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/module.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/splice.h>
  57#include <linux/security.h>
  58#include <linux/swapops.h>
  59#include <linux/mempolicy.h>
  60#include <linux/namei.h>
  61#include <linux/ctype.h>
  62#include <linux/migrate.h>
  63#include <linux/highmem.h>
  64#include <linux/seq_file.h>
  65#include <linux/magic.h>
  66
  67#include <asm/uaccess.h>
  68#include <asm/pgtable.h>
  69
  70#define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
  71#define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
  72
  73/* Pretend that each entry is of this size in directory's i_size */
  74#define BOGO_DIRENT_SIZE 20
  75
  76/* Symlink up to this size is kmalloc'ed instead of using a swappable page */
  77#define SHORT_SYMLINK_LEN 128
  78
  79struct shmem_xattr {
  80        struct list_head list;  /* anchored by shmem_inode_info->xattr_list */
  81        char *name;             /* xattr name */
  82        size_t size;
  83        char value[0];
  84};
  85
  86/* Flag allocation requirements to shmem_getpage */
  87enum sgp_type {
  88        SGP_READ,       /* don't exceed i_size, don't allocate page */
  89        SGP_CACHE,      /* don't exceed i_size, may allocate page */
  90        SGP_DIRTY,      /* like SGP_CACHE, but set new page dirty */
  91        SGP_WRITE,      /* may exceed i_size, may allocate page */
  92};
  93
  94#ifdef CONFIG_TMPFS
  95static unsigned long shmem_default_max_blocks(void)
  96{
  97        return totalram_pages / 2;
  98}
  99
 100static unsigned long shmem_default_max_inodes(void)
 101{
 102        return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
 103}
 104#endif
 105
 106static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
 107        struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
 108
 109static inline int shmem_getpage(struct inode *inode, pgoff_t index,
 110        struct page **pagep, enum sgp_type sgp, int *fault_type)
 111{
 112        return shmem_getpage_gfp(inode, index, pagep, sgp,
 113                        mapping_gfp_mask(inode->i_mapping), fault_type);
 114}
 115
 116static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
 117{
 118        return sb->s_fs_info;
 119}
 120
 121/*
 122 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
 123 * for shared memory and for shared anonymous (/dev/zero) mappings
 124 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
 125 * consistent with the pre-accounting of private mappings ...
 126 */
 127static inline int shmem_acct_size(unsigned long flags, loff_t size)
 128{
 129        return (flags & VM_NORESERVE) ?
 130                0 : security_vm_enough_memory_kern(VM_ACCT(size));
 131}
 132
 133static inline void shmem_unacct_size(unsigned long flags, loff_t size)
 134{
 135        if (!(flags & VM_NORESERVE))
 136                vm_unacct_memory(VM_ACCT(size));
 137}
 138
 139/*
 140 * ... whereas tmpfs objects are accounted incrementally as
 141 * pages are allocated, in order to allow huge sparse files.
 142 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
 143 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
 144 */
 145static inline int shmem_acct_block(unsigned long flags)
 146{
 147        return (flags & VM_NORESERVE) ?
 148                security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE)) : 0;
 149}
 150
 151static inline void shmem_unacct_blocks(unsigned long flags, long pages)
 152{
 153        if (flags & VM_NORESERVE)
 154                vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
 155}
 156
 157static const struct super_operations shmem_ops;
 158static const struct address_space_operations shmem_aops;
 159static const struct file_operations shmem_file_operations;
 160static const struct inode_operations shmem_inode_operations;
 161static const struct inode_operations shmem_dir_inode_operations;
 162static const struct inode_operations shmem_special_inode_operations;
 163static const struct vm_operations_struct shmem_vm_ops;
 164
 165static struct backing_dev_info shmem_backing_dev_info  __read_mostly = {
 166        .ra_pages       = 0,    /* No readahead */
 167        .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
 168};
 169
 170static LIST_HEAD(shmem_swaplist);
 171static DEFINE_MUTEX(shmem_swaplist_mutex);
 172
 173static int shmem_reserve_inode(struct super_block *sb)
 174{
 175        struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 176        if (sbinfo->max_inodes) {
 177                spin_lock(&sbinfo->stat_lock);
 178                if (!sbinfo->free_inodes) {
 179                        spin_unlock(&sbinfo->stat_lock);
 180                        return -ENOSPC;
 181                }
 182                sbinfo->free_inodes--;
 183                spin_unlock(&sbinfo->stat_lock);
 184        }
 185        return 0;
 186}
 187
 188static void shmem_free_inode(struct super_block *sb)
 189{
 190        struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 191        if (sbinfo->max_inodes) {
 192                spin_lock(&sbinfo->stat_lock);
 193                sbinfo->free_inodes++;
 194                spin_unlock(&sbinfo->stat_lock);
 195        }
 196}
 197
 198/**
 199 * shmem_recalc_inode - recalculate the block usage of an inode
 200 * @inode: inode to recalc
 201 *
 202 * We have to calculate the free blocks since the mm can drop
 203 * undirtied hole pages behind our back.
 204 *
 205 * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
 206 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
 207 *
 208 * It has to be called with the spinlock held.
 209 */
 210static void shmem_recalc_inode(struct inode *inode)
 211{
 212        struct shmem_inode_info *info = SHMEM_I(inode);
 213        long freed;
 214
 215        freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
 216        if (freed > 0) {
 217                struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 218                if (sbinfo->max_blocks)
 219                        percpu_counter_add(&sbinfo->used_blocks, -freed);
 220                info->alloced -= freed;
 221                inode->i_blocks -= freed * BLOCKS_PER_PAGE;
 222                shmem_unacct_blocks(info->flags, freed);
 223        }
 224}
 225
 226/*
 227 * Replace item expected in radix tree by a new item, while holding tree lock.
 228 */
 229static int shmem_radix_tree_replace(struct address_space *mapping,
 230                        pgoff_t index, void *expected, void *replacement)
 231{
 232        void **pslot;
 233        void *item = NULL;
 234
 235        VM_BUG_ON(!expected);
 236        pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
 237        if (pslot)
 238                item = radix_tree_deref_slot_protected(pslot,
 239                                                        &mapping->tree_lock);
 240        if (item != expected)
 241                return -ENOENT;
 242        if (replacement)
 243                radix_tree_replace_slot(pslot, replacement);
 244        else
 245                radix_tree_delete(&mapping->page_tree, index);
 246        return 0;
 247}
 248
 249/*
 250 * Like add_to_page_cache_locked, but error if expected item has gone.
 251 */
 252static int shmem_add_to_page_cache(struct page *page,
 253                                   struct address_space *mapping,
 254                                   pgoff_t index, gfp_t gfp, void *expected)
 255{
 256        int error = 0;
 257
 258        VM_BUG_ON(!PageLocked(page));
 259        VM_BUG_ON(!PageSwapBacked(page));
 260
 261        if (!expected)
 262                error = radix_tree_preload(gfp & GFP_RECLAIM_MASK);
 263        if (!error) {
 264                page_cache_get(page);
 265                page->mapping = mapping;
 266                page->index = index;
 267
 268                spin_lock_irq(&mapping->tree_lock);
 269                if (!expected)
 270                        error = radix_tree_insert(&mapping->page_tree,
 271                                                        index, page);
 272                else
 273                        error = shmem_radix_tree_replace(mapping, index,
 274                                                        expected, page);
 275                if (!error) {
 276                        mapping->nrpages++;
 277                        __inc_zone_page_state(page, NR_FILE_PAGES);
 278                        __inc_zone_page_state(page, NR_SHMEM);
 279                        spin_unlock_irq(&mapping->tree_lock);
 280                } else {
 281                        page->mapping = NULL;
 282                        spin_unlock_irq(&mapping->tree_lock);
 283                        page_cache_release(page);
 284                }
 285                if (!expected)
 286                        radix_tree_preload_end();
 287        }
 288        if (error)
 289                mem_cgroup_uncharge_cache_page(page);
 290        return error;
 291}
 292
 293/*
 294 * Like delete_from_page_cache, but substitutes swap for page.
 295 */
 296static void shmem_delete_from_page_cache(struct page *page, void *radswap)
 297{
 298        struct address_space *mapping = page->mapping;
 299        int error;
 300
 301        spin_lock_irq(&mapping->tree_lock);
 302        error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
 303        page->mapping = NULL;
 304        mapping->nrpages--;
 305        __dec_zone_page_state(page, NR_FILE_PAGES);
 306        __dec_zone_page_state(page, NR_SHMEM);
 307        spin_unlock_irq(&mapping->tree_lock);
 308        page_cache_release(page);
 309        BUG_ON(error);
 310}
 311
 312/*
 313 * Like find_get_pages, but collecting swap entries as well as pages.
 314 */
 315static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping,
 316                                        pgoff_t start, unsigned int nr_pages,
 317                                        struct page **pages, pgoff_t *indices)
 318{
 319        unsigned int i;
 320        unsigned int ret;
 321        unsigned int nr_found;
 322
 323        rcu_read_lock();
 324restart:
 325        nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
 326                                (void ***)pages, indices, start, nr_pages);
 327        ret = 0;
 328        for (i = 0; i < nr_found; i++) {
 329                struct page *page;
 330repeat:
 331                page = radix_tree_deref_slot((void **)pages[i]);
 332                if (unlikely(!page))
 333                        continue;
 334                if (radix_tree_exception(page)) {
 335                        if (radix_tree_deref_retry(page))
 336                                goto restart;
 337                        /*
 338                         * Otherwise, we must be storing a swap entry
 339                         * here as an exceptional entry: so return it
 340                         * without attempting to raise page count.
 341                         */
 342                        goto export;
 343                }
 344                if (!page_cache_get_speculative(page))
 345                        goto repeat;
 346
 347                /* Has the page moved? */
 348                if (unlikely(page != *((void **)pages[i]))) {
 349                        page_cache_release(page);
 350                        goto repeat;
 351                }
 352export:
 353                indices[ret] = indices[i];
 354                pages[ret] = page;
 355                ret++;
 356        }
 357        if (unlikely(!ret && nr_found))
 358                goto restart;
 359        rcu_read_unlock();
 360        return ret;
 361}
 362
 363/*
 364 * Remove swap entry from radix tree, free the swap and its page cache.
 365 */
 366static int shmem_free_swap(struct address_space *mapping,
 367                           pgoff_t index, void *radswap)
 368{
 369        int error;
 370
 371        spin_lock_irq(&mapping->tree_lock);
 372        error = shmem_radix_tree_replace(mapping, index, radswap, NULL);
 373        spin_unlock_irq(&mapping->tree_lock);
 374        if (!error)
 375                free_swap_and_cache(radix_to_swp_entry(radswap));
 376        return error;
 377}
 378
 379/*
 380 * Pagevec may contain swap entries, so shuffle up pages before releasing.
 381 */
 382static void shmem_pagevec_release(struct pagevec *pvec)
 383{
 384        int i, j;
 385
 386        for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
 387                struct page *page = pvec->pages[i];
 388                if (!radix_tree_exceptional_entry(page))
 389                        pvec->pages[j++] = page;
 390        }
 391        pvec->nr = j;
 392        pagevec_release(pvec);
 393}
 394
 395/*
 396 * Remove range of pages and swap entries from radix tree, and free them.
 397 */
 398void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
 399{
 400        struct address_space *mapping = inode->i_mapping;
 401        struct shmem_inode_info *info = SHMEM_I(inode);
 402        pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 403        unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
 404        pgoff_t end = (lend >> PAGE_CACHE_SHIFT);
 405        struct pagevec pvec;
 406        pgoff_t indices[PAGEVEC_SIZE];
 407        long nr_swaps_freed = 0;
 408        pgoff_t index;
 409        int i;
 410
 411        BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
 412
 413        pagevec_init(&pvec, 0);
 414        index = start;
 415        while (index <= end) {
 416                pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
 417                        min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
 418                                                        pvec.pages, indices);
 419                if (!pvec.nr)
 420                        break;
 421                mem_cgroup_uncharge_start();
 422                for (i = 0; i < pagevec_count(&pvec); i++) {
 423                        struct page *page = pvec.pages[i];
 424
 425                        index = indices[i];
 426                        if (index > end)
 427                                break;
 428
 429                        if (radix_tree_exceptional_entry(page)) {
 430                                nr_swaps_freed += !shmem_free_swap(mapping,
 431                                                                index, page);
 432                                continue;
 433                        }
 434
 435                        if (!trylock_page(page))
 436                                continue;
 437                        if (page->mapping == mapping) {
 438                                VM_BUG_ON(PageWriteback(page));
 439                                truncate_inode_page(mapping, page);
 440                        }
 441                        unlock_page(page);
 442                }
 443                shmem_pagevec_release(&pvec);
 444                mem_cgroup_uncharge_end();
 445                cond_resched();
 446                index++;
 447        }
 448
 449        if (partial) {
 450                struct page *page = NULL;
 451                shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
 452                if (page) {
 453                        zero_user_segment(page, partial, PAGE_CACHE_SIZE);
 454                        set_page_dirty(page);
 455                        unlock_page(page);
 456                        page_cache_release(page);
 457                }
 458        }
 459
 460        index = start;
 461        for ( ; ; ) {
 462                cond_resched();
 463                pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
 464                        min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
 465                                                        pvec.pages, indices);
 466                if (!pvec.nr) {
 467                        if (index == start)
 468                                break;
 469                        index = start;
 470                        continue;
 471                }
 472                if (index == start && indices[0] > end) {
 473                        shmem_pagevec_release(&pvec);
 474                        break;
 475                }
 476                mem_cgroup_uncharge_start();
 477                for (i = 0; i < pagevec_count(&pvec); i++) {
 478                        struct page *page = pvec.pages[i];
 479
 480                        index = indices[i];
 481                        if (index > end)
 482                                break;
 483
 484                        if (radix_tree_exceptional_entry(page)) {
 485                                nr_swaps_freed += !shmem_free_swap(mapping,
 486                                                                index, page);
 487                                continue;
 488                        }
 489
 490                        lock_page(page);
 491                        if (page->mapping == mapping) {
 492                                VM_BUG_ON(PageWriteback(page));
 493                                truncate_inode_page(mapping, page);
 494                        }
 495                        unlock_page(page);
 496                }
 497                shmem_pagevec_release(&pvec);
 498                mem_cgroup_uncharge_end();
 499                index++;
 500        }
 501
 502        spin_lock(&info->lock);
 503        info->swapped -= nr_swaps_freed;
 504        shmem_recalc_inode(inode);
 505        spin_unlock(&info->lock);
 506
 507        inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 508}
 509EXPORT_SYMBOL_GPL(shmem_truncate_range);
 510
 511static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
 512{
 513        struct inode *inode = dentry->d_inode;
 514        int error;
 515
 516        error = inode_change_ok(inode, attr);
 517        if (error)
 518                return error;
 519
 520        if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
 521                loff_t oldsize = inode->i_size;
 522                loff_t newsize = attr->ia_size;
 523
 524                if (newsize != oldsize) {
 525                        i_size_write(inode, newsize);
 526                        inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 527                }
 528                if (newsize < oldsize) {
 529                        loff_t holebegin = round_up(newsize, PAGE_SIZE);
 530                        unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
 531                        shmem_truncate_range(inode, newsize, (loff_t)-1);
 532                        /* unmap again to remove racily COWed private pages */
 533                        unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
 534                }
 535        }
 536
 537        setattr_copy(inode, attr);
 538#ifdef CONFIG_TMPFS_POSIX_ACL
 539        if (attr->ia_valid & ATTR_MODE)
 540                error = generic_acl_chmod(inode);
 541#endif
 542        return error;
 543}
 544
 545static void shmem_evict_inode(struct inode *inode)
 546{
 547        struct shmem_inode_info *info = SHMEM_I(inode);
 548        struct shmem_xattr *xattr, *nxattr;
 549
 550        if (inode->i_mapping->a_ops == &shmem_aops) {
 551                shmem_unacct_size(info->flags, inode->i_size);
 552                inode->i_size = 0;
 553                shmem_truncate_range(inode, 0, (loff_t)-1);
 554                if (!list_empty(&info->swaplist)) {
 555                        mutex_lock(&shmem_swaplist_mutex);
 556                        list_del_init(&info->swaplist);
 557                        mutex_unlock(&shmem_swaplist_mutex);
 558                }
 559        } else
 560                kfree(info->symlink);
 561
 562        list_for_each_entry_safe(xattr, nxattr, &info->xattr_list, list) {
 563                kfree(xattr->name);
 564                kfree(xattr);
 565        }
 566        BUG_ON(inode->i_blocks);
 567        shmem_free_inode(inode->i_sb);
 568        end_writeback(inode);
 569}
 570
 571/*
 572 * If swap found in inode, free it and move page from swapcache to filecache.
 573 */
 574static int shmem_unuse_inode(struct shmem_inode_info *info,
 575                             swp_entry_t swap, struct page *page)
 576{
 577        struct address_space *mapping = info->vfs_inode.i_mapping;
 578        void *radswap;
 579        pgoff_t index;
 580        int error;
 581
 582        radswap = swp_to_radix_entry(swap);
 583        index = radix_tree_locate_item(&mapping->page_tree, radswap);
 584        if (index == -1)
 585                return 0;
 586
 587        /*
 588         * Move _head_ to start search for next from here.
 589         * But be careful: shmem_evict_inode checks list_empty without taking
 590         * mutex, and there's an instant in list_move_tail when info->swaplist
 591         * would appear empty, if it were the only one on shmem_swaplist.
 592         */
 593        if (shmem_swaplist.next != &info->swaplist)
 594                list_move_tail(&shmem_swaplist, &info->swaplist);
 595
 596        /*
 597         * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
 598         * but also to hold up shmem_evict_inode(): so inode cannot be freed
 599         * beneath us (pagelock doesn't help until the page is in pagecache).
 600         */
 601        error = shmem_add_to_page_cache(page, mapping, index,
 602                                                GFP_NOWAIT, radswap);
 603        /* which does mem_cgroup_uncharge_cache_page on error */
 604
 605        if (error != -ENOMEM) {
 606                /*
 607                 * Truncation and eviction use free_swap_and_cache(), which
 608                 * only does trylock page: if we raced, best clean up here.
 609                 */
 610                delete_from_swap_cache(page);
 611                set_page_dirty(page);
 612                if (!error) {
 613                        spin_lock(&info->lock);
 614                        info->swapped--;
 615                        spin_unlock(&info->lock);
 616                        swap_free(swap);
 617                }
 618                error = 1;      /* not an error, but entry was found */
 619        }
 620        return error;
 621}
 622
 623/*
 624 * Search through swapped inodes to find and replace swap by page.
 625 */
 626int shmem_unuse(swp_entry_t swap, struct page *page)
 627{
 628        struct list_head *this, *next;
 629        struct shmem_inode_info *info;
 630        int found = 0;
 631        int error;
 632
 633        /*
 634         * Charge page using GFP_KERNEL while we can wait, before taking
 635         * the shmem_swaplist_mutex which might hold up shmem_writepage().
 636         * Charged back to the user (not to caller) when swap account is used.
 637         */
 638        error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
 639        if (error)
 640                goto out;
 641        /* No radix_tree_preload: swap entry keeps a place for page in tree */
 642
 643        mutex_lock(&shmem_swaplist_mutex);
 644        list_for_each_safe(this, next, &shmem_swaplist) {
 645                info = list_entry(this, struct shmem_inode_info, swaplist);
 646                if (info->swapped)
 647                        found = shmem_unuse_inode(info, swap, page);
 648                else
 649                        list_del_init(&info->swaplist);
 650                cond_resched();
 651                if (found)
 652                        break;
 653        }
 654        mutex_unlock(&shmem_swaplist_mutex);
 655
 656        if (!found)
 657                mem_cgroup_uncharge_cache_page(page);
 658        if (found < 0)
 659                error = found;
 660out:
 661        unlock_page(page);
 662        page_cache_release(page);
 663        return error;
 664}
 665
 666/*
 667 * Move the page from the page cache to the swap cache.
 668 */
 669static int shmem_writepage(struct page *page, struct writeback_control *wbc)
 670{
 671        struct shmem_inode_info *info;
 672        struct address_space *mapping;
 673        struct inode *inode;
 674        swp_entry_t swap;
 675        pgoff_t index;
 676
 677        BUG_ON(!PageLocked(page));
 678        mapping = page->mapping;
 679        index = page->index;
 680        inode = mapping->host;
 681        info = SHMEM_I(inode);
 682        if (info->flags & VM_LOCKED)
 683                goto redirty;
 684        if (!total_swap_pages)
 685                goto redirty;
 686
 687        /*
 688         * shmem_backing_dev_info's capabilities prevent regular writeback or
 689         * sync from ever calling shmem_writepage; but a stacking filesystem
 690         * might use ->writepage of its underlying filesystem, in which case
 691         * tmpfs should write out to swap only in response to memory pressure,
 692         * and not for the writeback threads or sync.
 693         */
 694        if (!wbc->for_reclaim) {
 695                WARN_ON_ONCE(1);        /* Still happens? Tell us about it! */
 696                goto redirty;
 697        }
 698        swap = get_swap_page();
 699        if (!swap.val)
 700                goto redirty;
 701
 702        /*
 703         * Add inode to shmem_unuse()'s list of swapped-out inodes,
 704         * if it's not already there.  Do it now before the page is
 705         * moved to swap cache, when its pagelock no longer protects
 706         * the inode from eviction.  But don't unlock the mutex until
 707         * we've incremented swapped, because shmem_unuse_inode() will
 708         * prune a !swapped inode from the swaplist under this mutex.
 709         */
 710        mutex_lock(&shmem_swaplist_mutex);
 711        if (list_empty(&info->swaplist))
 712                list_add_tail(&info->swaplist, &shmem_swaplist);
 713
 714        if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
 715                swap_shmem_alloc(swap);
 716                shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
 717
 718                spin_lock(&info->lock);
 719                info->swapped++;
 720                shmem_recalc_inode(inode);
 721                spin_unlock(&info->lock);
 722
 723                mutex_unlock(&shmem_swaplist_mutex);
 724                BUG_ON(page_mapped(page));
 725                swap_writepage(page, wbc);
 726                return 0;
 727        }
 728
 729        mutex_unlock(&shmem_swaplist_mutex);
 730        swapcache_free(swap, NULL);
 731redirty:
 732        set_page_dirty(page);
 733        if (wbc->for_reclaim)
 734                return AOP_WRITEPAGE_ACTIVATE;  /* Return with page locked */
 735        unlock_page(page);
 736        return 0;
 737}
 738
 739#ifdef CONFIG_NUMA
 740#ifdef CONFIG_TMPFS
 741static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
 742{
 743        char buffer[64];
 744
 745        if (!mpol || mpol->mode == MPOL_DEFAULT)
 746                return;         /* show nothing */
 747
 748        mpol_to_str(buffer, sizeof(buffer), mpol, 1);
 749
 750        seq_printf(seq, ",mpol=%s", buffer);
 751}
 752
 753static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
 754{
 755        struct mempolicy *mpol = NULL;
 756        if (sbinfo->mpol) {
 757                spin_lock(&sbinfo->stat_lock);  /* prevent replace/use races */
 758                mpol = sbinfo->mpol;
 759                mpol_get(mpol);
 760                spin_unlock(&sbinfo->stat_lock);
 761        }
 762        return mpol;
 763}
 764#endif /* CONFIG_TMPFS */
 765
 766static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
 767                        struct shmem_inode_info *info, pgoff_t index)
 768{
 769        struct mempolicy mpol, *spol;
 770        struct vm_area_struct pvma;
 771
 772        spol = mpol_cond_copy(&mpol,
 773                        mpol_shared_policy_lookup(&info->policy, index));
 774
 775        /* Create a pseudo vma that just contains the policy */
 776        pvma.vm_start = 0;
 777        pvma.vm_pgoff = index;
 778        pvma.vm_ops = NULL;
 779        pvma.vm_policy = spol;
 780        return swapin_readahead(swap, gfp, &pvma, 0);
 781}
 782
 783static struct page *shmem_alloc_page(gfp_t gfp,
 784                        struct shmem_inode_info *info, pgoff_t index)
 785{
 786        struct vm_area_struct pvma;
 787
 788        /* Create a pseudo vma that just contains the policy */
 789        pvma.vm_start = 0;
 790        pvma.vm_pgoff = index;
 791        pvma.vm_ops = NULL;
 792        pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
 793
 794        /*
 795         * alloc_page_vma() will drop the shared policy reference
 796         */
 797        return alloc_page_vma(gfp, &pvma, 0);
 798}
 799#else /* !CONFIG_NUMA */
 800#ifdef CONFIG_TMPFS
 801static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
 802{
 803}
 804#endif /* CONFIG_TMPFS */
 805
 806static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
 807                        struct shmem_inode_info *info, pgoff_t index)
 808{
 809        return swapin_readahead(swap, gfp, NULL, 0);
 810}
 811
 812static inline struct page *shmem_alloc_page(gfp_t gfp,
 813                        struct shmem_inode_info *info, pgoff_t index)
 814{
 815        return alloc_page(gfp);
 816}
 817#endif /* CONFIG_NUMA */
 818
 819#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
 820static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
 821{
 822        return NULL;
 823}
 824#endif
 825
 826/*
 827 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
 828 *
 829 * If we allocate a new one we do not mark it dirty. That's up to the
 830 * vm. If we swap it in we mark it dirty since we also free the swap
 831 * entry since a page cannot live in both the swap and page cache
 832 */
 833static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
 834        struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
 835{
 836        struct address_space *mapping = inode->i_mapping;
 837        struct shmem_inode_info *info;
 838        struct shmem_sb_info *sbinfo;
 839        struct page *page;
 840        swp_entry_t swap;
 841        int error;
 842        int once = 0;
 843
 844        if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
 845                return -EFBIG;
 846repeat:
 847        swap.val = 0;
 848        page = find_lock_page(mapping, index);
 849        if (radix_tree_exceptional_entry(page)) {
 850                swap = radix_to_swp_entry(page);
 851                page = NULL;
 852        }
 853
 854        if (sgp != SGP_WRITE &&
 855            ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
 856                error = -EINVAL;
 857                goto failed;
 858        }
 859
 860        if (page || (sgp == SGP_READ && !swap.val)) {
 861                /*
 862                 * Once we can get the page lock, it must be uptodate:
 863                 * if there were an error in reading back from swap,
 864                 * the page would not be inserted into the filecache.
 865                 */
 866                BUG_ON(page && !PageUptodate(page));
 867                *pagep = page;
 868                return 0;
 869        }
 870
 871        /*
 872         * Fast cache lookup did not find it:
 873         * bring it back from swap or allocate.
 874         */
 875        info = SHMEM_I(inode);
 876        sbinfo = SHMEM_SB(inode->i_sb);
 877
 878        if (swap.val) {
 879                /* Look it up and read it in.. */
 880                page = lookup_swap_cache(swap);
 881                if (!page) {
 882                        /* here we actually do the io */
 883                        if (fault_type)
 884                                *fault_type |= VM_FAULT_MAJOR;
 885                        page = shmem_swapin(swap, gfp, info, index);
 886                        if (!page) {
 887                                error = -ENOMEM;
 888                                goto failed;
 889                        }
 890                }
 891
 892                /* We have to do this with page locked to prevent races */
 893                lock_page(page);
 894                if (!PageUptodate(page)) {
 895                        error = -EIO;
 896                        goto failed;
 897                }
 898                wait_on_page_writeback(page);
 899
 900                /* Someone may have already done it for us */
 901                if (page->mapping) {
 902                        if (page->mapping == mapping &&
 903                            page->index == index)
 904                                goto done;
 905                        error = -EEXIST;
 906                        goto failed;
 907                }
 908
 909                error = mem_cgroup_cache_charge(page, current->mm,
 910                                                gfp & GFP_RECLAIM_MASK);
 911                if (!error)
 912                        error = shmem_add_to_page_cache(page, mapping, index,
 913                                                gfp, swp_to_radix_entry(swap));
 914                if (error)
 915                        goto failed;
 916
 917                spin_lock(&info->lock);
 918                info->swapped--;
 919                shmem_recalc_inode(inode);
 920                spin_unlock(&info->lock);
 921
 922                delete_from_swap_cache(page);
 923                set_page_dirty(page);
 924                swap_free(swap);
 925
 926        } else {
 927                if (shmem_acct_block(info->flags)) {
 928                        error = -ENOSPC;
 929                        goto failed;
 930                }
 931                if (sbinfo->max_blocks) {
 932                        if (percpu_counter_compare(&sbinfo->used_blocks,
 933                                                sbinfo->max_blocks) >= 0) {
 934                                error = -ENOSPC;
 935                                goto unacct;
 936                        }
 937                        percpu_counter_inc(&sbinfo->used_blocks);
 938                }
 939
 940                page = shmem_alloc_page(gfp, info, index);
 941                if (!page) {
 942                        error = -ENOMEM;
 943                        goto decused;
 944                }
 945
 946                SetPageSwapBacked(page);
 947                __set_page_locked(page);
 948                error = mem_cgroup_cache_charge(page, current->mm,
 949                                                gfp & GFP_RECLAIM_MASK);
 950                if (!error)
 951                        error = shmem_add_to_page_cache(page, mapping, index,
 952                                                gfp, NULL);
 953                if (error)
 954                        goto decused;
 955                lru_cache_add_anon(page);
 956
 957                spin_lock(&info->lock);
 958                info->alloced++;
 959                inode->i_blocks += BLOCKS_PER_PAGE;
 960                shmem_recalc_inode(inode);
 961                spin_unlock(&info->lock);
 962
 963                clear_highpage(page);
 964                flush_dcache_page(page);
 965                SetPageUptodate(page);
 966                if (sgp == SGP_DIRTY)
 967                        set_page_dirty(page);
 968        }
 969done:
 970        /* Perhaps the file has been truncated since we checked */
 971        if (sgp != SGP_WRITE &&
 972            ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
 973                error = -EINVAL;
 974                goto trunc;
 975        }
 976        *pagep = page;
 977        return 0;
 978
 979        /*
 980         * Error recovery.
 981         */
 982trunc:
 983        ClearPageDirty(page);
 984        delete_from_page_cache(page);
 985        spin_lock(&info->lock);
 986        info->alloced--;
 987        inode->i_blocks -= BLOCKS_PER_PAGE;
 988        spin_unlock(&info->lock);
 989decused:
 990        if (sbinfo->max_blocks)
 991                percpu_counter_add(&sbinfo->used_blocks, -1);
 992unacct:
 993        shmem_unacct_blocks(info->flags, 1);
 994failed:
 995        if (swap.val && error != -EINVAL) {
 996                struct page *test = find_get_page(mapping, index);
 997                if (test && !radix_tree_exceptional_entry(test))
 998                        page_cache_release(test);
 999                /* Have another try if the entry has changed */
1000                if (test != swp_to_radix_entry(swap))
1001                        error = -EEXIST;
1002        }
1003        if (page) {
1004                unlock_page(page);
1005                page_cache_release(page);
1006        }
1007        if (error == -ENOSPC && !once++) {
1008                info = SHMEM_I(inode);
1009                spin_lock(&info->lock);
1010                shmem_recalc_inode(inode);
1011                spin_unlock(&info->lock);
1012                goto repeat;
1013        }
1014        if (error == -EEXIST)
1015                goto repeat;
1016        return error;
1017}
1018
1019static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1020{
1021        struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1022        int error;
1023        int ret = VM_FAULT_LOCKED;
1024
1025        error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1026        if (error)
1027                return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1028
1029        if (ret & VM_FAULT_MAJOR) {
1030                count_vm_event(PGMAJFAULT);
1031                mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1032        }
1033        return ret;
1034}
1035
1036#ifdef CONFIG_NUMA
1037static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
1038{
1039        struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1040        return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
1041}
1042
1043static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1044                                          unsigned long addr)
1045{
1046        struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1047        pgoff_t index;
1048
1049        index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1050        return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
1051}
1052#endif
1053
1054int shmem_lock(struct file *file, int lock, struct user_struct *user)
1055{
1056        struct inode *inode = file->f_path.dentry->d_inode;
1057        struct shmem_inode_info *info = SHMEM_I(inode);
1058        int retval = -ENOMEM;
1059
1060        spin_lock(&info->lock);
1061        if (lock && !(info->flags & VM_LOCKED)) {
1062                if (!user_shm_lock(inode->i_size, user))
1063                        goto out_nomem;
1064                info->flags |= VM_LOCKED;
1065                mapping_set_unevictable(file->f_mapping);
1066        }
1067        if (!lock && (info->flags & VM_LOCKED) && user) {
1068                user_shm_unlock(inode->i_size, user);
1069                info->flags &= ~VM_LOCKED;
1070                mapping_clear_unevictable(file->f_mapping);
1071                scan_mapping_unevictable_pages(file->f_mapping);
1072        }
1073        retval = 0;
1074
1075out_nomem:
1076        spin_unlock(&info->lock);
1077        return retval;
1078}
1079
1080static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1081{
1082        file_accessed(file);
1083        vma->vm_ops = &shmem_vm_ops;
1084        vma->vm_flags |= VM_CAN_NONLINEAR;
1085        return 0;
1086}
1087
1088static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1089                                     int mode, dev_t dev, unsigned long flags)
1090{
1091        struct inode *inode;
1092        struct shmem_inode_info *info;
1093        struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1094
1095        if (shmem_reserve_inode(sb))
1096                return NULL;
1097
1098        inode = new_inode(sb);
1099        if (inode) {
1100                inode->i_ino = get_next_ino();
1101                inode_init_owner(inode, dir, mode);
1102                inode->i_blocks = 0;
1103                inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1104                inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1105                inode->i_generation = get_seconds();
1106                info = SHMEM_I(inode);
1107                memset(info, 0, (char *)inode - (char *)info);
1108                spin_lock_init(&info->lock);
1109                info->flags = flags & VM_NORESERVE;
1110                INIT_LIST_HEAD(&info->swaplist);
1111                INIT_LIST_HEAD(&info->xattr_list);
1112                cache_no_acl(inode);
1113
1114                switch (mode & S_IFMT) {
1115                default:
1116                        inode->i_op = &shmem_special_inode_operations;
1117                        init_special_inode(inode, mode, dev);
1118                        break;
1119                case S_IFREG:
1120                        inode->i_mapping->a_ops = &shmem_aops;
1121                        inode->i_op = &shmem_inode_operations;
1122                        inode->i_fop = &shmem_file_operations;
1123                        mpol_shared_policy_init(&info->policy,
1124                                                 shmem_get_sbmpol(sbinfo));
1125                        break;
1126                case S_IFDIR:
1127                        inc_nlink(inode);
1128                        /* Some things misbehave if size == 0 on a directory */
1129                        inode->i_size = 2 * BOGO_DIRENT_SIZE;
1130                        inode->i_op = &shmem_dir_inode_operations;
1131                        inode->i_fop = &simple_dir_operations;
1132                        break;
1133                case S_IFLNK:
1134                        /*
1135                         * Must not load anything in the rbtree,
1136                         * mpol_free_shared_policy will not be called.
1137                         */
1138                        mpol_shared_policy_init(&info->policy, NULL);
1139                        break;
1140                }
1141        } else
1142                shmem_free_inode(sb);
1143        return inode;
1144}
1145
1146#ifdef CONFIG_TMPFS
1147static const struct inode_operations shmem_symlink_inode_operations;
1148static const struct inode_operations shmem_short_symlink_operations;
1149
1150static int
1151shmem_write_begin(struct file *file, struct address_space *mapping,
1152                        loff_t pos, unsigned len, unsigned flags,
1153                        struct page **pagep, void **fsdata)
1154{
1155        struct inode *inode = mapping->host;
1156        pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1157        return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1158}
1159
1160static int
1161shmem_write_end(struct file *file, struct address_space *mapping,
1162                        loff_t pos, unsigned len, unsigned copied,
1163                        struct page *page, void *fsdata)
1164{
1165        struct inode *inode = mapping->host;
1166
1167        if (pos + copied > inode->i_size)
1168                i_size_write(inode, pos + copied);
1169
1170        set_page_dirty(page);
1171        unlock_page(page);
1172        page_cache_release(page);
1173
1174        return copied;
1175}
1176
1177static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1178{
1179        struct inode *inode = filp->f_path.dentry->d_inode;
1180        struct address_space *mapping = inode->i_mapping;
1181        pgoff_t index;
1182        unsigned long offset;
1183        enum sgp_type sgp = SGP_READ;
1184
1185        /*
1186         * Might this read be for a stacking filesystem?  Then when reading
1187         * holes of a sparse file, we actually need to allocate those pages,
1188         * and even mark them dirty, so it cannot exceed the max_blocks limit.
1189         */
1190        if (segment_eq(get_fs(), KERNEL_DS))
1191                sgp = SGP_DIRTY;
1192
1193        index = *ppos >> PAGE_CACHE_SHIFT;
1194        offset = *ppos & ~PAGE_CACHE_MASK;
1195
1196        for (;;) {
1197                struct page *page = NULL;
1198                pgoff_t end_index;
1199                unsigned long nr, ret;
1200                loff_t i_size = i_size_read(inode);
1201
1202                end_index = i_size >> PAGE_CACHE_SHIFT;
1203                if (index > end_index)
1204                        break;
1205                if (index == end_index) {
1206                        nr = i_size & ~PAGE_CACHE_MASK;
1207                        if (nr <= offset)
1208                                break;
1209                }
1210
1211                desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1212                if (desc->error) {
1213                        if (desc->error == -EINVAL)
1214                                desc->error = 0;
1215                        break;
1216                }
1217                if (page)
1218                        unlock_page(page);
1219
1220                /*
1221                 * We must evaluate after, since reads (unlike writes)
1222                 * are called without i_mutex protection against truncate
1223                 */
1224                nr = PAGE_CACHE_SIZE;
1225                i_size = i_size_read(inode);
1226                end_index = i_size >> PAGE_CACHE_SHIFT;
1227                if (index == end_index) {
1228                        nr = i_size & ~PAGE_CACHE_MASK;
1229                        if (nr <= offset) {
1230                                if (page)
1231                                        page_cache_release(page);
1232                                break;
1233                        }
1234                }
1235                nr -= offset;
1236
1237                if (page) {
1238                        /*
1239                         * If users can be writing to this page using arbitrary
1240                         * virtual addresses, take care about potential aliasing
1241                         * before reading the page on the kernel side.
1242                         */
1243                        if (mapping_writably_mapped(mapping))
1244                                flush_dcache_page(page);
1245                        /*
1246                         * Mark the page accessed if we read the beginning.
1247                         */
1248                        if (!offset)
1249                                mark_page_accessed(page);
1250                } else {
1251                        page = ZERO_PAGE(0);
1252                        page_cache_get(page);
1253                }
1254
1255                /*
1256                 * Ok, we have the page, and it's up-to-date, so
1257                 * now we can copy it to user space...
1258                 *
1259                 * The actor routine returns how many bytes were actually used..
1260                 * NOTE! This may not be the same as how much of a user buffer
1261                 * we filled up (we may be padding etc), so we can only update
1262                 * "pos" here (the actor routine has to update the user buffer
1263                 * pointers and the remaining count).
1264                 */
1265                ret = actor(desc, page, offset, nr);
1266                offset += ret;
1267                index += offset >> PAGE_CACHE_SHIFT;
1268                offset &= ~PAGE_CACHE_MASK;
1269
1270                page_cache_release(page);
1271                if (ret != nr || !desc->count)
1272                        break;
1273
1274                cond_resched();
1275        }
1276
1277        *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1278        file_accessed(filp);
1279}
1280
1281static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1282                const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1283{
1284        struct file *filp = iocb->ki_filp;
1285        ssize_t retval;
1286        unsigned long seg;
1287        size_t count;
1288        loff_t *ppos = &iocb->ki_pos;
1289
1290        retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1291        if (retval)
1292                return retval;
1293
1294        for (seg = 0; seg < nr_segs; seg++) {
1295                read_descriptor_t desc;
1296
1297                desc.written = 0;
1298                desc.arg.buf = iov[seg].iov_base;
1299                desc.count = iov[seg].iov_len;
1300                if (desc.count == 0)
1301                        continue;
1302                desc.error = 0;
1303                do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1304                retval += desc.written;
1305                if (desc.error) {
1306                        retval = retval ?: desc.error;
1307                        break;
1308                }
1309                if (desc.count > 0)
1310                        break;
1311        }
1312        return retval;
1313}
1314
1315static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1316                                struct pipe_inode_info *pipe, size_t len,
1317                                unsigned int flags)
1318{
1319        struct address_space *mapping = in->f_mapping;
1320        struct inode *inode = mapping->host;
1321        unsigned int loff, nr_pages, req_pages;
1322        struct page *pages[PIPE_DEF_BUFFERS];
1323        struct partial_page partial[PIPE_DEF_BUFFERS];
1324        struct page *page;
1325        pgoff_t index, end_index;
1326        loff_t isize, left;
1327        int error, page_nr;
1328        struct splice_pipe_desc spd = {
1329                .pages = pages,
1330                .partial = partial,
1331                .flags = flags,
1332                .ops = &page_cache_pipe_buf_ops,
1333                .spd_release = spd_release_page,
1334        };
1335
1336        isize = i_size_read(inode);
1337        if (unlikely(*ppos >= isize))
1338                return 0;
1339
1340        left = isize - *ppos;
1341        if (unlikely(left < len))
1342                len = left;
1343
1344        if (splice_grow_spd(pipe, &spd))
1345                return -ENOMEM;
1346
1347        index = *ppos >> PAGE_CACHE_SHIFT;
1348        loff = *ppos & ~PAGE_CACHE_MASK;
1349        req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1350        nr_pages = min(req_pages, pipe->buffers);
1351
1352        spd.nr_pages = find_get_pages_contig(mapping, index,
1353                                                nr_pages, spd.pages);
1354        index += spd.nr_pages;
1355        error = 0;
1356
1357        while (spd.nr_pages < nr_pages) {
1358                error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1359                if (error)
1360                        break;
1361                unlock_page(page);
1362                spd.pages[spd.nr_pages++] = page;
1363                index++;
1364        }
1365
1366        index = *ppos >> PAGE_CACHE_SHIFT;
1367        nr_pages = spd.nr_pages;
1368        spd.nr_pages = 0;
1369
1370        for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1371                unsigned int this_len;
1372
1373                if (!len)
1374                        break;
1375
1376                this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1377                page = spd.pages[page_nr];
1378
1379                if (!PageUptodate(page) || page->mapping != mapping) {
1380                        error = shmem_getpage(inode, index, &page,
1381                                                        SGP_CACHE, NULL);
1382                        if (error)
1383                                break;
1384                        unlock_page(page);
1385                        page_cache_release(spd.pages[page_nr]);
1386                        spd.pages[page_nr] = page;
1387                }
1388
1389                isize = i_size_read(inode);
1390                end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1391                if (unlikely(!isize || index > end_index))
1392                        break;
1393
1394                if (end_index == index) {
1395                        unsigned int plen;
1396
1397                        plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1398                        if (plen <= loff)
1399                                break;
1400
1401                        this_len = min(this_len, plen - loff);
1402                        len = this_len;
1403                }
1404
1405                spd.partial[page_nr].offset = loff;
1406                spd.partial[page_nr].len = this_len;
1407                len -= this_len;
1408                loff = 0;
1409                spd.nr_pages++;
1410                index++;
1411        }
1412
1413        while (page_nr < nr_pages)
1414                page_cache_release(spd.pages[page_nr++]);
1415
1416        if (spd.nr_pages)
1417                error = splice_to_pipe(pipe, &spd);
1418
1419        splice_shrink_spd(pipe, &spd);
1420
1421        if (error > 0) {
1422                *ppos += error;
1423                file_accessed(in);
1424        }
1425        return error;
1426}
1427
1428static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1429{
1430        struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1431
1432        buf->f_type = TMPFS_MAGIC;
1433        buf->f_bsize = PAGE_CACHE_SIZE;
1434        buf->f_namelen = NAME_MAX;
1435        if (sbinfo->max_blocks) {
1436                buf->f_blocks = sbinfo->max_blocks;
1437                buf->f_bavail =
1438                buf->f_bfree  = sbinfo->max_blocks -
1439                                percpu_counter_sum(&sbinfo->used_blocks);
1440        }
1441        if (sbinfo->max_inodes) {
1442                buf->f_files = sbinfo->max_inodes;
1443                buf->f_ffree = sbinfo->free_inodes;
1444        }
1445        /* else leave those fields 0 like simple_statfs */
1446        return 0;
1447}
1448
1449/*
1450 * File creation. Allocate an inode, and we're done..
1451 */
1452static int
1453shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1454{
1455        struct inode *inode;
1456        int error = -ENOSPC;
1457
1458        inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1459        if (inode) {
1460                error = security_inode_init_security(inode, dir,
1461                                                     &dentry->d_name, NULL,
1462                                                     NULL, NULL);
1463                if (error) {
1464                        if (error != -EOPNOTSUPP) {
1465                                iput(inode);
1466                                return error;
1467                        }
1468                }
1469#ifdef CONFIG_TMPFS_POSIX_ACL
1470                error = generic_acl_init(inode, dir);
1471                if (error) {
1472                        iput(inode);
1473                        return error;
1474                }
1475#else
1476                error = 0;
1477#endif
1478                dir->i_size += BOGO_DIRENT_SIZE;
1479                dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1480                d_instantiate(dentry, inode);
1481                dget(dentry); /* Extra count - pin the dentry in core */
1482        }
1483        return error;
1484}
1485
1486static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1487{
1488        int error;
1489
1490        if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1491                return error;
1492        inc_nlink(dir);
1493        return 0;
1494}
1495
1496static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1497                struct nameidata *nd)
1498{
1499        return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1500}
1501
1502/*
1503 * Link a file..
1504 */
1505static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1506{
1507        struct inode *inode = old_dentry->d_inode;
1508        int ret;
1509
1510        /*
1511         * No ordinary (disk based) filesystem counts links as inodes;
1512         * but each new link needs a new dentry, pinning lowmem, and
1513         * tmpfs dentries cannot be pruned until they are unlinked.
1514         */
1515        ret = shmem_reserve_inode(inode->i_sb);
1516        if (ret)
1517                goto out;
1518
1519        dir->i_size += BOGO_DIRENT_SIZE;
1520        inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1521        inc_nlink(inode);
1522        ihold(inode);   /* New dentry reference */
1523        dget(dentry);           /* Extra pinning count for the created dentry */
1524        d_instantiate(dentry, inode);
1525out:
1526        return ret;
1527}
1528
1529static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1530{
1531        struct inode *inode = dentry->d_inode;
1532
1533        if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1534                shmem_free_inode(inode->i_sb);
1535
1536        dir->i_size -= BOGO_DIRENT_SIZE;
1537        inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1538        drop_nlink(inode);
1539        dput(dentry);   /* Undo the count from "create" - this does all the work */
1540        return 0;
1541}
1542
1543static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1544{
1545        if (!simple_empty(dentry))
1546                return -ENOTEMPTY;
1547
1548        drop_nlink(dentry->d_inode);
1549        drop_nlink(dir);
1550        return shmem_unlink(dir, dentry);
1551}
1552
1553/*
1554 * The VFS layer already does all the dentry stuff for rename,
1555 * we just have to decrement the usage count for the target if
1556 * it exists so that the VFS layer correctly free's it when it
1557 * gets overwritten.
1558 */
1559static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1560{
1561        struct inode *inode = old_dentry->d_inode;
1562        int they_are_dirs = S_ISDIR(inode->i_mode);
1563
1564        if (!simple_empty(new_dentry))
1565                return -ENOTEMPTY;
1566
1567        if (new_dentry->d_inode) {
1568                (void) shmem_unlink(new_dir, new_dentry);
1569                if (they_are_dirs)
1570                        drop_nlink(old_dir);
1571        } else if (they_are_dirs) {
1572                drop_nlink(old_dir);
1573                inc_nlink(new_dir);
1574        }
1575
1576        old_dir->i_size -= BOGO_DIRENT_SIZE;
1577        new_dir->i_size += BOGO_DIRENT_SIZE;
1578        old_dir->i_ctime = old_dir->i_mtime =
1579        new_dir->i_ctime = new_dir->i_mtime =
1580        inode->i_ctime = CURRENT_TIME;
1581        return 0;
1582}
1583
1584static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1585{
1586        int error;
1587        int len;
1588        struct inode *inode;
1589        struct page *page;
1590        char *kaddr;
1591        struct shmem_inode_info *info;
1592
1593        len = strlen(symname) + 1;
1594        if (len > PAGE_CACHE_SIZE)
1595                return -ENAMETOOLONG;
1596
1597        inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
1598        if (!inode)
1599                return -ENOSPC;
1600
1601        error = security_inode_init_security(inode, dir, &dentry->d_name, NULL,
1602                                             NULL, NULL);
1603        if (error) {
1604                if (error != -EOPNOTSUPP) {
1605                        iput(inode);
1606                        return error;
1607                }
1608                error = 0;
1609        }
1610
1611        info = SHMEM_I(inode);
1612        inode->i_size = len-1;
1613        if (len <= SHORT_SYMLINK_LEN) {
1614                info->symlink = kmemdup(symname, len, GFP_KERNEL);
1615                if (!info->symlink) {
1616                        iput(inode);
1617                        return -ENOMEM;
1618                }
1619                inode->i_op = &shmem_short_symlink_operations;
1620        } else {
1621                error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1622                if (error) {
1623                        iput(inode);
1624                        return error;
1625                }
1626                inode->i_mapping->a_ops = &shmem_aops;
1627                inode->i_op = &shmem_symlink_inode_operations;
1628                kaddr = kmap_atomic(page, KM_USER0);
1629                memcpy(kaddr, symname, len);
1630                kunmap_atomic(kaddr, KM_USER0);
1631                set_page_dirty(page);
1632                unlock_page(page);
1633                page_cache_release(page);
1634        }
1635        dir->i_size += BOGO_DIRENT_SIZE;
1636        dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1637        d_instantiate(dentry, inode);
1638        dget(dentry);
1639        return 0;
1640}
1641
1642static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
1643{
1644        nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
1645        return NULL;
1646}
1647
1648static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1649{
1650        struct page *page = NULL;
1651        int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1652        nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
1653        if (page)
1654                unlock_page(page);
1655        return page;
1656}
1657
1658static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1659{
1660        if (!IS_ERR(nd_get_link(nd))) {
1661                struct page *page = cookie;
1662                kunmap(page);
1663                mark_page_accessed(page);
1664                page_cache_release(page);
1665        }
1666}
1667
1668#ifdef CONFIG_TMPFS_XATTR
1669/*
1670 * Superblocks without xattr inode operations may get some security.* xattr
1671 * support from the LSM "for free". As soon as we have any other xattrs
1672 * like ACLs, we also need to implement the security.* handlers at
1673 * filesystem level, though.
1674 */
1675
1676static int shmem_xattr_get(struct dentry *dentry, const char *name,
1677                           void *buffer, size_t size)
1678{
1679        struct shmem_inode_info *info;
1680        struct shmem_xattr *xattr;
1681        int ret = -ENODATA;
1682
1683        info = SHMEM_I(dentry->d_inode);
1684
1685        spin_lock(&info->lock);
1686        list_for_each_entry(xattr, &info->xattr_list, list) {
1687                if (strcmp(name, xattr->name))
1688                        continue;
1689
1690                ret = xattr->size;
1691                if (buffer) {
1692                        if (size < xattr->size)
1693                                ret = -ERANGE;
1694                        else
1695                                memcpy(buffer, xattr->value, xattr->size);
1696                }
1697                break;
1698        }
1699        spin_unlock(&info->lock);
1700        return ret;
1701}
1702
1703static int shmem_xattr_set(struct dentry *dentry, const char *name,
1704                           const void *value, size_t size, int flags)
1705{
1706        struct inode *inode = dentry->d_inode;
1707        struct shmem_inode_info *info = SHMEM_I(inode);
1708        struct shmem_xattr *xattr;
1709        struct shmem_xattr *new_xattr = NULL;
1710        size_t len;
1711        int err = 0;
1712
1713        /* value == NULL means remove */
1714        if (value) {
1715                /* wrap around? */
1716                len = sizeof(*new_xattr) + size;
1717                if (len <= sizeof(*new_xattr))
1718                        return -ENOMEM;
1719
1720                new_xattr = kmalloc(len, GFP_KERNEL);
1721                if (!new_xattr)
1722                        return -ENOMEM;
1723
1724                new_xattr->name = kstrdup(name, GFP_KERNEL);
1725                if (!new_xattr->name) {
1726                        kfree(new_xattr);
1727                        return -ENOMEM;
1728                }
1729
1730                new_xattr->size = size;
1731                memcpy(new_xattr->value, value, size);
1732        }
1733
1734        spin_lock(&info->lock);
1735        list_for_each_entry(xattr, &info->xattr_list, list) {
1736                if (!strcmp(name, xattr->name)) {
1737                        if (flags & XATTR_CREATE) {
1738                                xattr = new_xattr;
1739                                err = -EEXIST;
1740                        } else if (new_xattr) {
1741                                list_replace(&xattr->list, &new_xattr->list);
1742                        } else {
1743                                list_del(&xattr->list);
1744                        }
1745                        goto out;
1746                }
1747        }
1748        if (flags & XATTR_REPLACE) {
1749                xattr = new_xattr;
1750                err = -ENODATA;
1751        } else {
1752                list_add(&new_xattr->list, &info->xattr_list);
1753                xattr = NULL;
1754        }
1755out:
1756        spin_unlock(&info->lock);
1757        if (xattr)
1758                kfree(xattr->name);
1759        kfree(xattr);
1760        return err;
1761}
1762
1763static const struct xattr_handler *shmem_xattr_handlers[] = {
1764#ifdef CONFIG_TMPFS_POSIX_ACL
1765        &generic_acl_access_handler,
1766        &generic_acl_default_handler,
1767#endif
1768        NULL
1769};
1770
1771static int shmem_xattr_validate(const char *name)
1772{
1773        struct { const char *prefix; size_t len; } arr[] = {
1774                { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
1775                { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
1776        };
1777        int i;
1778
1779        for (i = 0; i < ARRAY_SIZE(arr); i++) {
1780                size_t preflen = arr[i].len;
1781                if (strncmp(name, arr[i].prefix, preflen) == 0) {
1782                        if (!name[preflen])
1783                                return -EINVAL;
1784                        return 0;
1785                }
1786        }
1787        return -EOPNOTSUPP;
1788}
1789
1790static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
1791                              void *buffer, size_t size)
1792{
1793        int err;
1794
1795        /*
1796         * If this is a request for a synthetic attribute in the system.*
1797         * namespace use the generic infrastructure to resolve a handler
1798         * for it via sb->s_xattr.
1799         */
1800        if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1801                return generic_getxattr(dentry, name, buffer, size);
1802
1803        err = shmem_xattr_validate(name);
1804        if (err)
1805                return err;
1806
1807        return shmem_xattr_get(dentry, name, buffer, size);
1808}
1809
1810static int shmem_setxattr(struct dentry *dentry, const char *name,
1811                          const void *value, size_t size, int flags)
1812{
1813        int err;
1814
1815        /*
1816         * If this is a request for a synthetic attribute in the system.*
1817         * namespace use the generic infrastructure to resolve a handler
1818         * for it via sb->s_xattr.
1819         */
1820        if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1821                return generic_setxattr(dentry, name, value, size, flags);
1822
1823        err = shmem_xattr_validate(name);
1824        if (err)
1825                return err;
1826
1827        if (size == 0)
1828                value = "";  /* empty EA, do not remove */
1829
1830        return shmem_xattr_set(dentry, name, value, size, flags);
1831
1832}
1833
1834static int shmem_removexattr(struct dentry *dentry, const char *name)
1835{
1836        int err;
1837
1838        /*
1839         * If this is a request for a synthetic attribute in the system.*
1840         * namespace use the generic infrastructure to resolve a handler
1841         * for it via sb->s_xattr.
1842         */
1843        if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1844                return generic_removexattr(dentry, name);
1845
1846        err = shmem_xattr_validate(name);
1847        if (err)
1848                return err;
1849
1850        return shmem_xattr_set(dentry, name, NULL, 0, XATTR_REPLACE);
1851}
1852
1853static bool xattr_is_trusted(const char *name)
1854{
1855        return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN);
1856}
1857
1858static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
1859{
1860        bool trusted = capable(CAP_SYS_ADMIN);
1861        struct shmem_xattr *xattr;
1862        struct shmem_inode_info *info;
1863        size_t used = 0;
1864
1865        info = SHMEM_I(dentry->d_inode);
1866
1867        spin_lock(&info->lock);
1868        list_for_each_entry(xattr, &info->xattr_list, list) {
1869                size_t len;
1870
1871                /* skip "trusted." attributes for unprivileged callers */
1872                if (!trusted && xattr_is_trusted(xattr->name))
1873                        continue;
1874
1875                len = strlen(xattr->name) + 1;
1876                used += len;
1877                if (buffer) {
1878                        if (size < used) {
1879                                used = -ERANGE;
1880                                break;
1881                        }
1882                        memcpy(buffer, xattr->name, len);
1883                        buffer += len;
1884                }
1885        }
1886        spin_unlock(&info->lock);
1887
1888        return used;
1889}
1890#endif /* CONFIG_TMPFS_XATTR */
1891
1892static const struct inode_operations shmem_short_symlink_operations = {
1893        .readlink       = generic_readlink,
1894        .follow_link    = shmem_follow_short_symlink,
1895#ifdef CONFIG_TMPFS_XATTR
1896        .setxattr       = shmem_setxattr,
1897        .getxattr       = shmem_getxattr,
1898        .listxattr      = shmem_listxattr,
1899        .removexattr    = shmem_removexattr,
1900#endif
1901};
1902
1903static const struct inode_operations shmem_symlink_inode_operations = {
1904        .readlink       = generic_readlink,
1905        .follow_link    = shmem_follow_link,
1906        .put_link       = shmem_put_link,
1907#ifdef CONFIG_TMPFS_XATTR
1908        .setxattr       = shmem_setxattr,
1909        .getxattr       = shmem_getxattr,
1910        .listxattr      = shmem_listxattr,
1911        .removexattr    = shmem_removexattr,
1912#endif
1913};
1914
1915static struct dentry *shmem_get_parent(struct dentry *child)
1916{
1917        return ERR_PTR(-ESTALE);
1918}
1919
1920static int shmem_match(struct inode *ino, void *vfh)
1921{
1922        __u32 *fh = vfh;
1923        __u64 inum = fh[2];
1924        inum = (inum << 32) | fh[1];
1925        return ino->i_ino == inum && fh[0] == ino->i_generation;
1926}
1927
1928static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
1929                struct fid *fid, int fh_len, int fh_type)
1930{
1931        struct inode *inode;
1932        struct dentry *dentry = NULL;
1933        u64 inum = fid->raw[2];
1934        inum = (inum << 32) | fid->raw[1];
1935
1936        if (fh_len < 3)
1937                return NULL;
1938
1939        inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
1940                        shmem_match, fid->raw);
1941        if (inode) {
1942                dentry = d_find_alias(inode);
1943                iput(inode);
1944        }
1945
1946        return dentry;
1947}
1948
1949static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
1950                                int connectable)
1951{
1952        struct inode *inode = dentry->d_inode;
1953
1954        if (*len < 3) {
1955                *len = 3;
1956                return 255;
1957        }
1958
1959        if (inode_unhashed(inode)) {
1960                /* Unfortunately insert_inode_hash is not idempotent,
1961                 * so as we hash inodes here rather than at creation
1962                 * time, we need a lock to ensure we only try
1963                 * to do it once
1964                 */
1965                static DEFINE_SPINLOCK(lock);
1966                spin_lock(&lock);
1967                if (inode_unhashed(inode))
1968                        __insert_inode_hash(inode,
1969                                            inode->i_ino + inode->i_generation);
1970                spin_unlock(&lock);
1971        }
1972
1973        fh[0] = inode->i_generation;
1974        fh[1] = inode->i_ino;
1975        fh[2] = ((__u64)inode->i_ino) >> 32;
1976
1977        *len = 3;
1978        return 1;
1979}
1980
1981static const struct export_operations shmem_export_ops = {
1982        .get_parent     = shmem_get_parent,
1983        .encode_fh      = shmem_encode_fh,
1984        .fh_to_dentry   = shmem_fh_to_dentry,
1985};
1986
1987static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
1988                               bool remount)
1989{
1990        char *this_char, *value, *rest;
1991
1992        while (options != NULL) {
1993                this_char = options;
1994                for (;;) {
1995                        /*
1996                         * NUL-terminate this option: unfortunately,
1997                         * mount options form a comma-separated list,
1998                         * but mpol's nodelist may also contain commas.
1999                         */
2000                        options = strchr(options, ',');
2001                        if (options == NULL)
2002                                break;
2003                        options++;
2004                        if (!isdigit(*options)) {
2005                                options[-1] = '\0';
2006                                break;
2007                        }
2008                }
2009                if (!*this_char)
2010                        continue;
2011                if ((value = strchr(this_char,'=')) != NULL) {
2012                        *value++ = 0;
2013                } else {
2014                        printk(KERN_ERR
2015                            "tmpfs: No value for mount option '%s'\n",
2016                            this_char);
2017                        return 1;
2018                }
2019
2020                if (!strcmp(this_char,"size")) {
2021                        unsigned long long size;
2022                        size = memparse(value,&rest);
2023                        if (*rest == '%') {
2024                                size <<= PAGE_SHIFT;
2025                                size *= totalram_pages;
2026                                do_div(size, 100);
2027                                rest++;
2028                        }
2029                        if (*rest)
2030                                goto bad_val;
2031                        sbinfo->max_blocks =
2032                                DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2033                } else if (!strcmp(this_char,"nr_blocks")) {
2034                        sbinfo->max_blocks = memparse(value, &rest);
2035                        if (*rest)
2036                                goto bad_val;
2037                } else if (!strcmp(this_char,"nr_inodes")) {
2038                        sbinfo->max_inodes = memparse(value, &rest);
2039                        if (*rest)
2040                                goto bad_val;
2041                } else if (!strcmp(this_char,"mode")) {
2042                        if (remount)
2043                                continue;
2044                        sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2045                        if (*rest)
2046                                goto bad_val;
2047                } else if (!strcmp(this_char,"uid")) {
2048                        if (remount)
2049                                continue;
2050                        sbinfo->uid = simple_strtoul(value, &rest, 0);
2051                        if (*rest)
2052                                goto bad_val;
2053                } else if (!strcmp(this_char,"gid")) {
2054                        if (remount)
2055                                continue;
2056                        sbinfo->gid = simple_strtoul(value, &rest, 0);
2057                        if (*rest)
2058                                goto bad_val;
2059                } else if (!strcmp(this_char,"mpol")) {
2060                        if (mpol_parse_str(value, &sbinfo->mpol, 1))
2061                                goto bad_val;
2062                } else {
2063                        printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2064                               this_char);
2065                        return 1;
2066                }
2067        }
2068        return 0;
2069
2070bad_val:
2071        printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2072               value, this_char);
2073        return 1;
2074
2075}
2076
2077static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2078{
2079        struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2080        struct shmem_sb_info config = *sbinfo;
2081        unsigned long inodes;
2082        int error = -EINVAL;
2083
2084        if (shmem_parse_options(data, &config, true))
2085                return error;
2086
2087        spin_lock(&sbinfo->stat_lock);
2088        inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2089        if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2090                goto out;
2091        if (config.max_inodes < inodes)
2092                goto out;
2093        /*
2094         * Those tests disallow limited->unlimited while any are in use;
2095         * but we must separately disallow unlimited->limited, because
2096         * in that case we have no record of how much is already in use.
2097         */
2098        if (config.max_blocks && !sbinfo->max_blocks)
2099                goto out;
2100        if (config.max_inodes && !sbinfo->max_inodes)
2101                goto out;
2102
2103        error = 0;
2104        sbinfo->max_blocks  = config.max_blocks;
2105        sbinfo->max_inodes  = config.max_inodes;
2106        sbinfo->free_inodes = config.max_inodes - inodes;
2107
2108        mpol_put(sbinfo->mpol);
2109        sbinfo->mpol        = config.mpol;      /* transfers initial ref */
2110out:
2111        spin_unlock(&sbinfo->stat_lock);
2112        return error;
2113}
2114
2115static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2116{
2117        struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2118
2119        if (sbinfo->max_blocks != shmem_default_max_blocks())
2120                seq_printf(seq, ",size=%luk",
2121                        sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2122        if (sbinfo->max_inodes != shmem_default_max_inodes())
2123                seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2124        if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2125                seq_printf(seq, ",mode=%03o", sbinfo->mode);
2126        if (sbinfo->uid != 0)
2127                seq_printf(seq, ",uid=%u", sbinfo->uid);
2128        if (sbinfo->gid != 0)
2129                seq_printf(seq, ",gid=%u", sbinfo->gid);
2130        shmem_show_mpol(seq, sbinfo->mpol);
2131        return 0;
2132}
2133#endif /* CONFIG_TMPFS */
2134
2135static void shmem_put_super(struct super_block *sb)
2136{
2137        struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2138
2139        percpu_counter_destroy(&sbinfo->used_blocks);
2140        kfree(sbinfo);
2141        sb->s_fs_info = NULL;
2142}
2143
2144int shmem_fill_super(struct super_block *sb, void *data, int silent)
2145{
2146        struct inode *inode;
2147        struct dentry *root;
2148        struct shmem_sb_info *sbinfo;
2149        int err = -ENOMEM;
2150
2151        /* Round up to L1_CACHE_BYTES to resist false sharing */
2152        sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2153                                L1_CACHE_BYTES), GFP_KERNEL);
2154        if (!sbinfo)
2155                return -ENOMEM;
2156
2157        sbinfo->mode = S_IRWXUGO | S_ISVTX;
2158        sbinfo->uid = current_fsuid();
2159        sbinfo->gid = current_fsgid();
2160        sb->s_fs_info = sbinfo;
2161
2162#ifdef CONFIG_TMPFS
2163        /*
2164         * Per default we only allow half of the physical ram per
2165         * tmpfs instance, limiting inodes to one per page of lowmem;
2166         * but the internal instance is left unlimited.
2167         */
2168        if (!(sb->s_flags & MS_NOUSER)) {
2169                sbinfo->max_blocks = shmem_default_max_blocks();
2170                sbinfo->max_inodes = shmem_default_max_inodes();
2171                if (shmem_parse_options(data, sbinfo, false)) {
2172                        err = -EINVAL;
2173                        goto failed;
2174                }
2175        }
2176        sb->s_export_op = &shmem_export_ops;
2177#else
2178        sb->s_flags |= MS_NOUSER;
2179#endif
2180
2181        spin_lock_init(&sbinfo->stat_lock);
2182        if (percpu_counter_init(&sbinfo->used_blocks, 0))
2183                goto failed;
2184        sbinfo->free_inodes = sbinfo->max_inodes;
2185
2186        sb->s_maxbytes = MAX_LFS_FILESIZE;
2187        sb->s_blocksize = PAGE_CACHE_SIZE;
2188        sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2189        sb->s_magic = TMPFS_MAGIC;
2190        sb->s_op = &shmem_ops;
2191        sb->s_time_gran = 1;
2192#ifdef CONFIG_TMPFS_XATTR
2193        sb->s_xattr = shmem_xattr_handlers;
2194#endif
2195#ifdef CONFIG_TMPFS_POSIX_ACL
2196        sb->s_flags |= MS_POSIXACL;
2197#endif
2198
2199        inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2200        if (!inode)
2201                goto failed;
2202        inode->i_uid = sbinfo->uid;
2203        inode->i_gid = sbinfo->gid;
2204        root = d_alloc_root(inode);
2205        if (!root)
2206                goto failed_iput;
2207        sb->s_root = root;
2208        return 0;
2209
2210failed_iput:
2211        iput(inode);
2212failed:
2213        shmem_put_super(sb);
2214        return err;
2215}
2216
2217static struct kmem_cache *shmem_inode_cachep;
2218
2219static struct inode *shmem_alloc_inode(struct super_block *sb)
2220{
2221        struct shmem_inode_info *info;
2222        info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2223        if (!info)
2224                return NULL;
2225        return &info->vfs_inode;
2226}
2227
2228static void shmem_destroy_callback(struct rcu_head *head)
2229{
2230        struct inode *inode = container_of(head, struct inode, i_rcu);
2231        INIT_LIST_HEAD(&inode->i_dentry);
2232        kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2233}
2234
2235static void shmem_destroy_inode(struct inode *inode)
2236{
2237        if ((inode->i_mode & S_IFMT) == S_IFREG)
2238                mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2239        call_rcu(&inode->i_rcu, shmem_destroy_callback);
2240}
2241
2242static void shmem_init_inode(void *foo)
2243{
2244        struct shmem_inode_info *info = foo;
2245        inode_init_once(&info->vfs_inode);
2246}
2247
2248static int shmem_init_inodecache(void)
2249{
2250        shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2251                                sizeof(struct shmem_inode_info),
2252                                0, SLAB_PANIC, shmem_init_inode);
2253        return 0;
2254}
2255
2256static void shmem_destroy_inodecache(void)
2257{
2258        kmem_cache_destroy(shmem_inode_cachep);
2259}
2260
2261static const struct address_space_operations shmem_aops = {
2262        .writepage      = shmem_writepage,
2263        .set_page_dirty = __set_page_dirty_no_writeback,
2264#ifdef CONFIG_TMPFS
2265        .write_begin    = shmem_write_begin,
2266        .write_end      = shmem_write_end,
2267#endif
2268        .migratepage    = migrate_page,
2269        .error_remove_page = generic_error_remove_page,
2270};
2271
2272static const struct file_operations shmem_file_operations = {
2273        .mmap           = shmem_mmap,
2274#ifdef CONFIG_TMPFS
2275        .llseek         = generic_file_llseek,
2276        .read           = do_sync_read,
2277        .write          = do_sync_write,
2278        .aio_read       = shmem_file_aio_read,
2279        .aio_write      = generic_file_aio_write,
2280        .fsync          = noop_fsync,
2281        .splice_read    = shmem_file_splice_read,
2282        .splice_write   = generic_file_splice_write,
2283#endif
2284};
2285
2286static const struct inode_operations shmem_inode_operations = {
2287        .setattr        = shmem_setattr,
2288        .truncate_range = shmem_truncate_range,
2289#ifdef CONFIG_TMPFS_XATTR
2290        .setxattr       = shmem_setxattr,
2291        .getxattr       = shmem_getxattr,
2292        .listxattr      = shmem_listxattr,
2293        .removexattr    = shmem_removexattr,
2294#endif
2295};
2296
2297static const struct inode_operations shmem_dir_inode_operations = {
2298#ifdef CONFIG_TMPFS
2299        .create         = shmem_create,
2300        .lookup         = simple_lookup,
2301        .link           = shmem_link,
2302        .unlink         = shmem_unlink,
2303        .symlink        = shmem_symlink,
2304        .mkdir          = shmem_mkdir,
2305        .rmdir          = shmem_rmdir,
2306        .mknod          = shmem_mknod,
2307        .rename         = shmem_rename,
2308#endif
2309#ifdef CONFIG_TMPFS_XATTR
2310        .setxattr       = shmem_setxattr,
2311        .getxattr       = shmem_getxattr,
2312        .listxattr      = shmem_listxattr,
2313        .removexattr    = shmem_removexattr,
2314#endif
2315#ifdef CONFIG_TMPFS_POSIX_ACL
2316        .setattr        = shmem_setattr,
2317#endif
2318};
2319
2320static const struct inode_operations shmem_special_inode_operations = {
2321#ifdef CONFIG_TMPFS_XATTR
2322        .setxattr       = shmem_setxattr,
2323        .getxattr       = shmem_getxattr,
2324        .listxattr      = shmem_listxattr,
2325        .removexattr    = shmem_removexattr,
2326#endif
2327#ifdef CONFIG_TMPFS_POSIX_ACL
2328        .setattr        = shmem_setattr,
2329#endif
2330};
2331
2332static const struct super_operations shmem_ops = {
2333        .alloc_inode    = shmem_alloc_inode,
2334        .destroy_inode  = shmem_destroy_inode,
2335#ifdef CONFIG_TMPFS
2336        .statfs         = shmem_statfs,
2337        .remount_fs     = shmem_remount_fs,
2338        .show_options   = shmem_show_options,
2339#endif
2340        .evict_inode    = shmem_evict_inode,
2341        .drop_inode     = generic_delete_inode,
2342        .put_super      = shmem_put_super,
2343};
2344
2345static const struct vm_operations_struct shmem_vm_ops = {
2346        .fault          = shmem_fault,
2347#ifdef CONFIG_NUMA
2348        .set_policy     = shmem_set_policy,
2349        .get_policy     = shmem_get_policy,
2350#endif
2351};
2352
2353static struct dentry *shmem_mount(struct file_system_type *fs_type,
2354        int flags, const char *dev_name, void *data)
2355{
2356        return mount_nodev(fs_type, flags, data, shmem_fill_super);
2357}
2358
2359static struct file_system_type shmem_fs_type = {
2360        .owner          = THIS_MODULE,
2361        .name           = "tmpfs",
2362        .mount          = shmem_mount,
2363        .kill_sb        = kill_litter_super,
2364};
2365
2366int __init shmem_init(void)
2367{
2368        int error;
2369
2370        error = bdi_init(&shmem_backing_dev_info);
2371        if (error)
2372                goto out4;
2373
2374        error = shmem_init_inodecache();
2375        if (error)
2376                goto out3;
2377
2378        error = register_filesystem(&shmem_fs_type);
2379        if (error) {
2380                printk(KERN_ERR "Could not register tmpfs\n");
2381                goto out2;
2382        }
2383
2384        shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER,
2385                                 shmem_fs_type.name, NULL);
2386        if (IS_ERR(shm_mnt)) {
2387                error = PTR_ERR(shm_mnt);
2388                printk(KERN_ERR "Could not kern_mount tmpfs\n");
2389                goto out1;
2390        }
2391        return 0;
2392
2393out1:
2394        unregister_filesystem(&shmem_fs_type);
2395out2:
2396        shmem_destroy_inodecache();
2397out3:
2398        bdi_destroy(&shmem_backing_dev_info);
2399out4:
2400        shm_mnt = ERR_PTR(error);
2401        return error;
2402}
2403
2404#else /* !CONFIG_SHMEM */
2405
2406/*
2407 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2408 *
2409 * This is intended for small system where the benefits of the full
2410 * shmem code (swap-backed and resource-limited) are outweighed by
2411 * their complexity. On systems without swap this code should be
2412 * effectively equivalent, but much lighter weight.
2413 */
2414
2415#include <linux/ramfs.h>
2416
2417static struct file_system_type shmem_fs_type = {
2418        .name           = "tmpfs",
2419        .mount          = ramfs_mount,
2420        .kill_sb        = kill_litter_super,
2421};
2422
2423int __init shmem_init(void)
2424{
2425        BUG_ON(register_filesystem(&shmem_fs_type) != 0);
2426
2427        shm_mnt = kern_mount(&shmem_fs_type);
2428        BUG_ON(IS_ERR(shm_mnt));
2429
2430        return 0;
2431}
2432
2433int shmem_unuse(swp_entry_t swap, struct page *page)
2434{
2435        return 0;
2436}
2437
2438int shmem_lock(struct file *file, int lock, struct user_struct *user)
2439{
2440        return 0;
2441}
2442
2443void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
2444{
2445        truncate_inode_pages_range(inode->i_mapping, lstart, lend);
2446}
2447EXPORT_SYMBOL_GPL(shmem_truncate_range);
2448
2449#define shmem_vm_ops                            generic_file_vm_ops
2450#define shmem_file_operations                   ramfs_file_operations
2451#define shmem_get_inode(sb, dir, mode, dev, flags)      ramfs_get_inode(sb, dir, mode, dev)
2452#define shmem_acct_size(flags, size)            0
2453#define shmem_unacct_size(flags, size)          do {} while (0)
2454
2455#endif /* CONFIG_SHMEM */
2456
2457/* common code */
2458
2459/**
2460 * shmem_file_setup - get an unlinked file living in tmpfs
2461 * @name: name for dentry (to be seen in /proc/<pid>/maps
2462 * @size: size to be set for the file
2463 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2464 */
2465struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2466{
2467        int error;
2468        struct file *file;
2469        struct inode *inode;
2470        struct path path;
2471        struct dentry *root;
2472        struct qstr this;
2473
2474        if (IS_ERR(shm_mnt))
2475                return (void *)shm_mnt;
2476
2477        if (size < 0 || size > MAX_LFS_FILESIZE)
2478                return ERR_PTR(-EINVAL);
2479
2480        if (shmem_acct_size(flags, size))
2481                return ERR_PTR(-ENOMEM);
2482
2483        error = -ENOMEM;
2484        this.name = name;
2485        this.len = strlen(name);
2486        this.hash = 0; /* will go */
2487        root = shm_mnt->mnt_root;
2488        path.dentry = d_alloc(root, &this);
2489        if (!path.dentry)
2490                goto put_memory;
2491        path.mnt = mntget(shm_mnt);
2492
2493        error = -ENOSPC;
2494        inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2495        if (!inode)
2496                goto put_dentry;
2497
2498        d_instantiate(path.dentry, inode);
2499        inode->i_size = size;
2500        inode->i_nlink = 0;     /* It is unlinked */
2501#ifndef CONFIG_MMU
2502        error = ramfs_nommu_expand_for_mapping(inode, size);
2503        if (error)
2504                goto put_dentry;
2505#endif
2506
2507        error = -ENFILE;
2508        file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2509                  &shmem_file_operations);
2510        if (!file)
2511                goto put_dentry;
2512
2513        return file;
2514
2515put_dentry:
2516        path_put(&path);
2517put_memory:
2518        shmem_unacct_size(flags, size);
2519        return ERR_PTR(error);
2520}
2521EXPORT_SYMBOL_GPL(shmem_file_setup);
2522
2523/**
2524 * shmem_zero_setup - setup a shared anonymous mapping
2525 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2526 */
2527int shmem_zero_setup(struct vm_area_struct *vma)
2528{
2529        struct file *file;
2530        loff_t size = vma->vm_end - vma->vm_start;
2531
2532        file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2533        if (IS_ERR(file))
2534                return PTR_ERR(file);
2535
2536        if (vma->vm_file)
2537                fput(vma->vm_file);
2538        vma->vm_file = file;
2539        vma->vm_ops = &shmem_vm_ops;
2540        vma->vm_flags |= VM_CAN_NONLINEAR;
2541        return 0;
2542}
2543
2544/**
2545 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2546 * @mapping:    the page's address_space
2547 * @index:      the page index
2548 * @gfp:        the page allocator flags to use if allocating
2549 *
2550 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2551 * with any new page allocations done using the specified allocation flags.
2552 * But read_cache_page_gfp() uses the ->readpage() method: which does not
2553 * suit tmpfs, since it may have pages in swapcache, and needs to find those
2554 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
2555 *
2556 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
2557 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
2558 */
2559struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
2560                                         pgoff_t index, gfp_t gfp)
2561{
2562#ifdef CONFIG_SHMEM
2563        struct inode *inode = mapping->host;
2564        struct page *page;
2565        int error;
2566
2567        BUG_ON(mapping->a_ops != &shmem_aops);
2568        error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
2569        if (error)
2570                page = ERR_PTR(error);
2571        else
2572                unlock_page(page);
2573        return page;
2574#else
2575        /*
2576         * The tiny !SHMEM case uses ramfs without swap
2577         */
2578        return read_cache_page_gfp(mapping, index, gfp);
2579#endif
2580}
2581EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);
2582
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.