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