
1317			int tag, unsigned int nr_pages, struct page **pages)
 
1318{
1319	struct radix_tree_iter iter;
1320	void **slot;
1321	unsigned ret = 0;
1322
1323	if (unlikely(!nr_pages))
1324		return 0;
1325
1326	rcu_read_lock();
1327restart:
1328	radix_tree_for_each_tagged(slot, &mapping->page_tree,
1329				   &iter, *index, tag) {
1330		struct page *page;
 
1331repeat:
1332		page = radix_tree_deref_slot(slot);
1333		if (unlikely(!page))
1334			continue;
1335
1336		if (radix_tree_exception(page)) {
1337			if (radix_tree_deref_retry(page)) {
1338				/*
1339				 * Transient condition which can only trigger
1340				 * when entry at index 0 moves out of or back
1341				 * to root: none yet gotten, safe to restart.
1342				 */
1343				goto restart;
1344			}
1345			/*
1346			 * A shadow entry of a recently evicted page.
1347			 *
1348			 * Those entries should never be tagged, but
1349			 * this tree walk is lockless and the tags are
1350			 * looked up in bulk, one radix tree node at a
1351			 * time, so there is a sizable window for page
1352			 * reclaim to evict a page we saw tagged.
1353			 *
1354			 * Skip over it.
1355			 */
1356			continue;
1357		}
1358
1359		if (!page_cache_get_speculative(page))
 
1360			goto repeat;
1361
 
 
 
 
 
 
1362		/* Has the page moved? */
1363		if (unlikely(page != *slot)) {
1364			page_cache_release(page);
1365			goto repeat;
1366		}
1367
1368		pages[ret] = page;
1369		if (++ret == nr_pages)
1370			break;
 
 
1371	}
1372
 
 
 
 
 
 
 
 
 
 
 
1373	rcu_read_unlock();
1374
1375	if (ret)
1376		*index = pages[ret - 1]->index + 1;
1377
1378	return ret;
1379}
1380EXPORT_SYMBOL(find_get_pages_tag);
1381
1382/**
1383 * grab_cache_page_nowait - returns locked page at given index in given cache
1384 * @mapping: target address_space
1385 * @index: the page index
1386 *
1387 * Same as grab_cache_page(), but do not wait if the page is unavailable.
1388 * This is intended for speculative data generators, where the data can
1389 * be regenerated if the page couldn't be grabbed.  This routine should
1390 * be safe to call while holding the lock for another page.
1391 *
1392 * Clear __GFP_FS when allocating the page to avoid recursion into the fs
1393 * and deadlock against the caller's locked page.
1394 */
1395struct page *
1396grab_cache_page_nowait(struct address_space *mapping, pgoff_t index)
1397{
1398	struct page *page = find_get_page(mapping, index);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1399
1400	if (page) {
1401		if (trylock_page(page))
1402			return page;
1403		page_cache_release(page);
1404		return NULL;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1405	}
1406	page = __page_cache_alloc(mapping_gfp_mask(mapping) & ~__GFP_FS);
1407	if (page && add_to_page_cache_lru(page, mapping, index, GFP_NOFS)) {
1408		page_cache_release(page);
1409		page = NULL;
1410	}
1411	return page;
1412}
1413EXPORT_SYMBOL(grab_cache_page_nowait);
1414
1415/*
1416 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
1417 * a _large_ part of the i/o request. Imagine the worst scenario:
1418 *
1419 *      ---R__________________________________________B__________
1420 *         ^ reading here                             ^ bad block(assume 4k)
1421 *
1422 * read(R) => miss => readahead(R...B) => media error => frustrating retries
1423 * => failing the whole request => read(R) => read(R+1) =>
1424 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
1425 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
1426 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
1427 *
1428 * It is going insane. Fix it by quickly scaling down the readahead size.
1429 */
1430static void shrink_readahead_size_eio(struct file *filp,
1431					struct file_ra_state *ra)
1432{
1433	ra->ra_pages /= 4;
1434}
1435
1436/**
1437 * do_generic_file_read - generic file read routine
1438 * @filp:	the file to read
1439 * @ppos:	current file position
1440 * @iter:	data destination
1441 * @written:	already copied
1442 *
1443 * This is a generic file read routine, and uses the
1444 * mapping->a_ops->readpage() function for the actual low-level stuff.
1445 *
1446 * This is really ugly. But the goto's actually try to clarify some
1447 * of the logic when it comes to error handling etc.
1448 */
1449static ssize_t do_generic_file_read(struct file *filp, loff_t *ppos,
1450		struct iov_iter *iter, ssize_t written)
1451{
 
1452	struct address_space *mapping = filp->f_mapping;
1453	struct inode *inode = mapping->host;
1454	struct file_ra_state *ra = &filp->f_ra;
 
1455	pgoff_t index;
1456	pgoff_t last_index;
1457	pgoff_t prev_index;
1458	unsigned long offset;      /* offset into pagecache page */
1459	unsigned int prev_offset;
1460	int error = 0;
1461
1462	index = *ppos >> PAGE_CACHE_SHIFT;
1463	prev_index = ra->prev_pos >> PAGE_CACHE_SHIFT;
1464	prev_offset = ra->prev_pos & (PAGE_CACHE_SIZE-1);
1465	last_index = (*ppos + iter->count + PAGE_CACHE_SIZE-1) >> PAGE_CACHE_SHIFT;
1466	offset = *ppos & ~PAGE_CACHE_MASK;
 
 
 
 
1467
1468	for (;;) {
1469		struct page *page;
1470		pgoff_t end_index;
1471		loff_t isize;
1472		unsigned long nr, ret;
1473
1474		cond_resched();
1475find_page:
 
 
 
 
 
1476		page = find_get_page(mapping, index);
1477		if (!page) {
 
 
1478			page_cache_sync_readahead(mapping,
1479					ra, filp,
1480					index, last_index - index);
1481			page = find_get_page(mapping, index);
1482			if (unlikely(page == NULL))
1483				goto no_cached_page;
1484		}
1485		if (PageReadahead(page)) {
1486			page_cache_async_readahead(mapping,
1487					ra, filp, page,
1488					index, last_index - index);
1489		}
1490		if (!PageUptodate(page)) {
1491			if (inode->i_blkbits == PAGE_CACHE_SHIFT ||
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1492					!mapping->a_ops->is_partially_uptodate)
1493				goto page_not_up_to_date;
 
 
 
1494			if (!trylock_page(page))
1495				goto page_not_up_to_date;
1496			/* Did it get truncated before we got the lock? */
1497			if (!page->mapping)
1498				goto page_not_up_to_date_locked;
1499			if (!mapping->a_ops->is_partially_uptodate(page,
1500							offset, iter->count))
1501				goto page_not_up_to_date_locked;
1502			unlock_page(page);
1503		}
1504page_ok:
1505		/*
1506		 * i_size must be checked after we know the page is Uptodate.
1507		 *
1508		 * Checking i_size after the check allows us to calculate
1509		 * the correct value for "nr", which means the zero-filled
1510		 * part of the page is not copied back to userspace (unless
1511		 * another truncate extends the file - this is desired though).
1512		 */
1513
1514		isize = i_size_read(inode);
1515		end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1516		if (unlikely(!isize || index > end_index)) {
1517			page_cache_release(page);
1518			goto out;
1519		}
1520
1521		/* nr is the maximum number of bytes to copy from this page */
1522		nr = PAGE_CACHE_SIZE;
1523		if (index == end_index) {
1524			nr = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1525			if (nr <= offset) {
1526				page_cache_release(page);
1527				goto out;
1528			}
1529		}
1530		nr = nr - offset;
1531
1532		/* If users can be writing to this page using arbitrary
1533		 * virtual addresses, take care about potential aliasing
1534		 * before reading the page on the kernel side.
1535		 */
1536		if (mapping_writably_mapped(mapping))
1537			flush_dcache_page(page);
1538
1539		/*
1540		 * When a sequential read accesses a page several times,
1541		 * only mark it as accessed the first time.
1542		 */
1543		if (prev_index != index || offset != prev_offset)
1544			mark_page_accessed(page);
1545		prev_index = index;
1546
1547		/*
1548		 * Ok, we have the page, and it's up-to-date, so
1549		 * now we can copy it to user space...
1550		 */
1551
1552		ret = copy_page_to_iter(page, offset, nr, iter);
1553		offset += ret;
1554		index += offset >> PAGE_CACHE_SHIFT;
1555		offset &= ~PAGE_CACHE_MASK;
1556		prev_offset = offset;
1557
1558		page_cache_release(page);
1559		written += ret;
1560		if (!iov_iter_count(iter))
1561			goto out;
1562		if (ret < nr) {
1563			error = -EFAULT;
1564			goto out;
1565		}
1566		continue;
1567
1568page_not_up_to_date:
1569		/* Get exclusive access to the page ... */
1570		error = lock_page_killable(page);
1571		if (unlikely(error))
1572			goto readpage_error;
1573
1574page_not_up_to_date_locked:
1575		/* Did it get truncated before we got the lock? */
1576		if (!page->mapping) {
1577			unlock_page(page);
1578			page_cache_release(page);
1579			continue;
1580		}
1581
1582		/* Did somebody else fill it already? */
1583		if (PageUptodate(page)) {
1584			unlock_page(page);
1585			goto page_ok;
1586		}
1587
1588readpage:
1589		/*
1590		 * A previous I/O error may have been due to temporary
1591		 * failures, eg. multipath errors.
1592		 * PG_error will be set again if readpage fails.
1593		 */
1594		ClearPageError(page);
1595		/* Start the actual read. The read will unlock the page. */
1596		error = mapping->a_ops->readpage(filp, page);
1597
1598		if (unlikely(error)) {
1599			if (error == AOP_TRUNCATED_PAGE) {
1600				page_cache_release(page);
1601				error = 0;
1602				goto find_page;
1603			}
1604			goto readpage_error;
1605		}
1606
1607		if (!PageUptodate(page)) {
1608			error = lock_page_killable(page);
1609			if (unlikely(error))
1610				goto readpage_error;
1611			if (!PageUptodate(page)) {
1612				if (page->mapping == NULL) {
1613					/*
1614					 * invalidate_mapping_pages got it
1615					 */
1616					unlock_page(page);
1617					page_cache_release(page);
1618					goto find_page;
1619				}
1620				unlock_page(page);
1621				shrink_readahead_size_eio(filp, ra);
1622				error = -EIO;
1623				goto readpage_error;
1624			}
1625			unlock_page(page);
1626		}
1627
1628		goto page_ok;
1629
1630readpage_error:
1631		/* UHHUH! A synchronous read error occurred. Report it */
1632		page_cache_release(page);
1633		goto out;
1634
1635no_cached_page:
1636		/*
1637		 * Ok, it wasn't cached, so we need to create a new
1638		 * page..
1639		 */
1640		page = page_cache_alloc_cold(mapping);
1641		if (!page) {
1642			error = -ENOMEM;
1643			goto out;
1644		}
1645		error = add_to_page_cache_lru(page, mapping,
1646						index, GFP_KERNEL);
1647		if (error) {
1648			page_cache_release(page);
1649			if (error == -EEXIST) {
1650				error = 0;
1651				goto find_page;
1652			}
1653			goto out;
1654		}
1655		goto readpage;
1656	}
1657
 
 
1658out:
1659	ra->prev_pos = prev_index;
1660	ra->prev_pos <<= PAGE_CACHE_SHIFT;
1661	ra->prev_pos |= prev_offset;
1662
1663	*ppos = ((loff_t)index << PAGE_CACHE_SHIFT) + offset;
1664	file_accessed(filp);
1665	return written ? written : error;
1666}
1667
1668/*
1669 * Performs necessary checks before doing a write
1670 * @iov:	io vector request
1671 * @nr_segs:	number of segments in the iovec
1672 * @count:	number of bytes to write
1673 * @access_flags: type of access: %VERIFY_READ or %VERIFY_WRITE
1674 *
1675 * Adjust number of segments and amount of bytes to write (nr_segs should be
1676 * properly initialized first). Returns appropriate error code that caller
1677 * should return or zero in case that write should be allowed.
1678 */
1679int generic_segment_checks(const struct iovec *iov,
1680			unsigned long *nr_segs, size_t *count, int access_flags)
1681{
1682	unsigned long   seg;
1683	size_t cnt = 0;
1684	for (seg = 0; seg < *nr_segs; seg++) {
1685		const struct iovec *iv = &iov[seg];
1686
1687		/*
1688		 * If any segment has a negative length, or the cumulative
1689		 * length ever wraps negative then return -EINVAL.
1690		 */
1691		cnt += iv->iov_len;
1692		if (unlikely((ssize_t)(cnt|iv->iov_len) < 0))
1693			return -EINVAL;
1694		if (access_ok(access_flags, iv->iov_base, iv->iov_len))
1695			continue;
1696		if (seg == 0)
1697			return -EFAULT;
1698		*nr_segs = seg;
1699		cnt -= iv->iov_len;	/* This segment is no good */
1700		break;
1701	}
1702	*count = cnt;
1703	return 0;
1704}
1705EXPORT_SYMBOL(generic_segment_checks);
1706
1707/**
1708 * generic_file_aio_read - generic filesystem read routine
1709 * @iocb:	kernel I/O control block
1710 * @iov:	io vector request
1711 * @nr_segs:	number of segments in the iovec
1712 * @pos:	current file position
1713 *
1714 * This is the "read()" routine for all filesystems
1715 * that can use the page cache directly.
1716 */
1717ssize_t
1718generic_file_aio_read(struct kiocb *iocb, const struct iovec *iov,
1719		unsigned long nr_segs, loff_t pos)
1720{
1721	struct file *filp = iocb->ki_filp;
1722	ssize_t retval;
1723	size_t count;
1724	loff_t *ppos = &iocb->ki_pos;
1725	struct iov_iter i;
1726
1727	count = 0;
1728	retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1729	if (retval)
1730		return retval;
1731	iov_iter_init(&i, iov, nr_segs, count, 0);
1732
1733	/* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
1734	if (filp->f_flags & O_DIRECT) {
 
 
1735		loff_t size;
1736		struct address_space *mapping;
1737		struct inode *inode;
1738
1739		mapping = filp->f_mapping;
1740		inode = mapping->host;
1741		if (!count)
1742			goto out; /* skip atime */
1743		size = i_size_read(inode);
1744		retval = filemap_write_and_wait_range(mapping, pos,
1745					pos + iov_length(iov, nr_segs) - 1);
1746		if (!retval) {
1747			retval = mapping->a_ops->direct_IO(READ, iocb,
1748							   iov, pos, nr_segs);
 
 
 
 
 
1749		}
1750		if (retval > 0) {
1751			*ppos = pos + retval;
 
 
 
 
1752			count -= retval;
1753			/*
1754			 * If we did a short DIO read we need to skip the
1755			 * section of the iov that we've already read data into.
1756			 */
1757			iov_iter_advance(&i, retval);
1758		}
 
1759
1760		/*
1761		 * Btrfs can have a short DIO read if we encounter
1762		 * compressed extents, so if there was an error, or if
1763		 * we've already read everything we wanted to, or if
1764		 * there was a short read because we hit EOF, go ahead
1765		 * and return.  Otherwise fallthrough to buffered io for
1766		 * the rest of the read.
 
1767		 */
1768		if (retval < 0 || !count || *ppos >= size) {
1769			file_accessed(filp);
1770			goto out;
1771		}
1772	}
1773
1774	retval = do_generic_file_read(filp, ppos, &i, retval);
1775out:
1776	return retval;
1777}
1778EXPORT_SYMBOL(generic_file_aio_read);
1779
1780#ifdef CONFIG_MMU
1781/**
1782 * page_cache_read - adds requested page to the page cache if not already there
1783 * @file:	file to read
1784 * @offset:	page index
 
1785 *
1786 * This adds the requested page to the page cache if it isn't already there,
1787 * and schedules an I/O to read in its contents from disk.
1788 */
1789static int page_cache_read(struct file *file, pgoff_t offset)
1790{
1791	struct address_space *mapping = file->f_mapping;
1792	struct page *page; 
1793	int ret;
1794
1795	do {
1796		page = page_cache_alloc_cold(mapping);
1797		if (!page)
1798			return -ENOMEM;
1799
1800		ret = add_to_page_cache_lru(page, mapping, offset, GFP_KERNEL);
1801		if (ret == 0)
1802			ret = mapping->a_ops->readpage(file, page);
1803		else if (ret == -EEXIST)
1804			ret = 0; /* losing race to add is OK */
1805
1806		page_cache_release(page);
1807
1808	} while (ret == AOP_TRUNCATED_PAGE);
1809		
1810	return ret;
1811}
1812
1813#define MMAP_LOTSAMISS  (100)
1814
1815/*
1816 * Synchronous readahead happens when we don't even find
1817 * a page in the page cache at all.
1818 */
1819static void do_sync_mmap_readahead(struct vm_area_struct *vma,
1820				   struct file_ra_state *ra,
1821				   struct file *file,
1822				   pgoff_t offset)
1823{
1824	unsigned long ra_pages;
1825	struct address_space *mapping = file->f_mapping;
1826
1827	/* If we don't want any read-ahead, don't bother */
1828	if (vma->vm_flags & VM_RAND_READ)
1829		return;
1830	if (!ra->ra_pages)
1831		return;
1832
1833	if (vma->vm_flags & VM_SEQ_READ) {
1834		page_cache_sync_readahead(mapping, ra, file, offset,
1835					  ra->ra_pages);
1836		return;
1837	}
1838
1839	/* Avoid banging the cache line if not needed */
1840	if (ra->mmap_miss < MMAP_LOTSAMISS * 10)
1841		ra->mmap_miss++;
1842
1843	/*
1844	 * Do we miss much more than hit in this file? If so,
1845	 * stop bothering with read-ahead. It will only hurt.
1846	 */
1847	if (ra->mmap_miss > MMAP_LOTSAMISS)
1848		return;
1849
1850	/*
1851	 * mmap read-around
1852	 */
1853	ra_pages = max_sane_readahead(ra->ra_pages);
1854	ra->start = max_t(long, 0, offset - ra_pages / 2);
1855	ra->size = ra_pages;
1856	ra->async_size = ra_pages / 4;
1857	ra_submit(ra, mapping, file);
1858}
1859
1860/*
1861 * Asynchronous readahead happens when we find the page and PG_readahead,
1862 * so we want to possibly extend the readahead further..
1863 */
1864static void do_async_mmap_readahead(struct vm_area_struct *vma,
1865				    struct file_ra_state *ra,
1866				    struct file *file,
1867				    struct page *page,
1868				    pgoff_t offset)
1869{
1870	struct address_space *mapping = file->f_mapping;
1871
1872	/* If we don't want any read-ahead, don't bother */
1873	if (vma->vm_flags & VM_RAND_READ)
1874		return;
1875	if (ra->mmap_miss > 0)
1876		ra->mmap_miss--;
1877	if (PageReadahead(page))
1878		page_cache_async_readahead(mapping, ra, file,
1879					   page, offset, ra->ra_pages);
1880}
1881
1882/**
1883 * filemap_fault - read in file data for page fault handling
1884 * @vma:	vma in which the fault was taken
1885 * @vmf:	struct vm_fault containing details of the fault
1886 *
1887 * filemap_fault() is invoked via the vma operations vector for a
1888 * mapped memory region to read in file data during a page fault.
1889 *
1890 * The goto's are kind of ugly, but this streamlines the normal case of having
1891 * it in the page cache, and handles the special cases reasonably without
1892 * having a lot of duplicated code.
 
 
 
 
 
 
 
 
 
 
 
 
1893 */
1894int filemap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1895{
1896	int error;
1897	struct file *file = vma->vm_file;
1898	struct address_space *mapping = file->f_mapping;
1899	struct file_ra_state *ra = &file->f_ra;
1900	struct inode *inode = mapping->host;
1901	pgoff_t offset = vmf->pgoff;
 
1902	struct page *page;
1903	loff_t size;
1904	int ret = 0;
1905
1906	size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1907	if (offset >= size >> PAGE_CACHE_SHIFT)
1908		return VM_FAULT_SIGBUS;
1909
1910	/*
1911	 * Do we have something in the page cache already?
1912	 */
1913	page = find_get_page(mapping, offset);
1914	if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
1915		/*
1916		 * We found the page, so try async readahead before
1917		 * waiting for the lock.
1918		 */
1919		do_async_mmap_readahead(vma, ra, file, page, offset);
1920	} else if (!page) {
1921		/* No page in the page cache at all */
1922		do_sync_mmap_readahead(vma, ra, file, offset);
1923		count_vm_event(PGMAJFAULT);
1924		mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1925		ret = VM_FAULT_MAJOR;
1926retry_find:
1927		page = find_get_page(mapping, offset);
1928		if (!page)
1929			goto no_cached_page;
1930	}
1931
1932	if (!lock_page_or_retry(page, vma->vm_mm, vmf->flags)) {
1933		page_cache_release(page);
1934		return ret | VM_FAULT_RETRY;
1935	}
1936
1937	/* Did it get truncated? */
1938	if (unlikely(page->mapping != mapping)) {
1939		unlock_page(page);
1940		put_page(page);
1941		goto retry_find;
1942	}
1943	VM_BUG_ON_PAGE(page->index != offset, page);
1944
1945	/*
1946	 * We have a locked page in the page cache, now we need to check
1947	 * that it's up-to-date. If not, it is going to be due to an error.
1948	 */
1949	if (unlikely(!PageUptodate(page)))
1950		goto page_not_uptodate;
1951
1952	/*
1953	 * Found the page and have a reference on it.
1954	 * We must recheck i_size under page lock.
1955	 */
1956	size = round_up(i_size_read(inode), PAGE_CACHE_SIZE);
1957	if (unlikely(offset >= size >> PAGE_CACHE_SHIFT)) {
1958		unlock_page(page);
1959		page_cache_release(page);
1960		return VM_FAULT_SIGBUS;
1961	}
1962
1963	vmf->page = page;
1964	return ret | VM_FAULT_LOCKED;
1965
1966no_cached_page:
1967	/*
1968	 * We're only likely to ever get here if MADV_RANDOM is in
1969	 * effect.
1970	 */
1971	error = page_cache_read(file, offset);
1972
1973	/*
1974	 * The page we want has now been added to the page cache.
1975	 * In the unlikely event that someone removed it in the
1976	 * meantime, we'll just come back here and read it again.
1977	 */
1978	if (error >= 0)
1979		goto retry_find;
1980
1981	/*
1982	 * An error return from page_cache_read can result if the
1983	 * system is low on memory, or a problem occurs while trying
1984	 * to schedule I/O.
1985	 */
1986	if (error == -ENOMEM)
1987		return VM_FAULT_OOM;
1988	return VM_FAULT_SIGBUS;
1989
1990page_not_uptodate:
1991	/*
1992	 * Umm, take care of errors if the page isn't up-to-date.
1993	 * Try to re-read it _once_. We do this synchronously,
1994	 * because there really aren't any performance issues here
1995	 * and we need to check for errors.
1996	 */
1997	ClearPageError(page);
1998	error = mapping->a_ops->readpage(file, page);
1999	if (!error) {
2000		wait_on_page_locked(page);
2001		if (!PageUptodate(page))
2002			error = -EIO;
2003	}
2004	page_cache_release(page);
2005
2006	if (!error || error == AOP_TRUNCATED_PAGE)
2007		goto retry_find;
2008
2009	/* Things didn't work out. Return zero to tell the mm layer so. */
2010	shrink_readahead_size_eio(file, ra);
2011	return VM_FAULT_SIGBUS;
2012}
2013EXPORT_SYMBOL(filemap_fault);
2014
2015void filemap_map_pages(struct vm_area_struct *vma, struct vm_fault *vmf)
 
2016{
2017	struct radix_tree_iter iter;
2018	void **slot;
2019	struct file *file = vma->vm_file;
2020	struct address_space *mapping = file->f_mapping;
2021	loff_t size;
2022	struct page *page;
2023	unsigned long address = (unsigned long) vmf->virtual_address;
2024	unsigned long addr;
2025	pte_t *pte;
2026
2027	rcu_read_lock();
2028	radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, vmf->pgoff) {
2029		if (iter.index > vmf->max_pgoff)
2030			break;
2031repeat:
2032		page = radix_tree_deref_slot(slot);
2033		if (unlikely(!page))
2034			goto next;
2035		if (radix_tree_exception(page)) {
2036			if (radix_tree_deref_retry(page))
2037				break;
2038			else
2039				goto next;
 
2040		}
2041
2042		if (!page_cache_get_speculative(page))
 
2043			goto repeat;
2044
 
 
 
 
 
 
2045		/* Has the page moved? */
2046		if (unlikely(page != *slot)) {
2047			page_cache_release(page);
2048			goto repeat;
2049		}
2050
2051		if (!PageUptodate(page) ||
2052				PageReadahead(page) ||
2053				PageHWPoison(page))
2054			goto skip;
2055		if (!trylock_page(page))
2056			goto skip;
2057
2058		if (page->mapping != mapping || !PageUptodate(page))
2059			goto unlock;
2060
2061		size = round_up(i_size_read(mapping->host), PAGE_CACHE_SIZE);
2062		if (page->index >= size >> PAGE_CACHE_SHIFT)
2063			goto unlock;
2064
2065		pte = vmf->pte + page->index - vmf->pgoff;
2066		if (!pte_none(*pte))
2067			goto unlock;
2068
2069		if (file->f_ra.mmap_miss > 0)
2070			file->f_ra.mmap_miss--;
2071		addr = address + (page->index - vmf->pgoff) * PAGE_SIZE;
2072		do_set_pte(vma, addr, page, pte, false, false);
 
 
 
 
 
2073		unlock_page(page);
2074		goto next;
2075unlock:
2076		unlock_page(page);
2077skip:
2078		page_cache_release(page);
2079next:
2080		if (iter.index == vmf->max_pgoff)
 
 
 
2081			break;
2082	}
2083	rcu_read_unlock();
2084}
2085EXPORT_SYMBOL(filemap_map_pages);
2086
2087int filemap_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf)
2088{
2089	struct page *page = vmf->page;
2090	struct inode *inode = file_inode(vma->vm_file);
2091	int ret = VM_FAULT_LOCKED;
2092
2093	sb_start_pagefault(inode->i_sb);
2094	file_update_time(vma->vm_file);
2095	lock_page(page);
2096	if (page->mapping != inode->i_mapping) {
2097		unlock_page(page);
2098		ret = VM_FAULT_NOPAGE;
2099		goto out;
2100	}
2101	/*
2102	 * We mark the page dirty already here so that when freeze is in
2103	 * progress, we are guaranteed that writeback during freezing will
2104	 * see the dirty page and writeprotect it again.
2105	 */
2106	set_page_dirty(page);
2107	wait_for_stable_page(page);
2108out:
2109	sb_end_pagefault(inode->i_sb);
2110	return ret;
2111}
2112EXPORT_SYMBOL(filemap_page_mkwrite);
2113
2114const struct vm_operations_struct generic_file_vm_ops = {
2115	.fault		= filemap_fault,
2116	.map_pages	= filemap_map_pages,
2117	.page_mkwrite	= filemap_page_mkwrite,
2118	.remap_pages	= generic_file_remap_pages,
2119};
2120
2121/* This is used for a general mmap of a disk file */
2122
2123int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2124{
2125	struct address_space *mapping = file->f_mapping;
2126
2127	if (!mapping->a_ops->readpage)
2128		return -ENOEXEC;
2129	file_accessed(file);
2130	vma->vm_ops = &generic_file_vm_ops;
2131	return 0;
2132}
2133
2134/*
2135 * This is for filesystems which do not implement ->writepage.
2136 */
2137int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2138{
2139	if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2140		return -EINVAL;
2141	return generic_file_mmap(file, vma);
2142}
2143#else
 
 
 
 
2144int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2145{
2146	return -ENOSYS;
2147}
2148int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2149{
2150	return -ENOSYS;
2151}
2152#endif /* CONFIG_MMU */
2153
 
2154EXPORT_SYMBOL(generic_file_mmap);
2155EXPORT_SYMBOL(generic_file_readonly_mmap);
2156
2157static struct page *wait_on_page_read(struct page *page)
2158{
2159	if (!IS_ERR(page)) {
2160		wait_on_page_locked(page);
2161		if (!PageUptodate(page)) {
2162			page_cache_release(page);
2163			page = ERR_PTR(-EIO);
2164		}
2165	}
2166	return page;
2167}
2168
2169static struct page *__read_cache_page(struct address_space *mapping,
2170				pgoff_t index,
2171				int (*filler)(void *, struct page *),
2172				void *data,
2173				gfp_t gfp)
2174{
2175	struct page *page;
2176	int err;
2177repeat:
2178	page = find_get_page(mapping, index);
2179	if (!page) {
2180		page = __page_cache_alloc(gfp | __GFP_COLD);
2181		if (!page)
2182			return ERR_PTR(-ENOMEM);
2183		err = add_to_page_cache_lru(page, mapping, index, gfp);
2184		if (unlikely(err)) {
2185			page_cache_release(page);
2186			if (err == -EEXIST)
2187				goto repeat;
2188			/* Presumably ENOMEM for radix tree node */
2189			return ERR_PTR(err);
2190		}
 
 
2191		err = filler(data, page);
2192		if (err < 0) {
2193			page_cache_release(page);
2194			page = ERR_PTR(err);
2195		} else {
2196			page = wait_on_page_read(page);
2197		}
 
 
 
 
 
2198	}
2199	return page;
2200}
2201
2202static struct page *do_read_cache_page(struct address_space *mapping,
2203				pgoff_t index,
2204				int (*filler)(void *, struct page *),
2205				void *data,
2206				gfp_t gfp)
2207
2208{
2209	struct page *page;
2210	int err;
2211
2212retry:
2213	page = __read_cache_page(mapping, index, filler, data, gfp);
2214	if (IS_ERR(page))
2215		return page;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2216	if (PageUptodate(page))
2217		goto out;
2218
 
2219	lock_page(page);
 
 
2220	if (!page->mapping) {
2221		unlock_page(page);
2222		page_cache_release(page);
2223		goto retry;
2224	}
 
 
2225	if (PageUptodate(page)) {
2226		unlock_page(page);
2227		goto out;
2228	}
2229	err = filler(data, page);
2230	if (err < 0) {
2231		page_cache_release(page);
2232		return ERR_PTR(err);
2233	} else {
2234		page = wait_on_page_read(page);
2235		if (IS_ERR(page))
2236			return page;
2237	}
2238out:
2239	mark_page_accessed(page);
2240	return page;
2241}
2242
2243/**
2244 * read_cache_page - read into page cache, fill it if needed
2245 * @mapping:	the page's address_space
2246 * @index:	the page index
2247 * @filler:	function to perform the read
2248 * @data:	first arg to filler(data, page) function, often left as NULL
2249 *
2250 * Read into the page cache. If a page already exists, and PageUptodate() is
2251 * not set, try to fill the page and wait for it to become unlocked.
2252 *
2253 * If the page does not get brought uptodate, return -EIO.
2254 */
2255struct page *read_cache_page(struct address_space *mapping,
2256				pgoff_t index,
2257				int (*filler)(void *, struct page *),
2258				void *data)
2259{
2260	return do_read_cache_page(mapping, index, filler, data, mapping_gfp_mask(mapping));
2261}
2262EXPORT_SYMBOL(read_cache_page);
2263
2264/**
2265 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
2266 * @mapping:	the page's address_space
2267 * @index:	the page index
2268 * @gfp:	the page allocator flags to use if allocating
2269 *
2270 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
2271 * any new page allocations done using the specified allocation flags.
2272 *
2273 * If the page does not get brought uptodate, return -EIO.
2274 */
2275struct page *read_cache_page_gfp(struct address_space *mapping,
2276				pgoff_t index,
2277				gfp_t gfp)
2278{
2279	filler_t *filler = (filler_t *)mapping->a_ops->readpage;
2280
2281	return do_read_cache_page(mapping, index, filler, NULL, gfp);
2282}
2283EXPORT_SYMBOL(read_cache_page_gfp);
2284
2285/*
2286 * Performs necessary checks before doing a write
2287 *
2288 * Can adjust writing position or amount of bytes to write.
2289 * Returns appropriate error code that caller should return or
2290 * zero in case that write should be allowed.
2291 */
2292inline int generic_write_checks(struct file *file, loff_t *pos, size_t *count, int isblk)
2293{
 
2294	struct inode *inode = file->f_mapping->host;
2295	unsigned long limit = rlimit(RLIMIT_FSIZE);
 
 
 
 
 
 
 
 
2296
2297        if (unlikely(*pos < 0))
2298                return -EINVAL;
2299
2300	if (!isblk) {
2301		/* FIXME: this is for backwards compatibility with 2.4 */
2302		if (file->f_flags & O_APPEND)
2303                        *pos = i_size_read(inode);
2304
2305		if (limit != RLIM_INFINITY) {
2306			if (*pos >= limit) {
2307				send_sig(SIGXFSZ, current, 0);
2308				return -EFBIG;
2309			}
2310			if (*count > limit - (typeof(limit))*pos) {
2311				*count = limit - (typeof(limit))*pos;
2312			}
2313		}
 
2314	}
2315
2316	/*
2317	 * LFS rule
2318	 */
2319	if (unlikely(*pos + *count > MAX_NON_LFS &&
2320				!(file->f_flags & O_LARGEFILE))) {
2321		if (*pos >= MAX_NON_LFS) {
2322			return -EFBIG;
2323		}
2324		if (*count > MAX_NON_LFS - (unsigned long)*pos) {
2325			*count = MAX_NON_LFS - (unsigned long)*pos;
2326		}
2327	}
2328
2329	/*
2330	 * Are we about to exceed the fs block limit ?
2331	 *
2332	 * If we have written data it becomes a short write.  If we have
2333	 * exceeded without writing data we send a signal and return EFBIG.
2334	 * Linus frestrict idea will clean these up nicely..
2335	 */
2336	if (likely(!isblk)) {
2337		if (unlikely(*pos >= inode->i_sb->s_maxbytes)) {
2338			if (*count || *pos > inode->i_sb->s_maxbytes) {
2339				return -EFBIG;
2340			}
2341			/* zero-length writes at ->s_maxbytes are OK */
2342		}
2343
2344		if (unlikely(*pos + *count > inode->i_sb->s_maxbytes))
2345			*count = inode->i_sb->s_maxbytes - *pos;
2346	} else {
2347#ifdef CONFIG_BLOCK
2348		loff_t isize;
2349		if (bdev_read_only(I_BDEV(inode)))
2350			return -EPERM;
2351		isize = i_size_read(inode);
2352		if (*pos >= isize) {
2353			if (*count || *pos > isize)
2354				return -ENOSPC;
2355		}
2356
2357		if (*pos + *count > isize)
2358			*count = isize - *pos;
2359#else
2360		return -EPERM;
2361#endif
2362	}
2363	return 0;
2364}
2365EXPORT_SYMBOL(generic_write_checks);
2366
2367int pagecache_write_begin(struct file *file, struct address_space *mapping,
2368				loff_t pos, unsigned len, unsigned flags,
2369				struct page **pagep, void **fsdata)
2370{
2371	const struct address_space_operations *aops = mapping->a_ops;
2372
2373	return aops->write_begin(file, mapping, pos, len, flags,
2374							pagep, fsdata);
2375}
2376EXPORT_SYMBOL(pagecache_write_begin);
2377
2378int pagecache_write_end(struct file *file, struct address_space *mapping,
2379				loff_t pos, unsigned len, unsigned copied,
2380				struct page *page, void *fsdata)
2381{
2382	const struct address_space_operations *aops = mapping->a_ops;
2383
2384	mark_page_accessed(page);
2385	return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
2386}
2387EXPORT_SYMBOL(pagecache_write_end);
2388
2389ssize_t
2390generic_file_direct_write(struct kiocb *iocb, const struct iovec *iov,
2391		unsigned long *nr_segs, loff_t pos,
2392		size_t count, size_t ocount)
2393{
2394	struct file	*file = iocb->ki_filp;
2395	struct address_space *mapping = file->f_mapping;
2396	struct inode	*inode = mapping->host;
 
2397	ssize_t		written;
2398	size_t		write_len;
2399	pgoff_t		end;
2400
2401	if (count != ocount)
2402		*nr_segs = iov_shorten((struct iovec *)iov, *nr_segs, count);
2403
2404	write_len = iov_length(iov, *nr_segs);
2405	end = (pos + write_len - 1) >> PAGE_CACHE_SHIFT;
2406
2407	written = filemap_write_and_wait_range(mapping, pos, pos + write_len - 1);
2408	if (written)
2409		goto out;
 
 
 
 
 
2410
2411	/*
2412	 * After a write we want buffered reads to be sure to go to disk to get
2413	 * the new data.  We invalidate clean cached page from the region we're
2414	 * about to write.  We do this *before* the write so that we can return
2415	 * without clobbering -EIOCBQUEUED from ->direct_IO().
2416	 */
2417	if (mapping->nrpages) {
2418		written = invalidate_inode_pages2_range(mapping,
2419					pos >> PAGE_CACHE_SHIFT, end);
2420		/*
2421		 * If a page can not be invalidated, return 0 to fall back
2422		 * to buffered write.
2423		 */
2424		if (written) {
2425			if (written == -EBUSY)
2426				return 0;
2427			goto out;
2428		}
2429	}
2430
2431	written = mapping->a_ops->direct_IO(WRITE, iocb, iov, pos, *nr_segs);
2432
2433	/*
2434	 * Finally, try again to invalidate clean pages which might have been
2435	 * cached by non-direct readahead, or faulted in by get_user_pages()
2436	 * if the source of the write was an mmap'ed region of the file
2437	 * we're writing.  Either one is a pretty crazy thing to do,
2438	 * so we don't support it 100%.  If this invalidation
2439	 * fails, tough, the write still worked...
 
 
 
 
 
2440	 */
2441	if (mapping->nrpages) {
2442		invalidate_inode_pages2_range(mapping,
2443					      pos >> PAGE_CACHE_SHIFT, end);
2444	}
2445
2446	if (written > 0) {
2447		pos += written;
 
2448		if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
2449			i_size_write(inode, pos);
2450			mark_inode_dirty(inode);
2451		}
2452		iocb->ki_pos = pos;
2453	}
 
2454out:
2455	return written;
2456}
2457EXPORT_SYMBOL(generic_file_direct_write);
2458
2459/*
2460 * Find or create a page at the given pagecache position. Return the locked
2461 * page. This function is specifically for buffered writes.
2462 */
2463struct page *grab_cache_page_write_begin(struct address_space *mapping,
2464					pgoff_t index, unsigned flags)
2465{
2466	int status;
2467	gfp_t gfp_mask;
2468	struct page *page;
2469	gfp_t gfp_notmask = 0;
2470
2471	gfp_mask = mapping_gfp_mask(mapping);
2472	if (mapping_cap_account_dirty(mapping))
2473		gfp_mask |= __GFP_WRITE;
2474	if (flags & AOP_FLAG_NOFS)
2475		gfp_notmask = __GFP_FS;
2476repeat:
2477	page = find_lock_page(mapping, index);
 
2478	if (page)
2479		goto found;
2480
2481	page = __page_cache_alloc(gfp_mask & ~gfp_notmask);
2482	if (!page)
2483		return NULL;
2484	status = add_to_page_cache_lru(page, mapping, index,
2485						GFP_KERNEL & ~gfp_notmask);
2486	if (unlikely(status)) {
2487		page_cache_release(page);
2488		if (status == -EEXIST)
2489			goto repeat;
2490		return NULL;
2491	}
2492found:
2493	wait_for_stable_page(page);
2494	return page;
2495}
2496EXPORT_SYMBOL(grab_cache_page_write_begin);
2497
2498ssize_t generic_perform_write(struct file *file,
2499				struct iov_iter *i, loff_t pos)
2500{
2501	struct address_space *mapping = file->f_mapping;
2502	const struct address_space_operations *a_ops = mapping->a_ops;
2503	long status = 0;
2504	ssize_t written = 0;
2505	unsigned int flags = 0;
2506
2507	/*
2508	 * Copies from kernel address space cannot fail (NFSD is a big user).
2509	 */
2510	if (segment_eq(get_fs(), KERNEL_DS))
2511		flags |= AOP_FLAG_UNINTERRUPTIBLE;
2512
2513	do {
2514		struct page *page;
2515		unsigned long offset;	/* Offset into pagecache page */
2516		unsigned long bytes;	/* Bytes to write to page */
2517		size_t copied;		/* Bytes copied from user */
2518		void *fsdata;
2519
2520		offset = (pos & (PAGE_CACHE_SIZE - 1));
2521		bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2522						iov_iter_count(i));
2523
2524again:
2525		/*
2526		 * Bring in the user page that we will copy from _first_.
2527		 * Otherwise there's a nasty deadlock on copying from the
2528		 * same page as we're writing to, without it being marked
2529		 * up-to-date.
2530		 *
2531		 * Not only is this an optimisation, but it is also required
2532		 * to check that the address is actually valid, when atomic
2533		 * usercopies are used, below.
2534		 */
2535		if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
2536			status = -EFAULT;
2537			break;
2538		}
2539
 
 
 
 
 
2540		status = a_ops->write_begin(file, mapping, pos, bytes, flags,
2541						&page, &fsdata);
2542		if (unlikely(status))
2543			break;
2544
2545		if (mapping_writably_mapped(mapping))
2546			flush_dcache_page(page);
2547
2548		copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
2549		flush_dcache_page(page);
2550
2551		mark_page_accessed(page);
2552		status = a_ops->write_end(file, mapping, pos, bytes, copied,
2553						page, fsdata);
2554		if (unlikely(status < 0))
2555			break;
2556		copied = status;
2557
2558		cond_resched();
2559
2560		iov_iter_advance(i, copied);
2561		if (unlikely(copied == 0)) {
2562			/*
2563			 * If we were unable to copy any data at all, we must
2564			 * fall back to a single segment length write.
2565			 *
2566			 * If we didn't fallback here, we could livelock
2567			 * because not all segments in the iov can be copied at
2568			 * once without a pagefault.
2569			 */
2570			bytes = min_t(unsigned long, PAGE_CACHE_SIZE - offset,
2571						iov_iter_single_seg_count(i));
2572			goto again;
2573		}
2574		pos += copied;
2575		written += copied;
2576
2577		balance_dirty_pages_ratelimited(mapping);
2578		if (fatal_signal_pending(current)) {
2579			status = -EINTR;
2580			break;
2581		}
2582	} while (iov_iter_count(i));
2583
2584	return written ? written : status;
2585}
2586EXPORT_SYMBOL(generic_perform_write);
2587
2588/**
2589 * __generic_file_aio_write - write data to a file
2590 * @iocb:	IO state structure (file, offset, etc.)
2591 * @iov:	vector with data to write
2592 * @nr_segs:	number of segments in the vector
2593 *
2594 * This function does all the work needed for actually writing data to a
2595 * file. It does all basic checks, removes SUID from the file, updates
2596 * modification times and calls proper subroutines depending on whether we
2597 * do direct IO or a standard buffered write.
2598 *
2599 * It expects i_mutex to be grabbed unless we work on a block device or similar
2600 * object which does not need locking at all.
2601 *
2602 * This function does *not* take care of syncing data in case of O_SYNC write.
2603 * A caller has to handle it. This is mainly due to the fact that we want to
2604 * avoid syncing under i_mutex.
2605 */
2606ssize_t __generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2607				 unsigned long nr_segs)
2608{
2609	struct file *file = iocb->ki_filp;
2610	struct address_space * mapping = file->f_mapping;
2611	size_t ocount;		/* original count */
2612	size_t count;		/* after file limit checks */
2613	struct inode 	*inode = mapping->host;
2614	loff_t		pos = iocb->ki_pos;
2615	ssize_t		written = 0;
2616	ssize_t		err;
2617	ssize_t		status;
2618	struct iov_iter from;
2619
2620	ocount = 0;
2621	err = generic_segment_checks(iov, &nr_segs, &ocount, VERIFY_READ);
2622	if (err)
2623		return err;
2624
2625	count = ocount;
2626
2627	/* We can write back this queue in page reclaim */
2628	current->backing_dev_info = mapping->backing_dev_info;
2629	err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
2630	if (err)
2631		goto out;
2632
2633	if (count == 0)
2634		goto out;
2635
2636	err = file_remove_suid(file);
2637	if (err)
2638		goto out;
2639
2640	err = file_update_time(file);
2641	if (err)
2642		goto out;
2643
2644	iov_iter_init(&from, iov, nr_segs, count, 0);
2645
2646	/* coalesce the iovecs and go direct-to-BIO for O_DIRECT */
2647	if (unlikely(file->f_flags & O_DIRECT)) {
2648		loff_t endbyte;
2649
2650		written = generic_file_direct_write(iocb, iov, &from.nr_segs, pos,
2651							count, ocount);
2652		if (written < 0 || written == count)
2653			goto out;
2654		iov_iter_advance(&from, written);
2655
 
2656		/*
2657		 * direct-io write to a hole: fall through to buffered I/O
2658		 * for completing the rest of the request.
 
 
 
2659		 */
2660		pos += written;
2661		count -= written;
2662
2663		status = generic_perform_write(file, &from, pos);
2664		/*
2665		 * If generic_perform_write() returned a synchronous error
2666		 * then we want to return the number of bytes which were
2667		 * direct-written, or the error code if that was zero.  Note
2668		 * that this differs from normal direct-io semantics, which
2669		 * will return -EFOO even if some bytes were written.
2670		 */
2671		if (unlikely(status < 0) && !written) {
2672			err = status;
2673			goto out;
2674		}
2675		iocb->ki_pos = pos + status;
2676		/*
2677		 * We need to ensure that the page cache pages are written to
2678		 * disk and invalidated to preserve the expected O_DIRECT
2679		 * semantics.
2680		 */
2681		endbyte = pos + status - 1;
2682		err = filemap_write_and_wait_range(file->f_mapping, pos, endbyte);
2683		if (err == 0) {
 
2684			written += status;
2685			invalidate_mapping_pages(mapping,
2686						 pos >> PAGE_CACHE_SHIFT,
2687						 endbyte >> PAGE_CACHE_SHIFT);
2688		} else {
2689			/*
2690			 * We don't know how much we wrote, so just return
2691			 * the number of bytes which were direct-written
2692			 */
2693		}
2694	} else {
2695		written = generic_perform_write(file, &from, pos);
2696		if (likely(written >= 0))
2697			iocb->ki_pos = pos + written;
2698	}
2699out:
2700	current->backing_dev_info = NULL;
2701	return written ? written : err;
2702}
2703EXPORT_SYMBOL(__generic_file_aio_write);
2704
2705/**
2706 * generic_file_aio_write - write data to a file
2707 * @iocb:	IO state structure
2708 * @iov:	vector with data to write
2709 * @nr_segs:	number of segments in the vector
2710 * @pos:	position in file where to write
2711 *
2712 * This is a wrapper around __generic_file_aio_write() to be used by most
2713 * filesystems. It takes care of syncing the file in case of O_SYNC file
2714 * and acquires i_mutex as needed.
2715 */
2716ssize_t generic_file_aio_write(struct kiocb *iocb, const struct iovec *iov,
2717		unsigned long nr_segs, loff_t pos)
2718{
2719	struct file *file = iocb->ki_filp;
2720	struct inode *inode = file->f_mapping->host;
2721	ssize_t ret;
2722
2723	BUG_ON(iocb->ki_pos != pos);
 
 
 
 
2724
2725	mutex_lock(&inode->i_mutex);
2726	ret = __generic_file_aio_write(iocb, iov, nr_segs);
2727	mutex_unlock(&inode->i_mutex);
2728
2729	if (ret > 0) {
2730		ssize_t err;
2731
2732		err = generic_write_sync(file, iocb->ki_pos - ret, ret);
2733		if (err < 0)
2734			ret = err;
2735	}
2736	return ret;
2737}
2738EXPORT_SYMBOL(generic_file_aio_write);
2739
2740/**
2741 * try_to_release_page() - release old fs-specific metadata on a page
2742 *
2743 * @page: the page which the kernel is trying to free
2744 * @gfp_mask: memory allocation flags (and I/O mode)
2745 *
2746 * The address_space is to try to release any data against the page
2747 * (presumably at page->private).  If the release was successful, return `1'.
2748 * Otherwise return zero.
2749 *
2750 * This may also be called if PG_fscache is set on a page, indicating that the
2751 * page is known to the local caching routines.
2752 *
2753 * The @gfp_mask argument specifies whether I/O may be performed to release
2754 * this page (__GFP_IO), and whether the call may block (__GFP_WAIT & __GFP_FS).
2755 *
2756 */
2757int try_to_release_page(struct page *page, gfp_t gfp_mask)
2758{
2759	struct address_space * const mapping = page->mapping;
2760
2761	BUG_ON(!PageLocked(page));
2762	if (PageWriteback(page))
2763		return 0;
2764
2765	if (mapping && mapping->a_ops->releasepage)
2766		return mapping->a_ops->releasepage(page, gfp_mask);
2767	return try_to_free_buffers(page);
2768}
2769
2770EXPORT_SYMBOL(try_to_release_page);