mm/shmem.c

   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) 2002-2005 Hugh Dickins.
  10  * Copyright (C) 2002-2005 VERITAS Software Corporation.
  11  * Copyright (C) 2004 Andi Kleen, SuSE Labs
  12  *
  13  * Extended attribute support for tmpfs:
  14  * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
  15  * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
  16  *
  17  * tiny-shmem:
  18  * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
  19  *
  20  * This file is released under the GPL.
  21  */
  22
  23 #include <linux/fs.h>
  24 #include <linux/init.h>
  25 #include <linux/vfs.h>
  26 #include <linux/mount.h>
  27 #include <linux/pagemap.h>
  28 #include <linux/file.h>
  29 #include <linux/mm.h>
  30 #include <linux/module.h>
  31 #include <linux/swap.h>
  32
  33 static struct vfsmount *shm_mnt;
  34
  35 #ifdef CONFIG_SHMEM
  36 /*
  37  * This virtual memory filesystem is heavily based on the ramfs. It
  38  * extends ramfs by the ability to use swap and honor resource limits
  39  * which makes it a completely usable filesystem.
  40  */
  41
  42 #include <linux/xattr.h>
  43 #include <linux/exportfs.h>
  44 #include <linux/posix_acl.h>
  45 #include <linux/generic_acl.h>
  46 #include <linux/mman.h>
  47 #include <linux/string.h>
  48 #include <linux/slab.h>
  49 #include <linux/backing-dev.h>
  50 #include <linux/shmem_fs.h>
  51 #include <linux/writeback.h>
  52 #include <linux/blkdev.h>
  53 #include <linux/security.h>
  54 #include <linux/swapops.h>
  55 #include <linux/mempolicy.h>
  56 #include <linux/namei.h>
  57 #include <linux/ctype.h>
  58 #include <linux/migrate.h>
  59 #include <linux/highmem.h>
  60 #include <linux/seq_file.h>
  61 #include <linux/magic.h>
  62
  63 #include <asm/uaccess.h>
  64 #include <asm/div64.h>
  65 #include <asm/pgtable.h>
  66
  67 /*
  68  * The maximum size of a shmem/tmpfs file is limited by the maximum size of
  69  * its triple-indirect swap vector - see illustration at shmem_swp_entry().
  70  *
  71  * With 4kB page size, maximum file size is just over 2TB on a 32-bit kernel,
  72  * but one eighth of that on a 64-bit kernel.  With 8kB page size, maximum
  73  * file size is just over 4TB on a 64-bit kernel, but 16TB on a 32-bit kernel,
  74  * MAX_LFS_FILESIZE being then more restrictive than swap vector layout.
  75  *
  76  * We use / and * instead of shifts in the definitions below, so that the swap
  77  * vector can be tested with small even values (e.g. 20) for ENTRIES_PER_PAGE.
  78  */
  79 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
  80 #define ENTRIES_PER_PAGEPAGE ((unsigned long long)ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
  81
  82 #define SHMSWP_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
  83 #define SHMSWP_MAX_BYTES (SHMSWP_MAX_INDEX << PAGE_CACHE_SHIFT)
  84
  85 #define SHMEM_MAX_BYTES  min_t(unsigned long long, SHMSWP_MAX_BYTES, MAX_LFS_FILESIZE)
  86 #define SHMEM_MAX_INDEX  ((unsigned long)((SHMEM_MAX_BYTES+1) >> PAGE_CACHE_SHIFT))
  87
  88 #define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
  89 #define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
  90
  91 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
  92 #define SHMEM_PAGEIN     VM_READ
  93 #define SHMEM_TRUNCATE   VM_WRITE
  94
  95 /* Definition to limit shmem_truncate's steps between cond_rescheds */
  96 #define LATENCY_LIMIT    64
  97
  98 /* Pretend that each entry is of this size in directory's i_size */
  99 #define BOGO_DIRENT_SIZE 20
 100
 101 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
 102 enum sgp_type {
 103         SGP_READ,       /* don't exceed i_size, don't allocate page */
 104         SGP_CACHE,      /* don't exceed i_size, may allocate page */
 105         SGP_DIRTY,      /* like SGP_CACHE, but set new page dirty */
 106         SGP_WRITE,      /* may exceed i_size, may allocate page */
 107 };
 108
 109 #ifdef CONFIG_TMPFS
 110 static unsigned long shmem_default_max_blocks(void)
 111 {
 112         return totalram_pages / 2;
 113 }
 114
 115 static unsigned long shmem_default_max_inodes(void)
 116 {
 117         return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
 118 }
 119 #endif
 120
 121 static int shmem_getpage(struct inode *inode, unsigned long idx,
 122                          struct page **pagep, enum sgp_type sgp, int *type);
 123
 124 static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
 125 {
 126         /*
 127          * The above definition of ENTRIES_PER_PAGE, and the use of
 128          * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
 129          * might be reconsidered if it ever diverges from PAGE_SIZE.
 130          *
 131          * Mobility flags are masked out as swap vectors cannot move
 132          */
 133         return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO,
 134                                 PAGE_CACHE_SHIFT-PAGE_SHIFT);
 135 }
 136
 137 static inline void shmem_dir_free(struct page *page)
 138 {
 139         __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
 140 }
 141
 142 static struct page **shmem_dir_map(struct page *page)
 143 {
 144         return (struct page **)kmap_atomic(page, KM_USER0);
 145 }
 146
 147 static inline void shmem_dir_unmap(struct page **dir)
 148 {
 149         kunmap_atomic(dir, KM_USER0);
 150 }
 151
 152 static swp_entry_t *shmem_swp_map(struct page *page)
 153 {
 154         return (swp_entry_t *)kmap_atomic(page, KM_USER1);
 155 }
 156
 157 static inline void shmem_swp_balance_unmap(void)
 158 {
 159         /*
 160          * When passing a pointer to an i_direct entry, to code which
 161          * also handles indirect entries and so will shmem_swp_unmap,
 162          * we must arrange for the preempt count to remain in balance.
 163          * What kmap_atomic of a lowmem page does depends on config
 164          * and architecture, so pretend to kmap_atomic some lowmem page.
 165          */
 166         (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
 167 }
 168
 169 static inline void shmem_swp_unmap(swp_entry_t *entry)
 170 {
 171         kunmap_atomic(entry, KM_USER1);
 172 }
 173
 174 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
 175 {
 176         return sb->s_fs_info;
 177 }
 178
 179 /*
 180  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
 181  * for shared memory and for shared anonymous (/dev/zero) mappings
 182  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
 183  * consistent with the pre-accounting of private mappings ...
 184  */
 185 static inline int shmem_acct_size(unsigned long flags, loff_t size)
 186 {
 187         return (flags & VM_NORESERVE) ?
 188                 0 : security_vm_enough_memory_kern(VM_ACCT(size));
 189 }
 190
 191 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
 192 {
 193         if (!(flags & VM_NORESERVE))
 194                 vm_unacct_memory(VM_ACCT(size));
 195 }
 196
 197 /*
 198  * ... whereas tmpfs objects are accounted incrementally as
 199  * pages are allocated, in order to allow huge sparse files.
 200  * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
 201  * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
 202  */
 203 static inline int shmem_acct_block(unsigned long flags)
 204 {
 205         return (flags & VM_NORESERVE) ?
 206                 security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE)) : 0;
 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_CACHE_SIZE));
 213 }
 214
 215 static const struct super_operations shmem_ops;
 216 static const struct address_space_operations shmem_aops;
 217 static const struct file_operations shmem_file_operations;
 218 static const struct inode_operations shmem_inode_operations;
 219 static const struct inode_operations shmem_dir_inode_operations;
 220 static const struct inode_operations shmem_special_inode_operations;
 221 static const struct vm_operations_struct shmem_vm_ops;
 222
 223 static struct backing_dev_info shmem_backing_dev_info  __read_mostly = {
 224         .ra_pages       = 0,    /* No readahead */
 225         .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
 226         .unplug_io_fn   = default_unplug_io_fn,
 227 };
 228
 229 static LIST_HEAD(shmem_swaplist);
 230 static DEFINE_MUTEX(shmem_swaplist_mutex);
 231
 232 static void shmem_free_blocks(struct inode *inode, long pages)
 233 {
 234         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 235         if (sbinfo->max_blocks) {
 236                 spin_lock(&sbinfo->stat_lock);
 237                 sbinfo->free_blocks += pages;
 238                 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
 239                 spin_unlock(&sbinfo->stat_lock);
 240         }
 241 }
 242
 243 static int shmem_reserve_inode(struct super_block *sb)
 244 {
 245         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 246         if (sbinfo->max_inodes) {
 247                 spin_lock(&sbinfo->stat_lock);
 248                 if (!sbinfo->free_inodes) {
 249                         spin_unlock(&sbinfo->stat_lock);
 250                         return -ENOSPC;
 251                 }
 252                 sbinfo->free_inodes--;
 253                 spin_unlock(&sbinfo->stat_lock);
 254         }
 255         return 0;
 256 }
 257
 258 static void shmem_free_inode(struct super_block *sb)
 259 {
 260         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 261         if (sbinfo->max_inodes) {
 262                 spin_lock(&sbinfo->stat_lock);
 263                 sbinfo->free_inodes++;
 264                 spin_unlock(&sbinfo->stat_lock);
 265         }
 266 }
 267
 268 /**
 269  * shmem_recalc_inode - recalculate the size of an inode
 270  * @inode: inode to recalc
 271  *
 272  * We have to calculate the free blocks since the mm can drop
 273  * undirtied hole pages behind our back.
 274  *
 275  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
 276  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
 277  *
 278  * It has to be called with the spinlock held.
 279  */
 280 static void shmem_recalc_inode(struct inode *inode)
 281 {
 282         struct shmem_inode_info *info = SHMEM_I(inode);
 283         long freed;
 284
 285         freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
 286         if (freed > 0) {
 287                 info->alloced -= freed;
 288                 shmem_unacct_blocks(info->flags, freed);
 289                 shmem_free_blocks(inode, freed);
 290         }
 291 }
 292
 293 /**
 294  * shmem_swp_entry - find the swap vector position in the info structure
 295  * @info:  info structure for the inode
 296  * @index: index of the page to find
 297  * @page:  optional page to add to the structure. Has to be preset to
 298  *         all zeros
 299  *
 300  * If there is no space allocated yet it will return NULL when
 301  * page is NULL, else it will use the page for the needed block,
 302  * setting it to NULL on return to indicate that it has been used.
 303  *
 304  * The swap vector is organized the following way:
 305  *
 306  * There are SHMEM_NR_DIRECT entries directly stored in the
 307  * shmem_inode_info structure. So small files do not need an addional
 308  * allocation.
 309  *
 310  * For pages with index > SHMEM_NR_DIRECT there is the pointer
 311  * i_indirect which points to a page which holds in the first half
 312  * doubly indirect blocks, in the second half triple indirect blocks:
 313  *
 314  * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
 315  * following layout (for SHMEM_NR_DIRECT == 16):
 316  *
 317  * i_indirect -> dir --> 16-19
 318  *            |      +-> 20-23
 319  *            |
 320  *            +-->dir2 --> 24-27
 321  *            |        +-> 28-31
 322  *            |        +-> 32-35
 323  *            |        +-> 36-39
 324  *            |
 325  *            +-->dir3 --> 40-43
 326  *                     +-> 44-47
 327  *                     +-> 48-51
 328  *                     +-> 52-55
 329  */
 330 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
 331 {
 332         unsigned long offset;
 333         struct page **dir;
 334         struct page *subdir;
 335
 336         if (index < SHMEM_NR_DIRECT) {
 337                 shmem_swp_balance_unmap();
 338                 return info->i_direct+index;
 339         }
 340         if (!info->i_indirect) {
 341                 if (page) {
 342                         info->i_indirect = *page;
 343                         *page = NULL;
 344                 }
 345                 return NULL;                    /* need another page */
 346         }
 347
 348         index -= SHMEM_NR_DIRECT;
 349         offset = index % ENTRIES_PER_PAGE;
 350         index /= ENTRIES_PER_PAGE;
 351         dir = shmem_dir_map(info->i_indirect);
 352
 353         if (index >= ENTRIES_PER_PAGE/2) {
 354                 index -= ENTRIES_PER_PAGE/2;
 355                 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
 356                 index %= ENTRIES_PER_PAGE;
 357                 subdir = *dir;
 358                 if (!subdir) {
 359                         if (page) {
 360                                 *dir = *page;
 361                                 *page = NULL;
 362                         }
 363                         shmem_dir_unmap(dir);
 364                         return NULL;            /* need another page */
 365                 }
 366                 shmem_dir_unmap(dir);
 367                 dir = shmem_dir_map(subdir);
 368         }
 369
 370         dir += index;
 371         subdir = *dir;
 372         if (!subdir) {
 373                 if (!page || !(subdir = *page)) {
 374                         shmem_dir_unmap(dir);
 375                         return NULL;            /* need a page */
 376                 }
 377                 *dir = subdir;
 378                 *page = NULL;
 379         }
 380         shmem_dir_unmap(dir);
 381         return shmem_swp_map(subdir) + offset;
 382 }
 383
 384 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
 385 {
 386         long incdec = value? 1: -1;
 387
 388         entry->val = value;
 389         info->swapped += incdec;
 390         if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
 391                 struct page *page = kmap_atomic_to_page(entry);
 392                 set_page_private(page, page_private(page) + incdec);
 393         }
 394 }
 395
 396 /**
 397  * shmem_swp_alloc - get the position of the swap entry for the page.
 398  * @info:       info structure for the inode
 399  * @index:      index of the page to find
 400  * @sgp:        check and recheck i_size? skip allocation?
 401  *
 402  * If the entry does not exist, allocate it.
 403  */
 404 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
 405 {
 406         struct inode *inode = &info->vfs_inode;
 407         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 408         struct page *page = NULL;
 409         swp_entry_t *entry;
 410
 411         if (sgp != SGP_WRITE &&
 412             ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
 413                 return ERR_PTR(-EINVAL);
 414
 415         while (!(entry = shmem_swp_entry(info, index, &page))) {
 416                 if (sgp == SGP_READ)
 417                         return shmem_swp_map(ZERO_PAGE(0));
 418                 /*
 419                  * Test free_blocks against 1 not 0, since we have 1 data
 420                  * page (and perhaps indirect index pages) yet to allocate:
 421                  * a waste to allocate index if we cannot allocate data.
 422                  */
 423                 if (sbinfo->max_blocks) {
 424                         spin_lock(&sbinfo->stat_lock);
 425                         if (sbinfo->free_blocks <= 1) {
 426                                 spin_unlock(&sbinfo->stat_lock);
 427                                 return ERR_PTR(-ENOSPC);
 428                         }
 429                         sbinfo->free_blocks--;
 430                         inode->i_blocks += BLOCKS_PER_PAGE;
 431                         spin_unlock(&sbinfo->stat_lock);
 432                 }
 433
 434                 spin_unlock(&info->lock);
 435                 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
 436                 spin_lock(&info->lock);
 437
 438                 if (!page) {
 439                         shmem_free_blocks(inode, 1);
 440                         return ERR_PTR(-ENOMEM);
 441                 }
 442                 if (sgp != SGP_WRITE &&
 443                     ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
 444                         entry = ERR_PTR(-EINVAL);
 445                         break;
 446                 }
 447                 if (info->next_index <= index)
 448                         info->next_index = index + 1;
 449         }
 450         if (page) {
 451                 /* another task gave its page, or truncated the file */
 452                 shmem_free_blocks(inode, 1);
 453                 shmem_dir_free(page);
 454         }
 455         if (info->next_index <= index && !IS_ERR(entry))
 456                 info->next_index = index + 1;
 457         return entry;
 458 }
 459
 460 /**
 461  * shmem_free_swp - free some swap entries in a directory
 462  * @dir:        pointer to the directory
 463  * @edir:       pointer after last entry of the directory
 464  * @punch_lock: pointer to spinlock when needed for the holepunch case
 465  */
 466 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
 467                                                 spinlock_t *punch_lock)
 468 {
 469         spinlock_t *punch_unlock = NULL;
 470         swp_entry_t *ptr;
 471         int freed = 0;
 472
 473         for (ptr = dir; ptr < edir; ptr++) {
 474                 if (ptr->val) {
 475                         if (unlikely(punch_lock)) {
 476                                 punch_unlock = punch_lock;
 477                                 punch_lock = NULL;
 478                                 spin_lock(punch_unlock);
 479                                 if (!ptr->val)
 480                                         continue;
 481                         }
 482                         free_swap_and_cache(*ptr);
 483                         *ptr = (swp_entry_t){0};
 484                         freed++;
 485                 }
 486         }
 487         if (punch_unlock)
 488                 spin_unlock(punch_unlock);
 489         return freed;
 490 }
 491
 492 static int shmem_map_and_free_swp(struct page *subdir, int offset,
 493                 int limit, struct page ***dir, spinlock_t *punch_lock)
 494 {
 495         swp_entry_t *ptr;
 496         int freed = 0;
 497
 498         ptr = shmem_swp_map(subdir);
 499         for (; offset < limit; offset += LATENCY_LIMIT) {
 500                 int size = limit - offset;
 501                 if (size > LATENCY_LIMIT)
 502                         size = LATENCY_LIMIT;
 503                 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
 504                                                         punch_lock);
 505                 if (need_resched()) {
 506                         shmem_swp_unmap(ptr);
 507                         if (*dir) {
 508                                 shmem_dir_unmap(*dir);
 509                                 *dir = NULL;
 510                         }
 511                         cond_resched();
 512                         ptr = shmem_swp_map(subdir);
 513                 }
 514         }
 515         shmem_swp_unmap(ptr);
 516         return freed;
 517 }
 518
 519 static void shmem_free_pages(struct list_head *next)
 520 {
 521         struct page *page;
 522         int freed = 0;
 523
 524         do {
 525                 page = container_of(next, struct page, lru);
 526                 next = next->next;
 527                 shmem_dir_free(page);
 528                 freed++;
 529                 if (freed >= LATENCY_LIMIT) {
 530                         cond_resched();
 531                         freed = 0;
 532                 }
 533         } while (next);
 534 }
 535
 536 static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
 537 {
 538         struct shmem_inode_info *info = SHMEM_I(inode);
 539         unsigned long idx;
 540         unsigned long size;
 541         unsigned long limit;
 542         unsigned long stage;
 543         unsigned long diroff;
 544         struct page **dir;
 545         struct page *topdir;
 546         struct page *middir;
 547         struct page *subdir;
 548         swp_entry_t *ptr;
 549         LIST_HEAD(pages_to_free);
 550         long nr_pages_to_free = 0;
 551         long nr_swaps_freed = 0;
 552         int offset;
 553         int freed;
 554         int punch_hole;
 555         spinlock_t *needs_lock;
 556         spinlock_t *punch_lock;
 557         unsigned long upper_limit;
 558
 559         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 560         idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 561         if (idx >= info->next_index)
 562                 return;
 563
 564         spin_lock(&info->lock);
 565         info->flags |= SHMEM_TRUNCATE;
 566         if (likely(end == (loff_t) -1)) {
 567                 limit = info->next_index;
 568                 upper_limit = SHMEM_MAX_INDEX;
 569                 info->next_index = idx;
 570                 needs_lock = NULL;
 571                 punch_hole = 0;
 572         } else {
 573                 if (end + 1 >= inode->i_size) { /* we may free a little more */
 574                         limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
 575                                                         PAGE_CACHE_SHIFT;
 576                         upper_limit = SHMEM_MAX_INDEX;
 577                 } else {
 578                         limit = (end + 1) >> PAGE_CACHE_SHIFT;
 579                         upper_limit = limit;
 580                 }
 581                 needs_lock = &info->lock;
 582                 punch_hole = 1;
 583         }
 584
 585         topdir = info->i_indirect;
 586         if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
 587                 info->i_indirect = NULL;
 588                 nr_pages_to_free++;
 589                 list_add(&topdir->lru, &pages_to_free);
 590         }
 591         spin_unlock(&info->lock);
 592
 593         if (info->swapped && idx < SHMEM_NR_DIRECT) {
 594                 ptr = info->i_direct;
 595                 size = limit;
 596                 if (size > SHMEM_NR_DIRECT)
 597                         size = SHMEM_NR_DIRECT;
 598                 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
 599         }
 600
 601         /*
 602          * If there are no indirect blocks or we are punching a hole
 603          * below indirect blocks, nothing to be done.
 604          */
 605         if (!topdir || limit <= SHMEM_NR_DIRECT)
 606                 goto done2;
 607
 608         /*
 609          * The truncation case has already dropped info->lock, and we're safe
 610          * because i_size and next_index have already been lowered, preventing
 611          * access beyond.  But in the punch_hole case, we still need to take
 612          * the lock when updating the swap directory, because there might be
 613          * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
 614          * shmem_writepage.  However, whenever we find we can remove a whole
 615          * directory page (not at the misaligned start or end of the range),
 616          * we first NULLify its pointer in the level above, and then have no
 617          * need to take the lock when updating its contents: needs_lock and
 618          * punch_lock (either pointing to info->lock or NULL) manage this.
 619          */
 620
 621         upper_limit -= SHMEM_NR_DIRECT;
 622         limit -= SHMEM_NR_DIRECT;
 623         idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
 624         offset = idx % ENTRIES_PER_PAGE;
 625         idx -= offset;
 626
 627         dir = shmem_dir_map(topdir);
 628         stage = ENTRIES_PER_PAGEPAGE/2;
 629         if (idx < ENTRIES_PER_PAGEPAGE/2) {
 630                 middir = topdir;
 631                 diroff = idx/ENTRIES_PER_PAGE;
 632         } else {
 633                 dir += ENTRIES_PER_PAGE/2;
 634                 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
 635                 while (stage <= idx)
 636                         stage += ENTRIES_PER_PAGEPAGE;
 637                 middir = *dir;
 638                 if (*dir) {
 639                         diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
 640                                 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
 641                         if (!diroff && !offset && upper_limit >= stage) {
 642                                 if (needs_lock) {
 643                                         spin_lock(needs_lock);
 644                                         *dir = NULL;
 645                                         spin_unlock(needs_lock);
 646                                         needs_lock = NULL;
 647                                 } else
 648                                         *dir = NULL;
 649                                 nr_pages_to_free++;
 650                                 list_add(&middir->lru, &pages_to_free);
 651                         }
 652                         shmem_dir_unmap(dir);
 653                         dir = shmem_dir_map(middir);
 654                 } else {
 655                         diroff = 0;
 656                         offset = 0;
 657                         idx = stage;
 658                 }
 659         }
 660
 661         for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
 662                 if (unlikely(idx == stage)) {
 663                         shmem_dir_unmap(dir);
 664                         dir = shmem_dir_map(topdir) +
 665                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
 666                         while (!*dir) {
 667                                 dir++;
 668                                 idx += ENTRIES_PER_PAGEPAGE;
 669                                 if (idx >= limit)
 670                                         goto done1;
 671                         }
 672                         stage = idx + ENTRIES_PER_PAGEPAGE;
 673                         middir = *dir;
 674                         if (punch_hole)
 675                                 needs_lock = &info->lock;
 676                         if (upper_limit >= stage) {
 677                                 if (needs_lock) {
 678                                         spin_lock(needs_lock);
 679                                         *dir = NULL;
 680                                         spin_unlock(needs_lock);
 681                                         needs_lock = NULL;
 682                                 } else
 683                                         *dir = NULL;
 684                                 nr_pages_to_free++;
 685                                 list_add(&middir->lru, &pages_to_free);
 686                         }
 687                         shmem_dir_unmap(dir);
 688                         cond_resched();
 689                         dir = shmem_dir_map(middir);
 690                         diroff = 0;
 691                 }
 692                 punch_lock = needs_lock;
 693                 subdir = dir[diroff];
 694                 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
 695                         if (needs_lock) {
 696                                 spin_lock(needs_lock);
 697                                 dir[diroff] = NULL;
 698                                 spin_unlock(needs_lock);
 699                                 punch_lock = NULL;
 700                         } else
 701                                 dir[diroff] = NULL;
 702                         nr_pages_to_free++;
 703                         list_add(&subdir->lru, &pages_to_free);
 704                 }
 705                 if (subdir && page_private(subdir) /* has swap entries */) {
 706                         size = limit - idx;
 707                         if (size > ENTRIES_PER_PAGE)
 708                                 size = ENTRIES_PER_PAGE;
 709                         freed = shmem_map_and_free_swp(subdir,
 710                                         offset, size, &dir, punch_lock);
 711                         if (!dir)
 712                                 dir = shmem_dir_map(middir);
 713                         nr_swaps_freed += freed;
 714                         if (offset || punch_lock) {
 715                                 spin_lock(&info->lock);
 716                                 set_page_private(subdir,
 717                                         page_private(subdir) - freed);
 718                                 spin_unlock(&info->lock);
 719                         } else
 720                                 BUG_ON(page_private(subdir) != freed);
 721                 }
 722                 offset = 0;
 723         }
 724 done1:
 725         shmem_dir_unmap(dir);
 726 done2:
 727         if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
 728                 /*
 729                  * Call truncate_inode_pages again: racing shmem_unuse_inode
 730                  * may have swizzled a page in from swap since vmtruncate or
 731                  * generic_delete_inode did it, before we lowered next_index.
 732                  * Also, though shmem_getpage checks i_size before adding to
 733                  * cache, no recheck after: so fix the narrow window there too.
 734                  *
 735                  * Recalling truncate_inode_pages_range and unmap_mapping_range
 736                  * every time for punch_hole (which never got a chance to clear
 737                  * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
 738                  * yet hardly ever necessary: try to optimize them out later.
 739                  */
 740                 truncate_inode_pages_range(inode->i_mapping, start, end);
 741                 if (punch_hole)
 742                         unmap_mapping_range(inode->i_mapping, start,
 743                                                         end - start, 1);
 744         }
 745
 746         spin_lock(&info->lock);
 747         info->flags &= ~SHMEM_TRUNCATE;
 748         info->swapped -= nr_swaps_freed;
 749         if (nr_pages_to_free)
 750                 shmem_free_blocks(inode, nr_pages_to_free);
 751         shmem_recalc_inode(inode);
 752         spin_unlock(&info->lock);
 753
 754         /*
 755          * Empty swap vector directory pages to be freed?
 756          */
 757         if (!list_empty(&pages_to_free)) {
 758                 pages_to_free.prev->next = NULL;
 759                 shmem_free_pages(pages_to_free.next);
 760         }
 761 }
 762
 763 static void shmem_truncate(struct inode *inode)
 764 {
 765         shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
 766 }
 767
 768 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
 769 {
 770         struct inode *inode = dentry->d_inode;
 771         struct page *page = NULL;
 772         int error;
 773
 774         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
 775                 if (attr->ia_size < inode->i_size) {
 776                         /*
 777                          * If truncating down to a partial page, then
 778                          * if that page is already allocated, hold it
 779                          * in memory until the truncation is over, so
 780                          * truncate_partial_page cannnot miss it were
 781                          * it assigned to swap.
 782                          */
 783                         if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
 784                                 (void) shmem_getpage(inode,
 785                                         attr->ia_size>>PAGE_CACHE_SHIFT,
 786                                                 &page, SGP_READ, NULL);
 787                                 if (page)
 788                                         unlock_page(page);
 789                         }
 790                         /*
 791                          * Reset SHMEM_PAGEIN flag so that shmem_truncate can
 792                          * detect if any pages might have been added to cache
 793                          * after truncate_inode_pages.  But we needn't bother
 794                          * if it's being fully truncated to zero-length: the
 795                          * nrpages check is efficient enough in that case.
 796                          */
 797                         if (attr->ia_size) {
 798                                 struct shmem_inode_info *info = SHMEM_I(inode);
 799                                 spin_lock(&info->lock);
 800                                 info->flags &= ~SHMEM_PAGEIN;
 801                                 spin_unlock(&info->lock);
 802                         }
 803                 }
 804         }
 805
 806         error = inode_change_ok(inode, attr);
 807         if (!error)
 808                 error = inode_setattr(inode, attr);
 809 #ifdef CONFIG_TMPFS_POSIX_ACL
 810         if (!error && (attr->ia_valid & ATTR_MODE))
 811                 error = generic_acl_chmod(inode);
 812 #endif
 813         if (page)
 814                 page_cache_release(page);
 815         return error;
 816 }
 817
 818 static void shmem_delete_inode(struct inode *inode)
 819 {
 820         struct shmem_inode_info *info = SHMEM_I(inode);
 821
 822         if (inode->i_op->truncate == shmem_truncate) {
 823                 truncate_inode_pages(inode->i_mapping, 0);
 824                 shmem_unacct_size(info->flags, inode->i_size);
 825                 inode->i_size = 0;
 826                 shmem_truncate(inode);
 827                 if (!list_empty(&info->swaplist)) {
 828                         mutex_lock(&shmem_swaplist_mutex);
 829                         list_del_init(&info->swaplist);
 830                         mutex_unlock(&shmem_swaplist_mutex);
 831                 }
 832         }
 833         BUG_ON(inode->i_blocks);
 834         shmem_free_inode(inode->i_sb);
 835         clear_inode(inode);
 836 }
 837
 838 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
 839 {
 840         swp_entry_t *ptr;
 841
 842         for (ptr = dir; ptr < edir; ptr++) {
 843                 if (ptr->val == entry.val)
 844                         return ptr - dir;
 845         }
 846         return -1;
 847 }
 848
 849 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
 850 {
 851         struct inode *inode;
 852         unsigned long idx;
 853         unsigned long size;
 854         unsigned long limit;
 855         unsigned long stage;
 856         struct page **dir;
 857         struct page *subdir;
 858         swp_entry_t *ptr;
 859         int offset;
 860         int error;
 861
 862         idx = 0;
 863         ptr = info->i_direct;
 864         spin_lock(&info->lock);
 865         if (!info->swapped) {
 866                 list_del_init(&info->swaplist);
 867                 goto lost2;
 868         }
 869         limit = info->next_index;
 870         size = limit;
 871         if (size > SHMEM_NR_DIRECT)
 872                 size = SHMEM_NR_DIRECT;
 873         offset = shmem_find_swp(entry, ptr, ptr+size);
 874         if (offset >= 0)
 875                 goto found;
 876         if (!info->i_indirect)
 877                 goto lost2;
 878
 879         dir = shmem_dir_map(info->i_indirect);
 880         stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
 881
 882         for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
 883                 if (unlikely(idx == stage)) {
 884                         shmem_dir_unmap(dir-1);
 885                         if (cond_resched_lock(&info->lock)) {
 886                                 /* check it has not been truncated */
 887                                 if (limit > info->next_index) {
 888                                         limit = info->next_index;
 889                                         if (idx >= limit)
 890                                                 goto lost2;
 891                                 }
 892                         }
 893                         dir = shmem_dir_map(info->i_indirect) +
 894                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
 895                         while (!*dir) {
 896                                 dir++;
 897                                 idx += ENTRIES_PER_PAGEPAGE;
 898                                 if (idx >= limit)
 899                                         goto lost1;
 900                         }
 901                         stage = idx + ENTRIES_PER_PAGEPAGE;
 902                         subdir = *dir;
 903                         shmem_dir_unmap(dir);
 904                         dir = shmem_dir_map(subdir);
 905                 }
 906                 subdir = *dir;
 907                 if (subdir && page_private(subdir)) {
 908                         ptr = shmem_swp_map(subdir);
 909                         size = limit - idx;
 910                         if (size > ENTRIES_PER_PAGE)
 911                                 size = ENTRIES_PER_PAGE;
 912                         offset = shmem_find_swp(entry, ptr, ptr+size);
 913                         shmem_swp_unmap(ptr);
 914                         if (offset >= 0) {
 915                                 shmem_dir_unmap(dir);
 916                                 goto found;
 917                         }
 918                 }
 919         }
 920 lost1:
 921         shmem_dir_unmap(dir-1);
 922 lost2:
 923         spin_unlock(&info->lock);
 924         return 0;
 925 found:
 926         idx += offset;
 927         inode = igrab(&info->vfs_inode);
 928         spin_unlock(&info->lock);
 929
 930         /*
 931          * Move _head_ to start search for next from here.
 932          * But be careful: shmem_delete_inode checks list_empty without taking
 933          * mutex, and there's an instant in list_move_tail when info->swaplist
 934          * would appear empty, if it were the only one on shmem_swaplist.  We
 935          * could avoid doing it if inode NULL; or use this minor optimization.
 936          */
 937         if (shmem_swaplist.next != &info->swaplist)
 938                 list_move_tail(&shmem_swaplist, &info->swaplist);
 939         mutex_unlock(&shmem_swaplist_mutex);
 940
 941         error = 1;
 942         if (!inode)
 943                 goto out;
 944         /*
 945          * Charge page using GFP_KERNEL while we can wait.
 946          * Charged back to the user(not to caller) when swap account is used.
 947          * add_to_page_cache() will be called with GFP_NOWAIT.
 948          */
 949         error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
 950         if (error)
 951                 goto out;
 952         error = radix_tree_preload(GFP_KERNEL);
 953         if (error) {
 954                 mem_cgroup_uncharge_cache_page(page);
 955                 goto out;
 956         }
 957         error = 1;
 958
 959         spin_lock(&info->lock);
 960         ptr = shmem_swp_entry(info, idx, NULL);
 961         if (ptr && ptr->val == entry.val) {
 962                 error = add_to_page_cache_locked(page, inode->i_mapping,
 963                                                 idx, GFP_NOWAIT);
 964                 /* does mem_cgroup_uncharge_cache_page on error */
 965         } else  /* we must compensate for our precharge above */
 966                 mem_cgroup_uncharge_cache_page(page);
 967
 968         if (error == -EEXIST) {
 969                 struct page *filepage = find_get_page(inode->i_mapping, idx);
 970                 error = 1;
 971                 if (filepage) {
 972                         /*
 973                          * There might be a more uptodate page coming down
 974                          * from a stacked writepage: forget our swappage if so.
 975                          */
 976                         if (PageUptodate(filepage))
 977                                 error = 0;
 978                         page_cache_release(filepage);
 979                 }
 980         }
 981         if (!error) {
 982                 delete_from_swap_cache(page);
 983                 set_page_dirty(page);
 984                 info->flags |= SHMEM_PAGEIN;
 985                 shmem_swp_set(info, ptr, 0);
 986                 swap_free(entry);
 987                 error = 1;      /* not an error, but entry was found */
 988         }
 989         if (ptr)
 990                 shmem_swp_unmap(ptr);
 991         spin_unlock(&info->lock);
 992         radix_tree_preload_end();
 993 out:
 994         unlock_page(page);
 995         page_cache_release(page);
 996         iput(inode);            /* allows for NULL */
 997         return error;
 998 }
 999
1000 /*
1001  * shmem_unuse() search for an eventually swapped out shmem page.
1002  */
1003 int shmem_unuse(swp_entry_t entry, struct page *page)
1004 {
1005         struct list_head *p, *next;
1006         struct shmem_inode_info *info;
1007         int found = 0;
1008
1009         mutex_lock(&shmem_swaplist_mutex);
1010         list_for_each_safe(p, next, &shmem_swaplist) {
1011                 info = list_entry(p, struct shmem_inode_info, swaplist);
1012                 found = shmem_unuse_inode(info, entry, page);
1013                 cond_resched();
1014                 if (found)
1015                         goto out;
1016         }
1017         mutex_unlock(&shmem_swaplist_mutex);
1018         /*
1019          * Can some race bring us here?  We've been holding page lock,
1020          * so I think not; but would rather try again later than BUG()
1021          */
1022         unlock_page(page);
1023         page_cache_release(page);
1024 out:
1025         return (found < 0) ? found : 0;
1026 }
1027
1028 /*
1029  * Move the page from the page cache to the swap cache.
1030  */
1031 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1032 {
1033         struct shmem_inode_info *info;
1034         swp_entry_t *entry, swap;
1035         struct address_space *mapping;
1036         unsigned long index;
1037         struct inode *inode;
1038
1039         BUG_ON(!PageLocked(page));
1040         mapping = page->mapping;
1041         index = page->index;
1042         inode = mapping->host;
1043         info = SHMEM_I(inode);
1044         if (info->flags & VM_LOCKED)
1045                 goto redirty;
1046         if (!total_swap_pages)
1047                 goto redirty;
1048
1049         /*
1050          * shmem_backing_dev_info's capabilities prevent regular writeback or
1051          * sync from ever calling shmem_writepage; but a stacking filesystem
1052          * may use the ->writepage of its underlying filesystem, in which case
1053          * tmpfs should write out to swap only in response to memory pressure,
1054          * and not for the writeback threads or sync.  However, in those cases,
1055          * we do still want to check if there's a redundant swappage to be
1056          * discarded.
1057          */
1058         if (wbc->for_reclaim)
1059                 swap = get_swap_page();
1060         else
1061                 swap.val = 0;
1062
1063         spin_lock(&info->lock);
1064         if (index >= info->next_index) {
1065                 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
1066                 goto unlock;
1067         }
1068         entry = shmem_swp_entry(info, index, NULL);
1069         if (entry->val) {
1070                 /*
1071                  * The more uptodate page coming down from a stacked
1072                  * writepage should replace our old swappage.
1073                  */
1074                 free_swap_and_cache(*entry);
1075                 shmem_swp_set(info, entry, 0);
1076         }
1077         shmem_recalc_inode(inode);
1078
1079         if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1080                 remove_from_page_cache(page);
1081                 shmem_swp_set(info, entry, swap.val);
1082                 shmem_swp_unmap(entry);
1083                 if (list_empty(&info->swaplist))
1084                         inode = igrab(inode);
1085                 else
1086                         inode = NULL;
1087                 spin_unlock(&info->lock);
1088                 swap_shmem_alloc(swap);
1089                 BUG_ON(page_mapped(page));
1090                 page_cache_release(page);       /* pagecache ref */
1091                 swap_writepage(page, wbc);
1092                 if (inode) {
1093                         mutex_lock(&shmem_swaplist_mutex);
1094                         /* move instead of add in case we're racing */
1095                         list_move_tail(&info->swaplist, &shmem_swaplist);
1096                         mutex_unlock(&shmem_swaplist_mutex);
1097                         iput(inode);
1098                 }
1099                 return 0;
1100         }
1101
1102         shmem_swp_unmap(entry);
1103 unlock:
1104         spin_unlock(&info->lock);
1105         /*
1106          * add_to_swap_cache() doesn't return -EEXIST, so we can safely
1107          * clear SWAP_HAS_CACHE flag.
1108          */
1109         swapcache_free(swap, NULL);
1110 redirty:
1111         set_page_dirty(page);
1112         if (wbc->for_reclaim)
1113                 return AOP_WRITEPAGE_ACTIVATE;  /* Return with page locked */
1114         unlock_page(page);
1115         return 0;
1116 }
1117
1118 #ifdef CONFIG_NUMA
1119 #ifdef CONFIG_TMPFS
1120 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1121 {
1122         char buffer[64];
1123
1124         if (!mpol || mpol->mode == MPOL_DEFAULT)
1125                 return;         /* show nothing */
1126
1127         mpol_to_str(buffer, sizeof(buffer), mpol, 1);
1128
1129         seq_printf(seq, ",mpol=%s", buffer);
1130 }
1131
1132 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1133 {
1134         struct mempolicy *mpol = NULL;
1135         if (sbinfo->mpol) {
1136                 spin_lock(&sbinfo->stat_lock);  /* prevent replace/use races */
1137                 mpol = sbinfo->mpol;
1138                 mpol_get(mpol);
1139                 spin_unlock(&sbinfo->stat_lock);
1140         }
1141         return mpol;
1142 }
1143 #endif /* CONFIG_TMPFS */
1144
1145 static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1146                         struct shmem_inode_info *info, unsigned long idx)
1147 {
1148         struct mempolicy mpol, *spol;
1149         struct vm_area_struct pvma;
1150         struct page *page;
1151
1152         spol = mpol_cond_copy(&mpol,
1153                                 mpol_shared_policy_lookup(&info->policy, idx));
1154
1155         /* Create a pseudo vma that just contains the policy */
1156         pvma.vm_start = 0;
1157         pvma.vm_pgoff = idx;
1158         pvma.vm_ops = NULL;
1159         pvma.vm_policy = spol;
1160         page = swapin_readahead(entry, gfp, &pvma, 0);
1161         return page;
1162 }
1163
1164 static struct page *shmem_alloc_page(gfp_t gfp,
1165                         struct shmem_inode_info *info, unsigned long idx)
1166 {
1167         struct vm_area_struct pvma;
1168
1169         /* Create a pseudo vma that just contains the policy */
1170         pvma.vm_start = 0;
1171         pvma.vm_pgoff = idx;
1172         pvma.vm_ops = NULL;
1173         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1174
1175         /*
1176          * alloc_page_vma() will drop the shared policy reference
1177          */
1178         return alloc_page_vma(gfp, &pvma, 0);
1179 }
1180 #else /* !CONFIG_NUMA */
1181 #ifdef CONFIG_TMPFS
1182 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p)
1183 {
1184 }
1185 #endif /* CONFIG_TMPFS */
1186
1187 static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1188                         struct shmem_inode_info *info, unsigned long idx)
1189 {
1190         return swapin_readahead(entry, gfp, NULL, 0);
1191 }
1192
1193 static inline struct page *shmem_alloc_page(gfp_t gfp,
1194                         struct shmem_inode_info *info, unsigned long idx)
1195 {
1196         return alloc_page(gfp);
1197 }
1198 #endif /* CONFIG_NUMA */
1199
1200 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1201 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1202 {
1203         return NULL;
1204 }
1205 #endif
1206
1207 /*
1208  * shmem_getpage - either get the page from swap or allocate a new one
1209  *
1210  * If we allocate a new one we do not mark it dirty. That's up to the
1211  * vm. If we swap it in we mark it dirty since we also free the swap
1212  * entry since a page cannot live in both the swap and page cache
1213  */
1214 static int shmem_getpage(struct inode *inode, unsigned long idx,
1215                         struct page **pagep, enum sgp_type sgp, int *type)
1216 {
1217         struct address_space *mapping = inode->i_mapping;
1218         struct shmem_inode_info *info = SHMEM_I(inode);
1219         struct shmem_sb_info *sbinfo;
1220         struct page *filepage = *pagep;
1221         struct page *swappage;
1222         swp_entry_t *entry;
1223         swp_entry_t swap;
1224         gfp_t gfp;
1225         int error;
1226
1227         if (idx >= SHMEM_MAX_INDEX)
1228                 return -EFBIG;
1229
1230         if (type)
1231                 *type = 0;
1232
1233         /*
1234          * Normally, filepage is NULL on entry, and either found
1235          * uptodate immediately, or allocated and zeroed, or read
1236          * in under swappage, which is then assigned to filepage.
1237          * But shmem_readpage (required for splice) passes in a locked
1238          * filepage, which may be found not uptodate by other callers
1239          * too, and may need to be copied from the swappage read in.
1240          */
1241 repeat:
1242         if (!filepage)
1243                 filepage = find_lock_page(mapping, idx);
1244         if (filepage && PageUptodate(filepage))
1245                 goto done;
1246         error = 0;
1247         gfp = mapping_gfp_mask(mapping);
1248         if (!filepage) {
1249                 /*
1250                  * Try to preload while we can wait, to not make a habit of
1251                  * draining atomic reserves; but don't latch on to this cpu.
1252                  */
1253                 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM);
1254                 if (error)
1255                         goto failed;
1256                 radix_tree_preload_end();
1257         }
1258
1259         spin_lock(&info->lock);
1260         shmem_recalc_inode(inode);
1261         entry = shmem_swp_alloc(info, idx, sgp);
1262         if (IS_ERR(entry)) {
1263                 spin_unlock(&info->lock);
1264                 error = PTR_ERR(entry);
1265                 goto failed;
1266         }
1267         swap = *entry;
1268
1269         if (swap.val) {
1270                 /* Look it up and read it in.. */
1271                 swappage = lookup_swap_cache(swap);
1272                 if (!swappage) {
1273                         shmem_swp_unmap(entry);
1274                         /* here we actually do the io */
1275                         if (type && !(*type & VM_FAULT_MAJOR)) {
1276                                 __count_vm_event(PGMAJFAULT);
1277                                 *type |= VM_FAULT_MAJOR;
1278                         }
1279                         spin_unlock(&info->lock);
1280                         swappage = shmem_swapin(swap, gfp, info, idx);
1281                         if (!swappage) {
1282                                 spin_lock(&info->lock);
1283                                 entry = shmem_swp_alloc(info, idx, sgp);
1284                                 if (IS_ERR(entry))
1285                                         error = PTR_ERR(entry);
1286                                 else {
1287                                         if (entry->val == swap.val)
1288                                                 error = -ENOMEM;
1289                                         shmem_swp_unmap(entry);
1290                                 }
1291                                 spin_unlock(&info->lock);
1292                                 if (error)
1293                                         goto failed;
1294                                 goto repeat;
1295                         }
1296                         wait_on_page_locked(swappage);
1297                         page_cache_release(swappage);
1298                         goto repeat;
1299                 }
1300
1301                 /* We have to do this with page locked to prevent races */
1302                 if (!trylock_page(swappage)) {
1303                         shmem_swp_unmap(entry);
1304                         spin_unlock(&info->lock);
1305                         wait_on_page_locked(swappage);
1306                         page_cache_release(swappage);
1307                         goto repeat;
1308                 }
1309                 if (PageWriteback(swappage)) {
1310                         shmem_swp_unmap(entry);
1311                         spin_unlock(&info->lock);
1312                         wait_on_page_writeback(swappage);
1313                         unlock_page(swappage);
1314                         page_cache_release(swappage);
1315                         goto repeat;
1316                 }
1317                 if (!PageUptodate(swappage)) {
1318                         shmem_swp_unmap(entry);
1319                         spin_unlock(&info->lock);
1320                         unlock_page(swappage);
1321                         page_cache_release(swappage);
1322                         error = -EIO;
1323                         goto failed;
1324                 }
1325
1326                 if (filepage) {
1327                         shmem_swp_set(info, entry, 0);
1328                         shmem_swp_unmap(entry);
1329                         delete_from_swap_cache(swappage);
1330                         spin_unlock(&info->lock);
1331                         copy_highpage(filepage, swappage);
1332                         unlock_page(swappage);
1333                         page_cache_release(swappage);
1334                         flush_dcache_page(filepage);
1335                         SetPageUptodate(filepage);
1336                         set_page_dirty(filepage);
1337                         swap_free(swap);
1338                 } else if (!(error = add_to_page_cache_locked(swappage, mapping,
1339                                         idx, GFP_NOWAIT))) {
1340                         info->flags |= SHMEM_PAGEIN;
1341                         shmem_swp_set(info, entry, 0);
1342                         shmem_swp_unmap(entry);
1343                         delete_from_swap_cache(swappage);
1344                         spin_unlock(&info->lock);
1345                         filepage = swappage;
1346                         set_page_dirty(filepage);
1347                         swap_free(swap);
1348                 } else {
1349                         shmem_swp_unmap(entry);
1350                         spin_unlock(&info->lock);
1351                         if (error == -ENOMEM) {
1352                                 /*
1353                                  * reclaim from proper memory cgroup and
1354                                  * call memcg's OOM if needed.
1355                                  */
1356                                 error = mem_cgroup_shmem_charge_fallback(
1357                                                                 swappage,
1358                                                                 current->mm,
1359                                                                 gfp);
1360                                 if (error) {
1361                                         unlock_page(swappage);
1362                                         page_cache_release(swappage);
1363                                         goto failed;
1364                                 }
1365                         }
1366                         unlock_page(swappage);
1367                         page_cache_release(swappage);
1368                         goto repeat;
1369                 }
1370         } else if (sgp == SGP_READ && !filepage) {
1371                 shmem_swp_unmap(entry);
1372                 filepage = find_get_page(mapping, idx);
1373                 if (filepage &&
1374                     (!PageUptodate(filepage) || !trylock_page(filepage))) {
1375                         spin_unlock(&info->lock);
1376                         wait_on_page_locked(filepage);
1377                         page_cache_release(filepage);
1378                         filepage = NULL;
1379                         goto repeat;
1380                 }
1381                 spin_unlock(&info->lock);
1382         } else {
1383                 shmem_swp_unmap(entry);
1384                 sbinfo = SHMEM_SB(inode->i_sb);
1385                 if (sbinfo->max_blocks) {
1386                         spin_lock(&sbinfo->stat_lock);
1387                         if (sbinfo->free_blocks == 0 ||
1388                             shmem_acct_block(info->flags)) {
1389                                 spin_unlock(&sbinfo->stat_lock);
1390                                 spin_unlock(&info->lock);
1391                                 error = -ENOSPC;
1392                                 goto failed;
1393                         }
1394                         sbinfo->free_blocks--;
1395                         inode->i_blocks += BLOCKS_PER_PAGE;
1396                         spin_unlock(&sbinfo->stat_lock);
1397                 } else if (shmem_acct_block(info->flags)) {
1398                         spin_unlock(&info->lock);
1399                         error = -ENOSPC;
1400                         goto failed;
1401                 }
1402
1403                 if (!filepage) {
1404                         int ret;
1405
1406                         spin_unlock(&info->lock);
1407                         filepage = shmem_alloc_page(gfp, info, idx);
1408                         if (!filepage) {
1409                                 shmem_unacct_blocks(info->flags, 1);
1410                                 shmem_free_blocks(inode, 1);
1411                                 error = -ENOMEM;
1412                                 goto failed;
1413                         }
1414                         SetPageSwapBacked(filepage);
1415
1416                         /* Precharge page while we can wait, compensate after */
1417                         error = mem_cgroup_cache_charge(filepage, current->mm,
1418                                         GFP_KERNEL);
1419                         if (error) {
1420                                 page_cache_release(filepage);
1421                                 shmem_unacct_blocks(info->flags, 1);
1422                                 shmem_free_blocks(inode, 1);
1423                                 filepage = NULL;
1424                                 goto failed;
1425                         }
1426
1427                         spin_lock(&info->lock);
1428                         entry = shmem_swp_alloc(info, idx, sgp);
1429                         if (IS_ERR(entry))
1430                                 error = PTR_ERR(entry);
1431                         else {
1432                                 swap = *entry;
1433                                 shmem_swp_unmap(entry);
1434                         }
1435                         ret = error || swap.val;
1436                         if (ret)
1437                                 mem_cgroup_uncharge_cache_page(filepage);
1438                         else
1439                                 ret = add_to_page_cache_lru(filepage, mapping,
1440                                                 idx, GFP_NOWAIT);
1441                         /*
1442                          * At add_to_page_cache_lru() failure, uncharge will
1443                          * be done automatically.
1444                          */
1445                         if (ret) {
1446                                 spin_unlock(&info->lock);
1447                                 page_cache_release(filepage);
1448                                 shmem_unacct_blocks(info->flags, 1);
1449                                 shmem_free_blocks(inode, 1);
1450                                 filepage = NULL;
1451                                 if (error)
1452                                         goto failed;
1453                                 goto repeat;
1454                         }
1455                         info->flags |= SHMEM_PAGEIN;
1456                 }
1457
1458                 info->alloced++;
1459                 spin_unlock(&info->lock);
1460                 clear_highpage(filepage);
1461                 flush_dcache_page(filepage);
1462                 SetPageUptodate(filepage);
1463                 if (sgp == SGP_DIRTY)
1464                         set_page_dirty(filepage);
1465         }
1466 done:
1467         *pagep = filepage;
1468         return 0;
1469
1470 failed:
1471         if (*pagep != filepage) {
1472                 unlock_page(filepage);
1473                 page_cache_release(filepage);
1474         }
1475         return error;
1476 }
1477
1478 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1479 {
1480         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1481         int error;
1482         int ret;
1483
1484         if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1485                 return VM_FAULT_SIGBUS;
1486
1487         error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1488         if (error)
1489                 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1490
1491         return ret | VM_FAULT_LOCKED;
1492 }
1493
1494 #ifdef CONFIG_NUMA
1495 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1496 {
1497         struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1498         return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1499 }
1500
1501 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1502                                           unsigned long addr)
1503 {
1504         struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1505         unsigned long idx;
1506
1507         idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1508         return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1509 }
1510 #endif
1511
1512 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1513 {
1514         struct inode *inode = file->f_path.dentry->d_inode;
1515         struct shmem_inode_info *info = SHMEM_I(inode);
1516         int retval = -ENOMEM;
1517
1518         spin_lock(&info->lock);
1519         if (lock && !(info->flags & VM_LOCKED)) {
1520                 if (!user_shm_lock(inode->i_size, user))
1521                         goto out_nomem;
1522                 info->flags |= VM_LOCKED;
1523                 mapping_set_unevictable(file->f_mapping);
1524         }
1525         if (!lock && (info->flags & VM_LOCKED) && user) {
1526                 user_shm_unlock(inode->i_size, user);
1527                 info->flags &= ~VM_LOCKED;
1528                 mapping_clear_unevictable(file->f_mapping);
1529                 scan_mapping_unevictable_pages(file->f_mapping);
1530         }
1531         retval = 0;
1532
1533 out_nomem:
1534         spin_unlock(&info->lock);
1535         return retval;
1536 }
1537
1538 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1539 {
1540         file_accessed(file);
1541         vma->vm_ops = &shmem_vm_ops;
1542         vma->vm_flags |= VM_CAN_NONLINEAR;
1543         return 0;
1544 }
1545
1546 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1547                                      int mode, dev_t dev, unsigned long flags)
1548 {
1549         struct inode *inode;
1550         struct shmem_inode_info *info;
1551         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1552
1553         if (shmem_reserve_inode(sb))
1554                 return NULL;
1555
1556         inode = new_inode(sb);
1557         if (inode) {
1558                 inode_init_owner(inode, dir, mode);
1559                 inode->i_blocks = 0;
1560                 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1561                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1562                 inode->i_generation = get_seconds();
1563                 info = SHMEM_I(inode);
1564                 memset(info, 0, (char *)inode - (char *)info);
1565                 spin_lock_init(&info->lock);
1566                 info->flags = flags & VM_NORESERVE;
1567                 INIT_LIST_HEAD(&info->swaplist);
1568                 cache_no_acl(inode);
1569
1570                 switch (mode & S_IFMT) {
1571                 default:
1572                         inode->i_op = &shmem_special_inode_operations;
1573                         init_special_inode(inode, mode, dev);
1574                         break;
1575                 case S_IFREG:
1576                         inode->i_mapping->a_ops = &shmem_aops;
1577                         inode->i_op = &shmem_inode_operations;
1578                         inode->i_fop = &shmem_file_operations;
1579                         mpol_shared_policy_init(&info->policy,
1580                                                  shmem_get_sbmpol(sbinfo));
1581                         break;
1582                 case S_IFDIR:
1583                         inc_nlink(inode);
1584                         /* Some things misbehave if size == 0 on a directory */
1585                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
1586                         inode->i_op = &shmem_dir_inode_operations;
1587                         inode->i_fop = &simple_dir_operations;
1588                         break;
1589                 case S_IFLNK:
1590                         /*
1591                          * Must not load anything in the rbtree,
1592                          * mpol_free_shared_policy will not be called.
1593                          */
1594                         mpol_shared_policy_init(&info->policy, NULL);
1595                         break;
1596                 }
1597         } else
1598                 shmem_free_inode(sb);
1599         return inode;
1600 }
1601
1602 #ifdef CONFIG_TMPFS
1603 static const struct inode_operations shmem_symlink_inode_operations;
1604 static const struct inode_operations shmem_symlink_inline_operations;
1605
1606 /*
1607  * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
1608  * but providing them allows a tmpfs file to be used for splice, sendfile, and
1609  * below the loop driver, in the generic fashion that many filesystems support.
1610  */
1611 static int shmem_readpage(struct file *file, struct page *page)
1612 {
1613         struct inode *inode = page->mapping->host;
1614         int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1615         unlock_page(page);
1616         return error;
1617 }
1618
1619 static int
1620 shmem_write_begin(struct file *file, struct address_space *mapping,
1621                         loff_t pos, unsigned len, unsigned flags,
1622                         struct page **pagep, void **fsdata)
1623 {
1624         struct inode *inode = mapping->host;
1625         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1626         *pagep = NULL;
1627         return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1628 }
1629
1630 static int
1631 shmem_write_end(struct file *file, struct address_space *mapping,
1632                         loff_t pos, unsigned len, unsigned copied,
1633                         struct page *page, void *fsdata)
1634 {
1635         struct inode *inode = mapping->host;
1636
1637         if (pos + copied > inode->i_size)
1638                 i_size_write(inode, pos + copied);
1639
1640         set_page_dirty(page);
1641         unlock_page(page);
1642         page_cache_release(page);
1643
1644         return copied;
1645 }
1646
1647 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1648 {
1649         struct inode *inode = filp->f_path.dentry->d_inode;
1650         struct address_space *mapping = inode->i_mapping;
1651         unsigned long index, offset;
1652         enum sgp_type sgp = SGP_READ;
1653
1654         /*
1655          * Might this read be for a stacking filesystem?  Then when reading
1656          * holes of a sparse file, we actually need to allocate those pages,
1657          * and even mark them dirty, so it cannot exceed the max_blocks limit.
1658          */
1659         if (segment_eq(get_fs(), KERNEL_DS))
1660                 sgp = SGP_DIRTY;
1661
1662         index = *ppos >> PAGE_CACHE_SHIFT;
1663         offset = *ppos & ~PAGE_CACHE_MASK;
1664
1665         for (;;) {
1666                 struct page *page = NULL;
1667                 unsigned long end_index, nr, ret;
1668                 loff_t i_size = i_size_read(inode);
1669
1670                 end_index = i_size >> PAGE_CACHE_SHIFT;
1671                 if (index > end_index)
1672                         break;
1673                 if (index == end_index) {
1674                         nr = i_size & ~PAGE_CACHE_MASK;
1675                         if (nr <= offset)
1676                                 break;
1677                 }
1678
1679                 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1680                 if (desc->error) {
1681                         if (desc->error == -EINVAL)
1682                                 desc->error = 0;
1683                         break;
1684                 }
1685                 if (page)
1686                         unlock_page(page);
1687
1688                 /*
1689                  * We must evaluate after, since reads (unlike writes)
1690                  * are called without i_mutex protection against truncate
1691                  */
1692                 nr = PAGE_CACHE_SIZE;
1693                 i_size = i_size_read(inode);
1694                 end_index = i_size >> PAGE_CACHE_SHIFT;
1695                 if (index == end_index) {
1696                         nr = i_size & ~PAGE_CACHE_MASK;
1697                         if (nr <= offset) {
1698                                 if (page)
1699                                         page_cache_release(page);
1700                                 break;
1701                         }
1702                 }
1703                 nr -= offset;
1704
1705                 if (page) {
1706                         /*
1707                          * If users can be writing to this page using arbitrary
1708                          * virtual addresses, take care about potential aliasing
1709                          * before reading the page on the kernel side.
1710                          */
1711                         if (mapping_writably_mapped(mapping))
1712                                 flush_dcache_page(page);
1713                         /*
1714                          * Mark the page accessed if we read the beginning.
1715                          */
1716                         if (!offset)
1717                                 mark_page_accessed(page);
1718                 } else {
1719                         page = ZERO_PAGE(0);
1720                         page_cache_get(page);
1721                 }
1722
1723                 /*
1724                  * Ok, we have the page, and it's up-to-date, so
1725                  * now we can copy it to user space...
1726                  *
1727                  * The actor routine returns how many bytes were actually used..
1728                  * NOTE! This may not be the same as how much of a user buffer
1729                  * we filled up (we may be padding etc), so we can only update
1730                  * "pos" here (the actor routine has to update the user buffer
1731                  * pointers and the remaining count).
1732                  */
1733                 ret = actor(desc, page, offset, nr);
1734                 offset += ret;
1735                 index += offset >> PAGE_CACHE_SHIFT;
1736                 offset &= ~PAGE_CACHE_MASK;
1737
1738                 page_cache_release(page);
1739                 if (ret != nr || !desc->count)
1740                         break;
1741
1742                 cond_resched();
1743         }
1744
1745         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1746         file_accessed(filp);
1747 }
1748
1749 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1750                 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1751 {
1752         struct file *filp = iocb->ki_filp;
1753         ssize_t retval;
1754         unsigned long seg;
1755         size_t count;
1756         loff_t *ppos = &iocb->ki_pos;
1757
1758         retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1759         if (retval)
1760                 return retval;
1761
1762         for (seg = 0; seg < nr_segs; seg++) {
1763                 read_descriptor_t desc;
1764
1765                 desc.written = 0;
1766                 desc.arg.buf = iov[seg].iov_base;
1767                 desc.count = iov[seg].iov_len;
1768                 if (desc.count == 0)
1769                         continue;
1770                 desc.error = 0;
1771                 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1772                 retval += desc.written;
1773                 if (desc.error) {
1774                         retval = retval ?: desc.error;
1775                         break;
1776                 }
1777                 if (desc.count > 0)
1778                         break;
1779         }
1780         return retval;
1781 }
1782
1783 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1784 {
1785         struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1786
1787         buf->f_type = TMPFS_MAGIC;
1788         buf->f_bsize = PAGE_CACHE_SIZE;
1789         buf->f_namelen = NAME_MAX;
1790         spin_lock(&sbinfo->stat_lock);
1791         if (sbinfo->max_blocks) {
1792                 buf->f_blocks = sbinfo->max_blocks;
1793                 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1794         }
1795         if (sbinfo->max_inodes) {
1796                 buf->f_files = sbinfo->max_inodes;
1797                 buf->f_ffree = sbinfo->free_inodes;
1798         }
1799         /* else leave those fields 0 like simple_statfs */
1800         spin_unlock(&sbinfo->stat_lock);
1801         return 0;
1802 }
1803
1804 /*
1805  * File creation. Allocate an inode, and we're done..
1806  */
1807 static int
1808 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1809 {
1810         struct inode *inode;
1811         int error = -ENOSPC;
1812
1813         inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1814         if (inode) {
1815                 error = security_inode_init_security(inode, dir, NULL, NULL,
1816                                                      NULL);
1817                 if (error) {
1818                         if (error != -EOPNOTSUPP) {
1819                                 iput(inode);
1820                                 return error;
1821                         }
1822                 }
1823 #ifdef CONFIG_TMPFS_POSIX_ACL
1824                 error = generic_acl_init(inode, dir);
1825                 if (error) {
1826                         iput(inode);
1827                         return error;
1828                 }
1829 #else
1830                 error = 0;
1831 #endif
1832                 dir->i_size += BOGO_DIRENT_SIZE;
1833                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1834                 d_instantiate(dentry, inode);
1835                 dget(dentry); /* Extra count - pin the dentry in core */
1836         }
1837         return error;
1838 }
1839
1840 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1841 {
1842         int error;
1843
1844         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1845                 return error;
1846         inc_nlink(dir);
1847         return 0;
1848 }
1849
1850 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1851                 struct nameidata *nd)
1852 {
1853         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1854 }
1855
1856 /*
1857  * Link a file..
1858  */
1859 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1860 {
1861         struct inode *inode = old_dentry->d_inode;
1862         int ret;
1863
1864         /*
1865          * No ordinary (disk based) filesystem counts links as inodes;
1866          * but each new link needs a new dentry, pinning lowmem, and
1867          * tmpfs dentries cannot be pruned until they are unlinked.
1868          */
1869         ret = shmem_reserve_inode(inode->i_sb);
1870         if (ret)
1871                 goto out;
1872
1873         dir->i_size += BOGO_DIRENT_SIZE;
1874         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1875         inc_nlink(inode);
1876         atomic_inc(&inode->i_count);    /* New dentry reference */
1877         dget(dentry);           /* Extra pinning count for the created dentry */
1878         d_instantiate(dentry, inode);
1879 out:
1880         return ret;
1881 }
1882
1883 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1884 {
1885         struct inode *inode = dentry->d_inode;
1886
1887         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1888                 shmem_free_inode(inode->i_sb);
1889
1890         dir->i_size -= BOGO_DIRENT_SIZE;
1891         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1892         drop_nlink(inode);
1893         dput(dentry);   /* Undo the count from "create" - this does all the work */
1894         return 0;
1895 }
1896
1897 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1898 {
1899         if (!simple_empty(dentry))
1900                 return -ENOTEMPTY;
1901
1902         drop_nlink(dentry->d_inode);
1903         drop_nlink(dir);
1904         return shmem_unlink(dir, dentry);
1905 }
1906
1907 /*
1908  * The VFS layer already does all the dentry stuff for rename,
1909  * we just have to decrement the usage count for the target if
1910  * it exists so that the VFS layer correctly free's it when it
1911  * gets overwritten.
1912  */
1913 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1914 {
1915         struct inode *inode = old_dentry->d_inode;
1916         int they_are_dirs = S_ISDIR(inode->i_mode);
1917
1918         if (!simple_empty(new_dentry))
1919                 return -ENOTEMPTY;
1920
1921         if (new_dentry->d_inode) {
1922                 (void) shmem_unlink(new_dir, new_dentry);
1923                 if (they_are_dirs)
1924                         drop_nlink(old_dir);
1925         } else if (they_are_dirs) {
1926                 drop_nlink(old_dir);
1927                 inc_nlink(new_dir);
1928         }
1929
1930         old_dir->i_size -= BOGO_DIRENT_SIZE;
1931         new_dir->i_size += BOGO_DIRENT_SIZE;
1932         old_dir->i_ctime = old_dir->i_mtime =
1933         new_dir->i_ctime = new_dir->i_mtime =
1934         inode->i_ctime = CURRENT_TIME;
1935         return 0;
1936 }
1937
1938 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1939 {
1940         int error;
1941         int len;
1942         struct inode *inode;
1943         struct page *page = NULL;
1944         char *kaddr;
1945         struct shmem_inode_info *info;
1946
1947         len = strlen(symname) + 1;
1948         if (len > PAGE_CACHE_SIZE)
1949                 return -ENAMETOOLONG;
1950
1951         inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
1952         if (!inode)
1953                 return -ENOSPC;
1954
1955         error = security_inode_init_security(inode, dir, NULL, NULL,
1956                                              NULL);
1957         if (error) {
1958                 if (error != -EOPNOTSUPP) {
1959                         iput(inode);
1960                         return error;
1961                 }
1962                 error = 0;
1963         }
1964
1965         info = SHMEM_I(inode);
1966         inode->i_size = len-1;
1967         if (len <= (char *)inode - (char *)info) {
1968                 /* do it inline */
1969                 memcpy(info, symname, len);
1970                 inode->i_op = &shmem_symlink_inline_operations;
1971         } else {
1972                 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1973                 if (error) {
1974                         iput(inode);
1975                         return error;
1976                 }
1977                 inode->i_mapping->a_ops = &shmem_aops;
1978                 inode->i_op = &shmem_symlink_inode_operations;
1979                 kaddr = kmap_atomic(page, KM_USER0);
1980                 memcpy(kaddr, symname, len);
1981                 kunmap_atomic(kaddr, KM_USER0);
1982                 set_page_dirty(page);
1983                 unlock_page(page);
1984                 page_cache_release(page);
1985         }
1986         dir->i_size += BOGO_DIRENT_SIZE;
1987         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1988         d_instantiate(dentry, inode);
1989         dget(dentry);
1990         return 0;
1991 }
1992
1993 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1994 {
1995         nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1996         return NULL;
1997 }
1998
1999 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
2000 {
2001         struct page *page = NULL;
2002         int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2003         nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
2004         if (page)
2005                 unlock_page(page);
2006         return page;
2007 }
2008
2009 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2010 {
2011         if (!IS_ERR(nd_get_link(nd))) {
2012                 struct page *page = cookie;
2013                 kunmap(page);
2014                 mark_page_accessed(page);
2015                 page_cache_release(page);
2016         }
2017 }
2018
2019 static const struct inode_operations shmem_symlink_inline_operations = {
2020         .readlink       = generic_readlink,
2021         .follow_link    = shmem_follow_link_inline,
2022 };
2023
2024 static const struct inode_operations shmem_symlink_inode_operations = {
2025         .truncate       = shmem_truncate,
2026         .readlink       = generic_readlink,
2027         .follow_link    = shmem_follow_link,
2028         .put_link       = shmem_put_link,
2029 };
2030
2031 #ifdef CONFIG_TMPFS_POSIX_ACL
2032 /*
2033  * Superblocks without xattr inode operations will get security.* xattr
2034  * support from the VFS "for free". As soon as we have any other xattrs
2035  * like ACLs, we also need to implement the security.* handlers at
2036  * filesystem level, though.
2037  */
2038
2039 static size_t shmem_xattr_security_list(struct dentry *dentry, char *list,
2040                                         size_t list_len, const char *name,
2041                                         size_t name_len, int handler_flags)
2042 {
2043         return security_inode_listsecurity(dentry->d_inode, list, list_len);
2044 }
2045
2046 static int shmem_xattr_security_get(struct dentry *dentry, const char *name,
2047                 void *buffer, size_t size, int handler_flags)
2048 {
2049         if (strcmp(name, "") == 0)
2050                 return -EINVAL;
2051         return xattr_getsecurity(dentry->d_inode, name, buffer, size);
2052 }
2053
2054 static int shmem_xattr_security_set(struct dentry *dentry, const char *name,
2055                 const void *value, size_t size, int flags, int handler_flags)
2056 {
2057         if (strcmp(name, "") == 0)
2058                 return -EINVAL;
2059         return security_inode_setsecurity(dentry->d_inode, name, value,
2060                                           size, flags);
2061 }
2062
2063 static const struct xattr_handler shmem_xattr_security_handler = {
2064         .prefix = XATTR_SECURITY_PREFIX,
2065         .list   = shmem_xattr_security_list,
2066         .get    = shmem_xattr_security_get,
2067         .set    = shmem_xattr_security_set,
2068 };
2069
2070 static const struct xattr_handler *shmem_xattr_handlers[] = {
2071         &generic_acl_access_handler,
2072         &generic_acl_default_handler,
2073         &shmem_xattr_security_handler,
2074         NULL
2075 };
2076 #endif
2077
2078 static struct dentry *shmem_get_parent(struct dentry *child)
2079 {
2080         return ERR_PTR(-ESTALE);
2081 }
2082
2083 static int shmem_match(struct inode *ino, void *vfh)
2084 {
2085         __u32 *fh = vfh;
2086         __u64 inum = fh[2];
2087         inum = (inum << 32) | fh[1];
2088         return ino->i_ino == inum && fh[0] == ino->i_generation;
2089 }
2090
2091 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2092                 struct fid *fid, int fh_len, int fh_type)
2093 {
2094         struct inode *inode;
2095         struct dentry *dentry = NULL;
2096         u64 inum = fid->raw[2];
2097         inum = (inum << 32) | fid->raw[1];
2098
2099         if (fh_len < 3)
2100                 return NULL;
2101
2102         inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2103                         shmem_match, fid->raw);
2104         if (inode) {
2105                 dentry = d_find_alias(inode);
2106                 iput(inode);
2107         }
2108
2109         return dentry;
2110 }
2111
2112 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2113                                 int connectable)
2114 {
2115         struct inode *inode = dentry->d_inode;
2116
2117         if (*len < 3)
2118                 return 255;
2119
2120         if (hlist_unhashed(&inode->i_hash)) {
2121                 /* Unfortunately insert_inode_hash is not idempotent,
2122                  * so as we hash inodes here rather than at creation
2123                  * time, we need a lock to ensure we only try
2124                  * to do it once
2125                  */
2126                 static DEFINE_SPINLOCK(lock);
2127                 spin_lock(&lock);
2128                 if (hlist_unhashed(&inode->i_hash))
2129                         __insert_inode_hash(inode,
2130                                             inode->i_ino + inode->i_generation);
2131                 spin_unlock(&lock);
2132         }
2133
2134         fh[0] = inode->i_generation;
2135         fh[1] = inode->i_ino;
2136         fh[2] = ((__u64)inode->i_ino) >> 32;
2137
2138         *len = 3;
2139         return 1;
2140 }
2141
2142 static const struct export_operations shmem_export_ops = {
2143         .get_parent     = shmem_get_parent,
2144         .encode_fh      = shmem_encode_fh,
2145         .fh_to_dentry   = shmem_fh_to_dentry,
2146 };
2147
2148 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2149                                bool remount)
2150 {
2151         char *this_char, *value, *rest;
2152
2153         while (options != NULL) {
2154                 this_char = options;
2155                 for (;;) {
2156                         /*
2157                          * NUL-terminate this option: unfortunately,
2158                          * mount options form a comma-separated list,
2159                          * but mpol's nodelist may also contain commas.
2160                          */
2161                         options = strchr(options, ',');
2162                         if (options == NULL)
2163                                 break;
2164                         options++;
2165                         if (!isdigit(*options)) {
2166                                 options[-1] = '\0';
2167                                 break;
2168                         }
2169                 }
2170                 if (!*this_char)
2171                         continue;
2172                 if ((value = strchr(this_char,'=')) != NULL) {
2173                         *value++ = 0;
2174                 } else {
2175                         printk(KERN_ERR
2176                             "tmpfs: No value for mount option '%s'\n",
2177                             this_char);
2178                         return 1;
2179                 }
2180
2181                 if (!strcmp(this_char,"size")) {
2182                         unsigned long long size;
2183                         size = memparse(value,&rest);
2184                         if (*rest == '%') {
2185                                 size <<= PAGE_SHIFT;
2186                                 size *= totalram_pages;
2187                                 do_div(size, 100);
2188                                 rest++;
2189                         }
2190                         if (*rest)
2191                                 goto bad_val;
2192                         sbinfo->max_blocks =
2193                                 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2194                 } else if (!strcmp(this_char,"nr_blocks")) {
2195                         sbinfo->max_blocks = memparse(value, &rest);
2196                         if (*rest)
2197                                 goto bad_val;
2198                 } else if (!strcmp(this_char,"nr_inodes")) {
2199                         sbinfo->max_inodes = memparse(value, &rest);
2200                         if (*rest)
2201                                 goto bad_val;
2202                 } else if (!strcmp(this_char,"mode")) {
2203                         if (remount)
2204                                 continue;
2205                         sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2206                         if (*rest)
2207                                 goto bad_val;
2208                 } else if (!strcmp(this_char,"uid")) {
2209                         if (remount)
2210                                 continue;
2211                         sbinfo->uid = simple_strtoul(value, &rest, 0);
2212                         if (*rest)
2213                                 goto bad_val;
2214                 } else if (!strcmp(this_char,"gid")) {
2215                         if (remount)
2216                                 continue;
2217                         sbinfo->gid = simple_strtoul(value, &rest, 0);
2218                         if (*rest)
2219                                 goto bad_val;
2220                 } else if (!strcmp(this_char,"mpol")) {
2221                         if (mpol_parse_str(value, &sbinfo->mpol, 1))
2222                                 goto bad_val;
2223                 } else {
2224                         printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2225                                this_char);
2226                         return 1;
2227                 }
2228         }
2229         return 0;
2230
2231 bad_val:
2232         printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2233                value, this_char);
2234         return 1;
2235
2236 }
2237
2238 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2239 {
2240         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2241         struct shmem_sb_info config = *sbinfo;
2242         unsigned long blocks;
2243         unsigned long inodes;
2244         int error = -EINVAL;
2245
2246         if (shmem_parse_options(data, &config, true))
2247                 return error;
2248
2249         spin_lock(&sbinfo->stat_lock);
2250         blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2251         inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2252         if (config.max_blocks < blocks)
2253                 goto out;
2254         if (config.max_inodes < inodes)
2255                 goto out;
2256         /*
2257          * Those tests also disallow limited->unlimited while any are in
2258          * use, so i_blocks will always be zero when max_blocks is zero;
2259          * but we must separately disallow unlimited->limited, because
2260          * in that case we have no record of how much is already in use.
2261          */
2262         if (config.max_blocks && !sbinfo->max_blocks)
2263                 goto out;
2264         if (config.max_inodes && !sbinfo->max_inodes)
2265                 goto out;
2266
2267         error = 0;
2268         sbinfo->max_blocks  = config.max_blocks;
2269         sbinfo->free_blocks = config.max_blocks - blocks;
2270         sbinfo->max_inodes  = config.max_inodes;
2271         sbinfo->free_inodes = config.max_inodes - inodes;
2272
2273         mpol_put(sbinfo->mpol);
2274         sbinfo->mpol        = config.mpol;      /* transfers initial ref */
2275 out:
2276         spin_unlock(&sbinfo->stat_lock);
2277         return error;
2278 }
2279
2280 static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2281 {
2282         struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2283
2284         if (sbinfo->max_blocks != shmem_default_max_blocks())
2285                 seq_printf(seq, ",size=%luk",
2286                         sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2287         if (sbinfo->max_inodes != shmem_default_max_inodes())
2288                 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2289         if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2290                 seq_printf(seq, ",mode=%03o", sbinfo->mode);
2291         if (sbinfo->uid != 0)
2292                 seq_printf(seq, ",uid=%u", sbinfo->uid);
2293         if (sbinfo->gid != 0)
2294                 seq_printf(seq, ",gid=%u", sbinfo->gid);
2295         shmem_show_mpol(seq, sbinfo->mpol);
2296         return 0;
2297 }
2298 #endif /* CONFIG_TMPFS */
2299
2300 static void shmem_put_super(struct super_block *sb)
2301 {
2302         kfree(sb->s_fs_info);
2303         sb->s_fs_info = NULL;
2304 }
2305
2306 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2307 {
2308         struct inode *inode;
2309         struct dentry *root;
2310         struct shmem_sb_info *sbinfo;
2311         int err = -ENOMEM;
2312
2313         /* Round up to L1_CACHE_BYTES to resist false sharing */
2314         sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2315                                 L1_CACHE_BYTES), GFP_KERNEL);
2316         if (!sbinfo)
2317                 return -ENOMEM;
2318
2319         sbinfo->mode = S_IRWXUGO | S_ISVTX;
2320         sbinfo->uid = current_fsuid();
2321         sbinfo->gid = current_fsgid();
2322         sb->s_fs_info = sbinfo;
2323
2324 #ifdef CONFIG_TMPFS
2325         /*
2326          * Per default we only allow half of the physical ram per
2327          * tmpfs instance, limiting inodes to one per page of lowmem;
2328          * but the internal instance is left unlimited.
2329          */
2330         if (!(sb->s_flags & MS_NOUSER)) {
2331                 sbinfo->max_blocks = shmem_default_max_blocks();
2332                 sbinfo->max_inodes = shmem_default_max_inodes();
2333                 if (shmem_parse_options(data, sbinfo, false)) {
2334                         err = -EINVAL;
2335                         goto failed;
2336                 }
2337         }
2338         sb->s_export_op = &shmem_export_ops;
2339 #else
2340         sb->s_flags |= MS_NOUSER;
2341 #endif
2342
2343         spin_lock_init(&sbinfo->stat_lock);
2344         sbinfo->free_blocks = sbinfo->max_blocks;
2345         sbinfo->free_inodes = sbinfo->max_inodes;
2346
2347         sb->s_maxbytes = SHMEM_MAX_BYTES;
2348         sb->s_blocksize = PAGE_CACHE_SIZE;
2349         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2350         sb->s_magic = TMPFS_MAGIC;
2351         sb->s_op = &shmem_ops;
2352         sb->s_time_gran = 1;
2353 #ifdef CONFIG_TMPFS_POSIX_ACL
2354         sb->s_xattr = shmem_xattr_handlers;
2355         sb->s_flags |= MS_POSIXACL;
2356 #endif
2357
2358         inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2359         if (!inode)
2360                 goto failed;
2361         inode->i_uid = sbinfo->uid;
2362         inode->i_gid = sbinfo->gid;
2363         root = d_alloc_root(inode);
2364         if (!root)
2365                 goto failed_iput;
2366         sb->s_root = root;
2367         return 0;
2368
2369 failed_iput:
2370         iput(inode);
2371 failed:
2372         shmem_put_super(sb);
2373         return err;
2374 }
2375
2376 static struct kmem_cache *shmem_inode_cachep;
2377
2378 static struct inode *shmem_alloc_inode(struct super_block *sb)
2379 {
2380         struct shmem_inode_info *p;
2381         p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2382         if (!p)
2383                 return NULL;
2384         return &p->vfs_inode;
2385 }
2386
2387 static void shmem_destroy_inode(struct inode *inode)
2388 {
2389         if ((inode->i_mode & S_IFMT) == S_IFREG) {
2390                 /* only struct inode is valid if it's an inline symlink */
2391                 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2392         }
2393         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2394 }
2395
2396 static void init_once(void *foo)
2397 {
2398         struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2399
2400         inode_init_once(&p->vfs_inode);
2401 }
2402
2403 static int init_inodecache(void)
2404 {
2405         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2406                                 sizeof(struct shmem_inode_info),
2407                                 0, SLAB_PANIC, init_once);
2408         return 0;
2409 }
2410
2411 static void destroy_inodecache(void)
2412 {
2413         kmem_cache_destroy(shmem_inode_cachep);
2414 }
2415
2416 static const struct address_space_operations shmem_aops = {
2417         .writepage      = shmem_writepage,
2418         .set_page_dirty = __set_page_dirty_no_writeback,
2419 #ifdef CONFIG_TMPFS
2420         .readpage       = shmem_readpage,
2421         .write_begin    = shmem_write_begin,
2422         .write_end      = shmem_write_end,
2423 #endif
2424         .migratepage    = migrate_page,
2425         .error_remove_page = generic_error_remove_page,
2426 };
2427
2428 static const struct file_operations shmem_file_operations = {
2429         .mmap           = shmem_mmap,
2430 #ifdef CONFIG_TMPFS
2431         .llseek         = generic_file_llseek,
2432         .read           = do_sync_read,
2433         .write          = do_sync_write,
2434         .aio_read       = shmem_file_aio_read,
2435         .aio_write      = generic_file_aio_write,
2436         .fsync          = simple_sync_file,
2437         .splice_read    = generic_file_splice_read,
2438         .splice_write   = generic_file_splice_write,
2439 #endif
2440 };
2441
2442 static const struct inode_operations shmem_inode_operations = {
2443         .truncate       = shmem_truncate,
2444         .setattr        = shmem_notify_change,
2445         .truncate_range = shmem_truncate_range,
2446 #ifdef CONFIG_TMPFS_POSIX_ACL
2447         .setxattr       = generic_setxattr,
2448         .getxattr       = generic_getxattr,
2449         .listxattr      = generic_listxattr,
2450         .removexattr    = generic_removexattr,
2451         .check_acl      = generic_check_acl,
2452 #endif
2453
2454 };
2455
2456 static const struct inode_operations shmem_dir_inode_operations = {
2457 #ifdef CONFIG_TMPFS
2458         .create         = shmem_create,
2459         .lookup         = simple_lookup,
2460         .link           = shmem_link,
2461         .unlink         = shmem_unlink,
2462         .symlink        = shmem_symlink,
2463         .mkdir          = shmem_mkdir,
2464         .rmdir          = shmem_rmdir,
2465         .mknod          = shmem_mknod,
2466         .rename         = shmem_rename,
2467 #endif
2468 #ifdef CONFIG_TMPFS_POSIX_ACL
2469         .setattr        = shmem_notify_change,
2470         .setxattr       = generic_setxattr,
2471         .getxattr       = generic_getxattr,
2472         .listxattr      = generic_listxattr,
2473         .removexattr    = generic_removexattr,
2474         .check_acl      = generic_check_acl,
2475 #endif
2476 };
2477
2478 static const struct inode_operations shmem_special_inode_operations = {
2479 #ifdef CONFIG_TMPFS_POSIX_ACL
2480         .setattr        = shmem_notify_change,
2481         .setxattr       = generic_setxattr,
2482         .getxattr       = generic_getxattr,
2483         .listxattr      = generic_listxattr,
2484         .removexattr    = generic_removexattr,
2485         .check_acl      = generic_check_acl,
2486 #endif
2487 };
2488
2489 static const struct super_operations shmem_ops = {
2490         .alloc_inode    = shmem_alloc_inode,
2491         .destroy_inode  = shmem_destroy_inode,
2492 #ifdef CONFIG_TMPFS
2493         .statfs         = shmem_statfs,
2494         .remount_fs     = shmem_remount_fs,
2495         .show_options   = shmem_show_options,
2496 #endif
2497         .delete_inode   = shmem_delete_inode,
2498         .drop_inode     = generic_delete_inode,
2499         .put_super      = shmem_put_super,
2500 };
2501
2502 static const struct vm_operations_struct shmem_vm_ops = {
2503         .fault          = shmem_fault,
2504 #ifdef CONFIG_NUMA
2505         .set_policy     = shmem_set_policy,
2506         .get_policy     = shmem_get_policy,
2507 #endif
2508 };
2509
2510
2511 static int shmem_get_sb(struct file_system_type *fs_type,
2512         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2513 {
2514         return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
2515 }
2516
2517 static struct file_system_type tmpfs_fs_type = {
2518         .owner          = THIS_MODULE,
2519         .name           = "tmpfs",
2520         .get_sb         = shmem_get_sb,
2521         .kill_sb        = kill_litter_super,
2522 };
2523
2524 int __init init_tmpfs(void)
2525 {
2526         int error;
2527
2528         error = bdi_init(&shmem_backing_dev_info);
2529         if (error)
2530                 goto out4;
2531
2532         error = init_inodecache();
2533         if (error)
2534                 goto out3;
2535
2536         error = register_filesystem(&tmpfs_fs_type);
2537         if (error) {
2538                 printk(KERN_ERR "Could not register tmpfs\n");
2539                 goto out2;
2540         }
2541
2542         shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
2543                                 tmpfs_fs_type.name, NULL);
2544         if (IS_ERR(shm_mnt)) {
2545                 error = PTR_ERR(shm_mnt);
2546                 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2547                 goto out1;
2548         }
2549         return 0;
2550
2551 out1:
2552         unregister_filesystem(&tmpfs_fs_type);
2553 out2:
2554         destroy_inodecache();
2555 out3:
2556         bdi_destroy(&shmem_backing_dev_info);
2557 out4:
2558         shm_mnt = ERR_PTR(error);
2559         return error;
2560 }
2561
2562 #else /* !CONFIG_SHMEM */
2563
2564 /*
2565  * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2566  *
2567  * This is intended for small system where the benefits of the full
2568  * shmem code (swap-backed and resource-limited) are outweighed by
2569  * their complexity. On systems without swap this code should be
2570  * effectively equivalent, but much lighter weight.
2571  */
2572
2573 #include <linux/ramfs.h>
2574
2575 static struct file_system_type tmpfs_fs_type = {
2576         .name           = "tmpfs",
2577         .get_sb         = ramfs_get_sb,
2578         .kill_sb        = kill_litter_super,
2579 };
2580
2581 int __init init_tmpfs(void)
2582 {
2583         BUG_ON(register_filesystem(&tmpfs_fs_type) != 0);
2584
2585         shm_mnt = kern_mount(&tmpfs_fs_type);
2586         BUG_ON(IS_ERR(shm_mnt));
2587
2588         return 0;
2589 }
2590
2591 int shmem_unuse(swp_entry_t entry, struct page *page)
2592 {
2593         return 0;
2594 }
2595
2596 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2597 {
2598         return 0;
2599 }
2600
2601 #define shmem_vm_ops                            generic_file_vm_ops
2602 #define shmem_file_operations                   ramfs_file_operations
2603 #define shmem_get_inode(sb, dir, mode, dev, flags)      ramfs_get_inode(sb, dir, mode, dev)
2604 #define shmem_acct_size(flags, size)            0
2605 #define shmem_unacct_size(flags, size)          do {} while (0)
2606 #define SHMEM_MAX_BYTES                         MAX_LFS_FILESIZE
2607
2608 #endif /* CONFIG_SHMEM */
2609
2610 /* common code */
2611
2612 /**
2613  * shmem_file_setup - get an unlinked file living in tmpfs
2614  * @name: name for dentry (to be seen in /proc/<pid>/maps
2615  * @size: size to be set for the file
2616  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2617  */
2618 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2619 {
2620         int error;
2621         struct file *file;
2622         struct inode *inode;
2623         struct path path;
2624         struct dentry *root;
2625         struct qstr this;
2626
2627         if (IS_ERR(shm_mnt))
2628                 return (void *)shm_mnt;
2629
2630         if (size < 0 || size > SHMEM_MAX_BYTES)
2631                 return ERR_PTR(-EINVAL);
2632
2633         if (shmem_acct_size(flags, size))
2634                 return ERR_PTR(-ENOMEM);
2635
2636         error = -ENOMEM;
2637         this.name = name;
2638         this.len = strlen(name);
2639         this.hash = 0; /* will go */
2640         root = shm_mnt->mnt_root;
2641         path.dentry = d_alloc(root, &this);
2642         if (!path.dentry)
2643                 goto put_memory;
2644         path.mnt = mntget(shm_mnt);
2645
2646         error = -ENOSPC;
2647         inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2648         if (!inode)
2649                 goto put_dentry;
2650
2651         d_instantiate(path.dentry, inode);
2652         inode->i_size = size;
2653         inode->i_nlink = 0;     /* It is unlinked */
2654 #ifndef CONFIG_MMU
2655         error = ramfs_nommu_expand_for_mapping(inode, size);
2656         if (error)
2657                 goto put_dentry;
2658 #endif
2659
2660         error = -ENFILE;
2661         file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2662                   &shmem_file_operations);
2663         if (!file)
2664                 goto put_dentry;
2665
2666         return file;
2667
2668 put_dentry:
2669         path_put(&path);
2670 put_memory:
2671         shmem_unacct_size(flags, size);
2672         return ERR_PTR(error);
2673 }
2674 EXPORT_SYMBOL_GPL(shmem_file_setup);
2675
2676 /**
2677  * shmem_zero_setup - setup a shared anonymous mapping
2678  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2679  */
2680 int shmem_zero_setup(struct vm_area_struct *vma)
2681 {
2682         struct file *file;
2683         loff_t size = vma->vm_end - vma->vm_start;
2684
2685         file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2686         if (IS_ERR(file))
2687                 return PTR_ERR(file);
2688
2689         if (vma->vm_file)
2690                 fput(vma->vm_file);
2691         vma->vm_file = file;
2692         vma->vm_ops = &shmem_vm_ops;
2693         return 0;
2694 }