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/pagemap.h>
  29 #include <linux/file.h>
  30 #include <linux/mm.h>
  31 #include <linux/export.h>
  32 #include <linux/swap.h>
  33
  34 static struct vfsmount *shm_mnt;
  35
  36 #ifdef CONFIG_SHMEM
  37 /*
  38  * This virtual memory filesystem is heavily based on the ramfs. It
  39  * extends ramfs by the ability to use swap and honor resource limits
  40  * which makes it a completely usable filesystem.
  41  */
  42
  43 #include <linux/xattr.h>
  44 #include <linux/exportfs.h>
  45 #include <linux/posix_acl.h>
  46 #include <linux/generic_acl.h>
  47 #include <linux/mman.h>
  48 #include <linux/string.h>
  49 #include <linux/slab.h>
  50 #include <linux/backing-dev.h>
  51 #include <linux/shmem_fs.h>
  52 #include <linux/writeback.h>
  53 #include <linux/blkdev.h>
  54 #include <linux/pagevec.h>
  55 #include <linux/percpu_counter.h>
  56 #include <linux/splice.h>
  57 #include <linux/security.h>
  58 #include <linux/swapops.h>
  59 #include <linux/mempolicy.h>
  60 #include <linux/namei.h>
  61 #include <linux/ctype.h>
  62 #include <linux/migrate.h>
  63 #include <linux/highmem.h>
  64 #include <linux/seq_file.h>
  65 #include <linux/magic.h>
  66
  67 #include <asm/uaccess.h>
  68 #include <asm/pgtable.h>
  69
  70 #define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
  71 #define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
  72
  73 /* Pretend that each entry is of this size in directory's i_size */
  74 #define BOGO_DIRENT_SIZE 20
  75
  76 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
  77 #define SHORT_SYMLINK_LEN 128
  78
  79 struct shmem_xattr {
  80         struct list_head list;  /* anchored by shmem_inode_info->xattr_list */
  81         char *name;             /* xattr name */
  82         size_t size;
  83         char value[0];
  84 };
  85
  86 /* Flag allocation requirements to shmem_getpage */
  87 enum sgp_type {
  88         SGP_READ,       /* don't exceed i_size, don't allocate page */
  89         SGP_CACHE,      /* don't exceed i_size, may allocate page */
  90         SGP_DIRTY,      /* like SGP_CACHE, but set new page dirty */
  91         SGP_WRITE,      /* may exceed i_size, may allocate page */
  92 };
  93
  94 #ifdef CONFIG_TMPFS
  95 static unsigned long shmem_default_max_blocks(void)
  96 {
  97         return totalram_pages / 2;
  98 }
  99
 100 static unsigned long shmem_default_max_inodes(void)
 101 {
 102         return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
 103 }
 104 #endif
 105
 106 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
 107         struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
 108
 109 static inline int shmem_getpage(struct inode *inode, pgoff_t index,
 110         struct page **pagep, enum sgp_type sgp, int *fault_type)
 111 {
 112         return shmem_getpage_gfp(inode, index, pagep, sgp,
 113                         mapping_gfp_mask(inode->i_mapping), fault_type);
 114 }
 115
 116 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
 117 {
 118         return sb->s_fs_info;
 119 }
 120
 121 /*
 122  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
 123  * for shared memory and for shared anonymous (/dev/zero) mappings
 124  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
 125  * consistent with the pre-accounting of private mappings ...
 126  */
 127 static inline int shmem_acct_size(unsigned long flags, loff_t size)
 128 {
 129         return (flags & VM_NORESERVE) ?
 130                 0 : security_vm_enough_memory_kern(VM_ACCT(size));
 131 }
 132
 133 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
 134 {
 135         if (!(flags & VM_NORESERVE))
 136                 vm_unacct_memory(VM_ACCT(size));
 137 }
 138
 139 /*
 140  * ... whereas tmpfs objects are accounted incrementally as
 141  * pages are allocated, in order to allow huge sparse files.
 142  * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
 143  * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
 144  */
 145 static inline int shmem_acct_block(unsigned long flags)
 146 {
 147         return (flags & VM_NORESERVE) ?
 148                 security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE)) : 0;
 149 }
 150
 151 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
 152 {
 153         if (flags & VM_NORESERVE)
 154                 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
 155 }
 156
 157 static const struct super_operations shmem_ops;
 158 static const struct address_space_operations shmem_aops;
 159 static const struct file_operations shmem_file_operations;
 160 static const struct inode_operations shmem_inode_operations;
 161 static const struct inode_operations shmem_dir_inode_operations;
 162 static const struct inode_operations shmem_special_inode_operations;
 163 static const struct vm_operations_struct shmem_vm_ops;
 164
 165 static struct backing_dev_info shmem_backing_dev_info  __read_mostly = {
 166         .ra_pages       = 0,    /* No readahead */
 167         .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
 168 };
 169
 170 static LIST_HEAD(shmem_swaplist);
 171 static DEFINE_MUTEX(shmem_swaplist_mutex);
 172
 173 static int shmem_reserve_inode(struct super_block *sb)
 174 {
 175         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 176         if (sbinfo->max_inodes) {
 177                 spin_lock(&sbinfo->stat_lock);
 178                 if (!sbinfo->free_inodes) {
 179                         spin_unlock(&sbinfo->stat_lock);
 180                         return -ENOSPC;
 181                 }
 182                 sbinfo->free_inodes--;
 183                 spin_unlock(&sbinfo->stat_lock);
 184         }
 185         return 0;
 186 }
 187
 188 static void shmem_free_inode(struct super_block *sb)
 189 {
 190         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 191         if (sbinfo->max_inodes) {
 192                 spin_lock(&sbinfo->stat_lock);
 193                 sbinfo->free_inodes++;
 194                 spin_unlock(&sbinfo->stat_lock);
 195         }
 196 }
 197
 198 /**
 199  * shmem_recalc_inode - recalculate the block usage of an inode
 200  * @inode: inode to recalc
 201  *
 202  * We have to calculate the free blocks since the mm can drop
 203  * undirtied hole pages behind our back.
 204  *
 205  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
 206  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
 207  *
 208  * It has to be called with the spinlock held.
 209  */
 210 static void shmem_recalc_inode(struct inode *inode)
 211 {
 212         struct shmem_inode_info *info = SHMEM_I(inode);
 213         long freed;
 214
 215         freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
 216         if (freed > 0) {
 217                 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 218                 if (sbinfo->max_blocks)
 219                         percpu_counter_add(&sbinfo->used_blocks, -freed);
 220                 info->alloced -= freed;
 221                 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
 222                 shmem_unacct_blocks(info->flags, freed);
 223         }
 224 }
 225
 226 /*
 227  * Replace item expected in radix tree by a new item, while holding tree lock.
 228  */
 229 static int shmem_radix_tree_replace(struct address_space *mapping,
 230                         pgoff_t index, void *expected, void *replacement)
 231 {
 232         void **pslot;
 233         void *item = NULL;
 234
 235         VM_BUG_ON(!expected);
 236         pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
 237         if (pslot)
 238                 item = radix_tree_deref_slot_protected(pslot,
 239                                                         &mapping->tree_lock);
 240         if (item != expected)
 241                 return -ENOENT;
 242         if (replacement)
 243                 radix_tree_replace_slot(pslot, replacement);
 244         else
 245                 radix_tree_delete(&mapping->page_tree, index);
 246         return 0;
 247 }
 248
 249 /*
 250  * Like add_to_page_cache_locked, but error if expected item has gone.
 251  */
 252 static int shmem_add_to_page_cache(struct page *page,
 253                                    struct address_space *mapping,
 254                                    pgoff_t index, gfp_t gfp, void *expected)
 255 {
 256         int error = 0;
 257
 258         VM_BUG_ON(!PageLocked(page));
 259         VM_BUG_ON(!PageSwapBacked(page));
 260
 261         if (!expected)
 262                 error = radix_tree_preload(gfp & GFP_RECLAIM_MASK);
 263         if (!error) {
 264                 page_cache_get(page);
 265                 page->mapping = mapping;
 266                 page->index = index;
 267
 268                 spin_lock_irq(&mapping->tree_lock);
 269                 if (!expected)
 270                         error = radix_tree_insert(&mapping->page_tree,
 271                                                         index, page);
 272                 else
 273                         error = shmem_radix_tree_replace(mapping, index,
 274                                                         expected, page);
 275                 if (!error) {
 276                         mapping->nrpages++;
 277                         __inc_zone_page_state(page, NR_FILE_PAGES);
 278                         __inc_zone_page_state(page, NR_SHMEM);
 279                         spin_unlock_irq(&mapping->tree_lock);
 280                 } else {
 281                         page->mapping = NULL;
 282                         spin_unlock_irq(&mapping->tree_lock);
 283                         page_cache_release(page);
 284                 }
 285                 if (!expected)
 286                         radix_tree_preload_end();
 287         }
 288         if (error)
 289                 mem_cgroup_uncharge_cache_page(page);
 290         return error;
 291 }
 292
 293 /*
 294  * Like delete_from_page_cache, but substitutes swap for page.
 295  */
 296 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
 297 {
 298         struct address_space *mapping = page->mapping;
 299         int error;
 300
 301         spin_lock_irq(&mapping->tree_lock);
 302         error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
 303         page->mapping = NULL;
 304         mapping->nrpages--;
 305         __dec_zone_page_state(page, NR_FILE_PAGES);
 306         __dec_zone_page_state(page, NR_SHMEM);
 307         spin_unlock_irq(&mapping->tree_lock);
 308         page_cache_release(page);
 309         BUG_ON(error);
 310 }
 311
 312 /*
 313  * Like find_get_pages, but collecting swap entries as well as pages.
 314  */
 315 static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping,
 316                                         pgoff_t start, unsigned int nr_pages,
 317                                         struct page **pages, pgoff_t *indices)
 318 {
 319         unsigned int i;
 320         unsigned int ret;
 321         unsigned int nr_found;
 322
 323         rcu_read_lock();
 324 restart:
 325         nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
 326                                 (void ***)pages, indices, start, nr_pages);
 327         ret = 0;
 328         for (i = 0; i < nr_found; i++) {
 329                 struct page *page;
 330 repeat:
 331                 page = radix_tree_deref_slot((void **)pages[i]);
 332                 if (unlikely(!page))
 333                         continue;
 334                 if (radix_tree_exception(page)) {
 335                         if (radix_tree_deref_retry(page))
 336                                 goto restart;
 337                         /*
 338                          * Otherwise, we must be storing a swap entry
 339                          * here as an exceptional entry: so return it
 340                          * without attempting to raise page count.
 341                          */
 342                         goto export;
 343                 }
 344                 if (!page_cache_get_speculative(page))
 345                         goto repeat;
 346
 347                 /* Has the page moved? */
 348                 if (unlikely(page != *((void **)pages[i]))) {
 349                         page_cache_release(page);
 350                         goto repeat;
 351                 }
 352 export:
 353                 indices[ret] = indices[i];
 354                 pages[ret] = page;
 355                 ret++;
 356         }
 357         if (unlikely(!ret && nr_found))
 358                 goto restart;
 359         rcu_read_unlock();
 360         return ret;
 361 }
 362
 363 /*
 364  * Remove swap entry from radix tree, free the swap and its page cache.
 365  */
 366 static int shmem_free_swap(struct address_space *mapping,
 367                            pgoff_t index, void *radswap)
 368 {
 369         int error;
 370
 371         spin_lock_irq(&mapping->tree_lock);
 372         error = shmem_radix_tree_replace(mapping, index, radswap, NULL);
 373         spin_unlock_irq(&mapping->tree_lock);
 374         if (!error)
 375                 free_swap_and_cache(radix_to_swp_entry(radswap));
 376         return error;
 377 }
 378
 379 /*
 380  * Pagevec may contain swap entries, so shuffle up pages before releasing.
 381  */
 382 static void shmem_pagevec_release(struct pagevec *pvec)
 383 {
 384         int i, j;
 385
 386         for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
 387                 struct page *page = pvec->pages[i];
 388                 if (!radix_tree_exceptional_entry(page))
 389                         pvec->pages[j++] = page;
 390         }
 391         pvec->nr = j;
 392         pagevec_release(pvec);
 393 }
 394
 395 /*
 396  * Remove range of pages and swap entries from radix tree, and free them.
 397  */
 398 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
 399 {
 400         struct address_space *mapping = inode->i_mapping;
 401         struct shmem_inode_info *info = SHMEM_I(inode);
 402         pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 403         unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
 404         pgoff_t end = (lend >> PAGE_CACHE_SHIFT);
 405         struct pagevec pvec;
 406         pgoff_t indices[PAGEVEC_SIZE];
 407         long nr_swaps_freed = 0;
 408         pgoff_t index;
 409         int i;
 410
 411         BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
 412
 413         pagevec_init(&pvec, 0);
 414         index = start;
 415         while (index <= end) {
 416                 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
 417                         min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
 418                                                         pvec.pages, indices);
 419                 if (!pvec.nr)
 420                         break;
 421                 mem_cgroup_uncharge_start();
 422                 for (i = 0; i < pagevec_count(&pvec); i++) {
 423                         struct page *page = pvec.pages[i];
 424
 425                         index = indices[i];
 426                         if (index > end)
 427                                 break;
 428
 429                         if (radix_tree_exceptional_entry(page)) {
 430                                 nr_swaps_freed += !shmem_free_swap(mapping,
 431                                                                 index, page);
 432                                 continue;
 433                         }
 434
 435                         if (!trylock_page(page))
 436                                 continue;
 437                         if (page->mapping == mapping) {
 438                                 VM_BUG_ON(PageWriteback(page));
 439                                 truncate_inode_page(mapping, page);
 440                         }
 441                         unlock_page(page);
 442                 }
 443                 shmem_pagevec_release(&pvec);
 444                 mem_cgroup_uncharge_end();
 445                 cond_resched();
 446                 index++;
 447         }
 448
 449         if (partial) {
 450                 struct page *page = NULL;
 451                 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
 452                 if (page) {
 453                         zero_user_segment(page, partial, PAGE_CACHE_SIZE);
 454                         set_page_dirty(page);
 455                         unlock_page(page);
 456                         page_cache_release(page);
 457                 }
 458         }
 459
 460         index = start;
 461         for ( ; ; ) {
 462                 cond_resched();
 463                 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
 464                         min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
 465                                                         pvec.pages, indices);
 466                 if (!pvec.nr) {
 467                         if (index == start)
 468                                 break;
 469                         index = start;
 470                         continue;
 471                 }
 472                 if (index == start && indices[0] > end) {
 473                         shmem_pagevec_release(&pvec);
 474                         break;
 475                 }
 476                 mem_cgroup_uncharge_start();
 477                 for (i = 0; i < pagevec_count(&pvec); i++) {
 478                         struct page *page = pvec.pages[i];
 479
 480                         index = indices[i];
 481                         if (index > end)
 482                                 break;
 483
 484                         if (radix_tree_exceptional_entry(page)) {
 485                                 nr_swaps_freed += !shmem_free_swap(mapping,
 486                                                                 index, page);
 487                                 continue;
 488                         }
 489
 490                         lock_page(page);
 491                         if (page->mapping == mapping) {
 492                                 VM_BUG_ON(PageWriteback(page));
 493                                 truncate_inode_page(mapping, page);
 494                         }
 495                         unlock_page(page);
 496                 }
 497                 shmem_pagevec_release(&pvec);
 498                 mem_cgroup_uncharge_end();
 499                 index++;
 500         }
 501
 502         spin_lock(&info->lock);
 503         info->swapped -= nr_swaps_freed;
 504         shmem_recalc_inode(inode);
 505         spin_unlock(&info->lock);
 506
 507         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 508 }
 509 EXPORT_SYMBOL_GPL(shmem_truncate_range);
 510
 511 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
 512 {
 513         struct inode *inode = dentry->d_inode;
 514         int error;
 515
 516         error = inode_change_ok(inode, attr);
 517         if (error)
 518                 return error;
 519
 520         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
 521                 loff_t oldsize = inode->i_size;
 522                 loff_t newsize = attr->ia_size;
 523
 524                 if (newsize != oldsize) {
 525                         i_size_write(inode, newsize);
 526                         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 527                 }
 528                 if (newsize < oldsize) {
 529                         loff_t holebegin = round_up(newsize, PAGE_SIZE);
 530                         unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
 531                         shmem_truncate_range(inode, newsize, (loff_t)-1);
 532                         /* unmap again to remove racily COWed private pages */
 533                         unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
 534                 }
 535         }
 536
 537         setattr_copy(inode, attr);
 538 #ifdef CONFIG_TMPFS_POSIX_ACL
 539         if (attr->ia_valid & ATTR_MODE)
 540                 error = generic_acl_chmod(inode);
 541 #endif
 542         return error;
 543 }
 544
 545 static void shmem_evict_inode(struct inode *inode)
 546 {
 547         struct shmem_inode_info *info = SHMEM_I(inode);
 548         struct shmem_xattr *xattr, *nxattr;
 549
 550         if (inode->i_mapping->a_ops == &shmem_aops) {
 551                 shmem_unacct_size(info->flags, inode->i_size);
 552                 inode->i_size = 0;
 553                 shmem_truncate_range(inode, 0, (loff_t)-1);
 554                 if (!list_empty(&info->swaplist)) {
 555                         mutex_lock(&shmem_swaplist_mutex);
 556                         list_del_init(&info->swaplist);
 557                         mutex_unlock(&shmem_swaplist_mutex);
 558                 }
 559         } else
 560                 kfree(info->symlink);
 561
 562         list_for_each_entry_safe(xattr, nxattr, &info->xattr_list, list) {
 563                 kfree(xattr->name);
 564                 kfree(xattr);
 565         }
 566         BUG_ON(inode->i_blocks);
 567         shmem_free_inode(inode->i_sb);
 568         end_writeback(inode);
 569 }
 570
 571 /*
 572  * If swap found in inode, free it and move page from swapcache to filecache.
 573  */
 574 static int shmem_unuse_inode(struct shmem_inode_info *info,
 575                              swp_entry_t swap, struct page *page)
 576 {
 577         struct address_space *mapping = info->vfs_inode.i_mapping;
 578         void *radswap;
 579         pgoff_t index;
 580         int error;
 581
 582         radswap = swp_to_radix_entry(swap);
 583         index = radix_tree_locate_item(&mapping->page_tree, radswap);
 584         if (index == -1)
 585                 return 0;
 586
 587         /*
 588          * Move _head_ to start search for next from here.
 589          * But be careful: shmem_evict_inode checks list_empty without taking
 590          * mutex, and there's an instant in list_move_tail when info->swaplist
 591          * would appear empty, if it were the only one on shmem_swaplist.
 592          */
 593         if (shmem_swaplist.next != &info->swaplist)
 594                 list_move_tail(&shmem_swaplist, &info->swaplist);
 595
 596         /*
 597          * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
 598          * but also to hold up shmem_evict_inode(): so inode cannot be freed
 599          * beneath us (pagelock doesn't help until the page is in pagecache).
 600          */
 601         error = shmem_add_to_page_cache(page, mapping, index,
 602                                                 GFP_NOWAIT, radswap);
 603         /* which does mem_cgroup_uncharge_cache_page on error */
 604
 605         if (error != -ENOMEM) {
 606                 /*
 607                  * Truncation and eviction use free_swap_and_cache(), which
 608                  * only does trylock page: if we raced, best clean up here.
 609                  */
 610                 delete_from_swap_cache(page);
 611                 set_page_dirty(page);
 612                 if (!error) {
 613                         spin_lock(&info->lock);
 614                         info->swapped--;
 615                         spin_unlock(&info->lock);
 616                         swap_free(swap);
 617                 }
 618                 error = 1;      /* not an error, but entry was found */
 619         }
 620         return error;
 621 }
 622
 623 /*
 624  * Search through swapped inodes to find and replace swap by page.
 625  */
 626 int shmem_unuse(swp_entry_t swap, struct page *page)
 627 {
 628         struct list_head *this, *next;
 629         struct shmem_inode_info *info;
 630         int found = 0;
 631         int error;
 632
 633         /*
 634          * Charge page using GFP_KERNEL while we can wait, before taking
 635          * the shmem_swaplist_mutex which might hold up shmem_writepage().
 636          * Charged back to the user (not to caller) when swap account is used.
 637          */
 638         error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
 639         if (error)
 640                 goto out;
 641         /* No radix_tree_preload: swap entry keeps a place for page in tree */
 642
 643         mutex_lock(&shmem_swaplist_mutex);
 644         list_for_each_safe(this, next, &shmem_swaplist) {
 645                 info = list_entry(this, struct shmem_inode_info, swaplist);
 646                 if (info->swapped)
 647                         found = shmem_unuse_inode(info, swap, page);
 648                 else
 649                         list_del_init(&info->swaplist);
 650                 cond_resched();
 651                 if (found)
 652                         break;
 653         }
 654         mutex_unlock(&shmem_swaplist_mutex);
 655
 656         if (!found)
 657                 mem_cgroup_uncharge_cache_page(page);
 658         if (found < 0)
 659                 error = found;
 660 out:
 661         unlock_page(page);
 662         page_cache_release(page);
 663         return error;
 664 }
 665
 666 /*
 667  * Move the page from the page cache to the swap cache.
 668  */
 669 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
 670 {
 671         struct shmem_inode_info *info;
 672         struct address_space *mapping;
 673         struct inode *inode;
 674         swp_entry_t swap;
 675         pgoff_t index;
 676
 677         BUG_ON(!PageLocked(page));
 678         mapping = page->mapping;
 679         index = page->index;
 680         inode = mapping->host;
 681         info = SHMEM_I(inode);
 682         if (info->flags & VM_LOCKED)
 683                 goto redirty;
 684         if (!total_swap_pages)
 685                 goto redirty;
 686
 687         /*
 688          * shmem_backing_dev_info's capabilities prevent regular writeback or
 689          * sync from ever calling shmem_writepage; but a stacking filesystem
 690          * might use ->writepage of its underlying filesystem, in which case
 691          * tmpfs should write out to swap only in response to memory pressure,
 692          * and not for the writeback threads or sync.
 693          */
 694         if (!wbc->for_reclaim) {
 695                 WARN_ON_ONCE(1);        /* Still happens? Tell us about it! */
 696                 goto redirty;
 697         }
 698         swap = get_swap_page();
 699         if (!swap.val)
 700                 goto redirty;
 701
 702         /*
 703          * Add inode to shmem_unuse()'s list of swapped-out inodes,
 704          * if it's not already there.  Do it now before the page is
 705          * moved to swap cache, when its pagelock no longer protects
 706          * the inode from eviction.  But don't unlock the mutex until
 707          * we've incremented swapped, because shmem_unuse_inode() will
 708          * prune a !swapped inode from the swaplist under this mutex.
 709          */
 710         mutex_lock(&shmem_swaplist_mutex);
 711         if (list_empty(&info->swaplist))
 712                 list_add_tail(&info->swaplist, &shmem_swaplist);
 713
 714         if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
 715                 swap_shmem_alloc(swap);
 716                 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
 717
 718                 spin_lock(&info->lock);
 719                 info->swapped++;
 720                 shmem_recalc_inode(inode);
 721                 spin_unlock(&info->lock);
 722
 723                 mutex_unlock(&shmem_swaplist_mutex);
 724                 BUG_ON(page_mapped(page));
 725                 swap_writepage(page, wbc);
 726                 return 0;
 727         }
 728
 729         mutex_unlock(&shmem_swaplist_mutex);
 730         swapcache_free(swap, NULL);
 731 redirty:
 732         set_page_dirty(page);
 733         if (wbc->for_reclaim)
 734                 return AOP_WRITEPAGE_ACTIVATE;  /* Return with page locked */
 735         unlock_page(page);
 736         return 0;
 737 }
 738
 739 #ifdef CONFIG_NUMA
 740 #ifdef CONFIG_TMPFS
 741 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
 742 {
 743         char buffer[64];
 744
 745         if (!mpol || mpol->mode == MPOL_DEFAULT)
 746                 return;         /* show nothing */
 747
 748         mpol_to_str(buffer, sizeof(buffer), mpol, 1);
 749
 750         seq_printf(seq, ",mpol=%s", buffer);
 751 }
 752
 753 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
 754 {
 755         struct mempolicy *mpol = NULL;
 756         if (sbinfo->mpol) {
 757                 spin_lock(&sbinfo->stat_lock);  /* prevent replace/use races */
 758                 mpol = sbinfo->mpol;
 759                 mpol_get(mpol);
 760                 spin_unlock(&sbinfo->stat_lock);
 761         }
 762         return mpol;
 763 }
 764 #endif /* CONFIG_TMPFS */
 765
 766 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
 767                         struct shmem_inode_info *info, pgoff_t index)
 768 {
 769         struct mempolicy mpol, *spol;
 770         struct vm_area_struct pvma;
 771
 772         spol = mpol_cond_copy(&mpol,
 773                         mpol_shared_policy_lookup(&info->policy, index));
 774
 775         /* Create a pseudo vma that just contains the policy */
 776         pvma.vm_start = 0;
 777         pvma.vm_pgoff = index;
 778         pvma.vm_ops = NULL;
 779         pvma.vm_policy = spol;
 780         return swapin_readahead(swap, gfp, &pvma, 0);
 781 }
 782
 783 static struct page *shmem_alloc_page(gfp_t gfp,
 784                         struct shmem_inode_info *info, pgoff_t index)
 785 {
 786         struct vm_area_struct pvma;
 787
 788         /* Create a pseudo vma that just contains the policy */
 789         pvma.vm_start = 0;
 790         pvma.vm_pgoff = index;
 791         pvma.vm_ops = NULL;
 792         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
 793
 794         /*
 795          * alloc_page_vma() will drop the shared policy reference
 796          */
 797         return alloc_page_vma(gfp, &pvma, 0);
 798 }
 799 #else /* !CONFIG_NUMA */
 800 #ifdef CONFIG_TMPFS
 801 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
 802 {
 803 }
 804 #endif /* CONFIG_TMPFS */
 805
 806 static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
 807                         struct shmem_inode_info *info, pgoff_t index)
 808 {
 809         return swapin_readahead(swap, gfp, NULL, 0);
 810 }
 811
 812 static inline struct page *shmem_alloc_page(gfp_t gfp,
 813                         struct shmem_inode_info *info, pgoff_t index)
 814 {
 815         return alloc_page(gfp);
 816 }
 817 #endif /* CONFIG_NUMA */
 818
 819 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
 820 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
 821 {
 822         return NULL;
 823 }
 824 #endif
 825
 826 /*
 827  * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
 828  *
 829  * If we allocate a new one we do not mark it dirty. That's up to the
 830  * vm. If we swap it in we mark it dirty since we also free the swap
 831  * entry since a page cannot live in both the swap and page cache
 832  */
 833 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
 834         struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
 835 {
 836         struct address_space *mapping = inode->i_mapping;
 837         struct shmem_inode_info *info;
 838         struct shmem_sb_info *sbinfo;
 839         struct page *page;
 840         swp_entry_t swap;
 841         int error;
 842         int once = 0;
 843
 844         if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
 845                 return -EFBIG;
 846 repeat:
 847         swap.val = 0;
 848         page = find_lock_page(mapping, index);
 849         if (radix_tree_exceptional_entry(page)) {
 850                 swap = radix_to_swp_entry(page);
 851                 page = NULL;
 852         }
 853
 854         if (sgp != SGP_WRITE &&
 855             ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
 856                 error = -EINVAL;
 857                 goto failed;
 858         }
 859
 860         if (page || (sgp == SGP_READ && !swap.val)) {
 861                 /*
 862                  * Once we can get the page lock, it must be uptodate:
 863                  * if there were an error in reading back from swap,
 864                  * the page would not be inserted into the filecache.
 865                  */
 866                 BUG_ON(page && !PageUptodate(page));
 867                 *pagep = page;
 868                 return 0;
 869         }
 870
 871         /*
 872          * Fast cache lookup did not find it:
 873          * bring it back from swap or allocate.
 874          */
 875         info = SHMEM_I(inode);
 876         sbinfo = SHMEM_SB(inode->i_sb);
 877
 878         if (swap.val) {
 879                 /* Look it up and read it in.. */
 880                 page = lookup_swap_cache(swap);
 881                 if (!page) {
 882                         /* here we actually do the io */
 883                         if (fault_type)
 884                                 *fault_type |= VM_FAULT_MAJOR;
 885                         page = shmem_swapin(swap, gfp, info, index);
 886                         if (!page) {
 887                                 error = -ENOMEM;
 888                                 goto failed;
 889                         }
 890                 }
 891
 892                 /* We have to do this with page locked to prevent races */
 893                 lock_page(page);
 894                 if (!PageUptodate(page)) {
 895                         error = -EIO;
 896                         goto failed;
 897                 }
 898                 wait_on_page_writeback(page);
 899
 900                 /* Someone may have already done it for us */
 901                 if (page->mapping) {
 902                         if (page->mapping == mapping &&
 903                             page->index == index)
 904                                 goto done;
 905                         error = -EEXIST;
 906                         goto failed;
 907                 }
 908
 909                 error = mem_cgroup_cache_charge(page, current->mm,
 910                                                 gfp & GFP_RECLAIM_MASK);
 911                 if (!error)
 912                         error = shmem_add_to_page_cache(page, mapping, index,
 913                                                 gfp, swp_to_radix_entry(swap));
 914                 if (error)
 915                         goto failed;
 916
 917                 spin_lock(&info->lock);
 918                 info->swapped--;
 919                 shmem_recalc_inode(inode);
 920                 spin_unlock(&info->lock);
 921
 922                 delete_from_swap_cache(page);
 923                 set_page_dirty(page);
 924                 swap_free(swap);
 925
 926         } else {
 927                 if (shmem_acct_block(info->flags)) {
 928                         error = -ENOSPC;
 929                         goto failed;
 930                 }
 931                 if (sbinfo->max_blocks) {
 932                         if (percpu_counter_compare(&sbinfo->used_blocks,
 933                                                 sbinfo->max_blocks) >= 0) {
 934                                 error = -ENOSPC;
 935                                 goto unacct;
 936                         }
 937                         percpu_counter_inc(&sbinfo->used_blocks);
 938                 }
 939
 940                 page = shmem_alloc_page(gfp, info, index);
 941                 if (!page) {
 942                         error = -ENOMEM;
 943                         goto decused;
 944                 }
 945
 946                 SetPageSwapBacked(page);
 947                 __set_page_locked(page);
 948                 error = mem_cgroup_cache_charge(page, current->mm,
 949                                                 gfp & GFP_RECLAIM_MASK);
 950                 if (!error)
 951                         error = shmem_add_to_page_cache(page, mapping, index,
 952                                                 gfp, NULL);
 953                 if (error)
 954                         goto decused;
 955                 lru_cache_add_anon(page);
 956
 957                 spin_lock(&info->lock);
 958                 info->alloced++;
 959                 inode->i_blocks += BLOCKS_PER_PAGE;
 960                 shmem_recalc_inode(inode);
 961                 spin_unlock(&info->lock);
 962
 963                 clear_highpage(page);
 964                 flush_dcache_page(page);
 965                 SetPageUptodate(page);
 966                 if (sgp == SGP_DIRTY)
 967                         set_page_dirty(page);
 968         }
 969 done:
 970         /* Perhaps the file has been truncated since we checked */
 971         if (sgp != SGP_WRITE &&
 972             ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
 973                 error = -EINVAL;
 974                 goto trunc;
 975         }
 976         *pagep = page;
 977         return 0;
 978
 979         /*
 980          * Error recovery.
 981          */
 982 trunc:
 983         ClearPageDirty(page);
 984         delete_from_page_cache(page);
 985         spin_lock(&info->lock);
 986         info->alloced--;
 987         inode->i_blocks -= BLOCKS_PER_PAGE;
 988         spin_unlock(&info->lock);
 989 decused:
 990         if (sbinfo->max_blocks)
 991                 percpu_counter_add(&sbinfo->used_blocks, -1);
 992 unacct:
 993         shmem_unacct_blocks(info->flags, 1);
 994 failed:
 995         if (swap.val && error != -EINVAL) {
 996                 struct page *test = find_get_page(mapping, index);
 997                 if (test && !radix_tree_exceptional_entry(test))
 998                         page_cache_release(test);
 999                 /* Have another try if the entry has changed */
1000                 if (test != swp_to_radix_entry(swap))
1001                         error = -EEXIST;
1002         }
1003         if (page) {
1004                 unlock_page(page);
1005                 page_cache_release(page);
1006         }
1007         if (error == -ENOSPC && !once++) {
1008                 info = SHMEM_I(inode);
1009                 spin_lock(&info->lock);
1010                 shmem_recalc_inode(inode);
1011                 spin_unlock(&info->lock);
1012                 goto repeat;
1013         }
1014         if (error == -EEXIST)
1015                 goto repeat;
1016         return error;
1017 }
1018
1019 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1020 {
1021         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1022         int error;
1023         int ret = VM_FAULT_LOCKED;
1024
1025         error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1026         if (error)
1027                 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1028
1029         if (ret & VM_FAULT_MAJOR) {
1030                 count_vm_event(PGMAJFAULT);
1031                 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1032         }
1033         return ret;
1034 }
1035
1036 #ifdef CONFIG_NUMA
1037 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
1038 {
1039         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1040         return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
1041 }
1042
1043 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1044                                           unsigned long addr)
1045 {
1046         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1047         pgoff_t index;
1048
1049         index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1050         return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
1051 }
1052 #endif
1053
1054 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1055 {
1056         struct inode *inode = file->f_path.dentry->d_inode;
1057         struct shmem_inode_info *info = SHMEM_I(inode);
1058         int retval = -ENOMEM;
1059
1060         spin_lock(&info->lock);
1061         if (lock && !(info->flags & VM_LOCKED)) {
1062                 if (!user_shm_lock(inode->i_size, user))
1063                         goto out_nomem;
1064                 info->flags |= VM_LOCKED;
1065                 mapping_set_unevictable(file->f_mapping);
1066         }
1067         if (!lock && (info->flags & VM_LOCKED) && user) {
1068                 user_shm_unlock(inode->i_size, user);
1069                 info->flags &= ~VM_LOCKED;
1070                 mapping_clear_unevictable(file->f_mapping);
1071         }
1072         retval = 0;
1073
1074 out_nomem:
1075         spin_unlock(&info->lock);
1076         return retval;
1077 }
1078
1079 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1080 {
1081         file_accessed(file);
1082         vma->vm_ops = &shmem_vm_ops;
1083         vma->vm_flags |= VM_CAN_NONLINEAR;
1084         return 0;
1085 }
1086
1087 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1088                                      int mode, dev_t dev, unsigned long flags)
1089 {
1090         struct inode *inode;
1091         struct shmem_inode_info *info;
1092         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1093
1094         if (shmem_reserve_inode(sb))
1095                 return NULL;
1096
1097         inode = new_inode(sb);
1098         if (inode) {
1099                 inode->i_ino = get_next_ino();
1100                 inode_init_owner(inode, dir, mode);
1101                 inode->i_blocks = 0;
1102                 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1103                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1104                 inode->i_generation = get_seconds();
1105                 info = SHMEM_I(inode);
1106                 memset(info, 0, (char *)inode - (char *)info);
1107                 spin_lock_init(&info->lock);
1108                 info->flags = flags & VM_NORESERVE;
1109                 INIT_LIST_HEAD(&info->swaplist);
1110                 INIT_LIST_HEAD(&info->xattr_list);
1111                 cache_no_acl(inode);
1112
1113                 switch (mode & S_IFMT) {
1114                 default:
1115                         inode->i_op = &shmem_special_inode_operations;
1116                         init_special_inode(inode, mode, dev);
1117                         break;
1118                 case S_IFREG:
1119                         inode->i_mapping->a_ops = &shmem_aops;
1120                         inode->i_op = &shmem_inode_operations;
1121                         inode->i_fop = &shmem_file_operations;
1122                         mpol_shared_policy_init(&info->policy,
1123                                                  shmem_get_sbmpol(sbinfo));
1124                         break;
1125                 case S_IFDIR:
1126                         inc_nlink(inode);
1127                         /* Some things misbehave if size == 0 on a directory */
1128                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
1129                         inode->i_op = &shmem_dir_inode_operations;
1130                         inode->i_fop = &simple_dir_operations;
1131                         break;
1132                 case S_IFLNK:
1133                         /*
1134                          * Must not load anything in the rbtree,
1135                          * mpol_free_shared_policy will not be called.
1136                          */
1137                         mpol_shared_policy_init(&info->policy, NULL);
1138                         break;
1139                 }
1140         } else
1141                 shmem_free_inode(sb);
1142         return inode;
1143 }
1144
1145 #ifdef CONFIG_TMPFS
1146 static const struct inode_operations shmem_symlink_inode_operations;
1147 static const struct inode_operations shmem_short_symlink_operations;
1148
1149 static int
1150 shmem_write_begin(struct file *file, struct address_space *mapping,
1151                         loff_t pos, unsigned len, unsigned flags,
1152                         struct page **pagep, void **fsdata)
1153 {
1154         struct inode *inode = mapping->host;
1155         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1156         return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1157 }
1158
1159 static int
1160 shmem_write_end(struct file *file, struct address_space *mapping,
1161                         loff_t pos, unsigned len, unsigned copied,
1162                         struct page *page, void *fsdata)
1163 {
1164         struct inode *inode = mapping->host;
1165
1166         if (pos + copied > inode->i_size)
1167                 i_size_write(inode, pos + copied);
1168
1169         set_page_dirty(page);
1170         unlock_page(page);
1171         page_cache_release(page);
1172
1173         return copied;
1174 }
1175
1176 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1177 {
1178         struct inode *inode = filp->f_path.dentry->d_inode;
1179         struct address_space *mapping = inode->i_mapping;
1180         pgoff_t index;
1181         unsigned long offset;
1182         enum sgp_type sgp = SGP_READ;
1183
1184         /*
1185          * Might this read be for a stacking filesystem?  Then when reading
1186          * holes of a sparse file, we actually need to allocate those pages,
1187          * and even mark them dirty, so it cannot exceed the max_blocks limit.
1188          */
1189         if (segment_eq(get_fs(), KERNEL_DS))
1190                 sgp = SGP_DIRTY;
1191
1192         index = *ppos >> PAGE_CACHE_SHIFT;
1193         offset = *ppos & ~PAGE_CACHE_MASK;
1194
1195         for (;;) {
1196                 struct page *page = NULL;
1197                 pgoff_t end_index;
1198                 unsigned long nr, ret;
1199                 loff_t i_size = i_size_read(inode);
1200
1201                 end_index = i_size >> PAGE_CACHE_SHIFT;
1202                 if (index > end_index)
1203                         break;
1204                 if (index == end_index) {
1205                         nr = i_size & ~PAGE_CACHE_MASK;
1206                         if (nr <= offset)
1207                                 break;
1208                 }
1209
1210                 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1211                 if (desc->error) {
1212                         if (desc->error == -EINVAL)
1213                                 desc->error = 0;
1214                         break;
1215                 }
1216                 if (page)
1217                         unlock_page(page);
1218
1219                 /*
1220                  * We must evaluate after, since reads (unlike writes)
1221                  * are called without i_mutex protection against truncate
1222                  */
1223                 nr = PAGE_CACHE_SIZE;
1224                 i_size = i_size_read(inode);
1225                 end_index = i_size >> PAGE_CACHE_SHIFT;
1226                 if (index == end_index) {
1227                         nr = i_size & ~PAGE_CACHE_MASK;
1228                         if (nr <= offset) {
1229                                 if (page)
1230                                         page_cache_release(page);
1231                                 break;
1232                         }
1233                 }
1234                 nr -= offset;
1235
1236                 if (page) {
1237                         /*
1238                          * If users can be writing to this page using arbitrary
1239                          * virtual addresses, take care about potential aliasing
1240                          * before reading the page on the kernel side.
1241                          */
1242                         if (mapping_writably_mapped(mapping))
1243                                 flush_dcache_page(page);
1244                         /*
1245                          * Mark the page accessed if we read the beginning.
1246                          */
1247                         if (!offset)
1248                                 mark_page_accessed(page);
1249                 } else {
1250                         page = ZERO_PAGE(0);
1251                         page_cache_get(page);
1252                 }
1253
1254                 /*
1255                  * Ok, we have the page, and it's up-to-date, so
1256                  * now we can copy it to user space...
1257                  *
1258                  * The actor routine returns how many bytes were actually used..
1259                  * NOTE! This may not be the same as how much of a user buffer
1260                  * we filled up (we may be padding etc), so we can only update
1261                  * "pos" here (the actor routine has to update the user buffer
1262                  * pointers and the remaining count).
1263                  */
1264                 ret = actor(desc, page, offset, nr);
1265                 offset += ret;
1266                 index += offset >> PAGE_CACHE_SHIFT;
1267                 offset &= ~PAGE_CACHE_MASK;
1268
1269                 page_cache_release(page);
1270                 if (ret != nr || !desc->count)
1271                         break;
1272
1273                 cond_resched();
1274         }
1275
1276         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1277         file_accessed(filp);
1278 }
1279
1280 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1281                 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1282 {
1283         struct file *filp = iocb->ki_filp;
1284         ssize_t retval;
1285         unsigned long seg;
1286         size_t count;
1287         loff_t *ppos = &iocb->ki_pos;
1288
1289         retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1290         if (retval)
1291                 return retval;
1292
1293         for (seg = 0; seg < nr_segs; seg++) {
1294                 read_descriptor_t desc;
1295
1296                 desc.written = 0;
1297                 desc.arg.buf = iov[seg].iov_base;
1298                 desc.count = iov[seg].iov_len;
1299                 if (desc.count == 0)
1300                         continue;
1301                 desc.error = 0;
1302                 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1303                 retval += desc.written;
1304                 if (desc.error) {
1305                         retval = retval ?: desc.error;
1306                         break;
1307                 }
1308                 if (desc.count > 0)
1309                         break;
1310         }
1311         return retval;
1312 }
1313
1314 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1315                                 struct pipe_inode_info *pipe, size_t len,
1316                                 unsigned int flags)
1317 {
1318         struct address_space *mapping = in->f_mapping;
1319         struct inode *inode = mapping->host;
1320         unsigned int loff, nr_pages, req_pages;
1321         struct page *pages[PIPE_DEF_BUFFERS];
1322         struct partial_page partial[PIPE_DEF_BUFFERS];
1323         struct page *page;
1324         pgoff_t index, end_index;
1325         loff_t isize, left;
1326         int error, page_nr;
1327         struct splice_pipe_desc spd = {
1328                 .pages = pages,
1329                 .partial = partial,
1330                 .flags = flags,
1331                 .ops = &page_cache_pipe_buf_ops,
1332                 .spd_release = spd_release_page,
1333         };
1334
1335         isize = i_size_read(inode);
1336         if (unlikely(*ppos >= isize))
1337                 return 0;
1338
1339         left = isize - *ppos;
1340         if (unlikely(left < len))
1341                 len = left;
1342
1343         if (splice_grow_spd(pipe, &spd))
1344                 return -ENOMEM;
1345
1346         index = *ppos >> PAGE_CACHE_SHIFT;
1347         loff = *ppos & ~PAGE_CACHE_MASK;
1348         req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1349         nr_pages = min(req_pages, pipe->buffers);
1350
1351         spd.nr_pages = find_get_pages_contig(mapping, index,
1352                                                 nr_pages, spd.pages);
1353         index += spd.nr_pages;
1354         error = 0;
1355
1356         while (spd.nr_pages < nr_pages) {
1357                 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1358                 if (error)
1359                         break;
1360                 unlock_page(page);
1361                 spd.pages[spd.nr_pages++] = page;
1362                 index++;
1363         }
1364
1365         index = *ppos >> PAGE_CACHE_SHIFT;
1366         nr_pages = spd.nr_pages;
1367         spd.nr_pages = 0;
1368
1369         for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1370                 unsigned int this_len;
1371
1372                 if (!len)
1373                         break;
1374
1375                 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1376                 page = spd.pages[page_nr];
1377
1378                 if (!PageUptodate(page) || page->mapping != mapping) {
1379                         error = shmem_getpage(inode, index, &page,
1380                                                         SGP_CACHE, NULL);
1381                         if (error)
1382                                 break;
1383                         unlock_page(page);
1384                         page_cache_release(spd.pages[page_nr]);
1385                         spd.pages[page_nr] = page;
1386                 }
1387
1388                 isize = i_size_read(inode);
1389                 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1390                 if (unlikely(!isize || index > end_index))
1391                         break;
1392
1393                 if (end_index == index) {
1394                         unsigned int plen;
1395
1396                         plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1397                         if (plen <= loff)
1398                                 break;
1399
1400                         this_len = min(this_len, plen - loff);
1401                         len = this_len;
1402                 }
1403
1404                 spd.partial[page_nr].offset = loff;
1405                 spd.partial[page_nr].len = this_len;
1406                 len -= this_len;
1407                 loff = 0;
1408                 spd.nr_pages++;
1409                 index++;
1410         }
1411
1412         while (page_nr < nr_pages)
1413                 page_cache_release(spd.pages[page_nr++]);
1414
1415         if (spd.nr_pages)
1416                 error = splice_to_pipe(pipe, &spd);
1417
1418         splice_shrink_spd(pipe, &spd);
1419
1420         if (error > 0) {
1421                 *ppos += error;
1422                 file_accessed(in);
1423         }
1424         return error;
1425 }
1426
1427 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1428 {
1429         struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1430
1431         buf->f_type = TMPFS_MAGIC;
1432         buf->f_bsize = PAGE_CACHE_SIZE;
1433         buf->f_namelen = NAME_MAX;
1434         if (sbinfo->max_blocks) {
1435                 buf->f_blocks = sbinfo->max_blocks;
1436                 buf->f_bavail =
1437                 buf->f_bfree  = sbinfo->max_blocks -
1438                                 percpu_counter_sum(&sbinfo->used_blocks);
1439         }
1440         if (sbinfo->max_inodes) {
1441                 buf->f_files = sbinfo->max_inodes;
1442                 buf->f_ffree = sbinfo->free_inodes;
1443         }
1444         /* else leave those fields 0 like simple_statfs */
1445         return 0;
1446 }
1447
1448 /*
1449  * File creation. Allocate an inode, and we're done..
1450  */
1451 static int
1452 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1453 {
1454         struct inode *inode;
1455         int error = -ENOSPC;
1456
1457         inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1458         if (inode) {
1459                 error = security_inode_init_security(inode, dir,
1460                                                      &dentry->d_name,
1461                                                      NULL, NULL);
1462                 if (error) {
1463                         if (error != -EOPNOTSUPP) {
1464                                 iput(inode);
1465                                 return error;
1466                         }
1467                 }
1468 #ifdef CONFIG_TMPFS_POSIX_ACL
1469                 error = generic_acl_init(inode, dir);
1470                 if (error) {
1471                         iput(inode);
1472                         return error;
1473                 }
1474 #else
1475                 error = 0;
1476 #endif
1477                 dir->i_size += BOGO_DIRENT_SIZE;
1478                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1479                 d_instantiate(dentry, inode);
1480                 dget(dentry); /* Extra count - pin the dentry in core */
1481         }
1482         return error;
1483 }
1484
1485 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1486 {
1487         int error;
1488
1489         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1490                 return error;
1491         inc_nlink(dir);
1492         return 0;
1493 }
1494
1495 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1496                 struct nameidata *nd)
1497 {
1498         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1499 }
1500
1501 /*
1502  * Link a file..
1503  */
1504 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1505 {
1506         struct inode *inode = old_dentry->d_inode;
1507         int ret;
1508
1509         /*
1510          * No ordinary (disk based) filesystem counts links as inodes;
1511          * but each new link needs a new dentry, pinning lowmem, and
1512          * tmpfs dentries cannot be pruned until they are unlinked.
1513          */
1514         ret = shmem_reserve_inode(inode->i_sb);
1515         if (ret)
1516                 goto out;
1517
1518         dir->i_size += BOGO_DIRENT_SIZE;
1519         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1520         inc_nlink(inode);
1521         ihold(inode);   /* New dentry reference */
1522         dget(dentry);           /* Extra pinning count for the created dentry */
1523         d_instantiate(dentry, inode);
1524 out:
1525         return ret;
1526 }
1527
1528 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1529 {
1530         struct inode *inode = dentry->d_inode;
1531
1532         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1533                 shmem_free_inode(inode->i_sb);
1534
1535         dir->i_size -= BOGO_DIRENT_SIZE;
1536         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1537         drop_nlink(inode);
1538         dput(dentry);   /* Undo the count from "create" - this does all the work */
1539         return 0;
1540 }
1541
1542 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1543 {
1544         if (!simple_empty(dentry))
1545                 return -ENOTEMPTY;
1546
1547         drop_nlink(dentry->d_inode);
1548         drop_nlink(dir);
1549         return shmem_unlink(dir, dentry);
1550 }
1551
1552 /*
1553  * The VFS layer already does all the dentry stuff for rename,
1554  * we just have to decrement the usage count for the target if
1555  * it exists so that the VFS layer correctly free's it when it
1556  * gets overwritten.
1557  */
1558 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1559 {
1560         struct inode *inode = old_dentry->d_inode;
1561         int they_are_dirs = S_ISDIR(inode->i_mode);
1562
1563         if (!simple_empty(new_dentry))
1564                 return -ENOTEMPTY;
1565
1566         if (new_dentry->d_inode) {
1567                 (void) shmem_unlink(new_dir, new_dentry);
1568                 if (they_are_dirs)
1569                         drop_nlink(old_dir);
1570         } else if (they_are_dirs) {
1571                 drop_nlink(old_dir);
1572                 inc_nlink(new_dir);
1573         }
1574
1575         old_dir->i_size -= BOGO_DIRENT_SIZE;
1576         new_dir->i_size += BOGO_DIRENT_SIZE;
1577         old_dir->i_ctime = old_dir->i_mtime =
1578         new_dir->i_ctime = new_dir->i_mtime =
1579         inode->i_ctime = CURRENT_TIME;
1580         return 0;
1581 }
1582
1583 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1584 {
1585         int error;
1586         int len;
1587         struct inode *inode;
1588         struct page *page;
1589         char *kaddr;
1590         struct shmem_inode_info *info;
1591
1592         len = strlen(symname) + 1;
1593         if (len > PAGE_CACHE_SIZE)
1594                 return -ENAMETOOLONG;
1595
1596         inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
1597         if (!inode)
1598                 return -ENOSPC;
1599
1600         error = security_inode_init_security(inode, dir, &dentry->d_name,
1601                                              NULL, NULL);
1602         if (error) {
1603                 if (error != -EOPNOTSUPP) {
1604                         iput(inode);
1605                         return error;
1606                 }
1607                 error = 0;
1608         }
1609
1610         info = SHMEM_I(inode);
1611         inode->i_size = len-1;
1612         if (len <= SHORT_SYMLINK_LEN) {
1613                 info->symlink = kmemdup(symname, len, GFP_KERNEL);
1614                 if (!info->symlink) {
1615                         iput(inode);
1616                         return -ENOMEM;
1617                 }
1618                 inode->i_op = &shmem_short_symlink_operations;
1619         } else {
1620                 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1621                 if (error) {
1622                         iput(inode);
1623                         return error;
1624                 }
1625                 inode->i_mapping->a_ops = &shmem_aops;
1626                 inode->i_op = &shmem_symlink_inode_operations;
1627                 kaddr = kmap_atomic(page, KM_USER0);
1628                 memcpy(kaddr, symname, len);
1629                 kunmap_atomic(kaddr, KM_USER0);
1630                 set_page_dirty(page);
1631                 unlock_page(page);
1632                 page_cache_release(page);
1633         }
1634         dir->i_size += BOGO_DIRENT_SIZE;
1635         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1636         d_instantiate(dentry, inode);
1637         dget(dentry);
1638         return 0;
1639 }
1640
1641 static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
1642 {
1643         nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
1644         return NULL;
1645 }
1646
1647 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1648 {
1649         struct page *page = NULL;
1650         int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1651         nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
1652         if (page)
1653                 unlock_page(page);
1654         return page;
1655 }
1656
1657 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1658 {
1659         if (!IS_ERR(nd_get_link(nd))) {
1660                 struct page *page = cookie;
1661                 kunmap(page);
1662                 mark_page_accessed(page);
1663                 page_cache_release(page);
1664         }
1665 }
1666
1667 #ifdef CONFIG_TMPFS_XATTR
1668 /*
1669  * Superblocks without xattr inode operations may get some security.* xattr
1670  * support from the LSM "for free". As soon as we have any other xattrs
1671  * like ACLs, we also need to implement the security.* handlers at
1672  * filesystem level, though.
1673  */
1674
1675 static int shmem_xattr_get(struct dentry *dentry, const char *name,
1676                            void *buffer, size_t size)
1677 {
1678         struct shmem_inode_info *info;
1679         struct shmem_xattr *xattr;
1680         int ret = -ENODATA;
1681
1682         info = SHMEM_I(dentry->d_inode);
1683
1684         spin_lock(&info->lock);
1685         list_for_each_entry(xattr, &info->xattr_list, list) {
1686                 if (strcmp(name, xattr->name))
1687                         continue;
1688
1689                 ret = xattr->size;
1690                 if (buffer) {
1691                         if (size < xattr->size)
1692                                 ret = -ERANGE;
1693                         else
1694                                 memcpy(buffer, xattr->value, xattr->size);
1695                 }
1696                 break;
1697         }
1698         spin_unlock(&info->lock);
1699         return ret;
1700 }
1701
1702 static int shmem_xattr_set(struct dentry *dentry, const char *name,
1703                            const void *value, size_t size, int flags)
1704 {
1705         struct inode *inode = dentry->d_inode;
1706         struct shmem_inode_info *info = SHMEM_I(inode);
1707         struct shmem_xattr *xattr;
1708         struct shmem_xattr *new_xattr = NULL;
1709         size_t len;
1710         int err = 0;
1711
1712         /* value == NULL means remove */
1713         if (value) {
1714                 /* wrap around? */
1715                 len = sizeof(*new_xattr) + size;
1716                 if (len <= sizeof(*new_xattr))
1717                         return -ENOMEM;
1718
1719                 new_xattr = kmalloc(len, GFP_KERNEL);
1720                 if (!new_xattr)
1721                         return -ENOMEM;
1722
1723                 new_xattr->name = kstrdup(name, GFP_KERNEL);
1724                 if (!new_xattr->name) {
1725                         kfree(new_xattr);
1726                         return -ENOMEM;
1727                 }
1728
1729                 new_xattr->size = size;
1730                 memcpy(new_xattr->value, value, size);
1731         }
1732
1733         spin_lock(&info->lock);
1734         list_for_each_entry(xattr, &info->xattr_list, list) {
1735                 if (!strcmp(name, xattr->name)) {
1736                         if (flags & XATTR_CREATE) {
1737                                 xattr = new_xattr;
1738                                 err = -EEXIST;
1739                         } else if (new_xattr) {
1740                                 list_replace(&xattr->list, &new_xattr->list);
1741                         } else {
1742                                 list_del(&xattr->list);
1743                         }
1744                         goto out;
1745                 }
1746         }
1747         if (flags & XATTR_REPLACE) {
1748                 xattr = new_xattr;
1749                 err = -ENODATA;
1750         } else {
1751                 list_add(&new_xattr->list, &info->xattr_list);
1752                 xattr = NULL;
1753         }
1754 out:
1755         spin_unlock(&info->lock);
1756         if (xattr)
1757                 kfree(xattr->name);
1758         kfree(xattr);
1759         return err;
1760 }
1761
1762 static const struct xattr_handler *shmem_xattr_handlers[] = {
1763 #ifdef CONFIG_TMPFS_POSIX_ACL
1764         &generic_acl_access_handler,
1765         &generic_acl_default_handler,
1766 #endif
1767         NULL
1768 };
1769
1770 static int shmem_xattr_validate(const char *name)
1771 {
1772         struct { const char *prefix; size_t len; } arr[] = {
1773                 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
1774                 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
1775         };
1776         int i;
1777
1778         for (i = 0; i < ARRAY_SIZE(arr); i++) {
1779                 size_t preflen = arr[i].len;
1780                 if (strncmp(name, arr[i].prefix, preflen) == 0) {
1781                         if (!name[preflen])
1782                                 return -EINVAL;
1783                         return 0;
1784                 }
1785         }
1786         return -EOPNOTSUPP;
1787 }
1788
1789 static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
1790                               void *buffer, size_t size)
1791 {
1792         int err;
1793
1794         /*
1795          * If this is a request for a synthetic attribute in the system.*
1796          * namespace use the generic infrastructure to resolve a handler
1797          * for it via sb->s_xattr.
1798          */
1799         if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1800                 return generic_getxattr(dentry, name, buffer, size);
1801
1802         err = shmem_xattr_validate(name);
1803         if (err)
1804                 return err;
1805
1806         return shmem_xattr_get(dentry, name, buffer, size);
1807 }
1808
1809 static int shmem_setxattr(struct dentry *dentry, const char *name,
1810                           const void *value, size_t size, int flags)
1811 {
1812         int err;
1813
1814         /*
1815          * If this is a request for a synthetic attribute in the system.*
1816          * namespace use the generic infrastructure to resolve a handler
1817          * for it via sb->s_xattr.
1818          */
1819         if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1820                 return generic_setxattr(dentry, name, value, size, flags);
1821
1822         err = shmem_xattr_validate(name);
1823         if (err)
1824                 return err;
1825
1826         if (size == 0)
1827                 value = "";  /* empty EA, do not remove */
1828
1829         return shmem_xattr_set(dentry, name, value, size, flags);
1830
1831 }
1832
1833 static int shmem_removexattr(struct dentry *dentry, const char *name)
1834 {
1835         int err;
1836
1837         /*
1838          * If this is a request for a synthetic attribute in the system.*
1839          * namespace use the generic infrastructure to resolve a handler
1840          * for it via sb->s_xattr.
1841          */
1842         if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1843                 return generic_removexattr(dentry, name);
1844
1845         err = shmem_xattr_validate(name);
1846         if (err)
1847                 return err;
1848
1849         return shmem_xattr_set(dentry, name, NULL, 0, XATTR_REPLACE);
1850 }
1851
1852 static bool xattr_is_trusted(const char *name)
1853 {
1854         return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN);
1855 }
1856
1857 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
1858 {
1859         bool trusted = capable(CAP_SYS_ADMIN);
1860         struct shmem_xattr *xattr;
1861         struct shmem_inode_info *info;
1862         size_t used = 0;
1863
1864         info = SHMEM_I(dentry->d_inode);
1865
1866         spin_lock(&info->lock);
1867         list_for_each_entry(xattr, &info->xattr_list, list) {
1868                 size_t len;
1869
1870                 /* skip "trusted." attributes for unprivileged callers */
1871                 if (!trusted && xattr_is_trusted(xattr->name))
1872                         continue;
1873
1874                 len = strlen(xattr->name) + 1;
1875                 used += len;
1876                 if (buffer) {
1877                         if (size < used) {
1878                                 used = -ERANGE;
1879                                 break;
1880                         }
1881                         memcpy(buffer, xattr->name, len);
1882                         buffer += len;
1883                 }
1884         }
1885         spin_unlock(&info->lock);
1886
1887         return used;
1888 }
1889 #endif /* CONFIG_TMPFS_XATTR */
1890
1891 static const struct inode_operations shmem_short_symlink_operations = {
1892         .readlink       = generic_readlink,
1893         .follow_link    = shmem_follow_short_symlink,
1894 #ifdef CONFIG_TMPFS_XATTR
1895         .setxattr       = shmem_setxattr,
1896         .getxattr       = shmem_getxattr,
1897         .listxattr      = shmem_listxattr,
1898         .removexattr    = shmem_removexattr,
1899 #endif
1900 };
1901
1902 static const struct inode_operations shmem_symlink_inode_operations = {
1903         .readlink       = generic_readlink,
1904         .follow_link    = shmem_follow_link,
1905         .put_link       = shmem_put_link,
1906 #ifdef CONFIG_TMPFS_XATTR
1907         .setxattr       = shmem_setxattr,
1908         .getxattr       = shmem_getxattr,
1909         .listxattr      = shmem_listxattr,
1910         .removexattr    = shmem_removexattr,
1911 #endif
1912 };
1913
1914 static struct dentry *shmem_get_parent(struct dentry *child)
1915 {
1916         return ERR_PTR(-ESTALE);
1917 }
1918
1919 static int shmem_match(struct inode *ino, void *vfh)
1920 {
1921         __u32 *fh = vfh;
1922         __u64 inum = fh[2];
1923         inum = (inum << 32) | fh[1];
1924         return ino->i_ino == inum && fh[0] == ino->i_generation;
1925 }
1926
1927 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
1928                 struct fid *fid, int fh_len, int fh_type)
1929 {
1930         struct inode *inode;
1931         struct dentry *dentry = NULL;
1932         u64 inum = fid->raw[2];
1933         inum = (inum << 32) | fid->raw[1];
1934
1935         if (fh_len < 3)
1936                 return NULL;
1937
1938         inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
1939                         shmem_match, fid->raw);
1940         if (inode) {
1941                 dentry = d_find_alias(inode);
1942                 iput(inode);
1943         }
1944
1945         return dentry;
1946 }
1947
1948 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
1949                                 int connectable)
1950 {
1951         struct inode *inode = dentry->d_inode;
1952
1953         if (*len < 3) {
1954                 *len = 3;
1955                 return 255;
1956         }
1957
1958         if (inode_unhashed(inode)) {
1959                 /* Unfortunately insert_inode_hash is not idempotent,
1960                  * so as we hash inodes here rather than at creation
1961                  * time, we need a lock to ensure we only try
1962                  * to do it once
1963                  */
1964                 static DEFINE_SPINLOCK(lock);
1965                 spin_lock(&lock);
1966                 if (inode_unhashed(inode))
1967                         __insert_inode_hash(inode,
1968                                             inode->i_ino + inode->i_generation);
1969                 spin_unlock(&lock);
1970         }
1971
1972         fh[0] = inode->i_generation;
1973         fh[1] = inode->i_ino;
1974         fh[2] = ((__u64)inode->i_ino) >> 32;
1975
1976         *len = 3;
1977         return 1;
1978 }
1979
1980 static const struct export_operations shmem_export_ops = {
1981         .get_parent     = shmem_get_parent,
1982         .encode_fh      = shmem_encode_fh,
1983         .fh_to_dentry   = shmem_fh_to_dentry,
1984 };
1985
1986 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
1987                                bool remount)
1988 {
1989         char *this_char, *value, *rest;
1990
1991         while (options != NULL) {
1992                 this_char = options;
1993                 for (;;) {
1994                         /*
1995                          * NUL-terminate this option: unfortunately,
1996                          * mount options form a comma-separated list,
1997                          * but mpol's nodelist may also contain commas.
1998                          */
1999                         options = strchr(options, ',');
2000                         if (options == NULL)
2001                                 break;
2002                         options++;
2003                         if (!isdigit(*options)) {
2004                                 options[-1] = '\0';
2005                                 break;
2006                         }
2007                 }
2008                 if (!*this_char)
2009                         continue;
2010                 if ((value = strchr(this_char,'=')) != NULL) {
2011                         *value++ = 0;
2012                 } else {
2013                         printk(KERN_ERR
2014                             "tmpfs: No value for mount option '%s'\n",
2015                             this_char);
2016                         return 1;
2017                 }
2018
2019                 if (!strcmp(this_char,"size")) {
2020                         unsigned long long size;
2021                         size = memparse(value,&rest);
2022                         if (*rest == '%') {
2023                                 size <<= PAGE_SHIFT;
2024                                 size *= totalram_pages;
2025                                 do_div(size, 100);
2026                                 rest++;
2027                         }
2028                         if (*rest)
2029                                 goto bad_val;
2030                         sbinfo->max_blocks =
2031                                 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2032                 } else if (!strcmp(this_char,"nr_blocks")) {
2033                         sbinfo->max_blocks = memparse(value, &rest);
2034                         if (*rest)
2035                                 goto bad_val;
2036                 } else if (!strcmp(this_char,"nr_inodes")) {
2037                         sbinfo->max_inodes = memparse(value, &rest);
2038                         if (*rest)
2039                                 goto bad_val;
2040                 } else if (!strcmp(this_char,"mode")) {
2041                         if (remount)
2042                                 continue;
2043                         sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2044                         if (*rest)
2045                                 goto bad_val;
2046                 } else if (!strcmp(this_char,"uid")) {
2047                         if (remount)
2048                                 continue;
2049                         sbinfo->uid = simple_strtoul(value, &rest, 0);
2050                         if (*rest)
2051                                 goto bad_val;
2052                 } else if (!strcmp(this_char,"gid")) {
2053                         if (remount)
2054                                 continue;
2055                         sbinfo->gid = simple_strtoul(value, &rest, 0);
2056                         if (*rest)
2057                                 goto bad_val;
2058                 } else if (!strcmp(this_char,"mpol")) {
2059                         if (mpol_parse_str(value, &sbinfo->mpol, 1))
2060                                 goto bad_val;
2061                 } else {
2062                         printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2063                                this_char);
2064                         return 1;
2065                 }
2066         }
2067         return 0;
2068
2069 bad_val:
2070         printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2071                value, this_char);
2072         return 1;
2073
2074 }
2075
2076 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2077 {
2078         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2079         struct shmem_sb_info config = *sbinfo;
2080         unsigned long inodes;
2081         int error = -EINVAL;
2082
2083         if (shmem_parse_options(data, &config, true))
2084                 return error;
2085
2086         spin_lock(&sbinfo->stat_lock);
2087         inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2088         if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2089                 goto out;
2090         if (config.max_inodes < inodes)
2091                 goto out;
2092         /*
2093          * Those tests disallow limited->unlimited while any are in use;
2094          * but we must separately disallow unlimited->limited, because
2095          * in that case we have no record of how much is already in use.
2096          */
2097         if (config.max_blocks && !sbinfo->max_blocks)
2098                 goto out;
2099         if (config.max_inodes && !sbinfo->max_inodes)
2100                 goto out;
2101
2102         error = 0;
2103         sbinfo->max_blocks  = config.max_blocks;
2104         sbinfo->max_inodes  = config.max_inodes;
2105         sbinfo->free_inodes = config.max_inodes - inodes;
2106
2107         mpol_put(sbinfo->mpol);
2108         sbinfo->mpol        = config.mpol;      /* transfers initial ref */
2109 out:
2110         spin_unlock(&sbinfo->stat_lock);
2111         return error;
2112 }
2113
2114 static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
2115 {
2116         struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
2117
2118         if (sbinfo->max_blocks != shmem_default_max_blocks())
2119                 seq_printf(seq, ",size=%luk",
2120                         sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2121         if (sbinfo->max_inodes != shmem_default_max_inodes())
2122                 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2123         if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2124                 seq_printf(seq, ",mode=%03o", sbinfo->mode);
2125         if (sbinfo->uid != 0)
2126                 seq_printf(seq, ",uid=%u", sbinfo->uid);
2127         if (sbinfo->gid != 0)
2128                 seq_printf(seq, ",gid=%u", sbinfo->gid);
2129         shmem_show_mpol(seq, sbinfo->mpol);
2130         return 0;
2131 }
2132 #endif /* CONFIG_TMPFS */
2133
2134 static void shmem_put_super(struct super_block *sb)
2135 {
2136         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2137
2138         percpu_counter_destroy(&sbinfo->used_blocks);
2139         kfree(sbinfo);
2140         sb->s_fs_info = NULL;
2141 }
2142
2143 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2144 {
2145         struct inode *inode;
2146         struct dentry *root;
2147         struct shmem_sb_info *sbinfo;
2148         int err = -ENOMEM;
2149
2150         /* Round up to L1_CACHE_BYTES to resist false sharing */
2151         sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2152                                 L1_CACHE_BYTES), GFP_KERNEL);
2153         if (!sbinfo)
2154                 return -ENOMEM;
2155
2156         sbinfo->mode = S_IRWXUGO | S_ISVTX;
2157         sbinfo->uid = current_fsuid();
2158         sbinfo->gid = current_fsgid();
2159         sb->s_fs_info = sbinfo;
2160
2161 #ifdef CONFIG_TMPFS
2162         /*
2163          * Per default we only allow half of the physical ram per
2164          * tmpfs instance, limiting inodes to one per page of lowmem;
2165          * but the internal instance is left unlimited.
2166          */
2167         if (!(sb->s_flags & MS_NOUSER)) {
2168                 sbinfo->max_blocks = shmem_default_max_blocks();
2169                 sbinfo->max_inodes = shmem_default_max_inodes();
2170                 if (shmem_parse_options(data, sbinfo, false)) {
2171                         err = -EINVAL;
2172                         goto failed;
2173                 }
2174         }
2175         sb->s_export_op = &shmem_export_ops;
2176 #else
2177         sb->s_flags |= MS_NOUSER;
2178 #endif
2179
2180         spin_lock_init(&sbinfo->stat_lock);
2181         if (percpu_counter_init(&sbinfo->used_blocks, 0))
2182                 goto failed;
2183         sbinfo->free_inodes = sbinfo->max_inodes;
2184
2185         sb->s_maxbytes = MAX_LFS_FILESIZE;
2186         sb->s_blocksize = PAGE_CACHE_SIZE;
2187         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2188         sb->s_magic = TMPFS_MAGIC;
2189         sb->s_op = &shmem_ops;
2190         sb->s_time_gran = 1;
2191 #ifdef CONFIG_TMPFS_XATTR
2192         sb->s_xattr = shmem_xattr_handlers;
2193 #endif
2194 #ifdef CONFIG_TMPFS_POSIX_ACL
2195         sb->s_flags |= MS_POSIXACL;
2196 #endif
2197
2198         inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2199         if (!inode)
2200                 goto failed;
2201         inode->i_uid = sbinfo->uid;
2202         inode->i_gid = sbinfo->gid;
2203         root = d_alloc_root(inode);
2204         if (!root)
2205                 goto failed_iput;
2206         sb->s_root = root;
2207         return 0;
2208
2209 failed_iput:
2210         iput(inode);
2211 failed:
2212         shmem_put_super(sb);
2213         return err;
2214 }
2215
2216 static struct kmem_cache *shmem_inode_cachep;
2217
2218 static struct inode *shmem_alloc_inode(struct super_block *sb)
2219 {
2220         struct shmem_inode_info *info;
2221         info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2222         if (!info)
2223                 return NULL;
2224         return &info->vfs_inode;
2225 }
2226
2227 static void shmem_destroy_callback(struct rcu_head *head)
2228 {
2229         struct inode *inode = container_of(head, struct inode, i_rcu);
2230         INIT_LIST_HEAD(&inode->i_dentry);
2231         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2232 }
2233
2234 static void shmem_destroy_inode(struct inode *inode)
2235 {
2236         if ((inode->i_mode & S_IFMT) == S_IFREG)
2237                 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2238         call_rcu(&inode->i_rcu, shmem_destroy_callback);
2239 }
2240
2241 static void shmem_init_inode(void *foo)
2242 {
2243         struct shmem_inode_info *info = foo;
2244         inode_init_once(&info->vfs_inode);
2245 }
2246
2247 static int shmem_init_inodecache(void)
2248 {
2249         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2250                                 sizeof(struct shmem_inode_info),
2251                                 0, SLAB_PANIC, shmem_init_inode);
2252         return 0;
2253 }
2254
2255 static void shmem_destroy_inodecache(void)
2256 {
2257         kmem_cache_destroy(shmem_inode_cachep);
2258 }
2259
2260 static const struct address_space_operations shmem_aops = {
2261         .writepage      = shmem_writepage,
2262         .set_page_dirty = __set_page_dirty_no_writeback,
2263 #ifdef CONFIG_TMPFS
2264         .write_begin    = shmem_write_begin,
2265         .write_end      = shmem_write_end,
2266 #endif
2267         .migratepage    = migrate_page,
2268         .error_remove_page = generic_error_remove_page,
2269 };
2270
2271 static const struct file_operations shmem_file_operations = {
2272         .mmap           = shmem_mmap,
2273 #ifdef CONFIG_TMPFS
2274         .llseek         = generic_file_llseek,
2275         .read           = do_sync_read,
2276         .write          = do_sync_write,
2277         .aio_read       = shmem_file_aio_read,
2278         .aio_write      = generic_file_aio_write,
2279         .fsync          = noop_fsync,
2280         .splice_read    = shmem_file_splice_read,
2281         .splice_write   = generic_file_splice_write,
2282 #endif
2283 };
2284
2285 static const struct inode_operations shmem_inode_operations = {
2286         .setattr        = shmem_setattr,
2287         .truncate_range = shmem_truncate_range,
2288 #ifdef CONFIG_TMPFS_XATTR
2289         .setxattr       = shmem_setxattr,
2290         .getxattr       = shmem_getxattr,
2291         .listxattr      = shmem_listxattr,
2292         .removexattr    = shmem_removexattr,
2293 #endif
2294 };
2295
2296 static const struct inode_operations shmem_dir_inode_operations = {
2297 #ifdef CONFIG_TMPFS
2298         .create         = shmem_create,
2299         .lookup         = simple_lookup,
2300         .link           = shmem_link,
2301         .unlink         = shmem_unlink,
2302         .symlink        = shmem_symlink,
2303         .mkdir          = shmem_mkdir,
2304         .rmdir          = shmem_rmdir,
2305         .mknod          = shmem_mknod,
2306         .rename         = shmem_rename,
2307 #endif
2308 #ifdef CONFIG_TMPFS_XATTR
2309         .setxattr       = shmem_setxattr,
2310         .getxattr       = shmem_getxattr,
2311         .listxattr      = shmem_listxattr,
2312         .removexattr    = shmem_removexattr,
2313 #endif
2314 #ifdef CONFIG_TMPFS_POSIX_ACL
2315         .setattr        = shmem_setattr,
2316 #endif
2317 };
2318
2319 static const struct inode_operations shmem_special_inode_operations = {
2320 #ifdef CONFIG_TMPFS_XATTR
2321         .setxattr       = shmem_setxattr,
2322         .getxattr       = shmem_getxattr,
2323         .listxattr      = shmem_listxattr,
2324         .removexattr    = shmem_removexattr,
2325 #endif
2326 #ifdef CONFIG_TMPFS_POSIX_ACL
2327         .setattr        = shmem_setattr,
2328 #endif
2329 };
2330
2331 static const struct super_operations shmem_ops = {
2332         .alloc_inode    = shmem_alloc_inode,
2333         .destroy_inode  = shmem_destroy_inode,
2334 #ifdef CONFIG_TMPFS
2335         .statfs         = shmem_statfs,
2336         .remount_fs     = shmem_remount_fs,
2337         .show_options   = shmem_show_options,
2338 #endif
2339         .evict_inode    = shmem_evict_inode,
2340         .drop_inode     = generic_delete_inode,
2341         .put_super      = shmem_put_super,
2342 };
2343
2344 static const struct vm_operations_struct shmem_vm_ops = {
2345         .fault          = shmem_fault,
2346 #ifdef CONFIG_NUMA
2347         .set_policy     = shmem_set_policy,
2348         .get_policy     = shmem_get_policy,
2349 #endif
2350 };
2351
2352 static struct dentry *shmem_mount(struct file_system_type *fs_type,
2353         int flags, const char *dev_name, void *data)
2354 {
2355         return mount_nodev(fs_type, flags, data, shmem_fill_super);
2356 }
2357
2358 static struct file_system_type shmem_fs_type = {
2359         .owner          = THIS_MODULE,
2360         .name           = "tmpfs",
2361         .mount          = shmem_mount,
2362         .kill_sb        = kill_litter_super,
2363 };
2364
2365 int __init shmem_init(void)
2366 {
2367         int error;
2368
2369         error = bdi_init(&shmem_backing_dev_info);
2370         if (error)
2371                 goto out4;
2372
2373         error = shmem_init_inodecache();
2374         if (error)
2375                 goto out3;
2376
2377         error = register_filesystem(&shmem_fs_type);
2378         if (error) {
2379                 printk(KERN_ERR "Could not register tmpfs\n");
2380                 goto out2;
2381         }
2382
2383         shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER,
2384                                  shmem_fs_type.name, NULL);
2385         if (IS_ERR(shm_mnt)) {
2386                 error = PTR_ERR(shm_mnt);
2387                 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2388                 goto out1;
2389         }
2390         return 0;
2391
2392 out1:
2393         unregister_filesystem(&shmem_fs_type);
2394 out2:
2395         shmem_destroy_inodecache();
2396 out3:
2397         bdi_destroy(&shmem_backing_dev_info);
2398 out4:
2399         shm_mnt = ERR_PTR(error);
2400         return error;
2401 }
2402
2403 #else /* !CONFIG_SHMEM */
2404
2405 /*
2406  * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2407  *
2408  * This is intended for small system where the benefits of the full
2409  * shmem code (swap-backed and resource-limited) are outweighed by
2410  * their complexity. On systems without swap this code should be
2411  * effectively equivalent, but much lighter weight.
2412  */
2413
2414 #include <linux/ramfs.h>
2415
2416 static struct file_system_type shmem_fs_type = {
2417         .name           = "tmpfs",
2418         .mount          = ramfs_mount,
2419         .kill_sb        = kill_litter_super,
2420 };
2421
2422 int __init shmem_init(void)
2423 {
2424         BUG_ON(register_filesystem(&shmem_fs_type) != 0);
2425
2426         shm_mnt = kern_mount(&shmem_fs_type);
2427         BUG_ON(IS_ERR(shm_mnt));
2428
2429         return 0;
2430 }
2431
2432 int shmem_unuse(swp_entry_t swap, struct page *page)
2433 {
2434         return 0;
2435 }
2436
2437 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2438 {
2439         return 0;
2440 }
2441
2442 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
2443 {
2444         truncate_inode_pages_range(inode->i_mapping, lstart, lend);
2445 }
2446 EXPORT_SYMBOL_GPL(shmem_truncate_range);
2447
2448 #define shmem_vm_ops                            generic_file_vm_ops
2449 #define shmem_file_operations                   ramfs_file_operations
2450 #define shmem_get_inode(sb, dir, mode, dev, flags)      ramfs_get_inode(sb, dir, mode, dev)
2451 #define shmem_acct_size(flags, size)            0
2452 #define shmem_unacct_size(flags, size)          do {} while (0)
2453
2454 #endif /* CONFIG_SHMEM */
2455
2456 /* common code */
2457
2458 /**
2459  * shmem_file_setup - get an unlinked file living in tmpfs
2460  * @name: name for dentry (to be seen in /proc/<pid>/maps
2461  * @size: size to be set for the file
2462  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2463  */
2464 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2465 {
2466         int error;
2467         struct file *file;
2468         struct inode *inode;
2469         struct path path;
2470         struct dentry *root;
2471         struct qstr this;
2472
2473         if (IS_ERR(shm_mnt))
2474                 return (void *)shm_mnt;
2475
2476         if (size < 0 || size > MAX_LFS_FILESIZE)
2477                 return ERR_PTR(-EINVAL);
2478
2479         if (shmem_acct_size(flags, size))
2480                 return ERR_PTR(-ENOMEM);
2481
2482         error = -ENOMEM;
2483         this.name = name;
2484         this.len = strlen(name);
2485         this.hash = 0; /* will go */
2486         root = shm_mnt->mnt_root;
2487         path.dentry = d_alloc(root, &this);
2488         if (!path.dentry)
2489                 goto put_memory;
2490         path.mnt = mntget(shm_mnt);
2491
2492         error = -ENOSPC;
2493         inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2494         if (!inode)
2495                 goto put_dentry;
2496
2497         d_instantiate(path.dentry, inode);
2498         inode->i_size = size;
2499         clear_nlink(inode);     /* It is unlinked */
2500 #ifndef CONFIG_MMU
2501         error = ramfs_nommu_expand_for_mapping(inode, size);
2502         if (error)
2503                 goto put_dentry;
2504 #endif
2505
2506         error = -ENFILE;
2507         file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2508                   &shmem_file_operations);
2509         if (!file)
2510                 goto put_dentry;
2511
2512         return file;
2513
2514 put_dentry:
2515         path_put(&path);
2516 put_memory:
2517         shmem_unacct_size(flags, size);
2518         return ERR_PTR(error);
2519 }
2520 EXPORT_SYMBOL_GPL(shmem_file_setup);
2521
2522 /**
2523  * shmem_zero_setup - setup a shared anonymous mapping
2524  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2525  */
2526 int shmem_zero_setup(struct vm_area_struct *vma)
2527 {
2528         struct file *file;
2529         loff_t size = vma->vm_end - vma->vm_start;
2530
2531         file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2532         if (IS_ERR(file))
2533                 return PTR_ERR(file);
2534
2535         if (vma->vm_file)
2536                 fput(vma->vm_file);
2537         vma->vm_file = file;
2538         vma->vm_ops = &shmem_vm_ops;
2539         vma->vm_flags |= VM_CAN_NONLINEAR;
2540         return 0;
2541 }
2542
2543 /**
2544  * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2545  * @mapping:    the page's address_space
2546  * @index:      the page index
2547  * @gfp:        the page allocator flags to use if allocating
2548  *
2549  * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2550  * with any new page allocations done using the specified allocation flags.
2551  * But read_cache_page_gfp() uses the ->readpage() method: which does not
2552  * suit tmpfs, since it may have pages in swapcache, and needs to find those
2553  * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
2554  *
2555  * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
2556  * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
2557  */
2558 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
2559                                          pgoff_t index, gfp_t gfp)
2560 {
2561 #ifdef CONFIG_SHMEM
2562         struct inode *inode = mapping->host;
2563         struct page *page;
2564         int error;
2565
2566         BUG_ON(mapping->a_ops != &shmem_aops);
2567         error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
2568         if (error)
2569                 page = ERR_PTR(error);
2570         else
2571                 unlock_page(page);
2572         return page;
2573 #else
2574         /*
2575          * The tiny !SHMEM case uses ramfs without swap
2576          */
2577         return read_cache_page_gfp(mapping, index, gfp);
2578 #endif
2579 }
2580 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);