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) 2002-2003 Hugh Dickins.
  10  * Copyright (C) 2002-2003 VERITAS Software Corporation.
  11  * Copyright (C) 2004 Andi Kleen, SuSE Labs
  12  *
  13  * This file is released under the GPL.
  14  */
  15
  16 /*
  17  * This virtual memory filesystem is heavily based on the ramfs. It
  18  * extends ramfs by the ability to use swap and honor resource limits
  19  * which makes it a completely usable filesystem.
  20  */
  21
  22 #include <linux/config.h>
  23 #include <linux/module.h>
  24 #include <linux/init.h>
  25 #include <linux/devfs_fs_kernel.h>
  26 #include <linux/fs.h>
  27 #include <linux/mm.h>
  28 #include <linux/mman.h>
  29 #include <linux/file.h>
  30 #include <linux/swap.h>
  31 #include <linux/pagemap.h>
  32 #include <linux/string.h>
  33 #include <linux/slab.h>
  34 #include <linux/backing-dev.h>
  35 #include <linux/shmem_fs.h>
  36 #include <linux/mount.h>
  37 #include <linux/writeback.h>
  38 #include <linux/vfs.h>
  39 #include <linux/blkdev.h>
  40 #include <linux/security.h>
  41 #include <linux/swapops.h>
  42 #include <linux/mempolicy.h>
  43 #include <linux/namei.h>
  44 #include <asm/uaccess.h>
  45 #include <asm/div64.h>
  46 #include <asm/pgtable.h>
  47
  48 /* This magic number is used in glibc for posix shared memory */
  49 #define TMPFS_MAGIC     0x01021994
  50
  51 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
  52 #define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
  53 #define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
  54
  55 #define SHMEM_MAX_INDEX  (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
  56 #define SHMEM_MAX_BYTES  ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
  57
  58 #define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
  59
  60 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
  61 #define SHMEM_PAGEIN     VM_READ
  62 #define SHMEM_TRUNCATE   VM_WRITE
  63
  64 /* Pretend that each entry is of this size in directory's i_size */
  65 #define BOGO_DIRENT_SIZE 20
  66
  67 /* Keep swapped page count in private field of indirect struct page */
  68 #define nr_swapped              private
  69
  70 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
  71 enum sgp_type {
  72         SGP_QUICK,      /* don't try more than file page cache lookup */
  73         SGP_READ,       /* don't exceed i_size, don't allocate page */
  74         SGP_CACHE,      /* don't exceed i_size, may allocate page */
  75         SGP_WRITE,      /* may exceed i_size, may allocate page */
  76 };
  77
  78 static int shmem_getpage(struct inode *inode, unsigned long idx,
  79                          struct page **pagep, enum sgp_type sgp, int *type);
  80
  81 static inline struct page *shmem_dir_alloc(unsigned int gfp_mask)
  82 {
  83         /*
  84          * The above definition of ENTRIES_PER_PAGE, and the use of
  85          * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
  86          * might be reconsidered if it ever diverges from PAGE_SIZE.
  87          */
  88         return alloc_pages(gfp_mask, PAGE_CACHE_SHIFT-PAGE_SHIFT);
  89 }
  90
  91 static inline void shmem_dir_free(struct page *page)
  92 {
  93         __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
  94 }
  95
  96 static struct page **shmem_dir_map(struct page *page)
  97 {
  98         return (struct page **)kmap_atomic(page, KM_USER0);
  99 }
 100
 101 static inline void shmem_dir_unmap(struct page **dir)
 102 {
 103         kunmap_atomic(dir, KM_USER0);
 104 }
 105
 106 static swp_entry_t *shmem_swp_map(struct page *page)
 107 {
 108         return (swp_entry_t *)kmap_atomic(page, KM_USER1);
 109 }
 110
 111 static inline void shmem_swp_balance_unmap(void)
 112 {
 113         /*
 114          * When passing a pointer to an i_direct entry, to code which
 115          * also handles indirect entries and so will shmem_swp_unmap,
 116          * we must arrange for the preempt count to remain in balance.
 117          * What kmap_atomic of a lowmem page does depends on config
 118          * and architecture, so pretend to kmap_atomic some lowmem page.
 119          */
 120         (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
 121 }
 122
 123 static inline void shmem_swp_unmap(swp_entry_t *entry)
 124 {
 125         kunmap_atomic(entry, KM_USER1);
 126 }
 127
 128 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
 129 {
 130         return sb->s_fs_info;
 131 }
 132
 133 /*
 134  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
 135  * for shared memory and for shared anonymous (/dev/zero) mappings
 136  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
 137  * consistent with the pre-accounting of private mappings ...
 138  */
 139 static inline int shmem_acct_size(unsigned long flags, loff_t size)
 140 {
 141         return (flags & VM_ACCOUNT)?
 142                 security_vm_enough_memory(VM_ACCT(size)): 0;
 143 }
 144
 145 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
 146 {
 147         if (flags & VM_ACCOUNT)
 148                 vm_unacct_memory(VM_ACCT(size));
 149 }
 150
 151 /*
 152  * ... whereas tmpfs objects are accounted incrementally as
 153  * pages are allocated, in order to allow huge sparse files.
 154  * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
 155  * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
 156  */
 157 static inline int shmem_acct_block(unsigned long flags)
 158 {
 159         return (flags & VM_ACCOUNT)?
 160                 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
 161 }
 162
 163 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
 164 {
 165         if (!(flags & VM_ACCOUNT))
 166                 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
 167 }
 168
 169 static struct super_operations shmem_ops;
 170 static struct address_space_operations shmem_aops;
 171 static struct file_operations shmem_file_operations;
 172 static struct inode_operations shmem_inode_operations;
 173 static struct inode_operations shmem_dir_inode_operations;
 174 static struct vm_operations_struct shmem_vm_ops;
 175
 176 static struct backing_dev_info shmem_backing_dev_info = {
 177         .ra_pages       = 0,    /* No readahead */
 178         .memory_backed  = 1,    /* Does not contribute to dirty memory */
 179         .unplug_io_fn = default_unplug_io_fn,
 180 };
 181
 182 LIST_HEAD(shmem_inodes);
 183 static spinlock_t shmem_ilock = SPIN_LOCK_UNLOCKED;
 184
 185 static void shmem_free_block(struct inode *inode)
 186 {
 187         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 188         spin_lock(&sbinfo->stat_lock);
 189         sbinfo->free_blocks++;
 190         inode->i_blocks -= BLOCKS_PER_PAGE;
 191         spin_unlock(&sbinfo->stat_lock);
 192 }
 193
 194 /*
 195  * shmem_recalc_inode - recalculate the size of an inode
 196  *
 197  * @inode: inode to recalc
 198  *
 199  * We have to calculate the free blocks since the mm can drop
 200  * undirtied hole pages behind our back.
 201  *
 202  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
 203  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
 204  *
 205  * It has to be called with the spinlock held.
 206  */
 207 static void shmem_recalc_inode(struct inode *inode)
 208 {
 209         struct shmem_inode_info *info = SHMEM_I(inode);
 210         long freed;
 211
 212         freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
 213         if (freed > 0) {
 214                 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 215                 info->alloced -= freed;
 216                 spin_lock(&sbinfo->stat_lock);
 217                 sbinfo->free_blocks += freed;
 218                 inode->i_blocks -= freed*BLOCKS_PER_PAGE;
 219                 spin_unlock(&sbinfo->stat_lock);
 220                 shmem_unacct_blocks(info->flags, freed);
 221         }
 222 }
 223
 224 /*
 225  * shmem_swp_entry - find the swap vector position in the info structure
 226  *
 227  * @info:  info structure for the inode
 228  * @index: index of the page to find
 229  * @page:  optional page to add to the structure. Has to be preset to
 230  *         all zeros
 231  *
 232  * If there is no space allocated yet it will return NULL when
 233  * page is NULL, else it will use the page for the needed block,
 234  * setting it to NULL on return to indicate that it has been used.
 235  *
 236  * The swap vector is organized the following way:
 237  *
 238  * There are SHMEM_NR_DIRECT entries directly stored in the
 239  * shmem_inode_info structure. So small files do not need an addional
 240  * allocation.
 241  *
 242  * For pages with index > SHMEM_NR_DIRECT there is the pointer
 243  * i_indirect which points to a page which holds in the first half
 244  * doubly indirect blocks, in the second half triple indirect blocks:
 245  *
 246  * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
 247  * following layout (for SHMEM_NR_DIRECT == 16):
 248  *
 249  * i_indirect -> dir --> 16-19
 250  *            |      +-> 20-23
 251  *            |
 252  *            +-->dir2 --> 24-27
 253  *            |        +-> 28-31
 254  *            |        +-> 32-35
 255  *            |        +-> 36-39
 256  *            |
 257  *            +-->dir3 --> 40-43
 258  *                     +-> 44-47
 259  *                     +-> 48-51
 260  *                     +-> 52-55
 261  */
 262 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
 263 {
 264         unsigned long offset;
 265         struct page **dir;
 266         struct page *subdir;
 267
 268         if (index < SHMEM_NR_DIRECT) {
 269                 shmem_swp_balance_unmap();
 270                 return info->i_direct+index;
 271         }
 272         if (!info->i_indirect) {
 273                 if (page) {
 274                         info->i_indirect = *page;
 275                         *page = NULL;
 276                 }
 277                 return NULL;                    /* need another page */
 278         }
 279
 280         index -= SHMEM_NR_DIRECT;
 281         offset = index % ENTRIES_PER_PAGE;
 282         index /= ENTRIES_PER_PAGE;
 283         dir = shmem_dir_map(info->i_indirect);
 284
 285         if (index >= ENTRIES_PER_PAGE/2) {
 286                 index -= ENTRIES_PER_PAGE/2;
 287                 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
 288                 index %= ENTRIES_PER_PAGE;
 289                 subdir = *dir;
 290                 if (!subdir) {
 291                         if (page) {
 292                                 *dir = *page;
 293                                 *page = NULL;
 294                         }
 295                         shmem_dir_unmap(dir);
 296                         return NULL;            /* need another page */
 297                 }
 298                 shmem_dir_unmap(dir);
 299                 dir = shmem_dir_map(subdir);
 300         }
 301
 302         dir += index;
 303         subdir = *dir;
 304         if (!subdir) {
 305                 if (!page || !(subdir = *page)) {
 306                         shmem_dir_unmap(dir);
 307                         return NULL;            /* need a page */
 308                 }
 309                 *dir = subdir;
 310                 *page = NULL;
 311         }
 312         shmem_dir_unmap(dir);
 313         return shmem_swp_map(subdir) + offset;
 314 }
 315
 316 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
 317 {
 318         long incdec = value? 1: -1;
 319
 320         entry->val = value;
 321         info->swapped += incdec;
 322         if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT)
 323                 kmap_atomic_to_page(entry)->nr_swapped += incdec;
 324 }
 325
 326 /*
 327  * shmem_swp_alloc - get the position of the swap entry for the page.
 328  *                   If it does not exist allocate the entry.
 329  *
 330  * @info:       info structure for the inode
 331  * @index:      index of the page to find
 332  * @sgp:        check and recheck i_size? skip allocation?
 333  */
 334 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
 335 {
 336         struct inode *inode = &info->vfs_inode;
 337         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 338         struct page *page = NULL;
 339         swp_entry_t *entry;
 340         static const swp_entry_t unswapped = { 0 };
 341
 342         if (sgp != SGP_WRITE &&
 343             ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
 344                 return ERR_PTR(-EINVAL);
 345
 346         while (!(entry = shmem_swp_entry(info, index, &page))) {
 347                 if (sgp == SGP_READ)
 348                         return (swp_entry_t *) &unswapped;
 349                 /*
 350                  * Test free_blocks against 1 not 0, since we have 1 data
 351                  * page (and perhaps indirect index pages) yet to allocate:
 352                  * a waste to allocate index if we cannot allocate data.
 353                  */
 354                 spin_lock(&sbinfo->stat_lock);
 355                 if (sbinfo->free_blocks <= 1) {
 356                         spin_unlock(&sbinfo->stat_lock);
 357                         return ERR_PTR(-ENOSPC);
 358                 }
 359                 sbinfo->free_blocks--;
 360                 inode->i_blocks += BLOCKS_PER_PAGE;
 361                 spin_unlock(&sbinfo->stat_lock);
 362
 363                 spin_unlock(&info->lock);
 364                 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
 365                 if (page) {
 366                         clear_highpage(page);
 367                         page->nr_swapped = 0;
 368                 }
 369                 spin_lock(&info->lock);
 370
 371                 if (!page) {
 372                         shmem_free_block(inode);
 373                         return ERR_PTR(-ENOMEM);
 374                 }
 375                 if (sgp != SGP_WRITE &&
 376                     ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
 377                         entry = ERR_PTR(-EINVAL);
 378                         break;
 379                 }
 380                 if (info->next_index <= index)
 381                         info->next_index = index + 1;
 382         }
 383         if (page) {
 384                 /* another task gave its page, or truncated the file */
 385                 shmem_free_block(inode);
 386                 shmem_dir_free(page);
 387         }
 388         if (info->next_index <= index && !IS_ERR(entry))
 389                 info->next_index = index + 1;
 390         return entry;
 391 }
 392
 393 /*
 394  * shmem_free_swp - free some swap entries in a directory
 395  *
 396  * @dir:   pointer to the directory
 397  * @edir:  pointer after last entry of the directory
 398  */
 399 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir)
 400 {
 401         swp_entry_t *ptr;
 402         int freed = 0;
 403
 404         for (ptr = dir; ptr < edir; ptr++) {
 405                 if (ptr->val) {
 406                         free_swap_and_cache(*ptr);
 407                         *ptr = (swp_entry_t){0};
 408                         freed++;
 409                 }
 410         }
 411         return freed;
 412 }
 413
 414 static void shmem_truncate(struct inode *inode)
 415 {
 416         struct shmem_inode_info *info = SHMEM_I(inode);
 417         unsigned long idx;
 418         unsigned long size;
 419         unsigned long limit;
 420         unsigned long stage;
 421         struct page **dir;
 422         struct page *subdir;
 423         struct page *empty;
 424         swp_entry_t *ptr;
 425         int offset;
 426         int freed;
 427
 428         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 429         idx = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 430         if (idx >= info->next_index)
 431                 return;
 432
 433         spin_lock(&info->lock);
 434         info->flags |= SHMEM_TRUNCATE;
 435         limit = info->next_index;
 436         info->next_index = idx;
 437         if (info->swapped && idx < SHMEM_NR_DIRECT) {
 438                 ptr = info->i_direct;
 439                 size = limit;
 440                 if (size > SHMEM_NR_DIRECT)
 441                         size = SHMEM_NR_DIRECT;
 442                 info->swapped -= shmem_free_swp(ptr+idx, ptr+size);
 443         }
 444         if (!info->i_indirect)
 445                 goto done2;
 446
 447         BUG_ON(limit <= SHMEM_NR_DIRECT);
 448         limit -= SHMEM_NR_DIRECT;
 449         idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
 450         offset = idx % ENTRIES_PER_PAGE;
 451         idx -= offset;
 452
 453         empty = NULL;
 454         dir = shmem_dir_map(info->i_indirect);
 455         stage = ENTRIES_PER_PAGEPAGE/2;
 456         if (idx < ENTRIES_PER_PAGEPAGE/2)
 457                 dir += idx/ENTRIES_PER_PAGE;
 458         else {
 459                 dir += ENTRIES_PER_PAGE/2;
 460                 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
 461                 while (stage <= idx)
 462                         stage += ENTRIES_PER_PAGEPAGE;
 463                 if (*dir) {
 464                         subdir = *dir;
 465                         size = ((idx - ENTRIES_PER_PAGEPAGE/2) %
 466                                 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
 467                         if (!size && !offset) {
 468                                 empty = subdir;
 469                                 *dir = NULL;
 470                         }
 471                         shmem_dir_unmap(dir);
 472                         dir = shmem_dir_map(subdir) + size;
 473                 } else {
 474                         offset = 0;
 475                         idx = stage;
 476                 }
 477         }
 478
 479         for (; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
 480                 if (unlikely(idx == stage)) {
 481                         shmem_dir_unmap(dir-1);
 482                         dir = shmem_dir_map(info->i_indirect) +
 483                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
 484                         while (!*dir) {
 485                                 dir++;
 486                                 idx += ENTRIES_PER_PAGEPAGE;
 487                                 if (idx >= limit)
 488                                         goto done1;
 489                         }
 490                         stage = idx + ENTRIES_PER_PAGEPAGE;
 491                         subdir = *dir;
 492                         *dir = NULL;
 493                         shmem_dir_unmap(dir);
 494                         if (empty) {
 495                                 shmem_dir_free(empty);
 496                                 shmem_free_block(inode);
 497                         }
 498                         empty = subdir;
 499                         cond_resched_lock(&info->lock);
 500                         dir = shmem_dir_map(subdir);
 501                 }
 502                 subdir = *dir;
 503                 if (subdir && subdir->nr_swapped) {
 504                         ptr = shmem_swp_map(subdir);
 505                         size = limit - idx;
 506                         if (size > ENTRIES_PER_PAGE)
 507                                 size = ENTRIES_PER_PAGE;
 508                         freed = shmem_free_swp(ptr+offset, ptr+size);
 509                         shmem_swp_unmap(ptr);
 510                         info->swapped -= freed;
 511                         subdir->nr_swapped -= freed;
 512                         BUG_ON(subdir->nr_swapped > offset);
 513                 }
 514                 if (offset)
 515                         offset = 0;
 516                 else if (subdir) {
 517                         *dir = NULL;
 518                         shmem_dir_free(subdir);
 519                         shmem_free_block(inode);
 520                 }
 521         }
 522 done1:
 523         shmem_dir_unmap(dir-1);
 524         if (empty) {
 525                 shmem_dir_free(empty);
 526                 shmem_free_block(inode);
 527         }
 528         if (info->next_index <= SHMEM_NR_DIRECT) {
 529                 shmem_dir_free(info->i_indirect);
 530                 info->i_indirect = NULL;
 531                 shmem_free_block(inode);
 532         }
 533 done2:
 534         BUG_ON(info->swapped > info->next_index);
 535         if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
 536                 /*
 537                  * Call truncate_inode_pages again: racing shmem_unuse_inode
 538                  * may have swizzled a page in from swap since vmtruncate or
 539                  * generic_delete_inode did it, before we lowered next_index.
 540                  * Also, though shmem_getpage checks i_size before adding to
 541                  * cache, no recheck after: so fix the narrow window there too.
 542                  */
 543                 spin_unlock(&info->lock);
 544                 truncate_inode_pages(inode->i_mapping, inode->i_size);
 545                 spin_lock(&info->lock);
 546         }
 547         info->flags &= ~SHMEM_TRUNCATE;
 548         shmem_recalc_inode(inode);
 549         spin_unlock(&info->lock);
 550 }
 551
 552 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
 553 {
 554         struct inode *inode = dentry->d_inode;
 555         struct page *page = NULL;
 556         int error;
 557
 558         if (attr->ia_valid & ATTR_SIZE) {
 559                 if (attr->ia_size < inode->i_size) {
 560                         /*
 561                          * If truncating down to a partial page, then
 562                          * if that page is already allocated, hold it
 563                          * in memory until the truncation is over, so
 564                          * truncate_partial_page cannnot miss it were
 565                          * it assigned to swap.
 566                          */
 567                         if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
 568                                 (void) shmem_getpage(inode,
 569                                         attr->ia_size>>PAGE_CACHE_SHIFT,
 570                                                 &page, SGP_READ, NULL);
 571                         }
 572                         /*
 573                          * Reset SHMEM_PAGEIN flag so that shmem_truncate can
 574                          * detect if any pages might have been added to cache
 575                          * after truncate_inode_pages.  But we needn't bother
 576                          * if it's being fully truncated to zero-length: the
 577                          * nrpages check is efficient enough in that case.
 578                          */
 579                         if (attr->ia_size) {
 580                                 struct shmem_inode_info *info = SHMEM_I(inode);
 581                                 spin_lock(&info->lock);
 582                                 info->flags &= ~SHMEM_PAGEIN;
 583                                 spin_unlock(&info->lock);
 584                         }
 585                 }
 586         }
 587
 588         error = inode_change_ok(inode, attr);
 589         if (!error)
 590                 error = inode_setattr(inode, attr);
 591         if (page)
 592                 page_cache_release(page);
 593         return error;
 594 }
 595
 596 static void shmem_delete_inode(struct inode *inode)
 597 {
 598         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 599         struct shmem_inode_info *info = SHMEM_I(inode);
 600
 601         if (inode->i_op->truncate == shmem_truncate) {
 602                 spin_lock(&shmem_ilock);
 603                 list_del(&info->list);
 604                 spin_unlock(&shmem_ilock);
 605                 shmem_unacct_size(info->flags, inode->i_size);
 606                 inode->i_size = 0;
 607                 shmem_truncate(inode);
 608         }
 609         BUG_ON(inode->i_blocks);
 610         spin_lock(&sbinfo->stat_lock);
 611         sbinfo->free_inodes++;
 612         spin_unlock(&sbinfo->stat_lock);
 613         clear_inode(inode);
 614 }
 615
 616 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
 617 {
 618         swp_entry_t *ptr;
 619
 620         for (ptr = dir; ptr < edir; ptr++) {
 621                 if (ptr->val == entry.val)
 622                         return ptr - dir;
 623         }
 624         return -1;
 625 }
 626
 627 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
 628 {
 629         struct inode *inode;
 630         unsigned long idx;
 631         unsigned long size;
 632         unsigned long limit;
 633         unsigned long stage;
 634         struct page **dir;
 635         struct page *subdir;
 636         swp_entry_t *ptr;
 637         int offset;
 638
 639         idx = 0;
 640         ptr = info->i_direct;
 641         spin_lock(&info->lock);
 642         limit = info->next_index;
 643         size = limit;
 644         if (size > SHMEM_NR_DIRECT)
 645                 size = SHMEM_NR_DIRECT;
 646         offset = shmem_find_swp(entry, ptr, ptr+size);
 647         if (offset >= 0) {
 648                 shmem_swp_balance_unmap();
 649                 goto found;
 650         }
 651         if (!info->i_indirect)
 652                 goto lost2;
 653         /* we might be racing with shmem_truncate */
 654         if (limit <= SHMEM_NR_DIRECT)
 655                 goto lost2;
 656
 657         dir = shmem_dir_map(info->i_indirect);
 658         stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
 659
 660         for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
 661                 if (unlikely(idx == stage)) {
 662                         shmem_dir_unmap(dir-1);
 663                         dir = shmem_dir_map(info->i_indirect) +
 664                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
 665                         while (!*dir) {
 666                                 dir++;
 667                                 idx += ENTRIES_PER_PAGEPAGE;
 668                                 if (idx >= limit)
 669                                         goto lost1;
 670                         }
 671                         stage = idx + ENTRIES_PER_PAGEPAGE;
 672                         subdir = *dir;
 673                         shmem_dir_unmap(dir);
 674                         dir = shmem_dir_map(subdir);
 675                 }
 676                 subdir = *dir;
 677                 if (subdir && subdir->nr_swapped) {
 678                         ptr = shmem_swp_map(subdir);
 679                         size = limit - idx;
 680                         if (size > ENTRIES_PER_PAGE)
 681                                 size = ENTRIES_PER_PAGE;
 682                         offset = shmem_find_swp(entry, ptr, ptr+size);
 683                         if (offset >= 0) {
 684                                 shmem_dir_unmap(dir);
 685                                 goto found;
 686                         }
 687                         shmem_swp_unmap(ptr);
 688                 }
 689         }
 690 lost1:
 691         shmem_dir_unmap(dir-1);
 692 lost2:
 693         spin_unlock(&info->lock);
 694         return 0;
 695 found:
 696         idx += offset;
 697         inode = &info->vfs_inode;
 698         if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
 699                 info->flags |= SHMEM_PAGEIN;
 700                 shmem_swp_set(info, ptr + offset, 0);
 701         }
 702         shmem_swp_unmap(ptr);
 703         spin_unlock(&info->lock);
 704         /*
 705          * Decrement swap count even when the entry is left behind:
 706          * try_to_unuse will skip over mms, then reincrement count.
 707          */
 708         swap_free(entry);
 709         return 1;
 710 }
 711
 712 /*
 713  * shmem_unuse() search for an eventually swapped out shmem page.
 714  */
 715 int shmem_unuse(swp_entry_t entry, struct page *page)
 716 {
 717         struct list_head *p;
 718         struct shmem_inode_info *info;
 719         int found = 0;
 720
 721         spin_lock(&shmem_ilock);
 722         list_for_each(p, &shmem_inodes) {
 723                 info = list_entry(p, struct shmem_inode_info, list);
 724
 725                 if (info->swapped && shmem_unuse_inode(info, entry, page)) {
 726                         /* move head to start search for next from here */
 727                         list_move_tail(&shmem_inodes, &info->list);
 728                         found = 1;
 729                         break;
 730                 }
 731         }
 732         spin_unlock(&shmem_ilock);
 733         return found;
 734 }
 735
 736 /*
 737  * Move the page from the page cache to the swap cache.
 738  */
 739 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
 740 {
 741         struct shmem_inode_info *info;
 742         swp_entry_t *entry, swap;
 743         struct address_space *mapping;
 744         unsigned long index;
 745         struct inode *inode;
 746
 747         BUG_ON(!PageLocked(page));
 748         BUG_ON(page_mapped(page));
 749
 750         mapping = page->mapping;
 751         index = page->index;
 752         inode = mapping->host;
 753         info = SHMEM_I(inode);
 754         if (info->flags & VM_LOCKED)
 755                 goto redirty;
 756         swap = get_swap_page();
 757         if (!swap.val)
 758                 goto redirty;
 759
 760         spin_lock(&info->lock);
 761         shmem_recalc_inode(inode);
 762         if (index >= info->next_index) {
 763                 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
 764                 goto unlock;
 765         }
 766         entry = shmem_swp_entry(info, index, NULL);
 767         BUG_ON(!entry);
 768         BUG_ON(entry->val);
 769
 770         if (move_to_swap_cache(page, swap) == 0) {
 771                 shmem_swp_set(info, entry, swap.val);
 772                 shmem_swp_unmap(entry);
 773                 spin_unlock(&info->lock);
 774                 unlock_page(page);
 775                 return 0;
 776         }
 777
 778         shmem_swp_unmap(entry);
 779 unlock:
 780         spin_unlock(&info->lock);
 781         swap_free(swap);
 782 redirty:
 783         set_page_dirty(page);
 784         return WRITEPAGE_ACTIVATE;      /* Return with the page locked */
 785 }
 786
 787 #ifdef CONFIG_NUMA
 788 static struct page *shmem_swapin_async(struct shared_policy *p,
 789                                        swp_entry_t entry, unsigned long idx)
 790 {
 791         struct page *page;
 792         struct vm_area_struct pvma;
 793
 794         /* Create a pseudo vma that just contains the policy */
 795         memset(&pvma, 0, sizeof(struct vm_area_struct));
 796         pvma.vm_end = PAGE_SIZE;
 797         pvma.vm_pgoff = idx;
 798         pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
 799         page = read_swap_cache_async(entry, &pvma, 0);
 800         mpol_free(pvma.vm_policy);
 801         return page;
 802 }
 803
 804 struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry,
 805                           unsigned long idx)
 806 {
 807         struct shared_policy *p = &info->policy;
 808         int i, num;
 809         struct page *page;
 810         unsigned long offset;
 811
 812         num = valid_swaphandles(entry, &offset);
 813         for (i = 0; i < num; offset++, i++) {
 814                 page = shmem_swapin_async(p,
 815                                 swp_entry(swp_type(entry), offset), idx);
 816                 if (!page)
 817                         break;
 818                 page_cache_release(page);
 819         }
 820         lru_add_drain();        /* Push any new pages onto the LRU now */
 821         return shmem_swapin_async(p, entry, idx);
 822 }
 823
 824 static struct page *
 825 shmem_alloc_page(unsigned long gfp, struct shmem_inode_info *info,
 826                  unsigned long idx)
 827 {
 828         struct vm_area_struct pvma;
 829         struct page *page;
 830
 831         memset(&pvma, 0, sizeof(struct vm_area_struct));
 832         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
 833         pvma.vm_pgoff = idx;
 834         pvma.vm_end = PAGE_SIZE;
 835         page = alloc_page_vma(gfp, &pvma, 0);
 836         mpol_free(pvma.vm_policy);
 837         return page;
 838 }
 839 #else
 840 static inline struct page *
 841 shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
 842 {
 843         swapin_readahead(entry, 0, NULL);
 844         return read_swap_cache_async(entry, NULL, 0);
 845 }
 846
 847 static inline struct page *
 848 shmem_alloc_page(unsigned long gfp,struct shmem_inode_info *info,
 849                                  unsigned long idx)
 850 {
 851         return alloc_page(gfp);
 852 }
 853 #endif
 854
 855 /*
 856  * shmem_getpage - either get the page from swap or allocate a new one
 857  *
 858  * If we allocate a new one we do not mark it dirty. That's up to the
 859  * vm. If we swap it in we mark it dirty since we also free the swap
 860  * entry since a page cannot live in both the swap and page cache
 861  */
 862 static int shmem_getpage(struct inode *inode, unsigned long idx,
 863                         struct page **pagep, enum sgp_type sgp, int *type)
 864 {
 865         struct address_space *mapping = inode->i_mapping;
 866         struct shmem_inode_info *info = SHMEM_I(inode);
 867         struct shmem_sb_info *sbinfo;
 868         struct page *filepage = *pagep;
 869         struct page *swappage;
 870         swp_entry_t *entry;
 871         swp_entry_t swap;
 872         int error, majmin = VM_FAULT_MINOR;
 873
 874         if (idx >= SHMEM_MAX_INDEX)
 875                 return -EFBIG;
 876         /*
 877          * Normally, filepage is NULL on entry, and either found
 878          * uptodate immediately, or allocated and zeroed, or read
 879          * in under swappage, which is then assigned to filepage.
 880          * But shmem_prepare_write passes in a locked filepage,
 881          * which may be found not uptodate by other callers too,
 882          * and may need to be copied from the swappage read in.
 883          */
 884 repeat:
 885         if (!filepage)
 886                 filepage = find_lock_page(mapping, idx);
 887         if (filepage && PageUptodate(filepage))
 888                 goto done;
 889         error = 0;
 890         if (sgp == SGP_QUICK)
 891                 goto failed;
 892
 893         spin_lock(&info->lock);
 894         shmem_recalc_inode(inode);
 895         entry = shmem_swp_alloc(info, idx, sgp);
 896         if (IS_ERR(entry)) {
 897                 spin_unlock(&info->lock);
 898                 error = PTR_ERR(entry);
 899                 goto failed;
 900         }
 901         swap = *entry;
 902
 903         if (swap.val) {
 904                 /* Look it up and read it in.. */
 905                 swappage = lookup_swap_cache(swap);
 906                 if (!swappage) {
 907                         shmem_swp_unmap(entry);
 908                         spin_unlock(&info->lock);
 909                         /* here we actually do the io */
 910                         if (majmin == VM_FAULT_MINOR && type)
 911                                 inc_page_state(pgmajfault);
 912                         majmin = VM_FAULT_MAJOR;
 913                         swappage = shmem_swapin(info, swap, idx);
 914                         if (!swappage) {
 915                                 spin_lock(&info->lock);
 916                                 entry = shmem_swp_alloc(info, idx, sgp);
 917                                 if (IS_ERR(entry))
 918                                         error = PTR_ERR(entry);
 919                                 else {
 920                                         if (entry->val == swap.val)
 921                                                 error = -ENOMEM;
 922                                         shmem_swp_unmap(entry);
 923                                 }
 924                                 spin_unlock(&info->lock);
 925                                 if (error)
 926                                         goto failed;
 927                                 goto repeat;
 928                         }
 929                         wait_on_page_locked(swappage);
 930                         page_cache_release(swappage);
 931                         goto repeat;
 932                 }
 933
 934                 /* We have to do this with page locked to prevent races */
 935                 if (TestSetPageLocked(swappage)) {
 936                         shmem_swp_unmap(entry);
 937                         spin_unlock(&info->lock);
 938                         wait_on_page_locked(swappage);
 939                         page_cache_release(swappage);
 940                         goto repeat;
 941                 }
 942                 if (PageWriteback(swappage)) {
 943                         shmem_swp_unmap(entry);
 944                         spin_unlock(&info->lock);
 945                         wait_on_page_writeback(swappage);
 946                         unlock_page(swappage);
 947                         page_cache_release(swappage);
 948                         goto repeat;
 949                 }
 950                 if (!PageUptodate(swappage)) {
 951                         shmem_swp_unmap(entry);
 952                         spin_unlock(&info->lock);
 953                         unlock_page(swappage);
 954                         page_cache_release(swappage);
 955                         error = -EIO;
 956                         goto failed;
 957                 }
 958
 959                 if (filepage) {
 960                         shmem_swp_set(info, entry, 0);
 961                         shmem_swp_unmap(entry);
 962                         delete_from_swap_cache(swappage);
 963                         spin_unlock(&info->lock);
 964                         copy_highpage(filepage, swappage);
 965                         unlock_page(swappage);
 966                         page_cache_release(swappage);
 967                         flush_dcache_page(filepage);
 968                         SetPageUptodate(filepage);
 969                         set_page_dirty(filepage);
 970                         swap_free(swap);
 971                 } else if (!(error = move_from_swap_cache(
 972                                 swappage, idx, mapping))) {
 973                         info->flags |= SHMEM_PAGEIN;
 974                         shmem_swp_set(info, entry, 0);
 975                         shmem_swp_unmap(entry);
 976                         spin_unlock(&info->lock);
 977                         filepage = swappage;
 978                         swap_free(swap);
 979                 } else {
 980                         shmem_swp_unmap(entry);
 981                         spin_unlock(&info->lock);
 982                         unlock_page(swappage);
 983                         page_cache_release(swappage);
 984                         if (error == -ENOMEM) {
 985                                 /* let kswapd refresh zone for GFP_ATOMICs */
 986                                 blk_congestion_wait(WRITE, HZ/50);
 987                         }
 988                         goto repeat;
 989                 }
 990         } else if (sgp == SGP_READ && !filepage) {
 991                 shmem_swp_unmap(entry);
 992                 filepage = find_get_page(mapping, idx);
 993                 if (filepage &&
 994                     (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
 995                         spin_unlock(&info->lock);
 996                         wait_on_page_locked(filepage);
 997                         page_cache_release(filepage);
 998                         filepage = NULL;
 999                         goto repeat;
1000                 }
1001                 spin_unlock(&info->lock);
1002         } else {
1003                 shmem_swp_unmap(entry);
1004                 sbinfo = SHMEM_SB(inode->i_sb);
1005                 spin_lock(&sbinfo->stat_lock);
1006                 if (sbinfo->free_blocks == 0 || shmem_acct_block(info->flags)) {
1007                         spin_unlock(&sbinfo->stat_lock);
1008                         spin_unlock(&info->lock);
1009                         error = -ENOSPC;
1010                         goto failed;
1011                 }
1012                 sbinfo->free_blocks--;
1013                 inode->i_blocks += BLOCKS_PER_PAGE;
1014                 spin_unlock(&sbinfo->stat_lock);
1015
1016                 if (!filepage) {
1017                         spin_unlock(&info->lock);
1018                         filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
1019                                                     info,
1020                                                     idx);
1021                         if (!filepage) {
1022                                 shmem_unacct_blocks(info->flags, 1);
1023                                 shmem_free_block(inode);
1024                                 error = -ENOMEM;
1025                                 goto failed;
1026                         }
1027
1028                         spin_lock(&info->lock);
1029                         entry = shmem_swp_alloc(info, idx, sgp);
1030                         if (IS_ERR(entry))
1031                                 error = PTR_ERR(entry);
1032                         else {
1033                                 swap = *entry;
1034                                 shmem_swp_unmap(entry);
1035                         }
1036                         if (error || swap.val || 0 != add_to_page_cache_lru(
1037                                         filepage, mapping, idx, GFP_ATOMIC)) {
1038                                 spin_unlock(&info->lock);
1039                                 page_cache_release(filepage);
1040                                 shmem_unacct_blocks(info->flags, 1);
1041                                 shmem_free_block(inode);
1042                                 filepage = NULL;
1043                                 if (error)
1044                                         goto failed;
1045                                 goto repeat;
1046                         }
1047                         info->flags |= SHMEM_PAGEIN;
1048                 }
1049
1050                 info->alloced++;
1051                 spin_unlock(&info->lock);
1052                 clear_highpage(filepage);
1053                 flush_dcache_page(filepage);
1054                 SetPageUptodate(filepage);
1055         }
1056 done:
1057         if (!*pagep) {
1058                 if (filepage) {
1059                         unlock_page(filepage);
1060                         *pagep = filepage;
1061                 } else
1062                         *pagep = ZERO_PAGE(0);
1063         }
1064         if (type)
1065                 *type = majmin;
1066         return 0;
1067
1068 failed:
1069         if (*pagep != filepage) {
1070                 unlock_page(filepage);
1071                 page_cache_release(filepage);
1072         }
1073         return error;
1074 }
1075
1076 struct page *shmem_nopage(struct vm_area_struct *vma, unsigned long address, int *type)
1077 {
1078         struct inode *inode = vma->vm_file->f_dentry->d_inode;
1079         struct page *page = NULL;
1080         unsigned long idx;
1081         int error;
1082
1083         idx = (address - vma->vm_start) >> PAGE_SHIFT;
1084         idx += vma->vm_pgoff;
1085         idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT;
1086
1087         error = shmem_getpage(inode, idx, &page, SGP_CACHE, type);
1088         if (error)
1089                 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS;
1090
1091         mark_page_accessed(page);
1092         return page;
1093 }
1094
1095 static int shmem_populate(struct vm_area_struct *vma,
1096         unsigned long addr, unsigned long len,
1097         pgprot_t prot, unsigned long pgoff, int nonblock)
1098 {
1099         struct inode *inode = vma->vm_file->f_dentry->d_inode;
1100         struct mm_struct *mm = vma->vm_mm;
1101         enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE;
1102         unsigned long size;
1103
1104         size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1105         if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size)
1106                 return -EINVAL;
1107
1108         while ((long) len > 0) {
1109                 struct page *page = NULL;
1110                 int err;
1111                 /*
1112                  * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE
1113                  */
1114                 err = shmem_getpage(inode, pgoff, &page, sgp, NULL);
1115                 if (err)
1116                         return err;
1117                 if (page) {
1118                         mark_page_accessed(page);
1119                         err = install_page(mm, vma, addr, page, prot);
1120                         if (err) {
1121                                 page_cache_release(page);
1122                                 return err;
1123                         }
1124                 } else if (nonblock) {
1125                         /*
1126                          * If a nonlinear mapping then store the file page
1127                          * offset in the pte.
1128                          */
1129                         if (pgoff != linear_page_index(vma, addr)) {
1130                                 err = install_file_pte(mm, vma, addr, pgoff, prot);
1131                                 if (err)
1132                                         return err;
1133                         }
1134                 }
1135
1136                 len -= PAGE_SIZE;
1137                 addr += PAGE_SIZE;
1138                 pgoff++;
1139         }
1140         return 0;
1141 }
1142
1143 #ifdef CONFIG_NUMA
1144 int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1145 {
1146         struct inode *i = vma->vm_file->f_dentry->d_inode;
1147         return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1148 }
1149
1150 struct mempolicy *
1151 shmem_get_policy(struct vm_area_struct *vma, unsigned long addr)
1152 {
1153         struct inode *i = vma->vm_file->f_dentry->d_inode;
1154         unsigned long idx;
1155
1156         idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1157         return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1158 }
1159 #endif
1160
1161 /* Protects current->user->locked_shm from concurrent access */
1162 static spinlock_t shmem_lock_user = SPIN_LOCK_UNLOCKED;
1163
1164 int shmem_lock(struct file *file, int lock, struct user_struct * user)
1165 {
1166         struct inode *inode = file->f_dentry->d_inode;
1167         struct shmem_inode_info *info = SHMEM_I(inode);
1168         unsigned long lock_limit, locked;
1169         int retval = -ENOMEM;
1170
1171         spin_lock(&info->lock);
1172         spin_lock(&shmem_lock_user);
1173         if (lock && !(info->flags & VM_LOCKED)) {
1174                 locked = inode->i_size >> PAGE_SHIFT;
1175                 locked += user->locked_shm;
1176                 lock_limit = current->rlim[RLIMIT_MEMLOCK].rlim_cur;
1177                 lock_limit >>= PAGE_SHIFT;
1178                 if ((locked > lock_limit) && !capable(CAP_IPC_LOCK))
1179                         goto out_nomem;
1180                 /* for this branch user == current->user so it won't go away under us */
1181                 atomic_inc(&user->__count);
1182                 user->locked_shm = locked;
1183         }
1184         if (!lock && (info->flags & VM_LOCKED) && user) {
1185                 locked = inode->i_size >> PAGE_SHIFT;
1186                 user->locked_shm -= locked;
1187                 free_uid(user);
1188         }
1189         if (lock)
1190                 info->flags |= VM_LOCKED;
1191         else
1192                 info->flags &= ~VM_LOCKED;
1193         retval = 0;
1194 out_nomem:
1195         spin_unlock(&shmem_lock_user);
1196         spin_unlock(&info->lock);
1197         return retval;
1198 }
1199
1200 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1201 {
1202         file_accessed(file);
1203         vma->vm_ops = &shmem_vm_ops;
1204         return 0;
1205 }
1206
1207 static struct inode *
1208 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1209 {
1210         struct inode *inode;
1211         struct shmem_inode_info *info;
1212         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1213
1214         spin_lock(&sbinfo->stat_lock);
1215         if (!sbinfo->free_inodes) {
1216                 spin_unlock(&sbinfo->stat_lock);
1217                 return NULL;
1218         }
1219         sbinfo->free_inodes--;
1220         spin_unlock(&sbinfo->stat_lock);
1221
1222         inode = new_inode(sb);
1223         if (inode) {
1224                 inode->i_mode = mode;
1225                 inode->i_uid = current->fsuid;
1226                 inode->i_gid = current->fsgid;
1227                 inode->i_blksize = PAGE_CACHE_SIZE;
1228                 inode->i_blocks = 0;
1229                 inode->i_mapping->a_ops = &shmem_aops;
1230                 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1231                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1232                 info = SHMEM_I(inode);
1233                 memset(info, 0, (char *)inode - (char *)info);
1234                 spin_lock_init(&info->lock);
1235                 mpol_shared_policy_init(&info->policy);
1236                 switch (mode & S_IFMT) {
1237                 default:
1238                         init_special_inode(inode, mode, dev);
1239                         break;
1240                 case S_IFREG:
1241                         inode->i_op = &shmem_inode_operations;
1242                         inode->i_fop = &shmem_file_operations;
1243                         spin_lock(&shmem_ilock);
1244                         list_add_tail(&info->list, &shmem_inodes);
1245                         spin_unlock(&shmem_ilock);
1246                         break;
1247                 case S_IFDIR:
1248                         inode->i_nlink++;
1249                         /* Some things misbehave if size == 0 on a directory */
1250                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
1251                         inode->i_op = &shmem_dir_inode_operations;
1252                         inode->i_fop = &simple_dir_operations;
1253                         break;
1254                 case S_IFLNK:
1255                         break;
1256                 }
1257         }
1258         return inode;
1259 }
1260
1261 static int shmem_set_size(struct shmem_sb_info *info,
1262                           unsigned long max_blocks, unsigned long max_inodes)
1263 {
1264         int error;
1265         unsigned long blocks, inodes;
1266
1267         spin_lock(&info->stat_lock);
1268         blocks = info->max_blocks - info->free_blocks;
1269         inodes = info->max_inodes - info->free_inodes;
1270         error = -EINVAL;
1271         if (max_blocks < blocks)
1272                 goto out;
1273         if (max_inodes < inodes)
1274                 goto out;
1275         error = 0;
1276         info->max_blocks  = max_blocks;
1277         info->free_blocks = max_blocks - blocks;
1278         info->max_inodes  = max_inodes;
1279         info->free_inodes = max_inodes - inodes;
1280 out:
1281         spin_unlock(&info->stat_lock);
1282         return error;
1283 }
1284
1285 #ifdef CONFIG_TMPFS
1286
1287 static struct inode_operations shmem_symlink_inode_operations;
1288 static struct inode_operations shmem_symlink_inline_operations;
1289
1290 /*
1291  * Normally tmpfs makes no use of shmem_prepare_write, but it
1292  * lets a tmpfs file be used read-write below the loop driver.
1293  */
1294 static int
1295 shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1296 {
1297         struct inode *inode = page->mapping->host;
1298         return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL);
1299 }
1300
1301 static ssize_t
1302 shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1303 {
1304         struct inode    *inode = file->f_dentry->d_inode;
1305         loff_t          pos;
1306         unsigned long   written;
1307         int             err;
1308
1309         if ((ssize_t) count < 0)
1310                 return -EINVAL;
1311
1312         if (!access_ok(VERIFY_READ, buf, count))
1313                 return -EFAULT;
1314
1315         down(&inode->i_sem);
1316
1317         pos = *ppos;
1318         written = 0;
1319
1320         err = generic_write_checks(file, &pos, &count, 0);
1321         if (err || !count)
1322                 goto out;
1323
1324         err = remove_suid(file->f_dentry);
1325         if (err)
1326                 goto out;
1327
1328         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1329
1330         do {
1331                 struct page *page = NULL;
1332                 unsigned long bytes, index, offset;
1333                 char *kaddr;
1334                 int left;
1335
1336                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1337                 index = pos >> PAGE_CACHE_SHIFT;
1338                 bytes = PAGE_CACHE_SIZE - offset;
1339                 if (bytes > count)
1340                         bytes = count;
1341
1342                 /*
1343                  * We don't hold page lock across copy from user -
1344                  * what would it guard against? - so no deadlock here.
1345                  * But it still may be a good idea to prefault below.
1346                  */
1347
1348                 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1349                 if (err)
1350                         break;
1351
1352                 left = bytes;
1353                 if (PageHighMem(page)) {
1354                         volatile unsigned char dummy;
1355                         __get_user(dummy, buf);
1356                         __get_user(dummy, buf + bytes - 1);
1357
1358                         kaddr = kmap_atomic(page, KM_USER0);
1359                         left = __copy_from_user(kaddr + offset, buf, bytes);
1360                         kunmap_atomic(kaddr, KM_USER0);
1361                 }
1362                 if (left) {
1363                         kaddr = kmap(page);
1364                         left = __copy_from_user(kaddr + offset, buf, bytes);
1365                         kunmap(page);
1366                 }
1367
1368                 written += bytes;
1369                 count -= bytes;
1370                 pos += bytes;
1371                 buf += bytes;
1372                 if (pos > inode->i_size)
1373                         i_size_write(inode, pos);
1374
1375                 flush_dcache_page(page);
1376                 set_page_dirty(page);
1377                 mark_page_accessed(page);
1378                 page_cache_release(page);
1379
1380                 if (left) {
1381                         pos -= left;
1382                         written -= left;
1383                         err = -EFAULT;
1384                         break;
1385                 }
1386
1387                 /*
1388                  * Our dirty pages are not counted in nr_dirty,
1389                  * and we do not attempt to balance dirty pages.
1390                  */
1391
1392                 cond_resched();
1393         } while (count);
1394
1395         *ppos = pos;
1396         if (written)
1397                 err = written;
1398 out:
1399         up(&inode->i_sem);
1400         return err;
1401 }
1402
1403 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1404 {
1405         struct inode *inode = filp->f_dentry->d_inode;
1406         struct address_space *mapping = inode->i_mapping;
1407         unsigned long index, offset;
1408
1409         index = *ppos >> PAGE_CACHE_SHIFT;
1410         offset = *ppos & ~PAGE_CACHE_MASK;
1411
1412         for (;;) {
1413                 struct page *page = NULL;
1414                 unsigned long end_index, nr, ret;
1415                 loff_t i_size = i_size_read(inode);
1416
1417                 end_index = i_size >> PAGE_CACHE_SHIFT;
1418                 if (index > end_index)
1419                         break;
1420                 if (index == end_index) {
1421                         nr = i_size & ~PAGE_CACHE_MASK;
1422                         if (nr <= offset)
1423                                 break;
1424                 }
1425
1426                 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1427                 if (desc->error) {
1428                         if (desc->error == -EINVAL)
1429                                 desc->error = 0;
1430                         break;
1431                 }
1432
1433                 /*
1434                  * We must evaluate after, since reads (unlike writes)
1435                  * are called without i_sem protection against truncate
1436                  */
1437                 nr = PAGE_CACHE_SIZE;
1438                 i_size = i_size_read(inode);
1439                 end_index = i_size >> PAGE_CACHE_SHIFT;
1440                 if (index == end_index) {
1441                         nr = i_size & ~PAGE_CACHE_MASK;
1442                         if (nr <= offset) {
1443                                 page_cache_release(page);
1444                                 break;
1445                         }
1446                 }
1447                 nr -= offset;
1448
1449                 if (page != ZERO_PAGE(0)) {
1450                         /*
1451                          * If users can be writing to this page using arbitrary
1452                          * virtual addresses, take care about potential aliasing
1453                          * before reading the page on the kernel side.
1454                          */
1455                         if (mapping_writably_mapped(mapping))
1456                                 flush_dcache_page(page);
1457                         /*
1458                          * Mark the page accessed if we read the beginning.
1459                          */
1460                         if (!offset)
1461                                 mark_page_accessed(page);
1462                 }
1463
1464                 /*
1465                  * Ok, we have the page, and it's up-to-date, so
1466                  * now we can copy it to user space...
1467                  *
1468                  * The actor routine returns how many bytes were actually used..
1469                  * NOTE! This may not be the same as how much of a user buffer
1470                  * we filled up (we may be padding etc), so we can only update
1471                  * "pos" here (the actor routine has to update the user buffer
1472                  * pointers and the remaining count).
1473                  */
1474                 ret = actor(desc, page, offset, nr);
1475                 offset += ret;
1476                 index += offset >> PAGE_CACHE_SHIFT;
1477                 offset &= ~PAGE_CACHE_MASK;
1478
1479                 page_cache_release(page);
1480                 if (ret != nr || !desc->count)
1481                         break;
1482
1483                 cond_resched();
1484         }
1485
1486         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1487         file_accessed(filp);
1488 }
1489
1490 static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1491 {
1492         read_descriptor_t desc;
1493
1494         if ((ssize_t) count < 0)
1495                 return -EINVAL;
1496         if (!access_ok(VERIFY_WRITE, buf, count))
1497                 return -EFAULT;
1498         if (!count)
1499                 return 0;
1500
1501         desc.written = 0;
1502         desc.count = count;
1503         desc.arg.buf = buf;
1504         desc.error = 0;
1505
1506         do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1507         if (desc.written)
1508                 return desc.written;
1509         return desc.error;
1510 }
1511
1512 static ssize_t shmem_file_sendfile(struct file *in_file, loff_t *ppos,
1513                          size_t count, read_actor_t actor, void *target)
1514 {
1515         read_descriptor_t desc;
1516
1517         if (!count)
1518                 return 0;
1519
1520         desc.written = 0;
1521         desc.count = count;
1522         desc.arg.data = target;
1523         desc.error = 0;
1524
1525         do_shmem_file_read(in_file, ppos, &desc, actor);
1526         if (desc.written)
1527                 return desc.written;
1528         return desc.error;
1529 }
1530
1531 static int shmem_statfs(struct super_block *sb, struct kstatfs *buf)
1532 {
1533         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1534
1535         buf->f_type = TMPFS_MAGIC;
1536         buf->f_bsize = PAGE_CACHE_SIZE;
1537         spin_lock(&sbinfo->stat_lock);
1538         buf->f_blocks = sbinfo->max_blocks;
1539         buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1540         buf->f_files = sbinfo->max_inodes;
1541         buf->f_ffree = sbinfo->free_inodes;
1542         spin_unlock(&sbinfo->stat_lock);
1543         buf->f_namelen = NAME_MAX;
1544         return 0;
1545 }
1546
1547 /*
1548  * File creation. Allocate an inode, and we're done..
1549  */
1550 static int
1551 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1552 {
1553         struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1554         int error = -ENOSPC;
1555
1556         if (inode) {
1557                 if (dir->i_mode & S_ISGID) {
1558                         inode->i_gid = dir->i_gid;
1559                         if (S_ISDIR(mode))
1560                                 inode->i_mode |= S_ISGID;
1561                 }
1562                 dir->i_size += BOGO_DIRENT_SIZE;
1563                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1564                 d_instantiate(dentry, inode);
1565                 dget(dentry); /* Extra count - pin the dentry in core */
1566                 error = 0;
1567         }
1568         return error;
1569 }
1570
1571 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1572 {
1573         int error;
1574
1575         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1576                 return error;
1577         dir->i_nlink++;
1578         return 0;
1579 }
1580
1581 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1582                 struct nameidata *nd)
1583 {
1584         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1585 }
1586
1587 /*
1588  * Link a file..
1589  */
1590 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1591 {
1592         struct inode *inode = old_dentry->d_inode;
1593
1594         dir->i_size += BOGO_DIRENT_SIZE;
1595         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1596         inode->i_nlink++;
1597         atomic_inc(&inode->i_count);    /* New dentry reference */
1598         dget(dentry);           /* Extra pinning count for the created dentry */
1599         d_instantiate(dentry, inode);
1600         return 0;
1601 }
1602
1603 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1604 {
1605         struct inode *inode = dentry->d_inode;
1606
1607         dir->i_size -= BOGO_DIRENT_SIZE;
1608         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1609         inode->i_nlink--;
1610         dput(dentry);   /* Undo the count from "create" - this does all the work */
1611         return 0;
1612 }
1613
1614 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1615 {
1616         if (!simple_empty(dentry))
1617                 return -ENOTEMPTY;
1618
1619         dir->i_nlink--;
1620         return shmem_unlink(dir, dentry);
1621 }
1622
1623 /*
1624  * The VFS layer already does all the dentry stuff for rename,
1625  * we just have to decrement the usage count for the target if
1626  * it exists so that the VFS layer correctly free's it when it
1627  * gets overwritten.
1628  */
1629 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1630 {
1631         struct inode *inode = old_dentry->d_inode;
1632         int they_are_dirs = S_ISDIR(inode->i_mode);
1633
1634         if (!simple_empty(new_dentry))
1635                 return -ENOTEMPTY;
1636
1637         if (new_dentry->d_inode) {
1638                 (void) shmem_unlink(new_dir, new_dentry);
1639                 if (they_are_dirs)
1640                         old_dir->i_nlink--;
1641         } else if (they_are_dirs) {
1642                 old_dir->i_nlink--;
1643                 new_dir->i_nlink++;
1644         }
1645
1646         old_dir->i_size -= BOGO_DIRENT_SIZE;
1647         new_dir->i_size += BOGO_DIRENT_SIZE;
1648         old_dir->i_ctime = old_dir->i_mtime =
1649         new_dir->i_ctime = new_dir->i_mtime =
1650         inode->i_ctime = CURRENT_TIME;
1651         return 0;
1652 }
1653
1654 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1655 {
1656         int error;
1657         int len;
1658         struct inode *inode;
1659         struct page *page = NULL;
1660         char *kaddr;
1661         struct shmem_inode_info *info;
1662
1663         len = strlen(symname) + 1;
1664         if (len > PAGE_CACHE_SIZE)
1665                 return -ENAMETOOLONG;
1666
1667         inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1668         if (!inode)
1669                 return -ENOSPC;
1670
1671         info = SHMEM_I(inode);
1672         inode->i_size = len-1;
1673         if (len <= (char *)inode - (char *)info) {
1674                 /* do it inline */
1675                 memcpy(info, symname, len);
1676                 inode->i_op = &shmem_symlink_inline_operations;
1677         } else {
1678                 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1679                 if (error) {
1680                         iput(inode);
1681                         return error;
1682                 }
1683                 inode->i_op = &shmem_symlink_inode_operations;
1684                 spin_lock(&shmem_ilock);
1685                 list_add_tail(&info->list, &shmem_inodes);
1686                 spin_unlock(&shmem_ilock);
1687                 kaddr = kmap_atomic(page, KM_USER0);
1688                 memcpy(kaddr, symname, len);
1689                 kunmap_atomic(kaddr, KM_USER0);
1690                 set_page_dirty(page);
1691                 page_cache_release(page);
1692         }
1693         if (dir->i_mode & S_ISGID)
1694                 inode->i_gid = dir->i_gid;
1695         dir->i_size += BOGO_DIRENT_SIZE;
1696         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1697         d_instantiate(dentry, inode);
1698         dget(dentry);
1699         return 0;
1700 }
1701
1702 static int shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1703 {
1704         nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1705         return 0;
1706 }
1707
1708 static int shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1709 {
1710         struct page *page = NULL;
1711         int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1712         nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1713         return 0;
1714 }
1715
1716 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd)
1717 {
1718         if (!IS_ERR(nd_get_link(nd))) {
1719                 struct page *page;
1720
1721                 page = find_get_page(dentry->d_inode->i_mapping, 0);
1722                 if (!page)
1723                         BUG();
1724                 kunmap(page);
1725                 mark_page_accessed(page);
1726                 page_cache_release(page);
1727                 page_cache_release(page);
1728         }
1729 }
1730
1731 static struct inode_operations shmem_symlink_inline_operations = {
1732         .readlink       = generic_readlink,
1733         .follow_link    = shmem_follow_link_inline,
1734 };
1735
1736 static struct inode_operations shmem_symlink_inode_operations = {
1737         .truncate       = shmem_truncate,
1738         .readlink       = generic_readlink,
1739         .follow_link    = shmem_follow_link,
1740         .put_link       = shmem_put_link,
1741 };
1742
1743 static int shmem_parse_options(char *options, int *mode, uid_t *uid, gid_t *gid, unsigned long *blocks, unsigned long *inodes)
1744 {
1745         char *this_char, *value, *rest;
1746
1747         while ((this_char = strsep(&options, ",")) != NULL) {
1748                 if (!*this_char)
1749                         continue;
1750                 if ((value = strchr(this_char,'=')) != NULL) {
1751                         *value++ = 0;
1752                 } else {
1753                         printk(KERN_ERR
1754                             "tmpfs: No value for mount option '%s'\n",
1755                             this_char);
1756                         return 1;
1757                 }
1758
1759                 if (!strcmp(this_char,"size")) {
1760                         unsigned long long size;
1761                         size = memparse(value,&rest);
1762                         if (*rest == '%') {
1763                                 size <<= PAGE_SHIFT;
1764                                 size *= totalram_pages;
1765                                 do_div(size, 100);
1766                                 rest++;
1767                         }
1768                         if (*rest)
1769                                 goto bad_val;
1770                         *blocks = size >> PAGE_CACHE_SHIFT;
1771                 } else if (!strcmp(this_char,"nr_blocks")) {
1772                         *blocks = memparse(value,&rest);
1773                         if (*rest)
1774                                 goto bad_val;
1775                 } else if (!strcmp(this_char,"nr_inodes")) {
1776                         *inodes = memparse(value,&rest);
1777                         if (*rest)
1778                                 goto bad_val;
1779                 } else if (!strcmp(this_char,"mode")) {
1780                         if (!mode)
1781                                 continue;
1782                         *mode = simple_strtoul(value,&rest,8);
1783                         if (*rest)
1784                                 goto bad_val;
1785                 } else if (!strcmp(this_char,"uid")) {
1786                         if (!uid)
1787                                 continue;
1788                         *uid = simple_strtoul(value,&rest,0);
1789                         if (*rest)
1790                                 goto bad_val;
1791                 } else if (!strcmp(this_char,"gid")) {
1792                         if (!gid)
1793                                 continue;
1794                         *gid = simple_strtoul(value,&rest,0);
1795                         if (*rest)
1796                                 goto bad_val;
1797                 } else {
1798                         printk(KERN_ERR "tmpfs: Bad mount option %s\n",
1799                                this_char);
1800                         return 1;
1801                 }
1802         }
1803         return 0;
1804
1805 bad_val:
1806         printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
1807                value, this_char);
1808         return 1;
1809
1810 }
1811
1812 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
1813 {
1814         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1815         unsigned long max_blocks = sbinfo->max_blocks;
1816         unsigned long max_inodes = sbinfo->max_inodes;
1817
1818         if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks, &max_inodes))
1819                 return -EINVAL;
1820         return shmem_set_size(sbinfo, max_blocks, max_inodes);
1821 }
1822 #endif
1823
1824 static int shmem_fill_super(struct super_block *sb,
1825                             void *data, int silent)
1826 {
1827         struct inode *inode;
1828         struct dentry *root;
1829         unsigned long blocks, inodes;
1830         int mode   = S_IRWXUGO | S_ISVTX;
1831         uid_t uid = current->fsuid;
1832         gid_t gid = current->fsgid;
1833         struct shmem_sb_info *sbinfo;
1834         int err = -ENOMEM;
1835
1836         sbinfo = kmalloc(sizeof(struct shmem_sb_info), GFP_KERNEL);
1837         if (!sbinfo)
1838                 return -ENOMEM;
1839         sb->s_fs_info = sbinfo;
1840         memset(sbinfo, 0, sizeof(struct shmem_sb_info));
1841
1842         /*
1843          * Per default we only allow half of the physical ram per
1844          * tmpfs instance
1845          */
1846         blocks = inodes = totalram_pages / 2;
1847
1848 #ifdef CONFIG_TMPFS
1849         if (shmem_parse_options(data, &mode, &uid, &gid, &blocks, &inodes)) {
1850                 err = -EINVAL;
1851                 goto failed;
1852         }
1853 #else
1854         sb->s_flags |= MS_NOUSER;
1855 #endif
1856
1857         spin_lock_init(&sbinfo->stat_lock);
1858         sbinfo->max_blocks = blocks;
1859         sbinfo->free_blocks = blocks;
1860         sbinfo->max_inodes = inodes;
1861         sbinfo->free_inodes = inodes;
1862         sb->s_maxbytes = SHMEM_MAX_BYTES;
1863         sb->s_blocksize = PAGE_CACHE_SIZE;
1864         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1865         sb->s_magic = TMPFS_MAGIC;
1866         sb->s_op = &shmem_ops;
1867         inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
1868         if (!inode)
1869                 goto failed;
1870         inode->i_uid = uid;
1871         inode->i_gid = gid;
1872         root = d_alloc_root(inode);
1873         if (!root)
1874                 goto failed_iput;
1875         sb->s_root = root;
1876         return 0;
1877
1878 failed_iput:
1879         iput(inode);
1880 failed:
1881         kfree(sbinfo);
1882         sb->s_fs_info = NULL;
1883         return err;
1884 }
1885
1886 static void shmem_put_super(struct super_block *sb)
1887 {
1888         kfree(sb->s_fs_info);
1889         sb->s_fs_info = NULL;
1890 }
1891
1892 static kmem_cache_t *shmem_inode_cachep;
1893
1894 static struct inode *shmem_alloc_inode(struct super_block *sb)
1895 {
1896         struct shmem_inode_info *p;
1897         p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, SLAB_KERNEL);
1898         if (!p)
1899                 return NULL;
1900         return &p->vfs_inode;
1901 }
1902
1903 static void shmem_destroy_inode(struct inode *inode)
1904 {
1905         mpol_free_shared_policy(&SHMEM_I(inode)->policy);
1906         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
1907 }
1908
1909 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
1910 {
1911         struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
1912
1913         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
1914             SLAB_CTOR_CONSTRUCTOR) {
1915                 inode_init_once(&p->vfs_inode);
1916         }
1917 }
1918
1919 static int init_inodecache(void)
1920 {
1921         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
1922                                 sizeof(struct shmem_inode_info),
1923                                 0, SLAB_HWCACHE_ALIGN|SLAB_RECLAIM_ACCOUNT,
1924                                 init_once, NULL);
1925         if (shmem_inode_cachep == NULL)
1926                 return -ENOMEM;
1927         return 0;
1928 }
1929
1930 static void destroy_inodecache(void)
1931 {
1932         if (kmem_cache_destroy(shmem_inode_cachep))
1933                 printk(KERN_INFO "shmem_inode_cache: not all structures were freed\n");
1934 }
1935
1936 static struct address_space_operations shmem_aops = {
1937         .writepage      = shmem_writepage,
1938         .set_page_dirty = __set_page_dirty_nobuffers,
1939 #ifdef CONFIG_TMPFS
1940         .prepare_write  = shmem_prepare_write,
1941         .commit_write   = simple_commit_write,
1942 #endif
1943 };
1944
1945 static struct file_operations shmem_file_operations = {
1946         .mmap           = shmem_mmap,
1947 #ifdef CONFIG_TMPFS
1948         .llseek         = generic_file_llseek,
1949         .read           = shmem_file_read,
1950         .write          = shmem_file_write,
1951         .fsync          = simple_sync_file,
1952         .sendfile       = shmem_file_sendfile,
1953 #endif
1954 };
1955
1956 static struct inode_operations shmem_inode_operations = {
1957         .truncate       = shmem_truncate,
1958         .setattr        = shmem_notify_change,
1959 };
1960
1961 static struct inode_operations shmem_dir_inode_operations = {
1962 #ifdef CONFIG_TMPFS
1963         .create         = shmem_create,
1964         .lookup         = simple_lookup,
1965         .link           = shmem_link,
1966         .unlink         = shmem_unlink,
1967         .symlink        = shmem_symlink,
1968         .mkdir          = shmem_mkdir,
1969         .rmdir          = shmem_rmdir,
1970         .mknod          = shmem_mknod,
1971         .rename         = shmem_rename,
1972 #endif
1973 };
1974
1975 static struct super_operations shmem_ops = {
1976         .alloc_inode    = shmem_alloc_inode,
1977         .destroy_inode  = shmem_destroy_inode,
1978 #ifdef CONFIG_TMPFS
1979         .statfs         = shmem_statfs,
1980         .remount_fs     = shmem_remount_fs,
1981 #endif
1982         .delete_inode   = shmem_delete_inode,
1983         .drop_inode     = generic_delete_inode,
1984         .put_super      = shmem_put_super,
1985 };
1986
1987 static struct vm_operations_struct shmem_vm_ops = {
1988         .nopage         = shmem_nopage,
1989         .populate       = shmem_populate,
1990 #ifdef CONFIG_NUMA
1991         .set_policy     = shmem_set_policy,
1992         .get_policy     = shmem_get_policy,
1993 #endif
1994 };
1995
1996 static struct super_block *shmem_get_sb(struct file_system_type *fs_type,
1997         int flags, const char *dev_name, void *data)
1998 {
1999         return get_sb_nodev(fs_type, flags, data, shmem_fill_super);
2000 }
2001
2002 static struct file_system_type tmpfs_fs_type = {
2003         .owner          = THIS_MODULE,
2004         .name           = "tmpfs",
2005         .get_sb         = shmem_get_sb,
2006         .kill_sb        = kill_litter_super,
2007 };
2008 static struct vfsmount *shm_mnt;
2009
2010 static int __init init_tmpfs(void)
2011 {
2012         int error;
2013
2014         error = init_inodecache();
2015         if (error)
2016                 goto out3;
2017
2018         error = register_filesystem(&tmpfs_fs_type);
2019         if (error) {
2020                 printk(KERN_ERR "Could not register tmpfs\n");
2021                 goto out2;
2022         }
2023 #ifdef CONFIG_TMPFS
2024         devfs_mk_dir("shm");
2025 #endif
2026         shm_mnt = kern_mount(&tmpfs_fs_type);
2027         if (IS_ERR(shm_mnt)) {
2028                 error = PTR_ERR(shm_mnt);
2029                 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2030                 goto out1;
2031         }
2032
2033         /* The internal instance should not do size checking */
2034         shmem_set_size(SHMEM_SB(shm_mnt->mnt_sb), ULONG_MAX, ULONG_MAX);
2035         return 0;
2036
2037 out1:
2038         unregister_filesystem(&tmpfs_fs_type);
2039 out2:
2040         destroy_inodecache();
2041 out3:
2042         shm_mnt = ERR_PTR(error);
2043         return error;
2044 }
2045 module_init(init_tmpfs)
2046
2047 /*
2048  * shmem_file_setup - get an unlinked file living in tmpfs
2049  *
2050  * @name: name for dentry (to be seen in /proc/<pid>/maps
2051  * @size: size to be set for the file
2052  *
2053  */
2054 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2055 {
2056         int error;
2057         struct file *file;
2058         struct inode *inode;
2059         struct dentry *dentry, *root;
2060         struct qstr this;
2061
2062         if (IS_ERR(shm_mnt))
2063                 return (void *)shm_mnt;
2064
2065         if (size > SHMEM_MAX_BYTES)
2066                 return ERR_PTR(-EINVAL);
2067
2068         if (shmem_acct_size(flags, size))
2069                 return ERR_PTR(-ENOMEM);
2070
2071         error = -ENOMEM;
2072         this.name = name;
2073         this.len = strlen(name);
2074         this.hash = 0; /* will go */
2075         root = shm_mnt->mnt_root;
2076         dentry = d_alloc(root, &this);
2077         if (!dentry)
2078                 goto put_memory;
2079
2080         error = -ENFILE;
2081         file = get_empty_filp();
2082         if (!file)
2083                 goto put_dentry;
2084
2085         error = -ENOSPC;
2086         inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2087         if (!inode)
2088                 goto close_file;
2089
2090         SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2091         d_instantiate(dentry, inode);
2092         inode->i_size = size;
2093         inode->i_nlink = 0;     /* It is unlinked */
2094         file->f_vfsmnt = mntget(shm_mnt);
2095         file->f_dentry = dentry;
2096         file->f_mapping = inode->i_mapping;
2097         file->f_op = &shmem_file_operations;
2098         file->f_mode = FMODE_WRITE | FMODE_READ;
2099         return(file);
2100
2101 close_file:
2102         put_filp(file);
2103 put_dentry:
2104         dput(dentry);
2105 put_memory:
2106         shmem_unacct_size(flags, size);
2107         return ERR_PTR(error);
2108 }
2109
2110 /*
2111  * shmem_zero_setup - setup a shared anonymous mapping
2112  *
2113  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2114  */
2115 int shmem_zero_setup(struct vm_area_struct *vma)
2116 {
2117         struct file *file;
2118         loff_t size = vma->vm_end - vma->vm_start;
2119
2120         file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2121         if (IS_ERR(file))
2122                 return PTR_ERR(file);
2123
2124         if (vma->vm_file)
2125                 fput(vma->vm_file);
2126         vma->vm_file = file;
2127         vma->vm_ops = &shmem_vm_ops;
2128         return 0;
2129 }
2130
2131 EXPORT_SYMBOL(shmem_file_setup);