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);
v3.15
   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/export.h>
  33#include <linux/swap.h>
  34#include <linux/aio.h>
 
 
 
 
 
 
  35
  36static struct vfsmount *shm_mnt;
  37
  38#ifdef CONFIG_SHMEM
  39/*
  40 * This virtual memory filesystem is heavily based on the ramfs. It
  41 * extends ramfs by the ability to use swap and honor resource limits
  42 * which makes it a completely usable filesystem.
  43 */
  44
  45#include <linux/xattr.h>
  46#include <linux/exportfs.h>
  47#include <linux/posix_acl.h>
  48#include <linux/posix_acl_xattr.h>
  49#include <linux/mman.h>
  50#include <linux/string.h>
  51#include <linux/slab.h>
  52#include <linux/backing-dev.h>
  53#include <linux/shmem_fs.h>
  54#include <linux/writeback.h>
  55#include <linux/blkdev.h>
  56#include <linux/pagevec.h>
  57#include <linux/percpu_counter.h>
  58#include <linux/falloc.h>
  59#include <linux/splice.h>
  60#include <linux/security.h>
  61#include <linux/swapops.h>
  62#include <linux/mempolicy.h>
  63#include <linux/namei.h>
  64#include <linux/ctype.h>
  65#include <linux/migrate.h>
  66#include <linux/highmem.h>
  67#include <linux/seq_file.h>
  68#include <linux/magic.h>
 
 
 
 
 
 
 
 
  69
  70#include <asm/uaccess.h>
  71#include <asm/pgtable.h>
  72
  73#define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
  74#define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
  75
  76/* Pretend that each entry is of this size in directory's i_size */
  77#define BOGO_DIRENT_SIZE 20
  78
  79/* Symlink up to this size is kmalloc'ed instead of using a swappable page */
  80#define SHORT_SYMLINK_LEN 128
  81
  82/*
  83 * shmem_fallocate and shmem_writepage communicate via inode->i_private
  84 * (with i_mutex making sure that it has only one user at a time):
  85 * we would prefer not to enlarge the shmem inode just for that.
  86 */
  87struct shmem_falloc {
 
  88	pgoff_t start;		/* start of range currently being fallocated */
  89	pgoff_t next;		/* the next page offset to be fallocated */
  90	pgoff_t nr_falloced;	/* how many new pages have been fallocated */
  91	pgoff_t nr_unswapped;	/* how often writepage refused to swap out */
  92};
  93
  94/* Flag allocation requirements to shmem_getpage */
  95enum sgp_type {
  96	SGP_READ,	/* don't exceed i_size, don't allocate page */
  97	SGP_CACHE,	/* don't exceed i_size, may allocate page */
  98	SGP_DIRTY,	/* like SGP_CACHE, but set new page dirty */
  99	SGP_WRITE,	/* may exceed i_size, may allocate !Uptodate page */
 100	SGP_FALLOC,	/* like SGP_WRITE, but make existing page Uptodate */
 
 
 
 
 
 
 
 101};
 102
 103#ifdef CONFIG_TMPFS
 104static unsigned long shmem_default_max_blocks(void)
 105{
 106	return totalram_pages / 2;
 107}
 108
 109static unsigned long shmem_default_max_inodes(void)
 110{
 111	return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
 
 
 112}
 113#endif
 114
 115static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
 116static int shmem_replace_page(struct page **pagep, gfp_t gfp,
 117				struct shmem_inode_info *info, pgoff_t index);
 
 
 
 
 118static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
 119	struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
 
 
 120
 121static inline int shmem_getpage(struct inode *inode, pgoff_t index,
 122	struct page **pagep, enum sgp_type sgp, int *fault_type)
 123{
 124	return shmem_getpage_gfp(inode, index, pagep, sgp,
 125			mapping_gfp_mask(inode->i_mapping), fault_type);
 126}
 127
 128static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
 129{
 130	return sb->s_fs_info;
 131}
 132
 133/*
 134 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
 135 * for shared memory and for shared anonymous (/dev/zero) mappings
 136 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
 137 * consistent with the pre-accounting of private mappings ...
 138 */
 139static inline int shmem_acct_size(unsigned long flags, loff_t size)
 140{
 141	return (flags & VM_NORESERVE) ?
 142		0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
 143}
 144
 145static inline void shmem_unacct_size(unsigned long flags, loff_t size)
 146{
 147	if (!(flags & VM_NORESERVE))
 148		vm_unacct_memory(VM_ACCT(size));
 149}
 150
 
 
 
 
 
 
 
 
 
 
 
 
 
 151/*
 152 * ... whereas tmpfs objects are accounted incrementally as
 153 * pages are allocated, in order to allow huge sparse files.
 154 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
 155 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
 156 */
 157static inline int shmem_acct_block(unsigned long flags)
 158{
 159	return (flags & VM_NORESERVE) ?
 160		security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0;
 
 
 
 161}
 162
 163static inline void shmem_unacct_blocks(unsigned long flags, long pages)
 164{
 165	if (flags & VM_NORESERVE)
 166		vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 167}
 168
 169static const struct super_operations shmem_ops;
 170static const struct address_space_operations shmem_aops;
 171static const struct file_operations shmem_file_operations;
 172static const struct inode_operations shmem_inode_operations;
 173static const struct inode_operations shmem_dir_inode_operations;
 174static const struct inode_operations shmem_special_inode_operations;
 175static const struct vm_operations_struct shmem_vm_ops;
 
 176
 177static struct backing_dev_info shmem_backing_dev_info  __read_mostly = {
 178	.ra_pages	= 0,	/* No readahead */
 179	.capabilities	= BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
 180};
 181
 182static LIST_HEAD(shmem_swaplist);
 183static DEFINE_MUTEX(shmem_swaplist_mutex);
 184
 185static int shmem_reserve_inode(struct super_block *sb)
 
 
 
 
 
 
 
 
 
 
 186{
 187	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 188	if (sbinfo->max_inodes) {
 
 
 189		spin_lock(&sbinfo->stat_lock);
 190		if (!sbinfo->free_inodes) {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 191			spin_unlock(&sbinfo->stat_lock);
 192			return -ENOSPC;
 
 193		}
 194		sbinfo->free_inodes--;
 195		spin_unlock(&sbinfo->stat_lock);
 
 196	}
 
 197	return 0;
 198}
 199
 200static void shmem_free_inode(struct super_block *sb)
 201{
 202	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 203	if (sbinfo->max_inodes) {
 204		spin_lock(&sbinfo->stat_lock);
 205		sbinfo->free_inodes++;
 206		spin_unlock(&sbinfo->stat_lock);
 207	}
 208}
 209
 210/**
 211 * shmem_recalc_inode - recalculate the block usage of an inode
 212 * @inode: inode to recalc
 213 *
 214 * We have to calculate the free blocks since the mm can drop
 215 * undirtied hole pages behind our back.
 216 *
 217 * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
 218 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
 219 *
 220 * It has to be called with the spinlock held.
 221 */
 222static void shmem_recalc_inode(struct inode *inode)
 223{
 224	struct shmem_inode_info *info = SHMEM_I(inode);
 225	long freed;
 226
 227	freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
 228	if (freed > 0) {
 229		struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 230		if (sbinfo->max_blocks)
 231			percpu_counter_add(&sbinfo->used_blocks, -freed);
 232		info->alloced -= freed;
 233		inode->i_blocks -= freed * BLOCKS_PER_PAGE;
 234		shmem_unacct_blocks(info->flags, freed);
 235	}
 236}
 237
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 238/*
 239 * Replace item expected in radix tree by a new item, while holding tree lock.
 240 */
 241static int shmem_radix_tree_replace(struct address_space *mapping,
 242			pgoff_t index, void *expected, void *replacement)
 243{
 244	void **pslot;
 245	void *item;
 246
 247	VM_BUG_ON(!expected);
 248	VM_BUG_ON(!replacement);
 249	pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
 250	if (!pslot)
 251		return -ENOENT;
 252	item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock);
 253	if (item != expected)
 254		return -ENOENT;
 255	radix_tree_replace_slot(pslot, replacement);
 256	return 0;
 257}
 258
 259/*
 260 * Sometimes, before we decide whether to proceed or to fail, we must check
 261 * that an entry was not already brought back from swap by a racing thread.
 262 *
 263 * Checking page is not enough: by the time a SwapCache page is locked, it
 264 * might be reused, and again be SwapCache, using the same swap as before.
 265 */
 266static bool shmem_confirm_swap(struct address_space *mapping,
 267			       pgoff_t index, swp_entry_t swap)
 268{
 269	void *item;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 270
 271	rcu_read_lock();
 272	item = radix_tree_lookup(&mapping->page_tree, index);
 273	rcu_read_unlock();
 274	return item == swp_to_radix_entry(swap);
 
 
 
 275}
 276
 277/*
 278 * Like add_to_page_cache_locked, but error if expected item has gone.
 279 */
 280static int shmem_add_to_page_cache(struct page *page,
 281				   struct address_space *mapping,
 282				   pgoff_t index, gfp_t gfp, void *expected)
 
 283{
 
 
 
 284	int error;
 285
 
 
 286	VM_BUG_ON_PAGE(!PageLocked(page), page);
 287	VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
 
 288
 289	page_cache_get(page);
 290	page->mapping = mapping;
 291	page->index = index;
 292
 293	spin_lock_irq(&mapping->tree_lock);
 294	if (!expected)
 295		error = radix_tree_insert(&mapping->page_tree, index, page);
 296	else
 297		error = shmem_radix_tree_replace(mapping, index, expected,
 298								 page);
 299	if (!error) {
 300		mapping->nrpages++;
 301		__inc_zone_page_state(page, NR_FILE_PAGES);
 302		__inc_zone_page_state(page, NR_SHMEM);
 303		spin_unlock_irq(&mapping->tree_lock);
 304	} else {
 305		page->mapping = NULL;
 306		spin_unlock_irq(&mapping->tree_lock);
 307		page_cache_release(page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 308	}
 
 
 
 
 
 309	return error;
 310}
 311
 312/*
 313 * Like delete_from_page_cache, but substitutes swap for page.
 314 */
 315static void shmem_delete_from_page_cache(struct page *page, void *radswap)
 316{
 317	struct address_space *mapping = page->mapping;
 318	int error;
 319
 320	spin_lock_irq(&mapping->tree_lock);
 321	error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
 
 
 322	page->mapping = NULL;
 323	mapping->nrpages--;
 324	__dec_zone_page_state(page, NR_FILE_PAGES);
 325	__dec_zone_page_state(page, NR_SHMEM);
 326	spin_unlock_irq(&mapping->tree_lock);
 327	page_cache_release(page);
 328	BUG_ON(error);
 329}
 330
 331/*
 332 * Remove swap entry from radix tree, free the swap and its page cache.
 333 */
 334static int shmem_free_swap(struct address_space *mapping,
 335			   pgoff_t index, void *radswap)
 336{
 337	void *old;
 338
 339	spin_lock_irq(&mapping->tree_lock);
 340	old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
 341	spin_unlock_irq(&mapping->tree_lock);
 342	if (old != radswap)
 343		return -ENOENT;
 344	free_swap_and_cache(radix_to_swp_entry(radswap));
 345	return 0;
 346}
 347
 348/*
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 349 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
 350 */
 351void shmem_unlock_mapping(struct address_space *mapping)
 352{
 353	struct pagevec pvec;
 354	pgoff_t indices[PAGEVEC_SIZE];
 355	pgoff_t index = 0;
 356
 357	pagevec_init(&pvec, 0);
 358	/*
 359	 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
 360	 */
 361	while (!mapping_unevictable(mapping)) {
 362		/*
 363		 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
 364		 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
 365		 */
 366		pvec.nr = find_get_entries(mapping, index,
 367					   PAGEVEC_SIZE, pvec.pages, indices);
 368		if (!pvec.nr)
 369			break;
 370		index = indices[pvec.nr - 1] + 1;
 371		pagevec_remove_exceptionals(&pvec);
 372		check_move_unevictable_pages(pvec.pages, pvec.nr);
 373		pagevec_release(&pvec);
 374		cond_resched();
 375	}
 376}
 377
 378/*
 379 * Remove range of pages and swap entries from radix tree, and free them.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 380 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
 381 */
 382static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
 383								 bool unfalloc)
 384{
 385	struct address_space *mapping = inode->i_mapping;
 386	struct shmem_inode_info *info = SHMEM_I(inode);
 387	pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 388	pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
 389	unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
 390	unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
 391	struct pagevec pvec;
 392	pgoff_t indices[PAGEVEC_SIZE];
 393	long nr_swaps_freed = 0;
 394	pgoff_t index;
 395	int i;
 396
 397	if (lend == -1)
 398		end = -1;	/* unsigned, so actually very big */
 399
 400	pagevec_init(&pvec, 0);
 401	index = start;
 402	while (index < end) {
 403		pvec.nr = find_get_entries(mapping, index,
 404			min(end - index, (pgoff_t)PAGEVEC_SIZE),
 405			pvec.pages, indices);
 406		if (!pvec.nr)
 407			break;
 408		mem_cgroup_uncharge_start();
 409		for (i = 0; i < pagevec_count(&pvec); i++) {
 410			struct page *page = pvec.pages[i];
 411
 412			index = indices[i];
 413			if (index >= end)
 414				break;
 415
 416			if (radix_tree_exceptional_entry(page)) {
 417				if (unfalloc)
 418					continue;
 419				nr_swaps_freed += !shmem_free_swap(mapping,
 420								index, page);
 421				continue;
 422			}
 423
 
 
 424			if (!trylock_page(page))
 425				continue;
 426			if (!unfalloc || !PageUptodate(page)) {
 427				if (page->mapping == mapping) {
 428					VM_BUG_ON_PAGE(PageWriteback(page), page);
 
 
 429					truncate_inode_page(mapping, page);
 430				}
 431			}
 432			unlock_page(page);
 433		}
 434		pagevec_remove_exceptionals(&pvec);
 435		pagevec_release(&pvec);
 436		mem_cgroup_uncharge_end();
 437		cond_resched();
 438		index++;
 439	}
 440
 441	if (partial_start) {
 442		struct page *page = NULL;
 443		shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
 444		if (page) {
 445			unsigned int top = PAGE_CACHE_SIZE;
 446			if (start > end) {
 447				top = partial_end;
 448				partial_end = 0;
 449			}
 450			zero_user_segment(page, partial_start, top);
 451			set_page_dirty(page);
 452			unlock_page(page);
 453			page_cache_release(page);
 454		}
 455	}
 456	if (partial_end) {
 457		struct page *page = NULL;
 458		shmem_getpage(inode, end, &page, SGP_READ, NULL);
 459		if (page) {
 460			zero_user_segment(page, 0, partial_end);
 461			set_page_dirty(page);
 462			unlock_page(page);
 463			page_cache_release(page);
 464		}
 465	}
 466	if (start >= end)
 467		return;
 468
 469	index = start;
 470	for ( ; ; ) {
 471		cond_resched();
 472
 473		pvec.nr = find_get_entries(mapping, index,
 474				min(end - index, (pgoff_t)PAGEVEC_SIZE),
 475				pvec.pages, indices);
 476		if (!pvec.nr) {
 477			if (index == start || unfalloc)
 
 478				break;
 
 479			index = start;
 480			continue;
 481		}
 482		if ((index == start || unfalloc) && indices[0] >= end) {
 483			pagevec_remove_exceptionals(&pvec);
 484			pagevec_release(&pvec);
 485			break;
 486		}
 487		mem_cgroup_uncharge_start();
 488		for (i = 0; i < pagevec_count(&pvec); i++) {
 489			struct page *page = pvec.pages[i];
 490
 491			index = indices[i];
 492			if (index >= end)
 493				break;
 494
 495			if (radix_tree_exceptional_entry(page)) {
 496				if (unfalloc)
 497					continue;
 498				nr_swaps_freed += !shmem_free_swap(mapping,
 499								index, page);
 
 
 
 
 500				continue;
 501			}
 502
 503			lock_page(page);
 
 504			if (!unfalloc || !PageUptodate(page)) {
 505				if (page->mapping == mapping) {
 506					VM_BUG_ON_PAGE(PageWriteback(page), page);
 
 
 
 
 
 
 507					truncate_inode_page(mapping, page);
 
 
 
 
 
 
 
 
 508				}
 509			}
 510			unlock_page(page);
 511		}
 512		pagevec_remove_exceptionals(&pvec);
 513		pagevec_release(&pvec);
 514		mem_cgroup_uncharge_end();
 515		index++;
 516	}
 517
 518	spin_lock(&info->lock);
 519	info->swapped -= nr_swaps_freed;
 520	shmem_recalc_inode(inode);
 521	spin_unlock(&info->lock);
 522}
 523
 524void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
 525{
 526	shmem_undo_range(inode, lstart, lend, false);
 527	inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 528}
 529EXPORT_SYMBOL_GPL(shmem_truncate_range);
 530
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 531static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
 532{
 533	struct inode *inode = dentry->d_inode;
 
 
 534	int error;
 535
 536	error = inode_change_ok(inode, attr);
 537	if (error)
 538		return error;
 539
 540	if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
 541		loff_t oldsize = inode->i_size;
 542		loff_t newsize = attr->ia_size;
 543
 
 
 
 
 
 544		if (newsize != oldsize) {
 
 
 
 
 545			i_size_write(inode, newsize);
 546			inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 547		}
 548		if (newsize < oldsize) {
 549			loff_t holebegin = round_up(newsize, PAGE_SIZE);
 550			unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
 551			shmem_truncate_range(inode, newsize, (loff_t)-1);
 
 
 
 
 552			/* unmap again to remove racily COWed private pages */
 553			unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 554		}
 555	}
 556
 557	setattr_copy(inode, attr);
 558	if (attr->ia_valid & ATTR_MODE)
 559		error = posix_acl_chmod(inode, inode->i_mode);
 560	return error;
 561}
 562
 563static void shmem_evict_inode(struct inode *inode)
 564{
 565	struct shmem_inode_info *info = SHMEM_I(inode);
 
 566
 567	if (inode->i_mapping->a_ops == &shmem_aops) {
 568		shmem_unacct_size(info->flags, inode->i_size);
 569		inode->i_size = 0;
 570		shmem_truncate_range(inode, 0, (loff_t)-1);
 571		if (!list_empty(&info->swaplist)) {
 
 
 
 
 
 
 
 
 
 
 
 572			mutex_lock(&shmem_swaplist_mutex);
 573			list_del_init(&info->swaplist);
 
 
 574			mutex_unlock(&shmem_swaplist_mutex);
 575		}
 576	} else
 577		kfree(info->symlink);
 578
 579	simple_xattrs_free(&info->xattrs);
 580	WARN_ON(inode->i_blocks);
 581	shmem_free_inode(inode->i_sb);
 582	clear_inode(inode);
 583}
 584
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 585/*
 586 * If swap found in inode, free it and move page from swapcache to filecache.
 
 587 */
 588static int shmem_unuse_inode(struct shmem_inode_info *info,
 589			     swp_entry_t swap, struct page **pagep)
 590{
 591	struct address_space *mapping = info->vfs_inode.i_mapping;
 592	void *radswap;
 593	pgoff_t index;
 594	gfp_t gfp;
 595	int error = 0;
 
 596
 597	radswap = swp_to_radix_entry(swap);
 598	index = radix_tree_locate_item(&mapping->page_tree, radswap);
 599	if (index == -1)
 600		return 0;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 601
 602	/*
 603	 * Move _head_ to start search for next from here.
 604	 * But be careful: shmem_evict_inode checks list_empty without taking
 605	 * mutex, and there's an instant in list_move_tail when info->swaplist
 606	 * would appear empty, if it were the only one on shmem_swaplist.
 607	 */
 608	if (shmem_swaplist.next != &info->swaplist)
 609		list_move_tail(&shmem_swaplist, &info->swaplist);
 
 
 
 
 
 
 610
 611	gfp = mapping_gfp_mask(mapping);
 612	if (shmem_should_replace_page(*pagep, gfp)) {
 613		mutex_unlock(&shmem_swaplist_mutex);
 614		error = shmem_replace_page(pagep, gfp, info, index);
 615		mutex_lock(&shmem_swaplist_mutex);
 616		/*
 617		 * We needed to drop mutex to make that restrictive page
 618		 * allocation, but the inode might have been freed while we
 619		 * dropped it: although a racing shmem_evict_inode() cannot
 620		 * complete without emptying the radix_tree, our page lock
 621		 * on this swapcache page is not enough to prevent that -
 622		 * free_swap_and_cache() of our swap entry will only
 623		 * trylock_page(), removing swap from radix_tree whatever.
 624		 *
 625		 * We must not proceed to shmem_add_to_page_cache() if the
 626		 * inode has been freed, but of course we cannot rely on
 627		 * inode or mapping or info to check that.  However, we can
 628		 * safely check if our swap entry is still in use (and here
 629		 * it can't have got reused for another page): if it's still
 630		 * in use, then the inode cannot have been freed yet, and we
 631		 * can safely proceed (if it's no longer in use, that tells
 632		 * nothing about the inode, but we don't need to unuse swap).
 633		 */
 634		if (!page_swapcount(*pagep))
 635			error = -ENOENT;
 636	}
 637
 638	/*
 639	 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
 640	 * but also to hold up shmem_evict_inode(): so inode cannot be freed
 641	 * beneath us (pagelock doesn't help until the page is in pagecache).
 642	 */
 643	if (!error)
 644		error = shmem_add_to_page_cache(*pagep, mapping, index,
 645						GFP_NOWAIT, radswap);
 646	if (error != -ENOMEM) {
 647		/*
 648		 * Truncation and eviction use free_swap_and_cache(), which
 649		 * only does trylock page: if we raced, best clean up here.
 650		 */
 651		delete_from_swap_cache(*pagep);
 652		set_page_dirty(*pagep);
 653		if (!error) {
 654			spin_lock(&info->lock);
 655			info->swapped--;
 656			spin_unlock(&info->lock);
 657			swap_free(swap);
 658		}
 659		error = 1;	/* not an error, but entry was found */
 660	}
 661	return error;
 
 
 662}
 663
 664/*
 665 * Search through swapped inodes to find and replace swap by page.
 
 
 666 */
 667int shmem_unuse(swp_entry_t swap, struct page *page)
 
 668{
 669	struct list_head *this, *next;
 670	struct shmem_inode_info *info;
 671	int found = 0;
 672	int error = 0;
 673
 674	/*
 675	 * There's a faint possibility that swap page was replaced before
 676	 * caller locked it: caller will come back later with the right page.
 677	 */
 678	if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
 679		goto out;
 680
 681	/*
 682	 * Charge page using GFP_KERNEL while we can wait, before taking
 683	 * the shmem_swaplist_mutex which might hold up shmem_writepage().
 684	 * Charged back to the user (not to caller) when swap account is used.
 685	 */
 686	error = mem_cgroup_charge_file(page, current->mm, GFP_KERNEL);
 687	if (error)
 688		goto out;
 689	/* No radix_tree_preload: swap entry keeps a place for page in tree */
 690
 691	mutex_lock(&shmem_swaplist_mutex);
 692	list_for_each_safe(this, next, &shmem_swaplist) {
 693		info = list_entry(this, struct shmem_inode_info, swaplist);
 694		if (info->swapped)
 695			found = shmem_unuse_inode(info, swap, &page);
 696		else
 697			list_del_init(&info->swaplist);
 
 
 
 
 
 
 
 
 
 
 
 
 
 698		cond_resched();
 699		if (found)
 
 
 
 
 
 
 
 700			break;
 701	}
 702	mutex_unlock(&shmem_swaplist_mutex);
 703
 704	if (found < 0)
 705		error = found;
 706out:
 707	unlock_page(page);
 708	page_cache_release(page);
 709	return error;
 710}
 711
 712/*
 713 * Move the page from the page cache to the swap cache.
 714 */
 715static int shmem_writepage(struct page *page, struct writeback_control *wbc)
 716{
 717	struct shmem_inode_info *info;
 718	struct address_space *mapping;
 719	struct inode *inode;
 720	swp_entry_t swap;
 721	pgoff_t index;
 722
 
 723	BUG_ON(!PageLocked(page));
 724	mapping = page->mapping;
 725	index = page->index;
 726	inode = mapping->host;
 727	info = SHMEM_I(inode);
 728	if (info->flags & VM_LOCKED)
 729		goto redirty;
 730	if (!total_swap_pages)
 731		goto redirty;
 732
 733	/*
 734	 * shmem_backing_dev_info's capabilities prevent regular writeback or
 735	 * sync from ever calling shmem_writepage; but a stacking filesystem
 736	 * might use ->writepage of its underlying filesystem, in which case
 737	 * tmpfs should write out to swap only in response to memory pressure,
 738	 * and not for the writeback threads or sync.
 739	 */
 740	if (!wbc->for_reclaim) {
 741		WARN_ON_ONCE(1);	/* Still happens? Tell us about it! */
 742		goto redirty;
 743	}
 744
 745	/*
 746	 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
 747	 * value into swapfile.c, the only way we can correctly account for a
 748	 * fallocated page arriving here is now to initialize it and write it.
 749	 *
 750	 * That's okay for a page already fallocated earlier, but if we have
 751	 * not yet completed the fallocation, then (a) we want to keep track
 752	 * of this page in case we have to undo it, and (b) it may not be a
 753	 * good idea to continue anyway, once we're pushing into swap.  So
 754	 * reactivate the page, and let shmem_fallocate() quit when too many.
 755	 */
 756	if (!PageUptodate(page)) {
 757		if (inode->i_private) {
 758			struct shmem_falloc *shmem_falloc;
 759			spin_lock(&inode->i_lock);
 760			shmem_falloc = inode->i_private;
 761			if (shmem_falloc &&
 
 762			    index >= shmem_falloc->start &&
 763			    index < shmem_falloc->next)
 764				shmem_falloc->nr_unswapped++;
 765			else
 766				shmem_falloc = NULL;
 767			spin_unlock(&inode->i_lock);
 768			if (shmem_falloc)
 769				goto redirty;
 770		}
 771		clear_highpage(page);
 772		flush_dcache_page(page);
 773		SetPageUptodate(page);
 774	}
 775
 776	swap = get_swap_page();
 777	if (!swap.val)
 778		goto redirty;
 779
 780	/*
 781	 * Add inode to shmem_unuse()'s list of swapped-out inodes,
 782	 * if it's not already there.  Do it now before the page is
 783	 * moved to swap cache, when its pagelock no longer protects
 784	 * the inode from eviction.  But don't unlock the mutex until
 785	 * we've incremented swapped, because shmem_unuse_inode() will
 786	 * prune a !swapped inode from the swaplist under this mutex.
 787	 */
 788	mutex_lock(&shmem_swaplist_mutex);
 789	if (list_empty(&info->swaplist))
 790		list_add_tail(&info->swaplist, &shmem_swaplist);
 
 
 
 
 
 
 
 
 791
 792	if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
 793		swap_shmem_alloc(swap);
 794		shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
 795
 796		spin_lock(&info->lock);
 797		info->swapped++;
 798		shmem_recalc_inode(inode);
 799		spin_unlock(&info->lock);
 800
 801		mutex_unlock(&shmem_swaplist_mutex);
 802		BUG_ON(page_mapped(page));
 803		swap_writepage(page, wbc);
 804		return 0;
 805	}
 806
 807	mutex_unlock(&shmem_swaplist_mutex);
 808	swapcache_free(swap, NULL);
 809redirty:
 810	set_page_dirty(page);
 811	if (wbc->for_reclaim)
 812		return AOP_WRITEPAGE_ACTIVATE;	/* Return with page locked */
 813	unlock_page(page);
 814	return 0;
 815}
 816
 817#ifdef CONFIG_NUMA
 818#ifdef CONFIG_TMPFS
 819static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
 820{
 821	char buffer[64];
 822
 823	if (!mpol || mpol->mode == MPOL_DEFAULT)
 824		return;		/* show nothing */
 825
 826	mpol_to_str(buffer, sizeof(buffer), mpol);
 827
 828	seq_printf(seq, ",mpol=%s", buffer);
 829}
 830
 831static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
 832{
 833	struct mempolicy *mpol = NULL;
 834	if (sbinfo->mpol) {
 835		spin_lock(&sbinfo->stat_lock);	/* prevent replace/use races */
 836		mpol = sbinfo->mpol;
 837		mpol_get(mpol);
 838		spin_unlock(&sbinfo->stat_lock);
 839	}
 840	return mpol;
 841}
 842#endif /* CONFIG_TMPFS */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 843
 844static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
 845			struct shmem_inode_info *info, pgoff_t index)
 846{
 847	struct vm_area_struct pvma;
 848	struct page *page;
 
 849
 850	/* Create a pseudo vma that just contains the policy */
 851	pvma.vm_start = 0;
 852	/* Bias interleave by inode number to distribute better across nodes */
 853	pvma.vm_pgoff = index + info->vfs_inode.i_ino;
 854	pvma.vm_ops = NULL;
 855	pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
 856
 857	page = swapin_readahead(swap, gfp, &pvma, 0);
 
 858
 859	/* Drop reference taken by mpol_shared_policy_lookup() */
 860	mpol_cond_put(pvma.vm_policy);
 
 
 
 
 
 
 
 
 
 
 861
 
 
 
 
 
 
 
 
 862	return page;
 863}
 864
 865static struct page *shmem_alloc_page(gfp_t gfp,
 866			struct shmem_inode_info *info, pgoff_t index)
 867{
 868	struct vm_area_struct pvma;
 869	struct page *page;
 870
 871	/* Create a pseudo vma that just contains the policy */
 872	pvma.vm_start = 0;
 873	/* Bias interleave by inode number to distribute better across nodes */
 874	pvma.vm_pgoff = index + info->vfs_inode.i_ino;
 875	pvma.vm_ops = NULL;
 876	pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
 877
 878	page = alloc_page_vma(gfp, &pvma, 0);
 879
 880	/* Drop reference taken by mpol_shared_policy_lookup() */
 881	mpol_cond_put(pvma.vm_policy);
 882
 883	return page;
 884}
 885#else /* !CONFIG_NUMA */
 886#ifdef CONFIG_TMPFS
 887static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
 
 888{
 889}
 890#endif /* CONFIG_TMPFS */
 
 
 
 
 
 
 891
 892static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
 893			struct shmem_inode_info *info, pgoff_t index)
 894{
 895	return swapin_readahead(swap, gfp, NULL, 0);
 896}
 897
 898static inline struct page *shmem_alloc_page(gfp_t gfp,
 899			struct shmem_inode_info *info, pgoff_t index)
 900{
 901	return alloc_page(gfp);
 902}
 903#endif /* CONFIG_NUMA */
 
 
 
 904
 905#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
 906static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
 907{
 908	return NULL;
 909}
 910#endif
 911
 912/*
 913 * When a page is moved from swapcache to shmem filecache (either by the
 914 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
 915 * shmem_unuse_inode()), it may have been read in earlier from swap, in
 916 * ignorance of the mapping it belongs to.  If that mapping has special
 917 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
 918 * we may need to copy to a suitable page before moving to filecache.
 919 *
 920 * In a future release, this may well be extended to respect cpuset and
 921 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
 922 * but for now it is a simple matter of zone.
 923 */
 924static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
 925{
 926	return page_zonenum(page) > gfp_zone(gfp);
 927}
 928
 929static int shmem_replace_page(struct page **pagep, gfp_t gfp,
 930				struct shmem_inode_info *info, pgoff_t index)
 931{
 932	struct page *oldpage, *newpage;
 933	struct address_space *swap_mapping;
 
 934	pgoff_t swap_index;
 935	int error;
 936
 937	oldpage = *pagep;
 938	swap_index = page_private(oldpage);
 
 939	swap_mapping = page_mapping(oldpage);
 940
 941	/*
 942	 * We have arrived here because our zones are constrained, so don't
 943	 * limit chance of success by further cpuset and node constraints.
 944	 */
 945	gfp &= ~GFP_CONSTRAINT_MASK;
 946	newpage = shmem_alloc_page(gfp, info, index);
 947	if (!newpage)
 948		return -ENOMEM;
 949
 950	page_cache_get(newpage);
 951	copy_highpage(newpage, oldpage);
 952	flush_dcache_page(newpage);
 953
 954	__set_page_locked(newpage);
 
 955	SetPageUptodate(newpage);
 956	SetPageSwapBacked(newpage);
 957	set_page_private(newpage, swap_index);
 958	SetPageSwapCache(newpage);
 959
 960	/*
 961	 * Our caller will very soon move newpage out of swapcache, but it's
 962	 * a nice clean interface for us to replace oldpage by newpage there.
 963	 */
 964	spin_lock_irq(&swap_mapping->tree_lock);
 965	error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
 966								   newpage);
 967	if (!error) {
 968		__inc_zone_page_state(newpage, NR_FILE_PAGES);
 969		__dec_zone_page_state(oldpage, NR_FILE_PAGES);
 
 970	}
 971	spin_unlock_irq(&swap_mapping->tree_lock);
 972
 973	if (unlikely(error)) {
 974		/*
 975		 * Is this possible?  I think not, now that our callers check
 976		 * both PageSwapCache and page_private after getting page lock;
 977		 * but be defensive.  Reverse old to newpage for clear and free.
 978		 */
 979		oldpage = newpage;
 980	} else {
 981		mem_cgroup_replace_page_cache(oldpage, newpage);
 982		lru_cache_add_anon(newpage);
 983		*pagep = newpage;
 984	}
 985
 986	ClearPageSwapCache(oldpage);
 987	set_page_private(oldpage, 0);
 988
 989	unlock_page(oldpage);
 990	page_cache_release(oldpage);
 991	page_cache_release(oldpage);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 992	return error;
 993}
 994
 995/*
 996 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
 997 *
 998 * If we allocate a new one we do not mark it dirty. That's up to the
 999 * vm. If we swap it in we mark it dirty since we also free the swap
1000 * entry since a page cannot live in both the swap and page cache
 
 
 
1001 */
1002static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1003	struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
 
 
1004{
1005	struct address_space *mapping = inode->i_mapping;
1006	struct shmem_inode_info *info;
1007	struct shmem_sb_info *sbinfo;
 
1008	struct page *page;
1009	swp_entry_t swap;
 
1010	int error;
1011	int once = 0;
1012	int alloced = 0;
1013
1014	if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
1015		return -EFBIG;
 
 
1016repeat:
1017	swap.val = 0;
 
 
 
 
 
 
 
1018	page = find_lock_entry(mapping, index);
1019	if (radix_tree_exceptional_entry(page)) {
1020		swap = radix_to_swp_entry(page);
1021		page = NULL;
 
 
 
 
 
1022	}
1023
1024	if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1025	    ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1026		error = -EINVAL;
1027		goto failed;
1028	}
1029
1030	/* fallocated page? */
1031	if (page && !PageUptodate(page)) {
1032		if (sgp != SGP_READ)
1033			goto clear;
1034		unlock_page(page);
1035		page_cache_release(page);
1036		page = NULL;
1037	}
1038	if (page || (sgp == SGP_READ && !swap.val)) {
1039		*pagep = page;
1040		return 0;
1041	}
1042
1043	/*
1044	 * Fast cache lookup did not find it:
1045	 * bring it back from swap or allocate.
1046	 */
1047	info = SHMEM_I(inode);
1048	sbinfo = SHMEM_SB(inode->i_sb);
1049
1050	if (swap.val) {
1051		/* Look it up and read it in.. */
1052		page = lookup_swap_cache(swap);
1053		if (!page) {
1054			/* here we actually do the io */
1055			if (fault_type)
1056				*fault_type |= VM_FAULT_MAJOR;
1057			page = shmem_swapin(swap, gfp, info, index);
1058			if (!page) {
1059				error = -ENOMEM;
1060				goto failed;
1061			}
1062		}
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1063
1064		/* We have to do this with page locked to prevent races */
1065		lock_page(page);
1066		if (!PageSwapCache(page) || page_private(page) != swap.val ||
1067		    !shmem_confirm_swap(mapping, index, swap)) {
1068			error = -EEXIST;	/* try again */
1069			goto unlock;
1070		}
1071		if (!PageUptodate(page)) {
1072			error = -EIO;
1073			goto failed;
1074		}
1075		wait_on_page_writeback(page);
1076
1077		if (shmem_should_replace_page(page, gfp)) {
1078			error = shmem_replace_page(&page, gfp, info, index);
1079			if (error)
1080				goto failed;
 
1081		}
 
 
1082
1083		error = mem_cgroup_charge_file(page, current->mm,
1084						gfp & GFP_RECLAIM_MASK);
1085		if (!error) {
1086			error = shmem_add_to_page_cache(page, mapping, index,
1087						gfp, swp_to_radix_entry(swap));
1088			/*
1089			 * We already confirmed swap under page lock, and make
1090			 * no memory allocation here, so usually no possibility
1091			 * of error; but free_swap_and_cache() only trylocks a
1092			 * page, so it is just possible that the entry has been
1093			 * truncated or holepunched since swap was confirmed.
1094			 * shmem_undo_range() will have done some of the
1095			 * unaccounting, now delete_from_swap_cache() will do
1096			 * the rest (including mem_cgroup_uncharge_swapcache).
1097			 * Reset swap.val? No, leave it so "failed" goes back to
1098			 * "repeat": reading a hole and writing should succeed.
1099			 */
1100			if (error)
1101				delete_from_swap_cache(page);
1102		}
1103		if (error)
1104			goto failed;
1105
1106		spin_lock(&info->lock);
1107		info->swapped--;
1108		shmem_recalc_inode(inode);
1109		spin_unlock(&info->lock);
1110
1111		delete_from_swap_cache(page);
1112		set_page_dirty(page);
1113		swap_free(swap);
 
 
 
1114
1115	} else {
1116		if (shmem_acct_block(info->flags)) {
1117			error = -ENOSPC;
1118			goto failed;
1119		}
1120		if (sbinfo->max_blocks) {
1121			if (percpu_counter_compare(&sbinfo->used_blocks,
1122						sbinfo->max_blocks) >= 0) {
1123				error = -ENOSPC;
1124				goto unacct;
1125			}
1126			percpu_counter_inc(&sbinfo->used_blocks);
1127		}
1128
1129		page = shmem_alloc_page(gfp, info, index);
1130		if (!page) {
1131			error = -ENOMEM;
1132			goto decused;
1133		}
1134
1135		SetPageSwapBacked(page);
1136		__set_page_locked(page);
1137		error = mem_cgroup_charge_file(page, current->mm,
1138						gfp & GFP_RECLAIM_MASK);
1139		if (error)
1140			goto decused;
1141		error = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK);
1142		if (!error) {
1143			error = shmem_add_to_page_cache(page, mapping, index,
1144							gfp, NULL);
1145			radix_tree_preload_end();
1146		}
1147		if (error) {
1148			mem_cgroup_uncharge_cache_page(page);
1149			goto decused;
1150		}
1151		lru_cache_add_anon(page);
1152
1153		spin_lock(&info->lock);
1154		info->alloced++;
1155		inode->i_blocks += BLOCKS_PER_PAGE;
1156		shmem_recalc_inode(inode);
1157		spin_unlock(&info->lock);
1158		alloced = true;
1159
 
 
 
1160		/*
1161		 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
 
1162		 */
1163		if (sgp == SGP_FALLOC)
1164			sgp = SGP_WRITE;
1165clear:
1166		/*
1167		 * Let SGP_WRITE caller clear ends if write does not fill page;
1168		 * but SGP_FALLOC on a page fallocated earlier must initialize
1169		 * it now, lest undo on failure cancel our earlier guarantee.
1170		 */
1171		if (sgp != SGP_WRITE) {
1172			clear_highpage(page);
1173			flush_dcache_page(page);
1174			SetPageUptodate(page);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1175		}
1176		if (sgp == SGP_DIRTY)
1177			set_page_dirty(page);
1178	}
1179
1180	/* Perhaps the file has been truncated since we checked */
1181	if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1182	    ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
 
 
 
 
 
 
 
1183		error = -EINVAL;
1184		if (alloced)
1185			goto trunc;
1186		else
1187			goto failed;
1188	}
1189	*pagep = page;
1190	return 0;
1191
1192	/*
1193	 * Error recovery.
1194	 */
1195trunc:
1196	info = SHMEM_I(inode);
1197	ClearPageDirty(page);
1198	delete_from_page_cache(page);
1199	spin_lock(&info->lock);
1200	info->alloced--;
1201	inode->i_blocks -= BLOCKS_PER_PAGE;
1202	spin_unlock(&info->lock);
1203decused:
1204	sbinfo = SHMEM_SB(inode->i_sb);
1205	if (sbinfo->max_blocks)
1206		percpu_counter_add(&sbinfo->used_blocks, -1);
1207unacct:
1208	shmem_unacct_blocks(info->flags, 1);
1209failed:
1210	if (swap.val && error != -EINVAL &&
1211	    !shmem_confirm_swap(mapping, index, swap))
1212		error = -EEXIST;
 
 
1213unlock:
1214	if (page) {
1215		unlock_page(page);
1216		page_cache_release(page);
1217	}
1218	if (error == -ENOSPC && !once++) {
1219		info = SHMEM_I(inode);
1220		spin_lock(&info->lock);
1221		shmem_recalc_inode(inode);
1222		spin_unlock(&info->lock);
1223		goto repeat;
1224	}
1225	if (error == -EEXIST)	/* from above or from radix_tree_insert */
1226		goto repeat;
1227	return error;
1228}
1229
1230static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 
 
 
 
 
 
 
 
 
 
 
 
1231{
 
1232	struct inode *inode = file_inode(vma->vm_file);
1233	int error;
1234	int ret = VM_FAULT_LOCKED;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1235
1236	error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1237	if (error)
1238		return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
 
 
 
 
 
 
 
1239
1240	if (ret & VM_FAULT_MAJOR) {
1241		count_vm_event(PGMAJFAULT);
1242		mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
 
 
1243	}
 
 
 
 
 
 
 
 
 
 
 
 
 
1244	return ret;
1245}
1246
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1247#ifdef CONFIG_NUMA
1248static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
1249{
1250	struct inode *inode = file_inode(vma->vm_file);
1251	return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
1252}
1253
1254static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1255					  unsigned long addr)
1256{
1257	struct inode *inode = file_inode(vma->vm_file);
1258	pgoff_t index;
1259
1260	index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1261	return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
1262}
1263#endif
1264
1265int shmem_lock(struct file *file, int lock, struct user_struct *user)
1266{
1267	struct inode *inode = file_inode(file);
1268	struct shmem_inode_info *info = SHMEM_I(inode);
1269	int retval = -ENOMEM;
1270
1271	spin_lock(&info->lock);
 
 
 
 
1272	if (lock && !(info->flags & VM_LOCKED)) {
1273		if (!user_shm_lock(inode->i_size, user))
1274			goto out_nomem;
1275		info->flags |= VM_LOCKED;
1276		mapping_set_unevictable(file->f_mapping);
1277	}
1278	if (!lock && (info->flags & VM_LOCKED) && user) {
1279		user_shm_unlock(inode->i_size, user);
1280		info->flags &= ~VM_LOCKED;
1281		mapping_clear_unevictable(file->f_mapping);
1282	}
1283	retval = 0;
1284
1285out_nomem:
1286	spin_unlock(&info->lock);
1287	return retval;
1288}
1289
1290static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1291{
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1292	file_accessed(file);
1293	vma->vm_ops = &shmem_vm_ops;
 
 
 
 
 
1294	return 0;
1295}
1296
1297static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1298				     umode_t mode, dev_t dev, unsigned long flags)
1299{
1300	struct inode *inode;
1301	struct shmem_inode_info *info;
1302	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 
1303
1304	if (shmem_reserve_inode(sb))
1305		return NULL;
1306
1307	inode = new_inode(sb);
1308	if (inode) {
1309		inode->i_ino = get_next_ino();
1310		inode_init_owner(inode, dir, mode);
1311		inode->i_blocks = 0;
1312		inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1313		inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1314		inode->i_generation = get_seconds();
1315		info = SHMEM_I(inode);
1316		memset(info, 0, (char *)inode - (char *)info);
1317		spin_lock_init(&info->lock);
 
 
1318		info->flags = flags & VM_NORESERVE;
 
1319		INIT_LIST_HEAD(&info->swaplist);
1320		simple_xattrs_init(&info->xattrs);
1321		cache_no_acl(inode);
1322
1323		switch (mode & S_IFMT) {
1324		default:
1325			inode->i_op = &shmem_special_inode_operations;
1326			init_special_inode(inode, mode, dev);
1327			break;
1328		case S_IFREG:
1329			inode->i_mapping->a_ops = &shmem_aops;
1330			inode->i_op = &shmem_inode_operations;
1331			inode->i_fop = &shmem_file_operations;
1332			mpol_shared_policy_init(&info->policy,
1333						 shmem_get_sbmpol(sbinfo));
1334			break;
1335		case S_IFDIR:
1336			inc_nlink(inode);
1337			/* Some things misbehave if size == 0 on a directory */
1338			inode->i_size = 2 * BOGO_DIRENT_SIZE;
1339			inode->i_op = &shmem_dir_inode_operations;
1340			inode->i_fop = &simple_dir_operations;
1341			break;
1342		case S_IFLNK:
1343			/*
1344			 * Must not load anything in the rbtree,
1345			 * mpol_free_shared_policy will not be called.
1346			 */
1347			mpol_shared_policy_init(&info->policy, NULL);
1348			break;
1349		}
 
 
1350	} else
1351		shmem_free_inode(sb);
1352	return inode;
1353}
1354
1355bool shmem_mapping(struct address_space *mapping)
1356{
1357	return mapping->backing_dev_info == &shmem_backing_dev_info;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1358}
1359
1360#ifdef CONFIG_TMPFS
1361static const struct inode_operations shmem_symlink_inode_operations;
1362static const struct inode_operations shmem_short_symlink_operations;
1363
1364#ifdef CONFIG_TMPFS_XATTR
1365static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1366#else
1367#define shmem_initxattrs NULL
1368#endif
1369
1370static int
1371shmem_write_begin(struct file *file, struct address_space *mapping,
1372			loff_t pos, unsigned len, unsigned flags,
1373			struct page **pagep, void **fsdata)
1374{
1375	struct inode *inode = mapping->host;
1376	pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1377	return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
 
 
 
 
 
 
 
 
 
 
 
1378}
1379
1380static int
1381shmem_write_end(struct file *file, struct address_space *mapping,
1382			loff_t pos, unsigned len, unsigned copied,
1383			struct page *page, void *fsdata)
1384{
1385	struct inode *inode = mapping->host;
1386
1387	if (pos + copied > inode->i_size)
1388		i_size_write(inode, pos + copied);
1389
1390	if (!PageUptodate(page)) {
1391		if (copied < PAGE_CACHE_SIZE) {
1392			unsigned from = pos & (PAGE_CACHE_SIZE - 1);
 
 
 
 
 
 
 
 
 
 
 
1393			zero_user_segments(page, 0, from,
1394					from + copied, PAGE_CACHE_SIZE);
1395		}
1396		SetPageUptodate(page);
1397	}
1398	set_page_dirty(page);
1399	unlock_page(page);
1400	page_cache_release(page);
1401
1402	return copied;
1403}
1404
1405static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1406		const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1407{
1408	struct file *file = iocb->ki_filp;
1409	struct inode *inode = file_inode(file);
1410	struct address_space *mapping = inode->i_mapping;
1411	pgoff_t index;
1412	unsigned long offset;
1413	enum sgp_type sgp = SGP_READ;
1414	int error = 0;
1415	ssize_t retval;
1416	size_t count;
1417	loff_t *ppos = &iocb->ki_pos;
1418	struct iov_iter iter;
1419
1420	retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1421	if (retval)
1422		return retval;
1423	iov_iter_init(&iter, iov, nr_segs, count, 0);
1424
1425	/*
1426	 * Might this read be for a stacking filesystem?  Then when reading
1427	 * holes of a sparse file, we actually need to allocate those pages,
1428	 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1429	 */
1430	if (segment_eq(get_fs(), KERNEL_DS))
1431		sgp = SGP_DIRTY;
1432
1433	index = *ppos >> PAGE_CACHE_SHIFT;
1434	offset = *ppos & ~PAGE_CACHE_MASK;
1435
1436	for (;;) {
1437		struct page *page = NULL;
1438		pgoff_t end_index;
1439		unsigned long nr, ret;
1440		loff_t i_size = i_size_read(inode);
1441
1442		end_index = i_size >> PAGE_CACHE_SHIFT;
1443		if (index > end_index)
1444			break;
1445		if (index == end_index) {
1446			nr = i_size & ~PAGE_CACHE_MASK;
1447			if (nr <= offset)
1448				break;
1449		}
1450
1451		error = shmem_getpage(inode, index, &page, sgp, NULL);
1452		if (error) {
1453			if (error == -EINVAL)
1454				error = 0;
1455			break;
1456		}
1457		if (page)
 
 
1458			unlock_page(page);
 
1459
1460		/*
1461		 * We must evaluate after, since reads (unlike writes)
1462		 * are called without i_mutex protection against truncate
1463		 */
1464		nr = PAGE_CACHE_SIZE;
1465		i_size = i_size_read(inode);
1466		end_index = i_size >> PAGE_CACHE_SHIFT;
1467		if (index == end_index) {
1468			nr = i_size & ~PAGE_CACHE_MASK;
1469			if (nr <= offset) {
1470				if (page)
1471					page_cache_release(page);
1472				break;
1473			}
1474		}
1475		nr -= offset;
1476
1477		if (page) {
1478			/*
1479			 * If users can be writing to this page using arbitrary
1480			 * virtual addresses, take care about potential aliasing
1481			 * before reading the page on the kernel side.
1482			 */
1483			if (mapping_writably_mapped(mapping))
1484				flush_dcache_page(page);
1485			/*
1486			 * Mark the page accessed if we read the beginning.
1487			 */
1488			if (!offset)
1489				mark_page_accessed(page);
1490		} else {
1491			page = ZERO_PAGE(0);
1492			page_cache_get(page);
1493		}
1494
1495		/*
1496		 * Ok, we have the page, and it's up-to-date, so
1497		 * now we can copy it to user space...
1498		 */
1499		ret = copy_page_to_iter(page, offset, nr, &iter);
1500		retval += ret;
1501		offset += ret;
1502		index += offset >> PAGE_CACHE_SHIFT;
1503		offset &= ~PAGE_CACHE_MASK;
1504
1505		page_cache_release(page);
1506		if (!iov_iter_count(&iter))
1507			break;
1508		if (ret < nr) {
1509			error = -EFAULT;
1510			break;
1511		}
1512		cond_resched();
1513	}
1514
1515	*ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1516	file_accessed(file);
1517	return retval ? retval : error;
1518}
1519
1520static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1521				struct pipe_inode_info *pipe, size_t len,
1522				unsigned int flags)
1523{
1524	struct address_space *mapping = in->f_mapping;
1525	struct inode *inode = mapping->host;
1526	unsigned int loff, nr_pages, req_pages;
1527	struct page *pages[PIPE_DEF_BUFFERS];
1528	struct partial_page partial[PIPE_DEF_BUFFERS];
1529	struct page *page;
1530	pgoff_t index, end_index;
1531	loff_t isize, left;
1532	int error, page_nr;
1533	struct splice_pipe_desc spd = {
1534		.pages = pages,
1535		.partial = partial,
1536		.nr_pages_max = PIPE_DEF_BUFFERS,
1537		.flags = flags,
1538		.ops = &page_cache_pipe_buf_ops,
1539		.spd_release = spd_release_page,
1540	};
1541
1542	isize = i_size_read(inode);
1543	if (unlikely(*ppos >= isize))
1544		return 0;
1545
1546	left = isize - *ppos;
1547	if (unlikely(left < len))
1548		len = left;
1549
1550	if (splice_grow_spd(pipe, &spd))
1551		return -ENOMEM;
1552
1553	index = *ppos >> PAGE_CACHE_SHIFT;
1554	loff = *ppos & ~PAGE_CACHE_MASK;
1555	req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1556	nr_pages = min(req_pages, spd.nr_pages_max);
1557
1558	spd.nr_pages = find_get_pages_contig(mapping, index,
1559						nr_pages, spd.pages);
1560	index += spd.nr_pages;
1561	error = 0;
1562
1563	while (spd.nr_pages < nr_pages) {
1564		error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1565		if (error)
1566			break;
1567		unlock_page(page);
1568		spd.pages[spd.nr_pages++] = page;
1569		index++;
1570	}
1571
1572	index = *ppos >> PAGE_CACHE_SHIFT;
1573	nr_pages = spd.nr_pages;
1574	spd.nr_pages = 0;
1575
1576	for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1577		unsigned int this_len;
1578
1579		if (!len)
1580			break;
1581
1582		this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1583		page = spd.pages[page_nr];
1584
1585		if (!PageUptodate(page) || page->mapping != mapping) {
1586			error = shmem_getpage(inode, index, &page,
1587							SGP_CACHE, NULL);
1588			if (error)
1589				break;
1590			unlock_page(page);
1591			page_cache_release(spd.pages[page_nr]);
1592			spd.pages[page_nr] = page;
1593		}
1594
1595		isize = i_size_read(inode);
1596		end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1597		if (unlikely(!isize || index > end_index))
1598			break;
1599
1600		if (end_index == index) {
1601			unsigned int plen;
1602
1603			plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1604			if (plen <= loff)
1605				break;
1606
1607			this_len = min(this_len, plen - loff);
1608			len = this_len;
1609		}
1610
1611		spd.partial[page_nr].offset = loff;
1612		spd.partial[page_nr].len = this_len;
1613		len -= this_len;
1614		loff = 0;
1615		spd.nr_pages++;
1616		index++;
1617	}
1618
1619	while (page_nr < nr_pages)
1620		page_cache_release(spd.pages[page_nr++]);
1621
1622	if (spd.nr_pages)
1623		error = splice_to_pipe(pipe, &spd);
1624
1625	splice_shrink_spd(&spd);
1626
1627	if (error > 0) {
1628		*ppos += error;
1629		file_accessed(in);
1630	}
1631	return error;
1632}
1633
1634/*
1635 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1636 */
1637static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
1638				    pgoff_t index, pgoff_t end, int whence)
1639{
1640	struct page *page;
1641	struct pagevec pvec;
1642	pgoff_t indices[PAGEVEC_SIZE];
1643	bool done = false;
1644	int i;
1645
1646	pagevec_init(&pvec, 0);
1647	pvec.nr = 1;		/* start small: we may be there already */
1648	while (!done) {
1649		pvec.nr = find_get_entries(mapping, index,
1650					pvec.nr, pvec.pages, indices);
1651		if (!pvec.nr) {
1652			if (whence == SEEK_DATA)
1653				index = end;
1654			break;
1655		}
1656		for (i = 0; i < pvec.nr; i++, index++) {
1657			if (index < indices[i]) {
1658				if (whence == SEEK_HOLE) {
1659					done = true;
1660					break;
1661				}
1662				index = indices[i];
1663			}
1664			page = pvec.pages[i];
1665			if (page && !radix_tree_exceptional_entry(page)) {
1666				if (!PageUptodate(page))
1667					page = NULL;
1668			}
1669			if (index >= end ||
1670			    (page && whence == SEEK_DATA) ||
1671			    (!page && whence == SEEK_HOLE)) {
1672				done = true;
1673				break;
1674			}
1675		}
1676		pagevec_remove_exceptionals(&pvec);
1677		pagevec_release(&pvec);
1678		pvec.nr = PAGEVEC_SIZE;
1679		cond_resched();
1680	}
1681	return index;
1682}
1683
1684static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
1685{
1686	struct address_space *mapping = file->f_mapping;
1687	struct inode *inode = mapping->host;
1688	pgoff_t start, end;
1689	loff_t new_offset;
1690
1691	if (whence != SEEK_DATA && whence != SEEK_HOLE)
1692		return generic_file_llseek_size(file, offset, whence,
1693					MAX_LFS_FILESIZE, i_size_read(inode));
1694	mutex_lock(&inode->i_mutex);
1695	/* We're holding i_mutex so we can access i_size directly */
1696
1697	if (offset < 0)
1698		offset = -EINVAL;
1699	else if (offset >= inode->i_size)
1700		offset = -ENXIO;
1701	else {
1702		start = offset >> PAGE_CACHE_SHIFT;
1703		end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1704		new_offset = shmem_seek_hole_data(mapping, start, end, whence);
1705		new_offset <<= PAGE_CACHE_SHIFT;
1706		if (new_offset > offset) {
1707			if (new_offset < inode->i_size)
1708				offset = new_offset;
1709			else if (whence == SEEK_DATA)
1710				offset = -ENXIO;
1711			else
1712				offset = inode->i_size;
1713		}
1714	}
1715
1716	if (offset >= 0)
1717		offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
1718	mutex_unlock(&inode->i_mutex);
1719	return offset;
1720}
1721
1722static long shmem_fallocate(struct file *file, int mode, loff_t offset,
1723							 loff_t len)
1724{
1725	struct inode *inode = file_inode(file);
1726	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 
1727	struct shmem_falloc shmem_falloc;
1728	pgoff_t start, index, end;
1729	int error;
1730
1731	mutex_lock(&inode->i_mutex);
 
 
 
1732
1733	if (mode & FALLOC_FL_PUNCH_HOLE) {
1734		struct address_space *mapping = file->f_mapping;
1735		loff_t unmap_start = round_up(offset, PAGE_SIZE);
1736		loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1737
1738		if ((u64)unmap_end > (u64)unmap_start)
1739			unmap_mapping_range(mapping, unmap_start,
1740					    1 + unmap_end - unmap_start, 0);
1741		shmem_truncate_range(inode, offset, offset + len - 1);
1742		/* No need to unmap again: hole-punching leaves COWed pages */
 
 
 
 
 
 
1743		error = 0;
1744		goto out;
1745	}
1746
1747	/* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1748	error = inode_newsize_ok(inode, offset + len);
1749	if (error)
1750		goto out;
1751
1752	start = offset >> PAGE_CACHE_SHIFT;
1753	end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 
 
 
 
 
1754	/* Try to avoid a swapstorm if len is impossible to satisfy */
1755	if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
1756		error = -ENOSPC;
1757		goto out;
1758	}
1759
 
1760	shmem_falloc.start = start;
1761	shmem_falloc.next  = start;
1762	shmem_falloc.nr_falloced = 0;
1763	shmem_falloc.nr_unswapped = 0;
1764	spin_lock(&inode->i_lock);
1765	inode->i_private = &shmem_falloc;
1766	spin_unlock(&inode->i_lock);
1767
1768	for (index = start; index < end; index++) {
1769		struct page *page;
1770
1771		/*
1772		 * Good, the fallocate(2) manpage permits EINTR: we may have
1773		 * been interrupted because we are using up too much memory.
1774		 */
1775		if (signal_pending(current))
1776			error = -EINTR;
1777		else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
1778			error = -ENOMEM;
1779		else
1780			error = shmem_getpage(inode, index, &page, SGP_FALLOC,
1781									NULL);
1782		if (error) {
1783			/* Remove the !PageUptodate pages we added */
1784			shmem_undo_range(inode,
1785				(loff_t)start << PAGE_CACHE_SHIFT,
1786				(loff_t)index << PAGE_CACHE_SHIFT, true);
 
 
1787			goto undone;
1788		}
1789
1790		/*
1791		 * Inform shmem_writepage() how far we have reached.
1792		 * No need for lock or barrier: we have the page lock.
1793		 */
1794		shmem_falloc.next++;
1795		if (!PageUptodate(page))
1796			shmem_falloc.nr_falloced++;
1797
1798		/*
1799		 * If !PageUptodate, leave it that way so that freeable pages
1800		 * can be recognized if we need to rollback on error later.
1801		 * But set_page_dirty so that memory pressure will swap rather
1802		 * than free the pages we are allocating (and SGP_CACHE pages
1803		 * might still be clean: we now need to mark those dirty too).
1804		 */
1805		set_page_dirty(page);
1806		unlock_page(page);
1807		page_cache_release(page);
1808		cond_resched();
1809	}
1810
1811	if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
1812		i_size_write(inode, offset + len);
1813	inode->i_ctime = CURRENT_TIME;
1814undone:
1815	spin_lock(&inode->i_lock);
1816	inode->i_private = NULL;
1817	spin_unlock(&inode->i_lock);
1818out:
1819	mutex_unlock(&inode->i_mutex);
1820	return error;
1821}
1822
1823static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1824{
1825	struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1826
1827	buf->f_type = TMPFS_MAGIC;
1828	buf->f_bsize = PAGE_CACHE_SIZE;
1829	buf->f_namelen = NAME_MAX;
1830	if (sbinfo->max_blocks) {
1831		buf->f_blocks = sbinfo->max_blocks;
1832		buf->f_bavail =
1833		buf->f_bfree  = sbinfo->max_blocks -
1834				percpu_counter_sum(&sbinfo->used_blocks);
1835	}
1836	if (sbinfo->max_inodes) {
1837		buf->f_files = sbinfo->max_inodes;
1838		buf->f_ffree = sbinfo->free_inodes;
1839	}
1840	/* else leave those fields 0 like simple_statfs */
1841	return 0;
1842}
1843
1844/*
1845 * File creation. Allocate an inode, and we're done..
1846 */
1847static int
1848shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
1849{
1850	struct inode *inode;
1851	int error = -ENOSPC;
1852
1853	inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1854	if (inode) {
1855		error = simple_acl_create(dir, inode);
1856		if (error)
1857			goto out_iput;
1858		error = security_inode_init_security(inode, dir,
1859						     &dentry->d_name,
1860						     shmem_initxattrs, NULL);
1861		if (error && error != -EOPNOTSUPP)
1862			goto out_iput;
1863
1864		error = 0;
1865		dir->i_size += BOGO_DIRENT_SIZE;
1866		dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1867		d_instantiate(dentry, inode);
1868		dget(dentry); /* Extra count - pin the dentry in core */
1869	}
1870	return error;
1871out_iput:
1872	iput(inode);
1873	return error;
1874}
1875
1876static int
1877shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
1878{
1879	struct inode *inode;
1880	int error = -ENOSPC;
1881
1882	inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
1883	if (inode) {
1884		error = security_inode_init_security(inode, dir,
1885						     NULL,
1886						     shmem_initxattrs, NULL);
1887		if (error && error != -EOPNOTSUPP)
1888			goto out_iput;
1889		error = simple_acl_create(dir, inode);
1890		if (error)
1891			goto out_iput;
1892		d_tmpfile(dentry, inode);
1893	}
1894	return error;
1895out_iput:
1896	iput(inode);
1897	return error;
1898}
1899
1900static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1901{
1902	int error;
1903
1904	if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1905		return error;
1906	inc_nlink(dir);
1907	return 0;
1908}
1909
1910static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1911		bool excl)
1912{
1913	return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1914}
1915
1916/*
1917 * Link a file..
1918 */
1919static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1920{
1921	struct inode *inode = old_dentry->d_inode;
1922	int ret;
1923
1924	/*
1925	 * No ordinary (disk based) filesystem counts links as inodes;
1926	 * but each new link needs a new dentry, pinning lowmem, and
1927	 * tmpfs dentries cannot be pruned until they are unlinked.
 
 
1928	 */
1929	ret = shmem_reserve_inode(inode->i_sb);
1930	if (ret)
1931		goto out;
 
 
1932
1933	dir->i_size += BOGO_DIRENT_SIZE;
1934	inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1935	inc_nlink(inode);
1936	ihold(inode);	/* New dentry reference */
1937	dget(dentry);		/* Extra pinning count for the created dentry */
1938	d_instantiate(dentry, inode);
1939out:
1940	return ret;
1941}
1942
1943static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1944{
1945	struct inode *inode = dentry->d_inode;
1946
1947	if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1948		shmem_free_inode(inode->i_sb);
1949
1950	dir->i_size -= BOGO_DIRENT_SIZE;
1951	inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1952	drop_nlink(inode);
1953	dput(dentry);	/* Undo the count from "create" - this does all the work */
1954	return 0;
1955}
1956
1957static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1958{
1959	if (!simple_empty(dentry))
1960		return -ENOTEMPTY;
1961
1962	drop_nlink(dentry->d_inode);
1963	drop_nlink(dir);
1964	return shmem_unlink(dir, dentry);
1965}
1966
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1967/*
1968 * The VFS layer already does all the dentry stuff for rename,
1969 * we just have to decrement the usage count for the target if
1970 * it exists so that the VFS layer correctly free's it when it
1971 * gets overwritten.
1972 */
1973static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1974{
1975	struct inode *inode = old_dentry->d_inode;
1976	int they_are_dirs = S_ISDIR(inode->i_mode);
1977
 
 
 
 
 
 
1978	if (!simple_empty(new_dentry))
1979		return -ENOTEMPTY;
1980
1981	if (new_dentry->d_inode) {
 
 
 
 
 
 
 
 
1982		(void) shmem_unlink(new_dir, new_dentry);
1983		if (they_are_dirs)
 
1984			drop_nlink(old_dir);
 
1985	} else if (they_are_dirs) {
1986		drop_nlink(old_dir);
1987		inc_nlink(new_dir);
1988	}
1989
1990	old_dir->i_size -= BOGO_DIRENT_SIZE;
1991	new_dir->i_size += BOGO_DIRENT_SIZE;
1992	old_dir->i_ctime = old_dir->i_mtime =
1993	new_dir->i_ctime = new_dir->i_mtime =
1994	inode->i_ctime = CURRENT_TIME;
1995	return 0;
1996}
1997
1998static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1999{
2000	int error;
2001	int len;
2002	struct inode *inode;
2003	struct page *page;
2004	char *kaddr;
2005	struct shmem_inode_info *info;
2006
2007	len = strlen(symname) + 1;
2008	if (len > PAGE_CACHE_SIZE)
2009		return -ENAMETOOLONG;
2010
2011	inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
 
2012	if (!inode)
2013		return -ENOSPC;
2014
2015	error = security_inode_init_security(inode, dir, &dentry->d_name,
2016					     shmem_initxattrs, NULL);
2017	if (error) {
2018		if (error != -EOPNOTSUPP) {
2019			iput(inode);
2020			return error;
2021		}
2022		error = 0;
2023	}
2024
2025	info = SHMEM_I(inode);
2026	inode->i_size = len-1;
2027	if (len <= SHORT_SYMLINK_LEN) {
2028		info->symlink = kmemdup(symname, len, GFP_KERNEL);
2029		if (!info->symlink) {
2030			iput(inode);
2031			return -ENOMEM;
2032		}
2033		inode->i_op = &shmem_short_symlink_operations;
2034	} else {
2035		error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
 
2036		if (error) {
2037			iput(inode);
2038			return error;
2039		}
2040		inode->i_mapping->a_ops = &shmem_aops;
2041		inode->i_op = &shmem_symlink_inode_operations;
2042		kaddr = kmap_atomic(page);
2043		memcpy(kaddr, symname, len);
2044		kunmap_atomic(kaddr);
2045		SetPageUptodate(page);
2046		set_page_dirty(page);
2047		unlock_page(page);
2048		page_cache_release(page);
2049	}
2050	dir->i_size += BOGO_DIRENT_SIZE;
2051	dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2052	d_instantiate(dentry, inode);
2053	dget(dentry);
2054	return 0;
2055}
2056
2057static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
2058{
2059	nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
2060	return NULL;
2061}
2062
2063static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
 
 
2064{
2065	struct page *page = NULL;
2066	int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2067	nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
2068	if (page)
 
 
 
 
 
 
 
 
 
 
2069		unlock_page(page);
2070	return page;
2071}
2072
2073static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2074{
2075	if (!IS_ERR(nd_get_link(nd))) {
2076		struct page *page = cookie;
2077		kunmap(page);
2078		mark_page_accessed(page);
2079		page_cache_release(page);
2080	}
 
 
2081}
2082
2083#ifdef CONFIG_TMPFS_XATTR
2084/*
2085 * Superblocks without xattr inode operations may get some security.* xattr
2086 * support from the LSM "for free". As soon as we have any other xattrs
2087 * like ACLs, we also need to implement the security.* handlers at
2088 * filesystem level, though.
2089 */
2090
2091/*
2092 * Callback for security_inode_init_security() for acquiring xattrs.
2093 */
2094static int shmem_initxattrs(struct inode *inode,
2095			    const struct xattr *xattr_array,
2096			    void *fs_info)
2097{
2098	struct shmem_inode_info *info = SHMEM_I(inode);
2099	const struct xattr *xattr;
2100	struct simple_xattr *new_xattr;
2101	size_t len;
2102
2103	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
2104		new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
2105		if (!new_xattr)
2106			return -ENOMEM;
2107
2108		len = strlen(xattr->name) + 1;
2109		new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2110					  GFP_KERNEL);
2111		if (!new_xattr->name) {
2112			kfree(new_xattr);
2113			return -ENOMEM;
2114		}
2115
2116		memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2117		       XATTR_SECURITY_PREFIX_LEN);
2118		memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2119		       xattr->name, len);
2120
2121		simple_xattr_list_add(&info->xattrs, new_xattr);
2122	}
2123
2124	return 0;
2125}
2126
2127static const struct xattr_handler *shmem_xattr_handlers[] = {
2128#ifdef CONFIG_TMPFS_POSIX_ACL
2129	&posix_acl_access_xattr_handler,
2130	&posix_acl_default_xattr_handler,
2131#endif
2132	NULL
2133};
2134
2135static int shmem_xattr_validate(const char *name)
2136{
2137	struct { const char *prefix; size_t len; } arr[] = {
2138		{ XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2139		{ XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2140	};
2141	int i;
2142
2143	for (i = 0; i < ARRAY_SIZE(arr); i++) {
2144		size_t preflen = arr[i].len;
2145		if (strncmp(name, arr[i].prefix, preflen) == 0) {
2146			if (!name[preflen])
2147				return -EINVAL;
2148			return 0;
2149		}
2150	}
2151	return -EOPNOTSUPP;
2152}
2153
2154static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2155			      void *buffer, size_t size)
2156{
2157	struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2158	int err;
2159
2160	/*
2161	 * If this is a request for a synthetic attribute in the system.*
2162	 * namespace use the generic infrastructure to resolve a handler
2163	 * for it via sb->s_xattr.
2164	 */
2165	if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2166		return generic_getxattr(dentry, name, buffer, size);
2167
2168	err = shmem_xattr_validate(name);
2169	if (err)
2170		return err;
2171
 
2172	return simple_xattr_get(&info->xattrs, name, buffer, size);
2173}
2174
2175static int shmem_setxattr(struct dentry *dentry, const char *name,
2176			  const void *value, size_t size, int flags)
 
 
2177{
2178	struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2179	int err;
2180
2181	/*
2182	 * If this is a request for a synthetic attribute in the system.*
2183	 * namespace use the generic infrastructure to resolve a handler
2184	 * for it via sb->s_xattr.
2185	 */
2186	if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2187		return generic_setxattr(dentry, name, value, size, flags);
2188
2189	err = shmem_xattr_validate(name);
2190	if (err)
2191		return err;
2192
2193	return simple_xattr_set(&info->xattrs, name, value, size, flags);
 
2194}
2195
2196static int shmem_removexattr(struct dentry *dentry, const char *name)
2197{
2198	struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2199	int err;
 
2200
2201	/*
2202	 * If this is a request for a synthetic attribute in the system.*
2203	 * namespace use the generic infrastructure to resolve a handler
2204	 * for it via sb->s_xattr.
2205	 */
2206	if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2207		return generic_removexattr(dentry, name);
2208
2209	err = shmem_xattr_validate(name);
2210	if (err)
2211		return err;
2212
2213	return simple_xattr_remove(&info->xattrs, name);
2214}
 
 
 
2215
2216static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2217{
2218	struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2219	return simple_xattr_list(&info->xattrs, buffer, size);
2220}
2221#endif /* CONFIG_TMPFS_XATTR */
2222
2223static const struct inode_operations shmem_short_symlink_operations = {
2224	.readlink	= generic_readlink,
2225	.follow_link	= shmem_follow_short_symlink,
2226#ifdef CONFIG_TMPFS_XATTR
2227	.setxattr	= shmem_setxattr,
2228	.getxattr	= shmem_getxattr,
2229	.listxattr	= shmem_listxattr,
2230	.removexattr	= shmem_removexattr,
2231#endif
2232};
2233
2234static const struct inode_operations shmem_symlink_inode_operations = {
2235	.readlink	= generic_readlink,
2236	.follow_link	= shmem_follow_link,
2237	.put_link	= shmem_put_link,
2238#ifdef CONFIG_TMPFS_XATTR
2239	.setxattr	= shmem_setxattr,
2240	.getxattr	= shmem_getxattr,
2241	.listxattr	= shmem_listxattr,
2242	.removexattr	= shmem_removexattr,
2243#endif
2244};
2245
2246static struct dentry *shmem_get_parent(struct dentry *child)
2247{
2248	return ERR_PTR(-ESTALE);
2249}
2250
2251static int shmem_match(struct inode *ino, void *vfh)
2252{
2253	__u32 *fh = vfh;
2254	__u64 inum = fh[2];
2255	inum = (inum << 32) | fh[1];
2256	return ino->i_ino == inum && fh[0] == ino->i_generation;
2257}
2258
 
 
 
 
 
 
 
 
 
2259static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2260		struct fid *fid, int fh_len, int fh_type)
2261{
2262	struct inode *inode;
2263	struct dentry *dentry = NULL;
2264	u64 inum;
2265
2266	if (fh_len < 3)
2267		return NULL;
2268
2269	inum = fid->raw[2];
2270	inum = (inum << 32) | fid->raw[1];
2271
2272	inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2273			shmem_match, fid->raw);
2274	if (inode) {
2275		dentry = d_find_alias(inode);
2276		iput(inode);
2277	}
2278
2279	return dentry;
2280}
2281
2282static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2283				struct inode *parent)
2284{
2285	if (*len < 3) {
2286		*len = 3;
2287		return FILEID_INVALID;
2288	}
2289
2290	if (inode_unhashed(inode)) {
2291		/* Unfortunately insert_inode_hash is not idempotent,
2292		 * so as we hash inodes here rather than at creation
2293		 * time, we need a lock to ensure we only try
2294		 * to do it once
2295		 */
2296		static DEFINE_SPINLOCK(lock);
2297		spin_lock(&lock);
2298		if (inode_unhashed(inode))
2299			__insert_inode_hash(inode,
2300					    inode->i_ino + inode->i_generation);
2301		spin_unlock(&lock);
2302	}
2303
2304	fh[0] = inode->i_generation;
2305	fh[1] = inode->i_ino;
2306	fh[2] = ((__u64)inode->i_ino) >> 32;
2307
2308	*len = 3;
2309	return 1;
2310}
2311
2312static const struct export_operations shmem_export_ops = {
2313	.get_parent     = shmem_get_parent,
2314	.encode_fh      = shmem_encode_fh,
2315	.fh_to_dentry	= shmem_fh_to_dentry,
2316};
2317
2318static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2319			       bool remount)
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2320{
2321	char *this_char, *value, *rest;
2322	struct mempolicy *mpol = NULL;
2323	uid_t uid;
2324	gid_t gid;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2325
2326	while (options != NULL) {
2327		this_char = options;
2328		for (;;) {
2329			/*
2330			 * NUL-terminate this option: unfortunately,
2331			 * mount options form a comma-separated list,
2332			 * but mpol's nodelist may also contain commas.
2333			 */
2334			options = strchr(options, ',');
2335			if (options == NULL)
2336				break;
2337			options++;
2338			if (!isdigit(*options)) {
2339				options[-1] = '\0';
2340				break;
2341			}
2342		}
2343		if (!*this_char)
2344			continue;
2345		if ((value = strchr(this_char,'=')) != NULL) {
2346			*value++ = 0;
2347		} else {
2348			printk(KERN_ERR
2349			    "tmpfs: No value for mount option '%s'\n",
2350			    this_char);
2351			goto error;
2352		}
2353
2354		if (!strcmp(this_char,"size")) {
2355			unsigned long long size;
2356			size = memparse(value,&rest);
2357			if (*rest == '%') {
2358				size <<= PAGE_SHIFT;
2359				size *= totalram_pages;
2360				do_div(size, 100);
2361				rest++;
2362			}
2363			if (*rest)
2364				goto bad_val;
2365			sbinfo->max_blocks =
2366				DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2367		} else if (!strcmp(this_char,"nr_blocks")) {
2368			sbinfo->max_blocks = memparse(value, &rest);
2369			if (*rest)
2370				goto bad_val;
2371		} else if (!strcmp(this_char,"nr_inodes")) {
2372			sbinfo->max_inodes = memparse(value, &rest);
2373			if (*rest)
2374				goto bad_val;
2375		} else if (!strcmp(this_char,"mode")) {
2376			if (remount)
2377				continue;
2378			sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2379			if (*rest)
2380				goto bad_val;
2381		} else if (!strcmp(this_char,"uid")) {
2382			if (remount)
2383				continue;
2384			uid = simple_strtoul(value, &rest, 0);
2385			if (*rest)
2386				goto bad_val;
2387			sbinfo->uid = make_kuid(current_user_ns(), uid);
2388			if (!uid_valid(sbinfo->uid))
2389				goto bad_val;
2390		} else if (!strcmp(this_char,"gid")) {
2391			if (remount)
2392				continue;
2393			gid = simple_strtoul(value, &rest, 0);
2394			if (*rest)
2395				goto bad_val;
2396			sbinfo->gid = make_kgid(current_user_ns(), gid);
2397			if (!gid_valid(sbinfo->gid))
2398				goto bad_val;
2399		} else if (!strcmp(this_char,"mpol")) {
2400			mpol_put(mpol);
2401			mpol = NULL;
2402			if (mpol_parse_str(value, &mpol))
2403				goto bad_val;
2404		} else {
2405			printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2406			       this_char);
2407			goto error;
2408		}
2409	}
2410	sbinfo->mpol = mpol;
2411	return 0;
2412
2413bad_val:
2414	printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2415	       value, this_char);
2416error:
2417	mpol_put(mpol);
2418	return 1;
2419
2420}
2421
2422static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
 
 
 
 
 
 
 
2423{
2424	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2425	struct shmem_sb_info config = *sbinfo;
2426	unsigned long inodes;
2427	int error = -EINVAL;
2428
2429	config.mpol = NULL;
2430	if (shmem_parse_options(data, &config, true))
2431		return error;
2432
2433	spin_lock(&sbinfo->stat_lock);
2434	inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2435	if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2436		goto out;
2437	if (config.max_inodes < inodes)
2438		goto out;
2439	/*
2440	 * Those tests disallow limited->unlimited while any are in use;
2441	 * but we must separately disallow unlimited->limited, because
2442	 * in that case we have no record of how much is already in use.
2443	 */
2444	if (config.max_blocks && !sbinfo->max_blocks)
2445		goto out;
2446	if (config.max_inodes && !sbinfo->max_inodes)
 
 
 
 
 
 
 
 
 
 
 
 
 
2447		goto out;
 
2448
2449	error = 0;
2450	sbinfo->max_blocks  = config.max_blocks;
2451	sbinfo->max_inodes  = config.max_inodes;
2452	sbinfo->free_inodes = config.max_inodes - inodes;
 
 
 
 
 
 
2453
2454	/*
2455	 * Preserve previous mempolicy unless mpol remount option was specified.
2456	 */
2457	if (config.mpol) {
2458		mpol_put(sbinfo->mpol);
2459		sbinfo->mpol = config.mpol;	/* transfers initial ref */
 
2460	}
 
 
2461out:
2462	spin_unlock(&sbinfo->stat_lock);
2463	return error;
2464}
2465
2466static int shmem_show_options(struct seq_file *seq, struct dentry *root)
2467{
2468	struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
2469
2470	if (sbinfo->max_blocks != shmem_default_max_blocks())
2471		seq_printf(seq, ",size=%luk",
2472			sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2473	if (sbinfo->max_inodes != shmem_default_max_inodes())
2474		seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2475	if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2476		seq_printf(seq, ",mode=%03ho", sbinfo->mode);
2477	if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2478		seq_printf(seq, ",uid=%u",
2479				from_kuid_munged(&init_user_ns, sbinfo->uid));
2480	if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2481		seq_printf(seq, ",gid=%u",
2482				from_kgid_munged(&init_user_ns, sbinfo->gid));
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2483	shmem_show_mpol(seq, sbinfo->mpol);
2484	return 0;
2485}
 
2486#endif /* CONFIG_TMPFS */
2487
2488static void shmem_put_super(struct super_block *sb)
2489{
2490	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2491
 
2492	percpu_counter_destroy(&sbinfo->used_blocks);
2493	mpol_put(sbinfo->mpol);
2494	kfree(sbinfo);
2495	sb->s_fs_info = NULL;
2496}
2497
2498int shmem_fill_super(struct super_block *sb, void *data, int silent)
2499{
 
2500	struct inode *inode;
2501	struct shmem_sb_info *sbinfo;
2502	int err = -ENOMEM;
2503
2504	/* Round up to L1_CACHE_BYTES to resist false sharing */
2505	sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2506				L1_CACHE_BYTES), GFP_KERNEL);
2507	if (!sbinfo)
2508		return -ENOMEM;
2509
2510	sbinfo->mode = S_IRWXUGO | S_ISVTX;
2511	sbinfo->uid = current_fsuid();
2512	sbinfo->gid = current_fsgid();
2513	sb->s_fs_info = sbinfo;
2514
2515#ifdef CONFIG_TMPFS
2516	/*
2517	 * Per default we only allow half of the physical ram per
2518	 * tmpfs instance, limiting inodes to one per page of lowmem;
2519	 * but the internal instance is left unlimited.
2520	 */
2521	if (!(sb->s_flags & MS_KERNMOUNT)) {
2522		sbinfo->max_blocks = shmem_default_max_blocks();
2523		sbinfo->max_inodes = shmem_default_max_inodes();
2524		if (shmem_parse_options(data, sbinfo, false)) {
2525			err = -EINVAL;
2526			goto failed;
2527		}
2528	} else {
2529		sb->s_flags |= MS_NOUSER;
2530	}
2531	sb->s_export_op = &shmem_export_ops;
2532	sb->s_flags |= MS_NOSEC;
2533#else
2534	sb->s_flags |= MS_NOUSER;
2535#endif
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2536
2537	spin_lock_init(&sbinfo->stat_lock);
2538	if (percpu_counter_init(&sbinfo->used_blocks, 0))
2539		goto failed;
2540	sbinfo->free_inodes = sbinfo->max_inodes;
 
2541
2542	sb->s_maxbytes = MAX_LFS_FILESIZE;
2543	sb->s_blocksize = PAGE_CACHE_SIZE;
2544	sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2545	sb->s_magic = TMPFS_MAGIC;
2546	sb->s_op = &shmem_ops;
2547	sb->s_time_gran = 1;
2548#ifdef CONFIG_TMPFS_XATTR
2549	sb->s_xattr = shmem_xattr_handlers;
2550#endif
2551#ifdef CONFIG_TMPFS_POSIX_ACL
2552	sb->s_flags |= MS_POSIXACL;
2553#endif
 
2554
2555	inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2556	if (!inode)
2557		goto failed;
2558	inode->i_uid = sbinfo->uid;
2559	inode->i_gid = sbinfo->gid;
2560	sb->s_root = d_make_root(inode);
2561	if (!sb->s_root)
2562		goto failed;
2563	return 0;
2564
2565failed:
2566	shmem_put_super(sb);
2567	return err;
2568}
2569
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2570static struct kmem_cache *shmem_inode_cachep;
2571
2572static struct inode *shmem_alloc_inode(struct super_block *sb)
2573{
2574	struct shmem_inode_info *info;
2575	info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2576	if (!info)
2577		return NULL;
2578	return &info->vfs_inode;
2579}
2580
2581static void shmem_destroy_callback(struct rcu_head *head)
2582{
2583	struct inode *inode = container_of(head, struct inode, i_rcu);
 
2584	kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2585}
2586
2587static void shmem_destroy_inode(struct inode *inode)
2588{
2589	if (S_ISREG(inode->i_mode))
2590		mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2591	call_rcu(&inode->i_rcu, shmem_destroy_callback);
2592}
2593
2594static void shmem_init_inode(void *foo)
2595{
2596	struct shmem_inode_info *info = foo;
2597	inode_init_once(&info->vfs_inode);
2598}
2599
2600static int shmem_init_inodecache(void)
2601{
2602	shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2603				sizeof(struct shmem_inode_info),
2604				0, SLAB_PANIC, shmem_init_inode);
2605	return 0;
2606}
2607
2608static void shmem_destroy_inodecache(void)
2609{
2610	kmem_cache_destroy(shmem_inode_cachep);
2611}
2612
2613static const struct address_space_operations shmem_aops = {
2614	.writepage	= shmem_writepage,
2615	.set_page_dirty	= __set_page_dirty_no_writeback,
2616#ifdef CONFIG_TMPFS
2617	.write_begin	= shmem_write_begin,
2618	.write_end	= shmem_write_end,
2619#endif
 
2620	.migratepage	= migrate_page,
 
2621	.error_remove_page = generic_error_remove_page,
2622};
2623
2624static const struct file_operations shmem_file_operations = {
2625	.mmap		= shmem_mmap,
 
2626#ifdef CONFIG_TMPFS
2627	.llseek		= shmem_file_llseek,
2628	.read		= do_sync_read,
2629	.write		= do_sync_write,
2630	.aio_read	= shmem_file_aio_read,
2631	.aio_write	= generic_file_aio_write,
2632	.fsync		= noop_fsync,
2633	.splice_read	= shmem_file_splice_read,
2634	.splice_write	= generic_file_splice_write,
2635	.fallocate	= shmem_fallocate,
2636#endif
2637};
2638
2639static const struct inode_operations shmem_inode_operations = {
 
2640	.setattr	= shmem_setattr,
2641#ifdef CONFIG_TMPFS_XATTR
2642	.setxattr	= shmem_setxattr,
2643	.getxattr	= shmem_getxattr,
2644	.listxattr	= shmem_listxattr,
2645	.removexattr	= shmem_removexattr,
2646	.set_acl	= simple_set_acl,
2647#endif
2648};
2649
2650static const struct inode_operations shmem_dir_inode_operations = {
2651#ifdef CONFIG_TMPFS
2652	.create		= shmem_create,
2653	.lookup		= simple_lookup,
2654	.link		= shmem_link,
2655	.unlink		= shmem_unlink,
2656	.symlink	= shmem_symlink,
2657	.mkdir		= shmem_mkdir,
2658	.rmdir		= shmem_rmdir,
2659	.mknod		= shmem_mknod,
2660	.rename		= shmem_rename,
2661	.tmpfile	= shmem_tmpfile,
2662#endif
2663#ifdef CONFIG_TMPFS_XATTR
2664	.setxattr	= shmem_setxattr,
2665	.getxattr	= shmem_getxattr,
2666	.listxattr	= shmem_listxattr,
2667	.removexattr	= shmem_removexattr,
2668#endif
2669#ifdef CONFIG_TMPFS_POSIX_ACL
2670	.setattr	= shmem_setattr,
2671	.set_acl	= simple_set_acl,
2672#endif
2673};
2674
2675static const struct inode_operations shmem_special_inode_operations = {
2676#ifdef CONFIG_TMPFS_XATTR
2677	.setxattr	= shmem_setxattr,
2678	.getxattr	= shmem_getxattr,
2679	.listxattr	= shmem_listxattr,
2680	.removexattr	= shmem_removexattr,
2681#endif
2682#ifdef CONFIG_TMPFS_POSIX_ACL
2683	.setattr	= shmem_setattr,
2684	.set_acl	= simple_set_acl,
2685#endif
2686};
2687
2688static const struct super_operations shmem_ops = {
2689	.alloc_inode	= shmem_alloc_inode,
 
2690	.destroy_inode	= shmem_destroy_inode,
2691#ifdef CONFIG_TMPFS
2692	.statfs		= shmem_statfs,
2693	.remount_fs	= shmem_remount_fs,
2694	.show_options	= shmem_show_options,
2695#endif
2696	.evict_inode	= shmem_evict_inode,
2697	.drop_inode	= generic_delete_inode,
2698	.put_super	= shmem_put_super,
 
 
 
 
2699};
2700
2701static const struct vm_operations_struct shmem_vm_ops = {
2702	.fault		= shmem_fault,
2703	.map_pages	= filemap_map_pages,
2704#ifdef CONFIG_NUMA
2705	.set_policy     = shmem_set_policy,
2706	.get_policy     = shmem_get_policy,
2707#endif
2708	.remap_pages	= generic_file_remap_pages,
2709};
2710
2711static struct dentry *shmem_mount(struct file_system_type *fs_type,
2712	int flags, const char *dev_name, void *data)
2713{
2714	return mount_nodev(fs_type, flags, data, shmem_fill_super);
 
 
 
 
 
 
 
 
 
 
 
 
2715}
2716
2717static struct file_system_type shmem_fs_type = {
2718	.owner		= THIS_MODULE,
2719	.name		= "tmpfs",
2720	.mount		= shmem_mount,
 
 
 
2721	.kill_sb	= kill_litter_super,
2722	.fs_flags	= FS_USERNS_MOUNT,
2723};
2724
2725int __init shmem_init(void)
2726{
2727	int error;
2728
2729	/* If rootfs called this, don't re-init */
2730	if (shmem_inode_cachep)
2731		return 0;
2732
2733	error = bdi_init(&shmem_backing_dev_info);
2734	if (error)
2735		goto out4;
2736
2737	error = shmem_init_inodecache();
2738	if (error)
2739		goto out3;
2740
2741	error = register_filesystem(&shmem_fs_type);
2742	if (error) {
2743		printk(KERN_ERR "Could not register tmpfs\n");
2744		goto out2;
2745	}
2746
2747	shm_mnt = kern_mount(&shmem_fs_type);
2748	if (IS_ERR(shm_mnt)) {
2749		error = PTR_ERR(shm_mnt);
2750		printk(KERN_ERR "Could not kern_mount tmpfs\n");
2751		goto out1;
2752	}
 
 
 
 
 
 
 
2753	return 0;
2754
2755out1:
2756	unregister_filesystem(&shmem_fs_type);
2757out2:
2758	shmem_destroy_inodecache();
2759out3:
2760	bdi_destroy(&shmem_backing_dev_info);
2761out4:
2762	shm_mnt = ERR_PTR(error);
2763	return error;
2764}
2765
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2766#else /* !CONFIG_SHMEM */
2767
2768/*
2769 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2770 *
2771 * This is intended for small system where the benefits of the full
2772 * shmem code (swap-backed and resource-limited) are outweighed by
2773 * their complexity. On systems without swap this code should be
2774 * effectively equivalent, but much lighter weight.
2775 */
2776
2777static struct file_system_type shmem_fs_type = {
2778	.name		= "tmpfs",
2779	.mount		= ramfs_mount,
 
2780	.kill_sb	= kill_litter_super,
2781	.fs_flags	= FS_USERNS_MOUNT,
2782};
2783
2784int __init shmem_init(void)
2785{
2786	BUG_ON(register_filesystem(&shmem_fs_type) != 0);
2787
2788	shm_mnt = kern_mount(&shmem_fs_type);
2789	BUG_ON(IS_ERR(shm_mnt));
2790
2791	return 0;
2792}
2793
2794int shmem_unuse(swp_entry_t swap, struct page *page)
 
2795{
2796	return 0;
2797}
2798
2799int shmem_lock(struct file *file, int lock, struct user_struct *user)
2800{
2801	return 0;
2802}
2803
2804void shmem_unlock_mapping(struct address_space *mapping)
2805{
2806}
2807
 
 
 
 
 
 
 
 
 
2808void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
2809{
2810	truncate_inode_pages_range(inode->i_mapping, lstart, lend);
2811}
2812EXPORT_SYMBOL_GPL(shmem_truncate_range);
2813
2814#define shmem_vm_ops				generic_file_vm_ops
2815#define shmem_file_operations			ramfs_file_operations
2816#define shmem_get_inode(sb, dir, mode, dev, flags)	ramfs_get_inode(sb, dir, mode, dev)
2817#define shmem_acct_size(flags, size)		0
2818#define shmem_unacct_size(flags, size)		do {} while (0)
2819
2820#endif /* CONFIG_SHMEM */
2821
2822/* common code */
2823
2824static struct dentry_operations anon_ops = {
2825	.d_dname = simple_dname
2826};
2827
2828static struct file *__shmem_file_setup(const char *name, loff_t size,
2829				       unsigned long flags, unsigned int i_flags)
2830{
 
2831	struct file *res;
2832	struct inode *inode;
2833	struct path path;
2834	struct super_block *sb;
2835	struct qstr this;
2836
2837	if (IS_ERR(shm_mnt))
2838		return ERR_CAST(shm_mnt);
2839
2840	if (size < 0 || size > MAX_LFS_FILESIZE)
2841		return ERR_PTR(-EINVAL);
2842
2843	if (shmem_acct_size(flags, size))
2844		return ERR_PTR(-ENOMEM);
2845
2846	res = ERR_PTR(-ENOMEM);
2847	this.name = name;
2848	this.len = strlen(name);
2849	this.hash = 0; /* will go */
2850	sb = shm_mnt->mnt_sb;
2851	path.dentry = d_alloc_pseudo(sb, &this);
2852	if (!path.dentry)
2853		goto put_memory;
2854	d_set_d_op(path.dentry, &anon_ops);
2855	path.mnt = mntget(shm_mnt);
2856
2857	res = ERR_PTR(-ENOSPC);
2858	inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2859	if (!inode)
2860		goto put_dentry;
2861
2862	inode->i_flags |= i_flags;
2863	d_instantiate(path.dentry, inode);
2864	inode->i_size = size;
2865	clear_nlink(inode);	/* It is unlinked */
2866	res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
 
 
 
2867	if (IS_ERR(res))
2868		goto put_dentry;
2869
2870	res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2871		  &shmem_file_operations);
2872	if (IS_ERR(res))
2873		goto put_dentry;
2874
2875	return res;
2876
2877put_dentry:
2878	path_put(&path);
2879put_memory:
2880	shmem_unacct_size(flags, size);
2881	return res;
2882}
2883
2884/**
2885 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
2886 * 	kernel internal.  There will be NO LSM permission checks against the
2887 * 	underlying inode.  So users of this interface must do LSM checks at a
2888 * 	higher layer.  The one user is the big_key implementation.  LSM checks
2889 * 	are provided at the key level rather than the inode level.
2890 * @name: name for dentry (to be seen in /proc/<pid>/maps
2891 * @size: size to be set for the file
2892 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2893 */
2894struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
2895{
2896	return __shmem_file_setup(name, size, flags, S_PRIVATE);
2897}
2898
2899/**
2900 * shmem_file_setup - get an unlinked file living in tmpfs
2901 * @name: name for dentry (to be seen in /proc/<pid>/maps
2902 * @size: size to be set for the file
2903 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2904 */
2905struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2906{
2907	return __shmem_file_setup(name, size, flags, 0);
2908}
2909EXPORT_SYMBOL_GPL(shmem_file_setup);
2910
2911/**
 
 
 
 
 
 
 
 
 
 
 
 
 
 
2912 * shmem_zero_setup - setup a shared anonymous mapping
2913 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2914 */
2915int shmem_zero_setup(struct vm_area_struct *vma)
2916{
2917	struct file *file;
2918	loff_t size = vma->vm_end - vma->vm_start;
2919
2920	file = shmem_file_setup("dev/zero", size, vma->vm_flags);
 
 
 
 
 
 
2921	if (IS_ERR(file))
2922		return PTR_ERR(file);
2923
2924	if (vma->vm_file)
2925		fput(vma->vm_file);
2926	vma->vm_file = file;
2927	vma->vm_ops = &shmem_vm_ops;
 
 
 
 
 
 
 
2928	return 0;
2929}
2930
2931/**
2932 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2933 * @mapping:	the page's address_space
2934 * @index:	the page index
2935 * @gfp:	the page allocator flags to use if allocating
2936 *
2937 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2938 * with any new page allocations done using the specified allocation flags.
2939 * But read_cache_page_gfp() uses the ->readpage() method: which does not
2940 * suit tmpfs, since it may have pages in swapcache, and needs to find those
2941 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
2942 *
2943 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
2944 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
2945 */
2946struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
2947					 pgoff_t index, gfp_t gfp)
2948{
2949#ifdef CONFIG_SHMEM
2950	struct inode *inode = mapping->host;
2951	struct page *page;
2952	int error;
2953
2954	BUG_ON(mapping->a_ops != &shmem_aops);
2955	error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
 
2956	if (error)
2957		page = ERR_PTR(error);
2958	else
2959		unlock_page(page);
2960	return page;
2961#else
2962	/*
2963	 * The tiny !SHMEM case uses ramfs without swap
2964	 */
2965	return read_cache_page_gfp(mapping, index, gfp);
2966#endif
2967}
2968EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);