mm/shmem.c

   1 /*
   2  * Resizable virtual memory filesystem for Linux.
   3  *
   4  * Copyright (C) 2000 Linus Torvalds.
   5  *               2000 Transmeta Corp.
   6  *               2000-2001 Christoph Rohland
   7  *               2000-2001 SAP AG
   8  *               2002 Red Hat Inc.
   9  * Copyright (C) 2002-2011 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/ramfs.h>
  29 #include <linux/pagemap.h>
  30 #include <linux/file.h>
  31 #include <linux/mm.h>
  32 #include <linux/export.h>
  33 #include <linux/swap.h>
  34 #include <linux/aio.h>
  35
  36 static struct vfsmount *shm_mnt;
  37
  38 #ifdef CONFIG_SHMEM
  39 /*
  40  * This virtual memory filesystem is heavily based on the ramfs. It
  41  * extends ramfs by the ability to use swap and honor resource limits
  42  * which makes it a completely usable filesystem.
  43  */
  44
  45 #include <linux/xattr.h>
  46 #include <linux/exportfs.h>
  47 #include <linux/posix_acl.h>
  48 #include <linux/posix_acl_xattr.h>
  49 #include <linux/mman.h>
  50 #include <linux/string.h>
  51 #include <linux/slab.h>
  52 #include <linux/backing-dev.h>
  53 #include <linux/shmem_fs.h>
  54 #include <linux/writeback.h>
  55 #include <linux/blkdev.h>
  56 #include <linux/pagevec.h>
  57 #include <linux/percpu_counter.h>
  58 #include <linux/falloc.h>
  59 #include <linux/splice.h>
  60 #include <linux/security.h>
  61 #include <linux/swapops.h>
  62 #include <linux/mempolicy.h>
  63 #include <linux/namei.h>
  64 #include <linux/ctype.h>
  65 #include <linux/migrate.h>
  66 #include <linux/highmem.h>
  67 #include <linux/seq_file.h>
  68 #include <linux/magic.h>
  69
  70 #include <asm/uaccess.h>
  71 #include <asm/pgtable.h>
  72
  73 #define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
  74 #define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
  75
  76 /* Pretend that each entry is of this size in directory's i_size */
  77 #define BOGO_DIRENT_SIZE 20
  78
  79 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
  80 #define SHORT_SYMLINK_LEN 128
  81
  82 /*
  83  * shmem_fallocate communicates with shmem_fault or shmem_writepage via
  84  * inode->i_private (with i_mutex making sure that it has only one user at
  85  * a time): we would prefer not to enlarge the shmem inode just for that.
  86  */
  87 struct shmem_falloc {
  88         int     mode;           /* FALLOC_FL mode currently operating */
  89         pgoff_t start;          /* start of range currently being fallocated */
  90         pgoff_t next;           /* the next page offset to be fallocated */
  91         pgoff_t nr_falloced;    /* how many new pages have been fallocated */
  92         pgoff_t nr_unswapped;   /* how often writepage refused to swap out */
  93 };
  94
  95 /* Flag allocation requirements to shmem_getpage */
  96 enum sgp_type {
  97         SGP_READ,       /* don't exceed i_size, don't allocate page */
  98         SGP_CACHE,      /* don't exceed i_size, may allocate page */
  99         SGP_DIRTY,      /* like SGP_CACHE, but set new page dirty */
 100         SGP_WRITE,      /* may exceed i_size, may allocate !Uptodate page */
 101         SGP_FALLOC,     /* like SGP_WRITE, but make existing page Uptodate */
 102 };
 103
 104 #ifdef CONFIG_TMPFS
 105 static unsigned long shmem_default_max_blocks(void)
 106 {
 107         return totalram_pages / 2;
 108 }
 109
 110 static unsigned long shmem_default_max_inodes(void)
 111 {
 112         return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
 113 }
 114 #endif
 115
 116 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
 117 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
 118                                 struct shmem_inode_info *info, pgoff_t index);
 119 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
 120         struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
 121
 122 static inline int shmem_getpage(struct inode *inode, pgoff_t index,
 123         struct page **pagep, enum sgp_type sgp, int *fault_type)
 124 {
 125         return shmem_getpage_gfp(inode, index, pagep, sgp,
 126                         mapping_gfp_mask(inode->i_mapping), fault_type);
 127 }
 128
 129 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
 130 {
 131         return sb->s_fs_info;
 132 }
 133
 134 /*
 135  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
 136  * for shared memory and for shared anonymous (/dev/zero) mappings
 137  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
 138  * consistent with the pre-accounting of private mappings ...
 139  */
 140 static inline int shmem_acct_size(unsigned long flags, loff_t size)
 141 {
 142         return (flags & VM_NORESERVE) ?
 143                 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
 144 }
 145
 146 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
 147 {
 148         if (!(flags & VM_NORESERVE))
 149                 vm_unacct_memory(VM_ACCT(size));
 150 }
 151
 152 /*
 153  * ... whereas tmpfs objects are accounted incrementally as
 154  * pages are allocated, in order to allow huge sparse files.
 155  * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
 156  * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
 157  */
 158 static inline int shmem_acct_block(unsigned long flags)
 159 {
 160         return (flags & VM_NORESERVE) ?
 161                 security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0;
 162 }
 163
 164 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
 165 {
 166         if (flags & VM_NORESERVE)
 167                 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
 168 }
 169
 170 static const struct super_operations shmem_ops;
 171 static const struct address_space_operations shmem_aops;
 172 static const struct file_operations shmem_file_operations;
 173 static const struct inode_operations shmem_inode_operations;
 174 static const struct inode_operations shmem_dir_inode_operations;
 175 static const struct inode_operations shmem_special_inode_operations;
 176 static const struct vm_operations_struct shmem_vm_ops;
 177
 178 static struct backing_dev_info shmem_backing_dev_info  __read_mostly = {
 179         .ra_pages       = 0,    /* No readahead */
 180         .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
 181 };
 182
 183 static LIST_HEAD(shmem_swaplist);
 184 static DEFINE_MUTEX(shmem_swaplist_mutex);
 185
 186 static int shmem_reserve_inode(struct super_block *sb)
 187 {
 188         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 189         if (sbinfo->max_inodes) {
 190                 spin_lock(&sbinfo->stat_lock);
 191                 if (!sbinfo->free_inodes) {
 192                         spin_unlock(&sbinfo->stat_lock);
 193                         return -ENOSPC;
 194                 }
 195                 sbinfo->free_inodes--;
 196                 spin_unlock(&sbinfo->stat_lock);
 197         }
 198         return 0;
 199 }
 200
 201 static void shmem_free_inode(struct super_block *sb)
 202 {
 203         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 204         if (sbinfo->max_inodes) {
 205                 spin_lock(&sbinfo->stat_lock);
 206                 sbinfo->free_inodes++;
 207                 spin_unlock(&sbinfo->stat_lock);
 208         }
 209 }
 210
 211 /**
 212  * shmem_recalc_inode - recalculate the block usage of an inode
 213  * @inode: inode to recalc
 214  *
 215  * We have to calculate the free blocks since the mm can drop
 216  * undirtied hole pages behind our back.
 217  *
 218  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
 219  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
 220  *
 221  * It has to be called with the spinlock held.
 222  */
 223 static void shmem_recalc_inode(struct inode *inode)
 224 {
 225         struct shmem_inode_info *info = SHMEM_I(inode);
 226         long freed;
 227
 228         freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
 229         if (freed > 0) {
 230                 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 231                 if (sbinfo->max_blocks)
 232                         percpu_counter_add(&sbinfo->used_blocks, -freed);
 233                 info->alloced -= freed;
 234                 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
 235                 shmem_unacct_blocks(info->flags, freed);
 236         }
 237 }
 238
 239 /*
 240  * Replace item expected in radix tree by a new item, while holding tree lock.
 241  */
 242 static int shmem_radix_tree_replace(struct address_space *mapping,
 243                         pgoff_t index, void *expected, void *replacement)
 244 {
 245         void **pslot;
 246         void *item;
 247
 248         VM_BUG_ON(!expected);
 249         VM_BUG_ON(!replacement);
 250         pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
 251         if (!pslot)
 252                 return -ENOENT;
 253         item = radix_tree_deref_slot_protected(pslot, &mapping->tree_lock);
 254         if (item != expected)
 255                 return -ENOENT;
 256         radix_tree_replace_slot(pslot, replacement);
 257         return 0;
 258 }
 259
 260 /*
 261  * Sometimes, before we decide whether to proceed or to fail, we must check
 262  * that an entry was not already brought back from swap by a racing thread.
 263  *
 264  * Checking page is not enough: by the time a SwapCache page is locked, it
 265  * might be reused, and again be SwapCache, using the same swap as before.
 266  */
 267 static bool shmem_confirm_swap(struct address_space *mapping,
 268                                pgoff_t index, swp_entry_t swap)
 269 {
 270         void *item;
 271
 272         rcu_read_lock();
 273         item = radix_tree_lookup(&mapping->page_tree, index);
 274         rcu_read_unlock();
 275         return item == swp_to_radix_entry(swap);
 276 }
 277
 278 /*
 279  * Like add_to_page_cache_locked, but error if expected item has gone.
 280  */
 281 static int shmem_add_to_page_cache(struct page *page,
 282                                    struct address_space *mapping,
 283                                    pgoff_t index, gfp_t gfp, void *expected)
 284 {
 285         int error;
 286
 287         VM_BUG_ON_PAGE(!PageLocked(page), page);
 288         VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
 289
 290         page_cache_get(page);
 291         page->mapping = mapping;
 292         page->index = index;
 293
 294         spin_lock_irq(&mapping->tree_lock);
 295         if (!expected)
 296                 error = radix_tree_insert(&mapping->page_tree, index, page);
 297         else
 298                 error = shmem_radix_tree_replace(mapping, index, expected,
 299                                                                  page);
 300         if (!error) {
 301                 mapping->nrpages++;
 302                 __inc_zone_page_state(page, NR_FILE_PAGES);
 303                 __inc_zone_page_state(page, NR_SHMEM);
 304                 spin_unlock_irq(&mapping->tree_lock);
 305         } else {
 306                 page->mapping = NULL;
 307                 spin_unlock_irq(&mapping->tree_lock);
 308                 page_cache_release(page);
 309         }
 310         return error;
 311 }
 312
 313 /*
 314  * Like delete_from_page_cache, but substitutes swap for page.
 315  */
 316 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
 317 {
 318         struct address_space *mapping = page->mapping;
 319         int error;
 320
 321         spin_lock_irq(&mapping->tree_lock);
 322         error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
 323         page->mapping = NULL;
 324         mapping->nrpages--;
 325         __dec_zone_page_state(page, NR_FILE_PAGES);
 326         __dec_zone_page_state(page, NR_SHMEM);
 327         spin_unlock_irq(&mapping->tree_lock);
 328         page_cache_release(page);
 329         BUG_ON(error);
 330 }
 331
 332 /*
 333  * Remove swap entry from radix tree, free the swap and its page cache.
 334  */
 335 static int shmem_free_swap(struct address_space *mapping,
 336                            pgoff_t index, void *radswap)
 337 {
 338         void *old;
 339
 340         spin_lock_irq(&mapping->tree_lock);
 341         old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
 342         spin_unlock_irq(&mapping->tree_lock);
 343         if (old != radswap)
 344                 return -ENOENT;
 345         free_swap_and_cache(radix_to_swp_entry(radswap));
 346         return 0;
 347 }
 348
 349 /*
 350  * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
 351  */
 352 void shmem_unlock_mapping(struct address_space *mapping)
 353 {
 354         struct pagevec pvec;
 355         pgoff_t indices[PAGEVEC_SIZE];
 356         pgoff_t index = 0;
 357
 358         pagevec_init(&pvec, 0);
 359         /*
 360          * Minor point, but we might as well stop if someone else SHM_LOCKs it.
 361          */
 362         while (!mapping_unevictable(mapping)) {
 363                 /*
 364                  * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
 365                  * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
 366                  */
 367                 pvec.nr = find_get_entries(mapping, index,
 368                                            PAGEVEC_SIZE, pvec.pages, indices);
 369                 if (!pvec.nr)
 370                         break;
 371                 index = indices[pvec.nr - 1] + 1;
 372                 pagevec_remove_exceptionals(&pvec);
 373                 check_move_unevictable_pages(pvec.pages, pvec.nr);
 374                 pagevec_release(&pvec);
 375                 cond_resched();
 376         }
 377 }
 378
 379 /*
 380  * Remove range of pages and swap entries from radix tree, and free them.
 381  * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
 382  */
 383 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
 384                                                                  bool unfalloc)
 385 {
 386         struct address_space *mapping = inode->i_mapping;
 387         struct shmem_inode_info *info = SHMEM_I(inode);
 388         pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 389         pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
 390         unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
 391         unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
 392         struct pagevec pvec;
 393         pgoff_t indices[PAGEVEC_SIZE];
 394         long nr_swaps_freed = 0;
 395         pgoff_t index;
 396         int i;
 397
 398         if (lend == -1)
 399                 end = -1;       /* unsigned, so actually very big */
 400
 401         pagevec_init(&pvec, 0);
 402         index = start;
 403         while (index < end) {
 404                 pvec.nr = find_get_entries(mapping, index,
 405                         min(end - index, (pgoff_t)PAGEVEC_SIZE),
 406                         pvec.pages, indices);
 407                 if (!pvec.nr)
 408                         break;
 409                 mem_cgroup_uncharge_start();
 410                 for (i = 0; i < pagevec_count(&pvec); i++) {
 411                         struct page *page = pvec.pages[i];
 412
 413                         index = indices[i];
 414                         if (index >= end)
 415                                 break;
 416
 417                         if (radix_tree_exceptional_entry(page)) {
 418                                 if (unfalloc)
 419                                         continue;
 420                                 nr_swaps_freed += !shmem_free_swap(mapping,
 421                                                                 index, page);
 422                                 continue;
 423                         }
 424
 425                         if (!trylock_page(page))
 426                                 continue;
 427                         if (!unfalloc || !PageUptodate(page)) {
 428                                 if (page->mapping == mapping) {
 429                                         VM_BUG_ON_PAGE(PageWriteback(page), page);
 430                                         truncate_inode_page(mapping, page);
 431                                 }
 432                         }
 433                         unlock_page(page);
 434                 }
 435                 pagevec_remove_exceptionals(&pvec);
 436                 pagevec_release(&pvec);
 437                 mem_cgroup_uncharge_end();
 438                 cond_resched();
 439                 index++;
 440         }
 441
 442         if (partial_start) {
 443                 struct page *page = NULL;
 444                 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
 445                 if (page) {
 446                         unsigned int top = PAGE_CACHE_SIZE;
 447                         if (start > end) {
 448                                 top = partial_end;
 449                                 partial_end = 0;
 450                         }
 451                         zero_user_segment(page, partial_start, top);
 452                         set_page_dirty(page);
 453                         unlock_page(page);
 454                         page_cache_release(page);
 455                 }
 456         }
 457         if (partial_end) {
 458                 struct page *page = NULL;
 459                 shmem_getpage(inode, end, &page, SGP_READ, NULL);
 460                 if (page) {
 461                         zero_user_segment(page, 0, partial_end);
 462                         set_page_dirty(page);
 463                         unlock_page(page);
 464                         page_cache_release(page);
 465                 }
 466         }
 467         if (start >= end)
 468                 return;
 469
 470         index = start;
 471         for ( ; ; ) {
 472                 cond_resched();
 473
 474                 pvec.nr = find_get_entries(mapping, index,
 475                                 min(end - index, (pgoff_t)PAGEVEC_SIZE),
 476                                 pvec.pages, indices);
 477                 if (!pvec.nr) {
 478                         if (index == start || unfalloc)
 479                                 break;
 480                         index = start;
 481                         continue;
 482                 }
 483                 if ((index == start || unfalloc) && indices[0] >= end) {
 484                         pagevec_remove_exceptionals(&pvec);
 485                         pagevec_release(&pvec);
 486                         break;
 487                 }
 488                 mem_cgroup_uncharge_start();
 489                 for (i = 0; i < pagevec_count(&pvec); i++) {
 490                         struct page *page = pvec.pages[i];
 491
 492                         index = indices[i];
 493                         if (index >= end)
 494                                 break;
 495
 496                         if (radix_tree_exceptional_entry(page)) {
 497                                 if (unfalloc)
 498                                         continue;
 499                                 nr_swaps_freed += !shmem_free_swap(mapping,
 500                                                                 index, page);
 501                                 continue;
 502                         }
 503
 504                         lock_page(page);
 505                         if (!unfalloc || !PageUptodate(page)) {
 506                                 if (page->mapping == mapping) {
 507                                         VM_BUG_ON_PAGE(PageWriteback(page), page);
 508                                         truncate_inode_page(mapping, page);
 509                                 }
 510                         }
 511                         unlock_page(page);
 512                 }
 513                 pagevec_remove_exceptionals(&pvec);
 514                 pagevec_release(&pvec);
 515                 mem_cgroup_uncharge_end();
 516                 index++;
 517         }
 518
 519         spin_lock(&info->lock);
 520         info->swapped -= nr_swaps_freed;
 521         shmem_recalc_inode(inode);
 522         spin_unlock(&info->lock);
 523 }
 524
 525 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
 526 {
 527         shmem_undo_range(inode, lstart, lend, false);
 528         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 529 }
 530 EXPORT_SYMBOL_GPL(shmem_truncate_range);
 531
 532 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
 533 {
 534         struct inode *inode = dentry->d_inode;
 535         int error;
 536
 537         error = inode_change_ok(inode, attr);
 538         if (error)
 539                 return error;
 540
 541         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
 542                 loff_t oldsize = inode->i_size;
 543                 loff_t newsize = attr->ia_size;
 544
 545                 if (newsize != oldsize) {
 546                         i_size_write(inode, newsize);
 547                         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 548                 }
 549                 if (newsize < oldsize) {
 550                         loff_t holebegin = round_up(newsize, PAGE_SIZE);
 551                         unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
 552                         shmem_truncate_range(inode, newsize, (loff_t)-1);
 553                         /* unmap again to remove racily COWed private pages */
 554                         unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
 555                 }
 556         }
 557
 558         setattr_copy(inode, attr);
 559         if (attr->ia_valid & ATTR_MODE)
 560                 error = posix_acl_chmod(inode, inode->i_mode);
 561         return error;
 562 }
 563
 564 static void shmem_evict_inode(struct inode *inode)
 565 {
 566         struct shmem_inode_info *info = SHMEM_I(inode);
 567
 568         if (inode->i_mapping->a_ops == &shmem_aops) {
 569                 shmem_unacct_size(info->flags, inode->i_size);
 570                 inode->i_size = 0;
 571                 shmem_truncate_range(inode, 0, (loff_t)-1);
 572                 if (!list_empty(&info->swaplist)) {
 573                         mutex_lock(&shmem_swaplist_mutex);
 574                         list_del_init(&info->swaplist);
 575                         mutex_unlock(&shmem_swaplist_mutex);
 576                 }
 577         } else
 578                 kfree(info->symlink);
 579
 580         simple_xattrs_free(&info->xattrs);
 581         WARN_ON(inode->i_blocks);
 582         shmem_free_inode(inode->i_sb);
 583         clear_inode(inode);
 584 }
 585
 586 /*
 587  * If swap found in inode, free it and move page from swapcache to filecache.
 588  */
 589 static int shmem_unuse_inode(struct shmem_inode_info *info,
 590                              swp_entry_t swap, struct page **pagep)
 591 {
 592         struct address_space *mapping = info->vfs_inode.i_mapping;
 593         void *radswap;
 594         pgoff_t index;
 595         gfp_t gfp;
 596         int error = 0;
 597
 598         radswap = swp_to_radix_entry(swap);
 599         index = radix_tree_locate_item(&mapping->page_tree, radswap);
 600         if (index == -1)
 601                 return 0;
 602
 603         /*
 604          * Move _head_ to start search for next from here.
 605          * But be careful: shmem_evict_inode checks list_empty without taking
 606          * mutex, and there's an instant in list_move_tail when info->swaplist
 607          * would appear empty, if it were the only one on shmem_swaplist.
 608          */
 609         if (shmem_swaplist.next != &info->swaplist)
 610                 list_move_tail(&shmem_swaplist, &info->swaplist);
 611
 612         gfp = mapping_gfp_mask(mapping);
 613         if (shmem_should_replace_page(*pagep, gfp)) {
 614                 mutex_unlock(&shmem_swaplist_mutex);
 615                 error = shmem_replace_page(pagep, gfp, info, index);
 616                 mutex_lock(&shmem_swaplist_mutex);
 617                 /*
 618                  * We needed to drop mutex to make that restrictive page
 619                  * allocation, but the inode might have been freed while we
 620                  * dropped it: although a racing shmem_evict_inode() cannot
 621                  * complete without emptying the radix_tree, our page lock
 622                  * on this swapcache page is not enough to prevent that -
 623                  * free_swap_and_cache() of our swap entry will only
 624                  * trylock_page(), removing swap from radix_tree whatever.
 625                  *
 626                  * We must not proceed to shmem_add_to_page_cache() if the
 627                  * inode has been freed, but of course we cannot rely on
 628                  * inode or mapping or info to check that.  However, we can
 629                  * safely check if our swap entry is still in use (and here
 630                  * it can't have got reused for another page): if it's still
 631                  * in use, then the inode cannot have been freed yet, and we
 632                  * can safely proceed (if it's no longer in use, that tells
 633                  * nothing about the inode, but we don't need to unuse swap).
 634                  */
 635                 if (!page_swapcount(*pagep))
 636                         error = -ENOENT;
 637         }
 638
 639         /*
 640          * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
 641          * but also to hold up shmem_evict_inode(): so inode cannot be freed
 642          * beneath us (pagelock doesn't help until the page is in pagecache).
 643          */
 644         if (!error)
 645                 error = shmem_add_to_page_cache(*pagep, mapping, index,
 646                                                 GFP_NOWAIT, radswap);
 647         if (error != -ENOMEM) {
 648                 /*
 649                  * Truncation and eviction use free_swap_and_cache(), which
 650                  * only does trylock page: if we raced, best clean up here.
 651                  */
 652                 delete_from_swap_cache(*pagep);
 653                 set_page_dirty(*pagep);
 654                 if (!error) {
 655                         spin_lock(&info->lock);
 656                         info->swapped--;
 657                         spin_unlock(&info->lock);
 658                         swap_free(swap);
 659                 }
 660                 error = 1;      /* not an error, but entry was found */
 661         }
 662         return error;
 663 }
 664
 665 /*
 666  * Search through swapped inodes to find and replace swap by page.
 667  */
 668 int shmem_unuse(swp_entry_t swap, struct page *page)
 669 {
 670         struct list_head *this, *next;
 671         struct shmem_inode_info *info;
 672         int found = 0;
 673         int error = 0;
 674
 675         /*
 676          * There's a faint possibility that swap page was replaced before
 677          * caller locked it: caller will come back later with the right page.
 678          */
 679         if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
 680                 goto out;
 681
 682         /*
 683          * Charge page using GFP_KERNEL while we can wait, before taking
 684          * the shmem_swaplist_mutex which might hold up shmem_writepage().
 685          * Charged back to the user (not to caller) when swap account is used.
 686          */
 687         error = mem_cgroup_charge_file(page, current->mm, GFP_KERNEL);
 688         if (error)
 689                 goto out;
 690         /* No radix_tree_preload: swap entry keeps a place for page in tree */
 691
 692         mutex_lock(&shmem_swaplist_mutex);
 693         list_for_each_safe(this, next, &shmem_swaplist) {
 694                 info = list_entry(this, struct shmem_inode_info, swaplist);
 695                 if (info->swapped)
 696                         found = shmem_unuse_inode(info, swap, &page);
 697                 else
 698                         list_del_init(&info->swaplist);
 699                 cond_resched();
 700                 if (found)
 701                         break;
 702         }
 703         mutex_unlock(&shmem_swaplist_mutex);
 704
 705         if (found < 0)
 706                 error = found;
 707 out:
 708         unlock_page(page);
 709         page_cache_release(page);
 710         return error;
 711 }
 712
 713 /*
 714  * Move the page from the page cache to the swap cache.
 715  */
 716 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
 717 {
 718         struct shmem_inode_info *info;
 719         struct address_space *mapping;
 720         struct inode *inode;
 721         swp_entry_t swap;
 722         pgoff_t index;
 723
 724         BUG_ON(!PageLocked(page));
 725         mapping = page->mapping;
 726         index = page->index;
 727         inode = mapping->host;
 728         info = SHMEM_I(inode);
 729         if (info->flags & VM_LOCKED)
 730                 goto redirty;
 731         if (!total_swap_pages)
 732                 goto redirty;
 733
 734         /*
 735          * shmem_backing_dev_info's capabilities prevent regular writeback or
 736          * sync from ever calling shmem_writepage; but a stacking filesystem
 737          * might use ->writepage of its underlying filesystem, in which case
 738          * tmpfs should write out to swap only in response to memory pressure,
 739          * and not for the writeback threads or sync.
 740          */
 741         if (!wbc->for_reclaim) {
 742                 WARN_ON_ONCE(1);        /* Still happens? Tell us about it! */
 743                 goto redirty;
 744         }
 745
 746         /*
 747          * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
 748          * value into swapfile.c, the only way we can correctly account for a
 749          * fallocated page arriving here is now to initialize it and write it.
 750          *
 751          * That's okay for a page already fallocated earlier, but if we have
 752          * not yet completed the fallocation, then (a) we want to keep track
 753          * of this page in case we have to undo it, and (b) it may not be a
 754          * good idea to continue anyway, once we're pushing into swap.  So
 755          * reactivate the page, and let shmem_fallocate() quit when too many.
 756          */
 757         if (!PageUptodate(page)) {
 758                 if (inode->i_private) {
 759                         struct shmem_falloc *shmem_falloc;
 760                         spin_lock(&inode->i_lock);
 761                         shmem_falloc = inode->i_private;
 762                         if (shmem_falloc &&
 763                             !shmem_falloc->mode &&
 764                             index >= shmem_falloc->start &&
 765                             index < shmem_falloc->next)
 766                                 shmem_falloc->nr_unswapped++;
 767                         else
 768                                 shmem_falloc = NULL;
 769                         spin_unlock(&inode->i_lock);
 770                         if (shmem_falloc)
 771                                 goto redirty;
 772                 }
 773                 clear_highpage(page);
 774                 flush_dcache_page(page);
 775                 SetPageUptodate(page);
 776         }
 777
 778         swap = get_swap_page();
 779         if (!swap.val)
 780                 goto redirty;
 781
 782         /*
 783          * Add inode to shmem_unuse()'s list of swapped-out inodes,
 784          * if it's not already there.  Do it now before the page is
 785          * moved to swap cache, when its pagelock no longer protects
 786          * the inode from eviction.  But don't unlock the mutex until
 787          * we've incremented swapped, because shmem_unuse_inode() will
 788          * prune a !swapped inode from the swaplist under this mutex.
 789          */
 790         mutex_lock(&shmem_swaplist_mutex);
 791         if (list_empty(&info->swaplist))
 792                 list_add_tail(&info->swaplist, &shmem_swaplist);
 793
 794         if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
 795                 swap_shmem_alloc(swap);
 796                 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
 797
 798                 spin_lock(&info->lock);
 799                 info->swapped++;
 800                 shmem_recalc_inode(inode);
 801                 spin_unlock(&info->lock);
 802
 803                 mutex_unlock(&shmem_swaplist_mutex);
 804                 BUG_ON(page_mapped(page));
 805                 swap_writepage(page, wbc);
 806                 return 0;
 807         }
 808
 809         mutex_unlock(&shmem_swaplist_mutex);
 810         swapcache_free(swap, NULL);
 811 redirty:
 812         set_page_dirty(page);
 813         if (wbc->for_reclaim)
 814                 return AOP_WRITEPAGE_ACTIVATE;  /* Return with page locked */
 815         unlock_page(page);
 816         return 0;
 817 }
 818
 819 #ifdef CONFIG_NUMA
 820 #ifdef CONFIG_TMPFS
 821 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
 822 {
 823         char buffer[64];
 824
 825         if (!mpol || mpol->mode == MPOL_DEFAULT)
 826                 return;         /* show nothing */
 827
 828         mpol_to_str(buffer, sizeof(buffer), mpol);
 829
 830         seq_printf(seq, ",mpol=%s", buffer);
 831 }
 832
 833 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
 834 {
 835         struct mempolicy *mpol = NULL;
 836         if (sbinfo->mpol) {
 837                 spin_lock(&sbinfo->stat_lock);  /* prevent replace/use races */
 838                 mpol = sbinfo->mpol;
 839                 mpol_get(mpol);
 840                 spin_unlock(&sbinfo->stat_lock);
 841         }
 842         return mpol;
 843 }
 844 #endif /* CONFIG_TMPFS */
 845
 846 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
 847                         struct shmem_inode_info *info, pgoff_t index)
 848 {
 849         struct vm_area_struct pvma;
 850         struct page *page;
 851
 852         /* Create a pseudo vma that just contains the policy */
 853         pvma.vm_start = 0;
 854         /* Bias interleave by inode number to distribute better across nodes */
 855         pvma.vm_pgoff = index + info->vfs_inode.i_ino;
 856         pvma.vm_ops = NULL;
 857         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
 858
 859         page = swapin_readahead(swap, gfp, &pvma, 0);
 860
 861         /* Drop reference taken by mpol_shared_policy_lookup() */
 862         mpol_cond_put(pvma.vm_policy);
 863
 864         return page;
 865 }
 866
 867 static struct page *shmem_alloc_page(gfp_t gfp,
 868                         struct shmem_inode_info *info, pgoff_t index)
 869 {
 870         struct vm_area_struct pvma;
 871         struct page *page;
 872
 873         /* Create a pseudo vma that just contains the policy */
 874         pvma.vm_start = 0;
 875         /* Bias interleave by inode number to distribute better across nodes */
 876         pvma.vm_pgoff = index + info->vfs_inode.i_ino;
 877         pvma.vm_ops = NULL;
 878         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
 879
 880         page = alloc_page_vma(gfp, &pvma, 0);
 881
 882         /* Drop reference taken by mpol_shared_policy_lookup() */
 883         mpol_cond_put(pvma.vm_policy);
 884
 885         return page;
 886 }
 887 #else /* !CONFIG_NUMA */
 888 #ifdef CONFIG_TMPFS
 889 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
 890 {
 891 }
 892 #endif /* CONFIG_TMPFS */
 893
 894 static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
 895                         struct shmem_inode_info *info, pgoff_t index)
 896 {
 897         return swapin_readahead(swap, gfp, NULL, 0);
 898 }
 899
 900 static inline struct page *shmem_alloc_page(gfp_t gfp,
 901                         struct shmem_inode_info *info, pgoff_t index)
 902 {
 903         return alloc_page(gfp);
 904 }
 905 #endif /* CONFIG_NUMA */
 906
 907 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
 908 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
 909 {
 910         return NULL;
 911 }
 912 #endif
 913
 914 /*
 915  * When a page is moved from swapcache to shmem filecache (either by the
 916  * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
 917  * shmem_unuse_inode()), it may have been read in earlier from swap, in
 918  * ignorance of the mapping it belongs to.  If that mapping has special
 919  * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
 920  * we may need to copy to a suitable page before moving to filecache.
 921  *
 922  * In a future release, this may well be extended to respect cpuset and
 923  * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
 924  * but for now it is a simple matter of zone.
 925  */
 926 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
 927 {
 928         return page_zonenum(page) > gfp_zone(gfp);
 929 }
 930
 931 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
 932                                 struct shmem_inode_info *info, pgoff_t index)
 933 {
 934         struct page *oldpage, *newpage;
 935         struct address_space *swap_mapping;
 936         pgoff_t swap_index;
 937         int error;
 938
 939         oldpage = *pagep;
 940         swap_index = page_private(oldpage);
 941         swap_mapping = page_mapping(oldpage);
 942
 943         /*
 944          * We have arrived here because our zones are constrained, so don't
 945          * limit chance of success by further cpuset and node constraints.
 946          */
 947         gfp &= ~GFP_CONSTRAINT_MASK;
 948         newpage = shmem_alloc_page(gfp, info, index);
 949         if (!newpage)
 950                 return -ENOMEM;
 951
 952         page_cache_get(newpage);
 953         copy_highpage(newpage, oldpage);
 954         flush_dcache_page(newpage);
 955
 956         __set_page_locked(newpage);
 957         SetPageUptodate(newpage);
 958         SetPageSwapBacked(newpage);
 959         set_page_private(newpage, swap_index);
 960         SetPageSwapCache(newpage);
 961
 962         /*
 963          * Our caller will very soon move newpage out of swapcache, but it's
 964          * a nice clean interface for us to replace oldpage by newpage there.
 965          */
 966         spin_lock_irq(&swap_mapping->tree_lock);
 967         error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
 968                                                                    newpage);
 969         if (!error) {
 970                 __inc_zone_page_state(newpage, NR_FILE_PAGES);
 971                 __dec_zone_page_state(oldpage, NR_FILE_PAGES);
 972         }
 973         spin_unlock_irq(&swap_mapping->tree_lock);
 974
 975         if (unlikely(error)) {
 976                 /*
 977                  * Is this possible?  I think not, now that our callers check
 978                  * both PageSwapCache and page_private after getting page lock;
 979                  * but be defensive.  Reverse old to newpage for clear and free.
 980                  */
 981                 oldpage = newpage;
 982         } else {
 983                 mem_cgroup_replace_page_cache(oldpage, newpage);
 984                 lru_cache_add_anon(newpage);
 985                 *pagep = newpage;
 986         }
 987
 988         ClearPageSwapCache(oldpage);
 989         set_page_private(oldpage, 0);
 990
 991         unlock_page(oldpage);
 992         page_cache_release(oldpage);
 993         page_cache_release(oldpage);
 994         return error;
 995 }
 996
 997 /*
 998  * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
 999  *
1000  * If we allocate a new one we do not mark it dirty. That's up to the
1001  * vm. If we swap it in we mark it dirty since we also free the swap
1002  * entry since a page cannot live in both the swap and page cache
1003  */
1004 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1005         struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
1006 {
1007         struct address_space *mapping = inode->i_mapping;
1008         struct shmem_inode_info *info;
1009         struct shmem_sb_info *sbinfo;
1010         struct page *page;
1011         swp_entry_t swap;
1012         int error;
1013         int once = 0;
1014         int alloced = 0;
1015
1016         if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
1017                 return -EFBIG;
1018 repeat:
1019         swap.val = 0;
1020         page = find_lock_entry(mapping, index);
1021         if (radix_tree_exceptional_entry(page)) {
1022                 swap = radix_to_swp_entry(page);
1023                 page = NULL;
1024         }
1025
1026         if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1027             ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1028                 error = -EINVAL;
1029                 goto failed;
1030         }
1031
1032         if (page && sgp == SGP_WRITE)
1033                 mark_page_accessed(page);
1034
1035         /* fallocated page? */
1036         if (page && !PageUptodate(page)) {
1037                 if (sgp != SGP_READ)
1038                         goto clear;
1039                 unlock_page(page);
1040                 page_cache_release(page);
1041                 page = NULL;
1042         }
1043         if (page || (sgp == SGP_READ && !swap.val)) {
1044                 *pagep = page;
1045                 return 0;
1046         }
1047
1048         /*
1049          * Fast cache lookup did not find it:
1050          * bring it back from swap or allocate.
1051          */
1052         info = SHMEM_I(inode);
1053         sbinfo = SHMEM_SB(inode->i_sb);
1054
1055         if (swap.val) {
1056                 /* Look it up and read it in.. */
1057                 page = lookup_swap_cache(swap);
1058                 if (!page) {
1059                         /* here we actually do the io */
1060                         if (fault_type)
1061                                 *fault_type |= VM_FAULT_MAJOR;
1062                         page = shmem_swapin(swap, gfp, info, index);
1063                         if (!page) {
1064                                 error = -ENOMEM;
1065                                 goto failed;
1066                         }
1067                 }
1068
1069                 /* We have to do this with page locked to prevent races */
1070                 lock_page(page);
1071                 if (!PageSwapCache(page) || page_private(page) != swap.val ||
1072                     !shmem_confirm_swap(mapping, index, swap)) {
1073                         error = -EEXIST;        /* try again */
1074                         goto unlock;
1075                 }
1076                 if (!PageUptodate(page)) {
1077                         error = -EIO;
1078                         goto failed;
1079                 }
1080                 wait_on_page_writeback(page);
1081
1082                 if (shmem_should_replace_page(page, gfp)) {
1083                         error = shmem_replace_page(&page, gfp, info, index);
1084                         if (error)
1085                                 goto failed;
1086                 }
1087
1088                 error = mem_cgroup_charge_file(page, current->mm,
1089                                                 gfp & GFP_RECLAIM_MASK);
1090                 if (!error) {
1091                         error = shmem_add_to_page_cache(page, mapping, index,
1092                                                 gfp, swp_to_radix_entry(swap));
1093                         /*
1094                          * We already confirmed swap under page lock, and make
1095                          * no memory allocation here, so usually no possibility
1096                          * of error; but free_swap_and_cache() only trylocks a
1097                          * page, so it is just possible that the entry has been
1098                          * truncated or holepunched since swap was confirmed.
1099                          * shmem_undo_range() will have done some of the
1100                          * unaccounting, now delete_from_swap_cache() will do
1101                          * the rest (including mem_cgroup_uncharge_swapcache).
1102                          * Reset swap.val? No, leave it so "failed" goes back to
1103                          * "repeat": reading a hole and writing should succeed.
1104                          */
1105                         if (error)
1106                                 delete_from_swap_cache(page);
1107                 }
1108                 if (error)
1109                         goto failed;
1110
1111                 spin_lock(&info->lock);
1112                 info->swapped--;
1113                 shmem_recalc_inode(inode);
1114                 spin_unlock(&info->lock);
1115
1116                 if (sgp == SGP_WRITE)
1117                         mark_page_accessed(page);
1118
1119                 delete_from_swap_cache(page);
1120                 set_page_dirty(page);
1121                 swap_free(swap);
1122
1123         } else {
1124                 if (shmem_acct_block(info->flags)) {
1125                         error = -ENOSPC;
1126                         goto failed;
1127                 }
1128                 if (sbinfo->max_blocks) {
1129                         if (percpu_counter_compare(&sbinfo->used_blocks,
1130                                                 sbinfo->max_blocks) >= 0) {
1131                                 error = -ENOSPC;
1132                                 goto unacct;
1133                         }
1134                         percpu_counter_inc(&sbinfo->used_blocks);
1135                 }
1136
1137                 page = shmem_alloc_page(gfp, info, index);
1138                 if (!page) {
1139                         error = -ENOMEM;
1140                         goto decused;
1141                 }
1142
1143                 __SetPageSwapBacked(page);
1144                 __set_page_locked(page);
1145                 if (sgp == SGP_WRITE)
1146                         init_page_accessed(page);
1147
1148                 error = mem_cgroup_charge_file(page, current->mm,
1149                                                 gfp & GFP_RECLAIM_MASK);
1150                 if (error)
1151                         goto decused;
1152                 error = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK);
1153                 if (!error) {
1154                         error = shmem_add_to_page_cache(page, mapping, index,
1155                                                         gfp, NULL);
1156                         radix_tree_preload_end();
1157                 }
1158                 if (error) {
1159                         mem_cgroup_uncharge_cache_page(page);
1160                         goto decused;
1161                 }
1162                 lru_cache_add_anon(page);
1163
1164                 spin_lock(&info->lock);
1165                 info->alloced++;
1166                 inode->i_blocks += BLOCKS_PER_PAGE;
1167                 shmem_recalc_inode(inode);
1168                 spin_unlock(&info->lock);
1169                 alloced = true;
1170
1171                 /*
1172                  * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1173                  */
1174                 if (sgp == SGP_FALLOC)
1175                         sgp = SGP_WRITE;
1176 clear:
1177                 /*
1178                  * Let SGP_WRITE caller clear ends if write does not fill page;
1179                  * but SGP_FALLOC on a page fallocated earlier must initialize
1180                  * it now, lest undo on failure cancel our earlier guarantee.
1181                  */
1182                 if (sgp != SGP_WRITE) {
1183                         clear_highpage(page);
1184                         flush_dcache_page(page);
1185                         SetPageUptodate(page);
1186                 }
1187                 if (sgp == SGP_DIRTY)
1188                         set_page_dirty(page);
1189         }
1190
1191         /* Perhaps the file has been truncated since we checked */
1192         if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1193             ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1194                 error = -EINVAL;
1195                 if (alloced)
1196                         goto trunc;
1197                 else
1198                         goto failed;
1199         }
1200         *pagep = page;
1201         return 0;
1202
1203         /*
1204          * Error recovery.
1205          */
1206 trunc:
1207         info = SHMEM_I(inode);
1208         ClearPageDirty(page);
1209         delete_from_page_cache(page);
1210         spin_lock(&info->lock);
1211         info->alloced--;
1212         inode->i_blocks -= BLOCKS_PER_PAGE;
1213         spin_unlock(&info->lock);
1214 decused:
1215         sbinfo = SHMEM_SB(inode->i_sb);
1216         if (sbinfo->max_blocks)
1217                 percpu_counter_add(&sbinfo->used_blocks, -1);
1218 unacct:
1219         shmem_unacct_blocks(info->flags, 1);
1220 failed:
1221         if (swap.val && error != -EINVAL &&
1222             !shmem_confirm_swap(mapping, index, swap))
1223                 error = -EEXIST;
1224 unlock:
1225         if (page) {
1226                 unlock_page(page);
1227                 page_cache_release(page);
1228         }
1229         if (error == -ENOSPC && !once++) {
1230                 info = SHMEM_I(inode);
1231                 spin_lock(&info->lock);
1232                 shmem_recalc_inode(inode);
1233                 spin_unlock(&info->lock);
1234                 goto repeat;
1235         }
1236         if (error == -EEXIST)   /* from above or from radix_tree_insert */
1237                 goto repeat;
1238         return error;
1239 }
1240
1241 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1242 {
1243         struct inode *inode = file_inode(vma->vm_file);
1244         int error;
1245         int ret = VM_FAULT_LOCKED;
1246
1247         /*
1248          * Trinity finds that probing a hole which tmpfs is punching can
1249          * prevent the hole-punch from ever completing: which in turn
1250          * locks writers out with its hold on i_mutex.  So refrain from
1251          * faulting pages into the hole while it's being punched, and
1252          * wait on i_mutex to be released if vmf->flags permits.
1253          */
1254         if (unlikely(inode->i_private)) {
1255                 struct shmem_falloc *shmem_falloc;
1256
1257                 spin_lock(&inode->i_lock);
1258                 shmem_falloc = inode->i_private;
1259                 if (!shmem_falloc ||
1260                     shmem_falloc->mode != FALLOC_FL_PUNCH_HOLE ||
1261                     vmf->pgoff < shmem_falloc->start ||
1262                     vmf->pgoff >= shmem_falloc->next)
1263                         shmem_falloc = NULL;
1264                 spin_unlock(&inode->i_lock);
1265                 /*
1266                  * i_lock has protected us from taking shmem_falloc seriously
1267                  * once return from shmem_fallocate() went back up that stack.
1268                  * i_lock does not serialize with i_mutex at all, but it does
1269                  * not matter if sometimes we wait unnecessarily, or sometimes
1270                  * miss out on waiting: we just need to make those cases rare.
1271                  */
1272                 if (shmem_falloc) {
1273                         if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1274                            !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
1275                                 up_read(&vma->vm_mm->mmap_sem);
1276                                 mutex_lock(&inode->i_mutex);
1277                                 mutex_unlock(&inode->i_mutex);
1278                                 return VM_FAULT_RETRY;
1279                         }
1280                         /* cond_resched? Leave that to GUP or return to user */
1281                         return VM_FAULT_NOPAGE;
1282                 }
1283         }
1284
1285         error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1286         if (error)
1287                 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1288
1289         if (ret & VM_FAULT_MAJOR) {
1290                 count_vm_event(PGMAJFAULT);
1291                 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1292         }
1293         return ret;
1294 }
1295
1296 #ifdef CONFIG_NUMA
1297 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
1298 {
1299         struct inode *inode = file_inode(vma->vm_file);
1300         return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
1301 }
1302
1303 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1304                                           unsigned long addr)
1305 {
1306         struct inode *inode = file_inode(vma->vm_file);
1307         pgoff_t index;
1308
1309         index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1310         return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
1311 }
1312 #endif
1313
1314 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1315 {
1316         struct inode *inode = file_inode(file);
1317         struct shmem_inode_info *info = SHMEM_I(inode);
1318         int retval = -ENOMEM;
1319
1320         spin_lock(&info->lock);
1321         if (lock && !(info->flags & VM_LOCKED)) {
1322                 if (!user_shm_lock(inode->i_size, user))
1323                         goto out_nomem;
1324                 info->flags |= VM_LOCKED;
1325                 mapping_set_unevictable(file->f_mapping);
1326         }
1327         if (!lock && (info->flags & VM_LOCKED) && user) {
1328                 user_shm_unlock(inode->i_size, user);
1329                 info->flags &= ~VM_LOCKED;
1330                 mapping_clear_unevictable(file->f_mapping);
1331         }
1332         retval = 0;
1333
1334 out_nomem:
1335         spin_unlock(&info->lock);
1336         return retval;
1337 }
1338
1339 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1340 {
1341         file_accessed(file);
1342         vma->vm_ops = &shmem_vm_ops;
1343         return 0;
1344 }
1345
1346 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1347                                      umode_t mode, dev_t dev, unsigned long flags)
1348 {
1349         struct inode *inode;
1350         struct shmem_inode_info *info;
1351         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1352
1353         if (shmem_reserve_inode(sb))
1354                 return NULL;
1355
1356         inode = new_inode(sb);
1357         if (inode) {
1358                 inode->i_ino = get_next_ino();
1359                 inode_init_owner(inode, dir, mode);
1360                 inode->i_blocks = 0;
1361                 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1362                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1363                 inode->i_generation = get_seconds();
1364                 info = SHMEM_I(inode);
1365                 memset(info, 0, (char *)inode - (char *)info);
1366                 spin_lock_init(&info->lock);
1367                 info->flags = flags & VM_NORESERVE;
1368                 INIT_LIST_HEAD(&info->swaplist);
1369                 simple_xattrs_init(&info->xattrs);
1370                 cache_no_acl(inode);
1371
1372                 switch (mode & S_IFMT) {
1373                 default:
1374                         inode->i_op = &shmem_special_inode_operations;
1375                         init_special_inode(inode, mode, dev);
1376                         break;
1377                 case S_IFREG:
1378                         inode->i_mapping->a_ops = &shmem_aops;
1379                         inode->i_op = &shmem_inode_operations;
1380                         inode->i_fop = &shmem_file_operations;
1381                         mpol_shared_policy_init(&info->policy,
1382                                                  shmem_get_sbmpol(sbinfo));
1383                         break;
1384                 case S_IFDIR:
1385                         inc_nlink(inode);
1386                         /* Some things misbehave if size == 0 on a directory */
1387                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
1388                         inode->i_op = &shmem_dir_inode_operations;
1389                         inode->i_fop = &simple_dir_operations;
1390                         break;
1391                 case S_IFLNK:
1392                         /*
1393                          * Must not load anything in the rbtree,
1394                          * mpol_free_shared_policy will not be called.
1395                          */
1396                         mpol_shared_policy_init(&info->policy, NULL);
1397                         break;
1398                 }
1399         } else
1400                 shmem_free_inode(sb);
1401         return inode;
1402 }
1403
1404 bool shmem_mapping(struct address_space *mapping)
1405 {
1406         return mapping->backing_dev_info == &shmem_backing_dev_info;
1407 }
1408
1409 #ifdef CONFIG_TMPFS
1410 static const struct inode_operations shmem_symlink_inode_operations;
1411 static const struct inode_operations shmem_short_symlink_operations;
1412
1413 #ifdef CONFIG_TMPFS_XATTR
1414 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1415 #else
1416 #define shmem_initxattrs NULL
1417 #endif
1418
1419 static int
1420 shmem_write_begin(struct file *file, struct address_space *mapping,
1421                         loff_t pos, unsigned len, unsigned flags,
1422                         struct page **pagep, void **fsdata)
1423 {
1424         struct inode *inode = mapping->host;
1425         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1426         return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1427 }
1428
1429 static int
1430 shmem_write_end(struct file *file, struct address_space *mapping,
1431                         loff_t pos, unsigned len, unsigned copied,
1432                         struct page *page, void *fsdata)
1433 {
1434         struct inode *inode = mapping->host;
1435
1436         if (pos + copied > inode->i_size)
1437                 i_size_write(inode, pos + copied);
1438
1439         if (!PageUptodate(page)) {
1440                 if (copied < PAGE_CACHE_SIZE) {
1441                         unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1442                         zero_user_segments(page, 0, from,
1443                                         from + copied, PAGE_CACHE_SIZE);
1444                 }
1445                 SetPageUptodate(page);
1446         }
1447         set_page_dirty(page);
1448         unlock_page(page);
1449         page_cache_release(page);
1450
1451         return copied;
1452 }
1453
1454 static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
1455 {
1456         struct file *file = iocb->ki_filp;
1457         struct inode *inode = file_inode(file);
1458         struct address_space *mapping = inode->i_mapping;
1459         pgoff_t index;
1460         unsigned long offset;
1461         enum sgp_type sgp = SGP_READ;
1462         int error = 0;
1463         ssize_t retval = 0;
1464         loff_t *ppos = &iocb->ki_pos;
1465
1466         /*
1467          * Might this read be for a stacking filesystem?  Then when reading
1468          * holes of a sparse file, we actually need to allocate those pages,
1469          * and even mark them dirty, so it cannot exceed the max_blocks limit.
1470          */
1471         if (segment_eq(get_fs(), KERNEL_DS))
1472                 sgp = SGP_DIRTY;
1473
1474         index = *ppos >> PAGE_CACHE_SHIFT;
1475         offset = *ppos & ~PAGE_CACHE_MASK;
1476
1477         for (;;) {
1478                 struct page *page = NULL;
1479                 pgoff_t end_index;
1480                 unsigned long nr, ret;
1481                 loff_t i_size = i_size_read(inode);
1482
1483                 end_index = i_size >> PAGE_CACHE_SHIFT;
1484                 if (index > end_index)
1485                         break;
1486                 if (index == end_index) {
1487                         nr = i_size & ~PAGE_CACHE_MASK;
1488                         if (nr <= offset)
1489                                 break;
1490                 }
1491
1492                 error = shmem_getpage(inode, index, &page, sgp, NULL);
1493                 if (error) {
1494                         if (error == -EINVAL)
1495                                 error = 0;
1496                         break;
1497                 }
1498                 if (page)
1499                         unlock_page(page);
1500
1501                 /*
1502                  * We must evaluate after, since reads (unlike writes)
1503                  * are called without i_mutex protection against truncate
1504                  */
1505                 nr = PAGE_CACHE_SIZE;
1506                 i_size = i_size_read(inode);
1507                 end_index = i_size >> PAGE_CACHE_SHIFT;
1508                 if (index == end_index) {
1509                         nr = i_size & ~PAGE_CACHE_MASK;
1510                         if (nr <= offset) {
1511                                 if (page)
1512                                         page_cache_release(page);
1513                                 break;
1514                         }
1515                 }
1516                 nr -= offset;
1517
1518                 if (page) {
1519                         /*
1520                          * If users can be writing to this page using arbitrary
1521                          * virtual addresses, take care about potential aliasing
1522                          * before reading the page on the kernel side.
1523                          */
1524                         if (mapping_writably_mapped(mapping))
1525                                 flush_dcache_page(page);
1526                         /*
1527                          * Mark the page accessed if we read the beginning.
1528                          */
1529                         if (!offset)
1530                                 mark_page_accessed(page);
1531                 } else {
1532                         page = ZERO_PAGE(0);
1533                         page_cache_get(page);
1534                 }
1535
1536                 /*
1537                  * Ok, we have the page, and it's up-to-date, so
1538                  * now we can copy it to user space...
1539                  */
1540                 ret = copy_page_to_iter(page, offset, nr, to);
1541                 retval += ret;
1542                 offset += ret;
1543                 index += offset >> PAGE_CACHE_SHIFT;
1544                 offset &= ~PAGE_CACHE_MASK;
1545
1546                 page_cache_release(page);
1547                 if (!iov_iter_count(to))
1548                         break;
1549                 if (ret < nr) {
1550                         error = -EFAULT;
1551                         break;
1552                 }
1553                 cond_resched();
1554         }
1555
1556         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1557         file_accessed(file);
1558         return retval ? retval : error;
1559 }
1560
1561 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1562                                 struct pipe_inode_info *pipe, size_t len,
1563                                 unsigned int flags)
1564 {
1565         struct address_space *mapping = in->f_mapping;
1566         struct inode *inode = mapping->host;
1567         unsigned int loff, nr_pages, req_pages;
1568         struct page *pages[PIPE_DEF_BUFFERS];
1569         struct partial_page partial[PIPE_DEF_BUFFERS];
1570         struct page *page;
1571         pgoff_t index, end_index;
1572         loff_t isize, left;
1573         int error, page_nr;
1574         struct splice_pipe_desc spd = {
1575                 .pages = pages,
1576                 .partial = partial,
1577                 .nr_pages_max = PIPE_DEF_BUFFERS,
1578                 .flags = flags,
1579                 .ops = &page_cache_pipe_buf_ops,
1580                 .spd_release = spd_release_page,
1581         };
1582
1583         isize = i_size_read(inode);
1584         if (unlikely(*ppos >= isize))
1585                 return 0;
1586
1587         left = isize - *ppos;
1588         if (unlikely(left < len))
1589                 len = left;
1590
1591         if (splice_grow_spd(pipe, &spd))
1592                 return -ENOMEM;
1593
1594         index = *ppos >> PAGE_CACHE_SHIFT;
1595         loff = *ppos & ~PAGE_CACHE_MASK;
1596         req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1597         nr_pages = min(req_pages, spd.nr_pages_max);
1598
1599         spd.nr_pages = find_get_pages_contig(mapping, index,
1600                                                 nr_pages, spd.pages);
1601         index += spd.nr_pages;
1602         error = 0;
1603
1604         while (spd.nr_pages < nr_pages) {
1605                 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1606                 if (error)
1607                         break;
1608                 unlock_page(page);
1609                 spd.pages[spd.nr_pages++] = page;
1610                 index++;
1611         }
1612
1613         index = *ppos >> PAGE_CACHE_SHIFT;
1614         nr_pages = spd.nr_pages;
1615         spd.nr_pages = 0;
1616
1617         for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1618                 unsigned int this_len;
1619
1620                 if (!len)
1621                         break;
1622
1623                 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1624                 page = spd.pages[page_nr];
1625
1626                 if (!PageUptodate(page) || page->mapping != mapping) {
1627                         error = shmem_getpage(inode, index, &page,
1628                                                         SGP_CACHE, NULL);
1629                         if (error)
1630                                 break;
1631                         unlock_page(page);
1632                         page_cache_release(spd.pages[page_nr]);
1633                         spd.pages[page_nr] = page;
1634                 }
1635
1636                 isize = i_size_read(inode);
1637                 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1638                 if (unlikely(!isize || index > end_index))
1639                         break;
1640
1641                 if (end_index == index) {
1642                         unsigned int plen;
1643
1644                         plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1645                         if (plen <= loff)
1646                                 break;
1647
1648                         this_len = min(this_len, plen - loff);
1649                         len = this_len;
1650                 }
1651
1652                 spd.partial[page_nr].offset = loff;
1653                 spd.partial[page_nr].len = this_len;
1654                 len -= this_len;
1655                 loff = 0;
1656                 spd.nr_pages++;
1657                 index++;
1658         }
1659
1660         while (page_nr < nr_pages)
1661                 page_cache_release(spd.pages[page_nr++]);
1662
1663         if (spd.nr_pages)
1664                 error = splice_to_pipe(pipe, &spd);
1665
1666         splice_shrink_spd(&spd);
1667
1668         if (error > 0) {
1669                 *ppos += error;
1670                 file_accessed(in);
1671         }
1672         return error;
1673 }
1674
1675 /*
1676  * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1677  */
1678 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
1679                                     pgoff_t index, pgoff_t end, int whence)
1680 {
1681         struct page *page;
1682         struct pagevec pvec;
1683         pgoff_t indices[PAGEVEC_SIZE];
1684         bool done = false;
1685         int i;
1686
1687         pagevec_init(&pvec, 0);
1688         pvec.nr = 1;            /* start small: we may be there already */
1689         while (!done) {
1690                 pvec.nr = find_get_entries(mapping, index,
1691                                         pvec.nr, pvec.pages, indices);
1692                 if (!pvec.nr) {
1693                         if (whence == SEEK_DATA)
1694                                 index = end;
1695                         break;
1696                 }
1697                 for (i = 0; i < pvec.nr; i++, index++) {
1698                         if (index < indices[i]) {
1699                                 if (whence == SEEK_HOLE) {
1700                                         done = true;
1701                                         break;
1702                                 }
1703                                 index = indices[i];
1704                         }
1705                         page = pvec.pages[i];
1706                         if (page && !radix_tree_exceptional_entry(page)) {
1707                                 if (!PageUptodate(page))
1708                                         page = NULL;
1709                         }
1710                         if (index >= end ||
1711                             (page && whence == SEEK_DATA) ||
1712                             (!page && whence == SEEK_HOLE)) {
1713                                 done = true;
1714                                 break;
1715                         }
1716                 }
1717                 pagevec_remove_exceptionals(&pvec);
1718                 pagevec_release(&pvec);
1719                 pvec.nr = PAGEVEC_SIZE;
1720                 cond_resched();
1721         }
1722         return index;
1723 }
1724
1725 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
1726 {
1727         struct address_space *mapping = file->f_mapping;
1728         struct inode *inode = mapping->host;
1729         pgoff_t start, end;
1730         loff_t new_offset;
1731
1732         if (whence != SEEK_DATA && whence != SEEK_HOLE)
1733                 return generic_file_llseek_size(file, offset, whence,
1734                                         MAX_LFS_FILESIZE, i_size_read(inode));
1735         mutex_lock(&inode->i_mutex);
1736         /* We're holding i_mutex so we can access i_size directly */
1737
1738         if (offset < 0)
1739                 offset = -EINVAL;
1740         else if (offset >= inode->i_size)
1741                 offset = -ENXIO;
1742         else {
1743                 start = offset >> PAGE_CACHE_SHIFT;
1744                 end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1745                 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
1746                 new_offset <<= PAGE_CACHE_SHIFT;
1747                 if (new_offset > offset) {
1748                         if (new_offset < inode->i_size)
1749                                 offset = new_offset;
1750                         else if (whence == SEEK_DATA)
1751                                 offset = -ENXIO;
1752                         else
1753                                 offset = inode->i_size;
1754                 }
1755         }
1756
1757         if (offset >= 0)
1758                 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
1759         mutex_unlock(&inode->i_mutex);
1760         return offset;
1761 }
1762
1763 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
1764                                                          loff_t len)
1765 {
1766         struct inode *inode = file_inode(file);
1767         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1768         struct shmem_falloc shmem_falloc;
1769         pgoff_t start, index, end;
1770         int error;
1771
1772         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
1773                 return -EOPNOTSUPP;
1774
1775         mutex_lock(&inode->i_mutex);
1776
1777         shmem_falloc.mode = mode & ~FALLOC_FL_KEEP_SIZE;
1778
1779         if (mode & FALLOC_FL_PUNCH_HOLE) {
1780                 struct address_space *mapping = file->f_mapping;
1781                 loff_t unmap_start = round_up(offset, PAGE_SIZE);
1782                 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
1783
1784                 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
1785                 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
1786                 spin_lock(&inode->i_lock);
1787                 inode->i_private = &shmem_falloc;
1788                 spin_unlock(&inode->i_lock);
1789
1790                 if ((u64)unmap_end > (u64)unmap_start)
1791                         unmap_mapping_range(mapping, unmap_start,
1792                                             1 + unmap_end - unmap_start, 0);
1793                 shmem_truncate_range(inode, offset, offset + len - 1);
1794                 /* No need to unmap again: hole-punching leaves COWed pages */
1795                 error = 0;
1796                 goto undone;
1797         }
1798
1799         /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1800         error = inode_newsize_ok(inode, offset + len);
1801         if (error)
1802                 goto out;
1803
1804         start = offset >> PAGE_CACHE_SHIFT;
1805         end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1806         /* Try to avoid a swapstorm if len is impossible to satisfy */
1807         if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
1808                 error = -ENOSPC;
1809                 goto out;
1810         }
1811
1812         shmem_falloc.start = start;
1813         shmem_falloc.next  = start;
1814         shmem_falloc.nr_falloced = 0;
1815         shmem_falloc.nr_unswapped = 0;
1816         spin_lock(&inode->i_lock);
1817         inode->i_private = &shmem_falloc;
1818         spin_unlock(&inode->i_lock);
1819
1820         for (index = start; index < end; index++) {
1821                 struct page *page;
1822
1823                 /*
1824                  * Good, the fallocate(2) manpage permits EINTR: we may have
1825                  * been interrupted because we are using up too much memory.
1826                  */
1827                 if (signal_pending(current))
1828                         error = -EINTR;
1829                 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
1830                         error = -ENOMEM;
1831                 else
1832                         error = shmem_getpage(inode, index, &page, SGP_FALLOC,
1833                                                                         NULL);
1834                 if (error) {
1835                         /* Remove the !PageUptodate pages we added */
1836                         shmem_undo_range(inode,
1837                                 (loff_t)start << PAGE_CACHE_SHIFT,
1838                                 (loff_t)index << PAGE_CACHE_SHIFT, true);
1839                         goto undone;
1840                 }
1841
1842                 /*
1843                  * Inform shmem_writepage() how far we have reached.
1844                  * No need for lock or barrier: we have the page lock.
1845                  */
1846                 shmem_falloc.next++;
1847                 if (!PageUptodate(page))
1848                         shmem_falloc.nr_falloced++;
1849
1850                 /*
1851                  * If !PageUptodate, leave it that way so that freeable pages
1852                  * can be recognized if we need to rollback on error later.
1853                  * But set_page_dirty so that memory pressure will swap rather
1854                  * than free the pages we are allocating (and SGP_CACHE pages
1855                  * might still be clean: we now need to mark those dirty too).
1856                  */
1857                 set_page_dirty(page);
1858                 unlock_page(page);
1859                 page_cache_release(page);
1860                 cond_resched();
1861         }
1862
1863         if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
1864                 i_size_write(inode, offset + len);
1865         inode->i_ctime = CURRENT_TIME;
1866 undone:
1867         spin_lock(&inode->i_lock);
1868         inode->i_private = NULL;
1869         spin_unlock(&inode->i_lock);
1870 out:
1871         mutex_unlock(&inode->i_mutex);
1872         return error;
1873 }
1874
1875 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1876 {
1877         struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1878
1879         buf->f_type = TMPFS_MAGIC;
1880         buf->f_bsize = PAGE_CACHE_SIZE;
1881         buf->f_namelen = NAME_MAX;
1882         if (sbinfo->max_blocks) {
1883                 buf->f_blocks = sbinfo->max_blocks;
1884                 buf->f_bavail =
1885                 buf->f_bfree  = sbinfo->max_blocks -
1886                                 percpu_counter_sum(&sbinfo->used_blocks);
1887         }
1888         if (sbinfo->max_inodes) {
1889                 buf->f_files = sbinfo->max_inodes;
1890                 buf->f_ffree = sbinfo->free_inodes;
1891         }
1892         /* else leave those fields 0 like simple_statfs */
1893         return 0;
1894 }
1895
1896 /*
1897  * File creation. Allocate an inode, and we're done..
1898  */
1899 static int
1900 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
1901 {
1902         struct inode *inode;
1903         int error = -ENOSPC;
1904
1905         inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1906         if (inode) {
1907                 error = simple_acl_create(dir, inode);
1908                 if (error)
1909                         goto out_iput;
1910                 error = security_inode_init_security(inode, dir,
1911                                                      &dentry->d_name,
1912                                                      shmem_initxattrs, NULL);
1913                 if (error && error != -EOPNOTSUPP)
1914                         goto out_iput;
1915
1916                 error = 0;
1917                 dir->i_size += BOGO_DIRENT_SIZE;
1918                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1919                 d_instantiate(dentry, inode);
1920                 dget(dentry); /* Extra count - pin the dentry in core */
1921         }
1922         return error;
1923 out_iput:
1924         iput(inode);
1925         return error;
1926 }
1927
1928 static int
1929 shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
1930 {
1931         struct inode *inode;
1932         int error = -ENOSPC;
1933
1934         inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
1935         if (inode) {
1936                 error = security_inode_init_security(inode, dir,
1937                                                      NULL,
1938                                                      shmem_initxattrs, NULL);
1939                 if (error && error != -EOPNOTSUPP)
1940                         goto out_iput;
1941                 error = simple_acl_create(dir, inode);
1942                 if (error)
1943                         goto out_iput;
1944                 d_tmpfile(dentry, inode);
1945         }
1946         return error;
1947 out_iput:
1948         iput(inode);
1949         return error;
1950 }
1951
1952 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1953 {
1954         int error;
1955
1956         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1957                 return error;
1958         inc_nlink(dir);
1959         return 0;
1960 }
1961
1962 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1963                 bool excl)
1964 {
1965         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1966 }
1967
1968 /*
1969  * Link a file..
1970  */
1971 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1972 {
1973         struct inode *inode = old_dentry->d_inode;
1974         int ret;
1975
1976         /*
1977          * No ordinary (disk based) filesystem counts links as inodes;
1978          * but each new link needs a new dentry, pinning lowmem, and
1979          * tmpfs dentries cannot be pruned until they are unlinked.
1980          */
1981         ret = shmem_reserve_inode(inode->i_sb);
1982         if (ret)
1983                 goto out;
1984
1985         dir->i_size += BOGO_DIRENT_SIZE;
1986         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1987         inc_nlink(inode);
1988         ihold(inode);   /* New dentry reference */
1989         dget(dentry);           /* Extra pinning count for the created dentry */
1990         d_instantiate(dentry, inode);
1991 out:
1992         return ret;
1993 }
1994
1995 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1996 {
1997         struct inode *inode = dentry->d_inode;
1998
1999         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2000                 shmem_free_inode(inode->i_sb);
2001
2002         dir->i_size -= BOGO_DIRENT_SIZE;
2003         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2004         drop_nlink(inode);
2005         dput(dentry);   /* Undo the count from "create" - this does all the work */
2006         return 0;
2007 }
2008
2009 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2010 {
2011         if (!simple_empty(dentry))
2012                 return -ENOTEMPTY;
2013
2014         drop_nlink(dentry->d_inode);
2015         drop_nlink(dir);
2016         return shmem_unlink(dir, dentry);
2017 }
2018
2019 /*
2020  * The VFS layer already does all the dentry stuff for rename,
2021  * we just have to decrement the usage count for the target if
2022  * it exists so that the VFS layer correctly free's it when it
2023  * gets overwritten.
2024  */
2025 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2026 {
2027         struct inode *inode = old_dentry->d_inode;
2028         int they_are_dirs = S_ISDIR(inode->i_mode);
2029
2030         if (!simple_empty(new_dentry))
2031                 return -ENOTEMPTY;
2032
2033         if (new_dentry->d_inode) {
2034                 (void) shmem_unlink(new_dir, new_dentry);
2035                 if (they_are_dirs)
2036                         drop_nlink(old_dir);
2037         } else if (they_are_dirs) {
2038                 drop_nlink(old_dir);
2039                 inc_nlink(new_dir);
2040         }
2041
2042         old_dir->i_size -= BOGO_DIRENT_SIZE;
2043         new_dir->i_size += BOGO_DIRENT_SIZE;
2044         old_dir->i_ctime = old_dir->i_mtime =
2045         new_dir->i_ctime = new_dir->i_mtime =
2046         inode->i_ctime = CURRENT_TIME;
2047         return 0;
2048 }
2049
2050 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2051 {
2052         int error;
2053         int len;
2054         struct inode *inode;
2055         struct page *page;
2056         char *kaddr;
2057         struct shmem_inode_info *info;
2058
2059         len = strlen(symname) + 1;
2060         if (len > PAGE_CACHE_SIZE)
2061                 return -ENAMETOOLONG;
2062
2063         inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
2064         if (!inode)
2065                 return -ENOSPC;
2066
2067         error = security_inode_init_security(inode, dir, &dentry->d_name,
2068                                              shmem_initxattrs, NULL);
2069         if (error) {
2070                 if (error != -EOPNOTSUPP) {
2071                         iput(inode);
2072                         return error;
2073                 }
2074                 error = 0;
2075         }
2076
2077         info = SHMEM_I(inode);
2078         inode->i_size = len-1;
2079         if (len <= SHORT_SYMLINK_LEN) {
2080                 info->symlink = kmemdup(symname, len, GFP_KERNEL);
2081                 if (!info->symlink) {
2082                         iput(inode);
2083                         return -ENOMEM;
2084                 }
2085                 inode->i_op = &shmem_short_symlink_operations;
2086         } else {
2087                 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2088                 if (error) {
2089                         iput(inode);
2090                         return error;
2091                 }
2092                 inode->i_mapping->a_ops = &shmem_aops;
2093                 inode->i_op = &shmem_symlink_inode_operations;
2094                 kaddr = kmap_atomic(page);
2095                 memcpy(kaddr, symname, len);
2096                 kunmap_atomic(kaddr);
2097                 SetPageUptodate(page);
2098                 set_page_dirty(page);
2099                 unlock_page(page);
2100                 page_cache_release(page);
2101         }
2102         dir->i_size += BOGO_DIRENT_SIZE;
2103         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2104         d_instantiate(dentry, inode);
2105         dget(dentry);
2106         return 0;
2107 }
2108
2109 static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
2110 {
2111         nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
2112         return NULL;
2113 }
2114
2115 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
2116 {
2117         struct page *page = NULL;
2118         int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2119         nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
2120         if (page)
2121                 unlock_page(page);
2122         return page;
2123 }
2124
2125 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2126 {
2127         if (!IS_ERR(nd_get_link(nd))) {
2128                 struct page *page = cookie;
2129                 kunmap(page);
2130                 mark_page_accessed(page);
2131                 page_cache_release(page);
2132         }
2133 }
2134
2135 #ifdef CONFIG_TMPFS_XATTR
2136 /*
2137  * Superblocks without xattr inode operations may get some security.* xattr
2138  * support from the LSM "for free". As soon as we have any other xattrs
2139  * like ACLs, we also need to implement the security.* handlers at
2140  * filesystem level, though.
2141  */
2142
2143 /*
2144  * Callback for security_inode_init_security() for acquiring xattrs.
2145  */
2146 static int shmem_initxattrs(struct inode *inode,
2147                             const struct xattr *xattr_array,
2148                             void *fs_info)
2149 {
2150         struct shmem_inode_info *info = SHMEM_I(inode);
2151         const struct xattr *xattr;
2152         struct simple_xattr *new_xattr;
2153         size_t len;
2154
2155         for (xattr = xattr_array; xattr->name != NULL; xattr++) {
2156                 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
2157                 if (!new_xattr)
2158                         return -ENOMEM;
2159
2160                 len = strlen(xattr->name) + 1;
2161                 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2162                                           GFP_KERNEL);
2163                 if (!new_xattr->name) {
2164                         kfree(new_xattr);
2165                         return -ENOMEM;
2166                 }
2167
2168                 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2169                        XATTR_SECURITY_PREFIX_LEN);
2170                 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2171                        xattr->name, len);
2172
2173                 simple_xattr_list_add(&info->xattrs, new_xattr);
2174         }
2175
2176         return 0;
2177 }
2178
2179 static const struct xattr_handler *shmem_xattr_handlers[] = {
2180 #ifdef CONFIG_TMPFS_POSIX_ACL
2181         &posix_acl_access_xattr_handler,
2182         &posix_acl_default_xattr_handler,
2183 #endif
2184         NULL
2185 };
2186
2187 static int shmem_xattr_validate(const char *name)
2188 {
2189         struct { const char *prefix; size_t len; } arr[] = {
2190                 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2191                 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2192         };
2193         int i;
2194
2195         for (i = 0; i < ARRAY_SIZE(arr); i++) {
2196                 size_t preflen = arr[i].len;
2197                 if (strncmp(name, arr[i].prefix, preflen) == 0) {
2198                         if (!name[preflen])
2199                                 return -EINVAL;
2200                         return 0;
2201                 }
2202         }
2203         return -EOPNOTSUPP;
2204 }
2205
2206 static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2207                               void *buffer, size_t size)
2208 {
2209         struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2210         int err;
2211
2212         /*
2213          * If this is a request for a synthetic attribute in the system.*
2214          * namespace use the generic infrastructure to resolve a handler
2215          * for it via sb->s_xattr.
2216          */
2217         if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2218                 return generic_getxattr(dentry, name, buffer, size);
2219
2220         err = shmem_xattr_validate(name);
2221         if (err)
2222                 return err;
2223
2224         return simple_xattr_get(&info->xattrs, name, buffer, size);
2225 }
2226
2227 static int shmem_setxattr(struct dentry *dentry, const char *name,
2228                           const void *value, size_t size, int flags)
2229 {
2230         struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2231         int err;
2232
2233         /*
2234          * If this is a request for a synthetic attribute in the system.*
2235          * namespace use the generic infrastructure to resolve a handler
2236          * for it via sb->s_xattr.
2237          */
2238         if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2239                 return generic_setxattr(dentry, name, value, size, flags);
2240
2241         err = shmem_xattr_validate(name);
2242         if (err)
2243                 return err;
2244
2245         return simple_xattr_set(&info->xattrs, name, value, size, flags);
2246 }
2247
2248 static int shmem_removexattr(struct dentry *dentry, const char *name)
2249 {
2250         struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2251         int err;
2252
2253         /*
2254          * If this is a request for a synthetic attribute in the system.*
2255          * namespace use the generic infrastructure to resolve a handler
2256          * for it via sb->s_xattr.
2257          */
2258         if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2259                 return generic_removexattr(dentry, name);
2260
2261         err = shmem_xattr_validate(name);
2262         if (err)
2263                 return err;
2264
2265         return simple_xattr_remove(&info->xattrs, name);
2266 }
2267
2268 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2269 {
2270         struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2271         return simple_xattr_list(&info->xattrs, buffer, size);
2272 }
2273 #endif /* CONFIG_TMPFS_XATTR */
2274
2275 static const struct inode_operations shmem_short_symlink_operations = {
2276         .readlink       = generic_readlink,
2277         .follow_link    = shmem_follow_short_symlink,
2278 #ifdef CONFIG_TMPFS_XATTR
2279         .setxattr       = shmem_setxattr,
2280         .getxattr       = shmem_getxattr,
2281         .listxattr      = shmem_listxattr,
2282         .removexattr    = shmem_removexattr,
2283 #endif
2284 };
2285
2286 static const struct inode_operations shmem_symlink_inode_operations = {
2287         .readlink       = generic_readlink,
2288         .follow_link    = shmem_follow_link,
2289         .put_link       = shmem_put_link,
2290 #ifdef CONFIG_TMPFS_XATTR
2291         .setxattr       = shmem_setxattr,
2292         .getxattr       = shmem_getxattr,
2293         .listxattr      = shmem_listxattr,
2294         .removexattr    = shmem_removexattr,
2295 #endif
2296 };
2297
2298 static struct dentry *shmem_get_parent(struct dentry *child)
2299 {
2300         return ERR_PTR(-ESTALE);
2301 }
2302
2303 static int shmem_match(struct inode *ino, void *vfh)
2304 {
2305         __u32 *fh = vfh;
2306         __u64 inum = fh[2];
2307         inum = (inum << 32) | fh[1];
2308         return ino->i_ino == inum && fh[0] == ino->i_generation;
2309 }
2310
2311 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2312                 struct fid *fid, int fh_len, int fh_type)
2313 {
2314         struct inode *inode;
2315         struct dentry *dentry = NULL;
2316         u64 inum;
2317
2318         if (fh_len < 3)
2319                 return NULL;
2320
2321         inum = fid->raw[2];
2322         inum = (inum << 32) | fid->raw[1];
2323
2324         inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2325                         shmem_match, fid->raw);
2326         if (inode) {
2327                 dentry = d_find_alias(inode);
2328                 iput(inode);
2329         }
2330
2331         return dentry;
2332 }
2333
2334 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2335                                 struct inode *parent)
2336 {
2337         if (*len < 3) {
2338                 *len = 3;
2339                 return FILEID_INVALID;
2340         }
2341
2342         if (inode_unhashed(inode)) {
2343                 /* Unfortunately insert_inode_hash is not idempotent,
2344                  * so as we hash inodes here rather than at creation
2345                  * time, we need a lock to ensure we only try
2346                  * to do it once
2347                  */
2348                 static DEFINE_SPINLOCK(lock);
2349                 spin_lock(&lock);
2350                 if (inode_unhashed(inode))
2351                         __insert_inode_hash(inode,
2352                                             inode->i_ino + inode->i_generation);
2353                 spin_unlock(&lock);
2354         }
2355
2356         fh[0] = inode->i_generation;
2357         fh[1] = inode->i_ino;
2358         fh[2] = ((__u64)inode->i_ino) >> 32;
2359
2360         *len = 3;
2361         return 1;
2362 }
2363
2364 static const struct export_operations shmem_export_ops = {
2365         .get_parent     = shmem_get_parent,
2366         .encode_fh      = shmem_encode_fh,
2367         .fh_to_dentry   = shmem_fh_to_dentry,
2368 };
2369
2370 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2371                                bool remount)
2372 {
2373         char *this_char, *value, *rest;
2374         struct mempolicy *mpol = NULL;
2375         uid_t uid;
2376         gid_t gid;
2377
2378         while (options != NULL) {
2379                 this_char = options;
2380                 for (;;) {
2381                         /*
2382                          * NUL-terminate this option: unfortunately,
2383                          * mount options form a comma-separated list,
2384                          * but mpol's nodelist may also contain commas.
2385                          */
2386                         options = strchr(options, ',');
2387                         if (options == NULL)
2388                                 break;
2389                         options++;
2390                         if (!isdigit(*options)) {
2391                                 options[-1] = '\0';
2392                                 break;
2393                         }
2394                 }
2395                 if (!*this_char)
2396                         continue;
2397                 if ((value = strchr(this_char,'=')) != NULL) {
2398                         *value++ = 0;
2399                 } else {
2400                         printk(KERN_ERR
2401                             "tmpfs: No value for mount option '%s'\n",
2402                             this_char);
2403                         goto error;
2404                 }
2405
2406                 if (!strcmp(this_char,"size")) {
2407                         unsigned long long size;
2408                         size = memparse(value,&rest);
2409                         if (*rest == '%') {
2410                                 size <<= PAGE_SHIFT;
2411                                 size *= totalram_pages;
2412                                 do_div(size, 100);
2413                                 rest++;
2414                         }
2415                         if (*rest)
2416                                 goto bad_val;
2417                         sbinfo->max_blocks =
2418                                 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2419                 } else if (!strcmp(this_char,"nr_blocks")) {
2420                         sbinfo->max_blocks = memparse(value, &rest);
2421                         if (*rest)
2422                                 goto bad_val;
2423                 } else if (!strcmp(this_char,"nr_inodes")) {
2424                         sbinfo->max_inodes = memparse(value, &rest);
2425                         if (*rest)
2426                                 goto bad_val;
2427                 } else if (!strcmp(this_char,"mode")) {
2428                         if (remount)
2429                                 continue;
2430                         sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2431                         if (*rest)
2432                                 goto bad_val;
2433                 } else if (!strcmp(this_char,"uid")) {
2434                         if (remount)
2435                                 continue;
2436                         uid = simple_strtoul(value, &rest, 0);
2437                         if (*rest)
2438                                 goto bad_val;
2439                         sbinfo->uid = make_kuid(current_user_ns(), uid);
2440                         if (!uid_valid(sbinfo->uid))
2441                                 goto bad_val;
2442                 } else if (!strcmp(this_char,"gid")) {
2443                         if (remount)
2444                                 continue;
2445                         gid = simple_strtoul(value, &rest, 0);
2446                         if (*rest)
2447                                 goto bad_val;
2448                         sbinfo->gid = make_kgid(current_user_ns(), gid);
2449                         if (!gid_valid(sbinfo->gid))
2450                                 goto bad_val;
2451                 } else if (!strcmp(this_char,"mpol")) {
2452                         mpol_put(mpol);
2453                         mpol = NULL;
2454                         if (mpol_parse_str(value, &mpol))
2455                                 goto bad_val;
2456                 } else {
2457                         printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2458                                this_char);
2459                         goto error;
2460                 }
2461         }
2462         sbinfo->mpol = mpol;
2463         return 0;
2464
2465 bad_val:
2466         printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2467                value, this_char);
2468 error:
2469         mpol_put(mpol);
2470         return 1;
2471
2472 }
2473
2474 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2475 {
2476         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2477         struct shmem_sb_info config = *sbinfo;
2478         unsigned long inodes;
2479         int error = -EINVAL;
2480
2481         config.mpol = NULL;
2482         if (shmem_parse_options(data, &config, true))
2483                 return error;
2484
2485         spin_lock(&sbinfo->stat_lock);
2486         inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2487         if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2488                 goto out;
2489         if (config.max_inodes < inodes)
2490                 goto out;
2491         /*
2492          * Those tests disallow limited->unlimited while any are in use;
2493          * but we must separately disallow unlimited->limited, because
2494          * in that case we have no record of how much is already in use.
2495          */
2496         if (config.max_blocks && !sbinfo->max_blocks)
2497                 goto out;
2498         if (config.max_inodes && !sbinfo->max_inodes)
2499                 goto out;
2500
2501         error = 0;
2502         sbinfo->max_blocks  = config.max_blocks;
2503         sbinfo->max_inodes  = config.max_inodes;
2504         sbinfo->free_inodes = config.max_inodes - inodes;
2505
2506         /*
2507          * Preserve previous mempolicy unless mpol remount option was specified.
2508          */
2509         if (config.mpol) {
2510                 mpol_put(sbinfo->mpol);
2511                 sbinfo->mpol = config.mpol;     /* transfers initial ref */
2512         }
2513 out:
2514         spin_unlock(&sbinfo->stat_lock);
2515         return error;
2516 }
2517
2518 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
2519 {
2520         struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
2521
2522         if (sbinfo->max_blocks != shmem_default_max_blocks())
2523                 seq_printf(seq, ",size=%luk",
2524                         sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2525         if (sbinfo->max_inodes != shmem_default_max_inodes())
2526                 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2527         if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2528                 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
2529         if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2530                 seq_printf(seq, ",uid=%u",
2531                                 from_kuid_munged(&init_user_ns, sbinfo->uid));
2532         if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2533                 seq_printf(seq, ",gid=%u",
2534                                 from_kgid_munged(&init_user_ns, sbinfo->gid));
2535         shmem_show_mpol(seq, sbinfo->mpol);
2536         return 0;
2537 }
2538 #endif /* CONFIG_TMPFS */
2539
2540 static void shmem_put_super(struct super_block *sb)
2541 {
2542         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2543
2544         percpu_counter_destroy(&sbinfo->used_blocks);
2545         mpol_put(sbinfo->mpol);
2546         kfree(sbinfo);
2547         sb->s_fs_info = NULL;
2548 }
2549
2550 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2551 {
2552         struct inode *inode;
2553         struct shmem_sb_info *sbinfo;
2554         int err = -ENOMEM;
2555
2556         /* Round up to L1_CACHE_BYTES to resist false sharing */
2557         sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2558                                 L1_CACHE_BYTES), GFP_KERNEL);
2559         if (!sbinfo)
2560                 return -ENOMEM;
2561
2562         sbinfo->mode = S_IRWXUGO | S_ISVTX;
2563         sbinfo->uid = current_fsuid();
2564         sbinfo->gid = current_fsgid();
2565         sb->s_fs_info = sbinfo;
2566
2567 #ifdef CONFIG_TMPFS
2568         /*
2569          * Per default we only allow half of the physical ram per
2570          * tmpfs instance, limiting inodes to one per page of lowmem;
2571          * but the internal instance is left unlimited.
2572          */
2573         if (!(sb->s_flags & MS_KERNMOUNT)) {
2574                 sbinfo->max_blocks = shmem_default_max_blocks();
2575                 sbinfo->max_inodes = shmem_default_max_inodes();
2576                 if (shmem_parse_options(data, sbinfo, false)) {
2577                         err = -EINVAL;
2578                         goto failed;
2579                 }
2580         } else {
2581                 sb->s_flags |= MS_NOUSER;
2582         }
2583         sb->s_export_op = &shmem_export_ops;
2584         sb->s_flags |= MS_NOSEC;
2585 #else
2586         sb->s_flags |= MS_NOUSER;
2587 #endif
2588
2589         spin_lock_init(&sbinfo->stat_lock);
2590         if (percpu_counter_init(&sbinfo->used_blocks, 0))
2591                 goto failed;
2592         sbinfo->free_inodes = sbinfo->max_inodes;
2593
2594         sb->s_maxbytes = MAX_LFS_FILESIZE;
2595         sb->s_blocksize = PAGE_CACHE_SIZE;
2596         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2597         sb->s_magic = TMPFS_MAGIC;
2598         sb->s_op = &shmem_ops;
2599         sb->s_time_gran = 1;
2600 #ifdef CONFIG_TMPFS_XATTR
2601         sb->s_xattr = shmem_xattr_handlers;
2602 #endif
2603 #ifdef CONFIG_TMPFS_POSIX_ACL
2604         sb->s_flags |= MS_POSIXACL;
2605 #endif
2606
2607         inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2608         if (!inode)
2609                 goto failed;
2610         inode->i_uid = sbinfo->uid;
2611         inode->i_gid = sbinfo->gid;
2612         sb->s_root = d_make_root(inode);
2613         if (!sb->s_root)
2614                 goto failed;
2615         return 0;
2616
2617 failed:
2618         shmem_put_super(sb);
2619         return err;
2620 }
2621
2622 static struct kmem_cache *shmem_inode_cachep;
2623
2624 static struct inode *shmem_alloc_inode(struct super_block *sb)
2625 {
2626         struct shmem_inode_info *info;
2627         info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2628         if (!info)
2629                 return NULL;
2630         return &info->vfs_inode;
2631 }
2632
2633 static void shmem_destroy_callback(struct rcu_head *head)
2634 {
2635         struct inode *inode = container_of(head, struct inode, i_rcu);
2636         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2637 }
2638
2639 static void shmem_destroy_inode(struct inode *inode)
2640 {
2641         if (S_ISREG(inode->i_mode))
2642                 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2643         call_rcu(&inode->i_rcu, shmem_destroy_callback);
2644 }
2645
2646 static void shmem_init_inode(void *foo)
2647 {
2648         struct shmem_inode_info *info = foo;
2649         inode_init_once(&info->vfs_inode);
2650 }
2651
2652 static int shmem_init_inodecache(void)
2653 {
2654         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2655                                 sizeof(struct shmem_inode_info),
2656                                 0, SLAB_PANIC, shmem_init_inode);
2657         return 0;
2658 }
2659
2660 static void shmem_destroy_inodecache(void)
2661 {
2662         kmem_cache_destroy(shmem_inode_cachep);
2663 }
2664
2665 static const struct address_space_operations shmem_aops = {
2666         .writepage      = shmem_writepage,
2667         .set_page_dirty = __set_page_dirty_no_writeback,
2668 #ifdef CONFIG_TMPFS
2669         .write_begin    = shmem_write_begin,
2670         .write_end      = shmem_write_end,
2671 #endif
2672         .migratepage    = migrate_page,
2673         .error_remove_page = generic_error_remove_page,
2674 };
2675
2676 static const struct file_operations shmem_file_operations = {
2677         .mmap           = shmem_mmap,
2678 #ifdef CONFIG_TMPFS
2679         .llseek         = shmem_file_llseek,
2680         .read           = new_sync_read,
2681         .write          = new_sync_write,
2682         .read_iter      = shmem_file_read_iter,
2683         .write_iter     = generic_file_write_iter,
2684         .fsync          = noop_fsync,
2685         .splice_read    = shmem_file_splice_read,
2686         .splice_write   = iter_file_splice_write,
2687         .fallocate      = shmem_fallocate,
2688 #endif
2689 };
2690
2691 static const struct inode_operations shmem_inode_operations = {
2692         .setattr        = shmem_setattr,
2693 #ifdef CONFIG_TMPFS_XATTR
2694         .setxattr       = shmem_setxattr,
2695         .getxattr       = shmem_getxattr,
2696         .listxattr      = shmem_listxattr,
2697         .removexattr    = shmem_removexattr,
2698         .set_acl        = simple_set_acl,
2699 #endif
2700 };
2701
2702 static const struct inode_operations shmem_dir_inode_operations = {
2703 #ifdef CONFIG_TMPFS
2704         .create         = shmem_create,
2705         .lookup         = simple_lookup,
2706         .link           = shmem_link,
2707         .unlink         = shmem_unlink,
2708         .symlink        = shmem_symlink,
2709         .mkdir          = shmem_mkdir,
2710         .rmdir          = shmem_rmdir,
2711         .mknod          = shmem_mknod,
2712         .rename         = shmem_rename,
2713         .tmpfile        = shmem_tmpfile,
2714 #endif
2715 #ifdef CONFIG_TMPFS_XATTR
2716         .setxattr       = shmem_setxattr,
2717         .getxattr       = shmem_getxattr,
2718         .listxattr      = shmem_listxattr,
2719         .removexattr    = shmem_removexattr,
2720 #endif
2721 #ifdef CONFIG_TMPFS_POSIX_ACL
2722         .setattr        = shmem_setattr,
2723         .set_acl        = simple_set_acl,
2724 #endif
2725 };
2726
2727 static const struct inode_operations shmem_special_inode_operations = {
2728 #ifdef CONFIG_TMPFS_XATTR
2729         .setxattr       = shmem_setxattr,
2730         .getxattr       = shmem_getxattr,
2731         .listxattr      = shmem_listxattr,
2732         .removexattr    = shmem_removexattr,
2733 #endif
2734 #ifdef CONFIG_TMPFS_POSIX_ACL
2735         .setattr        = shmem_setattr,
2736         .set_acl        = simple_set_acl,
2737 #endif
2738 };
2739
2740 static const struct super_operations shmem_ops = {
2741         .alloc_inode    = shmem_alloc_inode,
2742         .destroy_inode  = shmem_destroy_inode,
2743 #ifdef CONFIG_TMPFS
2744         .statfs         = shmem_statfs,
2745         .remount_fs     = shmem_remount_fs,
2746         .show_options   = shmem_show_options,
2747 #endif
2748         .evict_inode    = shmem_evict_inode,
2749         .drop_inode     = generic_delete_inode,
2750         .put_super      = shmem_put_super,
2751 };
2752
2753 static const struct vm_operations_struct shmem_vm_ops = {
2754         .fault          = shmem_fault,
2755         .map_pages      = filemap_map_pages,
2756 #ifdef CONFIG_NUMA
2757         .set_policy     = shmem_set_policy,
2758         .get_policy     = shmem_get_policy,
2759 #endif
2760         .remap_pages    = generic_file_remap_pages,
2761 };
2762
2763 static struct dentry *shmem_mount(struct file_system_type *fs_type,
2764         int flags, const char *dev_name, void *data)
2765 {
2766         return mount_nodev(fs_type, flags, data, shmem_fill_super);
2767 }
2768
2769 static struct file_system_type shmem_fs_type = {
2770         .owner          = THIS_MODULE,
2771         .name           = "tmpfs",
2772         .mount          = shmem_mount,
2773         .kill_sb        = kill_litter_super,
2774         .fs_flags       = FS_USERNS_MOUNT,
2775 };
2776
2777 int __init shmem_init(void)
2778 {
2779         int error;
2780
2781         /* If rootfs called this, don't re-init */
2782         if (shmem_inode_cachep)
2783                 return 0;
2784
2785         error = bdi_init(&shmem_backing_dev_info);
2786         if (error)
2787                 goto out4;
2788
2789         error = shmem_init_inodecache();
2790         if (error)
2791                 goto out3;
2792
2793         error = register_filesystem(&shmem_fs_type);
2794         if (error) {
2795                 printk(KERN_ERR "Could not register tmpfs\n");
2796                 goto out2;
2797         }
2798
2799         shm_mnt = kern_mount(&shmem_fs_type);
2800         if (IS_ERR(shm_mnt)) {
2801                 error = PTR_ERR(shm_mnt);
2802                 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2803                 goto out1;
2804         }
2805         return 0;
2806
2807 out1:
2808         unregister_filesystem(&shmem_fs_type);
2809 out2:
2810         shmem_destroy_inodecache();
2811 out3:
2812         bdi_destroy(&shmem_backing_dev_info);
2813 out4:
2814         shm_mnt = ERR_PTR(error);
2815         return error;
2816 }
2817
2818 #else /* !CONFIG_SHMEM */
2819
2820 /*
2821  * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2822  *
2823  * This is intended for small system where the benefits of the full
2824  * shmem code (swap-backed and resource-limited) are outweighed by
2825  * their complexity. On systems without swap this code should be
2826  * effectively equivalent, but much lighter weight.
2827  */
2828
2829 static struct file_system_type shmem_fs_type = {
2830         .name           = "tmpfs",
2831         .mount          = ramfs_mount,
2832         .kill_sb        = kill_litter_super,
2833         .fs_flags       = FS_USERNS_MOUNT,
2834 };
2835
2836 int __init shmem_init(void)
2837 {
2838         BUG_ON(register_filesystem(&shmem_fs_type) != 0);
2839
2840         shm_mnt = kern_mount(&shmem_fs_type);
2841         BUG_ON(IS_ERR(shm_mnt));
2842
2843         return 0;
2844 }
2845
2846 int shmem_unuse(swp_entry_t swap, struct page *page)
2847 {
2848         return 0;
2849 }
2850
2851 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2852 {
2853         return 0;
2854 }
2855
2856 void shmem_unlock_mapping(struct address_space *mapping)
2857 {
2858 }
2859
2860 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
2861 {
2862         truncate_inode_pages_range(inode->i_mapping, lstart, lend);
2863 }
2864 EXPORT_SYMBOL_GPL(shmem_truncate_range);
2865
2866 #define shmem_vm_ops                            generic_file_vm_ops
2867 #define shmem_file_operations                   ramfs_file_operations
2868 #define shmem_get_inode(sb, dir, mode, dev, flags)      ramfs_get_inode(sb, dir, mode, dev)
2869 #define shmem_acct_size(flags, size)            0
2870 #define shmem_unacct_size(flags, size)          do {} while (0)
2871
2872 #endif /* CONFIG_SHMEM */
2873
2874 /* common code */
2875
2876 static struct dentry_operations anon_ops = {
2877         .d_dname = simple_dname
2878 };
2879
2880 static struct file *__shmem_file_setup(const char *name, loff_t size,
2881                                        unsigned long flags, unsigned int i_flags)
2882 {
2883         struct file *res;
2884         struct inode *inode;
2885         struct path path;
2886         struct super_block *sb;
2887         struct qstr this;
2888
2889         if (IS_ERR(shm_mnt))
2890                 return ERR_CAST(shm_mnt);
2891
2892         if (size < 0 || size > MAX_LFS_FILESIZE)
2893                 return ERR_PTR(-EINVAL);
2894
2895         if (shmem_acct_size(flags, size))
2896                 return ERR_PTR(-ENOMEM);
2897
2898         res = ERR_PTR(-ENOMEM);
2899         this.name = name;
2900         this.len = strlen(name);
2901         this.hash = 0; /* will go */
2902         sb = shm_mnt->mnt_sb;
2903         path.dentry = d_alloc_pseudo(sb, &this);
2904         if (!path.dentry)
2905                 goto put_memory;
2906         d_set_d_op(path.dentry, &anon_ops);
2907         path.mnt = mntget(shm_mnt);
2908
2909         res = ERR_PTR(-ENOSPC);
2910         inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2911         if (!inode)
2912                 goto put_dentry;
2913
2914         inode->i_flags |= i_flags;
2915         d_instantiate(path.dentry, inode);
2916         inode->i_size = size;
2917         clear_nlink(inode);     /* It is unlinked */
2918         res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
2919         if (IS_ERR(res))
2920                 goto put_dentry;
2921
2922         res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2923                   &shmem_file_operations);
2924         if (IS_ERR(res))
2925                 goto put_dentry;
2926
2927         return res;
2928
2929 put_dentry:
2930         path_put(&path);
2931 put_memory:
2932         shmem_unacct_size(flags, size);
2933         return res;
2934 }
2935
2936 /**
2937  * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
2938  *      kernel internal.  There will be NO LSM permission checks against the
2939  *      underlying inode.  So users of this interface must do LSM checks at a
2940  *      higher layer.  The one user is the big_key implementation.  LSM checks
2941  *      are provided at the key level rather than the inode level.
2942  * @name: name for dentry (to be seen in /proc/<pid>/maps
2943  * @size: size to be set for the file
2944  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2945  */
2946 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
2947 {
2948         return __shmem_file_setup(name, size, flags, S_PRIVATE);
2949 }
2950
2951 /**
2952  * shmem_file_setup - get an unlinked file living in tmpfs
2953  * @name: name for dentry (to be seen in /proc/<pid>/maps
2954  * @size: size to be set for the file
2955  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2956  */
2957 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2958 {
2959         return __shmem_file_setup(name, size, flags, 0);
2960 }
2961 EXPORT_SYMBOL_GPL(shmem_file_setup);
2962
2963 /**
2964  * shmem_zero_setup - setup a shared anonymous mapping
2965  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2966  */
2967 int shmem_zero_setup(struct vm_area_struct *vma)
2968 {
2969         struct file *file;
2970         loff_t size = vma->vm_end - vma->vm_start;
2971
2972         file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2973         if (IS_ERR(file))
2974                 return PTR_ERR(file);
2975
2976         if (vma->vm_file)
2977                 fput(vma->vm_file);
2978         vma->vm_file = file;
2979         vma->vm_ops = &shmem_vm_ops;
2980         return 0;
2981 }
2982
2983 /**
2984  * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2985  * @mapping:    the page's address_space
2986  * @index:      the page index
2987  * @gfp:        the page allocator flags to use if allocating
2988  *
2989  * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2990  * with any new page allocations done using the specified allocation flags.
2991  * But read_cache_page_gfp() uses the ->readpage() method: which does not
2992  * suit tmpfs, since it may have pages in swapcache, and needs to find those
2993  * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
2994  *
2995  * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
2996  * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
2997  */
2998 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
2999                                          pgoff_t index, gfp_t gfp)
3000 {
3001 #ifdef CONFIG_SHMEM
3002         struct inode *inode = mapping->host;
3003         struct page *page;
3004         int error;
3005
3006         BUG_ON(mapping->a_ops != &shmem_aops);
3007         error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
3008         if (error)
3009                 page = ERR_PTR(error);
3010         else
3011                 unlock_page(page);
3012         return page;
3013 #else
3014         /*
3015          * The tiny !SHMEM case uses ramfs without swap
3016          */
3017         return read_cache_page_gfp(mapping, index, gfp);
3018 #endif
3019 }
3020 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);