Linux Audio

Check our new training course

Loading...
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *	linux/mm/filemap.c
   4 *
   5 * Copyright (C) 1994-1999  Linus Torvalds
   6 */
   7
   8/*
   9 * This file handles the generic file mmap semantics used by
  10 * most "normal" filesystems (but you don't /have/ to use this:
  11 * the NFS filesystem used to do this differently, for example)
  12 */
  13#include <linux/export.h>
  14#include <linux/compiler.h>
  15#include <linux/dax.h>
  16#include <linux/fs.h>
  17#include <linux/sched/signal.h>
  18#include <linux/uaccess.h>
  19#include <linux/capability.h>
  20#include <linux/kernel_stat.h>
  21#include <linux/gfp.h>
  22#include <linux/mm.h>
  23#include <linux/swap.h>
 
 
  24#include <linux/mman.h>
  25#include <linux/pagemap.h>
  26#include <linux/file.h>
  27#include <linux/uio.h>
  28#include <linux/error-injection.h>
  29#include <linux/hash.h>
  30#include <linux/writeback.h>
  31#include <linux/backing-dev.h>
  32#include <linux/pagevec.h>
  33#include <linux/blkdev.h>
  34#include <linux/security.h>
  35#include <linux/cpuset.h>
  36#include <linux/hugetlb.h>
  37#include <linux/memcontrol.h>
  38#include <linux/cleancache.h>
  39#include <linux/shmem_fs.h>
  40#include <linux/rmap.h>
  41#include <linux/delayacct.h>
  42#include <linux/psi.h>
  43#include <linux/ramfs.h>
  44#include <linux/page_idle.h>
 
 
 
 
 
 
  45#include "internal.h"
  46
  47#define CREATE_TRACE_POINTS
  48#include <trace/events/filemap.h>
  49
  50/*
  51 * FIXME: remove all knowledge of the buffer layer from the core VM
  52 */
  53#include <linux/buffer_head.h> /* for try_to_free_buffers */
  54
  55#include <asm/mman.h>
  56
 
 
  57/*
  58 * Shared mappings implemented 30.11.1994. It's not fully working yet,
  59 * though.
  60 *
  61 * Shared mappings now work. 15.8.1995  Bruno.
  62 *
  63 * finished 'unifying' the page and buffer cache and SMP-threaded the
  64 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
  65 *
  66 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
  67 */
  68
  69/*
  70 * Lock ordering:
  71 *
  72 *  ->i_mmap_rwsem		(truncate_pagecache)
  73 *    ->private_lock		(__free_pte->__set_page_dirty_buffers)
  74 *      ->swap_lock		(exclusive_swap_page, others)
  75 *        ->i_pages lock
  76 *
  77 *  ->i_mutex
  78 *    ->i_mmap_rwsem		(truncate->unmap_mapping_range)
 
  79 *
  80 *  ->mmap_lock
  81 *    ->i_mmap_rwsem
  82 *      ->page_table_lock or pte_lock	(various, mainly in memory.c)
  83 *        ->i_pages lock	(arch-dependent flush_dcache_mmap_lock)
  84 *
  85 *  ->mmap_lock
  86 *    ->lock_page		(access_process_vm)
 
  87 *
  88 *  ->i_mutex			(generic_perform_write)
  89 *    ->mmap_lock		(fault_in_pages_readable->do_page_fault)
  90 *
  91 *  bdi->wb.list_lock
  92 *    sb_lock			(fs/fs-writeback.c)
  93 *    ->i_pages lock		(__sync_single_inode)
  94 *
  95 *  ->i_mmap_rwsem
  96 *    ->anon_vma.lock		(vma_adjust)
  97 *
  98 *  ->anon_vma.lock
  99 *    ->page_table_lock or pte_lock	(anon_vma_prepare and various)
 100 *
 101 *  ->page_table_lock or pte_lock
 102 *    ->swap_lock		(try_to_unmap_one)
 103 *    ->private_lock		(try_to_unmap_one)
 104 *    ->i_pages lock		(try_to_unmap_one)
 105 *    ->pgdat->lru_lock		(follow_page->mark_page_accessed)
 106 *    ->pgdat->lru_lock		(check_pte_range->isolate_lru_page)
 107 *    ->private_lock		(page_remove_rmap->set_page_dirty)
 108 *    ->i_pages lock		(page_remove_rmap->set_page_dirty)
 109 *    bdi.wb->list_lock		(page_remove_rmap->set_page_dirty)
 110 *    ->inode->i_lock		(page_remove_rmap->set_page_dirty)
 111 *    ->memcg->move_lock	(page_remove_rmap->lock_page_memcg)
 112 *    bdi.wb->list_lock		(zap_pte_range->set_page_dirty)
 113 *    ->inode->i_lock		(zap_pte_range->set_page_dirty)
 114 *    ->private_lock		(zap_pte_range->__set_page_dirty_buffers)
 115 *
 116 * ->i_mmap_rwsem
 117 *   ->tasklist_lock            (memory_failure, collect_procs_ao)
 118 */
 119
 
 
 
 
 
 
 
 
 
 120static void page_cache_delete(struct address_space *mapping,
 121				   struct page *page, void *shadow)
 122{
 123	XA_STATE(xas, &mapping->i_pages, page->index);
 124	unsigned int nr = 1;
 125
 126	mapping_set_update(&xas, mapping);
 127
 128	/* hugetlb pages are represented by a single entry in the xarray */
 129	if (!PageHuge(page)) {
 130		xas_set_order(&xas, page->index, compound_order(page));
 131		nr = compound_nr(page);
 132	}
 133
 134	VM_BUG_ON_PAGE(!PageLocked(page), page);
 135	VM_BUG_ON_PAGE(PageTail(page), page);
 136	VM_BUG_ON_PAGE(nr != 1 && shadow, page);
 137
 138	xas_store(&xas, shadow);
 139	xas_init_marks(&xas);
 140
 141	page->mapping = NULL;
 142	/* Leave page->index set: truncation lookup relies upon it */
 143
 144	if (shadow) {
 145		mapping->nrexceptional += nr;
 146		/*
 147		 * Make sure the nrexceptional update is committed before
 148		 * the nrpages update so that final truncate racing
 149		 * with reclaim does not see both counters 0 at the
 150		 * same time and miss a shadow entry.
 151		 */
 152		smp_wmb();
 153	}
 154	mapping->nrpages -= nr;
 155}
 156
 157static void unaccount_page_cache_page(struct address_space *mapping,
 158				      struct page *page)
 159{
 160	int nr;
 161
 162	/*
 163	 * if we're uptodate, flush out into the cleancache, otherwise
 164	 * invalidate any existing cleancache entries.  We can't leave
 165	 * stale data around in the cleancache once our page is gone
 166	 */
 167	if (PageUptodate(page) && PageMappedToDisk(page))
 168		cleancache_put_page(page);
 169	else
 170		cleancache_invalidate_page(mapping, page);
 171
 172	VM_BUG_ON_PAGE(PageTail(page), page);
 173	VM_BUG_ON_PAGE(page_mapped(page), page);
 174	if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(page_mapped(page))) {
 175		int mapcount;
 176
 
 
 177		pr_alert("BUG: Bad page cache in process %s  pfn:%05lx\n",
 178			 current->comm, page_to_pfn(page));
 179		dump_page(page, "still mapped when deleted");
 180		dump_stack();
 181		add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
 182
 183		mapcount = page_mapcount(page);
 184		if (mapping_exiting(mapping) &&
 185		    page_count(page) >= mapcount + 2) {
 186			/*
 187			 * All vmas have already been torn down, so it's
 188			 * a good bet that actually the page is unmapped,
 189			 * and we'd prefer not to leak it: if we're wrong,
 190			 * some other bad page check should catch it later.
 191			 */
 192			page_mapcount_reset(page);
 193			page_ref_sub(page, mapcount);
 
 
 194		}
 195	}
 196
 197	/* hugetlb pages do not participate in page cache accounting. */
 198	if (PageHuge(page))
 199		return;
 200
 201	nr = thp_nr_pages(page);
 202
 203	__mod_lruvec_page_state(page, NR_FILE_PAGES, -nr);
 204	if (PageSwapBacked(page)) {
 205		__mod_lruvec_page_state(page, NR_SHMEM, -nr);
 206		if (PageTransHuge(page))
 207			__dec_node_page_state(page, NR_SHMEM_THPS);
 208	} else if (PageTransHuge(page)) {
 209		__dec_node_page_state(page, NR_FILE_THPS);
 210		filemap_nr_thps_dec(mapping);
 211	}
 212
 213	/*
 214	 * At this point page must be either written or cleaned by
 215	 * truncate.  Dirty page here signals a bug and loss of
 216	 * unwritten data.
 217	 *
 218	 * This fixes dirty accounting after removing the page entirely
 219	 * but leaves PageDirty set: it has no effect for truncated
 220	 * page and anyway will be cleared before returning page into
 
 
 
 
 221	 * buddy allocator.
 222	 */
 223	if (WARN_ON_ONCE(PageDirty(page)))
 224		account_page_cleaned(page, mapping, inode_to_wb(mapping->host));
 
 225}
 226
 227/*
 228 * Delete a page from the page cache and free it. Caller has to make
 229 * sure the page is locked and that nobody else uses it - or that usage
 230 * is safe.  The caller must hold the i_pages lock.
 231 */
 232void __delete_from_page_cache(struct page *page, void *shadow)
 233{
 234	struct address_space *mapping = page->mapping;
 235
 236	trace_mm_filemap_delete_from_page_cache(page);
 237
 238	unaccount_page_cache_page(mapping, page);
 239	page_cache_delete(mapping, page, shadow);
 
 240}
 241
 242static void page_cache_free_page(struct address_space *mapping,
 243				struct page *page)
 244{
 245	void (*freepage)(struct page *);
 
 246
 247	freepage = mapping->a_ops->freepage;
 248	if (freepage)
 249		freepage(page);
 250
 251	if (PageTransHuge(page) && !PageHuge(page)) {
 252		page_ref_sub(page, HPAGE_PMD_NR);
 253		VM_BUG_ON_PAGE(page_count(page) <= 0, page);
 254	} else {
 255		put_page(page);
 256	}
 257}
 258
 259/**
 260 * delete_from_page_cache - delete page from page cache
 261 * @page: the page which the kernel is trying to remove from page cache
 262 *
 263 * This must be called only on pages that have been verified to be in the page
 264 * cache and locked.  It will never put the page into the free list, the caller
 265 * has a reference on the page.
 266 */
 267void delete_from_page_cache(struct page *page)
 268{
 269	struct address_space *mapping = page_mapping(page);
 270	unsigned long flags;
 
 
 
 
 
 
 
 
 
 
 
 271
 272	BUG_ON(!PageLocked(page));
 273	xa_lock_irqsave(&mapping->i_pages, flags);
 274	__delete_from_page_cache(page, NULL);
 275	xa_unlock_irqrestore(&mapping->i_pages, flags);
 276
 277	page_cache_free_page(mapping, page);
 278}
 279EXPORT_SYMBOL(delete_from_page_cache);
 280
 281/*
 282 * page_cache_delete_batch - delete several pages from page cache
 283 * @mapping: the mapping to which pages belong
 284 * @pvec: pagevec with pages to delete
 285 *
 286 * The function walks over mapping->i_pages and removes pages passed in @pvec
 287 * from the mapping. The function expects @pvec to be sorted by page index
 288 * and is optimised for it to be dense.
 289 * It tolerates holes in @pvec (mapping entries at those indices are not
 290 * modified). The function expects only THP head pages to be present in the
 291 * @pvec.
 292 *
 293 * The function expects the i_pages lock to be held.
 294 */
 295static void page_cache_delete_batch(struct address_space *mapping,
 296			     struct pagevec *pvec)
 297{
 298	XA_STATE(xas, &mapping->i_pages, pvec->pages[0]->index);
 299	int total_pages = 0;
 300	int i = 0;
 301	struct page *page;
 302
 303	mapping_set_update(&xas, mapping);
 304	xas_for_each(&xas, page, ULONG_MAX) {
 305		if (i >= pagevec_count(pvec))
 306			break;
 307
 308		/* A swap/dax/shadow entry got inserted? Skip it. */
 309		if (xa_is_value(page))
 310			continue;
 311		/*
 312		 * A page got inserted in our range? Skip it. We have our
 313		 * pages locked so they are protected from being removed.
 314		 * If we see a page whose index is higher than ours, it
 315		 * means our page has been removed, which shouldn't be
 316		 * possible because we're holding the PageLock.
 317		 */
 318		if (page != pvec->pages[i]) {
 319			VM_BUG_ON_PAGE(page->index > pvec->pages[i]->index,
 320					page);
 321			continue;
 322		}
 323
 324		WARN_ON_ONCE(!PageLocked(page));
 325
 326		if (page->index == xas.xa_index)
 327			page->mapping = NULL;
 328		/* Leave page->index set: truncation lookup relies on it */
 329
 330		/*
 331		 * Move to the next page in the vector if this is a regular
 332		 * page or the index is of the last sub-page of this compound
 333		 * page.
 334		 */
 335		if (page->index + compound_nr(page) - 1 == xas.xa_index)
 336			i++;
 337		xas_store(&xas, NULL);
 338		total_pages++;
 339	}
 340	mapping->nrpages -= total_pages;
 341}
 342
 343void delete_from_page_cache_batch(struct address_space *mapping,
 344				  struct pagevec *pvec)
 345{
 346	int i;
 347	unsigned long flags;
 348
 349	if (!pagevec_count(pvec))
 350		return;
 351
 352	xa_lock_irqsave(&mapping->i_pages, flags);
 353	for (i = 0; i < pagevec_count(pvec); i++) {
 354		trace_mm_filemap_delete_from_page_cache(pvec->pages[i]);
 355
 356		unaccount_page_cache_page(mapping, pvec->pages[i]);
 357	}
 358	page_cache_delete_batch(mapping, pvec);
 359	xa_unlock_irqrestore(&mapping->i_pages, flags);
 
 
 
 
 
 360
 361	for (i = 0; i < pagevec_count(pvec); i++)
 362		page_cache_free_page(mapping, pvec->pages[i]);
 363}
 364
 365int filemap_check_errors(struct address_space *mapping)
 366{
 367	int ret = 0;
 368	/* Check for outstanding write errors */
 369	if (test_bit(AS_ENOSPC, &mapping->flags) &&
 370	    test_and_clear_bit(AS_ENOSPC, &mapping->flags))
 371		ret = -ENOSPC;
 372	if (test_bit(AS_EIO, &mapping->flags) &&
 373	    test_and_clear_bit(AS_EIO, &mapping->flags))
 374		ret = -EIO;
 375	return ret;
 376}
 377EXPORT_SYMBOL(filemap_check_errors);
 378
 379static int filemap_check_and_keep_errors(struct address_space *mapping)
 380{
 381	/* Check for outstanding write errors */
 382	if (test_bit(AS_EIO, &mapping->flags))
 383		return -EIO;
 384	if (test_bit(AS_ENOSPC, &mapping->flags))
 385		return -ENOSPC;
 386	return 0;
 387}
 388
 389/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 390 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
 391 * @mapping:	address space structure to write
 392 * @start:	offset in bytes where the range starts
 393 * @end:	offset in bytes where the range ends (inclusive)
 394 * @sync_mode:	enable synchronous operation
 395 *
 396 * Start writeback against all of a mapping's dirty pages that lie
 397 * within the byte offsets <start, end> inclusive.
 398 *
 399 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
 400 * opposed to a regular memory cleansing writeback.  The difference between
 401 * these two operations is that if a dirty page/buffer is encountered, it must
 402 * be waited upon, and not just skipped over.
 403 *
 404 * Return: %0 on success, negative error code otherwise.
 405 */
 406int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
 407				loff_t end, int sync_mode)
 408{
 409	int ret;
 410	struct writeback_control wbc = {
 411		.sync_mode = sync_mode,
 412		.nr_to_write = LONG_MAX,
 413		.range_start = start,
 414		.range_end = end,
 415	};
 416
 417	if (!mapping_cap_writeback_dirty(mapping) ||
 418	    !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
 419		return 0;
 420
 421	wbc_attach_fdatawrite_inode(&wbc, mapping->host);
 422	ret = do_writepages(mapping, &wbc);
 423	wbc_detach_inode(&wbc);
 424	return ret;
 425}
 426
 427static inline int __filemap_fdatawrite(struct address_space *mapping,
 428	int sync_mode)
 429{
 430	return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
 431}
 432
 433int filemap_fdatawrite(struct address_space *mapping)
 434{
 435	return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
 436}
 437EXPORT_SYMBOL(filemap_fdatawrite);
 438
 439int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
 440				loff_t end)
 441{
 442	return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
 443}
 444EXPORT_SYMBOL(filemap_fdatawrite_range);
 445
 446/**
 447 * filemap_flush - mostly a non-blocking flush
 448 * @mapping:	target address_space
 449 *
 450 * This is a mostly non-blocking flush.  Not suitable for data-integrity
 451 * purposes - I/O may not be started against all dirty pages.
 452 *
 453 * Return: %0 on success, negative error code otherwise.
 454 */
 455int filemap_flush(struct address_space *mapping)
 456{
 457	return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
 458}
 459EXPORT_SYMBOL(filemap_flush);
 460
 461/**
 462 * filemap_range_has_page - check if a page exists in range.
 463 * @mapping:           address space within which to check
 464 * @start_byte:        offset in bytes where the range starts
 465 * @end_byte:          offset in bytes where the range ends (inclusive)
 466 *
 467 * Find at least one page in the range supplied, usually used to check if
 468 * direct writing in this range will trigger a writeback.
 469 *
 470 * Return: %true if at least one page exists in the specified range,
 471 * %false otherwise.
 472 */
 473bool filemap_range_has_page(struct address_space *mapping,
 474			   loff_t start_byte, loff_t end_byte)
 475{
 476	struct page *page;
 477	XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
 478	pgoff_t max = end_byte >> PAGE_SHIFT;
 479
 480	if (end_byte < start_byte)
 481		return false;
 482
 483	rcu_read_lock();
 484	for (;;) {
 485		page = xas_find(&xas, max);
 486		if (xas_retry(&xas, page))
 487			continue;
 488		/* Shadow entries don't count */
 489		if (xa_is_value(page))
 490			continue;
 491		/*
 492		 * We don't need to try to pin this page; we're about to
 493		 * release the RCU lock anyway.  It is enough to know that
 494		 * there was a page here recently.
 495		 */
 496		break;
 497	}
 498	rcu_read_unlock();
 499
 500	return page != NULL;
 501}
 502EXPORT_SYMBOL(filemap_range_has_page);
 503
 504static void __filemap_fdatawait_range(struct address_space *mapping,
 505				     loff_t start_byte, loff_t end_byte)
 506{
 507	pgoff_t index = start_byte >> PAGE_SHIFT;
 508	pgoff_t end = end_byte >> PAGE_SHIFT;
 509	struct pagevec pvec;
 510	int nr_pages;
 511
 512	if (end_byte < start_byte)
 513		return;
 514
 515	pagevec_init(&pvec);
 516	while (index <= end) {
 517		unsigned i;
 518
 519		nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index,
 520				end, PAGECACHE_TAG_WRITEBACK);
 521		if (!nr_pages)
 
 522			break;
 523
 524		for (i = 0; i < nr_pages; i++) {
 525			struct page *page = pvec.pages[i];
 526
 527			wait_on_page_writeback(page);
 528			ClearPageError(page);
 529		}
 530		pagevec_release(&pvec);
 531		cond_resched();
 532	}
 533}
 534
 535/**
 536 * filemap_fdatawait_range - wait for writeback to complete
 537 * @mapping:		address space structure to wait for
 538 * @start_byte:		offset in bytes where the range starts
 539 * @end_byte:		offset in bytes where the range ends (inclusive)
 540 *
 541 * Walk the list of under-writeback pages of the given address space
 542 * in the given range and wait for all of them.  Check error status of
 543 * the address space and return it.
 544 *
 545 * Since the error status of the address space is cleared by this function,
 546 * callers are responsible for checking the return value and handling and/or
 547 * reporting the error.
 548 *
 549 * Return: error status of the address space.
 550 */
 551int filemap_fdatawait_range(struct address_space *mapping, loff_t start_byte,
 552			    loff_t end_byte)
 553{
 554	__filemap_fdatawait_range(mapping, start_byte, end_byte);
 555	return filemap_check_errors(mapping);
 556}
 557EXPORT_SYMBOL(filemap_fdatawait_range);
 558
 559/**
 560 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
 561 * @mapping:		address space structure to wait for
 562 * @start_byte:		offset in bytes where the range starts
 563 * @end_byte:		offset in bytes where the range ends (inclusive)
 564 *
 565 * Walk the list of under-writeback pages of the given address space in the
 566 * given range and wait for all of them.  Unlike filemap_fdatawait_range(),
 567 * this function does not clear error status of the address space.
 568 *
 569 * Use this function if callers don't handle errors themselves.  Expected
 570 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
 571 * fsfreeze(8)
 572 */
 573int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
 574		loff_t start_byte, loff_t end_byte)
 575{
 576	__filemap_fdatawait_range(mapping, start_byte, end_byte);
 577	return filemap_check_and_keep_errors(mapping);
 578}
 579EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
 580
 581/**
 582 * file_fdatawait_range - wait for writeback to complete
 583 * @file:		file pointing to address space structure to wait for
 584 * @start_byte:		offset in bytes where the range starts
 585 * @end_byte:		offset in bytes where the range ends (inclusive)
 586 *
 587 * Walk the list of under-writeback pages of the address space that file
 588 * refers to, in the given range and wait for all of them.  Check error
 589 * status of the address space vs. the file->f_wb_err cursor and return it.
 590 *
 591 * Since the error status of the file is advanced by this function,
 592 * callers are responsible for checking the return value and handling and/or
 593 * reporting the error.
 594 *
 595 * Return: error status of the address space vs. the file->f_wb_err cursor.
 596 */
 597int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
 598{
 599	struct address_space *mapping = file->f_mapping;
 600
 601	__filemap_fdatawait_range(mapping, start_byte, end_byte);
 602	return file_check_and_advance_wb_err(file);
 603}
 604EXPORT_SYMBOL(file_fdatawait_range);
 605
 606/**
 607 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
 608 * @mapping: address space structure to wait for
 609 *
 610 * Walk the list of under-writeback pages of the given address space
 611 * and wait for all of them.  Unlike filemap_fdatawait(), this function
 612 * does not clear error status of the address space.
 613 *
 614 * Use this function if callers don't handle errors themselves.  Expected
 615 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
 616 * fsfreeze(8)
 617 *
 618 * Return: error status of the address space.
 619 */
 620int filemap_fdatawait_keep_errors(struct address_space *mapping)
 621{
 622	__filemap_fdatawait_range(mapping, 0, LLONG_MAX);
 623	return filemap_check_and_keep_errors(mapping);
 624}
 625EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
 626
 627/* Returns true if writeback might be needed or already in progress. */
 628static bool mapping_needs_writeback(struct address_space *mapping)
 629{
 630	if (dax_mapping(mapping))
 631		return mapping->nrexceptional;
 632
 633	return mapping->nrpages;
 634}
 635
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 636/**
 637 * filemap_write_and_wait_range - write out & wait on a file range
 638 * @mapping:	the address_space for the pages
 639 * @lstart:	offset in bytes where the range starts
 640 * @lend:	offset in bytes where the range ends (inclusive)
 641 *
 642 * Write out and wait upon file offsets lstart->lend, inclusive.
 643 *
 644 * Note that @lend is inclusive (describes the last byte to be written) so
 645 * that this function can be used to write to the very end-of-file (end = -1).
 646 *
 647 * Return: error status of the address space.
 648 */
 649int filemap_write_and_wait_range(struct address_space *mapping,
 650				 loff_t lstart, loff_t lend)
 651{
 652	int err = 0;
 
 
 
 653
 654	if (mapping_needs_writeback(mapping)) {
 655		err = __filemap_fdatawrite_range(mapping, lstart, lend,
 656						 WB_SYNC_ALL);
 657		/*
 658		 * Even if the above returned error, the pages may be
 659		 * written partially (e.g. -ENOSPC), so we wait for it.
 660		 * But the -EIO is special case, it may indicate the worst
 661		 * thing (e.g. bug) happened, so we avoid waiting for it.
 662		 */
 663		if (err != -EIO) {
 664			int err2 = filemap_fdatawait_range(mapping,
 665						lstart, lend);
 666			if (!err)
 667				err = err2;
 668		} else {
 669			/* Clear any previously stored errors */
 670			filemap_check_errors(mapping);
 671		}
 672	} else {
 673		err = filemap_check_errors(mapping);
 674	}
 
 
 
 675	return err;
 676}
 677EXPORT_SYMBOL(filemap_write_and_wait_range);
 678
 679void __filemap_set_wb_err(struct address_space *mapping, int err)
 680{
 681	errseq_t eseq = errseq_set(&mapping->wb_err, err);
 682
 683	trace_filemap_set_wb_err(mapping, eseq);
 684}
 685EXPORT_SYMBOL(__filemap_set_wb_err);
 686
 687/**
 688 * file_check_and_advance_wb_err - report wb error (if any) that was previously
 689 * 				   and advance wb_err to current one
 690 * @file: struct file on which the error is being reported
 691 *
 692 * When userland calls fsync (or something like nfsd does the equivalent), we
 693 * want to report any writeback errors that occurred since the last fsync (or
 694 * since the file was opened if there haven't been any).
 695 *
 696 * Grab the wb_err from the mapping. If it matches what we have in the file,
 697 * then just quickly return 0. The file is all caught up.
 698 *
 699 * If it doesn't match, then take the mapping value, set the "seen" flag in
 700 * it and try to swap it into place. If it works, or another task beat us
 701 * to it with the new value, then update the f_wb_err and return the error
 702 * portion. The error at this point must be reported via proper channels
 703 * (a'la fsync, or NFS COMMIT operation, etc.).
 704 *
 705 * While we handle mapping->wb_err with atomic operations, the f_wb_err
 706 * value is protected by the f_lock since we must ensure that it reflects
 707 * the latest value swapped in for this file descriptor.
 708 *
 709 * Return: %0 on success, negative error code otherwise.
 710 */
 711int file_check_and_advance_wb_err(struct file *file)
 712{
 713	int err = 0;
 714	errseq_t old = READ_ONCE(file->f_wb_err);
 715	struct address_space *mapping = file->f_mapping;
 716
 717	/* Locklessly handle the common case where nothing has changed */
 718	if (errseq_check(&mapping->wb_err, old)) {
 719		/* Something changed, must use slow path */
 720		spin_lock(&file->f_lock);
 721		old = file->f_wb_err;
 722		err = errseq_check_and_advance(&mapping->wb_err,
 723						&file->f_wb_err);
 724		trace_file_check_and_advance_wb_err(file, old);
 725		spin_unlock(&file->f_lock);
 726	}
 727
 728	/*
 729	 * We're mostly using this function as a drop in replacement for
 730	 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
 731	 * that the legacy code would have had on these flags.
 732	 */
 733	clear_bit(AS_EIO, &mapping->flags);
 734	clear_bit(AS_ENOSPC, &mapping->flags);
 735	return err;
 736}
 737EXPORT_SYMBOL(file_check_and_advance_wb_err);
 738
 739/**
 740 * file_write_and_wait_range - write out & wait on a file range
 741 * @file:	file pointing to address_space with pages
 742 * @lstart:	offset in bytes where the range starts
 743 * @lend:	offset in bytes where the range ends (inclusive)
 744 *
 745 * Write out and wait upon file offsets lstart->lend, inclusive.
 746 *
 747 * Note that @lend is inclusive (describes the last byte to be written) so
 748 * that this function can be used to write to the very end-of-file (end = -1).
 749 *
 750 * After writing out and waiting on the data, we check and advance the
 751 * f_wb_err cursor to the latest value, and return any errors detected there.
 752 *
 753 * Return: %0 on success, negative error code otherwise.
 754 */
 755int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
 756{
 757	int err = 0, err2;
 758	struct address_space *mapping = file->f_mapping;
 759
 
 
 
 760	if (mapping_needs_writeback(mapping)) {
 761		err = __filemap_fdatawrite_range(mapping, lstart, lend,
 762						 WB_SYNC_ALL);
 763		/* See comment of filemap_write_and_wait() */
 764		if (err != -EIO)
 765			__filemap_fdatawait_range(mapping, lstart, lend);
 766	}
 767	err2 = file_check_and_advance_wb_err(file);
 768	if (!err)
 769		err = err2;
 770	return err;
 771}
 772EXPORT_SYMBOL(file_write_and_wait_range);
 773
 774/**
 775 * replace_page_cache_page - replace a pagecache page with a new one
 776 * @old:	page to be replaced
 777 * @new:	page to replace with
 778 * @gfp_mask:	allocation mode
 779 *
 780 * This function replaces a page in the pagecache with a new one.  On
 781 * success it acquires the pagecache reference for the new page and
 782 * drops it for the old page.  Both the old and new pages must be
 783 * locked.  This function does not add the new page to the LRU, the
 784 * caller must do that.
 785 *
 786 * The remove + add is atomic.  This function cannot fail.
 787 *
 788 * Return: %0
 789 */
 790int replace_page_cache_page(struct page *old, struct page *new, gfp_t gfp_mask)
 791{
 792	struct address_space *mapping = old->mapping;
 793	void (*freepage)(struct page *) = mapping->a_ops->freepage;
 794	pgoff_t offset = old->index;
 795	XA_STATE(xas, &mapping->i_pages, offset);
 796	unsigned long flags;
 797
 798	VM_BUG_ON_PAGE(!PageLocked(old), old);
 799	VM_BUG_ON_PAGE(!PageLocked(new), new);
 800	VM_BUG_ON_PAGE(new->mapping, new);
 801
 802	get_page(new);
 803	new->mapping = mapping;
 804	new->index = offset;
 805
 806	mem_cgroup_migrate(old, new);
 807
 808	xas_lock_irqsave(&xas, flags);
 809	xas_store(&xas, new);
 810
 811	old->mapping = NULL;
 812	/* hugetlb pages do not participate in page cache accounting. */
 813	if (!PageHuge(old))
 814		__dec_lruvec_page_state(old, NR_FILE_PAGES);
 815	if (!PageHuge(new))
 816		__inc_lruvec_page_state(new, NR_FILE_PAGES);
 817	if (PageSwapBacked(old))
 818		__dec_lruvec_page_state(old, NR_SHMEM);
 819	if (PageSwapBacked(new))
 820		__inc_lruvec_page_state(new, NR_SHMEM);
 821	xas_unlock_irqrestore(&xas, flags);
 822	if (freepage)
 823		freepage(old);
 824	put_page(old);
 825
 826	return 0;
 827}
 828EXPORT_SYMBOL_GPL(replace_page_cache_page);
 829
 830static int __add_to_page_cache_locked(struct page *page,
 831				      struct address_space *mapping,
 832				      pgoff_t offset, gfp_t gfp_mask,
 833				      void **shadowp)
 834{
 835	XA_STATE(xas, &mapping->i_pages, offset);
 836	int huge = PageHuge(page);
 837	int error;
 838	void *old;
 839
 840	VM_BUG_ON_PAGE(!PageLocked(page), page);
 841	VM_BUG_ON_PAGE(PageSwapBacked(page), page);
 842	mapping_set_update(&xas, mapping);
 843
 844	get_page(page);
 845	page->mapping = mapping;
 846	page->index = offset;
 847
 848	if (!huge) {
 849		error = mem_cgroup_charge(page, current->mm, gfp_mask);
 850		if (error)
 851			goto error;
 
 852	}
 853
 
 
 
 
 
 
 
 
 
 854	do {
 
 
 
 
 
 
 855		xas_lock_irq(&xas);
 856		old = xas_load(&xas);
 857		if (old && !xa_is_value(old))
 858			xas_set_err(&xas, -EEXIST);
 859		xas_store(&xas, page);
 860		if (xas_error(&xas))
 861			goto unlock;
 
 862
 863		if (xa_is_value(old)) {
 864			mapping->nrexceptional--;
 865			if (shadowp)
 866				*shadowp = old;
 
 
 
 
 
 
 
 
 867		}
 868		mapping->nrpages++;
 
 
 
 
 
 869
 870		/* hugetlb pages do not participate in page cache accounting */
 871		if (!huge)
 872			__inc_lruvec_page_state(page, NR_FILE_PAGES);
 
 
 
 
 873unlock:
 874		xas_unlock_irq(&xas);
 875	} while (xas_nomem(&xas, gfp_mask & GFP_RECLAIM_MASK));
 876
 877	if (xas_error(&xas)) {
 878		error = xas_error(&xas);
 879		goto error;
 880	}
 881
 882	trace_mm_filemap_add_to_page_cache(page);
 883	return 0;
 884error:
 885	page->mapping = NULL;
 
 
 886	/* Leave page->index set: truncation relies upon it */
 887	put_page(page);
 888	return error;
 889}
 890ALLOW_ERROR_INJECTION(__add_to_page_cache_locked, ERRNO);
 891
 892/**
 893 * add_to_page_cache_locked - add a locked page to the pagecache
 894 * @page:	page to add
 895 * @mapping:	the page's address_space
 896 * @offset:	page index
 897 * @gfp_mask:	page allocation mode
 898 *
 899 * This function is used to add a page to the pagecache. It must be locked.
 900 * This function does not add the page to the LRU.  The caller must do that.
 901 *
 902 * Return: %0 on success, negative error code otherwise.
 903 */
 904int add_to_page_cache_locked(struct page *page, struct address_space *mapping,
 905		pgoff_t offset, gfp_t gfp_mask)
 906{
 907	return __add_to_page_cache_locked(page, mapping, offset,
 908					  gfp_mask, NULL);
 909}
 910EXPORT_SYMBOL(add_to_page_cache_locked);
 911
 912int add_to_page_cache_lru(struct page *page, struct address_space *mapping,
 913				pgoff_t offset, gfp_t gfp_mask)
 914{
 915	void *shadow = NULL;
 916	int ret;
 917
 918	__SetPageLocked(page);
 919	ret = __add_to_page_cache_locked(page, mapping, offset,
 920					 gfp_mask, &shadow);
 921	if (unlikely(ret))
 922		__ClearPageLocked(page);
 923	else {
 924		/*
 925		 * The page might have been evicted from cache only
 926		 * recently, in which case it should be activated like
 927		 * any other repeatedly accessed page.
 928		 * The exception is pages getting rewritten; evicting other
 929		 * data from the working set, only to cache data that will
 930		 * get overwritten with something else, is a waste of memory.
 931		 */
 932		WARN_ON_ONCE(PageActive(page));
 933		if (!(gfp_mask & __GFP_WRITE) && shadow)
 934			workingset_refault(page, shadow);
 935		lru_cache_add(page);
 936	}
 937	return ret;
 938}
 939EXPORT_SYMBOL_GPL(add_to_page_cache_lru);
 940
 941#ifdef CONFIG_NUMA
 942struct page *__page_cache_alloc(gfp_t gfp)
 943{
 944	int n;
 945	struct page *page;
 946
 947	if (cpuset_do_page_mem_spread()) {
 948		unsigned int cpuset_mems_cookie;
 949		do {
 950			cpuset_mems_cookie = read_mems_allowed_begin();
 951			n = cpuset_mem_spread_node();
 952			page = __alloc_pages_node(n, gfp, 0);
 953		} while (!page && read_mems_allowed_retry(cpuset_mems_cookie));
 954
 955		return page;
 956	}
 957	return alloc_pages(gfp, 0);
 958}
 959EXPORT_SYMBOL(__page_cache_alloc);
 960#endif
 961
 962/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 963 * In order to wait for pages to become available there must be
 964 * waitqueues associated with pages. By using a hash table of
 965 * waitqueues where the bucket discipline is to maintain all
 966 * waiters on the same queue and wake all when any of the pages
 967 * become available, and for the woken contexts to check to be
 968 * sure the appropriate page became available, this saves space
 969 * at a cost of "thundering herd" phenomena during rare hash
 970 * collisions.
 971 */
 972#define PAGE_WAIT_TABLE_BITS 8
 973#define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
 974static wait_queue_head_t page_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
 975
 976static wait_queue_head_t *page_waitqueue(struct page *page)
 977{
 978	return &page_wait_table[hash_ptr(page, PAGE_WAIT_TABLE_BITS)];
 979}
 980
 981void __init pagecache_init(void)
 982{
 983	int i;
 984
 985	for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
 986		init_waitqueue_head(&page_wait_table[i]);
 987
 988	page_writeback_init();
 989}
 990
 991/*
 992 * The page wait code treats the "wait->flags" somewhat unusually, because
 993 * we have multiple different kinds of waits, not just the usual "exclusive"
 994 * one.
 995 *
 996 * We have:
 997 *
 998 *  (a) no special bits set:
 999 *
1000 *	We're just waiting for the bit to be released, and when a waker
1001 *	calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1002 *	and remove it from the wait queue.
1003 *
1004 *	Simple and straightforward.
1005 *
1006 *  (b) WQ_FLAG_EXCLUSIVE:
1007 *
1008 *	The waiter is waiting to get the lock, and only one waiter should
1009 *	be woken up to avoid any thundering herd behavior. We'll set the
1010 *	WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1011 *
1012 *	This is the traditional exclusive wait.
1013 *
1014 *  (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1015 *
1016 *	The waiter is waiting to get the bit, and additionally wants the
1017 *	lock to be transferred to it for fair lock behavior. If the lock
1018 *	cannot be taken, we stop walking the wait queue without waking
1019 *	the waiter.
1020 *
1021 *	This is the "fair lock handoff" case, and in addition to setting
1022 *	WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1023 *	that it now has the lock.
1024 */
1025static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1026{
1027	unsigned int flags;
1028	struct wait_page_key *key = arg;
1029	struct wait_page_queue *wait_page
1030		= container_of(wait, struct wait_page_queue, wait);
1031
1032	if (!wake_page_match(wait_page, key))
1033		return 0;
1034
1035	/*
1036	 * If it's a lock handoff wait, we get the bit for it, and
1037	 * stop walking (and do not wake it up) if we can't.
1038	 */
1039	flags = wait->flags;
1040	if (flags & WQ_FLAG_EXCLUSIVE) {
1041		if (test_bit(key->bit_nr, &key->page->flags))
1042			return -1;
1043		if (flags & WQ_FLAG_CUSTOM) {
1044			if (test_and_set_bit(key->bit_nr, &key->page->flags))
1045				return -1;
1046			flags |= WQ_FLAG_DONE;
1047		}
1048	}
1049
1050	/*
1051	 * We are holding the wait-queue lock, but the waiter that
1052	 * is waiting for this will be checking the flags without
1053	 * any locking.
1054	 *
1055	 * So update the flags atomically, and wake up the waiter
1056	 * afterwards to avoid any races. This store-release pairs
1057	 * with the load-acquire in wait_on_page_bit_common().
1058	 */
1059	smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1060	wake_up_state(wait->private, mode);
1061
1062	/*
1063	 * Ok, we have successfully done what we're waiting for,
1064	 * and we can unconditionally remove the wait entry.
1065	 *
1066	 * Note that this pairs with the "finish_wait()" in the
1067	 * waiter, and has to be the absolute last thing we do.
1068	 * After this list_del_init(&wait->entry) the wait entry
1069	 * might be de-allocated and the process might even have
1070	 * exited.
1071	 */
1072	list_del_init_careful(&wait->entry);
1073	return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1074}
1075
1076static void wake_up_page_bit(struct page *page, int bit_nr)
1077{
1078	wait_queue_head_t *q = page_waitqueue(page);
1079	struct wait_page_key key;
1080	unsigned long flags;
1081	wait_queue_entry_t bookmark;
1082
1083	key.page = page;
1084	key.bit_nr = bit_nr;
1085	key.page_match = 0;
1086
1087	bookmark.flags = 0;
1088	bookmark.private = NULL;
1089	bookmark.func = NULL;
1090	INIT_LIST_HEAD(&bookmark.entry);
1091
1092	spin_lock_irqsave(&q->lock, flags);
1093	__wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1094
1095	while (bookmark.flags & WQ_FLAG_BOOKMARK) {
1096		/*
1097		 * Take a breather from holding the lock,
1098		 * allow pages that finish wake up asynchronously
1099		 * to acquire the lock and remove themselves
1100		 * from wait queue
1101		 */
1102		spin_unlock_irqrestore(&q->lock, flags);
1103		cpu_relax();
1104		spin_lock_irqsave(&q->lock, flags);
1105		__wake_up_locked_key_bookmark(q, TASK_NORMAL, &key, &bookmark);
1106	}
1107
1108	/*
1109	 * It is possible for other pages to have collided on the waitqueue
1110	 * hash, so in that case check for a page match. That prevents a long-
1111	 * term waiter
1112	 *
1113	 * It is still possible to miss a case here, when we woke page waiters
1114	 * and removed them from the waitqueue, but there are still other
1115	 * page waiters.
1116	 */
1117	if (!waitqueue_active(q) || !key.page_match) {
1118		ClearPageWaiters(page);
1119		/*
1120		 * It's possible to miss clearing Waiters here, when we woke
1121		 * our page waiters, but the hashed waitqueue has waiters for
1122		 * other pages on it.
1123		 *
1124		 * That's okay, it's a rare case. The next waker will clear it.
1125		 */
1126	}
1127	spin_unlock_irqrestore(&q->lock, flags);
1128}
1129
1130static void wake_up_page(struct page *page, int bit)
1131{
1132	if (!PageWaiters(page))
1133		return;
1134	wake_up_page_bit(page, bit);
1135}
1136
1137/*
1138 * A choice of three behaviors for wait_on_page_bit_common():
1139 */
1140enum behavior {
1141	EXCLUSIVE,	/* Hold ref to page and take the bit when woken, like
1142			 * __lock_page() waiting on then setting PG_locked.
1143			 */
1144	SHARED,		/* Hold ref to page and check the bit when woken, like
1145			 * wait_on_page_writeback() waiting on PG_writeback.
1146			 */
1147	DROP,		/* Drop ref to page before wait, no check when woken,
1148			 * like put_and_wait_on_page_locked() on PG_locked.
1149			 */
1150};
1151
1152/*
1153 * Attempt to check (or get) the page bit, and mark us done
1154 * if successful.
1155 */
1156static inline bool trylock_page_bit_common(struct page *page, int bit_nr,
1157					struct wait_queue_entry *wait)
1158{
1159	if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1160		if (test_and_set_bit(bit_nr, &page->flags))
1161			return false;
1162	} else if (test_bit(bit_nr, &page->flags))
1163		return false;
1164
1165	wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1166	return true;
1167}
1168
1169/* How many times do we accept lock stealing from under a waiter? */
1170int sysctl_page_lock_unfairness = 5;
1171
1172static inline int wait_on_page_bit_common(wait_queue_head_t *q,
1173	struct page *page, int bit_nr, int state, enum behavior behavior)
1174{
 
1175	int unfairness = sysctl_page_lock_unfairness;
1176	struct wait_page_queue wait_page;
1177	wait_queue_entry_t *wait = &wait_page.wait;
1178	bool thrashing = false;
1179	bool delayacct = false;
1180	unsigned long pflags;
 
1181
1182	if (bit_nr == PG_locked &&
1183	    !PageUptodate(page) && PageWorkingset(page)) {
1184		if (!PageSwapBacked(page)) {
1185			delayacct_thrashing_start();
1186			delayacct = true;
1187		}
1188		psi_memstall_enter(&pflags);
1189		thrashing = true;
1190	}
1191
1192	init_wait(wait);
1193	wait->func = wake_page_function;
1194	wait_page.page = page;
1195	wait_page.bit_nr = bit_nr;
1196
1197repeat:
1198	wait->flags = 0;
1199	if (behavior == EXCLUSIVE) {
1200		wait->flags = WQ_FLAG_EXCLUSIVE;
1201		if (--unfairness < 0)
1202			wait->flags |= WQ_FLAG_CUSTOM;
1203	}
1204
1205	/*
1206	 * Do one last check whether we can get the
1207	 * page bit synchronously.
1208	 *
1209	 * Do the SetPageWaiters() marking before that
1210	 * to let any waker we _just_ missed know they
1211	 * need to wake us up (otherwise they'll never
1212	 * even go to the slow case that looks at the
1213	 * page queue), and add ourselves to the wait
1214	 * queue if we need to sleep.
1215	 *
1216	 * This part needs to be done under the queue
1217	 * lock to avoid races.
1218	 */
1219	spin_lock_irq(&q->lock);
1220	SetPageWaiters(page);
1221	if (!trylock_page_bit_common(page, bit_nr, wait))
1222		__add_wait_queue_entry_tail(q, wait);
1223	spin_unlock_irq(&q->lock);
1224
1225	/*
1226	 * From now on, all the logic will be based on
1227	 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1228	 * see whether the page bit testing has already
1229	 * been done by the wake function.
1230	 *
1231	 * We can drop our reference to the page.
1232	 */
1233	if (behavior == DROP)
1234		put_page(page);
1235
1236	/*
1237	 * Note that until the "finish_wait()", or until
1238	 * we see the WQ_FLAG_WOKEN flag, we need to
1239	 * be very careful with the 'wait->flags', because
1240	 * we may race with a waker that sets them.
1241	 */
1242	for (;;) {
1243		unsigned int flags;
1244
1245		set_current_state(state);
1246
1247		/* Loop until we've been woken or interrupted */
1248		flags = smp_load_acquire(&wait->flags);
1249		if (!(flags & WQ_FLAG_WOKEN)) {
1250			if (signal_pending_state(state, current))
1251				break;
1252
1253			io_schedule();
1254			continue;
1255		}
1256
1257		/* If we were non-exclusive, we're done */
1258		if (behavior != EXCLUSIVE)
1259			break;
1260
1261		/* If the waker got the lock for us, we're done */
1262		if (flags & WQ_FLAG_DONE)
1263			break;
1264
1265		/*
1266		 * Otherwise, if we're getting the lock, we need to
1267		 * try to get it ourselves.
1268		 *
1269		 * And if that fails, we'll have to retry this all.
1270		 */
1271		if (unlikely(test_and_set_bit(bit_nr, &page->flags)))
1272			goto repeat;
1273
1274		wait->flags |= WQ_FLAG_DONE;
1275		break;
1276	}
1277
1278	/*
1279	 * If a signal happened, this 'finish_wait()' may remove the last
1280	 * waiter from the wait-queues, but the PageWaiters bit will remain
1281	 * set. That's ok. The next wakeup will take care of it, and trying
1282	 * to do it here would be difficult and prone to races.
1283	 */
1284	finish_wait(q, wait);
1285
1286	if (thrashing) {
1287		if (delayacct)
1288			delayacct_thrashing_end();
1289		psi_memstall_leave(&pflags);
1290	}
1291
1292	/*
1293	 * NOTE! The wait->flags weren't stable until we've done the
1294	 * 'finish_wait()', and we could have exited the loop above due
1295	 * to a signal, and had a wakeup event happen after the signal
1296	 * test but before the 'finish_wait()'.
1297	 *
1298	 * So only after the finish_wait() can we reliably determine
1299	 * if we got woken up or not, so we can now figure out the final
1300	 * return value based on that state without races.
1301	 *
1302	 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1303	 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1304	 */
1305	if (behavior == EXCLUSIVE)
1306		return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1307
1308	return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1309}
1310
1311void wait_on_page_bit(struct page *page, int bit_nr)
1312{
1313	wait_queue_head_t *q = page_waitqueue(page);
1314	wait_on_page_bit_common(q, page, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
1315}
1316EXPORT_SYMBOL(wait_on_page_bit);
1317
1318int wait_on_page_bit_killable(struct page *page, int bit_nr)
 
 
 
 
 
 
 
 
 
 
 
1319{
1320	wait_queue_head_t *q = page_waitqueue(page);
1321	return wait_on_page_bit_common(q, page, bit_nr, TASK_KILLABLE, SHARED);
1322}
1323EXPORT_SYMBOL(wait_on_page_bit_killable);
 
 
 
1324
1325static int __wait_on_page_locked_async(struct page *page,
1326				       struct wait_page_queue *wait, bool set)
1327{
1328	struct wait_queue_head *q = page_waitqueue(page);
1329	int ret = 0;
 
1330
1331	wait->page = page;
1332	wait->bit_nr = PG_locked;
 
 
 
1333
1334	spin_lock_irq(&q->lock);
1335	__add_wait_queue_entry_tail(q, &wait->wait);
1336	SetPageWaiters(page);
1337	if (set)
1338		ret = !trylock_page(page);
1339	else
1340		ret = PageLocked(page);
1341	/*
1342	 * If we were succesful now, we know we're still on the
1343	 * waitqueue as we're still under the lock. This means it's
1344	 * safe to remove and return success, we know the callback
1345	 * isn't going to trigger.
1346	 */
1347	if (!ret)
1348		__remove_wait_queue(q, &wait->wait);
1349	else
1350		ret = -EIOCBQUEUED;
1351	spin_unlock_irq(&q->lock);
1352	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1353}
 
1354
1355static int wait_on_page_locked_async(struct page *page,
1356				     struct wait_page_queue *wait)
1357{
1358	if (!PageLocked(page))
1359		return 0;
1360	return __wait_on_page_locked_async(compound_head(page), wait, false);
1361}
 
 
 
 
 
 
 
1362
1363/**
1364 * put_and_wait_on_page_locked - Drop a reference and wait for it to be unlocked
1365 * @page: The page to wait for.
 
1366 *
1367 * The caller should hold a reference on @page.  They expect the page to
1368 * become unlocked relatively soon, but do not wish to hold up migration
1369 * (for example) by holding the reference while waiting for the page to
1370 * come unlocked.  After this function returns, the caller should not
1371 * dereference @page.
 
 
1372 */
1373void put_and_wait_on_page_locked(struct page *page)
1374{
1375	wait_queue_head_t *q;
1376
1377	page = compound_head(page);
1378	q = page_waitqueue(page);
1379	wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE, DROP);
1380}
1381
1382/**
1383 * add_page_wait_queue - Add an arbitrary waiter to a page's wait queue
1384 * @page: Page defining the wait queue of interest
1385 * @waiter: Waiter to add to the queue
1386 *
1387 * Add an arbitrary @waiter to the wait queue for the nominated @page.
1388 */
1389void add_page_wait_queue(struct page *page, wait_queue_entry_t *waiter)
1390{
1391	wait_queue_head_t *q = page_waitqueue(page);
1392	unsigned long flags;
1393
1394	spin_lock_irqsave(&q->lock, flags);
1395	__add_wait_queue_entry_tail(q, waiter);
1396	SetPageWaiters(page);
1397	spin_unlock_irqrestore(&q->lock, flags);
1398}
1399EXPORT_SYMBOL_GPL(add_page_wait_queue);
1400
1401#ifndef clear_bit_unlock_is_negative_byte
1402
1403/*
1404 * PG_waiters is the high bit in the same byte as PG_lock.
 
1405 *
1406 * On x86 (and on many other architectures), we can clear PG_lock and
1407 * test the sign bit at the same time. But if the architecture does
1408 * not support that special operation, we just do this all by hand
1409 * instead.
1410 *
1411 * The read of PG_waiters has to be after (or concurrently with) PG_locked
1412 * being cleared, but a memory barrier should be unnecessary since it is
1413 * in the same byte as PG_locked.
1414 */
1415static inline bool clear_bit_unlock_is_negative_byte(long nr, volatile void *mem)
1416{
1417	clear_bit_unlock(nr, mem);
1418	/* smp_mb__after_atomic(); */
1419	return test_bit(PG_waiters, mem);
 
 
 
1420}
 
1421
1422#endif
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1423
1424/**
1425 * unlock_page - unlock a locked page
1426 * @page: the page
1427 *
1428 * Unlocks the page and wakes up sleepers in ___wait_on_page_locked().
1429 * Also wakes sleepers in wait_on_page_writeback() because the wakeup
1430 * mechanism between PageLocked pages and PageWriteback pages is shared.
1431 * But that's OK - sleepers in wait_on_page_writeback() just go back to sleep.
1432 *
1433 * Note that this depends on PG_waiters being the sign bit in the byte
1434 * that contains PG_locked - thus the BUILD_BUG_ON(). That allows us to
1435 * clear the PG_locked bit and test PG_waiters at the same time fairly
1436 * portably (architectures that do LL/SC can test any bit, while x86 can
1437 * test the sign bit).
1438 */
1439void unlock_page(struct page *page)
1440{
1441	BUILD_BUG_ON(PG_waiters != 7);
1442	page = compound_head(page);
1443	VM_BUG_ON_PAGE(!PageLocked(page), page);
1444	if (clear_bit_unlock_is_negative_byte(PG_locked, &page->flags))
1445		wake_up_page_bit(page, PG_locked);
1446}
1447EXPORT_SYMBOL(unlock_page);
1448
1449/**
1450 * end_page_writeback - end writeback against a page
1451 * @page: the page
 
 
1452 */
1453void end_page_writeback(struct page *page)
1454{
1455	/*
1456	 * TestClearPageReclaim could be used here but it is an atomic
1457	 * operation and overkill in this particular case. Failing to
1458	 * shuffle a page marked for immediate reclaim is too mild to
1459	 * justify taking an atomic operation penalty at the end of
1460	 * ever page writeback.
1461	 */
1462	if (PageReclaim(page)) {
1463		ClearPageReclaim(page);
1464		rotate_reclaimable_page(page);
1465	}
 
 
 
 
 
 
 
 
1466
1467	if (!test_clear_page_writeback(page))
1468		BUG();
 
 
 
1469
1470	smp_mb__after_atomic();
1471	wake_up_page(page, PG_writeback);
1472}
1473EXPORT_SYMBOL(end_page_writeback);
1474
1475/*
1476 * After completing I/O on a page, call this routine to update the page
1477 * flags appropriately
 
 
 
 
1478 */
1479void page_endio(struct page *page, bool is_write, int err)
1480{
1481	if (!is_write) {
1482		if (!err) {
1483			SetPageUptodate(page);
1484		} else {
1485			ClearPageUptodate(page);
1486			SetPageError(page);
1487		}
1488		unlock_page(page);
1489	} else {
1490		if (err) {
1491			struct address_space *mapping;
1492
1493			SetPageError(page);
1494			mapping = page_mapping(page);
1495			if (mapping)
1496				mapping_set_error(mapping, err);
1497		}
1498		end_page_writeback(page);
 
 
 
 
1499	}
 
 
 
 
 
 
 
 
 
 
 
 
1500}
1501EXPORT_SYMBOL_GPL(page_endio);
1502
1503/**
1504 * __lock_page - get a lock on the page, assuming we need to sleep to get it
1505 * @__page: the page to lock
1506 */
1507void __lock_page(struct page *__page)
1508{
1509	struct page *page = compound_head(__page);
1510	wait_queue_head_t *q = page_waitqueue(page);
1511	wait_on_page_bit_common(q, page, PG_locked, TASK_UNINTERRUPTIBLE,
1512				EXCLUSIVE);
1513}
1514EXPORT_SYMBOL(__lock_page);
1515
1516int __lock_page_killable(struct page *__page)
1517{
1518	struct page *page = compound_head(__page);
1519	wait_queue_head_t *q = page_waitqueue(page);
1520	return wait_on_page_bit_common(q, page, PG_locked, TASK_KILLABLE,
1521					EXCLUSIVE);
1522}
1523EXPORT_SYMBOL_GPL(__lock_page_killable);
1524
1525int __lock_page_async(struct page *page, struct wait_page_queue *wait)
1526{
1527	return __wait_on_page_locked_async(page, wait, true);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1528}
1529
1530/*
1531 * Return values:
1532 * 1 - page is locked; mmap_lock is still held.
1533 * 0 - page is not locked.
1534 *     mmap_lock has been released (mmap_read_unlock(), unless flags had both
1535 *     FAULT_FLAG_ALLOW_RETRY and FAULT_FLAG_RETRY_NOWAIT set, in
1536 *     which case mmap_lock is still held.
1537 *
1538 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 1
1539 * with the page locked and the mmap_lock unperturbed.
1540 */
1541int __lock_page_or_retry(struct page *page, struct mm_struct *mm,
1542			 unsigned int flags)
1543{
 
 
1544	if (fault_flag_allow_retry_first(flags)) {
1545		/*
1546		 * CAUTION! In this case, mmap_lock is not released
1547		 * even though return 0.
1548		 */
1549		if (flags & FAULT_FLAG_RETRY_NOWAIT)
1550			return 0;
1551
1552		mmap_read_unlock(mm);
1553		if (flags & FAULT_FLAG_KILLABLE)
1554			wait_on_page_locked_killable(page);
1555		else
1556			wait_on_page_locked(page);
1557		return 0;
1558	} else {
1559		if (flags & FAULT_FLAG_KILLABLE) {
1560			int ret;
1561
1562			ret = __lock_page_killable(page);
1563			if (ret) {
1564				mmap_read_unlock(mm);
1565				return 0;
1566			}
1567		} else
1568			__lock_page(page);
1569		return 1;
1570	}
 
 
1571}
1572
1573/**
1574 * page_cache_next_miss() - Find the next gap in the page cache.
1575 * @mapping: Mapping.
1576 * @index: Index.
1577 * @max_scan: Maximum range to search.
1578 *
1579 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1580 * gap with the lowest index.
1581 *
1582 * This function may be called under the rcu_read_lock.  However, this will
1583 * not atomically search a snapshot of the cache at a single point in time.
1584 * For example, if a gap is created at index 5, then subsequently a gap is
1585 * created at index 10, page_cache_next_miss covering both indices may
1586 * return 10 if called under the rcu_read_lock.
1587 *
1588 * Return: The index of the gap if found, otherwise an index outside the
1589 * range specified (in which case 'return - index >= max_scan' will be true).
1590 * In the rare case of index wrap-around, 0 will be returned.
1591 */
1592pgoff_t page_cache_next_miss(struct address_space *mapping,
1593			     pgoff_t index, unsigned long max_scan)
1594{
1595	XA_STATE(xas, &mapping->i_pages, index);
1596
1597	while (max_scan--) {
1598		void *entry = xas_next(&xas);
1599		if (!entry || xa_is_value(entry))
1600			break;
1601		if (xas.xa_index == 0)
1602			break;
1603	}
1604
1605	return xas.xa_index;
1606}
1607EXPORT_SYMBOL(page_cache_next_miss);
1608
1609/**
1610 * page_cache_prev_miss() - Find the previous gap in the page cache.
1611 * @mapping: Mapping.
1612 * @index: Index.
1613 * @max_scan: Maximum range to search.
1614 *
1615 * Search the range [max(index - max_scan + 1, 0), index] for the
1616 * gap with the highest index.
1617 *
1618 * This function may be called under the rcu_read_lock.  However, this will
1619 * not atomically search a snapshot of the cache at a single point in time.
1620 * For example, if a gap is created at index 10, then subsequently a gap is
1621 * created at index 5, page_cache_prev_miss() covering both indices may
1622 * return 5 if called under the rcu_read_lock.
1623 *
1624 * Return: The index of the gap if found, otherwise an index outside the
1625 * range specified (in which case 'index - return >= max_scan' will be true).
1626 * In the rare case of wrap-around, ULONG_MAX will be returned.
1627 */
1628pgoff_t page_cache_prev_miss(struct address_space *mapping,
1629			     pgoff_t index, unsigned long max_scan)
1630{
1631	XA_STATE(xas, &mapping->i_pages, index);
1632
1633	while (max_scan--) {
1634		void *entry = xas_prev(&xas);
1635		if (!entry || xa_is_value(entry))
1636			break;
1637		if (xas.xa_index == ULONG_MAX)
1638			break;
1639	}
1640
1641	return xas.xa_index;
1642}
1643EXPORT_SYMBOL(page_cache_prev_miss);
1644
1645/**
1646 * find_get_entry - find and get a page cache entry
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1647 * @mapping: the address_space to search
1648 * @offset: the page cache index
1649 *
1650 * Looks up the page cache slot at @mapping & @offset.  If there is a
1651 * page cache page, it is returned with an increased refcount.
1652 *
1653 * If the slot holds a shadow entry of a previously evicted page, or a
1654 * swap entry from shmem/tmpfs, it is returned.
 
 
1655 *
1656 * Return: the found page or shadow entry, %NULL if nothing is found.
1657 */
1658struct page *find_get_entry(struct address_space *mapping, pgoff_t offset)
1659{
1660	XA_STATE(xas, &mapping->i_pages, offset);
1661	struct page *page;
1662
1663	rcu_read_lock();
1664repeat:
1665	xas_reset(&xas);
1666	page = xas_load(&xas);
1667	if (xas_retry(&xas, page))
1668		goto repeat;
1669	/*
1670	 * A shadow entry of a recently evicted page, or a swap entry from
1671	 * shmem/tmpfs.  Return it without attempting to raise page count.
1672	 */
1673	if (!page || xa_is_value(page))
1674		goto out;
1675
1676	if (!page_cache_get_speculative(page))
1677		goto repeat;
1678
1679	/*
1680	 * Has the page moved or been split?
1681	 * This is part of the lockless pagecache protocol. See
1682	 * include/linux/pagemap.h for details.
1683	 */
1684	if (unlikely(page != xas_reload(&xas))) {
1685		put_page(page);
1686		goto repeat;
1687	}
1688	page = find_subpage(page, offset);
1689out:
1690	rcu_read_unlock();
1691
1692	return page;
1693}
1694
1695/**
1696 * find_lock_entry - locate, pin and lock a page cache entry
1697 * @mapping: the address_space to search
1698 * @offset: the page cache index
1699 *
1700 * Looks up the page cache slot at @mapping & @offset.  If there is a
1701 * page cache page, it is returned locked and with an increased
1702 * refcount.
1703 *
1704 * If the slot holds a shadow entry of a previously evicted page, or a
1705 * swap entry from shmem/tmpfs, it is returned.
1706 *
1707 * find_lock_entry() may sleep.
1708 *
1709 * Return: the found page or shadow entry, %NULL if nothing is found.
1710 */
1711struct page *find_lock_entry(struct address_space *mapping, pgoff_t offset)
1712{
1713	struct page *page;
1714
1715repeat:
1716	page = find_get_entry(mapping, offset);
1717	if (page && !xa_is_value(page)) {
1718		lock_page(page);
1719		/* Has the page been truncated? */
1720		if (unlikely(page_mapping(page) != mapping)) {
1721			unlock_page(page);
1722			put_page(page);
1723			goto repeat;
1724		}
1725		VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
1726	}
1727	return page;
1728}
1729EXPORT_SYMBOL(find_lock_entry);
1730
1731/**
1732 * pagecache_get_page - Find and get a reference to a page.
1733 * @mapping: The address_space to search.
1734 * @index: The page index.
1735 * @fgp_flags: %FGP flags modify how the page is returned.
1736 * @gfp_mask: Memory allocation flags to use if %FGP_CREAT is specified.
1737 *
1738 * Looks up the page cache entry at @mapping & @index.
1739 *
1740 * @fgp_flags can be zero or more of these flags:
1741 *
1742 * * %FGP_ACCESSED - The page will be marked accessed.
1743 * * %FGP_LOCK - The page is returned locked.
1744 * * %FGP_CREAT - If no page is present then a new page is allocated using
1745 *   @gfp_mask and added to the page cache and the VM's LRU list.
1746 *   The page is returned locked and with an increased refcount.
1747 * * %FGP_FOR_MMAP - The caller wants to do its own locking dance if the
1748 *   page is already in cache.  If the page was allocated, unlock it before
1749 *   returning so the caller can do the same dance.
1750 * * %FGP_WRITE - The page will be written
1751 * * %FGP_NOFS - __GFP_FS will get cleared in gfp mask
1752 * * %FGP_NOWAIT - Don't get blocked by page lock
1753 *
1754 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1755 * if the %GFP flags specified for %FGP_CREAT are atomic.
1756 *
1757 * If there is a page cache page, it is returned with an increased refcount.
1758 *
1759 * Return: The found page or %NULL otherwise.
1760 */
1761struct page *pagecache_get_page(struct address_space *mapping, pgoff_t index,
1762		int fgp_flags, gfp_t gfp_mask)
1763{
1764	struct page *page;
1765
1766repeat:
1767	page = find_get_entry(mapping, index);
1768	if (xa_is_value(page))
1769		page = NULL;
1770	if (!page)
1771		goto no_page;
1772
1773	if (fgp_flags & FGP_LOCK) {
1774		if (fgp_flags & FGP_NOWAIT) {
1775			if (!trylock_page(page)) {
1776				put_page(page);
1777				return NULL;
1778			}
1779		} else {
1780			lock_page(page);
1781		}
1782
1783		/* Has the page been truncated? */
1784		if (unlikely(compound_head(page)->mapping != mapping)) {
1785			unlock_page(page);
1786			put_page(page);
1787			goto repeat;
1788		}
1789		VM_BUG_ON_PAGE(page->index != index, page);
1790	}
1791
1792	if (fgp_flags & FGP_ACCESSED)
1793		mark_page_accessed(page);
1794	else if (fgp_flags & FGP_WRITE) {
1795		/* Clear idle flag for buffer write */
1796		if (page_is_idle(page))
1797			clear_page_idle(page);
1798	}
1799
 
 
1800no_page:
1801	if (!page && (fgp_flags & FGP_CREAT)) {
 
1802		int err;
1803		if ((fgp_flags & FGP_WRITE) && mapping_cap_account_dirty(mapping))
1804			gfp_mask |= __GFP_WRITE;
1805		if (fgp_flags & FGP_NOFS)
1806			gfp_mask &= ~__GFP_FS;
1807
1808		page = __page_cache_alloc(gfp_mask);
1809		if (!page)
1810			return NULL;
1811
 
 
 
 
 
 
 
 
1812		if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1813			fgp_flags |= FGP_LOCK;
1814
1815		/* Init accessed so avoid atomic mark_page_accessed later */
1816		if (fgp_flags & FGP_ACCESSED)
1817			__SetPageReferenced(page);
 
 
 
 
1818
1819		err = add_to_page_cache_lru(page, mapping, index, gfp_mask);
1820		if (unlikely(err)) {
1821			put_page(page);
1822			page = NULL;
1823			if (err == -EEXIST)
1824				goto repeat;
1825		}
1826
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1827		/*
1828		 * add_to_page_cache_lru locks the page, and for mmap we expect
1829		 * an unlocked page.
1830		 */
1831		if (page && (fgp_flags & FGP_FOR_MMAP))
1832			unlock_page(page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1833	}
1834
1835	return page;
 
 
 
1836}
1837EXPORT_SYMBOL(pagecache_get_page);
1838
1839/**
1840 * find_get_entries - gang pagecache lookup
1841 * @mapping:	The address_space to search
1842 * @start:	The starting page cache index
1843 * @nr_entries:	The maximum number of entries
1844 * @entries:	Where the resulting entries are placed
1845 * @indices:	The cache indices corresponding to the entries in @entries
1846 *
1847 * find_get_entries() will search for and return a group of up to
1848 * @nr_entries entries in the mapping.  The entries are placed at
1849 * @entries.  find_get_entries() takes a reference against any actual
1850 * pages it returns.
1851 *
1852 * The search returns a group of mapping-contiguous page cache entries
1853 * with ascending indexes.  There may be holes in the indices due to
1854 * not-present pages.
1855 *
1856 * Any shadow entries of evicted pages, or swap entries from
1857 * shmem/tmpfs, are included in the returned array.
1858 *
1859 * If it finds a Transparent Huge Page, head or tail, find_get_entries()
1860 * stops at that page: the caller is likely to have a better way to handle
1861 * the compound page as a whole, and then skip its extent, than repeatedly
1862 * calling find_get_entries() to return all its tails.
1863 *
1864 * Return: the number of pages and shadow entries which were found.
1865 */
1866unsigned find_get_entries(struct address_space *mapping,
1867			  pgoff_t start, unsigned int nr_entries,
1868			  struct page **entries, pgoff_t *indices)
1869{
1870	XA_STATE(xas, &mapping->i_pages, start);
1871	struct page *page;
1872	unsigned int ret = 0;
1873
1874	if (!nr_entries)
1875		return 0;
1876
1877	rcu_read_lock();
1878	xas_for_each(&xas, page, ULONG_MAX) {
1879		if (xas_retry(&xas, page))
1880			continue;
1881		/*
1882		 * A shadow entry of a recently evicted page, a swap
1883		 * entry from shmem/tmpfs or a DAX entry.  Return it
1884		 * without attempting to raise page count.
1885		 */
1886		if (xa_is_value(page))
1887			goto export;
1888
1889		if (!page_cache_get_speculative(page))
1890			goto retry;
1891
1892		/* Has the page moved or been split? */
1893		if (unlikely(page != xas_reload(&xas)))
1894			goto put_page;
1895
1896		/*
1897		 * Terminate early on finding a THP, to allow the caller to
1898		 * handle it all at once; but continue if this is hugetlbfs.
1899		 */
1900		if (PageTransHuge(page) && !PageHuge(page)) {
1901			page = find_subpage(page, xas.xa_index);
1902			nr_entries = ret + 1;
1903		}
1904export:
1905		indices[ret] = xas.xa_index;
1906		entries[ret] = page;
1907		if (++ret == nr_entries)
1908			break;
1909		continue;
1910put_page:
1911		put_page(page);
1912retry:
1913		xas_reset(&xas);
1914	}
1915	rcu_read_unlock();
1916	return ret;
 
 
 
 
 
 
 
 
 
 
1917}
1918
1919/**
1920 * find_get_pages_range - gang pagecache lookup
1921 * @mapping:	The address_space to search
1922 * @start:	The starting page index
1923 * @end:	The final page index (inclusive)
1924 * @nr_pages:	The maximum number of pages
1925 * @pages:	Where the resulting pages are placed
 
 
 
 
 
 
 
 
 
 
1926 *
1927 * find_get_pages_range() will search for and return a group of up to @nr_pages
1928 * pages in the mapping starting at index @start and up to index @end
1929 * (inclusive).  The pages are placed at @pages.  find_get_pages_range() takes
1930 * a reference against the returned pages.
1931 *
1932 * The search returns a group of mapping-contiguous pages with ascending
1933 * indexes.  There may be holes in the indices due to not-present pages.
1934 * We also update @start to index the next page for the traversal.
1935 *
1936 * Return: the number of pages which were found. If this number is
1937 * smaller than @nr_pages, the end of specified range has been
1938 * reached.
1939 */
1940unsigned find_get_pages_range(struct address_space *mapping, pgoff_t *start,
1941			      pgoff_t end, unsigned int nr_pages,
1942			      struct page **pages)
1943{
1944	XA_STATE(xas, &mapping->i_pages, *start);
1945	struct page *page;
1946	unsigned ret = 0;
1947
1948	if (unlikely(!nr_pages))
1949		return 0;
1950
1951	rcu_read_lock();
1952	xas_for_each(&xas, page, end) {
1953		if (xas_retry(&xas, page))
1954			continue;
1955		/* Skip over shadow, swap and DAX entries */
1956		if (xa_is_value(page))
1957			continue;
1958
1959		if (!page_cache_get_speculative(page))
1960			goto retry;
1961
1962		/* Has the page moved or been split? */
1963		if (unlikely(page != xas_reload(&xas)))
1964			goto put_page;
1965
1966		pages[ret] = find_subpage(page, xas.xa_index);
1967		if (++ret == nr_pages) {
1968			*start = xas.xa_index + 1;
1969			goto out;
1970		}
 
 
 
1971		continue;
1972put_page:
1973		put_page(page);
1974retry:
1975		xas_reset(&xas);
1976	}
1977
1978	/*
1979	 * We come here when there is no page beyond @end. We take care to not
1980	 * overflow the index @start as it confuses some of the callers. This
1981	 * breaks the iteration when there is a page at index -1 but that is
1982	 * already broken anyway.
1983	 */
1984	if (end == (pgoff_t)-1)
1985		*start = (pgoff_t)-1;
1986	else
1987		*start = end + 1;
1988out:
1989	rcu_read_unlock();
1990
1991	return ret;
 
 
 
 
 
 
 
 
 
1992}
1993
1994/**
1995 * find_get_pages_contig - gang contiguous pagecache lookup
1996 * @mapping:	The address_space to search
1997 * @index:	The starting page index
1998 * @nr_pages:	The maximum number of pages
1999 * @pages:	Where the resulting pages are placed
2000 *
2001 * find_get_pages_contig() works exactly like find_get_pages(), except
2002 * that the returned number of pages are guaranteed to be contiguous.
 
2003 *
2004 * Return: the number of pages which were found.
 
2005 */
2006unsigned find_get_pages_contig(struct address_space *mapping, pgoff_t index,
2007			       unsigned int nr_pages, struct page **pages)
2008{
2009	XA_STATE(xas, &mapping->i_pages, index);
2010	struct page *page;
2011	unsigned int ret = 0;
2012
2013	if (unlikely(!nr_pages))
2014		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2015
2016	rcu_read_lock();
2017	for (page = xas_load(&xas); page; page = xas_next(&xas)) {
2018		if (xas_retry(&xas, page))
 
 
2019			continue;
2020		/*
2021		 * If the entry has been swapped out, we can stop looking.
2022		 * No current caller is looking for DAX entries.
2023		 */
2024		if (xa_is_value(page))
2025			break;
2026
2027		if (!page_cache_get_speculative(page))
2028			goto retry;
2029
2030		/* Has the page moved or been split? */
2031		if (unlikely(page != xas_reload(&xas)))
2032			goto put_page;
2033
2034		pages[ret] = find_subpage(page, xas.xa_index);
2035		if (++ret == nr_pages)
2036			break;
 
 
2037		continue;
2038put_page:
2039		put_page(page);
 
2040retry:
2041		xas_reset(&xas);
2042	}
 
 
 
 
 
 
 
 
 
2043	rcu_read_unlock();
2044	return ret;
2045}
2046EXPORT_SYMBOL(find_get_pages_contig);
2047
2048/**
2049 * find_get_pages_range_tag - find and return pages in given range matching @tag
2050 * @mapping:	the address_space to search
2051 * @index:	the starting page index
2052 * @end:	The final page index (inclusive)
2053 * @tag:	the tag index
2054 * @nr_pages:	the maximum number of pages
2055 * @pages:	where the resulting pages are placed
2056 *
2057 * Like find_get_pages, except we only return pages which are tagged with
2058 * @tag.   We update @index to index the next page for the traversal.
 
 
 
 
2059 *
2060 * Return: the number of pages which were found.
 
2061 */
2062unsigned find_get_pages_range_tag(struct address_space *mapping, pgoff_t *index,
2063			pgoff_t end, xa_mark_t tag, unsigned int nr_pages,
2064			struct page **pages)
2065{
2066	XA_STATE(xas, &mapping->i_pages, *index);
2067	struct page *page;
2068	unsigned ret = 0;
2069
2070	if (unlikely(!nr_pages))
2071		return 0;
2072
2073	rcu_read_lock();
2074	xas_for_each_marked(&xas, page, end, tag) {
2075		if (xas_retry(&xas, page))
2076			continue;
2077		/*
2078		 * Shadow entries should never be tagged, but this iteration
2079		 * is lockless so there is a window for page reclaim to evict
2080		 * a page we saw tagged.  Skip over it.
2081		 */
2082		if (xa_is_value(page))
2083			continue;
2084
2085		if (!page_cache_get_speculative(page))
2086			goto retry;
2087
2088		/* Has the page moved or been split? */
2089		if (unlikely(page != xas_reload(&xas)))
2090			goto put_page;
2091
2092		pages[ret] = find_subpage(page, xas.xa_index);
2093		if (++ret == nr_pages) {
2094			*index = xas.xa_index + 1;
2095			goto out;
2096		}
2097		continue;
2098put_page:
2099		put_page(page);
2100retry:
2101		xas_reset(&xas);
2102	}
2103
2104	/*
2105	 * We come here when we got to @end. We take care to not overflow the
2106	 * index @index as it confuses some of the callers. This breaks the
2107	 * iteration when there is a page at index -1 but that is already
2108	 * broken anyway.
2109	 */
2110	if (end == (pgoff_t)-1)
2111		*index = (pgoff_t)-1;
2112	else
2113		*index = end + 1;
2114out:
2115	rcu_read_unlock();
2116
2117	return ret;
2118}
2119EXPORT_SYMBOL(find_get_pages_range_tag);
2120
2121/*
2122 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2123 * a _large_ part of the i/o request. Imagine the worst scenario:
2124 *
2125 *      ---R__________________________________________B__________
2126 *         ^ reading here                             ^ bad block(assume 4k)
2127 *
2128 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2129 * => failing the whole request => read(R) => read(R+1) =>
2130 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2131 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2132 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2133 *
2134 * It is going insane. Fix it by quickly scaling down the readahead size.
2135 */
2136static void shrink_readahead_size_eio(struct file_ra_state *ra)
2137{
2138	ra->ra_pages /= 4;
2139}
2140
2141/**
2142 * generic_file_buffered_read - generic file read routine
2143 * @iocb:	the iocb to read
2144 * @iter:	data destination
2145 * @written:	already copied
2146 *
2147 * This is a generic file read routine, and uses the
2148 * mapping->a_ops->readpage() function for the actual low-level stuff.
2149 *
2150 * This is really ugly. But the goto's actually try to clarify some
2151 * of the logic when it comes to error handling etc.
2152 *
2153 * Return:
2154 * * total number of bytes copied, including those the were already @written
2155 * * negative error code if nothing was copied
 
 
2156 */
2157ssize_t generic_file_buffered_read(struct kiocb *iocb,
2158		struct iov_iter *iter, ssize_t written)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2159{
2160	struct file *filp = iocb->ki_filp;
2161	struct address_space *mapping = filp->f_mapping;
2162	struct inode *inode = mapping->host;
2163	struct file_ra_state *ra = &filp->f_ra;
2164	loff_t *ppos = &iocb->ki_pos;
2165	pgoff_t index;
2166	pgoff_t last_index;
2167	pgoff_t prev_index;
2168	unsigned long offset;      /* offset into pagecache page */
2169	unsigned int prev_offset;
2170	int error = 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2171
2172	if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2173		return 0;
2174	iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
 
 
 
2175
2176	index = *ppos >> PAGE_SHIFT;
2177	prev_index = ra->prev_pos >> PAGE_SHIFT;
2178	prev_offset = ra->prev_pos & (PAGE_SIZE-1);
2179	last_index = (*ppos + iter->count + PAGE_SIZE-1) >> PAGE_SHIFT;
2180	offset = *ppos & ~PAGE_MASK;
2181
2182	for (;;) {
2183		struct page *page;
2184		pgoff_t end_index;
2185		loff_t isize;
2186		unsigned long nr, ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2187
 
 
 
 
 
 
 
 
 
2188		cond_resched();
2189find_page:
2190		if (fatal_signal_pending(current)) {
2191			error = -EINTR;
2192			goto out;
2193		}
2194
2195		page = find_get_page(mapping, index);
2196		if (!page) {
2197			if (iocb->ki_flags & IOCB_NOIO)
2198				goto would_block;
2199			page_cache_sync_readahead(mapping,
2200					ra, filp,
2201					index, last_index - index);
2202			page = find_get_page(mapping, index);
2203			if (unlikely(page == NULL))
2204				goto no_cached_page;
2205		}
2206		if (PageReadahead(page)) {
2207			if (iocb->ki_flags & IOCB_NOIO) {
2208				put_page(page);
2209				goto out;
2210			}
2211			page_cache_async_readahead(mapping,
2212					ra, filp, page,
2213					index, last_index - index);
2214		}
2215		if (!PageUptodate(page)) {
2216			/*
2217			 * See comment in do_read_cache_page on why
2218			 * wait_on_page_locked is used to avoid unnecessarily
2219			 * serialisations and why it's safe.
2220			 */
2221			if (iocb->ki_flags & IOCB_WAITQ) {
2222				if (written) {
2223					put_page(page);
2224					goto out;
2225				}
2226				error = wait_on_page_locked_async(page,
2227								iocb->ki_waitq);
2228			} else {
2229				if (iocb->ki_flags & IOCB_NOWAIT) {
2230					put_page(page);
2231					goto would_block;
2232				}
2233				error = wait_on_page_locked_killable(page);
2234			}
2235			if (unlikely(error))
2236				goto readpage_error;
2237			if (PageUptodate(page))
2238				goto page_ok;
2239
2240			if (inode->i_blkbits == PAGE_SHIFT ||
2241					!mapping->a_ops->is_partially_uptodate)
2242				goto page_not_up_to_date;
2243			/* pipes can't handle partially uptodate pages */
2244			if (unlikely(iov_iter_is_pipe(iter)))
2245				goto page_not_up_to_date;
2246			if (!trylock_page(page))
2247				goto page_not_up_to_date;
2248			/* Did it get truncated before we got the lock? */
2249			if (!page->mapping)
2250				goto page_not_up_to_date_locked;
2251			if (!mapping->a_ops->is_partially_uptodate(page,
2252							offset, iter->count))
2253				goto page_not_up_to_date_locked;
2254			unlock_page(page);
2255		}
2256page_ok:
2257		/*
2258		 * i_size must be checked after we know the page is Uptodate.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2259		 *
2260		 * Checking i_size after the check allows us to calculate
2261		 * the correct value for "nr", which means the zero-filled
2262		 * part of the page is not copied back to userspace (unless
2263		 * another truncate extends the file - this is desired though).
2264		 */
2265
2266		isize = i_size_read(inode);
2267		end_index = (isize - 1) >> PAGE_SHIFT;
2268		if (unlikely(!isize || index > end_index)) {
2269			put_page(page);
2270			goto out;
2271		}
2272
2273		/* nr is the maximum number of bytes to copy from this page */
2274		nr = PAGE_SIZE;
2275		if (index == end_index) {
2276			nr = ((isize - 1) & ~PAGE_MASK) + 1;
2277			if (nr <= offset) {
2278				put_page(page);
2279				goto out;
2280			}
2281		}
2282		nr = nr - offset;
2283
2284		/* If users can be writing to this page using arbitrary
2285		 * virtual addresses, take care about potential aliasing
2286		 * before reading the page on the kernel side.
2287		 */
2288		if (mapping_writably_mapped(mapping))
2289			flush_dcache_page(page);
2290
2291		/*
2292		 * When a sequential read accesses a page several times,
2293		 * only mark it as accessed the first time.
2294		 */
2295		if (prev_index != index || offset != prev_offset)
2296			mark_page_accessed(page);
2297		prev_index = index;
2298
2299		/*
2300		 * Ok, we have the page, and it's up-to-date, so
2301		 * now we can copy it to user space...
2302		 */
 
 
 
 
 
 
 
 
 
 
 
2303
2304		ret = copy_page_to_iter(page, offset, nr, iter);
2305		offset += ret;
2306		index += offset >> PAGE_SHIFT;
2307		offset &= ~PAGE_MASK;
2308		prev_offset = offset;
2309
2310		put_page(page);
2311		written += ret;
2312		if (!iov_iter_count(iter))
2313			goto out;
2314		if (ret < nr) {
2315			error = -EFAULT;
2316			goto out;
2317		}
2318		continue;
2319
2320page_not_up_to_date:
2321		/* Get exclusive access to the page ... */
2322		if (iocb->ki_flags & IOCB_WAITQ)
2323			error = lock_page_async(page, iocb->ki_waitq);
2324		else
2325			error = lock_page_killable(page);
2326		if (unlikely(error))
2327			goto readpage_error;
2328
2329page_not_up_to_date_locked:
2330		/* Did it get truncated before we got the lock? */
2331		if (!page->mapping) {
2332			unlock_page(page);
2333			put_page(page);
2334			continue;
2335		}
2336
2337		/* Did somebody else fill it already? */
2338		if (PageUptodate(page)) {
2339			unlock_page(page);
2340			goto page_ok;
2341		}
2342
2343readpage:
2344		if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT)) {
2345			unlock_page(page);
2346			put_page(page);
2347			goto would_block;
2348		}
2349		/*
2350		 * A previous I/O error may have been due to temporary
2351		 * failures, eg. multipath errors.
2352		 * PG_error will be set again if readpage fails.
2353		 */
2354		ClearPageError(page);
2355		/* Start the actual read. The read will unlock the page. */
2356		error = mapping->a_ops->readpage(filp, page);
2357
2358		if (unlikely(error)) {
2359			if (error == AOP_TRUNCATED_PAGE) {
2360				put_page(page);
2361				error = 0;
2362				goto find_page;
2363			}
2364			goto readpage_error;
2365		}
2366
2367		if (!PageUptodate(page)) {
2368			if (iocb->ki_flags & IOCB_WAITQ)
2369				error = lock_page_async(page, iocb->ki_waitq);
2370			else
2371				error = lock_page_killable(page);
2372
2373			if (unlikely(error))
2374				goto readpage_error;
2375			if (!PageUptodate(page)) {
2376				if (page->mapping == NULL) {
2377					/*
2378					 * invalidate_mapping_pages got it
2379					 */
2380					unlock_page(page);
2381					put_page(page);
2382					goto find_page;
2383				}
2384				unlock_page(page);
2385				shrink_readahead_size_eio(ra);
2386				error = -EIO;
2387				goto readpage_error;
2388			}
2389			unlock_page(page);
2390		}
 
 
 
 
 
2391
2392		goto page_ok;
 
 
 
 
2393
2394readpage_error:
2395		/* UHHUH! A synchronous read error occurred. Report it */
2396		put_page(page);
2397		goto out;
 
2398
2399no_cached_page:
2400		/*
2401		 * Ok, it wasn't cached, so we need to create a new
2402		 * page..
2403		 */
2404		page = page_cache_alloc(mapping);
2405		if (!page) {
2406			error = -ENOMEM;
2407			goto out;
2408		}
2409		error = add_to_page_cache_lru(page, mapping, index,
2410				mapping_gfp_constraint(mapping, GFP_KERNEL));
2411		if (error) {
2412			put_page(page);
2413			if (error == -EEXIST) {
2414				error = 0;
2415				goto find_page;
2416			}
2417			goto out;
2418		}
2419		goto readpage;
2420	}
2421
2422would_block:
2423	error = -EAGAIN;
2424out:
2425	ra->prev_pos = prev_index;
2426	ra->prev_pos <<= PAGE_SHIFT;
2427	ra->prev_pos |= prev_offset;
2428
2429	*ppos = ((loff_t)index << PAGE_SHIFT) + offset;
2430	file_accessed(filp);
2431	return written ? written : error;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2432}
2433EXPORT_SYMBOL_GPL(generic_file_buffered_read);
2434
2435/**
2436 * generic_file_read_iter - generic filesystem read routine
2437 * @iocb:	kernel I/O control block
2438 * @iter:	destination for the data read
2439 *
2440 * This is the "read_iter()" routine for all filesystems
2441 * that can use the page cache directly.
2442 *
2443 * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2444 * be returned when no data can be read without waiting for I/O requests
2445 * to complete; it doesn't prevent readahead.
2446 *
2447 * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2448 * requests shall be made for the read or for readahead.  When no data
2449 * can be read, -EAGAIN shall be returned.  When readahead would be
2450 * triggered, a partial, possibly empty read shall be returned.
2451 *
2452 * Return:
2453 * * number of bytes copied, even for partial reads
2454 * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2455 */
2456ssize_t
2457generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2458{
2459	size_t count = iov_iter_count(iter);
2460	ssize_t retval = 0;
2461
2462	if (!count)
2463		goto out; /* skip atime */
2464
2465	if (iocb->ki_flags & IOCB_DIRECT) {
2466		struct file *file = iocb->ki_filp;
2467		struct address_space *mapping = file->f_mapping;
2468		struct inode *inode = mapping->host;
2469		loff_t size;
2470
2471		size = i_size_read(inode);
2472		if (iocb->ki_flags & IOCB_NOWAIT) {
2473			if (filemap_range_has_page(mapping, iocb->ki_pos,
2474						   iocb->ki_pos + count - 1))
2475				return -EAGAIN;
2476		} else {
2477			retval = filemap_write_and_wait_range(mapping,
2478						iocb->ki_pos,
2479					        iocb->ki_pos + count - 1);
2480			if (retval < 0)
2481				goto out;
2482		}
2483
 
 
 
2484		file_accessed(file);
2485
2486		retval = mapping->a_ops->direct_IO(iocb, iter);
2487		if (retval >= 0) {
2488			iocb->ki_pos += retval;
2489			count -= retval;
2490		}
2491		iov_iter_revert(iter, count - iov_iter_count(iter));
 
2492
2493		/*
2494		 * Btrfs can have a short DIO read if we encounter
2495		 * compressed extents, so if there was an error, or if
2496		 * we've already read everything we wanted to, or if
2497		 * there was a short read because we hit EOF, go ahead
2498		 * and return.  Otherwise fallthrough to buffered io for
2499		 * the rest of the read.  Buffered reads will not work for
2500		 * DAX files, so don't bother trying.
2501		 */
2502		if (retval < 0 || !count || iocb->ki_pos >= size ||
2503		    IS_DAX(inode))
2504			goto out;
 
2505	}
2506
2507	retval = generic_file_buffered_read(iocb, iter, retval);
2508out:
2509	return retval;
2510}
2511EXPORT_SYMBOL(generic_file_read_iter);
2512
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2513#ifdef CONFIG_MMU
2514#define MMAP_LOTSAMISS  (100)
2515/*
2516 * lock_page_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
2517 * @vmf - the vm_fault for this fault.
2518 * @page - the page to lock.
2519 * @fpin - the pointer to the file we may pin (or is already pinned).
2520 *
2521 * This works similar to lock_page_or_retry in that it can drop the mmap_lock.
2522 * It differs in that it actually returns the page locked if it returns 1 and 0
2523 * if it couldn't lock the page.  If we did have to drop the mmap_lock then fpin
2524 * will point to the pinned file and needs to be fput()'ed at a later point.
 
2525 */
2526static int lock_page_maybe_drop_mmap(struct vm_fault *vmf, struct page *page,
2527				     struct file **fpin)
2528{
2529	if (trylock_page(page))
2530		return 1;
2531
2532	/*
2533	 * NOTE! This will make us return with VM_FAULT_RETRY, but with
2534	 * the mmap_lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
2535	 * is supposed to work. We have way too many special cases..
2536	 */
2537	if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
2538		return 0;
2539
2540	*fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
2541	if (vmf->flags & FAULT_FLAG_KILLABLE) {
2542		if (__lock_page_killable(page)) {
2543			/*
2544			 * We didn't have the right flags to drop the mmap_lock,
2545			 * but all fault_handlers only check for fatal signals
2546			 * if we return VM_FAULT_RETRY, so we need to drop the
2547			 * mmap_lock here and return 0 if we don't have a fpin.
 
2548			 */
2549			if (*fpin == NULL)
2550				mmap_read_unlock(vmf->vma->vm_mm);
2551			return 0;
2552		}
2553	} else
2554		__lock_page(page);
 
2555	return 1;
2556}
2557
2558
2559/*
2560 * Synchronous readahead happens when we don't even find a page in the page
2561 * cache at all.  We don't want to perform IO under the mmap sem, so if we have
2562 * to drop the mmap sem we return the file that was pinned in order for us to do
2563 * that.  If we didn't pin a file then we return NULL.  The file that is
2564 * returned needs to be fput()'ed when we're done with it.
2565 */
2566static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
2567{
2568	struct file *file = vmf->vma->vm_file;
2569	struct file_ra_state *ra = &file->f_ra;
2570	struct address_space *mapping = file->f_mapping;
 
2571	struct file *fpin = NULL;
2572	pgoff_t offset = vmf->pgoff;
2573	unsigned int mmap_miss;
2574
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2575	/* If we don't want any read-ahead, don't bother */
2576	if (vmf->vma->vm_flags & VM_RAND_READ)
2577		return fpin;
2578	if (!ra->ra_pages)
2579		return fpin;
2580
2581	if (vmf->vma->vm_flags & VM_SEQ_READ) {
2582		fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2583		page_cache_sync_readahead(mapping, ra, file, offset,
2584					  ra->ra_pages);
2585		return fpin;
2586	}
2587
2588	/* Avoid banging the cache line if not needed */
2589	mmap_miss = READ_ONCE(ra->mmap_miss);
2590	if (mmap_miss < MMAP_LOTSAMISS * 10)
2591		WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
2592
2593	/*
2594	 * Do we miss much more than hit in this file? If so,
2595	 * stop bothering with read-ahead. It will only hurt.
2596	 */
2597	if (mmap_miss > MMAP_LOTSAMISS)
2598		return fpin;
2599
2600	/*
2601	 * mmap read-around
2602	 */
2603	fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2604	ra->start = max_t(long, 0, offset - ra->ra_pages / 2);
2605	ra->size = ra->ra_pages;
2606	ra->async_size = ra->ra_pages / 4;
2607	ra_submit(ra, mapping, file);
 
2608	return fpin;
2609}
2610
2611/*
2612 * Asynchronous readahead happens when we find the page and PG_readahead,
2613 * so we want to possibly extend the readahead further.  We return the file that
2614 * was pinned if we have to drop the mmap_lock in order to do IO.
2615 */
2616static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
2617					    struct page *page)
2618{
2619	struct file *file = vmf->vma->vm_file;
2620	struct file_ra_state *ra = &file->f_ra;
2621	struct address_space *mapping = file->f_mapping;
2622	struct file *fpin = NULL;
2623	unsigned int mmap_miss;
2624	pgoff_t offset = vmf->pgoff;
2625
2626	/* If we don't want any read-ahead, don't bother */
2627	if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
2628		return fpin;
 
2629	mmap_miss = READ_ONCE(ra->mmap_miss);
2630	if (mmap_miss)
2631		WRITE_ONCE(ra->mmap_miss, --mmap_miss);
2632	if (PageReadahead(page)) {
 
2633		fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2634		page_cache_async_readahead(mapping, ra, file,
2635					   page, offset, ra->ra_pages);
2636	}
2637	return fpin;
2638}
2639
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2640/**
2641 * filemap_fault - read in file data for page fault handling
2642 * @vmf:	struct vm_fault containing details of the fault
2643 *
2644 * filemap_fault() is invoked via the vma operations vector for a
2645 * mapped memory region to read in file data during a page fault.
2646 *
2647 * The goto's are kind of ugly, but this streamlines the normal case of having
2648 * it in the page cache, and handles the special cases reasonably without
2649 * having a lot of duplicated code.
2650 *
2651 * vma->vm_mm->mmap_lock must be held on entry.
2652 *
2653 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
2654 * may be dropped before doing I/O or by lock_page_maybe_drop_mmap().
2655 *
2656 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
2657 * has not been released.
2658 *
2659 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
2660 *
2661 * Return: bitwise-OR of %VM_FAULT_ codes.
2662 */
2663vm_fault_t filemap_fault(struct vm_fault *vmf)
2664{
2665	int error;
2666	struct file *file = vmf->vma->vm_file;
2667	struct file *fpin = NULL;
2668	struct address_space *mapping = file->f_mapping;
2669	struct file_ra_state *ra = &file->f_ra;
2670	struct inode *inode = mapping->host;
2671	pgoff_t offset = vmf->pgoff;
2672	pgoff_t max_off;
2673	struct page *page;
2674	vm_fault_t ret = 0;
 
2675
2676	max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2677	if (unlikely(offset >= max_off))
2678		return VM_FAULT_SIGBUS;
2679
2680	/*
2681	 * Do we have something in the page cache already?
2682	 */
2683	page = find_get_page(mapping, offset);
2684	if (likely(page) && !(vmf->flags & FAULT_FLAG_TRIED)) {
2685		/*
2686		 * We found the page, so try async readahead before
2687		 * waiting for the lock.
2688		 */
2689		fpin = do_async_mmap_readahead(vmf, page);
2690	} else if (!page) {
 
 
 
 
 
 
 
 
 
2691		/* No page in the page cache at all */
2692		count_vm_event(PGMAJFAULT);
2693		count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
2694		ret = VM_FAULT_MAJOR;
2695		fpin = do_sync_mmap_readahead(vmf);
2696retry_find:
2697		page = pagecache_get_page(mapping, offset,
 
 
 
 
 
 
 
 
2698					  FGP_CREAT|FGP_FOR_MMAP,
2699					  vmf->gfp_mask);
2700		if (!page) {
2701			if (fpin)
2702				goto out_retry;
 
2703			return VM_FAULT_OOM;
2704		}
2705	}
2706
2707	if (!lock_page_maybe_drop_mmap(vmf, page, &fpin))
2708		goto out_retry;
2709
2710	/* Did it get truncated? */
2711	if (unlikely(compound_head(page)->mapping != mapping)) {
2712		unlock_page(page);
2713		put_page(page);
2714		goto retry_find;
2715	}
2716	VM_BUG_ON_PAGE(page_to_pgoff(page) != offset, page);
2717
2718	/*
2719	 * We have a locked page in the page cache, now we need to check
2720	 * that it's up-to-date. If not, it is going to be due to an error.
 
2721	 */
2722	if (unlikely(!PageUptodate(page)))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2723		goto page_not_uptodate;
 
2724
2725	/*
2726	 * We've made it this far and we had to drop our mmap_lock, now is the
2727	 * time to return to the upper layer and have it re-find the vma and
2728	 * redo the fault.
2729	 */
2730	if (fpin) {
2731		unlock_page(page);
2732		goto out_retry;
2733	}
 
 
2734
2735	/*
2736	 * Found the page and have a reference on it.
2737	 * We must recheck i_size under page lock.
2738	 */
2739	max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2740	if (unlikely(offset >= max_off)) {
2741		unlock_page(page);
2742		put_page(page);
2743		return VM_FAULT_SIGBUS;
2744	}
2745
2746	vmf->page = page;
2747	return ret | VM_FAULT_LOCKED;
2748
2749page_not_uptodate:
2750	/*
2751	 * Umm, take care of errors if the page isn't up-to-date.
2752	 * Try to re-read it _once_. We do this synchronously,
2753	 * because there really aren't any performance issues here
2754	 * and we need to check for errors.
2755	 */
2756	ClearPageError(page);
2757	fpin = maybe_unlock_mmap_for_io(vmf, fpin);
2758	error = mapping->a_ops->readpage(file, page);
2759	if (!error) {
2760		wait_on_page_locked(page);
2761		if (!PageUptodate(page))
2762			error = -EIO;
2763	}
2764	if (fpin)
2765		goto out_retry;
2766	put_page(page);
2767
2768	if (!error || error == AOP_TRUNCATED_PAGE)
2769		goto retry_find;
 
2770
2771	shrink_readahead_size_eio(ra);
2772	return VM_FAULT_SIGBUS;
2773
2774out_retry:
2775	/*
2776	 * We dropped the mmap_lock, we need to return to the fault handler to
2777	 * re-find the vma and come back and find our hopefully still populated
2778	 * page.
2779	 */
2780	if (page)
2781		put_page(page);
 
 
2782	if (fpin)
2783		fput(fpin);
2784	return ret | VM_FAULT_RETRY;
2785}
2786EXPORT_SYMBOL(filemap_fault);
2787
2788void filemap_map_pages(struct vm_fault *vmf,
2789		pgoff_t start_pgoff, pgoff_t end_pgoff)
2790{
2791	struct file *file = vmf->vma->vm_file;
2792	struct address_space *mapping = file->f_mapping;
2793	pgoff_t last_pgoff = start_pgoff;
2794	unsigned long max_idx;
2795	XA_STATE(xas, &mapping->i_pages, start_pgoff);
2796	struct page *page;
2797	unsigned int mmap_miss = READ_ONCE(file->f_ra.mmap_miss);
2798
2799	rcu_read_lock();
2800	xas_for_each(&xas, page, end_pgoff) {
2801		if (xas_retry(&xas, page))
2802			continue;
2803		if (xa_is_value(page))
2804			goto next;
2805
2806		/*
2807		 * Check for a locked page first, as a speculative
2808		 * reference may adversely influence page migration.
2809		 */
2810		if (PageLocked(page))
2811			goto next;
2812		if (!page_cache_get_speculative(page))
2813			goto next;
 
 
 
 
2814
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2815		/* Has the page moved or been split? */
2816		if (unlikely(page != xas_reload(&xas)))
2817			goto skip;
2818		page = find_subpage(page, xas.xa_index);
2819
2820		if (!PageUptodate(page) ||
2821				PageReadahead(page) ||
2822				PageHWPoison(page))
2823			goto skip;
2824		if (!trylock_page(page))
2825			goto skip;
2826
2827		if (page->mapping != mapping || !PageUptodate(page))
 
2828			goto unlock;
2829
2830		max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
2831		if (page->index >= max_idx)
2832			goto unlock;
 
 
 
 
 
 
 
 
 
2833
2834		if (mmap_miss > 0)
2835			mmap_miss--;
 
 
 
 
 
 
 
 
 
 
 
2836
2837		vmf->address += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
2838		if (vmf->pte)
2839			vmf->pte += xas.xa_index - last_pgoff;
2840		last_pgoff = xas.xa_index;
2841		if (alloc_set_pte(vmf, page))
2842			goto unlock;
2843		unlock_page(page);
2844		goto next;
2845unlock:
2846		unlock_page(page);
 
 
 
 
 
 
2847skip:
2848		put_page(page);
2849next:
2850		/* Huge page is mapped? No need to proceed. */
2851		if (pmd_trans_huge(*vmf->pmd))
2852			break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2853	}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2854	rcu_read_unlock();
2855	WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss);
 
 
 
 
 
 
 
2856}
2857EXPORT_SYMBOL(filemap_map_pages);
2858
2859vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
2860{
2861	struct page *page = vmf->page;
2862	struct inode *inode = file_inode(vmf->vma->vm_file);
2863	vm_fault_t ret = VM_FAULT_LOCKED;
2864
2865	sb_start_pagefault(inode->i_sb);
2866	file_update_time(vmf->vma->vm_file);
2867	lock_page(page);
2868	if (page->mapping != inode->i_mapping) {
2869		unlock_page(page);
2870		ret = VM_FAULT_NOPAGE;
2871		goto out;
2872	}
2873	/*
2874	 * We mark the page dirty already here so that when freeze is in
2875	 * progress, we are guaranteed that writeback during freezing will
2876	 * see the dirty page and writeprotect it again.
2877	 */
2878	set_page_dirty(page);
2879	wait_for_stable_page(page);
2880out:
2881	sb_end_pagefault(inode->i_sb);
2882	return ret;
2883}
2884
2885const struct vm_operations_struct generic_file_vm_ops = {
2886	.fault		= filemap_fault,
2887	.map_pages	= filemap_map_pages,
2888	.page_mkwrite	= filemap_page_mkwrite,
2889};
2890
2891/* This is used for a general mmap of a disk file */
2892
2893int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2894{
2895	struct address_space *mapping = file->f_mapping;
2896
2897	if (!mapping->a_ops->readpage)
2898		return -ENOEXEC;
2899	file_accessed(file);
2900	vma->vm_ops = &generic_file_vm_ops;
2901	return 0;
2902}
2903
2904/*
2905 * This is for filesystems which do not implement ->writepage.
2906 */
2907int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
2908{
2909	if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_MAYWRITE))
2910		return -EINVAL;
2911	return generic_file_mmap(file, vma);
2912}
2913#else
2914vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
2915{
2916	return VM_FAULT_SIGBUS;
2917}
2918int generic_file_mmap(struct file * file, struct vm_area_struct * vma)
2919{
2920	return -ENOSYS;
2921}
2922int generic_file_readonly_mmap(struct file * file, struct vm_area_struct * vma)
2923{
2924	return -ENOSYS;
2925}
2926#endif /* CONFIG_MMU */
2927
2928EXPORT_SYMBOL(filemap_page_mkwrite);
2929EXPORT_SYMBOL(generic_file_mmap);
2930EXPORT_SYMBOL(generic_file_readonly_mmap);
2931
2932static struct page *wait_on_page_read(struct page *page)
2933{
2934	if (!IS_ERR(page)) {
2935		wait_on_page_locked(page);
2936		if (!PageUptodate(page)) {
2937			put_page(page);
2938			page = ERR_PTR(-EIO);
2939		}
2940	}
2941	return page;
2942}
2943
2944static struct page *do_read_cache_page(struct address_space *mapping,
2945				pgoff_t index,
2946				int (*filler)(void *, struct page *),
2947				void *data,
2948				gfp_t gfp)
2949{
2950	struct page *page;
2951	int err;
 
 
 
2952repeat:
2953	page = find_get_page(mapping, index);
2954	if (!page) {
2955		page = __page_cache_alloc(gfp);
2956		if (!page)
2957			return ERR_PTR(-ENOMEM);
2958		err = add_to_page_cache_lru(page, mapping, index, gfp);
2959		if (unlikely(err)) {
2960			put_page(page);
2961			if (err == -EEXIST)
2962				goto repeat;
2963			/* Presumably ENOMEM for xarray node */
2964			return ERR_PTR(err);
2965		}
2966
2967filler:
2968		if (filler)
2969			err = filler(data, page);
2970		else
2971			err = mapping->a_ops->readpage(data, page);
2972
2973		if (err < 0) {
2974			put_page(page);
2975			return ERR_PTR(err);
2976		}
2977
2978		page = wait_on_page_read(page);
2979		if (IS_ERR(page))
2980			return page;
2981		goto out;
2982	}
2983	if (PageUptodate(page))
2984		goto out;
2985
2986	/*
2987	 * Page is not up to date and may be locked due one of the following
2988	 * case a: Page is being filled and the page lock is held
2989	 * case b: Read/write error clearing the page uptodate status
2990	 * case c: Truncation in progress (page locked)
2991	 * case d: Reclaim in progress
2992	 *
2993	 * Case a, the page will be up to date when the page is unlocked.
2994	 *    There is no need to serialise on the page lock here as the page
2995	 *    is pinned so the lock gives no additional protection. Even if the
2996	 *    page is truncated, the data is still valid if PageUptodate as
2997	 *    it's a race vs truncate race.
2998	 * Case b, the page will not be up to date
2999	 * Case c, the page may be truncated but in itself, the data may still
3000	 *    be valid after IO completes as it's a read vs truncate race. The
3001	 *    operation must restart if the page is not uptodate on unlock but
3002	 *    otherwise serialising on page lock to stabilise the mapping gives
3003	 *    no additional guarantees to the caller as the page lock is
3004	 *    released before return.
3005	 * Case d, similar to truncation. If reclaim holds the page lock, it
3006	 *    will be a race with remove_mapping that determines if the mapping
3007	 *    is valid on unlock but otherwise the data is valid and there is
3008	 *    no need to serialise with page lock.
3009	 *
3010	 * As the page lock gives no additional guarantee, we optimistically
3011	 * wait on the page to be unlocked and check if it's up to date and
3012	 * use the page if it is. Otherwise, the page lock is required to
3013	 * distinguish between the different cases. The motivation is that we
3014	 * avoid spurious serialisations and wakeups when multiple processes
3015	 * wait on the same page for IO to complete.
3016	 */
3017	wait_on_page_locked(page);
3018	if (PageUptodate(page))
3019		goto out;
3020
3021	/* Distinguish between all the cases under the safety of the lock */
3022	lock_page(page);
 
 
3023
3024	/* Case c or d, restart the operation */
3025	if (!page->mapping) {
3026		unlock_page(page);
3027		put_page(page);
3028		goto repeat;
3029	}
3030
3031	/* Someone else locked and filled the page in a very small window */
3032	if (PageUptodate(page)) {
3033		unlock_page(page);
3034		goto out;
3035	}
3036
3037	/*
3038	 * A previous I/O error may have been due to temporary
3039	 * failures.
3040	 * Clear page error before actual read, PG_error will be
3041	 * set again if read page fails.
3042	 */
3043	ClearPageError(page);
3044	goto filler;
3045
3046out:
3047	mark_page_accessed(page);
3048	return page;
3049}
3050
3051/**
3052 * read_cache_page - read into page cache, fill it if needed
3053 * @mapping:	the page's address_space
3054 * @index:	the page index
3055 * @filler:	function to perform the read
3056 * @data:	first arg to filler(data, page) function, often left as NULL
3057 *
3058 * Read into the page cache. If a page already exists, and PageUptodate() is
3059 * not set, try to fill the page and wait for it to become unlocked.
3060 *
3061 * If the page does not get brought uptodate, return -EIO.
 
3062 *
3063 * Return: up to date page on success, ERR_PTR() on failure.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3064 */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3065struct page *read_cache_page(struct address_space *mapping,
3066				pgoff_t index,
3067				int (*filler)(void *, struct page *),
3068				void *data)
3069{
3070	return do_read_cache_page(mapping, index, filler, data,
3071			mapping_gfp_mask(mapping));
3072}
3073EXPORT_SYMBOL(read_cache_page);
3074
3075/**
3076 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3077 * @mapping:	the page's address_space
3078 * @index:	the page index
3079 * @gfp:	the page allocator flags to use if allocating
3080 *
3081 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3082 * any new page allocations done using the specified allocation flags.
3083 *
3084 * If the page does not get brought uptodate, return -EIO.
3085 *
 
 
3086 * Return: up to date page on success, ERR_PTR() on failure.
3087 */
3088struct page *read_cache_page_gfp(struct address_space *mapping,
3089				pgoff_t index,
3090				gfp_t gfp)
3091{
3092	return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3093}
3094EXPORT_SYMBOL(read_cache_page_gfp);
3095
3096/*
3097 * Don't operate on ranges the page cache doesn't support, and don't exceed the
3098 * LFS limits.  If pos is under the limit it becomes a short access.  If it
3099 * exceeds the limit we return -EFBIG.
3100 */
3101static int generic_write_check_limits(struct file *file, loff_t pos,
3102				      loff_t *count)
3103{
3104	struct inode *inode = file->f_mapping->host;
3105	loff_t max_size = inode->i_sb->s_maxbytes;
3106	loff_t limit = rlimit(RLIMIT_FSIZE);
3107
3108	if (limit != RLIM_INFINITY) {
3109		if (pos >= limit) {
3110			send_sig(SIGXFSZ, current, 0);
3111			return -EFBIG;
3112		}
3113		*count = min(*count, limit - pos);
3114	}
3115
3116	if (!(file->f_flags & O_LARGEFILE))
3117		max_size = MAX_NON_LFS;
3118
3119	if (unlikely(pos >= max_size))
3120		return -EFBIG;
3121
3122	*count = min(*count, max_size - pos);
3123
3124	return 0;
3125}
3126
3127/*
3128 * Performs necessary checks before doing a write
3129 *
3130 * Can adjust writing position or amount of bytes to write.
3131 * Returns appropriate error code that caller should return or
3132 * zero in case that write should be allowed.
3133 */
3134inline ssize_t generic_write_checks(struct kiocb *iocb, struct iov_iter *from)
3135{
3136	struct file *file = iocb->ki_filp;
3137	struct inode *inode = file->f_mapping->host;
3138	loff_t count;
3139	int ret;
3140
3141	if (IS_SWAPFILE(inode))
3142		return -ETXTBSY;
3143
3144	if (!iov_iter_count(from))
3145		return 0;
3146
3147	/* FIXME: this is for backwards compatibility with 2.4 */
3148	if (iocb->ki_flags & IOCB_APPEND)
3149		iocb->ki_pos = i_size_read(inode);
3150
3151	if ((iocb->ki_flags & IOCB_NOWAIT) && !(iocb->ki_flags & IOCB_DIRECT))
3152		return -EINVAL;
3153
3154	count = iov_iter_count(from);
3155	ret = generic_write_check_limits(file, iocb->ki_pos, &count);
3156	if (ret)
3157		return ret;
3158
3159	iov_iter_truncate(from, count);
3160	return iov_iter_count(from);
3161}
3162EXPORT_SYMBOL(generic_write_checks);
3163
3164/*
3165 * Performs necessary checks before doing a clone.
3166 *
3167 * Can adjust amount of bytes to clone via @req_count argument.
3168 * Returns appropriate error code that caller should return or
3169 * zero in case the clone should be allowed.
3170 */
3171int generic_remap_checks(struct file *file_in, loff_t pos_in,
3172			 struct file *file_out, loff_t pos_out,
3173			 loff_t *req_count, unsigned int remap_flags)
3174{
3175	struct inode *inode_in = file_in->f_mapping->host;
3176	struct inode *inode_out = file_out->f_mapping->host;
3177	uint64_t count = *req_count;
3178	uint64_t bcount;
3179	loff_t size_in, size_out;
3180	loff_t bs = inode_out->i_sb->s_blocksize;
3181	int ret;
3182
3183	/* The start of both ranges must be aligned to an fs block. */
3184	if (!IS_ALIGNED(pos_in, bs) || !IS_ALIGNED(pos_out, bs))
3185		return -EINVAL;
3186
3187	/* Ensure offsets don't wrap. */
3188	if (pos_in + count < pos_in || pos_out + count < pos_out)
3189		return -EINVAL;
3190
3191	size_in = i_size_read(inode_in);
3192	size_out = i_size_read(inode_out);
3193
3194	/* Dedupe requires both ranges to be within EOF. */
3195	if ((remap_flags & REMAP_FILE_DEDUP) &&
3196	    (pos_in >= size_in || pos_in + count > size_in ||
3197	     pos_out >= size_out || pos_out + count > size_out))
3198		return -EINVAL;
3199
3200	/* Ensure the infile range is within the infile. */
3201	if (pos_in >= size_in)
3202		return -EINVAL;
3203	count = min(count, size_in - (uint64_t)pos_in);
3204
3205	ret = generic_write_check_limits(file_out, pos_out, &count);
3206	if (ret)
3207		return ret;
3208
3209	/*
3210	 * If the user wanted us to link to the infile's EOF, round up to the
3211	 * next block boundary for this check.
3212	 *
3213	 * Otherwise, make sure the count is also block-aligned, having
3214	 * already confirmed the starting offsets' block alignment.
3215	 */
3216	if (pos_in + count == size_in) {
3217		bcount = ALIGN(size_in, bs) - pos_in;
3218	} else {
3219		if (!IS_ALIGNED(count, bs))
3220			count = ALIGN_DOWN(count, bs);
3221		bcount = count;
3222	}
3223
3224	/* Don't allow overlapped cloning within the same file. */
3225	if (inode_in == inode_out &&
3226	    pos_out + bcount > pos_in &&
3227	    pos_out < pos_in + bcount)
3228		return -EINVAL;
3229
3230	/*
3231	 * We shortened the request but the caller can't deal with that, so
3232	 * bounce the request back to userspace.
3233	 */
3234	if (*req_count != count && !(remap_flags & REMAP_FILE_CAN_SHORTEN))
3235		return -EINVAL;
3236
3237	*req_count = count;
3238	return 0;
3239}
3240
3241
3242/*
3243 * Performs common checks before doing a file copy/clone
3244 * from @file_in to @file_out.
3245 */
3246int generic_file_rw_checks(struct file *file_in, struct file *file_out)
3247{
3248	struct inode *inode_in = file_inode(file_in);
3249	struct inode *inode_out = file_inode(file_out);
3250
3251	/* Don't copy dirs, pipes, sockets... */
3252	if (S_ISDIR(inode_in->i_mode) || S_ISDIR(inode_out->i_mode))
3253		return -EISDIR;
3254	if (!S_ISREG(inode_in->i_mode) || !S_ISREG(inode_out->i_mode))
3255		return -EINVAL;
3256
3257	if (!(file_in->f_mode & FMODE_READ) ||
3258	    !(file_out->f_mode & FMODE_WRITE) ||
3259	    (file_out->f_flags & O_APPEND))
3260		return -EBADF;
3261
3262	return 0;
3263}
3264
3265/*
3266 * Performs necessary checks before doing a file copy
3267 *
3268 * Can adjust amount of bytes to copy via @req_count argument.
3269 * Returns appropriate error code that caller should return or
3270 * zero in case the copy should be allowed.
3271 */
3272int generic_copy_file_checks(struct file *file_in, loff_t pos_in,
3273			     struct file *file_out, loff_t pos_out,
3274			     size_t *req_count, unsigned int flags)
3275{
3276	struct inode *inode_in = file_inode(file_in);
3277	struct inode *inode_out = file_inode(file_out);
3278	uint64_t count = *req_count;
3279	loff_t size_in;
3280	int ret;
3281
3282	ret = generic_file_rw_checks(file_in, file_out);
3283	if (ret)
3284		return ret;
3285
3286	/* Don't touch certain kinds of inodes */
3287	if (IS_IMMUTABLE(inode_out))
3288		return -EPERM;
3289
3290	if (IS_SWAPFILE(inode_in) || IS_SWAPFILE(inode_out))
3291		return -ETXTBSY;
3292
3293	/* Ensure offsets don't wrap. */
3294	if (pos_in + count < pos_in || pos_out + count < pos_out)
3295		return -EOVERFLOW;
3296
3297	/* Shorten the copy to EOF */
3298	size_in = i_size_read(inode_in);
3299	if (pos_in >= size_in)
3300		count = 0;
3301	else
3302		count = min(count, size_in - (uint64_t)pos_in);
3303
3304	ret = generic_write_check_limits(file_out, pos_out, &count);
3305	if (ret)
3306		return ret;
3307
3308	/* Don't allow overlapped copying within the same file. */
3309	if (inode_in == inode_out &&
3310	    pos_out + count > pos_in &&
3311	    pos_out < pos_in + count)
3312		return -EINVAL;
3313
3314	*req_count = count;
3315	return 0;
3316}
3317
3318int pagecache_write_begin(struct file *file, struct address_space *mapping,
3319				loff_t pos, unsigned len, unsigned flags,
3320				struct page **pagep, void **fsdata)
3321{
3322	const struct address_space_operations *aops = mapping->a_ops;
3323
3324	return aops->write_begin(file, mapping, pos, len, flags,
3325							pagep, fsdata);
3326}
3327EXPORT_SYMBOL(pagecache_write_begin);
3328
3329int pagecache_write_end(struct file *file, struct address_space *mapping,
3330				loff_t pos, unsigned len, unsigned copied,
3331				struct page *page, void *fsdata)
3332{
3333	const struct address_space_operations *aops = mapping->a_ops;
3334
3335	return aops->write_end(file, mapping, pos, len, copied, page, fsdata);
3336}
3337EXPORT_SYMBOL(pagecache_write_end);
3338
3339/*
3340 * Warn about a page cache invalidation failure during a direct I/O write.
3341 */
3342void dio_warn_stale_pagecache(struct file *filp)
3343{
3344	static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3345	char pathname[128];
3346	struct inode *inode = file_inode(filp);
3347	char *path;
3348
3349	errseq_set(&inode->i_mapping->wb_err, -EIO);
3350	if (__ratelimit(&_rs)) {
3351		path = file_path(filp, pathname, sizeof(pathname));
3352		if (IS_ERR(path))
3353			path = "(unknown)";
3354		pr_crit("Page cache invalidation failure on direct I/O.  Possible data corruption due to collision with buffered I/O!\n");
3355		pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3356			current->comm);
3357	}
3358}
3359
3360ssize_t
3361generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3362{
3363	struct file	*file = iocb->ki_filp;
3364	struct address_space *mapping = file->f_mapping;
3365	struct inode	*inode = mapping->host;
3366	loff_t		pos = iocb->ki_pos;
3367	ssize_t		written;
3368	size_t		write_len;
3369	pgoff_t		end;
3370
3371	write_len = iov_iter_count(from);
3372	end = (pos + write_len - 1) >> PAGE_SHIFT;
 
 
 
 
3373
3374	if (iocb->ki_flags & IOCB_NOWAIT) {
3375		/* If there are pages to writeback, return */
3376		if (filemap_range_has_page(inode->i_mapping, pos,
3377					   pos + write_len - 1))
3378			return -EAGAIN;
3379	} else {
3380		written = filemap_write_and_wait_range(mapping, pos,
3381							pos + write_len - 1);
3382		if (written)
3383			goto out;
3384	}
3385
3386	/*
3387	 * After a write we want buffered reads to be sure to go to disk to get
3388	 * the new data.  We invalidate clean cached page from the region we're
3389	 * about to write.  We do this *before* the write so that we can return
3390	 * without clobbering -EIOCBQUEUED from ->direct_IO().
3391	 */
3392	written = invalidate_inode_pages2_range(mapping,
3393					pos >> PAGE_SHIFT, end);
3394	/*
3395	 * If a page can not be invalidated, return 0 to fall back
3396	 * to buffered write.
3397	 */
 
3398	if (written) {
3399		if (written == -EBUSY)
3400			return 0;
3401		goto out;
3402	}
3403
3404	written = mapping->a_ops->direct_IO(iocb, from);
3405
3406	/*
3407	 * Finally, try again to invalidate clean pages which might have been
3408	 * cached by non-direct readahead, or faulted in by get_user_pages()
3409	 * if the source of the write was an mmap'ed region of the file
3410	 * we're writing.  Either one is a pretty crazy thing to do,
3411	 * so we don't support it 100%.  If this invalidation
3412	 * fails, tough, the write still worked...
3413	 *
3414	 * Most of the time we do not need this since dio_complete() will do
3415	 * the invalidation for us. However there are some file systems that
3416	 * do not end up with dio_complete() being called, so let's not break
3417	 * them by removing it completely.
3418	 *
3419	 * Noticeable example is a blkdev_direct_IO().
3420	 *
3421	 * Skip invalidation for async writes or if mapping has no pages.
3422	 */
3423	if (written > 0 && mapping->nrpages &&
3424	    invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT, end))
3425		dio_warn_stale_pagecache(file);
3426
3427	if (written > 0) {
 
 
 
 
3428		pos += written;
3429		write_len -= written;
3430		if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3431			i_size_write(inode, pos);
3432			mark_inode_dirty(inode);
3433		}
3434		iocb->ki_pos = pos;
3435	}
3436	iov_iter_revert(from, write_len - iov_iter_count(from));
3437out:
3438	return written;
3439}
3440EXPORT_SYMBOL(generic_file_direct_write);
3441
3442/*
3443 * Find or create a page at the given pagecache position. Return the locked
3444 * page. This function is specifically for buffered writes.
3445 */
3446struct page *grab_cache_page_write_begin(struct address_space *mapping,
3447					pgoff_t index, unsigned flags)
3448{
3449	struct page *page;
3450	int fgp_flags = FGP_LOCK|FGP_WRITE|FGP_CREAT;
3451
3452	if (flags & AOP_FLAG_NOFS)
3453		fgp_flags |= FGP_NOFS;
3454
3455	page = pagecache_get_page(mapping, index, fgp_flags,
3456			mapping_gfp_mask(mapping));
3457	if (page)
3458		wait_for_stable_page(page);
3459
3460	return page;
3461}
3462EXPORT_SYMBOL(grab_cache_page_write_begin);
3463
3464ssize_t generic_perform_write(struct file *file,
3465				struct iov_iter *i, loff_t pos)
3466{
 
 
3467	struct address_space *mapping = file->f_mapping;
3468	const struct address_space_operations *a_ops = mapping->a_ops;
3469	long status = 0;
3470	ssize_t written = 0;
3471	unsigned int flags = 0;
3472
3473	do {
3474		struct page *page;
3475		unsigned long offset;	/* Offset into pagecache page */
3476		unsigned long bytes;	/* Bytes to write to page */
3477		size_t copied;		/* Bytes copied from user */
3478		void *fsdata;
3479
3480		offset = (pos & (PAGE_SIZE - 1));
3481		bytes = min_t(unsigned long, PAGE_SIZE - offset,
3482						iov_iter_count(i));
3483
3484again:
3485		/*
3486		 * Bring in the user page that we will copy from _first_.
3487		 * Otherwise there's a nasty deadlock on copying from the
3488		 * same page as we're writing to, without it being marked
3489		 * up-to-date.
3490		 *
3491		 * Not only is this an optimisation, but it is also required
3492		 * to check that the address is actually valid, when atomic
3493		 * usercopies are used, below.
3494		 */
3495		if (unlikely(iov_iter_fault_in_readable(i, bytes))) {
3496			status = -EFAULT;
3497			break;
3498		}
3499
3500		if (fatal_signal_pending(current)) {
3501			status = -EINTR;
3502			break;
3503		}
3504
3505		status = a_ops->write_begin(file, mapping, pos, bytes, flags,
3506						&page, &fsdata);
3507		if (unlikely(status < 0))
3508			break;
3509
3510		if (mapping_writably_mapped(mapping))
3511			flush_dcache_page(page);
3512
3513		copied = iov_iter_copy_from_user_atomic(page, i, offset, bytes);
3514		flush_dcache_page(page);
3515
3516		status = a_ops->write_end(file, mapping, pos, bytes, copied,
3517						page, fsdata);
3518		if (unlikely(status < 0))
3519			break;
3520		copied = status;
3521
 
3522		cond_resched();
3523
3524		iov_iter_advance(i, copied);
3525		if (unlikely(copied == 0)) {
3526			/*
3527			 * If we were unable to copy any data at all, we must
3528			 * fall back to a single segment length write.
3529			 *
3530			 * If we didn't fallback here, we could livelock
3531			 * because not all segments in the iov can be copied at
3532			 * once without a pagefault.
3533			 */
3534			bytes = min_t(unsigned long, PAGE_SIZE - offset,
3535						iov_iter_single_seg_count(i));
3536			goto again;
3537		}
3538		pos += copied;
3539		written += copied;
3540
3541		balance_dirty_pages_ratelimited(mapping);
3542	} while (iov_iter_count(i));
3543
3544	return written ? written : status;
 
 
 
3545}
3546EXPORT_SYMBOL(generic_perform_write);
3547
3548/**
3549 * __generic_file_write_iter - write data to a file
3550 * @iocb:	IO state structure (file, offset, etc.)
3551 * @from:	iov_iter with data to write
3552 *
3553 * This function does all the work needed for actually writing data to a
3554 * file. It does all basic checks, removes SUID from the file, updates
3555 * modification times and calls proper subroutines depending on whether we
3556 * do direct IO or a standard buffered write.
3557 *
3558 * It expects i_mutex to be grabbed unless we work on a block device or similar
3559 * object which does not need locking at all.
3560 *
3561 * This function does *not* take care of syncing data in case of O_SYNC write.
3562 * A caller has to handle it. This is mainly due to the fact that we want to
3563 * avoid syncing under i_mutex.
3564 *
3565 * Return:
3566 * * number of bytes written, even for truncated writes
3567 * * negative error code if no data has been written at all
3568 */
3569ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3570{
3571	struct file *file = iocb->ki_filp;
3572	struct address_space * mapping = file->f_mapping;
3573	struct inode 	*inode = mapping->host;
3574	ssize_t		written = 0;
3575	ssize_t		err;
3576	ssize_t		status;
3577
3578	/* We can write back this queue in page reclaim */
3579	current->backing_dev_info = inode_to_bdi(inode);
3580	err = file_remove_privs(file);
3581	if (err)
3582		goto out;
3583
3584	err = file_update_time(file);
3585	if (err)
3586		goto out;
3587
3588	if (iocb->ki_flags & IOCB_DIRECT) {
3589		loff_t pos, endbyte;
 
3590
3591		written = generic_file_direct_write(iocb, from);
 
3592		/*
3593		 * If the write stopped short of completing, fall back to
3594		 * buffered writes.  Some filesystems do this for writes to
3595		 * holes, for example.  For DAX files, a buffered write will
3596		 * not succeed (even if it did, DAX does not handle dirty
3597		 * page-cache pages correctly).
3598		 */
3599		if (written < 0 || !iov_iter_count(from) || IS_DAX(inode))
3600			goto out;
3601
3602		status = generic_perform_write(file, from, pos = iocb->ki_pos);
3603		/*
3604		 * If generic_perform_write() returned a synchronous error
3605		 * then we want to return the number of bytes which were
3606		 * direct-written, or the error code if that was zero.  Note
3607		 * that this differs from normal direct-io semantics, which
3608		 * will return -EFOO even if some bytes were written.
3609		 */
3610		if (unlikely(status < 0)) {
3611			err = status;
3612			goto out;
3613		}
3614		/*
3615		 * We need to ensure that the page cache pages are written to
3616		 * disk and invalidated to preserve the expected O_DIRECT
3617		 * semantics.
3618		 */
3619		endbyte = pos + status - 1;
3620		err = filemap_write_and_wait_range(mapping, pos, endbyte);
3621		if (err == 0) {
3622			iocb->ki_pos = endbyte + 1;
3623			written += status;
3624			invalidate_mapping_pages(mapping,
3625						 pos >> PAGE_SHIFT,
3626						 endbyte >> PAGE_SHIFT);
3627		} else {
3628			/*
3629			 * We don't know how much we wrote, so just return
3630			 * the number of bytes which were direct-written
3631			 */
3632		}
3633	} else {
3634		written = generic_perform_write(file, from, iocb->ki_pos);
3635		if (likely(written > 0))
3636			iocb->ki_pos += written;
3637	}
3638out:
3639	current->backing_dev_info = NULL;
3640	return written ? written : err;
3641}
3642EXPORT_SYMBOL(__generic_file_write_iter);
3643
3644/**
3645 * generic_file_write_iter - write data to a file
3646 * @iocb:	IO state structure
3647 * @from:	iov_iter with data to write
3648 *
3649 * This is a wrapper around __generic_file_write_iter() to be used by most
3650 * filesystems. It takes care of syncing the file in case of O_SYNC file
3651 * and acquires i_mutex as needed.
3652 * Return:
3653 * * negative error code if no data has been written at all of
3654 *   vfs_fsync_range() failed for a synchronous write
3655 * * number of bytes written, even for truncated writes
3656 */
3657ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
3658{
3659	struct file *file = iocb->ki_filp;
3660	struct inode *inode = file->f_mapping->host;
3661	ssize_t ret;
3662
3663	inode_lock(inode);
3664	ret = generic_write_checks(iocb, from);
3665	if (ret > 0)
3666		ret = __generic_file_write_iter(iocb, from);
3667	inode_unlock(inode);
3668
3669	if (ret > 0)
3670		ret = generic_write_sync(iocb, ret);
3671	return ret;
3672}
3673EXPORT_SYMBOL(generic_file_write_iter);
3674
3675/**
3676 * try_to_release_page() - release old fs-specific metadata on a page
3677 *
3678 * @page: the page which the kernel is trying to free
3679 * @gfp_mask: memory allocation flags (and I/O mode)
3680 *
3681 * The address_space is to try to release any data against the page
3682 * (presumably at page->private).
3683 *
3684 * This may also be called if PG_fscache is set on a page, indicating that the
3685 * page is known to the local caching routines.
3686 *
3687 * The @gfp_mask argument specifies whether I/O may be performed to release
3688 * this page (__GFP_IO), and whether the call may block (__GFP_RECLAIM & __GFP_FS).
 
3689 *
3690 * Return: %1 if the release was successful, otherwise return zero.
3691 */
3692int try_to_release_page(struct page *page, gfp_t gfp_mask)
3693{
3694	struct address_space * const mapping = page->mapping;
3695
3696	BUG_ON(!PageLocked(page));
3697	if (PageWriteback(page))
3698		return 0;
 
 
3699
3700	if (mapping && mapping->a_ops->releasepage)
3701		return mapping->a_ops->releasepage(page, gfp_mask);
3702	return try_to_free_buffers(page);
3703}
 
3704
3705EXPORT_SYMBOL(try_to_release_page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *	linux/mm/filemap.c
   4 *
   5 * Copyright (C) 1994-1999  Linus Torvalds
   6 */
   7
   8/*
   9 * This file handles the generic file mmap semantics used by
  10 * most "normal" filesystems (but you don't /have/ to use this:
  11 * the NFS filesystem used to do this differently, for example)
  12 */
  13#include <linux/export.h>
  14#include <linux/compiler.h>
  15#include <linux/dax.h>
  16#include <linux/fs.h>
  17#include <linux/sched/signal.h>
  18#include <linux/uaccess.h>
  19#include <linux/capability.h>
  20#include <linux/kernel_stat.h>
  21#include <linux/gfp.h>
  22#include <linux/mm.h>
  23#include <linux/swap.h>
  24#include <linux/swapops.h>
  25#include <linux/syscalls.h>
  26#include <linux/mman.h>
  27#include <linux/pagemap.h>
  28#include <linux/file.h>
  29#include <linux/uio.h>
  30#include <linux/error-injection.h>
  31#include <linux/hash.h>
  32#include <linux/writeback.h>
  33#include <linux/backing-dev.h>
  34#include <linux/pagevec.h>
 
  35#include <linux/security.h>
  36#include <linux/cpuset.h>
  37#include <linux/hugetlb.h>
  38#include <linux/memcontrol.h>
 
  39#include <linux/shmem_fs.h>
  40#include <linux/rmap.h>
  41#include <linux/delayacct.h>
  42#include <linux/psi.h>
  43#include <linux/ramfs.h>
  44#include <linux/page_idle.h>
  45#include <linux/migrate.h>
  46#include <linux/pipe_fs_i.h>
  47#include <linux/splice.h>
  48#include <linux/rcupdate_wait.h>
  49#include <asm/pgalloc.h>
  50#include <asm/tlbflush.h>
  51#include "internal.h"
  52
  53#define CREATE_TRACE_POINTS
  54#include <trace/events/filemap.h>
  55
  56/*
  57 * FIXME: remove all knowledge of the buffer layer from the core VM
  58 */
  59#include <linux/buffer_head.h> /* for try_to_free_buffers */
  60
  61#include <asm/mman.h>
  62
  63#include "swap.h"
  64
  65/*
  66 * Shared mappings implemented 30.11.1994. It's not fully working yet,
  67 * though.
  68 *
  69 * Shared mappings now work. 15.8.1995  Bruno.
  70 *
  71 * finished 'unifying' the page and buffer cache and SMP-threaded the
  72 * page-cache, 21.05.1999, Ingo Molnar <mingo@redhat.com>
  73 *
  74 * SMP-threaded pagemap-LRU 1999, Andrea Arcangeli <andrea@suse.de>
  75 */
  76
  77/*
  78 * Lock ordering:
  79 *
  80 *  ->i_mmap_rwsem		(truncate_pagecache)
  81 *    ->private_lock		(__free_pte->block_dirty_folio)
  82 *      ->swap_lock		(exclusive_swap_page, others)
  83 *        ->i_pages lock
  84 *
  85 *  ->i_rwsem
  86 *    ->invalidate_lock		(acquired by fs in truncate path)
  87 *      ->i_mmap_rwsem		(truncate->unmap_mapping_range)
  88 *
  89 *  ->mmap_lock
  90 *    ->i_mmap_rwsem
  91 *      ->page_table_lock or pte_lock	(various, mainly in memory.c)
  92 *        ->i_pages lock	(arch-dependent flush_dcache_mmap_lock)
  93 *
  94 *  ->mmap_lock
  95 *    ->invalidate_lock		(filemap_fault)
  96 *      ->lock_page		(filemap_fault, access_process_vm)
  97 *
  98 *  ->i_rwsem			(generic_perform_write)
  99 *    ->mmap_lock		(fault_in_readable->do_page_fault)
 100 *
 101 *  bdi->wb.list_lock
 102 *    sb_lock			(fs/fs-writeback.c)
 103 *    ->i_pages lock		(__sync_single_inode)
 104 *
 105 *  ->i_mmap_rwsem
 106 *    ->anon_vma.lock		(vma_merge)
 107 *
 108 *  ->anon_vma.lock
 109 *    ->page_table_lock or pte_lock	(anon_vma_prepare and various)
 110 *
 111 *  ->page_table_lock or pte_lock
 112 *    ->swap_lock		(try_to_unmap_one)
 113 *    ->private_lock		(try_to_unmap_one)
 114 *    ->i_pages lock		(try_to_unmap_one)
 115 *    ->lruvec->lru_lock	(follow_page->mark_page_accessed)
 116 *    ->lruvec->lru_lock	(check_pte_range->isolate_lru_page)
 117 *    ->private_lock		(folio_remove_rmap_pte->set_page_dirty)
 118 *    ->i_pages lock		(folio_remove_rmap_pte->set_page_dirty)
 119 *    bdi.wb->list_lock		(folio_remove_rmap_pte->set_page_dirty)
 120 *    ->inode->i_lock		(folio_remove_rmap_pte->set_page_dirty)
 121 *    ->memcg->move_lock	(folio_remove_rmap_pte->folio_memcg_lock)
 122 *    bdi.wb->list_lock		(zap_pte_range->set_page_dirty)
 123 *    ->inode->i_lock		(zap_pte_range->set_page_dirty)
 124 *    ->private_lock		(zap_pte_range->block_dirty_folio)
 
 
 
 125 */
 126
 127static void mapping_set_update(struct xa_state *xas,
 128		struct address_space *mapping)
 129{
 130	if (dax_mapping(mapping) || shmem_mapping(mapping))
 131		return;
 132	xas_set_update(xas, workingset_update_node);
 133	xas_set_lru(xas, &shadow_nodes);
 134}
 135
 136static void page_cache_delete(struct address_space *mapping,
 137				   struct folio *folio, void *shadow)
 138{
 139	XA_STATE(xas, &mapping->i_pages, folio->index);
 140	long nr = 1;
 141
 142	mapping_set_update(&xas, mapping);
 143
 144	xas_set_order(&xas, folio->index, folio_order(folio));
 145	nr = folio_nr_pages(folio);
 
 
 
 146
 147	VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
 
 
 148
 149	xas_store(&xas, shadow);
 150	xas_init_marks(&xas);
 151
 152	folio->mapping = NULL;
 153	/* Leave page->index set: truncation lookup relies upon it */
 
 
 
 
 
 
 
 
 
 
 
 154	mapping->nrpages -= nr;
 155}
 156
 157static void filemap_unaccount_folio(struct address_space *mapping,
 158		struct folio *folio)
 159{
 160	long nr;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 161
 162	VM_BUG_ON_FOLIO(folio_mapped(folio), folio);
 163	if (!IS_ENABLED(CONFIG_DEBUG_VM) && unlikely(folio_mapped(folio))) {
 164		pr_alert("BUG: Bad page cache in process %s  pfn:%05lx\n",
 165			 current->comm, folio_pfn(folio));
 166		dump_page(&folio->page, "still mapped when deleted");
 167		dump_stack();
 168		add_taint(TAINT_BAD_PAGE, LOCKDEP_NOW_UNRELIABLE);
 169
 170		if (mapping_exiting(mapping) && !folio_test_large(folio)) {
 171			int mapcount = page_mapcount(&folio->page);
 172
 173			if (folio_ref_count(folio) >= mapcount + 2) {
 174				/*
 175				 * All vmas have already been torn down, so it's
 176				 * a good bet that actually the page is unmapped
 177				 * and we'd rather not leak it: if we're wrong,
 178				 * another bad page check should catch it later.
 179				 */
 180				page_mapcount_reset(&folio->page);
 181				folio_ref_sub(folio, mapcount);
 182			}
 183		}
 184	}
 185
 186	/* hugetlb folios do not participate in page cache accounting. */
 187	if (folio_test_hugetlb(folio))
 188		return;
 189
 190	nr = folio_nr_pages(folio);
 191
 192	__lruvec_stat_mod_folio(folio, NR_FILE_PAGES, -nr);
 193	if (folio_test_swapbacked(folio)) {
 194		__lruvec_stat_mod_folio(folio, NR_SHMEM, -nr);
 195		if (folio_test_pmd_mappable(folio))
 196			__lruvec_stat_mod_folio(folio, NR_SHMEM_THPS, -nr);
 197	} else if (folio_test_pmd_mappable(folio)) {
 198		__lruvec_stat_mod_folio(folio, NR_FILE_THPS, -nr);
 199		filemap_nr_thps_dec(mapping);
 200	}
 201
 202	/*
 203	 * At this point folio must be either written or cleaned by
 204	 * truncate.  Dirty folio here signals a bug and loss of
 205	 * unwritten data - on ordinary filesystems.
 206	 *
 207	 * But it's harmless on in-memory filesystems like tmpfs; and can
 208	 * occur when a driver which did get_user_pages() sets page dirty
 209	 * before putting it, while the inode is being finally evicted.
 210	 *
 211	 * Below fixes dirty accounting after removing the folio entirely
 212	 * but leaves the dirty flag set: it has no effect for truncated
 213	 * folio and anyway will be cleared before returning folio to
 214	 * buddy allocator.
 215	 */
 216	if (WARN_ON_ONCE(folio_test_dirty(folio) &&
 217			 mapping_can_writeback(mapping)))
 218		folio_account_cleaned(folio, inode_to_wb(mapping->host));
 219}
 220
 221/*
 222 * Delete a page from the page cache and free it. Caller has to make
 223 * sure the page is locked and that nobody else uses it - or that usage
 224 * is safe.  The caller must hold the i_pages lock.
 225 */
 226void __filemap_remove_folio(struct folio *folio, void *shadow)
 227{
 228	struct address_space *mapping = folio->mapping;
 
 
 229
 230	trace_mm_filemap_delete_from_page_cache(folio);
 231	filemap_unaccount_folio(mapping, folio);
 232	page_cache_delete(mapping, folio, shadow);
 233}
 234
 235void filemap_free_folio(struct address_space *mapping, struct folio *folio)
 
 236{
 237	void (*free_folio)(struct folio *);
 238	int refs = 1;
 239
 240	free_folio = mapping->a_ops->free_folio;
 241	if (free_folio)
 242		free_folio(folio);
 243
 244	if (folio_test_large(folio))
 245		refs = folio_nr_pages(folio);
 246	folio_put_refs(folio, refs);
 
 
 
 247}
 248
 249/**
 250 * filemap_remove_folio - Remove folio from page cache.
 251 * @folio: The folio.
 252 *
 253 * This must be called only on folios that are locked and have been
 254 * verified to be in the page cache.  It will never put the folio into
 255 * the free list because the caller has a reference on the page.
 256 */
 257void filemap_remove_folio(struct folio *folio)
 258{
 259	struct address_space *mapping = folio->mapping;
 260
 261	BUG_ON(!folio_test_locked(folio));
 262	spin_lock(&mapping->host->i_lock);
 263	xa_lock_irq(&mapping->i_pages);
 264	__filemap_remove_folio(folio, NULL);
 265	xa_unlock_irq(&mapping->i_pages);
 266	if (mapping_shrinkable(mapping))
 267		inode_add_lru(mapping->host);
 268	spin_unlock(&mapping->host->i_lock);
 269
 270	filemap_free_folio(mapping, folio);
 271}
 272
 273/*
 274 * page_cache_delete_batch - delete several folios from page cache
 275 * @mapping: the mapping to which folios belong
 276 * @fbatch: batch of folios to delete
 277 *
 278 * The function walks over mapping->i_pages and removes folios passed in
 279 * @fbatch from the mapping. The function expects @fbatch to be sorted
 280 * by page index and is optimised for it to be dense.
 281 * It tolerates holes in @fbatch (mapping entries at those indices are not
 282 * modified).
 
 
 
 
 
 
 
 
 
 
 283 *
 284 * The function expects the i_pages lock to be held.
 285 */
 286static void page_cache_delete_batch(struct address_space *mapping,
 287			     struct folio_batch *fbatch)
 288{
 289	XA_STATE(xas, &mapping->i_pages, fbatch->folios[0]->index);
 290	long total_pages = 0;
 291	int i = 0;
 292	struct folio *folio;
 293
 294	mapping_set_update(&xas, mapping);
 295	xas_for_each(&xas, folio, ULONG_MAX) {
 296		if (i >= folio_batch_count(fbatch))
 297			break;
 298
 299		/* A swap/dax/shadow entry got inserted? Skip it. */
 300		if (xa_is_value(folio))
 301			continue;
 302		/*
 303		 * A page got inserted in our range? Skip it. We have our
 304		 * pages locked so they are protected from being removed.
 305		 * If we see a page whose index is higher than ours, it
 306		 * means our page has been removed, which shouldn't be
 307		 * possible because we're holding the PageLock.
 308		 */
 309		if (folio != fbatch->folios[i]) {
 310			VM_BUG_ON_FOLIO(folio->index >
 311					fbatch->folios[i]->index, folio);
 312			continue;
 313		}
 314
 315		WARN_ON_ONCE(!folio_test_locked(folio));
 316
 317		folio->mapping = NULL;
 318		/* Leave folio->index set: truncation lookup relies on it */
 
 319
 320		i++;
 
 
 
 
 
 
 321		xas_store(&xas, NULL);
 322		total_pages += folio_nr_pages(folio);
 323	}
 324	mapping->nrpages -= total_pages;
 325}
 326
 327void delete_from_page_cache_batch(struct address_space *mapping,
 328				  struct folio_batch *fbatch)
 329{
 330	int i;
 
 331
 332	if (!folio_batch_count(fbatch))
 333		return;
 334
 335	spin_lock(&mapping->host->i_lock);
 336	xa_lock_irq(&mapping->i_pages);
 337	for (i = 0; i < folio_batch_count(fbatch); i++) {
 338		struct folio *folio = fbatch->folios[i];
 339
 340		trace_mm_filemap_delete_from_page_cache(folio);
 341		filemap_unaccount_folio(mapping, folio);
 342	}
 343	page_cache_delete_batch(mapping, fbatch);
 344	xa_unlock_irq(&mapping->i_pages);
 345	if (mapping_shrinkable(mapping))
 346		inode_add_lru(mapping->host);
 347	spin_unlock(&mapping->host->i_lock);
 348
 349	for (i = 0; i < folio_batch_count(fbatch); i++)
 350		filemap_free_folio(mapping, fbatch->folios[i]);
 351}
 352
 353int filemap_check_errors(struct address_space *mapping)
 354{
 355	int ret = 0;
 356	/* Check for outstanding write errors */
 357	if (test_bit(AS_ENOSPC, &mapping->flags) &&
 358	    test_and_clear_bit(AS_ENOSPC, &mapping->flags))
 359		ret = -ENOSPC;
 360	if (test_bit(AS_EIO, &mapping->flags) &&
 361	    test_and_clear_bit(AS_EIO, &mapping->flags))
 362		ret = -EIO;
 363	return ret;
 364}
 365EXPORT_SYMBOL(filemap_check_errors);
 366
 367static int filemap_check_and_keep_errors(struct address_space *mapping)
 368{
 369	/* Check for outstanding write errors */
 370	if (test_bit(AS_EIO, &mapping->flags))
 371		return -EIO;
 372	if (test_bit(AS_ENOSPC, &mapping->flags))
 373		return -ENOSPC;
 374	return 0;
 375}
 376
 377/**
 378 * filemap_fdatawrite_wbc - start writeback on mapping dirty pages in range
 379 * @mapping:	address space structure to write
 380 * @wbc:	the writeback_control controlling the writeout
 381 *
 382 * Call writepages on the mapping using the provided wbc to control the
 383 * writeout.
 384 *
 385 * Return: %0 on success, negative error code otherwise.
 386 */
 387int filemap_fdatawrite_wbc(struct address_space *mapping,
 388			   struct writeback_control *wbc)
 389{
 390	int ret;
 391
 392	if (!mapping_can_writeback(mapping) ||
 393	    !mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
 394		return 0;
 395
 396	wbc_attach_fdatawrite_inode(wbc, mapping->host);
 397	ret = do_writepages(mapping, wbc);
 398	wbc_detach_inode(wbc);
 399	return ret;
 400}
 401EXPORT_SYMBOL(filemap_fdatawrite_wbc);
 402
 403/**
 404 * __filemap_fdatawrite_range - start writeback on mapping dirty pages in range
 405 * @mapping:	address space structure to write
 406 * @start:	offset in bytes where the range starts
 407 * @end:	offset in bytes where the range ends (inclusive)
 408 * @sync_mode:	enable synchronous operation
 409 *
 410 * Start writeback against all of a mapping's dirty pages that lie
 411 * within the byte offsets <start, end> inclusive.
 412 *
 413 * If sync_mode is WB_SYNC_ALL then this is a "data integrity" operation, as
 414 * opposed to a regular memory cleansing writeback.  The difference between
 415 * these two operations is that if a dirty page/buffer is encountered, it must
 416 * be waited upon, and not just skipped over.
 417 *
 418 * Return: %0 on success, negative error code otherwise.
 419 */
 420int __filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
 421				loff_t end, int sync_mode)
 422{
 
 423	struct writeback_control wbc = {
 424		.sync_mode = sync_mode,
 425		.nr_to_write = LONG_MAX,
 426		.range_start = start,
 427		.range_end = end,
 428	};
 429
 430	return filemap_fdatawrite_wbc(mapping, &wbc);
 
 
 
 
 
 
 
 431}
 432
 433static inline int __filemap_fdatawrite(struct address_space *mapping,
 434	int sync_mode)
 435{
 436	return __filemap_fdatawrite_range(mapping, 0, LLONG_MAX, sync_mode);
 437}
 438
 439int filemap_fdatawrite(struct address_space *mapping)
 440{
 441	return __filemap_fdatawrite(mapping, WB_SYNC_ALL);
 442}
 443EXPORT_SYMBOL(filemap_fdatawrite);
 444
 445int filemap_fdatawrite_range(struct address_space *mapping, loff_t start,
 446				loff_t end)
 447{
 448	return __filemap_fdatawrite_range(mapping, start, end, WB_SYNC_ALL);
 449}
 450EXPORT_SYMBOL(filemap_fdatawrite_range);
 451
 452/**
 453 * filemap_flush - mostly a non-blocking flush
 454 * @mapping:	target address_space
 455 *
 456 * This is a mostly non-blocking flush.  Not suitable for data-integrity
 457 * purposes - I/O may not be started against all dirty pages.
 458 *
 459 * Return: %0 on success, negative error code otherwise.
 460 */
 461int filemap_flush(struct address_space *mapping)
 462{
 463	return __filemap_fdatawrite(mapping, WB_SYNC_NONE);
 464}
 465EXPORT_SYMBOL(filemap_flush);
 466
 467/**
 468 * filemap_range_has_page - check if a page exists in range.
 469 * @mapping:           address space within which to check
 470 * @start_byte:        offset in bytes where the range starts
 471 * @end_byte:          offset in bytes where the range ends (inclusive)
 472 *
 473 * Find at least one page in the range supplied, usually used to check if
 474 * direct writing in this range will trigger a writeback.
 475 *
 476 * Return: %true if at least one page exists in the specified range,
 477 * %false otherwise.
 478 */
 479bool filemap_range_has_page(struct address_space *mapping,
 480			   loff_t start_byte, loff_t end_byte)
 481{
 482	struct folio *folio;
 483	XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
 484	pgoff_t max = end_byte >> PAGE_SHIFT;
 485
 486	if (end_byte < start_byte)
 487		return false;
 488
 489	rcu_read_lock();
 490	for (;;) {
 491		folio = xas_find(&xas, max);
 492		if (xas_retry(&xas, folio))
 493			continue;
 494		/* Shadow entries don't count */
 495		if (xa_is_value(folio))
 496			continue;
 497		/*
 498		 * We don't need to try to pin this page; we're about to
 499		 * release the RCU lock anyway.  It is enough to know that
 500		 * there was a page here recently.
 501		 */
 502		break;
 503	}
 504	rcu_read_unlock();
 505
 506	return folio != NULL;
 507}
 508EXPORT_SYMBOL(filemap_range_has_page);
 509
 510static void __filemap_fdatawait_range(struct address_space *mapping,
 511				     loff_t start_byte, loff_t end_byte)
 512{
 513	pgoff_t index = start_byte >> PAGE_SHIFT;
 514	pgoff_t end = end_byte >> PAGE_SHIFT;
 515	struct folio_batch fbatch;
 516	unsigned nr_folios;
 517
 518	folio_batch_init(&fbatch);
 
 519
 
 520	while (index <= end) {
 521		unsigned i;
 522
 523		nr_folios = filemap_get_folios_tag(mapping, &index, end,
 524				PAGECACHE_TAG_WRITEBACK, &fbatch);
 525
 526		if (!nr_folios)
 527			break;
 528
 529		for (i = 0; i < nr_folios; i++) {
 530			struct folio *folio = fbatch.folios[i];
 531
 532			folio_wait_writeback(folio);
 533			folio_clear_error(folio);
 534		}
 535		folio_batch_release(&fbatch);
 536		cond_resched();
 537	}
 538}
 539
 540/**
 541 * filemap_fdatawait_range - wait for writeback to complete
 542 * @mapping:		address space structure to wait for
 543 * @start_byte:		offset in bytes where the range starts
 544 * @end_byte:		offset in bytes where the range ends (inclusive)
 545 *
 546 * Walk the list of under-writeback pages of the given address space
 547 * in the given range and wait for all of them.  Check error status of
 548 * the address space and return it.
 549 *
 550 * Since the error status of the address space is cleared by this function,
 551 * callers are responsible for checking the return value and handling and/or
 552 * reporting the error.
 553 *
 554 * Return: error status of the address space.
 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 * filemap_fdatawait_range_keep_errors - wait for writeback to complete
 566 * @mapping:		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 given address space in the
 571 * given range and wait for all of them.  Unlike filemap_fdatawait_range(),
 572 * this function does not clear error status of the address space.
 573 *
 574 * Use this function if callers don't handle errors themselves.  Expected
 575 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
 576 * fsfreeze(8)
 577 */
 578int filemap_fdatawait_range_keep_errors(struct address_space *mapping,
 579		loff_t start_byte, loff_t end_byte)
 580{
 581	__filemap_fdatawait_range(mapping, start_byte, end_byte);
 582	return filemap_check_and_keep_errors(mapping);
 583}
 584EXPORT_SYMBOL(filemap_fdatawait_range_keep_errors);
 585
 586/**
 587 * file_fdatawait_range - wait for writeback to complete
 588 * @file:		file pointing to address space structure to wait for
 589 * @start_byte:		offset in bytes where the range starts
 590 * @end_byte:		offset in bytes where the range ends (inclusive)
 591 *
 592 * Walk the list of under-writeback pages of the address space that file
 593 * refers to, in the given range and wait for all of them.  Check error
 594 * status of the address space vs. the file->f_wb_err cursor and return it.
 595 *
 596 * Since the error status of the file is advanced by this function,
 597 * callers are responsible for checking the return value and handling and/or
 598 * reporting the error.
 599 *
 600 * Return: error status of the address space vs. the file->f_wb_err cursor.
 601 */
 602int file_fdatawait_range(struct file *file, loff_t start_byte, loff_t end_byte)
 603{
 604	struct address_space *mapping = file->f_mapping;
 605
 606	__filemap_fdatawait_range(mapping, start_byte, end_byte);
 607	return file_check_and_advance_wb_err(file);
 608}
 609EXPORT_SYMBOL(file_fdatawait_range);
 610
 611/**
 612 * filemap_fdatawait_keep_errors - wait for writeback without clearing errors
 613 * @mapping: address space structure to wait for
 614 *
 615 * Walk the list of under-writeback pages of the given address space
 616 * and wait for all of them.  Unlike filemap_fdatawait(), this function
 617 * does not clear error status of the address space.
 618 *
 619 * Use this function if callers don't handle errors themselves.  Expected
 620 * call sites are system-wide / filesystem-wide data flushers: e.g. sync(2),
 621 * fsfreeze(8)
 622 *
 623 * Return: error status of the address space.
 624 */
 625int filemap_fdatawait_keep_errors(struct address_space *mapping)
 626{
 627	__filemap_fdatawait_range(mapping, 0, LLONG_MAX);
 628	return filemap_check_and_keep_errors(mapping);
 629}
 630EXPORT_SYMBOL(filemap_fdatawait_keep_errors);
 631
 632/* Returns true if writeback might be needed or already in progress. */
 633static bool mapping_needs_writeback(struct address_space *mapping)
 634{
 
 
 
 635	return mapping->nrpages;
 636}
 637
 638bool filemap_range_has_writeback(struct address_space *mapping,
 639				 loff_t start_byte, loff_t end_byte)
 640{
 641	XA_STATE(xas, &mapping->i_pages, start_byte >> PAGE_SHIFT);
 642	pgoff_t max = end_byte >> PAGE_SHIFT;
 643	struct folio *folio;
 644
 645	if (end_byte < start_byte)
 646		return false;
 647
 648	rcu_read_lock();
 649	xas_for_each(&xas, folio, max) {
 650		if (xas_retry(&xas, folio))
 651			continue;
 652		if (xa_is_value(folio))
 653			continue;
 654		if (folio_test_dirty(folio) || folio_test_locked(folio) ||
 655				folio_test_writeback(folio))
 656			break;
 657	}
 658	rcu_read_unlock();
 659	return folio != NULL;
 660}
 661EXPORT_SYMBOL_GPL(filemap_range_has_writeback);
 662
 663/**
 664 * filemap_write_and_wait_range - write out & wait on a file range
 665 * @mapping:	the address_space for the pages
 666 * @lstart:	offset in bytes where the range starts
 667 * @lend:	offset in bytes where the range ends (inclusive)
 668 *
 669 * Write out and wait upon file offsets lstart->lend, inclusive.
 670 *
 671 * Note that @lend is inclusive (describes the last byte to be written) so
 672 * that this function can be used to write to the very end-of-file (end = -1).
 673 *
 674 * Return: error status of the address space.
 675 */
 676int filemap_write_and_wait_range(struct address_space *mapping,
 677				 loff_t lstart, loff_t lend)
 678{
 679	int err = 0, err2;
 680
 681	if (lend < lstart)
 682		return 0;
 683
 684	if (mapping_needs_writeback(mapping)) {
 685		err = __filemap_fdatawrite_range(mapping, lstart, lend,
 686						 WB_SYNC_ALL);
 687		/*
 688		 * Even if the above returned error, the pages may be
 689		 * written partially (e.g. -ENOSPC), so we wait for it.
 690		 * But the -EIO is special case, it may indicate the worst
 691		 * thing (e.g. bug) happened, so we avoid waiting for it.
 692		 */
 693		if (err != -EIO)
 694			__filemap_fdatawait_range(mapping, lstart, lend);
 
 
 
 
 
 
 
 
 
 695	}
 696	err2 = filemap_check_errors(mapping);
 697	if (!err)
 698		err = err2;
 699	return err;
 700}
 701EXPORT_SYMBOL(filemap_write_and_wait_range);
 702
 703void __filemap_set_wb_err(struct address_space *mapping, int err)
 704{
 705	errseq_t eseq = errseq_set(&mapping->wb_err, err);
 706
 707	trace_filemap_set_wb_err(mapping, eseq);
 708}
 709EXPORT_SYMBOL(__filemap_set_wb_err);
 710
 711/**
 712 * file_check_and_advance_wb_err - report wb error (if any) that was previously
 713 * 				   and advance wb_err to current one
 714 * @file: struct file on which the error is being reported
 715 *
 716 * When userland calls fsync (or something like nfsd does the equivalent), we
 717 * want to report any writeback errors that occurred since the last fsync (or
 718 * since the file was opened if there haven't been any).
 719 *
 720 * Grab the wb_err from the mapping. If it matches what we have in the file,
 721 * then just quickly return 0. The file is all caught up.
 722 *
 723 * If it doesn't match, then take the mapping value, set the "seen" flag in
 724 * it and try to swap it into place. If it works, or another task beat us
 725 * to it with the new value, then update the f_wb_err and return the error
 726 * portion. The error at this point must be reported via proper channels
 727 * (a'la fsync, or NFS COMMIT operation, etc.).
 728 *
 729 * While we handle mapping->wb_err with atomic operations, the f_wb_err
 730 * value is protected by the f_lock since we must ensure that it reflects
 731 * the latest value swapped in for this file descriptor.
 732 *
 733 * Return: %0 on success, negative error code otherwise.
 734 */
 735int file_check_and_advance_wb_err(struct file *file)
 736{
 737	int err = 0;
 738	errseq_t old = READ_ONCE(file->f_wb_err);
 739	struct address_space *mapping = file->f_mapping;
 740
 741	/* Locklessly handle the common case where nothing has changed */
 742	if (errseq_check(&mapping->wb_err, old)) {
 743		/* Something changed, must use slow path */
 744		spin_lock(&file->f_lock);
 745		old = file->f_wb_err;
 746		err = errseq_check_and_advance(&mapping->wb_err,
 747						&file->f_wb_err);
 748		trace_file_check_and_advance_wb_err(file, old);
 749		spin_unlock(&file->f_lock);
 750	}
 751
 752	/*
 753	 * We're mostly using this function as a drop in replacement for
 754	 * filemap_check_errors. Clear AS_EIO/AS_ENOSPC to emulate the effect
 755	 * that the legacy code would have had on these flags.
 756	 */
 757	clear_bit(AS_EIO, &mapping->flags);
 758	clear_bit(AS_ENOSPC, &mapping->flags);
 759	return err;
 760}
 761EXPORT_SYMBOL(file_check_and_advance_wb_err);
 762
 763/**
 764 * file_write_and_wait_range - write out & wait on a file range
 765 * @file:	file pointing to address_space with pages
 766 * @lstart:	offset in bytes where the range starts
 767 * @lend:	offset in bytes where the range ends (inclusive)
 768 *
 769 * Write out and wait upon file offsets lstart->lend, inclusive.
 770 *
 771 * Note that @lend is inclusive (describes the last byte to be written) so
 772 * that this function can be used to write to the very end-of-file (end = -1).
 773 *
 774 * After writing out and waiting on the data, we check and advance the
 775 * f_wb_err cursor to the latest value, and return any errors detected there.
 776 *
 777 * Return: %0 on success, negative error code otherwise.
 778 */
 779int file_write_and_wait_range(struct file *file, loff_t lstart, loff_t lend)
 780{
 781	int err = 0, err2;
 782	struct address_space *mapping = file->f_mapping;
 783
 784	if (lend < lstart)
 785		return 0;
 786
 787	if (mapping_needs_writeback(mapping)) {
 788		err = __filemap_fdatawrite_range(mapping, lstart, lend,
 789						 WB_SYNC_ALL);
 790		/* See comment of filemap_write_and_wait() */
 791		if (err != -EIO)
 792			__filemap_fdatawait_range(mapping, lstart, lend);
 793	}
 794	err2 = file_check_and_advance_wb_err(file);
 795	if (!err)
 796		err = err2;
 797	return err;
 798}
 799EXPORT_SYMBOL(file_write_and_wait_range);
 800
 801/**
 802 * replace_page_cache_folio - replace a pagecache folio with a new one
 803 * @old:	folio to be replaced
 804 * @new:	folio to replace with
 805 *
 806 * This function replaces a folio in the pagecache with a new one.  On
 807 * success it acquires the pagecache reference for the new folio and
 808 * drops it for the old folio.  Both the old and new folios must be
 809 * locked.  This function does not add the new folio to the LRU, the
 
 810 * caller must do that.
 811 *
 812 * The remove + add is atomic.  This function cannot fail.
 
 
 813 */
 814void replace_page_cache_folio(struct folio *old, struct folio *new)
 815{
 816	struct address_space *mapping = old->mapping;
 817	void (*free_folio)(struct folio *) = mapping->a_ops->free_folio;
 818	pgoff_t offset = old->index;
 819	XA_STATE(xas, &mapping->i_pages, offset);
 
 820
 821	VM_BUG_ON_FOLIO(!folio_test_locked(old), old);
 822	VM_BUG_ON_FOLIO(!folio_test_locked(new), new);
 823	VM_BUG_ON_FOLIO(new->mapping, new);
 824
 825	folio_get(new);
 826	new->mapping = mapping;
 827	new->index = offset;
 828
 829	mem_cgroup_replace_folio(old, new);
 830
 831	xas_lock_irq(&xas);
 832	xas_store(&xas, new);
 833
 834	old->mapping = NULL;
 835	/* hugetlb pages do not participate in page cache accounting. */
 836	if (!folio_test_hugetlb(old))
 837		__lruvec_stat_sub_folio(old, NR_FILE_PAGES);
 838	if (!folio_test_hugetlb(new))
 839		__lruvec_stat_add_folio(new, NR_FILE_PAGES);
 840	if (folio_test_swapbacked(old))
 841		__lruvec_stat_sub_folio(old, NR_SHMEM);
 842	if (folio_test_swapbacked(new))
 843		__lruvec_stat_add_folio(new, NR_SHMEM);
 844	xas_unlock_irq(&xas);
 845	if (free_folio)
 846		free_folio(old);
 847	folio_put(old);
 
 
 848}
 849EXPORT_SYMBOL_GPL(replace_page_cache_folio);
 850
 851noinline int __filemap_add_folio(struct address_space *mapping,
 852		struct folio *folio, pgoff_t index, gfp_t gfp, void **shadowp)
 
 
 853{
 854	XA_STATE(xas, &mapping->i_pages, index);
 855	bool huge = folio_test_hugetlb(folio);
 856	bool charged = false;
 857	long nr = 1;
 858
 859	VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
 860	VM_BUG_ON_FOLIO(folio_test_swapbacked(folio), folio);
 861	mapping_set_update(&xas, mapping);
 862
 
 
 
 
 863	if (!huge) {
 864		int error = mem_cgroup_charge(folio, NULL, gfp);
 865		if (error)
 866			return error;
 867		charged = true;
 868	}
 869
 870	VM_BUG_ON_FOLIO(index & (folio_nr_pages(folio) - 1), folio);
 871	xas_set_order(&xas, index, folio_order(folio));
 872	nr = folio_nr_pages(folio);
 873
 874	gfp &= GFP_RECLAIM_MASK;
 875	folio_ref_add(folio, nr);
 876	folio->mapping = mapping;
 877	folio->index = xas.xa_index;
 878
 879	do {
 880		unsigned int order = xa_get_order(xas.xa, xas.xa_index);
 881		void *entry, *old = NULL;
 882
 883		if (order > folio_order(folio))
 884			xas_split_alloc(&xas, xa_load(xas.xa, xas.xa_index),
 885					order, gfp);
 886		xas_lock_irq(&xas);
 887		xas_for_each_conflict(&xas, entry) {
 888			old = entry;
 889			if (!xa_is_value(entry)) {
 890				xas_set_err(&xas, -EEXIST);
 891				goto unlock;
 892			}
 893		}
 894
 895		if (old) {
 
 896			if (shadowp)
 897				*shadowp = old;
 898			/* entry may have been split before we acquired lock */
 899			order = xa_get_order(xas.xa, xas.xa_index);
 900			if (order > folio_order(folio)) {
 901				/* How to handle large swap entries? */
 902				BUG_ON(shmem_mapping(mapping));
 903				xas_split(&xas, old, order);
 904				xas_reset(&xas);
 905			}
 906		}
 907
 908		xas_store(&xas, folio);
 909		if (xas_error(&xas))
 910			goto unlock;
 911
 912		mapping->nrpages += nr;
 913
 914		/* hugetlb pages do not participate in page cache accounting */
 915		if (!huge) {
 916			__lruvec_stat_mod_folio(folio, NR_FILE_PAGES, nr);
 917			if (folio_test_pmd_mappable(folio))
 918				__lruvec_stat_mod_folio(folio,
 919						NR_FILE_THPS, nr);
 920		}
 921unlock:
 922		xas_unlock_irq(&xas);
 923	} while (xas_nomem(&xas, gfp));
 924
 925	if (xas_error(&xas))
 
 926		goto error;
 
 927
 928	trace_mm_filemap_add_to_page_cache(folio);
 929	return 0;
 930error:
 931	if (charged)
 932		mem_cgroup_uncharge(folio);
 933	folio->mapping = NULL;
 934	/* Leave page->index set: truncation relies upon it */
 935	folio_put_refs(folio, nr);
 936	return xas_error(&xas);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 937}
 938ALLOW_ERROR_INJECTION(__filemap_add_folio, ERRNO);
 939
 940int filemap_add_folio(struct address_space *mapping, struct folio *folio,
 941				pgoff_t index, gfp_t gfp)
 942{
 943	void *shadow = NULL;
 944	int ret;
 945
 946	__folio_set_locked(folio);
 947	ret = __filemap_add_folio(mapping, folio, index, gfp, &shadow);
 
 948	if (unlikely(ret))
 949		__folio_clear_locked(folio);
 950	else {
 951		/*
 952		 * The folio might have been evicted from cache only
 953		 * recently, in which case it should be activated like
 954		 * any other repeatedly accessed folio.
 955		 * The exception is folios getting rewritten; evicting other
 956		 * data from the working set, only to cache data that will
 957		 * get overwritten with something else, is a waste of memory.
 958		 */
 959		WARN_ON_ONCE(folio_test_active(folio));
 960		if (!(gfp & __GFP_WRITE) && shadow)
 961			workingset_refault(folio, shadow);
 962		folio_add_lru(folio);
 963	}
 964	return ret;
 965}
 966EXPORT_SYMBOL_GPL(filemap_add_folio);
 967
 968#ifdef CONFIG_NUMA
 969struct folio *filemap_alloc_folio(gfp_t gfp, unsigned int order)
 970{
 971	int n;
 972	struct folio *folio;
 973
 974	if (cpuset_do_page_mem_spread()) {
 975		unsigned int cpuset_mems_cookie;
 976		do {
 977			cpuset_mems_cookie = read_mems_allowed_begin();
 978			n = cpuset_mem_spread_node();
 979			folio = __folio_alloc_node(gfp, order, n);
 980		} while (!folio && read_mems_allowed_retry(cpuset_mems_cookie));
 981
 982		return folio;
 983	}
 984	return folio_alloc(gfp, order);
 985}
 986EXPORT_SYMBOL(filemap_alloc_folio);
 987#endif
 988
 989/*
 990 * filemap_invalidate_lock_two - lock invalidate_lock for two mappings
 991 *
 992 * Lock exclusively invalidate_lock of any passed mapping that is not NULL.
 993 *
 994 * @mapping1: the first mapping to lock
 995 * @mapping2: the second mapping to lock
 996 */
 997void filemap_invalidate_lock_two(struct address_space *mapping1,
 998				 struct address_space *mapping2)
 999{
1000	if (mapping1 > mapping2)
1001		swap(mapping1, mapping2);
1002	if (mapping1)
1003		down_write(&mapping1->invalidate_lock);
1004	if (mapping2 && mapping1 != mapping2)
1005		down_write_nested(&mapping2->invalidate_lock, 1);
1006}
1007EXPORT_SYMBOL(filemap_invalidate_lock_two);
1008
1009/*
1010 * filemap_invalidate_unlock_two - unlock invalidate_lock for two mappings
1011 *
1012 * Unlock exclusive invalidate_lock of any passed mapping that is not NULL.
1013 *
1014 * @mapping1: the first mapping to unlock
1015 * @mapping2: the second mapping to unlock
1016 */
1017void filemap_invalidate_unlock_two(struct address_space *mapping1,
1018				   struct address_space *mapping2)
1019{
1020	if (mapping1)
1021		up_write(&mapping1->invalidate_lock);
1022	if (mapping2 && mapping1 != mapping2)
1023		up_write(&mapping2->invalidate_lock);
1024}
1025EXPORT_SYMBOL(filemap_invalidate_unlock_two);
1026
1027/*
1028 * In order to wait for pages to become available there must be
1029 * waitqueues associated with pages. By using a hash table of
1030 * waitqueues where the bucket discipline is to maintain all
1031 * waiters on the same queue and wake all when any of the pages
1032 * become available, and for the woken contexts to check to be
1033 * sure the appropriate page became available, this saves space
1034 * at a cost of "thundering herd" phenomena during rare hash
1035 * collisions.
1036 */
1037#define PAGE_WAIT_TABLE_BITS 8
1038#define PAGE_WAIT_TABLE_SIZE (1 << PAGE_WAIT_TABLE_BITS)
1039static wait_queue_head_t folio_wait_table[PAGE_WAIT_TABLE_SIZE] __cacheline_aligned;
1040
1041static wait_queue_head_t *folio_waitqueue(struct folio *folio)
1042{
1043	return &folio_wait_table[hash_ptr(folio, PAGE_WAIT_TABLE_BITS)];
1044}
1045
1046void __init pagecache_init(void)
1047{
1048	int i;
1049
1050	for (i = 0; i < PAGE_WAIT_TABLE_SIZE; i++)
1051		init_waitqueue_head(&folio_wait_table[i]);
1052
1053	page_writeback_init();
1054}
1055
1056/*
1057 * The page wait code treats the "wait->flags" somewhat unusually, because
1058 * we have multiple different kinds of waits, not just the usual "exclusive"
1059 * one.
1060 *
1061 * We have:
1062 *
1063 *  (a) no special bits set:
1064 *
1065 *	We're just waiting for the bit to be released, and when a waker
1066 *	calls the wakeup function, we set WQ_FLAG_WOKEN and wake it up,
1067 *	and remove it from the wait queue.
1068 *
1069 *	Simple and straightforward.
1070 *
1071 *  (b) WQ_FLAG_EXCLUSIVE:
1072 *
1073 *	The waiter is waiting to get the lock, and only one waiter should
1074 *	be woken up to avoid any thundering herd behavior. We'll set the
1075 *	WQ_FLAG_WOKEN bit, wake it up, and remove it from the wait queue.
1076 *
1077 *	This is the traditional exclusive wait.
1078 *
1079 *  (c) WQ_FLAG_EXCLUSIVE | WQ_FLAG_CUSTOM:
1080 *
1081 *	The waiter is waiting to get the bit, and additionally wants the
1082 *	lock to be transferred to it for fair lock behavior. If the lock
1083 *	cannot be taken, we stop walking the wait queue without waking
1084 *	the waiter.
1085 *
1086 *	This is the "fair lock handoff" case, and in addition to setting
1087 *	WQ_FLAG_WOKEN, we set WQ_FLAG_DONE to let the waiter easily see
1088 *	that it now has the lock.
1089 */
1090static int wake_page_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *arg)
1091{
1092	unsigned int flags;
1093	struct wait_page_key *key = arg;
1094	struct wait_page_queue *wait_page
1095		= container_of(wait, struct wait_page_queue, wait);
1096
1097	if (!wake_page_match(wait_page, key))
1098		return 0;
1099
1100	/*
1101	 * If it's a lock handoff wait, we get the bit for it, and
1102	 * stop walking (and do not wake it up) if we can't.
1103	 */
1104	flags = wait->flags;
1105	if (flags & WQ_FLAG_EXCLUSIVE) {
1106		if (test_bit(key->bit_nr, &key->folio->flags))
1107			return -1;
1108		if (flags & WQ_FLAG_CUSTOM) {
1109			if (test_and_set_bit(key->bit_nr, &key->folio->flags))
1110				return -1;
1111			flags |= WQ_FLAG_DONE;
1112		}
1113	}
1114
1115	/*
1116	 * We are holding the wait-queue lock, but the waiter that
1117	 * is waiting for this will be checking the flags without
1118	 * any locking.
1119	 *
1120	 * So update the flags atomically, and wake up the waiter
1121	 * afterwards to avoid any races. This store-release pairs
1122	 * with the load-acquire in folio_wait_bit_common().
1123	 */
1124	smp_store_release(&wait->flags, flags | WQ_FLAG_WOKEN);
1125	wake_up_state(wait->private, mode);
1126
1127	/*
1128	 * Ok, we have successfully done what we're waiting for,
1129	 * and we can unconditionally remove the wait entry.
1130	 *
1131	 * Note that this pairs with the "finish_wait()" in the
1132	 * waiter, and has to be the absolute last thing we do.
1133	 * After this list_del_init(&wait->entry) the wait entry
1134	 * might be de-allocated and the process might even have
1135	 * exited.
1136	 */
1137	list_del_init_careful(&wait->entry);
1138	return (flags & WQ_FLAG_EXCLUSIVE) != 0;
1139}
1140
1141static void folio_wake_bit(struct folio *folio, int bit_nr)
1142{
1143	wait_queue_head_t *q = folio_waitqueue(folio);
1144	struct wait_page_key key;
1145	unsigned long flags;
 
1146
1147	key.folio = folio;
1148	key.bit_nr = bit_nr;
1149	key.page_match = 0;
1150
 
 
 
 
 
1151	spin_lock_irqsave(&q->lock, flags);
1152	__wake_up_locked_key(q, TASK_NORMAL, &key);
 
 
 
 
 
 
 
 
 
 
 
 
 
1153
1154	/*
1155	 * It's possible to miss clearing waiters here, when we woke our page
1156	 * waiters, but the hashed waitqueue has waiters for other pages on it.
1157	 * That's okay, it's a rare case. The next waker will clear it.
1158	 *
1159	 * Note that, depending on the page pool (buddy, hugetlb, ZONE_DEVICE,
1160	 * other), the flag may be cleared in the course of freeing the page;
1161	 * but that is not required for correctness.
1162	 */
1163	if (!waitqueue_active(q) || !key.page_match)
1164		folio_clear_waiters(folio);
 
 
 
 
 
 
 
 
 
 
1165
1166	spin_unlock_irqrestore(&q->lock, flags);
 
 
 
 
1167}
1168
1169/*
1170 * A choice of three behaviors for folio_wait_bit_common():
1171 */
1172enum behavior {
1173	EXCLUSIVE,	/* Hold ref to page and take the bit when woken, like
1174			 * __folio_lock() waiting on then setting PG_locked.
1175			 */
1176	SHARED,		/* Hold ref to page and check the bit when woken, like
1177			 * folio_wait_writeback() waiting on PG_writeback.
1178			 */
1179	DROP,		/* Drop ref to page before wait, no check when woken,
1180			 * like folio_put_wait_locked() on PG_locked.
1181			 */
1182};
1183
1184/*
1185 * Attempt to check (or get) the folio flag, and mark us done
1186 * if successful.
1187 */
1188static inline bool folio_trylock_flag(struct folio *folio, int bit_nr,
1189					struct wait_queue_entry *wait)
1190{
1191	if (wait->flags & WQ_FLAG_EXCLUSIVE) {
1192		if (test_and_set_bit(bit_nr, &folio->flags))
1193			return false;
1194	} else if (test_bit(bit_nr, &folio->flags))
1195		return false;
1196
1197	wait->flags |= WQ_FLAG_WOKEN | WQ_FLAG_DONE;
1198	return true;
1199}
1200
1201/* How many times do we accept lock stealing from under a waiter? */
1202int sysctl_page_lock_unfairness = 5;
1203
1204static inline int folio_wait_bit_common(struct folio *folio, int bit_nr,
1205		int state, enum behavior behavior)
1206{
1207	wait_queue_head_t *q = folio_waitqueue(folio);
1208	int unfairness = sysctl_page_lock_unfairness;
1209	struct wait_page_queue wait_page;
1210	wait_queue_entry_t *wait = &wait_page.wait;
1211	bool thrashing = false;
 
1212	unsigned long pflags;
1213	bool in_thrashing;
1214
1215	if (bit_nr == PG_locked &&
1216	    !folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1217		delayacct_thrashing_start(&in_thrashing);
 
 
 
1218		psi_memstall_enter(&pflags);
1219		thrashing = true;
1220	}
1221
1222	init_wait(wait);
1223	wait->func = wake_page_function;
1224	wait_page.folio = folio;
1225	wait_page.bit_nr = bit_nr;
1226
1227repeat:
1228	wait->flags = 0;
1229	if (behavior == EXCLUSIVE) {
1230		wait->flags = WQ_FLAG_EXCLUSIVE;
1231		if (--unfairness < 0)
1232			wait->flags |= WQ_FLAG_CUSTOM;
1233	}
1234
1235	/*
1236	 * Do one last check whether we can get the
1237	 * page bit synchronously.
1238	 *
1239	 * Do the folio_set_waiters() marking before that
1240	 * to let any waker we _just_ missed know they
1241	 * need to wake us up (otherwise they'll never
1242	 * even go to the slow case that looks at the
1243	 * page queue), and add ourselves to the wait
1244	 * queue if we need to sleep.
1245	 *
1246	 * This part needs to be done under the queue
1247	 * lock to avoid races.
1248	 */
1249	spin_lock_irq(&q->lock);
1250	folio_set_waiters(folio);
1251	if (!folio_trylock_flag(folio, bit_nr, wait))
1252		__add_wait_queue_entry_tail(q, wait);
1253	spin_unlock_irq(&q->lock);
1254
1255	/*
1256	 * From now on, all the logic will be based on
1257	 * the WQ_FLAG_WOKEN and WQ_FLAG_DONE flag, to
1258	 * see whether the page bit testing has already
1259	 * been done by the wake function.
1260	 *
1261	 * We can drop our reference to the folio.
1262	 */
1263	if (behavior == DROP)
1264		folio_put(folio);
1265
1266	/*
1267	 * Note that until the "finish_wait()", or until
1268	 * we see the WQ_FLAG_WOKEN flag, we need to
1269	 * be very careful with the 'wait->flags', because
1270	 * we may race with a waker that sets them.
1271	 */
1272	for (;;) {
1273		unsigned int flags;
1274
1275		set_current_state(state);
1276
1277		/* Loop until we've been woken or interrupted */
1278		flags = smp_load_acquire(&wait->flags);
1279		if (!(flags & WQ_FLAG_WOKEN)) {
1280			if (signal_pending_state(state, current))
1281				break;
1282
1283			io_schedule();
1284			continue;
1285		}
1286
1287		/* If we were non-exclusive, we're done */
1288		if (behavior != EXCLUSIVE)
1289			break;
1290
1291		/* If the waker got the lock for us, we're done */
1292		if (flags & WQ_FLAG_DONE)
1293			break;
1294
1295		/*
1296		 * Otherwise, if we're getting the lock, we need to
1297		 * try to get it ourselves.
1298		 *
1299		 * And if that fails, we'll have to retry this all.
1300		 */
1301		if (unlikely(test_and_set_bit(bit_nr, folio_flags(folio, 0))))
1302			goto repeat;
1303
1304		wait->flags |= WQ_FLAG_DONE;
1305		break;
1306	}
1307
1308	/*
1309	 * If a signal happened, this 'finish_wait()' may remove the last
1310	 * waiter from the wait-queues, but the folio waiters bit will remain
1311	 * set. That's ok. The next wakeup will take care of it, and trying
1312	 * to do it here would be difficult and prone to races.
1313	 */
1314	finish_wait(q, wait);
1315
1316	if (thrashing) {
1317		delayacct_thrashing_end(&in_thrashing);
 
1318		psi_memstall_leave(&pflags);
1319	}
1320
1321	/*
1322	 * NOTE! The wait->flags weren't stable until we've done the
1323	 * 'finish_wait()', and we could have exited the loop above due
1324	 * to a signal, and had a wakeup event happen after the signal
1325	 * test but before the 'finish_wait()'.
1326	 *
1327	 * So only after the finish_wait() can we reliably determine
1328	 * if we got woken up or not, so we can now figure out the final
1329	 * return value based on that state without races.
1330	 *
1331	 * Also note that WQ_FLAG_WOKEN is sufficient for a non-exclusive
1332	 * waiter, but an exclusive one requires WQ_FLAG_DONE.
1333	 */
1334	if (behavior == EXCLUSIVE)
1335		return wait->flags & WQ_FLAG_DONE ? 0 : -EINTR;
1336
1337	return wait->flags & WQ_FLAG_WOKEN ? 0 : -EINTR;
1338}
1339
1340#ifdef CONFIG_MIGRATION
1341/**
1342 * migration_entry_wait_on_locked - Wait for a migration entry to be removed
1343 * @entry: migration swap entry.
1344 * @ptl: already locked ptl. This function will drop the lock.
1345 *
1346 * Wait for a migration entry referencing the given page to be removed. This is
1347 * equivalent to put_and_wait_on_page_locked(page, TASK_UNINTERRUPTIBLE) except
1348 * this can be called without taking a reference on the page. Instead this
1349 * should be called while holding the ptl for the migration entry referencing
1350 * the page.
1351 *
1352 * Returns after unlocking the ptl.
1353 *
1354 * This follows the same logic as folio_wait_bit_common() so see the comments
1355 * there.
1356 */
1357void migration_entry_wait_on_locked(swp_entry_t entry, spinlock_t *ptl)
1358	__releases(ptl)
1359{
1360	struct wait_page_queue wait_page;
1361	wait_queue_entry_t *wait = &wait_page.wait;
1362	bool thrashing = false;
1363	unsigned long pflags;
1364	bool in_thrashing;
1365	wait_queue_head_t *q;
1366	struct folio *folio = pfn_swap_entry_folio(entry);
1367
1368	q = folio_waitqueue(folio);
1369	if (!folio_test_uptodate(folio) && folio_test_workingset(folio)) {
1370		delayacct_thrashing_start(&in_thrashing);
1371		psi_memstall_enter(&pflags);
1372		thrashing = true;
1373	}
1374
1375	init_wait(wait);
1376	wait->func = wake_page_function;
1377	wait_page.folio = folio;
1378	wait_page.bit_nr = PG_locked;
1379	wait->flags = 0;
1380
1381	spin_lock_irq(&q->lock);
1382	folio_set_waiters(folio);
1383	if (!folio_trylock_flag(folio, PG_locked, wait))
1384		__add_wait_queue_entry_tail(q, wait);
1385	spin_unlock_irq(&q->lock);
1386
 
1387	/*
1388	 * If a migration entry exists for the page the migration path must hold
1389	 * a valid reference to the page, and it must take the ptl to remove the
1390	 * migration entry. So the page is valid until the ptl is dropped.
 
1391	 */
1392	spin_unlock(ptl);
1393
1394	for (;;) {
1395		unsigned int flags;
1396
1397		set_current_state(TASK_UNINTERRUPTIBLE);
1398
1399		/* Loop until we've been woken or interrupted */
1400		flags = smp_load_acquire(&wait->flags);
1401		if (!(flags & WQ_FLAG_WOKEN)) {
1402			if (signal_pending_state(TASK_UNINTERRUPTIBLE, current))
1403				break;
1404
1405			io_schedule();
1406			continue;
1407		}
1408		break;
1409	}
1410
1411	finish_wait(q, wait);
1412
1413	if (thrashing) {
1414		delayacct_thrashing_end(&in_thrashing);
1415		psi_memstall_leave(&pflags);
1416	}
1417}
1418#endif
1419
1420void folio_wait_bit(struct folio *folio, int bit_nr)
 
1421{
1422	folio_wait_bit_common(folio, bit_nr, TASK_UNINTERRUPTIBLE, SHARED);
 
 
1423}
1424EXPORT_SYMBOL(folio_wait_bit);
1425
1426int folio_wait_bit_killable(struct folio *folio, int bit_nr)
1427{
1428	return folio_wait_bit_common(folio, bit_nr, TASK_KILLABLE, SHARED);
1429}
1430EXPORT_SYMBOL(folio_wait_bit_killable);
1431
1432/**
1433 * folio_put_wait_locked - Drop a reference and wait for it to be unlocked
1434 * @folio: The folio to wait for.
1435 * @state: The sleep state (TASK_KILLABLE, TASK_UNINTERRUPTIBLE, etc).
1436 *
1437 * The caller should hold a reference on @folio.  They expect the page to
1438 * become unlocked relatively soon, but do not wish to hold up migration
1439 * (for example) by holding the reference while waiting for the folio to
1440 * come unlocked.  After this function returns, the caller should not
1441 * dereference @folio.
1442 *
1443 * Return: 0 if the folio was unlocked or -EINTR if interrupted by a signal.
1444 */
1445static int folio_put_wait_locked(struct folio *folio, int state)
1446{
1447	return folio_wait_bit_common(folio, PG_locked, state, DROP);
 
 
 
 
1448}
1449
1450/**
1451 * folio_add_wait_queue - Add an arbitrary waiter to a folio's wait queue
1452 * @folio: Folio defining the wait queue of interest
1453 * @waiter: Waiter to add to the queue
1454 *
1455 * Add an arbitrary @waiter to the wait queue for the nominated @folio.
1456 */
1457void folio_add_wait_queue(struct folio *folio, wait_queue_entry_t *waiter)
1458{
1459	wait_queue_head_t *q = folio_waitqueue(folio);
1460	unsigned long flags;
1461
1462	spin_lock_irqsave(&q->lock, flags);
1463	__add_wait_queue_entry_tail(q, waiter);
1464	folio_set_waiters(folio);
1465	spin_unlock_irqrestore(&q->lock, flags);
1466}
1467EXPORT_SYMBOL_GPL(folio_add_wait_queue);
 
 
1468
1469/**
1470 * folio_unlock - Unlock a locked folio.
1471 * @folio: The folio.
1472 *
1473 * Unlocks the folio and wakes up any thread sleeping on the page lock.
 
 
 
1474 *
1475 * Context: May be called from interrupt or process context.  May not be
1476 * called from NMI context.
 
1477 */
1478void folio_unlock(struct folio *folio)
1479{
1480	/* Bit 7 allows x86 to check the byte's sign bit */
1481	BUILD_BUG_ON(PG_waiters != 7);
1482	BUILD_BUG_ON(PG_locked > 7);
1483	VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1484	if (folio_xor_flags_has_waiters(folio, 1 << PG_locked))
1485		folio_wake_bit(folio, PG_locked);
1486}
1487EXPORT_SYMBOL(folio_unlock);
1488
1489/**
1490 * folio_end_read - End read on a folio.
1491 * @folio: The folio.
1492 * @success: True if all reads completed successfully.
1493 *
1494 * When all reads against a folio have completed, filesystems should
1495 * call this function to let the pagecache know that no more reads
1496 * are outstanding.  This will unlock the folio and wake up any thread
1497 * sleeping on the lock.  The folio will also be marked uptodate if all
1498 * reads succeeded.
1499 *
1500 * Context: May be called from interrupt or process context.  May not be
1501 * called from NMI context.
1502 */
1503void folio_end_read(struct folio *folio, bool success)
1504{
1505	unsigned long mask = 1 << PG_locked;
1506
1507	/* Must be in bottom byte for x86 to work */
1508	BUILD_BUG_ON(PG_uptodate > 7);
1509	VM_BUG_ON_FOLIO(!folio_test_locked(folio), folio);
1510	VM_BUG_ON_FOLIO(folio_test_uptodate(folio), folio);
1511
1512	if (likely(success))
1513		mask |= 1 << PG_uptodate;
1514	if (folio_xor_flags_has_waiters(folio, mask))
1515		folio_wake_bit(folio, PG_locked);
1516}
1517EXPORT_SYMBOL(folio_end_read);
1518
1519/**
1520 * folio_end_private_2 - Clear PG_private_2 and wake any waiters.
1521 * @folio: The folio.
1522 *
1523 * Clear the PG_private_2 bit on a folio and wake up any sleepers waiting for
1524 * it.  The folio reference held for PG_private_2 being set is released.
1525 *
1526 * This is, for example, used when a netfs folio is being written to a local
1527 * disk cache, thereby allowing writes to the cache for the same folio to be
1528 * serialised.
 
 
 
 
1529 */
1530void folio_end_private_2(struct folio *folio)
1531{
1532	VM_BUG_ON_FOLIO(!folio_test_private_2(folio), folio);
1533	clear_bit_unlock(PG_private_2, folio_flags(folio, 0));
1534	folio_wake_bit(folio, PG_private_2);
1535	folio_put(folio);
 
1536}
1537EXPORT_SYMBOL(folio_end_private_2);
1538
1539/**
1540 * folio_wait_private_2 - Wait for PG_private_2 to be cleared on a folio.
1541 * @folio: The folio to wait on.
1542 *
1543 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio.
1544 */
1545void folio_wait_private_2(struct folio *folio)
1546{
1547	while (folio_test_private_2(folio))
1548		folio_wait_bit(folio, PG_private_2);
1549}
1550EXPORT_SYMBOL(folio_wait_private_2);
1551
1552/**
1553 * folio_wait_private_2_killable - Wait for PG_private_2 to be cleared on a folio.
1554 * @folio: The folio to wait on.
1555 *
1556 * Wait for PG_private_2 (aka PG_fscache) to be cleared on a folio or until a
1557 * fatal signal is received by the calling task.
1558 *
1559 * Return:
1560 * - 0 if successful.
1561 * - -EINTR if a fatal signal was encountered.
1562 */
1563int folio_wait_private_2_killable(struct folio *folio)
1564{
1565	int ret = 0;
1566
1567	while (folio_test_private_2(folio)) {
1568		ret = folio_wait_bit_killable(folio, PG_private_2);
1569		if (ret < 0)
1570			break;
1571	}
1572
1573	return ret;
 
1574}
1575EXPORT_SYMBOL(folio_wait_private_2_killable);
1576
1577/**
1578 * folio_end_writeback - End writeback against a folio.
1579 * @folio: The folio.
1580 *
1581 * The folio must actually be under writeback.
1582 *
1583 * Context: May be called from process or interrupt context.
1584 */
1585void folio_end_writeback(struct folio *folio)
1586{
1587	VM_BUG_ON_FOLIO(!folio_test_writeback(folio), folio);
 
 
 
 
 
 
 
 
 
 
1588
1589	/*
1590	 * folio_test_clear_reclaim() could be used here but it is an
1591	 * atomic operation and overkill in this particular case. Failing
1592	 * to shuffle a folio marked for immediate reclaim is too mild
1593	 * a gain to justify taking an atomic operation penalty at the
1594	 * end of every folio writeback.
1595	 */
1596	if (folio_test_reclaim(folio)) {
1597		folio_clear_reclaim(folio);
1598		folio_rotate_reclaimable(folio);
1599	}
1600
1601	/*
1602	 * Writeback does not hold a folio reference of its own, relying
1603	 * on truncation to wait for the clearing of PG_writeback.
1604	 * But here we must make sure that the folio is not freed and
1605	 * reused before the folio_wake_bit().
1606	 */
1607	folio_get(folio);
1608	if (__folio_end_writeback(folio))
1609		folio_wake_bit(folio, PG_writeback);
1610	acct_reclaim_writeback(folio);
1611	folio_put(folio);
1612}
1613EXPORT_SYMBOL(folio_end_writeback);
1614
1615/**
1616 * __folio_lock - Get a lock on the folio, assuming we need to sleep to get it.
1617 * @folio: The folio to lock
1618 */
1619void __folio_lock(struct folio *folio)
1620{
1621	folio_wait_bit_common(folio, PG_locked, TASK_UNINTERRUPTIBLE,
 
 
1622				EXCLUSIVE);
1623}
1624EXPORT_SYMBOL(__folio_lock);
1625
1626int __folio_lock_killable(struct folio *folio)
1627{
1628	return folio_wait_bit_common(folio, PG_locked, TASK_KILLABLE,
 
 
1629					EXCLUSIVE);
1630}
1631EXPORT_SYMBOL_GPL(__folio_lock_killable);
1632
1633static int __folio_lock_async(struct folio *folio, struct wait_page_queue *wait)
1634{
1635	struct wait_queue_head *q = folio_waitqueue(folio);
1636	int ret;
1637
1638	wait->folio = folio;
1639	wait->bit_nr = PG_locked;
1640
1641	spin_lock_irq(&q->lock);
1642	__add_wait_queue_entry_tail(q, &wait->wait);
1643	folio_set_waiters(folio);
1644	ret = !folio_trylock(folio);
1645	/*
1646	 * If we were successful now, we know we're still on the
1647	 * waitqueue as we're still under the lock. This means it's
1648	 * safe to remove and return success, we know the callback
1649	 * isn't going to trigger.
1650	 */
1651	if (!ret)
1652		__remove_wait_queue(q, &wait->wait);
1653	else
1654		ret = -EIOCBQUEUED;
1655	spin_unlock_irq(&q->lock);
1656	return ret;
1657}
1658
1659/*
1660 * Return values:
1661 * 0 - folio is locked.
1662 * non-zero - folio is not locked.
1663 *     mmap_lock or per-VMA lock has been released (mmap_read_unlock() or
1664 *     vma_end_read()), unless flags had both FAULT_FLAG_ALLOW_RETRY and
1665 *     FAULT_FLAG_RETRY_NOWAIT set, in which case the lock is still held.
1666 *
1667 * If neither ALLOW_RETRY nor KILLABLE are set, will always return 0
1668 * with the folio locked and the mmap_lock/per-VMA lock is left unperturbed.
1669 */
1670vm_fault_t __folio_lock_or_retry(struct folio *folio, struct vm_fault *vmf)
 
1671{
1672	unsigned int flags = vmf->flags;
1673
1674	if (fault_flag_allow_retry_first(flags)) {
1675		/*
1676		 * CAUTION! In this case, mmap_lock/per-VMA lock is not
1677		 * released even though returning VM_FAULT_RETRY.
1678		 */
1679		if (flags & FAULT_FLAG_RETRY_NOWAIT)
1680			return VM_FAULT_RETRY;
1681
1682		release_fault_lock(vmf);
1683		if (flags & FAULT_FLAG_KILLABLE)
1684			folio_wait_locked_killable(folio);
1685		else
1686			folio_wait_locked(folio);
1687		return VM_FAULT_RETRY;
1688	}
1689	if (flags & FAULT_FLAG_KILLABLE) {
1690		bool ret;
1691
1692		ret = __folio_lock_killable(folio);
1693		if (ret) {
1694			release_fault_lock(vmf);
1695			return VM_FAULT_RETRY;
1696		}
1697	} else {
1698		__folio_lock(folio);
 
1699	}
1700
1701	return 0;
1702}
1703
1704/**
1705 * page_cache_next_miss() - Find the next gap in the page cache.
1706 * @mapping: Mapping.
1707 * @index: Index.
1708 * @max_scan: Maximum range to search.
1709 *
1710 * Search the range [index, min(index + max_scan - 1, ULONG_MAX)] for the
1711 * gap with the lowest index.
1712 *
1713 * This function may be called under the rcu_read_lock.  However, this will
1714 * not atomically search a snapshot of the cache at a single point in time.
1715 * For example, if a gap is created at index 5, then subsequently a gap is
1716 * created at index 10, page_cache_next_miss covering both indices may
1717 * return 10 if called under the rcu_read_lock.
1718 *
1719 * Return: The index of the gap if found, otherwise an index outside the
1720 * range specified (in which case 'return - index >= max_scan' will be true).
1721 * In the rare case of index wrap-around, 0 will be returned.
1722 */
1723pgoff_t page_cache_next_miss(struct address_space *mapping,
1724			     pgoff_t index, unsigned long max_scan)
1725{
1726	XA_STATE(xas, &mapping->i_pages, index);
1727
1728	while (max_scan--) {
1729		void *entry = xas_next(&xas);
1730		if (!entry || xa_is_value(entry))
1731			break;
1732		if (xas.xa_index == 0)
1733			break;
1734	}
1735
1736	return xas.xa_index;
1737}
1738EXPORT_SYMBOL(page_cache_next_miss);
1739
1740/**
1741 * page_cache_prev_miss() - Find the previous gap in the page cache.
1742 * @mapping: Mapping.
1743 * @index: Index.
1744 * @max_scan: Maximum range to search.
1745 *
1746 * Search the range [max(index - max_scan + 1, 0), index] for the
1747 * gap with the highest index.
1748 *
1749 * This function may be called under the rcu_read_lock.  However, this will
1750 * not atomically search a snapshot of the cache at a single point in time.
1751 * For example, if a gap is created at index 10, then subsequently a gap is
1752 * created at index 5, page_cache_prev_miss() covering both indices may
1753 * return 5 if called under the rcu_read_lock.
1754 *
1755 * Return: The index of the gap if found, otherwise an index outside the
1756 * range specified (in which case 'index - return >= max_scan' will be true).
1757 * In the rare case of wrap-around, ULONG_MAX will be returned.
1758 */
1759pgoff_t page_cache_prev_miss(struct address_space *mapping,
1760			     pgoff_t index, unsigned long max_scan)
1761{
1762	XA_STATE(xas, &mapping->i_pages, index);
1763
1764	while (max_scan--) {
1765		void *entry = xas_prev(&xas);
1766		if (!entry || xa_is_value(entry))
1767			break;
1768		if (xas.xa_index == ULONG_MAX)
1769			break;
1770	}
1771
1772	return xas.xa_index;
1773}
1774EXPORT_SYMBOL(page_cache_prev_miss);
1775
1776/*
1777 * Lockless page cache protocol:
1778 * On the lookup side:
1779 * 1. Load the folio from i_pages
1780 * 2. Increment the refcount if it's not zero
1781 * 3. If the folio is not found by xas_reload(), put the refcount and retry
1782 *
1783 * On the removal side:
1784 * A. Freeze the page (by zeroing the refcount if nobody else has a reference)
1785 * B. Remove the page from i_pages
1786 * C. Return the page to the page allocator
1787 *
1788 * This means that any page may have its reference count temporarily
1789 * increased by a speculative page cache (or fast GUP) lookup as it can
1790 * be allocated by another user before the RCU grace period expires.
1791 * Because the refcount temporarily acquired here may end up being the
1792 * last refcount on the page, any page allocation must be freeable by
1793 * folio_put().
1794 */
1795
1796/*
1797 * filemap_get_entry - Get a page cache entry.
1798 * @mapping: the address_space to search
1799 * @index: The page cache index.
 
 
 
1800 *
1801 * Looks up the page cache entry at @mapping & @index.  If it is a folio,
1802 * it is returned with an increased refcount.  If it is a shadow entry
1803 * of a previously evicted folio, or a swap entry from shmem/tmpfs,
1804 * it is returned without further action.
1805 *
1806 * Return: The folio, swap or shadow entry, %NULL if nothing is found.
1807 */
1808void *filemap_get_entry(struct address_space *mapping, pgoff_t index)
1809{
1810	XA_STATE(xas, &mapping->i_pages, index);
1811	struct folio *folio;
1812
1813	rcu_read_lock();
1814repeat:
1815	xas_reset(&xas);
1816	folio = xas_load(&xas);
1817	if (xas_retry(&xas, folio))
1818		goto repeat;
1819	/*
1820	 * A shadow entry of a recently evicted page, or a swap entry from
1821	 * shmem/tmpfs.  Return it without attempting to raise page count.
1822	 */
1823	if (!folio || xa_is_value(folio))
1824		goto out;
1825
1826	if (!folio_try_get_rcu(folio))
1827		goto repeat;
1828
1829	if (unlikely(folio != xas_reload(&xas))) {
1830		folio_put(folio);
 
 
 
 
 
1831		goto repeat;
1832	}
 
1833out:
1834	rcu_read_unlock();
1835
1836	return folio;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1837}
 
1838
1839/**
1840 * __filemap_get_folio - Find and get a reference to a folio.
1841 * @mapping: The address_space to search.
1842 * @index: The page index.
1843 * @fgp_flags: %FGP flags modify how the folio is returned.
1844 * @gfp: Memory allocation flags to use if %FGP_CREAT is specified.
1845 *
1846 * Looks up the page cache entry at @mapping & @index.
1847 *
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1848 * If %FGP_LOCK or %FGP_CREAT are specified then the function may sleep even
1849 * if the %GFP flags specified for %FGP_CREAT are atomic.
1850 *
1851 * If this function returns a folio, it is returned with an increased refcount.
1852 *
1853 * Return: The found folio or an ERR_PTR() otherwise.
1854 */
1855struct folio *__filemap_get_folio(struct address_space *mapping, pgoff_t index,
1856		fgf_t fgp_flags, gfp_t gfp)
1857{
1858	struct folio *folio;
1859
1860repeat:
1861	folio = filemap_get_entry(mapping, index);
1862	if (xa_is_value(folio))
1863		folio = NULL;
1864	if (!folio)
1865		goto no_page;
1866
1867	if (fgp_flags & FGP_LOCK) {
1868		if (fgp_flags & FGP_NOWAIT) {
1869			if (!folio_trylock(folio)) {
1870				folio_put(folio);
1871				return ERR_PTR(-EAGAIN);
1872			}
1873		} else {
1874			folio_lock(folio);
1875		}
1876
1877		/* Has the page been truncated? */
1878		if (unlikely(folio->mapping != mapping)) {
1879			folio_unlock(folio);
1880			folio_put(folio);
1881			goto repeat;
1882		}
1883		VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
1884	}
1885
1886	if (fgp_flags & FGP_ACCESSED)
1887		folio_mark_accessed(folio);
1888	else if (fgp_flags & FGP_WRITE) {
1889		/* Clear idle flag for buffer write */
1890		if (folio_test_idle(folio))
1891			folio_clear_idle(folio);
1892	}
1893
1894	if (fgp_flags & FGP_STABLE)
1895		folio_wait_stable(folio);
1896no_page:
1897	if (!folio && (fgp_flags & FGP_CREAT)) {
1898		unsigned order = FGF_GET_ORDER(fgp_flags);
1899		int err;
 
 
 
 
 
 
 
 
1900
1901		if ((fgp_flags & FGP_WRITE) && mapping_can_writeback(mapping))
1902			gfp |= __GFP_WRITE;
1903		if (fgp_flags & FGP_NOFS)
1904			gfp &= ~__GFP_FS;
1905		if (fgp_flags & FGP_NOWAIT) {
1906			gfp &= ~GFP_KERNEL;
1907			gfp |= GFP_NOWAIT | __GFP_NOWARN;
1908		}
1909		if (WARN_ON_ONCE(!(fgp_flags & (FGP_LOCK | FGP_FOR_MMAP))))
1910			fgp_flags |= FGP_LOCK;
1911
1912		if (!mapping_large_folio_support(mapping))
1913			order = 0;
1914		if (order > MAX_PAGECACHE_ORDER)
1915			order = MAX_PAGECACHE_ORDER;
1916		/* If we're not aligned, allocate a smaller folio */
1917		if (index & ((1UL << order) - 1))
1918			order = __ffs(index);
1919
1920		do {
1921			gfp_t alloc_gfp = gfp;
 
 
 
 
 
1922
1923			err = -ENOMEM;
1924			if (order > 0)
1925				alloc_gfp |= __GFP_NORETRY | __GFP_NOWARN;
1926			folio = filemap_alloc_folio(alloc_gfp, order);
1927			if (!folio)
1928				continue;
1929
1930			/* Init accessed so avoid atomic mark_page_accessed later */
1931			if (fgp_flags & FGP_ACCESSED)
1932				__folio_set_referenced(folio);
1933
1934			err = filemap_add_folio(mapping, folio, index, gfp);
1935			if (!err)
1936				break;
1937			folio_put(folio);
1938			folio = NULL;
1939		} while (order-- > 0);
1940
1941		if (err == -EEXIST)
1942			goto repeat;
1943		if (err)
1944			return ERR_PTR(err);
1945		/*
1946		 * filemap_add_folio locks the page, and for mmap
1947		 * we expect an unlocked page.
1948		 */
1949		if (folio && (fgp_flags & FGP_FOR_MMAP))
1950			folio_unlock(folio);
1951	}
1952
1953	if (!folio)
1954		return ERR_PTR(-ENOENT);
1955	return folio;
1956}
1957EXPORT_SYMBOL(__filemap_get_folio);
1958
1959static inline struct folio *find_get_entry(struct xa_state *xas, pgoff_t max,
1960		xa_mark_t mark)
1961{
1962	struct folio *folio;
1963
1964retry:
1965	if (mark == XA_PRESENT)
1966		folio = xas_find(xas, max);
1967	else
1968		folio = xas_find_marked(xas, max, mark);
1969
1970	if (xas_retry(xas, folio))
1971		goto retry;
1972	/*
1973	 * A shadow entry of a recently evicted page, a swap
1974	 * entry from shmem/tmpfs or a DAX entry.  Return it
1975	 * without attempting to raise page count.
1976	 */
1977	if (!folio || xa_is_value(folio))
1978		return folio;
1979
1980	if (!folio_try_get_rcu(folio))
1981		goto reset;
1982
1983	if (unlikely(folio != xas_reload(xas))) {
1984		folio_put(folio);
1985		goto reset;
1986	}
1987
1988	return folio;
1989reset:
1990	xas_reset(xas);
1991	goto retry;
1992}
 
1993
1994/**
1995 * find_get_entries - gang pagecache lookup
1996 * @mapping:	The address_space to search
1997 * @start:	The starting page cache index
1998 * @end:	The final page index (inclusive).
1999 * @fbatch:	Where the resulting entries are placed.
2000 * @indices:	The cache indices corresponding to the entries in @entries
2001 *
2002 * find_get_entries() will search for and return a batch of entries in
2003 * the mapping.  The entries are placed in @fbatch.  find_get_entries()
2004 * takes a reference on any actual folios it returns.
2005 *
2006 * The entries have ascending indexes.  The indices may not be consecutive
2007 * due to not-present entries or large folios.
 
 
2008 *
2009 * Any shadow entries of evicted folios, or swap entries from
2010 * shmem/tmpfs, are included in the returned array.
2011 *
2012 * Return: The number of entries which were found.
2013 */
2014unsigned find_get_entries(struct address_space *mapping, pgoff_t *start,
2015		pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
 
 
 
 
 
 
2016{
2017	XA_STATE(xas, &mapping->i_pages, *start);
2018	struct folio *folio;
 
 
 
 
2019
2020	rcu_read_lock();
2021	while ((folio = find_get_entry(&xas, end, XA_PRESENT)) != NULL) {
2022		indices[fbatch->nr] = xas.xa_index;
2023		if (!folio_batch_add(fbatch, folio))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2024			break;
 
 
 
 
 
2025	}
2026	rcu_read_unlock();
2027
2028	if (folio_batch_count(fbatch)) {
2029		unsigned long nr = 1;
2030		int idx = folio_batch_count(fbatch) - 1;
2031
2032		folio = fbatch->folios[idx];
2033		if (!xa_is_value(folio))
2034			nr = folio_nr_pages(folio);
2035		*start = indices[idx] + nr;
2036	}
2037	return folio_batch_count(fbatch);
2038}
2039
2040/**
2041 * find_lock_entries - Find a batch of pagecache entries.
2042 * @mapping:	The address_space to search.
2043 * @start:	The starting page cache index.
2044 * @end:	The final page index (inclusive).
2045 * @fbatch:	Where the resulting entries are placed.
2046 * @indices:	The cache indices of the entries in @fbatch.
2047 *
2048 * find_lock_entries() will return a batch of entries from @mapping.
2049 * Swap, shadow and DAX entries are included.  Folios are returned
2050 * locked and with an incremented refcount.  Folios which are locked
2051 * by somebody else or under writeback are skipped.  Folios which are
2052 * partially outside the range are not returned.
2053 *
2054 * The entries have ascending indexes.  The indices may not be consecutive
2055 * due to not-present entries, large folios, folios which could not be
2056 * locked or folios under writeback.
2057 *
2058 * Return: The number of entries which were found.
2059 */
2060unsigned find_lock_entries(struct address_space *mapping, pgoff_t *start,
2061		pgoff_t end, struct folio_batch *fbatch, pgoff_t *indices)
 
 
 
 
 
 
 
 
 
 
 
 
2062{
2063	XA_STATE(xas, &mapping->i_pages, *start);
2064	struct folio *folio;
 
 
 
 
2065
2066	rcu_read_lock();
2067	while ((folio = find_get_entry(&xas, end, XA_PRESENT))) {
2068		if (!xa_is_value(folio)) {
2069			if (folio->index < *start)
2070				goto put;
2071			if (folio_next_index(folio) - 1 > end)
2072				goto put;
2073			if (!folio_trylock(folio))
2074				goto put;
2075			if (folio->mapping != mapping ||
2076			    folio_test_writeback(folio))
2077				goto unlock;
2078			VM_BUG_ON_FOLIO(!folio_contains(folio, xas.xa_index),
2079					folio);
 
 
 
 
 
2080		}
2081		indices[fbatch->nr] = xas.xa_index;
2082		if (!folio_batch_add(fbatch, folio))
2083			break;
2084		continue;
2085unlock:
2086		folio_unlock(folio);
2087put:
2088		folio_put(folio);
2089	}
 
 
 
 
 
 
 
 
 
 
 
 
2090	rcu_read_unlock();
2091
2092	if (folio_batch_count(fbatch)) {
2093		unsigned long nr = 1;
2094		int idx = folio_batch_count(fbatch) - 1;
2095
2096		folio = fbatch->folios[idx];
2097		if (!xa_is_value(folio))
2098			nr = folio_nr_pages(folio);
2099		*start = indices[idx] + nr;
2100	}
2101	return folio_batch_count(fbatch);
2102}
2103
2104/**
2105 * filemap_get_folios - Get a batch of folios
2106 * @mapping:	The address_space to search
2107 * @start:	The starting page index
2108 * @end:	The final page index (inclusive)
2109 * @fbatch:	The batch to fill.
2110 *
2111 * Search for and return a batch of folios in the mapping starting at
2112 * index @start and up to index @end (inclusive).  The folios are returned
2113 * in @fbatch with an elevated reference count.
2114 *
2115 * Return: The number of folios which were found.
2116 * We also update @start to index the next folio for the traversal.
2117 */
2118unsigned filemap_get_folios(struct address_space *mapping, pgoff_t *start,
2119		pgoff_t end, struct folio_batch *fbatch)
2120{
2121	return filemap_get_folios_tag(mapping, start, end, XA_PRESENT, fbatch);
2122}
2123EXPORT_SYMBOL(filemap_get_folios);
2124
2125/**
2126 * filemap_get_folios_contig - Get a batch of contiguous folios
2127 * @mapping:	The address_space to search
2128 * @start:	The starting page index
2129 * @end:	The final page index (inclusive)
2130 * @fbatch:	The batch to fill
2131 *
2132 * filemap_get_folios_contig() works exactly like filemap_get_folios(),
2133 * except the returned folios are guaranteed to be contiguous. This may
2134 * not return all contiguous folios if the batch gets filled up.
2135 *
2136 * Return: The number of folios found.
2137 * Also update @start to be positioned for traversal of the next folio.
2138 */
2139
2140unsigned filemap_get_folios_contig(struct address_space *mapping,
2141		pgoff_t *start, pgoff_t end, struct folio_batch *fbatch)
2142{
2143	XA_STATE(xas, &mapping->i_pages, *start);
2144	unsigned long nr;
2145	struct folio *folio;
2146
2147	rcu_read_lock();
2148
2149	for (folio = xas_load(&xas); folio && xas.xa_index <= end;
2150			folio = xas_next(&xas)) {
2151		if (xas_retry(&xas, folio))
2152			continue;
2153		/*
2154		 * If the entry has been swapped out, we can stop looking.
2155		 * No current caller is looking for DAX entries.
2156		 */
2157		if (xa_is_value(folio))
2158			goto update_start;
2159
2160		if (!folio_try_get_rcu(folio))
2161			goto retry;
2162
2163		if (unlikely(folio != xas_reload(&xas)))
2164			goto put_folio;
 
2165
2166		if (!folio_batch_add(fbatch, folio)) {
2167			nr = folio_nr_pages(folio);
2168			*start = folio->index + nr;
2169			goto out;
2170		}
2171		continue;
2172put_folio:
2173		folio_put(folio);
2174
2175retry:
2176		xas_reset(&xas);
2177	}
2178
2179update_start:
2180	nr = folio_batch_count(fbatch);
2181
2182	if (nr) {
2183		folio = fbatch->folios[nr - 1];
2184		*start = folio_next_index(folio);
2185	}
2186out:
2187	rcu_read_unlock();
2188	return folio_batch_count(fbatch);
2189}
2190EXPORT_SYMBOL(filemap_get_folios_contig);
2191
2192/**
2193 * filemap_get_folios_tag - Get a batch of folios matching @tag
2194 * @mapping:    The address_space to search
2195 * @start:      The starting page index
2196 * @end:        The final page index (inclusive)
2197 * @tag:        The tag index
2198 * @fbatch:     The batch to fill
2199 *
2200 * The first folio may start before @start; if it does, it will contain
2201 * @start.  The final folio may extend beyond @end; if it does, it will
2202 * contain @end.  The folios have ascending indices.  There may be gaps
2203 * between the folios if there are indices which have no folio in the
2204 * page cache.  If folios are added to or removed from the page cache
2205 * while this is running, they may or may not be found by this call.
2206 * Only returns folios that are tagged with @tag.
2207 *
2208 * Return: The number of folios found.
2209 * Also update @start to index the next folio for traversal.
2210 */
2211unsigned filemap_get_folios_tag(struct address_space *mapping, pgoff_t *start,
2212			pgoff_t end, xa_mark_t tag, struct folio_batch *fbatch)
 
2213{
2214	XA_STATE(xas, &mapping->i_pages, *start);
2215	struct folio *folio;
 
 
 
 
2216
2217	rcu_read_lock();
2218	while ((folio = find_get_entry(&xas, end, tag)) != NULL) {
 
 
2219		/*
2220		 * Shadow entries should never be tagged, but this iteration
2221		 * is lockless so there is a window for page reclaim to evict
2222		 * a page we saw tagged. Skip over it.
2223		 */
2224		if (xa_is_value(folio))
2225			continue;
2226		if (!folio_batch_add(fbatch, folio)) {
2227			unsigned long nr = folio_nr_pages(folio);
2228			*start = folio->index + nr;
 
 
 
 
 
 
 
 
2229			goto out;
2230		}
 
 
 
 
 
2231	}
 
2232	/*
2233	 * We come here when there is no page beyond @end. We take care to not
2234	 * overflow the index @start as it confuses some of the callers. This
2235	 * breaks the iteration when there is a page at index -1 but that is
2236	 * already broke anyway.
2237	 */
2238	if (end == (pgoff_t)-1)
2239		*start = (pgoff_t)-1;
2240	else
2241		*start = end + 1;
2242out:
2243	rcu_read_unlock();
2244
2245	return folio_batch_count(fbatch);
2246}
2247EXPORT_SYMBOL(filemap_get_folios_tag);
2248
2249/*
2250 * CD/DVDs are error prone. When a medium error occurs, the driver may fail
2251 * a _large_ part of the i/o request. Imagine the worst scenario:
2252 *
2253 *      ---R__________________________________________B__________
2254 *         ^ reading here                             ^ bad block(assume 4k)
2255 *
2256 * read(R) => miss => readahead(R...B) => media error => frustrating retries
2257 * => failing the whole request => read(R) => read(R+1) =>
2258 * readahead(R+1...B+1) => bang => read(R+2) => read(R+3) =>
2259 * readahead(R+3...B+2) => bang => read(R+3) => read(R+4) =>
2260 * readahead(R+4...B+3) => bang => read(R+4) => read(R+5) => ......
2261 *
2262 * It is going insane. Fix it by quickly scaling down the readahead size.
2263 */
2264static void shrink_readahead_size_eio(struct file_ra_state *ra)
2265{
2266	ra->ra_pages /= 4;
2267}
2268
2269/*
2270 * filemap_get_read_batch - Get a batch of folios for read
 
 
 
 
 
 
 
 
 
2271 *
2272 * Get a batch of folios which represent a contiguous range of bytes in
2273 * the file.  No exceptional entries will be returned.  If @index is in
2274 * the middle of a folio, the entire folio will be returned.  The last
2275 * folio in the batch may have the readahead flag set or the uptodate flag
2276 * clear so that the caller can take the appropriate action.
2277 */
2278static void filemap_get_read_batch(struct address_space *mapping,
2279		pgoff_t index, pgoff_t max, struct folio_batch *fbatch)
2280{
2281	XA_STATE(xas, &mapping->i_pages, index);
2282	struct folio *folio;
2283
2284	rcu_read_lock();
2285	for (folio = xas_load(&xas); folio; folio = xas_next(&xas)) {
2286		if (xas_retry(&xas, folio))
2287			continue;
2288		if (xas.xa_index > max || xa_is_value(folio))
2289			break;
2290		if (xa_is_sibling(folio))
2291			break;
2292		if (!folio_try_get_rcu(folio))
2293			goto retry;
2294
2295		if (unlikely(folio != xas_reload(&xas)))
2296			goto put_folio;
2297
2298		if (!folio_batch_add(fbatch, folio))
2299			break;
2300		if (!folio_test_uptodate(folio))
2301			break;
2302		if (folio_test_readahead(folio))
2303			break;
2304		xas_advance(&xas, folio_next_index(folio) - 1);
2305		continue;
2306put_folio:
2307		folio_put(folio);
2308retry:
2309		xas_reset(&xas);
2310	}
2311	rcu_read_unlock();
2312}
2313
2314static int filemap_read_folio(struct file *file, filler_t filler,
2315		struct folio *folio)
2316{
2317	bool workingset = folio_test_workingset(folio);
2318	unsigned long pflags;
2319	int error;
2320
2321	/*
2322	 * A previous I/O error may have been due to temporary failures,
2323	 * eg. multipath errors.  PG_error will be set again if read_folio
2324	 * fails.
2325	 */
2326	folio_clear_error(folio);
2327
2328	/* Start the actual read. The read will unlock the page. */
2329	if (unlikely(workingset))
2330		psi_memstall_enter(&pflags);
2331	error = filler(file, folio);
2332	if (unlikely(workingset))
2333		psi_memstall_leave(&pflags);
2334	if (error)
2335		return error;
2336
2337	error = folio_wait_locked_killable(folio);
2338	if (error)
2339		return error;
2340	if (folio_test_uptodate(folio))
2341		return 0;
2342	if (file)
2343		shrink_readahead_size_eio(&file->f_ra);
2344	return -EIO;
2345}
2346
2347static bool filemap_range_uptodate(struct address_space *mapping,
2348		loff_t pos, size_t count, struct folio *folio,
2349		bool need_uptodate)
2350{
2351	if (folio_test_uptodate(folio))
2352		return true;
2353	/* pipes can't handle partially uptodate pages */
2354	if (need_uptodate)
2355		return false;
2356	if (!mapping->a_ops->is_partially_uptodate)
2357		return false;
2358	if (mapping->host->i_blkbits >= folio_shift(folio))
2359		return false;
2360
2361	if (folio_pos(folio) > pos) {
2362		count -= folio_pos(folio) - pos;
2363		pos = 0;
2364	} else {
2365		pos -= folio_pos(folio);
2366	}
2367
2368	return mapping->a_ops->is_partially_uptodate(folio, pos, count);
2369}
2370
2371static int filemap_update_page(struct kiocb *iocb,
2372		struct address_space *mapping, size_t count,
2373		struct folio *folio, bool need_uptodate)
2374{
2375	int error;
2376
2377	if (iocb->ki_flags & IOCB_NOWAIT) {
2378		if (!filemap_invalidate_trylock_shared(mapping))
2379			return -EAGAIN;
2380	} else {
2381		filemap_invalidate_lock_shared(mapping);
2382	}
2383
2384	if (!folio_trylock(folio)) {
2385		error = -EAGAIN;
2386		if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_NOIO))
2387			goto unlock_mapping;
2388		if (!(iocb->ki_flags & IOCB_WAITQ)) {
2389			filemap_invalidate_unlock_shared(mapping);
2390			/*
2391			 * This is where we usually end up waiting for a
2392			 * previously submitted readahead to finish.
2393			 */
2394			folio_put_wait_locked(folio, TASK_KILLABLE);
2395			return AOP_TRUNCATED_PAGE;
2396		}
2397		error = __folio_lock_async(folio, iocb->ki_waitq);
2398		if (error)
2399			goto unlock_mapping;
2400	}
2401
2402	error = AOP_TRUNCATED_PAGE;
2403	if (!folio->mapping)
2404		goto unlock;
2405
2406	error = 0;
2407	if (filemap_range_uptodate(mapping, iocb->ki_pos, count, folio,
2408				   need_uptodate))
2409		goto unlock;
2410
2411	error = -EAGAIN;
2412	if (iocb->ki_flags & (IOCB_NOIO | IOCB_NOWAIT | IOCB_WAITQ))
2413		goto unlock;
2414
2415	error = filemap_read_folio(iocb->ki_filp, mapping->a_ops->read_folio,
2416			folio);
2417	goto unlock_mapping;
2418unlock:
2419	folio_unlock(folio);
2420unlock_mapping:
2421	filemap_invalidate_unlock_shared(mapping);
2422	if (error == AOP_TRUNCATED_PAGE)
2423		folio_put(folio);
2424	return error;
2425}
2426
2427static int filemap_create_folio(struct file *file,
2428		struct address_space *mapping, pgoff_t index,
2429		struct folio_batch *fbatch)
2430{
2431	struct folio *folio;
2432	int error;
2433
2434	folio = filemap_alloc_folio(mapping_gfp_mask(mapping), 0);
2435	if (!folio)
2436		return -ENOMEM;
2437
2438	/*
2439	 * Protect against truncate / hole punch. Grabbing invalidate_lock
2440	 * here assures we cannot instantiate and bring uptodate new
2441	 * pagecache folios after evicting page cache during truncate
2442	 * and before actually freeing blocks.	Note that we could
2443	 * release invalidate_lock after inserting the folio into
2444	 * the page cache as the locked folio would then be enough to
2445	 * synchronize with hole punching. But there are code paths
2446	 * such as filemap_update_page() filling in partially uptodate
2447	 * pages or ->readahead() that need to hold invalidate_lock
2448	 * while mapping blocks for IO so let's hold the lock here as
2449	 * well to keep locking rules simple.
2450	 */
2451	filemap_invalidate_lock_shared(mapping);
2452	error = filemap_add_folio(mapping, folio, index,
2453			mapping_gfp_constraint(mapping, GFP_KERNEL));
2454	if (error == -EEXIST)
2455		error = AOP_TRUNCATED_PAGE;
2456	if (error)
2457		goto error;
2458
2459	error = filemap_read_folio(file, mapping->a_ops->read_folio, folio);
2460	if (error)
2461		goto error;
2462
2463	filemap_invalidate_unlock_shared(mapping);
2464	folio_batch_add(fbatch, folio);
2465	return 0;
2466error:
2467	filemap_invalidate_unlock_shared(mapping);
2468	folio_put(folio);
2469	return error;
2470}
2471
2472static int filemap_readahead(struct kiocb *iocb, struct file *file,
2473		struct address_space *mapping, struct folio *folio,
2474		pgoff_t last_index)
2475{
2476	DEFINE_READAHEAD(ractl, file, &file->f_ra, mapping, folio->index);
2477
2478	if (iocb->ki_flags & IOCB_NOIO)
2479		return -EAGAIN;
2480	page_cache_async_ra(&ractl, folio, last_index - folio->index);
2481	return 0;
2482}
2483
2484static int filemap_get_pages(struct kiocb *iocb, size_t count,
2485		struct folio_batch *fbatch, bool need_uptodate)
2486{
2487	struct file *filp = iocb->ki_filp;
2488	struct address_space *mapping = filp->f_mapping;
 
2489	struct file_ra_state *ra = &filp->f_ra;
2490	pgoff_t index = iocb->ki_pos >> PAGE_SHIFT;
 
2491	pgoff_t last_index;
2492	struct folio *folio;
2493	int err = 0;
2494
2495	/* "last_index" is the index of the page beyond the end of the read */
2496	last_index = DIV_ROUND_UP(iocb->ki_pos + count, PAGE_SIZE);
2497retry:
2498	if (fatal_signal_pending(current))
2499		return -EINTR;
2500
2501	filemap_get_read_batch(mapping, index, last_index - 1, fbatch);
2502	if (!folio_batch_count(fbatch)) {
2503		if (iocb->ki_flags & IOCB_NOIO)
2504			return -EAGAIN;
2505		page_cache_sync_readahead(mapping, ra, filp, index,
2506				last_index - index);
2507		filemap_get_read_batch(mapping, index, last_index - 1, fbatch);
2508	}
2509	if (!folio_batch_count(fbatch)) {
2510		if (iocb->ki_flags & (IOCB_NOWAIT | IOCB_WAITQ))
2511			return -EAGAIN;
2512		err = filemap_create_folio(filp, mapping,
2513				iocb->ki_pos >> PAGE_SHIFT, fbatch);
2514		if (err == AOP_TRUNCATED_PAGE)
2515			goto retry;
2516		return err;
2517	}
2518
2519	folio = fbatch->folios[folio_batch_count(fbatch) - 1];
2520	if (folio_test_readahead(folio)) {
2521		err = filemap_readahead(iocb, filp, mapping, folio, last_index);
2522		if (err)
2523			goto err;
2524	}
2525	if (!folio_test_uptodate(folio)) {
2526		if ((iocb->ki_flags & IOCB_WAITQ) &&
2527		    folio_batch_count(fbatch) > 1)
2528			iocb->ki_flags |= IOCB_NOWAIT;
2529		err = filemap_update_page(iocb, mapping, count, folio,
2530					  need_uptodate);
2531		if (err)
2532			goto err;
2533	}
2534
2535	return 0;
2536err:
2537	if (err < 0)
2538		folio_put(folio);
2539	if (likely(--fbatch->nr))
2540		return 0;
2541	if (err == AOP_TRUNCATED_PAGE)
2542		goto retry;
2543	return err;
2544}
2545
2546static inline bool pos_same_folio(loff_t pos1, loff_t pos2, struct folio *folio)
2547{
2548	unsigned int shift = folio_shift(folio);
 
 
2549
2550	return (pos1 >> shift == pos2 >> shift);
2551}
2552
2553/**
2554 * filemap_read - Read data from the page cache.
2555 * @iocb: The iocb to read.
2556 * @iter: Destination for the data.
2557 * @already_read: Number of bytes already read by the caller.
2558 *
2559 * Copies data from the page cache.  If the data is not currently present,
2560 * uses the readahead and read_folio address_space operations to fetch it.
2561 *
2562 * Return: Total number of bytes copied, including those already read by
2563 * the caller.  If an error happens before any bytes are copied, returns
2564 * a negative error number.
2565 */
2566ssize_t filemap_read(struct kiocb *iocb, struct iov_iter *iter,
2567		ssize_t already_read)
2568{
2569	struct file *filp = iocb->ki_filp;
2570	struct file_ra_state *ra = &filp->f_ra;
2571	struct address_space *mapping = filp->f_mapping;
2572	struct inode *inode = mapping->host;
2573	struct folio_batch fbatch;
2574	int i, error = 0;
2575	bool writably_mapped;
2576	loff_t isize, end_offset;
2577	loff_t last_pos = ra->prev_pos;
2578
2579	if (unlikely(iocb->ki_pos >= inode->i_sb->s_maxbytes))
2580		return 0;
2581	if (unlikely(!iov_iter_count(iter)))
2582		return 0;
2583
2584	iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
2585	folio_batch_init(&fbatch);
2586
2587	do {
2588		cond_resched();
 
 
 
 
 
2589
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2590		/*
2591		 * If we've already successfully copied some data, then we
2592		 * can no longer safely return -EIOCBQUEUED. Hence mark
2593		 * an async read NOWAIT at that point.
2594		 */
2595		if ((iocb->ki_flags & IOCB_WAITQ) && already_read)
2596			iocb->ki_flags |= IOCB_NOWAIT;
2597
2598		if (unlikely(iocb->ki_pos >= i_size_read(inode)))
2599			break;
2600
2601		error = filemap_get_pages(iocb, iter->count, &fbatch, false);
2602		if (error < 0)
2603			break;
2604
2605		/*
2606		 * i_size must be checked after we know the pages are Uptodate.
2607		 *
2608		 * Checking i_size after the check allows us to calculate
2609		 * the correct value for "nr", which means the zero-filled
2610		 * part of the page is not copied back to userspace (unless
2611		 * another truncate extends the file - this is desired though).
2612		 */
 
2613		isize = i_size_read(inode);
2614		if (unlikely(iocb->ki_pos >= isize))
2615			goto put_folios;
2616		end_offset = min_t(loff_t, isize, iocb->ki_pos + iter->count);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2617
2618		/*
2619		 * Once we start copying data, we don't want to be touching any
2620		 * cachelines that might be contended:
2621		 */
2622		writably_mapped = mapping_writably_mapped(mapping);
 
 
2623
2624		/*
2625		 * When a read accesses the same folio several times, only
2626		 * mark it as accessed the first time.
2627		 */
2628		if (!pos_same_folio(iocb->ki_pos, last_pos - 1,
2629				    fbatch.folios[0]))
2630			folio_mark_accessed(fbatch.folios[0]);
2631
2632		for (i = 0; i < folio_batch_count(&fbatch); i++) {
2633			struct folio *folio = fbatch.folios[i];
2634			size_t fsize = folio_size(folio);
2635			size_t offset = iocb->ki_pos & (fsize - 1);
2636			size_t bytes = min_t(loff_t, end_offset - iocb->ki_pos,
2637					     fsize - offset);
2638			size_t copied;
2639
2640			if (end_offset < folio_pos(folio))
2641				break;
2642			if (i > 0)
2643				folio_mark_accessed(folio);
2644			/*
2645			 * If users can be writing to this folio using arbitrary
2646			 * virtual addresses, take care of potential aliasing
2647			 * before reading the folio on the kernel side.
2648			 */
2649			if (writably_mapped)
2650				flush_dcache_folio(folio);
 
 
 
 
2651
2652			copied = copy_folio_to_iter(folio, offset, bytes, iter);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2653
2654			already_read += copied;
2655			iocb->ki_pos += copied;
2656			last_pos = iocb->ki_pos;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2657
2658			if (copied < bytes) {
2659				error = -EFAULT;
2660				break;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2661			}
 
2662		}
2663put_folios:
2664		for (i = 0; i < folio_batch_count(&fbatch); i++)
2665			folio_put(fbatch.folios[i]);
2666		folio_batch_init(&fbatch);
2667	} while (iov_iter_count(iter) && iocb->ki_pos < isize && !error);
2668
2669	file_accessed(filp);
2670	ra->prev_pos = last_pos;
2671	return already_read ? already_read : error;
2672}
2673EXPORT_SYMBOL_GPL(filemap_read);
2674
2675int kiocb_write_and_wait(struct kiocb *iocb, size_t count)
2676{
2677	struct address_space *mapping = iocb->ki_filp->f_mapping;
2678	loff_t pos = iocb->ki_pos;
2679	loff_t end = pos + count - 1;
2680
2681	if (iocb->ki_flags & IOCB_NOWAIT) {
2682		if (filemap_range_needs_writeback(mapping, pos, end))
2683			return -EAGAIN;
2684		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2685	}
2686
2687	return filemap_write_and_wait_range(mapping, pos, end);
2688}
2689EXPORT_SYMBOL_GPL(kiocb_write_and_wait);
 
 
 
2690
2691int kiocb_invalidate_pages(struct kiocb *iocb, size_t count)
2692{
2693	struct address_space *mapping = iocb->ki_filp->f_mapping;
2694	loff_t pos = iocb->ki_pos;
2695	loff_t end = pos + count - 1;
2696	int ret;
2697
2698	if (iocb->ki_flags & IOCB_NOWAIT) {
2699		/* we could block if there are any pages in the range */
2700		if (filemap_range_has_page(mapping, pos, end))
2701			return -EAGAIN;
2702	} else {
2703		ret = filemap_write_and_wait_range(mapping, pos, end);
2704		if (ret)
2705			return ret;
2706	}
2707
2708	/*
2709	 * After a write we want buffered reads to be sure to go to disk to get
2710	 * the new data.  We invalidate clean cached page from the region we're
2711	 * about to write.  We do this *before* the write so that we can return
2712	 * without clobbering -EIOCBQUEUED from ->direct_IO().
2713	 */
2714	return invalidate_inode_pages2_range(mapping, pos >> PAGE_SHIFT,
2715					     end >> PAGE_SHIFT);
2716}
2717EXPORT_SYMBOL_GPL(kiocb_invalidate_pages);
2718
2719/**
2720 * generic_file_read_iter - generic filesystem read routine
2721 * @iocb:	kernel I/O control block
2722 * @iter:	destination for the data read
2723 *
2724 * This is the "read_iter()" routine for all filesystems
2725 * that can use the page cache directly.
2726 *
2727 * The IOCB_NOWAIT flag in iocb->ki_flags indicates that -EAGAIN shall
2728 * be returned when no data can be read without waiting for I/O requests
2729 * to complete; it doesn't prevent readahead.
2730 *
2731 * The IOCB_NOIO flag in iocb->ki_flags indicates that no new I/O
2732 * requests shall be made for the read or for readahead.  When no data
2733 * can be read, -EAGAIN shall be returned.  When readahead would be
2734 * triggered, a partial, possibly empty read shall be returned.
2735 *
2736 * Return:
2737 * * number of bytes copied, even for partial reads
2738 * * negative error code (or 0 if IOCB_NOIO) if nothing was read
2739 */
2740ssize_t
2741generic_file_read_iter(struct kiocb *iocb, struct iov_iter *iter)
2742{
2743	size_t count = iov_iter_count(iter);
2744	ssize_t retval = 0;
2745
2746	if (!count)
2747		return 0; /* skip atime */
2748
2749	if (iocb->ki_flags & IOCB_DIRECT) {
2750		struct file *file = iocb->ki_filp;
2751		struct address_space *mapping = file->f_mapping;
2752		struct inode *inode = mapping->host;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2753
2754		retval = kiocb_write_and_wait(iocb, count);
2755		if (retval < 0)
2756			return retval;
2757		file_accessed(file);
2758
2759		retval = mapping->a_ops->direct_IO(iocb, iter);
2760		if (retval >= 0) {
2761			iocb->ki_pos += retval;
2762			count -= retval;
2763		}
2764		if (retval != -EIOCBQUEUED)
2765			iov_iter_revert(iter, count - iov_iter_count(iter));
2766
2767		/*
2768		 * Btrfs can have a short DIO read if we encounter
2769		 * compressed extents, so if there was an error, or if
2770		 * we've already read everything we wanted to, or if
2771		 * there was a short read because we hit EOF, go ahead
2772		 * and return.  Otherwise fallthrough to buffered io for
2773		 * the rest of the read.  Buffered reads will not work for
2774		 * DAX files, so don't bother trying.
2775		 */
2776		if (retval < 0 || !count || IS_DAX(inode))
2777			return retval;
2778		if (iocb->ki_pos >= i_size_read(inode))
2779			return retval;
2780	}
2781
2782	return filemap_read(iocb, iter, retval);
 
 
2783}
2784EXPORT_SYMBOL(generic_file_read_iter);
2785
2786/*
2787 * Splice subpages from a folio into a pipe.
2788 */
2789size_t splice_folio_into_pipe(struct pipe_inode_info *pipe,
2790			      struct folio *folio, loff_t fpos, size_t size)
2791{
2792	struct page *page;
2793	size_t spliced = 0, offset = offset_in_folio(folio, fpos);
2794
2795	page = folio_page(folio, offset / PAGE_SIZE);
2796	size = min(size, folio_size(folio) - offset);
2797	offset %= PAGE_SIZE;
2798
2799	while (spliced < size &&
2800	       !pipe_full(pipe->head, pipe->tail, pipe->max_usage)) {
2801		struct pipe_buffer *buf = pipe_head_buf(pipe);
2802		size_t part = min_t(size_t, PAGE_SIZE - offset, size - spliced);
2803
2804		*buf = (struct pipe_buffer) {
2805			.ops	= &page_cache_pipe_buf_ops,
2806			.page	= page,
2807			.offset	= offset,
2808			.len	= part,
2809		};
2810		folio_get(folio);
2811		pipe->head++;
2812		page++;
2813		spliced += part;
2814		offset = 0;
2815	}
2816
2817	return spliced;
2818}
2819
2820/**
2821 * filemap_splice_read -  Splice data from a file's pagecache into a pipe
2822 * @in: The file to read from
2823 * @ppos: Pointer to the file position to read from
2824 * @pipe: The pipe to splice into
2825 * @len: The amount to splice
2826 * @flags: The SPLICE_F_* flags
2827 *
2828 * This function gets folios from a file's pagecache and splices them into the
2829 * pipe.  Readahead will be called as necessary to fill more folios.  This may
2830 * be used for blockdevs also.
2831 *
2832 * Return: On success, the number of bytes read will be returned and *@ppos
2833 * will be updated if appropriate; 0 will be returned if there is no more data
2834 * to be read; -EAGAIN will be returned if the pipe had no space, and some
2835 * other negative error code will be returned on error.  A short read may occur
2836 * if the pipe has insufficient space, we reach the end of the data or we hit a
2837 * hole.
2838 */
2839ssize_t filemap_splice_read(struct file *in, loff_t *ppos,
2840			    struct pipe_inode_info *pipe,
2841			    size_t len, unsigned int flags)
2842{
2843	struct folio_batch fbatch;
2844	struct kiocb iocb;
2845	size_t total_spliced = 0, used, npages;
2846	loff_t isize, end_offset;
2847	bool writably_mapped;
2848	int i, error = 0;
2849
2850	if (unlikely(*ppos >= in->f_mapping->host->i_sb->s_maxbytes))
2851		return 0;
2852
2853	init_sync_kiocb(&iocb, in);
2854	iocb.ki_pos = *ppos;
2855
2856	/* Work out how much data we can actually add into the pipe */
2857	used = pipe_occupancy(pipe->head, pipe->tail);
2858	npages = max_t(ssize_t, pipe->max_usage - used, 0);
2859	len = min_t(size_t, len, npages * PAGE_SIZE);
2860
2861	folio_batch_init(&fbatch);
2862
2863	do {
2864		cond_resched();
2865
2866		if (*ppos >= i_size_read(in->f_mapping->host))
2867			break;
2868
2869		iocb.ki_pos = *ppos;
2870		error = filemap_get_pages(&iocb, len, &fbatch, true);
2871		if (error < 0)
2872			break;
2873
2874		/*
2875		 * i_size must be checked after we know the pages are Uptodate.
2876		 *
2877		 * Checking i_size after the check allows us to calculate
2878		 * the correct value for "nr", which means the zero-filled
2879		 * part of the page is not copied back to userspace (unless
2880		 * another truncate extends the file - this is desired though).
2881		 */
2882		isize = i_size_read(in->f_mapping->host);
2883		if (unlikely(*ppos >= isize))
2884			break;
2885		end_offset = min_t(loff_t, isize, *ppos + len);
2886
2887		/*
2888		 * Once we start copying data, we don't want to be touching any
2889		 * cachelines that might be contended:
2890		 */
2891		writably_mapped = mapping_writably_mapped(in->f_mapping);
2892
2893		for (i = 0; i < folio_batch_count(&fbatch); i++) {
2894			struct folio *folio = fbatch.folios[i];
2895			size_t n;
2896
2897			if (folio_pos(folio) >= end_offset)
2898				goto out;
2899			folio_mark_accessed(folio);
2900
2901			/*
2902			 * If users can be writing to this folio using arbitrary
2903			 * virtual addresses, take care of potential aliasing
2904			 * before reading the folio on the kernel side.
2905			 */
2906			if (writably_mapped)
2907				flush_dcache_folio(folio);
2908
2909			n = min_t(loff_t, len, isize - *ppos);
2910			n = splice_folio_into_pipe(pipe, folio, *ppos, n);
2911			if (!n)
2912				goto out;
2913			len -= n;
2914			total_spliced += n;
2915			*ppos += n;
2916			in->f_ra.prev_pos = *ppos;
2917			if (pipe_full(pipe->head, pipe->tail, pipe->max_usage))
2918				goto out;
2919		}
2920
2921		folio_batch_release(&fbatch);
2922	} while (len);
2923
2924out:
2925	folio_batch_release(&fbatch);
2926	file_accessed(in);
2927
2928	return total_spliced ? total_spliced : error;
2929}
2930EXPORT_SYMBOL(filemap_splice_read);
2931
2932static inline loff_t folio_seek_hole_data(struct xa_state *xas,
2933		struct address_space *mapping, struct folio *folio,
2934		loff_t start, loff_t end, bool seek_data)
2935{
2936	const struct address_space_operations *ops = mapping->a_ops;
2937	size_t offset, bsz = i_blocksize(mapping->host);
2938
2939	if (xa_is_value(folio) || folio_test_uptodate(folio))
2940		return seek_data ? start : end;
2941	if (!ops->is_partially_uptodate)
2942		return seek_data ? end : start;
2943
2944	xas_pause(xas);
2945	rcu_read_unlock();
2946	folio_lock(folio);
2947	if (unlikely(folio->mapping != mapping))
2948		goto unlock;
2949
2950	offset = offset_in_folio(folio, start) & ~(bsz - 1);
2951
2952	do {
2953		if (ops->is_partially_uptodate(folio, offset, bsz) ==
2954							seek_data)
2955			break;
2956		start = (start + bsz) & ~(bsz - 1);
2957		offset += bsz;
2958	} while (offset < folio_size(folio));
2959unlock:
2960	folio_unlock(folio);
2961	rcu_read_lock();
2962	return start;
2963}
2964
2965static inline size_t seek_folio_size(struct xa_state *xas, struct folio *folio)
2966{
2967	if (xa_is_value(folio))
2968		return PAGE_SIZE << xa_get_order(xas->xa, xas->xa_index);
2969	return folio_size(folio);
2970}
2971
2972/**
2973 * mapping_seek_hole_data - Seek for SEEK_DATA / SEEK_HOLE in the page cache.
2974 * @mapping: Address space to search.
2975 * @start: First byte to consider.
2976 * @end: Limit of search (exclusive).
2977 * @whence: Either SEEK_HOLE or SEEK_DATA.
2978 *
2979 * If the page cache knows which blocks contain holes and which blocks
2980 * contain data, your filesystem can use this function to implement
2981 * SEEK_HOLE and SEEK_DATA.  This is useful for filesystems which are
2982 * entirely memory-based such as tmpfs, and filesystems which support
2983 * unwritten extents.
2984 *
2985 * Return: The requested offset on success, or -ENXIO if @whence specifies
2986 * SEEK_DATA and there is no data after @start.  There is an implicit hole
2987 * after @end - 1, so SEEK_HOLE returns @end if all the bytes between @start
2988 * and @end contain data.
2989 */
2990loff_t mapping_seek_hole_data(struct address_space *mapping, loff_t start,
2991		loff_t end, int whence)
2992{
2993	XA_STATE(xas, &mapping->i_pages, start >> PAGE_SHIFT);
2994	pgoff_t max = (end - 1) >> PAGE_SHIFT;
2995	bool seek_data = (whence == SEEK_DATA);
2996	struct folio *folio;
2997
2998	if (end <= start)
2999		return -ENXIO;
3000
3001	rcu_read_lock();
3002	while ((folio = find_get_entry(&xas, max, XA_PRESENT))) {
3003		loff_t pos = (u64)xas.xa_index << PAGE_SHIFT;
3004		size_t seek_size;
3005
3006		if (start < pos) {
3007			if (!seek_data)
3008				goto unlock;
3009			start = pos;
3010		}
3011
3012		seek_size = seek_folio_size(&xas, folio);
3013		pos = round_up((u64)pos + 1, seek_size);
3014		start = folio_seek_hole_data(&xas, mapping, folio, start, pos,
3015				seek_data);
3016		if (start < pos)
3017			goto unlock;
3018		if (start >= end)
3019			break;
3020		if (seek_size > PAGE_SIZE)
3021			xas_set(&xas, pos >> PAGE_SHIFT);
3022		if (!xa_is_value(folio))
3023			folio_put(folio);
3024	}
3025	if (seek_data)
3026		start = -ENXIO;
3027unlock:
3028	rcu_read_unlock();
3029	if (folio && !xa_is_value(folio))
3030		folio_put(folio);
3031	if (start > end)
3032		return end;
3033	return start;
3034}
3035
3036#ifdef CONFIG_MMU
3037#define MMAP_LOTSAMISS  (100)
3038/*
3039 * lock_folio_maybe_drop_mmap - lock the page, possibly dropping the mmap_lock
3040 * @vmf - the vm_fault for this fault.
3041 * @folio - the folio to lock.
3042 * @fpin - the pointer to the file we may pin (or is already pinned).
3043 *
3044 * This works similar to lock_folio_or_retry in that it can drop the
3045 * mmap_lock.  It differs in that it actually returns the folio locked
3046 * if it returns 1 and 0 if it couldn't lock the folio.  If we did have
3047 * to drop the mmap_lock then fpin will point to the pinned file and
3048 * needs to be fput()'ed at a later point.
3049 */
3050static int lock_folio_maybe_drop_mmap(struct vm_fault *vmf, struct folio *folio,
3051				     struct file **fpin)
3052{
3053	if (folio_trylock(folio))
3054		return 1;
3055
3056	/*
3057	 * NOTE! This will make us return with VM_FAULT_RETRY, but with
3058	 * the fault lock still held. That's how FAULT_FLAG_RETRY_NOWAIT
3059	 * is supposed to work. We have way too many special cases..
3060	 */
3061	if (vmf->flags & FAULT_FLAG_RETRY_NOWAIT)
3062		return 0;
3063
3064	*fpin = maybe_unlock_mmap_for_io(vmf, *fpin);
3065	if (vmf->flags & FAULT_FLAG_KILLABLE) {
3066		if (__folio_lock_killable(folio)) {
3067			/*
3068			 * We didn't have the right flags to drop the
3069			 * fault lock, but all fault_handlers only check
3070			 * for fatal signals if we return VM_FAULT_RETRY,
3071			 * so we need to drop the fault lock here and
3072			 * return 0 if we don't have a fpin.
3073			 */
3074			if (*fpin == NULL)
3075				release_fault_lock(vmf);
3076			return 0;
3077		}
3078	} else
3079		__folio_lock(folio);
3080
3081	return 1;
3082}
3083
 
3084/*
3085 * Synchronous readahead happens when we don't even find a page in the page
3086 * cache at all.  We don't want to perform IO under the mmap sem, so if we have
3087 * to drop the mmap sem we return the file that was pinned in order for us to do
3088 * that.  If we didn't pin a file then we return NULL.  The file that is
3089 * returned needs to be fput()'ed when we're done with it.
3090 */
3091static struct file *do_sync_mmap_readahead(struct vm_fault *vmf)
3092{
3093	struct file *file = vmf->vma->vm_file;
3094	struct file_ra_state *ra = &file->f_ra;
3095	struct address_space *mapping = file->f_mapping;
3096	DEFINE_READAHEAD(ractl, file, ra, mapping, vmf->pgoff);
3097	struct file *fpin = NULL;
3098	unsigned long vm_flags = vmf->vma->vm_flags;
3099	unsigned int mmap_miss;
3100
3101#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3102	/* Use the readahead code, even if readahead is disabled */
3103	if (vm_flags & VM_HUGEPAGE) {
3104		fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3105		ractl._index &= ~((unsigned long)HPAGE_PMD_NR - 1);
3106		ra->size = HPAGE_PMD_NR;
3107		/*
3108		 * Fetch two PMD folios, so we get the chance to actually
3109		 * readahead, unless we've been told not to.
3110		 */
3111		if (!(vm_flags & VM_RAND_READ))
3112			ra->size *= 2;
3113		ra->async_size = HPAGE_PMD_NR;
3114		page_cache_ra_order(&ractl, ra, HPAGE_PMD_ORDER);
3115		return fpin;
3116	}
3117#endif
3118
3119	/* If we don't want any read-ahead, don't bother */
3120	if (vm_flags & VM_RAND_READ)
3121		return fpin;
3122	if (!ra->ra_pages)
3123		return fpin;
3124
3125	if (vm_flags & VM_SEQ_READ) {
3126		fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3127		page_cache_sync_ra(&ractl, ra->ra_pages);
 
3128		return fpin;
3129	}
3130
3131	/* Avoid banging the cache line if not needed */
3132	mmap_miss = READ_ONCE(ra->mmap_miss);
3133	if (mmap_miss < MMAP_LOTSAMISS * 10)
3134		WRITE_ONCE(ra->mmap_miss, ++mmap_miss);
3135
3136	/*
3137	 * Do we miss much more than hit in this file? If so,
3138	 * stop bothering with read-ahead. It will only hurt.
3139	 */
3140	if (mmap_miss > MMAP_LOTSAMISS)
3141		return fpin;
3142
3143	/*
3144	 * mmap read-around
3145	 */
3146	fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3147	ra->start = max_t(long, 0, vmf->pgoff - ra->ra_pages / 2);
3148	ra->size = ra->ra_pages;
3149	ra->async_size = ra->ra_pages / 4;
3150	ractl._index = ra->start;
3151	page_cache_ra_order(&ractl, ra, 0);
3152	return fpin;
3153}
3154
3155/*
3156 * Asynchronous readahead happens when we find the page and PG_readahead,
3157 * so we want to possibly extend the readahead further.  We return the file that
3158 * was pinned if we have to drop the mmap_lock in order to do IO.
3159 */
3160static struct file *do_async_mmap_readahead(struct vm_fault *vmf,
3161					    struct folio *folio)
3162{
3163	struct file *file = vmf->vma->vm_file;
3164	struct file_ra_state *ra = &file->f_ra;
3165	DEFINE_READAHEAD(ractl, file, ra, file->f_mapping, vmf->pgoff);
3166	struct file *fpin = NULL;
3167	unsigned int mmap_miss;
 
3168
3169	/* If we don't want any read-ahead, don't bother */
3170	if (vmf->vma->vm_flags & VM_RAND_READ || !ra->ra_pages)
3171		return fpin;
3172
3173	mmap_miss = READ_ONCE(ra->mmap_miss);
3174	if (mmap_miss)
3175		WRITE_ONCE(ra->mmap_miss, --mmap_miss);
3176
3177	if (folio_test_readahead(folio)) {
3178		fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3179		page_cache_async_ra(&ractl, folio, ra->ra_pages);
 
3180	}
3181	return fpin;
3182}
3183
3184static vm_fault_t filemap_fault_recheck_pte_none(struct vm_fault *vmf)
3185{
3186	struct vm_area_struct *vma = vmf->vma;
3187	vm_fault_t ret = 0;
3188	pte_t *ptep;
3189
3190	/*
3191	 * We might have COW'ed a pagecache folio and might now have an mlocked
3192	 * anon folio mapped. The original pagecache folio is not mlocked and
3193	 * might have been evicted. During a read+clear/modify/write update of
3194	 * the PTE, such as done in do_numa_page()/change_pte_range(), we
3195	 * temporarily clear the PTE under PT lock and might detect it here as
3196	 * "none" when not holding the PT lock.
3197	 *
3198	 * Not rechecking the PTE under PT lock could result in an unexpected
3199	 * major fault in an mlock'ed region. Recheck only for this special
3200	 * scenario while holding the PT lock, to not degrade non-mlocked
3201	 * scenarios. Recheck the PTE without PT lock firstly, thereby reducing
3202	 * the number of times we hold PT lock.
3203	 */
3204	if (!(vma->vm_flags & VM_LOCKED))
3205		return 0;
3206
3207	if (!(vmf->flags & FAULT_FLAG_ORIG_PTE_VALID))
3208		return 0;
3209
3210	ptep = pte_offset_map(vmf->pmd, vmf->address);
3211	if (unlikely(!ptep))
3212		return VM_FAULT_NOPAGE;
3213
3214	if (unlikely(!pte_none(ptep_get_lockless(ptep)))) {
3215		ret = VM_FAULT_NOPAGE;
3216	} else {
3217		spin_lock(vmf->ptl);
3218		if (unlikely(!pte_none(ptep_get(ptep))))
3219			ret = VM_FAULT_NOPAGE;
3220		spin_unlock(vmf->ptl);
3221	}
3222	pte_unmap(ptep);
3223	return ret;
3224}
3225
3226/**
3227 * filemap_fault - read in file data for page fault handling
3228 * @vmf:	struct vm_fault containing details of the fault
3229 *
3230 * filemap_fault() is invoked via the vma operations vector for a
3231 * mapped memory region to read in file data during a page fault.
3232 *
3233 * The goto's are kind of ugly, but this streamlines the normal case of having
3234 * it in the page cache, and handles the special cases reasonably without
3235 * having a lot of duplicated code.
3236 *
3237 * vma->vm_mm->mmap_lock must be held on entry.
3238 *
3239 * If our return value has VM_FAULT_RETRY set, it's because the mmap_lock
3240 * may be dropped before doing I/O or by lock_folio_maybe_drop_mmap().
3241 *
3242 * If our return value does not have VM_FAULT_RETRY set, the mmap_lock
3243 * has not been released.
3244 *
3245 * We never return with VM_FAULT_RETRY and a bit from VM_FAULT_ERROR set.
3246 *
3247 * Return: bitwise-OR of %VM_FAULT_ codes.
3248 */
3249vm_fault_t filemap_fault(struct vm_fault *vmf)
3250{
3251	int error;
3252	struct file *file = vmf->vma->vm_file;
3253	struct file *fpin = NULL;
3254	struct address_space *mapping = file->f_mapping;
 
3255	struct inode *inode = mapping->host;
3256	pgoff_t max_idx, index = vmf->pgoff;
3257	struct folio *folio;
 
3258	vm_fault_t ret = 0;
3259	bool mapping_locked = false;
3260
3261	max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3262	if (unlikely(index >= max_idx))
3263		return VM_FAULT_SIGBUS;
3264
3265	/*
3266	 * Do we have something in the page cache already?
3267	 */
3268	folio = filemap_get_folio(mapping, index);
3269	if (likely(!IS_ERR(folio))) {
3270		/*
3271		 * We found the page, so try async readahead before waiting for
3272		 * the lock.
3273		 */
3274		if (!(vmf->flags & FAULT_FLAG_TRIED))
3275			fpin = do_async_mmap_readahead(vmf, folio);
3276		if (unlikely(!folio_test_uptodate(folio))) {
3277			filemap_invalidate_lock_shared(mapping);
3278			mapping_locked = true;
3279		}
3280	} else {
3281		ret = filemap_fault_recheck_pte_none(vmf);
3282		if (unlikely(ret))
3283			return ret;
3284
3285		/* No page in the page cache at all */
3286		count_vm_event(PGMAJFAULT);
3287		count_memcg_event_mm(vmf->vma->vm_mm, PGMAJFAULT);
3288		ret = VM_FAULT_MAJOR;
3289		fpin = do_sync_mmap_readahead(vmf);
3290retry_find:
3291		/*
3292		 * See comment in filemap_create_folio() why we need
3293		 * invalidate_lock
3294		 */
3295		if (!mapping_locked) {
3296			filemap_invalidate_lock_shared(mapping);
3297			mapping_locked = true;
3298		}
3299		folio = __filemap_get_folio(mapping, index,
3300					  FGP_CREAT|FGP_FOR_MMAP,
3301					  vmf->gfp_mask);
3302		if (IS_ERR(folio)) {
3303			if (fpin)
3304				goto out_retry;
3305			filemap_invalidate_unlock_shared(mapping);
3306			return VM_FAULT_OOM;
3307		}
3308	}
3309
3310	if (!lock_folio_maybe_drop_mmap(vmf, folio, &fpin))
3311		goto out_retry;
3312
3313	/* Did it get truncated? */
3314	if (unlikely(folio->mapping != mapping)) {
3315		folio_unlock(folio);
3316		folio_put(folio);
3317		goto retry_find;
3318	}
3319	VM_BUG_ON_FOLIO(!folio_contains(folio, index), folio);
3320
3321	/*
3322	 * We have a locked folio in the page cache, now we need to check
3323	 * that it's up-to-date. If not, it is going to be due to an error,
3324	 * or because readahead was otherwise unable to retrieve it.
3325	 */
3326	if (unlikely(!folio_test_uptodate(folio))) {
3327		/*
3328		 * If the invalidate lock is not held, the folio was in cache
3329		 * and uptodate and now it is not. Strange but possible since we
3330		 * didn't hold the page lock all the time. Let's drop
3331		 * everything, get the invalidate lock and try again.
3332		 */
3333		if (!mapping_locked) {
3334			folio_unlock(folio);
3335			folio_put(folio);
3336			goto retry_find;
3337		}
3338
3339		/*
3340		 * OK, the folio is really not uptodate. This can be because the
3341		 * VMA has the VM_RAND_READ flag set, or because an error
3342		 * arose. Let's read it in directly.
3343		 */
3344		goto page_not_uptodate;
3345	}
3346
3347	/*
3348	 * We've made it this far and we had to drop our mmap_lock, now is the
3349	 * time to return to the upper layer and have it re-find the vma and
3350	 * redo the fault.
3351	 */
3352	if (fpin) {
3353		folio_unlock(folio);
3354		goto out_retry;
3355	}
3356	if (mapping_locked)
3357		filemap_invalidate_unlock_shared(mapping);
3358
3359	/*
3360	 * Found the page and have a reference on it.
3361	 * We must recheck i_size under page lock.
3362	 */
3363	max_idx = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
3364	if (unlikely(index >= max_idx)) {
3365		folio_unlock(folio);
3366		folio_put(folio);
3367		return VM_FAULT_SIGBUS;
3368	}
3369
3370	vmf->page = folio_file_page(folio, index);
3371	return ret | VM_FAULT_LOCKED;
3372
3373page_not_uptodate:
3374	/*
3375	 * Umm, take care of errors if the page isn't up-to-date.
3376	 * Try to re-read it _once_. We do this synchronously,
3377	 * because there really aren't any performance issues here
3378	 * and we need to check for errors.
3379	 */
 
3380	fpin = maybe_unlock_mmap_for_io(vmf, fpin);
3381	error = filemap_read_folio(file, mapping->a_ops->read_folio, folio);
 
 
 
 
 
3382	if (fpin)
3383		goto out_retry;
3384	folio_put(folio);
3385
3386	if (!error || error == AOP_TRUNCATED_PAGE)
3387		goto retry_find;
3388	filemap_invalidate_unlock_shared(mapping);
3389
 
3390	return VM_FAULT_SIGBUS;
3391
3392out_retry:
3393	/*
3394	 * We dropped the mmap_lock, we need to return to the fault handler to
3395	 * re-find the vma and come back and find our hopefully still populated
3396	 * page.
3397	 */
3398	if (!IS_ERR(folio))
3399		folio_put(folio);
3400	if (mapping_locked)
3401		filemap_invalidate_unlock_shared(mapping);
3402	if (fpin)
3403		fput(fpin);
3404	return ret | VM_FAULT_RETRY;
3405}
3406EXPORT_SYMBOL(filemap_fault);
3407
3408static bool filemap_map_pmd(struct vm_fault *vmf, struct folio *folio,
3409		pgoff_t start)
3410{
3411	struct mm_struct *mm = vmf->vma->vm_mm;
 
 
 
 
 
 
3412
3413	/* Huge page is mapped? No need to proceed. */
3414	if (pmd_trans_huge(*vmf->pmd)) {
3415		folio_unlock(folio);
3416		folio_put(folio);
3417		return true;
3418	}
3419
3420	if (pmd_none(*vmf->pmd) && folio_test_pmd_mappable(folio)) {
3421		struct page *page = folio_file_page(folio, start);
3422		vm_fault_t ret = do_set_pmd(vmf, page);
3423		if (!ret) {
3424			/* The page is mapped successfully, reference consumed. */
3425			folio_unlock(folio);
3426			return true;
3427		}
3428	}
3429
3430	if (pmd_none(*vmf->pmd) && vmf->prealloc_pte)
3431		pmd_install(mm, vmf->pmd, &vmf->prealloc_pte);
3432
3433	return false;
3434}
3435
3436static struct folio *next_uptodate_folio(struct xa_state *xas,
3437		struct address_space *mapping, pgoff_t end_pgoff)
3438{
3439	struct folio *folio = xas_next_entry(xas, end_pgoff);
3440	unsigned long max_idx;
3441
3442	do {
3443		if (!folio)
3444			return NULL;
3445		if (xas_retry(xas, folio))
3446			continue;
3447		if (xa_is_value(folio))
3448			continue;
3449		if (folio_test_locked(folio))
3450			continue;
3451		if (!folio_try_get_rcu(folio))
3452			continue;
3453		/* Has the page moved or been split? */
3454		if (unlikely(folio != xas_reload(xas)))
3455			goto skip;
3456		if (!folio_test_uptodate(folio) || folio_test_readahead(folio))
 
 
 
 
3457			goto skip;
3458		if (!folio_trylock(folio))
3459			goto skip;
3460		if (folio->mapping != mapping)
3461			goto unlock;
3462		if (!folio_test_uptodate(folio))
3463			goto unlock;
 
3464		max_idx = DIV_ROUND_UP(i_size_read(mapping->host), PAGE_SIZE);
3465		if (xas->xa_index >= max_idx)
3466			goto unlock;
3467		return folio;
3468unlock:
3469		folio_unlock(folio);
3470skip:
3471		folio_put(folio);
3472	} while ((folio = xas_next_entry(xas, end_pgoff)) != NULL);
3473
3474	return NULL;
3475}
3476
3477/*
3478 * Map page range [start_page, start_page + nr_pages) of folio.
3479 * start_page is gotten from start by folio_page(folio, start)
3480 */
3481static vm_fault_t filemap_map_folio_range(struct vm_fault *vmf,
3482			struct folio *folio, unsigned long start,
3483			unsigned long addr, unsigned int nr_pages,
3484			unsigned int *mmap_miss)
3485{
3486	vm_fault_t ret = 0;
3487	struct page *page = folio_page(folio, start);
3488	unsigned int count = 0;
3489	pte_t *old_ptep = vmf->pte;
3490
3491	do {
3492		if (PageHWPoison(page + count))
3493			goto skip;
3494
3495		(*mmap_miss)++;
3496
3497		/*
3498		 * NOTE: If there're PTE markers, we'll leave them to be
3499		 * handled in the specific fault path, and it'll prohibit the
3500		 * fault-around logic.
3501		 */
3502		if (!pte_none(ptep_get(&vmf->pte[count])))
3503			goto skip;
3504
3505		count++;
3506		continue;
3507skip:
3508		if (count) {
3509			set_pte_range(vmf, folio, page, count, addr);
3510			folio_ref_add(folio, count);
3511			if (in_range(vmf->address, addr, count * PAGE_SIZE))
3512				ret = VM_FAULT_NOPAGE;
3513		}
3514
3515		count++;
3516		page += count;
3517		vmf->pte += count;
3518		addr += count * PAGE_SIZE;
3519		count = 0;
3520	} while (--nr_pages > 0);
3521
3522	if (count) {
3523		set_pte_range(vmf, folio, page, count, addr);
3524		folio_ref_add(folio, count);
3525		if (in_range(vmf->address, addr, count * PAGE_SIZE))
3526			ret = VM_FAULT_NOPAGE;
3527	}
3528
3529	vmf->pte = old_ptep;
3530
3531	return ret;
3532}
3533
3534static vm_fault_t filemap_map_order0_folio(struct vm_fault *vmf,
3535		struct folio *folio, unsigned long addr,
3536		unsigned int *mmap_miss)
3537{
3538	vm_fault_t ret = 0;
3539	struct page *page = &folio->page;
3540
3541	if (PageHWPoison(page))
3542		return ret;
3543
3544	(*mmap_miss)++;
3545
3546	/*
3547	 * NOTE: If there're PTE markers, we'll leave them to be
3548	 * handled in the specific fault path, and it'll prohibit
3549	 * the fault-around logic.
3550	 */
3551	if (!pte_none(ptep_get(vmf->pte)))
3552		return ret;
3553
3554	if (vmf->address == addr)
3555		ret = VM_FAULT_NOPAGE;
3556
3557	set_pte_range(vmf, folio, page, 1, addr);
3558	folio_ref_inc(folio);
3559
3560	return ret;
3561}
3562
3563vm_fault_t filemap_map_pages(struct vm_fault *vmf,
3564			     pgoff_t start_pgoff, pgoff_t end_pgoff)
3565{
3566	struct vm_area_struct *vma = vmf->vma;
3567	struct file *file = vma->vm_file;
3568	struct address_space *mapping = file->f_mapping;
3569	pgoff_t last_pgoff = start_pgoff;
3570	unsigned long addr;
3571	XA_STATE(xas, &mapping->i_pages, start_pgoff);
3572	struct folio *folio;
3573	vm_fault_t ret = 0;
3574	unsigned int nr_pages = 0, mmap_miss = 0, mmap_miss_saved;
3575
3576	rcu_read_lock();
3577	folio = next_uptodate_folio(&xas, mapping, end_pgoff);
3578	if (!folio)
3579		goto out;
3580
3581	if (filemap_map_pmd(vmf, folio, start_pgoff)) {
3582		ret = VM_FAULT_NOPAGE;
3583		goto out;
3584	}
3585
3586	addr = vma->vm_start + ((start_pgoff - vma->vm_pgoff) << PAGE_SHIFT);
3587	vmf->pte = pte_offset_map_lock(vma->vm_mm, vmf->pmd, addr, &vmf->ptl);
3588	if (!vmf->pte) {
3589		folio_unlock(folio);
3590		folio_put(folio);
3591		goto out;
3592	}
3593	do {
3594		unsigned long end;
3595
3596		addr += (xas.xa_index - last_pgoff) << PAGE_SHIFT;
3597		vmf->pte += xas.xa_index - last_pgoff;
3598		last_pgoff = xas.xa_index;
3599		end = folio_next_index(folio) - 1;
3600		nr_pages = min(end, end_pgoff) - xas.xa_index + 1;
3601
3602		if (!folio_test_large(folio))
3603			ret |= filemap_map_order0_folio(vmf,
3604					folio, addr, &mmap_miss);
3605		else
3606			ret |= filemap_map_folio_range(vmf, folio,
3607					xas.xa_index - folio->index, addr,
3608					nr_pages, &mmap_miss);
3609
3610		folio_unlock(folio);
3611		folio_put(folio);
3612	} while ((folio = next_uptodate_folio(&xas, mapping, end_pgoff)) != NULL);
3613	pte_unmap_unlock(vmf->pte, vmf->ptl);
3614out:
3615	rcu_read_unlock();
3616
3617	mmap_miss_saved = READ_ONCE(file->f_ra.mmap_miss);
3618	if (mmap_miss >= mmap_miss_saved)
3619		WRITE_ONCE(file->f_ra.mmap_miss, 0);
3620	else
3621		WRITE_ONCE(file->f_ra.mmap_miss, mmap_miss_saved - mmap_miss);
3622
3623	return ret;
3624}
3625EXPORT_SYMBOL(filemap_map_pages);
3626
3627vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3628{
3629	struct address_space *mapping = vmf->vma->vm_file->f_mapping;
3630	struct folio *folio = page_folio(vmf->page);
3631	vm_fault_t ret = VM_FAULT_LOCKED;
3632
3633	sb_start_pagefault(mapping->host->i_sb);
3634	file_update_time(vmf->vma->vm_file);
3635	folio_lock(folio);
3636	if (folio->mapping != mapping) {
3637		folio_unlock(folio);
3638		ret = VM_FAULT_NOPAGE;
3639		goto out;
3640	}
3641	/*
3642	 * We mark the folio dirty already here so that when freeze is in
3643	 * progress, we are guaranteed that writeback during freezing will
3644	 * see the dirty folio and writeprotect it again.
3645	 */
3646	folio_mark_dirty(folio);
3647	folio_wait_stable(folio);
3648out:
3649	sb_end_pagefault(mapping->host->i_sb);
3650	return ret;
3651}
3652
3653const struct vm_operations_struct generic_file_vm_ops = {
3654	.fault		= filemap_fault,
3655	.map_pages	= filemap_map_pages,
3656	.page_mkwrite	= filemap_page_mkwrite,
3657};
3658
3659/* This is used for a general mmap of a disk file */
3660
3661int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3662{
3663	struct address_space *mapping = file->f_mapping;
3664
3665	if (!mapping->a_ops->read_folio)
3666		return -ENOEXEC;
3667	file_accessed(file);
3668	vma->vm_ops = &generic_file_vm_ops;
3669	return 0;
3670}
3671
3672/*
3673 * This is for filesystems which do not implement ->writepage.
3674 */
3675int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3676{
3677	if (vma_is_shared_maywrite(vma))
3678		return -EINVAL;
3679	return generic_file_mmap(file, vma);
3680}
3681#else
3682vm_fault_t filemap_page_mkwrite(struct vm_fault *vmf)
3683{
3684	return VM_FAULT_SIGBUS;
3685}
3686int generic_file_mmap(struct file *file, struct vm_area_struct *vma)
3687{
3688	return -ENOSYS;
3689}
3690int generic_file_readonly_mmap(struct file *file, struct vm_area_struct *vma)
3691{
3692	return -ENOSYS;
3693}
3694#endif /* CONFIG_MMU */
3695
3696EXPORT_SYMBOL(filemap_page_mkwrite);
3697EXPORT_SYMBOL(generic_file_mmap);
3698EXPORT_SYMBOL(generic_file_readonly_mmap);
3699
3700static struct folio *do_read_cache_folio(struct address_space *mapping,
3701		pgoff_t index, filler_t filler, struct file *file, gfp_t gfp)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3702{
3703	struct folio *folio;
3704	int err;
3705
3706	if (!filler)
3707		filler = mapping->a_ops->read_folio;
3708repeat:
3709	folio = filemap_get_folio(mapping, index);
3710	if (IS_ERR(folio)) {
3711		folio = filemap_alloc_folio(gfp, 0);
3712		if (!folio)
3713			return ERR_PTR(-ENOMEM);
3714		err = filemap_add_folio(mapping, folio, index, gfp);
3715		if (unlikely(err)) {
3716			folio_put(folio);
3717			if (err == -EEXIST)
3718				goto repeat;
3719			/* Presumably ENOMEM for xarray node */
3720			return ERR_PTR(err);
3721		}
3722
3723		goto filler;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3724	}
3725	if (folio_test_uptodate(folio))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3726		goto out;
3727
3728	if (!folio_trylock(folio)) {
3729		folio_put_wait_locked(folio, TASK_UNINTERRUPTIBLE);
3730		goto repeat;
3731	}
3732
3733	/* Folio was truncated from mapping */
3734	if (!folio->mapping) {
3735		folio_unlock(folio);
3736		folio_put(folio);
3737		goto repeat;
3738	}
3739
3740	/* Someone else locked and filled the page in a very small window */
3741	if (folio_test_uptodate(folio)) {
3742		folio_unlock(folio);
3743		goto out;
3744	}
3745
3746filler:
3747	err = filemap_read_folio(file, filler, folio);
3748	if (err) {
3749		folio_put(folio);
3750		if (err == AOP_TRUNCATED_PAGE)
3751			goto repeat;
3752		return ERR_PTR(err);
3753	}
3754
3755out:
3756	folio_mark_accessed(folio);
3757	return folio;
3758}
3759
3760/**
3761 * read_cache_folio - Read into page cache, fill it if needed.
3762 * @mapping: The address_space to read from.
3763 * @index: The index to read.
3764 * @filler: Function to perform the read, or NULL to use aops->read_folio().
3765 * @file: Passed to filler function, may be NULL if not required.
3766 *
3767 * Read one page into the page cache.  If it succeeds, the folio returned
3768 * will contain @index, but it may not be the first page of the folio.
3769 *
3770 * If the filler function returns an error, it will be returned to the
3771 * caller.
3772 *
3773 * Context: May sleep.  Expects mapping->invalidate_lock to be held.
3774 * Return: An uptodate folio on success, ERR_PTR() on failure.
3775 */
3776struct folio *read_cache_folio(struct address_space *mapping, pgoff_t index,
3777		filler_t filler, struct file *file)
3778{
3779	return do_read_cache_folio(mapping, index, filler, file,
3780			mapping_gfp_mask(mapping));
3781}
3782EXPORT_SYMBOL(read_cache_folio);
3783
3784/**
3785 * mapping_read_folio_gfp - Read into page cache, using specified allocation flags.
3786 * @mapping:	The address_space for the folio.
3787 * @index:	The index that the allocated folio will contain.
3788 * @gfp:	The page allocator flags to use if allocating.
3789 *
3790 * This is the same as "read_cache_folio(mapping, index, NULL, NULL)", but with
3791 * any new memory allocations done using the specified allocation flags.
3792 *
3793 * The most likely error from this function is EIO, but ENOMEM is
3794 * possible and so is EINTR.  If ->read_folio returns another error,
3795 * that will be returned to the caller.
3796 *
3797 * The function expects mapping->invalidate_lock to be already held.
3798 *
3799 * Return: Uptodate folio on success, ERR_PTR() on failure.
3800 */
3801struct folio *mapping_read_folio_gfp(struct address_space *mapping,
3802		pgoff_t index, gfp_t gfp)
3803{
3804	return do_read_cache_folio(mapping, index, NULL, NULL, gfp);
3805}
3806EXPORT_SYMBOL(mapping_read_folio_gfp);
3807
3808static struct page *do_read_cache_page(struct address_space *mapping,
3809		pgoff_t index, filler_t *filler, struct file *file, gfp_t gfp)
3810{
3811	struct folio *folio;
3812
3813	folio = do_read_cache_folio(mapping, index, filler, file, gfp);
3814	if (IS_ERR(folio))
3815		return &folio->page;
3816	return folio_file_page(folio, index);
3817}
3818
3819struct page *read_cache_page(struct address_space *mapping,
3820			pgoff_t index, filler_t *filler, struct file *file)
 
 
3821{
3822	return do_read_cache_page(mapping, index, filler, file,
3823			mapping_gfp_mask(mapping));
3824}
3825EXPORT_SYMBOL(read_cache_page);
3826
3827/**
3828 * read_cache_page_gfp - read into page cache, using specified page allocation flags.
3829 * @mapping:	the page's address_space
3830 * @index:	the page index
3831 * @gfp:	the page allocator flags to use if allocating
3832 *
3833 * This is the same as "read_mapping_page(mapping, index, NULL)", but with
3834 * any new page allocations done using the specified allocation flags.
3835 *
3836 * If the page does not get brought uptodate, return -EIO.
3837 *
3838 * The function expects mapping->invalidate_lock to be already held.
3839 *
3840 * Return: up to date page on success, ERR_PTR() on failure.
3841 */
3842struct page *read_cache_page_gfp(struct address_space *mapping,
3843				pgoff_t index,
3844				gfp_t gfp)
3845{
3846	return do_read_cache_page(mapping, index, NULL, NULL, gfp);
3847}
3848EXPORT_SYMBOL(read_cache_page_gfp);
3849
3850/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3851 * Warn about a page cache invalidation failure during a direct I/O write.
3852 */
3853static void dio_warn_stale_pagecache(struct file *filp)
3854{
3855	static DEFINE_RATELIMIT_STATE(_rs, 86400 * HZ, DEFAULT_RATELIMIT_BURST);
3856	char pathname[128];
 
3857	char *path;
3858
3859	errseq_set(&filp->f_mapping->wb_err, -EIO);
3860	if (__ratelimit(&_rs)) {
3861		path = file_path(filp, pathname, sizeof(pathname));
3862		if (IS_ERR(path))
3863			path = "(unknown)";
3864		pr_crit("Page cache invalidation failure on direct I/O.  Possible data corruption due to collision with buffered I/O!\n");
3865		pr_crit("File: %s PID: %d Comm: %.20s\n", path, current->pid,
3866			current->comm);
3867	}
3868}
3869
3870void kiocb_invalidate_post_direct_write(struct kiocb *iocb, size_t count)
 
3871{
3872	struct address_space *mapping = iocb->ki_filp->f_mapping;
 
 
 
 
 
 
3873
3874	if (mapping->nrpages &&
3875	    invalidate_inode_pages2_range(mapping,
3876			iocb->ki_pos >> PAGE_SHIFT,
3877			(iocb->ki_pos + count - 1) >> PAGE_SHIFT))
3878		dio_warn_stale_pagecache(iocb->ki_filp);
3879}
3880
3881ssize_t
3882generic_file_direct_write(struct kiocb *iocb, struct iov_iter *from)
3883{
3884	struct address_space *mapping = iocb->ki_filp->f_mapping;
3885	size_t write_len = iov_iter_count(from);
3886	ssize_t written;
 
 
 
 
 
3887
3888	/*
 
 
 
 
 
 
 
 
3889	 * If a page can not be invalidated, return 0 to fall back
3890	 * to buffered write.
3891	 */
3892	written = kiocb_invalidate_pages(iocb, write_len);
3893	if (written) {
3894		if (written == -EBUSY)
3895			return 0;
3896		return written;
3897	}
3898
3899	written = mapping->a_ops->direct_IO(iocb, from);
3900
3901	/*
3902	 * Finally, try again to invalidate clean pages which might have been
3903	 * cached by non-direct readahead, or faulted in by get_user_pages()
3904	 * if the source of the write was an mmap'ed region of the file
3905	 * we're writing.  Either one is a pretty crazy thing to do,
3906	 * so we don't support it 100%.  If this invalidation
3907	 * fails, tough, the write still worked...
3908	 *
3909	 * Most of the time we do not need this since dio_complete() will do
3910	 * the invalidation for us. However there are some file systems that
3911	 * do not end up with dio_complete() being called, so let's not break
3912	 * them by removing it completely.
3913	 *
3914	 * Noticeable example is a blkdev_direct_IO().
3915	 *
3916	 * Skip invalidation for async writes or if mapping has no pages.
3917	 */
 
 
 
 
3918	if (written > 0) {
3919		struct inode *inode = mapping->host;
3920		loff_t pos = iocb->ki_pos;
3921
3922		kiocb_invalidate_post_direct_write(iocb, written);
3923		pos += written;
3924		write_len -= written;
3925		if (pos > i_size_read(inode) && !S_ISBLK(inode->i_mode)) {
3926			i_size_write(inode, pos);
3927			mark_inode_dirty(inode);
3928		}
3929		iocb->ki_pos = pos;
3930	}
3931	if (written != -EIOCBQUEUED)
3932		iov_iter_revert(from, write_len - iov_iter_count(from));
3933	return written;
3934}
3935EXPORT_SYMBOL(generic_file_direct_write);
3936
3937ssize_t generic_perform_write(struct kiocb *iocb, struct iov_iter *i)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3938{
3939	struct file *file = iocb->ki_filp;
3940	loff_t pos = iocb->ki_pos;
3941	struct address_space *mapping = file->f_mapping;
3942	const struct address_space_operations *a_ops = mapping->a_ops;
3943	long status = 0;
3944	ssize_t written = 0;
 
3945
3946	do {
3947		struct page *page;
3948		unsigned long offset;	/* Offset into pagecache page */
3949		unsigned long bytes;	/* Bytes to write to page */
3950		size_t copied;		/* Bytes copied from user */
3951		void *fsdata = NULL;
3952
3953		offset = (pos & (PAGE_SIZE - 1));
3954		bytes = min_t(unsigned long, PAGE_SIZE - offset,
3955						iov_iter_count(i));
3956
3957again:
3958		/*
3959		 * Bring in the user page that we will copy from _first_.
3960		 * Otherwise there's a nasty deadlock on copying from the
3961		 * same page as we're writing to, without it being marked
3962		 * up-to-date.
 
 
 
 
3963		 */
3964		if (unlikely(fault_in_iov_iter_readable(i, bytes) == bytes)) {
3965			status = -EFAULT;
3966			break;
3967		}
3968
3969		if (fatal_signal_pending(current)) {
3970			status = -EINTR;
3971			break;
3972		}
3973
3974		status = a_ops->write_begin(file, mapping, pos, bytes,
3975						&page, &fsdata);
3976		if (unlikely(status < 0))
3977			break;
3978
3979		if (mapping_writably_mapped(mapping))
3980			flush_dcache_page(page);
3981
3982		copied = copy_page_from_iter_atomic(page, offset, bytes, i);
3983		flush_dcache_page(page);
3984
3985		status = a_ops->write_end(file, mapping, pos, bytes, copied,
3986						page, fsdata);
3987		if (unlikely(status != copied)) {
3988			iov_iter_revert(i, copied - max(status, 0L));
3989			if (unlikely(status < 0))
3990				break;
3991		}
3992		cond_resched();
3993
3994		if (unlikely(status == 0)) {
 
3995			/*
3996			 * A short copy made ->write_end() reject the
3997			 * thing entirely.  Might be memory poisoning
3998			 * halfway through, might be a race with munmap,
3999			 * might be severe memory pressure.
 
 
4000			 */
4001			if (copied)
4002				bytes = copied;
4003			goto again;
4004		}
4005		pos += status;
4006		written += status;
4007
4008		balance_dirty_pages_ratelimited(mapping);
4009	} while (iov_iter_count(i));
4010
4011	if (!written)
4012		return status;
4013	iocb->ki_pos += written;
4014	return written;
4015}
4016EXPORT_SYMBOL(generic_perform_write);
4017
4018/**
4019 * __generic_file_write_iter - write data to a file
4020 * @iocb:	IO state structure (file, offset, etc.)
4021 * @from:	iov_iter with data to write
4022 *
4023 * This function does all the work needed for actually writing data to a
4024 * file. It does all basic checks, removes SUID from the file, updates
4025 * modification times and calls proper subroutines depending on whether we
4026 * do direct IO or a standard buffered write.
4027 *
4028 * It expects i_rwsem to be grabbed unless we work on a block device or similar
4029 * object which does not need locking at all.
4030 *
4031 * This function does *not* take care of syncing data in case of O_SYNC write.
4032 * A caller has to handle it. This is mainly due to the fact that we want to
4033 * avoid syncing under i_rwsem.
4034 *
4035 * Return:
4036 * * number of bytes written, even for truncated writes
4037 * * negative error code if no data has been written at all
4038 */
4039ssize_t __generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
4040{
4041	struct file *file = iocb->ki_filp;
4042	struct address_space *mapping = file->f_mapping;
4043	struct inode *inode = mapping->host;
4044	ssize_t ret;
 
 
 
 
 
 
 
 
4045
4046	ret = file_remove_privs(file);
4047	if (ret)
4048		return ret;
4049
4050	ret = file_update_time(file);
4051	if (ret)
4052		return ret;
4053
4054	if (iocb->ki_flags & IOCB_DIRECT) {
4055		ret = generic_file_direct_write(iocb, from);
4056		/*
4057		 * If the write stopped short of completing, fall back to
4058		 * buffered writes.  Some filesystems do this for writes to
4059		 * holes, for example.  For DAX files, a buffered write will
4060		 * not succeed (even if it did, DAX does not handle dirty
4061		 * page-cache pages correctly).
4062		 */
4063		if (ret < 0 || !iov_iter_count(from) || IS_DAX(inode))
4064			return ret;
4065		return direct_write_fallback(iocb, from, ret,
4066				generic_perform_write(iocb, from));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
4067	}
4068
4069	return generic_perform_write(iocb, from);
 
4070}
4071EXPORT_SYMBOL(__generic_file_write_iter);
4072
4073/**
4074 * generic_file_write_iter - write data to a file
4075 * @iocb:	IO state structure
4076 * @from:	iov_iter with data to write
4077 *
4078 * This is a wrapper around __generic_file_write_iter() to be used by most
4079 * filesystems. It takes care of syncing the file in case of O_SYNC file
4080 * and acquires i_rwsem as needed.
4081 * Return:
4082 * * negative error code if no data has been written at all of
4083 *   vfs_fsync_range() failed for a synchronous write
4084 * * number of bytes written, even for truncated writes
4085 */
4086ssize_t generic_file_write_iter(struct kiocb *iocb, struct iov_iter *from)
4087{
4088	struct file *file = iocb->ki_filp;
4089	struct inode *inode = file->f_mapping->host;
4090	ssize_t ret;
4091
4092	inode_lock(inode);
4093	ret = generic_write_checks(iocb, from);
4094	if (ret > 0)
4095		ret = __generic_file_write_iter(iocb, from);
4096	inode_unlock(inode);
4097
4098	if (ret > 0)
4099		ret = generic_write_sync(iocb, ret);
4100	return ret;
4101}
4102EXPORT_SYMBOL(generic_file_write_iter);
4103
4104/**
4105 * filemap_release_folio() - Release fs-specific metadata on a folio.
4106 * @folio: The folio which the kernel is trying to free.
4107 * @gfp: Memory allocation flags (and I/O mode).
 
4108 *
4109 * The address_space is trying to release any data attached to a folio
4110 * (presumably at folio->private).
4111 *
4112 * This will also be called if the private_2 flag is set on a page,
4113 * indicating that the folio has other metadata associated with it.
4114 *
4115 * The @gfp argument specifies whether I/O may be performed to release
4116 * this page (__GFP_IO), and whether the call may block
4117 * (__GFP_RECLAIM & __GFP_FS).
4118 *
4119 * Return: %true if the release was successful, otherwise %false.
4120 */
4121bool filemap_release_folio(struct folio *folio, gfp_t gfp)
4122{
4123	struct address_space * const mapping = folio->mapping;
4124
4125	BUG_ON(!folio_test_locked(folio));
4126	if (!folio_needs_release(folio))
4127		return true;
4128	if (folio_test_writeback(folio))
4129		return false;
4130
4131	if (mapping && mapping->a_ops->release_folio)
4132		return mapping->a_ops->release_folio(folio, gfp);
4133	return try_to_free_buffers(folio);
4134}
4135EXPORT_SYMBOL(filemap_release_folio);
4136
4137#ifdef CONFIG_CACHESTAT_SYSCALL
4138/**
4139 * filemap_cachestat() - compute the page cache statistics of a mapping
4140 * @mapping:	The mapping to compute the statistics for.
4141 * @first_index:	The starting page cache index.
4142 * @last_index:	The final page index (inclusive).
4143 * @cs:	the cachestat struct to write the result to.
4144 *
4145 * This will query the page cache statistics of a mapping in the
4146 * page range of [first_index, last_index] (inclusive). The statistics
4147 * queried include: number of dirty pages, number of pages marked for
4148 * writeback, and the number of (recently) evicted pages.
4149 */
4150static void filemap_cachestat(struct address_space *mapping,
4151		pgoff_t first_index, pgoff_t last_index, struct cachestat *cs)
4152{
4153	XA_STATE(xas, &mapping->i_pages, first_index);
4154	struct folio *folio;
4155
4156	rcu_read_lock();
4157	xas_for_each(&xas, folio, last_index) {
4158		int order;
4159		unsigned long nr_pages;
4160		pgoff_t folio_first_index, folio_last_index;
4161
4162		/*
4163		 * Don't deref the folio. It is not pinned, and might
4164		 * get freed (and reused) underneath us.
4165		 *
4166		 * We *could* pin it, but that would be expensive for
4167		 * what should be a fast and lightweight syscall.
4168		 *
4169		 * Instead, derive all information of interest from
4170		 * the rcu-protected xarray.
4171		 */
4172
4173		if (xas_retry(&xas, folio))
4174			continue;
4175
4176		order = xa_get_order(xas.xa, xas.xa_index);
4177		nr_pages = 1 << order;
4178		folio_first_index = round_down(xas.xa_index, 1 << order);
4179		folio_last_index = folio_first_index + nr_pages - 1;
4180
4181		/* Folios might straddle the range boundaries, only count covered pages */
4182		if (folio_first_index < first_index)
4183			nr_pages -= first_index - folio_first_index;
4184
4185		if (folio_last_index > last_index)
4186			nr_pages -= folio_last_index - last_index;
4187
4188		if (xa_is_value(folio)) {
4189			/* page is evicted */
4190			void *shadow = (void *)folio;
4191			bool workingset; /* not used */
4192
4193			cs->nr_evicted += nr_pages;
4194
4195#ifdef CONFIG_SWAP /* implies CONFIG_MMU */
4196			if (shmem_mapping(mapping)) {
4197				/* shmem file - in swap cache */
4198				swp_entry_t swp = radix_to_swp_entry(folio);
4199
4200				/* swapin error results in poisoned entry */
4201				if (non_swap_entry(swp))
4202					goto resched;
4203
4204				/*
4205				 * Getting a swap entry from the shmem
4206				 * inode means we beat
4207				 * shmem_unuse(). rcu_read_lock()
4208				 * ensures swapoff waits for us before
4209				 * freeing the swapper space. However,
4210				 * we can race with swapping and
4211				 * invalidation, so there might not be
4212				 * a shadow in the swapcache (yet).
4213				 */
4214				shadow = get_shadow_from_swap_cache(swp);
4215				if (!shadow)
4216					goto resched;
4217			}
4218#endif
4219			if (workingset_test_recent(shadow, true, &workingset))
4220				cs->nr_recently_evicted += nr_pages;
4221
4222			goto resched;
4223		}
4224
4225		/* page is in cache */
4226		cs->nr_cache += nr_pages;
4227
4228		if (xas_get_mark(&xas, PAGECACHE_TAG_DIRTY))
4229			cs->nr_dirty += nr_pages;
4230
4231		if (xas_get_mark(&xas, PAGECACHE_TAG_WRITEBACK))
4232			cs->nr_writeback += nr_pages;
4233
4234resched:
4235		if (need_resched()) {
4236			xas_pause(&xas);
4237			cond_resched_rcu();
4238		}
4239	}
4240	rcu_read_unlock();
4241}
4242
4243/*
4244 * The cachestat(2) system call.
4245 *
4246 * cachestat() returns the page cache statistics of a file in the
4247 * bytes range specified by `off` and `len`: number of cached pages,
4248 * number of dirty pages, number of pages marked for writeback,
4249 * number of evicted pages, and number of recently evicted pages.
4250 *
4251 * An evicted page is a page that is previously in the page cache
4252 * but has been evicted since. A page is recently evicted if its last
4253 * eviction was recent enough that its reentry to the cache would
4254 * indicate that it is actively being used by the system, and that
4255 * there is memory pressure on the system.
4256 *
4257 * `off` and `len` must be non-negative integers. If `len` > 0,
4258 * the queried range is [`off`, `off` + `len`]. If `len` == 0,
4259 * we will query in the range from `off` to the end of the file.
4260 *
4261 * The `flags` argument is unused for now, but is included for future
4262 * extensibility. User should pass 0 (i.e no flag specified).
4263 *
4264 * Currently, hugetlbfs is not supported.
4265 *
4266 * Because the status of a page can change after cachestat() checks it
4267 * but before it returns to the application, the returned values may
4268 * contain stale information.
4269 *
4270 * return values:
4271 *  zero        - success
4272 *  -EFAULT     - cstat or cstat_range points to an illegal address
4273 *  -EINVAL     - invalid flags
4274 *  -EBADF      - invalid file descriptor
4275 *  -EOPNOTSUPP - file descriptor is of a hugetlbfs file
4276 */
4277SYSCALL_DEFINE4(cachestat, unsigned int, fd,
4278		struct cachestat_range __user *, cstat_range,
4279		struct cachestat __user *, cstat, unsigned int, flags)
4280{
4281	struct fd f = fdget(fd);
4282	struct address_space *mapping;
4283	struct cachestat_range csr;
4284	struct cachestat cs;
4285	pgoff_t first_index, last_index;
4286
4287	if (!f.file)
4288		return -EBADF;
4289
4290	if (copy_from_user(&csr, cstat_range,
4291			sizeof(struct cachestat_range))) {
4292		fdput(f);
4293		return -EFAULT;
4294	}
4295
4296	/* hugetlbfs is not supported */
4297	if (is_file_hugepages(f.file)) {
4298		fdput(f);
4299		return -EOPNOTSUPP;
4300	}
4301
4302	if (flags != 0) {
4303		fdput(f);
4304		return -EINVAL;
4305	}
4306
4307	first_index = csr.off >> PAGE_SHIFT;
4308	last_index =
4309		csr.len == 0 ? ULONG_MAX : (csr.off + csr.len - 1) >> PAGE_SHIFT;
4310	memset(&cs, 0, sizeof(struct cachestat));
4311	mapping = f.file->f_mapping;
4312	filemap_cachestat(mapping, first_index, last_index, &cs);
4313	fdput(f);
4314
4315	if (copy_to_user(cstat, &cs, sizeof(struct cachestat)))
4316		return -EFAULT;
4317
4318	return 0;
4319}
4320#endif /* CONFIG_CACHESTAT_SYSCALL */