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