Linux Audio

Check our new training course

Loading...
v5.9
   1/*
   2 * Resizable virtual memory filesystem for Linux.
   3 *
   4 * Copyright (C) 2000 Linus Torvalds.
   5 *		 2000 Transmeta Corp.
   6 *		 2000-2001 Christoph Rohland
   7 *		 2000-2001 SAP AG
   8 *		 2002 Red Hat Inc.
   9 * Copyright (C) 2002-2011 Hugh Dickins.
  10 * Copyright (C) 2011 Google Inc.
  11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
  12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
  13 *
  14 * Extended attribute support for tmpfs:
  15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
  16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
  17 *
  18 * tiny-shmem:
  19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
  20 *
  21 * This file is released under the GPL.
  22 */
  23
  24#include <linux/fs.h>
  25#include <linux/init.h>
  26#include <linux/vfs.h>
  27#include <linux/mount.h>
  28#include <linux/ramfs.h>
  29#include <linux/pagemap.h>
  30#include <linux/file.h>
  31#include <linux/mm.h>
  32#include <linux/random.h>
  33#include <linux/sched/signal.h>
  34#include <linux/export.h>
  35#include <linux/swap.h>
  36#include <linux/uio.h>
  37#include <linux/khugepaged.h>
  38#include <linux/hugetlb.h>
  39#include <linux/frontswap.h>
  40#include <linux/fs_parser.h>
  41
  42#include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */
  43
  44static struct vfsmount *shm_mnt;
  45
  46#ifdef CONFIG_SHMEM
  47/*
  48 * This virtual memory filesystem is heavily based on the ramfs. It
  49 * extends ramfs by the ability to use swap and honor resource limits
  50 * which makes it a completely usable filesystem.
  51 */
  52
  53#include <linux/xattr.h>
  54#include <linux/exportfs.h>
  55#include <linux/posix_acl.h>
  56#include <linux/posix_acl_xattr.h>
  57#include <linux/mman.h>
  58#include <linux/string.h>
  59#include <linux/slab.h>
  60#include <linux/backing-dev.h>
  61#include <linux/shmem_fs.h>
  62#include <linux/writeback.h>
  63#include <linux/blkdev.h>
  64#include <linux/pagevec.h>
  65#include <linux/percpu_counter.h>
  66#include <linux/falloc.h>
  67#include <linux/splice.h>
  68#include <linux/security.h>
  69#include <linux/swapops.h>
  70#include <linux/mempolicy.h>
  71#include <linux/namei.h>
  72#include <linux/ctype.h>
  73#include <linux/migrate.h>
  74#include <linux/highmem.h>
  75#include <linux/seq_file.h>
  76#include <linux/magic.h>
  77#include <linux/syscalls.h>
  78#include <linux/fcntl.h>
  79#include <uapi/linux/memfd.h>
  80#include <linux/userfaultfd_k.h>
  81#include <linux/rmap.h>
  82#include <linux/uuid.h>
  83
  84#include <linux/uaccess.h>
 
  85
  86#include "internal.h"
  87
  88#define BLOCKS_PER_PAGE  (PAGE_SIZE/512)
  89#define VM_ACCT(size)    (PAGE_ALIGN(size) >> PAGE_SHIFT)
  90
  91/* Pretend that each entry is of this size in directory's i_size */
  92#define BOGO_DIRENT_SIZE 20
  93
  94/* Symlink up to this size is kmalloc'ed instead of using a swappable page */
  95#define SHORT_SYMLINK_LEN 128
  96
  97/*
  98 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
  99 * inode->i_private (with i_mutex making sure that it has only one user at
 100 * a time): we would prefer not to enlarge the shmem inode just for that.
 101 */
 102struct shmem_falloc {
 103	wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
 104	pgoff_t start;		/* start of range currently being fallocated */
 105	pgoff_t next;		/* the next page offset to be fallocated */
 106	pgoff_t nr_falloced;	/* how many new pages have been fallocated */
 107	pgoff_t nr_unswapped;	/* how often writepage refused to swap out */
 108};
 109
 110struct shmem_options {
 111	unsigned long long blocks;
 112	unsigned long long inodes;
 113	struct mempolicy *mpol;
 114	kuid_t uid;
 115	kgid_t gid;
 116	umode_t mode;
 117	bool full_inums;
 118	int huge;
 119	int seen;
 120#define SHMEM_SEEN_BLOCKS 1
 121#define SHMEM_SEEN_INODES 2
 122#define SHMEM_SEEN_HUGE 4
 123#define SHMEM_SEEN_INUMS 8
 124};
 125
 126#ifdef CONFIG_TMPFS
 127static unsigned long shmem_default_max_blocks(void)
 128{
 129	return totalram_pages() / 2;
 130}
 131
 132static unsigned long shmem_default_max_inodes(void)
 133{
 134	unsigned long nr_pages = totalram_pages();
 135
 136	return min(nr_pages - totalhigh_pages(), nr_pages / 2);
 137}
 138#endif
 139
 140static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
 141static int shmem_replace_page(struct page **pagep, gfp_t gfp,
 142				struct shmem_inode_info *info, pgoff_t index);
 143static int shmem_swapin_page(struct inode *inode, pgoff_t index,
 144			     struct page **pagep, enum sgp_type sgp,
 145			     gfp_t gfp, struct vm_area_struct *vma,
 146			     vm_fault_t *fault_type);
 147static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
 148		struct page **pagep, enum sgp_type sgp,
 149		gfp_t gfp, struct vm_area_struct *vma,
 150		struct vm_fault *vmf, vm_fault_t *fault_type);
 151
 152int shmem_getpage(struct inode *inode, pgoff_t index,
 153		struct page **pagep, enum sgp_type sgp)
 154{
 155	return shmem_getpage_gfp(inode, index, pagep, sgp,
 156		mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
 157}
 158
 159static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
 160{
 161	return sb->s_fs_info;
 162}
 163
 164/*
 165 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
 166 * for shared memory and for shared anonymous (/dev/zero) mappings
 167 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
 168 * consistent with the pre-accounting of private mappings ...
 169 */
 170static inline int shmem_acct_size(unsigned long flags, loff_t size)
 171{
 172	return (flags & VM_NORESERVE) ?
 173		0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
 174}
 175
 176static inline void shmem_unacct_size(unsigned long flags, loff_t size)
 177{
 178	if (!(flags & VM_NORESERVE))
 179		vm_unacct_memory(VM_ACCT(size));
 180}
 181
 182static inline int shmem_reacct_size(unsigned long flags,
 183		loff_t oldsize, loff_t newsize)
 184{
 185	if (!(flags & VM_NORESERVE)) {
 186		if (VM_ACCT(newsize) > VM_ACCT(oldsize))
 187			return security_vm_enough_memory_mm(current->mm,
 188					VM_ACCT(newsize) - VM_ACCT(oldsize));
 189		else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
 190			vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
 191	}
 192	return 0;
 193}
 194
 195/*
 196 * ... whereas tmpfs objects are accounted incrementally as
 197 * pages are allocated, in order to allow large sparse files.
 198 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
 199 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
 200 */
 201static inline int shmem_acct_block(unsigned long flags, long pages)
 202{
 203	if (!(flags & VM_NORESERVE))
 204		return 0;
 205
 206	return security_vm_enough_memory_mm(current->mm,
 207			pages * VM_ACCT(PAGE_SIZE));
 208}
 209
 210static inline void shmem_unacct_blocks(unsigned long flags, long pages)
 211{
 212	if (flags & VM_NORESERVE)
 213		vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
 214}
 215
 216static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
 217{
 218	struct shmem_inode_info *info = SHMEM_I(inode);
 219	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 220
 221	if (shmem_acct_block(info->flags, pages))
 222		return false;
 223
 224	if (sbinfo->max_blocks) {
 225		if (percpu_counter_compare(&sbinfo->used_blocks,
 226					   sbinfo->max_blocks - pages) > 0)
 227			goto unacct;
 228		percpu_counter_add(&sbinfo->used_blocks, pages);
 229	}
 230
 231	return true;
 232
 233unacct:
 234	shmem_unacct_blocks(info->flags, pages);
 235	return false;
 236}
 237
 238static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
 239{
 240	struct shmem_inode_info *info = SHMEM_I(inode);
 241	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 242
 243	if (sbinfo->max_blocks)
 244		percpu_counter_sub(&sbinfo->used_blocks, pages);
 245	shmem_unacct_blocks(info->flags, pages);
 246}
 247
 248static const struct super_operations shmem_ops;
 249static const struct address_space_operations shmem_aops;
 250static const struct file_operations shmem_file_operations;
 251static const struct inode_operations shmem_inode_operations;
 252static const struct inode_operations shmem_dir_inode_operations;
 253static const struct inode_operations shmem_special_inode_operations;
 254static const struct vm_operations_struct shmem_vm_ops;
 255static struct file_system_type shmem_fs_type;
 256
 257bool vma_is_shmem(struct vm_area_struct *vma)
 258{
 259	return vma->vm_ops == &shmem_vm_ops;
 260}
 261
 262static LIST_HEAD(shmem_swaplist);
 263static DEFINE_MUTEX(shmem_swaplist_mutex);
 264
 265/*
 266 * shmem_reserve_inode() performs bookkeeping to reserve a shmem inode, and
 267 * produces a novel ino for the newly allocated inode.
 268 *
 269 * It may also be called when making a hard link to permit the space needed by
 270 * each dentry. However, in that case, no new inode number is needed since that
 271 * internally draws from another pool of inode numbers (currently global
 272 * get_next_ino()). This case is indicated by passing NULL as inop.
 273 */
 274#define SHMEM_INO_BATCH 1024
 275static int shmem_reserve_inode(struct super_block *sb, ino_t *inop)
 276{
 277	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 278	ino_t ino;
 279
 280	if (!(sb->s_flags & SB_KERNMOUNT)) {
 281		spin_lock(&sbinfo->stat_lock);
 282		if (sbinfo->max_inodes) {
 283			if (!sbinfo->free_inodes) {
 284				spin_unlock(&sbinfo->stat_lock);
 285				return -ENOSPC;
 286			}
 287			sbinfo->free_inodes--;
 288		}
 289		if (inop) {
 290			ino = sbinfo->next_ino++;
 291			if (unlikely(is_zero_ino(ino)))
 292				ino = sbinfo->next_ino++;
 293			if (unlikely(!sbinfo->full_inums &&
 294				     ino > UINT_MAX)) {
 295				/*
 296				 * Emulate get_next_ino uint wraparound for
 297				 * compatibility
 298				 */
 299				if (IS_ENABLED(CONFIG_64BIT))
 300					pr_warn("%s: inode number overflow on device %d, consider using inode64 mount option\n",
 301						__func__, MINOR(sb->s_dev));
 302				sbinfo->next_ino = 1;
 303				ino = sbinfo->next_ino++;
 304			}
 305			*inop = ino;
 306		}
 307		spin_unlock(&sbinfo->stat_lock);
 308	} else if (inop) {
 309		/*
 310		 * __shmem_file_setup, one of our callers, is lock-free: it
 311		 * doesn't hold stat_lock in shmem_reserve_inode since
 312		 * max_inodes is always 0, and is called from potentially
 313		 * unknown contexts. As such, use a per-cpu batched allocator
 314		 * which doesn't require the per-sb stat_lock unless we are at
 315		 * the batch boundary.
 316		 *
 317		 * We don't need to worry about inode{32,64} since SB_KERNMOUNT
 318		 * shmem mounts are not exposed to userspace, so we don't need
 319		 * to worry about things like glibc compatibility.
 320		 */
 321		ino_t *next_ino;
 322		next_ino = per_cpu_ptr(sbinfo->ino_batch, get_cpu());
 323		ino = *next_ino;
 324		if (unlikely(ino % SHMEM_INO_BATCH == 0)) {
 325			spin_lock(&sbinfo->stat_lock);
 326			ino = sbinfo->next_ino;
 327			sbinfo->next_ino += SHMEM_INO_BATCH;
 328			spin_unlock(&sbinfo->stat_lock);
 329			if (unlikely(is_zero_ino(ino)))
 330				ino++;
 331		}
 332		*inop = ino;
 333		*next_ino = ++ino;
 334		put_cpu();
 335	}
 336
 337	return 0;
 338}
 339
 340static void shmem_free_inode(struct super_block *sb)
 341{
 342	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 343	if (sbinfo->max_inodes) {
 344		spin_lock(&sbinfo->stat_lock);
 345		sbinfo->free_inodes++;
 346		spin_unlock(&sbinfo->stat_lock);
 347	}
 348}
 349
 350/**
 351 * shmem_recalc_inode - recalculate the block usage of an inode
 352 * @inode: inode to recalc
 353 *
 354 * We have to calculate the free blocks since the mm can drop
 355 * undirtied hole pages behind our back.
 356 *
 357 * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
 358 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
 359 *
 360 * It has to be called with the spinlock held.
 361 */
 362static void shmem_recalc_inode(struct inode *inode)
 363{
 364	struct shmem_inode_info *info = SHMEM_I(inode);
 365	long freed;
 366
 367	freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
 368	if (freed > 0) {
 369		info->alloced -= freed;
 370		inode->i_blocks -= freed * BLOCKS_PER_PAGE;
 371		shmem_inode_unacct_blocks(inode, freed);
 372	}
 373}
 374
 375bool shmem_charge(struct inode *inode, long pages)
 376{
 377	struct shmem_inode_info *info = SHMEM_I(inode);
 378	unsigned long flags;
 379
 380	if (!shmem_inode_acct_block(inode, pages))
 381		return false;
 382
 383	/* nrpages adjustment first, then shmem_recalc_inode() when balanced */
 384	inode->i_mapping->nrpages += pages;
 385
 386	spin_lock_irqsave(&info->lock, flags);
 387	info->alloced += pages;
 388	inode->i_blocks += pages * BLOCKS_PER_PAGE;
 389	shmem_recalc_inode(inode);
 390	spin_unlock_irqrestore(&info->lock, flags);
 391
 392	return true;
 393}
 394
 395void shmem_uncharge(struct inode *inode, long pages)
 396{
 397	struct shmem_inode_info *info = SHMEM_I(inode);
 398	unsigned long flags;
 399
 400	/* nrpages adjustment done by __delete_from_page_cache() or caller */
 401
 402	spin_lock_irqsave(&info->lock, flags);
 403	info->alloced -= pages;
 404	inode->i_blocks -= pages * BLOCKS_PER_PAGE;
 405	shmem_recalc_inode(inode);
 406	spin_unlock_irqrestore(&info->lock, flags);
 407
 408	shmem_inode_unacct_blocks(inode, pages);
 409}
 410
 411/*
 412 * Replace item expected in xarray by a new item, while holding xa_lock.
 413 */
 414static int shmem_replace_entry(struct address_space *mapping,
 415			pgoff_t index, void *expected, void *replacement)
 416{
 417	XA_STATE(xas, &mapping->i_pages, index);
 418	void *item;
 419
 420	VM_BUG_ON(!expected);
 421	VM_BUG_ON(!replacement);
 422	item = xas_load(&xas);
 423	if (item != expected)
 424		return -ENOENT;
 425	xas_store(&xas, replacement);
 426	return 0;
 427}
 428
 429/*
 430 * Sometimes, before we decide whether to proceed or to fail, we must check
 431 * that an entry was not already brought back from swap by a racing thread.
 432 *
 433 * Checking page is not enough: by the time a SwapCache page is locked, it
 434 * might be reused, and again be SwapCache, using the same swap as before.
 435 */
 436static bool shmem_confirm_swap(struct address_space *mapping,
 437			       pgoff_t index, swp_entry_t swap)
 438{
 439	return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
 440}
 441
 442/*
 443 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
 444 *
 445 * SHMEM_HUGE_NEVER:
 446 *	disables huge pages for the mount;
 447 * SHMEM_HUGE_ALWAYS:
 448 *	enables huge pages for the mount;
 449 * SHMEM_HUGE_WITHIN_SIZE:
 450 *	only allocate huge pages if the page will be fully within i_size,
 451 *	also respect fadvise()/madvise() hints;
 452 * SHMEM_HUGE_ADVISE:
 453 *	only allocate huge pages if requested with fadvise()/madvise();
 454 */
 455
 456#define SHMEM_HUGE_NEVER	0
 457#define SHMEM_HUGE_ALWAYS	1
 458#define SHMEM_HUGE_WITHIN_SIZE	2
 459#define SHMEM_HUGE_ADVISE	3
 460
 461/*
 462 * Special values.
 463 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
 464 *
 465 * SHMEM_HUGE_DENY:
 466 *	disables huge on shm_mnt and all mounts, for emergency use;
 467 * SHMEM_HUGE_FORCE:
 468 *	enables huge on shm_mnt and all mounts, w/o needing option, for testing;
 469 *
 470 */
 471#define SHMEM_HUGE_DENY		(-1)
 472#define SHMEM_HUGE_FORCE	(-2)
 473
 474#ifdef CONFIG_TRANSPARENT_HUGEPAGE
 475/* ifdef here to avoid bloating shmem.o when not necessary */
 476
 477static int shmem_huge __read_mostly;
 478
 479#if defined(CONFIG_SYSFS)
 480static int shmem_parse_huge(const char *str)
 481{
 482	if (!strcmp(str, "never"))
 483		return SHMEM_HUGE_NEVER;
 484	if (!strcmp(str, "always"))
 485		return SHMEM_HUGE_ALWAYS;
 486	if (!strcmp(str, "within_size"))
 487		return SHMEM_HUGE_WITHIN_SIZE;
 488	if (!strcmp(str, "advise"))
 489		return SHMEM_HUGE_ADVISE;
 490	if (!strcmp(str, "deny"))
 491		return SHMEM_HUGE_DENY;
 492	if (!strcmp(str, "force"))
 493		return SHMEM_HUGE_FORCE;
 494	return -EINVAL;
 495}
 496#endif
 497
 498#if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
 499static const char *shmem_format_huge(int huge)
 500{
 501	switch (huge) {
 502	case SHMEM_HUGE_NEVER:
 503		return "never";
 504	case SHMEM_HUGE_ALWAYS:
 505		return "always";
 506	case SHMEM_HUGE_WITHIN_SIZE:
 507		return "within_size";
 508	case SHMEM_HUGE_ADVISE:
 509		return "advise";
 510	case SHMEM_HUGE_DENY:
 511		return "deny";
 512	case SHMEM_HUGE_FORCE:
 513		return "force";
 514	default:
 515		VM_BUG_ON(1);
 516		return "bad_val";
 517	}
 518}
 519#endif
 520
 521static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
 522		struct shrink_control *sc, unsigned long nr_to_split)
 523{
 524	LIST_HEAD(list), *pos, *next;
 525	LIST_HEAD(to_remove);
 526	struct inode *inode;
 527	struct shmem_inode_info *info;
 528	struct page *page;
 529	unsigned long batch = sc ? sc->nr_to_scan : 128;
 530	int removed = 0, split = 0;
 531
 532	if (list_empty(&sbinfo->shrinklist))
 533		return SHRINK_STOP;
 534
 535	spin_lock(&sbinfo->shrinklist_lock);
 536	list_for_each_safe(pos, next, &sbinfo->shrinklist) {
 537		info = list_entry(pos, struct shmem_inode_info, shrinklist);
 538
 539		/* pin the inode */
 540		inode = igrab(&info->vfs_inode);
 541
 542		/* inode is about to be evicted */
 543		if (!inode) {
 544			list_del_init(&info->shrinklist);
 545			removed++;
 546			goto next;
 547		}
 548
 549		/* Check if there's anything to gain */
 550		if (round_up(inode->i_size, PAGE_SIZE) ==
 551				round_up(inode->i_size, HPAGE_PMD_SIZE)) {
 552			list_move(&info->shrinklist, &to_remove);
 553			removed++;
 554			goto next;
 555		}
 556
 557		list_move(&info->shrinklist, &list);
 558next:
 559		if (!--batch)
 560			break;
 561	}
 562	spin_unlock(&sbinfo->shrinklist_lock);
 563
 564	list_for_each_safe(pos, next, &to_remove) {
 565		info = list_entry(pos, struct shmem_inode_info, shrinklist);
 566		inode = &info->vfs_inode;
 567		list_del_init(&info->shrinklist);
 568		iput(inode);
 569	}
 570
 571	list_for_each_safe(pos, next, &list) {
 572		int ret;
 573
 574		info = list_entry(pos, struct shmem_inode_info, shrinklist);
 575		inode = &info->vfs_inode;
 576
 577		if (nr_to_split && split >= nr_to_split)
 578			goto leave;
 579
 580		page = find_get_page(inode->i_mapping,
 581				(inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
 582		if (!page)
 583			goto drop;
 584
 585		/* No huge page at the end of the file: nothing to split */
 586		if (!PageTransHuge(page)) {
 587			put_page(page);
 588			goto drop;
 589		}
 590
 591		/*
 592		 * Leave the inode on the list if we failed to lock
 593		 * the page at this time.
 594		 *
 595		 * Waiting for the lock may lead to deadlock in the
 596		 * reclaim path.
 597		 */
 598		if (!trylock_page(page)) {
 599			put_page(page);
 600			goto leave;
 601		}
 602
 603		ret = split_huge_page(page);
 604		unlock_page(page);
 605		put_page(page);
 606
 607		/* If split failed leave the inode on the list */
 608		if (ret)
 609			goto leave;
 610
 611		split++;
 612drop:
 613		list_del_init(&info->shrinklist);
 614		removed++;
 615leave:
 616		iput(inode);
 617	}
 618
 619	spin_lock(&sbinfo->shrinklist_lock);
 620	list_splice_tail(&list, &sbinfo->shrinklist);
 621	sbinfo->shrinklist_len -= removed;
 622	spin_unlock(&sbinfo->shrinklist_lock);
 623
 624	return split;
 625}
 626
 627static long shmem_unused_huge_scan(struct super_block *sb,
 628		struct shrink_control *sc)
 629{
 630	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 631
 632	if (!READ_ONCE(sbinfo->shrinklist_len))
 633		return SHRINK_STOP;
 634
 635	return shmem_unused_huge_shrink(sbinfo, sc, 0);
 636}
 637
 638static long shmem_unused_huge_count(struct super_block *sb,
 639		struct shrink_control *sc)
 640{
 641	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 642	return READ_ONCE(sbinfo->shrinklist_len);
 643}
 644#else /* !CONFIG_TRANSPARENT_HUGEPAGE */
 645
 646#define shmem_huge SHMEM_HUGE_DENY
 647
 648static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
 649		struct shrink_control *sc, unsigned long nr_to_split)
 650{
 651	return 0;
 652}
 653#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 654
 655static inline bool is_huge_enabled(struct shmem_sb_info *sbinfo)
 656{
 657	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
 658	    (shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge) &&
 659	    shmem_huge != SHMEM_HUGE_DENY)
 660		return true;
 661	return false;
 662}
 663
 664/*
 665 * Like add_to_page_cache_locked, but error if expected item has gone.
 666 */
 667static int shmem_add_to_page_cache(struct page *page,
 668				   struct address_space *mapping,
 669				   pgoff_t index, void *expected, gfp_t gfp,
 670				   struct mm_struct *charge_mm)
 671{
 672	XA_STATE_ORDER(xas, &mapping->i_pages, index, compound_order(page));
 673	unsigned long i = 0;
 674	unsigned long nr = compound_nr(page);
 675	int error;
 676
 677	VM_BUG_ON_PAGE(PageTail(page), page);
 678	VM_BUG_ON_PAGE(index != round_down(index, nr), page);
 679	VM_BUG_ON_PAGE(!PageLocked(page), page);
 680	VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
 681	VM_BUG_ON(expected && PageTransHuge(page));
 682
 683	page_ref_add(page, nr);
 684	page->mapping = mapping;
 685	page->index = index;
 686
 687	if (!PageSwapCache(page)) {
 688		error = mem_cgroup_charge(page, charge_mm, gfp);
 689		if (error) {
 690			if (PageTransHuge(page)) {
 691				count_vm_event(THP_FILE_FALLBACK);
 692				count_vm_event(THP_FILE_FALLBACK_CHARGE);
 693			}
 694			goto error;
 695		}
 696	}
 697	cgroup_throttle_swaprate(page, gfp);
 698
 699	do {
 700		void *entry;
 701		xas_lock_irq(&xas);
 702		entry = xas_find_conflict(&xas);
 703		if (entry != expected)
 704			xas_set_err(&xas, -EEXIST);
 705		xas_create_range(&xas);
 706		if (xas_error(&xas))
 707			goto unlock;
 708next:
 709		xas_store(&xas, page);
 710		if (++i < nr) {
 711			xas_next(&xas);
 712			goto next;
 713		}
 714		if (PageTransHuge(page)) {
 715			count_vm_event(THP_FILE_ALLOC);
 716			__inc_node_page_state(page, NR_SHMEM_THPS);
 717		}
 718		mapping->nrpages += nr;
 719		__mod_lruvec_page_state(page, NR_FILE_PAGES, nr);
 720		__mod_lruvec_page_state(page, NR_SHMEM, nr);
 721unlock:
 722		xas_unlock_irq(&xas);
 723	} while (xas_nomem(&xas, gfp));
 724
 725	if (xas_error(&xas)) {
 726		error = xas_error(&xas);
 727		goto error;
 
 728	}
 729
 730	return 0;
 731error:
 732	page->mapping = NULL;
 733	page_ref_sub(page, nr);
 734	return error;
 735}
 736
 737/*
 738 * Like delete_from_page_cache, but substitutes swap for page.
 739 */
 740static void shmem_delete_from_page_cache(struct page *page, void *radswap)
 741{
 742	struct address_space *mapping = page->mapping;
 743	int error;
 744
 745	VM_BUG_ON_PAGE(PageCompound(page), page);
 746
 747	xa_lock_irq(&mapping->i_pages);
 748	error = shmem_replace_entry(mapping, page->index, page, radswap);
 749	page->mapping = NULL;
 750	mapping->nrpages--;
 751	__dec_lruvec_page_state(page, NR_FILE_PAGES);
 752	__dec_lruvec_page_state(page, NR_SHMEM);
 753	xa_unlock_irq(&mapping->i_pages);
 754	put_page(page);
 755	BUG_ON(error);
 756}
 757
 758/*
 759 * Remove swap entry from page cache, free the swap and its page cache.
 760 */
 761static int shmem_free_swap(struct address_space *mapping,
 762			   pgoff_t index, void *radswap)
 763{
 764	void *old;
 765
 766	old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
 767	if (old != radswap)
 768		return -ENOENT;
 769	free_swap_and_cache(radix_to_swp_entry(radswap));
 770	return 0;
 771}
 772
 773/*
 774 * Determine (in bytes) how many of the shmem object's pages mapped by the
 775 * given offsets are swapped out.
 776 *
 777 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
 778 * as long as the inode doesn't go away and racy results are not a problem.
 779 */
 780unsigned long shmem_partial_swap_usage(struct address_space *mapping,
 781						pgoff_t start, pgoff_t end)
 782{
 783	XA_STATE(xas, &mapping->i_pages, start);
 784	struct page *page;
 785	unsigned long swapped = 0;
 786
 787	rcu_read_lock();
 788	xas_for_each(&xas, page, end - 1) {
 789		if (xas_retry(&xas, page))
 790			continue;
 791		if (xa_is_value(page))
 792			swapped++;
 793
 794		if (need_resched()) {
 795			xas_pause(&xas);
 796			cond_resched_rcu();
 797		}
 798	}
 799
 800	rcu_read_unlock();
 801
 802	return swapped << PAGE_SHIFT;
 803}
 804
 805/*
 806 * Determine (in bytes) how many of the shmem object's pages mapped by the
 807 * given vma is swapped out.
 808 *
 809 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
 810 * as long as the inode doesn't go away and racy results are not a problem.
 811 */
 812unsigned long shmem_swap_usage(struct vm_area_struct *vma)
 813{
 814	struct inode *inode = file_inode(vma->vm_file);
 815	struct shmem_inode_info *info = SHMEM_I(inode);
 816	struct address_space *mapping = inode->i_mapping;
 817	unsigned long swapped;
 818
 819	/* Be careful as we don't hold info->lock */
 820	swapped = READ_ONCE(info->swapped);
 821
 822	/*
 823	 * The easier cases are when the shmem object has nothing in swap, or
 824	 * the vma maps it whole. Then we can simply use the stats that we
 825	 * already track.
 826	 */
 827	if (!swapped)
 828		return 0;
 829
 830	if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
 831		return swapped << PAGE_SHIFT;
 832
 833	/* Here comes the more involved part */
 834	return shmem_partial_swap_usage(mapping,
 835			linear_page_index(vma, vma->vm_start),
 836			linear_page_index(vma, vma->vm_end));
 837}
 838
 839/*
 840 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
 841 */
 842void shmem_unlock_mapping(struct address_space *mapping)
 843{
 844	struct pagevec pvec;
 845	pgoff_t indices[PAGEVEC_SIZE];
 846	pgoff_t index = 0;
 847
 848	pagevec_init(&pvec);
 849	/*
 850	 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
 851	 */
 852	while (!mapping_unevictable(mapping)) {
 853		/*
 854		 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
 855		 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
 856		 */
 857		pvec.nr = find_get_entries(mapping, index,
 858					   PAGEVEC_SIZE, pvec.pages, indices);
 859		if (!pvec.nr)
 860			break;
 861		index = indices[pvec.nr - 1] + 1;
 862		pagevec_remove_exceptionals(&pvec);
 863		check_move_unevictable_pages(&pvec);
 864		pagevec_release(&pvec);
 865		cond_resched();
 866	}
 867}
 868
 869/*
 870 * Check whether a hole-punch or truncation needs to split a huge page,
 871 * returning true if no split was required, or the split has been successful.
 872 *
 873 * Eviction (or truncation to 0 size) should never need to split a huge page;
 874 * but in rare cases might do so, if shmem_undo_range() failed to trylock on
 875 * head, and then succeeded to trylock on tail.
 876 *
 877 * A split can only succeed when there are no additional references on the
 878 * huge page: so the split below relies upon find_get_entries() having stopped
 879 * when it found a subpage of the huge page, without getting further references.
 880 */
 881static bool shmem_punch_compound(struct page *page, pgoff_t start, pgoff_t end)
 882{
 883	if (!PageTransCompound(page))
 884		return true;
 885
 886	/* Just proceed to delete a huge page wholly within the range punched */
 887	if (PageHead(page) &&
 888	    page->index >= start && page->index + HPAGE_PMD_NR <= end)
 889		return true;
 890
 891	/* Try to split huge page, so we can truly punch the hole or truncate */
 892	return split_huge_page(page) >= 0;
 893}
 894
 895/*
 896 * Remove range of pages and swap entries from page cache, and free them.
 897 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
 898 */
 899static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
 900								 bool unfalloc)
 901{
 902	struct address_space *mapping = inode->i_mapping;
 903	struct shmem_inode_info *info = SHMEM_I(inode);
 904	pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
 905	pgoff_t end = (lend + 1) >> PAGE_SHIFT;
 906	unsigned int partial_start = lstart & (PAGE_SIZE - 1);
 907	unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
 908	struct pagevec pvec;
 909	pgoff_t indices[PAGEVEC_SIZE];
 910	long nr_swaps_freed = 0;
 911	pgoff_t index;
 912	int i;
 913
 914	if (lend == -1)
 915		end = -1;	/* unsigned, so actually very big */
 916
 917	pagevec_init(&pvec);
 918	index = start;
 919	while (index < end) {
 920		pvec.nr = find_get_entries(mapping, index,
 921			min(end - index, (pgoff_t)PAGEVEC_SIZE),
 922			pvec.pages, indices);
 923		if (!pvec.nr)
 924			break;
 925		for (i = 0; i < pagevec_count(&pvec); i++) {
 926			struct page *page = pvec.pages[i];
 927
 928			index = indices[i];
 929			if (index >= end)
 930				break;
 931
 932			if (xa_is_value(page)) {
 933				if (unfalloc)
 934					continue;
 935				nr_swaps_freed += !shmem_free_swap(mapping,
 936								index, page);
 937				continue;
 938			}
 939
 940			VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
 941
 942			if (!trylock_page(page))
 943				continue;
 944
 945			if ((!unfalloc || !PageUptodate(page)) &&
 946			    page_mapping(page) == mapping) {
 947				VM_BUG_ON_PAGE(PageWriteback(page), page);
 948				if (shmem_punch_compound(page, start, end))
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 949					truncate_inode_page(mapping, page);
 
 950			}
 951			unlock_page(page);
 952		}
 953		pagevec_remove_exceptionals(&pvec);
 954		pagevec_release(&pvec);
 955		cond_resched();
 956		index++;
 957	}
 958
 959	if (partial_start) {
 960		struct page *page = NULL;
 961		shmem_getpage(inode, start - 1, &page, SGP_READ);
 962		if (page) {
 963			unsigned int top = PAGE_SIZE;
 964			if (start > end) {
 965				top = partial_end;
 966				partial_end = 0;
 967			}
 968			zero_user_segment(page, partial_start, top);
 969			set_page_dirty(page);
 970			unlock_page(page);
 971			put_page(page);
 972		}
 973	}
 974	if (partial_end) {
 975		struct page *page = NULL;
 976		shmem_getpage(inode, end, &page, SGP_READ);
 977		if (page) {
 978			zero_user_segment(page, 0, partial_end);
 979			set_page_dirty(page);
 980			unlock_page(page);
 981			put_page(page);
 982		}
 983	}
 984	if (start >= end)
 985		return;
 986
 987	index = start;
 988	while (index < end) {
 989		cond_resched();
 990
 991		pvec.nr = find_get_entries(mapping, index,
 992				min(end - index, (pgoff_t)PAGEVEC_SIZE),
 993				pvec.pages, indices);
 994		if (!pvec.nr) {
 995			/* If all gone or hole-punch or unfalloc, we're done */
 996			if (index == start || end != -1)
 997				break;
 998			/* But if truncating, restart to make sure all gone */
 999			index = start;
1000			continue;
1001		}
1002		for (i = 0; i < pagevec_count(&pvec); i++) {
1003			struct page *page = pvec.pages[i];
1004
1005			index = indices[i];
1006			if (index >= end)
1007				break;
1008
1009			if (xa_is_value(page)) {
1010				if (unfalloc)
1011					continue;
1012				if (shmem_free_swap(mapping, index, page)) {
1013					/* Swap was replaced by page: retry */
1014					index--;
1015					break;
1016				}
1017				nr_swaps_freed++;
1018				continue;
1019			}
1020
1021			lock_page(page);
1022
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1023			if (!unfalloc || !PageUptodate(page)) {
1024				if (page_mapping(page) != mapping) {
 
 
 
 
1025					/* Page was replaced by swap: retry */
1026					unlock_page(page);
1027					index--;
1028					break;
1029				}
1030				VM_BUG_ON_PAGE(PageWriteback(page), page);
1031				if (shmem_punch_compound(page, start, end))
1032					truncate_inode_page(mapping, page);
1033				else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
1034					/* Wipe the page and don't get stuck */
1035					clear_highpage(page);
1036					flush_dcache_page(page);
1037					set_page_dirty(page);
1038					if (index <
1039					    round_up(start, HPAGE_PMD_NR))
1040						start = index + 1;
1041				}
1042			}
1043			unlock_page(page);
1044		}
1045		pagevec_remove_exceptionals(&pvec);
1046		pagevec_release(&pvec);
1047		index++;
1048	}
1049
1050	spin_lock_irq(&info->lock);
1051	info->swapped -= nr_swaps_freed;
1052	shmem_recalc_inode(inode);
1053	spin_unlock_irq(&info->lock);
1054}
1055
1056void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
1057{
1058	shmem_undo_range(inode, lstart, lend, false);
1059	inode->i_ctime = inode->i_mtime = current_time(inode);
1060}
1061EXPORT_SYMBOL_GPL(shmem_truncate_range);
1062
1063static int shmem_getattr(const struct path *path, struct kstat *stat,
1064			 u32 request_mask, unsigned int query_flags)
1065{
1066	struct inode *inode = path->dentry->d_inode;
1067	struct shmem_inode_info *info = SHMEM_I(inode);
1068	struct shmem_sb_info *sb_info = SHMEM_SB(inode->i_sb);
1069
1070	if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
1071		spin_lock_irq(&info->lock);
1072		shmem_recalc_inode(inode);
1073		spin_unlock_irq(&info->lock);
1074	}
1075	generic_fillattr(inode, stat);
1076
1077	if (is_huge_enabled(sb_info))
1078		stat->blksize = HPAGE_PMD_SIZE;
1079
1080	return 0;
1081}
1082
1083static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
1084{
1085	struct inode *inode = d_inode(dentry);
1086	struct shmem_inode_info *info = SHMEM_I(inode);
1087	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1088	int error;
1089
1090	error = setattr_prepare(dentry, attr);
1091	if (error)
1092		return error;
1093
1094	if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1095		loff_t oldsize = inode->i_size;
1096		loff_t newsize = attr->ia_size;
1097
1098		/* protected by i_mutex */
1099		if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1100		    (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1101			return -EPERM;
1102
1103		if (newsize != oldsize) {
1104			error = shmem_reacct_size(SHMEM_I(inode)->flags,
1105					oldsize, newsize);
1106			if (error)
1107				return error;
1108			i_size_write(inode, newsize);
1109			inode->i_ctime = inode->i_mtime = current_time(inode);
1110		}
1111		if (newsize <= oldsize) {
1112			loff_t holebegin = round_up(newsize, PAGE_SIZE);
1113			if (oldsize > holebegin)
1114				unmap_mapping_range(inode->i_mapping,
1115							holebegin, 0, 1);
1116			if (info->alloced)
1117				shmem_truncate_range(inode,
1118							newsize, (loff_t)-1);
1119			/* unmap again to remove racily COWed private pages */
1120			if (oldsize > holebegin)
1121				unmap_mapping_range(inode->i_mapping,
1122							holebegin, 0, 1);
1123
1124			/*
1125			 * Part of the huge page can be beyond i_size: subject
1126			 * to shrink under memory pressure.
1127			 */
1128			if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE)) {
1129				spin_lock(&sbinfo->shrinklist_lock);
1130				/*
1131				 * _careful to defend against unlocked access to
1132				 * ->shrink_list in shmem_unused_huge_shrink()
1133				 */
1134				if (list_empty_careful(&info->shrinklist)) {
1135					list_add_tail(&info->shrinklist,
1136							&sbinfo->shrinklist);
1137					sbinfo->shrinklist_len++;
1138				}
1139				spin_unlock(&sbinfo->shrinklist_lock);
1140			}
1141		}
1142	}
1143
1144	setattr_copy(inode, attr);
1145	if (attr->ia_valid & ATTR_MODE)
1146		error = posix_acl_chmod(inode, inode->i_mode);
1147	return error;
1148}
1149
1150static void shmem_evict_inode(struct inode *inode)
1151{
1152	struct shmem_inode_info *info = SHMEM_I(inode);
1153	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1154
1155	if (inode->i_mapping->a_ops == &shmem_aops) {
1156		shmem_unacct_size(info->flags, inode->i_size);
1157		inode->i_size = 0;
1158		shmem_truncate_range(inode, 0, (loff_t)-1);
1159		if (!list_empty(&info->shrinklist)) {
1160			spin_lock(&sbinfo->shrinklist_lock);
1161			if (!list_empty(&info->shrinklist)) {
1162				list_del_init(&info->shrinklist);
1163				sbinfo->shrinklist_len--;
1164			}
1165			spin_unlock(&sbinfo->shrinklist_lock);
1166		}
1167		while (!list_empty(&info->swaplist)) {
1168			/* Wait while shmem_unuse() is scanning this inode... */
1169			wait_var_event(&info->stop_eviction,
1170				       !atomic_read(&info->stop_eviction));
1171			mutex_lock(&shmem_swaplist_mutex);
1172			/* ...but beware of the race if we peeked too early */
1173			if (!atomic_read(&info->stop_eviction))
1174				list_del_init(&info->swaplist);
1175			mutex_unlock(&shmem_swaplist_mutex);
1176		}
1177	}
1178
1179	simple_xattrs_free(&info->xattrs);
1180	WARN_ON(inode->i_blocks);
1181	shmem_free_inode(inode->i_sb);
1182	clear_inode(inode);
1183}
1184
1185extern struct swap_info_struct *swap_info[];
1186
1187static int shmem_find_swap_entries(struct address_space *mapping,
1188				   pgoff_t start, unsigned int nr_entries,
1189				   struct page **entries, pgoff_t *indices,
1190				   unsigned int type, bool frontswap)
1191{
1192	XA_STATE(xas, &mapping->i_pages, start);
1193	struct page *page;
1194	swp_entry_t entry;
1195	unsigned int ret = 0;
1196
1197	if (!nr_entries)
1198		return 0;
1199
1200	rcu_read_lock();
1201	xas_for_each(&xas, page, ULONG_MAX) {
1202		if (xas_retry(&xas, page))
1203			continue;
1204
1205		if (!xa_is_value(page))
1206			continue;
1207
1208		entry = radix_to_swp_entry(page);
1209		if (swp_type(entry) != type)
1210			continue;
1211		if (frontswap &&
1212		    !frontswap_test(swap_info[type], swp_offset(entry)))
1213			continue;
1214
1215		indices[ret] = xas.xa_index;
1216		entries[ret] = page;
1217
1218		if (need_resched()) {
1219			xas_pause(&xas);
1220			cond_resched_rcu();
1221		}
1222		if (++ret == nr_entries)
1223			break;
1224	}
1225	rcu_read_unlock();
1226
1227	return ret;
1228}
1229
1230/*
1231 * Move the swapped pages for an inode to page cache. Returns the count
1232 * of pages swapped in, or the error in case of failure.
1233 */
1234static int shmem_unuse_swap_entries(struct inode *inode, struct pagevec pvec,
1235				    pgoff_t *indices)
1236{
1237	int i = 0;
1238	int ret = 0;
1239	int error = 0;
1240	struct address_space *mapping = inode->i_mapping;
1241
1242	for (i = 0; i < pvec.nr; i++) {
1243		struct page *page = pvec.pages[i];
1244
1245		if (!xa_is_value(page))
1246			continue;
1247		error = shmem_swapin_page(inode, indices[i],
1248					  &page, SGP_CACHE,
1249					  mapping_gfp_mask(mapping),
1250					  NULL, NULL);
1251		if (error == 0) {
1252			unlock_page(page);
1253			put_page(page);
1254			ret++;
1255		}
1256		if (error == -ENOMEM)
1257			break;
1258		error = 0;
1259	}
1260	return error ? error : ret;
1261}
1262
1263/*
1264 * If swap found in inode, free it and move page from swapcache to filecache.
1265 */
1266static int shmem_unuse_inode(struct inode *inode, unsigned int type,
1267			     bool frontswap, unsigned long *fs_pages_to_unuse)
1268{
1269	struct address_space *mapping = inode->i_mapping;
1270	pgoff_t start = 0;
1271	struct pagevec pvec;
1272	pgoff_t indices[PAGEVEC_SIZE];
1273	bool frontswap_partial = (frontswap && *fs_pages_to_unuse > 0);
1274	int ret = 0;
1275
1276	pagevec_init(&pvec);
1277	do {
1278		unsigned int nr_entries = PAGEVEC_SIZE;
1279
1280		if (frontswap_partial && *fs_pages_to_unuse < PAGEVEC_SIZE)
1281			nr_entries = *fs_pages_to_unuse;
1282
1283		pvec.nr = shmem_find_swap_entries(mapping, start, nr_entries,
1284						  pvec.pages, indices,
1285						  type, frontswap);
1286		if (pvec.nr == 0) {
1287			ret = 0;
1288			break;
1289		}
1290
1291		ret = shmem_unuse_swap_entries(inode, pvec, indices);
1292		if (ret < 0)
1293			break;
1294
1295		if (frontswap_partial) {
1296			*fs_pages_to_unuse -= ret;
1297			if (*fs_pages_to_unuse == 0) {
1298				ret = FRONTSWAP_PAGES_UNUSED;
1299				break;
1300			}
1301		}
1302
1303		start = indices[pvec.nr - 1];
1304	} while (true);
1305
1306	return ret;
1307}
1308
1309/*
1310 * Read all the shared memory data that resides in the swap
1311 * device 'type' back into memory, so the swap device can be
1312 * unused.
1313 */
1314int shmem_unuse(unsigned int type, bool frontswap,
1315		unsigned long *fs_pages_to_unuse)
1316{
1317	struct shmem_inode_info *info, *next;
1318	int error = 0;
1319
1320	if (list_empty(&shmem_swaplist))
1321		return 0;
1322
1323	mutex_lock(&shmem_swaplist_mutex);
1324	list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1325		if (!info->swapped) {
1326			list_del_init(&info->swaplist);
1327			continue;
1328		}
1329		/*
1330		 * Drop the swaplist mutex while searching the inode for swap;
1331		 * but before doing so, make sure shmem_evict_inode() will not
1332		 * remove placeholder inode from swaplist, nor let it be freed
1333		 * (igrab() would protect from unlink, but not from unmount).
1334		 */
1335		atomic_inc(&info->stop_eviction);
1336		mutex_unlock(&shmem_swaplist_mutex);
1337
1338		error = shmem_unuse_inode(&info->vfs_inode, type, frontswap,
1339					  fs_pages_to_unuse);
1340		cond_resched();
1341
1342		mutex_lock(&shmem_swaplist_mutex);
1343		next = list_next_entry(info, swaplist);
1344		if (!info->swapped)
1345			list_del_init(&info->swaplist);
1346		if (atomic_dec_and_test(&info->stop_eviction))
1347			wake_up_var(&info->stop_eviction);
1348		if (error)
1349			break;
1350	}
1351	mutex_unlock(&shmem_swaplist_mutex);
1352
1353	return error;
1354}
1355
1356/*
1357 * Move the page from the page cache to the swap cache.
1358 */
1359static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1360{
1361	struct shmem_inode_info *info;
1362	struct address_space *mapping;
1363	struct inode *inode;
1364	swp_entry_t swap;
1365	pgoff_t index;
1366
1367	VM_BUG_ON_PAGE(PageCompound(page), page);
1368	BUG_ON(!PageLocked(page));
1369	mapping = page->mapping;
1370	index = page->index;
1371	inode = mapping->host;
1372	info = SHMEM_I(inode);
1373	if (info->flags & VM_LOCKED)
1374		goto redirty;
1375	if (!total_swap_pages)
1376		goto redirty;
1377
1378	/*
1379	 * Our capabilities prevent regular writeback or sync from ever calling
1380	 * shmem_writepage; but a stacking filesystem might use ->writepage of
1381	 * its underlying filesystem, in which case tmpfs should write out to
1382	 * swap only in response to memory pressure, and not for the writeback
1383	 * threads or sync.
1384	 */
1385	if (!wbc->for_reclaim) {
1386		WARN_ON_ONCE(1);	/* Still happens? Tell us about it! */
1387		goto redirty;
1388	}
1389
1390	/*
1391	 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1392	 * value into swapfile.c, the only way we can correctly account for a
1393	 * fallocated page arriving here is now to initialize it and write it.
1394	 *
1395	 * That's okay for a page already fallocated earlier, but if we have
1396	 * not yet completed the fallocation, then (a) we want to keep track
1397	 * of this page in case we have to undo it, and (b) it may not be a
1398	 * good idea to continue anyway, once we're pushing into swap.  So
1399	 * reactivate the page, and let shmem_fallocate() quit when too many.
1400	 */
1401	if (!PageUptodate(page)) {
1402		if (inode->i_private) {
1403			struct shmem_falloc *shmem_falloc;
1404			spin_lock(&inode->i_lock);
1405			shmem_falloc = inode->i_private;
1406			if (shmem_falloc &&
1407			    !shmem_falloc->waitq &&
1408			    index >= shmem_falloc->start &&
1409			    index < shmem_falloc->next)
1410				shmem_falloc->nr_unswapped++;
1411			else
1412				shmem_falloc = NULL;
1413			spin_unlock(&inode->i_lock);
1414			if (shmem_falloc)
1415				goto redirty;
1416		}
1417		clear_highpage(page);
1418		flush_dcache_page(page);
1419		SetPageUptodate(page);
1420	}
1421
1422	swap = get_swap_page(page);
1423	if (!swap.val)
1424		goto redirty;
1425
1426	/*
1427	 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1428	 * if it's not already there.  Do it now before the page is
1429	 * moved to swap cache, when its pagelock no longer protects
1430	 * the inode from eviction.  But don't unlock the mutex until
1431	 * we've incremented swapped, because shmem_unuse_inode() will
1432	 * prune a !swapped inode from the swaplist under this mutex.
1433	 */
1434	mutex_lock(&shmem_swaplist_mutex);
1435	if (list_empty(&info->swaplist))
1436		list_add(&info->swaplist, &shmem_swaplist);
1437
1438	if (add_to_swap_cache(page, swap,
1439			__GFP_HIGH | __GFP_NOMEMALLOC | __GFP_NOWARN,
1440			NULL) == 0) {
1441		spin_lock_irq(&info->lock);
1442		shmem_recalc_inode(inode);
1443		info->swapped++;
1444		spin_unlock_irq(&info->lock);
1445
1446		swap_shmem_alloc(swap);
1447		shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1448
1449		mutex_unlock(&shmem_swaplist_mutex);
1450		BUG_ON(page_mapped(page));
1451		swap_writepage(page, wbc);
1452		return 0;
1453	}
1454
1455	mutex_unlock(&shmem_swaplist_mutex);
1456	put_swap_page(page, swap);
1457redirty:
1458	set_page_dirty(page);
1459	if (wbc->for_reclaim)
1460		return AOP_WRITEPAGE_ACTIVATE;	/* Return with page locked */
1461	unlock_page(page);
1462	return 0;
1463}
1464
1465#if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1466static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1467{
1468	char buffer[64];
1469
1470	if (!mpol || mpol->mode == MPOL_DEFAULT)
1471		return;		/* show nothing */
1472
1473	mpol_to_str(buffer, sizeof(buffer), mpol);
1474
1475	seq_printf(seq, ",mpol=%s", buffer);
1476}
1477
1478static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1479{
1480	struct mempolicy *mpol = NULL;
1481	if (sbinfo->mpol) {
1482		spin_lock(&sbinfo->stat_lock);	/* prevent replace/use races */
1483		mpol = sbinfo->mpol;
1484		mpol_get(mpol);
1485		spin_unlock(&sbinfo->stat_lock);
1486	}
1487	return mpol;
1488}
1489#else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1490static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1491{
1492}
1493static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1494{
1495	return NULL;
1496}
1497#endif /* CONFIG_NUMA && CONFIG_TMPFS */
1498#ifndef CONFIG_NUMA
1499#define vm_policy vm_private_data
1500#endif
1501
1502static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1503		struct shmem_inode_info *info, pgoff_t index)
1504{
1505	/* Create a pseudo vma that just contains the policy */
1506	vma_init(vma, NULL);
1507	/* Bias interleave by inode number to distribute better across nodes */
1508	vma->vm_pgoff = index + info->vfs_inode.i_ino;
1509	vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1510}
1511
1512static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1513{
1514	/* Drop reference taken by mpol_shared_policy_lookup() */
1515	mpol_cond_put(vma->vm_policy);
1516}
1517
1518static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1519			struct shmem_inode_info *info, pgoff_t index)
1520{
1521	struct vm_area_struct pvma;
1522	struct page *page;
1523	struct vm_fault vmf;
1524
1525	shmem_pseudo_vma_init(&pvma, info, index);
1526	vmf.vma = &pvma;
1527	vmf.address = 0;
1528	page = swap_cluster_readahead(swap, gfp, &vmf);
1529	shmem_pseudo_vma_destroy(&pvma);
1530
1531	return page;
1532}
1533
1534static struct page *shmem_alloc_hugepage(gfp_t gfp,
1535		struct shmem_inode_info *info, pgoff_t index)
1536{
1537	struct vm_area_struct pvma;
1538	struct address_space *mapping = info->vfs_inode.i_mapping;
1539	pgoff_t hindex;
1540	struct page *page;
1541
 
 
 
1542	hindex = round_down(index, HPAGE_PMD_NR);
1543	if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1544								XA_PRESENT))
1545		return NULL;
1546
1547	shmem_pseudo_vma_init(&pvma, info, hindex);
1548	page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1549			HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1550	shmem_pseudo_vma_destroy(&pvma);
1551	if (page)
1552		prep_transhuge_page(page);
1553	else
1554		count_vm_event(THP_FILE_FALLBACK);
1555	return page;
1556}
1557
1558static struct page *shmem_alloc_page(gfp_t gfp,
1559			struct shmem_inode_info *info, pgoff_t index)
1560{
1561	struct vm_area_struct pvma;
1562	struct page *page;
1563
1564	shmem_pseudo_vma_init(&pvma, info, index);
1565	page = alloc_page_vma(gfp, &pvma, 0);
1566	shmem_pseudo_vma_destroy(&pvma);
1567
1568	return page;
1569}
1570
1571static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1572		struct inode *inode,
1573		pgoff_t index, bool huge)
1574{
1575	struct shmem_inode_info *info = SHMEM_I(inode);
1576	struct page *page;
1577	int nr;
1578	int err = -ENOSPC;
1579
1580	if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
1581		huge = false;
1582	nr = huge ? HPAGE_PMD_NR : 1;
1583
1584	if (!shmem_inode_acct_block(inode, nr))
1585		goto failed;
1586
1587	if (huge)
1588		page = shmem_alloc_hugepage(gfp, info, index);
1589	else
1590		page = shmem_alloc_page(gfp, info, index);
1591	if (page) {
1592		__SetPageLocked(page);
1593		__SetPageSwapBacked(page);
1594		return page;
1595	}
1596
1597	err = -ENOMEM;
1598	shmem_inode_unacct_blocks(inode, nr);
1599failed:
1600	return ERR_PTR(err);
1601}
1602
1603/*
1604 * When a page is moved from swapcache to shmem filecache (either by the
1605 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1606 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1607 * ignorance of the mapping it belongs to.  If that mapping has special
1608 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1609 * we may need to copy to a suitable page before moving to filecache.
1610 *
1611 * In a future release, this may well be extended to respect cpuset and
1612 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1613 * but for now it is a simple matter of zone.
1614 */
1615static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1616{
1617	return page_zonenum(page) > gfp_zone(gfp);
1618}
1619
1620static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1621				struct shmem_inode_info *info, pgoff_t index)
1622{
1623	struct page *oldpage, *newpage;
1624	struct address_space *swap_mapping;
1625	swp_entry_t entry;
1626	pgoff_t swap_index;
1627	int error;
1628
1629	oldpage = *pagep;
1630	entry.val = page_private(oldpage);
1631	swap_index = swp_offset(entry);
1632	swap_mapping = page_mapping(oldpage);
1633
1634	/*
1635	 * We have arrived here because our zones are constrained, so don't
1636	 * limit chance of success by further cpuset and node constraints.
1637	 */
1638	gfp &= ~GFP_CONSTRAINT_MASK;
1639	newpage = shmem_alloc_page(gfp, info, index);
1640	if (!newpage)
1641		return -ENOMEM;
1642
1643	get_page(newpage);
1644	copy_highpage(newpage, oldpage);
1645	flush_dcache_page(newpage);
1646
1647	__SetPageLocked(newpage);
1648	__SetPageSwapBacked(newpage);
1649	SetPageUptodate(newpage);
1650	set_page_private(newpage, entry.val);
1651	SetPageSwapCache(newpage);
1652
1653	/*
1654	 * Our caller will very soon move newpage out of swapcache, but it's
1655	 * a nice clean interface for us to replace oldpage by newpage there.
1656	 */
1657	xa_lock_irq(&swap_mapping->i_pages);
1658	error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage);
1659	if (!error) {
1660		mem_cgroup_migrate(oldpage, newpage);
1661		__inc_lruvec_page_state(newpage, NR_FILE_PAGES);
1662		__dec_lruvec_page_state(oldpage, NR_FILE_PAGES);
1663	}
1664	xa_unlock_irq(&swap_mapping->i_pages);
1665
1666	if (unlikely(error)) {
1667		/*
1668		 * Is this possible?  I think not, now that our callers check
1669		 * both PageSwapCache and page_private after getting page lock;
1670		 * but be defensive.  Reverse old to newpage for clear and free.
1671		 */
1672		oldpage = newpage;
1673	} else {
1674		lru_cache_add(newpage);
 
1675		*pagep = newpage;
1676	}
1677
1678	ClearPageSwapCache(oldpage);
1679	set_page_private(oldpage, 0);
1680
1681	unlock_page(oldpage);
1682	put_page(oldpage);
1683	put_page(oldpage);
1684	return error;
1685}
1686
1687/*
1688 * Swap in the page pointed to by *pagep.
1689 * Caller has to make sure that *pagep contains a valid swapped page.
1690 * Returns 0 and the page in pagep if success. On failure, returns the
1691 * error code and NULL in *pagep.
1692 */
1693static int shmem_swapin_page(struct inode *inode, pgoff_t index,
1694			     struct page **pagep, enum sgp_type sgp,
1695			     gfp_t gfp, struct vm_area_struct *vma,
1696			     vm_fault_t *fault_type)
1697{
1698	struct address_space *mapping = inode->i_mapping;
1699	struct shmem_inode_info *info = SHMEM_I(inode);
1700	struct mm_struct *charge_mm = vma ? vma->vm_mm : current->mm;
 
1701	struct page *page;
1702	swp_entry_t swap;
1703	int error;
1704
1705	VM_BUG_ON(!*pagep || !xa_is_value(*pagep));
1706	swap = radix_to_swp_entry(*pagep);
1707	*pagep = NULL;
1708
1709	/* Look it up and read it in.. */
1710	page = lookup_swap_cache(swap, NULL, 0);
1711	if (!page) {
1712		/* Or update major stats only when swapin succeeds?? */
1713		if (fault_type) {
1714			*fault_type |= VM_FAULT_MAJOR;
1715			count_vm_event(PGMAJFAULT);
1716			count_memcg_event_mm(charge_mm, PGMAJFAULT);
1717		}
1718		/* Here we actually start the io */
1719		page = shmem_swapin(swap, gfp, info, index);
1720		if (!page) {
1721			error = -ENOMEM;
1722			goto failed;
1723		}
1724	}
1725
1726	/* We have to do this with page locked to prevent races */
1727	lock_page(page);
1728	if (!PageSwapCache(page) || page_private(page) != swap.val ||
1729	    !shmem_confirm_swap(mapping, index, swap)) {
1730		error = -EEXIST;
1731		goto unlock;
1732	}
1733	if (!PageUptodate(page)) {
1734		error = -EIO;
1735		goto failed;
1736	}
1737	wait_on_page_writeback(page);
1738
1739	if (shmem_should_replace_page(page, gfp)) {
1740		error = shmem_replace_page(&page, gfp, info, index);
1741		if (error)
1742			goto failed;
1743	}
1744
1745	error = shmem_add_to_page_cache(page, mapping, index,
1746					swp_to_radix_entry(swap), gfp,
1747					charge_mm);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1748	if (error)
1749		goto failed;
1750
 
 
1751	spin_lock_irq(&info->lock);
1752	info->swapped--;
1753	shmem_recalc_inode(inode);
1754	spin_unlock_irq(&info->lock);
1755
1756	if (sgp == SGP_WRITE)
1757		mark_page_accessed(page);
1758
1759	delete_from_swap_cache(page);
1760	set_page_dirty(page);
1761	swap_free(swap);
1762
1763	*pagep = page;
1764	return 0;
1765failed:
1766	if (!shmem_confirm_swap(mapping, index, swap))
1767		error = -EEXIST;
1768unlock:
1769	if (page) {
1770		unlock_page(page);
1771		put_page(page);
1772	}
1773
1774	return error;
1775}
1776
1777/*
1778 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1779 *
1780 * If we allocate a new one we do not mark it dirty. That's up to the
1781 * vm. If we swap it in we mark it dirty since we also free the swap
1782 * entry since a page cannot live in both the swap and page cache.
1783 *
1784 * vmf and fault_type are only supplied by shmem_fault:
1785 * otherwise they are NULL.
1786 */
1787static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1788	struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1789	struct vm_area_struct *vma, struct vm_fault *vmf,
1790			vm_fault_t *fault_type)
1791{
1792	struct address_space *mapping = inode->i_mapping;
1793	struct shmem_inode_info *info = SHMEM_I(inode);
1794	struct shmem_sb_info *sbinfo;
1795	struct mm_struct *charge_mm;
 
1796	struct page *page;
1797	enum sgp_type sgp_huge = sgp;
1798	pgoff_t hindex = index;
1799	int error;
1800	int once = 0;
1801	int alloced = 0;
1802
1803	if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1804		return -EFBIG;
1805	if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1806		sgp = SGP_CACHE;
1807repeat:
1808	if (sgp <= SGP_CACHE &&
1809	    ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1810		return -EINVAL;
1811	}
1812
1813	sbinfo = SHMEM_SB(inode->i_sb);
1814	charge_mm = vma ? vma->vm_mm : current->mm;
1815
1816	page = find_lock_entry(mapping, index);
1817	if (xa_is_value(page)) {
1818		error = shmem_swapin_page(inode, index, &page,
1819					  sgp, gfp, vma, fault_type);
1820		if (error == -EEXIST)
1821			goto repeat;
1822
1823		*pagep = page;
1824		return error;
1825	}
1826
1827	if (page && sgp == SGP_WRITE)
1828		mark_page_accessed(page);
1829
1830	/* fallocated page? */
1831	if (page && !PageUptodate(page)) {
1832		if (sgp != SGP_READ)
1833			goto clear;
1834		unlock_page(page);
1835		put_page(page);
1836		page = NULL;
1837	}
1838	if (page || sgp == SGP_READ) {
1839		*pagep = page;
1840		return 0;
1841	}
1842
1843	/*
1844	 * Fast cache lookup did not find it:
1845	 * bring it back from swap or allocate.
1846	 */
1847
1848	if (vma && userfaultfd_missing(vma)) {
1849		*fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1850		return 0;
1851	}
1852
1853	/* shmem_symlink() */
1854	if (mapping->a_ops != &shmem_aops)
1855		goto alloc_nohuge;
1856	if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1857		goto alloc_nohuge;
1858	if (shmem_huge == SHMEM_HUGE_FORCE)
1859		goto alloc_huge;
1860	switch (sbinfo->huge) {
1861	case SHMEM_HUGE_NEVER:
1862		goto alloc_nohuge;
1863	case SHMEM_HUGE_WITHIN_SIZE: {
1864		loff_t i_size;
1865		pgoff_t off;
1866
 
 
1867		off = round_up(index, HPAGE_PMD_NR);
1868		i_size = round_up(i_size_read(inode), PAGE_SIZE);
1869		if (i_size >= HPAGE_PMD_SIZE &&
1870		    i_size >> PAGE_SHIFT >= off)
1871			goto alloc_huge;
1872
1873		fallthrough;
1874	}
1875	case SHMEM_HUGE_ADVISE:
1876		if (sgp_huge == SGP_HUGE)
1877			goto alloc_huge;
1878		/* TODO: implement fadvise() hints */
1879		goto alloc_nohuge;
1880	}
1881
1882alloc_huge:
1883	page = shmem_alloc_and_acct_page(gfp, inode, index, true);
1884	if (IS_ERR(page)) {
1885alloc_nohuge:
1886		page = shmem_alloc_and_acct_page(gfp, inode,
1887						 index, false);
1888	}
1889	if (IS_ERR(page)) {
1890		int retry = 5;
1891
1892		error = PTR_ERR(page);
1893		page = NULL;
1894		if (error != -ENOSPC)
1895			goto unlock;
1896		/*
1897		 * Try to reclaim some space by splitting a huge page
1898		 * beyond i_size on the filesystem.
1899		 */
1900		while (retry--) {
1901			int ret;
1902
1903			ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1904			if (ret == SHRINK_STOP)
1905				break;
1906			if (ret)
1907				goto alloc_nohuge;
1908		}
1909		goto unlock;
1910	}
1911
1912	if (PageTransHuge(page))
1913		hindex = round_down(index, HPAGE_PMD_NR);
1914	else
1915		hindex = index;
1916
1917	if (sgp == SGP_WRITE)
1918		__SetPageReferenced(page);
1919
1920	error = shmem_add_to_page_cache(page, mapping, hindex,
1921					NULL, gfp & GFP_RECLAIM_MASK,
1922					charge_mm);
1923	if (error)
1924		goto unacct;
1925	lru_cache_add(page);
 
 
 
 
 
 
 
 
 
1926
1927	spin_lock_irq(&info->lock);
1928	info->alloced += compound_nr(page);
1929	inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1930	shmem_recalc_inode(inode);
1931	spin_unlock_irq(&info->lock);
1932	alloced = true;
1933
1934	if (PageTransHuge(page) &&
1935	    DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1936			hindex + HPAGE_PMD_NR - 1) {
1937		/*
1938		 * Part of the huge page is beyond i_size: subject
1939		 * to shrink under memory pressure.
1940		 */
1941		spin_lock(&sbinfo->shrinklist_lock);
1942		/*
1943		 * _careful to defend against unlocked access to
1944		 * ->shrink_list in shmem_unused_huge_shrink()
1945		 */
1946		if (list_empty_careful(&info->shrinklist)) {
1947			list_add_tail(&info->shrinklist,
1948				      &sbinfo->shrinklist);
1949			sbinfo->shrinklist_len++;
1950		}
1951		spin_unlock(&sbinfo->shrinklist_lock);
1952	}
1953
1954	/*
1955	 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1956	 */
1957	if (sgp == SGP_FALLOC)
1958		sgp = SGP_WRITE;
1959clear:
1960	/*
1961	 * Let SGP_WRITE caller clear ends if write does not fill page;
1962	 * but SGP_FALLOC on a page fallocated earlier must initialize
1963	 * it now, lest undo on failure cancel our earlier guarantee.
1964	 */
1965	if (sgp != SGP_WRITE && !PageUptodate(page)) {
1966		struct page *head = compound_head(page);
1967		int i;
1968
1969		for (i = 0; i < compound_nr(head); i++) {
1970			clear_highpage(head + i);
1971			flush_dcache_page(head + i);
1972		}
1973		SetPageUptodate(head);
1974	}
1975
1976	/* Perhaps the file has been truncated since we checked */
1977	if (sgp <= SGP_CACHE &&
1978	    ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1979		if (alloced) {
1980			ClearPageDirty(page);
1981			delete_from_page_cache(page);
1982			spin_lock_irq(&info->lock);
1983			shmem_recalc_inode(inode);
1984			spin_unlock_irq(&info->lock);
1985		}
1986		error = -EINVAL;
1987		goto unlock;
1988	}
1989	*pagep = page + index - hindex;
1990	return 0;
1991
1992	/*
1993	 * Error recovery.
1994	 */
1995unacct:
1996	shmem_inode_unacct_blocks(inode, compound_nr(page));
1997
1998	if (PageTransHuge(page)) {
1999		unlock_page(page);
2000		put_page(page);
2001		goto alloc_nohuge;
2002	}
2003unlock:
2004	if (page) {
2005		unlock_page(page);
2006		put_page(page);
2007	}
2008	if (error == -ENOSPC && !once++) {
2009		spin_lock_irq(&info->lock);
2010		shmem_recalc_inode(inode);
2011		spin_unlock_irq(&info->lock);
2012		goto repeat;
2013	}
2014	if (error == -EEXIST)
2015		goto repeat;
2016	return error;
2017}
2018
2019/*
2020 * This is like autoremove_wake_function, but it removes the wait queue
2021 * entry unconditionally - even if something else had already woken the
2022 * target.
2023 */
2024static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
2025{
2026	int ret = default_wake_function(wait, mode, sync, key);
2027	list_del_init(&wait->entry);
2028	return ret;
2029}
2030
2031static vm_fault_t shmem_fault(struct vm_fault *vmf)
2032{
2033	struct vm_area_struct *vma = vmf->vma;
2034	struct inode *inode = file_inode(vma->vm_file);
2035	gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
2036	enum sgp_type sgp;
2037	int err;
2038	vm_fault_t ret = VM_FAULT_LOCKED;
2039
2040	/*
2041	 * Trinity finds that probing a hole which tmpfs is punching can
2042	 * prevent the hole-punch from ever completing: which in turn
2043	 * locks writers out with its hold on i_mutex.  So refrain from
2044	 * faulting pages into the hole while it's being punched.  Although
2045	 * shmem_undo_range() does remove the additions, it may be unable to
2046	 * keep up, as each new page needs its own unmap_mapping_range() call,
2047	 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2048	 *
2049	 * It does not matter if we sometimes reach this check just before the
2050	 * hole-punch begins, so that one fault then races with the punch:
2051	 * we just need to make racing faults a rare case.
2052	 *
2053	 * The implementation below would be much simpler if we just used a
2054	 * standard mutex or completion: but we cannot take i_mutex in fault,
2055	 * and bloating every shmem inode for this unlikely case would be sad.
2056	 */
2057	if (unlikely(inode->i_private)) {
2058		struct shmem_falloc *shmem_falloc;
2059
2060		spin_lock(&inode->i_lock);
2061		shmem_falloc = inode->i_private;
2062		if (shmem_falloc &&
2063		    shmem_falloc->waitq &&
2064		    vmf->pgoff >= shmem_falloc->start &&
2065		    vmf->pgoff < shmem_falloc->next) {
2066			struct file *fpin;
2067			wait_queue_head_t *shmem_falloc_waitq;
2068			DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2069
2070			ret = VM_FAULT_NOPAGE;
2071			fpin = maybe_unlock_mmap_for_io(vmf, NULL);
2072			if (fpin)
 
 
2073				ret = VM_FAULT_RETRY;
 
2074
2075			shmem_falloc_waitq = shmem_falloc->waitq;
2076			prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2077					TASK_UNINTERRUPTIBLE);
2078			spin_unlock(&inode->i_lock);
2079			schedule();
2080
2081			/*
2082			 * shmem_falloc_waitq points into the shmem_fallocate()
2083			 * stack of the hole-punching task: shmem_falloc_waitq
2084			 * is usually invalid by the time we reach here, but
2085			 * finish_wait() does not dereference it in that case;
2086			 * though i_lock needed lest racing with wake_up_all().
2087			 */
2088			spin_lock(&inode->i_lock);
2089			finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2090			spin_unlock(&inode->i_lock);
2091
2092			if (fpin)
2093				fput(fpin);
2094			return ret;
2095		}
2096		spin_unlock(&inode->i_lock);
2097	}
2098
2099	sgp = SGP_CACHE;
2100
2101	if ((vma->vm_flags & VM_NOHUGEPAGE) ||
2102	    test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
2103		sgp = SGP_NOHUGE;
2104	else if (vma->vm_flags & VM_HUGEPAGE)
2105		sgp = SGP_HUGE;
2106
2107	err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
2108				  gfp, vma, vmf, &ret);
2109	if (err)
2110		return vmf_error(err);
2111	return ret;
2112}
2113
2114unsigned long shmem_get_unmapped_area(struct file *file,
2115				      unsigned long uaddr, unsigned long len,
2116				      unsigned long pgoff, unsigned long flags)
2117{
2118	unsigned long (*get_area)(struct file *,
2119		unsigned long, unsigned long, unsigned long, unsigned long);
2120	unsigned long addr;
2121	unsigned long offset;
2122	unsigned long inflated_len;
2123	unsigned long inflated_addr;
2124	unsigned long inflated_offset;
2125
2126	if (len > TASK_SIZE)
2127		return -ENOMEM;
2128
2129	get_area = current->mm->get_unmapped_area;
2130	addr = get_area(file, uaddr, len, pgoff, flags);
2131
2132	if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE))
2133		return addr;
2134	if (IS_ERR_VALUE(addr))
2135		return addr;
2136	if (addr & ~PAGE_MASK)
2137		return addr;
2138	if (addr > TASK_SIZE - len)
2139		return addr;
2140
2141	if (shmem_huge == SHMEM_HUGE_DENY)
2142		return addr;
2143	if (len < HPAGE_PMD_SIZE)
2144		return addr;
2145	if (flags & MAP_FIXED)
2146		return addr;
2147	/*
2148	 * Our priority is to support MAP_SHARED mapped hugely;
2149	 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2150	 * But if caller specified an address hint and we allocated area there
2151	 * successfully, respect that as before.
2152	 */
2153	if (uaddr == addr)
2154		return addr;
2155
2156	if (shmem_huge != SHMEM_HUGE_FORCE) {
2157		struct super_block *sb;
2158
2159		if (file) {
2160			VM_BUG_ON(file->f_op != &shmem_file_operations);
2161			sb = file_inode(file)->i_sb;
2162		} else {
2163			/*
2164			 * Called directly from mm/mmap.c, or drivers/char/mem.c
2165			 * for "/dev/zero", to create a shared anonymous object.
2166			 */
2167			if (IS_ERR(shm_mnt))
2168				return addr;
2169			sb = shm_mnt->mnt_sb;
2170		}
2171		if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2172			return addr;
2173	}
2174
2175	offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2176	if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2177		return addr;
2178	if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2179		return addr;
2180
2181	inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2182	if (inflated_len > TASK_SIZE)
2183		return addr;
2184	if (inflated_len < len)
2185		return addr;
2186
2187	inflated_addr = get_area(NULL, uaddr, inflated_len, 0, flags);
2188	if (IS_ERR_VALUE(inflated_addr))
2189		return addr;
2190	if (inflated_addr & ~PAGE_MASK)
2191		return addr;
2192
2193	inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2194	inflated_addr += offset - inflated_offset;
2195	if (inflated_offset > offset)
2196		inflated_addr += HPAGE_PMD_SIZE;
2197
2198	if (inflated_addr > TASK_SIZE - len)
2199		return addr;
2200	return inflated_addr;
2201}
2202
2203#ifdef CONFIG_NUMA
2204static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2205{
2206	struct inode *inode = file_inode(vma->vm_file);
2207	return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2208}
2209
2210static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2211					  unsigned long addr)
2212{
2213	struct inode *inode = file_inode(vma->vm_file);
2214	pgoff_t index;
2215
2216	index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2217	return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2218}
2219#endif
2220
2221int shmem_lock(struct file *file, int lock, struct user_struct *user)
2222{
2223	struct inode *inode = file_inode(file);
2224	struct shmem_inode_info *info = SHMEM_I(inode);
2225	int retval = -ENOMEM;
2226
2227	/*
2228	 * What serializes the accesses to info->flags?
2229	 * ipc_lock_object() when called from shmctl_do_lock(),
2230	 * no serialization needed when called from shm_destroy().
2231	 */
2232	if (lock && !(info->flags & VM_LOCKED)) {
2233		if (!user_shm_lock(inode->i_size, user))
2234			goto out_nomem;
2235		info->flags |= VM_LOCKED;
2236		mapping_set_unevictable(file->f_mapping);
2237	}
2238	if (!lock && (info->flags & VM_LOCKED) && user) {
2239		user_shm_unlock(inode->i_size, user);
2240		info->flags &= ~VM_LOCKED;
2241		mapping_clear_unevictable(file->f_mapping);
2242	}
2243	retval = 0;
2244
2245out_nomem:
 
2246	return retval;
2247}
2248
2249static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2250{
2251	struct shmem_inode_info *info = SHMEM_I(file_inode(file));
2252
2253	if (info->seals & F_SEAL_FUTURE_WRITE) {
2254		/*
2255		 * New PROT_WRITE and MAP_SHARED mmaps are not allowed when
2256		 * "future write" seal active.
2257		 */
2258		if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_WRITE))
2259			return -EPERM;
2260
2261		/*
2262		 * Since an F_SEAL_FUTURE_WRITE sealed memfd can be mapped as
2263		 * MAP_SHARED and read-only, take care to not allow mprotect to
2264		 * revert protections on such mappings. Do this only for shared
2265		 * mappings. For private mappings, don't need to mask
2266		 * VM_MAYWRITE as we still want them to be COW-writable.
2267		 */
2268		if (vma->vm_flags & VM_SHARED)
2269			vma->vm_flags &= ~(VM_MAYWRITE);
2270	}
2271
2272	file_accessed(file);
2273	vma->vm_ops = &shmem_vm_ops;
2274	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
2275			((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2276			(vma->vm_end & HPAGE_PMD_MASK)) {
2277		khugepaged_enter(vma, vma->vm_flags);
2278	}
2279	return 0;
2280}
2281
2282static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2283				     umode_t mode, dev_t dev, unsigned long flags)
2284{
2285	struct inode *inode;
2286	struct shmem_inode_info *info;
2287	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2288	ino_t ino;
2289
2290	if (shmem_reserve_inode(sb, &ino))
2291		return NULL;
2292
2293	inode = new_inode(sb);
2294	if (inode) {
2295		inode->i_ino = ino;
2296		inode_init_owner(inode, dir, mode);
2297		inode->i_blocks = 0;
2298		inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2299		inode->i_generation = prandom_u32();
2300		info = SHMEM_I(inode);
2301		memset(info, 0, (char *)inode - (char *)info);
2302		spin_lock_init(&info->lock);
2303		atomic_set(&info->stop_eviction, 0);
2304		info->seals = F_SEAL_SEAL;
2305		info->flags = flags & VM_NORESERVE;
2306		INIT_LIST_HEAD(&info->shrinklist);
2307		INIT_LIST_HEAD(&info->swaplist);
2308		simple_xattrs_init(&info->xattrs);
2309		cache_no_acl(inode);
2310
2311		switch (mode & S_IFMT) {
2312		default:
2313			inode->i_op = &shmem_special_inode_operations;
2314			init_special_inode(inode, mode, dev);
2315			break;
2316		case S_IFREG:
2317			inode->i_mapping->a_ops = &shmem_aops;
2318			inode->i_op = &shmem_inode_operations;
2319			inode->i_fop = &shmem_file_operations;
2320			mpol_shared_policy_init(&info->policy,
2321						 shmem_get_sbmpol(sbinfo));
2322			break;
2323		case S_IFDIR:
2324			inc_nlink(inode);
2325			/* Some things misbehave if size == 0 on a directory */
2326			inode->i_size = 2 * BOGO_DIRENT_SIZE;
2327			inode->i_op = &shmem_dir_inode_operations;
2328			inode->i_fop = &simple_dir_operations;
2329			break;
2330		case S_IFLNK:
2331			/*
2332			 * Must not load anything in the rbtree,
2333			 * mpol_free_shared_policy will not be called.
2334			 */
2335			mpol_shared_policy_init(&info->policy, NULL);
2336			break;
2337		}
2338
2339		lockdep_annotate_inode_mutex_key(inode);
2340	} else
2341		shmem_free_inode(sb);
2342	return inode;
2343}
2344
2345bool shmem_mapping(struct address_space *mapping)
2346{
2347	return mapping->a_ops == &shmem_aops;
2348}
2349
2350static int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2351				  pmd_t *dst_pmd,
2352				  struct vm_area_struct *dst_vma,
2353				  unsigned long dst_addr,
2354				  unsigned long src_addr,
2355				  bool zeropage,
2356				  struct page **pagep)
2357{
2358	struct inode *inode = file_inode(dst_vma->vm_file);
2359	struct shmem_inode_info *info = SHMEM_I(inode);
2360	struct address_space *mapping = inode->i_mapping;
2361	gfp_t gfp = mapping_gfp_mask(mapping);
2362	pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
 
2363	spinlock_t *ptl;
2364	void *page_kaddr;
2365	struct page *page;
2366	pte_t _dst_pte, *dst_pte;
2367	int ret;
2368	pgoff_t offset, max_off;
2369
2370	ret = -ENOMEM;
2371	if (!shmem_inode_acct_block(inode, 1))
2372		goto out;
2373
2374	if (!*pagep) {
2375		page = shmem_alloc_page(gfp, info, pgoff);
2376		if (!page)
2377			goto out_unacct_blocks;
2378
2379		if (!zeropage) {	/* mcopy_atomic */
2380			page_kaddr = kmap_atomic(page);
2381			ret = copy_from_user(page_kaddr,
2382					     (const void __user *)src_addr,
2383					     PAGE_SIZE);
2384			kunmap_atomic(page_kaddr);
2385
2386			/* fallback to copy_from_user outside mmap_lock */
2387			if (unlikely(ret)) {
2388				*pagep = page;
2389				shmem_inode_unacct_blocks(inode, 1);
2390				/* don't free the page */
2391				return -ENOENT;
2392			}
2393		} else {		/* mfill_zeropage_atomic */
2394			clear_highpage(page);
2395		}
2396	} else {
2397		page = *pagep;
2398		*pagep = NULL;
2399	}
2400
2401	VM_BUG_ON(PageLocked(page) || PageSwapBacked(page));
2402	__SetPageLocked(page);
2403	__SetPageSwapBacked(page);
2404	__SetPageUptodate(page);
2405
2406	ret = -EFAULT;
2407	offset = linear_page_index(dst_vma, dst_addr);
2408	max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2409	if (unlikely(offset >= max_off))
2410		goto out_release;
2411
2412	ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL,
2413				      gfp & GFP_RECLAIM_MASK, dst_mm);
2414	if (ret)
2415		goto out_release;
2416
 
 
 
 
 
 
 
2417	_dst_pte = mk_pte(page, dst_vma->vm_page_prot);
2418	if (dst_vma->vm_flags & VM_WRITE)
2419		_dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte));
2420	else {
2421		/*
2422		 * We don't set the pte dirty if the vma has no
2423		 * VM_WRITE permission, so mark the page dirty or it
2424		 * could be freed from under us. We could do it
2425		 * unconditionally before unlock_page(), but doing it
2426		 * only if VM_WRITE is not set is faster.
2427		 */
2428		set_page_dirty(page);
2429	}
2430
2431	dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);
2432
2433	ret = -EFAULT;
2434	max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2435	if (unlikely(offset >= max_off))
2436		goto out_release_unlock;
2437
2438	ret = -EEXIST;
2439	if (!pte_none(*dst_pte))
2440		goto out_release_unlock;
2441
2442	lru_cache_add(page);
2443
2444	spin_lock_irq(&info->lock);
2445	info->alloced++;
2446	inode->i_blocks += BLOCKS_PER_PAGE;
2447	shmem_recalc_inode(inode);
2448	spin_unlock_irq(&info->lock);
2449
2450	inc_mm_counter(dst_mm, mm_counter_file(page));
2451	page_add_file_rmap(page, false);
2452	set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
2453
2454	/* No need to invalidate - it was non-present before */
2455	update_mmu_cache(dst_vma, dst_addr, dst_pte);
2456	pte_unmap_unlock(dst_pte, ptl);
2457	unlock_page(page);
2458	ret = 0;
2459out:
2460	return ret;
2461out_release_unlock:
2462	pte_unmap_unlock(dst_pte, ptl);
2463	ClearPageDirty(page);
2464	delete_from_page_cache(page);
 
 
2465out_release:
2466	unlock_page(page);
2467	put_page(page);
2468out_unacct_blocks:
2469	shmem_inode_unacct_blocks(inode, 1);
2470	goto out;
2471}
2472
2473int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm,
2474			   pmd_t *dst_pmd,
2475			   struct vm_area_struct *dst_vma,
2476			   unsigned long dst_addr,
2477			   unsigned long src_addr,
2478			   struct page **pagep)
2479{
2480	return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2481				      dst_addr, src_addr, false, pagep);
2482}
2483
2484int shmem_mfill_zeropage_pte(struct mm_struct *dst_mm,
2485			     pmd_t *dst_pmd,
2486			     struct vm_area_struct *dst_vma,
2487			     unsigned long dst_addr)
2488{
2489	struct page *page = NULL;
2490
2491	return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2492				      dst_addr, 0, true, &page);
2493}
2494
2495#ifdef CONFIG_TMPFS
2496static const struct inode_operations shmem_symlink_inode_operations;
2497static const struct inode_operations shmem_short_symlink_operations;
2498
2499#ifdef CONFIG_TMPFS_XATTR
2500static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2501#else
2502#define shmem_initxattrs NULL
2503#endif
2504
2505static int
2506shmem_write_begin(struct file *file, struct address_space *mapping,
2507			loff_t pos, unsigned len, unsigned flags,
2508			struct page **pagep, void **fsdata)
2509{
2510	struct inode *inode = mapping->host;
2511	struct shmem_inode_info *info = SHMEM_I(inode);
2512	pgoff_t index = pos >> PAGE_SHIFT;
2513
2514	/* i_mutex is held by caller */
2515	if (unlikely(info->seals & (F_SEAL_GROW |
2516				   F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2517		if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
2518			return -EPERM;
2519		if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2520			return -EPERM;
2521	}
2522
2523	return shmem_getpage(inode, index, pagep, SGP_WRITE);
2524}
2525
2526static int
2527shmem_write_end(struct file *file, struct address_space *mapping,
2528			loff_t pos, unsigned len, unsigned copied,
2529			struct page *page, void *fsdata)
2530{
2531	struct inode *inode = mapping->host;
2532
2533	if (pos + copied > inode->i_size)
2534		i_size_write(inode, pos + copied);
2535
2536	if (!PageUptodate(page)) {
2537		struct page *head = compound_head(page);
2538		if (PageTransCompound(page)) {
2539			int i;
2540
2541			for (i = 0; i < HPAGE_PMD_NR; i++) {
2542				if (head + i == page)
2543					continue;
2544				clear_highpage(head + i);
2545				flush_dcache_page(head + i);
2546			}
2547		}
2548		if (copied < PAGE_SIZE) {
2549			unsigned from = pos & (PAGE_SIZE - 1);
2550			zero_user_segments(page, 0, from,
2551					from + copied, PAGE_SIZE);
2552		}
2553		SetPageUptodate(head);
2554	}
2555	set_page_dirty(page);
2556	unlock_page(page);
2557	put_page(page);
2558
2559	return copied;
2560}
2561
2562static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2563{
2564	struct file *file = iocb->ki_filp;
2565	struct inode *inode = file_inode(file);
2566	struct address_space *mapping = inode->i_mapping;
2567	pgoff_t index;
2568	unsigned long offset;
2569	enum sgp_type sgp = SGP_READ;
2570	int error = 0;
2571	ssize_t retval = 0;
2572	loff_t *ppos = &iocb->ki_pos;
2573
2574	/*
2575	 * Might this read be for a stacking filesystem?  Then when reading
2576	 * holes of a sparse file, we actually need to allocate those pages,
2577	 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2578	 */
2579	if (!iter_is_iovec(to))
2580		sgp = SGP_CACHE;
2581
2582	index = *ppos >> PAGE_SHIFT;
2583	offset = *ppos & ~PAGE_MASK;
2584
2585	for (;;) {
2586		struct page *page = NULL;
2587		pgoff_t end_index;
2588		unsigned long nr, ret;
2589		loff_t i_size = i_size_read(inode);
2590
2591		end_index = i_size >> PAGE_SHIFT;
2592		if (index > end_index)
2593			break;
2594		if (index == end_index) {
2595			nr = i_size & ~PAGE_MASK;
2596			if (nr <= offset)
2597				break;
2598		}
2599
2600		error = shmem_getpage(inode, index, &page, sgp);
2601		if (error) {
2602			if (error == -EINVAL)
2603				error = 0;
2604			break;
2605		}
2606		if (page) {
2607			if (sgp == SGP_CACHE)
2608				set_page_dirty(page);
2609			unlock_page(page);
2610		}
2611
2612		/*
2613		 * We must evaluate after, since reads (unlike writes)
2614		 * are called without i_mutex protection against truncate
2615		 */
2616		nr = PAGE_SIZE;
2617		i_size = i_size_read(inode);
2618		end_index = i_size >> PAGE_SHIFT;
2619		if (index == end_index) {
2620			nr = i_size & ~PAGE_MASK;
2621			if (nr <= offset) {
2622				if (page)
2623					put_page(page);
2624				break;
2625			}
2626		}
2627		nr -= offset;
2628
2629		if (page) {
2630			/*
2631			 * If users can be writing to this page using arbitrary
2632			 * virtual addresses, take care about potential aliasing
2633			 * before reading the page on the kernel side.
2634			 */
2635			if (mapping_writably_mapped(mapping))
2636				flush_dcache_page(page);
2637			/*
2638			 * Mark the page accessed if we read the beginning.
2639			 */
2640			if (!offset)
2641				mark_page_accessed(page);
2642		} else {
2643			page = ZERO_PAGE(0);
2644			get_page(page);
2645		}
2646
2647		/*
2648		 * Ok, we have the page, and it's up-to-date, so
2649		 * now we can copy it to user space...
2650		 */
2651		ret = copy_page_to_iter(page, offset, nr, to);
2652		retval += ret;
2653		offset += ret;
2654		index += offset >> PAGE_SHIFT;
2655		offset &= ~PAGE_MASK;
2656
2657		put_page(page);
2658		if (!iov_iter_count(to))
2659			break;
2660		if (ret < nr) {
2661			error = -EFAULT;
2662			break;
2663		}
2664		cond_resched();
2665	}
2666
2667	*ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2668	file_accessed(file);
2669	return retval ? retval : error;
2670}
2671
2672/*
2673 * llseek SEEK_DATA or SEEK_HOLE through the page cache.
2674 */
2675static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
2676				    pgoff_t index, pgoff_t end, int whence)
2677{
2678	struct page *page;
2679	struct pagevec pvec;
2680	pgoff_t indices[PAGEVEC_SIZE];
2681	bool done = false;
2682	int i;
2683
2684	pagevec_init(&pvec);
2685	pvec.nr = 1;		/* start small: we may be there already */
2686	while (!done) {
2687		pvec.nr = find_get_entries(mapping, index,
2688					pvec.nr, pvec.pages, indices);
2689		if (!pvec.nr) {
2690			if (whence == SEEK_DATA)
2691				index = end;
2692			break;
2693		}
2694		for (i = 0; i < pvec.nr; i++, index++) {
2695			if (index < indices[i]) {
2696				if (whence == SEEK_HOLE) {
2697					done = true;
2698					break;
2699				}
2700				index = indices[i];
2701			}
2702			page = pvec.pages[i];
2703			if (page && !xa_is_value(page)) {
2704				if (!PageUptodate(page))
2705					page = NULL;
2706			}
2707			if (index >= end ||
2708			    (page && whence == SEEK_DATA) ||
2709			    (!page && whence == SEEK_HOLE)) {
2710				done = true;
2711				break;
2712			}
2713		}
2714		pagevec_remove_exceptionals(&pvec);
2715		pagevec_release(&pvec);
2716		pvec.nr = PAGEVEC_SIZE;
2717		cond_resched();
2718	}
2719	return index;
2720}
2721
2722static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2723{
2724	struct address_space *mapping = file->f_mapping;
2725	struct inode *inode = mapping->host;
2726	pgoff_t start, end;
2727	loff_t new_offset;
2728
2729	if (whence != SEEK_DATA && whence != SEEK_HOLE)
2730		return generic_file_llseek_size(file, offset, whence,
2731					MAX_LFS_FILESIZE, i_size_read(inode));
2732	inode_lock(inode);
2733	/* We're holding i_mutex so we can access i_size directly */
2734
2735	if (offset < 0 || offset >= inode->i_size)
2736		offset = -ENXIO;
2737	else {
2738		start = offset >> PAGE_SHIFT;
2739		end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2740		new_offset = shmem_seek_hole_data(mapping, start, end, whence);
2741		new_offset <<= PAGE_SHIFT;
2742		if (new_offset > offset) {
2743			if (new_offset < inode->i_size)
2744				offset = new_offset;
2745			else if (whence == SEEK_DATA)
2746				offset = -ENXIO;
2747			else
2748				offset = inode->i_size;
2749		}
2750	}
2751
2752	if (offset >= 0)
2753		offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2754	inode_unlock(inode);
2755	return offset;
2756}
2757
2758static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2759							 loff_t len)
2760{
2761	struct inode *inode = file_inode(file);
2762	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2763	struct shmem_inode_info *info = SHMEM_I(inode);
2764	struct shmem_falloc shmem_falloc;
2765	pgoff_t start, index, end;
2766	int error;
2767
2768	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2769		return -EOPNOTSUPP;
2770
2771	inode_lock(inode);
2772
2773	if (mode & FALLOC_FL_PUNCH_HOLE) {
2774		struct address_space *mapping = file->f_mapping;
2775		loff_t unmap_start = round_up(offset, PAGE_SIZE);
2776		loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2777		DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2778
2779		/* protected by i_mutex */
2780		if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
2781			error = -EPERM;
2782			goto out;
2783		}
2784
2785		shmem_falloc.waitq = &shmem_falloc_waitq;
2786		shmem_falloc.start = (u64)unmap_start >> PAGE_SHIFT;
2787		shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2788		spin_lock(&inode->i_lock);
2789		inode->i_private = &shmem_falloc;
2790		spin_unlock(&inode->i_lock);
2791
2792		if ((u64)unmap_end > (u64)unmap_start)
2793			unmap_mapping_range(mapping, unmap_start,
2794					    1 + unmap_end - unmap_start, 0);
2795		shmem_truncate_range(inode, offset, offset + len - 1);
2796		/* No need to unmap again: hole-punching leaves COWed pages */
2797
2798		spin_lock(&inode->i_lock);
2799		inode->i_private = NULL;
2800		wake_up_all(&shmem_falloc_waitq);
2801		WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
2802		spin_unlock(&inode->i_lock);
2803		error = 0;
2804		goto out;
2805	}
2806
2807	/* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2808	error = inode_newsize_ok(inode, offset + len);
2809	if (error)
2810		goto out;
2811
2812	if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2813		error = -EPERM;
2814		goto out;
2815	}
2816
2817	start = offset >> PAGE_SHIFT;
2818	end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2819	/* Try to avoid a swapstorm if len is impossible to satisfy */
2820	if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2821		error = -ENOSPC;
2822		goto out;
2823	}
2824
2825	shmem_falloc.waitq = NULL;
2826	shmem_falloc.start = start;
2827	shmem_falloc.next  = start;
2828	shmem_falloc.nr_falloced = 0;
2829	shmem_falloc.nr_unswapped = 0;
2830	spin_lock(&inode->i_lock);
2831	inode->i_private = &shmem_falloc;
2832	spin_unlock(&inode->i_lock);
2833
2834	for (index = start; index < end; index++) {
2835		struct page *page;
2836
2837		/*
2838		 * Good, the fallocate(2) manpage permits EINTR: we may have
2839		 * been interrupted because we are using up too much memory.
2840		 */
2841		if (signal_pending(current))
2842			error = -EINTR;
2843		else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2844			error = -ENOMEM;
2845		else
2846			error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2847		if (error) {
2848			/* Remove the !PageUptodate pages we added */
2849			if (index > start) {
2850				shmem_undo_range(inode,
2851				    (loff_t)start << PAGE_SHIFT,
2852				    ((loff_t)index << PAGE_SHIFT) - 1, true);
2853			}
2854			goto undone;
2855		}
2856
2857		/*
2858		 * Inform shmem_writepage() how far we have reached.
2859		 * No need for lock or barrier: we have the page lock.
2860		 */
2861		shmem_falloc.next++;
2862		if (!PageUptodate(page))
2863			shmem_falloc.nr_falloced++;
2864
2865		/*
2866		 * If !PageUptodate, leave it that way so that freeable pages
2867		 * can be recognized if we need to rollback on error later.
2868		 * But set_page_dirty so that memory pressure will swap rather
2869		 * than free the pages we are allocating (and SGP_CACHE pages
2870		 * might still be clean: we now need to mark those dirty too).
2871		 */
2872		set_page_dirty(page);
2873		unlock_page(page);
2874		put_page(page);
2875		cond_resched();
2876	}
2877
2878	if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2879		i_size_write(inode, offset + len);
2880	inode->i_ctime = current_time(inode);
2881undone:
2882	spin_lock(&inode->i_lock);
2883	inode->i_private = NULL;
2884	spin_unlock(&inode->i_lock);
2885out:
2886	inode_unlock(inode);
2887	return error;
2888}
2889
2890static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2891{
2892	struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2893
2894	buf->f_type = TMPFS_MAGIC;
2895	buf->f_bsize = PAGE_SIZE;
2896	buf->f_namelen = NAME_MAX;
2897	if (sbinfo->max_blocks) {
2898		buf->f_blocks = sbinfo->max_blocks;
2899		buf->f_bavail =
2900		buf->f_bfree  = sbinfo->max_blocks -
2901				percpu_counter_sum(&sbinfo->used_blocks);
2902	}
2903	if (sbinfo->max_inodes) {
2904		buf->f_files = sbinfo->max_inodes;
2905		buf->f_ffree = sbinfo->free_inodes;
2906	}
2907	/* else leave those fields 0 like simple_statfs */
2908	return 0;
2909}
2910
2911/*
2912 * File creation. Allocate an inode, and we're done..
2913 */
2914static int
2915shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2916{
2917	struct inode *inode;
2918	int error = -ENOSPC;
2919
2920	inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2921	if (inode) {
2922		error = simple_acl_create(dir, inode);
2923		if (error)
2924			goto out_iput;
2925		error = security_inode_init_security(inode, dir,
2926						     &dentry->d_name,
2927						     shmem_initxattrs, NULL);
2928		if (error && error != -EOPNOTSUPP)
2929			goto out_iput;
2930
2931		error = 0;
2932		dir->i_size += BOGO_DIRENT_SIZE;
2933		dir->i_ctime = dir->i_mtime = current_time(dir);
2934		d_instantiate(dentry, inode);
2935		dget(dentry); /* Extra count - pin the dentry in core */
2936	}
2937	return error;
2938out_iput:
2939	iput(inode);
2940	return error;
2941}
2942
2943static int
2944shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2945{
2946	struct inode *inode;
2947	int error = -ENOSPC;
2948
2949	inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2950	if (inode) {
2951		error = security_inode_init_security(inode, dir,
2952						     NULL,
2953						     shmem_initxattrs, NULL);
2954		if (error && error != -EOPNOTSUPP)
2955			goto out_iput;
2956		error = simple_acl_create(dir, inode);
2957		if (error)
2958			goto out_iput;
2959		d_tmpfile(dentry, inode);
2960	}
2961	return error;
2962out_iput:
2963	iput(inode);
2964	return error;
2965}
2966
2967static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2968{
2969	int error;
2970
2971	if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2972		return error;
2973	inc_nlink(dir);
2974	return 0;
2975}
2976
2977static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2978		bool excl)
2979{
2980	return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2981}
2982
2983/*
2984 * Link a file..
2985 */
2986static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2987{
2988	struct inode *inode = d_inode(old_dentry);
2989	int ret = 0;
2990
2991	/*
2992	 * No ordinary (disk based) filesystem counts links as inodes;
2993	 * but each new link needs a new dentry, pinning lowmem, and
2994	 * tmpfs dentries cannot be pruned until they are unlinked.
2995	 * But if an O_TMPFILE file is linked into the tmpfs, the
2996	 * first link must skip that, to get the accounting right.
2997	 */
2998	if (inode->i_nlink) {
2999		ret = shmem_reserve_inode(inode->i_sb, NULL);
3000		if (ret)
3001			goto out;
3002	}
3003
3004	dir->i_size += BOGO_DIRENT_SIZE;
3005	inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3006	inc_nlink(inode);
3007	ihold(inode);	/* New dentry reference */
3008	dget(dentry);		/* Extra pinning count for the created dentry */
3009	d_instantiate(dentry, inode);
3010out:
3011	return ret;
3012}
3013
3014static int shmem_unlink(struct inode *dir, struct dentry *dentry)
3015{
3016	struct inode *inode = d_inode(dentry);
3017
3018	if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
3019		shmem_free_inode(inode->i_sb);
3020
3021	dir->i_size -= BOGO_DIRENT_SIZE;
3022	inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
3023	drop_nlink(inode);
3024	dput(dentry);	/* Undo the count from "create" - this does all the work */
3025	return 0;
3026}
3027
3028static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3029{
3030	if (!simple_empty(dentry))
3031		return -ENOTEMPTY;
3032
3033	drop_nlink(d_inode(dentry));
3034	drop_nlink(dir);
3035	return shmem_unlink(dir, dentry);
3036}
3037
3038static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
3039{
3040	bool old_is_dir = d_is_dir(old_dentry);
3041	bool new_is_dir = d_is_dir(new_dentry);
3042
3043	if (old_dir != new_dir && old_is_dir != new_is_dir) {
3044		if (old_is_dir) {
3045			drop_nlink(old_dir);
3046			inc_nlink(new_dir);
3047		} else {
3048			drop_nlink(new_dir);
3049			inc_nlink(old_dir);
3050		}
3051	}
3052	old_dir->i_ctime = old_dir->i_mtime =
3053	new_dir->i_ctime = new_dir->i_mtime =
3054	d_inode(old_dentry)->i_ctime =
3055	d_inode(new_dentry)->i_ctime = current_time(old_dir);
3056
3057	return 0;
3058}
3059
3060static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
3061{
3062	struct dentry *whiteout;
3063	int error;
3064
3065	whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3066	if (!whiteout)
3067		return -ENOMEM;
3068
3069	error = shmem_mknod(old_dir, whiteout,
3070			    S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3071	dput(whiteout);
3072	if (error)
3073		return error;
3074
3075	/*
3076	 * Cheat and hash the whiteout while the old dentry is still in
3077	 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3078	 *
3079	 * d_lookup() will consistently find one of them at this point,
3080	 * not sure which one, but that isn't even important.
3081	 */
3082	d_rehash(whiteout);
3083	return 0;
3084}
3085
3086/*
3087 * The VFS layer already does all the dentry stuff for rename,
3088 * we just have to decrement the usage count for the target if
3089 * it exists so that the VFS layer correctly free's it when it
3090 * gets overwritten.
3091 */
3092static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags)
3093{
3094	struct inode *inode = d_inode(old_dentry);
3095	int they_are_dirs = S_ISDIR(inode->i_mode);
3096
3097	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3098		return -EINVAL;
3099
3100	if (flags & RENAME_EXCHANGE)
3101		return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3102
3103	if (!simple_empty(new_dentry))
3104		return -ENOTEMPTY;
3105
3106	if (flags & RENAME_WHITEOUT) {
3107		int error;
3108
3109		error = shmem_whiteout(old_dir, old_dentry);
3110		if (error)
3111			return error;
3112	}
3113
3114	if (d_really_is_positive(new_dentry)) {
3115		(void) shmem_unlink(new_dir, new_dentry);
3116		if (they_are_dirs) {
3117			drop_nlink(d_inode(new_dentry));
3118			drop_nlink(old_dir);
3119		}
3120	} else if (they_are_dirs) {
3121		drop_nlink(old_dir);
3122		inc_nlink(new_dir);
3123	}
3124
3125	old_dir->i_size -= BOGO_DIRENT_SIZE;
3126	new_dir->i_size += BOGO_DIRENT_SIZE;
3127	old_dir->i_ctime = old_dir->i_mtime =
3128	new_dir->i_ctime = new_dir->i_mtime =
3129	inode->i_ctime = current_time(old_dir);
3130	return 0;
3131}
3132
3133static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3134{
3135	int error;
3136	int len;
3137	struct inode *inode;
3138	struct page *page;
3139
3140	len = strlen(symname) + 1;
3141	if (len > PAGE_SIZE)
3142		return -ENAMETOOLONG;
3143
3144	inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0,
3145				VM_NORESERVE);
3146	if (!inode)
3147		return -ENOSPC;
3148
3149	error = security_inode_init_security(inode, dir, &dentry->d_name,
3150					     shmem_initxattrs, NULL);
3151	if (error && error != -EOPNOTSUPP) {
3152		iput(inode);
3153		return error;
 
 
 
3154	}
3155
3156	inode->i_size = len-1;
3157	if (len <= SHORT_SYMLINK_LEN) {
3158		inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3159		if (!inode->i_link) {
3160			iput(inode);
3161			return -ENOMEM;
3162		}
3163		inode->i_op = &shmem_short_symlink_operations;
3164	} else {
3165		inode_nohighmem(inode);
3166		error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3167		if (error) {
3168			iput(inode);
3169			return error;
3170		}
3171		inode->i_mapping->a_ops = &shmem_aops;
3172		inode->i_op = &shmem_symlink_inode_operations;
3173		memcpy(page_address(page), symname, len);
3174		SetPageUptodate(page);
3175		set_page_dirty(page);
3176		unlock_page(page);
3177		put_page(page);
3178	}
3179	dir->i_size += BOGO_DIRENT_SIZE;
3180	dir->i_ctime = dir->i_mtime = current_time(dir);
3181	d_instantiate(dentry, inode);
3182	dget(dentry);
3183	return 0;
3184}
3185
3186static void shmem_put_link(void *arg)
3187{
3188	mark_page_accessed(arg);
3189	put_page(arg);
3190}
3191
3192static const char *shmem_get_link(struct dentry *dentry,
3193				  struct inode *inode,
3194				  struct delayed_call *done)
3195{
3196	struct page *page = NULL;
3197	int error;
3198	if (!dentry) {
3199		page = find_get_page(inode->i_mapping, 0);
3200		if (!page)
3201			return ERR_PTR(-ECHILD);
3202		if (!PageUptodate(page)) {
3203			put_page(page);
3204			return ERR_PTR(-ECHILD);
3205		}
3206	} else {
3207		error = shmem_getpage(inode, 0, &page, SGP_READ);
3208		if (error)
3209			return ERR_PTR(error);
3210		unlock_page(page);
3211	}
3212	set_delayed_call(done, shmem_put_link, page);
3213	return page_address(page);
3214}
3215
3216#ifdef CONFIG_TMPFS_XATTR
3217/*
3218 * Superblocks without xattr inode operations may get some security.* xattr
3219 * support from the LSM "for free". As soon as we have any other xattrs
3220 * like ACLs, we also need to implement the security.* handlers at
3221 * filesystem level, though.
3222 */
3223
3224/*
3225 * Callback for security_inode_init_security() for acquiring xattrs.
3226 */
3227static int shmem_initxattrs(struct inode *inode,
3228			    const struct xattr *xattr_array,
3229			    void *fs_info)
3230{
3231	struct shmem_inode_info *info = SHMEM_I(inode);
3232	const struct xattr *xattr;
3233	struct simple_xattr *new_xattr;
3234	size_t len;
3235
3236	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3237		new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3238		if (!new_xattr)
3239			return -ENOMEM;
3240
3241		len = strlen(xattr->name) + 1;
3242		new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3243					  GFP_KERNEL);
3244		if (!new_xattr->name) {
3245			kvfree(new_xattr);
3246			return -ENOMEM;
3247		}
3248
3249		memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3250		       XATTR_SECURITY_PREFIX_LEN);
3251		memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3252		       xattr->name, len);
3253
3254		simple_xattr_list_add(&info->xattrs, new_xattr);
3255	}
3256
3257	return 0;
3258}
3259
3260static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3261				   struct dentry *unused, struct inode *inode,
3262				   const char *name, void *buffer, size_t size)
3263{
3264	struct shmem_inode_info *info = SHMEM_I(inode);
3265
3266	name = xattr_full_name(handler, name);
3267	return simple_xattr_get(&info->xattrs, name, buffer, size);
3268}
3269
3270static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3271				   struct dentry *unused, struct inode *inode,
3272				   const char *name, const void *value,
3273				   size_t size, int flags)
3274{
3275	struct shmem_inode_info *info = SHMEM_I(inode);
3276
3277	name = xattr_full_name(handler, name);
3278	return simple_xattr_set(&info->xattrs, name, value, size, flags, NULL);
3279}
3280
3281static const struct xattr_handler shmem_security_xattr_handler = {
3282	.prefix = XATTR_SECURITY_PREFIX,
3283	.get = shmem_xattr_handler_get,
3284	.set = shmem_xattr_handler_set,
3285};
3286
3287static const struct xattr_handler shmem_trusted_xattr_handler = {
3288	.prefix = XATTR_TRUSTED_PREFIX,
3289	.get = shmem_xattr_handler_get,
3290	.set = shmem_xattr_handler_set,
3291};
3292
3293static const struct xattr_handler *shmem_xattr_handlers[] = {
3294#ifdef CONFIG_TMPFS_POSIX_ACL
3295	&posix_acl_access_xattr_handler,
3296	&posix_acl_default_xattr_handler,
3297#endif
3298	&shmem_security_xattr_handler,
3299	&shmem_trusted_xattr_handler,
3300	NULL
3301};
3302
3303static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3304{
3305	struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3306	return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3307}
3308#endif /* CONFIG_TMPFS_XATTR */
3309
3310static const struct inode_operations shmem_short_symlink_operations = {
3311	.get_link	= simple_get_link,
3312#ifdef CONFIG_TMPFS_XATTR
3313	.listxattr	= shmem_listxattr,
3314#endif
3315};
3316
3317static const struct inode_operations shmem_symlink_inode_operations = {
3318	.get_link	= shmem_get_link,
3319#ifdef CONFIG_TMPFS_XATTR
3320	.listxattr	= shmem_listxattr,
3321#endif
3322};
3323
3324static struct dentry *shmem_get_parent(struct dentry *child)
3325{
3326	return ERR_PTR(-ESTALE);
3327}
3328
3329static int shmem_match(struct inode *ino, void *vfh)
3330{
3331	__u32 *fh = vfh;
3332	__u64 inum = fh[2];
3333	inum = (inum << 32) | fh[1];
3334	return ino->i_ino == inum && fh[0] == ino->i_generation;
3335}
3336
3337/* Find any alias of inode, but prefer a hashed alias */
3338static struct dentry *shmem_find_alias(struct inode *inode)
3339{
3340	struct dentry *alias = d_find_alias(inode);
3341
3342	return alias ?: d_find_any_alias(inode);
3343}
3344
3345
3346static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3347		struct fid *fid, int fh_len, int fh_type)
3348{
3349	struct inode *inode;
3350	struct dentry *dentry = NULL;
3351	u64 inum;
3352
3353	if (fh_len < 3)
3354		return NULL;
3355
3356	inum = fid->raw[2];
3357	inum = (inum << 32) | fid->raw[1];
3358
3359	inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3360			shmem_match, fid->raw);
3361	if (inode) {
3362		dentry = shmem_find_alias(inode);
3363		iput(inode);
3364	}
3365
3366	return dentry;
3367}
3368
3369static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3370				struct inode *parent)
3371{
3372	if (*len < 3) {
3373		*len = 3;
3374		return FILEID_INVALID;
3375	}
3376
3377	if (inode_unhashed(inode)) {
3378		/* Unfortunately insert_inode_hash is not idempotent,
3379		 * so as we hash inodes here rather than at creation
3380		 * time, we need a lock to ensure we only try
3381		 * to do it once
3382		 */
3383		static DEFINE_SPINLOCK(lock);
3384		spin_lock(&lock);
3385		if (inode_unhashed(inode))
3386			__insert_inode_hash(inode,
3387					    inode->i_ino + inode->i_generation);
3388		spin_unlock(&lock);
3389	}
3390
3391	fh[0] = inode->i_generation;
3392	fh[1] = inode->i_ino;
3393	fh[2] = ((__u64)inode->i_ino) >> 32;
3394
3395	*len = 3;
3396	return 1;
3397}
3398
3399static const struct export_operations shmem_export_ops = {
3400	.get_parent     = shmem_get_parent,
3401	.encode_fh      = shmem_encode_fh,
3402	.fh_to_dentry	= shmem_fh_to_dentry,
3403};
3404
3405enum shmem_param {
3406	Opt_gid,
3407	Opt_huge,
3408	Opt_mode,
3409	Opt_mpol,
3410	Opt_nr_blocks,
3411	Opt_nr_inodes,
3412	Opt_size,
3413	Opt_uid,
3414	Opt_inode32,
3415	Opt_inode64,
3416};
3417
3418static const struct constant_table shmem_param_enums_huge[] = {
3419	{"never",	SHMEM_HUGE_NEVER },
3420	{"always",	SHMEM_HUGE_ALWAYS },
3421	{"within_size",	SHMEM_HUGE_WITHIN_SIZE },
3422	{"advise",	SHMEM_HUGE_ADVISE },
3423	{}
3424};
3425
3426const struct fs_parameter_spec shmem_fs_parameters[] = {
3427	fsparam_u32   ("gid",		Opt_gid),
3428	fsparam_enum  ("huge",		Opt_huge,  shmem_param_enums_huge),
3429	fsparam_u32oct("mode",		Opt_mode),
3430	fsparam_string("mpol",		Opt_mpol),
3431	fsparam_string("nr_blocks",	Opt_nr_blocks),
3432	fsparam_string("nr_inodes",	Opt_nr_inodes),
3433	fsparam_string("size",		Opt_size),
3434	fsparam_u32   ("uid",		Opt_uid),
3435	fsparam_flag  ("inode32",	Opt_inode32),
3436	fsparam_flag  ("inode64",	Opt_inode64),
3437	{}
3438};
3439
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3440static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
3441{
3442	struct shmem_options *ctx = fc->fs_private;
3443	struct fs_parse_result result;
3444	unsigned long long size;
3445	char *rest;
3446	int opt;
3447
3448	opt = fs_parse(fc, shmem_fs_parameters, param, &result);
3449	if (opt < 0)
3450		return opt;
3451
3452	switch (opt) {
3453	case Opt_size:
3454		size = memparse(param->string, &rest);
3455		if (*rest == '%') {
3456			size <<= PAGE_SHIFT;
3457			size *= totalram_pages();
3458			do_div(size, 100);
3459			rest++;
3460		}
3461		if (*rest)
3462			goto bad_value;
3463		ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3464		ctx->seen |= SHMEM_SEEN_BLOCKS;
3465		break;
3466	case Opt_nr_blocks:
3467		ctx->blocks = memparse(param->string, &rest);
3468		if (*rest)
3469			goto bad_value;
3470		ctx->seen |= SHMEM_SEEN_BLOCKS;
3471		break;
3472	case Opt_nr_inodes:
3473		ctx->inodes = memparse(param->string, &rest);
3474		if (*rest)
3475			goto bad_value;
3476		ctx->seen |= SHMEM_SEEN_INODES;
3477		break;
3478	case Opt_mode:
3479		ctx->mode = result.uint_32 & 07777;
3480		break;
3481	case Opt_uid:
3482		ctx->uid = make_kuid(current_user_ns(), result.uint_32);
3483		if (!uid_valid(ctx->uid))
3484			goto bad_value;
3485		break;
3486	case Opt_gid:
3487		ctx->gid = make_kgid(current_user_ns(), result.uint_32);
3488		if (!gid_valid(ctx->gid))
3489			goto bad_value;
3490		break;
3491	case Opt_huge:
3492		ctx->huge = result.uint_32;
3493		if (ctx->huge != SHMEM_HUGE_NEVER &&
3494		    !(IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
3495		      has_transparent_hugepage()))
3496			goto unsupported_parameter;
3497		ctx->seen |= SHMEM_SEEN_HUGE;
3498		break;
3499	case Opt_mpol:
3500		if (IS_ENABLED(CONFIG_NUMA)) {
3501			mpol_put(ctx->mpol);
3502			ctx->mpol = NULL;
3503			if (mpol_parse_str(param->string, &ctx->mpol))
3504				goto bad_value;
3505			break;
3506		}
3507		goto unsupported_parameter;
3508	case Opt_inode32:
3509		ctx->full_inums = false;
3510		ctx->seen |= SHMEM_SEEN_INUMS;
3511		break;
3512	case Opt_inode64:
3513		if (sizeof(ino_t) < 8) {
3514			return invalfc(fc,
3515				       "Cannot use inode64 with <64bit inums in kernel\n");
3516		}
3517		ctx->full_inums = true;
3518		ctx->seen |= SHMEM_SEEN_INUMS;
3519		break;
3520	}
3521	return 0;
3522
3523unsupported_parameter:
3524	return invalfc(fc, "Unsupported parameter '%s'", param->key);
3525bad_value:
3526	return invalfc(fc, "Bad value for '%s'", param->key);
3527}
3528
3529static int shmem_parse_options(struct fs_context *fc, void *data)
3530{
3531	char *options = data;
3532
3533	if (options) {
3534		int err = security_sb_eat_lsm_opts(options, &fc->security);
3535		if (err)
3536			return err;
3537	}
3538
3539	while (options != NULL) {
3540		char *this_char = options;
3541		for (;;) {
3542			/*
3543			 * NUL-terminate this option: unfortunately,
3544			 * mount options form a comma-separated list,
3545			 * but mpol's nodelist may also contain commas.
3546			 */
3547			options = strchr(options, ',');
3548			if (options == NULL)
3549				break;
3550			options++;
3551			if (!isdigit(*options)) {
3552				options[-1] = '\0';
3553				break;
3554			}
3555		}
3556		if (*this_char) {
3557			char *value = strchr(this_char,'=');
3558			size_t len = 0;
3559			int err;
3560
3561			if (value) {
3562				*value++ = '\0';
3563				len = strlen(value);
3564			}
3565			err = vfs_parse_fs_string(fc, this_char, value, len);
3566			if (err < 0)
3567				return err;
3568		}
3569	}
3570	return 0;
3571}
3572
3573/*
3574 * Reconfigure a shmem filesystem.
3575 *
3576 * Note that we disallow change from limited->unlimited blocks/inodes while any
3577 * are in use; but we must separately disallow unlimited->limited, because in
3578 * that case we have no record of how much is already in use.
3579 */
3580static int shmem_reconfigure(struct fs_context *fc)
3581{
3582	struct shmem_options *ctx = fc->fs_private;
3583	struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
3584	unsigned long inodes;
3585	const char *err;
3586
3587	spin_lock(&sbinfo->stat_lock);
3588	inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3589	if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
3590		if (!sbinfo->max_blocks) {
3591			err = "Cannot retroactively limit size";
3592			goto out;
3593		}
3594		if (percpu_counter_compare(&sbinfo->used_blocks,
3595					   ctx->blocks) > 0) {
3596			err = "Too small a size for current use";
3597			goto out;
3598		}
3599	}
3600	if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
3601		if (!sbinfo->max_inodes) {
3602			err = "Cannot retroactively limit inodes";
3603			goto out;
3604		}
3605		if (ctx->inodes < inodes) {
3606			err = "Too few inodes for current use";
3607			goto out;
3608		}
3609	}
3610
3611	if ((ctx->seen & SHMEM_SEEN_INUMS) && !ctx->full_inums &&
3612	    sbinfo->next_ino > UINT_MAX) {
3613		err = "Current inum too high to switch to 32-bit inums";
3614		goto out;
3615	}
3616
3617	if (ctx->seen & SHMEM_SEEN_HUGE)
3618		sbinfo->huge = ctx->huge;
3619	if (ctx->seen & SHMEM_SEEN_INUMS)
3620		sbinfo->full_inums = ctx->full_inums;
3621	if (ctx->seen & SHMEM_SEEN_BLOCKS)
3622		sbinfo->max_blocks  = ctx->blocks;
3623	if (ctx->seen & SHMEM_SEEN_INODES) {
3624		sbinfo->max_inodes  = ctx->inodes;
3625		sbinfo->free_inodes = ctx->inodes - inodes;
3626	}
3627
3628	/*
3629	 * Preserve previous mempolicy unless mpol remount option was specified.
3630	 */
3631	if (ctx->mpol) {
3632		mpol_put(sbinfo->mpol);
3633		sbinfo->mpol = ctx->mpol;	/* transfers initial ref */
3634		ctx->mpol = NULL;
3635	}
3636	spin_unlock(&sbinfo->stat_lock);
3637	return 0;
3638out:
3639	spin_unlock(&sbinfo->stat_lock);
3640	return invalfc(fc, "%s", err);
3641}
3642
3643static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3644{
3645	struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3646
3647	if (sbinfo->max_blocks != shmem_default_max_blocks())
3648		seq_printf(seq, ",size=%luk",
3649			sbinfo->max_blocks << (PAGE_SHIFT - 10));
3650	if (sbinfo->max_inodes != shmem_default_max_inodes())
3651		seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3652	if (sbinfo->mode != (0777 | S_ISVTX))
3653		seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3654	if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3655		seq_printf(seq, ",uid=%u",
3656				from_kuid_munged(&init_user_ns, sbinfo->uid));
3657	if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3658		seq_printf(seq, ",gid=%u",
3659				from_kgid_munged(&init_user_ns, sbinfo->gid));
3660
3661	/*
3662	 * Showing inode{64,32} might be useful even if it's the system default,
3663	 * since then people don't have to resort to checking both here and
3664	 * /proc/config.gz to confirm 64-bit inums were successfully applied
3665	 * (which may not even exist if IKCONFIG_PROC isn't enabled).
3666	 *
3667	 * We hide it when inode64 isn't the default and we are using 32-bit
3668	 * inodes, since that probably just means the feature isn't even under
3669	 * consideration.
3670	 *
3671	 * As such:
3672	 *
3673	 *                     +-----------------+-----------------+
3674	 *                     | TMPFS_INODE64=y | TMPFS_INODE64=n |
3675	 *  +------------------+-----------------+-----------------+
3676	 *  | full_inums=true  | show            | show            |
3677	 *  | full_inums=false | show            | hide            |
3678	 *  +------------------+-----------------+-----------------+
3679	 *
3680	 */
3681	if (IS_ENABLED(CONFIG_TMPFS_INODE64) || sbinfo->full_inums)
3682		seq_printf(seq, ",inode%d", (sbinfo->full_inums ? 64 : 32));
3683#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3684	/* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3685	if (sbinfo->huge)
3686		seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3687#endif
3688	shmem_show_mpol(seq, sbinfo->mpol);
3689	return 0;
3690}
3691
3692#endif /* CONFIG_TMPFS */
3693
3694static void shmem_put_super(struct super_block *sb)
3695{
3696	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3697
3698	free_percpu(sbinfo->ino_batch);
3699	percpu_counter_destroy(&sbinfo->used_blocks);
3700	mpol_put(sbinfo->mpol);
3701	kfree(sbinfo);
3702	sb->s_fs_info = NULL;
3703}
3704
3705static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
3706{
3707	struct shmem_options *ctx = fc->fs_private;
3708	struct inode *inode;
3709	struct shmem_sb_info *sbinfo;
3710	int err = -ENOMEM;
3711
3712	/* Round up to L1_CACHE_BYTES to resist false sharing */
3713	sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3714				L1_CACHE_BYTES), GFP_KERNEL);
3715	if (!sbinfo)
3716		return -ENOMEM;
3717
3718	sb->s_fs_info = sbinfo;
3719
3720#ifdef CONFIG_TMPFS
3721	/*
3722	 * Per default we only allow half of the physical ram per
3723	 * tmpfs instance, limiting inodes to one per page of lowmem;
3724	 * but the internal instance is left unlimited.
3725	 */
3726	if (!(sb->s_flags & SB_KERNMOUNT)) {
3727		if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
3728			ctx->blocks = shmem_default_max_blocks();
3729		if (!(ctx->seen & SHMEM_SEEN_INODES))
3730			ctx->inodes = shmem_default_max_inodes();
3731		if (!(ctx->seen & SHMEM_SEEN_INUMS))
3732			ctx->full_inums = IS_ENABLED(CONFIG_TMPFS_INODE64);
3733	} else {
3734		sb->s_flags |= SB_NOUSER;
3735	}
3736	sb->s_export_op = &shmem_export_ops;
3737	sb->s_flags |= SB_NOSEC;
3738#else
3739	sb->s_flags |= SB_NOUSER;
3740#endif
3741	sbinfo->max_blocks = ctx->blocks;
3742	sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes;
3743	if (sb->s_flags & SB_KERNMOUNT) {
3744		sbinfo->ino_batch = alloc_percpu(ino_t);
3745		if (!sbinfo->ino_batch)
3746			goto failed;
3747	}
3748	sbinfo->uid = ctx->uid;
3749	sbinfo->gid = ctx->gid;
3750	sbinfo->full_inums = ctx->full_inums;
3751	sbinfo->mode = ctx->mode;
3752	sbinfo->huge = ctx->huge;
3753	sbinfo->mpol = ctx->mpol;
3754	ctx->mpol = NULL;
3755
3756	spin_lock_init(&sbinfo->stat_lock);
3757	if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3758		goto failed;
3759	spin_lock_init(&sbinfo->shrinklist_lock);
3760	INIT_LIST_HEAD(&sbinfo->shrinklist);
3761
3762	sb->s_maxbytes = MAX_LFS_FILESIZE;
3763	sb->s_blocksize = PAGE_SIZE;
3764	sb->s_blocksize_bits = PAGE_SHIFT;
3765	sb->s_magic = TMPFS_MAGIC;
3766	sb->s_op = &shmem_ops;
3767	sb->s_time_gran = 1;
3768#ifdef CONFIG_TMPFS_XATTR
3769	sb->s_xattr = shmem_xattr_handlers;
3770#endif
3771#ifdef CONFIG_TMPFS_POSIX_ACL
3772	sb->s_flags |= SB_POSIXACL;
3773#endif
3774	uuid_gen(&sb->s_uuid);
3775
3776	inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3777	if (!inode)
3778		goto failed;
3779	inode->i_uid = sbinfo->uid;
3780	inode->i_gid = sbinfo->gid;
3781	sb->s_root = d_make_root(inode);
3782	if (!sb->s_root)
3783		goto failed;
3784	return 0;
3785
3786failed:
3787	shmem_put_super(sb);
3788	return err;
3789}
3790
3791static int shmem_get_tree(struct fs_context *fc)
3792{
3793	return get_tree_nodev(fc, shmem_fill_super);
3794}
3795
3796static void shmem_free_fc(struct fs_context *fc)
3797{
3798	struct shmem_options *ctx = fc->fs_private;
3799
3800	if (ctx) {
3801		mpol_put(ctx->mpol);
3802		kfree(ctx);
3803	}
3804}
3805
3806static const struct fs_context_operations shmem_fs_context_ops = {
3807	.free			= shmem_free_fc,
3808	.get_tree		= shmem_get_tree,
3809#ifdef CONFIG_TMPFS
3810	.parse_monolithic	= shmem_parse_options,
3811	.parse_param		= shmem_parse_one,
3812	.reconfigure		= shmem_reconfigure,
3813#endif
3814};
3815
3816static struct kmem_cache *shmem_inode_cachep;
3817
3818static struct inode *shmem_alloc_inode(struct super_block *sb)
3819{
3820	struct shmem_inode_info *info;
3821	info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3822	if (!info)
3823		return NULL;
3824	return &info->vfs_inode;
3825}
3826
3827static void shmem_free_in_core_inode(struct inode *inode)
3828{
3829	if (S_ISLNK(inode->i_mode))
3830		kfree(inode->i_link);
3831	kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3832}
3833
3834static void shmem_destroy_inode(struct inode *inode)
3835{
3836	if (S_ISREG(inode->i_mode))
3837		mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3838}
3839
3840static void shmem_init_inode(void *foo)
3841{
3842	struct shmem_inode_info *info = foo;
3843	inode_init_once(&info->vfs_inode);
3844}
3845
3846static void shmem_init_inodecache(void)
3847{
3848	shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3849				sizeof(struct shmem_inode_info),
3850				0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3851}
3852
3853static void shmem_destroy_inodecache(void)
3854{
3855	kmem_cache_destroy(shmem_inode_cachep);
3856}
3857
3858static const struct address_space_operations shmem_aops = {
3859	.writepage	= shmem_writepage,
3860	.set_page_dirty	= __set_page_dirty_no_writeback,
3861#ifdef CONFIG_TMPFS
3862	.write_begin	= shmem_write_begin,
3863	.write_end	= shmem_write_end,
3864#endif
3865#ifdef CONFIG_MIGRATION
3866	.migratepage	= migrate_page,
3867#endif
3868	.error_remove_page = generic_error_remove_page,
3869};
3870
3871static const struct file_operations shmem_file_operations = {
3872	.mmap		= shmem_mmap,
3873	.get_unmapped_area = shmem_get_unmapped_area,
3874#ifdef CONFIG_TMPFS
3875	.llseek		= shmem_file_llseek,
3876	.read_iter	= shmem_file_read_iter,
3877	.write_iter	= generic_file_write_iter,
3878	.fsync		= noop_fsync,
3879	.splice_read	= generic_file_splice_read,
3880	.splice_write	= iter_file_splice_write,
3881	.fallocate	= shmem_fallocate,
3882#endif
3883};
3884
3885static const struct inode_operations shmem_inode_operations = {
3886	.getattr	= shmem_getattr,
3887	.setattr	= shmem_setattr,
3888#ifdef CONFIG_TMPFS_XATTR
3889	.listxattr	= shmem_listxattr,
3890	.set_acl	= simple_set_acl,
3891#endif
3892};
3893
3894static const struct inode_operations shmem_dir_inode_operations = {
3895#ifdef CONFIG_TMPFS
3896	.create		= shmem_create,
3897	.lookup		= simple_lookup,
3898	.link		= shmem_link,
3899	.unlink		= shmem_unlink,
3900	.symlink	= shmem_symlink,
3901	.mkdir		= shmem_mkdir,
3902	.rmdir		= shmem_rmdir,
3903	.mknod		= shmem_mknod,
3904	.rename		= shmem_rename2,
3905	.tmpfile	= shmem_tmpfile,
3906#endif
3907#ifdef CONFIG_TMPFS_XATTR
3908	.listxattr	= shmem_listxattr,
3909#endif
3910#ifdef CONFIG_TMPFS_POSIX_ACL
3911	.setattr	= shmem_setattr,
3912	.set_acl	= simple_set_acl,
3913#endif
3914};
3915
3916static const struct inode_operations shmem_special_inode_operations = {
3917#ifdef CONFIG_TMPFS_XATTR
3918	.listxattr	= shmem_listxattr,
3919#endif
3920#ifdef CONFIG_TMPFS_POSIX_ACL
3921	.setattr	= shmem_setattr,
3922	.set_acl	= simple_set_acl,
3923#endif
3924};
3925
3926static const struct super_operations shmem_ops = {
3927	.alloc_inode	= shmem_alloc_inode,
3928	.free_inode	= shmem_free_in_core_inode,
3929	.destroy_inode	= shmem_destroy_inode,
3930#ifdef CONFIG_TMPFS
3931	.statfs		= shmem_statfs,
3932	.show_options	= shmem_show_options,
3933#endif
3934	.evict_inode	= shmem_evict_inode,
3935	.drop_inode	= generic_delete_inode,
3936	.put_super	= shmem_put_super,
3937#ifdef CONFIG_TRANSPARENT_HUGEPAGE
3938	.nr_cached_objects	= shmem_unused_huge_count,
3939	.free_cached_objects	= shmem_unused_huge_scan,
3940#endif
3941};
3942
3943static const struct vm_operations_struct shmem_vm_ops = {
3944	.fault		= shmem_fault,
3945	.map_pages	= filemap_map_pages,
3946#ifdef CONFIG_NUMA
3947	.set_policy     = shmem_set_policy,
3948	.get_policy     = shmem_get_policy,
3949#endif
3950};
3951
3952int shmem_init_fs_context(struct fs_context *fc)
3953{
3954	struct shmem_options *ctx;
3955
3956	ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
3957	if (!ctx)
3958		return -ENOMEM;
3959
3960	ctx->mode = 0777 | S_ISVTX;
3961	ctx->uid = current_fsuid();
3962	ctx->gid = current_fsgid();
3963
3964	fc->fs_private = ctx;
3965	fc->ops = &shmem_fs_context_ops;
3966	return 0;
3967}
3968
3969static struct file_system_type shmem_fs_type = {
3970	.owner		= THIS_MODULE,
3971	.name		= "tmpfs",
3972	.init_fs_context = shmem_init_fs_context,
3973#ifdef CONFIG_TMPFS
3974	.parameters	= shmem_fs_parameters,
3975#endif
3976	.kill_sb	= kill_litter_super,
3977	.fs_flags	= FS_USERNS_MOUNT,
3978};
3979
3980int __init shmem_init(void)
3981{
3982	int error;
3983
3984	shmem_init_inodecache();
3985
3986	error = register_filesystem(&shmem_fs_type);
3987	if (error) {
3988		pr_err("Could not register tmpfs\n");
3989		goto out2;
3990	}
3991
3992	shm_mnt = kern_mount(&shmem_fs_type);
3993	if (IS_ERR(shm_mnt)) {
3994		error = PTR_ERR(shm_mnt);
3995		pr_err("Could not kern_mount tmpfs\n");
3996		goto out1;
3997	}
3998
3999#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4000	if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
4001		SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4002	else
4003		shmem_huge = 0; /* just in case it was patched */
4004#endif
4005	return 0;
4006
4007out1:
4008	unregister_filesystem(&shmem_fs_type);
4009out2:
4010	shmem_destroy_inodecache();
4011	shm_mnt = ERR_PTR(error);
4012	return error;
4013}
4014
4015#if defined(CONFIG_TRANSPARENT_HUGEPAGE) && defined(CONFIG_SYSFS)
4016static ssize_t shmem_enabled_show(struct kobject *kobj,
4017		struct kobj_attribute *attr, char *buf)
4018{
4019	static const int values[] = {
4020		SHMEM_HUGE_ALWAYS,
4021		SHMEM_HUGE_WITHIN_SIZE,
4022		SHMEM_HUGE_ADVISE,
4023		SHMEM_HUGE_NEVER,
4024		SHMEM_HUGE_DENY,
4025		SHMEM_HUGE_FORCE,
4026	};
4027	int i, count;
4028
4029	for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
4030		const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
4031
4032		count += sprintf(buf + count, fmt,
4033				shmem_format_huge(values[i]));
4034	}
4035	buf[count - 1] = '\n';
4036	return count;
4037}
4038
4039static ssize_t shmem_enabled_store(struct kobject *kobj,
4040		struct kobj_attribute *attr, const char *buf, size_t count)
4041{
4042	char tmp[16];
4043	int huge;
4044
4045	if (count + 1 > sizeof(tmp))
4046		return -EINVAL;
4047	memcpy(tmp, buf, count);
4048	tmp[count] = '\0';
4049	if (count && tmp[count - 1] == '\n')
4050		tmp[count - 1] = '\0';
4051
4052	huge = shmem_parse_huge(tmp);
4053	if (huge == -EINVAL)
4054		return -EINVAL;
4055	if (!has_transparent_hugepage() &&
4056			huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
4057		return -EINVAL;
4058
4059	shmem_huge = huge;
4060	if (shmem_huge > SHMEM_HUGE_DENY)
4061		SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4062	return count;
4063}
4064
4065struct kobj_attribute shmem_enabled_attr =
4066	__ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
4067#endif /* CONFIG_TRANSPARENT_HUGEPAGE && CONFIG_SYSFS */
4068
4069#ifdef CONFIG_TRANSPARENT_HUGEPAGE
4070bool shmem_huge_enabled(struct vm_area_struct *vma)
4071{
4072	struct inode *inode = file_inode(vma->vm_file);
4073	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
4074	loff_t i_size;
4075	pgoff_t off;
4076
4077	if ((vma->vm_flags & VM_NOHUGEPAGE) ||
4078	    test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
4079		return false;
4080	if (shmem_huge == SHMEM_HUGE_FORCE)
4081		return true;
4082	if (shmem_huge == SHMEM_HUGE_DENY)
4083		return false;
4084	switch (sbinfo->huge) {
4085		case SHMEM_HUGE_NEVER:
4086			return false;
4087		case SHMEM_HUGE_ALWAYS:
4088			return true;
4089		case SHMEM_HUGE_WITHIN_SIZE:
4090			off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
4091			i_size = round_up(i_size_read(inode), PAGE_SIZE);
4092			if (i_size >= HPAGE_PMD_SIZE &&
4093					i_size >> PAGE_SHIFT >= off)
4094				return true;
4095			fallthrough;
4096		case SHMEM_HUGE_ADVISE:
4097			/* TODO: implement fadvise() hints */
4098			return (vma->vm_flags & VM_HUGEPAGE);
4099		default:
4100			VM_BUG_ON(1);
4101			return false;
4102	}
4103}
4104#endif /* CONFIG_TRANSPARENT_HUGEPAGE */
4105
4106#else /* !CONFIG_SHMEM */
4107
4108/*
4109 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4110 *
4111 * This is intended for small system where the benefits of the full
4112 * shmem code (swap-backed and resource-limited) are outweighed by
4113 * their complexity. On systems without swap this code should be
4114 * effectively equivalent, but much lighter weight.
4115 */
4116
4117static struct file_system_type shmem_fs_type = {
4118	.name		= "tmpfs",
4119	.init_fs_context = ramfs_init_fs_context,
4120	.parameters	= ramfs_fs_parameters,
4121	.kill_sb	= kill_litter_super,
4122	.fs_flags	= FS_USERNS_MOUNT,
4123};
4124
4125int __init shmem_init(void)
4126{
4127	BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4128
4129	shm_mnt = kern_mount(&shmem_fs_type);
4130	BUG_ON(IS_ERR(shm_mnt));
4131
4132	return 0;
4133}
4134
4135int shmem_unuse(unsigned int type, bool frontswap,
4136		unsigned long *fs_pages_to_unuse)
4137{
4138	return 0;
4139}
4140
4141int shmem_lock(struct file *file, int lock, struct user_struct *user)
4142{
4143	return 0;
4144}
4145
4146void shmem_unlock_mapping(struct address_space *mapping)
4147{
4148}
4149
4150#ifdef CONFIG_MMU
4151unsigned long shmem_get_unmapped_area(struct file *file,
4152				      unsigned long addr, unsigned long len,
4153				      unsigned long pgoff, unsigned long flags)
4154{
4155	return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4156}
4157#endif
4158
4159void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4160{
4161	truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4162}
4163EXPORT_SYMBOL_GPL(shmem_truncate_range);
4164
4165#define shmem_vm_ops				generic_file_vm_ops
4166#define shmem_file_operations			ramfs_file_operations
4167#define shmem_get_inode(sb, dir, mode, dev, flags)	ramfs_get_inode(sb, dir, mode, dev)
4168#define shmem_acct_size(flags, size)		0
4169#define shmem_unacct_size(flags, size)		do {} while (0)
4170
4171#endif /* CONFIG_SHMEM */
4172
4173/* common code */
4174
4175static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4176				       unsigned long flags, unsigned int i_flags)
4177{
4178	struct inode *inode;
4179	struct file *res;
4180
4181	if (IS_ERR(mnt))
4182		return ERR_CAST(mnt);
4183
4184	if (size < 0 || size > MAX_LFS_FILESIZE)
4185		return ERR_PTR(-EINVAL);
4186
4187	if (shmem_acct_size(flags, size))
4188		return ERR_PTR(-ENOMEM);
4189
4190	inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0,
4191				flags);
4192	if (unlikely(!inode)) {
4193		shmem_unacct_size(flags, size);
4194		return ERR_PTR(-ENOSPC);
4195	}
4196	inode->i_flags |= i_flags;
4197	inode->i_size = size;
4198	clear_nlink(inode);	/* It is unlinked */
4199	res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4200	if (!IS_ERR(res))
4201		res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4202				&shmem_file_operations);
4203	if (IS_ERR(res))
4204		iput(inode);
4205	return res;
4206}
4207
4208/**
4209 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4210 * 	kernel internal.  There will be NO LSM permission checks against the
4211 * 	underlying inode.  So users of this interface must do LSM checks at a
4212 *	higher layer.  The users are the big_key and shm implementations.  LSM
4213 *	checks are provided at the key or shm level rather than the inode.
4214 * @name: name for dentry (to be seen in /proc/<pid>/maps
4215 * @size: size to be set for the file
4216 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4217 */
4218struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4219{
4220	return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4221}
4222
4223/**
4224 * shmem_file_setup - get an unlinked file living in tmpfs
4225 * @name: name for dentry (to be seen in /proc/<pid>/maps
4226 * @size: size to be set for the file
4227 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4228 */
4229struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4230{
4231	return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4232}
4233EXPORT_SYMBOL_GPL(shmem_file_setup);
4234
4235/**
4236 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4237 * @mnt: the tmpfs mount where the file will be created
4238 * @name: name for dentry (to be seen in /proc/<pid>/maps
4239 * @size: size to be set for the file
4240 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4241 */
4242struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4243				       loff_t size, unsigned long flags)
4244{
4245	return __shmem_file_setup(mnt, name, size, flags, 0);
4246}
4247EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4248
4249/**
4250 * shmem_zero_setup - setup a shared anonymous mapping
4251 * @vma: the vma to be mmapped is prepared by do_mmap
4252 */
4253int shmem_zero_setup(struct vm_area_struct *vma)
4254{
4255	struct file *file;
4256	loff_t size = vma->vm_end - vma->vm_start;
4257
4258	/*
4259	 * Cloning a new file under mmap_lock leads to a lock ordering conflict
4260	 * between XFS directory reading and selinux: since this file is only
4261	 * accessible to the user through its mapping, use S_PRIVATE flag to
4262	 * bypass file security, in the same way as shmem_kernel_file_setup().
4263	 */
4264	file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4265	if (IS_ERR(file))
4266		return PTR_ERR(file);
4267
4268	if (vma->vm_file)
4269		fput(vma->vm_file);
4270	vma->vm_file = file;
4271	vma->vm_ops = &shmem_vm_ops;
4272
4273	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE) &&
4274			((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4275			(vma->vm_end & HPAGE_PMD_MASK)) {
4276		khugepaged_enter(vma, vma->vm_flags);
4277	}
4278
4279	return 0;
4280}
4281
4282/**
4283 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4284 * @mapping:	the page's address_space
4285 * @index:	the page index
4286 * @gfp:	the page allocator flags to use if allocating
4287 *
4288 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4289 * with any new page allocations done using the specified allocation flags.
4290 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4291 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4292 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4293 *
4294 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4295 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4296 */
4297struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4298					 pgoff_t index, gfp_t gfp)
4299{
4300#ifdef CONFIG_SHMEM
4301	struct inode *inode = mapping->host;
4302	struct page *page;
4303	int error;
4304
4305	BUG_ON(mapping->a_ops != &shmem_aops);
4306	error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4307				  gfp, NULL, NULL, NULL);
4308	if (error)
4309		page = ERR_PTR(error);
4310	else
4311		unlock_page(page);
4312	return page;
4313#else
4314	/*
4315	 * The tiny !SHMEM case uses ramfs without swap
4316	 */
4317	return read_cache_page_gfp(mapping, index, gfp);
4318#endif
4319}
4320EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);
v5.4
   1/*
   2 * Resizable virtual memory filesystem for Linux.
   3 *
   4 * Copyright (C) 2000 Linus Torvalds.
   5 *		 2000 Transmeta Corp.
   6 *		 2000-2001 Christoph Rohland
   7 *		 2000-2001 SAP AG
   8 *		 2002 Red Hat Inc.
   9 * Copyright (C) 2002-2011 Hugh Dickins.
  10 * Copyright (C) 2011 Google Inc.
  11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
  12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
  13 *
  14 * Extended attribute support for tmpfs:
  15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
  16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
  17 *
  18 * tiny-shmem:
  19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
  20 *
  21 * This file is released under the GPL.
  22 */
  23
  24#include <linux/fs.h>
  25#include <linux/init.h>
  26#include <linux/vfs.h>
  27#include <linux/mount.h>
  28#include <linux/ramfs.h>
  29#include <linux/pagemap.h>
  30#include <linux/file.h>
  31#include <linux/mm.h>
  32#include <linux/random.h>
  33#include <linux/sched/signal.h>
  34#include <linux/export.h>
  35#include <linux/swap.h>
  36#include <linux/uio.h>
  37#include <linux/khugepaged.h>
  38#include <linux/hugetlb.h>
  39#include <linux/frontswap.h>
  40#include <linux/fs_parser.h>
  41
  42#include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */
  43
  44static struct vfsmount *shm_mnt;
  45
  46#ifdef CONFIG_SHMEM
  47/*
  48 * This virtual memory filesystem is heavily based on the ramfs. It
  49 * extends ramfs by the ability to use swap and honor resource limits
  50 * which makes it a completely usable filesystem.
  51 */
  52
  53#include <linux/xattr.h>
  54#include <linux/exportfs.h>
  55#include <linux/posix_acl.h>
  56#include <linux/posix_acl_xattr.h>
  57#include <linux/mman.h>
  58#include <linux/string.h>
  59#include <linux/slab.h>
  60#include <linux/backing-dev.h>
  61#include <linux/shmem_fs.h>
  62#include <linux/writeback.h>
  63#include <linux/blkdev.h>
  64#include <linux/pagevec.h>
  65#include <linux/percpu_counter.h>
  66#include <linux/falloc.h>
  67#include <linux/splice.h>
  68#include <linux/security.h>
  69#include <linux/swapops.h>
  70#include <linux/mempolicy.h>
  71#include <linux/namei.h>
  72#include <linux/ctype.h>
  73#include <linux/migrate.h>
  74#include <linux/highmem.h>
  75#include <linux/seq_file.h>
  76#include <linux/magic.h>
  77#include <linux/syscalls.h>
  78#include <linux/fcntl.h>
  79#include <uapi/linux/memfd.h>
  80#include <linux/userfaultfd_k.h>
  81#include <linux/rmap.h>
  82#include <linux/uuid.h>
  83
  84#include <linux/uaccess.h>
  85#include <asm/pgtable.h>
  86
  87#include "internal.h"
  88
  89#define BLOCKS_PER_PAGE  (PAGE_SIZE/512)
  90#define VM_ACCT(size)    (PAGE_ALIGN(size) >> PAGE_SHIFT)
  91
  92/* Pretend that each entry is of this size in directory's i_size */
  93#define BOGO_DIRENT_SIZE 20
  94
  95/* Symlink up to this size is kmalloc'ed instead of using a swappable page */
  96#define SHORT_SYMLINK_LEN 128
  97
  98/*
  99 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
 100 * inode->i_private (with i_mutex making sure that it has only one user at
 101 * a time): we would prefer not to enlarge the shmem inode just for that.
 102 */
 103struct shmem_falloc {
 104	wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
 105	pgoff_t start;		/* start of range currently being fallocated */
 106	pgoff_t next;		/* the next page offset to be fallocated */
 107	pgoff_t nr_falloced;	/* how many new pages have been fallocated */
 108	pgoff_t nr_unswapped;	/* how often writepage refused to swap out */
 109};
 110
 111struct shmem_options {
 112	unsigned long long blocks;
 113	unsigned long long inodes;
 114	struct mempolicy *mpol;
 115	kuid_t uid;
 116	kgid_t gid;
 117	umode_t mode;
 
 118	int huge;
 119	int seen;
 120#define SHMEM_SEEN_BLOCKS 1
 121#define SHMEM_SEEN_INODES 2
 122#define SHMEM_SEEN_HUGE 4
 
 123};
 124
 125#ifdef CONFIG_TMPFS
 126static unsigned long shmem_default_max_blocks(void)
 127{
 128	return totalram_pages() / 2;
 129}
 130
 131static unsigned long shmem_default_max_inodes(void)
 132{
 133	unsigned long nr_pages = totalram_pages();
 134
 135	return min(nr_pages - totalhigh_pages(), nr_pages / 2);
 136}
 137#endif
 138
 139static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
 140static int shmem_replace_page(struct page **pagep, gfp_t gfp,
 141				struct shmem_inode_info *info, pgoff_t index);
 142static int shmem_swapin_page(struct inode *inode, pgoff_t index,
 143			     struct page **pagep, enum sgp_type sgp,
 144			     gfp_t gfp, struct vm_area_struct *vma,
 145			     vm_fault_t *fault_type);
 146static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
 147		struct page **pagep, enum sgp_type sgp,
 148		gfp_t gfp, struct vm_area_struct *vma,
 149		struct vm_fault *vmf, vm_fault_t *fault_type);
 150
 151int shmem_getpage(struct inode *inode, pgoff_t index,
 152		struct page **pagep, enum sgp_type sgp)
 153{
 154	return shmem_getpage_gfp(inode, index, pagep, sgp,
 155		mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
 156}
 157
 158static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
 159{
 160	return sb->s_fs_info;
 161}
 162
 163/*
 164 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
 165 * for shared memory and for shared anonymous (/dev/zero) mappings
 166 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
 167 * consistent with the pre-accounting of private mappings ...
 168 */
 169static inline int shmem_acct_size(unsigned long flags, loff_t size)
 170{
 171	return (flags & VM_NORESERVE) ?
 172		0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
 173}
 174
 175static inline void shmem_unacct_size(unsigned long flags, loff_t size)
 176{
 177	if (!(flags & VM_NORESERVE))
 178		vm_unacct_memory(VM_ACCT(size));
 179}
 180
 181static inline int shmem_reacct_size(unsigned long flags,
 182		loff_t oldsize, loff_t newsize)
 183{
 184	if (!(flags & VM_NORESERVE)) {
 185		if (VM_ACCT(newsize) > VM_ACCT(oldsize))
 186			return security_vm_enough_memory_mm(current->mm,
 187					VM_ACCT(newsize) - VM_ACCT(oldsize));
 188		else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
 189			vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
 190	}
 191	return 0;
 192}
 193
 194/*
 195 * ... whereas tmpfs objects are accounted incrementally as
 196 * pages are allocated, in order to allow large sparse files.
 197 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
 198 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
 199 */
 200static inline int shmem_acct_block(unsigned long flags, long pages)
 201{
 202	if (!(flags & VM_NORESERVE))
 203		return 0;
 204
 205	return security_vm_enough_memory_mm(current->mm,
 206			pages * VM_ACCT(PAGE_SIZE));
 207}
 208
 209static inline void shmem_unacct_blocks(unsigned long flags, long pages)
 210{
 211	if (flags & VM_NORESERVE)
 212		vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
 213}
 214
 215static inline bool shmem_inode_acct_block(struct inode *inode, long pages)
 216{
 217	struct shmem_inode_info *info = SHMEM_I(inode);
 218	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 219
 220	if (shmem_acct_block(info->flags, pages))
 221		return false;
 222
 223	if (sbinfo->max_blocks) {
 224		if (percpu_counter_compare(&sbinfo->used_blocks,
 225					   sbinfo->max_blocks - pages) > 0)
 226			goto unacct;
 227		percpu_counter_add(&sbinfo->used_blocks, pages);
 228	}
 229
 230	return true;
 231
 232unacct:
 233	shmem_unacct_blocks(info->flags, pages);
 234	return false;
 235}
 236
 237static inline void shmem_inode_unacct_blocks(struct inode *inode, long pages)
 238{
 239	struct shmem_inode_info *info = SHMEM_I(inode);
 240	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 241
 242	if (sbinfo->max_blocks)
 243		percpu_counter_sub(&sbinfo->used_blocks, pages);
 244	shmem_unacct_blocks(info->flags, pages);
 245}
 246
 247static const struct super_operations shmem_ops;
 248static const struct address_space_operations shmem_aops;
 249static const struct file_operations shmem_file_operations;
 250static const struct inode_operations shmem_inode_operations;
 251static const struct inode_operations shmem_dir_inode_operations;
 252static const struct inode_operations shmem_special_inode_operations;
 253static const struct vm_operations_struct shmem_vm_ops;
 254static struct file_system_type shmem_fs_type;
 255
 256bool vma_is_shmem(struct vm_area_struct *vma)
 257{
 258	return vma->vm_ops == &shmem_vm_ops;
 259}
 260
 261static LIST_HEAD(shmem_swaplist);
 262static DEFINE_MUTEX(shmem_swaplist_mutex);
 263
 264static int shmem_reserve_inode(struct super_block *sb)
 
 
 
 
 
 
 
 
 
 
 265{
 266	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 267	if (sbinfo->max_inodes) {
 
 
 268		spin_lock(&sbinfo->stat_lock);
 269		if (!sbinfo->free_inodes) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 270			spin_unlock(&sbinfo->stat_lock);
 271			return -ENOSPC;
 
 272		}
 273		sbinfo->free_inodes--;
 274		spin_unlock(&sbinfo->stat_lock);
 
 275	}
 
 276	return 0;
 277}
 278
 279static void shmem_free_inode(struct super_block *sb)
 280{
 281	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 282	if (sbinfo->max_inodes) {
 283		spin_lock(&sbinfo->stat_lock);
 284		sbinfo->free_inodes++;
 285		spin_unlock(&sbinfo->stat_lock);
 286	}
 287}
 288
 289/**
 290 * shmem_recalc_inode - recalculate the block usage of an inode
 291 * @inode: inode to recalc
 292 *
 293 * We have to calculate the free blocks since the mm can drop
 294 * undirtied hole pages behind our back.
 295 *
 296 * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
 297 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
 298 *
 299 * It has to be called with the spinlock held.
 300 */
 301static void shmem_recalc_inode(struct inode *inode)
 302{
 303	struct shmem_inode_info *info = SHMEM_I(inode);
 304	long freed;
 305
 306	freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
 307	if (freed > 0) {
 308		info->alloced -= freed;
 309		inode->i_blocks -= freed * BLOCKS_PER_PAGE;
 310		shmem_inode_unacct_blocks(inode, freed);
 311	}
 312}
 313
 314bool shmem_charge(struct inode *inode, long pages)
 315{
 316	struct shmem_inode_info *info = SHMEM_I(inode);
 317	unsigned long flags;
 318
 319	if (!shmem_inode_acct_block(inode, pages))
 320		return false;
 321
 322	/* nrpages adjustment first, then shmem_recalc_inode() when balanced */
 323	inode->i_mapping->nrpages += pages;
 324
 325	spin_lock_irqsave(&info->lock, flags);
 326	info->alloced += pages;
 327	inode->i_blocks += pages * BLOCKS_PER_PAGE;
 328	shmem_recalc_inode(inode);
 329	spin_unlock_irqrestore(&info->lock, flags);
 330
 331	return true;
 332}
 333
 334void shmem_uncharge(struct inode *inode, long pages)
 335{
 336	struct shmem_inode_info *info = SHMEM_I(inode);
 337	unsigned long flags;
 338
 339	/* nrpages adjustment done by __delete_from_page_cache() or caller */
 340
 341	spin_lock_irqsave(&info->lock, flags);
 342	info->alloced -= pages;
 343	inode->i_blocks -= pages * BLOCKS_PER_PAGE;
 344	shmem_recalc_inode(inode);
 345	spin_unlock_irqrestore(&info->lock, flags);
 346
 347	shmem_inode_unacct_blocks(inode, pages);
 348}
 349
 350/*
 351 * Replace item expected in xarray by a new item, while holding xa_lock.
 352 */
 353static int shmem_replace_entry(struct address_space *mapping,
 354			pgoff_t index, void *expected, void *replacement)
 355{
 356	XA_STATE(xas, &mapping->i_pages, index);
 357	void *item;
 358
 359	VM_BUG_ON(!expected);
 360	VM_BUG_ON(!replacement);
 361	item = xas_load(&xas);
 362	if (item != expected)
 363		return -ENOENT;
 364	xas_store(&xas, replacement);
 365	return 0;
 366}
 367
 368/*
 369 * Sometimes, before we decide whether to proceed or to fail, we must check
 370 * that an entry was not already brought back from swap by a racing thread.
 371 *
 372 * Checking page is not enough: by the time a SwapCache page is locked, it
 373 * might be reused, and again be SwapCache, using the same swap as before.
 374 */
 375static bool shmem_confirm_swap(struct address_space *mapping,
 376			       pgoff_t index, swp_entry_t swap)
 377{
 378	return xa_load(&mapping->i_pages, index) == swp_to_radix_entry(swap);
 379}
 380
 381/*
 382 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
 383 *
 384 * SHMEM_HUGE_NEVER:
 385 *	disables huge pages for the mount;
 386 * SHMEM_HUGE_ALWAYS:
 387 *	enables huge pages for the mount;
 388 * SHMEM_HUGE_WITHIN_SIZE:
 389 *	only allocate huge pages if the page will be fully within i_size,
 390 *	also respect fadvise()/madvise() hints;
 391 * SHMEM_HUGE_ADVISE:
 392 *	only allocate huge pages if requested with fadvise()/madvise();
 393 */
 394
 395#define SHMEM_HUGE_NEVER	0
 396#define SHMEM_HUGE_ALWAYS	1
 397#define SHMEM_HUGE_WITHIN_SIZE	2
 398#define SHMEM_HUGE_ADVISE	3
 399
 400/*
 401 * Special values.
 402 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
 403 *
 404 * SHMEM_HUGE_DENY:
 405 *	disables huge on shm_mnt and all mounts, for emergency use;
 406 * SHMEM_HUGE_FORCE:
 407 *	enables huge on shm_mnt and all mounts, w/o needing option, for testing;
 408 *
 409 */
 410#define SHMEM_HUGE_DENY		(-1)
 411#define SHMEM_HUGE_FORCE	(-2)
 412
 413#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
 414/* ifdef here to avoid bloating shmem.o when not necessary */
 415
 416static int shmem_huge __read_mostly;
 417
 418#if defined(CONFIG_SYSFS)
 419static int shmem_parse_huge(const char *str)
 420{
 421	if (!strcmp(str, "never"))
 422		return SHMEM_HUGE_NEVER;
 423	if (!strcmp(str, "always"))
 424		return SHMEM_HUGE_ALWAYS;
 425	if (!strcmp(str, "within_size"))
 426		return SHMEM_HUGE_WITHIN_SIZE;
 427	if (!strcmp(str, "advise"))
 428		return SHMEM_HUGE_ADVISE;
 429	if (!strcmp(str, "deny"))
 430		return SHMEM_HUGE_DENY;
 431	if (!strcmp(str, "force"))
 432		return SHMEM_HUGE_FORCE;
 433	return -EINVAL;
 434}
 435#endif
 436
 437#if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
 438static const char *shmem_format_huge(int huge)
 439{
 440	switch (huge) {
 441	case SHMEM_HUGE_NEVER:
 442		return "never";
 443	case SHMEM_HUGE_ALWAYS:
 444		return "always";
 445	case SHMEM_HUGE_WITHIN_SIZE:
 446		return "within_size";
 447	case SHMEM_HUGE_ADVISE:
 448		return "advise";
 449	case SHMEM_HUGE_DENY:
 450		return "deny";
 451	case SHMEM_HUGE_FORCE:
 452		return "force";
 453	default:
 454		VM_BUG_ON(1);
 455		return "bad_val";
 456	}
 457}
 458#endif
 459
 460static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
 461		struct shrink_control *sc, unsigned long nr_to_split)
 462{
 463	LIST_HEAD(list), *pos, *next;
 464	LIST_HEAD(to_remove);
 465	struct inode *inode;
 466	struct shmem_inode_info *info;
 467	struct page *page;
 468	unsigned long batch = sc ? sc->nr_to_scan : 128;
 469	int removed = 0, split = 0;
 470
 471	if (list_empty(&sbinfo->shrinklist))
 472		return SHRINK_STOP;
 473
 474	spin_lock(&sbinfo->shrinklist_lock);
 475	list_for_each_safe(pos, next, &sbinfo->shrinklist) {
 476		info = list_entry(pos, struct shmem_inode_info, shrinklist);
 477
 478		/* pin the inode */
 479		inode = igrab(&info->vfs_inode);
 480
 481		/* inode is about to be evicted */
 482		if (!inode) {
 483			list_del_init(&info->shrinklist);
 484			removed++;
 485			goto next;
 486		}
 487
 488		/* Check if there's anything to gain */
 489		if (round_up(inode->i_size, PAGE_SIZE) ==
 490				round_up(inode->i_size, HPAGE_PMD_SIZE)) {
 491			list_move(&info->shrinklist, &to_remove);
 492			removed++;
 493			goto next;
 494		}
 495
 496		list_move(&info->shrinklist, &list);
 497next:
 498		if (!--batch)
 499			break;
 500	}
 501	spin_unlock(&sbinfo->shrinklist_lock);
 502
 503	list_for_each_safe(pos, next, &to_remove) {
 504		info = list_entry(pos, struct shmem_inode_info, shrinklist);
 505		inode = &info->vfs_inode;
 506		list_del_init(&info->shrinklist);
 507		iput(inode);
 508	}
 509
 510	list_for_each_safe(pos, next, &list) {
 511		int ret;
 512
 513		info = list_entry(pos, struct shmem_inode_info, shrinklist);
 514		inode = &info->vfs_inode;
 515
 516		if (nr_to_split && split >= nr_to_split)
 517			goto leave;
 518
 519		page = find_get_page(inode->i_mapping,
 520				(inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
 521		if (!page)
 522			goto drop;
 523
 524		/* No huge page at the end of the file: nothing to split */
 525		if (!PageTransHuge(page)) {
 526			put_page(page);
 527			goto drop;
 528		}
 529
 530		/*
 531		 * Leave the inode on the list if we failed to lock
 532		 * the page at this time.
 533		 *
 534		 * Waiting for the lock may lead to deadlock in the
 535		 * reclaim path.
 536		 */
 537		if (!trylock_page(page)) {
 538			put_page(page);
 539			goto leave;
 540		}
 541
 542		ret = split_huge_page(page);
 543		unlock_page(page);
 544		put_page(page);
 545
 546		/* If split failed leave the inode on the list */
 547		if (ret)
 548			goto leave;
 549
 550		split++;
 551drop:
 552		list_del_init(&info->shrinklist);
 553		removed++;
 554leave:
 555		iput(inode);
 556	}
 557
 558	spin_lock(&sbinfo->shrinklist_lock);
 559	list_splice_tail(&list, &sbinfo->shrinklist);
 560	sbinfo->shrinklist_len -= removed;
 561	spin_unlock(&sbinfo->shrinklist_lock);
 562
 563	return split;
 564}
 565
 566static long shmem_unused_huge_scan(struct super_block *sb,
 567		struct shrink_control *sc)
 568{
 569	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 570
 571	if (!READ_ONCE(sbinfo->shrinklist_len))
 572		return SHRINK_STOP;
 573
 574	return shmem_unused_huge_shrink(sbinfo, sc, 0);
 575}
 576
 577static long shmem_unused_huge_count(struct super_block *sb,
 578		struct shrink_control *sc)
 579{
 580	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 581	return READ_ONCE(sbinfo->shrinklist_len);
 582}
 583#else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
 584
 585#define shmem_huge SHMEM_HUGE_DENY
 586
 587static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
 588		struct shrink_control *sc, unsigned long nr_to_split)
 589{
 590	return 0;
 591}
 592#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
 593
 594static inline bool is_huge_enabled(struct shmem_sb_info *sbinfo)
 595{
 596	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
 597	    (shmem_huge == SHMEM_HUGE_FORCE || sbinfo->huge) &&
 598	    shmem_huge != SHMEM_HUGE_DENY)
 599		return true;
 600	return false;
 601}
 602
 603/*
 604 * Like add_to_page_cache_locked, but error if expected item has gone.
 605 */
 606static int shmem_add_to_page_cache(struct page *page,
 607				   struct address_space *mapping,
 608				   pgoff_t index, void *expected, gfp_t gfp)
 
 609{
 610	XA_STATE_ORDER(xas, &mapping->i_pages, index, compound_order(page));
 611	unsigned long i = 0;
 612	unsigned long nr = compound_nr(page);
 
 613
 614	VM_BUG_ON_PAGE(PageTail(page), page);
 615	VM_BUG_ON_PAGE(index != round_down(index, nr), page);
 616	VM_BUG_ON_PAGE(!PageLocked(page), page);
 617	VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
 618	VM_BUG_ON(expected && PageTransHuge(page));
 619
 620	page_ref_add(page, nr);
 621	page->mapping = mapping;
 622	page->index = index;
 623
 
 
 
 
 
 
 
 
 
 
 
 
 624	do {
 625		void *entry;
 626		xas_lock_irq(&xas);
 627		entry = xas_find_conflict(&xas);
 628		if (entry != expected)
 629			xas_set_err(&xas, -EEXIST);
 630		xas_create_range(&xas);
 631		if (xas_error(&xas))
 632			goto unlock;
 633next:
 634		xas_store(&xas, page);
 635		if (++i < nr) {
 636			xas_next(&xas);
 637			goto next;
 638		}
 639		if (PageTransHuge(page)) {
 640			count_vm_event(THP_FILE_ALLOC);
 641			__inc_node_page_state(page, NR_SHMEM_THPS);
 642		}
 643		mapping->nrpages += nr;
 644		__mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
 645		__mod_node_page_state(page_pgdat(page), NR_SHMEM, nr);
 646unlock:
 647		xas_unlock_irq(&xas);
 648	} while (xas_nomem(&xas, gfp));
 649
 650	if (xas_error(&xas)) {
 651		page->mapping = NULL;
 652		page_ref_sub(page, nr);
 653		return xas_error(&xas);
 654	}
 655
 656	return 0;
 
 
 
 
 657}
 658
 659/*
 660 * Like delete_from_page_cache, but substitutes swap for page.
 661 */
 662static void shmem_delete_from_page_cache(struct page *page, void *radswap)
 663{
 664	struct address_space *mapping = page->mapping;
 665	int error;
 666
 667	VM_BUG_ON_PAGE(PageCompound(page), page);
 668
 669	xa_lock_irq(&mapping->i_pages);
 670	error = shmem_replace_entry(mapping, page->index, page, radswap);
 671	page->mapping = NULL;
 672	mapping->nrpages--;
 673	__dec_node_page_state(page, NR_FILE_PAGES);
 674	__dec_node_page_state(page, NR_SHMEM);
 675	xa_unlock_irq(&mapping->i_pages);
 676	put_page(page);
 677	BUG_ON(error);
 678}
 679
 680/*
 681 * Remove swap entry from page cache, free the swap and its page cache.
 682 */
 683static int shmem_free_swap(struct address_space *mapping,
 684			   pgoff_t index, void *radswap)
 685{
 686	void *old;
 687
 688	old = xa_cmpxchg_irq(&mapping->i_pages, index, radswap, NULL, 0);
 689	if (old != radswap)
 690		return -ENOENT;
 691	free_swap_and_cache(radix_to_swp_entry(radswap));
 692	return 0;
 693}
 694
 695/*
 696 * Determine (in bytes) how many of the shmem object's pages mapped by the
 697 * given offsets are swapped out.
 698 *
 699 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
 700 * as long as the inode doesn't go away and racy results are not a problem.
 701 */
 702unsigned long shmem_partial_swap_usage(struct address_space *mapping,
 703						pgoff_t start, pgoff_t end)
 704{
 705	XA_STATE(xas, &mapping->i_pages, start);
 706	struct page *page;
 707	unsigned long swapped = 0;
 708
 709	rcu_read_lock();
 710	xas_for_each(&xas, page, end - 1) {
 711		if (xas_retry(&xas, page))
 712			continue;
 713		if (xa_is_value(page))
 714			swapped++;
 715
 716		if (need_resched()) {
 717			xas_pause(&xas);
 718			cond_resched_rcu();
 719		}
 720	}
 721
 722	rcu_read_unlock();
 723
 724	return swapped << PAGE_SHIFT;
 725}
 726
 727/*
 728 * Determine (in bytes) how many of the shmem object's pages mapped by the
 729 * given vma is swapped out.
 730 *
 731 * This is safe to call without i_mutex or the i_pages lock thanks to RCU,
 732 * as long as the inode doesn't go away and racy results are not a problem.
 733 */
 734unsigned long shmem_swap_usage(struct vm_area_struct *vma)
 735{
 736	struct inode *inode = file_inode(vma->vm_file);
 737	struct shmem_inode_info *info = SHMEM_I(inode);
 738	struct address_space *mapping = inode->i_mapping;
 739	unsigned long swapped;
 740
 741	/* Be careful as we don't hold info->lock */
 742	swapped = READ_ONCE(info->swapped);
 743
 744	/*
 745	 * The easier cases are when the shmem object has nothing in swap, or
 746	 * the vma maps it whole. Then we can simply use the stats that we
 747	 * already track.
 748	 */
 749	if (!swapped)
 750		return 0;
 751
 752	if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
 753		return swapped << PAGE_SHIFT;
 754
 755	/* Here comes the more involved part */
 756	return shmem_partial_swap_usage(mapping,
 757			linear_page_index(vma, vma->vm_start),
 758			linear_page_index(vma, vma->vm_end));
 759}
 760
 761/*
 762 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
 763 */
 764void shmem_unlock_mapping(struct address_space *mapping)
 765{
 766	struct pagevec pvec;
 767	pgoff_t indices[PAGEVEC_SIZE];
 768	pgoff_t index = 0;
 769
 770	pagevec_init(&pvec);
 771	/*
 772	 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
 773	 */
 774	while (!mapping_unevictable(mapping)) {
 775		/*
 776		 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
 777		 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
 778		 */
 779		pvec.nr = find_get_entries(mapping, index,
 780					   PAGEVEC_SIZE, pvec.pages, indices);
 781		if (!pvec.nr)
 782			break;
 783		index = indices[pvec.nr - 1] + 1;
 784		pagevec_remove_exceptionals(&pvec);
 785		check_move_unevictable_pages(&pvec);
 786		pagevec_release(&pvec);
 787		cond_resched();
 788	}
 789}
 790
 791/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 792 * Remove range of pages and swap entries from page cache, and free them.
 793 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
 794 */
 795static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
 796								 bool unfalloc)
 797{
 798	struct address_space *mapping = inode->i_mapping;
 799	struct shmem_inode_info *info = SHMEM_I(inode);
 800	pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
 801	pgoff_t end = (lend + 1) >> PAGE_SHIFT;
 802	unsigned int partial_start = lstart & (PAGE_SIZE - 1);
 803	unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
 804	struct pagevec pvec;
 805	pgoff_t indices[PAGEVEC_SIZE];
 806	long nr_swaps_freed = 0;
 807	pgoff_t index;
 808	int i;
 809
 810	if (lend == -1)
 811		end = -1;	/* unsigned, so actually very big */
 812
 813	pagevec_init(&pvec);
 814	index = start;
 815	while (index < end) {
 816		pvec.nr = find_get_entries(mapping, index,
 817			min(end - index, (pgoff_t)PAGEVEC_SIZE),
 818			pvec.pages, indices);
 819		if (!pvec.nr)
 820			break;
 821		for (i = 0; i < pagevec_count(&pvec); i++) {
 822			struct page *page = pvec.pages[i];
 823
 824			index = indices[i];
 825			if (index >= end)
 826				break;
 827
 828			if (xa_is_value(page)) {
 829				if (unfalloc)
 830					continue;
 831				nr_swaps_freed += !shmem_free_swap(mapping,
 832								index, page);
 833				continue;
 834			}
 835
 836			VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
 837
 838			if (!trylock_page(page))
 839				continue;
 840
 841			if (PageTransTail(page)) {
 842				/* Middle of THP: zero out the page */
 843				clear_highpage(page);
 844				unlock_page(page);
 845				continue;
 846			} else if (PageTransHuge(page)) {
 847				if (index == round_down(end, HPAGE_PMD_NR)) {
 848					/*
 849					 * Range ends in the middle of THP:
 850					 * zero out the page
 851					 */
 852					clear_highpage(page);
 853					unlock_page(page);
 854					continue;
 855				}
 856				index += HPAGE_PMD_NR - 1;
 857				i += HPAGE_PMD_NR - 1;
 858			}
 859
 860			if (!unfalloc || !PageUptodate(page)) {
 861				VM_BUG_ON_PAGE(PageTail(page), page);
 862				if (page_mapping(page) == mapping) {
 863					VM_BUG_ON_PAGE(PageWriteback(page), page);
 864					truncate_inode_page(mapping, page);
 865				}
 866			}
 867			unlock_page(page);
 868		}
 869		pagevec_remove_exceptionals(&pvec);
 870		pagevec_release(&pvec);
 871		cond_resched();
 872		index++;
 873	}
 874
 875	if (partial_start) {
 876		struct page *page = NULL;
 877		shmem_getpage(inode, start - 1, &page, SGP_READ);
 878		if (page) {
 879			unsigned int top = PAGE_SIZE;
 880			if (start > end) {
 881				top = partial_end;
 882				partial_end = 0;
 883			}
 884			zero_user_segment(page, partial_start, top);
 885			set_page_dirty(page);
 886			unlock_page(page);
 887			put_page(page);
 888		}
 889	}
 890	if (partial_end) {
 891		struct page *page = NULL;
 892		shmem_getpage(inode, end, &page, SGP_READ);
 893		if (page) {
 894			zero_user_segment(page, 0, partial_end);
 895			set_page_dirty(page);
 896			unlock_page(page);
 897			put_page(page);
 898		}
 899	}
 900	if (start >= end)
 901		return;
 902
 903	index = start;
 904	while (index < end) {
 905		cond_resched();
 906
 907		pvec.nr = find_get_entries(mapping, index,
 908				min(end - index, (pgoff_t)PAGEVEC_SIZE),
 909				pvec.pages, indices);
 910		if (!pvec.nr) {
 911			/* If all gone or hole-punch or unfalloc, we're done */
 912			if (index == start || end != -1)
 913				break;
 914			/* But if truncating, restart to make sure all gone */
 915			index = start;
 916			continue;
 917		}
 918		for (i = 0; i < pagevec_count(&pvec); i++) {
 919			struct page *page = pvec.pages[i];
 920
 921			index = indices[i];
 922			if (index >= end)
 923				break;
 924
 925			if (xa_is_value(page)) {
 926				if (unfalloc)
 927					continue;
 928				if (shmem_free_swap(mapping, index, page)) {
 929					/* Swap was replaced by page: retry */
 930					index--;
 931					break;
 932				}
 933				nr_swaps_freed++;
 934				continue;
 935			}
 936
 937			lock_page(page);
 938
 939			if (PageTransTail(page)) {
 940				/* Middle of THP: zero out the page */
 941				clear_highpage(page);
 942				unlock_page(page);
 943				/*
 944				 * Partial thp truncate due 'start' in middle
 945				 * of THP: don't need to look on these pages
 946				 * again on !pvec.nr restart.
 947				 */
 948				if (index != round_down(end, HPAGE_PMD_NR))
 949					start++;
 950				continue;
 951			} else if (PageTransHuge(page)) {
 952				if (index == round_down(end, HPAGE_PMD_NR)) {
 953					/*
 954					 * Range ends in the middle of THP:
 955					 * zero out the page
 956					 */
 957					clear_highpage(page);
 958					unlock_page(page);
 959					continue;
 960				}
 961				index += HPAGE_PMD_NR - 1;
 962				i += HPAGE_PMD_NR - 1;
 963			}
 964
 965			if (!unfalloc || !PageUptodate(page)) {
 966				VM_BUG_ON_PAGE(PageTail(page), page);
 967				if (page_mapping(page) == mapping) {
 968					VM_BUG_ON_PAGE(PageWriteback(page), page);
 969					truncate_inode_page(mapping, page);
 970				} else {
 971					/* Page was replaced by swap: retry */
 972					unlock_page(page);
 973					index--;
 974					break;
 975				}
 
 
 
 
 
 
 
 
 
 
 
 
 976			}
 977			unlock_page(page);
 978		}
 979		pagevec_remove_exceptionals(&pvec);
 980		pagevec_release(&pvec);
 981		index++;
 982	}
 983
 984	spin_lock_irq(&info->lock);
 985	info->swapped -= nr_swaps_freed;
 986	shmem_recalc_inode(inode);
 987	spin_unlock_irq(&info->lock);
 988}
 989
 990void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
 991{
 992	shmem_undo_range(inode, lstart, lend, false);
 993	inode->i_ctime = inode->i_mtime = current_time(inode);
 994}
 995EXPORT_SYMBOL_GPL(shmem_truncate_range);
 996
 997static int shmem_getattr(const struct path *path, struct kstat *stat,
 998			 u32 request_mask, unsigned int query_flags)
 999{
1000	struct inode *inode = path->dentry->d_inode;
1001	struct shmem_inode_info *info = SHMEM_I(inode);
1002	struct shmem_sb_info *sb_info = SHMEM_SB(inode->i_sb);
1003
1004	if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
1005		spin_lock_irq(&info->lock);
1006		shmem_recalc_inode(inode);
1007		spin_unlock_irq(&info->lock);
1008	}
1009	generic_fillattr(inode, stat);
1010
1011	if (is_huge_enabled(sb_info))
1012		stat->blksize = HPAGE_PMD_SIZE;
1013
1014	return 0;
1015}
1016
1017static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
1018{
1019	struct inode *inode = d_inode(dentry);
1020	struct shmem_inode_info *info = SHMEM_I(inode);
1021	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1022	int error;
1023
1024	error = setattr_prepare(dentry, attr);
1025	if (error)
1026		return error;
1027
1028	if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
1029		loff_t oldsize = inode->i_size;
1030		loff_t newsize = attr->ia_size;
1031
1032		/* protected by i_mutex */
1033		if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
1034		    (newsize > oldsize && (info->seals & F_SEAL_GROW)))
1035			return -EPERM;
1036
1037		if (newsize != oldsize) {
1038			error = shmem_reacct_size(SHMEM_I(inode)->flags,
1039					oldsize, newsize);
1040			if (error)
1041				return error;
1042			i_size_write(inode, newsize);
1043			inode->i_ctime = inode->i_mtime = current_time(inode);
1044		}
1045		if (newsize <= oldsize) {
1046			loff_t holebegin = round_up(newsize, PAGE_SIZE);
1047			if (oldsize > holebegin)
1048				unmap_mapping_range(inode->i_mapping,
1049							holebegin, 0, 1);
1050			if (info->alloced)
1051				shmem_truncate_range(inode,
1052							newsize, (loff_t)-1);
1053			/* unmap again to remove racily COWed private pages */
1054			if (oldsize > holebegin)
1055				unmap_mapping_range(inode->i_mapping,
1056							holebegin, 0, 1);
1057
1058			/*
1059			 * Part of the huge page can be beyond i_size: subject
1060			 * to shrink under memory pressure.
1061			 */
1062			if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) {
1063				spin_lock(&sbinfo->shrinklist_lock);
1064				/*
1065				 * _careful to defend against unlocked access to
1066				 * ->shrink_list in shmem_unused_huge_shrink()
1067				 */
1068				if (list_empty_careful(&info->shrinklist)) {
1069					list_add_tail(&info->shrinklist,
1070							&sbinfo->shrinklist);
1071					sbinfo->shrinklist_len++;
1072				}
1073				spin_unlock(&sbinfo->shrinklist_lock);
1074			}
1075		}
1076	}
1077
1078	setattr_copy(inode, attr);
1079	if (attr->ia_valid & ATTR_MODE)
1080		error = posix_acl_chmod(inode, inode->i_mode);
1081	return error;
1082}
1083
1084static void shmem_evict_inode(struct inode *inode)
1085{
1086	struct shmem_inode_info *info = SHMEM_I(inode);
1087	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1088
1089	if (inode->i_mapping->a_ops == &shmem_aops) {
1090		shmem_unacct_size(info->flags, inode->i_size);
1091		inode->i_size = 0;
1092		shmem_truncate_range(inode, 0, (loff_t)-1);
1093		if (!list_empty(&info->shrinklist)) {
1094			spin_lock(&sbinfo->shrinklist_lock);
1095			if (!list_empty(&info->shrinklist)) {
1096				list_del_init(&info->shrinklist);
1097				sbinfo->shrinklist_len--;
1098			}
1099			spin_unlock(&sbinfo->shrinklist_lock);
1100		}
1101		while (!list_empty(&info->swaplist)) {
1102			/* Wait while shmem_unuse() is scanning this inode... */
1103			wait_var_event(&info->stop_eviction,
1104				       !atomic_read(&info->stop_eviction));
1105			mutex_lock(&shmem_swaplist_mutex);
1106			/* ...but beware of the race if we peeked too early */
1107			if (!atomic_read(&info->stop_eviction))
1108				list_del_init(&info->swaplist);
1109			mutex_unlock(&shmem_swaplist_mutex);
1110		}
1111	}
1112
1113	simple_xattrs_free(&info->xattrs);
1114	WARN_ON(inode->i_blocks);
1115	shmem_free_inode(inode->i_sb);
1116	clear_inode(inode);
1117}
1118
1119extern struct swap_info_struct *swap_info[];
1120
1121static int shmem_find_swap_entries(struct address_space *mapping,
1122				   pgoff_t start, unsigned int nr_entries,
1123				   struct page **entries, pgoff_t *indices,
1124				   unsigned int type, bool frontswap)
1125{
1126	XA_STATE(xas, &mapping->i_pages, start);
1127	struct page *page;
1128	swp_entry_t entry;
1129	unsigned int ret = 0;
1130
1131	if (!nr_entries)
1132		return 0;
1133
1134	rcu_read_lock();
1135	xas_for_each(&xas, page, ULONG_MAX) {
1136		if (xas_retry(&xas, page))
1137			continue;
1138
1139		if (!xa_is_value(page))
1140			continue;
1141
1142		entry = radix_to_swp_entry(page);
1143		if (swp_type(entry) != type)
1144			continue;
1145		if (frontswap &&
1146		    !frontswap_test(swap_info[type], swp_offset(entry)))
1147			continue;
1148
1149		indices[ret] = xas.xa_index;
1150		entries[ret] = page;
1151
1152		if (need_resched()) {
1153			xas_pause(&xas);
1154			cond_resched_rcu();
1155		}
1156		if (++ret == nr_entries)
1157			break;
1158	}
1159	rcu_read_unlock();
1160
1161	return ret;
1162}
1163
1164/*
1165 * Move the swapped pages for an inode to page cache. Returns the count
1166 * of pages swapped in, or the error in case of failure.
1167 */
1168static int shmem_unuse_swap_entries(struct inode *inode, struct pagevec pvec,
1169				    pgoff_t *indices)
1170{
1171	int i = 0;
1172	int ret = 0;
1173	int error = 0;
1174	struct address_space *mapping = inode->i_mapping;
1175
1176	for (i = 0; i < pvec.nr; i++) {
1177		struct page *page = pvec.pages[i];
1178
1179		if (!xa_is_value(page))
1180			continue;
1181		error = shmem_swapin_page(inode, indices[i],
1182					  &page, SGP_CACHE,
1183					  mapping_gfp_mask(mapping),
1184					  NULL, NULL);
1185		if (error == 0) {
1186			unlock_page(page);
1187			put_page(page);
1188			ret++;
1189		}
1190		if (error == -ENOMEM)
1191			break;
1192		error = 0;
1193	}
1194	return error ? error : ret;
1195}
1196
1197/*
1198 * If swap found in inode, free it and move page from swapcache to filecache.
1199 */
1200static int shmem_unuse_inode(struct inode *inode, unsigned int type,
1201			     bool frontswap, unsigned long *fs_pages_to_unuse)
1202{
1203	struct address_space *mapping = inode->i_mapping;
1204	pgoff_t start = 0;
1205	struct pagevec pvec;
1206	pgoff_t indices[PAGEVEC_SIZE];
1207	bool frontswap_partial = (frontswap && *fs_pages_to_unuse > 0);
1208	int ret = 0;
1209
1210	pagevec_init(&pvec);
1211	do {
1212		unsigned int nr_entries = PAGEVEC_SIZE;
1213
1214		if (frontswap_partial && *fs_pages_to_unuse < PAGEVEC_SIZE)
1215			nr_entries = *fs_pages_to_unuse;
1216
1217		pvec.nr = shmem_find_swap_entries(mapping, start, nr_entries,
1218						  pvec.pages, indices,
1219						  type, frontswap);
1220		if (pvec.nr == 0) {
1221			ret = 0;
1222			break;
1223		}
1224
1225		ret = shmem_unuse_swap_entries(inode, pvec, indices);
1226		if (ret < 0)
1227			break;
1228
1229		if (frontswap_partial) {
1230			*fs_pages_to_unuse -= ret;
1231			if (*fs_pages_to_unuse == 0) {
1232				ret = FRONTSWAP_PAGES_UNUSED;
1233				break;
1234			}
1235		}
1236
1237		start = indices[pvec.nr - 1];
1238	} while (true);
1239
1240	return ret;
1241}
1242
1243/*
1244 * Read all the shared memory data that resides in the swap
1245 * device 'type' back into memory, so the swap device can be
1246 * unused.
1247 */
1248int shmem_unuse(unsigned int type, bool frontswap,
1249		unsigned long *fs_pages_to_unuse)
1250{
1251	struct shmem_inode_info *info, *next;
1252	int error = 0;
1253
1254	if (list_empty(&shmem_swaplist))
1255		return 0;
1256
1257	mutex_lock(&shmem_swaplist_mutex);
1258	list_for_each_entry_safe(info, next, &shmem_swaplist, swaplist) {
1259		if (!info->swapped) {
1260			list_del_init(&info->swaplist);
1261			continue;
1262		}
1263		/*
1264		 * Drop the swaplist mutex while searching the inode for swap;
1265		 * but before doing so, make sure shmem_evict_inode() will not
1266		 * remove placeholder inode from swaplist, nor let it be freed
1267		 * (igrab() would protect from unlink, but not from unmount).
1268		 */
1269		atomic_inc(&info->stop_eviction);
1270		mutex_unlock(&shmem_swaplist_mutex);
1271
1272		error = shmem_unuse_inode(&info->vfs_inode, type, frontswap,
1273					  fs_pages_to_unuse);
1274		cond_resched();
1275
1276		mutex_lock(&shmem_swaplist_mutex);
1277		next = list_next_entry(info, swaplist);
1278		if (!info->swapped)
1279			list_del_init(&info->swaplist);
1280		if (atomic_dec_and_test(&info->stop_eviction))
1281			wake_up_var(&info->stop_eviction);
1282		if (error)
1283			break;
1284	}
1285	mutex_unlock(&shmem_swaplist_mutex);
1286
1287	return error;
1288}
1289
1290/*
1291 * Move the page from the page cache to the swap cache.
1292 */
1293static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1294{
1295	struct shmem_inode_info *info;
1296	struct address_space *mapping;
1297	struct inode *inode;
1298	swp_entry_t swap;
1299	pgoff_t index;
1300
1301	VM_BUG_ON_PAGE(PageCompound(page), page);
1302	BUG_ON(!PageLocked(page));
1303	mapping = page->mapping;
1304	index = page->index;
1305	inode = mapping->host;
1306	info = SHMEM_I(inode);
1307	if (info->flags & VM_LOCKED)
1308		goto redirty;
1309	if (!total_swap_pages)
1310		goto redirty;
1311
1312	/*
1313	 * Our capabilities prevent regular writeback or sync from ever calling
1314	 * shmem_writepage; but a stacking filesystem might use ->writepage of
1315	 * its underlying filesystem, in which case tmpfs should write out to
1316	 * swap only in response to memory pressure, and not for the writeback
1317	 * threads or sync.
1318	 */
1319	if (!wbc->for_reclaim) {
1320		WARN_ON_ONCE(1);	/* Still happens? Tell us about it! */
1321		goto redirty;
1322	}
1323
1324	/*
1325	 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1326	 * value into swapfile.c, the only way we can correctly account for a
1327	 * fallocated page arriving here is now to initialize it and write it.
1328	 *
1329	 * That's okay for a page already fallocated earlier, but if we have
1330	 * not yet completed the fallocation, then (a) we want to keep track
1331	 * of this page in case we have to undo it, and (b) it may not be a
1332	 * good idea to continue anyway, once we're pushing into swap.  So
1333	 * reactivate the page, and let shmem_fallocate() quit when too many.
1334	 */
1335	if (!PageUptodate(page)) {
1336		if (inode->i_private) {
1337			struct shmem_falloc *shmem_falloc;
1338			spin_lock(&inode->i_lock);
1339			shmem_falloc = inode->i_private;
1340			if (shmem_falloc &&
1341			    !shmem_falloc->waitq &&
1342			    index >= shmem_falloc->start &&
1343			    index < shmem_falloc->next)
1344				shmem_falloc->nr_unswapped++;
1345			else
1346				shmem_falloc = NULL;
1347			spin_unlock(&inode->i_lock);
1348			if (shmem_falloc)
1349				goto redirty;
1350		}
1351		clear_highpage(page);
1352		flush_dcache_page(page);
1353		SetPageUptodate(page);
1354	}
1355
1356	swap = get_swap_page(page);
1357	if (!swap.val)
1358		goto redirty;
1359
1360	/*
1361	 * Add inode to shmem_unuse()'s list of swapped-out inodes,
1362	 * if it's not already there.  Do it now before the page is
1363	 * moved to swap cache, when its pagelock no longer protects
1364	 * the inode from eviction.  But don't unlock the mutex until
1365	 * we've incremented swapped, because shmem_unuse_inode() will
1366	 * prune a !swapped inode from the swaplist under this mutex.
1367	 */
1368	mutex_lock(&shmem_swaplist_mutex);
1369	if (list_empty(&info->swaplist))
1370		list_add(&info->swaplist, &shmem_swaplist);
1371
1372	if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
 
 
1373		spin_lock_irq(&info->lock);
1374		shmem_recalc_inode(inode);
1375		info->swapped++;
1376		spin_unlock_irq(&info->lock);
1377
1378		swap_shmem_alloc(swap);
1379		shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1380
1381		mutex_unlock(&shmem_swaplist_mutex);
1382		BUG_ON(page_mapped(page));
1383		swap_writepage(page, wbc);
1384		return 0;
1385	}
1386
1387	mutex_unlock(&shmem_swaplist_mutex);
1388	put_swap_page(page, swap);
1389redirty:
1390	set_page_dirty(page);
1391	if (wbc->for_reclaim)
1392		return AOP_WRITEPAGE_ACTIVATE;	/* Return with page locked */
1393	unlock_page(page);
1394	return 0;
1395}
1396
1397#if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
1398static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1399{
1400	char buffer[64];
1401
1402	if (!mpol || mpol->mode == MPOL_DEFAULT)
1403		return;		/* show nothing */
1404
1405	mpol_to_str(buffer, sizeof(buffer), mpol);
1406
1407	seq_printf(seq, ",mpol=%s", buffer);
1408}
1409
1410static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1411{
1412	struct mempolicy *mpol = NULL;
1413	if (sbinfo->mpol) {
1414		spin_lock(&sbinfo->stat_lock);	/* prevent replace/use races */
1415		mpol = sbinfo->mpol;
1416		mpol_get(mpol);
1417		spin_unlock(&sbinfo->stat_lock);
1418	}
1419	return mpol;
1420}
1421#else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1422static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1423{
1424}
1425static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1426{
1427	return NULL;
1428}
1429#endif /* CONFIG_NUMA && CONFIG_TMPFS */
1430#ifndef CONFIG_NUMA
1431#define vm_policy vm_private_data
1432#endif
1433
1434static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1435		struct shmem_inode_info *info, pgoff_t index)
1436{
1437	/* Create a pseudo vma that just contains the policy */
1438	vma_init(vma, NULL);
1439	/* Bias interleave by inode number to distribute better across nodes */
1440	vma->vm_pgoff = index + info->vfs_inode.i_ino;
1441	vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1442}
1443
1444static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1445{
1446	/* Drop reference taken by mpol_shared_policy_lookup() */
1447	mpol_cond_put(vma->vm_policy);
1448}
1449
1450static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1451			struct shmem_inode_info *info, pgoff_t index)
1452{
1453	struct vm_area_struct pvma;
1454	struct page *page;
1455	struct vm_fault vmf;
1456
1457	shmem_pseudo_vma_init(&pvma, info, index);
1458	vmf.vma = &pvma;
1459	vmf.address = 0;
1460	page = swap_cluster_readahead(swap, gfp, &vmf);
1461	shmem_pseudo_vma_destroy(&pvma);
1462
1463	return page;
1464}
1465
1466static struct page *shmem_alloc_hugepage(gfp_t gfp,
1467		struct shmem_inode_info *info, pgoff_t index)
1468{
1469	struct vm_area_struct pvma;
1470	struct address_space *mapping = info->vfs_inode.i_mapping;
1471	pgoff_t hindex;
1472	struct page *page;
1473
1474	if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1475		return NULL;
1476
1477	hindex = round_down(index, HPAGE_PMD_NR);
1478	if (xa_find(&mapping->i_pages, &hindex, hindex + HPAGE_PMD_NR - 1,
1479								XA_PRESENT))
1480		return NULL;
1481
1482	shmem_pseudo_vma_init(&pvma, info, hindex);
1483	page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1484			HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1485	shmem_pseudo_vma_destroy(&pvma);
1486	if (page)
1487		prep_transhuge_page(page);
 
 
1488	return page;
1489}
1490
1491static struct page *shmem_alloc_page(gfp_t gfp,
1492			struct shmem_inode_info *info, pgoff_t index)
1493{
1494	struct vm_area_struct pvma;
1495	struct page *page;
1496
1497	shmem_pseudo_vma_init(&pvma, info, index);
1498	page = alloc_page_vma(gfp, &pvma, 0);
1499	shmem_pseudo_vma_destroy(&pvma);
1500
1501	return page;
1502}
1503
1504static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1505		struct inode *inode,
1506		pgoff_t index, bool huge)
1507{
1508	struct shmem_inode_info *info = SHMEM_I(inode);
1509	struct page *page;
1510	int nr;
1511	int err = -ENOSPC;
1512
1513	if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
1514		huge = false;
1515	nr = huge ? HPAGE_PMD_NR : 1;
1516
1517	if (!shmem_inode_acct_block(inode, nr))
1518		goto failed;
1519
1520	if (huge)
1521		page = shmem_alloc_hugepage(gfp, info, index);
1522	else
1523		page = shmem_alloc_page(gfp, info, index);
1524	if (page) {
1525		__SetPageLocked(page);
1526		__SetPageSwapBacked(page);
1527		return page;
1528	}
1529
1530	err = -ENOMEM;
1531	shmem_inode_unacct_blocks(inode, nr);
1532failed:
1533	return ERR_PTR(err);
1534}
1535
1536/*
1537 * When a page is moved from swapcache to shmem filecache (either by the
1538 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1539 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1540 * ignorance of the mapping it belongs to.  If that mapping has special
1541 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1542 * we may need to copy to a suitable page before moving to filecache.
1543 *
1544 * In a future release, this may well be extended to respect cpuset and
1545 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1546 * but for now it is a simple matter of zone.
1547 */
1548static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1549{
1550	return page_zonenum(page) > gfp_zone(gfp);
1551}
1552
1553static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1554				struct shmem_inode_info *info, pgoff_t index)
1555{
1556	struct page *oldpage, *newpage;
1557	struct address_space *swap_mapping;
1558	swp_entry_t entry;
1559	pgoff_t swap_index;
1560	int error;
1561
1562	oldpage = *pagep;
1563	entry.val = page_private(oldpage);
1564	swap_index = swp_offset(entry);
1565	swap_mapping = page_mapping(oldpage);
1566
1567	/*
1568	 * We have arrived here because our zones are constrained, so don't
1569	 * limit chance of success by further cpuset and node constraints.
1570	 */
1571	gfp &= ~GFP_CONSTRAINT_MASK;
1572	newpage = shmem_alloc_page(gfp, info, index);
1573	if (!newpage)
1574		return -ENOMEM;
1575
1576	get_page(newpage);
1577	copy_highpage(newpage, oldpage);
1578	flush_dcache_page(newpage);
1579
1580	__SetPageLocked(newpage);
1581	__SetPageSwapBacked(newpage);
1582	SetPageUptodate(newpage);
1583	set_page_private(newpage, entry.val);
1584	SetPageSwapCache(newpage);
1585
1586	/*
1587	 * Our caller will very soon move newpage out of swapcache, but it's
1588	 * a nice clean interface for us to replace oldpage by newpage there.
1589	 */
1590	xa_lock_irq(&swap_mapping->i_pages);
1591	error = shmem_replace_entry(swap_mapping, swap_index, oldpage, newpage);
1592	if (!error) {
1593		__inc_node_page_state(newpage, NR_FILE_PAGES);
1594		__dec_node_page_state(oldpage, NR_FILE_PAGES);
 
1595	}
1596	xa_unlock_irq(&swap_mapping->i_pages);
1597
1598	if (unlikely(error)) {
1599		/*
1600		 * Is this possible?  I think not, now that our callers check
1601		 * both PageSwapCache and page_private after getting page lock;
1602		 * but be defensive.  Reverse old to newpage for clear and free.
1603		 */
1604		oldpage = newpage;
1605	} else {
1606		mem_cgroup_migrate(oldpage, newpage);
1607		lru_cache_add_anon(newpage);
1608		*pagep = newpage;
1609	}
1610
1611	ClearPageSwapCache(oldpage);
1612	set_page_private(oldpage, 0);
1613
1614	unlock_page(oldpage);
1615	put_page(oldpage);
1616	put_page(oldpage);
1617	return error;
1618}
1619
1620/*
1621 * Swap in the page pointed to by *pagep.
1622 * Caller has to make sure that *pagep contains a valid swapped page.
1623 * Returns 0 and the page in pagep if success. On failure, returns the
1624 * the error code and NULL in *pagep.
1625 */
1626static int shmem_swapin_page(struct inode *inode, pgoff_t index,
1627			     struct page **pagep, enum sgp_type sgp,
1628			     gfp_t gfp, struct vm_area_struct *vma,
1629			     vm_fault_t *fault_type)
1630{
1631	struct address_space *mapping = inode->i_mapping;
1632	struct shmem_inode_info *info = SHMEM_I(inode);
1633	struct mm_struct *charge_mm = vma ? vma->vm_mm : current->mm;
1634	struct mem_cgroup *memcg;
1635	struct page *page;
1636	swp_entry_t swap;
1637	int error;
1638
1639	VM_BUG_ON(!*pagep || !xa_is_value(*pagep));
1640	swap = radix_to_swp_entry(*pagep);
1641	*pagep = NULL;
1642
1643	/* Look it up and read it in.. */
1644	page = lookup_swap_cache(swap, NULL, 0);
1645	if (!page) {
1646		/* Or update major stats only when swapin succeeds?? */
1647		if (fault_type) {
1648			*fault_type |= VM_FAULT_MAJOR;
1649			count_vm_event(PGMAJFAULT);
1650			count_memcg_event_mm(charge_mm, PGMAJFAULT);
1651		}
1652		/* Here we actually start the io */
1653		page = shmem_swapin(swap, gfp, info, index);
1654		if (!page) {
1655			error = -ENOMEM;
1656			goto failed;
1657		}
1658	}
1659
1660	/* We have to do this with page locked to prevent races */
1661	lock_page(page);
1662	if (!PageSwapCache(page) || page_private(page) != swap.val ||
1663	    !shmem_confirm_swap(mapping, index, swap)) {
1664		error = -EEXIST;
1665		goto unlock;
1666	}
1667	if (!PageUptodate(page)) {
1668		error = -EIO;
1669		goto failed;
1670	}
1671	wait_on_page_writeback(page);
1672
1673	if (shmem_should_replace_page(page, gfp)) {
1674		error = shmem_replace_page(&page, gfp, info, index);
1675		if (error)
1676			goto failed;
1677	}
1678
1679	error = mem_cgroup_try_charge_delay(page, charge_mm, gfp, &memcg,
1680					    false);
1681	if (!error) {
1682		error = shmem_add_to_page_cache(page, mapping, index,
1683						swp_to_radix_entry(swap), gfp);
1684		/*
1685		 * We already confirmed swap under page lock, and make
1686		 * no memory allocation here, so usually no possibility
1687		 * of error; but free_swap_and_cache() only trylocks a
1688		 * page, so it is just possible that the entry has been
1689		 * truncated or holepunched since swap was confirmed.
1690		 * shmem_undo_range() will have done some of the
1691		 * unaccounting, now delete_from_swap_cache() will do
1692		 * the rest.
1693		 */
1694		if (error) {
1695			mem_cgroup_cancel_charge(page, memcg, false);
1696			delete_from_swap_cache(page);
1697		}
1698	}
1699	if (error)
1700		goto failed;
1701
1702	mem_cgroup_commit_charge(page, memcg, true, false);
1703
1704	spin_lock_irq(&info->lock);
1705	info->swapped--;
1706	shmem_recalc_inode(inode);
1707	spin_unlock_irq(&info->lock);
1708
1709	if (sgp == SGP_WRITE)
1710		mark_page_accessed(page);
1711
1712	delete_from_swap_cache(page);
1713	set_page_dirty(page);
1714	swap_free(swap);
1715
1716	*pagep = page;
1717	return 0;
1718failed:
1719	if (!shmem_confirm_swap(mapping, index, swap))
1720		error = -EEXIST;
1721unlock:
1722	if (page) {
1723		unlock_page(page);
1724		put_page(page);
1725	}
1726
1727	return error;
1728}
1729
1730/*
1731 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1732 *
1733 * If we allocate a new one we do not mark it dirty. That's up to the
1734 * vm. If we swap it in we mark it dirty since we also free the swap
1735 * entry since a page cannot live in both the swap and page cache.
1736 *
1737 * vmf and fault_type are only supplied by shmem_fault:
1738 * otherwise they are NULL.
1739 */
1740static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1741	struct page **pagep, enum sgp_type sgp, gfp_t gfp,
1742	struct vm_area_struct *vma, struct vm_fault *vmf,
1743			vm_fault_t *fault_type)
1744{
1745	struct address_space *mapping = inode->i_mapping;
1746	struct shmem_inode_info *info = SHMEM_I(inode);
1747	struct shmem_sb_info *sbinfo;
1748	struct mm_struct *charge_mm;
1749	struct mem_cgroup *memcg;
1750	struct page *page;
1751	enum sgp_type sgp_huge = sgp;
1752	pgoff_t hindex = index;
1753	int error;
1754	int once = 0;
1755	int alloced = 0;
1756
1757	if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
1758		return -EFBIG;
1759	if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1760		sgp = SGP_CACHE;
1761repeat:
1762	if (sgp <= SGP_CACHE &&
1763	    ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1764		return -EINVAL;
1765	}
1766
1767	sbinfo = SHMEM_SB(inode->i_sb);
1768	charge_mm = vma ? vma->vm_mm : current->mm;
1769
1770	page = find_lock_entry(mapping, index);
1771	if (xa_is_value(page)) {
1772		error = shmem_swapin_page(inode, index, &page,
1773					  sgp, gfp, vma, fault_type);
1774		if (error == -EEXIST)
1775			goto repeat;
1776
1777		*pagep = page;
1778		return error;
1779	}
1780
1781	if (page && sgp == SGP_WRITE)
1782		mark_page_accessed(page);
1783
1784	/* fallocated page? */
1785	if (page && !PageUptodate(page)) {
1786		if (sgp != SGP_READ)
1787			goto clear;
1788		unlock_page(page);
1789		put_page(page);
1790		page = NULL;
1791	}
1792	if (page || sgp == SGP_READ) {
1793		*pagep = page;
1794		return 0;
1795	}
1796
1797	/*
1798	 * Fast cache lookup did not find it:
1799	 * bring it back from swap or allocate.
1800	 */
1801
1802	if (vma && userfaultfd_missing(vma)) {
1803		*fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1804		return 0;
1805	}
1806
1807	/* shmem_symlink() */
1808	if (mapping->a_ops != &shmem_aops)
1809		goto alloc_nohuge;
1810	if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
1811		goto alloc_nohuge;
1812	if (shmem_huge == SHMEM_HUGE_FORCE)
1813		goto alloc_huge;
1814	switch (sbinfo->huge) {
 
 
 
1815		loff_t i_size;
1816		pgoff_t off;
1817	case SHMEM_HUGE_NEVER:
1818		goto alloc_nohuge;
1819	case SHMEM_HUGE_WITHIN_SIZE:
1820		off = round_up(index, HPAGE_PMD_NR);
1821		i_size = round_up(i_size_read(inode), PAGE_SIZE);
1822		if (i_size >= HPAGE_PMD_SIZE &&
1823		    i_size >> PAGE_SHIFT >= off)
1824			goto alloc_huge;
1825		/* fallthrough */
 
 
1826	case SHMEM_HUGE_ADVISE:
1827		if (sgp_huge == SGP_HUGE)
1828			goto alloc_huge;
1829		/* TODO: implement fadvise() hints */
1830		goto alloc_nohuge;
1831	}
1832
1833alloc_huge:
1834	page = shmem_alloc_and_acct_page(gfp, inode, index, true);
1835	if (IS_ERR(page)) {
1836alloc_nohuge:
1837		page = shmem_alloc_and_acct_page(gfp, inode,
1838						 index, false);
1839	}
1840	if (IS_ERR(page)) {
1841		int retry = 5;
1842
1843		error = PTR_ERR(page);
1844		page = NULL;
1845		if (error != -ENOSPC)
1846			goto unlock;
1847		/*
1848		 * Try to reclaim some space by splitting a huge page
1849		 * beyond i_size on the filesystem.
1850		 */
1851		while (retry--) {
1852			int ret;
1853
1854			ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1855			if (ret == SHRINK_STOP)
1856				break;
1857			if (ret)
1858				goto alloc_nohuge;
1859		}
1860		goto unlock;
1861	}
1862
1863	if (PageTransHuge(page))
1864		hindex = round_down(index, HPAGE_PMD_NR);
1865	else
1866		hindex = index;
1867
1868	if (sgp == SGP_WRITE)
1869		__SetPageReferenced(page);
1870
1871	error = mem_cgroup_try_charge_delay(page, charge_mm, gfp, &memcg,
1872					    PageTransHuge(page));
 
1873	if (error)
1874		goto unacct;
1875	error = shmem_add_to_page_cache(page, mapping, hindex,
1876					NULL, gfp & GFP_RECLAIM_MASK);
1877	if (error) {
1878		mem_cgroup_cancel_charge(page, memcg,
1879					 PageTransHuge(page));
1880		goto unacct;
1881	}
1882	mem_cgroup_commit_charge(page, memcg, false,
1883				 PageTransHuge(page));
1884	lru_cache_add_anon(page);
1885
1886	spin_lock_irq(&info->lock);
1887	info->alloced += compound_nr(page);
1888	inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
1889	shmem_recalc_inode(inode);
1890	spin_unlock_irq(&info->lock);
1891	alloced = true;
1892
1893	if (PageTransHuge(page) &&
1894	    DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1895			hindex + HPAGE_PMD_NR - 1) {
1896		/*
1897		 * Part of the huge page is beyond i_size: subject
1898		 * to shrink under memory pressure.
1899		 */
1900		spin_lock(&sbinfo->shrinklist_lock);
1901		/*
1902		 * _careful to defend against unlocked access to
1903		 * ->shrink_list in shmem_unused_huge_shrink()
1904		 */
1905		if (list_empty_careful(&info->shrinklist)) {
1906			list_add_tail(&info->shrinklist,
1907				      &sbinfo->shrinklist);
1908			sbinfo->shrinklist_len++;
1909		}
1910		spin_unlock(&sbinfo->shrinklist_lock);
1911	}
1912
1913	/*
1914	 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1915	 */
1916	if (sgp == SGP_FALLOC)
1917		sgp = SGP_WRITE;
1918clear:
1919	/*
1920	 * Let SGP_WRITE caller clear ends if write does not fill page;
1921	 * but SGP_FALLOC on a page fallocated earlier must initialize
1922	 * it now, lest undo on failure cancel our earlier guarantee.
1923	 */
1924	if (sgp != SGP_WRITE && !PageUptodate(page)) {
1925		struct page *head = compound_head(page);
1926		int i;
1927
1928		for (i = 0; i < compound_nr(head); i++) {
1929			clear_highpage(head + i);
1930			flush_dcache_page(head + i);
1931		}
1932		SetPageUptodate(head);
1933	}
1934
1935	/* Perhaps the file has been truncated since we checked */
1936	if (sgp <= SGP_CACHE &&
1937	    ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
1938		if (alloced) {
1939			ClearPageDirty(page);
1940			delete_from_page_cache(page);
1941			spin_lock_irq(&info->lock);
1942			shmem_recalc_inode(inode);
1943			spin_unlock_irq(&info->lock);
1944		}
1945		error = -EINVAL;
1946		goto unlock;
1947	}
1948	*pagep = page + index - hindex;
1949	return 0;
1950
1951	/*
1952	 * Error recovery.
1953	 */
1954unacct:
1955	shmem_inode_unacct_blocks(inode, compound_nr(page));
1956
1957	if (PageTransHuge(page)) {
1958		unlock_page(page);
1959		put_page(page);
1960		goto alloc_nohuge;
1961	}
1962unlock:
1963	if (page) {
1964		unlock_page(page);
1965		put_page(page);
1966	}
1967	if (error == -ENOSPC && !once++) {
1968		spin_lock_irq(&info->lock);
1969		shmem_recalc_inode(inode);
1970		spin_unlock_irq(&info->lock);
1971		goto repeat;
1972	}
1973	if (error == -EEXIST)
1974		goto repeat;
1975	return error;
1976}
1977
1978/*
1979 * This is like autoremove_wake_function, but it removes the wait queue
1980 * entry unconditionally - even if something else had already woken the
1981 * target.
1982 */
1983static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
1984{
1985	int ret = default_wake_function(wait, mode, sync, key);
1986	list_del_init(&wait->entry);
1987	return ret;
1988}
1989
1990static vm_fault_t shmem_fault(struct vm_fault *vmf)
1991{
1992	struct vm_area_struct *vma = vmf->vma;
1993	struct inode *inode = file_inode(vma->vm_file);
1994	gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
1995	enum sgp_type sgp;
1996	int err;
1997	vm_fault_t ret = VM_FAULT_LOCKED;
1998
1999	/*
2000	 * Trinity finds that probing a hole which tmpfs is punching can
2001	 * prevent the hole-punch from ever completing: which in turn
2002	 * locks writers out with its hold on i_mutex.  So refrain from
2003	 * faulting pages into the hole while it's being punched.  Although
2004	 * shmem_undo_range() does remove the additions, it may be unable to
2005	 * keep up, as each new page needs its own unmap_mapping_range() call,
2006	 * and the i_mmap tree grows ever slower to scan if new vmas are added.
2007	 *
2008	 * It does not matter if we sometimes reach this check just before the
2009	 * hole-punch begins, so that one fault then races with the punch:
2010	 * we just need to make racing faults a rare case.
2011	 *
2012	 * The implementation below would be much simpler if we just used a
2013	 * standard mutex or completion: but we cannot take i_mutex in fault,
2014	 * and bloating every shmem inode for this unlikely case would be sad.
2015	 */
2016	if (unlikely(inode->i_private)) {
2017		struct shmem_falloc *shmem_falloc;
2018
2019		spin_lock(&inode->i_lock);
2020		shmem_falloc = inode->i_private;
2021		if (shmem_falloc &&
2022		    shmem_falloc->waitq &&
2023		    vmf->pgoff >= shmem_falloc->start &&
2024		    vmf->pgoff < shmem_falloc->next) {
 
2025			wait_queue_head_t *shmem_falloc_waitq;
2026			DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
2027
2028			ret = VM_FAULT_NOPAGE;
2029			if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
2030			   !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
2031				/* It's polite to up mmap_sem if we can */
2032				up_read(&vma->vm_mm->mmap_sem);
2033				ret = VM_FAULT_RETRY;
2034			}
2035
2036			shmem_falloc_waitq = shmem_falloc->waitq;
2037			prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
2038					TASK_UNINTERRUPTIBLE);
2039			spin_unlock(&inode->i_lock);
2040			schedule();
2041
2042			/*
2043			 * shmem_falloc_waitq points into the shmem_fallocate()
2044			 * stack of the hole-punching task: shmem_falloc_waitq
2045			 * is usually invalid by the time we reach here, but
2046			 * finish_wait() does not dereference it in that case;
2047			 * though i_lock needed lest racing with wake_up_all().
2048			 */
2049			spin_lock(&inode->i_lock);
2050			finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
2051			spin_unlock(&inode->i_lock);
 
 
 
2052			return ret;
2053		}
2054		spin_unlock(&inode->i_lock);
2055	}
2056
2057	sgp = SGP_CACHE;
2058
2059	if ((vma->vm_flags & VM_NOHUGEPAGE) ||
2060	    test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
2061		sgp = SGP_NOHUGE;
2062	else if (vma->vm_flags & VM_HUGEPAGE)
2063		sgp = SGP_HUGE;
2064
2065	err = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
2066				  gfp, vma, vmf, &ret);
2067	if (err)
2068		return vmf_error(err);
2069	return ret;
2070}
2071
2072unsigned long shmem_get_unmapped_area(struct file *file,
2073				      unsigned long uaddr, unsigned long len,
2074				      unsigned long pgoff, unsigned long flags)
2075{
2076	unsigned long (*get_area)(struct file *,
2077		unsigned long, unsigned long, unsigned long, unsigned long);
2078	unsigned long addr;
2079	unsigned long offset;
2080	unsigned long inflated_len;
2081	unsigned long inflated_addr;
2082	unsigned long inflated_offset;
2083
2084	if (len > TASK_SIZE)
2085		return -ENOMEM;
2086
2087	get_area = current->mm->get_unmapped_area;
2088	addr = get_area(file, uaddr, len, pgoff, flags);
2089
2090	if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
2091		return addr;
2092	if (IS_ERR_VALUE(addr))
2093		return addr;
2094	if (addr & ~PAGE_MASK)
2095		return addr;
2096	if (addr > TASK_SIZE - len)
2097		return addr;
2098
2099	if (shmem_huge == SHMEM_HUGE_DENY)
2100		return addr;
2101	if (len < HPAGE_PMD_SIZE)
2102		return addr;
2103	if (flags & MAP_FIXED)
2104		return addr;
2105	/*
2106	 * Our priority is to support MAP_SHARED mapped hugely;
2107	 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2108	 * But if caller specified an address hint, respect that as before.
 
2109	 */
2110	if (uaddr)
2111		return addr;
2112
2113	if (shmem_huge != SHMEM_HUGE_FORCE) {
2114		struct super_block *sb;
2115
2116		if (file) {
2117			VM_BUG_ON(file->f_op != &shmem_file_operations);
2118			sb = file_inode(file)->i_sb;
2119		} else {
2120			/*
2121			 * Called directly from mm/mmap.c, or drivers/char/mem.c
2122			 * for "/dev/zero", to create a shared anonymous object.
2123			 */
2124			if (IS_ERR(shm_mnt))
2125				return addr;
2126			sb = shm_mnt->mnt_sb;
2127		}
2128		if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
2129			return addr;
2130	}
2131
2132	offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2133	if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2134		return addr;
2135	if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2136		return addr;
2137
2138	inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2139	if (inflated_len > TASK_SIZE)
2140		return addr;
2141	if (inflated_len < len)
2142		return addr;
2143
2144	inflated_addr = get_area(NULL, 0, inflated_len, 0, flags);
2145	if (IS_ERR_VALUE(inflated_addr))
2146		return addr;
2147	if (inflated_addr & ~PAGE_MASK)
2148		return addr;
2149
2150	inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2151	inflated_addr += offset - inflated_offset;
2152	if (inflated_offset > offset)
2153		inflated_addr += HPAGE_PMD_SIZE;
2154
2155	if (inflated_addr > TASK_SIZE - len)
2156		return addr;
2157	return inflated_addr;
2158}
2159
2160#ifdef CONFIG_NUMA
2161static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
2162{
2163	struct inode *inode = file_inode(vma->vm_file);
2164	return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
2165}
2166
2167static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2168					  unsigned long addr)
2169{
2170	struct inode *inode = file_inode(vma->vm_file);
2171	pgoff_t index;
2172
2173	index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2174	return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
2175}
2176#endif
2177
2178int shmem_lock(struct file *file, int lock, struct user_struct *user)
2179{
2180	struct inode *inode = file_inode(file);
2181	struct shmem_inode_info *info = SHMEM_I(inode);
2182	int retval = -ENOMEM;
2183
2184	spin_lock_irq(&info->lock);
 
 
 
 
2185	if (lock && !(info->flags & VM_LOCKED)) {
2186		if (!user_shm_lock(inode->i_size, user))
2187			goto out_nomem;
2188		info->flags |= VM_LOCKED;
2189		mapping_set_unevictable(file->f_mapping);
2190	}
2191	if (!lock && (info->flags & VM_LOCKED) && user) {
2192		user_shm_unlock(inode->i_size, user);
2193		info->flags &= ~VM_LOCKED;
2194		mapping_clear_unevictable(file->f_mapping);
2195	}
2196	retval = 0;
2197
2198out_nomem:
2199	spin_unlock_irq(&info->lock);
2200	return retval;
2201}
2202
2203static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
2204{
2205	struct shmem_inode_info *info = SHMEM_I(file_inode(file));
2206
2207	if (info->seals & F_SEAL_FUTURE_WRITE) {
2208		/*
2209		 * New PROT_WRITE and MAP_SHARED mmaps are not allowed when
2210		 * "future write" seal active.
2211		 */
2212		if ((vma->vm_flags & VM_SHARED) && (vma->vm_flags & VM_WRITE))
2213			return -EPERM;
2214
2215		/*
2216		 * Since the F_SEAL_FUTURE_WRITE seals allow for a MAP_SHARED
2217		 * read-only mapping, take care to not allow mprotect to revert
2218		 * protections.
 
 
2219		 */
2220		vma->vm_flags &= ~(VM_MAYWRITE);
 
2221	}
2222
2223	file_accessed(file);
2224	vma->vm_ops = &shmem_vm_ops;
2225	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
2226			((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2227			(vma->vm_end & HPAGE_PMD_MASK)) {
2228		khugepaged_enter(vma, vma->vm_flags);
2229	}
2230	return 0;
2231}
2232
2233static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
2234				     umode_t mode, dev_t dev, unsigned long flags)
2235{
2236	struct inode *inode;
2237	struct shmem_inode_info *info;
2238	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 
2239
2240	if (shmem_reserve_inode(sb))
2241		return NULL;
2242
2243	inode = new_inode(sb);
2244	if (inode) {
2245		inode->i_ino = get_next_ino();
2246		inode_init_owner(inode, dir, mode);
2247		inode->i_blocks = 0;
2248		inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
2249		inode->i_generation = prandom_u32();
2250		info = SHMEM_I(inode);
2251		memset(info, 0, (char *)inode - (char *)info);
2252		spin_lock_init(&info->lock);
2253		atomic_set(&info->stop_eviction, 0);
2254		info->seals = F_SEAL_SEAL;
2255		info->flags = flags & VM_NORESERVE;
2256		INIT_LIST_HEAD(&info->shrinklist);
2257		INIT_LIST_HEAD(&info->swaplist);
2258		simple_xattrs_init(&info->xattrs);
2259		cache_no_acl(inode);
2260
2261		switch (mode & S_IFMT) {
2262		default:
2263			inode->i_op = &shmem_special_inode_operations;
2264			init_special_inode(inode, mode, dev);
2265			break;
2266		case S_IFREG:
2267			inode->i_mapping->a_ops = &shmem_aops;
2268			inode->i_op = &shmem_inode_operations;
2269			inode->i_fop = &shmem_file_operations;
2270			mpol_shared_policy_init(&info->policy,
2271						 shmem_get_sbmpol(sbinfo));
2272			break;
2273		case S_IFDIR:
2274			inc_nlink(inode);
2275			/* Some things misbehave if size == 0 on a directory */
2276			inode->i_size = 2 * BOGO_DIRENT_SIZE;
2277			inode->i_op = &shmem_dir_inode_operations;
2278			inode->i_fop = &simple_dir_operations;
2279			break;
2280		case S_IFLNK:
2281			/*
2282			 * Must not load anything in the rbtree,
2283			 * mpol_free_shared_policy will not be called.
2284			 */
2285			mpol_shared_policy_init(&info->policy, NULL);
2286			break;
2287		}
2288
2289		lockdep_annotate_inode_mutex_key(inode);
2290	} else
2291		shmem_free_inode(sb);
2292	return inode;
2293}
2294
2295bool shmem_mapping(struct address_space *mapping)
2296{
2297	return mapping->a_ops == &shmem_aops;
2298}
2299
2300static int shmem_mfill_atomic_pte(struct mm_struct *dst_mm,
2301				  pmd_t *dst_pmd,
2302				  struct vm_area_struct *dst_vma,
2303				  unsigned long dst_addr,
2304				  unsigned long src_addr,
2305				  bool zeropage,
2306				  struct page **pagep)
2307{
2308	struct inode *inode = file_inode(dst_vma->vm_file);
2309	struct shmem_inode_info *info = SHMEM_I(inode);
2310	struct address_space *mapping = inode->i_mapping;
2311	gfp_t gfp = mapping_gfp_mask(mapping);
2312	pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2313	struct mem_cgroup *memcg;
2314	spinlock_t *ptl;
2315	void *page_kaddr;
2316	struct page *page;
2317	pte_t _dst_pte, *dst_pte;
2318	int ret;
2319	pgoff_t offset, max_off;
2320
2321	ret = -ENOMEM;
2322	if (!shmem_inode_acct_block(inode, 1))
2323		goto out;
2324
2325	if (!*pagep) {
2326		page = shmem_alloc_page(gfp, info, pgoff);
2327		if (!page)
2328			goto out_unacct_blocks;
2329
2330		if (!zeropage) {	/* mcopy_atomic */
2331			page_kaddr = kmap_atomic(page);
2332			ret = copy_from_user(page_kaddr,
2333					     (const void __user *)src_addr,
2334					     PAGE_SIZE);
2335			kunmap_atomic(page_kaddr);
2336
2337			/* fallback to copy_from_user outside mmap_sem */
2338			if (unlikely(ret)) {
2339				*pagep = page;
2340				shmem_inode_unacct_blocks(inode, 1);
2341				/* don't free the page */
2342				return -ENOENT;
2343			}
2344		} else {		/* mfill_zeropage_atomic */
2345			clear_highpage(page);
2346		}
2347	} else {
2348		page = *pagep;
2349		*pagep = NULL;
2350	}
2351
2352	VM_BUG_ON(PageLocked(page) || PageSwapBacked(page));
2353	__SetPageLocked(page);
2354	__SetPageSwapBacked(page);
2355	__SetPageUptodate(page);
2356
2357	ret = -EFAULT;
2358	offset = linear_page_index(dst_vma, dst_addr);
2359	max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2360	if (unlikely(offset >= max_off))
2361		goto out_release;
2362
2363	ret = mem_cgroup_try_charge_delay(page, dst_mm, gfp, &memcg, false);
 
2364	if (ret)
2365		goto out_release;
2366
2367	ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL,
2368						gfp & GFP_RECLAIM_MASK);
2369	if (ret)
2370		goto out_release_uncharge;
2371
2372	mem_cgroup_commit_charge(page, memcg, false, false);
2373
2374	_dst_pte = mk_pte(page, dst_vma->vm_page_prot);
2375	if (dst_vma->vm_flags & VM_WRITE)
2376		_dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte));
2377	else {
2378		/*
2379		 * We don't set the pte dirty if the vma has no
2380		 * VM_WRITE permission, so mark the page dirty or it
2381		 * could be freed from under us. We could do it
2382		 * unconditionally before unlock_page(), but doing it
2383		 * only if VM_WRITE is not set is faster.
2384		 */
2385		set_page_dirty(page);
2386	}
2387
2388	dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);
2389
2390	ret = -EFAULT;
2391	max_off = DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE);
2392	if (unlikely(offset >= max_off))
2393		goto out_release_uncharge_unlock;
2394
2395	ret = -EEXIST;
2396	if (!pte_none(*dst_pte))
2397		goto out_release_uncharge_unlock;
2398
2399	lru_cache_add_anon(page);
2400
2401	spin_lock(&info->lock);
2402	info->alloced++;
2403	inode->i_blocks += BLOCKS_PER_PAGE;
2404	shmem_recalc_inode(inode);
2405	spin_unlock(&info->lock);
2406
2407	inc_mm_counter(dst_mm, mm_counter_file(page));
2408	page_add_file_rmap(page, false);
2409	set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
2410
2411	/* No need to invalidate - it was non-present before */
2412	update_mmu_cache(dst_vma, dst_addr, dst_pte);
2413	pte_unmap_unlock(dst_pte, ptl);
2414	unlock_page(page);
2415	ret = 0;
2416out:
2417	return ret;
2418out_release_uncharge_unlock:
2419	pte_unmap_unlock(dst_pte, ptl);
2420	ClearPageDirty(page);
2421	delete_from_page_cache(page);
2422out_release_uncharge:
2423	mem_cgroup_cancel_charge(page, memcg, false);
2424out_release:
2425	unlock_page(page);
2426	put_page(page);
2427out_unacct_blocks:
2428	shmem_inode_unacct_blocks(inode, 1);
2429	goto out;
2430}
2431
2432int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm,
2433			   pmd_t *dst_pmd,
2434			   struct vm_area_struct *dst_vma,
2435			   unsigned long dst_addr,
2436			   unsigned long src_addr,
2437			   struct page **pagep)
2438{
2439	return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2440				      dst_addr, src_addr, false, pagep);
2441}
2442
2443int shmem_mfill_zeropage_pte(struct mm_struct *dst_mm,
2444			     pmd_t *dst_pmd,
2445			     struct vm_area_struct *dst_vma,
2446			     unsigned long dst_addr)
2447{
2448	struct page *page = NULL;
2449
2450	return shmem_mfill_atomic_pte(dst_mm, dst_pmd, dst_vma,
2451				      dst_addr, 0, true, &page);
2452}
2453
2454#ifdef CONFIG_TMPFS
2455static const struct inode_operations shmem_symlink_inode_operations;
2456static const struct inode_operations shmem_short_symlink_operations;
2457
2458#ifdef CONFIG_TMPFS_XATTR
2459static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2460#else
2461#define shmem_initxattrs NULL
2462#endif
2463
2464static int
2465shmem_write_begin(struct file *file, struct address_space *mapping,
2466			loff_t pos, unsigned len, unsigned flags,
2467			struct page **pagep, void **fsdata)
2468{
2469	struct inode *inode = mapping->host;
2470	struct shmem_inode_info *info = SHMEM_I(inode);
2471	pgoff_t index = pos >> PAGE_SHIFT;
2472
2473	/* i_mutex is held by caller */
2474	if (unlikely(info->seals & (F_SEAL_GROW |
2475				   F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
2476		if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))
2477			return -EPERM;
2478		if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2479			return -EPERM;
2480	}
2481
2482	return shmem_getpage(inode, index, pagep, SGP_WRITE);
2483}
2484
2485static int
2486shmem_write_end(struct file *file, struct address_space *mapping,
2487			loff_t pos, unsigned len, unsigned copied,
2488			struct page *page, void *fsdata)
2489{
2490	struct inode *inode = mapping->host;
2491
2492	if (pos + copied > inode->i_size)
2493		i_size_write(inode, pos + copied);
2494
2495	if (!PageUptodate(page)) {
2496		struct page *head = compound_head(page);
2497		if (PageTransCompound(page)) {
2498			int i;
2499
2500			for (i = 0; i < HPAGE_PMD_NR; i++) {
2501				if (head + i == page)
2502					continue;
2503				clear_highpage(head + i);
2504				flush_dcache_page(head + i);
2505			}
2506		}
2507		if (copied < PAGE_SIZE) {
2508			unsigned from = pos & (PAGE_SIZE - 1);
2509			zero_user_segments(page, 0, from,
2510					from + copied, PAGE_SIZE);
2511		}
2512		SetPageUptodate(head);
2513	}
2514	set_page_dirty(page);
2515	unlock_page(page);
2516	put_page(page);
2517
2518	return copied;
2519}
2520
2521static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
2522{
2523	struct file *file = iocb->ki_filp;
2524	struct inode *inode = file_inode(file);
2525	struct address_space *mapping = inode->i_mapping;
2526	pgoff_t index;
2527	unsigned long offset;
2528	enum sgp_type sgp = SGP_READ;
2529	int error = 0;
2530	ssize_t retval = 0;
2531	loff_t *ppos = &iocb->ki_pos;
2532
2533	/*
2534	 * Might this read be for a stacking filesystem?  Then when reading
2535	 * holes of a sparse file, we actually need to allocate those pages,
2536	 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2537	 */
2538	if (!iter_is_iovec(to))
2539		sgp = SGP_CACHE;
2540
2541	index = *ppos >> PAGE_SHIFT;
2542	offset = *ppos & ~PAGE_MASK;
2543
2544	for (;;) {
2545		struct page *page = NULL;
2546		pgoff_t end_index;
2547		unsigned long nr, ret;
2548		loff_t i_size = i_size_read(inode);
2549
2550		end_index = i_size >> PAGE_SHIFT;
2551		if (index > end_index)
2552			break;
2553		if (index == end_index) {
2554			nr = i_size & ~PAGE_MASK;
2555			if (nr <= offset)
2556				break;
2557		}
2558
2559		error = shmem_getpage(inode, index, &page, sgp);
2560		if (error) {
2561			if (error == -EINVAL)
2562				error = 0;
2563			break;
2564		}
2565		if (page) {
2566			if (sgp == SGP_CACHE)
2567				set_page_dirty(page);
2568			unlock_page(page);
2569		}
2570
2571		/*
2572		 * We must evaluate after, since reads (unlike writes)
2573		 * are called without i_mutex protection against truncate
2574		 */
2575		nr = PAGE_SIZE;
2576		i_size = i_size_read(inode);
2577		end_index = i_size >> PAGE_SHIFT;
2578		if (index == end_index) {
2579			nr = i_size & ~PAGE_MASK;
2580			if (nr <= offset) {
2581				if (page)
2582					put_page(page);
2583				break;
2584			}
2585		}
2586		nr -= offset;
2587
2588		if (page) {
2589			/*
2590			 * If users can be writing to this page using arbitrary
2591			 * virtual addresses, take care about potential aliasing
2592			 * before reading the page on the kernel side.
2593			 */
2594			if (mapping_writably_mapped(mapping))
2595				flush_dcache_page(page);
2596			/*
2597			 * Mark the page accessed if we read the beginning.
2598			 */
2599			if (!offset)
2600				mark_page_accessed(page);
2601		} else {
2602			page = ZERO_PAGE(0);
2603			get_page(page);
2604		}
2605
2606		/*
2607		 * Ok, we have the page, and it's up-to-date, so
2608		 * now we can copy it to user space...
2609		 */
2610		ret = copy_page_to_iter(page, offset, nr, to);
2611		retval += ret;
2612		offset += ret;
2613		index += offset >> PAGE_SHIFT;
2614		offset &= ~PAGE_MASK;
2615
2616		put_page(page);
2617		if (!iov_iter_count(to))
2618			break;
2619		if (ret < nr) {
2620			error = -EFAULT;
2621			break;
2622		}
2623		cond_resched();
2624	}
2625
2626	*ppos = ((loff_t) index << PAGE_SHIFT) + offset;
2627	file_accessed(file);
2628	return retval ? retval : error;
2629}
2630
2631/*
2632 * llseek SEEK_DATA or SEEK_HOLE through the page cache.
2633 */
2634static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
2635				    pgoff_t index, pgoff_t end, int whence)
2636{
2637	struct page *page;
2638	struct pagevec pvec;
2639	pgoff_t indices[PAGEVEC_SIZE];
2640	bool done = false;
2641	int i;
2642
2643	pagevec_init(&pvec);
2644	pvec.nr = 1;		/* start small: we may be there already */
2645	while (!done) {
2646		pvec.nr = find_get_entries(mapping, index,
2647					pvec.nr, pvec.pages, indices);
2648		if (!pvec.nr) {
2649			if (whence == SEEK_DATA)
2650				index = end;
2651			break;
2652		}
2653		for (i = 0; i < pvec.nr; i++, index++) {
2654			if (index < indices[i]) {
2655				if (whence == SEEK_HOLE) {
2656					done = true;
2657					break;
2658				}
2659				index = indices[i];
2660			}
2661			page = pvec.pages[i];
2662			if (page && !xa_is_value(page)) {
2663				if (!PageUptodate(page))
2664					page = NULL;
2665			}
2666			if (index >= end ||
2667			    (page && whence == SEEK_DATA) ||
2668			    (!page && whence == SEEK_HOLE)) {
2669				done = true;
2670				break;
2671			}
2672		}
2673		pagevec_remove_exceptionals(&pvec);
2674		pagevec_release(&pvec);
2675		pvec.nr = PAGEVEC_SIZE;
2676		cond_resched();
2677	}
2678	return index;
2679}
2680
2681static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
2682{
2683	struct address_space *mapping = file->f_mapping;
2684	struct inode *inode = mapping->host;
2685	pgoff_t start, end;
2686	loff_t new_offset;
2687
2688	if (whence != SEEK_DATA && whence != SEEK_HOLE)
2689		return generic_file_llseek_size(file, offset, whence,
2690					MAX_LFS_FILESIZE, i_size_read(inode));
2691	inode_lock(inode);
2692	/* We're holding i_mutex so we can access i_size directly */
2693
2694	if (offset < 0 || offset >= inode->i_size)
2695		offset = -ENXIO;
2696	else {
2697		start = offset >> PAGE_SHIFT;
2698		end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
2699		new_offset = shmem_seek_hole_data(mapping, start, end, whence);
2700		new_offset <<= PAGE_SHIFT;
2701		if (new_offset > offset) {
2702			if (new_offset < inode->i_size)
2703				offset = new_offset;
2704			else if (whence == SEEK_DATA)
2705				offset = -ENXIO;
2706			else
2707				offset = inode->i_size;
2708		}
2709	}
2710
2711	if (offset >= 0)
2712		offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
2713	inode_unlock(inode);
2714	return offset;
2715}
2716
2717static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2718							 loff_t len)
2719{
2720	struct inode *inode = file_inode(file);
2721	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2722	struct shmem_inode_info *info = SHMEM_I(inode);
2723	struct shmem_falloc shmem_falloc;
2724	pgoff_t start, index, end;
2725	int error;
2726
2727	if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2728		return -EOPNOTSUPP;
2729
2730	inode_lock(inode);
2731
2732	if (mode & FALLOC_FL_PUNCH_HOLE) {
2733		struct address_space *mapping = file->f_mapping;
2734		loff_t unmap_start = round_up(offset, PAGE_SIZE);
2735		loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
2736		DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
2737
2738		/* protected by i_mutex */
2739		if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
2740			error = -EPERM;
2741			goto out;
2742		}
2743
2744		shmem_falloc.waitq = &shmem_falloc_waitq;
2745		shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2746		shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2747		spin_lock(&inode->i_lock);
2748		inode->i_private = &shmem_falloc;
2749		spin_unlock(&inode->i_lock);
2750
2751		if ((u64)unmap_end > (u64)unmap_start)
2752			unmap_mapping_range(mapping, unmap_start,
2753					    1 + unmap_end - unmap_start, 0);
2754		shmem_truncate_range(inode, offset, offset + len - 1);
2755		/* No need to unmap again: hole-punching leaves COWed pages */
2756
2757		spin_lock(&inode->i_lock);
2758		inode->i_private = NULL;
2759		wake_up_all(&shmem_falloc_waitq);
2760		WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
2761		spin_unlock(&inode->i_lock);
2762		error = 0;
2763		goto out;
2764	}
2765
2766	/* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2767	error = inode_newsize_ok(inode, offset + len);
2768	if (error)
2769		goto out;
2770
2771	if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2772		error = -EPERM;
2773		goto out;
2774	}
2775
2776	start = offset >> PAGE_SHIFT;
2777	end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
2778	/* Try to avoid a swapstorm if len is impossible to satisfy */
2779	if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2780		error = -ENOSPC;
2781		goto out;
2782	}
2783
2784	shmem_falloc.waitq = NULL;
2785	shmem_falloc.start = start;
2786	shmem_falloc.next  = start;
2787	shmem_falloc.nr_falloced = 0;
2788	shmem_falloc.nr_unswapped = 0;
2789	spin_lock(&inode->i_lock);
2790	inode->i_private = &shmem_falloc;
2791	spin_unlock(&inode->i_lock);
2792
2793	for (index = start; index < end; index++) {
2794		struct page *page;
2795
2796		/*
2797		 * Good, the fallocate(2) manpage permits EINTR: we may have
2798		 * been interrupted because we are using up too much memory.
2799		 */
2800		if (signal_pending(current))
2801			error = -EINTR;
2802		else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2803			error = -ENOMEM;
2804		else
2805			error = shmem_getpage(inode, index, &page, SGP_FALLOC);
2806		if (error) {
2807			/* Remove the !PageUptodate pages we added */
2808			if (index > start) {
2809				shmem_undo_range(inode,
2810				    (loff_t)start << PAGE_SHIFT,
2811				    ((loff_t)index << PAGE_SHIFT) - 1, true);
2812			}
2813			goto undone;
2814		}
2815
2816		/*
2817		 * Inform shmem_writepage() how far we have reached.
2818		 * No need for lock or barrier: we have the page lock.
2819		 */
2820		shmem_falloc.next++;
2821		if (!PageUptodate(page))
2822			shmem_falloc.nr_falloced++;
2823
2824		/*
2825		 * If !PageUptodate, leave it that way so that freeable pages
2826		 * can be recognized if we need to rollback on error later.
2827		 * But set_page_dirty so that memory pressure will swap rather
2828		 * than free the pages we are allocating (and SGP_CACHE pages
2829		 * might still be clean: we now need to mark those dirty too).
2830		 */
2831		set_page_dirty(page);
2832		unlock_page(page);
2833		put_page(page);
2834		cond_resched();
2835	}
2836
2837	if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2838		i_size_write(inode, offset + len);
2839	inode->i_ctime = current_time(inode);
2840undone:
2841	spin_lock(&inode->i_lock);
2842	inode->i_private = NULL;
2843	spin_unlock(&inode->i_lock);
2844out:
2845	inode_unlock(inode);
2846	return error;
2847}
2848
2849static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
2850{
2851	struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
2852
2853	buf->f_type = TMPFS_MAGIC;
2854	buf->f_bsize = PAGE_SIZE;
2855	buf->f_namelen = NAME_MAX;
2856	if (sbinfo->max_blocks) {
2857		buf->f_blocks = sbinfo->max_blocks;
2858		buf->f_bavail =
2859		buf->f_bfree  = sbinfo->max_blocks -
2860				percpu_counter_sum(&sbinfo->used_blocks);
2861	}
2862	if (sbinfo->max_inodes) {
2863		buf->f_files = sbinfo->max_inodes;
2864		buf->f_ffree = sbinfo->free_inodes;
2865	}
2866	/* else leave those fields 0 like simple_statfs */
2867	return 0;
2868}
2869
2870/*
2871 * File creation. Allocate an inode, and we're done..
2872 */
2873static int
2874shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
2875{
2876	struct inode *inode;
2877	int error = -ENOSPC;
2878
2879	inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
2880	if (inode) {
2881		error = simple_acl_create(dir, inode);
2882		if (error)
2883			goto out_iput;
2884		error = security_inode_init_security(inode, dir,
2885						     &dentry->d_name,
2886						     shmem_initxattrs, NULL);
2887		if (error && error != -EOPNOTSUPP)
2888			goto out_iput;
2889
2890		error = 0;
2891		dir->i_size += BOGO_DIRENT_SIZE;
2892		dir->i_ctime = dir->i_mtime = current_time(dir);
2893		d_instantiate(dentry, inode);
2894		dget(dentry); /* Extra count - pin the dentry in core */
2895	}
2896	return error;
2897out_iput:
2898	iput(inode);
2899	return error;
2900}
2901
2902static int
2903shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2904{
2905	struct inode *inode;
2906	int error = -ENOSPC;
2907
2908	inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2909	if (inode) {
2910		error = security_inode_init_security(inode, dir,
2911						     NULL,
2912						     shmem_initxattrs, NULL);
2913		if (error && error != -EOPNOTSUPP)
2914			goto out_iput;
2915		error = simple_acl_create(dir, inode);
2916		if (error)
2917			goto out_iput;
2918		d_tmpfile(dentry, inode);
2919	}
2920	return error;
2921out_iput:
2922	iput(inode);
2923	return error;
2924}
2925
2926static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
2927{
2928	int error;
2929
2930	if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2931		return error;
2932	inc_nlink(dir);
2933	return 0;
2934}
2935
2936static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
2937		bool excl)
2938{
2939	return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2940}
2941
2942/*
2943 * Link a file..
2944 */
2945static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2946{
2947	struct inode *inode = d_inode(old_dentry);
2948	int ret = 0;
2949
2950	/*
2951	 * No ordinary (disk based) filesystem counts links as inodes;
2952	 * but each new link needs a new dentry, pinning lowmem, and
2953	 * tmpfs dentries cannot be pruned until they are unlinked.
2954	 * But if an O_TMPFILE file is linked into the tmpfs, the
2955	 * first link must skip that, to get the accounting right.
2956	 */
2957	if (inode->i_nlink) {
2958		ret = shmem_reserve_inode(inode->i_sb);
2959		if (ret)
2960			goto out;
2961	}
2962
2963	dir->i_size += BOGO_DIRENT_SIZE;
2964	inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2965	inc_nlink(inode);
2966	ihold(inode);	/* New dentry reference */
2967	dget(dentry);		/* Extra pinning count for the created dentry */
2968	d_instantiate(dentry, inode);
2969out:
2970	return ret;
2971}
2972
2973static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2974{
2975	struct inode *inode = d_inode(dentry);
2976
2977	if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2978		shmem_free_inode(inode->i_sb);
2979
2980	dir->i_size -= BOGO_DIRENT_SIZE;
2981	inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
2982	drop_nlink(inode);
2983	dput(dentry);	/* Undo the count from "create" - this does all the work */
2984	return 0;
2985}
2986
2987static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2988{
2989	if (!simple_empty(dentry))
2990		return -ENOTEMPTY;
2991
2992	drop_nlink(d_inode(dentry));
2993	drop_nlink(dir);
2994	return shmem_unlink(dir, dentry);
2995}
2996
2997static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2998{
2999	bool old_is_dir = d_is_dir(old_dentry);
3000	bool new_is_dir = d_is_dir(new_dentry);
3001
3002	if (old_dir != new_dir && old_is_dir != new_is_dir) {
3003		if (old_is_dir) {
3004			drop_nlink(old_dir);
3005			inc_nlink(new_dir);
3006		} else {
3007			drop_nlink(new_dir);
3008			inc_nlink(old_dir);
3009		}
3010	}
3011	old_dir->i_ctime = old_dir->i_mtime =
3012	new_dir->i_ctime = new_dir->i_mtime =
3013	d_inode(old_dentry)->i_ctime =
3014	d_inode(new_dentry)->i_ctime = current_time(old_dir);
3015
3016	return 0;
3017}
3018
3019static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
3020{
3021	struct dentry *whiteout;
3022	int error;
3023
3024	whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3025	if (!whiteout)
3026		return -ENOMEM;
3027
3028	error = shmem_mknod(old_dir, whiteout,
3029			    S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3030	dput(whiteout);
3031	if (error)
3032		return error;
3033
3034	/*
3035	 * Cheat and hash the whiteout while the old dentry is still in
3036	 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3037	 *
3038	 * d_lookup() will consistently find one of them at this point,
3039	 * not sure which one, but that isn't even important.
3040	 */
3041	d_rehash(whiteout);
3042	return 0;
3043}
3044
3045/*
3046 * The VFS layer already does all the dentry stuff for rename,
3047 * we just have to decrement the usage count for the target if
3048 * it exists so that the VFS layer correctly free's it when it
3049 * gets overwritten.
3050 */
3051static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags)
3052{
3053	struct inode *inode = d_inode(old_dentry);
3054	int they_are_dirs = S_ISDIR(inode->i_mode);
3055
3056	if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
3057		return -EINVAL;
3058
3059	if (flags & RENAME_EXCHANGE)
3060		return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3061
3062	if (!simple_empty(new_dentry))
3063		return -ENOTEMPTY;
3064
3065	if (flags & RENAME_WHITEOUT) {
3066		int error;
3067
3068		error = shmem_whiteout(old_dir, old_dentry);
3069		if (error)
3070			return error;
3071	}
3072
3073	if (d_really_is_positive(new_dentry)) {
3074		(void) shmem_unlink(new_dir, new_dentry);
3075		if (they_are_dirs) {
3076			drop_nlink(d_inode(new_dentry));
3077			drop_nlink(old_dir);
3078		}
3079	} else if (they_are_dirs) {
3080		drop_nlink(old_dir);
3081		inc_nlink(new_dir);
3082	}
3083
3084	old_dir->i_size -= BOGO_DIRENT_SIZE;
3085	new_dir->i_size += BOGO_DIRENT_SIZE;
3086	old_dir->i_ctime = old_dir->i_mtime =
3087	new_dir->i_ctime = new_dir->i_mtime =
3088	inode->i_ctime = current_time(old_dir);
3089	return 0;
3090}
3091
3092static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3093{
3094	int error;
3095	int len;
3096	struct inode *inode;
3097	struct page *page;
3098
3099	len = strlen(symname) + 1;
3100	if (len > PAGE_SIZE)
3101		return -ENAMETOOLONG;
3102
3103	inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK | 0777, 0,
3104				VM_NORESERVE);
3105	if (!inode)
3106		return -ENOSPC;
3107
3108	error = security_inode_init_security(inode, dir, &dentry->d_name,
3109					     shmem_initxattrs, NULL);
3110	if (error) {
3111		if (error != -EOPNOTSUPP) {
3112			iput(inode);
3113			return error;
3114		}
3115		error = 0;
3116	}
3117
3118	inode->i_size = len-1;
3119	if (len <= SHORT_SYMLINK_LEN) {
3120		inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3121		if (!inode->i_link) {
3122			iput(inode);
3123			return -ENOMEM;
3124		}
3125		inode->i_op = &shmem_short_symlink_operations;
3126	} else {
3127		inode_nohighmem(inode);
3128		error = shmem_getpage(inode, 0, &page, SGP_WRITE);
3129		if (error) {
3130			iput(inode);
3131			return error;
3132		}
3133		inode->i_mapping->a_ops = &shmem_aops;
3134		inode->i_op = &shmem_symlink_inode_operations;
3135		memcpy(page_address(page), symname, len);
3136		SetPageUptodate(page);
3137		set_page_dirty(page);
3138		unlock_page(page);
3139		put_page(page);
3140	}
3141	dir->i_size += BOGO_DIRENT_SIZE;
3142	dir->i_ctime = dir->i_mtime = current_time(dir);
3143	d_instantiate(dentry, inode);
3144	dget(dentry);
3145	return 0;
3146}
3147
3148static void shmem_put_link(void *arg)
3149{
3150	mark_page_accessed(arg);
3151	put_page(arg);
3152}
3153
3154static const char *shmem_get_link(struct dentry *dentry,
3155				  struct inode *inode,
3156				  struct delayed_call *done)
3157{
3158	struct page *page = NULL;
3159	int error;
3160	if (!dentry) {
3161		page = find_get_page(inode->i_mapping, 0);
3162		if (!page)
3163			return ERR_PTR(-ECHILD);
3164		if (!PageUptodate(page)) {
3165			put_page(page);
3166			return ERR_PTR(-ECHILD);
3167		}
3168	} else {
3169		error = shmem_getpage(inode, 0, &page, SGP_READ);
3170		if (error)
3171			return ERR_PTR(error);
3172		unlock_page(page);
3173	}
3174	set_delayed_call(done, shmem_put_link, page);
3175	return page_address(page);
3176}
3177
3178#ifdef CONFIG_TMPFS_XATTR
3179/*
3180 * Superblocks without xattr inode operations may get some security.* xattr
3181 * support from the LSM "for free". As soon as we have any other xattrs
3182 * like ACLs, we also need to implement the security.* handlers at
3183 * filesystem level, though.
3184 */
3185
3186/*
3187 * Callback for security_inode_init_security() for acquiring xattrs.
3188 */
3189static int shmem_initxattrs(struct inode *inode,
3190			    const struct xattr *xattr_array,
3191			    void *fs_info)
3192{
3193	struct shmem_inode_info *info = SHMEM_I(inode);
3194	const struct xattr *xattr;
3195	struct simple_xattr *new_xattr;
3196	size_t len;
3197
3198	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
3199		new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
3200		if (!new_xattr)
3201			return -ENOMEM;
3202
3203		len = strlen(xattr->name) + 1;
3204		new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3205					  GFP_KERNEL);
3206		if (!new_xattr->name) {
3207			kfree(new_xattr);
3208			return -ENOMEM;
3209		}
3210
3211		memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3212		       XATTR_SECURITY_PREFIX_LEN);
3213		memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3214		       xattr->name, len);
3215
3216		simple_xattr_list_add(&info->xattrs, new_xattr);
3217	}
3218
3219	return 0;
3220}
3221
3222static int shmem_xattr_handler_get(const struct xattr_handler *handler,
3223				   struct dentry *unused, struct inode *inode,
3224				   const char *name, void *buffer, size_t size)
3225{
3226	struct shmem_inode_info *info = SHMEM_I(inode);
3227
3228	name = xattr_full_name(handler, name);
3229	return simple_xattr_get(&info->xattrs, name, buffer, size);
3230}
3231
3232static int shmem_xattr_handler_set(const struct xattr_handler *handler,
3233				   struct dentry *unused, struct inode *inode,
3234				   const char *name, const void *value,
3235				   size_t size, int flags)
3236{
3237	struct shmem_inode_info *info = SHMEM_I(inode);
3238
3239	name = xattr_full_name(handler, name);
3240	return simple_xattr_set(&info->xattrs, name, value, size, flags);
3241}
3242
3243static const struct xattr_handler shmem_security_xattr_handler = {
3244	.prefix = XATTR_SECURITY_PREFIX,
3245	.get = shmem_xattr_handler_get,
3246	.set = shmem_xattr_handler_set,
3247};
3248
3249static const struct xattr_handler shmem_trusted_xattr_handler = {
3250	.prefix = XATTR_TRUSTED_PREFIX,
3251	.get = shmem_xattr_handler_get,
3252	.set = shmem_xattr_handler_set,
3253};
3254
3255static const struct xattr_handler *shmem_xattr_handlers[] = {
3256#ifdef CONFIG_TMPFS_POSIX_ACL
3257	&posix_acl_access_xattr_handler,
3258	&posix_acl_default_xattr_handler,
3259#endif
3260	&shmem_security_xattr_handler,
3261	&shmem_trusted_xattr_handler,
3262	NULL
3263};
3264
3265static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3266{
3267	struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
3268	return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
3269}
3270#endif /* CONFIG_TMPFS_XATTR */
3271
3272static const struct inode_operations shmem_short_symlink_operations = {
3273	.get_link	= simple_get_link,
3274#ifdef CONFIG_TMPFS_XATTR
3275	.listxattr	= shmem_listxattr,
3276#endif
3277};
3278
3279static const struct inode_operations shmem_symlink_inode_operations = {
3280	.get_link	= shmem_get_link,
3281#ifdef CONFIG_TMPFS_XATTR
3282	.listxattr	= shmem_listxattr,
3283#endif
3284};
3285
3286static struct dentry *shmem_get_parent(struct dentry *child)
3287{
3288	return ERR_PTR(-ESTALE);
3289}
3290
3291static int shmem_match(struct inode *ino, void *vfh)
3292{
3293	__u32 *fh = vfh;
3294	__u64 inum = fh[2];
3295	inum = (inum << 32) | fh[1];
3296	return ino->i_ino == inum && fh[0] == ino->i_generation;
3297}
3298
3299/* Find any alias of inode, but prefer a hashed alias */
3300static struct dentry *shmem_find_alias(struct inode *inode)
3301{
3302	struct dentry *alias = d_find_alias(inode);
3303
3304	return alias ?: d_find_any_alias(inode);
3305}
3306
3307
3308static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3309		struct fid *fid, int fh_len, int fh_type)
3310{
3311	struct inode *inode;
3312	struct dentry *dentry = NULL;
3313	u64 inum;
3314
3315	if (fh_len < 3)
3316		return NULL;
3317
3318	inum = fid->raw[2];
3319	inum = (inum << 32) | fid->raw[1];
3320
3321	inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3322			shmem_match, fid->raw);
3323	if (inode) {
3324		dentry = shmem_find_alias(inode);
3325		iput(inode);
3326	}
3327
3328	return dentry;
3329}
3330
3331static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3332				struct inode *parent)
3333{
3334	if (*len < 3) {
3335		*len = 3;
3336		return FILEID_INVALID;
3337	}
3338
3339	if (inode_unhashed(inode)) {
3340		/* Unfortunately insert_inode_hash is not idempotent,
3341		 * so as we hash inodes here rather than at creation
3342		 * time, we need a lock to ensure we only try
3343		 * to do it once
3344		 */
3345		static DEFINE_SPINLOCK(lock);
3346		spin_lock(&lock);
3347		if (inode_unhashed(inode))
3348			__insert_inode_hash(inode,
3349					    inode->i_ino + inode->i_generation);
3350		spin_unlock(&lock);
3351	}
3352
3353	fh[0] = inode->i_generation;
3354	fh[1] = inode->i_ino;
3355	fh[2] = ((__u64)inode->i_ino) >> 32;
3356
3357	*len = 3;
3358	return 1;
3359}
3360
3361static const struct export_operations shmem_export_ops = {
3362	.get_parent     = shmem_get_parent,
3363	.encode_fh      = shmem_encode_fh,
3364	.fh_to_dentry	= shmem_fh_to_dentry,
3365};
3366
3367enum shmem_param {
3368	Opt_gid,
3369	Opt_huge,
3370	Opt_mode,
3371	Opt_mpol,
3372	Opt_nr_blocks,
3373	Opt_nr_inodes,
3374	Opt_size,
3375	Opt_uid,
 
 
3376};
3377
3378static const struct fs_parameter_spec shmem_param_specs[] = {
 
 
 
 
 
 
 
 
3379	fsparam_u32   ("gid",		Opt_gid),
3380	fsparam_enum  ("huge",		Opt_huge),
3381	fsparam_u32oct("mode",		Opt_mode),
3382	fsparam_string("mpol",		Opt_mpol),
3383	fsparam_string("nr_blocks",	Opt_nr_blocks),
3384	fsparam_string("nr_inodes",	Opt_nr_inodes),
3385	fsparam_string("size",		Opt_size),
3386	fsparam_u32   ("uid",		Opt_uid),
 
 
3387	{}
3388};
3389
3390static const struct fs_parameter_enum shmem_param_enums[] = {
3391	{ Opt_huge,	"never",	SHMEM_HUGE_NEVER },
3392	{ Opt_huge,	"always",	SHMEM_HUGE_ALWAYS },
3393	{ Opt_huge,	"within_size",	SHMEM_HUGE_WITHIN_SIZE },
3394	{ Opt_huge,	"advise",	SHMEM_HUGE_ADVISE },
3395	{}
3396};
3397
3398const struct fs_parameter_description shmem_fs_parameters = {
3399	.name		= "tmpfs",
3400	.specs		= shmem_param_specs,
3401	.enums		= shmem_param_enums,
3402};
3403
3404static int shmem_parse_one(struct fs_context *fc, struct fs_parameter *param)
3405{
3406	struct shmem_options *ctx = fc->fs_private;
3407	struct fs_parse_result result;
3408	unsigned long long size;
3409	char *rest;
3410	int opt;
3411
3412	opt = fs_parse(fc, &shmem_fs_parameters, param, &result);
3413	if (opt < 0)
3414		return opt;
3415
3416	switch (opt) {
3417	case Opt_size:
3418		size = memparse(param->string, &rest);
3419		if (*rest == '%') {
3420			size <<= PAGE_SHIFT;
3421			size *= totalram_pages();
3422			do_div(size, 100);
3423			rest++;
3424		}
3425		if (*rest)
3426			goto bad_value;
3427		ctx->blocks = DIV_ROUND_UP(size, PAGE_SIZE);
3428		ctx->seen |= SHMEM_SEEN_BLOCKS;
3429		break;
3430	case Opt_nr_blocks:
3431		ctx->blocks = memparse(param->string, &rest);
3432		if (*rest)
3433			goto bad_value;
3434		ctx->seen |= SHMEM_SEEN_BLOCKS;
3435		break;
3436	case Opt_nr_inodes:
3437		ctx->inodes = memparse(param->string, &rest);
3438		if (*rest)
3439			goto bad_value;
3440		ctx->seen |= SHMEM_SEEN_INODES;
3441		break;
3442	case Opt_mode:
3443		ctx->mode = result.uint_32 & 07777;
3444		break;
3445	case Opt_uid:
3446		ctx->uid = make_kuid(current_user_ns(), result.uint_32);
3447		if (!uid_valid(ctx->uid))
3448			goto bad_value;
3449		break;
3450	case Opt_gid:
3451		ctx->gid = make_kgid(current_user_ns(), result.uint_32);
3452		if (!gid_valid(ctx->gid))
3453			goto bad_value;
3454		break;
3455	case Opt_huge:
3456		ctx->huge = result.uint_32;
3457		if (ctx->huge != SHMEM_HUGE_NEVER &&
3458		    !(IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
3459		      has_transparent_hugepage()))
3460			goto unsupported_parameter;
3461		ctx->seen |= SHMEM_SEEN_HUGE;
3462		break;
3463	case Opt_mpol:
3464		if (IS_ENABLED(CONFIG_NUMA)) {
3465			mpol_put(ctx->mpol);
3466			ctx->mpol = NULL;
3467			if (mpol_parse_str(param->string, &ctx->mpol))
3468				goto bad_value;
3469			break;
3470		}
3471		goto unsupported_parameter;
 
 
 
 
 
 
 
 
 
 
 
 
3472	}
3473	return 0;
3474
3475unsupported_parameter:
3476	return invalf(fc, "tmpfs: Unsupported parameter '%s'", param->key);
3477bad_value:
3478	return invalf(fc, "tmpfs: Bad value for '%s'", param->key);
3479}
3480
3481static int shmem_parse_options(struct fs_context *fc, void *data)
3482{
3483	char *options = data;
3484
3485	if (options) {
3486		int err = security_sb_eat_lsm_opts(options, &fc->security);
3487		if (err)
3488			return err;
3489	}
3490
3491	while (options != NULL) {
3492		char *this_char = options;
3493		for (;;) {
3494			/*
3495			 * NUL-terminate this option: unfortunately,
3496			 * mount options form a comma-separated list,
3497			 * but mpol's nodelist may also contain commas.
3498			 */
3499			options = strchr(options, ',');
3500			if (options == NULL)
3501				break;
3502			options++;
3503			if (!isdigit(*options)) {
3504				options[-1] = '\0';
3505				break;
3506			}
3507		}
3508		if (*this_char) {
3509			char *value = strchr(this_char,'=');
3510			size_t len = 0;
3511			int err;
3512
3513			if (value) {
3514				*value++ = '\0';
3515				len = strlen(value);
3516			}
3517			err = vfs_parse_fs_string(fc, this_char, value, len);
3518			if (err < 0)
3519				return err;
3520		}
3521	}
3522	return 0;
3523}
3524
3525/*
3526 * Reconfigure a shmem filesystem.
3527 *
3528 * Note that we disallow change from limited->unlimited blocks/inodes while any
3529 * are in use; but we must separately disallow unlimited->limited, because in
3530 * that case we have no record of how much is already in use.
3531 */
3532static int shmem_reconfigure(struct fs_context *fc)
3533{
3534	struct shmem_options *ctx = fc->fs_private;
3535	struct shmem_sb_info *sbinfo = SHMEM_SB(fc->root->d_sb);
3536	unsigned long inodes;
3537	const char *err;
3538
3539	spin_lock(&sbinfo->stat_lock);
3540	inodes = sbinfo->max_inodes - sbinfo->free_inodes;
3541	if ((ctx->seen & SHMEM_SEEN_BLOCKS) && ctx->blocks) {
3542		if (!sbinfo->max_blocks) {
3543			err = "Cannot retroactively limit size";
3544			goto out;
3545		}
3546		if (percpu_counter_compare(&sbinfo->used_blocks,
3547					   ctx->blocks) > 0) {
3548			err = "Too small a size for current use";
3549			goto out;
3550		}
3551	}
3552	if ((ctx->seen & SHMEM_SEEN_INODES) && ctx->inodes) {
3553		if (!sbinfo->max_inodes) {
3554			err = "Cannot retroactively limit inodes";
3555			goto out;
3556		}
3557		if (ctx->inodes < inodes) {
3558			err = "Too few inodes for current use";
3559			goto out;
3560		}
3561	}
3562
 
 
 
 
 
 
3563	if (ctx->seen & SHMEM_SEEN_HUGE)
3564		sbinfo->huge = ctx->huge;
 
 
3565	if (ctx->seen & SHMEM_SEEN_BLOCKS)
3566		sbinfo->max_blocks  = ctx->blocks;
3567	if (ctx->seen & SHMEM_SEEN_INODES) {
3568		sbinfo->max_inodes  = ctx->inodes;
3569		sbinfo->free_inodes = ctx->inodes - inodes;
3570	}
3571
3572	/*
3573	 * Preserve previous mempolicy unless mpol remount option was specified.
3574	 */
3575	if (ctx->mpol) {
3576		mpol_put(sbinfo->mpol);
3577		sbinfo->mpol = ctx->mpol;	/* transfers initial ref */
3578		ctx->mpol = NULL;
3579	}
3580	spin_unlock(&sbinfo->stat_lock);
3581	return 0;
3582out:
3583	spin_unlock(&sbinfo->stat_lock);
3584	return invalf(fc, "tmpfs: %s", err);
3585}
3586
3587static int shmem_show_options(struct seq_file *seq, struct dentry *root)
3588{
3589	struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
3590
3591	if (sbinfo->max_blocks != shmem_default_max_blocks())
3592		seq_printf(seq, ",size=%luk",
3593			sbinfo->max_blocks << (PAGE_SHIFT - 10));
3594	if (sbinfo->max_inodes != shmem_default_max_inodes())
3595		seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3596	if (sbinfo->mode != (0777 | S_ISVTX))
3597		seq_printf(seq, ",mode=%03ho", sbinfo->mode);
3598	if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3599		seq_printf(seq, ",uid=%u",
3600				from_kuid_munged(&init_user_ns, sbinfo->uid));
3601	if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3602		seq_printf(seq, ",gid=%u",
3603				from_kgid_munged(&init_user_ns, sbinfo->gid));
3604#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
3605	/* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3606	if (sbinfo->huge)
3607		seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3608#endif
3609	shmem_show_mpol(seq, sbinfo->mpol);
3610	return 0;
3611}
3612
3613#endif /* CONFIG_TMPFS */
3614
3615static void shmem_put_super(struct super_block *sb)
3616{
3617	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3618
 
3619	percpu_counter_destroy(&sbinfo->used_blocks);
3620	mpol_put(sbinfo->mpol);
3621	kfree(sbinfo);
3622	sb->s_fs_info = NULL;
3623}
3624
3625static int shmem_fill_super(struct super_block *sb, struct fs_context *fc)
3626{
3627	struct shmem_options *ctx = fc->fs_private;
3628	struct inode *inode;
3629	struct shmem_sb_info *sbinfo;
3630	int err = -ENOMEM;
3631
3632	/* Round up to L1_CACHE_BYTES to resist false sharing */
3633	sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
3634				L1_CACHE_BYTES), GFP_KERNEL);
3635	if (!sbinfo)
3636		return -ENOMEM;
3637
3638	sb->s_fs_info = sbinfo;
3639
3640#ifdef CONFIG_TMPFS
3641	/*
3642	 * Per default we only allow half of the physical ram per
3643	 * tmpfs instance, limiting inodes to one per page of lowmem;
3644	 * but the internal instance is left unlimited.
3645	 */
3646	if (!(sb->s_flags & SB_KERNMOUNT)) {
3647		if (!(ctx->seen & SHMEM_SEEN_BLOCKS))
3648			ctx->blocks = shmem_default_max_blocks();
3649		if (!(ctx->seen & SHMEM_SEEN_INODES))
3650			ctx->inodes = shmem_default_max_inodes();
 
 
3651	} else {
3652		sb->s_flags |= SB_NOUSER;
3653	}
3654	sb->s_export_op = &shmem_export_ops;
3655	sb->s_flags |= SB_NOSEC;
3656#else
3657	sb->s_flags |= SB_NOUSER;
3658#endif
3659	sbinfo->max_blocks = ctx->blocks;
3660	sbinfo->free_inodes = sbinfo->max_inodes = ctx->inodes;
 
 
 
 
 
3661	sbinfo->uid = ctx->uid;
3662	sbinfo->gid = ctx->gid;
 
3663	sbinfo->mode = ctx->mode;
3664	sbinfo->huge = ctx->huge;
3665	sbinfo->mpol = ctx->mpol;
3666	ctx->mpol = NULL;
3667
3668	spin_lock_init(&sbinfo->stat_lock);
3669	if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
3670		goto failed;
3671	spin_lock_init(&sbinfo->shrinklist_lock);
3672	INIT_LIST_HEAD(&sbinfo->shrinklist);
3673
3674	sb->s_maxbytes = MAX_LFS_FILESIZE;
3675	sb->s_blocksize = PAGE_SIZE;
3676	sb->s_blocksize_bits = PAGE_SHIFT;
3677	sb->s_magic = TMPFS_MAGIC;
3678	sb->s_op = &shmem_ops;
3679	sb->s_time_gran = 1;
3680#ifdef CONFIG_TMPFS_XATTR
3681	sb->s_xattr = shmem_xattr_handlers;
3682#endif
3683#ifdef CONFIG_TMPFS_POSIX_ACL
3684	sb->s_flags |= SB_POSIXACL;
3685#endif
3686	uuid_gen(&sb->s_uuid);
3687
3688	inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
3689	if (!inode)
3690		goto failed;
3691	inode->i_uid = sbinfo->uid;
3692	inode->i_gid = sbinfo->gid;
3693	sb->s_root = d_make_root(inode);
3694	if (!sb->s_root)
3695		goto failed;
3696	return 0;
3697
3698failed:
3699	shmem_put_super(sb);
3700	return err;
3701}
3702
3703static int shmem_get_tree(struct fs_context *fc)
3704{
3705	return get_tree_nodev(fc, shmem_fill_super);
3706}
3707
3708static void shmem_free_fc(struct fs_context *fc)
3709{
3710	struct shmem_options *ctx = fc->fs_private;
3711
3712	if (ctx) {
3713		mpol_put(ctx->mpol);
3714		kfree(ctx);
3715	}
3716}
3717
3718static const struct fs_context_operations shmem_fs_context_ops = {
3719	.free			= shmem_free_fc,
3720	.get_tree		= shmem_get_tree,
3721#ifdef CONFIG_TMPFS
3722	.parse_monolithic	= shmem_parse_options,
3723	.parse_param		= shmem_parse_one,
3724	.reconfigure		= shmem_reconfigure,
3725#endif
3726};
3727
3728static struct kmem_cache *shmem_inode_cachep;
3729
3730static struct inode *shmem_alloc_inode(struct super_block *sb)
3731{
3732	struct shmem_inode_info *info;
3733	info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3734	if (!info)
3735		return NULL;
3736	return &info->vfs_inode;
3737}
3738
3739static void shmem_free_in_core_inode(struct inode *inode)
3740{
3741	if (S_ISLNK(inode->i_mode))
3742		kfree(inode->i_link);
3743	kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3744}
3745
3746static void shmem_destroy_inode(struct inode *inode)
3747{
3748	if (S_ISREG(inode->i_mode))
3749		mpol_free_shared_policy(&SHMEM_I(inode)->policy);
3750}
3751
3752static void shmem_init_inode(void *foo)
3753{
3754	struct shmem_inode_info *info = foo;
3755	inode_init_once(&info->vfs_inode);
3756}
3757
3758static void shmem_init_inodecache(void)
3759{
3760	shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3761				sizeof(struct shmem_inode_info),
3762				0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
3763}
3764
3765static void shmem_destroy_inodecache(void)
3766{
3767	kmem_cache_destroy(shmem_inode_cachep);
3768}
3769
3770static const struct address_space_operations shmem_aops = {
3771	.writepage	= shmem_writepage,
3772	.set_page_dirty	= __set_page_dirty_no_writeback,
3773#ifdef CONFIG_TMPFS
3774	.write_begin	= shmem_write_begin,
3775	.write_end	= shmem_write_end,
3776#endif
3777#ifdef CONFIG_MIGRATION
3778	.migratepage	= migrate_page,
3779#endif
3780	.error_remove_page = generic_error_remove_page,
3781};
3782
3783static const struct file_operations shmem_file_operations = {
3784	.mmap		= shmem_mmap,
3785	.get_unmapped_area = shmem_get_unmapped_area,
3786#ifdef CONFIG_TMPFS
3787	.llseek		= shmem_file_llseek,
3788	.read_iter	= shmem_file_read_iter,
3789	.write_iter	= generic_file_write_iter,
3790	.fsync		= noop_fsync,
3791	.splice_read	= generic_file_splice_read,
3792	.splice_write	= iter_file_splice_write,
3793	.fallocate	= shmem_fallocate,
3794#endif
3795};
3796
3797static const struct inode_operations shmem_inode_operations = {
3798	.getattr	= shmem_getattr,
3799	.setattr	= shmem_setattr,
3800#ifdef CONFIG_TMPFS_XATTR
3801	.listxattr	= shmem_listxattr,
3802	.set_acl	= simple_set_acl,
3803#endif
3804};
3805
3806static const struct inode_operations shmem_dir_inode_operations = {
3807#ifdef CONFIG_TMPFS
3808	.create		= shmem_create,
3809	.lookup		= simple_lookup,
3810	.link		= shmem_link,
3811	.unlink		= shmem_unlink,
3812	.symlink	= shmem_symlink,
3813	.mkdir		= shmem_mkdir,
3814	.rmdir		= shmem_rmdir,
3815	.mknod		= shmem_mknod,
3816	.rename		= shmem_rename2,
3817	.tmpfile	= shmem_tmpfile,
3818#endif
3819#ifdef CONFIG_TMPFS_XATTR
3820	.listxattr	= shmem_listxattr,
3821#endif
3822#ifdef CONFIG_TMPFS_POSIX_ACL
3823	.setattr	= shmem_setattr,
3824	.set_acl	= simple_set_acl,
3825#endif
3826};
3827
3828static const struct inode_operations shmem_special_inode_operations = {
3829#ifdef CONFIG_TMPFS_XATTR
3830	.listxattr	= shmem_listxattr,
3831#endif
3832#ifdef CONFIG_TMPFS_POSIX_ACL
3833	.setattr	= shmem_setattr,
3834	.set_acl	= simple_set_acl,
3835#endif
3836};
3837
3838static const struct super_operations shmem_ops = {
3839	.alloc_inode	= shmem_alloc_inode,
3840	.free_inode	= shmem_free_in_core_inode,
3841	.destroy_inode	= shmem_destroy_inode,
3842#ifdef CONFIG_TMPFS
3843	.statfs		= shmem_statfs,
3844	.show_options	= shmem_show_options,
3845#endif
3846	.evict_inode	= shmem_evict_inode,
3847	.drop_inode	= generic_delete_inode,
3848	.put_super	= shmem_put_super,
3849#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3850	.nr_cached_objects	= shmem_unused_huge_count,
3851	.free_cached_objects	= shmem_unused_huge_scan,
3852#endif
3853};
3854
3855static const struct vm_operations_struct shmem_vm_ops = {
3856	.fault		= shmem_fault,
3857	.map_pages	= filemap_map_pages,
3858#ifdef CONFIG_NUMA
3859	.set_policy     = shmem_set_policy,
3860	.get_policy     = shmem_get_policy,
3861#endif
3862};
3863
3864int shmem_init_fs_context(struct fs_context *fc)
3865{
3866	struct shmem_options *ctx;
3867
3868	ctx = kzalloc(sizeof(struct shmem_options), GFP_KERNEL);
3869	if (!ctx)
3870		return -ENOMEM;
3871
3872	ctx->mode = 0777 | S_ISVTX;
3873	ctx->uid = current_fsuid();
3874	ctx->gid = current_fsgid();
3875
3876	fc->fs_private = ctx;
3877	fc->ops = &shmem_fs_context_ops;
3878	return 0;
3879}
3880
3881static struct file_system_type shmem_fs_type = {
3882	.owner		= THIS_MODULE,
3883	.name		= "tmpfs",
3884	.init_fs_context = shmem_init_fs_context,
3885#ifdef CONFIG_TMPFS
3886	.parameters	= &shmem_fs_parameters,
3887#endif
3888	.kill_sb	= kill_litter_super,
3889	.fs_flags	= FS_USERNS_MOUNT,
3890};
3891
3892int __init shmem_init(void)
3893{
3894	int error;
3895
3896	shmem_init_inodecache();
3897
3898	error = register_filesystem(&shmem_fs_type);
3899	if (error) {
3900		pr_err("Could not register tmpfs\n");
3901		goto out2;
3902	}
3903
3904	shm_mnt = kern_mount(&shmem_fs_type);
3905	if (IS_ERR(shm_mnt)) {
3906		error = PTR_ERR(shm_mnt);
3907		pr_err("Could not kern_mount tmpfs\n");
3908		goto out1;
3909	}
3910
3911#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3912	if (has_transparent_hugepage() && shmem_huge > SHMEM_HUGE_DENY)
3913		SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3914	else
3915		shmem_huge = 0; /* just in case it was patched */
3916#endif
3917	return 0;
3918
3919out1:
3920	unregister_filesystem(&shmem_fs_type);
3921out2:
3922	shmem_destroy_inodecache();
3923	shm_mnt = ERR_PTR(error);
3924	return error;
3925}
3926
3927#if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
3928static ssize_t shmem_enabled_show(struct kobject *kobj,
3929		struct kobj_attribute *attr, char *buf)
3930{
3931	int values[] = {
3932		SHMEM_HUGE_ALWAYS,
3933		SHMEM_HUGE_WITHIN_SIZE,
3934		SHMEM_HUGE_ADVISE,
3935		SHMEM_HUGE_NEVER,
3936		SHMEM_HUGE_DENY,
3937		SHMEM_HUGE_FORCE,
3938	};
3939	int i, count;
3940
3941	for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
3942		const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
3943
3944		count += sprintf(buf + count, fmt,
3945				shmem_format_huge(values[i]));
3946	}
3947	buf[count - 1] = '\n';
3948	return count;
3949}
3950
3951static ssize_t shmem_enabled_store(struct kobject *kobj,
3952		struct kobj_attribute *attr, const char *buf, size_t count)
3953{
3954	char tmp[16];
3955	int huge;
3956
3957	if (count + 1 > sizeof(tmp))
3958		return -EINVAL;
3959	memcpy(tmp, buf, count);
3960	tmp[count] = '\0';
3961	if (count && tmp[count - 1] == '\n')
3962		tmp[count - 1] = '\0';
3963
3964	huge = shmem_parse_huge(tmp);
3965	if (huge == -EINVAL)
3966		return -EINVAL;
3967	if (!has_transparent_hugepage() &&
3968			huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
3969		return -EINVAL;
3970
3971	shmem_huge = huge;
3972	if (shmem_huge > SHMEM_HUGE_DENY)
3973		SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3974	return count;
3975}
3976
3977struct kobj_attribute shmem_enabled_attr =
3978	__ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
3979#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
3980
3981#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3982bool shmem_huge_enabled(struct vm_area_struct *vma)
3983{
3984	struct inode *inode = file_inode(vma->vm_file);
3985	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
3986	loff_t i_size;
3987	pgoff_t off;
3988
3989	if ((vma->vm_flags & VM_NOHUGEPAGE) ||
3990	    test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
3991		return false;
3992	if (shmem_huge == SHMEM_HUGE_FORCE)
3993		return true;
3994	if (shmem_huge == SHMEM_HUGE_DENY)
3995		return false;
3996	switch (sbinfo->huge) {
3997		case SHMEM_HUGE_NEVER:
3998			return false;
3999		case SHMEM_HUGE_ALWAYS:
4000			return true;
4001		case SHMEM_HUGE_WITHIN_SIZE:
4002			off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
4003			i_size = round_up(i_size_read(inode), PAGE_SIZE);
4004			if (i_size >= HPAGE_PMD_SIZE &&
4005					i_size >> PAGE_SHIFT >= off)
4006				return true;
4007			/* fall through */
4008		case SHMEM_HUGE_ADVISE:
4009			/* TODO: implement fadvise() hints */
4010			return (vma->vm_flags & VM_HUGEPAGE);
4011		default:
4012			VM_BUG_ON(1);
4013			return false;
4014	}
4015}
4016#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
4017
4018#else /* !CONFIG_SHMEM */
4019
4020/*
4021 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4022 *
4023 * This is intended for small system where the benefits of the full
4024 * shmem code (swap-backed and resource-limited) are outweighed by
4025 * their complexity. On systems without swap this code should be
4026 * effectively equivalent, but much lighter weight.
4027 */
4028
4029static struct file_system_type shmem_fs_type = {
4030	.name		= "tmpfs",
4031	.init_fs_context = ramfs_init_fs_context,
4032	.parameters	= &ramfs_fs_parameters,
4033	.kill_sb	= kill_litter_super,
4034	.fs_flags	= FS_USERNS_MOUNT,
4035};
4036
4037int __init shmem_init(void)
4038{
4039	BUG_ON(register_filesystem(&shmem_fs_type) != 0);
4040
4041	shm_mnt = kern_mount(&shmem_fs_type);
4042	BUG_ON(IS_ERR(shm_mnt));
4043
4044	return 0;
4045}
4046
4047int shmem_unuse(unsigned int type, bool frontswap,
4048		unsigned long *fs_pages_to_unuse)
4049{
4050	return 0;
4051}
4052
4053int shmem_lock(struct file *file, int lock, struct user_struct *user)
4054{
4055	return 0;
4056}
4057
4058void shmem_unlock_mapping(struct address_space *mapping)
4059{
4060}
4061
4062#ifdef CONFIG_MMU
4063unsigned long shmem_get_unmapped_area(struct file *file,
4064				      unsigned long addr, unsigned long len,
4065				      unsigned long pgoff, unsigned long flags)
4066{
4067	return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4068}
4069#endif
4070
4071void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
4072{
4073	truncate_inode_pages_range(inode->i_mapping, lstart, lend);
4074}
4075EXPORT_SYMBOL_GPL(shmem_truncate_range);
4076
4077#define shmem_vm_ops				generic_file_vm_ops
4078#define shmem_file_operations			ramfs_file_operations
4079#define shmem_get_inode(sb, dir, mode, dev, flags)	ramfs_get_inode(sb, dir, mode, dev)
4080#define shmem_acct_size(flags, size)		0
4081#define shmem_unacct_size(flags, size)		do {} while (0)
4082
4083#endif /* CONFIG_SHMEM */
4084
4085/* common code */
4086
4087static struct file *__shmem_file_setup(struct vfsmount *mnt, const char *name, loff_t size,
4088				       unsigned long flags, unsigned int i_flags)
4089{
4090	struct inode *inode;
4091	struct file *res;
4092
4093	if (IS_ERR(mnt))
4094		return ERR_CAST(mnt);
4095
4096	if (size < 0 || size > MAX_LFS_FILESIZE)
4097		return ERR_PTR(-EINVAL);
4098
4099	if (shmem_acct_size(flags, size))
4100		return ERR_PTR(-ENOMEM);
4101
4102	inode = shmem_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0,
4103				flags);
4104	if (unlikely(!inode)) {
4105		shmem_unacct_size(flags, size);
4106		return ERR_PTR(-ENOSPC);
4107	}
4108	inode->i_flags |= i_flags;
4109	inode->i_size = size;
4110	clear_nlink(inode);	/* It is unlinked */
4111	res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4112	if (!IS_ERR(res))
4113		res = alloc_file_pseudo(inode, mnt, name, O_RDWR,
4114				&shmem_file_operations);
4115	if (IS_ERR(res))
4116		iput(inode);
4117	return res;
4118}
4119
4120/**
4121 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4122 * 	kernel internal.  There will be NO LSM permission checks against the
4123 * 	underlying inode.  So users of this interface must do LSM checks at a
4124 *	higher layer.  The users are the big_key and shm implementations.  LSM
4125 *	checks are provided at the key or shm level rather than the inode.
4126 * @name: name for dentry (to be seen in /proc/<pid>/maps
4127 * @size: size to be set for the file
4128 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4129 */
4130struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4131{
4132	return __shmem_file_setup(shm_mnt, name, size, flags, S_PRIVATE);
4133}
4134
4135/**
4136 * shmem_file_setup - get an unlinked file living in tmpfs
4137 * @name: name for dentry (to be seen in /proc/<pid>/maps
4138 * @size: size to be set for the file
4139 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4140 */
4141struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4142{
4143	return __shmem_file_setup(shm_mnt, name, size, flags, 0);
4144}
4145EXPORT_SYMBOL_GPL(shmem_file_setup);
4146
4147/**
4148 * shmem_file_setup_with_mnt - get an unlinked file living in tmpfs
4149 * @mnt: the tmpfs mount where the file will be created
4150 * @name: name for dentry (to be seen in /proc/<pid>/maps
4151 * @size: size to be set for the file
4152 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4153 */
4154struct file *shmem_file_setup_with_mnt(struct vfsmount *mnt, const char *name,
4155				       loff_t size, unsigned long flags)
4156{
4157	return __shmem_file_setup(mnt, name, size, flags, 0);
4158}
4159EXPORT_SYMBOL_GPL(shmem_file_setup_with_mnt);
4160
4161/**
4162 * shmem_zero_setup - setup a shared anonymous mapping
4163 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4164 */
4165int shmem_zero_setup(struct vm_area_struct *vma)
4166{
4167	struct file *file;
4168	loff_t size = vma->vm_end - vma->vm_start;
4169
4170	/*
4171	 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4172	 * between XFS directory reading and selinux: since this file is only
4173	 * accessible to the user through its mapping, use S_PRIVATE flag to
4174	 * bypass file security, in the same way as shmem_kernel_file_setup().
4175	 */
4176	file = shmem_kernel_file_setup("dev/zero", size, vma->vm_flags);
4177	if (IS_ERR(file))
4178		return PTR_ERR(file);
4179
4180	if (vma->vm_file)
4181		fput(vma->vm_file);
4182	vma->vm_file = file;
4183	vma->vm_ops = &shmem_vm_ops;
4184
4185	if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
4186			((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4187			(vma->vm_end & HPAGE_PMD_MASK)) {
4188		khugepaged_enter(vma, vma->vm_flags);
4189	}
4190
4191	return 0;
4192}
4193
4194/**
4195 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4196 * @mapping:	the page's address_space
4197 * @index:	the page index
4198 * @gfp:	the page allocator flags to use if allocating
4199 *
4200 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4201 * with any new page allocations done using the specified allocation flags.
4202 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4203 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4204 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4205 *
4206 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4207 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
4208 */
4209struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4210					 pgoff_t index, gfp_t gfp)
4211{
4212#ifdef CONFIG_SHMEM
4213	struct inode *inode = mapping->host;
4214	struct page *page;
4215	int error;
4216
4217	BUG_ON(mapping->a_ops != &shmem_aops);
4218	error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
4219				  gfp, NULL, NULL, NULL);
4220	if (error)
4221		page = ERR_PTR(error);
4222	else
4223		unlock_page(page);
4224	return page;
4225#else
4226	/*
4227	 * The tiny !SHMEM case uses ramfs without swap
4228	 */
4229	return read_cache_page_gfp(mapping, index, gfp);
4230#endif
4231}
4232EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);