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 #include <linux/ima.h>
  33
  34 static struct vfsmount *shm_mnt;
  35
  36 #ifdef CONFIG_SHMEM
  37 /*
  38  * This virtual memory filesystem is heavily based on the ramfs. It
  39  * extends ramfs by the ability to use swap and honor resource limits
  40  * which makes it a completely usable filesystem.
  41  */
  42
  43 #include <linux/xattr.h>
  44 #include <linux/exportfs.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/vfs.h>
  53 #include <linux/blkdev.h>
  54 #include <linux/security.h>
  55 #include <linux/swapops.h>
  56 #include <linux/mempolicy.h>
  57 #include <linux/namei.h>
  58 #include <linux/ctype.h>
  59 #include <linux/migrate.h>
  60 #include <linux/highmem.h>
  61 #include <linux/seq_file.h>
  62 #include <linux/magic.h>
  63
  64 #include <asm/uaccess.h>
  65 #include <asm/div64.h>
  66 #include <asm/pgtable.h>
  67
  68 /*
  69  * The maximum size of a shmem/tmpfs file is limited by the maximum size of
  70  * its triple-indirect swap vector - see illustration at shmem_swp_entry().
  71  *
  72  * With 4kB page size, maximum file size is just over 2TB on a 32-bit kernel,
  73  * but one eighth of that on a 64-bit kernel.  With 8kB page size, maximum
  74  * file size is just over 4TB on a 64-bit kernel, but 16TB on a 32-bit kernel,
  75  * MAX_LFS_FILESIZE being then more restrictive than swap vector layout.
  76  *
  77  * We use / and * instead of shifts in the definitions below, so that the swap
  78  * vector can be tested with small even values (e.g. 20) for ENTRIES_PER_PAGE.
  79  */
  80 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
  81 #define ENTRIES_PER_PAGEPAGE ((unsigned long long)ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
  82
  83 #define SHMSWP_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
  84 #define SHMSWP_MAX_BYTES (SHMSWP_MAX_INDEX << PAGE_CACHE_SHIFT)
  85
  86 #define SHMEM_MAX_BYTES  min_t(unsigned long long, SHMSWP_MAX_BYTES, MAX_LFS_FILESIZE)
  87 #define SHMEM_MAX_INDEX  ((unsigned long)((SHMEM_MAX_BYTES+1) >> PAGE_CACHE_SHIFT))
  88
  89 #define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
  90 #define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
  91
  92 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
  93 #define SHMEM_PAGEIN     VM_READ
  94 #define SHMEM_TRUNCATE   VM_WRITE
  95
  96 /* Definition to limit shmem_truncate's steps between cond_rescheds */
  97 #define LATENCY_LIMIT    64
  98
  99 /* Pretend that each entry is of this size in directory's i_size */
 100 #define BOGO_DIRENT_SIZE 20
 101
 102 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
 103 enum sgp_type {
 104         SGP_READ,       /* don't exceed i_size, don't allocate page */
 105         SGP_CACHE,      /* don't exceed i_size, may allocate page */
 106         SGP_DIRTY,      /* like SGP_CACHE, but set new page dirty */
 107         SGP_WRITE,      /* may exceed i_size, may allocate page */
 108 };
 109
 110 #ifdef CONFIG_TMPFS
 111 static unsigned long shmem_default_max_blocks(void)
 112 {
 113         return totalram_pages / 2;
 114 }
 115
 116 static unsigned long shmem_default_max_inodes(void)
 117 {
 118         return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
 119 }
 120 #endif
 121
 122 static int shmem_getpage(struct inode *inode, unsigned long idx,
 123                          struct page **pagep, enum sgp_type sgp, int *type);
 124
 125 static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
 126 {
 127         /*
 128          * The above definition of ENTRIES_PER_PAGE, and the use of
 129          * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
 130          * might be reconsidered if it ever diverges from PAGE_SIZE.
 131          *
 132          * Mobility flags are masked out as swap vectors cannot move
 133          */
 134         return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO,
 135                                 PAGE_CACHE_SHIFT-PAGE_SHIFT);
 136 }
 137
 138 static inline void shmem_dir_free(struct page *page)
 139 {
 140         __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
 141 }
 142
 143 static struct page **shmem_dir_map(struct page *page)
 144 {
 145         return (struct page **)kmap_atomic(page, KM_USER0);
 146 }
 147
 148 static inline void shmem_dir_unmap(struct page **dir)
 149 {
 150         kunmap_atomic(dir, KM_USER0);
 151 }
 152
 153 static swp_entry_t *shmem_swp_map(struct page *page)
 154 {
 155         return (swp_entry_t *)kmap_atomic(page, KM_USER1);
 156 }
 157
 158 static inline void shmem_swp_balance_unmap(void)
 159 {
 160         /*
 161          * When passing a pointer to an i_direct entry, to code which
 162          * also handles indirect entries and so will shmem_swp_unmap,
 163          * we must arrange for the preempt count to remain in balance.
 164          * What kmap_atomic of a lowmem page does depends on config
 165          * and architecture, so pretend to kmap_atomic some lowmem page.
 166          */
 167         (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
 168 }
 169
 170 static inline void shmem_swp_unmap(swp_entry_t *entry)
 171 {
 172         kunmap_atomic(entry, KM_USER1);
 173 }
 174
 175 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
 176 {
 177         return sb->s_fs_info;
 178 }
 179
 180 /*
 181  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
 182  * for shared memory and for shared anonymous (/dev/zero) mappings
 183  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
 184  * consistent with the pre-accounting of private mappings ...
 185  */
 186 static inline int shmem_acct_size(unsigned long flags, loff_t size)
 187 {
 188         return (flags & VM_NORESERVE) ?
 189                 0 : security_vm_enough_memory_kern(VM_ACCT(size));
 190 }
 191
 192 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
 193 {
 194         if (!(flags & VM_NORESERVE))
 195                 vm_unacct_memory(VM_ACCT(size));
 196 }
 197
 198 /*
 199  * ... whereas tmpfs objects are accounted incrementally as
 200  * pages are allocated, in order to allow huge sparse files.
 201  * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
 202  * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
 203  */
 204 static inline int shmem_acct_block(unsigned long flags)
 205 {
 206         return (flags & VM_NORESERVE) ?
 207                 security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE)) : 0;
 208 }
 209
 210 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
 211 {
 212         if (flags & VM_NORESERVE)
 213                 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
 214 }
 215
 216 static const struct super_operations shmem_ops;
 217 static const struct address_space_operations shmem_aops;
 218 static const struct file_operations shmem_file_operations;
 219 static const struct inode_operations shmem_inode_operations;
 220 static const struct inode_operations shmem_dir_inode_operations;
 221 static const struct inode_operations shmem_special_inode_operations;
 222 static struct vm_operations_struct shmem_vm_ops;
 223
 224 static struct backing_dev_info shmem_backing_dev_info  __read_mostly = {
 225         .ra_pages       = 0,    /* No readahead */
 226         .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
 227         .unplug_io_fn   = default_unplug_io_fn,
 228 };
 229
 230 static LIST_HEAD(shmem_swaplist);
 231 static DEFINE_MUTEX(shmem_swaplist_mutex);
 232
 233 static void shmem_free_blocks(struct inode *inode, long pages)
 234 {
 235         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 236         if (sbinfo->max_blocks) {
 237                 spin_lock(&sbinfo->stat_lock);
 238                 sbinfo->free_blocks += pages;
 239                 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
 240                 spin_unlock(&sbinfo->stat_lock);
 241         }
 242 }
 243
 244 static int shmem_reserve_inode(struct super_block *sb)
 245 {
 246         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 247         if (sbinfo->max_inodes) {
 248                 spin_lock(&sbinfo->stat_lock);
 249                 if (!sbinfo->free_inodes) {
 250                         spin_unlock(&sbinfo->stat_lock);
 251                         return -ENOSPC;
 252                 }
 253                 sbinfo->free_inodes--;
 254                 spin_unlock(&sbinfo->stat_lock);
 255         }
 256         return 0;
 257 }
 258
 259 static void shmem_free_inode(struct super_block *sb)
 260 {
 261         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 262         if (sbinfo->max_inodes) {
 263                 spin_lock(&sbinfo->stat_lock);
 264                 sbinfo->free_inodes++;
 265                 spin_unlock(&sbinfo->stat_lock);
 266         }
 267 }
 268
 269 /**
 270  * shmem_recalc_inode - recalculate the size of an inode
 271  * @inode: inode to recalc
 272  *
 273  * We have to calculate the free blocks since the mm can drop
 274  * undirtied hole pages behind our back.
 275  *
 276  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
 277  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
 278  *
 279  * It has to be called with the spinlock held.
 280  */
 281 static void shmem_recalc_inode(struct inode *inode)
 282 {
 283         struct shmem_inode_info *info = SHMEM_I(inode);
 284         long freed;
 285
 286         freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
 287         if (freed > 0) {
 288                 info->alloced -= freed;
 289                 shmem_unacct_blocks(info->flags, freed);
 290                 shmem_free_blocks(inode, freed);
 291         }
 292 }
 293
 294 /**
 295  * shmem_swp_entry - find the swap vector position in the info structure
 296  * @info:  info structure for the inode
 297  * @index: index of the page to find
 298  * @page:  optional page to add to the structure. Has to be preset to
 299  *         all zeros
 300  *
 301  * If there is no space allocated yet it will return NULL when
 302  * page is NULL, else it will use the page for the needed block,
 303  * setting it to NULL on return to indicate that it has been used.
 304  *
 305  * The swap vector is organized the following way:
 306  *
 307  * There are SHMEM_NR_DIRECT entries directly stored in the
 308  * shmem_inode_info structure. So small files do not need an addional
 309  * allocation.
 310  *
 311  * For pages with index > SHMEM_NR_DIRECT there is the pointer
 312  * i_indirect which points to a page which holds in the first half
 313  * doubly indirect blocks, in the second half triple indirect blocks:
 314  *
 315  * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
 316  * following layout (for SHMEM_NR_DIRECT == 16):
 317  *
 318  * i_indirect -> dir --> 16-19
 319  *            |      +-> 20-23
 320  *            |
 321  *            +-->dir2 --> 24-27
 322  *            |        +-> 28-31
 323  *            |        +-> 32-35
 324  *            |        +-> 36-39
 325  *            |
 326  *            +-->dir3 --> 40-43
 327  *                     +-> 44-47
 328  *                     +-> 48-51
 329  *                     +-> 52-55
 330  */
 331 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
 332 {
 333         unsigned long offset;
 334         struct page **dir;
 335         struct page *subdir;
 336
 337         if (index < SHMEM_NR_DIRECT) {
 338                 shmem_swp_balance_unmap();
 339                 return info->i_direct+index;
 340         }
 341         if (!info->i_indirect) {
 342                 if (page) {
 343                         info->i_indirect = *page;
 344                         *page = NULL;
 345                 }
 346                 return NULL;                    /* need another page */
 347         }
 348
 349         index -= SHMEM_NR_DIRECT;
 350         offset = index % ENTRIES_PER_PAGE;
 351         index /= ENTRIES_PER_PAGE;
 352         dir = shmem_dir_map(info->i_indirect);
 353
 354         if (index >= ENTRIES_PER_PAGE/2) {
 355                 index -= ENTRIES_PER_PAGE/2;
 356                 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
 357                 index %= ENTRIES_PER_PAGE;
 358                 subdir = *dir;
 359                 if (!subdir) {
 360                         if (page) {
 361                                 *dir = *page;
 362                                 *page = NULL;
 363                         }
 364                         shmem_dir_unmap(dir);
 365                         return NULL;            /* need another page */
 366                 }
 367                 shmem_dir_unmap(dir);
 368                 dir = shmem_dir_map(subdir);
 369         }
 370
 371         dir += index;
 372         subdir = *dir;
 373         if (!subdir) {
 374                 if (!page || !(subdir = *page)) {
 375                         shmem_dir_unmap(dir);
 376                         return NULL;            /* need a page */
 377                 }
 378                 *dir = subdir;
 379                 *page = NULL;
 380         }
 381         shmem_dir_unmap(dir);
 382         return shmem_swp_map(subdir) + offset;
 383 }
 384
 385 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
 386 {
 387         long incdec = value? 1: -1;
 388
 389         entry->val = value;
 390         info->swapped += incdec;
 391         if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
 392                 struct page *page = kmap_atomic_to_page(entry);
 393                 set_page_private(page, page_private(page) + incdec);
 394         }
 395 }
 396
 397 /**
 398  * shmem_swp_alloc - get the position of the swap entry for the page.
 399  * @info:       info structure for the inode
 400  * @index:      index of the page to find
 401  * @sgp:        check and recheck i_size? skip allocation?
 402  *
 403  * If the entry does not exist, allocate it.
 404  */
 405 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
 406 {
 407         struct inode *inode = &info->vfs_inode;
 408         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 409         struct page *page = NULL;
 410         swp_entry_t *entry;
 411
 412         if (sgp != SGP_WRITE &&
 413             ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
 414                 return ERR_PTR(-EINVAL);
 415
 416         while (!(entry = shmem_swp_entry(info, index, &page))) {
 417                 if (sgp == SGP_READ)
 418                         return shmem_swp_map(ZERO_PAGE(0));
 419                 /*
 420                  * Test free_blocks against 1 not 0, since we have 1 data
 421                  * page (and perhaps indirect index pages) yet to allocate:
 422                  * a waste to allocate index if we cannot allocate data.
 423                  */
 424                 if (sbinfo->max_blocks) {
 425                         spin_lock(&sbinfo->stat_lock);
 426                         if (sbinfo->free_blocks <= 1) {
 427                                 spin_unlock(&sbinfo->stat_lock);
 428                                 return ERR_PTR(-ENOSPC);
 429                         }
 430                         sbinfo->free_blocks--;
 431                         inode->i_blocks += BLOCKS_PER_PAGE;
 432                         spin_unlock(&sbinfo->stat_lock);
 433                 }
 434
 435                 spin_unlock(&info->lock);
 436                 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
 437                 if (page)
 438                         set_page_private(page, 0);
 439                 spin_lock(&info->lock);
 440
 441                 if (!page) {
 442                         shmem_free_blocks(inode, 1);
 443                         return ERR_PTR(-ENOMEM);
 444                 }
 445                 if (sgp != SGP_WRITE &&
 446                     ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
 447                         entry = ERR_PTR(-EINVAL);
 448                         break;
 449                 }
 450                 if (info->next_index <= index)
 451                         info->next_index = index + 1;
 452         }
 453         if (page) {
 454                 /* another task gave its page, or truncated the file */
 455                 shmem_free_blocks(inode, 1);
 456                 shmem_dir_free(page);
 457         }
 458         if (info->next_index <= index && !IS_ERR(entry))
 459                 info->next_index = index + 1;
 460         return entry;
 461 }
 462
 463 /**
 464  * shmem_free_swp - free some swap entries in a directory
 465  * @dir:        pointer to the directory
 466  * @edir:       pointer after last entry of the directory
 467  * @punch_lock: pointer to spinlock when needed for the holepunch case
 468  */
 469 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
 470                                                 spinlock_t *punch_lock)
 471 {
 472         spinlock_t *punch_unlock = NULL;
 473         swp_entry_t *ptr;
 474         int freed = 0;
 475
 476         for (ptr = dir; ptr < edir; ptr++) {
 477                 if (ptr->val) {
 478                         if (unlikely(punch_lock)) {
 479                                 punch_unlock = punch_lock;
 480                                 punch_lock = NULL;
 481                                 spin_lock(punch_unlock);
 482                                 if (!ptr->val)
 483                                         continue;
 484                         }
 485                         free_swap_and_cache(*ptr);
 486                         *ptr = (swp_entry_t){0};
 487                         freed++;
 488                 }
 489         }
 490         if (punch_unlock)
 491                 spin_unlock(punch_unlock);
 492         return freed;
 493 }
 494
 495 static int shmem_map_and_free_swp(struct page *subdir, int offset,
 496                 int limit, struct page ***dir, spinlock_t *punch_lock)
 497 {
 498         swp_entry_t *ptr;
 499         int freed = 0;
 500
 501         ptr = shmem_swp_map(subdir);
 502         for (; offset < limit; offset += LATENCY_LIMIT) {
 503                 int size = limit - offset;
 504                 if (size > LATENCY_LIMIT)
 505                         size = LATENCY_LIMIT;
 506                 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
 507                                                         punch_lock);
 508                 if (need_resched()) {
 509                         shmem_swp_unmap(ptr);
 510                         if (*dir) {
 511                                 shmem_dir_unmap(*dir);
 512                                 *dir = NULL;
 513                         }
 514                         cond_resched();
 515                         ptr = shmem_swp_map(subdir);
 516                 }
 517         }
 518         shmem_swp_unmap(ptr);
 519         return freed;
 520 }
 521
 522 static void shmem_free_pages(struct list_head *next)
 523 {
 524         struct page *page;
 525         int freed = 0;
 526
 527         do {
 528                 page = container_of(next, struct page, lru);
 529                 next = next->next;
 530                 shmem_dir_free(page);
 531                 freed++;
 532                 if (freed >= LATENCY_LIMIT) {
 533                         cond_resched();
 534                         freed = 0;
 535                 }
 536         } while (next);
 537 }
 538
 539 static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
 540 {
 541         struct shmem_inode_info *info = SHMEM_I(inode);
 542         unsigned long idx;
 543         unsigned long size;
 544         unsigned long limit;
 545         unsigned long stage;
 546         unsigned long diroff;
 547         struct page **dir;
 548         struct page *topdir;
 549         struct page *middir;
 550         struct page *subdir;
 551         swp_entry_t *ptr;
 552         LIST_HEAD(pages_to_free);
 553         long nr_pages_to_free = 0;
 554         long nr_swaps_freed = 0;
 555         int offset;
 556         int freed;
 557         int punch_hole;
 558         spinlock_t *needs_lock;
 559         spinlock_t *punch_lock;
 560         unsigned long upper_limit;
 561
 562         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 563         idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 564         if (idx >= info->next_index)
 565                 return;
 566
 567         spin_lock(&info->lock);
 568         info->flags |= SHMEM_TRUNCATE;
 569         if (likely(end == (loff_t) -1)) {
 570                 limit = info->next_index;
 571                 upper_limit = SHMEM_MAX_INDEX;
 572                 info->next_index = idx;
 573                 needs_lock = NULL;
 574                 punch_hole = 0;
 575         } else {
 576                 if (end + 1 >= inode->i_size) { /* we may free a little more */
 577                         limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
 578                                                         PAGE_CACHE_SHIFT;
 579                         upper_limit = SHMEM_MAX_INDEX;
 580                 } else {
 581                         limit = (end + 1) >> PAGE_CACHE_SHIFT;
 582                         upper_limit = limit;
 583                 }
 584                 needs_lock = &info->lock;
 585                 punch_hole = 1;
 586         }
 587
 588         topdir = info->i_indirect;
 589         if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
 590                 info->i_indirect = NULL;
 591                 nr_pages_to_free++;
 592                 list_add(&topdir->lru, &pages_to_free);
 593         }
 594         spin_unlock(&info->lock);
 595
 596         if (info->swapped && idx < SHMEM_NR_DIRECT) {
 597                 ptr = info->i_direct;
 598                 size = limit;
 599                 if (size > SHMEM_NR_DIRECT)
 600                         size = SHMEM_NR_DIRECT;
 601                 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
 602         }
 603
 604         /*
 605          * If there are no indirect blocks or we are punching a hole
 606          * below indirect blocks, nothing to be done.
 607          */
 608         if (!topdir || limit <= SHMEM_NR_DIRECT)
 609                 goto done2;
 610
 611         /*
 612          * The truncation case has already dropped info->lock, and we're safe
 613          * because i_size and next_index have already been lowered, preventing
 614          * access beyond.  But in the punch_hole case, we still need to take
 615          * the lock when updating the swap directory, because there might be
 616          * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
 617          * shmem_writepage.  However, whenever we find we can remove a whole
 618          * directory page (not at the misaligned start or end of the range),
 619          * we first NULLify its pointer in the level above, and then have no
 620          * need to take the lock when updating its contents: needs_lock and
 621          * punch_lock (either pointing to info->lock or NULL) manage this.
 622          */
 623
 624         upper_limit -= SHMEM_NR_DIRECT;
 625         limit -= SHMEM_NR_DIRECT;
 626         idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
 627         offset = idx % ENTRIES_PER_PAGE;
 628         idx -= offset;
 629
 630         dir = shmem_dir_map(topdir);
 631         stage = ENTRIES_PER_PAGEPAGE/2;
 632         if (idx < ENTRIES_PER_PAGEPAGE/2) {
 633                 middir = topdir;
 634                 diroff = idx/ENTRIES_PER_PAGE;
 635         } else {
 636                 dir += ENTRIES_PER_PAGE/2;
 637                 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
 638                 while (stage <= idx)
 639                         stage += ENTRIES_PER_PAGEPAGE;
 640                 middir = *dir;
 641                 if (*dir) {
 642                         diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
 643                                 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
 644                         if (!diroff && !offset && upper_limit >= stage) {
 645                                 if (needs_lock) {
 646                                         spin_lock(needs_lock);
 647                                         *dir = NULL;
 648                                         spin_unlock(needs_lock);
 649                                         needs_lock = NULL;
 650                                 } else
 651                                         *dir = NULL;
 652                                 nr_pages_to_free++;
 653                                 list_add(&middir->lru, &pages_to_free);
 654                         }
 655                         shmem_dir_unmap(dir);
 656                         dir = shmem_dir_map(middir);
 657                 } else {
 658                         diroff = 0;
 659                         offset = 0;
 660                         idx = stage;
 661                 }
 662         }
 663
 664         for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
 665                 if (unlikely(idx == stage)) {
 666                         shmem_dir_unmap(dir);
 667                         dir = shmem_dir_map(topdir) +
 668                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
 669                         while (!*dir) {
 670                                 dir++;
 671                                 idx += ENTRIES_PER_PAGEPAGE;
 672                                 if (idx >= limit)
 673                                         goto done1;
 674                         }
 675                         stage = idx + ENTRIES_PER_PAGEPAGE;
 676                         middir = *dir;
 677                         if (punch_hole)
 678                                 needs_lock = &info->lock;
 679                         if (upper_limit >= stage) {
 680                                 if (needs_lock) {
 681                                         spin_lock(needs_lock);
 682                                         *dir = NULL;
 683                                         spin_unlock(needs_lock);
 684                                         needs_lock = NULL;
 685                                 } else
 686                                         *dir = NULL;
 687                                 nr_pages_to_free++;
 688                                 list_add(&middir->lru, &pages_to_free);
 689                         }
 690                         shmem_dir_unmap(dir);
 691                         cond_resched();
 692                         dir = shmem_dir_map(middir);
 693                         diroff = 0;
 694                 }
 695                 punch_lock = needs_lock;
 696                 subdir = dir[diroff];
 697                 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
 698                         if (needs_lock) {
 699                                 spin_lock(needs_lock);
 700                                 dir[diroff] = NULL;
 701                                 spin_unlock(needs_lock);
 702                                 punch_lock = NULL;
 703                         } else
 704                                 dir[diroff] = NULL;
 705                         nr_pages_to_free++;
 706                         list_add(&subdir->lru, &pages_to_free);
 707                 }
 708                 if (subdir && page_private(subdir) /* has swap entries */) {
 709                         size = limit - idx;
 710                         if (size > ENTRIES_PER_PAGE)
 711                                 size = ENTRIES_PER_PAGE;
 712                         freed = shmem_map_and_free_swp(subdir,
 713                                         offset, size, &dir, punch_lock);
 714                         if (!dir)
 715                                 dir = shmem_dir_map(middir);
 716                         nr_swaps_freed += freed;
 717                         if (offset || punch_lock) {
 718                                 spin_lock(&info->lock);
 719                                 set_page_private(subdir,
 720                                         page_private(subdir) - freed);
 721                                 spin_unlock(&info->lock);
 722                         } else
 723                                 BUG_ON(page_private(subdir) != freed);
 724                 }
 725                 offset = 0;
 726         }
 727 done1:
 728         shmem_dir_unmap(dir);
 729 done2:
 730         if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
 731                 /*
 732                  * Call truncate_inode_pages again: racing shmem_unuse_inode
 733                  * may have swizzled a page in from swap since vmtruncate or
 734                  * generic_delete_inode did it, before we lowered next_index.
 735                  * Also, though shmem_getpage checks i_size before adding to
 736                  * cache, no recheck after: so fix the narrow window there too.
 737                  *
 738                  * Recalling truncate_inode_pages_range and unmap_mapping_range
 739                  * every time for punch_hole (which never got a chance to clear
 740                  * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
 741                  * yet hardly ever necessary: try to optimize them out later.
 742                  */
 743                 truncate_inode_pages_range(inode->i_mapping, start, end);
 744                 if (punch_hole)
 745                         unmap_mapping_range(inode->i_mapping, start,
 746                                                         end - start, 1);
 747         }
 748
 749         spin_lock(&info->lock);
 750         info->flags &= ~SHMEM_TRUNCATE;
 751         info->swapped -= nr_swaps_freed;
 752         if (nr_pages_to_free)
 753                 shmem_free_blocks(inode, nr_pages_to_free);
 754         shmem_recalc_inode(inode);
 755         spin_unlock(&info->lock);
 756
 757         /*
 758          * Empty swap vector directory pages to be freed?
 759          */
 760         if (!list_empty(&pages_to_free)) {
 761                 pages_to_free.prev->next = NULL;
 762                 shmem_free_pages(pages_to_free.next);
 763         }
 764 }
 765
 766 static void shmem_truncate(struct inode *inode)
 767 {
 768         shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
 769 }
 770
 771 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
 772 {
 773         struct inode *inode = dentry->d_inode;
 774         struct page *page = NULL;
 775         int error;
 776
 777         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
 778                 if (attr->ia_size < inode->i_size) {
 779                         /*
 780                          * If truncating down to a partial page, then
 781                          * if that page is already allocated, hold it
 782                          * in memory until the truncation is over, so
 783                          * truncate_partial_page cannnot miss it were
 784                          * it assigned to swap.
 785                          */
 786                         if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
 787                                 (void) shmem_getpage(inode,
 788                                         attr->ia_size>>PAGE_CACHE_SHIFT,
 789                                                 &page, SGP_READ, NULL);
 790                                 if (page)
 791                                         unlock_page(page);
 792                         }
 793                         /*
 794                          * Reset SHMEM_PAGEIN flag so that shmem_truncate can
 795                          * detect if any pages might have been added to cache
 796                          * after truncate_inode_pages.  But we needn't bother
 797                          * if it's being fully truncated to zero-length: the
 798                          * nrpages check is efficient enough in that case.
 799                          */
 800                         if (attr->ia_size) {
 801                                 struct shmem_inode_info *info = SHMEM_I(inode);
 802                                 spin_lock(&info->lock);
 803                                 info->flags &= ~SHMEM_PAGEIN;
 804                                 spin_unlock(&info->lock);
 805                         }
 806                 }
 807         }
 808
 809         error = inode_change_ok(inode, attr);
 810         if (!error)
 811                 error = inode_setattr(inode, attr);
 812 #ifdef CONFIG_TMPFS_POSIX_ACL
 813         if (!error && (attr->ia_valid & ATTR_MODE))
 814                 error = generic_acl_chmod(inode, &shmem_acl_ops);
 815 #endif
 816         if (page)
 817                 page_cache_release(page);
 818         return error;
 819 }
 820
 821 static void shmem_delete_inode(struct inode *inode)
 822 {
 823         struct shmem_inode_info *info = SHMEM_I(inode);
 824
 825         if (inode->i_op->truncate == shmem_truncate) {
 826                 truncate_inode_pages(inode->i_mapping, 0);
 827                 shmem_unacct_size(info->flags, inode->i_size);
 828                 inode->i_size = 0;
 829                 shmem_truncate(inode);
 830                 if (!list_empty(&info->swaplist)) {
 831                         mutex_lock(&shmem_swaplist_mutex);
 832                         list_del_init(&info->swaplist);
 833                         mutex_unlock(&shmem_swaplist_mutex);
 834                 }
 835         }
 836         BUG_ON(inode->i_blocks);
 837         shmem_free_inode(inode->i_sb);
 838         clear_inode(inode);
 839 }
 840
 841 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
 842 {
 843         swp_entry_t *ptr;
 844
 845         for (ptr = dir; ptr < edir; ptr++) {
 846                 if (ptr->val == entry.val)
 847                         return ptr - dir;
 848         }
 849         return -1;
 850 }
 851
 852 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
 853 {
 854         struct inode *inode;
 855         unsigned long idx;
 856         unsigned long size;
 857         unsigned long limit;
 858         unsigned long stage;
 859         struct page **dir;
 860         struct page *subdir;
 861         swp_entry_t *ptr;
 862         int offset;
 863         int error;
 864
 865         idx = 0;
 866         ptr = info->i_direct;
 867         spin_lock(&info->lock);
 868         if (!info->swapped) {
 869                 list_del_init(&info->swaplist);
 870                 goto lost2;
 871         }
 872         limit = info->next_index;
 873         size = limit;
 874         if (size > SHMEM_NR_DIRECT)
 875                 size = SHMEM_NR_DIRECT;
 876         offset = shmem_find_swp(entry, ptr, ptr+size);
 877         if (offset >= 0)
 878                 goto found;
 879         if (!info->i_indirect)
 880                 goto lost2;
 881
 882         dir = shmem_dir_map(info->i_indirect);
 883         stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
 884
 885         for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
 886                 if (unlikely(idx == stage)) {
 887                         shmem_dir_unmap(dir-1);
 888                         if (cond_resched_lock(&info->lock)) {
 889                                 /* check it has not been truncated */
 890                                 if (limit > info->next_index) {
 891                                         limit = info->next_index;
 892                                         if (idx >= limit)
 893                                                 goto lost2;
 894                                 }
 895                         }
 896                         dir = shmem_dir_map(info->i_indirect) +
 897                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
 898                         while (!*dir) {
 899                                 dir++;
 900                                 idx += ENTRIES_PER_PAGEPAGE;
 901                                 if (idx >= limit)
 902                                         goto lost1;
 903                         }
 904                         stage = idx + ENTRIES_PER_PAGEPAGE;
 905                         subdir = *dir;
 906                         shmem_dir_unmap(dir);
 907                         dir = shmem_dir_map(subdir);
 908                 }
 909                 subdir = *dir;
 910                 if (subdir && page_private(subdir)) {
 911                         ptr = shmem_swp_map(subdir);
 912                         size = limit - idx;
 913                         if (size > ENTRIES_PER_PAGE)
 914                                 size = ENTRIES_PER_PAGE;
 915                         offset = shmem_find_swp(entry, ptr, ptr+size);
 916                         shmem_swp_unmap(ptr);
 917                         if (offset >= 0) {
 918                                 shmem_dir_unmap(dir);
 919                                 goto found;
 920                         }
 921                 }
 922         }
 923 lost1:
 924         shmem_dir_unmap(dir-1);
 925 lost2:
 926         spin_unlock(&info->lock);
 927         return 0;
 928 found:
 929         idx += offset;
 930         inode = igrab(&info->vfs_inode);
 931         spin_unlock(&info->lock);
 932
 933         /*
 934          * Move _head_ to start search for next from here.
 935          * But be careful: shmem_delete_inode checks list_empty without taking
 936          * mutex, and there's an instant in list_move_tail when info->swaplist
 937          * would appear empty, if it were the only one on shmem_swaplist.  We
 938          * could avoid doing it if inode NULL; or use this minor optimization.
 939          */
 940         if (shmem_swaplist.next != &info->swaplist)
 941                 list_move_tail(&shmem_swaplist, &info->swaplist);
 942         mutex_unlock(&shmem_swaplist_mutex);
 943
 944         error = 1;
 945         if (!inode)
 946                 goto out;
 947         /*
 948          * Charge page using GFP_KERNEL while we can wait.
 949          * Charged back to the user(not to caller) when swap account is used.
 950          * add_to_page_cache() will be called with GFP_NOWAIT.
 951          */
 952         error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
 953         if (error)
 954                 goto out;
 955         error = radix_tree_preload(GFP_KERNEL);
 956         if (error) {
 957                 mem_cgroup_uncharge_cache_page(page);
 958                 goto out;
 959         }
 960         error = 1;
 961
 962         spin_lock(&info->lock);
 963         ptr = shmem_swp_entry(info, idx, NULL);
 964         if (ptr && ptr->val == entry.val) {
 965                 error = add_to_page_cache_locked(page, inode->i_mapping,
 966                                                 idx, GFP_NOWAIT);
 967                 /* does mem_cgroup_uncharge_cache_page on error */
 968         } else  /* we must compensate for our precharge above */
 969                 mem_cgroup_uncharge_cache_page(page);
 970
 971         if (error == -EEXIST) {
 972                 struct page *filepage = find_get_page(inode->i_mapping, idx);
 973                 error = 1;
 974                 if (filepage) {
 975                         /*
 976                          * There might be a more uptodate page coming down
 977                          * from a stacked writepage: forget our swappage if so.
 978                          */
 979                         if (PageUptodate(filepage))
 980                                 error = 0;
 981                         page_cache_release(filepage);
 982                 }
 983         }
 984         if (!error) {
 985                 delete_from_swap_cache(page);
 986                 set_page_dirty(page);
 987                 info->flags |= SHMEM_PAGEIN;
 988                 shmem_swp_set(info, ptr, 0);
 989                 swap_free(entry);
 990                 error = 1;      /* not an error, but entry was found */
 991         }
 992         if (ptr)
 993                 shmem_swp_unmap(ptr);
 994         spin_unlock(&info->lock);
 995         radix_tree_preload_end();
 996 out:
 997         unlock_page(page);
 998         page_cache_release(page);
 999         iput(inode);            /* allows for NULL */
1000         return error;
1001 }
1002
1003 /*
1004  * shmem_unuse() search for an eventually swapped out shmem page.
1005  */
1006 int shmem_unuse(swp_entry_t entry, struct page *page)
1007 {
1008         struct list_head *p, *next;
1009         struct shmem_inode_info *info;
1010         int found = 0;
1011
1012         mutex_lock(&shmem_swaplist_mutex);
1013         list_for_each_safe(p, next, &shmem_swaplist) {
1014                 info = list_entry(p, struct shmem_inode_info, swaplist);
1015                 found = shmem_unuse_inode(info, entry, page);
1016                 cond_resched();
1017                 if (found)
1018                         goto out;
1019         }
1020         mutex_unlock(&shmem_swaplist_mutex);
1021 out:    return found;   /* 0 or 1 or -ENOMEM */
1022 }
1023
1024 /*
1025  * Move the page from the page cache to the swap cache.
1026  */
1027 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1028 {
1029         struct shmem_inode_info *info;
1030         swp_entry_t *entry, swap;
1031         struct address_space *mapping;
1032         unsigned long index;
1033         struct inode *inode;
1034
1035         BUG_ON(!PageLocked(page));
1036         mapping = page->mapping;
1037         index = page->index;
1038         inode = mapping->host;
1039         info = SHMEM_I(inode);
1040         if (info->flags & VM_LOCKED)
1041                 goto redirty;
1042         if (!total_swap_pages)
1043                 goto redirty;
1044
1045         /*
1046          * shmem_backing_dev_info's capabilities prevent regular writeback or
1047          * sync from ever calling shmem_writepage; but a stacking filesystem
1048          * may use the ->writepage of its underlying filesystem, in which case
1049          * tmpfs should write out to swap only in response to memory pressure,
1050          * and not for pdflush or sync.  However, in those cases, we do still
1051          * want to check if there's a redundant swappage to be discarded.
1052          */
1053         if (wbc->for_reclaim)
1054                 swap = get_swap_page();
1055         else
1056                 swap.val = 0;
1057
1058         spin_lock(&info->lock);
1059         if (index >= info->next_index) {
1060                 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
1061                 goto unlock;
1062         }
1063         entry = shmem_swp_entry(info, index, NULL);
1064         if (entry->val) {
1065                 /*
1066                  * The more uptodate page coming down from a stacked
1067                  * writepage should replace our old swappage.
1068                  */
1069                 free_swap_and_cache(*entry);
1070                 shmem_swp_set(info, entry, 0);
1071         }
1072         shmem_recalc_inode(inode);
1073
1074         if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
1075                 remove_from_page_cache(page);
1076                 shmem_swp_set(info, entry, swap.val);
1077                 shmem_swp_unmap(entry);
1078                 if (list_empty(&info->swaplist))
1079                         inode = igrab(inode);
1080                 else
1081                         inode = NULL;
1082                 spin_unlock(&info->lock);
1083                 swap_duplicate(swap);
1084                 BUG_ON(page_mapped(page));
1085                 page_cache_release(page);       /* pagecache ref */
1086                 swap_writepage(page, wbc);
1087                 if (inode) {
1088                         mutex_lock(&shmem_swaplist_mutex);
1089                         /* move instead of add in case we're racing */
1090                         list_move_tail(&info->swaplist, &shmem_swaplist);
1091                         mutex_unlock(&shmem_swaplist_mutex);
1092                         iput(inode);
1093                 }
1094                 return 0;
1095         }
1096
1097         shmem_swp_unmap(entry);
1098 unlock:
1099         spin_unlock(&info->lock);
1100         swap_free(swap);
1101 redirty:
1102         set_page_dirty(page);
1103         if (wbc->for_reclaim)
1104                 return AOP_WRITEPAGE_ACTIVATE;  /* Return with page locked */
1105         unlock_page(page);
1106         return 0;
1107 }
1108
1109 #ifdef CONFIG_NUMA
1110 #ifdef CONFIG_TMPFS
1111 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1112 {
1113         char buffer[64];
1114
1115         if (!mpol || mpol->mode == MPOL_DEFAULT)
1116                 return;         /* show nothing */
1117
1118         mpol_to_str(buffer, sizeof(buffer), mpol, 1);
1119
1120         seq_printf(seq, ",mpol=%s", buffer);
1121 }
1122
1123 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1124 {
1125         struct mempolicy *mpol = NULL;
1126         if (sbinfo->mpol) {
1127                 spin_lock(&sbinfo->stat_lock);  /* prevent replace/use races */
1128                 mpol = sbinfo->mpol;
1129                 mpol_get(mpol);
1130                 spin_unlock(&sbinfo->stat_lock);
1131         }
1132         return mpol;
1133 }
1134 #endif /* CONFIG_TMPFS */
1135
1136 static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1137                         struct shmem_inode_info *info, unsigned long idx)
1138 {
1139         struct mempolicy mpol, *spol;
1140         struct vm_area_struct pvma;
1141         struct page *page;
1142
1143         spol = mpol_cond_copy(&mpol,
1144                                 mpol_shared_policy_lookup(&info->policy, idx));
1145
1146         /* Create a pseudo vma that just contains the policy */
1147         pvma.vm_start = 0;
1148         pvma.vm_pgoff = idx;
1149         pvma.vm_ops = NULL;
1150         pvma.vm_policy = spol;
1151         page = swapin_readahead(entry, gfp, &pvma, 0);
1152         return page;
1153 }
1154
1155 static struct page *shmem_alloc_page(gfp_t gfp,
1156                         struct shmem_inode_info *info, unsigned long idx)
1157 {
1158         struct vm_area_struct pvma;
1159
1160         /* Create a pseudo vma that just contains the policy */
1161         pvma.vm_start = 0;
1162         pvma.vm_pgoff = idx;
1163         pvma.vm_ops = NULL;
1164         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1165
1166         /*
1167          * alloc_page_vma() will drop the shared policy reference
1168          */
1169         return alloc_page_vma(gfp, &pvma, 0);
1170 }
1171 #else /* !CONFIG_NUMA */
1172 #ifdef CONFIG_TMPFS
1173 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p)
1174 {
1175 }
1176 #endif /* CONFIG_TMPFS */
1177
1178 static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1179                         struct shmem_inode_info *info, unsigned long idx)
1180 {
1181         return swapin_readahead(entry, gfp, NULL, 0);
1182 }
1183
1184 static inline struct page *shmem_alloc_page(gfp_t gfp,
1185                         struct shmem_inode_info *info, unsigned long idx)
1186 {
1187         return alloc_page(gfp);
1188 }
1189 #endif /* CONFIG_NUMA */
1190
1191 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
1192 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1193 {
1194         return NULL;
1195 }
1196 #endif
1197
1198 /*
1199  * shmem_getpage - either get the page from swap or allocate a new one
1200  *
1201  * If we allocate a new one we do not mark it dirty. That's up to the
1202  * vm. If we swap it in we mark it dirty since we also free the swap
1203  * entry since a page cannot live in both the swap and page cache
1204  */
1205 static int shmem_getpage(struct inode *inode, unsigned long idx,
1206                         struct page **pagep, enum sgp_type sgp, int *type)
1207 {
1208         struct address_space *mapping = inode->i_mapping;
1209         struct shmem_inode_info *info = SHMEM_I(inode);
1210         struct shmem_sb_info *sbinfo;
1211         struct page *filepage = *pagep;
1212         struct page *swappage;
1213         swp_entry_t *entry;
1214         swp_entry_t swap;
1215         gfp_t gfp;
1216         int error;
1217
1218         if (idx >= SHMEM_MAX_INDEX)
1219                 return -EFBIG;
1220
1221         if (type)
1222                 *type = 0;
1223
1224         /*
1225          * Normally, filepage is NULL on entry, and either found
1226          * uptodate immediately, or allocated and zeroed, or read
1227          * in under swappage, which is then assigned to filepage.
1228          * But shmem_readpage (required for splice) passes in a locked
1229          * filepage, which may be found not uptodate by other callers
1230          * too, and may need to be copied from the swappage read in.
1231          */
1232 repeat:
1233         if (!filepage)
1234                 filepage = find_lock_page(mapping, idx);
1235         if (filepage && PageUptodate(filepage))
1236                 goto done;
1237         error = 0;
1238         gfp = mapping_gfp_mask(mapping);
1239         if (!filepage) {
1240                 /*
1241                  * Try to preload while we can wait, to not make a habit of
1242                  * draining atomic reserves; but don't latch on to this cpu.
1243                  */
1244                 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM);
1245                 if (error)
1246                         goto failed;
1247                 radix_tree_preload_end();
1248         }
1249
1250         spin_lock(&info->lock);
1251         shmem_recalc_inode(inode);
1252         entry = shmem_swp_alloc(info, idx, sgp);
1253         if (IS_ERR(entry)) {
1254                 spin_unlock(&info->lock);
1255                 error = PTR_ERR(entry);
1256                 goto failed;
1257         }
1258         swap = *entry;
1259
1260         if (swap.val) {
1261                 /* Look it up and read it in.. */
1262                 swappage = lookup_swap_cache(swap);
1263                 if (!swappage) {
1264                         shmem_swp_unmap(entry);
1265                         /* here we actually do the io */
1266                         if (type && !(*type & VM_FAULT_MAJOR)) {
1267                                 __count_vm_event(PGMAJFAULT);
1268                                 *type |= VM_FAULT_MAJOR;
1269                         }
1270                         spin_unlock(&info->lock);
1271                         swappage = shmem_swapin(swap, gfp, info, idx);
1272                         if (!swappage) {
1273                                 spin_lock(&info->lock);
1274                                 entry = shmem_swp_alloc(info, idx, sgp);
1275                                 if (IS_ERR(entry))
1276                                         error = PTR_ERR(entry);
1277                                 else {
1278                                         if (entry->val == swap.val)
1279                                                 error = -ENOMEM;
1280                                         shmem_swp_unmap(entry);
1281                                 }
1282                                 spin_unlock(&info->lock);
1283                                 if (error)
1284                                         goto failed;
1285                                 goto repeat;
1286                         }
1287                         wait_on_page_locked(swappage);
1288                         page_cache_release(swappage);
1289                         goto repeat;
1290                 }
1291
1292                 /* We have to do this with page locked to prevent races */
1293                 if (!trylock_page(swappage)) {
1294                         shmem_swp_unmap(entry);
1295                         spin_unlock(&info->lock);
1296                         wait_on_page_locked(swappage);
1297                         page_cache_release(swappage);
1298                         goto repeat;
1299                 }
1300                 if (PageWriteback(swappage)) {
1301                         shmem_swp_unmap(entry);
1302                         spin_unlock(&info->lock);
1303                         wait_on_page_writeback(swappage);
1304                         unlock_page(swappage);
1305                         page_cache_release(swappage);
1306                         goto repeat;
1307                 }
1308                 if (!PageUptodate(swappage)) {
1309                         shmem_swp_unmap(entry);
1310                         spin_unlock(&info->lock);
1311                         unlock_page(swappage);
1312                         page_cache_release(swappage);
1313                         error = -EIO;
1314                         goto failed;
1315                 }
1316
1317                 if (filepage) {
1318                         shmem_swp_set(info, entry, 0);
1319                         shmem_swp_unmap(entry);
1320                         delete_from_swap_cache(swappage);
1321                         spin_unlock(&info->lock);
1322                         copy_highpage(filepage, swappage);
1323                         unlock_page(swappage);
1324                         page_cache_release(swappage);
1325                         flush_dcache_page(filepage);
1326                         SetPageUptodate(filepage);
1327                         set_page_dirty(filepage);
1328                         swap_free(swap);
1329                 } else if (!(error = add_to_page_cache_locked(swappage, mapping,
1330                                         idx, GFP_NOWAIT))) {
1331                         info->flags |= SHMEM_PAGEIN;
1332                         shmem_swp_set(info, entry, 0);
1333                         shmem_swp_unmap(entry);
1334                         delete_from_swap_cache(swappage);
1335                         spin_unlock(&info->lock);
1336                         filepage = swappage;
1337                         set_page_dirty(filepage);
1338                         swap_free(swap);
1339                 } else {
1340                         shmem_swp_unmap(entry);
1341                         spin_unlock(&info->lock);
1342                         if (error == -ENOMEM) {
1343                                 /*
1344                                  * reclaim from proper memory cgroup and
1345                                  * call memcg's OOM if needed.
1346                                  */
1347                                 error = mem_cgroup_shmem_charge_fallback(
1348                                                                 swappage,
1349                                                                 current->mm,
1350                                                                 gfp);
1351                                 if (error) {
1352                                         unlock_page(swappage);
1353                                         page_cache_release(swappage);
1354                                         goto failed;
1355                                 }
1356                         }
1357                         unlock_page(swappage);
1358                         page_cache_release(swappage);
1359                         goto repeat;
1360                 }
1361         } else if (sgp == SGP_READ && !filepage) {
1362                 shmem_swp_unmap(entry);
1363                 filepage = find_get_page(mapping, idx);
1364                 if (filepage &&
1365                     (!PageUptodate(filepage) || !trylock_page(filepage))) {
1366                         spin_unlock(&info->lock);
1367                         wait_on_page_locked(filepage);
1368                         page_cache_release(filepage);
1369                         filepage = NULL;
1370                         goto repeat;
1371                 }
1372                 spin_unlock(&info->lock);
1373         } else {
1374                 shmem_swp_unmap(entry);
1375                 sbinfo = SHMEM_SB(inode->i_sb);
1376                 if (sbinfo->max_blocks) {
1377                         spin_lock(&sbinfo->stat_lock);
1378                         if (sbinfo->free_blocks == 0 ||
1379                             shmem_acct_block(info->flags)) {
1380                                 spin_unlock(&sbinfo->stat_lock);
1381                                 spin_unlock(&info->lock);
1382                                 error = -ENOSPC;
1383                                 goto failed;
1384                         }
1385                         sbinfo->free_blocks--;
1386                         inode->i_blocks += BLOCKS_PER_PAGE;
1387                         spin_unlock(&sbinfo->stat_lock);
1388                 } else if (shmem_acct_block(info->flags)) {
1389                         spin_unlock(&info->lock);
1390                         error = -ENOSPC;
1391                         goto failed;
1392                 }
1393
1394                 if (!filepage) {
1395                         int ret;
1396
1397                         spin_unlock(&info->lock);
1398                         filepage = shmem_alloc_page(gfp, info, idx);
1399                         if (!filepage) {
1400                                 shmem_unacct_blocks(info->flags, 1);
1401                                 shmem_free_blocks(inode, 1);
1402                                 error = -ENOMEM;
1403                                 goto failed;
1404                         }
1405                         SetPageSwapBacked(filepage);
1406
1407                         /* Precharge page while we can wait, compensate after */
1408                         error = mem_cgroup_cache_charge(filepage, current->mm,
1409                                         GFP_KERNEL);
1410                         if (error) {
1411                                 page_cache_release(filepage);
1412                                 shmem_unacct_blocks(info->flags, 1);
1413                                 shmem_free_blocks(inode, 1);
1414                                 filepage = NULL;
1415                                 goto failed;
1416                         }
1417
1418                         spin_lock(&info->lock);
1419                         entry = shmem_swp_alloc(info, idx, sgp);
1420                         if (IS_ERR(entry))
1421                                 error = PTR_ERR(entry);
1422                         else {
1423                                 swap = *entry;
1424                                 shmem_swp_unmap(entry);
1425                         }
1426                         ret = error || swap.val;
1427                         if (ret)
1428                                 mem_cgroup_uncharge_cache_page(filepage);
1429                         else
1430                                 ret = add_to_page_cache_lru(filepage, mapping,
1431                                                 idx, GFP_NOWAIT);
1432                         /*
1433                          * At add_to_page_cache_lru() failure, uncharge will
1434                          * be done automatically.
1435                          */
1436                         if (ret) {
1437                                 spin_unlock(&info->lock);
1438                                 page_cache_release(filepage);
1439                                 shmem_unacct_blocks(info->flags, 1);
1440                                 shmem_free_blocks(inode, 1);
1441                                 filepage = NULL;
1442                                 if (error)
1443                                         goto failed;
1444                                 goto repeat;
1445                         }
1446                         info->flags |= SHMEM_PAGEIN;
1447                 }
1448
1449                 info->alloced++;
1450                 spin_unlock(&info->lock);
1451                 clear_highpage(filepage);
1452                 flush_dcache_page(filepage);
1453                 SetPageUptodate(filepage);
1454                 if (sgp == SGP_DIRTY)
1455                         set_page_dirty(filepage);
1456         }
1457 done:
1458         *pagep = filepage;
1459         return 0;
1460
1461 failed:
1462         if (*pagep != filepage) {
1463                 unlock_page(filepage);
1464                 page_cache_release(filepage);
1465         }
1466         return error;
1467 }
1468
1469 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1470 {
1471         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1472         int error;
1473         int ret;
1474
1475         if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1476                 return VM_FAULT_SIGBUS;
1477
1478         error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1479         if (error)
1480                 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1481
1482         return ret | VM_FAULT_LOCKED;
1483 }
1484
1485 #ifdef CONFIG_NUMA
1486 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1487 {
1488         struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1489         return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1490 }
1491
1492 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1493                                           unsigned long addr)
1494 {
1495         struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1496         unsigned long idx;
1497
1498         idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1499         return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1500 }
1501 #endif
1502
1503 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1504 {
1505         struct inode *inode = file->f_path.dentry->d_inode;
1506         struct shmem_inode_info *info = SHMEM_I(inode);
1507         int retval = -ENOMEM;
1508
1509         spin_lock(&info->lock);
1510         if (lock && !(info->flags & VM_LOCKED)) {
1511                 if (!user_shm_lock(inode->i_size, user))
1512                         goto out_nomem;
1513                 info->flags |= VM_LOCKED;
1514                 mapping_set_unevictable(file->f_mapping);
1515         }
1516         if (!lock && (info->flags & VM_LOCKED) && user) {
1517                 user_shm_unlock(inode->i_size, user);
1518                 info->flags &= ~VM_LOCKED;
1519                 mapping_clear_unevictable(file->f_mapping);
1520                 scan_mapping_unevictable_pages(file->f_mapping);
1521         }
1522         retval = 0;
1523
1524 out_nomem:
1525         spin_unlock(&info->lock);
1526         return retval;
1527 }
1528
1529 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1530 {
1531         file_accessed(file);
1532         vma->vm_ops = &shmem_vm_ops;
1533         vma->vm_flags |= VM_CAN_NONLINEAR;
1534         return 0;
1535 }
1536
1537 static struct inode *shmem_get_inode(struct super_block *sb, int mode,
1538                                         dev_t dev, unsigned long flags)
1539 {
1540         struct inode *inode;
1541         struct shmem_inode_info *info;
1542         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1543
1544         if (shmem_reserve_inode(sb))
1545                 return NULL;
1546
1547         inode = new_inode(sb);
1548         if (inode) {
1549                 inode->i_mode = mode;
1550                 inode->i_uid = current_fsuid();
1551                 inode->i_gid = current_fsgid();
1552                 inode->i_blocks = 0;
1553                 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1554                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1555                 inode->i_generation = get_seconds();
1556                 info = SHMEM_I(inode);
1557                 memset(info, 0, (char *)inode - (char *)info);
1558                 spin_lock_init(&info->lock);
1559                 info->flags = flags & VM_NORESERVE;
1560                 INIT_LIST_HEAD(&info->swaplist);
1561
1562                 switch (mode & S_IFMT) {
1563                 default:
1564                         inode->i_op = &shmem_special_inode_operations;
1565                         init_special_inode(inode, mode, dev);
1566                         break;
1567                 case S_IFREG:
1568                         inode->i_mapping->a_ops = &shmem_aops;
1569                         inode->i_op = &shmem_inode_operations;
1570                         inode->i_fop = &shmem_file_operations;
1571                         mpol_shared_policy_init(&info->policy,
1572                                                  shmem_get_sbmpol(sbinfo));
1573                         break;
1574                 case S_IFDIR:
1575                         inc_nlink(inode);
1576                         /* Some things misbehave if size == 0 on a directory */
1577                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
1578                         inode->i_op = &shmem_dir_inode_operations;
1579                         inode->i_fop = &simple_dir_operations;
1580                         break;
1581                 case S_IFLNK:
1582                         /*
1583                          * Must not load anything in the rbtree,
1584                          * mpol_free_shared_policy will not be called.
1585                          */
1586                         mpol_shared_policy_init(&info->policy, NULL);
1587                         break;
1588                 }
1589         } else
1590                 shmem_free_inode(sb);
1591         return inode;
1592 }
1593
1594 #ifdef CONFIG_TMPFS
1595 static const struct inode_operations shmem_symlink_inode_operations;
1596 static const struct inode_operations shmem_symlink_inline_operations;
1597
1598 /*
1599  * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
1600  * but providing them allows a tmpfs file to be used for splice, sendfile, and
1601  * below the loop driver, in the generic fashion that many filesystems support.
1602  */
1603 static int shmem_readpage(struct file *file, struct page *page)
1604 {
1605         struct inode *inode = page->mapping->host;
1606         int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1607         unlock_page(page);
1608         return error;
1609 }
1610
1611 static int
1612 shmem_write_begin(struct file *file, struct address_space *mapping,
1613                         loff_t pos, unsigned len, unsigned flags,
1614                         struct page **pagep, void **fsdata)
1615 {
1616         struct inode *inode = mapping->host;
1617         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1618         *pagep = NULL;
1619         return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1620 }
1621
1622 static int
1623 shmem_write_end(struct file *file, struct address_space *mapping,
1624                         loff_t pos, unsigned len, unsigned copied,
1625                         struct page *page, void *fsdata)
1626 {
1627         struct inode *inode = mapping->host;
1628
1629         if (pos + copied > inode->i_size)
1630                 i_size_write(inode, pos + copied);
1631
1632         unlock_page(page);
1633         set_page_dirty(page);
1634         page_cache_release(page);
1635
1636         return copied;
1637 }
1638
1639 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1640 {
1641         struct inode *inode = filp->f_path.dentry->d_inode;
1642         struct address_space *mapping = inode->i_mapping;
1643         unsigned long index, offset;
1644         enum sgp_type sgp = SGP_READ;
1645
1646         /*
1647          * Might this read be for a stacking filesystem?  Then when reading
1648          * holes of a sparse file, we actually need to allocate those pages,
1649          * and even mark them dirty, so it cannot exceed the max_blocks limit.
1650          */
1651         if (segment_eq(get_fs(), KERNEL_DS))
1652                 sgp = SGP_DIRTY;
1653
1654         index = *ppos >> PAGE_CACHE_SHIFT;
1655         offset = *ppos & ~PAGE_CACHE_MASK;
1656
1657         for (;;) {
1658                 struct page *page = NULL;
1659                 unsigned long end_index, nr, ret;
1660                 loff_t i_size = i_size_read(inode);
1661
1662                 end_index = i_size >> PAGE_CACHE_SHIFT;
1663                 if (index > end_index)
1664                         break;
1665                 if (index == end_index) {
1666                         nr = i_size & ~PAGE_CACHE_MASK;
1667                         if (nr <= offset)
1668                                 break;
1669                 }
1670
1671                 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1672                 if (desc->error) {
1673                         if (desc->error == -EINVAL)
1674                                 desc->error = 0;
1675                         break;
1676                 }
1677                 if (page)
1678                         unlock_page(page);
1679
1680                 /*
1681                  * We must evaluate after, since reads (unlike writes)
1682                  * are called without i_mutex protection against truncate
1683                  */
1684                 nr = PAGE_CACHE_SIZE;
1685                 i_size = i_size_read(inode);
1686                 end_index = i_size >> PAGE_CACHE_SHIFT;
1687                 if (index == end_index) {
1688                         nr = i_size & ~PAGE_CACHE_MASK;
1689                         if (nr <= offset) {
1690                                 if (page)
1691                                         page_cache_release(page);
1692                                 break;
1693                         }
1694                 }
1695                 nr -= offset;
1696
1697                 if (page) {
1698                         /*
1699                          * If users can be writing to this page using arbitrary
1700                          * virtual addresses, take care about potential aliasing
1701                          * before reading the page on the kernel side.
1702                          */
1703                         if (mapping_writably_mapped(mapping))
1704                                 flush_dcache_page(page);
1705                         /*
1706                          * Mark the page accessed if we read the beginning.
1707                          */
1708                         if (!offset)
1709                                 mark_page_accessed(page);
1710                 } else {
1711                         page = ZERO_PAGE(0);
1712                         page_cache_get(page);
1713                 }
1714
1715                 /*
1716                  * Ok, we have the page, and it's up-to-date, so
1717                  * now we can copy it to user space...
1718                  *
1719                  * The actor routine returns how many bytes were actually used..
1720                  * NOTE! This may not be the same as how much of a user buffer
1721                  * we filled up (we may be padding etc), so we can only update
1722                  * "pos" here (the actor routine has to update the user buffer
1723                  * pointers and the remaining count).
1724                  */
1725                 ret = actor(desc, page, offset, nr);
1726                 offset += ret;
1727                 index += offset >> PAGE_CACHE_SHIFT;
1728                 offset &= ~PAGE_CACHE_MASK;
1729
1730                 page_cache_release(page);
1731                 if (ret != nr || !desc->count)
1732                         break;
1733
1734                 cond_resched();
1735         }
1736
1737         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1738         file_accessed(filp);
1739 }
1740
1741 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1742                 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1743 {
1744         struct file *filp = iocb->ki_filp;
1745         ssize_t retval;
1746         unsigned long seg;
1747         size_t count;
1748         loff_t *ppos = &iocb->ki_pos;
1749
1750         retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1751         if (retval)
1752                 return retval;
1753
1754         for (seg = 0; seg < nr_segs; seg++) {
1755                 read_descriptor_t desc;
1756
1757                 desc.written = 0;
1758                 desc.arg.buf = iov[seg].iov_base;
1759                 desc.count = iov[seg].iov_len;
1760                 if (desc.count == 0)
1761                         continue;
1762                 desc.error = 0;
1763                 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1764                 retval += desc.written;
1765                 if (desc.error) {
1766                         retval = retval ?: desc.error;
1767                         break;
1768                 }
1769                 if (desc.count > 0)
1770                         break;
1771         }
1772         return retval;
1773 }
1774
1775 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1776 {
1777         struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1778
1779         buf->f_type = TMPFS_MAGIC;
1780         buf->f_bsize = PAGE_CACHE_SIZE;
1781         buf->f_namelen = NAME_MAX;
1782         spin_lock(&sbinfo->stat_lock);
1783         if (sbinfo->max_blocks) {
1784                 buf->f_blocks = sbinfo->max_blocks;
1785                 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1786         }
1787         if (sbinfo->max_inodes) {
1788                 buf->f_files = sbinfo->max_inodes;
1789                 buf->f_ffree = sbinfo->free_inodes;
1790         }
1791         /* else leave those fields 0 like simple_statfs */
1792         spin_unlock(&sbinfo->stat_lock);
1793         return 0;
1794 }
1795
1796 /*
1797  * File creation. Allocate an inode, and we're done..
1798  */
1799 static int
1800 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1801 {
1802         struct inode *inode;
1803         int error = -ENOSPC;
1804
1805         inode = shmem_get_inode(dir->i_sb, mode, dev, VM_NORESERVE);
1806         if (inode) {
1807                 error = security_inode_init_security(inode, dir, NULL, NULL,
1808                                                      NULL);
1809                 if (error) {
1810                         if (error != -EOPNOTSUPP) {
1811                                 iput(inode);
1812                                 return error;
1813                         }
1814                 }
1815                 error = shmem_acl_init(inode, dir);
1816                 if (error) {
1817                         iput(inode);
1818                         return error;
1819                 }
1820                 if (dir->i_mode & S_ISGID) {
1821                         inode->i_gid = dir->i_gid;
1822                         if (S_ISDIR(mode))
1823                                 inode->i_mode |= S_ISGID;
1824                 }
1825                 dir->i_size += BOGO_DIRENT_SIZE;
1826                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1827                 d_instantiate(dentry, inode);
1828                 dget(dentry); /* Extra count - pin the dentry in core */
1829         }
1830         return error;
1831 }
1832
1833 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1834 {
1835         int error;
1836
1837         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1838                 return error;
1839         inc_nlink(dir);
1840         return 0;
1841 }
1842
1843 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1844                 struct nameidata *nd)
1845 {
1846         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1847 }
1848
1849 /*
1850  * Link a file..
1851  */
1852 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1853 {
1854         struct inode *inode = old_dentry->d_inode;
1855         int ret;
1856
1857         /*
1858          * No ordinary (disk based) filesystem counts links as inodes;
1859          * but each new link needs a new dentry, pinning lowmem, and
1860          * tmpfs dentries cannot be pruned until they are unlinked.
1861          */
1862         ret = shmem_reserve_inode(inode->i_sb);
1863         if (ret)
1864                 goto out;
1865
1866         dir->i_size += BOGO_DIRENT_SIZE;
1867         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1868         inc_nlink(inode);
1869         atomic_inc(&inode->i_count);    /* New dentry reference */
1870         dget(dentry);           /* Extra pinning count for the created dentry */
1871         d_instantiate(dentry, inode);
1872 out:
1873         return ret;
1874 }
1875
1876 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1877 {
1878         struct inode *inode = dentry->d_inode;
1879
1880         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1881                 shmem_free_inode(inode->i_sb);
1882
1883         dir->i_size -= BOGO_DIRENT_SIZE;
1884         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1885         drop_nlink(inode);
1886         dput(dentry);   /* Undo the count from "create" - this does all the work */
1887         return 0;
1888 }
1889
1890 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1891 {
1892         if (!simple_empty(dentry))
1893                 return -ENOTEMPTY;
1894
1895         drop_nlink(dentry->d_inode);
1896         drop_nlink(dir);
1897         return shmem_unlink(dir, dentry);
1898 }
1899
1900 /*
1901  * The VFS layer already does all the dentry stuff for rename,
1902  * we just have to decrement the usage count for the target if
1903  * it exists so that the VFS layer correctly free's it when it
1904  * gets overwritten.
1905  */
1906 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1907 {
1908         struct inode *inode = old_dentry->d_inode;
1909         int they_are_dirs = S_ISDIR(inode->i_mode);
1910
1911         if (!simple_empty(new_dentry))
1912                 return -ENOTEMPTY;
1913
1914         if (new_dentry->d_inode) {
1915                 (void) shmem_unlink(new_dir, new_dentry);
1916                 if (they_are_dirs)
1917                         drop_nlink(old_dir);
1918         } else if (they_are_dirs) {
1919                 drop_nlink(old_dir);
1920                 inc_nlink(new_dir);
1921         }
1922
1923         old_dir->i_size -= BOGO_DIRENT_SIZE;
1924         new_dir->i_size += BOGO_DIRENT_SIZE;
1925         old_dir->i_ctime = old_dir->i_mtime =
1926         new_dir->i_ctime = new_dir->i_mtime =
1927         inode->i_ctime = CURRENT_TIME;
1928         return 0;
1929 }
1930
1931 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1932 {
1933         int error;
1934         int len;
1935         struct inode *inode;
1936         struct page *page = NULL;
1937         char *kaddr;
1938         struct shmem_inode_info *info;
1939
1940         len = strlen(symname) + 1;
1941         if (len > PAGE_CACHE_SIZE)
1942                 return -ENAMETOOLONG;
1943
1944         inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
1945         if (!inode)
1946                 return -ENOSPC;
1947
1948         error = security_inode_init_security(inode, dir, NULL, NULL,
1949                                              NULL);
1950         if (error) {
1951                 if (error != -EOPNOTSUPP) {
1952                         iput(inode);
1953                         return error;
1954                 }
1955                 error = 0;
1956         }
1957
1958         info = SHMEM_I(inode);
1959         inode->i_size = len-1;
1960         if (len <= (char *)inode - (char *)info) {
1961                 /* do it inline */
1962                 memcpy(info, symname, len);
1963                 inode->i_op = &shmem_symlink_inline_operations;
1964         } else {
1965                 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1966                 if (error) {
1967                         iput(inode);
1968                         return error;
1969                 }
1970                 unlock_page(page);
1971                 inode->i_mapping->a_ops = &shmem_aops;
1972                 inode->i_op = &shmem_symlink_inode_operations;
1973                 kaddr = kmap_atomic(page, KM_USER0);
1974                 memcpy(kaddr, symname, len);
1975                 kunmap_atomic(kaddr, KM_USER0);
1976                 set_page_dirty(page);
1977                 page_cache_release(page);
1978         }
1979         if (dir->i_mode & S_ISGID)
1980                 inode->i_gid = dir->i_gid;
1981         dir->i_size += BOGO_DIRENT_SIZE;
1982         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1983         d_instantiate(dentry, inode);
1984         dget(dentry);
1985         return 0;
1986 }
1987
1988 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1989 {
1990         nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1991         return NULL;
1992 }
1993
1994 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1995 {
1996         struct page *page = NULL;
1997         int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1998         nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1999         if (page)
2000                 unlock_page(page);
2001         return page;
2002 }
2003
2004 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2005 {
2006         if (!IS_ERR(nd_get_link(nd))) {
2007                 struct page *page = cookie;
2008                 kunmap(page);
2009                 mark_page_accessed(page);
2010                 page_cache_release(page);
2011         }
2012 }
2013
2014 static const struct inode_operations shmem_symlink_inline_operations = {
2015         .readlink       = generic_readlink,
2016         .follow_link    = shmem_follow_link_inline,
2017 };
2018
2019 static const struct inode_operations shmem_symlink_inode_operations = {
2020         .truncate       = shmem_truncate,
2021         .readlink       = generic_readlink,
2022         .follow_link    = shmem_follow_link,
2023         .put_link       = shmem_put_link,
2024 };
2025
2026 #ifdef CONFIG_TMPFS_POSIX_ACL
2027 /*
2028  * Superblocks without xattr inode operations will get security.* xattr
2029  * support from the VFS "for free". As soon as we have any other xattrs
2030  * like ACLs, we also need to implement the security.* handlers at
2031  * filesystem level, though.
2032  */
2033
2034 static size_t shmem_xattr_security_list(struct inode *inode, char *list,
2035                                         size_t list_len, const char *name,
2036                                         size_t name_len)
2037 {
2038         return security_inode_listsecurity(inode, list, list_len);
2039 }
2040
2041 static int shmem_xattr_security_get(struct inode *inode, const char *name,
2042                                     void *buffer, size_t size)
2043 {
2044         if (strcmp(name, "") == 0)
2045                 return -EINVAL;
2046         return xattr_getsecurity(inode, name, buffer, size);
2047 }
2048
2049 static int shmem_xattr_security_set(struct inode *inode, const char *name,
2050                                     const void *value, size_t size, int flags)
2051 {
2052         if (strcmp(name, "") == 0)
2053                 return -EINVAL;
2054         return security_inode_setsecurity(inode, name, value, size, flags);
2055 }
2056
2057 static struct xattr_handler shmem_xattr_security_handler = {
2058         .prefix = XATTR_SECURITY_PREFIX,
2059         .list   = shmem_xattr_security_list,
2060         .get    = shmem_xattr_security_get,
2061         .set    = shmem_xattr_security_set,
2062 };
2063
2064 static struct xattr_handler *shmem_xattr_handlers[] = {
2065         &shmem_xattr_acl_access_handler,
2066         &shmem_xattr_acl_default_handler,
2067         &shmem_xattr_security_handler,
2068         NULL
2069 };
2070 #endif
2071
2072 static struct dentry *shmem_get_parent(struct dentry *child)
2073 {
2074         return ERR_PTR(-ESTALE);
2075 }
2076
2077 static int shmem_match(struct inode *ino, void *vfh)
2078 {
2079         __u32 *fh = vfh;
2080         __u64 inum = fh[2];
2081         inum = (inum << 32) | fh[1];
2082         return ino->i_ino == inum && fh[0] == ino->i_generation;
2083 }
2084
2085 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2086                 struct fid *fid, int fh_len, int fh_type)
2087 {
2088         struct inode *inode;
2089         struct dentry *dentry = NULL;
2090         u64 inum = fid->raw[2];
2091         inum = (inum << 32) | fid->raw[1];
2092
2093         if (fh_len < 3)
2094                 return NULL;
2095
2096         inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2097                         shmem_match, fid->raw);
2098         if (inode) {
2099                 dentry = d_find_alias(inode);
2100                 iput(inode);
2101         }
2102
2103         return dentry;
2104 }
2105
2106 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2107                                 int connectable)
2108 {
2109         struct inode *inode = dentry->d_inode;
2110
2111         if (*len < 3)
2112                 return 255;
2113
2114         if (hlist_unhashed(&inode->i_hash)) {
2115                 /* Unfortunately insert_inode_hash is not idempotent,
2116                  * so as we hash inodes here rather than at creation
2117                  * time, we need a lock to ensure we only try
2118                  * to do it once
2119                  */
2120                 static DEFINE_SPINLOCK(lock);
2121                 spin_lock(&lock);
2122                 if (hlist_unhashed(&inode->i_hash))
2123                         __insert_inode_hash(inode,
2124                                             inode->i_ino + inode->i_generation);
2125                 spin_unlock(&lock);
2126         }
2127
2128         fh[0] = inode->i_generation;
2129         fh[1] = inode->i_ino;
2130         fh[2] = ((__u64)inode->i_ino) >> 32;
2131
2132         *len = 3;
2133         return 1;
2134 }
2135
2136 static const struct export_operations shmem_export_ops = {
2137         .get_parent     = shmem_get_parent,
2138         .encode_fh      = shmem_encode_fh,
2139         .fh_to_dentry   = shmem_fh_to_dentry,
2140 };
2141
2142 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2143                                bool remount)
2144 {
2145         char *this_char, *value, *rest;
2146
2147         while (options != NULL) {
2148                 this_char = options;
2149                 for (;;) {
2150                         /*
2151                          * NUL-terminate this option: unfortunately,
2152                          * mount options form a comma-separated list,
2153                          * but mpol's nodelist may also contain commas.
2154                          */
2155                         options = strchr(options, ',');
2156                         if (options == NULL)
2157                                 break;
2158                         options++;
2159                         if (!isdigit(*options)) {
2160                                 options[-1] = '\0';
2161                                 break;
2162                         }
2163                 }
2164                 if (!*this_char)
2165                         continue;
2166                 if ((value = strchr(this_char,'=')) != NULL) {
2167                         *value++ = 0;
2168                 } else {
2169                         printk(KERN_ERR
2170                             "tmpfs: No value for mount option '%s'\n",
2171                             this_char);
2172                         return 1;
2173                 }
2174
2175                 if (!strcmp(this_char,"size")) {
2176                         unsigned long long size;
2177                         size = memparse(value,&rest);
2178                         if (*rest == '%') {
2179                                 size <<= PAGE_SHIFT;
2180                                 size *= totalram_pages;
2181                                 do_div(size, 100);
2182                                 rest++;
2183                         }
2184                         if (*rest)
2185                                 goto bad_val;
2186                         sbinfo->max_blocks =
2187                                 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2188                 } else if (!strcmp(this_char,"nr_blocks")) {
2189                         sbinfo->max_blocks = memparse(value, &rest);
2190                         if (*rest)
2191                                 goto bad_val;
2192                 } else if (!strcmp(this_char,"nr_inodes")) {
2193                         sbinfo->max_inodes = memparse(value, &rest);
2194                         if (*rest)
2195                                 goto bad_val;
2196                 } else if (!strcmp(this_char,"mode")) {
2197                         if (remount)
2198                                 continue;
2199                         sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2200                         if (*rest)
2201                                 goto bad_val;
2202                 } else if (!strcmp(this_char,"uid")) {
2203                         if (remount)
2204                                 continue;
2205                         sbinfo->uid = simple_strtoul(value, &rest, 0);
2206                         if (*rest)
2207                                 goto bad_val;
2208                 } else if (!strcmp(this_char,"gid")) {
2209                         if (remount)
2210                                 continue;
2211                         sbinfo->gid = simple_strtoul(value, &rest, 0);
2212                         if (*rest)
2213                                 goto bad_val;
2214                 } else if (!strcmp(this_char,"mpol")) {
2215                         if (mpol_parse_str(value, &sbinfo->mpol, 1))
2216                                 goto bad_val;
2217                 } else {
2218                         printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2219                                this_char);
2220                         return 1;
2221                 }
2222         }
2223         return 0;
2224
2225 bad_val:
2226         printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2227                value, this_char);
2228         return 1;
2229
2230 }
2231
2232 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2233 {
2234         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2235         struct shmem_sb_info config = *sbinfo;
2236         unsigned long blocks;
2237         unsigned long inodes;
2238         int error = -EINVAL;
2239
2240         if (shmem_parse_options(data, &config, true))
2241                 return error;
2242
2243         spin_lock(&sbinfo->stat_lock);
2244         blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2245         inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2246         if (config.max_blocks < blocks)
2247                 goto out;
2248         if (config.max_inodes < inodes)
2249                 goto out;
2250         /*
2251          * Those tests also disallow limited->unlimited while any are in
2252          * use, so i_blocks will always be zero when max_blocks is zero;
2253          * but we must separately disallow unlimited->limited, because
2254          * in that case we have no record of how much is already in use.
2255          */
2256         if (config.max_blocks && !sbinfo->max_blocks)
2257                 goto out;
2258         if (config.max_inodes && !sbinfo->max_inodes)
2259                 goto out;
2260
2261         error = 0;
2262         sbinfo->max_blocks  = config.max_blocks;
2263         sbinfo->free_blocks = config.max_blocks - blocks;
2264         sbinfo->max_inodes  = config.max_inodes;
2265         sbinfo->free_inodes = config.max_inodes - inodes;
2266
2267         mpol_put(sbinfo->mpol);
2268         sbinfo->mpol        = config.mpol;      /* transfers initial ref */
2269 out:
2270         spin_unlock(&sbinfo->stat_lock);
2271         return error;
2272 }
2273
2274 static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2275 {
2276         struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2277
2278         if (sbinfo->max_blocks != shmem_default_max_blocks())
2279                 seq_printf(seq, ",size=%luk",
2280                         sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2281         if (sbinfo->max_inodes != shmem_default_max_inodes())
2282                 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2283         if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2284                 seq_printf(seq, ",mode=%03o", sbinfo->mode);
2285         if (sbinfo->uid != 0)
2286                 seq_printf(seq, ",uid=%u", sbinfo->uid);
2287         if (sbinfo->gid != 0)
2288                 seq_printf(seq, ",gid=%u", sbinfo->gid);
2289         shmem_show_mpol(seq, sbinfo->mpol);
2290         return 0;
2291 }
2292 #endif /* CONFIG_TMPFS */
2293
2294 static void shmem_put_super(struct super_block *sb)
2295 {
2296         kfree(sb->s_fs_info);
2297         sb->s_fs_info = NULL;
2298 }
2299
2300 static int shmem_fill_super(struct super_block *sb,
2301                             void *data, int silent)
2302 {
2303         struct inode *inode;
2304         struct dentry *root;
2305         struct shmem_sb_info *sbinfo;
2306         int err = -ENOMEM;
2307
2308         /* Round up to L1_CACHE_BYTES to resist false sharing */
2309         sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
2310                                 L1_CACHE_BYTES), GFP_KERNEL);
2311         if (!sbinfo)
2312                 return -ENOMEM;
2313
2314         sbinfo->max_blocks = 0;
2315         sbinfo->max_inodes = 0;
2316         sbinfo->mode = S_IRWXUGO | S_ISVTX;
2317         sbinfo->uid = current_fsuid();
2318         sbinfo->gid = current_fsgid();
2319         sbinfo->mpol = NULL;
2320         sb->s_fs_info = sbinfo;
2321
2322 #ifdef CONFIG_TMPFS
2323         /*
2324          * Per default we only allow half of the physical ram per
2325          * tmpfs instance, limiting inodes to one per page of lowmem;
2326          * but the internal instance is left unlimited.
2327          */
2328         if (!(sb->s_flags & MS_NOUSER)) {
2329                 sbinfo->max_blocks = shmem_default_max_blocks();
2330                 sbinfo->max_inodes = shmem_default_max_inodes();
2331                 if (shmem_parse_options(data, sbinfo, false)) {
2332                         err = -EINVAL;
2333                         goto failed;
2334                 }
2335         }
2336         sb->s_export_op = &shmem_export_ops;
2337 #else
2338         sb->s_flags |= MS_NOUSER;
2339 #endif
2340
2341         spin_lock_init(&sbinfo->stat_lock);
2342         sbinfo->free_blocks = sbinfo->max_blocks;
2343         sbinfo->free_inodes = sbinfo->max_inodes;
2344
2345         sb->s_maxbytes = SHMEM_MAX_BYTES;
2346         sb->s_blocksize = PAGE_CACHE_SIZE;
2347         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2348         sb->s_magic = TMPFS_MAGIC;
2349         sb->s_op = &shmem_ops;
2350         sb->s_time_gran = 1;
2351 #ifdef CONFIG_TMPFS_POSIX_ACL
2352         sb->s_xattr = shmem_xattr_handlers;
2353         sb->s_flags |= MS_POSIXACL;
2354 #endif
2355
2356         inode = shmem_get_inode(sb, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2357         if (!inode)
2358                 goto failed;
2359         inode->i_uid = sbinfo->uid;
2360         inode->i_gid = sbinfo->gid;
2361         root = d_alloc_root(inode);
2362         if (!root)
2363                 goto failed_iput;
2364         sb->s_root = root;
2365         return 0;
2366
2367 failed_iput:
2368         iput(inode);
2369 failed:
2370         shmem_put_super(sb);
2371         return err;
2372 }
2373
2374 static struct kmem_cache *shmem_inode_cachep;
2375
2376 static struct inode *shmem_alloc_inode(struct super_block *sb)
2377 {
2378         struct shmem_inode_info *p;
2379         p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2380         if (!p)
2381                 return NULL;
2382         return &p->vfs_inode;
2383 }
2384
2385 static void shmem_destroy_inode(struct inode *inode)
2386 {
2387         if ((inode->i_mode & S_IFMT) == S_IFREG) {
2388                 /* only struct inode is valid if it's an inline symlink */
2389                 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2390         }
2391         shmem_acl_destroy_inode(inode);
2392         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2393 }
2394
2395 static void init_once(void *foo)
2396 {
2397         struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2398
2399         inode_init_once(&p->vfs_inode);
2400 #ifdef CONFIG_TMPFS_POSIX_ACL
2401         p->i_acl = NULL;
2402         p->i_default_acl = NULL;
2403 #endif
2404 }
2405
2406 static int init_inodecache(void)
2407 {
2408         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2409                                 sizeof(struct shmem_inode_info),
2410                                 0, SLAB_PANIC, init_once);
2411         return 0;
2412 }
2413
2414 static void destroy_inodecache(void)
2415 {
2416         kmem_cache_destroy(shmem_inode_cachep);
2417 }
2418
2419 static const struct address_space_operations shmem_aops = {
2420         .writepage      = shmem_writepage,
2421         .set_page_dirty = __set_page_dirty_no_writeback,
2422 #ifdef CONFIG_TMPFS
2423         .readpage       = shmem_readpage,
2424         .write_begin    = shmem_write_begin,
2425         .write_end      = shmem_write_end,
2426 #endif
2427         .migratepage    = migrate_page,
2428 };
2429
2430 static const struct file_operations shmem_file_operations = {
2431         .mmap           = shmem_mmap,
2432 #ifdef CONFIG_TMPFS
2433         .llseek         = generic_file_llseek,
2434         .read           = do_sync_read,
2435         .write          = do_sync_write,
2436         .aio_read       = shmem_file_aio_read,
2437         .aio_write      = generic_file_aio_write,
2438         .fsync          = simple_sync_file,
2439         .splice_read    = generic_file_splice_read,
2440         .splice_write   = generic_file_splice_write,
2441 #endif
2442 };
2443
2444 static const struct inode_operations shmem_inode_operations = {
2445         .truncate       = shmem_truncate,
2446         .setattr        = shmem_notify_change,
2447         .truncate_range = shmem_truncate_range,
2448 #ifdef CONFIG_TMPFS_POSIX_ACL
2449         .setxattr       = generic_setxattr,
2450         .getxattr       = generic_getxattr,
2451         .listxattr      = generic_listxattr,
2452         .removexattr    = generic_removexattr,
2453         .permission     = shmem_permission,
2454 #endif
2455
2456 };
2457
2458 static const struct inode_operations shmem_dir_inode_operations = {
2459 #ifdef CONFIG_TMPFS
2460         .create         = shmem_create,
2461         .lookup         = simple_lookup,
2462         .link           = shmem_link,
2463         .unlink         = shmem_unlink,
2464         .symlink        = shmem_symlink,
2465         .mkdir          = shmem_mkdir,
2466         .rmdir          = shmem_rmdir,
2467         .mknod          = shmem_mknod,
2468         .rename         = shmem_rename,
2469 #endif
2470 #ifdef CONFIG_TMPFS_POSIX_ACL
2471         .setattr        = shmem_notify_change,
2472         .setxattr       = generic_setxattr,
2473         .getxattr       = generic_getxattr,
2474         .listxattr      = generic_listxattr,
2475         .removexattr    = generic_removexattr,
2476         .permission     = shmem_permission,
2477 #endif
2478 };
2479
2480 static const struct inode_operations shmem_special_inode_operations = {
2481 #ifdef CONFIG_TMPFS_POSIX_ACL
2482         .setattr        = shmem_notify_change,
2483         .setxattr       = generic_setxattr,
2484         .getxattr       = generic_getxattr,
2485         .listxattr      = generic_listxattr,
2486         .removexattr    = generic_removexattr,
2487         .permission     = shmem_permission,
2488 #endif
2489 };
2490
2491 static const struct super_operations shmem_ops = {
2492         .alloc_inode    = shmem_alloc_inode,
2493         .destroy_inode  = shmem_destroy_inode,
2494 #ifdef CONFIG_TMPFS
2495         .statfs         = shmem_statfs,
2496         .remount_fs     = shmem_remount_fs,
2497         .show_options   = shmem_show_options,
2498 #endif
2499         .delete_inode   = shmem_delete_inode,
2500         .drop_inode     = generic_delete_inode,
2501         .put_super      = shmem_put_super,
2502 };
2503
2504 static struct vm_operations_struct shmem_vm_ops = {
2505         .fault          = shmem_fault,
2506 #ifdef CONFIG_NUMA
2507         .set_policy     = shmem_set_policy,
2508         .get_policy     = shmem_get_policy,
2509 #endif
2510 };
2511
2512
2513 static int shmem_get_sb(struct file_system_type *fs_type,
2514         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2515 {
2516         return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
2517 }
2518
2519 static struct file_system_type tmpfs_fs_type = {
2520         .owner          = THIS_MODULE,
2521         .name           = "tmpfs",
2522         .get_sb         = shmem_get_sb,
2523         .kill_sb        = kill_litter_super,
2524 };
2525
2526 static int __init init_tmpfs(void)
2527 {
2528         int error;
2529
2530         error = bdi_init(&shmem_backing_dev_info);
2531         if (error)
2532                 goto out4;
2533
2534         error = init_inodecache();
2535         if (error)
2536                 goto out3;
2537
2538         error = register_filesystem(&tmpfs_fs_type);
2539         if (error) {
2540                 printk(KERN_ERR "Could not register tmpfs\n");
2541                 goto out2;
2542         }
2543
2544         shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
2545                                 tmpfs_fs_type.name, NULL);
2546         if (IS_ERR(shm_mnt)) {
2547                 error = PTR_ERR(shm_mnt);
2548                 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2549                 goto out1;
2550         }
2551         return 0;
2552
2553 out1:
2554         unregister_filesystem(&tmpfs_fs_type);
2555 out2:
2556         destroy_inodecache();
2557 out3:
2558         bdi_destroy(&shmem_backing_dev_info);
2559 out4:
2560         shm_mnt = ERR_PTR(error);
2561         return error;
2562 }
2563
2564 #else /* !CONFIG_SHMEM */
2565
2566 /*
2567  * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2568  *
2569  * This is intended for small system where the benefits of the full
2570  * shmem code (swap-backed and resource-limited) are outweighed by
2571  * their complexity. On systems without swap this code should be
2572  * effectively equivalent, but much lighter weight.
2573  */
2574
2575 #include <linux/ramfs.h>
2576
2577 static struct file_system_type tmpfs_fs_type = {
2578         .name           = "tmpfs",
2579         .get_sb         = ramfs_get_sb,
2580         .kill_sb        = kill_litter_super,
2581 };
2582
2583 static int __init init_tmpfs(void)
2584 {
2585         BUG_ON(register_filesystem(&tmpfs_fs_type) != 0);
2586
2587         shm_mnt = kern_mount(&tmpfs_fs_type);
2588         BUG_ON(IS_ERR(shm_mnt));
2589
2590         return 0;
2591 }
2592
2593 int shmem_unuse(swp_entry_t entry, struct page *page)
2594 {
2595         return 0;
2596 }
2597
2598 #define shmem_vm_ops                            generic_file_vm_ops
2599 #define shmem_file_operations                   ramfs_file_operations
2600 #define shmem_get_inode(sb, mode, dev, flags)   ramfs_get_inode(sb, mode, dev)
2601 #define shmem_acct_size(flags, size)            0
2602 #define shmem_unacct_size(flags, size)          do {} while (0)
2603 #define SHMEM_MAX_BYTES                         MAX_LFS_FILESIZE
2604
2605 #endif /* CONFIG_SHMEM */
2606
2607 /* common code */
2608
2609 /**
2610  * shmem_file_setup - get an unlinked file living in tmpfs
2611  * @name: name for dentry (to be seen in /proc/<pid>/maps
2612  * @size: size to be set for the file
2613  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2614  */
2615 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2616 {
2617         int error;
2618         struct file *file;
2619         struct inode *inode;
2620         struct dentry *dentry, *root;
2621         struct qstr this;
2622
2623         if (IS_ERR(shm_mnt))
2624                 return (void *)shm_mnt;
2625
2626         if (size < 0 || size > SHMEM_MAX_BYTES)
2627                 return ERR_PTR(-EINVAL);
2628
2629         if (shmem_acct_size(flags, size))
2630                 return ERR_PTR(-ENOMEM);
2631
2632         error = -ENOMEM;
2633         this.name = name;
2634         this.len = strlen(name);
2635         this.hash = 0; /* will go */
2636         root = shm_mnt->mnt_root;
2637         dentry = d_alloc(root, &this);
2638         if (!dentry)
2639                 goto put_memory;
2640
2641         error = -ENFILE;
2642         file = get_empty_filp();
2643         if (!file)
2644                 goto put_dentry;
2645
2646         error = -ENOSPC;
2647         inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0, flags);
2648         if (!inode)
2649                 goto close_file;
2650
2651         d_instantiate(dentry, inode);
2652         inode->i_size = size;
2653         inode->i_nlink = 0;     /* It is unlinked */
2654         init_file(file, shm_mnt, dentry, FMODE_WRITE | FMODE_READ,
2655                   &shmem_file_operations);
2656
2657 #ifndef CONFIG_MMU
2658         error = ramfs_nommu_expand_for_mapping(inode, size);
2659         if (error)
2660                 goto close_file;
2661 #endif
2662         return file;
2663
2664 close_file:
2665         put_filp(file);
2666 put_dentry:
2667         dput(dentry);
2668 put_memory:
2669         shmem_unacct_size(flags, size);
2670         return ERR_PTR(error);
2671 }
2672 EXPORT_SYMBOL_GPL(shmem_file_setup);
2673
2674 /**
2675  * shmem_zero_setup - setup a shared anonymous mapping
2676  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2677  */
2678 int shmem_zero_setup(struct vm_area_struct *vma)
2679 {
2680         struct file *file;
2681         loff_t size = vma->vm_end - vma->vm_start;
2682
2683         file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2684         if (IS_ERR(file))
2685                 return PTR_ERR(file);
2686
2687         ima_shm_check(file);
2688         if (vma->vm_file)
2689                 fput(vma->vm_file);
2690         vma->vm_file = file;
2691         vma->vm_ops = &shmem_vm_ops;
2692         return 0;
2693 }
2694
2695 module_init(init_tmpfs)