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/module.h>
  27 #include <linux/init.h>
  28 #include <linux/fs.h>
  29 #include <linux/xattr.h>
  30 #include <linux/generic_acl.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 <linux/ctype.h>
  49 #include <linux/migrate.h>
  50 #include <linux/highmem.h>
  51 #include <linux/backing-dev.h>
  52
  53 #include <asm/uaccess.h>
  54 #include <asm/div64.h>
  55 #include <asm/pgtable.h>
  56
  57 /* This magic number is used in glibc for posix shared memory */
  58 #define TMPFS_MAGIC     0x01021994
  59
  60 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
  61 #define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
  62 #define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
  63
  64 #define SHMEM_MAX_INDEX  (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
  65 #define SHMEM_MAX_BYTES  ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
  66
  67 #define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
  68
  69 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
  70 #define SHMEM_PAGEIN     VM_READ
  71 #define SHMEM_TRUNCATE   VM_WRITE
  72
  73 /* Definition to limit shmem_truncate's steps between cond_rescheds */
  74 #define LATENCY_LIMIT    64
  75
  76 /* Pretend that each entry is of this size in directory's i_size */
  77 #define BOGO_DIRENT_SIZE 20
  78
  79 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
  80 enum sgp_type {
  81         SGP_QUICK,      /* don't try more than file page cache lookup */
  82         SGP_READ,       /* don't exceed i_size, don't allocate page */
  83         SGP_CACHE,      /* don't exceed i_size, may allocate page */
  84         SGP_WRITE,      /* may exceed i_size, may allocate page */
  85 };
  86
  87 static int shmem_getpage(struct inode *inode, unsigned long idx,
  88                          struct page **pagep, enum sgp_type sgp, int *type);
  89
  90 static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
  91 {
  92         /*
  93          * The above definition of ENTRIES_PER_PAGE, and the use of
  94          * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
  95          * might be reconsidered if it ever diverges from PAGE_SIZE.
  96          */
  97         return alloc_pages(gfp_mask, PAGE_CACHE_SHIFT-PAGE_SHIFT);
  98 }
  99
 100 static inline void shmem_dir_free(struct page *page)
 101 {
 102         __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
 103 }
 104
 105 static struct page **shmem_dir_map(struct page *page)
 106 {
 107         return (struct page **)kmap_atomic(page, KM_USER0);
 108 }
 109
 110 static inline void shmem_dir_unmap(struct page **dir)
 111 {
 112         kunmap_atomic(dir, KM_USER0);
 113 }
 114
 115 static swp_entry_t *shmem_swp_map(struct page *page)
 116 {
 117         return (swp_entry_t *)kmap_atomic(page, KM_USER1);
 118 }
 119
 120 static inline void shmem_swp_balance_unmap(void)
 121 {
 122         /*
 123          * When passing a pointer to an i_direct entry, to code which
 124          * also handles indirect entries and so will shmem_swp_unmap,
 125          * we must arrange for the preempt count to remain in balance.
 126          * What kmap_atomic of a lowmem page does depends on config
 127          * and architecture, so pretend to kmap_atomic some lowmem page.
 128          */
 129         (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
 130 }
 131
 132 static inline void shmem_swp_unmap(swp_entry_t *entry)
 133 {
 134         kunmap_atomic(entry, KM_USER1);
 135 }
 136
 137 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
 138 {
 139         return sb->s_fs_info;
 140 }
 141
 142 /*
 143  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
 144  * for shared memory and for shared anonymous (/dev/zero) mappings
 145  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
 146  * consistent with the pre-accounting of private mappings ...
 147  */
 148 static inline int shmem_acct_size(unsigned long flags, loff_t size)
 149 {
 150         return (flags & VM_ACCOUNT)?
 151                 security_vm_enough_memory(VM_ACCT(size)): 0;
 152 }
 153
 154 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
 155 {
 156         if (flags & VM_ACCOUNT)
 157                 vm_unacct_memory(VM_ACCT(size));
 158 }
 159
 160 /*
 161  * ... whereas tmpfs objects are accounted incrementally as
 162  * pages are allocated, in order to allow huge sparse files.
 163  * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
 164  * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
 165  */
 166 static inline int shmem_acct_block(unsigned long flags)
 167 {
 168         return (flags & VM_ACCOUNT)?
 169                 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
 170 }
 171
 172 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
 173 {
 174         if (!(flags & VM_ACCOUNT))
 175                 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
 176 }
 177
 178 static const struct super_operations shmem_ops;
 179 static const struct address_space_operations shmem_aops;
 180 static const struct file_operations shmem_file_operations;
 181 static const struct inode_operations shmem_inode_operations;
 182 static const struct inode_operations shmem_dir_inode_operations;
 183 static const struct inode_operations shmem_special_inode_operations;
 184 static struct vm_operations_struct shmem_vm_ops;
 185
 186 static struct backing_dev_info shmem_backing_dev_info  __read_mostly = {
 187         .ra_pages       = 0,    /* No readahead */
 188         .capabilities   = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
 189         .unplug_io_fn   = default_unplug_io_fn,
 190 };
 191
 192 static LIST_HEAD(shmem_swaplist);
 193 static DEFINE_SPINLOCK(shmem_swaplist_lock);
 194
 195 static void shmem_free_blocks(struct inode *inode, long pages)
 196 {
 197         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 198         if (sbinfo->max_blocks) {
 199                 spin_lock(&sbinfo->stat_lock);
 200                 sbinfo->free_blocks += pages;
 201                 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
 202                 spin_unlock(&sbinfo->stat_lock);
 203         }
 204 }
 205
 206 /*
 207  * shmem_recalc_inode - recalculate the size of an inode
 208  *
 209  * @inode: inode to recalc
 210  *
 211  * We have to calculate the free blocks since the mm can drop
 212  * undirtied hole pages behind our back.
 213  *
 214  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
 215  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
 216  *
 217  * It has to be called with the spinlock held.
 218  */
 219 static void shmem_recalc_inode(struct inode *inode)
 220 {
 221         struct shmem_inode_info *info = SHMEM_I(inode);
 222         long freed;
 223
 224         freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
 225         if (freed > 0) {
 226                 info->alloced -= freed;
 227                 shmem_unacct_blocks(info->flags, freed);
 228                 shmem_free_blocks(inode, freed);
 229         }
 230 }
 231
 232 /*
 233  * shmem_swp_entry - find the swap vector position in the info structure
 234  *
 235  * @info:  info structure for the inode
 236  * @index: index of the page to find
 237  * @page:  optional page to add to the structure. Has to be preset to
 238  *         all zeros
 239  *
 240  * If there is no space allocated yet it will return NULL when
 241  * page is NULL, else it will use the page for the needed block,
 242  * setting it to NULL on return to indicate that it has been used.
 243  *
 244  * The swap vector is organized the following way:
 245  *
 246  * There are SHMEM_NR_DIRECT entries directly stored in the
 247  * shmem_inode_info structure. So small files do not need an addional
 248  * allocation.
 249  *
 250  * For pages with index > SHMEM_NR_DIRECT there is the pointer
 251  * i_indirect which points to a page which holds in the first half
 252  * doubly indirect blocks, in the second half triple indirect blocks:
 253  *
 254  * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
 255  * following layout (for SHMEM_NR_DIRECT == 16):
 256  *
 257  * i_indirect -> dir --> 16-19
 258  *            |      +-> 20-23
 259  *            |
 260  *            +-->dir2 --> 24-27
 261  *            |        +-> 28-31
 262  *            |        +-> 32-35
 263  *            |        +-> 36-39
 264  *            |
 265  *            +-->dir3 --> 40-43
 266  *                     +-> 44-47
 267  *                     +-> 48-51
 268  *                     +-> 52-55
 269  */
 270 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
 271 {
 272         unsigned long offset;
 273         struct page **dir;
 274         struct page *subdir;
 275
 276         if (index < SHMEM_NR_DIRECT) {
 277                 shmem_swp_balance_unmap();
 278                 return info->i_direct+index;
 279         }
 280         if (!info->i_indirect) {
 281                 if (page) {
 282                         info->i_indirect = *page;
 283                         *page = NULL;
 284                 }
 285                 return NULL;                    /* need another page */
 286         }
 287
 288         index -= SHMEM_NR_DIRECT;
 289         offset = index % ENTRIES_PER_PAGE;
 290         index /= ENTRIES_PER_PAGE;
 291         dir = shmem_dir_map(info->i_indirect);
 292
 293         if (index >= ENTRIES_PER_PAGE/2) {
 294                 index -= ENTRIES_PER_PAGE/2;
 295                 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
 296                 index %= ENTRIES_PER_PAGE;
 297                 subdir = *dir;
 298                 if (!subdir) {
 299                         if (page) {
 300                                 *dir = *page;
 301                                 *page = NULL;
 302                         }
 303                         shmem_dir_unmap(dir);
 304                         return NULL;            /* need another page */
 305                 }
 306                 shmem_dir_unmap(dir);
 307                 dir = shmem_dir_map(subdir);
 308         }
 309
 310         dir += index;
 311         subdir = *dir;
 312         if (!subdir) {
 313                 if (!page || !(subdir = *page)) {
 314                         shmem_dir_unmap(dir);
 315                         return NULL;            /* need a page */
 316                 }
 317                 *dir = subdir;
 318                 *page = NULL;
 319         }
 320         shmem_dir_unmap(dir);
 321         return shmem_swp_map(subdir) + offset;
 322 }
 323
 324 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
 325 {
 326         long incdec = value? 1: -1;
 327
 328         entry->val = value;
 329         info->swapped += incdec;
 330         if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
 331                 struct page *page = kmap_atomic_to_page(entry);
 332                 set_page_private(page, page_private(page) + incdec);
 333         }
 334 }
 335
 336 /*
 337  * shmem_swp_alloc - get the position of the swap entry for the page.
 338  *                   If it does not exist allocate the entry.
 339  *
 340  * @info:       info structure for the inode
 341  * @index:      index of the page to find
 342  * @sgp:        check and recheck i_size? skip allocation?
 343  */
 344 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
 345 {
 346         struct inode *inode = &info->vfs_inode;
 347         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 348         struct page *page = NULL;
 349         swp_entry_t *entry;
 350
 351         if (sgp != SGP_WRITE &&
 352             ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
 353                 return ERR_PTR(-EINVAL);
 354
 355         while (!(entry = shmem_swp_entry(info, index, &page))) {
 356                 if (sgp == SGP_READ)
 357                         return shmem_swp_map(ZERO_PAGE(0));
 358                 /*
 359                  * Test free_blocks against 1 not 0, since we have 1 data
 360                  * page (and perhaps indirect index pages) yet to allocate:
 361                  * a waste to allocate index if we cannot allocate data.
 362                  */
 363                 if (sbinfo->max_blocks) {
 364                         spin_lock(&sbinfo->stat_lock);
 365                         if (sbinfo->free_blocks <= 1) {
 366                                 spin_unlock(&sbinfo->stat_lock);
 367                                 return ERR_PTR(-ENOSPC);
 368                         }
 369                         sbinfo->free_blocks--;
 370                         inode->i_blocks += BLOCKS_PER_PAGE;
 371                         spin_unlock(&sbinfo->stat_lock);
 372                 }
 373
 374                 spin_unlock(&info->lock);
 375                 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping) | __GFP_ZERO);
 376                 if (page)
 377                         set_page_private(page, 0);
 378                 spin_lock(&info->lock);
 379
 380                 if (!page) {
 381                         shmem_free_blocks(inode, 1);
 382                         return ERR_PTR(-ENOMEM);
 383                 }
 384                 if (sgp != SGP_WRITE &&
 385                     ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
 386                         entry = ERR_PTR(-EINVAL);
 387                         break;
 388                 }
 389                 if (info->next_index <= index)
 390                         info->next_index = index + 1;
 391         }
 392         if (page) {
 393                 /* another task gave its page, or truncated the file */
 394                 shmem_free_blocks(inode, 1);
 395                 shmem_dir_free(page);
 396         }
 397         if (info->next_index <= index && !IS_ERR(entry))
 398                 info->next_index = index + 1;
 399         return entry;
 400 }
 401
 402 /*
 403  * shmem_free_swp - free some swap entries in a directory
 404  *
 405  * @dir:   pointer to the directory
 406  * @edir:  pointer after last entry of the directory
 407  */
 408 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir)
 409 {
 410         swp_entry_t *ptr;
 411         int freed = 0;
 412
 413         for (ptr = dir; ptr < edir; ptr++) {
 414                 if (ptr->val) {
 415                         free_swap_and_cache(*ptr);
 416                         *ptr = (swp_entry_t){0};
 417                         freed++;
 418                 }
 419         }
 420         return freed;
 421 }
 422
 423 static int shmem_map_and_free_swp(struct page *subdir,
 424                 int offset, int limit, struct page ***dir)
 425 {
 426         swp_entry_t *ptr;
 427         int freed = 0;
 428
 429         ptr = shmem_swp_map(subdir);
 430         for (; offset < limit; offset += LATENCY_LIMIT) {
 431                 int size = limit - offset;
 432                 if (size > LATENCY_LIMIT)
 433                         size = LATENCY_LIMIT;
 434                 freed += shmem_free_swp(ptr+offset, ptr+offset+size);
 435                 if (need_resched()) {
 436                         shmem_swp_unmap(ptr);
 437                         if (*dir) {
 438                                 shmem_dir_unmap(*dir);
 439                                 *dir = NULL;
 440                         }
 441                         cond_resched();
 442                         ptr = shmem_swp_map(subdir);
 443                 }
 444         }
 445         shmem_swp_unmap(ptr);
 446         return freed;
 447 }
 448
 449 static void shmem_free_pages(struct list_head *next)
 450 {
 451         struct page *page;
 452         int freed = 0;
 453
 454         do {
 455                 page = container_of(next, struct page, lru);
 456                 next = next->next;
 457                 shmem_dir_free(page);
 458                 freed++;
 459                 if (freed >= LATENCY_LIMIT) {
 460                         cond_resched();
 461                         freed = 0;
 462                 }
 463         } while (next);
 464 }
 465
 466 static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
 467 {
 468         struct shmem_inode_info *info = SHMEM_I(inode);
 469         unsigned long idx;
 470         unsigned long size;
 471         unsigned long limit;
 472         unsigned long stage;
 473         unsigned long diroff;
 474         struct page **dir;
 475         struct page *topdir;
 476         struct page *middir;
 477         struct page *subdir;
 478         swp_entry_t *ptr;
 479         LIST_HEAD(pages_to_free);
 480         long nr_pages_to_free = 0;
 481         long nr_swaps_freed = 0;
 482         int offset;
 483         int freed;
 484         int punch_hole = 0;
 485
 486         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 487         idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 488         if (idx >= info->next_index)
 489                 return;
 490
 491         spin_lock(&info->lock);
 492         info->flags |= SHMEM_TRUNCATE;
 493         if (likely(end == (loff_t) -1)) {
 494                 limit = info->next_index;
 495                 info->next_index = idx;
 496         } else {
 497                 limit = (end + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 498                 if (limit > info->next_index)
 499                         limit = info->next_index;
 500                 punch_hole = 1;
 501         }
 502
 503         topdir = info->i_indirect;
 504         if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
 505                 info->i_indirect = NULL;
 506                 nr_pages_to_free++;
 507                 list_add(&topdir->lru, &pages_to_free);
 508         }
 509         spin_unlock(&info->lock);
 510
 511         if (info->swapped && idx < SHMEM_NR_DIRECT) {
 512                 ptr = info->i_direct;
 513                 size = limit;
 514                 if (size > SHMEM_NR_DIRECT)
 515                         size = SHMEM_NR_DIRECT;
 516                 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size);
 517         }
 518
 519         /*
 520          * If there are no indirect blocks or we are punching a hole
 521          * below indirect blocks, nothing to be done.
 522          */
 523         if (!topdir || (punch_hole && (limit <= SHMEM_NR_DIRECT)))
 524                 goto done2;
 525
 526         BUG_ON(limit <= SHMEM_NR_DIRECT);
 527         limit -= SHMEM_NR_DIRECT;
 528         idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
 529         offset = idx % ENTRIES_PER_PAGE;
 530         idx -= offset;
 531
 532         dir = shmem_dir_map(topdir);
 533         stage = ENTRIES_PER_PAGEPAGE/2;
 534         if (idx < ENTRIES_PER_PAGEPAGE/2) {
 535                 middir = topdir;
 536                 diroff = idx/ENTRIES_PER_PAGE;
 537         } else {
 538                 dir += ENTRIES_PER_PAGE/2;
 539                 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
 540                 while (stage <= idx)
 541                         stage += ENTRIES_PER_PAGEPAGE;
 542                 middir = *dir;
 543                 if (*dir) {
 544                         diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
 545                                 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
 546                         if (!diroff && !offset) {
 547                                 *dir = NULL;
 548                                 nr_pages_to_free++;
 549                                 list_add(&middir->lru, &pages_to_free);
 550                         }
 551                         shmem_dir_unmap(dir);
 552                         dir = shmem_dir_map(middir);
 553                 } else {
 554                         diroff = 0;
 555                         offset = 0;
 556                         idx = stage;
 557                 }
 558         }
 559
 560         for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
 561                 if (unlikely(idx == stage)) {
 562                         shmem_dir_unmap(dir);
 563                         dir = shmem_dir_map(topdir) +
 564                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
 565                         while (!*dir) {
 566                                 dir++;
 567                                 idx += ENTRIES_PER_PAGEPAGE;
 568                                 if (idx >= limit)
 569                                         goto done1;
 570                         }
 571                         stage = idx + ENTRIES_PER_PAGEPAGE;
 572                         middir = *dir;
 573                         *dir = NULL;
 574                         nr_pages_to_free++;
 575                         list_add(&middir->lru, &pages_to_free);
 576                         shmem_dir_unmap(dir);
 577                         cond_resched();
 578                         dir = shmem_dir_map(middir);
 579                         diroff = 0;
 580                 }
 581                 subdir = dir[diroff];
 582                 if (subdir && page_private(subdir)) {
 583                         size = limit - idx;
 584                         if (size > ENTRIES_PER_PAGE)
 585                                 size = ENTRIES_PER_PAGE;
 586                         freed = shmem_map_and_free_swp(subdir,
 587                                                 offset, size, &dir);
 588                         if (!dir)
 589                                 dir = shmem_dir_map(middir);
 590                         nr_swaps_freed += freed;
 591                         if (offset)
 592                                 spin_lock(&info->lock);
 593                         set_page_private(subdir, page_private(subdir) - freed);
 594                         if (offset)
 595                                 spin_unlock(&info->lock);
 596                         if (!punch_hole)
 597                                 BUG_ON(page_private(subdir) > offset);
 598                 }
 599                 if (offset)
 600                         offset = 0;
 601                 else if (subdir && !page_private(subdir)) {
 602                         dir[diroff] = NULL;
 603                         nr_pages_to_free++;
 604                         list_add(&subdir->lru, &pages_to_free);
 605                 }
 606         }
 607 done1:
 608         shmem_dir_unmap(dir);
 609 done2:
 610         if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
 611                 /*
 612                  * Call truncate_inode_pages again: racing shmem_unuse_inode
 613                  * may have swizzled a page in from swap since vmtruncate or
 614                  * generic_delete_inode did it, before we lowered next_index.
 615                  * Also, though shmem_getpage checks i_size before adding to
 616                  * cache, no recheck after: so fix the narrow window there too.
 617                  */
 618                 truncate_inode_pages_range(inode->i_mapping, start, end);
 619         }
 620
 621         spin_lock(&info->lock);
 622         info->flags &= ~SHMEM_TRUNCATE;
 623         info->swapped -= nr_swaps_freed;
 624         if (nr_pages_to_free)
 625                 shmem_free_blocks(inode, nr_pages_to_free);
 626         shmem_recalc_inode(inode);
 627         spin_unlock(&info->lock);
 628
 629         /*
 630          * Empty swap vector directory pages to be freed?
 631          */
 632         if (!list_empty(&pages_to_free)) {
 633                 pages_to_free.prev->next = NULL;
 634                 shmem_free_pages(pages_to_free.next);
 635         }
 636 }
 637
 638 static void shmem_truncate(struct inode *inode)
 639 {
 640         shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
 641 }
 642
 643 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
 644 {
 645         struct inode *inode = dentry->d_inode;
 646         struct page *page = NULL;
 647         int error;
 648
 649         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
 650                 if (attr->ia_size < inode->i_size) {
 651                         /*
 652                          * If truncating down to a partial page, then
 653                          * if that page is already allocated, hold it
 654                          * in memory until the truncation is over, so
 655                          * truncate_partial_page cannnot miss it were
 656                          * it assigned to swap.
 657                          */
 658                         if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
 659                                 (void) shmem_getpage(inode,
 660                                         attr->ia_size>>PAGE_CACHE_SHIFT,
 661                                                 &page, SGP_READ, NULL);
 662                         }
 663                         /*
 664                          * Reset SHMEM_PAGEIN flag so that shmem_truncate can
 665                          * detect if any pages might have been added to cache
 666                          * after truncate_inode_pages.  But we needn't bother
 667                          * if it's being fully truncated to zero-length: the
 668                          * nrpages check is efficient enough in that case.
 669                          */
 670                         if (attr->ia_size) {
 671                                 struct shmem_inode_info *info = SHMEM_I(inode);
 672                                 spin_lock(&info->lock);
 673                                 info->flags &= ~SHMEM_PAGEIN;
 674                                 spin_unlock(&info->lock);
 675                         }
 676                 }
 677         }
 678
 679         error = inode_change_ok(inode, attr);
 680         if (!error)
 681                 error = inode_setattr(inode, attr);
 682 #ifdef CONFIG_TMPFS_POSIX_ACL
 683         if (!error && (attr->ia_valid & ATTR_MODE))
 684                 error = generic_acl_chmod(inode, &shmem_acl_ops);
 685 #endif
 686         if (page)
 687                 page_cache_release(page);
 688         return error;
 689 }
 690
 691 static void shmem_delete_inode(struct inode *inode)
 692 {
 693         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 694         struct shmem_inode_info *info = SHMEM_I(inode);
 695
 696         if (inode->i_op->truncate == shmem_truncate) {
 697                 truncate_inode_pages(inode->i_mapping, 0);
 698                 shmem_unacct_size(info->flags, inode->i_size);
 699                 inode->i_size = 0;
 700                 shmem_truncate(inode);
 701                 if (!list_empty(&info->swaplist)) {
 702                         spin_lock(&shmem_swaplist_lock);
 703                         list_del_init(&info->swaplist);
 704                         spin_unlock(&shmem_swaplist_lock);
 705                 }
 706         }
 707         BUG_ON(inode->i_blocks);
 708         if (sbinfo->max_inodes) {
 709                 spin_lock(&sbinfo->stat_lock);
 710                 sbinfo->free_inodes++;
 711                 spin_unlock(&sbinfo->stat_lock);
 712         }
 713         clear_inode(inode);
 714 }
 715
 716 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
 717 {
 718         swp_entry_t *ptr;
 719
 720         for (ptr = dir; ptr < edir; ptr++) {
 721                 if (ptr->val == entry.val)
 722                         return ptr - dir;
 723         }
 724         return -1;
 725 }
 726
 727 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
 728 {
 729         struct inode *inode;
 730         unsigned long idx;
 731         unsigned long size;
 732         unsigned long limit;
 733         unsigned long stage;
 734         struct page **dir;
 735         struct page *subdir;
 736         swp_entry_t *ptr;
 737         int offset;
 738
 739         idx = 0;
 740         ptr = info->i_direct;
 741         spin_lock(&info->lock);
 742         limit = info->next_index;
 743         size = limit;
 744         if (size > SHMEM_NR_DIRECT)
 745                 size = SHMEM_NR_DIRECT;
 746         offset = shmem_find_swp(entry, ptr, ptr+size);
 747         if (offset >= 0) {
 748                 shmem_swp_balance_unmap();
 749                 goto found;
 750         }
 751         if (!info->i_indirect)
 752                 goto lost2;
 753
 754         dir = shmem_dir_map(info->i_indirect);
 755         stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
 756
 757         for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
 758                 if (unlikely(idx == stage)) {
 759                         shmem_dir_unmap(dir-1);
 760                         dir = shmem_dir_map(info->i_indirect) +
 761                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
 762                         while (!*dir) {
 763                                 dir++;
 764                                 idx += ENTRIES_PER_PAGEPAGE;
 765                                 if (idx >= limit)
 766                                         goto lost1;
 767                         }
 768                         stage = idx + ENTRIES_PER_PAGEPAGE;
 769                         subdir = *dir;
 770                         shmem_dir_unmap(dir);
 771                         dir = shmem_dir_map(subdir);
 772                 }
 773                 subdir = *dir;
 774                 if (subdir && page_private(subdir)) {
 775                         ptr = shmem_swp_map(subdir);
 776                         size = limit - idx;
 777                         if (size > ENTRIES_PER_PAGE)
 778                                 size = ENTRIES_PER_PAGE;
 779                         offset = shmem_find_swp(entry, ptr, ptr+size);
 780                         if (offset >= 0) {
 781                                 shmem_dir_unmap(dir);
 782                                 goto found;
 783                         }
 784                         shmem_swp_unmap(ptr);
 785                 }
 786         }
 787 lost1:
 788         shmem_dir_unmap(dir-1);
 789 lost2:
 790         spin_unlock(&info->lock);
 791         return 0;
 792 found:
 793         idx += offset;
 794         inode = &info->vfs_inode;
 795         if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
 796                 info->flags |= SHMEM_PAGEIN;
 797                 shmem_swp_set(info, ptr + offset, 0);
 798         }
 799         shmem_swp_unmap(ptr);
 800         spin_unlock(&info->lock);
 801         /*
 802          * Decrement swap count even when the entry is left behind:
 803          * try_to_unuse will skip over mms, then reincrement count.
 804          */
 805         swap_free(entry);
 806         return 1;
 807 }
 808
 809 /*
 810  * shmem_unuse() search for an eventually swapped out shmem page.
 811  */
 812 int shmem_unuse(swp_entry_t entry, struct page *page)
 813 {
 814         struct list_head *p, *next;
 815         struct shmem_inode_info *info;
 816         int found = 0;
 817
 818         spin_lock(&shmem_swaplist_lock);
 819         list_for_each_safe(p, next, &shmem_swaplist) {
 820                 info = list_entry(p, struct shmem_inode_info, swaplist);
 821                 if (!info->swapped)
 822                         list_del_init(&info->swaplist);
 823                 else if (shmem_unuse_inode(info, entry, page)) {
 824                         /* move head to start search for next from here */
 825                         list_move_tail(&shmem_swaplist, &info->swaplist);
 826                         found = 1;
 827                         break;
 828                 }
 829         }
 830         spin_unlock(&shmem_swaplist_lock);
 831         return found;
 832 }
 833
 834 /*
 835  * Move the page from the page cache to the swap cache.
 836  */
 837 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
 838 {
 839         struct shmem_inode_info *info;
 840         swp_entry_t *entry, swap;
 841         struct address_space *mapping;
 842         unsigned long index;
 843         struct inode *inode;
 844
 845         BUG_ON(!PageLocked(page));
 846         BUG_ON(page_mapped(page));
 847
 848         mapping = page->mapping;
 849         index = page->index;
 850         inode = mapping->host;
 851         info = SHMEM_I(inode);
 852         if (info->flags & VM_LOCKED)
 853                 goto redirty;
 854         swap = get_swap_page();
 855         if (!swap.val)
 856                 goto redirty;
 857
 858         spin_lock(&info->lock);
 859         shmem_recalc_inode(inode);
 860         if (index >= info->next_index) {
 861                 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
 862                 goto unlock;
 863         }
 864         entry = shmem_swp_entry(info, index, NULL);
 865         BUG_ON(!entry);
 866         BUG_ON(entry->val);
 867
 868         if (move_to_swap_cache(page, swap) == 0) {
 869                 shmem_swp_set(info, entry, swap.val);
 870                 shmem_swp_unmap(entry);
 871                 spin_unlock(&info->lock);
 872                 if (list_empty(&info->swaplist)) {
 873                         spin_lock(&shmem_swaplist_lock);
 874                         /* move instead of add in case we're racing */
 875                         list_move_tail(&info->swaplist, &shmem_swaplist);
 876                         spin_unlock(&shmem_swaplist_lock);
 877                 }
 878                 unlock_page(page);
 879                 return 0;
 880         }
 881
 882         shmem_swp_unmap(entry);
 883 unlock:
 884         spin_unlock(&info->lock);
 885         swap_free(swap);
 886 redirty:
 887         set_page_dirty(page);
 888         return AOP_WRITEPAGE_ACTIVATE;  /* Return with the page locked */
 889 }
 890
 891 #ifdef CONFIG_NUMA
 892 static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
 893 {
 894         char *nodelist = strchr(value, ':');
 895         int err = 1;
 896
 897         if (nodelist) {
 898                 /* NUL-terminate policy string */
 899                 *nodelist++ = '\0';
 900                 if (nodelist_parse(nodelist, *policy_nodes))
 901                         goto out;
 902         }
 903         if (!strcmp(value, "default")) {
 904                 *policy = MPOL_DEFAULT;
 905                 /* Don't allow a nodelist */
 906                 if (!nodelist)
 907                         err = 0;
 908         } else if (!strcmp(value, "prefer")) {
 909                 *policy = MPOL_PREFERRED;
 910                 /* Insist on a nodelist of one node only */
 911                 if (nodelist) {
 912                         char *rest = nodelist;
 913                         while (isdigit(*rest))
 914                                 rest++;
 915                         if (!*rest)
 916                                 err = 0;
 917                 }
 918         } else if (!strcmp(value, "bind")) {
 919                 *policy = MPOL_BIND;
 920                 /* Insist on a nodelist */
 921                 if (nodelist)
 922                         err = 0;
 923         } else if (!strcmp(value, "interleave")) {
 924                 *policy = MPOL_INTERLEAVE;
 925                 /* Default to nodes online if no nodelist */
 926                 if (!nodelist)
 927                         *policy_nodes = node_online_map;
 928                 err = 0;
 929         }
 930 out:
 931         /* Restore string for error message */
 932         if (nodelist)
 933                 *--nodelist = ':';
 934         return err;
 935 }
 936
 937 static struct page *shmem_swapin_async(struct shared_policy *p,
 938                                        swp_entry_t entry, unsigned long idx)
 939 {
 940         struct page *page;
 941         struct vm_area_struct pvma;
 942
 943         /* Create a pseudo vma that just contains the policy */
 944         memset(&pvma, 0, sizeof(struct vm_area_struct));
 945         pvma.vm_end = PAGE_SIZE;
 946         pvma.vm_pgoff = idx;
 947         pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
 948         page = read_swap_cache_async(entry, &pvma, 0);
 949         mpol_free(pvma.vm_policy);
 950         return page;
 951 }
 952
 953 struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry,
 954                           unsigned long idx)
 955 {
 956         struct shared_policy *p = &info->policy;
 957         int i, num;
 958         struct page *page;
 959         unsigned long offset;
 960
 961         num = valid_swaphandles(entry, &offset);
 962         for (i = 0; i < num; offset++, i++) {
 963                 page = shmem_swapin_async(p,
 964                                 swp_entry(swp_type(entry), offset), idx);
 965                 if (!page)
 966                         break;
 967                 page_cache_release(page);
 968         }
 969         lru_add_drain();        /* Push any new pages onto the LRU now */
 970         return shmem_swapin_async(p, entry, idx);
 971 }
 972
 973 static struct page *
 974 shmem_alloc_page(gfp_t gfp, struct shmem_inode_info *info,
 975                  unsigned long idx)
 976 {
 977         struct vm_area_struct pvma;
 978         struct page *page;
 979
 980         memset(&pvma, 0, sizeof(struct vm_area_struct));
 981         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
 982         pvma.vm_pgoff = idx;
 983         pvma.vm_end = PAGE_SIZE;
 984         page = alloc_page_vma(gfp | __GFP_ZERO, &pvma, 0);
 985         mpol_free(pvma.vm_policy);
 986         return page;
 987 }
 988 #else
 989 static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
 990 {
 991         return 1;
 992 }
 993
 994 static inline struct page *
 995 shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
 996 {
 997         swapin_readahead(entry, 0, NULL);
 998         return read_swap_cache_async(entry, NULL, 0);
 999 }
1000
1001 static inline struct page *
1002 shmem_alloc_page(gfp_t gfp,struct shmem_inode_info *info, unsigned long idx)
1003 {
1004         return alloc_page(gfp | __GFP_ZERO);
1005 }
1006 #endif
1007
1008 /*
1009  * shmem_getpage - either get the page from swap or allocate a new one
1010  *
1011  * If we allocate a new one we do not mark it dirty. That's up to the
1012  * vm. If we swap it in we mark it dirty since we also free the swap
1013  * entry since a page cannot live in both the swap and page cache
1014  */
1015 static int shmem_getpage(struct inode *inode, unsigned long idx,
1016                         struct page **pagep, enum sgp_type sgp, int *type)
1017 {
1018         struct address_space *mapping = inode->i_mapping;
1019         struct shmem_inode_info *info = SHMEM_I(inode);
1020         struct shmem_sb_info *sbinfo;
1021         struct page *filepage = *pagep;
1022         struct page *swappage;
1023         swp_entry_t *entry;
1024         swp_entry_t swap;
1025         int error;
1026
1027         if (idx >= SHMEM_MAX_INDEX)
1028                 return -EFBIG;
1029         /*
1030          * Normally, filepage is NULL on entry, and either found
1031          * uptodate immediately, or allocated and zeroed, or read
1032          * in under swappage, which is then assigned to filepage.
1033          * But shmem_prepare_write passes in a locked filepage,
1034          * which may be found not uptodate by other callers too,
1035          * and may need to be copied from the swappage read in.
1036          */
1037 repeat:
1038         if (!filepage)
1039                 filepage = find_lock_page(mapping, idx);
1040         if (filepage && PageUptodate(filepage))
1041                 goto done;
1042         error = 0;
1043         if (sgp == SGP_QUICK)
1044                 goto failed;
1045
1046         spin_lock(&info->lock);
1047         shmem_recalc_inode(inode);
1048         entry = shmem_swp_alloc(info, idx, sgp);
1049         if (IS_ERR(entry)) {
1050                 spin_unlock(&info->lock);
1051                 error = PTR_ERR(entry);
1052                 goto failed;
1053         }
1054         swap = *entry;
1055
1056         if (swap.val) {
1057                 /* Look it up and read it in.. */
1058                 swappage = lookup_swap_cache(swap);
1059                 if (!swappage) {
1060                         shmem_swp_unmap(entry);
1061                         /* here we actually do the io */
1062                         if (type && *type == VM_FAULT_MINOR) {
1063                                 __count_vm_event(PGMAJFAULT);
1064                                 *type = VM_FAULT_MAJOR;
1065                         }
1066                         spin_unlock(&info->lock);
1067                         swappage = shmem_swapin(info, swap, idx);
1068                         if (!swappage) {
1069                                 spin_lock(&info->lock);
1070                                 entry = shmem_swp_alloc(info, idx, sgp);
1071                                 if (IS_ERR(entry))
1072                                         error = PTR_ERR(entry);
1073                                 else {
1074                                         if (entry->val == swap.val)
1075                                                 error = -ENOMEM;
1076                                         shmem_swp_unmap(entry);
1077                                 }
1078                                 spin_unlock(&info->lock);
1079                                 if (error)
1080                                         goto failed;
1081                                 goto repeat;
1082                         }
1083                         wait_on_page_locked(swappage);
1084                         page_cache_release(swappage);
1085                         goto repeat;
1086                 }
1087
1088                 /* We have to do this with page locked to prevent races */
1089                 if (TestSetPageLocked(swappage)) {
1090                         shmem_swp_unmap(entry);
1091                         spin_unlock(&info->lock);
1092                         wait_on_page_locked(swappage);
1093                         page_cache_release(swappage);
1094                         goto repeat;
1095                 }
1096                 if (PageWriteback(swappage)) {
1097                         shmem_swp_unmap(entry);
1098                         spin_unlock(&info->lock);
1099                         wait_on_page_writeback(swappage);
1100                         unlock_page(swappage);
1101                         page_cache_release(swappage);
1102                         goto repeat;
1103                 }
1104                 if (!PageUptodate(swappage)) {
1105                         shmem_swp_unmap(entry);
1106                         spin_unlock(&info->lock);
1107                         unlock_page(swappage);
1108                         page_cache_release(swappage);
1109                         error = -EIO;
1110                         goto failed;
1111                 }
1112
1113                 if (filepage) {
1114                         shmem_swp_set(info, entry, 0);
1115                         shmem_swp_unmap(entry);
1116                         delete_from_swap_cache(swappage);
1117                         spin_unlock(&info->lock);
1118                         copy_highpage(filepage, swappage);
1119                         unlock_page(swappage);
1120                         page_cache_release(swappage);
1121                         flush_dcache_page(filepage);
1122                         SetPageUptodate(filepage);
1123                         set_page_dirty(filepage);
1124                         swap_free(swap);
1125                 } else if (!(error = move_from_swap_cache(
1126                                 swappage, idx, mapping))) {
1127                         info->flags |= SHMEM_PAGEIN;
1128                         shmem_swp_set(info, entry, 0);
1129                         shmem_swp_unmap(entry);
1130                         spin_unlock(&info->lock);
1131                         filepage = swappage;
1132                         swap_free(swap);
1133                 } else {
1134                         shmem_swp_unmap(entry);
1135                         spin_unlock(&info->lock);
1136                         unlock_page(swappage);
1137                         page_cache_release(swappage);
1138                         if (error == -ENOMEM) {
1139                                 /* let kswapd refresh zone for GFP_ATOMICs */
1140                                 congestion_wait(WRITE, HZ/50);
1141                         }
1142                         goto repeat;
1143                 }
1144         } else if (sgp == SGP_READ && !filepage) {
1145                 shmem_swp_unmap(entry);
1146                 filepage = find_get_page(mapping, idx);
1147                 if (filepage &&
1148                     (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1149                         spin_unlock(&info->lock);
1150                         wait_on_page_locked(filepage);
1151                         page_cache_release(filepage);
1152                         filepage = NULL;
1153                         goto repeat;
1154                 }
1155                 spin_unlock(&info->lock);
1156         } else {
1157                 shmem_swp_unmap(entry);
1158                 sbinfo = SHMEM_SB(inode->i_sb);
1159                 if (sbinfo->max_blocks) {
1160                         spin_lock(&sbinfo->stat_lock);
1161                         if (sbinfo->free_blocks == 0 ||
1162                             shmem_acct_block(info->flags)) {
1163                                 spin_unlock(&sbinfo->stat_lock);
1164                                 spin_unlock(&info->lock);
1165                                 error = -ENOSPC;
1166                                 goto failed;
1167                         }
1168                         sbinfo->free_blocks--;
1169                         inode->i_blocks += BLOCKS_PER_PAGE;
1170                         spin_unlock(&sbinfo->stat_lock);
1171                 } else if (shmem_acct_block(info->flags)) {
1172                         spin_unlock(&info->lock);
1173                         error = -ENOSPC;
1174                         goto failed;
1175                 }
1176
1177                 if (!filepage) {
1178                         spin_unlock(&info->lock);
1179                         filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
1180                                                     info,
1181                                                     idx);
1182                         if (!filepage) {
1183                                 shmem_unacct_blocks(info->flags, 1);
1184                                 shmem_free_blocks(inode, 1);
1185                                 error = -ENOMEM;
1186                                 goto failed;
1187                         }
1188
1189                         spin_lock(&info->lock);
1190                         entry = shmem_swp_alloc(info, idx, sgp);
1191                         if (IS_ERR(entry))
1192                                 error = PTR_ERR(entry);
1193                         else {
1194                                 swap = *entry;
1195                                 shmem_swp_unmap(entry);
1196                         }
1197                         if (error || swap.val || 0 != add_to_page_cache_lru(
1198                                         filepage, mapping, idx, GFP_ATOMIC)) {
1199                                 spin_unlock(&info->lock);
1200                                 page_cache_release(filepage);
1201                                 shmem_unacct_blocks(info->flags, 1);
1202                                 shmem_free_blocks(inode, 1);
1203                                 filepage = NULL;
1204                                 if (error)
1205                                         goto failed;
1206                                 goto repeat;
1207                         }
1208                         info->flags |= SHMEM_PAGEIN;
1209                 }
1210
1211                 info->alloced++;
1212                 spin_unlock(&info->lock);
1213                 flush_dcache_page(filepage);
1214                 SetPageUptodate(filepage);
1215         }
1216 done:
1217         if (*pagep != filepage) {
1218                 unlock_page(filepage);
1219                 *pagep = filepage;
1220         }
1221         return 0;
1222
1223 failed:
1224         if (*pagep != filepage) {
1225                 unlock_page(filepage);
1226                 page_cache_release(filepage);
1227         }
1228         return error;
1229 }
1230
1231 static struct page *shmem_nopage(struct vm_area_struct *vma,
1232                                  unsigned long address, int *type)
1233 {
1234         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1235         struct page *page = NULL;
1236         unsigned long idx;
1237         int error;
1238
1239         idx = (address - vma->vm_start) >> PAGE_SHIFT;
1240         idx += vma->vm_pgoff;
1241         idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT;
1242         if (((loff_t) idx << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1243                 return NOPAGE_SIGBUS;
1244
1245         error = shmem_getpage(inode, idx, &page, SGP_CACHE, type);
1246         if (error)
1247                 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS;
1248
1249         mark_page_accessed(page);
1250         return page;
1251 }
1252
1253 static int shmem_populate(struct vm_area_struct *vma,
1254         unsigned long addr, unsigned long len,
1255         pgprot_t prot, unsigned long pgoff, int nonblock)
1256 {
1257         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1258         struct mm_struct *mm = vma->vm_mm;
1259         enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE;
1260         unsigned long size;
1261
1262         size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1263         if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size)
1264                 return -EINVAL;
1265
1266         while ((long) len > 0) {
1267                 struct page *page = NULL;
1268                 int err;
1269                 /*
1270                  * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE
1271                  */
1272                 err = shmem_getpage(inode, pgoff, &page, sgp, NULL);
1273                 if (err)
1274                         return err;
1275                 /* Page may still be null, but only if nonblock was set. */
1276                 if (page) {
1277                         mark_page_accessed(page);
1278                         err = install_page(mm, vma, addr, page, prot);
1279                         if (err) {
1280                                 page_cache_release(page);
1281                                 return err;
1282                         }
1283                 } else if (vma->vm_flags & VM_NONLINEAR) {
1284                         /* No page was found just because we can't read it in
1285                          * now (being here implies nonblock != 0), but the page
1286                          * may exist, so set the PTE to fault it in later. */
1287                         err = install_file_pte(mm, vma, addr, pgoff, prot);
1288                         if (err)
1289                                 return err;
1290                 }
1291
1292                 len -= PAGE_SIZE;
1293                 addr += PAGE_SIZE;
1294                 pgoff++;
1295         }
1296         return 0;
1297 }
1298
1299 #ifdef CONFIG_NUMA
1300 int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1301 {
1302         struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1303         return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1304 }
1305
1306 struct mempolicy *
1307 shmem_get_policy(struct vm_area_struct *vma, unsigned long addr)
1308 {
1309         struct inode *i = vma->vm_file->f_path.dentry->d_inode;
1310         unsigned long idx;
1311
1312         idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1313         return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1314 }
1315 #endif
1316
1317 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1318 {
1319         struct inode *inode = file->f_path.dentry->d_inode;
1320         struct shmem_inode_info *info = SHMEM_I(inode);
1321         int retval = -ENOMEM;
1322
1323         spin_lock(&info->lock);
1324         if (lock && !(info->flags & VM_LOCKED)) {
1325                 if (!user_shm_lock(inode->i_size, user))
1326                         goto out_nomem;
1327                 info->flags |= VM_LOCKED;
1328         }
1329         if (!lock && (info->flags & VM_LOCKED) && user) {
1330                 user_shm_unlock(inode->i_size, user);
1331                 info->flags &= ~VM_LOCKED;
1332         }
1333         retval = 0;
1334 out_nomem:
1335         spin_unlock(&info->lock);
1336         return retval;
1337 }
1338
1339 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1340 {
1341         file_accessed(file);
1342         vma->vm_ops = &shmem_vm_ops;
1343         return 0;
1344 }
1345
1346 static struct inode *
1347 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1348 {
1349         struct inode *inode;
1350         struct shmem_inode_info *info;
1351         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1352
1353         if (sbinfo->max_inodes) {
1354                 spin_lock(&sbinfo->stat_lock);
1355                 if (!sbinfo->free_inodes) {
1356                         spin_unlock(&sbinfo->stat_lock);
1357                         return NULL;
1358                 }
1359                 sbinfo->free_inodes--;
1360                 spin_unlock(&sbinfo->stat_lock);
1361         }
1362
1363         inode = new_inode(sb);
1364         if (inode) {
1365                 inode->i_mode = mode;
1366                 inode->i_uid = current->fsuid;
1367                 inode->i_gid = current->fsgid;
1368                 inode->i_blocks = 0;
1369                 inode->i_mapping->a_ops = &shmem_aops;
1370                 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1371                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1372                 inode->i_generation = get_seconds();
1373                 info = SHMEM_I(inode);
1374                 memset(info, 0, (char *)inode - (char *)info);
1375                 spin_lock_init(&info->lock);
1376                 INIT_LIST_HEAD(&info->swaplist);
1377
1378                 switch (mode & S_IFMT) {
1379                 default:
1380                         inode->i_op = &shmem_special_inode_operations;
1381                         init_special_inode(inode, mode, dev);
1382                         break;
1383                 case S_IFREG:
1384                         inode->i_op = &shmem_inode_operations;
1385                         inode->i_fop = &shmem_file_operations;
1386                         mpol_shared_policy_init(&info->policy, sbinfo->policy,
1387                                                         &sbinfo->policy_nodes);
1388                         break;
1389                 case S_IFDIR:
1390                         inc_nlink(inode);
1391                         /* Some things misbehave if size == 0 on a directory */
1392                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
1393                         inode->i_op = &shmem_dir_inode_operations;
1394                         inode->i_fop = &simple_dir_operations;
1395                         break;
1396                 case S_IFLNK:
1397                         /*
1398                          * Must not load anything in the rbtree,
1399                          * mpol_free_shared_policy will not be called.
1400                          */
1401                         mpol_shared_policy_init(&info->policy, MPOL_DEFAULT,
1402                                                 NULL);
1403                         break;
1404                 }
1405         } else if (sbinfo->max_inodes) {
1406                 spin_lock(&sbinfo->stat_lock);
1407                 sbinfo->free_inodes++;
1408                 spin_unlock(&sbinfo->stat_lock);
1409         }
1410         return inode;
1411 }
1412
1413 #ifdef CONFIG_TMPFS
1414 static const struct inode_operations shmem_symlink_inode_operations;
1415 static const struct inode_operations shmem_symlink_inline_operations;
1416
1417 /*
1418  * Normally tmpfs makes no use of shmem_prepare_write, but it
1419  * lets a tmpfs file be used read-write below the loop driver.
1420  */
1421 static int
1422 shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1423 {
1424         struct inode *inode = page->mapping->host;
1425         return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL);
1426 }
1427
1428 static ssize_t
1429 shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1430 {
1431         struct inode    *inode = file->f_path.dentry->d_inode;
1432         loff_t          pos;
1433         unsigned long   written;
1434         ssize_t         err;
1435
1436         if ((ssize_t) count < 0)
1437                 return -EINVAL;
1438
1439         if (!access_ok(VERIFY_READ, buf, count))
1440                 return -EFAULT;
1441
1442         mutex_lock(&inode->i_mutex);
1443
1444         pos = *ppos;
1445         written = 0;
1446
1447         err = generic_write_checks(file, &pos, &count, 0);
1448         if (err || !count)
1449                 goto out;
1450
1451         err = remove_suid(file->f_path.dentry);
1452         if (err)
1453                 goto out;
1454
1455         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1456
1457         do {
1458                 struct page *page = NULL;
1459                 unsigned long bytes, index, offset;
1460                 char *kaddr;
1461                 int left;
1462
1463                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1464                 index = pos >> PAGE_CACHE_SHIFT;
1465                 bytes = PAGE_CACHE_SIZE - offset;
1466                 if (bytes > count)
1467                         bytes = count;
1468
1469                 /*
1470                  * We don't hold page lock across copy from user -
1471                  * what would it guard against? - so no deadlock here.
1472                  * But it still may be a good idea to prefault below.
1473                  */
1474
1475                 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1476                 if (err)
1477                         break;
1478
1479                 left = bytes;
1480                 if (PageHighMem(page)) {
1481                         volatile unsigned char dummy;
1482                         __get_user(dummy, buf);
1483                         __get_user(dummy, buf + bytes - 1);
1484
1485                         kaddr = kmap_atomic(page, KM_USER0);
1486                         left = __copy_from_user_inatomic(kaddr + offset,
1487                                                         buf, bytes);
1488                         kunmap_atomic(kaddr, KM_USER0);
1489                 }
1490                 if (left) {
1491                         kaddr = kmap(page);
1492                         left = __copy_from_user(kaddr + offset, buf, bytes);
1493                         kunmap(page);
1494                 }
1495
1496                 written += bytes;
1497                 count -= bytes;
1498                 pos += bytes;
1499                 buf += bytes;
1500                 if (pos > inode->i_size)
1501                         i_size_write(inode, pos);
1502
1503                 flush_dcache_page(page);
1504                 set_page_dirty(page);
1505                 mark_page_accessed(page);
1506                 page_cache_release(page);
1507
1508                 if (left) {
1509                         pos -= left;
1510                         written -= left;
1511                         err = -EFAULT;
1512                         break;
1513                 }
1514
1515                 /*
1516                  * Our dirty pages are not counted in nr_dirty,
1517                  * and we do not attempt to balance dirty pages.
1518                  */
1519
1520                 cond_resched();
1521         } while (count);
1522
1523         *ppos = pos;
1524         if (written)
1525                 err = written;
1526 out:
1527         mutex_unlock(&inode->i_mutex);
1528         return err;
1529 }
1530
1531 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1532 {
1533         struct inode *inode = filp->f_path.dentry->d_inode;
1534         struct address_space *mapping = inode->i_mapping;
1535         unsigned long index, offset;
1536
1537         index = *ppos >> PAGE_CACHE_SHIFT;
1538         offset = *ppos & ~PAGE_CACHE_MASK;
1539
1540         for (;;) {
1541                 struct page *page = NULL;
1542                 unsigned long end_index, nr, ret;
1543                 loff_t i_size = i_size_read(inode);
1544
1545                 end_index = i_size >> PAGE_CACHE_SHIFT;
1546                 if (index > end_index)
1547                         break;
1548                 if (index == end_index) {
1549                         nr = i_size & ~PAGE_CACHE_MASK;
1550                         if (nr <= offset)
1551                                 break;
1552                 }
1553
1554                 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1555                 if (desc->error) {
1556                         if (desc->error == -EINVAL)
1557                                 desc->error = 0;
1558                         break;
1559                 }
1560
1561                 /*
1562                  * We must evaluate after, since reads (unlike writes)
1563                  * are called without i_mutex protection against truncate
1564                  */
1565                 nr = PAGE_CACHE_SIZE;
1566                 i_size = i_size_read(inode);
1567                 end_index = i_size >> PAGE_CACHE_SHIFT;
1568                 if (index == end_index) {
1569                         nr = i_size & ~PAGE_CACHE_MASK;
1570                         if (nr <= offset) {
1571                                 if (page)
1572                                         page_cache_release(page);
1573                                 break;
1574                         }
1575                 }
1576                 nr -= offset;
1577
1578                 if (page) {
1579                         /*
1580                          * If users can be writing to this page using arbitrary
1581                          * virtual addresses, take care about potential aliasing
1582                          * before reading the page on the kernel side.
1583                          */
1584                         if (mapping_writably_mapped(mapping))
1585                                 flush_dcache_page(page);
1586                         /*
1587                          * Mark the page accessed if we read the beginning.
1588                          */
1589                         if (!offset)
1590                                 mark_page_accessed(page);
1591                 } else {
1592                         page = ZERO_PAGE(0);
1593                         page_cache_get(page);
1594                 }
1595
1596                 /*
1597                  * Ok, we have the page, and it's up-to-date, so
1598                  * now we can copy it to user space...
1599                  *
1600                  * The actor routine returns how many bytes were actually used..
1601                  * NOTE! This may not be the same as how much of a user buffer
1602                  * we filled up (we may be padding etc), so we can only update
1603                  * "pos" here (the actor routine has to update the user buffer
1604                  * pointers and the remaining count).
1605                  */
1606                 ret = actor(desc, page, offset, nr);
1607                 offset += ret;
1608                 index += offset >> PAGE_CACHE_SHIFT;
1609                 offset &= ~PAGE_CACHE_MASK;
1610
1611                 page_cache_release(page);
1612                 if (ret != nr || !desc->count)
1613                         break;
1614
1615                 cond_resched();
1616         }
1617
1618         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1619         file_accessed(filp);
1620 }
1621
1622 static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1623 {
1624         read_descriptor_t desc;
1625
1626         if ((ssize_t) count < 0)
1627                 return -EINVAL;
1628         if (!access_ok(VERIFY_WRITE, buf, count))
1629                 return -EFAULT;
1630         if (!count)
1631                 return 0;
1632
1633         desc.written = 0;
1634         desc.count = count;
1635         desc.arg.buf = buf;
1636         desc.error = 0;
1637
1638         do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1639         if (desc.written)
1640                 return desc.written;
1641         return desc.error;
1642 }
1643
1644 static ssize_t shmem_file_sendfile(struct file *in_file, loff_t *ppos,
1645                          size_t count, read_actor_t actor, void *target)
1646 {
1647         read_descriptor_t desc;
1648
1649         if (!count)
1650                 return 0;
1651
1652         desc.written = 0;
1653         desc.count = count;
1654         desc.arg.data = target;
1655         desc.error = 0;
1656
1657         do_shmem_file_read(in_file, ppos, &desc, actor);
1658         if (desc.written)
1659                 return desc.written;
1660         return desc.error;
1661 }
1662
1663 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1664 {
1665         struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1666
1667         buf->f_type = TMPFS_MAGIC;
1668         buf->f_bsize = PAGE_CACHE_SIZE;
1669         buf->f_namelen = NAME_MAX;
1670         spin_lock(&sbinfo->stat_lock);
1671         if (sbinfo->max_blocks) {
1672                 buf->f_blocks = sbinfo->max_blocks;
1673                 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1674         }
1675         if (sbinfo->max_inodes) {
1676                 buf->f_files = sbinfo->max_inodes;
1677                 buf->f_ffree = sbinfo->free_inodes;
1678         }
1679         /* else leave those fields 0 like simple_statfs */
1680         spin_unlock(&sbinfo->stat_lock);
1681         return 0;
1682 }
1683
1684 /*
1685  * File creation. Allocate an inode, and we're done..
1686  */
1687 static int
1688 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1689 {
1690         struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1691         int error = -ENOSPC;
1692
1693         if (inode) {
1694                 error = security_inode_init_security(inode, dir, NULL, NULL,
1695                                                      NULL);
1696                 if (error) {
1697                         if (error != -EOPNOTSUPP) {
1698                                 iput(inode);
1699                                 return error;
1700                         }
1701                 }
1702                 error = shmem_acl_init(inode, dir);
1703                 if (error) {
1704                         iput(inode);
1705                         return error;
1706                 }
1707                 if (dir->i_mode & S_ISGID) {
1708                         inode->i_gid = dir->i_gid;
1709                         if (S_ISDIR(mode))
1710                                 inode->i_mode |= S_ISGID;
1711                 }
1712                 dir->i_size += BOGO_DIRENT_SIZE;
1713                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1714                 d_instantiate(dentry, inode);
1715                 dget(dentry); /* Extra count - pin the dentry in core */
1716         }
1717         return error;
1718 }
1719
1720 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1721 {
1722         int error;
1723
1724         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1725                 return error;
1726         inc_nlink(dir);
1727         return 0;
1728 }
1729
1730 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1731                 struct nameidata *nd)
1732 {
1733         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1734 }
1735
1736 /*
1737  * Link a file..
1738  */
1739 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1740 {
1741         struct inode *inode = old_dentry->d_inode;
1742         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1743
1744         /*
1745          * No ordinary (disk based) filesystem counts links as inodes;
1746          * but each new link needs a new dentry, pinning lowmem, and
1747          * tmpfs dentries cannot be pruned until they are unlinked.
1748          */
1749         if (sbinfo->max_inodes) {
1750                 spin_lock(&sbinfo->stat_lock);
1751                 if (!sbinfo->free_inodes) {
1752                         spin_unlock(&sbinfo->stat_lock);
1753                         return -ENOSPC;
1754                 }
1755                 sbinfo->free_inodes--;
1756                 spin_unlock(&sbinfo->stat_lock);
1757         }
1758
1759         dir->i_size += BOGO_DIRENT_SIZE;
1760         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1761         inc_nlink(inode);
1762         atomic_inc(&inode->i_count);    /* New dentry reference */
1763         dget(dentry);           /* Extra pinning count for the created dentry */
1764         d_instantiate(dentry, inode);
1765         return 0;
1766 }
1767
1768 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1769 {
1770         struct inode *inode = dentry->d_inode;
1771
1772         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) {
1773                 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1774                 if (sbinfo->max_inodes) {
1775                         spin_lock(&sbinfo->stat_lock);
1776                         sbinfo->free_inodes++;
1777                         spin_unlock(&sbinfo->stat_lock);
1778                 }
1779         }
1780
1781         dir->i_size -= BOGO_DIRENT_SIZE;
1782         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1783         drop_nlink(inode);
1784         dput(dentry);   /* Undo the count from "create" - this does all the work */
1785         return 0;
1786 }
1787
1788 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1789 {
1790         if (!simple_empty(dentry))
1791                 return -ENOTEMPTY;
1792
1793         drop_nlink(dentry->d_inode);
1794         drop_nlink(dir);
1795         return shmem_unlink(dir, dentry);
1796 }
1797
1798 /*
1799  * The VFS layer already does all the dentry stuff for rename,
1800  * we just have to decrement the usage count for the target if
1801  * it exists so that the VFS layer correctly free's it when it
1802  * gets overwritten.
1803  */
1804 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1805 {
1806         struct inode *inode = old_dentry->d_inode;
1807         int they_are_dirs = S_ISDIR(inode->i_mode);
1808
1809         if (!simple_empty(new_dentry))
1810                 return -ENOTEMPTY;
1811
1812         if (new_dentry->d_inode) {
1813                 (void) shmem_unlink(new_dir, new_dentry);
1814                 if (they_are_dirs)
1815                         drop_nlink(old_dir);
1816         } else if (they_are_dirs) {
1817                 drop_nlink(old_dir);
1818                 inc_nlink(new_dir);
1819         }
1820
1821         old_dir->i_size -= BOGO_DIRENT_SIZE;
1822         new_dir->i_size += BOGO_DIRENT_SIZE;
1823         old_dir->i_ctime = old_dir->i_mtime =
1824         new_dir->i_ctime = new_dir->i_mtime =
1825         inode->i_ctime = CURRENT_TIME;
1826         return 0;
1827 }
1828
1829 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1830 {
1831         int error;
1832         int len;
1833         struct inode *inode;
1834         struct page *page = NULL;
1835         char *kaddr;
1836         struct shmem_inode_info *info;
1837
1838         len = strlen(symname) + 1;
1839         if (len > PAGE_CACHE_SIZE)
1840                 return -ENAMETOOLONG;
1841
1842         inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1843         if (!inode)
1844                 return -ENOSPC;
1845
1846         error = security_inode_init_security(inode, dir, NULL, NULL,
1847                                              NULL);
1848         if (error) {
1849                 if (error != -EOPNOTSUPP) {
1850                         iput(inode);
1851                         return error;
1852                 }
1853                 error = 0;
1854         }
1855
1856         info = SHMEM_I(inode);
1857         inode->i_size = len-1;
1858         if (len <= (char *)inode - (char *)info) {
1859                 /* do it inline */
1860                 memcpy(info, symname, len);
1861                 inode->i_op = &shmem_symlink_inline_operations;
1862         } else {
1863                 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1864                 if (error) {
1865                         iput(inode);
1866                         return error;
1867                 }
1868                 inode->i_op = &shmem_symlink_inode_operations;
1869                 kaddr = kmap_atomic(page, KM_USER0);
1870                 memcpy(kaddr, symname, len);
1871                 kunmap_atomic(kaddr, KM_USER0);
1872                 set_page_dirty(page);
1873                 page_cache_release(page);
1874         }
1875         if (dir->i_mode & S_ISGID)
1876                 inode->i_gid = dir->i_gid;
1877         dir->i_size += BOGO_DIRENT_SIZE;
1878         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1879         d_instantiate(dentry, inode);
1880         dget(dentry);
1881         return 0;
1882 }
1883
1884 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1885 {
1886         nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1887         return NULL;
1888 }
1889
1890 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1891 {
1892         struct page *page = NULL;
1893         int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1894         nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1895         return page;
1896 }
1897
1898 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1899 {
1900         if (!IS_ERR(nd_get_link(nd))) {
1901                 struct page *page = cookie;
1902                 kunmap(page);
1903                 mark_page_accessed(page);
1904                 page_cache_release(page);
1905         }
1906 }
1907
1908 static const struct inode_operations shmem_symlink_inline_operations = {
1909         .readlink       = generic_readlink,
1910         .follow_link    = shmem_follow_link_inline,
1911 };
1912
1913 static const struct inode_operations shmem_symlink_inode_operations = {
1914         .truncate       = shmem_truncate,
1915         .readlink       = generic_readlink,
1916         .follow_link    = shmem_follow_link,
1917         .put_link       = shmem_put_link,
1918 };
1919
1920 #ifdef CONFIG_TMPFS_POSIX_ACL
1921 /**
1922  * Superblocks without xattr inode operations will get security.* xattr
1923  * support from the VFS "for free". As soon as we have any other xattrs
1924  * like ACLs, we also need to implement the security.* handlers at
1925  * filesystem level, though.
1926  */
1927
1928 static size_t shmem_xattr_security_list(struct inode *inode, char *list,
1929                                         size_t list_len, const char *name,
1930                                         size_t name_len)
1931 {
1932         return security_inode_listsecurity(inode, list, list_len);
1933 }
1934
1935 static int shmem_xattr_security_get(struct inode *inode, const char *name,
1936                                     void *buffer, size_t size)
1937 {
1938         if (strcmp(name, "") == 0)
1939                 return -EINVAL;
1940         return security_inode_getsecurity(inode, name, buffer, size,
1941                                           -EOPNOTSUPP);
1942 }
1943
1944 static int shmem_xattr_security_set(struct inode *inode, const char *name,
1945                                     const void *value, size_t size, int flags)
1946 {
1947         if (strcmp(name, "") == 0)
1948                 return -EINVAL;
1949         return security_inode_setsecurity(inode, name, value, size, flags);
1950 }
1951
1952 static struct xattr_handler shmem_xattr_security_handler = {
1953         .prefix = XATTR_SECURITY_PREFIX,
1954         .list   = shmem_xattr_security_list,
1955         .get    = shmem_xattr_security_get,
1956         .set    = shmem_xattr_security_set,
1957 };
1958
1959 static struct xattr_handler *shmem_xattr_handlers[] = {
1960         &shmem_xattr_acl_access_handler,
1961         &shmem_xattr_acl_default_handler,
1962         &shmem_xattr_security_handler,
1963         NULL
1964 };
1965 #endif
1966
1967 static struct dentry *shmem_get_parent(struct dentry *child)
1968 {
1969         return ERR_PTR(-ESTALE);
1970 }
1971
1972 static int shmem_match(struct inode *ino, void *vfh)
1973 {
1974         __u32 *fh = vfh;
1975         __u64 inum = fh[2];
1976         inum = (inum << 32) | fh[1];
1977         return ino->i_ino == inum && fh[0] == ino->i_generation;
1978 }
1979
1980 static struct dentry *shmem_get_dentry(struct super_block *sb, void *vfh)
1981 {
1982         struct dentry *de = NULL;
1983         struct inode *inode;
1984         __u32 *fh = vfh;
1985         __u64 inum = fh[2];
1986         inum = (inum << 32) | fh[1];
1987
1988         inode = ilookup5(sb, (unsigned long)(inum+fh[0]), shmem_match, vfh);
1989         if (inode) {
1990                 de = d_find_alias(inode);
1991                 iput(inode);
1992         }
1993
1994         return de? de: ERR_PTR(-ESTALE);
1995 }
1996
1997 static struct dentry *shmem_decode_fh(struct super_block *sb, __u32 *fh,
1998                 int len, int type,
1999                 int (*acceptable)(void *context, struct dentry *de),
2000                 void *context)
2001 {
2002         if (len < 3)
2003                 return ERR_PTR(-ESTALE);
2004
2005         return sb->s_export_op->find_exported_dentry(sb, fh, NULL, acceptable,
2006                                                         context);
2007 }
2008
2009 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2010                                 int connectable)
2011 {
2012         struct inode *inode = dentry->d_inode;
2013
2014         if (*len < 3)
2015                 return 255;
2016
2017         if (hlist_unhashed(&inode->i_hash)) {
2018                 /* Unfortunately insert_inode_hash is not idempotent,
2019                  * so as we hash inodes here rather than at creation
2020                  * time, we need a lock to ensure we only try
2021                  * to do it once
2022                  */
2023                 static DEFINE_SPINLOCK(lock);
2024                 spin_lock(&lock);
2025                 if (hlist_unhashed(&inode->i_hash))
2026                         __insert_inode_hash(inode,
2027                                             inode->i_ino + inode->i_generation);
2028                 spin_unlock(&lock);
2029         }
2030
2031         fh[0] = inode->i_generation;
2032         fh[1] = inode->i_ino;
2033         fh[2] = ((__u64)inode->i_ino) >> 32;
2034
2035         *len = 3;
2036         return 1;
2037 }
2038
2039 static struct export_operations shmem_export_ops = {
2040         .get_parent     = shmem_get_parent,
2041         .get_dentry     = shmem_get_dentry,
2042         .encode_fh      = shmem_encode_fh,
2043         .decode_fh      = shmem_decode_fh,
2044 };
2045
2046 static int shmem_parse_options(char *options, int *mode, uid_t *uid,
2047         gid_t *gid, unsigned long *blocks, unsigned long *inodes,
2048         int *policy, nodemask_t *policy_nodes)
2049 {
2050         char *this_char, *value, *rest;
2051
2052         while (options != NULL) {
2053                 this_char = options;
2054                 for (;;) {
2055                         /*
2056                          * NUL-terminate this option: unfortunately,
2057                          * mount options form a comma-separated list,
2058                          * but mpol's nodelist may also contain commas.
2059                          */
2060                         options = strchr(options, ',');
2061                         if (options == NULL)
2062                                 break;
2063                         options++;
2064                         if (!isdigit(*options)) {
2065                                 options[-1] = '\0';
2066                                 break;
2067                         }
2068                 }
2069                 if (!*this_char)
2070                         continue;
2071                 if ((value = strchr(this_char,'=')) != NULL) {
2072                         *value++ = 0;
2073                 } else {
2074                         printk(KERN_ERR
2075                             "tmpfs: No value for mount option '%s'\n",
2076                             this_char);
2077                         return 1;
2078                 }
2079
2080                 if (!strcmp(this_char,"size")) {
2081                         unsigned long long size;
2082                         size = memparse(value,&rest);
2083                         if (*rest == '%') {
2084                                 size <<= PAGE_SHIFT;
2085                                 size *= totalram_pages;
2086                                 do_div(size, 100);
2087                                 rest++;
2088                         }
2089                         if (*rest)
2090                                 goto bad_val;
2091                         *blocks = size >> PAGE_CACHE_SHIFT;
2092                 } else if (!strcmp(this_char,"nr_blocks")) {
2093                         *blocks = memparse(value,&rest);
2094                         if (*rest)
2095                                 goto bad_val;
2096                 } else if (!strcmp(this_char,"nr_inodes")) {
2097                         *inodes = memparse(value,&rest);
2098                         if (*rest)
2099                                 goto bad_val;
2100                 } else if (!strcmp(this_char,"mode")) {
2101                         if (!mode)
2102                                 continue;
2103                         *mode = simple_strtoul(value,&rest,8);
2104                         if (*rest)
2105                                 goto bad_val;
2106                 } else if (!strcmp(this_char,"uid")) {
2107                         if (!uid)
2108                                 continue;
2109                         *uid = simple_strtoul(value,&rest,0);
2110                         if (*rest)
2111                                 goto bad_val;
2112                 } else if (!strcmp(this_char,"gid")) {
2113                         if (!gid)
2114                                 continue;
2115                         *gid = simple_strtoul(value,&rest,0);
2116                         if (*rest)
2117                                 goto bad_val;
2118                 } else if (!strcmp(this_char,"mpol")) {
2119                         if (shmem_parse_mpol(value,policy,policy_nodes))
2120                                 goto bad_val;
2121                 } else {
2122                         printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2123                                this_char);
2124                         return 1;
2125                 }
2126         }
2127         return 0;
2128
2129 bad_val:
2130         printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2131                value, this_char);
2132         return 1;
2133
2134 }
2135
2136 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2137 {
2138         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2139         unsigned long max_blocks = sbinfo->max_blocks;
2140         unsigned long max_inodes = sbinfo->max_inodes;
2141         int policy = sbinfo->policy;
2142         nodemask_t policy_nodes = sbinfo->policy_nodes;
2143         unsigned long blocks;
2144         unsigned long inodes;
2145         int error = -EINVAL;
2146
2147         if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks,
2148                                 &max_inodes, &policy, &policy_nodes))
2149                 return error;
2150
2151         spin_lock(&sbinfo->stat_lock);
2152         blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2153         inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2154         if (max_blocks < blocks)
2155                 goto out;
2156         if (max_inodes < inodes)
2157                 goto out;
2158         /*
2159          * Those tests also disallow limited->unlimited while any are in
2160          * use, so i_blocks will always be zero when max_blocks is zero;
2161          * but we must separately disallow unlimited->limited, because
2162          * in that case we have no record of how much is already in use.
2163          */
2164         if (max_blocks && !sbinfo->max_blocks)
2165                 goto out;
2166         if (max_inodes && !sbinfo->max_inodes)
2167                 goto out;
2168
2169         error = 0;
2170         sbinfo->max_blocks  = max_blocks;
2171         sbinfo->free_blocks = max_blocks - blocks;
2172         sbinfo->max_inodes  = max_inodes;
2173         sbinfo->free_inodes = max_inodes - inodes;
2174         sbinfo->policy = policy;
2175         sbinfo->policy_nodes = policy_nodes;
2176 out:
2177         spin_unlock(&sbinfo->stat_lock);
2178         return error;
2179 }
2180 #endif
2181
2182 static void shmem_put_super(struct super_block *sb)
2183 {
2184         kfree(sb->s_fs_info);
2185         sb->s_fs_info = NULL;
2186 }
2187
2188 static int shmem_fill_super(struct super_block *sb,
2189                             void *data, int silent)
2190 {
2191         struct inode *inode;
2192         struct dentry *root;
2193         int mode   = S_IRWXUGO | S_ISVTX;
2194         uid_t uid = current->fsuid;
2195         gid_t gid = current->fsgid;
2196         int err = -ENOMEM;
2197         struct shmem_sb_info *sbinfo;
2198         unsigned long blocks = 0;
2199         unsigned long inodes = 0;
2200         int policy = MPOL_DEFAULT;
2201         nodemask_t policy_nodes = node_online_map;
2202
2203 #ifdef CONFIG_TMPFS
2204         /*
2205          * Per default we only allow half of the physical ram per
2206          * tmpfs instance, limiting inodes to one per page of lowmem;
2207          * but the internal instance is left unlimited.
2208          */
2209         if (!(sb->s_flags & MS_NOUSER)) {
2210                 blocks = totalram_pages / 2;
2211                 inodes = totalram_pages - totalhigh_pages;
2212                 if (inodes > blocks)
2213                         inodes = blocks;
2214                 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks,
2215                                         &inodes, &policy, &policy_nodes))
2216                         return -EINVAL;
2217         }
2218         sb->s_export_op = &shmem_export_ops;
2219 #else
2220         sb->s_flags |= MS_NOUSER;
2221 #endif
2222
2223         /* Round up to L1_CACHE_BYTES to resist false sharing */
2224         sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
2225                                 L1_CACHE_BYTES), GFP_KERNEL);
2226         if (!sbinfo)
2227                 return -ENOMEM;
2228
2229         spin_lock_init(&sbinfo->stat_lock);
2230         sbinfo->max_blocks = blocks;
2231         sbinfo->free_blocks = blocks;
2232         sbinfo->max_inodes = inodes;
2233         sbinfo->free_inodes = inodes;
2234         sbinfo->policy = policy;
2235         sbinfo->policy_nodes = policy_nodes;
2236
2237         sb->s_fs_info = sbinfo;
2238         sb->s_maxbytes = SHMEM_MAX_BYTES;
2239         sb->s_blocksize = PAGE_CACHE_SIZE;
2240         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2241         sb->s_magic = TMPFS_MAGIC;
2242         sb->s_op = &shmem_ops;
2243         sb->s_time_gran = 1;
2244 #ifdef CONFIG_TMPFS_POSIX_ACL
2245         sb->s_xattr = shmem_xattr_handlers;
2246         sb->s_flags |= MS_POSIXACL;
2247 #endif
2248
2249         inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
2250         if (!inode)
2251                 goto failed;
2252         inode->i_uid = uid;
2253         inode->i_gid = gid;
2254         root = d_alloc_root(inode);
2255         if (!root)
2256                 goto failed_iput;
2257         sb->s_root = root;
2258         return 0;
2259
2260 failed_iput:
2261         iput(inode);
2262 failed:
2263         shmem_put_super(sb);
2264         return err;
2265 }
2266
2267 static struct kmem_cache *shmem_inode_cachep;
2268
2269 static struct inode *shmem_alloc_inode(struct super_block *sb)
2270 {
2271         struct shmem_inode_info *p;
2272         p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2273         if (!p)
2274                 return NULL;
2275         return &p->vfs_inode;
2276 }
2277
2278 static void shmem_destroy_inode(struct inode *inode)
2279 {
2280         if ((inode->i_mode & S_IFMT) == S_IFREG) {
2281                 /* only struct inode is valid if it's an inline symlink */
2282                 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2283         }
2284         shmem_acl_destroy_inode(inode);
2285         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2286 }
2287
2288 static void init_once(void *foo, struct kmem_cache *cachep,
2289                       unsigned long flags)
2290 {
2291         struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2292
2293         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2294             SLAB_CTOR_CONSTRUCTOR) {
2295                 inode_init_once(&p->vfs_inode);
2296 #ifdef CONFIG_TMPFS_POSIX_ACL
2297                 p->i_acl = NULL;
2298                 p->i_default_acl = NULL;
2299 #endif
2300         }
2301 }
2302
2303 static int init_inodecache(void)
2304 {
2305         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2306                                 sizeof(struct shmem_inode_info),
2307                                 0, 0, init_once, NULL);
2308         if (shmem_inode_cachep == NULL)
2309                 return -ENOMEM;
2310         return 0;
2311 }
2312
2313 static void destroy_inodecache(void)
2314 {
2315         kmem_cache_destroy(shmem_inode_cachep);
2316 }
2317
2318 static const struct address_space_operations shmem_aops = {
2319         .writepage      = shmem_writepage,
2320         .set_page_dirty = __set_page_dirty_no_writeback,
2321 #ifdef CONFIG_TMPFS
2322         .prepare_write  = shmem_prepare_write,
2323         .commit_write   = simple_commit_write,
2324 #endif
2325         .migratepage    = migrate_page,
2326 };
2327
2328 static const struct file_operations shmem_file_operations = {
2329         .mmap           = shmem_mmap,
2330 #ifdef CONFIG_TMPFS
2331         .llseek         = generic_file_llseek,
2332         .read           = shmem_file_read,
2333         .write          = shmem_file_write,
2334         .fsync          = simple_sync_file,
2335         .sendfile       = shmem_file_sendfile,
2336 #endif
2337 };
2338
2339 static const struct inode_operations shmem_inode_operations = {
2340         .truncate       = shmem_truncate,
2341         .setattr        = shmem_notify_change,
2342         .truncate_range = shmem_truncate_range,
2343 #ifdef CONFIG_TMPFS_POSIX_ACL
2344         .setxattr       = generic_setxattr,
2345         .getxattr       = generic_getxattr,
2346         .listxattr      = generic_listxattr,
2347         .removexattr    = generic_removexattr,
2348         .permission     = shmem_permission,
2349 #endif
2350
2351 };
2352
2353 static const struct inode_operations shmem_dir_inode_operations = {
2354 #ifdef CONFIG_TMPFS
2355         .create         = shmem_create,
2356         .lookup         = simple_lookup,
2357         .link           = shmem_link,
2358         .unlink         = shmem_unlink,
2359         .symlink        = shmem_symlink,
2360         .mkdir          = shmem_mkdir,
2361         .rmdir          = shmem_rmdir,
2362         .mknod          = shmem_mknod,
2363         .rename         = shmem_rename,
2364 #endif
2365 #ifdef CONFIG_TMPFS_POSIX_ACL
2366         .setattr        = shmem_notify_change,
2367         .setxattr       = generic_setxattr,
2368         .getxattr       = generic_getxattr,
2369         .listxattr      = generic_listxattr,
2370         .removexattr    = generic_removexattr,
2371         .permission     = shmem_permission,
2372 #endif
2373 };
2374
2375 static const struct inode_operations shmem_special_inode_operations = {
2376 #ifdef CONFIG_TMPFS_POSIX_ACL
2377         .setattr        = shmem_notify_change,
2378         .setxattr       = generic_setxattr,
2379         .getxattr       = generic_getxattr,
2380         .listxattr      = generic_listxattr,
2381         .removexattr    = generic_removexattr,
2382         .permission     = shmem_permission,
2383 #endif
2384 };
2385
2386 static const struct super_operations shmem_ops = {
2387         .alloc_inode    = shmem_alloc_inode,
2388         .destroy_inode  = shmem_destroy_inode,
2389 #ifdef CONFIG_TMPFS
2390         .statfs         = shmem_statfs,
2391         .remount_fs     = shmem_remount_fs,
2392 #endif
2393         .delete_inode   = shmem_delete_inode,
2394         .drop_inode     = generic_delete_inode,
2395         .put_super      = shmem_put_super,
2396 };
2397
2398 static struct vm_operations_struct shmem_vm_ops = {
2399         .nopage         = shmem_nopage,
2400         .populate       = shmem_populate,
2401 #ifdef CONFIG_NUMA
2402         .set_policy     = shmem_set_policy,
2403         .get_policy     = shmem_get_policy,
2404 #endif
2405 };
2406
2407
2408 static int shmem_get_sb(struct file_system_type *fs_type,
2409         int flags, const char *dev_name, void *data, struct vfsmount *mnt)
2410 {
2411         return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
2412 }
2413
2414 static struct file_system_type tmpfs_fs_type = {
2415         .owner          = THIS_MODULE,
2416         .name           = "tmpfs",
2417         .get_sb         = shmem_get_sb,
2418         .kill_sb        = kill_litter_super,
2419 };
2420 static struct vfsmount *shm_mnt;
2421
2422 static int __init init_tmpfs(void)
2423 {
2424         int error;
2425
2426         error = init_inodecache();
2427         if (error)
2428                 goto out3;
2429
2430         error = register_filesystem(&tmpfs_fs_type);
2431         if (error) {
2432                 printk(KERN_ERR "Could not register tmpfs\n");
2433                 goto out2;
2434         }
2435
2436         shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
2437                                 tmpfs_fs_type.name, NULL);
2438         if (IS_ERR(shm_mnt)) {
2439                 error = PTR_ERR(shm_mnt);
2440                 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2441                 goto out1;
2442         }
2443         return 0;
2444
2445 out1:
2446         unregister_filesystem(&tmpfs_fs_type);
2447 out2:
2448         destroy_inodecache();
2449 out3:
2450         shm_mnt = ERR_PTR(error);
2451         return error;
2452 }
2453 module_init(init_tmpfs)
2454
2455 /*
2456  * shmem_file_setup - get an unlinked file living in tmpfs
2457  *
2458  * @name: name for dentry (to be seen in /proc/<pid>/maps
2459  * @size: size to be set for the file
2460  *
2461  */
2462 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2463 {
2464         int error;
2465         struct file *file;
2466         struct inode *inode;
2467         struct dentry *dentry, *root;
2468         struct qstr this;
2469
2470         if (IS_ERR(shm_mnt))
2471                 return (void *)shm_mnt;
2472
2473         if (size < 0 || size > SHMEM_MAX_BYTES)
2474                 return ERR_PTR(-EINVAL);
2475
2476         if (shmem_acct_size(flags, size))
2477                 return ERR_PTR(-ENOMEM);
2478
2479         error = -ENOMEM;
2480         this.name = name;
2481         this.len = strlen(name);
2482         this.hash = 0; /* will go */
2483         root = shm_mnt->mnt_root;
2484         dentry = d_alloc(root, &this);
2485         if (!dentry)
2486                 goto put_memory;
2487
2488         error = -ENFILE;
2489         file = get_empty_filp();
2490         if (!file)
2491                 goto put_dentry;
2492
2493         error = -ENOSPC;
2494         inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2495         if (!inode)
2496                 goto close_file;
2497
2498         SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2499         d_instantiate(dentry, inode);
2500         inode->i_size = size;
2501         inode->i_nlink = 0;     /* It is unlinked */
2502         file->f_path.mnt = mntget(shm_mnt);
2503         file->f_path.dentry = dentry;
2504         file->f_mapping = inode->i_mapping;
2505         file->f_op = &shmem_file_operations;
2506         file->f_mode = FMODE_WRITE | FMODE_READ;
2507         return file;
2508
2509 close_file:
2510         put_filp(file);
2511 put_dentry:
2512         dput(dentry);
2513 put_memory:
2514         shmem_unacct_size(flags, size);
2515         return ERR_PTR(error);
2516 }
2517
2518 /*
2519  * shmem_zero_setup - setup a shared anonymous mapping
2520  *
2521  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2522  */
2523 int shmem_zero_setup(struct vm_area_struct *vma)
2524 {
2525         struct file *file;
2526         loff_t size = vma->vm_end - vma->vm_start;
2527
2528         file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2529         if (IS_ERR(file))
2530                 return PTR_ERR(file);
2531
2532         if (vma->vm_file)
2533                 fput(vma->vm_file);
2534         vma->vm_file = file;
2535         vma->vm_ops = &shmem_vm_ops;
2536         return 0;
2537 }