mm/shmem.c

   1 /*
   2  * Resizable virtual memory filesystem for Linux.
   3  *
   4  * Copyright (C) 2000 Linus Torvalds.
   5  *               2000 Transmeta Corp.
   6  *               2000-2001 Christoph Rohland
   7  *               2000-2001 SAP AG
   8  *               2002 Red Hat Inc.
   9  * Copyright (C) 2002-2004 Hugh Dickins.
  10  * Copyright (C) 2002-2004 VERITAS Software Corporation.
  11  * Copyright (C) 2004 Andi Kleen, SuSE Labs
  12  *
  13  * 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 static LIST_HEAD(shmem_swaplist);
 183 static spinlock_t shmem_swaplist_lock = 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         if (sbinfo) {
 189                 spin_lock(&sbinfo->stat_lock);
 190                 sbinfo->free_blocks++;
 191                 inode->i_blocks -= BLOCKS_PER_PAGE;
 192                 spin_unlock(&sbinfo->stat_lock);
 193         }
 194 }
 195
 196 /*
 197  * shmem_recalc_inode - recalculate the size of an inode
 198  *
 199  * @inode: inode to recalc
 200  *
 201  * We have to calculate the free blocks since the mm can drop
 202  * undirtied hole pages behind our back.
 203  *
 204  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
 205  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
 206  *
 207  * It has to be called with the spinlock held.
 208  */
 209 static void shmem_recalc_inode(struct inode *inode)
 210 {
 211         struct shmem_inode_info *info = SHMEM_I(inode);
 212         long freed;
 213
 214         freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
 215         if (freed > 0) {
 216                 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 217                 info->alloced -= freed;
 218                 shmem_unacct_blocks(info->flags, freed);
 219                 if (sbinfo) {
 220                         spin_lock(&sbinfo->stat_lock);
 221                         sbinfo->free_blocks += freed;
 222                         inode->i_blocks -= freed*BLOCKS_PER_PAGE;
 223                         spin_unlock(&sbinfo->stat_lock);
 224                 }
 225         }
 226 }
 227
 228 /*
 229  * shmem_swp_entry - find the swap vector position in the info structure
 230  *
 231  * @info:  info structure for the inode
 232  * @index: index of the page to find
 233  * @page:  optional page to add to the structure. Has to be preset to
 234  *         all zeros
 235  *
 236  * If there is no space allocated yet it will return NULL when
 237  * page is NULL, else it will use the page for the needed block,
 238  * setting it to NULL on return to indicate that it has been used.
 239  *
 240  * The swap vector is organized the following way:
 241  *
 242  * There are SHMEM_NR_DIRECT entries directly stored in the
 243  * shmem_inode_info structure. So small files do not need an addional
 244  * allocation.
 245  *
 246  * For pages with index > SHMEM_NR_DIRECT there is the pointer
 247  * i_indirect which points to a page which holds in the first half
 248  * doubly indirect blocks, in the second half triple indirect blocks:
 249  *
 250  * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
 251  * following layout (for SHMEM_NR_DIRECT == 16):
 252  *
 253  * i_indirect -> dir --> 16-19
 254  *            |      +-> 20-23
 255  *            |
 256  *            +-->dir2 --> 24-27
 257  *            |        +-> 28-31
 258  *            |        +-> 32-35
 259  *            |        +-> 36-39
 260  *            |
 261  *            +-->dir3 --> 40-43
 262  *                     +-> 44-47
 263  *                     +-> 48-51
 264  *                     +-> 52-55
 265  */
 266 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
 267 {
 268         unsigned long offset;
 269         struct page **dir;
 270         struct page *subdir;
 271
 272         if (index < SHMEM_NR_DIRECT) {
 273                 shmem_swp_balance_unmap();
 274                 return info->i_direct+index;
 275         }
 276         if (!info->i_indirect) {
 277                 if (page) {
 278                         info->i_indirect = *page;
 279                         *page = NULL;
 280                 }
 281                 return NULL;                    /* need another page */
 282         }
 283
 284         index -= SHMEM_NR_DIRECT;
 285         offset = index % ENTRIES_PER_PAGE;
 286         index /= ENTRIES_PER_PAGE;
 287         dir = shmem_dir_map(info->i_indirect);
 288
 289         if (index >= ENTRIES_PER_PAGE/2) {
 290                 index -= ENTRIES_PER_PAGE/2;
 291                 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
 292                 index %= ENTRIES_PER_PAGE;
 293                 subdir = *dir;
 294                 if (!subdir) {
 295                         if (page) {
 296                                 *dir = *page;
 297                                 *page = NULL;
 298                         }
 299                         shmem_dir_unmap(dir);
 300                         return NULL;            /* need another page */
 301                 }
 302                 shmem_dir_unmap(dir);
 303                 dir = shmem_dir_map(subdir);
 304         }
 305
 306         dir += index;
 307         subdir = *dir;
 308         if (!subdir) {
 309                 if (!page || !(subdir = *page)) {
 310                         shmem_dir_unmap(dir);
 311                         return NULL;            /* need a page */
 312                 }
 313                 *dir = subdir;
 314                 *page = NULL;
 315         }
 316         shmem_dir_unmap(dir);
 317         return shmem_swp_map(subdir) + offset;
 318 }
 319
 320 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
 321 {
 322         long incdec = value? 1: -1;
 323
 324         entry->val = value;
 325         info->swapped += incdec;
 326         if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT)
 327                 kmap_atomic_to_page(entry)->nr_swapped += incdec;
 328 }
 329
 330 /*
 331  * shmem_swp_alloc - get the position of the swap entry for the page.
 332  *                   If it does not exist allocate the entry.
 333  *
 334  * @info:       info structure for the inode
 335  * @index:      index of the page to find
 336  * @sgp:        check and recheck i_size? skip allocation?
 337  */
 338 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
 339 {
 340         struct inode *inode = &info->vfs_inode;
 341         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 342         struct page *page = NULL;
 343         swp_entry_t *entry;
 344
 345         if (sgp != SGP_WRITE &&
 346             ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
 347                 return ERR_PTR(-EINVAL);
 348
 349         while (!(entry = shmem_swp_entry(info, index, &page))) {
 350                 if (sgp == SGP_READ)
 351                         return shmem_swp_map(ZERO_PAGE(0));
 352                 /*
 353                  * Test free_blocks against 1 not 0, since we have 1 data
 354                  * page (and perhaps indirect index pages) yet to allocate:
 355                  * a waste to allocate index if we cannot allocate data.
 356                  */
 357                 if (sbinfo) {
 358                         spin_lock(&sbinfo->stat_lock);
 359                         if (sbinfo->free_blocks <= 1) {
 360                                 spin_unlock(&sbinfo->stat_lock);
 361                                 return ERR_PTR(-ENOSPC);
 362                         }
 363                         sbinfo->free_blocks--;
 364                         inode->i_blocks += BLOCKS_PER_PAGE;
 365                         spin_unlock(&sbinfo->stat_lock);
 366                 }
 367
 368                 spin_unlock(&info->lock);
 369                 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
 370                 if (page) {
 371                         clear_highpage(page);
 372                         page->nr_swapped = 0;
 373                 }
 374                 spin_lock(&info->lock);
 375
 376                 if (!page) {
 377                         shmem_free_block(inode);
 378                         return ERR_PTR(-ENOMEM);
 379                 }
 380                 if (sgp != SGP_WRITE &&
 381                     ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
 382                         entry = ERR_PTR(-EINVAL);
 383                         break;
 384                 }
 385                 if (info->next_index <= index)
 386                         info->next_index = index + 1;
 387         }
 388         if (page) {
 389                 /* another task gave its page, or truncated the file */
 390                 shmem_free_block(inode);
 391                 shmem_dir_free(page);
 392         }
 393         if (info->next_index <= index && !IS_ERR(entry))
 394                 info->next_index = index + 1;
 395         return entry;
 396 }
 397
 398 /*
 399  * shmem_free_swp - free some swap entries in a directory
 400  *
 401  * @dir:   pointer to the directory
 402  * @edir:  pointer after last entry of the directory
 403  */
 404 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir)
 405 {
 406         swp_entry_t *ptr;
 407         int freed = 0;
 408
 409         for (ptr = dir; ptr < edir; ptr++) {
 410                 if (ptr->val) {
 411                         free_swap_and_cache(*ptr);
 412                         *ptr = (swp_entry_t){0};
 413                         freed++;
 414                 }
 415         }
 416         return freed;
 417 }
 418
 419 static void shmem_truncate(struct inode *inode)
 420 {
 421         struct shmem_inode_info *info = SHMEM_I(inode);
 422         unsigned long idx;
 423         unsigned long size;
 424         unsigned long limit;
 425         unsigned long stage;
 426         struct page **dir;
 427         struct page *subdir;
 428         struct page *empty;
 429         swp_entry_t *ptr;
 430         int offset;
 431         int freed;
 432
 433         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 434         idx = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 435         if (idx >= info->next_index)
 436                 return;
 437
 438         spin_lock(&info->lock);
 439         info->flags |= SHMEM_TRUNCATE;
 440         limit = info->next_index;
 441         info->next_index = idx;
 442         if (info->swapped && idx < SHMEM_NR_DIRECT) {
 443                 ptr = info->i_direct;
 444                 size = limit;
 445                 if (size > SHMEM_NR_DIRECT)
 446                         size = SHMEM_NR_DIRECT;
 447                 info->swapped -= shmem_free_swp(ptr+idx, ptr+size);
 448         }
 449         if (!info->i_indirect)
 450                 goto done2;
 451
 452         BUG_ON(limit <= SHMEM_NR_DIRECT);
 453         limit -= SHMEM_NR_DIRECT;
 454         idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
 455         offset = idx % ENTRIES_PER_PAGE;
 456         idx -= offset;
 457
 458         empty = NULL;
 459         dir = shmem_dir_map(info->i_indirect);
 460         stage = ENTRIES_PER_PAGEPAGE/2;
 461         if (idx < ENTRIES_PER_PAGEPAGE/2)
 462                 dir += idx/ENTRIES_PER_PAGE;
 463         else {
 464                 dir += ENTRIES_PER_PAGE/2;
 465                 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
 466                 while (stage <= idx)
 467                         stage += ENTRIES_PER_PAGEPAGE;
 468                 if (*dir) {
 469                         subdir = *dir;
 470                         size = ((idx - ENTRIES_PER_PAGEPAGE/2) %
 471                                 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
 472                         if (!size && !offset) {
 473                                 empty = subdir;
 474                                 *dir = NULL;
 475                         }
 476                         shmem_dir_unmap(dir);
 477                         dir = shmem_dir_map(subdir) + size;
 478                 } else {
 479                         offset = 0;
 480                         idx = stage;
 481                 }
 482         }
 483
 484         for (; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
 485                 if (unlikely(idx == stage)) {
 486                         shmem_dir_unmap(dir-1);
 487                         dir = shmem_dir_map(info->i_indirect) +
 488                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
 489                         while (!*dir) {
 490                                 dir++;
 491                                 idx += ENTRIES_PER_PAGEPAGE;
 492                                 if (idx >= limit)
 493                                         goto done1;
 494                         }
 495                         stage = idx + ENTRIES_PER_PAGEPAGE;
 496                         subdir = *dir;
 497                         *dir = NULL;
 498                         shmem_dir_unmap(dir);
 499                         if (empty) {
 500                                 shmem_dir_free(empty);
 501                                 shmem_free_block(inode);
 502                         }
 503                         empty = subdir;
 504                         cond_resched_lock(&info->lock);
 505                         dir = shmem_dir_map(subdir);
 506                 }
 507                 subdir = *dir;
 508                 if (subdir && subdir->nr_swapped) {
 509                         ptr = shmem_swp_map(subdir);
 510                         size = limit - idx;
 511                         if (size > ENTRIES_PER_PAGE)
 512                                 size = ENTRIES_PER_PAGE;
 513                         freed = shmem_free_swp(ptr+offset, ptr+size);
 514                         shmem_swp_unmap(ptr);
 515                         info->swapped -= freed;
 516                         subdir->nr_swapped -= freed;
 517                         BUG_ON(subdir->nr_swapped > offset);
 518                 }
 519                 if (offset)
 520                         offset = 0;
 521                 else if (subdir) {
 522                         *dir = NULL;
 523                         shmem_dir_free(subdir);
 524                         shmem_free_block(inode);
 525                 }
 526         }
 527 done1:
 528         shmem_dir_unmap(dir-1);
 529         if (empty) {
 530                 shmem_dir_free(empty);
 531                 shmem_free_block(inode);
 532         }
 533         if (info->next_index <= SHMEM_NR_DIRECT) {
 534                 shmem_dir_free(info->i_indirect);
 535                 info->i_indirect = NULL;
 536                 shmem_free_block(inode);
 537         }
 538 done2:
 539         BUG_ON(info->swapped > info->next_index);
 540         if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
 541                 /*
 542                  * Call truncate_inode_pages again: racing shmem_unuse_inode
 543                  * may have swizzled a page in from swap since vmtruncate or
 544                  * generic_delete_inode did it, before we lowered next_index.
 545                  * Also, though shmem_getpage checks i_size before adding to
 546                  * cache, no recheck after: so fix the narrow window there too.
 547                  */
 548                 spin_unlock(&info->lock);
 549                 truncate_inode_pages(inode->i_mapping, inode->i_size);
 550                 spin_lock(&info->lock);
 551         }
 552         info->flags &= ~SHMEM_TRUNCATE;
 553         shmem_recalc_inode(inode);
 554         spin_unlock(&info->lock);
 555 }
 556
 557 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
 558 {
 559         struct inode *inode = dentry->d_inode;
 560         struct page *page = NULL;
 561         int error;
 562
 563         if (attr->ia_valid & ATTR_SIZE) {
 564                 if (attr->ia_size < inode->i_size) {
 565                         /*
 566                          * If truncating down to a partial page, then
 567                          * if that page is already allocated, hold it
 568                          * in memory until the truncation is over, so
 569                          * truncate_partial_page cannnot miss it were
 570                          * it assigned to swap.
 571                          */
 572                         if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
 573                                 (void) shmem_getpage(inode,
 574                                         attr->ia_size>>PAGE_CACHE_SHIFT,
 575                                                 &page, SGP_READ, NULL);
 576                         }
 577                         /*
 578                          * Reset SHMEM_PAGEIN flag so that shmem_truncate can
 579                          * detect if any pages might have been added to cache
 580                          * after truncate_inode_pages.  But we needn't bother
 581                          * if it's being fully truncated to zero-length: the
 582                          * nrpages check is efficient enough in that case.
 583                          */
 584                         if (attr->ia_size) {
 585                                 struct shmem_inode_info *info = SHMEM_I(inode);
 586                                 spin_lock(&info->lock);
 587                                 info->flags &= ~SHMEM_PAGEIN;
 588                                 spin_unlock(&info->lock);
 589                         }
 590                 }
 591         }
 592
 593         error = inode_change_ok(inode, attr);
 594         if (!error)
 595                 error = inode_setattr(inode, attr);
 596         if (page)
 597                 page_cache_release(page);
 598         return error;
 599 }
 600
 601 static void shmem_delete_inode(struct inode *inode)
 602 {
 603         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 604         struct shmem_inode_info *info = SHMEM_I(inode);
 605
 606         if (inode->i_op->truncate == shmem_truncate) {
 607                 shmem_unacct_size(info->flags, inode->i_size);
 608                 inode->i_size = 0;
 609                 shmem_truncate(inode);
 610                 if (!list_empty(&info->swaplist)) {
 611                         spin_lock(&shmem_swaplist_lock);
 612                         list_del_init(&info->swaplist);
 613                         spin_unlock(&shmem_swaplist_lock);
 614                 }
 615         }
 616         if (sbinfo) {
 617                 BUG_ON(inode->i_blocks);
 618                 spin_lock(&sbinfo->stat_lock);
 619                 sbinfo->free_inodes++;
 620                 spin_unlock(&sbinfo->stat_lock);
 621         }
 622         clear_inode(inode);
 623 }
 624
 625 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
 626 {
 627         swp_entry_t *ptr;
 628
 629         for (ptr = dir; ptr < edir; ptr++) {
 630                 if (ptr->val == entry.val)
 631                         return ptr - dir;
 632         }
 633         return -1;
 634 }
 635
 636 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
 637 {
 638         struct inode *inode;
 639         unsigned long idx;
 640         unsigned long size;
 641         unsigned long limit;
 642         unsigned long stage;
 643         struct page **dir;
 644         struct page *subdir;
 645         swp_entry_t *ptr;
 646         int offset;
 647
 648         idx = 0;
 649         ptr = info->i_direct;
 650         spin_lock(&info->lock);
 651         limit = info->next_index;
 652         size = limit;
 653         if (size > SHMEM_NR_DIRECT)
 654                 size = SHMEM_NR_DIRECT;
 655         offset = shmem_find_swp(entry, ptr, ptr+size);
 656         if (offset >= 0) {
 657                 shmem_swp_balance_unmap();
 658                 goto found;
 659         }
 660         if (!info->i_indirect)
 661                 goto lost2;
 662         /* we might be racing with shmem_truncate */
 663         if (limit <= SHMEM_NR_DIRECT)
 664                 goto lost2;
 665
 666         dir = shmem_dir_map(info->i_indirect);
 667         stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
 668
 669         for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
 670                 if (unlikely(idx == stage)) {
 671                         shmem_dir_unmap(dir-1);
 672                         dir = shmem_dir_map(info->i_indirect) +
 673                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
 674                         while (!*dir) {
 675                                 dir++;
 676                                 idx += ENTRIES_PER_PAGEPAGE;
 677                                 if (idx >= limit)
 678                                         goto lost1;
 679                         }
 680                         stage = idx + ENTRIES_PER_PAGEPAGE;
 681                         subdir = *dir;
 682                         shmem_dir_unmap(dir);
 683                         dir = shmem_dir_map(subdir);
 684                 }
 685                 subdir = *dir;
 686                 if (subdir && subdir->nr_swapped) {
 687                         ptr = shmem_swp_map(subdir);
 688                         size = limit - idx;
 689                         if (size > ENTRIES_PER_PAGE)
 690                                 size = ENTRIES_PER_PAGE;
 691                         offset = shmem_find_swp(entry, ptr, ptr+size);
 692                         if (offset >= 0) {
 693                                 shmem_dir_unmap(dir);
 694                                 goto found;
 695                         }
 696                         shmem_swp_unmap(ptr);
 697                 }
 698         }
 699 lost1:
 700         shmem_dir_unmap(dir-1);
 701 lost2:
 702         spin_unlock(&info->lock);
 703         return 0;
 704 found:
 705         idx += offset;
 706         inode = &info->vfs_inode;
 707         if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
 708                 info->flags |= SHMEM_PAGEIN;
 709                 shmem_swp_set(info, ptr + offset, 0);
 710         }
 711         shmem_swp_unmap(ptr);
 712         spin_unlock(&info->lock);
 713         /*
 714          * Decrement swap count even when the entry is left behind:
 715          * try_to_unuse will skip over mms, then reincrement count.
 716          */
 717         swap_free(entry);
 718         return 1;
 719 }
 720
 721 /*
 722  * shmem_unuse() search for an eventually swapped out shmem page.
 723  */
 724 int shmem_unuse(swp_entry_t entry, struct page *page)
 725 {
 726         struct list_head *p, *next;
 727         struct shmem_inode_info *info;
 728         int found = 0;
 729
 730         spin_lock(&shmem_swaplist_lock);
 731         list_for_each_safe(p, next, &shmem_swaplist) {
 732                 info = list_entry(p, struct shmem_inode_info, swaplist);
 733                 if (!info->swapped)
 734                         list_del_init(&info->swaplist);
 735                 else if (shmem_unuse_inode(info, entry, page)) {
 736                         /* move head to start search for next from here */
 737                         list_move_tail(&shmem_swaplist, &info->swaplist);
 738                         found = 1;
 739                         break;
 740                 }
 741         }
 742         spin_unlock(&shmem_swaplist_lock);
 743         return found;
 744 }
 745
 746 /*
 747  * Move the page from the page cache to the swap cache.
 748  */
 749 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
 750 {
 751         struct shmem_inode_info *info;
 752         swp_entry_t *entry, swap;
 753         struct address_space *mapping;
 754         unsigned long index;
 755         struct inode *inode;
 756
 757         BUG_ON(!PageLocked(page));
 758         BUG_ON(page_mapped(page));
 759
 760         mapping = page->mapping;
 761         index = page->index;
 762         inode = mapping->host;
 763         info = SHMEM_I(inode);
 764         if (info->flags & VM_LOCKED)
 765                 goto redirty;
 766         swap = get_swap_page();
 767         if (!swap.val)
 768                 goto redirty;
 769
 770         spin_lock(&info->lock);
 771         shmem_recalc_inode(inode);
 772         if (index >= info->next_index) {
 773                 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
 774                 goto unlock;
 775         }
 776         entry = shmem_swp_entry(info, index, NULL);
 777         BUG_ON(!entry);
 778         BUG_ON(entry->val);
 779
 780         if (move_to_swap_cache(page, swap) == 0) {
 781                 shmem_swp_set(info, entry, swap.val);
 782                 shmem_swp_unmap(entry);
 783                 spin_unlock(&info->lock);
 784                 if (list_empty(&info->swaplist)) {
 785                         spin_lock(&shmem_swaplist_lock);
 786                         /* move instead of add in case we're racing */
 787                         list_move_tail(&info->swaplist, &shmem_swaplist);
 788                         spin_unlock(&shmem_swaplist_lock);
 789                 }
 790                 unlock_page(page);
 791                 return 0;
 792         }
 793
 794         shmem_swp_unmap(entry);
 795 unlock:
 796         spin_unlock(&info->lock);
 797         swap_free(swap);
 798 redirty:
 799         set_page_dirty(page);
 800         return WRITEPAGE_ACTIVATE;      /* Return with the page locked */
 801 }
 802
 803 #ifdef CONFIG_NUMA
 804 static struct page *shmem_swapin_async(struct shared_policy *p,
 805                                        swp_entry_t entry, unsigned long idx)
 806 {
 807         struct page *page;
 808         struct vm_area_struct pvma;
 809
 810         /* Create a pseudo vma that just contains the policy */
 811         memset(&pvma, 0, sizeof(struct vm_area_struct));
 812         pvma.vm_end = PAGE_SIZE;
 813         pvma.vm_pgoff = idx;
 814         pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
 815         page = read_swap_cache_async(entry, &pvma, 0);
 816         mpol_free(pvma.vm_policy);
 817         return page;
 818 }
 819
 820 struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry,
 821                           unsigned long idx)
 822 {
 823         struct shared_policy *p = &info->policy;
 824         int i, num;
 825         struct page *page;
 826         unsigned long offset;
 827
 828         num = valid_swaphandles(entry, &offset);
 829         for (i = 0; i < num; offset++, i++) {
 830                 page = shmem_swapin_async(p,
 831                                 swp_entry(swp_type(entry), offset), idx);
 832                 if (!page)
 833                         break;
 834                 page_cache_release(page);
 835         }
 836         lru_add_drain();        /* Push any new pages onto the LRU now */
 837         return shmem_swapin_async(p, entry, idx);
 838 }
 839
 840 static struct page *
 841 shmem_alloc_page(unsigned long gfp, struct shmem_inode_info *info,
 842                  unsigned long idx)
 843 {
 844         struct vm_area_struct pvma;
 845         struct page *page;
 846
 847         memset(&pvma, 0, sizeof(struct vm_area_struct));
 848         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
 849         pvma.vm_pgoff = idx;
 850         pvma.vm_end = PAGE_SIZE;
 851         page = alloc_page_vma(gfp, &pvma, 0);
 852         mpol_free(pvma.vm_policy);
 853         return page;
 854 }
 855 #else
 856 static inline struct page *
 857 shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
 858 {
 859         swapin_readahead(entry, 0, NULL);
 860         return read_swap_cache_async(entry, NULL, 0);
 861 }
 862
 863 static inline struct page *
 864 shmem_alloc_page(unsigned long gfp,struct shmem_inode_info *info,
 865                                  unsigned long idx)
 866 {
 867         return alloc_page(gfp);
 868 }
 869 #endif
 870
 871 /*
 872  * shmem_getpage - either get the page from swap or allocate a new one
 873  *
 874  * If we allocate a new one we do not mark it dirty. That's up to the
 875  * vm. If we swap it in we mark it dirty since we also free the swap
 876  * entry since a page cannot live in both the swap and page cache
 877  */
 878 static int shmem_getpage(struct inode *inode, unsigned long idx,
 879                         struct page **pagep, enum sgp_type sgp, int *type)
 880 {
 881         struct address_space *mapping = inode->i_mapping;
 882         struct shmem_inode_info *info = SHMEM_I(inode);
 883         struct shmem_sb_info *sbinfo;
 884         struct page *filepage = *pagep;
 885         struct page *swappage;
 886         swp_entry_t *entry;
 887         swp_entry_t swap;
 888         int error;
 889
 890         if (idx >= SHMEM_MAX_INDEX)
 891                 return -EFBIG;
 892         /*
 893          * Normally, filepage is NULL on entry, and either found
 894          * uptodate immediately, or allocated and zeroed, or read
 895          * in under swappage, which is then assigned to filepage.
 896          * But shmem_prepare_write passes in a locked filepage,
 897          * which may be found not uptodate by other callers too,
 898          * and may need to be copied from the swappage read in.
 899          */
 900 repeat:
 901         if (!filepage)
 902                 filepage = find_lock_page(mapping, idx);
 903         if (filepage && PageUptodate(filepage))
 904                 goto done;
 905         error = 0;
 906         if (sgp == SGP_QUICK)
 907                 goto failed;
 908
 909         spin_lock(&info->lock);
 910         shmem_recalc_inode(inode);
 911         entry = shmem_swp_alloc(info, idx, sgp);
 912         if (IS_ERR(entry)) {
 913                 spin_unlock(&info->lock);
 914                 error = PTR_ERR(entry);
 915                 goto failed;
 916         }
 917         swap = *entry;
 918
 919         if (swap.val) {
 920                 /* Look it up and read it in.. */
 921                 swappage = lookup_swap_cache(swap);
 922                 if (!swappage) {
 923                         shmem_swp_unmap(entry);
 924                         spin_unlock(&info->lock);
 925                         /* here we actually do the io */
 926                         if (type && *type == VM_FAULT_MINOR) {
 927                                 inc_page_state(pgmajfault);
 928                                 *type = VM_FAULT_MAJOR;
 929                         }
 930                         swappage = shmem_swapin(info, swap, idx);
 931                         if (!swappage) {
 932                                 spin_lock(&info->lock);
 933                                 entry = shmem_swp_alloc(info, idx, sgp);
 934                                 if (IS_ERR(entry))
 935                                         error = PTR_ERR(entry);
 936                                 else {
 937                                         if (entry->val == swap.val)
 938                                                 error = -ENOMEM;
 939                                         shmem_swp_unmap(entry);
 940                                 }
 941                                 spin_unlock(&info->lock);
 942                                 if (error)
 943                                         goto failed;
 944                                 goto repeat;
 945                         }
 946                         wait_on_page_locked(swappage);
 947                         page_cache_release(swappage);
 948                         goto repeat;
 949                 }
 950
 951                 /* We have to do this with page locked to prevent races */
 952                 if (TestSetPageLocked(swappage)) {
 953                         shmem_swp_unmap(entry);
 954                         spin_unlock(&info->lock);
 955                         wait_on_page_locked(swappage);
 956                         page_cache_release(swappage);
 957                         goto repeat;
 958                 }
 959                 if (PageWriteback(swappage)) {
 960                         shmem_swp_unmap(entry);
 961                         spin_unlock(&info->lock);
 962                         wait_on_page_writeback(swappage);
 963                         unlock_page(swappage);
 964                         page_cache_release(swappage);
 965                         goto repeat;
 966                 }
 967                 if (!PageUptodate(swappage)) {
 968                         shmem_swp_unmap(entry);
 969                         spin_unlock(&info->lock);
 970                         unlock_page(swappage);
 971                         page_cache_release(swappage);
 972                         error = -EIO;
 973                         goto failed;
 974                 }
 975
 976                 if (filepage) {
 977                         shmem_swp_set(info, entry, 0);
 978                         shmem_swp_unmap(entry);
 979                         delete_from_swap_cache(swappage);
 980                         spin_unlock(&info->lock);
 981                         copy_highpage(filepage, swappage);
 982                         unlock_page(swappage);
 983                         page_cache_release(swappage);
 984                         flush_dcache_page(filepage);
 985                         SetPageUptodate(filepage);
 986                         set_page_dirty(filepage);
 987                         swap_free(swap);
 988                 } else if (!(error = move_from_swap_cache(
 989                                 swappage, idx, mapping))) {
 990                         info->flags |= SHMEM_PAGEIN;
 991                         shmem_swp_set(info, entry, 0);
 992                         shmem_swp_unmap(entry);
 993                         spin_unlock(&info->lock);
 994                         filepage = swappage;
 995                         swap_free(swap);
 996                 } else {
 997                         shmem_swp_unmap(entry);
 998                         spin_unlock(&info->lock);
 999                         unlock_page(swappage);
1000                         page_cache_release(swappage);
1001                         if (error == -ENOMEM) {
1002                                 /* let kswapd refresh zone for GFP_ATOMICs */
1003                                 blk_congestion_wait(WRITE, HZ/50);
1004                         }
1005                         goto repeat;
1006                 }
1007         } else if (sgp == SGP_READ && !filepage) {
1008                 shmem_swp_unmap(entry);
1009                 filepage = find_get_page(mapping, idx);
1010                 if (filepage &&
1011                     (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1012                         spin_unlock(&info->lock);
1013                         wait_on_page_locked(filepage);
1014                         page_cache_release(filepage);
1015                         filepage = NULL;
1016                         goto repeat;
1017                 }
1018                 spin_unlock(&info->lock);
1019         } else {
1020                 shmem_swp_unmap(entry);
1021                 sbinfo = SHMEM_SB(inode->i_sb);
1022                 if (sbinfo) {
1023                         spin_lock(&sbinfo->stat_lock);
1024                         if (sbinfo->free_blocks == 0 ||
1025                             shmem_acct_block(info->flags)) {
1026                                 spin_unlock(&sbinfo->stat_lock);
1027                                 spin_unlock(&info->lock);
1028                                 error = -ENOSPC;
1029                                 goto failed;
1030                         }
1031                         sbinfo->free_blocks--;
1032                         inode->i_blocks += BLOCKS_PER_PAGE;
1033                         spin_unlock(&sbinfo->stat_lock);
1034                 } else if (shmem_acct_block(info->flags)) {
1035                         spin_unlock(&info->lock);
1036                         error = -ENOSPC;
1037                         goto failed;
1038                 }
1039
1040                 if (!filepage) {
1041                         spin_unlock(&info->lock);
1042                         filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
1043                                                     info,
1044                                                     idx);
1045                         if (!filepage) {
1046                                 shmem_unacct_blocks(info->flags, 1);
1047                                 shmem_free_block(inode);
1048                                 error = -ENOMEM;
1049                                 goto failed;
1050                         }
1051
1052                         spin_lock(&info->lock);
1053                         entry = shmem_swp_alloc(info, idx, sgp);
1054                         if (IS_ERR(entry))
1055                                 error = PTR_ERR(entry);
1056                         else {
1057                                 swap = *entry;
1058                                 shmem_swp_unmap(entry);
1059                         }
1060                         if (error || swap.val || 0 != add_to_page_cache_lru(
1061                                         filepage, mapping, idx, GFP_ATOMIC)) {
1062                                 spin_unlock(&info->lock);
1063                                 page_cache_release(filepage);
1064                                 shmem_unacct_blocks(info->flags, 1);
1065                                 shmem_free_block(inode);
1066                                 filepage = NULL;
1067                                 if (error)
1068                                         goto failed;
1069                                 goto repeat;
1070                         }
1071                         info->flags |= SHMEM_PAGEIN;
1072                 }
1073
1074                 info->alloced++;
1075                 spin_unlock(&info->lock);
1076                 clear_highpage(filepage);
1077                 flush_dcache_page(filepage);
1078                 SetPageUptodate(filepage);
1079         }
1080 done:
1081         if (*pagep != filepage) {
1082                 unlock_page(filepage);
1083                 *pagep = filepage;
1084         }
1085         return 0;
1086
1087 failed:
1088         if (*pagep != filepage) {
1089                 unlock_page(filepage);
1090                 page_cache_release(filepage);
1091         }
1092         return error;
1093 }
1094
1095 struct page *shmem_nopage(struct vm_area_struct *vma, unsigned long address, int *type)
1096 {
1097         struct inode *inode = vma->vm_file->f_dentry->d_inode;
1098         struct page *page = NULL;
1099         unsigned long idx;
1100         int error;
1101
1102         idx = (address - vma->vm_start) >> PAGE_SHIFT;
1103         idx += vma->vm_pgoff;
1104         idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT;
1105
1106         error = shmem_getpage(inode, idx, &page, SGP_CACHE, type);
1107         if (error)
1108                 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS;
1109
1110         mark_page_accessed(page);
1111         return page;
1112 }
1113
1114 static int shmem_populate(struct vm_area_struct *vma,
1115         unsigned long addr, unsigned long len,
1116         pgprot_t prot, unsigned long pgoff, int nonblock)
1117 {
1118         struct inode *inode = vma->vm_file->f_dentry->d_inode;
1119         struct mm_struct *mm = vma->vm_mm;
1120         enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE;
1121         unsigned long size;
1122
1123         size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1124         if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size)
1125                 return -EINVAL;
1126
1127         while ((long) len > 0) {
1128                 struct page *page = NULL;
1129                 int err;
1130                 /*
1131                  * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE
1132                  */
1133                 err = shmem_getpage(inode, pgoff, &page, sgp, NULL);
1134                 if (err)
1135                         return err;
1136                 if (page) {
1137                         mark_page_accessed(page);
1138                         err = install_page(mm, vma, addr, page, prot);
1139                         if (err) {
1140                                 page_cache_release(page);
1141                                 return err;
1142                         }
1143                 } else if (nonblock) {
1144                         err = install_file_pte(mm, vma, addr, pgoff, prot);
1145                         if (err)
1146                                 return err;
1147                 }
1148
1149                 len -= PAGE_SIZE;
1150                 addr += PAGE_SIZE;
1151                 pgoff++;
1152         }
1153         return 0;
1154 }
1155
1156 #ifdef CONFIG_NUMA
1157 int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1158 {
1159         struct inode *i = vma->vm_file->f_dentry->d_inode;
1160         return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1161 }
1162
1163 struct mempolicy *
1164 shmem_get_policy(struct vm_area_struct *vma, unsigned long addr)
1165 {
1166         struct inode *i = vma->vm_file->f_dentry->d_inode;
1167         unsigned long idx;
1168
1169         idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1170         return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1171 }
1172 #endif
1173
1174 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1175 {
1176         struct inode *inode = file->f_dentry->d_inode;
1177         struct shmem_inode_info *info = SHMEM_I(inode);
1178         int retval = -ENOMEM;
1179
1180         spin_lock(&info->lock);
1181         if (lock && !(info->flags & VM_LOCKED)) {
1182                 if (!user_shm_lock(inode->i_size, user))
1183                         goto out_nomem;
1184                 info->flags |= VM_LOCKED;
1185         }
1186         if (!lock && (info->flags & VM_LOCKED) && user) {
1187                 user_shm_unlock(inode->i_size, user);
1188                 info->flags &= ~VM_LOCKED;
1189         }
1190         retval = 0;
1191 out_nomem:
1192         spin_unlock(&info->lock);
1193         return retval;
1194 }
1195
1196 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1197 {
1198         file_accessed(file);
1199         vma->vm_ops = &shmem_vm_ops;
1200         return 0;
1201 }
1202
1203 static struct inode *
1204 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1205 {
1206         struct inode *inode;
1207         struct shmem_inode_info *info;
1208         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1209
1210         if (sbinfo) {
1211                 spin_lock(&sbinfo->stat_lock);
1212                 if (!sbinfo->free_inodes) {
1213                         spin_unlock(&sbinfo->stat_lock);
1214                         return NULL;
1215                 }
1216                 sbinfo->free_inodes--;
1217                 spin_unlock(&sbinfo->stat_lock);
1218         }
1219
1220         inode = new_inode(sb);
1221         if (inode) {
1222                 inode->i_mode = mode;
1223                 inode->i_uid = current->fsuid;
1224                 inode->i_gid = current->fsgid;
1225                 inode->i_blksize = PAGE_CACHE_SIZE;
1226                 inode->i_blocks = 0;
1227                 inode->i_mapping->a_ops = &shmem_aops;
1228                 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1229                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1230                 info = SHMEM_I(inode);
1231                 memset(info, 0, (char *)inode - (char *)info);
1232                 spin_lock_init(&info->lock);
1233                 mpol_shared_policy_init(&info->policy);
1234                 INIT_LIST_HEAD(&info->swaplist);
1235
1236                 switch (mode & S_IFMT) {
1237                 default:
1238                         init_special_inode(inode, mode, dev);
1239                         break;
1240                 case S_IFREG:
1241                         inode->i_op = &shmem_inode_operations;
1242                         inode->i_fop = &shmem_file_operations;
1243                         break;
1244                 case S_IFDIR:
1245                         inode->i_nlink++;
1246                         /* Some things misbehave if size == 0 on a directory */
1247                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
1248                         inode->i_op = &shmem_dir_inode_operations;
1249                         inode->i_fop = &simple_dir_operations;
1250                         break;
1251                 case S_IFLNK:
1252                         break;
1253                 }
1254         }
1255         return inode;
1256 }
1257
1258 #ifdef CONFIG_TMPFS
1259
1260 static int shmem_set_size(struct shmem_sb_info *sbinfo,
1261                           unsigned long max_blocks, unsigned long max_inodes)
1262 {
1263         int error;
1264         unsigned long blocks, inodes;
1265
1266         spin_lock(&sbinfo->stat_lock);
1267         blocks = sbinfo->max_blocks - sbinfo->free_blocks;
1268         inodes = sbinfo->max_inodes - sbinfo->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         sbinfo->max_blocks  = max_blocks;
1276         sbinfo->free_blocks = max_blocks - blocks;
1277         sbinfo->max_inodes  = max_inodes;
1278         sbinfo->free_inodes = max_inodes - inodes;
1279 out:
1280         spin_unlock(&sbinfo->stat_lock);
1281         return error;
1282 }
1283
1284 static struct inode_operations shmem_symlink_inode_operations;
1285 static struct inode_operations shmem_symlink_inline_operations;
1286
1287 /*
1288  * Normally tmpfs makes no use of shmem_prepare_write, but it
1289  * lets a tmpfs file be used read-write below the loop driver.
1290  */
1291 static int
1292 shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1293 {
1294         struct inode *inode = page->mapping->host;
1295         return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL);
1296 }
1297
1298 static ssize_t
1299 shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1300 {
1301         struct inode    *inode = file->f_dentry->d_inode;
1302         loff_t          pos;
1303         unsigned long   written;
1304         ssize_t         err;
1305
1306         if ((ssize_t) count < 0)
1307                 return -EINVAL;
1308
1309         if (!access_ok(VERIFY_READ, buf, count))
1310                 return -EFAULT;
1311
1312         down(&inode->i_sem);
1313
1314         pos = *ppos;
1315         written = 0;
1316
1317         err = generic_write_checks(file, &pos, &count, 0);
1318         if (err || !count)
1319                 goto out;
1320
1321         err = remove_suid(file->f_dentry);
1322         if (err)
1323                 goto out;
1324
1325         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1326
1327         do {
1328                 struct page *page = NULL;
1329                 unsigned long bytes, index, offset;
1330                 char *kaddr;
1331                 int left;
1332
1333                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1334                 index = pos >> PAGE_CACHE_SHIFT;
1335                 bytes = PAGE_CACHE_SIZE - offset;
1336                 if (bytes > count)
1337                         bytes = count;
1338
1339                 /*
1340                  * We don't hold page lock across copy from user -
1341                  * what would it guard against? - so no deadlock here.
1342                  * But it still may be a good idea to prefault below.
1343                  */
1344
1345                 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1346                 if (err)
1347                         break;
1348
1349                 left = bytes;
1350                 if (PageHighMem(page)) {
1351                         volatile unsigned char dummy;
1352                         __get_user(dummy, buf);
1353                         __get_user(dummy, buf + bytes - 1);
1354
1355                         kaddr = kmap_atomic(page, KM_USER0);
1356                         left = __copy_from_user_inatomic(kaddr + offset,
1357                                                         buf, bytes);
1358                         kunmap_atomic(kaddr, KM_USER0);
1359                 }
1360                 if (left) {
1361                         kaddr = kmap(page);
1362                         left = __copy_from_user(kaddr + offset, buf, bytes);
1363                         kunmap(page);
1364                 }
1365
1366                 written += bytes;
1367                 count -= bytes;
1368                 pos += bytes;
1369                 buf += bytes;
1370                 if (pos > inode->i_size)
1371                         i_size_write(inode, pos);
1372
1373                 flush_dcache_page(page);
1374                 set_page_dirty(page);
1375                 mark_page_accessed(page);
1376                 page_cache_release(page);
1377
1378                 if (left) {
1379                         pos -= left;
1380                         written -= left;
1381                         err = -EFAULT;
1382                         break;
1383                 }
1384
1385                 /*
1386                  * Our dirty pages are not counted in nr_dirty,
1387                  * and we do not attempt to balance dirty pages.
1388                  */
1389
1390                 cond_resched();
1391         } while (count);
1392
1393         *ppos = pos;
1394         if (written)
1395                 err = written;
1396 out:
1397         up(&inode->i_sem);
1398         return err;
1399 }
1400
1401 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1402 {
1403         struct inode *inode = filp->f_dentry->d_inode;
1404         struct address_space *mapping = inode->i_mapping;
1405         unsigned long index, offset;
1406
1407         index = *ppos >> PAGE_CACHE_SHIFT;
1408         offset = *ppos & ~PAGE_CACHE_MASK;
1409
1410         for (;;) {
1411                 struct page *page = NULL;
1412                 unsigned long end_index, nr, ret;
1413                 loff_t i_size = i_size_read(inode);
1414
1415                 end_index = i_size >> PAGE_CACHE_SHIFT;
1416                 if (index > end_index)
1417                         break;
1418                 if (index == end_index) {
1419                         nr = i_size & ~PAGE_CACHE_MASK;
1420                         if (nr <= offset)
1421                                 break;
1422                 }
1423
1424                 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1425                 if (desc->error) {
1426                         if (desc->error == -EINVAL)
1427                                 desc->error = 0;
1428                         break;
1429                 }
1430
1431                 /*
1432                  * We must evaluate after, since reads (unlike writes)
1433                  * are called without i_sem protection against truncate
1434                  */
1435                 nr = PAGE_CACHE_SIZE;
1436                 i_size = i_size_read(inode);
1437                 end_index = i_size >> PAGE_CACHE_SHIFT;
1438                 if (index == end_index) {
1439                         nr = i_size & ~PAGE_CACHE_MASK;
1440                         if (nr <= offset) {
1441                                 if (page)
1442                                         page_cache_release(page);
1443                                 break;
1444                         }
1445                 }
1446                 nr -= offset;
1447
1448                 if (page) {
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                 } else
1462                         page = ZERO_PAGE(0);
1463
1464                 /*
1465                  * Ok, we have the page, and it's up-to-date, so
1466                  * now we can copy it to user space...
1467                  *
1468                  * The actor routine returns how many bytes were actually used..
1469                  * NOTE! This may not be the same as how much of a user buffer
1470                  * we filled up (we may be padding etc), so we can only update
1471                  * "pos" here (the actor routine has to update the user buffer
1472                  * pointers and the remaining count).
1473                  */
1474                 ret = actor(desc, page, offset, nr);
1475                 offset += ret;
1476                 index += offset >> PAGE_CACHE_SHIFT;
1477                 offset &= ~PAGE_CACHE_MASK;
1478
1479                 page_cache_release(page);
1480                 if (ret != nr || !desc->count)
1481                         break;
1482
1483                 cond_resched();
1484         }
1485
1486         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1487         file_accessed(filp);
1488 }
1489
1490 static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1491 {
1492         read_descriptor_t desc;
1493
1494         if ((ssize_t) count < 0)
1495                 return -EINVAL;
1496         if (!access_ok(VERIFY_WRITE, buf, count))
1497                 return -EFAULT;
1498         if (!count)
1499                 return 0;
1500
1501         desc.written = 0;
1502         desc.count = count;
1503         desc.arg.buf = buf;
1504         desc.error = 0;
1505
1506         do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1507         if (desc.written)
1508                 return desc.written;
1509         return desc.error;
1510 }
1511
1512 static ssize_t shmem_file_sendfile(struct file *in_file, loff_t *ppos,
1513                          size_t count, read_actor_t actor, void *target)
1514 {
1515         read_descriptor_t desc;
1516
1517         if (!count)
1518                 return 0;
1519
1520         desc.written = 0;
1521         desc.count = count;
1522         desc.arg.data = target;
1523         desc.error = 0;
1524
1525         do_shmem_file_read(in_file, ppos, &desc, actor);
1526         if (desc.written)
1527                 return desc.written;
1528         return desc.error;
1529 }
1530
1531 static int shmem_statfs(struct super_block *sb, struct kstatfs *buf)
1532 {
1533         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1534
1535         buf->f_type = TMPFS_MAGIC;
1536         buf->f_bsize = PAGE_CACHE_SIZE;
1537         buf->f_namelen = NAME_MAX;
1538         if (sbinfo) {
1539                 spin_lock(&sbinfo->stat_lock);
1540                 buf->f_blocks = sbinfo->max_blocks;
1541                 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1542                 buf->f_files = sbinfo->max_inodes;
1543                 buf->f_ffree = sbinfo->free_inodes;
1544                 spin_unlock(&sbinfo->stat_lock);
1545         }
1546         /* else leave those fields 0 like simple_statfs */
1547         return 0;
1548 }
1549
1550 /*
1551  * File creation. Allocate an inode, and we're done..
1552  */
1553 static int
1554 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1555 {
1556         struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1557         int error = -ENOSPC;
1558
1559         if (inode) {
1560                 if (dir->i_mode & S_ISGID) {
1561                         inode->i_gid = dir->i_gid;
1562                         if (S_ISDIR(mode))
1563                                 inode->i_mode |= S_ISGID;
1564                 }
1565                 dir->i_size += BOGO_DIRENT_SIZE;
1566                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1567                 d_instantiate(dentry, inode);
1568                 dget(dentry); /* Extra count - pin the dentry in core */
1569                 error = 0;
1570         }
1571         return error;
1572 }
1573
1574 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1575 {
1576         int error;
1577
1578         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1579                 return error;
1580         dir->i_nlink++;
1581         return 0;
1582 }
1583
1584 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1585                 struct nameidata *nd)
1586 {
1587         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1588 }
1589
1590 /*
1591  * Link a file..
1592  */
1593 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1594 {
1595         struct inode *inode = old_dentry->d_inode;
1596         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1597
1598         /*
1599          * No ordinary (disk based) filesystem counts links as inodes;
1600          * but each new link needs a new dentry, pinning lowmem, and
1601          * tmpfs dentries cannot be pruned until they are unlinked.
1602          */
1603         if (sbinfo) {
1604                 spin_lock(&sbinfo->stat_lock);
1605                 if (!sbinfo->free_inodes) {
1606                         spin_unlock(&sbinfo->stat_lock);
1607                         return -ENOSPC;
1608                 }
1609                 sbinfo->free_inodes--;
1610                 spin_unlock(&sbinfo->stat_lock);
1611         }
1612
1613         dir->i_size += BOGO_DIRENT_SIZE;
1614         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1615         inode->i_nlink++;
1616         atomic_inc(&inode->i_count);    /* New dentry reference */
1617         dget(dentry);           /* Extra pinning count for the created dentry */
1618         d_instantiate(dentry, inode);
1619         return 0;
1620 }
1621
1622 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1623 {
1624         struct inode *inode = dentry->d_inode;
1625
1626         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) {
1627                 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1628                 if (sbinfo) {
1629                         spin_lock(&sbinfo->stat_lock);
1630                         sbinfo->free_inodes++;
1631                         spin_unlock(&sbinfo->stat_lock);
1632                 }
1633         }
1634
1635         dir->i_size -= BOGO_DIRENT_SIZE;
1636         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1637         inode->i_nlink--;
1638         dput(dentry);   /* Undo the count from "create" - this does all the work */
1639         return 0;
1640 }
1641
1642 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1643 {
1644         if (!simple_empty(dentry))
1645                 return -ENOTEMPTY;
1646
1647         dir->i_nlink--;
1648         return shmem_unlink(dir, dentry);
1649 }
1650
1651 /*
1652  * The VFS layer already does all the dentry stuff for rename,
1653  * we just have to decrement the usage count for the target if
1654  * it exists so that the VFS layer correctly free's it when it
1655  * gets overwritten.
1656  */
1657 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1658 {
1659         struct inode *inode = old_dentry->d_inode;
1660         int they_are_dirs = S_ISDIR(inode->i_mode);
1661
1662         if (!simple_empty(new_dentry))
1663                 return -ENOTEMPTY;
1664
1665         if (new_dentry->d_inode) {
1666                 (void) shmem_unlink(new_dir, new_dentry);
1667                 if (they_are_dirs)
1668                         old_dir->i_nlink--;
1669         } else if (they_are_dirs) {
1670                 old_dir->i_nlink--;
1671                 new_dir->i_nlink++;
1672         }
1673
1674         old_dir->i_size -= BOGO_DIRENT_SIZE;
1675         new_dir->i_size += BOGO_DIRENT_SIZE;
1676         old_dir->i_ctime = old_dir->i_mtime =
1677         new_dir->i_ctime = new_dir->i_mtime =
1678         inode->i_ctime = CURRENT_TIME;
1679         return 0;
1680 }
1681
1682 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1683 {
1684         int error;
1685         int len;
1686         struct inode *inode;
1687         struct page *page = NULL;
1688         char *kaddr;
1689         struct shmem_inode_info *info;
1690
1691         len = strlen(symname) + 1;
1692         if (len > PAGE_CACHE_SIZE)
1693                 return -ENAMETOOLONG;
1694
1695         inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1696         if (!inode)
1697                 return -ENOSPC;
1698
1699         info = SHMEM_I(inode);
1700         inode->i_size = len-1;
1701         if (len <= (char *)inode - (char *)info) {
1702                 /* do it inline */
1703                 memcpy(info, symname, len);
1704                 inode->i_op = &shmem_symlink_inline_operations;
1705         } else {
1706                 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1707                 if (error) {
1708                         iput(inode);
1709                         return error;
1710                 }
1711                 inode->i_op = &shmem_symlink_inode_operations;
1712                 kaddr = kmap_atomic(page, KM_USER0);
1713                 memcpy(kaddr, symname, len);
1714                 kunmap_atomic(kaddr, KM_USER0);
1715                 set_page_dirty(page);
1716                 page_cache_release(page);
1717         }
1718         if (dir->i_mode & S_ISGID)
1719                 inode->i_gid = dir->i_gid;
1720         dir->i_size += BOGO_DIRENT_SIZE;
1721         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1722         d_instantiate(dentry, inode);
1723         dget(dentry);
1724         return 0;
1725 }
1726
1727 static int shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1728 {
1729         nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1730         return 0;
1731 }
1732
1733 static int shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1734 {
1735         struct page *page = NULL;
1736         int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1737         nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1738         return 0;
1739 }
1740
1741 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd)
1742 {
1743         if (!IS_ERR(nd_get_link(nd))) {
1744                 struct page *page;
1745
1746                 page = find_get_page(dentry->d_inode->i_mapping, 0);
1747                 if (!page)
1748                         BUG();
1749                 kunmap(page);
1750                 mark_page_accessed(page);
1751                 page_cache_release(page);
1752                 page_cache_release(page);
1753         }
1754 }
1755
1756 static struct inode_operations shmem_symlink_inline_operations = {
1757         .readlink       = generic_readlink,
1758         .follow_link    = shmem_follow_link_inline,
1759 };
1760
1761 static struct inode_operations shmem_symlink_inode_operations = {
1762         .truncate       = shmem_truncate,
1763         .readlink       = generic_readlink,
1764         .follow_link    = shmem_follow_link,
1765         .put_link       = shmem_put_link,
1766 };
1767
1768 static int shmem_parse_options(char *options, int *mode, uid_t *uid, gid_t *gid, unsigned long *blocks, unsigned long *inodes)
1769 {
1770         char *this_char, *value, *rest;
1771
1772         while ((this_char = strsep(&options, ",")) != NULL) {
1773                 if (!*this_char)
1774                         continue;
1775                 if ((value = strchr(this_char,'=')) != NULL) {
1776                         *value++ = 0;
1777                 } else {
1778                         printk(KERN_ERR
1779                             "tmpfs: No value for mount option '%s'\n",
1780                             this_char);
1781                         return 1;
1782                 }
1783
1784                 if (!strcmp(this_char,"size")) {
1785                         unsigned long long size;
1786                         size = memparse(value,&rest);
1787                         if (*rest == '%') {
1788                                 size <<= PAGE_SHIFT;
1789                                 size *= totalram_pages;
1790                                 do_div(size, 100);
1791                                 rest++;
1792                         }
1793                         if (*rest)
1794                                 goto bad_val;
1795                         *blocks = size >> PAGE_CACHE_SHIFT;
1796                 } else if (!strcmp(this_char,"nr_blocks")) {
1797                         *blocks = memparse(value,&rest);
1798                         if (*rest)
1799                                 goto bad_val;
1800                 } else if (!strcmp(this_char,"nr_inodes")) {
1801                         *inodes = memparse(value,&rest);
1802                         if (*rest)
1803                                 goto bad_val;
1804                 } else if (!strcmp(this_char,"mode")) {
1805                         if (!mode)
1806                                 continue;
1807                         *mode = simple_strtoul(value,&rest,8);
1808                         if (*rest)
1809                                 goto bad_val;
1810                 } else if (!strcmp(this_char,"uid")) {
1811                         if (!uid)
1812                                 continue;
1813                         *uid = simple_strtoul(value,&rest,0);
1814                         if (*rest)
1815                                 goto bad_val;
1816                 } else if (!strcmp(this_char,"gid")) {
1817                         if (!gid)
1818                                 continue;
1819                         *gid = simple_strtoul(value,&rest,0);
1820                         if (*rest)
1821                                 goto bad_val;
1822                 } else {
1823                         printk(KERN_ERR "tmpfs: Bad mount option %s\n",
1824                                this_char);
1825                         return 1;
1826                 }
1827         }
1828         return 0;
1829
1830 bad_val:
1831         printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
1832                value, this_char);
1833         return 1;
1834
1835 }
1836
1837 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
1838 {
1839         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1840         unsigned long max_blocks = 0;
1841         unsigned long max_inodes = 0;
1842
1843         if (sbinfo) {
1844                 max_blocks = sbinfo->max_blocks;
1845                 max_inodes = sbinfo->max_inodes;
1846         }
1847         if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks, &max_inodes))
1848                 return -EINVAL;
1849         /* Keep it simple: disallow limited <-> unlimited remount */
1850         if ((max_blocks || max_inodes) == !sbinfo)
1851                 return -EINVAL;
1852         /* But allow the pointless unlimited -> unlimited remount */
1853         if (!sbinfo)
1854                 return 0;
1855         return shmem_set_size(sbinfo, max_blocks, max_inodes);
1856 }
1857 #endif
1858
1859 static void shmem_put_super(struct super_block *sb)
1860 {
1861         kfree(sb->s_fs_info);
1862         sb->s_fs_info = NULL;
1863 }
1864
1865 static int shmem_fill_super(struct super_block *sb,
1866                             void *data, int silent)
1867 {
1868         struct inode *inode;
1869         struct dentry *root;
1870         int mode   = S_IRWXUGO | S_ISVTX;
1871         uid_t uid = current->fsuid;
1872         gid_t gid = current->fsgid;
1873         int err = -ENOMEM;
1874
1875 #ifdef CONFIG_TMPFS
1876         unsigned long blocks = 0;
1877         unsigned long inodes = 0;
1878
1879         /*
1880          * Per default we only allow half of the physical ram per
1881          * tmpfs instance, limiting inodes to one per page of lowmem;
1882          * but the internal instance is left unlimited.
1883          */
1884         if (!(sb->s_flags & MS_NOUSER)) {
1885                 blocks = totalram_pages / 2;
1886                 inodes = totalram_pages - totalhigh_pages;
1887                 if (inodes > blocks)
1888                         inodes = blocks;
1889
1890                 if (shmem_parse_options(data, &mode,
1891                                         &uid, &gid, &blocks, &inodes))
1892                         return -EINVAL;
1893         }
1894
1895         if (blocks || inodes) {
1896                 struct shmem_sb_info *sbinfo;
1897                 sbinfo = kmalloc(sizeof(struct shmem_sb_info), GFP_KERNEL);
1898                 if (!sbinfo)
1899                         return -ENOMEM;
1900                 sb->s_fs_info = sbinfo;
1901                 spin_lock_init(&sbinfo->stat_lock);
1902                 sbinfo->max_blocks = blocks;
1903                 sbinfo->free_blocks = blocks;
1904                 sbinfo->max_inodes = inodes;
1905                 sbinfo->free_inodes = inodes;
1906         }
1907 #endif
1908
1909         sb->s_maxbytes = SHMEM_MAX_BYTES;
1910         sb->s_blocksize = PAGE_CACHE_SIZE;
1911         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1912         sb->s_magic = TMPFS_MAGIC;
1913         sb->s_op = &shmem_ops;
1914         inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
1915         if (!inode)
1916                 goto failed;
1917         inode->i_uid = uid;
1918         inode->i_gid = gid;
1919         root = d_alloc_root(inode);
1920         if (!root)
1921                 goto failed_iput;
1922         sb->s_root = root;
1923         return 0;
1924
1925 failed_iput:
1926         iput(inode);
1927 failed:
1928         shmem_put_super(sb);
1929         return err;
1930 }
1931
1932 static kmem_cache_t *shmem_inode_cachep;
1933
1934 static struct inode *shmem_alloc_inode(struct super_block *sb)
1935 {
1936         struct shmem_inode_info *p;
1937         p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, SLAB_KERNEL);
1938         if (!p)
1939                 return NULL;
1940         return &p->vfs_inode;
1941 }
1942
1943 static void shmem_destroy_inode(struct inode *inode)
1944 {
1945         mpol_free_shared_policy(&SHMEM_I(inode)->policy);
1946         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
1947 }
1948
1949 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
1950 {
1951         struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
1952
1953         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
1954             SLAB_CTOR_CONSTRUCTOR) {
1955                 inode_init_once(&p->vfs_inode);
1956         }
1957 }
1958
1959 static int init_inodecache(void)
1960 {
1961         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
1962                                 sizeof(struct shmem_inode_info),
1963                                 0, 0, init_once, NULL);
1964         if (shmem_inode_cachep == NULL)
1965                 return -ENOMEM;
1966         return 0;
1967 }
1968
1969 static void destroy_inodecache(void)
1970 {
1971         if (kmem_cache_destroy(shmem_inode_cachep))
1972                 printk(KERN_INFO "shmem_inode_cache: not all structures were freed\n");
1973 }
1974
1975 static struct address_space_operations shmem_aops = {
1976         .writepage      = shmem_writepage,
1977         .set_page_dirty = __set_page_dirty_nobuffers,
1978 #ifdef CONFIG_TMPFS
1979         .prepare_write  = shmem_prepare_write,
1980         .commit_write   = simple_commit_write,
1981 #endif
1982 };
1983
1984 static struct file_operations shmem_file_operations = {
1985         .mmap           = shmem_mmap,
1986 #ifdef CONFIG_TMPFS
1987         .llseek         = generic_file_llseek,
1988         .read           = shmem_file_read,
1989         .write          = shmem_file_write,
1990         .fsync          = simple_sync_file,
1991         .sendfile       = shmem_file_sendfile,
1992 #endif
1993 };
1994
1995 static struct inode_operations shmem_inode_operations = {
1996         .truncate       = shmem_truncate,
1997         .setattr        = shmem_notify_change,
1998 };
1999
2000 static struct inode_operations shmem_dir_inode_operations = {
2001 #ifdef CONFIG_TMPFS
2002         .create         = shmem_create,
2003         .lookup         = simple_lookup,
2004         .link           = shmem_link,
2005         .unlink         = shmem_unlink,
2006         .symlink        = shmem_symlink,
2007         .mkdir          = shmem_mkdir,
2008         .rmdir          = shmem_rmdir,
2009         .mknod          = shmem_mknod,
2010         .rename         = shmem_rename,
2011 #endif
2012 };
2013
2014 static struct super_operations shmem_ops = {
2015         .alloc_inode    = shmem_alloc_inode,
2016         .destroy_inode  = shmem_destroy_inode,
2017 #ifdef CONFIG_TMPFS
2018         .statfs         = shmem_statfs,
2019         .remount_fs     = shmem_remount_fs,
2020 #endif
2021         .delete_inode   = shmem_delete_inode,
2022         .drop_inode     = generic_delete_inode,
2023         .put_super      = shmem_put_super,
2024 };
2025
2026 static struct vm_operations_struct shmem_vm_ops = {
2027         .nopage         = shmem_nopage,
2028         .populate       = shmem_populate,
2029 #ifdef CONFIG_NUMA
2030         .set_policy     = shmem_set_policy,
2031         .get_policy     = shmem_get_policy,
2032 #endif
2033 };
2034
2035 static struct super_block *shmem_get_sb(struct file_system_type *fs_type,
2036         int flags, const char *dev_name, void *data)
2037 {
2038         return get_sb_nodev(fs_type, flags, data, shmem_fill_super);
2039 }
2040
2041 static struct file_system_type tmpfs_fs_type = {
2042         .owner          = THIS_MODULE,
2043         .name           = "tmpfs",
2044         .get_sb         = shmem_get_sb,
2045         .kill_sb        = kill_litter_super,
2046 };
2047 static struct vfsmount *shm_mnt;
2048
2049 static int __init init_tmpfs(void)
2050 {
2051         int error;
2052
2053         error = init_inodecache();
2054         if (error)
2055                 goto out3;
2056
2057         error = register_filesystem(&tmpfs_fs_type);
2058         if (error) {
2059                 printk(KERN_ERR "Could not register tmpfs\n");
2060                 goto out2;
2061         }
2062 #ifdef CONFIG_TMPFS
2063         devfs_mk_dir("shm");
2064 #endif
2065         shm_mnt = do_kern_mount(tmpfs_fs_type.name, MS_NOUSER,
2066                                 tmpfs_fs_type.name, NULL);
2067         if (IS_ERR(shm_mnt)) {
2068                 error = PTR_ERR(shm_mnt);
2069                 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2070                 goto out1;
2071         }
2072         return 0;
2073
2074 out1:
2075         unregister_filesystem(&tmpfs_fs_type);
2076 out2:
2077         destroy_inodecache();
2078 out3:
2079         shm_mnt = ERR_PTR(error);
2080         return error;
2081 }
2082 module_init(init_tmpfs)
2083
2084 /*
2085  * shmem_file_setup - get an unlinked file living in tmpfs
2086  *
2087  * @name: name for dentry (to be seen in /proc/<pid>/maps
2088  * @size: size to be set for the file
2089  *
2090  */
2091 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2092 {
2093         int error;
2094         struct file *file;
2095         struct inode *inode;
2096         struct dentry *dentry, *root;
2097         struct qstr this;
2098
2099         if (IS_ERR(shm_mnt))
2100                 return (void *)shm_mnt;
2101
2102         if (size < 0 || size > SHMEM_MAX_BYTES)
2103                 return ERR_PTR(-EINVAL);
2104
2105         if (shmem_acct_size(flags, size))
2106                 return ERR_PTR(-ENOMEM);
2107
2108         error = -ENOMEM;
2109         this.name = name;
2110         this.len = strlen(name);
2111         this.hash = 0; /* will go */
2112         root = shm_mnt->mnt_root;
2113         dentry = d_alloc(root, &this);
2114         if (!dentry)
2115                 goto put_memory;
2116
2117         error = -ENFILE;
2118         file = get_empty_filp();
2119         if (!file)
2120                 goto put_dentry;
2121
2122         error = -ENOSPC;
2123         inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2124         if (!inode)
2125                 goto close_file;
2126
2127         SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2128         d_instantiate(dentry, inode);
2129         inode->i_size = size;
2130         inode->i_nlink = 0;     /* It is unlinked */
2131         file->f_vfsmnt = mntget(shm_mnt);
2132         file->f_dentry = dentry;
2133         file->f_mapping = inode->i_mapping;
2134         file->f_op = &shmem_file_operations;
2135         file->f_mode = FMODE_WRITE | FMODE_READ;
2136         return file;
2137
2138 close_file:
2139         put_filp(file);
2140 put_dentry:
2141         dput(dentry);
2142 put_memory:
2143         shmem_unacct_size(flags, size);
2144         return ERR_PTR(error);
2145 }
2146
2147 /*
2148  * shmem_zero_setup - setup a shared anonymous mapping
2149  *
2150  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2151  */
2152 int shmem_zero_setup(struct vm_area_struct *vma)
2153 {
2154         struct file *file;
2155         loff_t size = vma->vm_end - vma->vm_start;
2156
2157         file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2158         if (IS_ERR(file))
2159                 return PTR_ERR(file);
2160
2161         if (vma->vm_file)
2162                 fput(vma->vm_file);
2163         vma->vm_file = file;
2164         vma->vm_ops = &shmem_vm_ops;
2165         return 0;
2166 }
2167
2168 EXPORT_SYMBOL(shmem_file_setup);