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) 2002-2005 Hugh Dickins.
  10  * Copyright (C) 2002-2005 VERITAS Software Corporation.
  11  * Copyright (C) 2004 Andi Kleen, SuSE Labs
  12  *
  13  * Extended attribute support for tmpfs:
  14  * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
  15  * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
  16  *
  17  * This file is released under the GPL.
  18  */
  19
  20 /*
  21  * This virtual memory filesystem is heavily based on the ramfs. It
  22  * extends ramfs by the ability to use swap and honor resource limits
  23  * which makes it a completely usable filesystem.
  24  */
  25
  26 #include <linux/config.h>
  27 #include <linux/module.h>
  28 #include <linux/init.h>
  29 #include <linux/devfs_fs_kernel.h>
  30 #include <linux/fs.h>
  31 #include <linux/mm.h>
  32 #include <linux/mman.h>
  33 #include <linux/file.h>
  34 #include <linux/swap.h>
  35 #include <linux/pagemap.h>
  36 #include <linux/string.h>
  37 #include <linux/slab.h>
  38 #include <linux/backing-dev.h>
  39 #include <linux/shmem_fs.h>
  40 #include <linux/mount.h>
  41 #include <linux/writeback.h>
  42 #include <linux/vfs.h>
  43 #include <linux/blkdev.h>
  44 #include <linux/security.h>
  45 #include <linux/swapops.h>
  46 #include <linux/mempolicy.h>
  47 #include <linux/namei.h>
  48 #include <asm/uaccess.h>
  49 #include <asm/div64.h>
  50 #include <asm/pgtable.h>
  51
  52 /* This magic number is used in glibc for posix shared memory */
  53 #define TMPFS_MAGIC     0x01021994
  54
  55 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
  56 #define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
  57 #define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
  58
  59 #define SHMEM_MAX_INDEX  (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
  60 #define SHMEM_MAX_BYTES  ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
  61
  62 #define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
  63
  64 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
  65 #define SHMEM_PAGEIN     VM_READ
  66 #define SHMEM_TRUNCATE   VM_WRITE
  67
  68 /* Definition to limit shmem_truncate's steps between cond_rescheds */
  69 #define LATENCY_LIMIT    64
  70
  71 /* Pretend that each entry is of this size in directory's i_size */
  72 #define BOGO_DIRENT_SIZE 20
  73
  74 /* Keep swapped page count in private field of indirect struct page */
  75 #define nr_swapped              private
  76
  77 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
  78 enum sgp_type {
  79         SGP_QUICK,      /* don't try more than file page cache lookup */
  80         SGP_READ,       /* don't exceed i_size, don't allocate page */
  81         SGP_CACHE,      /* don't exceed i_size, may allocate page */
  82         SGP_WRITE,      /* may exceed i_size, may allocate page */
  83 };
  84
  85 static int shmem_getpage(struct inode *inode, unsigned long idx,
  86                          struct page **pagep, enum sgp_type sgp, int *type);
  87
  88 static inline struct page *shmem_dir_alloc(unsigned int gfp_mask)
  89 {
  90         /*
  91          * The above definition of ENTRIES_PER_PAGE, and the use of
  92          * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
  93          * might be reconsidered if it ever diverges from PAGE_SIZE.
  94          */
  95         return alloc_pages(gfp_mask, PAGE_CACHE_SHIFT-PAGE_SHIFT);
  96 }
  97
  98 static inline void shmem_dir_free(struct page *page)
  99 {
 100         __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
 101 }
 102
 103 static struct page **shmem_dir_map(struct page *page)
 104 {
 105         return (struct page **)kmap_atomic(page, KM_USER0);
 106 }
 107
 108 static inline void shmem_dir_unmap(struct page **dir)
 109 {
 110         kunmap_atomic(dir, KM_USER0);
 111 }
 112
 113 static swp_entry_t *shmem_swp_map(struct page *page)
 114 {
 115         return (swp_entry_t *)kmap_atomic(page, KM_USER1);
 116 }
 117
 118 static inline void shmem_swp_balance_unmap(void)
 119 {
 120         /*
 121          * When passing a pointer to an i_direct entry, to code which
 122          * also handles indirect entries and so will shmem_swp_unmap,
 123          * we must arrange for the preempt count to remain in balance.
 124          * What kmap_atomic of a lowmem page does depends on config
 125          * and architecture, so pretend to kmap_atomic some lowmem page.
 126          */
 127         (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
 128 }
 129
 130 static inline void shmem_swp_unmap(swp_entry_t *entry)
 131 {
 132         kunmap_atomic(entry, KM_USER1);
 133 }
 134
 135 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
 136 {
 137         return sb->s_fs_info;
 138 }
 139
 140 /*
 141  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
 142  * for shared memory and for shared anonymous (/dev/zero) mappings
 143  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
 144  * consistent with the pre-accounting of private mappings ...
 145  */
 146 static inline int shmem_acct_size(unsigned long flags, loff_t size)
 147 {
 148         return (flags & VM_ACCOUNT)?
 149                 security_vm_enough_memory(VM_ACCT(size)): 0;
 150 }
 151
 152 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
 153 {
 154         if (flags & VM_ACCOUNT)
 155                 vm_unacct_memory(VM_ACCT(size));
 156 }
 157
 158 /*
 159  * ... whereas tmpfs objects are accounted incrementally as
 160  * pages are allocated, in order to allow huge sparse files.
 161  * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
 162  * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
 163  */
 164 static inline int shmem_acct_block(unsigned long flags)
 165 {
 166         return (flags & VM_ACCOUNT)?
 167                 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
 168 }
 169
 170 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
 171 {
 172         if (!(flags & VM_ACCOUNT))
 173                 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
 174 }
 175
 176 static struct super_operations shmem_ops;
 177 static struct address_space_operations shmem_aops;
 178 static struct file_operations shmem_file_operations;
 179 static struct inode_operations shmem_inode_operations;
 180 static struct inode_operations shmem_dir_inode_operations;
 181 static struct vm_operations_struct shmem_vm_ops;
 182
 183 static struct backing_dev_info shmem_backing_dev_info  __read_mostly = {
 184         .ra_pages       = 0,    /* No readahead */
 185         .capabilities   = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
 186         .unplug_io_fn   = default_unplug_io_fn,
 187 };
 188
 189 static LIST_HEAD(shmem_swaplist);
 190 static DEFINE_SPINLOCK(shmem_swaplist_lock);
 191
 192 static void shmem_free_blocks(struct inode *inode, long pages)
 193 {
 194         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 195         if (sbinfo->max_blocks) {
 196                 spin_lock(&sbinfo->stat_lock);
 197                 sbinfo->free_blocks += pages;
 198                 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
 199                 spin_unlock(&sbinfo->stat_lock);
 200         }
 201 }
 202
 203 /*
 204  * shmem_recalc_inode - recalculate the size of an inode
 205  *
 206  * @inode: inode to recalc
 207  *
 208  * We have to calculate the free blocks since the mm can drop
 209  * undirtied hole pages behind our back.
 210  *
 211  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
 212  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
 213  *
 214  * It has to be called with the spinlock held.
 215  */
 216 static void shmem_recalc_inode(struct inode *inode)
 217 {
 218         struct shmem_inode_info *info = SHMEM_I(inode);
 219         long freed;
 220
 221         freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
 222         if (freed > 0) {
 223                 info->alloced -= freed;
 224                 shmem_unacct_blocks(info->flags, freed);
 225                 shmem_free_blocks(inode, freed);
 226         }
 227 }
 228
 229 /*
 230  * shmem_swp_entry - find the swap vector position in the info structure
 231  *
 232  * @info:  info structure for the inode
 233  * @index: index of the page to find
 234  * @page:  optional page to add to the structure. Has to be preset to
 235  *         all zeros
 236  *
 237  * If there is no space allocated yet it will return NULL when
 238  * page is NULL, else it will use the page for the needed block,
 239  * setting it to NULL on return to indicate that it has been used.
 240  *
 241  * The swap vector is organized the following way:
 242  *
 243  * There are SHMEM_NR_DIRECT entries directly stored in the
 244  * shmem_inode_info structure. So small files do not need an addional
 245  * allocation.
 246  *
 247  * For pages with index > SHMEM_NR_DIRECT there is the pointer
 248  * i_indirect which points to a page which holds in the first half
 249  * doubly indirect blocks, in the second half triple indirect blocks:
 250  *
 251  * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
 252  * following layout (for SHMEM_NR_DIRECT == 16):
 253  *
 254  * i_indirect -> dir --> 16-19
 255  *            |      +-> 20-23
 256  *            |
 257  *            +-->dir2 --> 24-27
 258  *            |        +-> 28-31
 259  *            |        +-> 32-35
 260  *            |        +-> 36-39
 261  *            |
 262  *            +-->dir3 --> 40-43
 263  *                     +-> 44-47
 264  *                     +-> 48-51
 265  *                     +-> 52-55
 266  */
 267 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
 268 {
 269         unsigned long offset;
 270         struct page **dir;
 271         struct page *subdir;
 272
 273         if (index < SHMEM_NR_DIRECT) {
 274                 shmem_swp_balance_unmap();
 275                 return info->i_direct+index;
 276         }
 277         if (!info->i_indirect) {
 278                 if (page) {
 279                         info->i_indirect = *page;
 280                         *page = NULL;
 281                 }
 282                 return NULL;                    /* need another page */
 283         }
 284
 285         index -= SHMEM_NR_DIRECT;
 286         offset = index % ENTRIES_PER_PAGE;
 287         index /= ENTRIES_PER_PAGE;
 288         dir = shmem_dir_map(info->i_indirect);
 289
 290         if (index >= ENTRIES_PER_PAGE/2) {
 291                 index -= ENTRIES_PER_PAGE/2;
 292                 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
 293                 index %= ENTRIES_PER_PAGE;
 294                 subdir = *dir;
 295                 if (!subdir) {
 296                         if (page) {
 297                                 *dir = *page;
 298                                 *page = NULL;
 299                         }
 300                         shmem_dir_unmap(dir);
 301                         return NULL;            /* need another page */
 302                 }
 303                 shmem_dir_unmap(dir);
 304                 dir = shmem_dir_map(subdir);
 305         }
 306
 307         dir += index;
 308         subdir = *dir;
 309         if (!subdir) {
 310                 if (!page || !(subdir = *page)) {
 311                         shmem_dir_unmap(dir);
 312                         return NULL;            /* need a page */
 313                 }
 314                 *dir = subdir;
 315                 *page = NULL;
 316         }
 317         shmem_dir_unmap(dir);
 318         return shmem_swp_map(subdir) + offset;
 319 }
 320
 321 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
 322 {
 323         long incdec = value? 1: -1;
 324
 325         entry->val = value;
 326         info->swapped += incdec;
 327         if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT)
 328                 kmap_atomic_to_page(entry)->nr_swapped += incdec;
 329 }
 330
 331 /*
 332  * shmem_swp_alloc - get the position of the swap entry for the page.
 333  *                   If it does not exist allocate the entry.
 334  *
 335  * @info:       info structure for the inode
 336  * @index:      index of the page to find
 337  * @sgp:        check and recheck i_size? skip allocation?
 338  */
 339 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
 340 {
 341         struct inode *inode = &info->vfs_inode;
 342         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 343         struct page *page = NULL;
 344         swp_entry_t *entry;
 345
 346         if (sgp != SGP_WRITE &&
 347             ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
 348                 return ERR_PTR(-EINVAL);
 349
 350         while (!(entry = shmem_swp_entry(info, index, &page))) {
 351                 if (sgp == SGP_READ)
 352                         return shmem_swp_map(ZERO_PAGE(0));
 353                 /*
 354                  * Test free_blocks against 1 not 0, since we have 1 data
 355                  * page (and perhaps indirect index pages) yet to allocate:
 356                  * a waste to allocate index if we cannot allocate data.
 357                  */
 358                 if (sbinfo->max_blocks) {
 359                         spin_lock(&sbinfo->stat_lock);
 360                         if (sbinfo->free_blocks <= 1) {
 361                                 spin_unlock(&sbinfo->stat_lock);
 362                                 return ERR_PTR(-ENOSPC);
 363                         }
 364                         sbinfo->free_blocks--;
 365                         inode->i_blocks += BLOCKS_PER_PAGE;
 366                         spin_unlock(&sbinfo->stat_lock);
 367                 }
 368
 369                 spin_unlock(&info->lock);
 370                 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping) | __GFP_ZERO);
 371                 if (page) {
 372                         page->nr_swapped = 0;
 373                 }
 374                 spin_lock(&info->lock);
 375
 376                 if (!page) {
 377                         shmem_free_blocks(inode, 1);
 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_blocks(inode, 1);
 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 int shmem_map_and_free_swp(struct page *subdir,
 420                 int offset, int limit, struct page ***dir)
 421 {
 422         swp_entry_t *ptr;
 423         int freed = 0;
 424
 425         ptr = shmem_swp_map(subdir);
 426         for (; offset < limit; offset += LATENCY_LIMIT) {
 427                 int size = limit - offset;
 428                 if (size > LATENCY_LIMIT)
 429                         size = LATENCY_LIMIT;
 430                 freed += shmem_free_swp(ptr+offset, ptr+offset+size);
 431                 if (need_resched()) {
 432                         shmem_swp_unmap(ptr);
 433                         if (*dir) {
 434                                 shmem_dir_unmap(*dir);
 435                                 *dir = NULL;
 436                         }
 437                         cond_resched();
 438                         ptr = shmem_swp_map(subdir);
 439                 }
 440         }
 441         shmem_swp_unmap(ptr);
 442         return freed;
 443 }
 444
 445 static void shmem_free_pages(struct list_head *next)
 446 {
 447         struct page *page;
 448         int freed = 0;
 449
 450         do {
 451                 page = container_of(next, struct page, lru);
 452                 next = next->next;
 453                 shmem_dir_free(page);
 454                 freed++;
 455                 if (freed >= LATENCY_LIMIT) {
 456                         cond_resched();
 457                         freed = 0;
 458                 }
 459         } while (next);
 460 }
 461
 462 static void shmem_truncate(struct inode *inode)
 463 {
 464         struct shmem_inode_info *info = SHMEM_I(inode);
 465         unsigned long idx;
 466         unsigned long size;
 467         unsigned long limit;
 468         unsigned long stage;
 469         unsigned long diroff;
 470         struct page **dir;
 471         struct page *topdir;
 472         struct page *middir;
 473         struct page *subdir;
 474         swp_entry_t *ptr;
 475         LIST_HEAD(pages_to_free);
 476         long nr_pages_to_free = 0;
 477         long nr_swaps_freed = 0;
 478         int offset;
 479         int freed;
 480
 481         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 482         idx = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 483         if (idx >= info->next_index)
 484                 return;
 485
 486         spin_lock(&info->lock);
 487         info->flags |= SHMEM_TRUNCATE;
 488         limit = info->next_index;
 489         info->next_index = idx;
 490         topdir = info->i_indirect;
 491         if (topdir && idx <= SHMEM_NR_DIRECT) {
 492                 info->i_indirect = NULL;
 493                 nr_pages_to_free++;
 494                 list_add(&topdir->lru, &pages_to_free);
 495         }
 496         spin_unlock(&info->lock);
 497
 498         if (info->swapped && idx < SHMEM_NR_DIRECT) {
 499                 ptr = info->i_direct;
 500                 size = limit;
 501                 if (size > SHMEM_NR_DIRECT)
 502                         size = SHMEM_NR_DIRECT;
 503                 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size);
 504         }
 505         if (!topdir)
 506                 goto done2;
 507
 508         BUG_ON(limit <= SHMEM_NR_DIRECT);
 509         limit -= SHMEM_NR_DIRECT;
 510         idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
 511         offset = idx % ENTRIES_PER_PAGE;
 512         idx -= offset;
 513
 514         dir = shmem_dir_map(topdir);
 515         stage = ENTRIES_PER_PAGEPAGE/2;
 516         if (idx < ENTRIES_PER_PAGEPAGE/2) {
 517                 middir = topdir;
 518                 diroff = idx/ENTRIES_PER_PAGE;
 519         } else {
 520                 dir += ENTRIES_PER_PAGE/2;
 521                 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
 522                 while (stage <= idx)
 523                         stage += ENTRIES_PER_PAGEPAGE;
 524                 middir = *dir;
 525                 if (*dir) {
 526                         diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
 527                                 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
 528                         if (!diroff && !offset) {
 529                                 *dir = NULL;
 530                                 nr_pages_to_free++;
 531                                 list_add(&middir->lru, &pages_to_free);
 532                         }
 533                         shmem_dir_unmap(dir);
 534                         dir = shmem_dir_map(middir);
 535                 } else {
 536                         diroff = 0;
 537                         offset = 0;
 538                         idx = stage;
 539                 }
 540         }
 541
 542         for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
 543                 if (unlikely(idx == stage)) {
 544                         shmem_dir_unmap(dir);
 545                         dir = shmem_dir_map(topdir) +
 546                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
 547                         while (!*dir) {
 548                                 dir++;
 549                                 idx += ENTRIES_PER_PAGEPAGE;
 550                                 if (idx >= limit)
 551                                         goto done1;
 552                         }
 553                         stage = idx + ENTRIES_PER_PAGEPAGE;
 554                         middir = *dir;
 555                         *dir = NULL;
 556                         nr_pages_to_free++;
 557                         list_add(&middir->lru, &pages_to_free);
 558                         shmem_dir_unmap(dir);
 559                         cond_resched();
 560                         dir = shmem_dir_map(middir);
 561                         diroff = 0;
 562                 }
 563                 subdir = dir[diroff];
 564                 if (subdir && subdir->nr_swapped) {
 565                         size = limit - idx;
 566                         if (size > ENTRIES_PER_PAGE)
 567                                 size = ENTRIES_PER_PAGE;
 568                         freed = shmem_map_and_free_swp(subdir,
 569                                                 offset, size, &dir);
 570                         if (!dir)
 571                                 dir = shmem_dir_map(middir);
 572                         nr_swaps_freed += freed;
 573                         if (offset)
 574                                 spin_lock(&info->lock);
 575                         subdir->nr_swapped -= freed;
 576                         if (offset)
 577                                 spin_unlock(&info->lock);
 578                         BUG_ON(subdir->nr_swapped > offset);
 579                 }
 580                 if (offset)
 581                         offset = 0;
 582                 else if (subdir) {
 583                         dir[diroff] = NULL;
 584                         nr_pages_to_free++;
 585                         list_add(&subdir->lru, &pages_to_free);
 586                 }
 587         }
 588 done1:
 589         shmem_dir_unmap(dir);
 590 done2:
 591         if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
 592                 /*
 593                  * Call truncate_inode_pages again: racing shmem_unuse_inode
 594                  * may have swizzled a page in from swap since vmtruncate or
 595                  * generic_delete_inode did it, before we lowered next_index.
 596                  * Also, though shmem_getpage checks i_size before adding to
 597                  * cache, no recheck after: so fix the narrow window there too.
 598                  */
 599                 truncate_inode_pages(inode->i_mapping, inode->i_size);
 600         }
 601
 602         spin_lock(&info->lock);
 603         info->flags &= ~SHMEM_TRUNCATE;
 604         info->swapped -= nr_swaps_freed;
 605         if (nr_pages_to_free)
 606                 shmem_free_blocks(inode, nr_pages_to_free);
 607         shmem_recalc_inode(inode);
 608         spin_unlock(&info->lock);
 609
 610         /*
 611          * Empty swap vector directory pages to be freed?
 612          */
 613         if (!list_empty(&pages_to_free)) {
 614                 pages_to_free.prev->next = NULL;
 615                 shmem_free_pages(pages_to_free.next);
 616         }
 617 }
 618
 619 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
 620 {
 621         struct inode *inode = dentry->d_inode;
 622         struct page *page = NULL;
 623         int error;
 624
 625         if (attr->ia_valid & ATTR_SIZE) {
 626                 if (attr->ia_size < inode->i_size) {
 627                         /*
 628                          * If truncating down to a partial page, then
 629                          * if that page is already allocated, hold it
 630                          * in memory until the truncation is over, so
 631                          * truncate_partial_page cannnot miss it were
 632                          * it assigned to swap.
 633                          */
 634                         if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
 635                                 (void) shmem_getpage(inode,
 636                                         attr->ia_size>>PAGE_CACHE_SHIFT,
 637                                                 &page, SGP_READ, NULL);
 638                         }
 639                         /*
 640                          * Reset SHMEM_PAGEIN flag so that shmem_truncate can
 641                          * detect if any pages might have been added to cache
 642                          * after truncate_inode_pages.  But we needn't bother
 643                          * if it's being fully truncated to zero-length: the
 644                          * nrpages check is efficient enough in that case.
 645                          */
 646                         if (attr->ia_size) {
 647                                 struct shmem_inode_info *info = SHMEM_I(inode);
 648                                 spin_lock(&info->lock);
 649                                 info->flags &= ~SHMEM_PAGEIN;
 650                                 spin_unlock(&info->lock);
 651                         }
 652                 }
 653         }
 654
 655         error = inode_change_ok(inode, attr);
 656         if (!error)
 657                 error = inode_setattr(inode, attr);
 658         if (page)
 659                 page_cache_release(page);
 660         return error;
 661 }
 662
 663 static void shmem_delete_inode(struct inode *inode)
 664 {
 665         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 666         struct shmem_inode_info *info = SHMEM_I(inode);
 667
 668         if (inode->i_op->truncate == shmem_truncate) {
 669                 shmem_unacct_size(info->flags, inode->i_size);
 670                 inode->i_size = 0;
 671                 shmem_truncate(inode);
 672                 if (!list_empty(&info->swaplist)) {
 673                         spin_lock(&shmem_swaplist_lock);
 674                         list_del_init(&info->swaplist);
 675                         spin_unlock(&shmem_swaplist_lock);
 676                 }
 677         }
 678         BUG_ON(inode->i_blocks);
 679         if (sbinfo->max_inodes) {
 680                 spin_lock(&sbinfo->stat_lock);
 681                 sbinfo->free_inodes++;
 682                 spin_unlock(&sbinfo->stat_lock);
 683         }
 684         clear_inode(inode);
 685 }
 686
 687 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
 688 {
 689         swp_entry_t *ptr;
 690
 691         for (ptr = dir; ptr < edir; ptr++) {
 692                 if (ptr->val == entry.val)
 693                         return ptr - dir;
 694         }
 695         return -1;
 696 }
 697
 698 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
 699 {
 700         struct inode *inode;
 701         unsigned long idx;
 702         unsigned long size;
 703         unsigned long limit;
 704         unsigned long stage;
 705         struct page **dir;
 706         struct page *subdir;
 707         swp_entry_t *ptr;
 708         int offset;
 709
 710         idx = 0;
 711         ptr = info->i_direct;
 712         spin_lock(&info->lock);
 713         limit = info->next_index;
 714         size = limit;
 715         if (size > SHMEM_NR_DIRECT)
 716                 size = SHMEM_NR_DIRECT;
 717         offset = shmem_find_swp(entry, ptr, ptr+size);
 718         if (offset >= 0) {
 719                 shmem_swp_balance_unmap();
 720                 goto found;
 721         }
 722         if (!info->i_indirect)
 723                 goto lost2;
 724
 725         dir = shmem_dir_map(info->i_indirect);
 726         stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
 727
 728         for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
 729                 if (unlikely(idx == stage)) {
 730                         shmem_dir_unmap(dir-1);
 731                         dir = shmem_dir_map(info->i_indirect) +
 732                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
 733                         while (!*dir) {
 734                                 dir++;
 735                                 idx += ENTRIES_PER_PAGEPAGE;
 736                                 if (idx >= limit)
 737                                         goto lost1;
 738                         }
 739                         stage = idx + ENTRIES_PER_PAGEPAGE;
 740                         subdir = *dir;
 741                         shmem_dir_unmap(dir);
 742                         dir = shmem_dir_map(subdir);
 743                 }
 744                 subdir = *dir;
 745                 if (subdir && subdir->nr_swapped) {
 746                         ptr = shmem_swp_map(subdir);
 747                         size = limit - idx;
 748                         if (size > ENTRIES_PER_PAGE)
 749                                 size = ENTRIES_PER_PAGE;
 750                         offset = shmem_find_swp(entry, ptr, ptr+size);
 751                         if (offset >= 0) {
 752                                 shmem_dir_unmap(dir);
 753                                 goto found;
 754                         }
 755                         shmem_swp_unmap(ptr);
 756                 }
 757         }
 758 lost1:
 759         shmem_dir_unmap(dir-1);
 760 lost2:
 761         spin_unlock(&info->lock);
 762         return 0;
 763 found:
 764         idx += offset;
 765         inode = &info->vfs_inode;
 766         if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
 767                 info->flags |= SHMEM_PAGEIN;
 768                 shmem_swp_set(info, ptr + offset, 0);
 769         }
 770         shmem_swp_unmap(ptr);
 771         spin_unlock(&info->lock);
 772         /*
 773          * Decrement swap count even when the entry is left behind:
 774          * try_to_unuse will skip over mms, then reincrement count.
 775          */
 776         swap_free(entry);
 777         return 1;
 778 }
 779
 780 /*
 781  * shmem_unuse() search for an eventually swapped out shmem page.
 782  */
 783 int shmem_unuse(swp_entry_t entry, struct page *page)
 784 {
 785         struct list_head *p, *next;
 786         struct shmem_inode_info *info;
 787         int found = 0;
 788
 789         spin_lock(&shmem_swaplist_lock);
 790         list_for_each_safe(p, next, &shmem_swaplist) {
 791                 info = list_entry(p, struct shmem_inode_info, swaplist);
 792                 if (!info->swapped)
 793                         list_del_init(&info->swaplist);
 794                 else if (shmem_unuse_inode(info, entry, page)) {
 795                         /* move head to start search for next from here */
 796                         list_move_tail(&shmem_swaplist, &info->swaplist);
 797                         found = 1;
 798                         break;
 799                 }
 800         }
 801         spin_unlock(&shmem_swaplist_lock);
 802         return found;
 803 }
 804
 805 /*
 806  * Move the page from the page cache to the swap cache.
 807  */
 808 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
 809 {
 810         struct shmem_inode_info *info;
 811         swp_entry_t *entry, swap;
 812         struct address_space *mapping;
 813         unsigned long index;
 814         struct inode *inode;
 815
 816         BUG_ON(!PageLocked(page));
 817         BUG_ON(page_mapped(page));
 818
 819         mapping = page->mapping;
 820         index = page->index;
 821         inode = mapping->host;
 822         info = SHMEM_I(inode);
 823         if (info->flags & VM_LOCKED)
 824                 goto redirty;
 825         swap = get_swap_page();
 826         if (!swap.val)
 827                 goto redirty;
 828
 829         spin_lock(&info->lock);
 830         shmem_recalc_inode(inode);
 831         if (index >= info->next_index) {
 832                 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
 833                 goto unlock;
 834         }
 835         entry = shmem_swp_entry(info, index, NULL);
 836         BUG_ON(!entry);
 837         BUG_ON(entry->val);
 838
 839         if (move_to_swap_cache(page, swap) == 0) {
 840                 shmem_swp_set(info, entry, swap.val);
 841                 shmem_swp_unmap(entry);
 842                 spin_unlock(&info->lock);
 843                 if (list_empty(&info->swaplist)) {
 844                         spin_lock(&shmem_swaplist_lock);
 845                         /* move instead of add in case we're racing */
 846                         list_move_tail(&info->swaplist, &shmem_swaplist);
 847                         spin_unlock(&shmem_swaplist_lock);
 848                 }
 849                 unlock_page(page);
 850                 return 0;
 851         }
 852
 853         shmem_swp_unmap(entry);
 854 unlock:
 855         spin_unlock(&info->lock);
 856         swap_free(swap);
 857 redirty:
 858         set_page_dirty(page);
 859         return WRITEPAGE_ACTIVATE;      /* Return with the page locked */
 860 }
 861
 862 #ifdef CONFIG_NUMA
 863 static struct page *shmem_swapin_async(struct shared_policy *p,
 864                                        swp_entry_t entry, unsigned long idx)
 865 {
 866         struct page *page;
 867         struct vm_area_struct pvma;
 868
 869         /* Create a pseudo vma that just contains the policy */
 870         memset(&pvma, 0, sizeof(struct vm_area_struct));
 871         pvma.vm_end = PAGE_SIZE;
 872         pvma.vm_pgoff = idx;
 873         pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
 874         page = read_swap_cache_async(entry, &pvma, 0);
 875         mpol_free(pvma.vm_policy);
 876         return page;
 877 }
 878
 879 struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry,
 880                           unsigned long idx)
 881 {
 882         struct shared_policy *p = &info->policy;
 883         int i, num;
 884         struct page *page;
 885         unsigned long offset;
 886
 887         num = valid_swaphandles(entry, &offset);
 888         for (i = 0; i < num; offset++, i++) {
 889                 page = shmem_swapin_async(p,
 890                                 swp_entry(swp_type(entry), offset), idx);
 891                 if (!page)
 892                         break;
 893                 page_cache_release(page);
 894         }
 895         lru_add_drain();        /* Push any new pages onto the LRU now */
 896         return shmem_swapin_async(p, entry, idx);
 897 }
 898
 899 static struct page *
 900 shmem_alloc_page(unsigned long gfp, struct shmem_inode_info *info,
 901                  unsigned long idx)
 902 {
 903         struct vm_area_struct pvma;
 904         struct page *page;
 905
 906         memset(&pvma, 0, sizeof(struct vm_area_struct));
 907         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
 908         pvma.vm_pgoff = idx;
 909         pvma.vm_end = PAGE_SIZE;
 910         page = alloc_page_vma(gfp | __GFP_ZERO, &pvma, 0);
 911         mpol_free(pvma.vm_policy);
 912         return page;
 913 }
 914 #else
 915 static inline struct page *
 916 shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
 917 {
 918         swapin_readahead(entry, 0, NULL);
 919         return read_swap_cache_async(entry, NULL, 0);
 920 }
 921
 922 static inline struct page *
 923 shmem_alloc_page(unsigned int __nocast gfp,struct shmem_inode_info *info,
 924                                  unsigned long idx)
 925 {
 926         return alloc_page(gfp | __GFP_ZERO);
 927 }
 928 #endif
 929
 930 /*
 931  * shmem_getpage - either get the page from swap or allocate a new one
 932  *
 933  * If we allocate a new one we do not mark it dirty. That's up to the
 934  * vm. If we swap it in we mark it dirty since we also free the swap
 935  * entry since a page cannot live in both the swap and page cache
 936  */
 937 static int shmem_getpage(struct inode *inode, unsigned long idx,
 938                         struct page **pagep, enum sgp_type sgp, int *type)
 939 {
 940         struct address_space *mapping = inode->i_mapping;
 941         struct shmem_inode_info *info = SHMEM_I(inode);
 942         struct shmem_sb_info *sbinfo;
 943         struct page *filepage = *pagep;
 944         struct page *swappage;
 945         swp_entry_t *entry;
 946         swp_entry_t swap;
 947         int error;
 948
 949         if (idx >= SHMEM_MAX_INDEX)
 950                 return -EFBIG;
 951         /*
 952          * Normally, filepage is NULL on entry, and either found
 953          * uptodate immediately, or allocated and zeroed, or read
 954          * in under swappage, which is then assigned to filepage.
 955          * But shmem_prepare_write passes in a locked filepage,
 956          * which may be found not uptodate by other callers too,
 957          * and may need to be copied from the swappage read in.
 958          */
 959 repeat:
 960         if (!filepage)
 961                 filepage = find_lock_page(mapping, idx);
 962         if (filepage && PageUptodate(filepage))
 963                 goto done;
 964         error = 0;
 965         if (sgp == SGP_QUICK)
 966                 goto failed;
 967
 968         spin_lock(&info->lock);
 969         shmem_recalc_inode(inode);
 970         entry = shmem_swp_alloc(info, idx, sgp);
 971         if (IS_ERR(entry)) {
 972                 spin_unlock(&info->lock);
 973                 error = PTR_ERR(entry);
 974                 goto failed;
 975         }
 976         swap = *entry;
 977
 978         if (swap.val) {
 979                 /* Look it up and read it in.. */
 980                 swappage = lookup_swap_cache(swap);
 981                 if (!swappage) {
 982                         shmem_swp_unmap(entry);
 983                         spin_unlock(&info->lock);
 984                         /* here we actually do the io */
 985                         if (type && *type == VM_FAULT_MINOR) {
 986                                 inc_page_state(pgmajfault);
 987                                 *type = VM_FAULT_MAJOR;
 988                         }
 989                         swappage = shmem_swapin(info, swap, idx);
 990                         if (!swappage) {
 991                                 spin_lock(&info->lock);
 992                                 entry = shmem_swp_alloc(info, idx, sgp);
 993                                 if (IS_ERR(entry))
 994                                         error = PTR_ERR(entry);
 995                                 else {
 996                                         if (entry->val == swap.val)
 997                                                 error = -ENOMEM;
 998                                         shmem_swp_unmap(entry);
 999                                 }
1000                                 spin_unlock(&info->lock);
1001                                 if (error)
1002                                         goto failed;
1003                                 goto repeat;
1004                         }
1005                         wait_on_page_locked(swappage);
1006                         page_cache_release(swappage);
1007                         goto repeat;
1008                 }
1009
1010                 /* We have to do this with page locked to prevent races */
1011                 if (TestSetPageLocked(swappage)) {
1012                         shmem_swp_unmap(entry);
1013                         spin_unlock(&info->lock);
1014                         wait_on_page_locked(swappage);
1015                         page_cache_release(swappage);
1016                         goto repeat;
1017                 }
1018                 if (PageWriteback(swappage)) {
1019                         shmem_swp_unmap(entry);
1020                         spin_unlock(&info->lock);
1021                         wait_on_page_writeback(swappage);
1022                         unlock_page(swappage);
1023                         page_cache_release(swappage);
1024                         goto repeat;
1025                 }
1026                 if (!PageUptodate(swappage)) {
1027                         shmem_swp_unmap(entry);
1028                         spin_unlock(&info->lock);
1029                         unlock_page(swappage);
1030                         page_cache_release(swappage);
1031                         error = -EIO;
1032                         goto failed;
1033                 }
1034
1035                 if (filepage) {
1036                         shmem_swp_set(info, entry, 0);
1037                         shmem_swp_unmap(entry);
1038                         delete_from_swap_cache(swappage);
1039                         spin_unlock(&info->lock);
1040                         copy_highpage(filepage, swappage);
1041                         unlock_page(swappage);
1042                         page_cache_release(swappage);
1043                         flush_dcache_page(filepage);
1044                         SetPageUptodate(filepage);
1045                         set_page_dirty(filepage);
1046                         swap_free(swap);
1047                 } else if (!(error = move_from_swap_cache(
1048                                 swappage, idx, mapping))) {
1049                         info->flags |= SHMEM_PAGEIN;
1050                         shmem_swp_set(info, entry, 0);
1051                         shmem_swp_unmap(entry);
1052                         spin_unlock(&info->lock);
1053                         filepage = swappage;
1054                         swap_free(swap);
1055                 } else {
1056                         shmem_swp_unmap(entry);
1057                         spin_unlock(&info->lock);
1058                         unlock_page(swappage);
1059                         page_cache_release(swappage);
1060                         if (error == -ENOMEM) {
1061                                 /* let kswapd refresh zone for GFP_ATOMICs */
1062                                 blk_congestion_wait(WRITE, HZ/50);
1063                         }
1064                         goto repeat;
1065                 }
1066         } else if (sgp == SGP_READ && !filepage) {
1067                 shmem_swp_unmap(entry);
1068                 filepage = find_get_page(mapping, idx);
1069                 if (filepage &&
1070                     (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1071                         spin_unlock(&info->lock);
1072                         wait_on_page_locked(filepage);
1073                         page_cache_release(filepage);
1074                         filepage = NULL;
1075                         goto repeat;
1076                 }
1077                 spin_unlock(&info->lock);
1078         } else {
1079                 shmem_swp_unmap(entry);
1080                 sbinfo = SHMEM_SB(inode->i_sb);
1081                 if (sbinfo->max_blocks) {
1082                         spin_lock(&sbinfo->stat_lock);
1083                         if (sbinfo->free_blocks == 0 ||
1084                             shmem_acct_block(info->flags)) {
1085                                 spin_unlock(&sbinfo->stat_lock);
1086                                 spin_unlock(&info->lock);
1087                                 error = -ENOSPC;
1088                                 goto failed;
1089                         }
1090                         sbinfo->free_blocks--;
1091                         inode->i_blocks += BLOCKS_PER_PAGE;
1092                         spin_unlock(&sbinfo->stat_lock);
1093                 } else if (shmem_acct_block(info->flags)) {
1094                         spin_unlock(&info->lock);
1095                         error = -ENOSPC;
1096                         goto failed;
1097                 }
1098
1099                 if (!filepage) {
1100                         spin_unlock(&info->lock);
1101                         filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
1102                                                     info,
1103                                                     idx);
1104                         if (!filepage) {
1105                                 shmem_unacct_blocks(info->flags, 1);
1106                                 shmem_free_blocks(inode, 1);
1107                                 error = -ENOMEM;
1108                                 goto failed;
1109                         }
1110
1111                         spin_lock(&info->lock);
1112                         entry = shmem_swp_alloc(info, idx, sgp);
1113                         if (IS_ERR(entry))
1114                                 error = PTR_ERR(entry);
1115                         else {
1116                                 swap = *entry;
1117                                 shmem_swp_unmap(entry);
1118                         }
1119                         if (error || swap.val || 0 != add_to_page_cache_lru(
1120                                         filepage, mapping, idx, GFP_ATOMIC)) {
1121                                 spin_unlock(&info->lock);
1122                                 page_cache_release(filepage);
1123                                 shmem_unacct_blocks(info->flags, 1);
1124                                 shmem_free_blocks(inode, 1);
1125                                 filepage = NULL;
1126                                 if (error)
1127                                         goto failed;
1128                                 goto repeat;
1129                         }
1130                         info->flags |= SHMEM_PAGEIN;
1131                 }
1132
1133                 info->alloced++;
1134                 spin_unlock(&info->lock);
1135                 flush_dcache_page(filepage);
1136                 SetPageUptodate(filepage);
1137         }
1138 done:
1139         if (*pagep != filepage) {
1140                 unlock_page(filepage);
1141                 *pagep = filepage;
1142         }
1143         return 0;
1144
1145 failed:
1146         if (*pagep != filepage) {
1147                 unlock_page(filepage);
1148                 page_cache_release(filepage);
1149         }
1150         return error;
1151 }
1152
1153 struct page *shmem_nopage(struct vm_area_struct *vma, unsigned long address, int *type)
1154 {
1155         struct inode *inode = vma->vm_file->f_dentry->d_inode;
1156         struct page *page = NULL;
1157         unsigned long idx;
1158         int error;
1159
1160         idx = (address - vma->vm_start) >> PAGE_SHIFT;
1161         idx += vma->vm_pgoff;
1162         idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT;
1163         if (((loff_t) idx << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1164                 return NOPAGE_SIGBUS;
1165
1166         error = shmem_getpage(inode, idx, &page, SGP_CACHE, type);
1167         if (error)
1168                 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS;
1169
1170         mark_page_accessed(page);
1171         return page;
1172 }
1173
1174 static int shmem_populate(struct vm_area_struct *vma,
1175         unsigned long addr, unsigned long len,
1176         pgprot_t prot, unsigned long pgoff, int nonblock)
1177 {
1178         struct inode *inode = vma->vm_file->f_dentry->d_inode;
1179         struct mm_struct *mm = vma->vm_mm;
1180         enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE;
1181         unsigned long size;
1182
1183         size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1184         if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size)
1185                 return -EINVAL;
1186
1187         while ((long) len > 0) {
1188                 struct page *page = NULL;
1189                 int err;
1190                 /*
1191                  * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE
1192                  */
1193                 err = shmem_getpage(inode, pgoff, &page, sgp, NULL);
1194                 if (err)
1195                         return err;
1196                 /* Page may still be null, but only if nonblock was set. */
1197                 if (page) {
1198                         mark_page_accessed(page);
1199                         err = install_page(mm, vma, addr, page, prot);
1200                         if (err) {
1201                                 page_cache_release(page);
1202                                 return err;
1203                         }
1204                 } else {
1205                         /* No page was found just because we can't read it in
1206                          * now (being here implies nonblock != 0), but the page
1207                          * may exist, so set the PTE to fault it in later. */
1208                         err = install_file_pte(mm, vma, addr, pgoff, prot);
1209                         if (err)
1210                                 return err;
1211                 }
1212
1213                 len -= PAGE_SIZE;
1214                 addr += PAGE_SIZE;
1215                 pgoff++;
1216         }
1217         return 0;
1218 }
1219
1220 #ifdef CONFIG_NUMA
1221 int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1222 {
1223         struct inode *i = vma->vm_file->f_dentry->d_inode;
1224         return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1225 }
1226
1227 struct mempolicy *
1228 shmem_get_policy(struct vm_area_struct *vma, unsigned long addr)
1229 {
1230         struct inode *i = vma->vm_file->f_dentry->d_inode;
1231         unsigned long idx;
1232
1233         idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1234         return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1235 }
1236 #endif
1237
1238 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1239 {
1240         struct inode *inode = file->f_dentry->d_inode;
1241         struct shmem_inode_info *info = SHMEM_I(inode);
1242         int retval = -ENOMEM;
1243
1244         spin_lock(&info->lock);
1245         if (lock && !(info->flags & VM_LOCKED)) {
1246                 if (!user_shm_lock(inode->i_size, user))
1247                         goto out_nomem;
1248                 info->flags |= VM_LOCKED;
1249         }
1250         if (!lock && (info->flags & VM_LOCKED) && user) {
1251                 user_shm_unlock(inode->i_size, user);
1252                 info->flags &= ~VM_LOCKED;
1253         }
1254         retval = 0;
1255 out_nomem:
1256         spin_unlock(&info->lock);
1257         return retval;
1258 }
1259
1260 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1261 {
1262         file_accessed(file);
1263         vma->vm_ops = &shmem_vm_ops;
1264         return 0;
1265 }
1266
1267 static struct inode *
1268 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1269 {
1270         struct inode *inode;
1271         struct shmem_inode_info *info;
1272         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1273
1274         if (sbinfo->max_inodes) {
1275                 spin_lock(&sbinfo->stat_lock);
1276                 if (!sbinfo->free_inodes) {
1277                         spin_unlock(&sbinfo->stat_lock);
1278                         return NULL;
1279                 }
1280                 sbinfo->free_inodes--;
1281                 spin_unlock(&sbinfo->stat_lock);
1282         }
1283
1284         inode = new_inode(sb);
1285         if (inode) {
1286                 inode->i_mode = mode;
1287                 inode->i_uid = current->fsuid;
1288                 inode->i_gid = current->fsgid;
1289                 inode->i_blksize = PAGE_CACHE_SIZE;
1290                 inode->i_blocks = 0;
1291                 inode->i_mapping->a_ops = &shmem_aops;
1292                 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1293                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1294                 info = SHMEM_I(inode);
1295                 memset(info, 0, (char *)inode - (char *)info);
1296                 spin_lock_init(&info->lock);
1297                 INIT_LIST_HEAD(&info->swaplist);
1298
1299                 switch (mode & S_IFMT) {
1300                 default:
1301                         init_special_inode(inode, mode, dev);
1302                         break;
1303                 case S_IFREG:
1304                         inode->i_op = &shmem_inode_operations;
1305                         inode->i_fop = &shmem_file_operations;
1306                         mpol_shared_policy_init(&info->policy);
1307                         break;
1308                 case S_IFDIR:
1309                         inode->i_nlink++;
1310                         /* Some things misbehave if size == 0 on a directory */
1311                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
1312                         inode->i_op = &shmem_dir_inode_operations;
1313                         inode->i_fop = &simple_dir_operations;
1314                         break;
1315                 case S_IFLNK:
1316                         /*
1317                          * Must not load anything in the rbtree,
1318                          * mpol_free_shared_policy will not be called.
1319                          */
1320                         mpol_shared_policy_init(&info->policy);
1321                         break;
1322                 }
1323         } else if (sbinfo->max_inodes) {
1324                 spin_lock(&sbinfo->stat_lock);
1325                 sbinfo->free_inodes++;
1326                 spin_unlock(&sbinfo->stat_lock);
1327         }
1328         return inode;
1329 }
1330
1331 #ifdef CONFIG_TMPFS
1332 static struct inode_operations shmem_symlink_inode_operations;
1333 static struct inode_operations shmem_symlink_inline_operations;
1334
1335 /*
1336  * Normally tmpfs makes no use of shmem_prepare_write, but it
1337  * lets a tmpfs file be used read-write below the loop driver.
1338  */
1339 static int
1340 shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1341 {
1342         struct inode *inode = page->mapping->host;
1343         return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL);
1344 }
1345
1346 static ssize_t
1347 shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1348 {
1349         struct inode    *inode = file->f_dentry->d_inode;
1350         loff_t          pos;
1351         unsigned long   written;
1352         ssize_t         err;
1353
1354         if ((ssize_t) count < 0)
1355                 return -EINVAL;
1356
1357         if (!access_ok(VERIFY_READ, buf, count))
1358                 return -EFAULT;
1359
1360         down(&inode->i_sem);
1361
1362         pos = *ppos;
1363         written = 0;
1364
1365         err = generic_write_checks(file, &pos, &count, 0);
1366         if (err || !count)
1367                 goto out;
1368
1369         err = remove_suid(file->f_dentry);
1370         if (err)
1371                 goto out;
1372
1373         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1374
1375         do {
1376                 struct page *page = NULL;
1377                 unsigned long bytes, index, offset;
1378                 char *kaddr;
1379                 int left;
1380
1381                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1382                 index = pos >> PAGE_CACHE_SHIFT;
1383                 bytes = PAGE_CACHE_SIZE - offset;
1384                 if (bytes > count)
1385                         bytes = count;
1386
1387                 /*
1388                  * We don't hold page lock across copy from user -
1389                  * what would it guard against? - so no deadlock here.
1390                  * But it still may be a good idea to prefault below.
1391                  */
1392
1393                 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1394                 if (err)
1395                         break;
1396
1397                 left = bytes;
1398                 if (PageHighMem(page)) {
1399                         volatile unsigned char dummy;
1400                         __get_user(dummy, buf);
1401                         __get_user(dummy, buf + bytes - 1);
1402
1403                         kaddr = kmap_atomic(page, KM_USER0);
1404                         left = __copy_from_user_inatomic(kaddr + offset,
1405                                                         buf, bytes);
1406                         kunmap_atomic(kaddr, KM_USER0);
1407                 }
1408                 if (left) {
1409                         kaddr = kmap(page);
1410                         left = __copy_from_user(kaddr + offset, buf, bytes);
1411                         kunmap(page);
1412                 }
1413
1414                 written += bytes;
1415                 count -= bytes;
1416                 pos += bytes;
1417                 buf += bytes;
1418                 if (pos > inode->i_size)
1419                         i_size_write(inode, pos);
1420
1421                 flush_dcache_page(page);
1422                 set_page_dirty(page);
1423                 mark_page_accessed(page);
1424                 page_cache_release(page);
1425
1426                 if (left) {
1427                         pos -= left;
1428                         written -= left;
1429                         err = -EFAULT;
1430                         break;
1431                 }
1432
1433                 /*
1434                  * Our dirty pages are not counted in nr_dirty,
1435                  * and we do not attempt to balance dirty pages.
1436                  */
1437
1438                 cond_resched();
1439         } while (count);
1440
1441         *ppos = pos;
1442         if (written)
1443                 err = written;
1444 out:
1445         up(&inode->i_sem);
1446         return err;
1447 }
1448
1449 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1450 {
1451         struct inode *inode = filp->f_dentry->d_inode;
1452         struct address_space *mapping = inode->i_mapping;
1453         unsigned long index, offset;
1454
1455         index = *ppos >> PAGE_CACHE_SHIFT;
1456         offset = *ppos & ~PAGE_CACHE_MASK;
1457
1458         for (;;) {
1459                 struct page *page = NULL;
1460                 unsigned long end_index, nr, ret;
1461                 loff_t i_size = i_size_read(inode);
1462
1463                 end_index = i_size >> PAGE_CACHE_SHIFT;
1464                 if (index > end_index)
1465                         break;
1466                 if (index == end_index) {
1467                         nr = i_size & ~PAGE_CACHE_MASK;
1468                         if (nr <= offset)
1469                                 break;
1470                 }
1471
1472                 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1473                 if (desc->error) {
1474                         if (desc->error == -EINVAL)
1475                                 desc->error = 0;
1476                         break;
1477                 }
1478
1479                 /*
1480                  * We must evaluate after, since reads (unlike writes)
1481                  * are called without i_sem protection against truncate
1482                  */
1483                 nr = PAGE_CACHE_SIZE;
1484                 i_size = i_size_read(inode);
1485                 end_index = i_size >> PAGE_CACHE_SHIFT;
1486                 if (index == end_index) {
1487                         nr = i_size & ~PAGE_CACHE_MASK;
1488                         if (nr <= offset) {
1489                                 if (page)
1490                                         page_cache_release(page);
1491                                 break;
1492                         }
1493                 }
1494                 nr -= offset;
1495
1496                 if (page) {
1497                         /*
1498                          * If users can be writing to this page using arbitrary
1499                          * virtual addresses, take care about potential aliasing
1500                          * before reading the page on the kernel side.
1501                          */
1502                         if (mapping_writably_mapped(mapping))
1503                                 flush_dcache_page(page);
1504                         /*
1505                          * Mark the page accessed if we read the beginning.
1506                          */
1507                         if (!offset)
1508                                 mark_page_accessed(page);
1509                 } else
1510                         page = ZERO_PAGE(0);
1511
1512                 /*
1513                  * Ok, we have the page, and it's up-to-date, so
1514                  * now we can copy it to user space...
1515                  *
1516                  * The actor routine returns how many bytes were actually used..
1517                  * NOTE! This may not be the same as how much of a user buffer
1518                  * we filled up (we may be padding etc), so we can only update
1519                  * "pos" here (the actor routine has to update the user buffer
1520                  * pointers and the remaining count).
1521                  */
1522                 ret = actor(desc, page, offset, nr);
1523                 offset += ret;
1524                 index += offset >> PAGE_CACHE_SHIFT;
1525                 offset &= ~PAGE_CACHE_MASK;
1526
1527                 page_cache_release(page);
1528                 if (ret != nr || !desc->count)
1529                         break;
1530
1531                 cond_resched();
1532         }
1533
1534         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1535         file_accessed(filp);
1536 }
1537
1538 static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1539 {
1540         read_descriptor_t desc;
1541
1542         if ((ssize_t) count < 0)
1543                 return -EINVAL;
1544         if (!access_ok(VERIFY_WRITE, buf, count))
1545                 return -EFAULT;
1546         if (!count)
1547                 return 0;
1548
1549         desc.written = 0;
1550         desc.count = count;
1551         desc.arg.buf = buf;
1552         desc.error = 0;
1553
1554         do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1555         if (desc.written)
1556                 return desc.written;
1557         return desc.error;
1558 }
1559
1560 static ssize_t shmem_file_sendfile(struct file *in_file, loff_t *ppos,
1561                          size_t count, read_actor_t actor, void *target)
1562 {
1563         read_descriptor_t desc;
1564
1565         if (!count)
1566                 return 0;
1567
1568         desc.written = 0;
1569         desc.count = count;
1570         desc.arg.data = target;
1571         desc.error = 0;
1572
1573         do_shmem_file_read(in_file, ppos, &desc, actor);
1574         if (desc.written)
1575                 return desc.written;
1576         return desc.error;
1577 }
1578
1579 static int shmem_statfs(struct super_block *sb, struct kstatfs *buf)
1580 {
1581         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1582
1583         buf->f_type = TMPFS_MAGIC;
1584         buf->f_bsize = PAGE_CACHE_SIZE;
1585         buf->f_namelen = NAME_MAX;
1586         spin_lock(&sbinfo->stat_lock);
1587         if (sbinfo->max_blocks) {
1588                 buf->f_blocks = sbinfo->max_blocks;
1589                 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1590         }
1591         if (sbinfo->max_inodes) {
1592                 buf->f_files = sbinfo->max_inodes;
1593                 buf->f_ffree = sbinfo->free_inodes;
1594         }
1595         /* else leave those fields 0 like simple_statfs */
1596         spin_unlock(&sbinfo->stat_lock);
1597         return 0;
1598 }
1599
1600 /*
1601  * File creation. Allocate an inode, and we're done..
1602  */
1603 static int
1604 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1605 {
1606         struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1607         int error = -ENOSPC;
1608
1609         if (inode) {
1610                 if (dir->i_mode & S_ISGID) {
1611                         inode->i_gid = dir->i_gid;
1612                         if (S_ISDIR(mode))
1613                                 inode->i_mode |= S_ISGID;
1614                 }
1615                 dir->i_size += BOGO_DIRENT_SIZE;
1616                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1617                 d_instantiate(dentry, inode);
1618                 dget(dentry); /* Extra count - pin the dentry in core */
1619                 error = 0;
1620         }
1621         return error;
1622 }
1623
1624 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1625 {
1626         int error;
1627
1628         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1629                 return error;
1630         dir->i_nlink++;
1631         return 0;
1632 }
1633
1634 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1635                 struct nameidata *nd)
1636 {
1637         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1638 }
1639
1640 /*
1641  * Link a file..
1642  */
1643 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1644 {
1645         struct inode *inode = old_dentry->d_inode;
1646         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1647
1648         /*
1649          * No ordinary (disk based) filesystem counts links as inodes;
1650          * but each new link needs a new dentry, pinning lowmem, and
1651          * tmpfs dentries cannot be pruned until they are unlinked.
1652          */
1653         if (sbinfo->max_inodes) {
1654                 spin_lock(&sbinfo->stat_lock);
1655                 if (!sbinfo->free_inodes) {
1656                         spin_unlock(&sbinfo->stat_lock);
1657                         return -ENOSPC;
1658                 }
1659                 sbinfo->free_inodes--;
1660                 spin_unlock(&sbinfo->stat_lock);
1661         }
1662
1663         dir->i_size += BOGO_DIRENT_SIZE;
1664         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1665         inode->i_nlink++;
1666         atomic_inc(&inode->i_count);    /* New dentry reference */
1667         dget(dentry);           /* Extra pinning count for the created dentry */
1668         d_instantiate(dentry, inode);
1669         return 0;
1670 }
1671
1672 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1673 {
1674         struct inode *inode = dentry->d_inode;
1675
1676         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) {
1677                 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1678                 if (sbinfo->max_inodes) {
1679                         spin_lock(&sbinfo->stat_lock);
1680                         sbinfo->free_inodes++;
1681                         spin_unlock(&sbinfo->stat_lock);
1682                 }
1683         }
1684
1685         dir->i_size -= BOGO_DIRENT_SIZE;
1686         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1687         inode->i_nlink--;
1688         dput(dentry);   /* Undo the count from "create" - this does all the work */
1689         return 0;
1690 }
1691
1692 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1693 {
1694         if (!simple_empty(dentry))
1695                 return -ENOTEMPTY;
1696
1697         dir->i_nlink--;
1698         return shmem_unlink(dir, dentry);
1699 }
1700
1701 /*
1702  * The VFS layer already does all the dentry stuff for rename,
1703  * we just have to decrement the usage count for the target if
1704  * it exists so that the VFS layer correctly free's it when it
1705  * gets overwritten.
1706  */
1707 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1708 {
1709         struct inode *inode = old_dentry->d_inode;
1710         int they_are_dirs = S_ISDIR(inode->i_mode);
1711
1712         if (!simple_empty(new_dentry))
1713                 return -ENOTEMPTY;
1714
1715         if (new_dentry->d_inode) {
1716                 (void) shmem_unlink(new_dir, new_dentry);
1717                 if (they_are_dirs)
1718                         old_dir->i_nlink--;
1719         } else if (they_are_dirs) {
1720                 old_dir->i_nlink--;
1721                 new_dir->i_nlink++;
1722         }
1723
1724         old_dir->i_size -= BOGO_DIRENT_SIZE;
1725         new_dir->i_size += BOGO_DIRENT_SIZE;
1726         old_dir->i_ctime = old_dir->i_mtime =
1727         new_dir->i_ctime = new_dir->i_mtime =
1728         inode->i_ctime = CURRENT_TIME;
1729         return 0;
1730 }
1731
1732 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1733 {
1734         int error;
1735         int len;
1736         struct inode *inode;
1737         struct page *page = NULL;
1738         char *kaddr;
1739         struct shmem_inode_info *info;
1740
1741         len = strlen(symname) + 1;
1742         if (len > PAGE_CACHE_SIZE)
1743                 return -ENAMETOOLONG;
1744
1745         inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1746         if (!inode)
1747                 return -ENOSPC;
1748
1749         info = SHMEM_I(inode);
1750         inode->i_size = len-1;
1751         if (len <= (char *)inode - (char *)info) {
1752                 /* do it inline */
1753                 memcpy(info, symname, len);
1754                 inode->i_op = &shmem_symlink_inline_operations;
1755         } else {
1756                 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1757                 if (error) {
1758                         iput(inode);
1759                         return error;
1760                 }
1761                 inode->i_op = &shmem_symlink_inode_operations;
1762                 kaddr = kmap_atomic(page, KM_USER0);
1763                 memcpy(kaddr, symname, len);
1764                 kunmap_atomic(kaddr, KM_USER0);
1765                 set_page_dirty(page);
1766                 page_cache_release(page);
1767         }
1768         if (dir->i_mode & S_ISGID)
1769                 inode->i_gid = dir->i_gid;
1770         dir->i_size += BOGO_DIRENT_SIZE;
1771         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1772         d_instantiate(dentry, inode);
1773         dget(dentry);
1774         return 0;
1775 }
1776
1777 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1778 {
1779         nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1780         return NULL;
1781 }
1782
1783 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1784 {
1785         struct page *page = NULL;
1786         int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1787         nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1788         return page;
1789 }
1790
1791 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1792 {
1793         if (!IS_ERR(nd_get_link(nd))) {
1794                 struct page *page = cookie;
1795                 kunmap(page);
1796                 mark_page_accessed(page);
1797                 page_cache_release(page);
1798         }
1799 }
1800
1801 static struct inode_operations shmem_symlink_inline_operations = {
1802         .readlink       = generic_readlink,
1803         .follow_link    = shmem_follow_link_inline,
1804 };
1805
1806 static struct inode_operations shmem_symlink_inode_operations = {
1807         .truncate       = shmem_truncate,
1808         .readlink       = generic_readlink,
1809         .follow_link    = shmem_follow_link,
1810         .put_link       = shmem_put_link,
1811 };
1812
1813 static int shmem_parse_options(char *options, int *mode, uid_t *uid, gid_t *gid, unsigned long *blocks, unsigned long *inodes)
1814 {
1815         char *this_char, *value, *rest;
1816
1817         while ((this_char = strsep(&options, ",")) != NULL) {
1818                 if (!*this_char)
1819                         continue;
1820                 if ((value = strchr(this_char,'=')) != NULL) {
1821                         *value++ = 0;
1822                 } else {
1823                         printk(KERN_ERR
1824                             "tmpfs: No value for mount option '%s'\n",
1825                             this_char);
1826                         return 1;
1827                 }
1828
1829                 if (!strcmp(this_char,"size")) {
1830                         unsigned long long size;
1831                         size = memparse(value,&rest);
1832                         if (*rest == '%') {
1833                                 size <<= PAGE_SHIFT;
1834                                 size *= totalram_pages;
1835                                 do_div(size, 100);
1836                                 rest++;
1837                         }
1838                         if (*rest)
1839                                 goto bad_val;
1840                         *blocks = size >> PAGE_CACHE_SHIFT;
1841                 } else if (!strcmp(this_char,"nr_blocks")) {
1842                         *blocks = memparse(value,&rest);
1843                         if (*rest)
1844                                 goto bad_val;
1845                 } else if (!strcmp(this_char,"nr_inodes")) {
1846                         *inodes = memparse(value,&rest);
1847                         if (*rest)
1848                                 goto bad_val;
1849                 } else if (!strcmp(this_char,"mode")) {
1850                         if (!mode)
1851                                 continue;
1852                         *mode = simple_strtoul(value,&rest,8);
1853                         if (*rest)
1854                                 goto bad_val;
1855                 } else if (!strcmp(this_char,"uid")) {
1856                         if (!uid)
1857                                 continue;
1858                         *uid = simple_strtoul(value,&rest,0);
1859                         if (*rest)
1860                                 goto bad_val;
1861                 } else if (!strcmp(this_char,"gid")) {
1862                         if (!gid)
1863                                 continue;
1864                         *gid = simple_strtoul(value,&rest,0);
1865                         if (*rest)
1866                                 goto bad_val;
1867                 } else {
1868                         printk(KERN_ERR "tmpfs: Bad mount option %s\n",
1869                                this_char);
1870                         return 1;
1871                 }
1872         }
1873         return 0;
1874
1875 bad_val:
1876         printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
1877                value, this_char);
1878         return 1;
1879
1880 }
1881
1882 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
1883 {
1884         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1885         unsigned long max_blocks = sbinfo->max_blocks;
1886         unsigned long max_inodes = sbinfo->max_inodes;
1887         unsigned long blocks;
1888         unsigned long inodes;
1889         int error = -EINVAL;
1890
1891         if (shmem_parse_options(data, NULL, NULL, NULL,
1892                                 &max_blocks, &max_inodes))
1893                 return error;
1894
1895         spin_lock(&sbinfo->stat_lock);
1896         blocks = sbinfo->max_blocks - sbinfo->free_blocks;
1897         inodes = sbinfo->max_inodes - sbinfo->free_inodes;
1898         if (max_blocks < blocks)
1899                 goto out;
1900         if (max_inodes < inodes)
1901                 goto out;
1902         /*
1903          * Those tests also disallow limited->unlimited while any are in
1904          * use, so i_blocks will always be zero when max_blocks is zero;
1905          * but we must separately disallow unlimited->limited, because
1906          * in that case we have no record of how much is already in use.
1907          */
1908         if (max_blocks && !sbinfo->max_blocks)
1909                 goto out;
1910         if (max_inodes && !sbinfo->max_inodes)
1911                 goto out;
1912
1913         error = 0;
1914         sbinfo->max_blocks  = max_blocks;
1915         sbinfo->free_blocks = max_blocks - blocks;
1916         sbinfo->max_inodes  = max_inodes;
1917         sbinfo->free_inodes = max_inodes - inodes;
1918 out:
1919         spin_unlock(&sbinfo->stat_lock);
1920         return error;
1921 }
1922 #endif
1923
1924 static void shmem_put_super(struct super_block *sb)
1925 {
1926         kfree(sb->s_fs_info);
1927         sb->s_fs_info = NULL;
1928 }
1929
1930 static int shmem_fill_super(struct super_block *sb,
1931                             void *data, int silent)
1932 {
1933         struct inode *inode;
1934         struct dentry *root;
1935         int mode   = S_IRWXUGO | S_ISVTX;
1936         uid_t uid = current->fsuid;
1937         gid_t gid = current->fsgid;
1938         int err = -ENOMEM;
1939         struct shmem_sb_info *sbinfo;
1940         unsigned long blocks = 0;
1941         unsigned long inodes = 0;
1942
1943 #ifdef CONFIG_TMPFS
1944         /*
1945          * Per default we only allow half of the physical ram per
1946          * tmpfs instance, limiting inodes to one per page of lowmem;
1947          * but the internal instance is left unlimited.
1948          */
1949         if (!(sb->s_flags & MS_NOUSER)) {
1950                 blocks = totalram_pages / 2;
1951                 inodes = totalram_pages - totalhigh_pages;
1952                 if (inodes > blocks)
1953                         inodes = blocks;
1954                 if (shmem_parse_options(data, &mode, &uid, &gid,
1955                                         &blocks, &inodes))
1956                         return -EINVAL;
1957         }
1958 #else
1959         sb->s_flags |= MS_NOUSER;
1960 #endif
1961
1962         /* Round up to L1_CACHE_BYTES to resist false sharing */
1963         sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
1964                                 L1_CACHE_BYTES), GFP_KERNEL);
1965         if (!sbinfo)
1966                 return -ENOMEM;
1967
1968         spin_lock_init(&sbinfo->stat_lock);
1969         sbinfo->max_blocks = blocks;
1970         sbinfo->free_blocks = blocks;
1971         sbinfo->max_inodes = inodes;
1972         sbinfo->free_inodes = inodes;
1973
1974         sb->s_fs_info = sbinfo;
1975         sb->s_maxbytes = SHMEM_MAX_BYTES;
1976         sb->s_blocksize = PAGE_CACHE_SIZE;
1977         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1978         sb->s_magic = TMPFS_MAGIC;
1979         sb->s_op = &shmem_ops;
1980
1981         inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
1982         if (!inode)
1983                 goto failed;
1984         inode->i_uid = uid;
1985         inode->i_gid = gid;
1986         root = d_alloc_root(inode);
1987         if (!root)
1988                 goto failed_iput;
1989         sb->s_root = root;
1990         return 0;
1991
1992 failed_iput:
1993         iput(inode);
1994 failed:
1995         shmem_put_super(sb);
1996         return err;
1997 }
1998
1999 static kmem_cache_t *shmem_inode_cachep;
2000
2001 static struct inode *shmem_alloc_inode(struct super_block *sb)
2002 {
2003         struct shmem_inode_info *p;
2004         p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, SLAB_KERNEL);
2005         if (!p)
2006                 return NULL;
2007         return &p->vfs_inode;
2008 }
2009
2010 static void shmem_destroy_inode(struct inode *inode)
2011 {
2012         if ((inode->i_mode & S_IFMT) == S_IFREG) {
2013                 /* only struct inode is valid if it's an inline symlink */
2014                 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2015         }
2016         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2017 }
2018
2019 static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
2020 {
2021         struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2022
2023         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2024             SLAB_CTOR_CONSTRUCTOR) {
2025                 inode_init_once(&p->vfs_inode);
2026         }
2027 }
2028
2029 static int init_inodecache(void)
2030 {
2031         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2032                                 sizeof(struct shmem_inode_info),
2033                                 0, 0, init_once, NULL);
2034         if (shmem_inode_cachep == NULL)
2035                 return -ENOMEM;
2036         return 0;
2037 }
2038
2039 static void destroy_inodecache(void)
2040 {
2041         if (kmem_cache_destroy(shmem_inode_cachep))
2042                 printk(KERN_INFO "shmem_inode_cache: not all structures were freed\n");
2043 }
2044
2045 static struct address_space_operations shmem_aops = {
2046         .writepage      = shmem_writepage,
2047         .set_page_dirty = __set_page_dirty_nobuffers,
2048 #ifdef CONFIG_TMPFS
2049         .prepare_write  = shmem_prepare_write,
2050         .commit_write   = simple_commit_write,
2051 #endif
2052 };
2053
2054 static struct file_operations shmem_file_operations = {
2055         .mmap           = shmem_mmap,
2056 #ifdef CONFIG_TMPFS
2057         .llseek         = generic_file_llseek,
2058         .read           = shmem_file_read,
2059         .write          = shmem_file_write,
2060         .fsync          = simple_sync_file,
2061         .sendfile       = shmem_file_sendfile,
2062 #endif
2063 };
2064
2065 static struct inode_operations shmem_inode_operations = {
2066         .truncate       = shmem_truncate,
2067         .setattr        = shmem_notify_change,
2068 };
2069
2070 static struct inode_operations shmem_dir_inode_operations = {
2071 #ifdef CONFIG_TMPFS
2072         .create         = shmem_create,
2073         .lookup         = simple_lookup,
2074         .link           = shmem_link,
2075         .unlink         = shmem_unlink,
2076         .symlink        = shmem_symlink,
2077         .mkdir          = shmem_mkdir,
2078         .rmdir          = shmem_rmdir,
2079         .mknod          = shmem_mknod,
2080         .rename         = shmem_rename,
2081 #endif
2082 };
2083
2084 static struct super_operations shmem_ops = {
2085         .alloc_inode    = shmem_alloc_inode,
2086         .destroy_inode  = shmem_destroy_inode,
2087 #ifdef CONFIG_TMPFS
2088         .statfs         = shmem_statfs,
2089         .remount_fs     = shmem_remount_fs,
2090 #endif
2091         .delete_inode   = shmem_delete_inode,
2092         .drop_inode     = generic_delete_inode,
2093         .put_super      = shmem_put_super,
2094 };
2095
2096 static struct vm_operations_struct shmem_vm_ops = {
2097         .nopage         = shmem_nopage,
2098         .populate       = shmem_populate,
2099 #ifdef CONFIG_NUMA
2100         .set_policy     = shmem_set_policy,
2101         .get_policy     = shmem_get_policy,
2102 #endif
2103 };
2104
2105
2106 static struct super_block *shmem_get_sb(struct file_system_type *fs_type,
2107         int flags, const char *dev_name, void *data)
2108 {
2109         return get_sb_nodev(fs_type, flags, data, shmem_fill_super);
2110 }
2111
2112 static struct file_system_type tmpfs_fs_type = {
2113         .owner          = THIS_MODULE,
2114         .name           = "tmpfs",
2115         .get_sb         = shmem_get_sb,
2116         .kill_sb        = kill_litter_super,
2117 };
2118 static struct vfsmount *shm_mnt;
2119
2120 static int __init init_tmpfs(void)
2121 {
2122         int error;
2123
2124         error = init_inodecache();
2125         if (error)
2126                 goto out3;
2127
2128         error = register_filesystem(&tmpfs_fs_type);
2129         if (error) {
2130                 printk(KERN_ERR "Could not register tmpfs\n");
2131                 goto out2;
2132         }
2133 #ifdef CONFIG_TMPFS
2134         devfs_mk_dir("shm");
2135 #endif
2136         shm_mnt = do_kern_mount(tmpfs_fs_type.name, MS_NOUSER,
2137                                 tmpfs_fs_type.name, NULL);
2138         if (IS_ERR(shm_mnt)) {
2139                 error = PTR_ERR(shm_mnt);
2140                 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2141                 goto out1;
2142         }
2143         return 0;
2144
2145 out1:
2146         unregister_filesystem(&tmpfs_fs_type);
2147 out2:
2148         destroy_inodecache();
2149 out3:
2150         shm_mnt = ERR_PTR(error);
2151         return error;
2152 }
2153 module_init(init_tmpfs)
2154
2155 /*
2156  * shmem_file_setup - get an unlinked file living in tmpfs
2157  *
2158  * @name: name for dentry (to be seen in /proc/<pid>/maps
2159  * @size: size to be set for the file
2160  *
2161  */
2162 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2163 {
2164         int error;
2165         struct file *file;
2166         struct inode *inode;
2167         struct dentry *dentry, *root;
2168         struct qstr this;
2169
2170         if (IS_ERR(shm_mnt))
2171                 return (void *)shm_mnt;
2172
2173         if (size < 0 || size > SHMEM_MAX_BYTES)
2174                 return ERR_PTR(-EINVAL);
2175
2176         if (shmem_acct_size(flags, size))
2177                 return ERR_PTR(-ENOMEM);
2178
2179         error = -ENOMEM;
2180         this.name = name;
2181         this.len = strlen(name);
2182         this.hash = 0; /* will go */
2183         root = shm_mnt->mnt_root;
2184         dentry = d_alloc(root, &this);
2185         if (!dentry)
2186                 goto put_memory;
2187
2188         error = -ENFILE;
2189         file = get_empty_filp();
2190         if (!file)
2191                 goto put_dentry;
2192
2193         error = -ENOSPC;
2194         inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2195         if (!inode)
2196                 goto close_file;
2197
2198         SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2199         d_instantiate(dentry, inode);
2200         inode->i_size = size;
2201         inode->i_nlink = 0;     /* It is unlinked */
2202         file->f_vfsmnt = mntget(shm_mnt);
2203         file->f_dentry = dentry;
2204         file->f_mapping = inode->i_mapping;
2205         file->f_op = &shmem_file_operations;
2206         file->f_mode = FMODE_WRITE | FMODE_READ;
2207         return file;
2208
2209 close_file:
2210         put_filp(file);
2211 put_dentry:
2212         dput(dentry);
2213 put_memory:
2214         shmem_unacct_size(flags, size);
2215         return ERR_PTR(error);
2216 }
2217
2218 /*
2219  * shmem_zero_setup - setup a shared anonymous mapping
2220  *
2221  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2222  */
2223 int shmem_zero_setup(struct vm_area_struct *vma)
2224 {
2225         struct file *file;
2226         loff_t size = vma->vm_end - vma->vm_start;
2227
2228         file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2229         if (IS_ERR(file))
2230                 return PTR_ERR(file);
2231
2232         if (vma->vm_file)
2233                 fput(vma->vm_file);
2234         vma->vm_file = file;
2235         vma->vm_ops = &shmem_vm_ops;
2236         return 0;
2237 }