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