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) 2002-2004 Hugh Dickins.
  10  * Copyright (C) 2002-2004 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  * This file is released under the GPL.
  18  */
  19
  20 /*
  21  * This virtual memory filesystem is heavily based on the ramfs. It
  22  * extends ramfs by the ability to use swap and honor resource limits
  23  * which makes it a completely usable filesystem.
  24  */
  25
  26 #include <linux/config.h>
  27 #include <linux/module.h>
  28 #include <linux/init.h>
  29 #include <linux/devfs_fs_kernel.h>
  30 #include <linux/fs.h>
  31 #include <linux/mm.h>
  32 #include <linux/mman.h>
  33 #include <linux/file.h>
  34 #include <linux/swap.h>
  35 #include <linux/pagemap.h>
  36 #include <linux/string.h>
  37 #include <linux/slab.h>
  38 #include <linux/backing-dev.h>
  39 #include <linux/shmem_fs.h>
  40 #include <linux/mount.h>
  41 #include <linux/writeback.h>
  42 #include <linux/vfs.h>
  43 #include <linux/blkdev.h>
  44 #include <linux/security.h>
  45 #include <linux/swapops.h>
  46 #include <linux/mempolicy.h>
  47 #include <linux/namei.h>
  48 #include <linux/xattr.h>
  49 #include <asm/uaccess.h>
  50 #include <asm/div64.h>
  51 #include <asm/pgtable.h>
  52
  53 /* This magic number is used in glibc for posix shared memory */
  54
  55 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
  56 #define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
  57 #define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
  58
  59 #define SHMEM_MAX_INDEX  (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
  60 #define SHMEM_MAX_BYTES  ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
  61
  62 #define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
  63
  64 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
  65 #define SHMEM_PAGEIN     VM_READ
  66 #define SHMEM_TRUNCATE   VM_WRITE
  67
  68 /* Pretend that each entry is of this size in directory's i_size */
  69 #define BOGO_DIRENT_SIZE 20
  70
  71 /* Keep swapped page count in private field of indirect struct page */
  72 #define nr_swapped              private
  73
  74 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
  75 enum sgp_type {
  76         SGP_QUICK,      /* don't try more than file page cache lookup */
  77         SGP_READ,       /* don't exceed i_size, don't allocate page */
  78         SGP_CACHE,      /* don't exceed i_size, may allocate page */
  79         SGP_WRITE,      /* may exceed i_size, may allocate page */
  80 };
  81
  82 static int shmem_getpage(struct inode *inode, unsigned long idx,
  83                          struct page **pagep, enum sgp_type sgp, int *type);
  84
  85 static inline struct page *shmem_dir_alloc(unsigned int gfp_mask)
  86 {
  87         /*
  88          * The above definition of ENTRIES_PER_PAGE, and the use of
  89          * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
  90          * might be reconsidered if it ever diverges from PAGE_SIZE.
  91          */
  92         return alloc_pages(gfp_mask, PAGE_CACHE_SHIFT-PAGE_SHIFT);
  93 }
  94
  95 static inline void shmem_dir_free(struct page *page)
  96 {
  97         __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
  98 }
  99
 100 static struct page **shmem_dir_map(struct page *page)
 101 {
 102         return (struct page **)kmap_atomic(page, KM_USER0);
 103 }
 104
 105 static inline void shmem_dir_unmap(struct page **dir)
 106 {
 107         kunmap_atomic(dir, KM_USER0);
 108 }
 109
 110 static swp_entry_t *shmem_swp_map(struct page *page)
 111 {
 112         return (swp_entry_t *)kmap_atomic(page, KM_USER1);
 113 }
 114
 115 static inline void shmem_swp_balance_unmap(void)
 116 {
 117         /*
 118          * When passing a pointer to an i_direct entry, to code which
 119          * also handles indirect entries and so will shmem_swp_unmap,
 120          * we must arrange for the preempt count to remain in balance.
 121          * What kmap_atomic of a lowmem page does depends on config
 122          * and architecture, so pretend to kmap_atomic some lowmem page.
 123          */
 124         (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
 125 }
 126
 127 static inline void shmem_swp_unmap(swp_entry_t *entry)
 128 {
 129         kunmap_atomic(entry, KM_USER1);
 130 }
 131
 132 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
 133 {
 134         return sb->s_fs_info;
 135 }
 136
 137 /*
 138  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
 139  * for shared memory and for shared anonymous (/dev/zero) mappings
 140  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
 141  * consistent with the pre-accounting of private mappings ...
 142  */
 143 static inline int shmem_acct_size(unsigned long flags, loff_t size)
 144 {
 145         return (flags & VM_ACCOUNT)?
 146                 security_vm_enough_memory(VM_ACCT(size)): 0;
 147 }
 148
 149 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
 150 {
 151         if (flags & VM_ACCOUNT)
 152                 vm_unacct_memory(VM_ACCT(size));
 153 }
 154
 155 /*
 156  * ... whereas tmpfs objects are accounted incrementally as
 157  * pages are allocated, in order to allow huge sparse files.
 158  * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
 159  * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
 160  */
 161 static inline int shmem_acct_block(unsigned long flags)
 162 {
 163         return (flags & VM_ACCOUNT)?
 164                 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
 165 }
 166
 167 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
 168 {
 169         if (!(flags & VM_ACCOUNT))
 170                 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
 171 }
 172
 173 static struct super_operations shmem_ops;
 174 static struct address_space_operations shmem_aops;
 175 static struct file_operations shmem_file_operations;
 176 static struct inode_operations shmem_inode_operations;
 177 static struct inode_operations shmem_dir_inode_operations;
 178 static struct inode_operations shmem_special_inode_operations;
 179 static struct vm_operations_struct shmem_vm_ops;
 180
 181 static struct backing_dev_info shmem_backing_dev_info = {
 182         .ra_pages       = 0,    /* No readahead */
 183         .memory_backed  = 1,    /* Does not contribute to dirty memory */
 184         .unplug_io_fn = default_unplug_io_fn,
 185 };
 186
 187 static LIST_HEAD(shmem_swaplist);
 188 static spinlock_t shmem_swaplist_lock = SPIN_LOCK_UNLOCKED;
 189
 190 static void shmem_free_block(struct inode *inode)
 191 {
 192         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 193         if (sbinfo) {
 194                 spin_lock(&sbinfo->stat_lock);
 195                 sbinfo->free_blocks++;
 196                 inode->i_blocks -= BLOCKS_PER_PAGE;
 197                 spin_unlock(&sbinfo->stat_lock);
 198         }
 199 }
 200
 201 /*
 202  * shmem_recalc_inode - recalculate the size of an inode
 203  *
 204  * @inode: inode to recalc
 205  *
 206  * We have to calculate the free blocks since the mm can drop
 207  * undirtied hole pages behind our back.
 208  *
 209  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
 210  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
 211  *
 212  * It has to be called with the spinlock held.
 213  */
 214 static void shmem_recalc_inode(struct inode *inode)
 215 {
 216         struct shmem_inode_info *info = SHMEM_I(inode);
 217         long freed;
 218
 219         freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
 220         if (freed > 0) {
 221                 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 222                 info->alloced -= freed;
 223                 shmem_unacct_blocks(info->flags, freed);
 224                 if (sbinfo) {
 225                         spin_lock(&sbinfo->stat_lock);
 226                         sbinfo->free_blocks += freed;
 227                         inode->i_blocks -= freed*BLOCKS_PER_PAGE;
 228                         spin_unlock(&sbinfo->stat_lock);
 229                 }
 230         }
 231 }
 232
 233 /*
 234  * shmem_swp_entry - find the swap vector position in the info structure
 235  *
 236  * @info:  info structure for the inode
 237  * @index: index of the page to find
 238  * @page:  optional page to add to the structure. Has to be preset to
 239  *         all zeros
 240  *
 241  * If there is no space allocated yet it will return NULL when
 242  * page is NULL, else it will use the page for the needed block,
 243  * setting it to NULL on return to indicate that it has been used.
 244  *
 245  * The swap vector is organized the following way:
 246  *
 247  * There are SHMEM_NR_DIRECT entries directly stored in the
 248  * shmem_inode_info structure. So small files do not need an addional
 249  * allocation.
 250  *
 251  * For pages with index > SHMEM_NR_DIRECT there is the pointer
 252  * i_indirect which points to a page which holds in the first half
 253  * doubly indirect blocks, in the second half triple indirect blocks:
 254  *
 255  * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
 256  * following layout (for SHMEM_NR_DIRECT == 16):
 257  *
 258  * i_indirect -> dir --> 16-19
 259  *            |      +-> 20-23
 260  *            |
 261  *            +-->dir2 --> 24-27
 262  *            |        +-> 28-31
 263  *            |        +-> 32-35
 264  *            |        +-> 36-39
 265  *            |
 266  *            +-->dir3 --> 40-43
 267  *                     +-> 44-47
 268  *                     +-> 48-51
 269  *                     +-> 52-55
 270  */
 271 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
 272 {
 273         unsigned long offset;
 274         struct page **dir;
 275         struct page *subdir;
 276
 277         if (index < SHMEM_NR_DIRECT) {
 278                 shmem_swp_balance_unmap();
 279                 return info->i_direct+index;
 280         }
 281         if (!info->i_indirect) {
 282                 if (page) {
 283                         info->i_indirect = *page;
 284                         *page = NULL;
 285                 }
 286                 return NULL;                    /* need another page */
 287         }
 288
 289         index -= SHMEM_NR_DIRECT;
 290         offset = index % ENTRIES_PER_PAGE;
 291         index /= ENTRIES_PER_PAGE;
 292         dir = shmem_dir_map(info->i_indirect);
 293
 294         if (index >= ENTRIES_PER_PAGE/2) {
 295                 index -= ENTRIES_PER_PAGE/2;
 296                 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
 297                 index %= ENTRIES_PER_PAGE;
 298                 subdir = *dir;
 299                 if (!subdir) {
 300                         if (page) {
 301                                 *dir = *page;
 302                                 *page = NULL;
 303                         }
 304                         shmem_dir_unmap(dir);
 305                         return NULL;            /* need another page */
 306                 }
 307                 shmem_dir_unmap(dir);
 308                 dir = shmem_dir_map(subdir);
 309         }
 310
 311         dir += index;
 312         subdir = *dir;
 313         if (!subdir) {
 314                 if (!page || !(subdir = *page)) {
 315                         shmem_dir_unmap(dir);
 316                         return NULL;            /* need a page */
 317                 }
 318                 *dir = subdir;
 319                 *page = NULL;
 320         }
 321         shmem_dir_unmap(dir);
 322         return shmem_swp_map(subdir) + offset;
 323 }
 324
 325 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
 326 {
 327         long incdec = value? 1: -1;
 328
 329         entry->val = value;
 330         info->swapped += incdec;
 331         if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT)
 332                 kmap_atomic_to_page(entry)->nr_swapped += incdec;
 333 }
 334
 335 /*
 336  * shmem_swp_alloc - get the position of the swap entry for the page.
 337  *                   If it does not exist allocate the entry.
 338  *
 339  * @info:       info structure for the inode
 340  * @index:      index of the page to find
 341  * @sgp:        check and recheck i_size? skip allocation?
 342  */
 343 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
 344 {
 345         struct inode *inode = &info->vfs_inode;
 346         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 347         struct page *page = NULL;
 348         swp_entry_t *entry;
 349
 350         if (sgp != SGP_WRITE &&
 351             ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
 352                 return ERR_PTR(-EINVAL);
 353
 354         while (!(entry = shmem_swp_entry(info, index, &page))) {
 355                 if (sgp == SGP_READ)
 356                         return shmem_swp_map(ZERO_PAGE(0));
 357                 /*
 358                  * Test free_blocks against 1 not 0, since we have 1 data
 359                  * page (and perhaps indirect index pages) yet to allocate:
 360                  * a waste to allocate index if we cannot allocate data.
 361                  */
 362                 if (sbinfo) {
 363                         spin_lock(&sbinfo->stat_lock);
 364                         if (sbinfo->free_blocks <= 1) {
 365                                 spin_unlock(&sbinfo->stat_lock);
 366                                 return ERR_PTR(-ENOSPC);
 367                         }
 368                         sbinfo->free_blocks--;
 369                         inode->i_blocks += BLOCKS_PER_PAGE;
 370                         spin_unlock(&sbinfo->stat_lock);
 371                 }
 372
 373                 spin_unlock(&info->lock);
 374                 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
 375                 if (page) {
 376                         clear_highpage(page);
 377                         page->nr_swapped = 0;
 378                 }
 379                 spin_lock(&info->lock);
 380
 381                 if (!page) {
 382                         shmem_free_block(inode);
 383                         return ERR_PTR(-ENOMEM);
 384                 }
 385                 if (sgp != SGP_WRITE &&
 386                     ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
 387                         entry = ERR_PTR(-EINVAL);
 388                         break;
 389                 }
 390                 if (info->next_index <= index)
 391                         info->next_index = index + 1;
 392         }
 393         if (page) {
 394                 /* another task gave its page, or truncated the file */
 395                 shmem_free_block(inode);
 396                 shmem_dir_free(page);
 397         }
 398         if (info->next_index <= index && !IS_ERR(entry))
 399                 info->next_index = index + 1;
 400         return entry;
 401 }
 402
 403 /*
 404  * shmem_free_swp - free some swap entries in a directory
 405  *
 406  * @dir:   pointer to the directory
 407  * @edir:  pointer after last entry of the directory
 408  */
 409 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir)
 410 {
 411         swp_entry_t *ptr;
 412         int freed = 0;
 413
 414         for (ptr = dir; ptr < edir; ptr++) {
 415                 if (ptr->val) {
 416                         free_swap_and_cache(*ptr);
 417                         *ptr = (swp_entry_t){0};
 418                         freed++;
 419                 }
 420         }
 421         return freed;
 422 }
 423
 424 static void shmem_truncate(struct inode *inode)
 425 {
 426         struct shmem_inode_info *info = SHMEM_I(inode);
 427         unsigned long idx;
 428         unsigned long size;
 429         unsigned long limit;
 430         unsigned long stage;
 431         struct page **dir;
 432         struct page *subdir;
 433         struct page *empty;
 434         swp_entry_t *ptr;
 435         int offset;
 436         int freed;
 437
 438         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 439         idx = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 440         if (idx >= info->next_index)
 441                 return;
 442
 443         spin_lock(&info->lock);
 444         info->flags |= SHMEM_TRUNCATE;
 445         limit = info->next_index;
 446         info->next_index = idx;
 447         if (info->swapped && idx < SHMEM_NR_DIRECT) {
 448                 ptr = info->i_direct;
 449                 size = limit;
 450                 if (size > SHMEM_NR_DIRECT)
 451                         size = SHMEM_NR_DIRECT;
 452                 info->swapped -= shmem_free_swp(ptr+idx, ptr+size);
 453         }
 454         if (!info->i_indirect)
 455                 goto done2;
 456
 457         BUG_ON(limit <= SHMEM_NR_DIRECT);
 458         limit -= SHMEM_NR_DIRECT;
 459         idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
 460         offset = idx % ENTRIES_PER_PAGE;
 461         idx -= offset;
 462
 463         empty = NULL;
 464         dir = shmem_dir_map(info->i_indirect);
 465         stage = ENTRIES_PER_PAGEPAGE/2;
 466         if (idx < ENTRIES_PER_PAGEPAGE/2)
 467                 dir += idx/ENTRIES_PER_PAGE;
 468         else {
 469                 dir += ENTRIES_PER_PAGE/2;
 470                 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
 471                 while (stage <= idx)
 472                         stage += ENTRIES_PER_PAGEPAGE;
 473                 if (*dir) {
 474                         subdir = *dir;
 475                         size = ((idx - ENTRIES_PER_PAGEPAGE/2) %
 476                                 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
 477                         if (!size && !offset) {
 478                                 empty = subdir;
 479                                 *dir = NULL;
 480                         }
 481                         shmem_dir_unmap(dir);
 482                         dir = shmem_dir_map(subdir) + size;
 483                 } else {
 484                         offset = 0;
 485                         idx = stage;
 486                 }
 487         }
 488
 489         for (; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
 490                 if (unlikely(idx == stage)) {
 491                         shmem_dir_unmap(dir-1);
 492                         dir = shmem_dir_map(info->i_indirect) +
 493                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
 494                         while (!*dir) {
 495                                 dir++;
 496                                 idx += ENTRIES_PER_PAGEPAGE;
 497                                 if (idx >= limit)
 498                                         goto done1;
 499                         }
 500                         stage = idx + ENTRIES_PER_PAGEPAGE;
 501                         subdir = *dir;
 502                         *dir = NULL;
 503                         shmem_dir_unmap(dir);
 504                         if (empty) {
 505                                 shmem_dir_free(empty);
 506                                 shmem_free_block(inode);
 507                         }
 508                         empty = subdir;
 509                         cond_resched_lock(&info->lock);
 510                         dir = shmem_dir_map(subdir);
 511                 }
 512                 subdir = *dir;
 513                 if (subdir && subdir->nr_swapped) {
 514                         ptr = shmem_swp_map(subdir);
 515                         size = limit - idx;
 516                         if (size > ENTRIES_PER_PAGE)
 517                                 size = ENTRIES_PER_PAGE;
 518                         freed = shmem_free_swp(ptr+offset, ptr+size);
 519                         shmem_swp_unmap(ptr);
 520                         info->swapped -= freed;
 521                         subdir->nr_swapped -= freed;
 522                         BUG_ON(subdir->nr_swapped > offset);
 523                 }
 524                 if (offset)
 525                         offset = 0;
 526                 else if (subdir) {
 527                         *dir = NULL;
 528                         shmem_dir_free(subdir);
 529                         shmem_free_block(inode);
 530                 }
 531         }
 532 done1:
 533         shmem_dir_unmap(dir-1);
 534         if (empty) {
 535                 shmem_dir_free(empty);
 536                 shmem_free_block(inode);
 537         }
 538         if (info->next_index <= SHMEM_NR_DIRECT) {
 539                 shmem_dir_free(info->i_indirect);
 540                 info->i_indirect = NULL;
 541                 shmem_free_block(inode);
 542         }
 543 done2:
 544         BUG_ON(info->swapped > info->next_index);
 545         if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
 546                 /*
 547                  * Call truncate_inode_pages again: racing shmem_unuse_inode
 548                  * may have swizzled a page in from swap since vmtruncate or
 549                  * generic_delete_inode did it, before we lowered next_index.
 550                  * Also, though shmem_getpage checks i_size before adding to
 551                  * cache, no recheck after: so fix the narrow window there too.
 552                  */
 553                 spin_unlock(&info->lock);
 554                 truncate_inode_pages(inode->i_mapping, inode->i_size);
 555                 spin_lock(&info->lock);
 556         }
 557         info->flags &= ~SHMEM_TRUNCATE;
 558         shmem_recalc_inode(inode);
 559         spin_unlock(&info->lock);
 560 }
 561
 562 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
 563 {
 564         struct inode *inode = dentry->d_inode;
 565         struct page *page = NULL;
 566         int error;
 567
 568         if (attr->ia_valid & ATTR_SIZE) {
 569                 if (attr->ia_size < inode->i_size) {
 570                         /*
 571                          * If truncating down to a partial page, then
 572                          * if that page is already allocated, hold it
 573                          * in memory until the truncation is over, so
 574                          * truncate_partial_page cannnot miss it were
 575                          * it assigned to swap.
 576                          */
 577                         if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
 578                                 (void) shmem_getpage(inode,
 579                                         attr->ia_size>>PAGE_CACHE_SHIFT,
 580                                                 &page, SGP_READ, NULL);
 581                         }
 582                         /*
 583                          * Reset SHMEM_PAGEIN flag so that shmem_truncate can
 584                          * detect if any pages might have been added to cache
 585                          * after truncate_inode_pages.  But we needn't bother
 586                          * if it's being fully truncated to zero-length: the
 587                          * nrpages check is efficient enough in that case.
 588                          */
 589                         if (attr->ia_size) {
 590                                 struct shmem_inode_info *info = SHMEM_I(inode);
 591                                 spin_lock(&info->lock);
 592                                 info->flags &= ~SHMEM_PAGEIN;
 593                                 spin_unlock(&info->lock);
 594                         }
 595                 }
 596         }
 597
 598         error = inode_change_ok(inode, attr);
 599         if (!error)
 600                 error = inode_setattr(inode, attr);
 601         if (page)
 602                 page_cache_release(page);
 603         return error;
 604 }
 605
 606 static void shmem_delete_inode(struct inode *inode)
 607 {
 608         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 609         struct shmem_inode_info *info = SHMEM_I(inode);
 610
 611         if (inode->i_op->truncate == shmem_truncate) {
 612                 shmem_unacct_size(info->flags, inode->i_size);
 613                 inode->i_size = 0;
 614                 shmem_truncate(inode);
 615                 if (!list_empty(&info->swaplist)) {
 616                         spin_lock(&shmem_swaplist_lock);
 617                         list_del_init(&info->swaplist);
 618                         spin_unlock(&shmem_swaplist_lock);
 619                 }
 620         }
 621         if (sbinfo) {
 622                 BUG_ON(inode->i_blocks);
 623                 spin_lock(&sbinfo->stat_lock);
 624                 sbinfo->free_inodes++;
 625                 spin_unlock(&sbinfo->stat_lock);
 626         }
 627         clear_inode(inode);
 628 }
 629
 630 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
 631 {
 632         swp_entry_t *ptr;
 633
 634         for (ptr = dir; ptr < edir; ptr++) {
 635                 if (ptr->val == entry.val)
 636                         return ptr - dir;
 637         }
 638         return -1;
 639 }
 640
 641 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
 642 {
 643         struct inode *inode;
 644         unsigned long idx;
 645         unsigned long size;
 646         unsigned long limit;
 647         unsigned long stage;
 648         struct page **dir;
 649         struct page *subdir;
 650         swp_entry_t *ptr;
 651         int offset;
 652
 653         idx = 0;
 654         ptr = info->i_direct;
 655         spin_lock(&info->lock);
 656         limit = info->next_index;
 657         size = limit;
 658         if (size > SHMEM_NR_DIRECT)
 659                 size = SHMEM_NR_DIRECT;
 660         offset = shmem_find_swp(entry, ptr, ptr+size);
 661         if (offset >= 0) {
 662                 shmem_swp_balance_unmap();
 663                 goto found;
 664         }
 665         if (!info->i_indirect)
 666                 goto lost2;
 667         /* we might be racing with shmem_truncate */
 668         if (limit <= SHMEM_NR_DIRECT)
 669                 goto lost2;
 670
 671         dir = shmem_dir_map(info->i_indirect);
 672         stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
 673
 674         for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
 675                 if (unlikely(idx == stage)) {
 676                         shmem_dir_unmap(dir-1);
 677                         dir = shmem_dir_map(info->i_indirect) +
 678                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
 679                         while (!*dir) {
 680                                 dir++;
 681                                 idx += ENTRIES_PER_PAGEPAGE;
 682                                 if (idx >= limit)
 683                                         goto lost1;
 684                         }
 685                         stage = idx + ENTRIES_PER_PAGEPAGE;
 686                         subdir = *dir;
 687                         shmem_dir_unmap(dir);
 688                         dir = shmem_dir_map(subdir);
 689                 }
 690                 subdir = *dir;
 691                 if (subdir && subdir->nr_swapped) {
 692                         ptr = shmem_swp_map(subdir);
 693                         size = limit - idx;
 694                         if (size > ENTRIES_PER_PAGE)
 695                                 size = ENTRIES_PER_PAGE;
 696                         offset = shmem_find_swp(entry, ptr, ptr+size);
 697                         if (offset >= 0) {
 698                                 shmem_dir_unmap(dir);
 699                                 goto found;
 700                         }
 701                         shmem_swp_unmap(ptr);
 702                 }
 703         }
 704 lost1:
 705         shmem_dir_unmap(dir-1);
 706 lost2:
 707         spin_unlock(&info->lock);
 708         return 0;
 709 found:
 710         idx += offset;
 711         inode = &info->vfs_inode;
 712         if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
 713                 info->flags |= SHMEM_PAGEIN;
 714                 shmem_swp_set(info, ptr + offset, 0);
 715         }
 716         shmem_swp_unmap(ptr);
 717         spin_unlock(&info->lock);
 718         /*
 719          * Decrement swap count even when the entry is left behind:
 720          * try_to_unuse will skip over mms, then reincrement count.
 721          */
 722         swap_free(entry);
 723         return 1;
 724 }
 725
 726 /*
 727  * shmem_unuse() search for an eventually swapped out shmem page.
 728  */
 729 int shmem_unuse(swp_entry_t entry, struct page *page)
 730 {
 731         struct list_head *p, *next;
 732         struct shmem_inode_info *info;
 733         int found = 0;
 734
 735         spin_lock(&shmem_swaplist_lock);
 736         list_for_each_safe(p, next, &shmem_swaplist) {
 737                 info = list_entry(p, struct shmem_inode_info, swaplist);
 738                 if (!info->swapped)
 739                         list_del_init(&info->swaplist);
 740                 else if (shmem_unuse_inode(info, entry, page)) {
 741                         /* move head to start search for next from here */
 742                         list_move_tail(&shmem_swaplist, &info->swaplist);
 743                         found = 1;
 744                         break;
 745                 }
 746         }
 747         spin_unlock(&shmem_swaplist_lock);
 748         return found;
 749 }
 750
 751 /*
 752  * Move the page from the page cache to the swap cache.
 753  */
 754 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
 755 {
 756         struct shmem_inode_info *info;
 757         swp_entry_t *entry, swap;
 758         struct address_space *mapping;
 759         unsigned long index;
 760         struct inode *inode;
 761
 762         BUG_ON(!PageLocked(page));
 763         BUG_ON(page_mapped(page));
 764
 765         mapping = page->mapping;
 766         index = page->index;
 767         inode = mapping->host;
 768         info = SHMEM_I(inode);
 769         if (info->flags & VM_LOCKED)
 770                 goto redirty;
 771         swap = get_swap_page();
 772         if (!swap.val)
 773                 goto redirty;
 774
 775         spin_lock(&info->lock);
 776         shmem_recalc_inode(inode);
 777         if (index >= info->next_index) {
 778                 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
 779                 goto unlock;
 780         }
 781         entry = shmem_swp_entry(info, index, NULL);
 782         BUG_ON(!entry);
 783         BUG_ON(entry->val);
 784
 785         if (move_to_swap_cache(page, swap) == 0) {
 786                 shmem_swp_set(info, entry, swap.val);
 787                 shmem_swp_unmap(entry);
 788                 spin_unlock(&info->lock);
 789                 if (list_empty(&info->swaplist)) {
 790                         spin_lock(&shmem_swaplist_lock);
 791                         /* move instead of add in case we're racing */
 792                         list_move_tail(&info->swaplist, &shmem_swaplist);
 793                         spin_unlock(&shmem_swaplist_lock);
 794                 }
 795                 unlock_page(page);
 796                 return 0;
 797         }
 798
 799         shmem_swp_unmap(entry);
 800 unlock:
 801         spin_unlock(&info->lock);
 802         swap_free(swap);
 803 redirty:
 804         set_page_dirty(page);
 805         return WRITEPAGE_ACTIVATE;      /* Return with the page locked */
 806 }
 807
 808 #ifdef CONFIG_NUMA
 809 static struct page *shmem_swapin_async(struct shared_policy *p,
 810                                        swp_entry_t entry, unsigned long idx)
 811 {
 812         struct page *page;
 813         struct vm_area_struct pvma;
 814
 815         /* Create a pseudo vma that just contains the policy */
 816         memset(&pvma, 0, sizeof(struct vm_area_struct));
 817         pvma.vm_end = PAGE_SIZE;
 818         pvma.vm_pgoff = idx;
 819         pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
 820         page = read_swap_cache_async(entry, &pvma, 0);
 821         mpol_free(pvma.vm_policy);
 822         return page;
 823 }
 824
 825 struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry,
 826                           unsigned long idx)
 827 {
 828         struct shared_policy *p = &info->policy;
 829         int i, num;
 830         struct page *page;
 831         unsigned long offset;
 832
 833         num = valid_swaphandles(entry, &offset);
 834         for (i = 0; i < num; offset++, i++) {
 835                 page = shmem_swapin_async(p,
 836                                 swp_entry(swp_type(entry), offset), idx);
 837                 if (!page)
 838                         break;
 839                 page_cache_release(page);
 840         }
 841         lru_add_drain();        /* Push any new pages onto the LRU now */
 842         return shmem_swapin_async(p, entry, idx);
 843 }
 844
 845 static struct page *
 846 shmem_alloc_page(unsigned long gfp, struct shmem_inode_info *info,
 847                  unsigned long idx)
 848 {
 849         struct vm_area_struct pvma;
 850         struct page *page;
 851
 852         memset(&pvma, 0, sizeof(struct vm_area_struct));
 853         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
 854         pvma.vm_pgoff = idx;
 855         pvma.vm_end = PAGE_SIZE;
 856         page = alloc_page_vma(gfp, &pvma, 0);
 857         mpol_free(pvma.vm_policy);
 858         return page;
 859 }
 860 #else
 861 static inline struct page *
 862 shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
 863 {
 864         swapin_readahead(entry, 0, NULL);
 865         return read_swap_cache_async(entry, NULL, 0);
 866 }
 867
 868 static inline struct page *
 869 shmem_alloc_page(unsigned long gfp,struct shmem_inode_info *info,
 870                                  unsigned long idx)
 871 {
 872         return alloc_page(gfp);
 873 }
 874 #endif
 875
 876 /*
 877  * shmem_getpage - either get the page from swap or allocate a new one
 878  *
 879  * If we allocate a new one we do not mark it dirty. That's up to the
 880  * vm. If we swap it in we mark it dirty since we also free the swap
 881  * entry since a page cannot live in both the swap and page cache
 882  */
 883 static int shmem_getpage(struct inode *inode, unsigned long idx,
 884                         struct page **pagep, enum sgp_type sgp, int *type)
 885 {
 886         struct address_space *mapping = inode->i_mapping;
 887         struct shmem_inode_info *info = SHMEM_I(inode);
 888         struct shmem_sb_info *sbinfo;
 889         struct page *filepage = *pagep;
 890         struct page *swappage;
 891         swp_entry_t *entry;
 892         swp_entry_t swap;
 893         int error;
 894
 895         if (idx >= SHMEM_MAX_INDEX)
 896                 return -EFBIG;
 897         /*
 898          * Normally, filepage is NULL on entry, and either found
 899          * uptodate immediately, or allocated and zeroed, or read
 900          * in under swappage, which is then assigned to filepage.
 901          * But shmem_prepare_write passes in a locked filepage,
 902          * which may be found not uptodate by other callers too,
 903          * and may need to be copied from the swappage read in.
 904          */
 905 repeat:
 906         if (!filepage)
 907                 filepage = find_lock_page(mapping, idx);
 908         if (filepage && PageUptodate(filepage))
 909                 goto done;
 910         error = 0;
 911         if (sgp == SGP_QUICK)
 912                 goto failed;
 913
 914         spin_lock(&info->lock);
 915         shmem_recalc_inode(inode);
 916         entry = shmem_swp_alloc(info, idx, sgp);
 917         if (IS_ERR(entry)) {
 918                 spin_unlock(&info->lock);
 919                 error = PTR_ERR(entry);
 920                 goto failed;
 921         }
 922         swap = *entry;
 923
 924         if (swap.val) {
 925                 /* Look it up and read it in.. */
 926                 swappage = lookup_swap_cache(swap);
 927                 if (!swappage) {
 928                         shmem_swp_unmap(entry);
 929                         spin_unlock(&info->lock);
 930                         /* here we actually do the io */
 931                         if (type && *type == VM_FAULT_MINOR) {
 932                                 inc_page_state(pgmajfault);
 933                                 *type = VM_FAULT_MAJOR;
 934                         }
 935                         swappage = shmem_swapin(info, swap, idx);
 936                         if (!swappage) {
 937                                 spin_lock(&info->lock);
 938                                 entry = shmem_swp_alloc(info, idx, sgp);
 939                                 if (IS_ERR(entry))
 940                                         error = PTR_ERR(entry);
 941                                 else {
 942                                         if (entry->val == swap.val)
 943                                                 error = -ENOMEM;
 944                                         shmem_swp_unmap(entry);
 945                                 }
 946                                 spin_unlock(&info->lock);
 947                                 if (error)
 948                                         goto failed;
 949                                 goto repeat;
 950                         }
 951                         wait_on_page_locked(swappage);
 952                         page_cache_release(swappage);
 953                         goto repeat;
 954                 }
 955
 956                 /* We have to do this with page locked to prevent races */
 957                 if (TestSetPageLocked(swappage)) {
 958                         shmem_swp_unmap(entry);
 959                         spin_unlock(&info->lock);
 960                         wait_on_page_locked(swappage);
 961                         page_cache_release(swappage);
 962                         goto repeat;
 963                 }
 964                 if (PageWriteback(swappage)) {
 965                         shmem_swp_unmap(entry);
 966                         spin_unlock(&info->lock);
 967                         wait_on_page_writeback(swappage);
 968                         unlock_page(swappage);
 969                         page_cache_release(swappage);
 970                         goto repeat;
 971                 }
 972                 if (!PageUptodate(swappage)) {
 973                         shmem_swp_unmap(entry);
 974                         spin_unlock(&info->lock);
 975                         unlock_page(swappage);
 976                         page_cache_release(swappage);
 977                         error = -EIO;
 978                         goto failed;
 979                 }
 980
 981                 if (filepage) {
 982                         shmem_swp_set(info, entry, 0);
 983                         shmem_swp_unmap(entry);
 984                         delete_from_swap_cache(swappage);
 985                         spin_unlock(&info->lock);
 986                         copy_highpage(filepage, swappage);
 987                         unlock_page(swappage);
 988                         page_cache_release(swappage);
 989                         flush_dcache_page(filepage);
 990                         SetPageUptodate(filepage);
 991                         set_page_dirty(filepage);
 992                         swap_free(swap);
 993                 } else if (!(error = move_from_swap_cache(
 994                                 swappage, idx, mapping))) {
 995                         info->flags |= SHMEM_PAGEIN;
 996                         shmem_swp_set(info, entry, 0);
 997                         shmem_swp_unmap(entry);
 998                         spin_unlock(&info->lock);
 999                         filepage = swappage;
1000                         swap_free(swap);
1001                 } else {
1002                         shmem_swp_unmap(entry);
1003                         spin_unlock(&info->lock);
1004                         unlock_page(swappage);
1005                         page_cache_release(swappage);
1006                         if (error == -ENOMEM) {
1007                                 /* let kswapd refresh zone for GFP_ATOMICs */
1008                                 blk_congestion_wait(WRITE, HZ/50);
1009                         }
1010                         goto repeat;
1011                 }
1012         } else if (sgp == SGP_READ && !filepage) {
1013                 shmem_swp_unmap(entry);
1014                 filepage = find_get_page(mapping, idx);
1015                 if (filepage &&
1016                     (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1017                         spin_unlock(&info->lock);
1018                         wait_on_page_locked(filepage);
1019                         page_cache_release(filepage);
1020                         filepage = NULL;
1021                         goto repeat;
1022                 }
1023                 spin_unlock(&info->lock);
1024         } else {
1025                 shmem_swp_unmap(entry);
1026                 sbinfo = SHMEM_SB(inode->i_sb);
1027                 if (sbinfo) {
1028                         spin_lock(&sbinfo->stat_lock);
1029                         if (sbinfo->free_blocks == 0 ||
1030                             shmem_acct_block(info->flags)) {
1031                                 spin_unlock(&sbinfo->stat_lock);
1032                                 spin_unlock(&info->lock);
1033                                 error = -ENOSPC;
1034                                 goto failed;
1035                         }
1036                         sbinfo->free_blocks--;
1037                         inode->i_blocks += BLOCKS_PER_PAGE;
1038                         spin_unlock(&sbinfo->stat_lock);
1039                 } else if (shmem_acct_block(info->flags)) {
1040                         spin_unlock(&info->lock);
1041                         error = -ENOSPC;
1042                         goto failed;
1043                 }
1044
1045                 if (!filepage) {
1046                         spin_unlock(&info->lock);
1047                         filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
1048                                                     info,
1049                                                     idx);
1050                         if (!filepage) {
1051                                 shmem_unacct_blocks(info->flags, 1);
1052                                 shmem_free_block(inode);
1053                                 error = -ENOMEM;
1054                                 goto failed;
1055                         }
1056
1057                         spin_lock(&info->lock);
1058                         entry = shmem_swp_alloc(info, idx, sgp);
1059                         if (IS_ERR(entry))
1060                                 error = PTR_ERR(entry);
1061                         else {
1062                                 swap = *entry;
1063                                 shmem_swp_unmap(entry);
1064                         }
1065                         if (error || swap.val || 0 != add_to_page_cache_lru(
1066                                         filepage, mapping, idx, GFP_ATOMIC)) {
1067                                 spin_unlock(&info->lock);
1068                                 page_cache_release(filepage);
1069                                 shmem_unacct_blocks(info->flags, 1);
1070                                 shmem_free_block(inode);
1071                                 filepage = NULL;
1072                                 if (error)
1073                                         goto failed;
1074                                 goto repeat;
1075                         }
1076                         info->flags |= SHMEM_PAGEIN;
1077                 }
1078
1079                 info->alloced++;
1080                 spin_unlock(&info->lock);
1081                 clear_highpage(filepage);
1082                 flush_dcache_page(filepage);
1083                 SetPageUptodate(filepage);
1084         }
1085 done:
1086         if (*pagep != filepage) {
1087                 unlock_page(filepage);
1088                 *pagep = filepage;
1089         }
1090         return 0;
1091
1092 failed:
1093         if (*pagep != filepage) {
1094                 unlock_page(filepage);
1095                 page_cache_release(filepage);
1096         }
1097         return error;
1098 }
1099
1100 struct page *shmem_nopage(struct vm_area_struct *vma, unsigned long address, int *type)
1101 {
1102         struct inode *inode = vma->vm_file->f_dentry->d_inode;
1103         struct page *page = NULL;
1104         unsigned long idx;
1105         int error;
1106
1107         idx = (address - vma->vm_start) >> PAGE_SHIFT;
1108         idx += vma->vm_pgoff;
1109         idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT;
1110
1111         error = shmem_getpage(inode, idx, &page, SGP_CACHE, type);
1112         if (error)
1113                 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS;
1114
1115         mark_page_accessed(page);
1116         return page;
1117 }
1118
1119 static int shmem_populate(struct vm_area_struct *vma,
1120         unsigned long addr, unsigned long len,
1121         pgprot_t prot, unsigned long pgoff, int nonblock)
1122 {
1123         struct inode *inode = vma->vm_file->f_dentry->d_inode;
1124         struct mm_struct *mm = vma->vm_mm;
1125         enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE;
1126         unsigned long size;
1127
1128         size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1129         if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size)
1130                 return -EINVAL;
1131
1132         while ((long) len > 0) {
1133                 struct page *page = NULL;
1134                 int err;
1135                 /*
1136                  * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE
1137                  */
1138                 err = shmem_getpage(inode, pgoff, &page, sgp, NULL);
1139                 if (err)
1140                         return err;
1141                 if (page) {
1142                         mark_page_accessed(page);
1143                         err = install_page(mm, vma, addr, page, prot);
1144                         if (err) {
1145                                 page_cache_release(page);
1146                                 return err;
1147                         }
1148                 } else if (nonblock) {
1149                         err = install_file_pte(mm, vma, addr, pgoff, prot);
1150                         if (err)
1151                                 return err;
1152                 }
1153
1154                 len -= PAGE_SIZE;
1155                 addr += PAGE_SIZE;
1156                 pgoff++;
1157         }
1158         return 0;
1159 }
1160
1161 #ifdef CONFIG_NUMA
1162 int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1163 {
1164         struct inode *i = vma->vm_file->f_dentry->d_inode;
1165         return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1166 }
1167
1168 struct mempolicy *
1169 shmem_get_policy(struct vm_area_struct *vma, unsigned long addr)
1170 {
1171         struct inode *i = vma->vm_file->f_dentry->d_inode;
1172         unsigned long idx;
1173
1174         idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1175         return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1176 }
1177 #endif
1178
1179 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1180 {
1181         struct inode *inode = file->f_dentry->d_inode;
1182         struct shmem_inode_info *info = SHMEM_I(inode);
1183         int retval = -ENOMEM;
1184
1185         spin_lock(&info->lock);
1186         if (lock && !(info->flags & VM_LOCKED)) {
1187                 if (!user_shm_lock(inode->i_size, user))
1188                         goto out_nomem;
1189                 info->flags |= VM_LOCKED;
1190         }
1191         if (!lock && (info->flags & VM_LOCKED) && user) {
1192                 user_shm_unlock(inode->i_size, user);
1193                 info->flags &= ~VM_LOCKED;
1194         }
1195         retval = 0;
1196 out_nomem:
1197         spin_unlock(&info->lock);
1198         return retval;
1199 }
1200
1201 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1202 {
1203         file_accessed(file);
1204         vma->vm_ops = &shmem_vm_ops;
1205         return 0;
1206 }
1207
1208 static struct inode *
1209 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1210 {
1211         struct inode *inode;
1212         struct shmem_inode_info *info;
1213         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1214
1215         if (sbinfo) {
1216                 spin_lock(&sbinfo->stat_lock);
1217                 if (!sbinfo->free_inodes) {
1218                         spin_unlock(&sbinfo->stat_lock);
1219                         return NULL;
1220                 }
1221                 sbinfo->free_inodes--;
1222                 spin_unlock(&sbinfo->stat_lock);
1223         }
1224
1225         inode = new_inode(sb);
1226         if (inode) {
1227                 inode->i_mode = mode;
1228                 inode->i_uid = current->fsuid;
1229                 inode->i_gid = current->fsgid;
1230                 inode->i_blksize = PAGE_CACHE_SIZE;
1231                 inode->i_blocks = 0;
1232                 inode->i_mapping->a_ops = &shmem_aops;
1233                 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1234                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1235                 info = SHMEM_I(inode);
1236                 memset(info, 0, (char *)inode - (char *)info);
1237                 spin_lock_init(&info->lock);
1238                 mpol_shared_policy_init(&info->policy);
1239                 INIT_LIST_HEAD(&info->swaplist);
1240
1241                 switch (mode & S_IFMT) {
1242                 default:
1243                         inode->i_op = &shmem_special_inode_operations;
1244                         init_special_inode(inode, mode, dev);
1245                         break;
1246                 case S_IFREG:
1247                         inode->i_op = &shmem_inode_operations;
1248                         inode->i_fop = &shmem_file_operations;
1249                         break;
1250                 case S_IFDIR:
1251                         inode->i_nlink++;
1252                         /* Some things misbehave if size == 0 on a directory */
1253                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
1254                         inode->i_op = &shmem_dir_inode_operations;
1255                         inode->i_fop = &simple_dir_operations;
1256                         break;
1257                 case S_IFLNK:
1258                         break;
1259                 }
1260         } else if (sbinfo) {
1261                 spin_lock(&sbinfo->stat_lock);
1262                 sbinfo->free_inodes++;
1263                 spin_unlock(&sbinfo->stat_lock);
1264         }
1265         return inode;
1266 }
1267
1268 #ifdef CONFIG_TMPFS
1269
1270 static int shmem_set_size(struct shmem_sb_info *sbinfo,
1271                           unsigned long max_blocks, unsigned long max_inodes)
1272 {
1273         int error;
1274         unsigned long blocks, inodes;
1275
1276         spin_lock(&sbinfo->stat_lock);
1277         blocks = sbinfo->max_blocks - sbinfo->free_blocks;
1278         inodes = sbinfo->max_inodes - sbinfo->free_inodes;
1279         error = -EINVAL;
1280         if (max_blocks < blocks)
1281                 goto out;
1282         if (max_inodes < inodes)
1283                 goto out;
1284         error = 0;
1285         sbinfo->max_blocks  = max_blocks;
1286         sbinfo->free_blocks = max_blocks - blocks;
1287         sbinfo->max_inodes  = max_inodes;
1288         sbinfo->free_inodes = max_inodes - inodes;
1289 out:
1290         spin_unlock(&sbinfo->stat_lock);
1291         return error;
1292 }
1293
1294 static struct inode_operations shmem_symlink_inode_operations;
1295 static struct inode_operations shmem_symlink_inline_operations;
1296
1297 /*
1298  * Normally tmpfs makes no use of shmem_prepare_write, but it
1299  * lets a tmpfs file be used read-write below the loop driver.
1300  */
1301 static int
1302 shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1303 {
1304         struct inode *inode = page->mapping->host;
1305         return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL);
1306 }
1307
1308 static ssize_t
1309 shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1310 {
1311         struct inode    *inode = file->f_dentry->d_inode;
1312         loff_t          pos;
1313         unsigned long   written;
1314         ssize_t         err;
1315
1316         if ((ssize_t) count < 0)
1317                 return -EINVAL;
1318
1319         if (!access_ok(VERIFY_READ, buf, count))
1320                 return -EFAULT;
1321
1322         down(&inode->i_sem);
1323
1324         pos = *ppos;
1325         written = 0;
1326
1327         err = generic_write_checks(file, &pos, &count, 0);
1328         if (err || !count)
1329                 goto out;
1330
1331         err = remove_suid(file->f_dentry);
1332         if (err)
1333                 goto out;
1334
1335         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1336
1337         do {
1338                 struct page *page = NULL;
1339                 unsigned long bytes, index, offset;
1340                 char *kaddr;
1341                 int left;
1342
1343                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1344                 index = pos >> PAGE_CACHE_SHIFT;
1345                 bytes = PAGE_CACHE_SIZE - offset;
1346                 if (bytes > count)
1347                         bytes = count;
1348
1349                 /*
1350                  * We don't hold page lock across copy from user -
1351                  * what would it guard against? - so no deadlock here.
1352                  * But it still may be a good idea to prefault below.
1353                  */
1354
1355                 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1356                 if (err)
1357                         break;
1358
1359                 left = bytes;
1360                 if (PageHighMem(page)) {
1361                         volatile unsigned char dummy;
1362                         __get_user(dummy, buf);
1363                         __get_user(dummy, buf + bytes - 1);
1364
1365                         kaddr = kmap_atomic(page, KM_USER0);
1366                         left = __copy_from_user_inatomic(kaddr + offset,
1367                                                         buf, bytes);
1368                         kunmap_atomic(kaddr, KM_USER0);
1369                 }
1370                 if (left) {
1371                         kaddr = kmap(page);
1372                         left = __copy_from_user(kaddr + offset, buf, bytes);
1373                         kunmap(page);
1374                 }
1375
1376                 written += bytes;
1377                 count -= bytes;
1378                 pos += bytes;
1379                 buf += bytes;
1380                 if (pos > inode->i_size)
1381                         i_size_write(inode, pos);
1382
1383                 flush_dcache_page(page);
1384                 set_page_dirty(page);
1385                 mark_page_accessed(page);
1386                 page_cache_release(page);
1387
1388                 if (left) {
1389                         pos -= left;
1390                         written -= left;
1391                         err = -EFAULT;
1392                         break;
1393                 }
1394
1395                 /*
1396                  * Our dirty pages are not counted in nr_dirty,
1397                  * and we do not attempt to balance dirty pages.
1398                  */
1399
1400                 cond_resched();
1401         } while (count);
1402
1403         *ppos = pos;
1404         if (written)
1405                 err = written;
1406 out:
1407         up(&inode->i_sem);
1408         return err;
1409 }
1410
1411 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1412 {
1413         struct inode *inode = filp->f_dentry->d_inode;
1414         struct address_space *mapping = inode->i_mapping;
1415         unsigned long index, offset;
1416
1417         index = *ppos >> PAGE_CACHE_SHIFT;
1418         offset = *ppos & ~PAGE_CACHE_MASK;
1419
1420         for (;;) {
1421                 struct page *page = NULL;
1422                 unsigned long end_index, nr, ret;
1423                 loff_t i_size = i_size_read(inode);
1424
1425                 end_index = i_size >> PAGE_CACHE_SHIFT;
1426                 if (index > end_index)
1427                         break;
1428                 if (index == end_index) {
1429                         nr = i_size & ~PAGE_CACHE_MASK;
1430                         if (nr <= offset)
1431                                 break;
1432                 }
1433
1434                 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1435                 if (desc->error) {
1436                         if (desc->error == -EINVAL)
1437                                 desc->error = 0;
1438                         break;
1439                 }
1440
1441                 /*
1442                  * We must evaluate after, since reads (unlike writes)
1443                  * are called without i_sem protection against truncate
1444                  */
1445                 nr = PAGE_CACHE_SIZE;
1446                 i_size = i_size_read(inode);
1447                 end_index = i_size >> PAGE_CACHE_SHIFT;
1448                 if (index == end_index) {
1449                         nr = i_size & ~PAGE_CACHE_MASK;
1450                         if (nr <= offset) {
1451                                 if (page)
1452                                         page_cache_release(page);
1453                                 break;
1454                         }
1455                 }
1456                 nr -= offset;
1457
1458                 if (page) {
1459                         /*
1460                          * If users can be writing to this page using arbitrary
1461                          * virtual addresses, take care about potential aliasing
1462                          * before reading the page on the kernel side.
1463                          */
1464                         if (mapping_writably_mapped(mapping))
1465                                 flush_dcache_page(page);
1466                         /*
1467                          * Mark the page accessed if we read the beginning.
1468                          */
1469                         if (!offset)
1470                                 mark_page_accessed(page);
1471                 } else
1472                         page = ZERO_PAGE(0);
1473
1474                 /*
1475                  * Ok, we have the page, and it's up-to-date, so
1476                  * now we can copy it to user space...
1477                  *
1478                  * The actor routine returns how many bytes were actually used..
1479                  * NOTE! This may not be the same as how much of a user buffer
1480                  * we filled up (we may be padding etc), so we can only update
1481                  * "pos" here (the actor routine has to update the user buffer
1482                  * pointers and the remaining count).
1483                  */
1484                 ret = actor(desc, page, offset, nr);
1485                 offset += ret;
1486                 index += offset >> PAGE_CACHE_SHIFT;
1487                 offset &= ~PAGE_CACHE_MASK;
1488
1489                 page_cache_release(page);
1490                 if (ret != nr || !desc->count)
1491                         break;
1492
1493                 cond_resched();
1494         }
1495
1496         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1497         file_accessed(filp);
1498 }
1499
1500 static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1501 {
1502         read_descriptor_t desc;
1503
1504         if ((ssize_t) count < 0)
1505                 return -EINVAL;
1506         if (!access_ok(VERIFY_WRITE, buf, count))
1507                 return -EFAULT;
1508         if (!count)
1509                 return 0;
1510
1511         desc.written = 0;
1512         desc.count = count;
1513         desc.arg.buf = buf;
1514         desc.error = 0;
1515
1516         do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1517         if (desc.written)
1518                 return desc.written;
1519         return desc.error;
1520 }
1521
1522 static ssize_t shmem_file_sendfile(struct file *in_file, loff_t *ppos,
1523                          size_t count, read_actor_t actor, void *target)
1524 {
1525         read_descriptor_t desc;
1526
1527         if (!count)
1528                 return 0;
1529
1530         desc.written = 0;
1531         desc.count = count;
1532         desc.arg.data = target;
1533         desc.error = 0;
1534
1535         do_shmem_file_read(in_file, ppos, &desc, actor);
1536         if (desc.written)
1537                 return desc.written;
1538         return desc.error;
1539 }
1540
1541 static int shmem_statfs(struct super_block *sb, struct kstatfs *buf)
1542 {
1543         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1544
1545         buf->f_type = TMPFS_SUPER_MAGIC;
1546         buf->f_bsize = PAGE_CACHE_SIZE;
1547         buf->f_namelen = NAME_MAX;
1548         if (sbinfo) {
1549                 spin_lock(&sbinfo->stat_lock);
1550                 buf->f_blocks = sbinfo->max_blocks;
1551                 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1552                 buf->f_files = sbinfo->max_inodes;
1553                 buf->f_ffree = sbinfo->free_inodes;
1554                 spin_unlock(&sbinfo->stat_lock);
1555         }
1556         /* else leave those fields 0 like simple_statfs */
1557         return 0;
1558 }
1559
1560 /*
1561  * File creation. Allocate an inode, and we're done..
1562  */
1563 static int
1564 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1565 {
1566         struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1567         int error = -ENOSPC;
1568
1569         if (inode) {
1570                 if (dir->i_mode & S_ISGID) {
1571                         inode->i_gid = dir->i_gid;
1572                         if (S_ISDIR(mode))
1573                                 inode->i_mode |= S_ISGID;
1574                 }
1575                 dir->i_size += BOGO_DIRENT_SIZE;
1576                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1577                 d_instantiate(dentry, inode);
1578                 dget(dentry); /* Extra count - pin the dentry in core */
1579                 error = 0;
1580         }
1581         return error;
1582 }
1583
1584 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1585 {
1586         int error;
1587
1588         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1589                 return error;
1590         dir->i_nlink++;
1591         return 0;
1592 }
1593
1594 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1595                 struct nameidata *nd)
1596 {
1597         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1598 }
1599
1600 /*
1601  * Link a file..
1602  */
1603 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1604 {
1605         struct inode *inode = old_dentry->d_inode;
1606         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1607
1608         /*
1609          * No ordinary (disk based) filesystem counts links as inodes;
1610          * but each new link needs a new dentry, pinning lowmem, and
1611          * tmpfs dentries cannot be pruned until they are unlinked.
1612          */
1613         if (sbinfo) {
1614                 spin_lock(&sbinfo->stat_lock);
1615                 if (!sbinfo->free_inodes) {
1616                         spin_unlock(&sbinfo->stat_lock);
1617                         return -ENOSPC;
1618                 }
1619                 sbinfo->free_inodes--;
1620                 spin_unlock(&sbinfo->stat_lock);
1621         }
1622
1623         dir->i_size += BOGO_DIRENT_SIZE;
1624         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1625         inode->i_nlink++;
1626         atomic_inc(&inode->i_count);    /* New dentry reference */
1627         dget(dentry);           /* Extra pinning count for the created dentry */
1628         d_instantiate(dentry, inode);
1629         return 0;
1630 }
1631
1632 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1633 {
1634         struct inode *inode = dentry->d_inode;
1635
1636         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) {
1637                 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1638                 if (sbinfo) {
1639                         spin_lock(&sbinfo->stat_lock);
1640                         sbinfo->free_inodes++;
1641                         spin_unlock(&sbinfo->stat_lock);
1642                 }
1643         }
1644
1645         dir->i_size -= BOGO_DIRENT_SIZE;
1646         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1647         inode->i_nlink--;
1648         dput(dentry);   /* Undo the count from "create" - this does all the work */
1649         return 0;
1650 }
1651
1652 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1653 {
1654         if (!simple_empty(dentry))
1655                 return -ENOTEMPTY;
1656
1657         dir->i_nlink--;
1658         return shmem_unlink(dir, dentry);
1659 }
1660
1661 /*
1662  * The VFS layer already does all the dentry stuff for rename,
1663  * we just have to decrement the usage count for the target if
1664  * it exists so that the VFS layer correctly free's it when it
1665  * gets overwritten.
1666  */
1667 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1668 {
1669         struct inode *inode = old_dentry->d_inode;
1670         int they_are_dirs = S_ISDIR(inode->i_mode);
1671
1672         if (!simple_empty(new_dentry))
1673                 return -ENOTEMPTY;
1674
1675         if (new_dentry->d_inode) {
1676                 (void) shmem_unlink(new_dir, new_dentry);
1677                 if (they_are_dirs)
1678                         old_dir->i_nlink--;
1679         } else if (they_are_dirs) {
1680                 old_dir->i_nlink--;
1681                 new_dir->i_nlink++;
1682         }
1683
1684         old_dir->i_size -= BOGO_DIRENT_SIZE;
1685         new_dir->i_size += BOGO_DIRENT_SIZE;
1686         old_dir->i_ctime = old_dir->i_mtime =
1687         new_dir->i_ctime = new_dir->i_mtime =
1688         inode->i_ctime = CURRENT_TIME;
1689         return 0;
1690 }
1691
1692 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1693 {
1694         int error;
1695         int len;
1696         struct inode *inode;
1697         struct page *page = NULL;
1698         char *kaddr;
1699         struct shmem_inode_info *info;
1700
1701         len = strlen(symname) + 1;
1702         if (len > PAGE_CACHE_SIZE)
1703                 return -ENAMETOOLONG;
1704
1705         inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1706         if (!inode)
1707                 return -ENOSPC;
1708
1709         info = SHMEM_I(inode);
1710         inode->i_size = len-1;
1711         if (len <= (char *)inode - (char *)info) {
1712                 /* do it inline */
1713                 memcpy(info, symname, len);
1714                 inode->i_op = &shmem_symlink_inline_operations;
1715         } else {
1716                 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1717                 if (error) {
1718                         iput(inode);
1719                         return error;
1720                 }
1721                 inode->i_op = &shmem_symlink_inode_operations;
1722                 kaddr = kmap_atomic(page, KM_USER0);
1723                 memcpy(kaddr, symname, len);
1724                 kunmap_atomic(kaddr, KM_USER0);
1725                 set_page_dirty(page);
1726                 page_cache_release(page);
1727         }
1728         if (dir->i_mode & S_ISGID)
1729                 inode->i_gid = dir->i_gid;
1730         dir->i_size += BOGO_DIRENT_SIZE;
1731         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1732         d_instantiate(dentry, inode);
1733         dget(dentry);
1734         return 0;
1735 }
1736
1737 static int shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1738 {
1739         nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1740         return 0;
1741 }
1742
1743 static int shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1744 {
1745         struct page *page = NULL;
1746         int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1747         nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1748         return 0;
1749 }
1750
1751 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd)
1752 {
1753         if (!IS_ERR(nd_get_link(nd))) {
1754                 struct page *page;
1755
1756                 page = find_get_page(dentry->d_inode->i_mapping, 0);
1757                 if (!page)
1758                         BUG();
1759                 kunmap(page);
1760                 mark_page_accessed(page);
1761                 page_cache_release(page);
1762                 page_cache_release(page);
1763         }
1764 }
1765
1766 static struct inode_operations shmem_symlink_inline_operations = {
1767         .readlink       = generic_readlink,
1768         .follow_link    = shmem_follow_link_inline,
1769 #ifdef CONFIG_TMPFS_XATTR
1770         .setxattr       = generic_setxattr,
1771         .getxattr       = generic_getxattr,
1772         .listxattr      = generic_listxattr,
1773         .removexattr    = generic_removexattr,
1774 #endif
1775 };
1776
1777 static struct inode_operations shmem_symlink_inode_operations = {
1778         .truncate       = shmem_truncate,
1779         .readlink       = generic_readlink,
1780         .follow_link    = shmem_follow_link,
1781         .put_link       = shmem_put_link,
1782 #ifdef CONFIG_TMPFS_XATTR
1783         .setxattr       = generic_setxattr,
1784         .getxattr       = generic_getxattr,
1785         .listxattr      = generic_listxattr,
1786         .removexattr    = generic_removexattr,
1787 #endif
1788 };
1789
1790 static int shmem_parse_options(char *options, int *mode, uid_t *uid, gid_t *gid, unsigned long *blocks, unsigned long *inodes)
1791 {
1792         char *this_char, *value, *rest;
1793
1794         while ((this_char = strsep(&options, ",")) != NULL) {
1795                 if (!*this_char)
1796                         continue;
1797                 if ((value = strchr(this_char,'=')) != NULL) {
1798                         *value++ = 0;
1799                 } else {
1800                         printk(KERN_ERR
1801                             "tmpfs: No value for mount option '%s'\n",
1802                             this_char);
1803                         return 1;
1804                 }
1805
1806                 if (!strcmp(this_char,"size")) {
1807                         unsigned long long size;
1808                         size = memparse(value,&rest);
1809                         if (*rest == '%') {
1810                                 size <<= PAGE_SHIFT;
1811                                 size *= totalram_pages;
1812                                 do_div(size, 100);
1813                                 rest++;
1814                         }
1815                         if (*rest)
1816                                 goto bad_val;
1817                         *blocks = size >> PAGE_CACHE_SHIFT;
1818                 } else if (!strcmp(this_char,"nr_blocks")) {
1819                         *blocks = memparse(value,&rest);
1820                         if (*rest)
1821                                 goto bad_val;
1822                 } else if (!strcmp(this_char,"nr_inodes")) {
1823                         *inodes = memparse(value,&rest);
1824                         if (*rest)
1825                                 goto bad_val;
1826                 } else if (!strcmp(this_char,"mode")) {
1827                         if (!mode)
1828                                 continue;
1829                         *mode = simple_strtoul(value,&rest,8);
1830                         if (*rest)
1831                                 goto bad_val;
1832                 } else if (!strcmp(this_char,"uid")) {
1833                         if (!uid)
1834                                 continue;
1835                         *uid = simple_strtoul(value,&rest,0);
1836                         if (*rest)
1837                                 goto bad_val;
1838                 } else if (!strcmp(this_char,"gid")) {
1839                         if (!gid)
1840                                 continue;
1841                         *gid = simple_strtoul(value,&rest,0);
1842                         if (*rest)
1843                                 goto bad_val;
1844                 } else {
1845                         printk(KERN_ERR "tmpfs: Bad mount option %s\n",
1846                                this_char);
1847                         return 1;
1848                 }
1849         }
1850         return 0;
1851
1852 bad_val:
1853         printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
1854                value, this_char);
1855         return 1;
1856
1857 }
1858
1859 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
1860 {
1861         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1862         unsigned long max_blocks = 0;
1863         unsigned long max_inodes = 0;
1864
1865         if (sbinfo) {
1866                 max_blocks = sbinfo->max_blocks;
1867                 max_inodes = sbinfo->max_inodes;
1868         }
1869         if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks, &max_inodes))
1870                 return -EINVAL;
1871         /* Keep it simple: disallow limited <-> unlimited remount */
1872         if ((max_blocks || max_inodes) == !sbinfo)
1873                 return -EINVAL;
1874         /* But allow the pointless unlimited -> unlimited remount */
1875         if (!sbinfo)
1876                 return 0;
1877         return shmem_set_size(sbinfo, max_blocks, max_inodes);
1878 }
1879 #endif
1880
1881 static void shmem_put_super(struct super_block *sb)
1882 {
1883         kfree(sb->s_fs_info);
1884         sb->s_fs_info = NULL;
1885 }
1886
1887 #ifdef CONFIG_TMPFS_XATTR
1888 static struct xattr_handler *shmem_xattr_handlers[];
1889 #else
1890 #define shmem_xattr_handlers NULL
1891 #endif
1892
1893 static int shmem_fill_super(struct super_block *sb,
1894                             void *data, int silent)
1895 {
1896         struct inode *inode;
1897         struct dentry *root;
1898         int mode   = S_IRWXUGO | S_ISVTX;
1899         uid_t uid = current->fsuid;
1900         gid_t gid = current->fsgid;
1901         int err = -ENOMEM;
1902
1903 #ifdef CONFIG_TMPFS
1904         unsigned long blocks = 0;
1905         unsigned long inodes = 0;
1906
1907         /*
1908          * Per default we only allow half of the physical ram per
1909          * tmpfs instance, limiting inodes to one per page of lowmem;
1910          * but the internal instance is left unlimited.
1911          */
1912         if (!(sb->s_flags & MS_NOUSER)) {
1913                 blocks = totalram_pages / 2;
1914                 inodes = totalram_pages - totalhigh_pages;
1915                 if (inodes > blocks)
1916                         inodes = blocks;
1917
1918                 if (shmem_parse_options(data, &mode,
1919                                         &uid, &gid, &blocks, &inodes))
1920                         return -EINVAL;
1921         }
1922
1923         if (blocks || inodes) {
1924                 struct shmem_sb_info *sbinfo;
1925                 sbinfo = kmalloc(sizeof(struct shmem_sb_info), GFP_KERNEL);
1926                 if (!sbinfo)
1927                         return -ENOMEM;
1928                 sb->s_fs_info = sbinfo;
1929                 spin_lock_init(&sbinfo->stat_lock);
1930                 sbinfo->max_blocks = blocks;
1931                 sbinfo->free_blocks = blocks;
1932                 sbinfo->max_inodes = inodes;
1933                 sbinfo->free_inodes = inodes;
1934         }
1935         sb->s_xattr = shmem_xattr_handlers;
1936 #endif
1937
1938         sb->s_maxbytes = SHMEM_MAX_BYTES;
1939         sb->s_blocksize = PAGE_CACHE_SIZE;
1940         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1941         sb->s_magic = TMPFS_SUPER_MAGIC;
1942         sb->s_op = &shmem_ops;
1943         inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
1944         if (!inode)
1945                 goto failed;
1946         inode->i_uid = uid;
1947         inode->i_gid = gid;
1948         root = d_alloc_root(inode);
1949         if (!root)
1950                 goto failed_iput;
1951         sb->s_root = root;
1952         return 0;
1953
1954 failed_iput:
1955         iput(inode);
1956 failed:
1957         shmem_put_super(sb);
1958         return err;
1959 }
1960
1961 static kmem_cache_t *shmem_inode_cachep;
1962
1963 static struct inode *shmem_alloc_inode(struct super_block *sb)
1964 {
1965         struct shmem_inode_info *p;
1966         p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, SLAB_KERNEL);
1967         if (!p)
1968                 return NULL;
1969         return &p->vfs_inode;
1970 }
1971
1972 static void shmem_destroy_inode(struct inode *inode)
1973 {
1974         mpol_free_shared_policy(&SHMEM_I(inode)->policy);
1975         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
1976 }
1977
1978 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
1979 {
1980         struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
1981
1982         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
1983             SLAB_CTOR_CONSTRUCTOR) {
1984                 inode_init_once(&p->vfs_inode);
1985         }
1986 }
1987
1988 static int init_inodecache(void)
1989 {
1990         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
1991                                 sizeof(struct shmem_inode_info),
1992                                 0, 0, init_once, NULL);
1993         if (shmem_inode_cachep == NULL)
1994                 return -ENOMEM;
1995         return 0;
1996 }
1997
1998 static void destroy_inodecache(void)
1999 {
2000         if (kmem_cache_destroy(shmem_inode_cachep))
2001                 printk(KERN_INFO "shmem_inode_cache: not all structures were freed\n");
2002 }
2003
2004 static struct address_space_operations shmem_aops = {
2005         .writepage      = shmem_writepage,
2006         .set_page_dirty = __set_page_dirty_nobuffers,
2007 #ifdef CONFIG_TMPFS
2008         .prepare_write  = shmem_prepare_write,
2009         .commit_write   = simple_commit_write,
2010 #endif
2011 };
2012
2013 static struct file_operations shmem_file_operations = {
2014         .mmap           = shmem_mmap,
2015 #ifdef CONFIG_TMPFS
2016         .llseek         = generic_file_llseek,
2017         .read           = shmem_file_read,
2018         .write          = shmem_file_write,
2019         .fsync          = simple_sync_file,
2020         .sendfile       = shmem_file_sendfile,
2021 #endif
2022 };
2023
2024 static struct inode_operations shmem_inode_operations = {
2025         .truncate       = shmem_truncate,
2026         .setattr        = shmem_notify_change,
2027 #ifdef CONFIG_TMPFS_XATTR
2028         .setxattr       = generic_setxattr,
2029         .getxattr       = generic_getxattr,
2030         .listxattr      = generic_listxattr,
2031         .removexattr    = generic_removexattr,
2032 #endif
2033 };
2034
2035 static struct inode_operations shmem_dir_inode_operations = {
2036 #ifdef CONFIG_TMPFS
2037         .create         = shmem_create,
2038         .lookup         = simple_lookup,
2039         .link           = shmem_link,
2040         .unlink         = shmem_unlink,
2041         .symlink        = shmem_symlink,
2042         .mkdir          = shmem_mkdir,
2043         .rmdir          = shmem_rmdir,
2044         .mknod          = shmem_mknod,
2045         .rename         = shmem_rename,
2046 #ifdef CONFIG_TMPFS_XATTR
2047         .setxattr       = generic_setxattr,
2048         .getxattr       = generic_getxattr,
2049         .listxattr      = generic_listxattr,
2050         .removexattr    = generic_removexattr,
2051 #endif
2052 #endif
2053 };
2054
2055 static struct inode_operations shmem_special_inode_operations = {
2056 #ifdef CONFIG_TMPFS_XATTR
2057         .setxattr       = generic_setxattr,
2058         .getxattr       = generic_getxattr,
2059         .listxattr      = generic_listxattr,
2060         .removexattr    = generic_removexattr,
2061 #endif
2062 };
2063
2064 static struct super_operations shmem_ops = {
2065         .alloc_inode    = shmem_alloc_inode,
2066         .destroy_inode  = shmem_destroy_inode,
2067 #ifdef CONFIG_TMPFS
2068         .statfs         = shmem_statfs,
2069         .remount_fs     = shmem_remount_fs,
2070 #endif
2071         .delete_inode   = shmem_delete_inode,
2072         .drop_inode     = generic_delete_inode,
2073         .put_super      = shmem_put_super,
2074 };
2075
2076 static struct vm_operations_struct shmem_vm_ops = {
2077         .nopage         = shmem_nopage,
2078         .populate       = shmem_populate,
2079 #ifdef CONFIG_NUMA
2080         .set_policy     = shmem_set_policy,
2081         .get_policy     = shmem_get_policy,
2082 #endif
2083 };
2084
2085
2086 #ifdef CONFIG_TMPFS_SECURITY
2087
2088 static size_t shmem_xattr_security_list(struct inode *inode, char *list, size_t list_len,
2089                                         const char *name, size_t name_len)
2090 {
2091         return security_inode_listsecurity(inode, list, list_len);
2092 }
2093
2094 static int shmem_xattr_security_get(struct inode *inode, const char *name, void *buffer, size_t size)
2095 {
2096         if (strcmp(name, "") == 0)
2097                 return -EINVAL;
2098         return security_inode_getsecurity(inode, name, buffer, size);
2099 }
2100
2101 static int shmem_xattr_security_set(struct inode *inode, const char *name, const void *value, size_t size, int flags)
2102 {
2103         if (strcmp(name, "") == 0)
2104                 return -EINVAL;
2105         return security_inode_setsecurity(inode, name, value, size, flags);
2106 }
2107
2108 struct xattr_handler shmem_xattr_security_handler = {
2109         .prefix = XATTR_SECURITY_PREFIX,
2110         .list   = shmem_xattr_security_list,
2111         .get    = shmem_xattr_security_get,
2112         .set    = shmem_xattr_security_set,
2113 };
2114
2115 #endif  /* CONFIG_TMPFS_SECURITY */
2116
2117 #ifdef CONFIG_TMPFS_XATTR
2118
2119 static struct xattr_handler *shmem_xattr_handlers[] = {
2120 #ifdef CONFIG_TMPFS_SECURITY
2121         &shmem_xattr_security_handler,
2122 #endif
2123         NULL
2124 };
2125
2126 #endif  /* CONFIG_TMPFS_XATTR */
2127
2128 static struct super_block *shmem_get_sb(struct file_system_type *fs_type,
2129         int flags, const char *dev_name, void *data)
2130 {
2131         return get_sb_nodev(fs_type, flags, data, shmem_fill_super);
2132 }
2133
2134 static struct file_system_type tmpfs_fs_type = {
2135         .owner          = THIS_MODULE,
2136         .name           = "tmpfs",
2137         .get_sb         = shmem_get_sb,
2138         .kill_sb        = kill_litter_super,
2139 };
2140 static struct vfsmount *shm_mnt;
2141
2142 static int __init init_tmpfs(void)
2143 {
2144         int error;
2145
2146         error = init_inodecache();
2147         if (error)
2148                 goto out3;
2149
2150         error = register_filesystem(&tmpfs_fs_type);
2151         if (error) {
2152                 printk(KERN_ERR "Could not register tmpfs\n");
2153                 goto out2;
2154         }
2155 #ifdef CONFIG_TMPFS
2156         devfs_mk_dir("shm");
2157 #endif
2158         shm_mnt = do_kern_mount(tmpfs_fs_type.name, MS_NOUSER,
2159                                 tmpfs_fs_type.name, NULL);
2160         if (IS_ERR(shm_mnt)) {
2161                 error = PTR_ERR(shm_mnt);
2162                 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2163                 goto out1;
2164         }
2165         return 0;
2166
2167 out1:
2168         unregister_filesystem(&tmpfs_fs_type);
2169 out2:
2170         destroy_inodecache();
2171 out3:
2172         shm_mnt = ERR_PTR(error);
2173         return error;
2174 }
2175 module_init(init_tmpfs)
2176
2177 /*
2178  * shmem_file_setup - get an unlinked file living in tmpfs
2179  *
2180  * @name: name for dentry (to be seen in /proc/<pid>/maps
2181  * @size: size to be set for the file
2182  *
2183  */
2184 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2185 {
2186         int error;
2187         struct file *file;
2188         struct inode *inode;
2189         struct dentry *dentry, *root;
2190         struct qstr this;
2191
2192         if (IS_ERR(shm_mnt))
2193                 return (void *)shm_mnt;
2194
2195         if (size < 0 || size > SHMEM_MAX_BYTES)
2196                 return ERR_PTR(-EINVAL);
2197
2198         if (shmem_acct_size(flags, size))
2199                 return ERR_PTR(-ENOMEM);
2200
2201         error = -ENOMEM;
2202         this.name = name;
2203         this.len = strlen(name);
2204         this.hash = 0; /* will go */
2205         root = shm_mnt->mnt_root;
2206         dentry = d_alloc(root, &this);
2207         if (!dentry)
2208                 goto put_memory;
2209
2210         error = -ENFILE;
2211         file = get_empty_filp();
2212         if (!file)
2213                 goto put_dentry;
2214
2215         error = -ENOSPC;
2216         inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2217         if (!inode)
2218                 goto close_file;
2219
2220         SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2221         d_instantiate(dentry, inode);
2222         inode->i_size = size;
2223         inode->i_nlink = 0;     /* It is unlinked */
2224         file->f_vfsmnt = mntget(shm_mnt);
2225         file->f_dentry = dentry;
2226         file->f_mapping = inode->i_mapping;
2227         file->f_op = &shmem_file_operations;
2228         file->f_mode = FMODE_WRITE | FMODE_READ;
2229         return file;
2230
2231 close_file:
2232         put_filp(file);
2233 put_dentry:
2234         dput(dentry);
2235 put_memory:
2236         shmem_unacct_size(flags, size);
2237         return ERR_PTR(error);
2238 }
2239
2240 /*
2241  * shmem_zero_setup - setup a shared anonymous mapping
2242  *
2243  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2244  */
2245 int shmem_zero_setup(struct vm_area_struct *vma)
2246 {
2247         struct file *file;
2248         loff_t size = vma->vm_end - vma->vm_start;
2249
2250         file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2251         if (IS_ERR(file))
2252                 return PTR_ERR(file);
2253
2254         if (vma->vm_file)
2255                 fput(vma->vm_file);
2256         vma->vm_file = file;
2257         vma->vm_ops = &shmem_vm_ops;
2258         return 0;
2259 }