linux/mm/shmem.c
<<
>>
Prefs
   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) 2002-2011 Hugh Dickins.
  10 * Copyright (C) 2011 Google Inc.
  11 * Copyright (C) 2002-2005 VERITAS Software Corporation.
  12 * Copyright (C) 2004 Andi Kleen, SuSE Labs
  13 *
  14 * Extended attribute support for tmpfs:
  15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
  16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
  17 *
  18 * tiny-shmem:
  19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
  20 *
  21 * This file is released under the GPL.
  22 */
  23
  24#include <linux/fs.h>
  25#include <linux/init.h>
  26#include <linux/vfs.h>
  27#include <linux/mount.h>
  28#include <linux/pagemap.h>
  29#include <linux/file.h>
  30#include <linux/mm.h>
  31#include <linux/export.h>
  32#include <linux/swap.h>
  33
  34static struct vfsmount *shm_mnt;
  35
  36#ifdef CONFIG_SHMEM
  37/*
  38 * This virtual memory filesystem is heavily based on the ramfs. It
  39 * extends ramfs by the ability to use swap and honor resource limits
  40 * which makes it a completely usable filesystem.
  41 */
  42
  43#include <linux/xattr.h>
  44#include <linux/exportfs.h>
  45#include <linux/posix_acl.h>
  46#include <linux/generic_acl.h>
  47#include <linux/mman.h>
  48#include <linux/string.h>
  49#include <linux/slab.h>
  50#include <linux/backing-dev.h>
  51#include <linux/shmem_fs.h>
  52#include <linux/writeback.h>
  53#include <linux/blkdev.h>
  54#include <linux/pagevec.h>
  55#include <linux/percpu_counter.h>
  56#include <linux/falloc.h>
  57#include <linux/splice.h>
  58#include <linux/security.h>
  59#include <linux/swapops.h>
  60#include <linux/mempolicy.h>
  61#include <linux/namei.h>
  62#include <linux/ctype.h>
  63#include <linux/migrate.h>
  64#include <linux/highmem.h>
  65#include <linux/seq_file.h>
  66#include <linux/magic.h>
  67
  68#include <asm/uaccess.h>
  69#include <asm/pgtable.h>
  70
  71#define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
  72#define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
  73
  74/* Pretend that each entry is of this size in directory's i_size */
  75#define BOGO_DIRENT_SIZE 20
  76
  77/* Symlink up to this size is kmalloc'ed instead of using a swappable page */
  78#define SHORT_SYMLINK_LEN 128
  79
  80struct shmem_xattr {
  81        struct list_head list;  /* anchored by shmem_inode_info->xattr_list */
  82        char *name;             /* xattr name */
  83        size_t size;
  84        char value[0];
  85};
  86
  87/*
  88 * shmem_fallocate and shmem_writepage communicate via inode->i_private
  89 * (with i_mutex making sure that it has only one user at a time):
  90 * we would prefer not to enlarge the shmem inode just for that.
  91 */
  92struct shmem_falloc {
  93        pgoff_t start;          /* start of range currently being fallocated */
  94        pgoff_t next;           /* the next page offset to be fallocated */
  95        pgoff_t nr_falloced;    /* how many new pages have been fallocated */
  96        pgoff_t nr_unswapped;   /* how often writepage refused to swap out */
  97};
  98
  99/* Flag allocation requirements to shmem_getpage */
 100enum sgp_type {
 101        SGP_READ,       /* don't exceed i_size, don't allocate page */
 102        SGP_CACHE,      /* don't exceed i_size, may allocate page */
 103        SGP_DIRTY,      /* like SGP_CACHE, but set new page dirty */
 104        SGP_WRITE,      /* may exceed i_size, may allocate !Uptodate page */
 105        SGP_FALLOC,     /* like SGP_WRITE, but make existing page Uptodate */
 106};
 107
 108#ifdef CONFIG_TMPFS
 109static unsigned long shmem_default_max_blocks(void)
 110{
 111        return totalram_pages / 2;
 112}
 113
 114static unsigned long shmem_default_max_inodes(void)
 115{
 116        return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
 117}
 118#endif
 119
 120static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
 121static int shmem_replace_page(struct page **pagep, gfp_t gfp,
 122                                struct shmem_inode_info *info, pgoff_t index);
 123static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
 124        struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
 125
 126static inline int shmem_getpage(struct inode *inode, pgoff_t index,
 127        struct page **pagep, enum sgp_type sgp, int *fault_type)
 128{
 129        return shmem_getpage_gfp(inode, index, pagep, sgp,
 130                        mapping_gfp_mask(inode->i_mapping), fault_type);
 131}
 132
 133static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
 134{
 135        return sb->s_fs_info;
 136}
 137
 138/*
 139 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
 140 * for shared memory and for shared anonymous (/dev/zero) mappings
 141 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
 142 * consistent with the pre-accounting of private mappings ...
 143 */
 144static inline int shmem_acct_size(unsigned long flags, loff_t size)
 145{
 146        return (flags & VM_NORESERVE) ?
 147                0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
 148}
 149
 150static inline void shmem_unacct_size(unsigned long flags, loff_t size)
 151{
 152        if (!(flags & VM_NORESERVE))
 153                vm_unacct_memory(VM_ACCT(size));
 154}
 155
 156/*
 157 * ... whereas tmpfs objects are accounted incrementally as
 158 * pages are allocated, in order to allow huge sparse files.
 159 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
 160 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
 161 */
 162static inline int shmem_acct_block(unsigned long flags)
 163{
 164        return (flags & VM_NORESERVE) ?
 165                security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0;
 166}
 167
 168static inline void shmem_unacct_blocks(unsigned long flags, long pages)
 169{
 170        if (flags & VM_NORESERVE)
 171                vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
 172}
 173
 174static const struct super_operations shmem_ops;
 175static const struct address_space_operations shmem_aops;
 176static const struct file_operations shmem_file_operations;
 177static const struct inode_operations shmem_inode_operations;
 178static const struct inode_operations shmem_dir_inode_operations;
 179static const struct inode_operations shmem_special_inode_operations;
 180static const struct vm_operations_struct shmem_vm_ops;
 181
 182static struct backing_dev_info shmem_backing_dev_info  __read_mostly = {
 183        .ra_pages       = 0,    /* No readahead */
 184        .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
 185};
 186
 187static LIST_HEAD(shmem_swaplist);
 188static DEFINE_MUTEX(shmem_swaplist_mutex);
 189
 190static int shmem_reserve_inode(struct super_block *sb)
 191{
 192        struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 193        if (sbinfo->max_inodes) {
 194                spin_lock(&sbinfo->stat_lock);
 195                if (!sbinfo->free_inodes) {
 196                        spin_unlock(&sbinfo->stat_lock);
 197                        return -ENOSPC;
 198                }
 199                sbinfo->free_inodes--;
 200                spin_unlock(&sbinfo->stat_lock);
 201        }
 202        return 0;
 203}
 204
 205static void shmem_free_inode(struct super_block *sb)
 206{
 207        struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 208        if (sbinfo->max_inodes) {
 209                spin_lock(&sbinfo->stat_lock);
 210                sbinfo->free_inodes++;
 211                spin_unlock(&sbinfo->stat_lock);
 212        }
 213}
 214
 215/**
 216 * shmem_recalc_inode - recalculate the block usage of an inode
 217 * @inode: inode to recalc
 218 *
 219 * We have to calculate the free blocks since the mm can drop
 220 * undirtied hole pages behind our back.
 221 *
 222 * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
 223 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
 224 *
 225 * It has to be called with the spinlock held.
 226 */
 227static void shmem_recalc_inode(struct inode *inode)
 228{
 229        struct shmem_inode_info *info = SHMEM_I(inode);
 230        long freed;
 231
 232        freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
 233        if (freed > 0) {
 234                struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 235                if (sbinfo->max_blocks)
 236                        percpu_counter_add(&sbinfo->used_blocks, -freed);
 237                info->alloced -= freed;
 238                inode->i_blocks -= freed * BLOCKS_PER_PAGE;
 239                shmem_unacct_blocks(info->flags, freed);
 240        }
 241}
 242
 243/*
 244 * Replace item expected in radix tree by a new item, while holding tree lock.
 245 */
 246static int shmem_radix_tree_replace(struct address_space *mapping,
 247                        pgoff_t index, void *expected, void *replacement)
 248{
 249        void **pslot;
 250        void *item = NULL;
 251
 252        VM_BUG_ON(!expected);
 253        pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
 254        if (pslot)
 255                item = radix_tree_deref_slot_protected(pslot,
 256                                                        &mapping->tree_lock);
 257        if (item != expected)
 258                return -ENOENT;
 259        if (replacement)
 260                radix_tree_replace_slot(pslot, replacement);
 261        else
 262                radix_tree_delete(&mapping->page_tree, index);
 263        return 0;
 264}
 265
 266/*
 267 * Sometimes, before we decide whether to proceed or to fail, we must check
 268 * that an entry was not already brought back from swap by a racing thread.
 269 *
 270 * Checking page is not enough: by the time a SwapCache page is locked, it
 271 * might be reused, and again be SwapCache, using the same swap as before.
 272 */
 273static bool shmem_confirm_swap(struct address_space *mapping,
 274                               pgoff_t index, swp_entry_t swap)
 275{
 276        void *item;
 277
 278        rcu_read_lock();
 279        item = radix_tree_lookup(&mapping->page_tree, index);
 280        rcu_read_unlock();
 281        return item == swp_to_radix_entry(swap);
 282}
 283
 284/*
 285 * Like add_to_page_cache_locked, but error if expected item has gone.
 286 */
 287static int shmem_add_to_page_cache(struct page *page,
 288                                   struct address_space *mapping,
 289                                   pgoff_t index, gfp_t gfp, void *expected)
 290{
 291        int error;
 292
 293        VM_BUG_ON(!PageLocked(page));
 294        VM_BUG_ON(!PageSwapBacked(page));
 295
 296        page_cache_get(page);
 297        page->mapping = mapping;
 298        page->index = index;
 299
 300        spin_lock_irq(&mapping->tree_lock);
 301        if (!expected)
 302                error = radix_tree_insert(&mapping->page_tree, index, page);
 303        else
 304                error = shmem_radix_tree_replace(mapping, index, expected,
 305                                                                 page);
 306        if (!error) {
 307                mapping->nrpages++;
 308                __inc_zone_page_state(page, NR_FILE_PAGES);
 309                __inc_zone_page_state(page, NR_SHMEM);
 310                spin_unlock_irq(&mapping->tree_lock);
 311        } else {
 312                page->mapping = NULL;
 313                spin_unlock_irq(&mapping->tree_lock);
 314                page_cache_release(page);
 315        }
 316        return error;
 317}
 318
 319/*
 320 * Like delete_from_page_cache, but substitutes swap for page.
 321 */
 322static void shmem_delete_from_page_cache(struct page *page, void *radswap)
 323{
 324        struct address_space *mapping = page->mapping;
 325        int error;
 326
 327        spin_lock_irq(&mapping->tree_lock);
 328        error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
 329        page->mapping = NULL;
 330        mapping->nrpages--;
 331        __dec_zone_page_state(page, NR_FILE_PAGES);
 332        __dec_zone_page_state(page, NR_SHMEM);
 333        spin_unlock_irq(&mapping->tree_lock);
 334        page_cache_release(page);
 335        BUG_ON(error);
 336}
 337
 338/*
 339 * Like find_get_pages, but collecting swap entries as well as pages.
 340 */
 341static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping,
 342                                        pgoff_t start, unsigned int nr_pages,
 343                                        struct page **pages, pgoff_t *indices)
 344{
 345        unsigned int i;
 346        unsigned int ret;
 347        unsigned int nr_found;
 348
 349        rcu_read_lock();
 350restart:
 351        nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
 352                                (void ***)pages, indices, start, nr_pages);
 353        ret = 0;
 354        for (i = 0; i < nr_found; i++) {
 355                struct page *page;
 356repeat:
 357                page = radix_tree_deref_slot((void **)pages[i]);
 358                if (unlikely(!page))
 359                        continue;
 360                if (radix_tree_exception(page)) {
 361                        if (radix_tree_deref_retry(page))
 362                                goto restart;
 363                        /*
 364                         * Otherwise, we must be storing a swap entry
 365                         * here as an exceptional entry: so return it
 366                         * without attempting to raise page count.
 367                         */
 368                        goto export;
 369                }
 370                if (!page_cache_get_speculative(page))
 371                        goto repeat;
 372
 373                /* Has the page moved? */
 374                if (unlikely(page != *((void **)pages[i]))) {
 375                        page_cache_release(page);
 376                        goto repeat;
 377                }
 378export:
 379                indices[ret] = indices[i];
 380                pages[ret] = page;
 381                ret++;
 382        }
 383        if (unlikely(!ret && nr_found))
 384                goto restart;
 385        rcu_read_unlock();
 386        return ret;
 387}
 388
 389/*
 390 * Remove swap entry from radix tree, free the swap and its page cache.
 391 */
 392static int shmem_free_swap(struct address_space *mapping,
 393                           pgoff_t index, void *radswap)
 394{
 395        int error;
 396
 397        spin_lock_irq(&mapping->tree_lock);
 398        error = shmem_radix_tree_replace(mapping, index, radswap, NULL);
 399        spin_unlock_irq(&mapping->tree_lock);
 400        if (!error)
 401                free_swap_and_cache(radix_to_swp_entry(radswap));
 402        return error;
 403}
 404
 405/*
 406 * Pagevec may contain swap entries, so shuffle up pages before releasing.
 407 */
 408static void shmem_deswap_pagevec(struct pagevec *pvec)
 409{
 410        int i, j;
 411
 412        for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
 413                struct page *page = pvec->pages[i];
 414                if (!radix_tree_exceptional_entry(page))
 415                        pvec->pages[j++] = page;
 416        }
 417        pvec->nr = j;
 418}
 419
 420/*
 421 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
 422 */
 423void shmem_unlock_mapping(struct address_space *mapping)
 424{
 425        struct pagevec pvec;
 426        pgoff_t indices[PAGEVEC_SIZE];
 427        pgoff_t index = 0;
 428
 429        pagevec_init(&pvec, 0);
 430        /*
 431         * Minor point, but we might as well stop if someone else SHM_LOCKs it.
 432         */
 433        while (!mapping_unevictable(mapping)) {
 434                /*
 435                 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
 436                 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
 437                 */
 438                pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
 439                                        PAGEVEC_SIZE, pvec.pages, indices);
 440                if (!pvec.nr)
 441                        break;
 442                index = indices[pvec.nr - 1] + 1;
 443                shmem_deswap_pagevec(&pvec);
 444                check_move_unevictable_pages(pvec.pages, pvec.nr);
 445                pagevec_release(&pvec);
 446                cond_resched();
 447        }
 448}
 449
 450/*
 451 * Remove range of pages and swap entries from radix tree, and free them.
 452 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
 453 */
 454static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
 455                                                                 bool unfalloc)
 456{
 457        struct address_space *mapping = inode->i_mapping;
 458        struct shmem_inode_info *info = SHMEM_I(inode);
 459        pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 460        pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
 461        unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
 462        unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
 463        struct pagevec pvec;
 464        pgoff_t indices[PAGEVEC_SIZE];
 465        long nr_swaps_freed = 0;
 466        pgoff_t index;
 467        int i;
 468
 469        if (lend == -1)
 470                end = -1;       /* unsigned, so actually very big */
 471
 472        pagevec_init(&pvec, 0);
 473        index = start;
 474        while (index < end) {
 475                pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
 476                                min(end - index, (pgoff_t)PAGEVEC_SIZE),
 477                                                        pvec.pages, indices);
 478                if (!pvec.nr)
 479                        break;
 480                mem_cgroup_uncharge_start();
 481                for (i = 0; i < pagevec_count(&pvec); i++) {
 482                        struct page *page = pvec.pages[i];
 483
 484                        index = indices[i];
 485                        if (index >= end)
 486                                break;
 487
 488                        if (radix_tree_exceptional_entry(page)) {
 489                                if (unfalloc)
 490                                        continue;
 491                                nr_swaps_freed += !shmem_free_swap(mapping,
 492                                                                index, page);
 493                                continue;
 494                        }
 495
 496                        if (!trylock_page(page))
 497                                continue;
 498                        if (!unfalloc || !PageUptodate(page)) {
 499                                if (page->mapping == mapping) {
 500                                        VM_BUG_ON(PageWriteback(page));
 501                                        truncate_inode_page(mapping, page);
 502                                }
 503                        }
 504                        unlock_page(page);
 505                }
 506                shmem_deswap_pagevec(&pvec);
 507                pagevec_release(&pvec);
 508                mem_cgroup_uncharge_end();
 509                cond_resched();
 510                index++;
 511        }
 512
 513        if (partial_start) {
 514                struct page *page = NULL;
 515                shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
 516                if (page) {
 517                        unsigned int top = PAGE_CACHE_SIZE;
 518                        if (start > end) {
 519                                top = partial_end;
 520                                partial_end = 0;
 521                        }
 522                        zero_user_segment(page, partial_start, top);
 523                        set_page_dirty(page);
 524                        unlock_page(page);
 525                        page_cache_release(page);
 526                }
 527        }
 528        if (partial_end) {
 529                struct page *page = NULL;
 530                shmem_getpage(inode, end, &page, SGP_READ, NULL);
 531                if (page) {
 532                        zero_user_segment(page, 0, partial_end);
 533                        set_page_dirty(page);
 534                        unlock_page(page);
 535                        page_cache_release(page);
 536                }
 537        }
 538        if (start >= end)
 539                return;
 540
 541        index = start;
 542        for ( ; ; ) {
 543                cond_resched();
 544                pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
 545                                min(end - index, (pgoff_t)PAGEVEC_SIZE),
 546                                                        pvec.pages, indices);
 547                if (!pvec.nr) {
 548                        if (index == start || unfalloc)
 549                                break;
 550                        index = start;
 551                        continue;
 552                }
 553                if ((index == start || unfalloc) && indices[0] >= end) {
 554                        shmem_deswap_pagevec(&pvec);
 555                        pagevec_release(&pvec);
 556                        break;
 557                }
 558                mem_cgroup_uncharge_start();
 559                for (i = 0; i < pagevec_count(&pvec); i++) {
 560                        struct page *page = pvec.pages[i];
 561
 562                        index = indices[i];
 563                        if (index >= end)
 564                                break;
 565
 566                        if (radix_tree_exceptional_entry(page)) {
 567                                if (unfalloc)
 568                                        continue;
 569                                nr_swaps_freed += !shmem_free_swap(mapping,
 570                                                                index, page);
 571                                continue;
 572                        }
 573
 574                        lock_page(page);
 575                        if (!unfalloc || !PageUptodate(page)) {
 576                                if (page->mapping == mapping) {
 577                                        VM_BUG_ON(PageWriteback(page));
 578                                        truncate_inode_page(mapping, page);
 579                                }
 580                        }
 581                        unlock_page(page);
 582                }
 583                shmem_deswap_pagevec(&pvec);
 584                pagevec_release(&pvec);
 585                mem_cgroup_uncharge_end();
 586                index++;
 587        }
 588
 589        spin_lock(&info->lock);
 590        info->swapped -= nr_swaps_freed;
 591        shmem_recalc_inode(inode);
 592        spin_unlock(&info->lock);
 593}
 594
 595void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
 596{
 597        shmem_undo_range(inode, lstart, lend, false);
 598        inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 599}
 600EXPORT_SYMBOL_GPL(shmem_truncate_range);
 601
 602static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
 603{
 604        struct inode *inode = dentry->d_inode;
 605        int error;
 606
 607        error = inode_change_ok(inode, attr);
 608        if (error)
 609                return error;
 610
 611        if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
 612                loff_t oldsize = inode->i_size;
 613                loff_t newsize = attr->ia_size;
 614
 615                if (newsize != oldsize) {
 616                        i_size_write(inode, newsize);
 617                        inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 618                }
 619                if (newsize < oldsize) {
 620                        loff_t holebegin = round_up(newsize, PAGE_SIZE);
 621                        unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
 622                        shmem_truncate_range(inode, newsize, (loff_t)-1);
 623                        /* unmap again to remove racily COWed private pages */
 624                        unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
 625                }
 626        }
 627
 628        setattr_copy(inode, attr);
 629#ifdef CONFIG_TMPFS_POSIX_ACL
 630        if (attr->ia_valid & ATTR_MODE)
 631                error = generic_acl_chmod(inode);
 632#endif
 633        return error;
 634}
 635
 636static void shmem_evict_inode(struct inode *inode)
 637{
 638        struct shmem_inode_info *info = SHMEM_I(inode);
 639        struct shmem_xattr *xattr, *nxattr;
 640
 641        if (inode->i_mapping->a_ops == &shmem_aops) {
 642                shmem_unacct_size(info->flags, inode->i_size);
 643                inode->i_size = 0;
 644                shmem_truncate_range(inode, 0, (loff_t)-1);
 645                if (!list_empty(&info->swaplist)) {
 646                        mutex_lock(&shmem_swaplist_mutex);
 647                        list_del_init(&info->swaplist);
 648                        mutex_unlock(&shmem_swaplist_mutex);
 649                }
 650        } else
 651                kfree(info->symlink);
 652
 653        list_for_each_entry_safe(xattr, nxattr, &info->xattr_list, list) {
 654                kfree(xattr->name);
 655                kfree(xattr);
 656        }
 657        BUG_ON(inode->i_blocks);
 658        shmem_free_inode(inode->i_sb);
 659        clear_inode(inode);
 660}
 661
 662/*
 663 * If swap found in inode, free it and move page from swapcache to filecache.
 664 */
 665static int shmem_unuse_inode(struct shmem_inode_info *info,
 666                             swp_entry_t swap, struct page **pagep)
 667{
 668        struct address_space *mapping = info->vfs_inode.i_mapping;
 669        void *radswap;
 670        pgoff_t index;
 671        gfp_t gfp;
 672        int error = 0;
 673
 674        radswap = swp_to_radix_entry(swap);
 675        index = radix_tree_locate_item(&mapping->page_tree, radswap);
 676        if (index == -1)
 677                return 0;
 678
 679        /*
 680         * Move _head_ to start search for next from here.
 681         * But be careful: shmem_evict_inode checks list_empty without taking
 682         * mutex, and there's an instant in list_move_tail when info->swaplist
 683         * would appear empty, if it were the only one on shmem_swaplist.
 684         */
 685        if (shmem_swaplist.next != &info->swaplist)
 686                list_move_tail(&shmem_swaplist, &info->swaplist);
 687
 688        gfp = mapping_gfp_mask(mapping);
 689        if (shmem_should_replace_page(*pagep, gfp)) {
 690                mutex_unlock(&shmem_swaplist_mutex);
 691                error = shmem_replace_page(pagep, gfp, info, index);
 692                mutex_lock(&shmem_swaplist_mutex);
 693                /*
 694                 * We needed to drop mutex to make that restrictive page
 695                 * allocation, but the inode might have been freed while we
 696                 * dropped it: although a racing shmem_evict_inode() cannot
 697                 * complete without emptying the radix_tree, our page lock
 698                 * on this swapcache page is not enough to prevent that -
 699                 * free_swap_and_cache() of our swap entry will only
 700                 * trylock_page(), removing swap from radix_tree whatever.
 701                 *
 702                 * We must not proceed to shmem_add_to_page_cache() if the
 703                 * inode has been freed, but of course we cannot rely on
 704                 * inode or mapping or info to check that.  However, we can
 705                 * safely check if our swap entry is still in use (and here
 706                 * it can't have got reused for another page): if it's still
 707                 * in use, then the inode cannot have been freed yet, and we
 708                 * can safely proceed (if it's no longer in use, that tells
 709                 * nothing about the inode, but we don't need to unuse swap).
 710                 */
 711                if (!page_swapcount(*pagep))
 712                        error = -ENOENT;
 713        }
 714
 715        /*
 716         * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
 717         * but also to hold up shmem_evict_inode(): so inode cannot be freed
 718         * beneath us (pagelock doesn't help until the page is in pagecache).
 719         */
 720        if (!error)
 721                error = shmem_add_to_page_cache(*pagep, mapping, index,
 722                                                GFP_NOWAIT, radswap);
 723        if (error != -ENOMEM) {
 724                /*
 725                 * Truncation and eviction use free_swap_and_cache(), which
 726                 * only does trylock page: if we raced, best clean up here.
 727                 */
 728                delete_from_swap_cache(*pagep);
 729                set_page_dirty(*pagep);
 730                if (!error) {
 731                        spin_lock(&info->lock);
 732                        info->swapped--;
 733                        spin_unlock(&info->lock);
 734                        swap_free(swap);
 735                }
 736                error = 1;      /* not an error, but entry was found */
 737        }
 738        return error;
 739}
 740
 741/*
 742 * Search through swapped inodes to find and replace swap by page.
 743 */
 744int shmem_unuse(swp_entry_t swap, struct page *page)
 745{
 746        struct list_head *this, *next;
 747        struct shmem_inode_info *info;
 748        int found = 0;
 749        int error = 0;
 750
 751        /*
 752         * There's a faint possibility that swap page was replaced before
 753         * caller locked it: caller will come back later with the right page.
 754         */
 755        if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
 756                goto out;
 757
 758        /*
 759         * Charge page using GFP_KERNEL while we can wait, before taking
 760         * the shmem_swaplist_mutex which might hold up shmem_writepage().
 761         * Charged back to the user (not to caller) when swap account is used.
 762         */
 763        error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
 764        if (error)
 765                goto out;
 766        /* No radix_tree_preload: swap entry keeps a place for page in tree */
 767
 768        mutex_lock(&shmem_swaplist_mutex);
 769        list_for_each_safe(this, next, &shmem_swaplist) {
 770                info = list_entry(this, struct shmem_inode_info, swaplist);
 771                if (info->swapped)
 772                        found = shmem_unuse_inode(info, swap, &page);
 773                else
 774                        list_del_init(&info->swaplist);
 775                cond_resched();
 776                if (found)
 777                        break;
 778        }
 779        mutex_unlock(&shmem_swaplist_mutex);
 780
 781        if (found < 0)
 782                error = found;
 783out:
 784        unlock_page(page);
 785        page_cache_release(page);
 786        return error;
 787}
 788
 789/*
 790 * Move the page from the page cache to the swap cache.
 791 */
 792static int shmem_writepage(struct page *page, struct writeback_control *wbc)
 793{
 794        struct shmem_inode_info *info;
 795        struct address_space *mapping;
 796        struct inode *inode;
 797        swp_entry_t swap;
 798        pgoff_t index;
 799
 800        BUG_ON(!PageLocked(page));
 801        mapping = page->mapping;
 802        index = page->index;
 803        inode = mapping->host;
 804        info = SHMEM_I(inode);
 805        if (info->flags & VM_LOCKED)
 806                goto redirty;
 807        if (!total_swap_pages)
 808                goto redirty;
 809
 810        /*
 811         * shmem_backing_dev_info's capabilities prevent regular writeback or
 812         * sync from ever calling shmem_writepage; but a stacking filesystem
 813         * might use ->writepage of its underlying filesystem, in which case
 814         * tmpfs should write out to swap only in response to memory pressure,
 815         * and not for the writeback threads or sync.
 816         */
 817        if (!wbc->for_reclaim) {
 818                WARN_ON_ONCE(1);        /* Still happens? Tell us about it! */
 819                goto redirty;
 820        }
 821
 822        /*
 823         * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
 824         * value into swapfile.c, the only way we can correctly account for a
 825         * fallocated page arriving here is now to initialize it and write it.
 826         *
 827         * That's okay for a page already fallocated earlier, but if we have
 828         * not yet completed the fallocation, then (a) we want to keep track
 829         * of this page in case we have to undo it, and (b) it may not be a
 830         * good idea to continue anyway, once we're pushing into swap.  So
 831         * reactivate the page, and let shmem_fallocate() quit when too many.
 832         */
 833        if (!PageUptodate(page)) {
 834                if (inode->i_private) {
 835                        struct shmem_falloc *shmem_falloc;
 836                        spin_lock(&inode->i_lock);
 837                        shmem_falloc = inode->i_private;
 838                        if (shmem_falloc &&
 839                            index >= shmem_falloc->start &&
 840                            index < shmem_falloc->next)
 841                                shmem_falloc->nr_unswapped++;
 842                        else
 843                                shmem_falloc = NULL;
 844                        spin_unlock(&inode->i_lock);
 845                        if (shmem_falloc)
 846                                goto redirty;
 847                }
 848                clear_highpage(page);
 849                flush_dcache_page(page);
 850                SetPageUptodate(page);
 851        }
 852
 853        swap = get_swap_page();
 854        if (!swap.val)
 855                goto redirty;
 856
 857        /*
 858         * Add inode to shmem_unuse()'s list of swapped-out inodes,
 859         * if it's not already there.  Do it now before the page is
 860         * moved to swap cache, when its pagelock no longer protects
 861         * the inode from eviction.  But don't unlock the mutex until
 862         * we've incremented swapped, because shmem_unuse_inode() will
 863         * prune a !swapped inode from the swaplist under this mutex.
 864         */
 865        mutex_lock(&shmem_swaplist_mutex);
 866        if (list_empty(&info->swaplist))
 867                list_add_tail(&info->swaplist, &shmem_swaplist);
 868
 869        if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
 870                swap_shmem_alloc(swap);
 871                shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
 872
 873                spin_lock(&info->lock);
 874                info->swapped++;
 875                shmem_recalc_inode(inode);
 876                spin_unlock(&info->lock);
 877
 878                mutex_unlock(&shmem_swaplist_mutex);
 879                BUG_ON(page_mapped(page));
 880                swap_writepage(page, wbc);
 881                return 0;
 882        }
 883
 884        mutex_unlock(&shmem_swaplist_mutex);
 885        swapcache_free(swap, NULL);
 886redirty:
 887        set_page_dirty(page);
 888        if (wbc->for_reclaim)
 889                return AOP_WRITEPAGE_ACTIVATE;  /* Return with page locked */
 890        unlock_page(page);
 891        return 0;
 892}
 893
 894#ifdef CONFIG_NUMA
 895#ifdef CONFIG_TMPFS
 896static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
 897{
 898        char buffer[64];
 899
 900        if (!mpol || mpol->mode == MPOL_DEFAULT)
 901                return;         /* show nothing */
 902
 903        mpol_to_str(buffer, sizeof(buffer), mpol, 1);
 904
 905        seq_printf(seq, ",mpol=%s", buffer);
 906}
 907
 908static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
 909{
 910        struct mempolicy *mpol = NULL;
 911        if (sbinfo->mpol) {
 912                spin_lock(&sbinfo->stat_lock);  /* prevent replace/use races */
 913                mpol = sbinfo->mpol;
 914                mpol_get(mpol);
 915                spin_unlock(&sbinfo->stat_lock);
 916        }
 917        return mpol;
 918}
 919#endif /* CONFIG_TMPFS */
 920
 921static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
 922                        struct shmem_inode_info *info, pgoff_t index)
 923{
 924        struct mempolicy mpol, *spol;
 925        struct vm_area_struct pvma;
 926
 927        spol = mpol_cond_copy(&mpol,
 928                        mpol_shared_policy_lookup(&info->policy, index));
 929
 930        /* Create a pseudo vma that just contains the policy */
 931        pvma.vm_start = 0;
 932        /* Bias interleave by inode number to distribute better across nodes */
 933        pvma.vm_pgoff = index + info->vfs_inode.i_ino;
 934        pvma.vm_ops = NULL;
 935        pvma.vm_policy = spol;
 936        return swapin_readahead(swap, gfp, &pvma, 0);
 937}
 938
 939static struct page *shmem_alloc_page(gfp_t gfp,
 940                        struct shmem_inode_info *info, pgoff_t index)
 941{
 942        struct vm_area_struct pvma;
 943
 944        /* Create a pseudo vma that just contains the policy */
 945        pvma.vm_start = 0;
 946        /* Bias interleave by inode number to distribute better across nodes */
 947        pvma.vm_pgoff = index + info->vfs_inode.i_ino;
 948        pvma.vm_ops = NULL;
 949        pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
 950
 951        /*
 952         * alloc_page_vma() will drop the shared policy reference
 953         */
 954        return alloc_page_vma(gfp, &pvma, 0);
 955}
 956#else /* !CONFIG_NUMA */
 957#ifdef CONFIG_TMPFS
 958static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
 959{
 960}
 961#endif /* CONFIG_TMPFS */
 962
 963static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
 964                        struct shmem_inode_info *info, pgoff_t index)
 965{
 966        return swapin_readahead(swap, gfp, NULL, 0);
 967}
 968
 969static inline struct page *shmem_alloc_page(gfp_t gfp,
 970                        struct shmem_inode_info *info, pgoff_t index)
 971{
 972        return alloc_page(gfp);
 973}
 974#endif /* CONFIG_NUMA */
 975
 976#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
 977static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
 978{
 979        return NULL;
 980}
 981#endif
 982
 983/*
 984 * When a page is moved from swapcache to shmem filecache (either by the
 985 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
 986 * shmem_unuse_inode()), it may have been read in earlier from swap, in
 987 * ignorance of the mapping it belongs to.  If that mapping has special
 988 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
 989 * we may need to copy to a suitable page before moving to filecache.
 990 *
 991 * In a future release, this may well be extended to respect cpuset and
 992 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
 993 * but for now it is a simple matter of zone.
 994 */
 995static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
 996{
 997        return page_zonenum(page) > gfp_zone(gfp);
 998}
 999
1000static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1001                                struct shmem_inode_info *info, pgoff_t index)
1002{
1003        struct page *oldpage, *newpage;
1004        struct address_space *swap_mapping;
1005        pgoff_t swap_index;
1006        int error;
1007
1008        oldpage = *pagep;
1009        swap_index = page_private(oldpage);
1010        swap_mapping = page_mapping(oldpage);
1011
1012        /*
1013         * We have arrived here because our zones are constrained, so don't
1014         * limit chance of success by further cpuset and node constraints.
1015         */
1016        gfp &= ~GFP_CONSTRAINT_MASK;
1017        newpage = shmem_alloc_page(gfp, info, index);
1018        if (!newpage)
1019                return -ENOMEM;
1020
1021        page_cache_get(newpage);
1022        copy_highpage(newpage, oldpage);
1023        flush_dcache_page(newpage);
1024
1025        __set_page_locked(newpage);
1026        SetPageUptodate(newpage);
1027        SetPageSwapBacked(newpage);
1028        set_page_private(newpage, swap_index);
1029        SetPageSwapCache(newpage);
1030
1031        /*
1032         * Our caller will very soon move newpage out of swapcache, but it's
1033         * a nice clean interface for us to replace oldpage by newpage there.
1034         */
1035        spin_lock_irq(&swap_mapping->tree_lock);
1036        error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1037                                                                   newpage);
1038        if (!error) {
1039                __inc_zone_page_state(newpage, NR_FILE_PAGES);
1040                __dec_zone_page_state(oldpage, NR_FILE_PAGES);
1041        }
1042        spin_unlock_irq(&swap_mapping->tree_lock);
1043
1044        if (unlikely(error)) {
1045                /*
1046                 * Is this possible?  I think not, now that our callers check
1047                 * both PageSwapCache and page_private after getting page lock;
1048                 * but be defensive.  Reverse old to newpage for clear and free.
1049                 */
1050                oldpage = newpage;
1051        } else {
1052                mem_cgroup_replace_page_cache(oldpage, newpage);
1053                lru_cache_add_anon(newpage);
1054                *pagep = newpage;
1055        }
1056
1057        ClearPageSwapCache(oldpage);
1058        set_page_private(oldpage, 0);
1059
1060        unlock_page(oldpage);
1061        page_cache_release(oldpage);
1062        page_cache_release(oldpage);
1063        return error;
1064}
1065
1066/*
1067 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1068 *
1069 * If we allocate a new one we do not mark it dirty. That's up to the
1070 * vm. If we swap it in we mark it dirty since we also free the swap
1071 * entry since a page cannot live in both the swap and page cache
1072 */
1073static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1074        struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
1075{
1076        struct address_space *mapping = inode->i_mapping;
1077        struct shmem_inode_info *info;
1078        struct shmem_sb_info *sbinfo;
1079        struct page *page;
1080        swp_entry_t swap;
1081        int error;
1082        int once = 0;
1083        int alloced = 0;
1084
1085        if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
1086                return -EFBIG;
1087repeat:
1088        swap.val = 0;
1089        page = find_lock_page(mapping, index);
1090        if (radix_tree_exceptional_entry(page)) {
1091                swap = radix_to_swp_entry(page);
1092                page = NULL;
1093        }
1094
1095        if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1096            ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1097                error = -EINVAL;
1098                goto failed;
1099        }
1100
1101        /* fallocated page? */
1102        if (page && !PageUptodate(page)) {
1103                if (sgp != SGP_READ)
1104                        goto clear;
1105                unlock_page(page);
1106                page_cache_release(page);
1107                page = NULL;
1108        }
1109        if (page || (sgp == SGP_READ && !swap.val)) {
1110                *pagep = page;
1111                return 0;
1112        }
1113
1114        /*
1115         * Fast cache lookup did not find it:
1116         * bring it back from swap or allocate.
1117         */
1118        info = SHMEM_I(inode);
1119        sbinfo = SHMEM_SB(inode->i_sb);
1120
1121        if (swap.val) {
1122                /* Look it up and read it in.. */
1123                page = lookup_swap_cache(swap);
1124                if (!page) {
1125                        /* here we actually do the io */
1126                        if (fault_type)
1127                                *fault_type |= VM_FAULT_MAJOR;
1128                        page = shmem_swapin(swap, gfp, info, index);
1129                        if (!page) {
1130                                error = -ENOMEM;
1131                                goto failed;
1132                        }
1133                }
1134
1135                /* We have to do this with page locked to prevent races */
1136                lock_page(page);
1137                if (!PageSwapCache(page) || page_private(page) != swap.val ||
1138                    !shmem_confirm_swap(mapping, index, swap)) {
1139                        error = -EEXIST;        /* try again */
1140                        goto unlock;
1141                }
1142                if (!PageUptodate(page)) {
1143                        error = -EIO;
1144                        goto failed;
1145                }
1146                wait_on_page_writeback(page);
1147
1148                if (shmem_should_replace_page(page, gfp)) {
1149                        error = shmem_replace_page(&page, gfp, info, index);
1150                        if (error)
1151                                goto failed;
1152                }
1153
1154                error = mem_cgroup_cache_charge(page, current->mm,
1155                                                gfp & GFP_RECLAIM_MASK);
1156                if (!error) {
1157                        error = shmem_add_to_page_cache(page, mapping, index,
1158                                                gfp, swp_to_radix_entry(swap));
1159                        /* We already confirmed swap, and make no allocation */
1160                        VM_BUG_ON(error);
1161                }
1162                if (error)
1163                        goto failed;
1164
1165                spin_lock(&info->lock);
1166                info->swapped--;
1167                shmem_recalc_inode(inode);
1168                spin_unlock(&info->lock);
1169
1170                delete_from_swap_cache(page);
1171                set_page_dirty(page);
1172                swap_free(swap);
1173
1174        } else {
1175                if (shmem_acct_block(info->flags)) {
1176                        error = -ENOSPC;
1177                        goto failed;
1178                }
1179                if (sbinfo->max_blocks) {
1180                        if (percpu_counter_compare(&sbinfo->used_blocks,
1181                                                sbinfo->max_blocks) >= 0) {
1182                                error = -ENOSPC;
1183                                goto unacct;
1184                        }
1185                        percpu_counter_inc(&sbinfo->used_blocks);
1186                }
1187
1188                page = shmem_alloc_page(gfp, info, index);
1189                if (!page) {
1190                        error = -ENOMEM;
1191                        goto decused;
1192                }
1193
1194                SetPageSwapBacked(page);
1195                __set_page_locked(page);
1196                error = mem_cgroup_cache_charge(page, current->mm,
1197                                                gfp & GFP_RECLAIM_MASK);
1198                if (error)
1199                        goto decused;
1200                error = radix_tree_preload(gfp & GFP_RECLAIM_MASK);
1201                if (!error) {
1202                        error = shmem_add_to_page_cache(page, mapping, index,
1203                                                        gfp, NULL);
1204                        radix_tree_preload_end();
1205                }
1206                if (error) {
1207                        mem_cgroup_uncharge_cache_page(page);
1208                        goto decused;
1209                }
1210                lru_cache_add_anon(page);
1211
1212                spin_lock(&info->lock);
1213                info->alloced++;
1214                inode->i_blocks += BLOCKS_PER_PAGE;
1215                shmem_recalc_inode(inode);
1216                spin_unlock(&info->lock);
1217                alloced = true;
1218
1219                /*
1220                 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1221                 */
1222                if (sgp == SGP_FALLOC)
1223                        sgp = SGP_WRITE;
1224clear:
1225                /*
1226                 * Let SGP_WRITE caller clear ends if write does not fill page;
1227                 * but SGP_FALLOC on a page fallocated earlier must initialize
1228                 * it now, lest undo on failure cancel our earlier guarantee.
1229                 */
1230                if (sgp != SGP_WRITE) {
1231                        clear_highpage(page);
1232                        flush_dcache_page(page);
1233                        SetPageUptodate(page);
1234                }
1235                if (sgp == SGP_DIRTY)
1236                        set_page_dirty(page);
1237        }
1238
1239        /* Perhaps the file has been truncated since we checked */
1240        if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1241            ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1242                error = -EINVAL;
1243                if (alloced)
1244                        goto trunc;
1245                else
1246                        goto failed;
1247        }
1248        *pagep = page;
1249        return 0;
1250
1251        /*
1252         * Error recovery.
1253         */
1254trunc:
1255        info = SHMEM_I(inode);
1256        ClearPageDirty(page);
1257        delete_from_page_cache(page);
1258        spin_lock(&info->lock);
1259        info->alloced--;
1260        inode->i_blocks -= BLOCKS_PER_PAGE;
1261        spin_unlock(&info->lock);
1262decused:
1263        sbinfo = SHMEM_SB(inode->i_sb);
1264        if (sbinfo->max_blocks)
1265                percpu_counter_add(&sbinfo->used_blocks, -1);
1266unacct:
1267        shmem_unacct_blocks(info->flags, 1);
1268failed:
1269        if (swap.val && error != -EINVAL &&
1270            !shmem_confirm_swap(mapping, index, swap))
1271                error = -EEXIST;
1272unlock:
1273        if (page) {
1274                unlock_page(page);
1275                page_cache_release(page);
1276        }
1277        if (error == -ENOSPC && !once++) {
1278                info = SHMEM_I(inode);
1279                spin_lock(&info->lock);
1280                shmem_recalc_inode(inode);
1281                spin_unlock(&info->lock);
1282                goto repeat;
1283        }
1284        if (error == -EEXIST)   /* from above or from radix_tree_insert */
1285                goto repeat;
1286        return error;
1287}
1288
1289static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1290{
1291        struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1292        int error;
1293        int ret = VM_FAULT_LOCKED;
1294
1295        error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1296        if (error)
1297                return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1298
1299        if (ret & VM_FAULT_MAJOR) {
1300                count_vm_event(PGMAJFAULT);
1301                mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1302        }
1303        return ret;
1304}
1305
1306#ifdef CONFIG_NUMA
1307static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
1308{
1309        struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1310        return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
1311}
1312
1313static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1314                                          unsigned long addr)
1315{
1316        struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1317        pgoff_t index;
1318
1319        index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1320        return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
1321}
1322#endif
1323
1324int shmem_lock(struct file *file, int lock, struct user_struct *user)
1325{
1326        struct inode *inode = file->f_path.dentry->d_inode;
1327        struct shmem_inode_info *info = SHMEM_I(inode);
1328        int retval = -ENOMEM;
1329
1330        spin_lock(&info->lock);
1331        if (lock && !(info->flags & VM_LOCKED)) {
1332                if (!user_shm_lock(inode->i_size, user))
1333                        goto out_nomem;
1334                info->flags |= VM_LOCKED;
1335                mapping_set_unevictable(file->f_mapping);
1336        }
1337        if (!lock && (info->flags & VM_LOCKED) && user) {
1338                user_shm_unlock(inode->i_size, user);
1339                info->flags &= ~VM_LOCKED;
1340                mapping_clear_unevictable(file->f_mapping);
1341        }
1342        retval = 0;
1343
1344out_nomem:
1345        spin_unlock(&info->lock);
1346        return retval;
1347}
1348
1349static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1350{
1351        file_accessed(file);
1352        vma->vm_ops = &shmem_vm_ops;
1353        vma->vm_flags |= VM_CAN_NONLINEAR;
1354        return 0;
1355}
1356
1357static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1358                                     umode_t mode, dev_t dev, unsigned long flags)
1359{
1360        struct inode *inode;
1361        struct shmem_inode_info *info;
1362        struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1363
1364        if (shmem_reserve_inode(sb))
1365                return NULL;
1366
1367        inode = new_inode(sb);
1368        if (inode) {
1369                inode->i_ino = get_next_ino();
1370                inode_init_owner(inode, dir, mode);
1371                inode->i_blocks = 0;
1372                inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1373                inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1374                inode->i_generation = get_seconds();
1375                info = SHMEM_I(inode);
1376                memset(info, 0, (char *)inode - (char *)info);
1377                spin_lock_init(&info->lock);
1378                info->flags = flags & VM_NORESERVE;
1379                INIT_LIST_HEAD(&info->swaplist);
1380                INIT_LIST_HEAD(&info->xattr_list);
1381                cache_no_acl(inode);
1382
1383                switch (mode & S_IFMT) {
1384                default:
1385                        inode->i_op = &shmem_special_inode_operations;
1386                        init_special_inode(inode, mode, dev);
1387                        break;
1388                case S_IFREG:
1389                        inode->i_mapping->a_ops = &shmem_aops;
1390                        inode->i_op = &shmem_inode_operations;
1391                        inode->i_fop = &shmem_file_operations;
1392                        mpol_shared_policy_init(&info->policy,
1393                                                 shmem_get_sbmpol(sbinfo));
1394                        break;
1395                case S_IFDIR:
1396                        inc_nlink(inode);
1397                        /* Some things misbehave if size == 0 on a directory */
1398                        inode->i_size = 2 * BOGO_DIRENT_SIZE;
1399                        inode->i_op = &shmem_dir_inode_operations;
1400                        inode->i_fop = &simple_dir_operations;
1401                        break;
1402                case S_IFLNK:
1403                        /*
1404                         * Must not load anything in the rbtree,
1405                         * mpol_free_shared_policy will not be called.
1406                         */
1407                        mpol_shared_policy_init(&info->policy, NULL);
1408                        break;
1409                }
1410        } else
1411                shmem_free_inode(sb);
1412        return inode;
1413}
1414
1415#ifdef CONFIG_TMPFS
1416static const struct inode_operations shmem_symlink_inode_operations;
1417static const struct inode_operations shmem_short_symlink_operations;
1418
1419#ifdef CONFIG_TMPFS_XATTR
1420static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1421#else
1422#define shmem_initxattrs NULL
1423#endif
1424
1425static int
1426shmem_write_begin(struct file *file, struct address_space *mapping,
1427                        loff_t pos, unsigned len, unsigned flags,
1428                        struct page **pagep, void **fsdata)
1429{
1430        struct inode *inode = mapping->host;
1431        pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1432        return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1433}
1434
1435static int
1436shmem_write_end(struct file *file, struct address_space *mapping,
1437                        loff_t pos, unsigned len, unsigned copied,
1438                        struct page *page, void *fsdata)
1439{
1440        struct inode *inode = mapping->host;
1441
1442        if (pos + copied > inode->i_size)
1443                i_size_write(inode, pos + copied);
1444
1445        if (!PageUptodate(page)) {
1446                if (copied < PAGE_CACHE_SIZE) {
1447                        unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1448                        zero_user_segments(page, 0, from,
1449                                        from + copied, PAGE_CACHE_SIZE);
1450                }
1451                SetPageUptodate(page);
1452        }
1453        set_page_dirty(page);
1454        unlock_page(page);
1455        page_cache_release(page);
1456
1457        return copied;
1458}
1459
1460static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1461{
1462        struct inode *inode = filp->f_path.dentry->d_inode;
1463        struct address_space *mapping = inode->i_mapping;
1464        pgoff_t index;
1465        unsigned long offset;
1466        enum sgp_type sgp = SGP_READ;
1467
1468        /*
1469         * Might this read be for a stacking filesystem?  Then when reading
1470         * holes of a sparse file, we actually need to allocate those pages,
1471         * and even mark them dirty, so it cannot exceed the max_blocks limit.
1472         */
1473        if (segment_eq(get_fs(), KERNEL_DS))
1474                sgp = SGP_DIRTY;
1475
1476        index = *ppos >> PAGE_CACHE_SHIFT;
1477        offset = *ppos & ~PAGE_CACHE_MASK;
1478
1479        for (;;) {
1480                struct page *page = NULL;
1481                pgoff_t end_index;
1482                unsigned long nr, ret;
1483                loff_t i_size = i_size_read(inode);
1484
1485                end_index = i_size >> PAGE_CACHE_SHIFT;
1486                if (index > end_index)
1487                        break;
1488                if (index == end_index) {
1489                        nr = i_size & ~PAGE_CACHE_MASK;
1490                        if (nr <= offset)
1491                                break;
1492                }
1493
1494                desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1495                if (desc->error) {
1496                        if (desc->error == -EINVAL)
1497                                desc->error = 0;
1498                        break;
1499                }
1500                if (page)
1501                        unlock_page(page);
1502
1503                /*
1504                 * We must evaluate after, since reads (unlike writes)
1505                 * are called without i_mutex protection against truncate
1506                 */
1507                nr = PAGE_CACHE_SIZE;
1508                i_size = i_size_read(inode);
1509                end_index = i_size >> PAGE_CACHE_SHIFT;
1510                if (index == end_index) {
1511                        nr = i_size & ~PAGE_CACHE_MASK;
1512                        if (nr <= offset) {
1513                                if (page)
1514                                        page_cache_release(page);
1515                                break;
1516                        }
1517                }
1518                nr -= offset;
1519
1520                if (page) {
1521                        /*
1522                         * If users can be writing to this page using arbitrary
1523                         * virtual addresses, take care about potential aliasing
1524                         * before reading the page on the kernel side.
1525                         */
1526                        if (mapping_writably_mapped(mapping))
1527                                flush_dcache_page(page);
1528                        /*
1529                         * Mark the page accessed if we read the beginning.
1530                         */
1531                        if (!offset)
1532                                mark_page_accessed(page);
1533                } else {
1534                        page = ZERO_PAGE(0);
1535                        page_cache_get(page);
1536                }
1537
1538                /*
1539                 * Ok, we have the page, and it's up-to-date, so
1540                 * now we can copy it to user space...
1541                 *
1542                 * The actor routine returns how many bytes were actually used..
1543                 * NOTE! This may not be the same as how much of a user buffer
1544                 * we filled up (we may be padding etc), so we can only update
1545                 * "pos" here (the actor routine has to update the user buffer
1546                 * pointers and the remaining count).
1547                 */
1548                ret = actor(desc, page, offset, nr);
1549                offset += ret;
1550                index += offset >> PAGE_CACHE_SHIFT;
1551                offset &= ~PAGE_CACHE_MASK;
1552
1553                page_cache_release(page);
1554                if (ret != nr || !desc->count)
1555                        break;
1556
1557                cond_resched();
1558        }
1559
1560        *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1561        file_accessed(filp);
1562}
1563
1564static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1565                const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1566{
1567        struct file *filp = iocb->ki_filp;
1568        ssize_t retval;
1569        unsigned long seg;
1570        size_t count;
1571        loff_t *ppos = &iocb->ki_pos;
1572
1573        retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1574        if (retval)
1575                return retval;
1576
1577        for (seg = 0; seg < nr_segs; seg++) {
1578                read_descriptor_t desc;
1579
1580                desc.written = 0;
1581                desc.arg.buf = iov[seg].iov_base;
1582                desc.count = iov[seg].iov_len;
1583                if (desc.count == 0)
1584                        continue;
1585                desc.error = 0;
1586                do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1587                retval += desc.written;
1588                if (desc.error) {
1589                        retval = retval ?: desc.error;
1590                        break;
1591                }
1592                if (desc.count > 0)
1593                        break;
1594        }
1595        return retval;
1596}
1597
1598static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1599                                struct pipe_inode_info *pipe, size_t len,
1600                                unsigned int flags)
1601{
1602        struct address_space *mapping = in->f_mapping;
1603        struct inode *inode = mapping->host;
1604        unsigned int loff, nr_pages, req_pages;
1605        struct page *pages[PIPE_DEF_BUFFERS];
1606        struct partial_page partial[PIPE_DEF_BUFFERS];
1607        struct page *page;
1608        pgoff_t index, end_index;
1609        loff_t isize, left;
1610        int error, page_nr;
1611        struct splice_pipe_desc spd = {
1612                .pages = pages,
1613                .partial = partial,
1614                .nr_pages_max = PIPE_DEF_BUFFERS,
1615                .flags = flags,
1616                .ops = &page_cache_pipe_buf_ops,
1617                .spd_release = spd_release_page,
1618        };
1619
1620        isize = i_size_read(inode);
1621        if (unlikely(*ppos >= isize))
1622                return 0;
1623
1624        left = isize - *ppos;
1625        if (unlikely(left < len))
1626                len = left;
1627
1628        if (splice_grow_spd(pipe, &spd))
1629                return -ENOMEM;
1630
1631        index = *ppos >> PAGE_CACHE_SHIFT;
1632        loff = *ppos & ~PAGE_CACHE_MASK;
1633        req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1634        nr_pages = min(req_pages, pipe->buffers);
1635
1636        spd.nr_pages = find_get_pages_contig(mapping, index,
1637                                                nr_pages, spd.pages);
1638        index += spd.nr_pages;
1639        error = 0;
1640
1641        while (spd.nr_pages < nr_pages) {
1642                error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1643                if (error)
1644                        break;
1645                unlock_page(page);
1646                spd.pages[spd.nr_pages++] = page;
1647                index++;
1648        }
1649
1650        index = *ppos >> PAGE_CACHE_SHIFT;
1651        nr_pages = spd.nr_pages;
1652        spd.nr_pages = 0;
1653
1654        for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1655                unsigned int this_len;
1656
1657                if (!len)
1658                        break;
1659
1660                this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1661                page = spd.pages[page_nr];
1662
1663                if (!PageUptodate(page) || page->mapping != mapping) {
1664                        error = shmem_getpage(inode, index, &page,
1665                                                        SGP_CACHE, NULL);
1666                        if (error)
1667                                break;
1668                        unlock_page(page);
1669                        page_cache_release(spd.pages[page_nr]);
1670                        spd.pages[page_nr] = page;
1671                }
1672
1673                isize = i_size_read(inode);
1674                end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1675                if (unlikely(!isize || index > end_index))
1676                        break;
1677
1678                if (end_index == index) {
1679                        unsigned int plen;
1680
1681                        plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1682                        if (plen <= loff)
1683                                break;
1684
1685                        this_len = min(this_len, plen - loff);
1686                        len = this_len;
1687                }
1688
1689                spd.partial[page_nr].offset = loff;
1690                spd.partial[page_nr].len = this_len;
1691                len -= this_len;
1692                loff = 0;
1693                spd.nr_pages++;
1694                index++;
1695        }
1696
1697        while (page_nr < nr_pages)
1698                page_cache_release(spd.pages[page_nr++]);
1699
1700        if (spd.nr_pages)
1701                error = splice_to_pipe(pipe, &spd);
1702
1703        splice_shrink_spd(&spd);
1704
1705        if (error > 0) {
1706                *ppos += error;
1707                file_accessed(in);
1708        }
1709        return error;
1710}
1711
1712static long shmem_fallocate(struct file *file, int mode, loff_t offset,
1713                                                         loff_t len)
1714{
1715        struct inode *inode = file->f_path.dentry->d_inode;
1716        struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1717        struct shmem_falloc shmem_falloc;
1718        pgoff_t start, index, end;
1719        int error;
1720
1721        mutex_lock(&inode->i_mutex);
1722
1723        if (mode & FALLOC_FL_PUNCH_HOLE) {
1724                struct address_space *mapping = file->f_mapping;
1725                loff_t unmap_start = round_up(offset, PAGE_SIZE);
1726                loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
1727
1728                if ((u64)unmap_end > (u64)unmap_start)
1729                        unmap_mapping_range(mapping, unmap_start,
1730                                            1 + unmap_end - unmap_start, 0);
1731                shmem_truncate_range(inode, offset, offset + len - 1);
1732                /* No need to unmap again: hole-punching leaves COWed pages */
1733                error = 0;
1734                goto out;
1735        }
1736
1737        /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1738        error = inode_newsize_ok(inode, offset + len);
1739        if (error)
1740                goto out;
1741
1742        start = offset >> PAGE_CACHE_SHIFT;
1743        end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1744        /* Try to avoid a swapstorm if len is impossible to satisfy */
1745        if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
1746                error = -ENOSPC;
1747                goto out;
1748        }
1749
1750        shmem_falloc.start = start;
1751        shmem_falloc.next  = start;
1752        shmem_falloc.nr_falloced = 0;
1753        shmem_falloc.nr_unswapped = 0;
1754        spin_lock(&inode->i_lock);
1755        inode->i_private = &shmem_falloc;
1756        spin_unlock(&inode->i_lock);
1757
1758        for (index = start; index < end; index++) {
1759                struct page *page;
1760
1761                /*
1762                 * Good, the fallocate(2) manpage permits EINTR: we may have
1763                 * been interrupted because we are using up too much memory.
1764                 */
1765                if (signal_pending(current))
1766                        error = -EINTR;
1767                else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
1768                        error = -ENOMEM;
1769                else
1770                        error = shmem_getpage(inode, index, &page, SGP_FALLOC,
1771                                                                        NULL);
1772                if (error) {
1773                        /* Remove the !PageUptodate pages we added */
1774                        shmem_undo_range(inode,
1775                                (loff_t)start << PAGE_CACHE_SHIFT,
1776                                (loff_t)index << PAGE_CACHE_SHIFT, true);
1777                        goto undone;
1778                }
1779
1780                /*
1781                 * Inform shmem_writepage() how far we have reached.
1782                 * No need for lock or barrier: we have the page lock.
1783                 */
1784                shmem_falloc.next++;
1785                if (!PageUptodate(page))
1786                        shmem_falloc.nr_falloced++;
1787
1788                /*
1789                 * If !PageUptodate, leave it that way so that freeable pages
1790                 * can be recognized if we need to rollback on error later.
1791                 * But set_page_dirty so that memory pressure will swap rather
1792                 * than free the pages we are allocating (and SGP_CACHE pages
1793                 * might still be clean: we now need to mark those dirty too).
1794                 */
1795                set_page_dirty(page);
1796                unlock_page(page);
1797                page_cache_release(page);
1798                cond_resched();
1799        }
1800
1801        if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
1802                i_size_write(inode, offset + len);
1803        inode->i_ctime = CURRENT_TIME;
1804undone:
1805        spin_lock(&inode->i_lock);
1806        inode->i_private = NULL;
1807        spin_unlock(&inode->i_lock);
1808out:
1809        mutex_unlock(&inode->i_mutex);
1810        return error;
1811}
1812
1813static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1814{
1815        struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1816
1817        buf->f_type = TMPFS_MAGIC;
1818        buf->f_bsize = PAGE_CACHE_SIZE;
1819        buf->f_namelen = NAME_MAX;
1820        if (sbinfo->max_blocks) {
1821                buf->f_blocks = sbinfo->max_blocks;
1822                buf->f_bavail =
1823                buf->f_bfree  = sbinfo->max_blocks -
1824                                percpu_counter_sum(&sbinfo->used_blocks);
1825        }
1826        if (sbinfo->max_inodes) {
1827                buf->f_files = sbinfo->max_inodes;
1828                buf->f_ffree = sbinfo->free_inodes;
1829        }
1830        /* else leave those fields 0 like simple_statfs */
1831        return 0;
1832}
1833
1834/*
1835 * File creation. Allocate an inode, and we're done..
1836 */
1837static int
1838shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
1839{
1840        struct inode *inode;
1841        int error = -ENOSPC;
1842
1843        inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1844        if (inode) {
1845                error = security_inode_init_security(inode, dir,
1846                                                     &dentry->d_name,
1847                                                     shmem_initxattrs, NULL);
1848                if (error) {
1849                        if (error != -EOPNOTSUPP) {
1850                                iput(inode);
1851                                return error;
1852                        }
1853                }
1854#ifdef CONFIG_TMPFS_POSIX_ACL
1855                error = generic_acl_init(inode, dir);
1856                if (error) {
1857                        iput(inode);
1858                        return error;
1859                }
1860#else
1861                error = 0;
1862#endif
1863                dir->i_size += BOGO_DIRENT_SIZE;
1864                dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1865                d_instantiate(dentry, inode);
1866                dget(dentry); /* Extra count - pin the dentry in core */
1867        }
1868        return error;
1869}
1870
1871static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1872{
1873        int error;
1874
1875        if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1876                return error;
1877        inc_nlink(dir);
1878        return 0;
1879}
1880
1881static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1882                bool excl)
1883{
1884        return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1885}
1886
1887/*
1888 * Link a file..
1889 */
1890static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1891{
1892        struct inode *inode = old_dentry->d_inode;
1893        int ret;
1894
1895        /*
1896         * No ordinary (disk based) filesystem counts links as inodes;
1897         * but each new link needs a new dentry, pinning lowmem, and
1898         * tmpfs dentries cannot be pruned until they are unlinked.
1899         */
1900        ret = shmem_reserve_inode(inode->i_sb);
1901        if (ret)
1902                goto out;
1903
1904        dir->i_size += BOGO_DIRENT_SIZE;
1905        inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1906        inc_nlink(inode);
1907        ihold(inode);   /* New dentry reference */
1908        dget(dentry);           /* Extra pinning count for the created dentry */
1909        d_instantiate(dentry, inode);
1910out:
1911        return ret;
1912}
1913
1914static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1915{
1916        struct inode *inode = dentry->d_inode;
1917
1918        if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1919                shmem_free_inode(inode->i_sb);
1920
1921        dir->i_size -= BOGO_DIRENT_SIZE;
1922        inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1923        drop_nlink(inode);
1924        dput(dentry);   /* Undo the count from "create" - this does all the work */
1925        return 0;
1926}
1927
1928static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1929{
1930        if (!simple_empty(dentry))
1931                return -ENOTEMPTY;
1932
1933        drop_nlink(dentry->d_inode);
1934        drop_nlink(dir);
1935        return shmem_unlink(dir, dentry);
1936}
1937
1938/*
1939 * The VFS layer already does all the dentry stuff for rename,
1940 * we just have to decrement the usage count for the target if
1941 * it exists so that the VFS layer correctly free's it when it
1942 * gets overwritten.
1943 */
1944static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1945{
1946        struct inode *inode = old_dentry->d_inode;
1947        int they_are_dirs = S_ISDIR(inode->i_mode);
1948
1949        if (!simple_empty(new_dentry))
1950                return -ENOTEMPTY;
1951
1952        if (new_dentry->d_inode) {
1953                (void) shmem_unlink(new_dir, new_dentry);
1954                if (they_are_dirs)
1955                        drop_nlink(old_dir);
1956        } else if (they_are_dirs) {
1957                drop_nlink(old_dir);
1958                inc_nlink(new_dir);
1959        }
1960
1961        old_dir->i_size -= BOGO_DIRENT_SIZE;
1962        new_dir->i_size += BOGO_DIRENT_SIZE;
1963        old_dir->i_ctime = old_dir->i_mtime =
1964        new_dir->i_ctime = new_dir->i_mtime =
1965        inode->i_ctime = CURRENT_TIME;
1966        return 0;
1967}
1968
1969static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1970{
1971        int error;
1972        int len;
1973        struct inode *inode;
1974        struct page *page;
1975        char *kaddr;
1976        struct shmem_inode_info *info;
1977
1978        len = strlen(symname) + 1;
1979        if (len > PAGE_CACHE_SIZE)
1980                return -ENAMETOOLONG;
1981
1982        inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
1983        if (!inode)
1984                return -ENOSPC;
1985
1986        error = security_inode_init_security(inode, dir, &dentry->d_name,
1987                                             shmem_initxattrs, NULL);
1988        if (error) {
1989                if (error != -EOPNOTSUPP) {
1990                        iput(inode);
1991                        return error;
1992                }
1993                error = 0;
1994        }
1995
1996        info = SHMEM_I(inode);
1997        inode->i_size = len-1;
1998        if (len <= SHORT_SYMLINK_LEN) {
1999                info->symlink = kmemdup(symname, len, GFP_KERNEL);
2000                if (!info->symlink) {
2001                        iput(inode);
2002                        return -ENOMEM;
2003                }
2004                inode->i_op = &shmem_short_symlink_operations;
2005        } else {
2006                error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2007                if (error) {
2008                        iput(inode);
2009                        return error;
2010                }
2011                inode->i_mapping->a_ops = &shmem_aops;
2012                inode->i_op = &shmem_symlink_inode_operations;
2013                kaddr = kmap_atomic(page);
2014                memcpy(kaddr, symname, len);
2015                kunmap_atomic(kaddr);
2016                SetPageUptodate(page);
2017                set_page_dirty(page);
2018                unlock_page(page);
2019                page_cache_release(page);
2020        }
2021        dir->i_size += BOGO_DIRENT_SIZE;
2022        dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2023        d_instantiate(dentry, inode);
2024        dget(dentry);
2025        return 0;
2026}
2027
2028static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
2029{
2030        nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
2031        return NULL;
2032}
2033
2034static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
2035{
2036        struct page *page = NULL;
2037        int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2038        nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
2039        if (page)
2040                unlock_page(page);
2041        return page;
2042}
2043
2044static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2045{
2046        if (!IS_ERR(nd_get_link(nd))) {
2047                struct page *page = cookie;
2048                kunmap(page);
2049                mark_page_accessed(page);
2050                page_cache_release(page);
2051        }
2052}
2053
2054#ifdef CONFIG_TMPFS_XATTR
2055/*
2056 * Superblocks without xattr inode operations may get some security.* xattr
2057 * support from the LSM "for free". As soon as we have any other xattrs
2058 * like ACLs, we also need to implement the security.* handlers at
2059 * filesystem level, though.
2060 */
2061
2062/*
2063 * Allocate new xattr and copy in the value; but leave the name to callers.
2064 */
2065static struct shmem_xattr *shmem_xattr_alloc(const void *value, size_t size)
2066{
2067        struct shmem_xattr *new_xattr;
2068        size_t len;
2069
2070        /* wrap around? */
2071        len = sizeof(*new_xattr) + size;
2072        if (len <= sizeof(*new_xattr))
2073                return NULL;
2074
2075        new_xattr = kmalloc(len, GFP_KERNEL);
2076        if (!new_xattr)
2077                return NULL;
2078
2079        new_xattr->size = size;
2080        memcpy(new_xattr->value, value, size);
2081        return new_xattr;
2082}
2083
2084/*
2085 * Callback for security_inode_init_security() for acquiring xattrs.
2086 */
2087static int shmem_initxattrs(struct inode *inode,
2088                            const struct xattr *xattr_array,
2089                            void *fs_info)
2090{
2091        struct shmem_inode_info *info = SHMEM_I(inode);
2092        const struct xattr *xattr;
2093        struct shmem_xattr *new_xattr;
2094        size_t len;
2095
2096        for (xattr = xattr_array; xattr->name != NULL; xattr++) {
2097                new_xattr = shmem_xattr_alloc(xattr->value, xattr->value_len);
2098                if (!new_xattr)
2099                        return -ENOMEM;
2100
2101                len = strlen(xattr->name) + 1;
2102                new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2103                                          GFP_KERNEL);
2104                if (!new_xattr->name) {
2105                        kfree(new_xattr);
2106                        return -ENOMEM;
2107                }
2108
2109                memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2110                       XATTR_SECURITY_PREFIX_LEN);
2111                memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2112                       xattr->name, len);
2113
2114                spin_lock(&info->lock);
2115                list_add(&new_xattr->list, &info->xattr_list);
2116                spin_unlock(&info->lock);
2117        }
2118
2119        return 0;
2120}
2121
2122static int shmem_xattr_get(struct dentry *dentry, const char *name,
2123                           void *buffer, size_t size)
2124{
2125        struct shmem_inode_info *info;
2126        struct shmem_xattr *xattr;
2127        int ret = -ENODATA;
2128
2129        info = SHMEM_I(dentry->d_inode);
2130
2131        spin_lock(&info->lock);
2132        list_for_each_entry(xattr, &info->xattr_list, list) {
2133                if (strcmp(name, xattr->name))
2134                        continue;
2135
2136                ret = xattr->size;
2137                if (buffer) {
2138                        if (size < xattr->size)
2139                                ret = -ERANGE;
2140                        else
2141                                memcpy(buffer, xattr->value, xattr->size);
2142                }
2143                break;
2144        }
2145        spin_unlock(&info->lock);
2146        return ret;
2147}
2148
2149static int shmem_xattr_set(struct inode *inode, const char *name,
2150                           const void *value, size_t size, int flags)
2151{
2152        struct shmem_inode_info *info = SHMEM_I(inode);
2153        struct shmem_xattr *xattr;
2154        struct shmem_xattr *new_xattr = NULL;
2155        int err = 0;
2156
2157        /* value == NULL means remove */
2158        if (value) {
2159                new_xattr = shmem_xattr_alloc(value, size);
2160                if (!new_xattr)
2161                        return -ENOMEM;
2162
2163                new_xattr->name = kstrdup(name, GFP_KERNEL);
2164                if (!new_xattr->name) {
2165                        kfree(new_xattr);
2166                        return -ENOMEM;
2167                }
2168        }
2169
2170        spin_lock(&info->lock);
2171        list_for_each_entry(xattr, &info->xattr_list, list) {
2172                if (!strcmp(name, xattr->name)) {
2173                        if (flags & XATTR_CREATE) {
2174                                xattr = new_xattr;
2175                                err = -EEXIST;
2176                        } else if (new_xattr) {
2177                                list_replace(&xattr->list, &new_xattr->list);
2178                        } else {
2179                                list_del(&xattr->list);
2180                        }
2181                        goto out;
2182                }
2183        }
2184        if (flags & XATTR_REPLACE) {
2185                xattr = new_xattr;
2186                err = -ENODATA;
2187        } else {
2188                list_add(&new_xattr->list, &info->xattr_list);
2189                xattr = NULL;
2190        }
2191out:
2192        spin_unlock(&info->lock);
2193        if (xattr)
2194                kfree(xattr->name);
2195        kfree(xattr);
2196        return err;
2197}
2198
2199static const struct xattr_handler *shmem_xattr_handlers[] = {
2200#ifdef CONFIG_TMPFS_POSIX_ACL
2201        &generic_acl_access_handler,
2202        &generic_acl_default_handler,
2203#endif
2204        NULL
2205};
2206
2207static int shmem_xattr_validate(const char *name)
2208{
2209        struct { const char *prefix; size_t len; } arr[] = {
2210                { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2211                { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2212        };
2213        int i;
2214
2215        for (i = 0; i < ARRAY_SIZE(arr); i++) {
2216                size_t preflen = arr[i].len;
2217                if (strncmp(name, arr[i].prefix, preflen) == 0) {
2218                        if (!name[preflen])
2219                                return -EINVAL;
2220                        return 0;
2221                }
2222        }
2223        return -EOPNOTSUPP;
2224}
2225
2226static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2227                              void *buffer, size_t size)
2228{
2229        int err;
2230
2231        /*
2232         * If this is a request for a synthetic attribute in the system.*
2233         * namespace use the generic infrastructure to resolve a handler
2234         * for it via sb->s_xattr.
2235         */
2236        if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2237                return generic_getxattr(dentry, name, buffer, size);
2238
2239        err = shmem_xattr_validate(name);
2240        if (err)
2241                return err;
2242
2243        return shmem_xattr_get(dentry, name, buffer, size);
2244}
2245
2246static int shmem_setxattr(struct dentry *dentry, const char *name,
2247                          const void *value, size_t size, int flags)
2248{
2249        int err;
2250
2251        /*
2252         * If this is a request for a synthetic attribute in the system.*
2253         * namespace use the generic infrastructure to resolve a handler
2254         * for it via sb->s_xattr.
2255         */
2256        if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2257                return generic_setxattr(dentry, name, value, size, flags);
2258
2259        err = shmem_xattr_validate(name);
2260        if (err)
2261                return err;
2262
2263        if (size == 0)
2264                value = "";  /* empty EA, do not remove */
2265
2266        return shmem_xattr_set(dentry->d_inode, name, value, size, flags);
2267
2268}
2269
2270static int shmem_removexattr(struct dentry *dentry, const char *name)
2271{
2272        int err;
2273
2274        /*
2275         * If this is a request for a synthetic attribute in the system.*
2276         * namespace use the generic infrastructure to resolve a handler
2277         * for it via sb->s_xattr.
2278         */
2279        if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2280                return generic_removexattr(dentry, name);
2281
2282        err = shmem_xattr_validate(name);
2283        if (err)
2284                return err;
2285
2286        return shmem_xattr_set(dentry->d_inode, name, NULL, 0, XATTR_REPLACE);
2287}
2288
2289static bool xattr_is_trusted(const char *name)
2290{
2291        return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN);
2292}
2293
2294static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2295{
2296        bool trusted = capable(CAP_SYS_ADMIN);
2297        struct shmem_xattr *xattr;
2298        struct shmem_inode_info *info;
2299        size_t used = 0;
2300
2301        info = SHMEM_I(dentry->d_inode);
2302
2303        spin_lock(&info->lock);
2304        list_for_each_entry(xattr, &info->xattr_list, list) {
2305                size_t len;
2306
2307                /* skip "trusted." attributes for unprivileged callers */
2308                if (!trusted && xattr_is_trusted(xattr->name))
2309                        continue;
2310
2311                len = strlen(xattr->name) + 1;
2312                used += len;
2313                if (buffer) {
2314                        if (size < used) {
2315                                used = -ERANGE;
2316                                break;
2317                        }
2318                        memcpy(buffer, xattr->name, len);
2319                        buffer += len;
2320                }
2321        }
2322        spin_unlock(&info->lock);
2323
2324        return used;
2325}
2326#endif /* CONFIG_TMPFS_XATTR */
2327
2328static const struct inode_operations shmem_short_symlink_operations = {
2329        .readlink       = generic_readlink,
2330        .follow_link    = shmem_follow_short_symlink,
2331#ifdef CONFIG_TMPFS_XATTR
2332        .setxattr       = shmem_setxattr,
2333        .getxattr       = shmem_getxattr,
2334        .listxattr      = shmem_listxattr,
2335        .removexattr    = shmem_removexattr,
2336#endif
2337};
2338
2339static const struct inode_operations shmem_symlink_inode_operations = {
2340        .readlink       = generic_readlink,
2341        .follow_link    = shmem_follow_link,
2342        .put_link       = shmem_put_link,
2343#ifdef CONFIG_TMPFS_XATTR
2344        .setxattr       = shmem_setxattr,
2345        .getxattr       = shmem_getxattr,
2346        .listxattr      = shmem_listxattr,
2347        .removexattr    = shmem_removexattr,
2348#endif
2349};
2350
2351static struct dentry *shmem_get_parent(struct dentry *child)
2352{
2353        return ERR_PTR(-ESTALE);
2354}
2355
2356static int shmem_match(struct inode *ino, void *vfh)
2357{
2358        __u32 *fh = vfh;
2359        __u64 inum = fh[2];
2360        inum = (inum << 32) | fh[1];
2361        return ino->i_ino == inum && fh[0] == ino->i_generation;
2362}
2363
2364static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2365                struct fid *fid, int fh_len, int fh_type)
2366{
2367        struct inode *inode;
2368        struct dentry *dentry = NULL;
2369        u64 inum;
2370
2371        if (fh_len < 3)
2372                return NULL;
2373
2374        inum = fid->raw[2];
2375        inum = (inum << 32) | fid->raw[1];
2376
2377        inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2378                        shmem_match, fid->raw);
2379        if (inode) {
2380                dentry = d_find_alias(inode);
2381                iput(inode);
2382        }
2383
2384        return dentry;
2385}
2386
2387static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2388                                struct inode *parent)
2389{
2390        if (*len < 3) {
2391                *len = 3;
2392                return 255;
2393        }
2394
2395        if (inode_unhashed(inode)) {
2396                /* Unfortunately insert_inode_hash is not idempotent,
2397                 * so as we hash inodes here rather than at creation
2398                 * time, we need a lock to ensure we only try
2399                 * to do it once
2400                 */
2401                static DEFINE_SPINLOCK(lock);
2402                spin_lock(&lock);
2403                if (inode_unhashed(inode))
2404                        __insert_inode_hash(inode,
2405                                            inode->i_ino + inode->i_generation);
2406                spin_unlock(&lock);
2407        }
2408
2409        fh[0] = inode->i_generation;
2410        fh[1] = inode->i_ino;
2411        fh[2] = ((__u64)inode->i_ino) >> 32;
2412
2413        *len = 3;
2414        return 1;
2415}
2416
2417static const struct export_operations shmem_export_ops = {
2418        .get_parent     = shmem_get_parent,
2419        .encode_fh      = shmem_encode_fh,
2420        .fh_to_dentry   = shmem_fh_to_dentry,
2421};
2422
2423static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2424                               bool remount)
2425{
2426        char *this_char, *value, *rest;
2427        uid_t uid;
2428        gid_t gid;
2429
2430        while (options != NULL) {
2431                this_char = options;
2432                for (;;) {
2433                        /*
2434                         * NUL-terminate this option: unfortunately,
2435                         * mount options form a comma-separated list,
2436                         * but mpol's nodelist may also contain commas.
2437                         */
2438                        options = strchr(options, ',');
2439                        if (options == NULL)
2440                                break;
2441                        options++;
2442                        if (!isdigit(*options)) {
2443                                options[-1] = '\0';
2444                                break;
2445                        }
2446                }
2447                if (!*this_char)
2448                        continue;
2449                if ((value = strchr(this_char,'=')) != NULL) {
2450                        *value++ = 0;
2451                } else {
2452                        printk(KERN_ERR
2453                            "tmpfs: No value for mount option '%s'\n",
2454                            this_char);
2455                        return 1;
2456                }
2457
2458                if (!strcmp(this_char,"size")) {
2459                        unsigned long long size;
2460                        size = memparse(value,&rest);
2461                        if (*rest == '%') {
2462                                size <<= PAGE_SHIFT;
2463                                size *= totalram_pages;
2464                                do_div(size, 100);
2465                                rest++;
2466                        }
2467                        if (*rest)
2468                                goto bad_val;
2469                        sbinfo->max_blocks =
2470                                DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2471                } else if (!strcmp(this_char,"nr_blocks")) {
2472                        sbinfo->max_blocks = memparse(value, &rest);
2473                        if (*rest)
2474                                goto bad_val;
2475                } else if (!strcmp(this_char,"nr_inodes")) {
2476                        sbinfo->max_inodes = memparse(value, &rest);
2477                        if (*rest)
2478                                goto bad_val;
2479                } else if (!strcmp(this_char,"mode")) {
2480                        if (remount)
2481                                continue;
2482                        sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2483                        if (*rest)
2484                                goto bad_val;
2485                } else if (!strcmp(this_char,"uid")) {
2486                        if (remount)
2487                                continue;
2488                        uid = simple_strtoul(value, &rest, 0);
2489                        if (*rest)
2490                                goto bad_val;
2491                        sbinfo->uid = make_kuid(current_user_ns(), uid);
2492                        if (!uid_valid(sbinfo->uid))
2493                                goto bad_val;
2494                } else if (!strcmp(this_char,"gid")) {
2495                        if (remount)
2496                                continue;
2497                        gid = simple_strtoul(value, &rest, 0);
2498                        if (*rest)
2499                                goto bad_val;
2500                        sbinfo->gid = make_kgid(current_user_ns(), gid);
2501                        if (!gid_valid(sbinfo->gid))
2502                                goto bad_val;
2503                } else if (!strcmp(this_char,"mpol")) {
2504                        if (mpol_parse_str(value, &sbinfo->mpol, 1))
2505                                goto bad_val;
2506                } else {
2507                        printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2508                               this_char);
2509                        return 1;
2510                }
2511        }
2512        return 0;
2513
2514bad_val:
2515        printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2516               value, this_char);
2517        return 1;
2518
2519}
2520
2521static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2522{
2523        struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2524        struct shmem_sb_info config = *sbinfo;
2525        unsigned long inodes;
2526        int error = -EINVAL;
2527
2528        if (shmem_parse_options(data, &config, true))
2529                return error;
2530
2531        spin_lock(&sbinfo->stat_lock);
2532        inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2533        if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2534                goto out;
2535        if (config.max_inodes < inodes)
2536                goto out;
2537        /*
2538         * Those tests disallow limited->unlimited while any are in use;
2539         * but we must separately disallow unlimited->limited, because
2540         * in that case we have no record of how much is already in use.
2541         */
2542        if (config.max_blocks && !sbinfo->max_blocks)
2543                goto out;
2544        if (config.max_inodes && !sbinfo->max_inodes)
2545                goto out;
2546
2547        error = 0;
2548        sbinfo->max_blocks  = config.max_blocks;
2549        sbinfo->max_inodes  = config.max_inodes;
2550        sbinfo->free_inodes = config.max_inodes - inodes;
2551
2552        mpol_put(sbinfo->mpol);
2553        sbinfo->mpol        = config.mpol;      /* transfers initial ref */
2554out:
2555        spin_unlock(&sbinfo->stat_lock);
2556        return error;
2557}
2558
2559static int shmem_show_options(struct seq_file *seq, struct dentry *root)
2560{
2561        struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
2562
2563        if (sbinfo->max_blocks != shmem_default_max_blocks())
2564                seq_printf(seq, ",size=%luk",
2565                        sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2566        if (sbinfo->max_inodes != shmem_default_max_inodes())
2567                seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2568        if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2569                seq_printf(seq, ",mode=%03ho", sbinfo->mode);
2570        if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2571                seq_printf(seq, ",uid=%u",
2572                                from_kuid_munged(&init_user_ns, sbinfo->uid));
2573        if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2574                seq_printf(seq, ",gid=%u",
2575                                from_kgid_munged(&init_user_ns, sbinfo->gid));
2576        shmem_show_mpol(seq, sbinfo->mpol);
2577        return 0;
2578}
2579#endif /* CONFIG_TMPFS */
2580
2581static void shmem_put_super(struct super_block *sb)
2582{
2583        struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2584
2585        percpu_counter_destroy(&sbinfo->used_blocks);
2586        kfree(sbinfo);
2587        sb->s_fs_info = NULL;
2588}
2589
2590int shmem_fill_super(struct super_block *sb, void *data, int silent)
2591{
2592        struct inode *inode;
2593        struct shmem_sb_info *sbinfo;
2594        int err = -ENOMEM;
2595
2596        /* Round up to L1_CACHE_BYTES to resist false sharing */
2597        sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2598                                L1_CACHE_BYTES), GFP_KERNEL);
2599        if (!sbinfo)
2600                return -ENOMEM;
2601
2602        sbinfo->mode = S_IRWXUGO | S_ISVTX;
2603        sbinfo->uid = current_fsuid();
2604        sbinfo->gid = current_fsgid();
2605        sb->s_fs_info = sbinfo;
2606
2607#ifdef CONFIG_TMPFS
2608        /*
2609         * Per default we only allow half of the physical ram per
2610         * tmpfs instance, limiting inodes to one per page of lowmem;
2611         * but the internal instance is left unlimited.
2612         */
2613        if (!(sb->s_flags & MS_NOUSER)) {
2614                sbinfo->max_blocks = shmem_default_max_blocks();
2615                sbinfo->max_inodes = shmem_default_max_inodes();
2616                if (shmem_parse_options(data, sbinfo, false)) {
2617                        err = -EINVAL;
2618                        goto failed;
2619                }
2620        }
2621        sb->s_export_op = &shmem_export_ops;
2622        sb->s_flags |= MS_NOSEC;
2623#else
2624        sb->s_flags |= MS_NOUSER;
2625#endif
2626
2627        spin_lock_init(&sbinfo->stat_lock);
2628        if (percpu_counter_init(&sbinfo->used_blocks, 0))
2629                goto failed;
2630        sbinfo->free_inodes = sbinfo->max_inodes;
2631
2632        sb->s_maxbytes = MAX_LFS_FILESIZE;
2633        sb->s_blocksize = PAGE_CACHE_SIZE;
2634        sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2635        sb->s_magic = TMPFS_MAGIC;
2636        sb->s_op = &shmem_ops;
2637        sb->s_time_gran = 1;
2638#ifdef CONFIG_TMPFS_XATTR
2639        sb->s_xattr = shmem_xattr_handlers;
2640#endif
2641#ifdef CONFIG_TMPFS_POSIX_ACL
2642        sb->s_flags |= MS_POSIXACL;
2643#endif
2644
2645        inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2646        if (!inode)
2647                goto failed;
2648        inode->i_uid = sbinfo->uid;
2649        inode->i_gid = sbinfo->gid;
2650        sb->s_root = d_make_root(inode);
2651        if (!sb->s_root)
2652                goto failed;
2653        return 0;
2654
2655failed:
2656        shmem_put_super(sb);
2657        return err;
2658}
2659
2660static struct kmem_cache *shmem_inode_cachep;
2661
2662static struct inode *shmem_alloc_inode(struct super_block *sb)
2663{
2664        struct shmem_inode_info *info;
2665        info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2666        if (!info)
2667                return NULL;
2668        return &info->vfs_inode;
2669}
2670
2671static void shmem_destroy_callback(struct rcu_head *head)
2672{
2673        struct inode *inode = container_of(head, struct inode, i_rcu);
2674        kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2675}
2676
2677static void shmem_destroy_inode(struct inode *inode)
2678{
2679        if (S_ISREG(inode->i_mode))
2680                mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2681        call_rcu(&inode->i_rcu, shmem_destroy_callback);
2682}
2683
2684static void shmem_init_inode(void *foo)
2685{
2686        struct shmem_inode_info *info = foo;
2687        inode_init_once(&info->vfs_inode);
2688}
2689
2690static int shmem_init_inodecache(void)
2691{
2692        shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2693                                sizeof(struct shmem_inode_info),
2694                                0, SLAB_PANIC, shmem_init_inode);
2695        return 0;
2696}
2697
2698static void shmem_destroy_inodecache(void)
2699{
2700        kmem_cache_destroy(shmem_inode_cachep);
2701}
2702
2703static const struct address_space_operations shmem_aops = {
2704        .writepage      = shmem_writepage,
2705        .set_page_dirty = __set_page_dirty_no_writeback,
2706#ifdef CONFIG_TMPFS
2707        .write_begin    = shmem_write_begin,
2708        .write_end      = shmem_write_end,
2709#endif
2710        .migratepage    = migrate_page,
2711        .error_remove_page = generic_error_remove_page,
2712};
2713
2714static const struct file_operations shmem_file_operations = {
2715        .mmap           = shmem_mmap,
2716#ifdef CONFIG_TMPFS
2717        .llseek         = generic_file_llseek,
2718        .read           = do_sync_read,
2719        .write          = do_sync_write,
2720        .aio_read       = shmem_file_aio_read,
2721        .aio_write      = generic_file_aio_write,
2722        .fsync          = noop_fsync,
2723        .splice_read    = shmem_file_splice_read,
2724        .splice_write   = generic_file_splice_write,
2725        .fallocate      = shmem_fallocate,
2726#endif
2727};
2728
2729static const struct inode_operations shmem_inode_operations = {
2730        .setattr        = shmem_setattr,
2731#ifdef CONFIG_TMPFS_XATTR
2732        .setxattr       = shmem_setxattr,
2733        .getxattr       = shmem_getxattr,
2734        .listxattr      = shmem_listxattr,
2735        .removexattr    = shmem_removexattr,
2736#endif
2737};
2738
2739static const struct inode_operations shmem_dir_inode_operations = {
2740#ifdef CONFIG_TMPFS
2741        .create         = shmem_create,
2742        .lookup         = simple_lookup,
2743        .link           = shmem_link,
2744        .unlink         = shmem_unlink,
2745        .symlink        = shmem_symlink,
2746        .mkdir          = shmem_mkdir,
2747        .rmdir          = shmem_rmdir,
2748        .mknod          = shmem_mknod,
2749        .rename         = shmem_rename,
2750#endif
2751#ifdef CONFIG_TMPFS_XATTR
2752        .setxattr       = shmem_setxattr,
2753        .getxattr       = shmem_getxattr,
2754        .listxattr      = shmem_listxattr,
2755        .removexattr    = shmem_removexattr,
2756#endif
2757#ifdef CONFIG_TMPFS_POSIX_ACL
2758        .setattr        = shmem_setattr,
2759#endif
2760};
2761
2762static const struct inode_operations shmem_special_inode_operations = {
2763#ifdef CONFIG_TMPFS_XATTR
2764        .setxattr       = shmem_setxattr,
2765        .getxattr       = shmem_getxattr,
2766        .listxattr      = shmem_listxattr,
2767        .removexattr    = shmem_removexattr,
2768#endif
2769#ifdef CONFIG_TMPFS_POSIX_ACL
2770        .setattr        = shmem_setattr,
2771#endif
2772};
2773
2774static const struct super_operations shmem_ops = {
2775        .alloc_inode    = shmem_alloc_inode,
2776        .destroy_inode  = shmem_destroy_inode,
2777#ifdef CONFIG_TMPFS
2778        .statfs         = shmem_statfs,
2779        .remount_fs     = shmem_remount_fs,
2780        .show_options   = shmem_show_options,
2781#endif
2782        .evict_inode    = shmem_evict_inode,
2783        .drop_inode     = generic_delete_inode,
2784        .put_super      = shmem_put_super,
2785};
2786
2787static const struct vm_operations_struct shmem_vm_ops = {
2788        .fault          = shmem_fault,
2789#ifdef CONFIG_NUMA
2790        .set_policy     = shmem_set_policy,
2791        .get_policy     = shmem_get_policy,
2792#endif
2793};
2794
2795static struct dentry *shmem_mount(struct file_system_type *fs_type,
2796        int flags, const char *dev_name, void *data)
2797{
2798        return mount_nodev(fs_type, flags, data, shmem_fill_super);
2799}
2800
2801static struct file_system_type shmem_fs_type = {
2802        .owner          = THIS_MODULE,
2803        .name           = "tmpfs",
2804        .mount          = shmem_mount,
2805        .kill_sb        = kill_litter_super,
2806};
2807
2808int __init shmem_init(void)
2809{
2810        int error;
2811
2812        error = bdi_init(&shmem_backing_dev_info);
2813        if (error)
2814                goto out4;
2815
2816        error = shmem_init_inodecache();
2817        if (error)
2818                goto out3;
2819
2820        error = register_filesystem(&shmem_fs_type);
2821        if (error) {
2822                printk(KERN_ERR "Could not register tmpfs\n");
2823                goto out2;
2824        }
2825
2826        shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER,
2827                                 shmem_fs_type.name, NULL);
2828        if (IS_ERR(shm_mnt)) {
2829                error = PTR_ERR(shm_mnt);
2830                printk(KERN_ERR "Could not kern_mount tmpfs\n");
2831                goto out1;
2832        }
2833        return 0;
2834
2835out1:
2836        unregister_filesystem(&shmem_fs_type);
2837out2:
2838        shmem_destroy_inodecache();
2839out3:
2840        bdi_destroy(&shmem_backing_dev_info);
2841out4:
2842        shm_mnt = ERR_PTR(error);
2843        return error;
2844}
2845
2846#else /* !CONFIG_SHMEM */
2847
2848/*
2849 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2850 *
2851 * This is intended for small system where the benefits of the full
2852 * shmem code (swap-backed and resource-limited) are outweighed by
2853 * their complexity. On systems without swap this code should be
2854 * effectively equivalent, but much lighter weight.
2855 */
2856
2857#include <linux/ramfs.h>
2858
2859static struct file_system_type shmem_fs_type = {
2860        .name           = "tmpfs",
2861        .mount          = ramfs_mount,
2862        .kill_sb        = kill_litter_super,
2863};
2864
2865int __init shmem_init(void)
2866{
2867        BUG_ON(register_filesystem(&shmem_fs_type) != 0);
2868
2869        shm_mnt = kern_mount(&shmem_fs_type);
2870        BUG_ON(IS_ERR(shm_mnt));
2871
2872        return 0;
2873}
2874
2875int shmem_unuse(swp_entry_t swap, struct page *page)
2876{
2877        return 0;
2878}
2879
2880int shmem_lock(struct file *file, int lock, struct user_struct *user)
2881{
2882        return 0;
2883}
2884
2885void shmem_unlock_mapping(struct address_space *mapping)
2886{
2887}
2888
2889void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
2890{
2891        truncate_inode_pages_range(inode->i_mapping, lstart, lend);
2892}
2893EXPORT_SYMBOL_GPL(shmem_truncate_range);
2894
2895#define shmem_vm_ops                            generic_file_vm_ops
2896#define shmem_file_operations                   ramfs_file_operations
2897#define shmem_get_inode(sb, dir, mode, dev, flags)      ramfs_get_inode(sb, dir, mode, dev)
2898#define shmem_acct_size(flags, size)            0
2899#define shmem_unacct_size(flags, size)          do {} while (0)
2900
2901#endif /* CONFIG_SHMEM */
2902
2903/* common code */
2904
2905/**
2906 * shmem_file_setup - get an unlinked file living in tmpfs
2907 * @name: name for dentry (to be seen in /proc/<pid>/maps
2908 * @size: size to be set for the file
2909 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2910 */
2911struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2912{
2913        int error;
2914        struct file *file;
2915        struct inode *inode;
2916        struct path path;
2917        struct dentry *root;
2918        struct qstr this;
2919
2920        if (IS_ERR(shm_mnt))
2921                return (void *)shm_mnt;
2922
2923        if (size < 0 || size > MAX_LFS_FILESIZE)
2924                return ERR_PTR(-EINVAL);
2925
2926        if (shmem_acct_size(flags, size))
2927                return ERR_PTR(-ENOMEM);
2928
2929        error = -ENOMEM;
2930        this.name = name;
2931        this.len = strlen(name);
2932        this.hash = 0; /* will go */
2933        root = shm_mnt->mnt_root;
2934        path.dentry = d_alloc(root, &this);
2935        if (!path.dentry)
2936                goto put_memory;
2937        path.mnt = mntget(shm_mnt);
2938
2939        error = -ENOSPC;
2940        inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2941        if (!inode)
2942                goto put_dentry;
2943
2944        d_instantiate(path.dentry, inode);
2945        inode->i_size = size;
2946        clear_nlink(inode);     /* It is unlinked */
2947#ifndef CONFIG_MMU
2948        error = ramfs_nommu_expand_for_mapping(inode, size);
2949        if (error)
2950                goto put_dentry;
2951#endif
2952
2953        error = -ENFILE;
2954        file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2955                  &shmem_file_operations);
2956        if (!file)
2957                goto put_dentry;
2958
2959        return file;
2960
2961put_dentry:
2962        path_put(&path);
2963put_memory:
2964        shmem_unacct_size(flags, size);
2965        return ERR_PTR(error);
2966}
2967EXPORT_SYMBOL_GPL(shmem_file_setup);
2968
2969/**
2970 * shmem_zero_setup - setup a shared anonymous mapping
2971 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2972 */
2973int shmem_zero_setup(struct vm_area_struct *vma)
2974{
2975        struct file *file;
2976        loff_t size = vma->vm_end - vma->vm_start;
2977
2978        file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2979        if (IS_ERR(file))
2980                return PTR_ERR(file);
2981
2982        if (vma->vm_file)
2983                fput(vma->vm_file);
2984        vma->vm_file = file;
2985        vma->vm_ops = &shmem_vm_ops;
2986        vma->vm_flags |= VM_CAN_NONLINEAR;
2987        return 0;
2988}
2989
2990/**
2991 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2992 * @mapping:    the page's address_space
2993 * @index:      the page index
2994 * @gfp:        the page allocator flags to use if allocating
2995 *
2996 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2997 * with any new page allocations done using the specified allocation flags.
2998 * But read_cache_page_gfp() uses the ->readpage() method: which does not
2999 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3000 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3001 *
3002 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3003 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
3004 */
3005struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
3006                                         pgoff_t index, gfp_t gfp)
3007{
3008#ifdef CONFIG_SHMEM
3009        struct inode *inode = mapping->host;
3010        struct page *page;
3011        int error;
3012
3013        BUG_ON(mapping->a_ops != &shmem_aops);
3014        error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
3015        if (error)
3016                page = ERR_PTR(error);
3017        else
3018                unlock_page(page);
3019        return page;
3020#else
3021        /*
3022         * The tiny !SHMEM case uses ramfs without swap
3023         */
3024        return read_cache_page_gfp(mapping, index, gfp);
3025#endif
3026}
3027EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);
3028
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.