root/mm/shmem.c

DEFINITIONS

1. shmem_default_max_blocks
2. shmem_default_max_inodes
3. shmem_getpage
4. SHMEM_SB
5. shmem_acct_size
6. shmem_unacct_size
7. shmem_reacct_size
8. shmem_acct_block
9. shmem_unacct_blocks
10. shmem_inode_acct_block
11. shmem_inode_unacct_blocks
12. vma_is_shmem
13. shmem_reserve_inode
14. shmem_free_inode
15. shmem_recalc_inode
16. shmem_charge
17. shmem_uncharge
18. shmem_replace_entry
19. shmem_confirm_swap
20. shmem_parse_huge
21. shmem_format_huge
22. shmem_unused_huge_shrink
23. shmem_unused_huge_scan
24. shmem_unused_huge_count
25. shmem_unused_huge_shrink
26. is_huge_enabled
27. shmem_add_to_page_cache
28. shmem_delete_from_page_cache
29. shmem_free_swap
30. shmem_partial_swap_usage
31. shmem_swap_usage
32. shmem_unlock_mapping
33. shmem_undo_range
34. shmem_truncate_range
35. shmem_getattr
36. shmem_setattr
37. shmem_evict_inode
38. shmem_find_swap_entries
39. shmem_unuse_swap_entries
40. shmem_unuse_inode
41. shmem_unuse
42. shmem_writepage
43. shmem_show_mpol
44. shmem_get_sbmpol
45. shmem_show_mpol
46. shmem_get_sbmpol
47. shmem_pseudo_vma_init
48. shmem_pseudo_vma_destroy
49. shmem_swapin
50. shmem_alloc_hugepage
51. shmem_alloc_page
52. shmem_alloc_and_acct_page
53. shmem_should_replace_page
54. shmem_replace_page
55. shmem_swapin_page
56. shmem_getpage_gfp
57. synchronous_wake_function
58. shmem_fault
59. shmem_get_unmapped_area
60. shmem_set_policy
61. shmem_get_policy
62. shmem_lock
63. shmem_mmap
64. shmem_get_inode
65. shmem_mapping
66. shmem_mfill_atomic_pte
67. shmem_mcopy_atomic_pte
68. shmem_mfill_zeropage_pte
69. shmem_write_begin
70. shmem_write_end
71. shmem_file_read_iter
72. shmem_seek_hole_data
73. shmem_file_llseek
74. shmem_fallocate
75. shmem_statfs
76. shmem_mknod
77. shmem_tmpfile
78. shmem_mkdir
79. shmem_create
80. shmem_link
81. shmem_unlink
82. shmem_rmdir
83. shmem_exchange
84. shmem_whiteout
85. shmem_rename2
86. shmem_symlink
87. shmem_put_link
88. shmem_get_link
89. shmem_initxattrs
90. shmem_xattr_handler_get
91. shmem_xattr_handler_set
92. shmem_listxattr
93. shmem_get_parent
94. shmem_match
95. shmem_find_alias
96. shmem_fh_to_dentry
97. shmem_encode_fh
98. shmem_parse_one
99. shmem_parse_options
100. shmem_reconfigure
101. shmem_show_options
102. shmem_put_super
103. shmem_fill_super
104. shmem_get_tree
105. shmem_free_fc
106. shmem_alloc_inode
107. shmem_free_in_core_inode
108. shmem_destroy_inode
109. shmem_init_inode
110. shmem_init_inodecache
111. shmem_destroy_inodecache
112. shmem_init_fs_context
113. shmem_init
114. shmem_enabled_show
115. shmem_enabled_store
116. shmem_huge_enabled
117. shmem_init
118. shmem_unuse
119. shmem_lock
120. shmem_unlock_mapping
121. shmem_get_unmapped_area
122. shmem_truncate_range
123. __shmem_file_setup
124. shmem_kernel_file_setup
125. shmem_file_setup
126. shmem_file_setup_with_mnt
127. shmem_zero_setup
128. shmem_read_mapping_page_gfp

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