mm/shmem.c

   1 /*
   2  * Resizable virtual memory filesystem for Linux.
   3  *
   4  * Copyright (C) 2000 Linus Torvalds.
   5  *               2000 Transmeta Corp.
   6  *               2000-2001 Christoph Rohland
   7  *               2000-2001 SAP AG
   8  *               2002 Red Hat Inc.
   9  * Copyright (C) 2002-2011 Hugh Dickins.
  10  * Copyright (C) 2011 Google Inc.
  11  * Copyright (C) 2002-2005 VERITAS Software Corporation.
  12  * Copyright (C) 2004 Andi Kleen, SuSE Labs
  13  *
  14  * Extended attribute support for tmpfs:
  15  * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
  16  * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
  17  *
  18  * tiny-shmem:
  19  * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
  20  *
  21  * This file is released under the GPL.
  22  */
  23
  24 #include <linux/fs.h>
  25 #include <linux/init.h>
  26 #include <linux/vfs.h>
  27 #include <linux/mount.h>
  28 #include <linux/pagemap.h>
  29 #include <linux/file.h>
  30 #include <linux/mm.h>
  31 #include <linux/export.h>
  32 #include <linux/swap.h>
  33
  34 static struct vfsmount *shm_mnt;
  35
  36 #ifdef CONFIG_SHMEM
  37 /*
  38  * This virtual memory filesystem is heavily based on the ramfs. It
  39  * extends ramfs by the ability to use swap and honor resource limits
  40  * which makes it a completely usable filesystem.
  41  */
  42
  43 #include <linux/xattr.h>
  44 #include <linux/exportfs.h>
  45 #include <linux/posix_acl.h>
  46 #include <linux/generic_acl.h>
  47 #include <linux/mman.h>
  48 #include <linux/string.h>
  49 #include <linux/slab.h>
  50 #include <linux/backing-dev.h>
  51 #include <linux/shmem_fs.h>
  52 #include <linux/writeback.h>
  53 #include <linux/blkdev.h>
  54 #include <linux/pagevec.h>
  55 #include <linux/percpu_counter.h>
  56 #include <linux/splice.h>
  57 #include <linux/security.h>
  58 #include <linux/swapops.h>
  59 #include <linux/mempolicy.h>
  60 #include <linux/namei.h>
  61 #include <linux/ctype.h>
  62 #include <linux/migrate.h>
  63 #include <linux/highmem.h>
  64 #include <linux/seq_file.h>
  65 #include <linux/magic.h>
  66
  67 #include <asm/uaccess.h>
  68 #include <asm/pgtable.h>
  69
  70 #define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
  71 #define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
  72
  73 /* Pretend that each entry is of this size in directory's i_size */
  74 #define BOGO_DIRENT_SIZE 20
  75
  76 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
  77 #define SHORT_SYMLINK_LEN 128
  78
  79 struct shmem_xattr {
  80         struct list_head list;  /* anchored by shmem_inode_info->xattr_list */
  81         char *name;             /* xattr name */
  82         size_t size;
  83         char value[0];
  84 };
  85
  86 /* Flag allocation requirements to shmem_getpage */
  87 enum sgp_type {
  88         SGP_READ,       /* don't exceed i_size, don't allocate page */
  89         SGP_CACHE,      /* don't exceed i_size, may allocate page */
  90         SGP_DIRTY,      /* like SGP_CACHE, but set new page dirty */
  91         SGP_WRITE,      /* may exceed i_size, may allocate page */
  92 };
  93
  94 #ifdef CONFIG_TMPFS
  95 static unsigned long shmem_default_max_blocks(void)
  96 {
  97         return totalram_pages / 2;
  98 }
  99
 100 static unsigned long shmem_default_max_inodes(void)
 101 {
 102         return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
 103 }
 104 #endif
 105
 106 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
 107         struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
 108
 109 static inline int shmem_getpage(struct inode *inode, pgoff_t index,
 110         struct page **pagep, enum sgp_type sgp, int *fault_type)
 111 {
 112         return shmem_getpage_gfp(inode, index, pagep, sgp,
 113                         mapping_gfp_mask(inode->i_mapping), fault_type);
 114 }
 115
 116 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
 117 {
 118         return sb->s_fs_info;
 119 }
 120
 121 /*
 122  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
 123  * for shared memory and for shared anonymous (/dev/zero) mappings
 124  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
 125  * consistent with the pre-accounting of private mappings ...
 126  */
 127 static inline int shmem_acct_size(unsigned long flags, loff_t size)
 128 {
 129         return (flags & VM_NORESERVE) ?
 130                 0 : security_vm_enough_memory_mm(current->mm, 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_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0;
 149 }
 150
 151 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
 152 {
 153         if (flags & VM_NORESERVE)
 154                 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
 155 }
 156
 157 static const struct super_operations shmem_ops;
 158 static const struct address_space_operations shmem_aops;
 159 static const struct file_operations shmem_file_operations;
 160 static const struct inode_operations shmem_inode_operations;
 161 static const struct inode_operations shmem_dir_inode_operations;
 162 static const struct inode_operations shmem_special_inode_operations;
 163 static const struct vm_operations_struct shmem_vm_ops;
 164
 165 static struct backing_dev_info shmem_backing_dev_info  __read_mostly = {
 166         .ra_pages       = 0,    /* No readahead */
 167         .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
 168 };
 169
 170 static LIST_HEAD(shmem_swaplist);
 171 static DEFINE_MUTEX(shmem_swaplist_mutex);
 172
 173 static int shmem_reserve_inode(struct super_block *sb)
 174 {
 175         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 176         if (sbinfo->max_inodes) {
 177                 spin_lock(&sbinfo->stat_lock);
 178                 if (!sbinfo->free_inodes) {
 179                         spin_unlock(&sbinfo->stat_lock);
 180                         return -ENOSPC;
 181                 }
 182                 sbinfo->free_inodes--;
 183                 spin_unlock(&sbinfo->stat_lock);
 184         }
 185         return 0;
 186 }
 187
 188 static void shmem_free_inode(struct super_block *sb)
 189 {
 190         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
 191         if (sbinfo->max_inodes) {
 192                 spin_lock(&sbinfo->stat_lock);
 193                 sbinfo->free_inodes++;
 194                 spin_unlock(&sbinfo->stat_lock);
 195         }
 196 }
 197
 198 /**
 199  * shmem_recalc_inode - recalculate the block usage of an inode
 200  * @inode: inode to recalc
 201  *
 202  * We have to calculate the free blocks since the mm can drop
 203  * undirtied hole pages behind our back.
 204  *
 205  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
 206  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
 207  *
 208  * It has to be called with the spinlock held.
 209  */
 210 static void shmem_recalc_inode(struct inode *inode)
 211 {
 212         struct shmem_inode_info *info = SHMEM_I(inode);
 213         long freed;
 214
 215         freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
 216         if (freed > 0) {
 217                 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 218                 if (sbinfo->max_blocks)
 219                         percpu_counter_add(&sbinfo->used_blocks, -freed);
 220                 info->alloced -= freed;
 221                 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
 222                 shmem_unacct_blocks(info->flags, freed);
 223         }
 224 }
 225
 226 /*
 227  * Replace item expected in radix tree by a new item, while holding tree lock.
 228  */
 229 static int shmem_radix_tree_replace(struct address_space *mapping,
 230                         pgoff_t index, void *expected, void *replacement)
 231 {
 232         void **pslot;
 233         void *item = NULL;
 234
 235         VM_BUG_ON(!expected);
 236         pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
 237         if (pslot)
 238                 item = radix_tree_deref_slot_protected(pslot,
 239                                                         &mapping->tree_lock);
 240         if (item != expected)
 241                 return -ENOENT;
 242         if (replacement)
 243                 radix_tree_replace_slot(pslot, replacement);
 244         else
 245                 radix_tree_delete(&mapping->page_tree, index);
 246         return 0;
 247 }
 248
 249 /*
 250  * Like add_to_page_cache_locked, but error if expected item has gone.
 251  */
 252 static int shmem_add_to_page_cache(struct page *page,
 253                                    struct address_space *mapping,
 254                                    pgoff_t index, gfp_t gfp, void *expected)
 255 {
 256         int error = 0;
 257
 258         VM_BUG_ON(!PageLocked(page));
 259         VM_BUG_ON(!PageSwapBacked(page));
 260
 261         if (!expected)
 262                 error = radix_tree_preload(gfp & GFP_RECLAIM_MASK);
 263         if (!error) {
 264                 page_cache_get(page);
 265                 page->mapping = mapping;
 266                 page->index = index;
 267
 268                 spin_lock_irq(&mapping->tree_lock);
 269                 if (!expected)
 270                         error = radix_tree_insert(&mapping->page_tree,
 271                                                         index, page);
 272                 else
 273                         error = shmem_radix_tree_replace(mapping, index,
 274                                                         expected, page);
 275                 if (!error) {
 276                         mapping->nrpages++;
 277                         __inc_zone_page_state(page, NR_FILE_PAGES);
 278                         __inc_zone_page_state(page, NR_SHMEM);
 279                         spin_unlock_irq(&mapping->tree_lock);
 280                 } else {
 281                         page->mapping = NULL;
 282                         spin_unlock_irq(&mapping->tree_lock);
 283                         page_cache_release(page);
 284                 }
 285                 if (!expected)
 286                         radix_tree_preload_end();
 287         }
 288         if (error)
 289                 mem_cgroup_uncharge_cache_page(page);
 290         return error;
 291 }
 292
 293 /*
 294  * Like delete_from_page_cache, but substitutes swap for page.
 295  */
 296 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
 297 {
 298         struct address_space *mapping = page->mapping;
 299         int error;
 300
 301         spin_lock_irq(&mapping->tree_lock);
 302         error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
 303         page->mapping = NULL;
 304         mapping->nrpages--;
 305         __dec_zone_page_state(page, NR_FILE_PAGES);
 306         __dec_zone_page_state(page, NR_SHMEM);
 307         spin_unlock_irq(&mapping->tree_lock);
 308         page_cache_release(page);
 309         BUG_ON(error);
 310 }
 311
 312 /*
 313  * Like find_get_pages, but collecting swap entries as well as pages.
 314  */
 315 static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping,
 316                                         pgoff_t start, unsigned int nr_pages,
 317                                         struct page **pages, pgoff_t *indices)
 318 {
 319         unsigned int i;
 320         unsigned int ret;
 321         unsigned int nr_found;
 322
 323         rcu_read_lock();
 324 restart:
 325         nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
 326                                 (void ***)pages, indices, start, nr_pages);
 327         ret = 0;
 328         for (i = 0; i < nr_found; i++) {
 329                 struct page *page;
 330 repeat:
 331                 page = radix_tree_deref_slot((void **)pages[i]);
 332                 if (unlikely(!page))
 333                         continue;
 334                 if (radix_tree_exception(page)) {
 335                         if (radix_tree_deref_retry(page))
 336                                 goto restart;
 337                         /*
 338                          * Otherwise, we must be storing a swap entry
 339                          * here as an exceptional entry: so return it
 340                          * without attempting to raise page count.
 341                          */
 342                         goto export;
 343                 }
 344                 if (!page_cache_get_speculative(page))
 345                         goto repeat;
 346
 347                 /* Has the page moved? */
 348                 if (unlikely(page != *((void **)pages[i]))) {
 349                         page_cache_release(page);
 350                         goto repeat;
 351                 }
 352 export:
 353                 indices[ret] = indices[i];
 354                 pages[ret] = page;
 355                 ret++;
 356         }
 357         if (unlikely(!ret && nr_found))
 358                 goto restart;
 359         rcu_read_unlock();
 360         return ret;
 361 }
 362
 363 /*
 364  * Remove swap entry from radix tree, free the swap and its page cache.
 365  */
 366 static int shmem_free_swap(struct address_space *mapping,
 367                            pgoff_t index, void *radswap)
 368 {
 369         int error;
 370
 371         spin_lock_irq(&mapping->tree_lock);
 372         error = shmem_radix_tree_replace(mapping, index, radswap, NULL);
 373         spin_unlock_irq(&mapping->tree_lock);
 374         if (!error)
 375                 free_swap_and_cache(radix_to_swp_entry(radswap));
 376         return error;
 377 }
 378
 379 /*
 380  * Pagevec may contain swap entries, so shuffle up pages before releasing.
 381  */
 382 static void shmem_deswap_pagevec(struct pagevec *pvec)
 383 {
 384         int i, j;
 385
 386         for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
 387                 struct page *page = pvec->pages[i];
 388                 if (!radix_tree_exceptional_entry(page))
 389                         pvec->pages[j++] = page;
 390         }
 391         pvec->nr = j;
 392 }
 393
 394 /*
 395  * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
 396  */
 397 void shmem_unlock_mapping(struct address_space *mapping)
 398 {
 399         struct pagevec pvec;
 400         pgoff_t indices[PAGEVEC_SIZE];
 401         pgoff_t index = 0;
 402
 403         pagevec_init(&pvec, 0);
 404         /*
 405          * Minor point, but we might as well stop if someone else SHM_LOCKs it.
 406          */
 407         while (!mapping_unevictable(mapping)) {
 408                 /*
 409                  * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
 410                  * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
 411                  */
 412                 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
 413                                         PAGEVEC_SIZE, pvec.pages, indices);
 414                 if (!pvec.nr)
 415                         break;
 416                 index = indices[pvec.nr - 1] + 1;
 417                 shmem_deswap_pagevec(&pvec);
 418                 check_move_unevictable_pages(pvec.pages, pvec.nr);
 419                 pagevec_release(&pvec);
 420                 cond_resched();
 421         }
 422 }
 423
 424 /*
 425  * Remove range of pages and swap entries from radix tree, and free them.
 426  */
 427 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
 428 {
 429         struct address_space *mapping = inode->i_mapping;
 430         struct shmem_inode_info *info = SHMEM_I(inode);
 431         pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 432         unsigned partial = lstart & (PAGE_CACHE_SIZE - 1);
 433         pgoff_t end = (lend >> PAGE_CACHE_SHIFT);
 434         struct pagevec pvec;
 435         pgoff_t indices[PAGEVEC_SIZE];
 436         long nr_swaps_freed = 0;
 437         pgoff_t index;
 438         int i;
 439
 440         BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1));
 441
 442         pagevec_init(&pvec, 0);
 443         index = start;
 444         while (index <= end) {
 445                 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
 446                         min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
 447                                                         pvec.pages, indices);
 448                 if (!pvec.nr)
 449                         break;
 450                 mem_cgroup_uncharge_start();
 451                 for (i = 0; i < pagevec_count(&pvec); i++) {
 452                         struct page *page = pvec.pages[i];
 453
 454                         index = indices[i];
 455                         if (index > end)
 456                                 break;
 457
 458                         if (radix_tree_exceptional_entry(page)) {
 459                                 nr_swaps_freed += !shmem_free_swap(mapping,
 460                                                                 index, page);
 461                                 continue;
 462                         }
 463
 464                         if (!trylock_page(page))
 465                                 continue;
 466                         if (page->mapping == mapping) {
 467                                 VM_BUG_ON(PageWriteback(page));
 468                                 truncate_inode_page(mapping, page);
 469                         }
 470                         unlock_page(page);
 471                 }
 472                 shmem_deswap_pagevec(&pvec);
 473                 pagevec_release(&pvec);
 474                 mem_cgroup_uncharge_end();
 475                 cond_resched();
 476                 index++;
 477         }
 478
 479         if (partial) {
 480                 struct page *page = NULL;
 481                 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
 482                 if (page) {
 483                         zero_user_segment(page, partial, PAGE_CACHE_SIZE);
 484                         set_page_dirty(page);
 485                         unlock_page(page);
 486                         page_cache_release(page);
 487                 }
 488         }
 489
 490         index = start;
 491         for ( ; ; ) {
 492                 cond_resched();
 493                 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
 494                         min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1,
 495                                                         pvec.pages, indices);
 496                 if (!pvec.nr) {
 497                         if (index == start)
 498                                 break;
 499                         index = start;
 500                         continue;
 501                 }
 502                 if (index == start && indices[0] > end) {
 503                         shmem_deswap_pagevec(&pvec);
 504                         pagevec_release(&pvec);
 505                         break;
 506                 }
 507                 mem_cgroup_uncharge_start();
 508                 for (i = 0; i < pagevec_count(&pvec); i++) {
 509                         struct page *page = pvec.pages[i];
 510
 511                         index = indices[i];
 512                         if (index > end)
 513                                 break;
 514
 515                         if (radix_tree_exceptional_entry(page)) {
 516                                 nr_swaps_freed += !shmem_free_swap(mapping,
 517                                                                 index, page);
 518                                 continue;
 519                         }
 520
 521                         lock_page(page);
 522                         if (page->mapping == mapping) {
 523                                 VM_BUG_ON(PageWriteback(page));
 524                                 truncate_inode_page(mapping, page);
 525                         }
 526                         unlock_page(page);
 527                 }
 528                 shmem_deswap_pagevec(&pvec);
 529                 pagevec_release(&pvec);
 530                 mem_cgroup_uncharge_end();
 531                 index++;
 532         }
 533
 534         spin_lock(&info->lock);
 535         info->swapped -= nr_swaps_freed;
 536         shmem_recalc_inode(inode);
 537         spin_unlock(&info->lock);
 538
 539         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 540 }
 541 EXPORT_SYMBOL_GPL(shmem_truncate_range);
 542
 543 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
 544 {
 545         struct inode *inode = dentry->d_inode;
 546         int error;
 547
 548         error = inode_change_ok(inode, attr);
 549         if (error)
 550                 return error;
 551
 552         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
 553                 loff_t oldsize = inode->i_size;
 554                 loff_t newsize = attr->ia_size;
 555
 556                 if (newsize != oldsize) {
 557                         i_size_write(inode, newsize);
 558                         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 559                 }
 560                 if (newsize < oldsize) {
 561                         loff_t holebegin = round_up(newsize, PAGE_SIZE);
 562                         unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
 563                         shmem_truncate_range(inode, newsize, (loff_t)-1);
 564                         /* unmap again to remove racily COWed private pages */
 565                         unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
 566                 }
 567         }
 568
 569         setattr_copy(inode, attr);
 570 #ifdef CONFIG_TMPFS_POSIX_ACL
 571         if (attr->ia_valid & ATTR_MODE)
 572                 error = generic_acl_chmod(inode);
 573 #endif
 574         return error;
 575 }
 576
 577 static void shmem_evict_inode(struct inode *inode)
 578 {
 579         struct shmem_inode_info *info = SHMEM_I(inode);
 580         struct shmem_xattr *xattr, *nxattr;
 581
 582         if (inode->i_mapping->a_ops == &shmem_aops) {
 583                 shmem_unacct_size(info->flags, inode->i_size);
 584                 inode->i_size = 0;
 585                 shmem_truncate_range(inode, 0, (loff_t)-1);
 586                 if (!list_empty(&info->swaplist)) {
 587                         mutex_lock(&shmem_swaplist_mutex);
 588                         list_del_init(&info->swaplist);
 589                         mutex_unlock(&shmem_swaplist_mutex);
 590                 }
 591         } else
 592                 kfree(info->symlink);
 593
 594         list_for_each_entry_safe(xattr, nxattr, &info->xattr_list, list) {
 595                 kfree(xattr->name);
 596                 kfree(xattr);
 597         }
 598         BUG_ON(inode->i_blocks);
 599         shmem_free_inode(inode->i_sb);
 600         end_writeback(inode);
 601 }
 602
 603 /*
 604  * If swap found in inode, free it and move page from swapcache to filecache.
 605  */
 606 static int shmem_unuse_inode(struct shmem_inode_info *info,
 607                              swp_entry_t swap, struct page *page)
 608 {
 609         struct address_space *mapping = info->vfs_inode.i_mapping;
 610         void *radswap;
 611         pgoff_t index;
 612         int error;
 613
 614         radswap = swp_to_radix_entry(swap);
 615         index = radix_tree_locate_item(&mapping->page_tree, radswap);
 616         if (index == -1)
 617                 return 0;
 618
 619         /*
 620          * Move _head_ to start search for next from here.
 621          * But be careful: shmem_evict_inode checks list_empty without taking
 622          * mutex, and there's an instant in list_move_tail when info->swaplist
 623          * would appear empty, if it were the only one on shmem_swaplist.
 624          */
 625         if (shmem_swaplist.next != &info->swaplist)
 626                 list_move_tail(&shmem_swaplist, &info->swaplist);
 627
 628         /*
 629          * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
 630          * but also to hold up shmem_evict_inode(): so inode cannot be freed
 631          * beneath us (pagelock doesn't help until the page is in pagecache).
 632          */
 633         error = shmem_add_to_page_cache(page, mapping, index,
 634                                                 GFP_NOWAIT, radswap);
 635         /* which does mem_cgroup_uncharge_cache_page on error */
 636
 637         if (error != -ENOMEM) {
 638                 /*
 639                  * Truncation and eviction use free_swap_and_cache(), which
 640                  * only does trylock page: if we raced, best clean up here.
 641                  */
 642                 delete_from_swap_cache(page);
 643                 set_page_dirty(page);
 644                 if (!error) {
 645                         spin_lock(&info->lock);
 646                         info->swapped--;
 647                         spin_unlock(&info->lock);
 648                         swap_free(swap);
 649                 }
 650                 error = 1;      /* not an error, but entry was found */
 651         }
 652         return error;
 653 }
 654
 655 /*
 656  * Search through swapped inodes to find and replace swap by page.
 657  */
 658 int shmem_unuse(swp_entry_t swap, struct page *page)
 659 {
 660         struct list_head *this, *next;
 661         struct shmem_inode_info *info;
 662         int found = 0;
 663         int error;
 664
 665         /*
 666          * Charge page using GFP_KERNEL while we can wait, before taking
 667          * the shmem_swaplist_mutex which might hold up shmem_writepage().
 668          * Charged back to the user (not to caller) when swap account is used.
 669          */
 670         error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
 671         if (error)
 672                 goto out;
 673         /* No radix_tree_preload: swap entry keeps a place for page in tree */
 674
 675         mutex_lock(&shmem_swaplist_mutex);
 676         list_for_each_safe(this, next, &shmem_swaplist) {
 677                 info = list_entry(this, struct shmem_inode_info, swaplist);
 678                 if (info->swapped)
 679                         found = shmem_unuse_inode(info, swap, page);
 680                 else
 681                         list_del_init(&info->swaplist);
 682                 cond_resched();
 683                 if (found)
 684                         break;
 685         }
 686         mutex_unlock(&shmem_swaplist_mutex);
 687
 688         if (!found)
 689                 mem_cgroup_uncharge_cache_page(page);
 690         if (found < 0)
 691                 error = found;
 692 out:
 693         unlock_page(page);
 694         page_cache_release(page);
 695         return error;
 696 }
 697
 698 /*
 699  * Move the page from the page cache to the swap cache.
 700  */
 701 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
 702 {
 703         struct shmem_inode_info *info;
 704         struct address_space *mapping;
 705         struct inode *inode;
 706         swp_entry_t swap;
 707         pgoff_t index;
 708
 709         BUG_ON(!PageLocked(page));
 710         mapping = page->mapping;
 711         index = page->index;
 712         inode = mapping->host;
 713         info = SHMEM_I(inode);
 714         if (info->flags & VM_LOCKED)
 715                 goto redirty;
 716         if (!total_swap_pages)
 717                 goto redirty;
 718
 719         /*
 720          * shmem_backing_dev_info's capabilities prevent regular writeback or
 721          * sync from ever calling shmem_writepage; but a stacking filesystem
 722          * might use ->writepage of its underlying filesystem, in which case
 723          * tmpfs should write out to swap only in response to memory pressure,
 724          * and not for the writeback threads or sync.
 725          */
 726         if (!wbc->for_reclaim) {
 727                 WARN_ON_ONCE(1);        /* Still happens? Tell us about it! */
 728                 goto redirty;
 729         }
 730         swap = get_swap_page();
 731         if (!swap.val)
 732                 goto redirty;
 733
 734         /*
 735          * Add inode to shmem_unuse()'s list of swapped-out inodes,
 736          * if it's not already there.  Do it now before the page is
 737          * moved to swap cache, when its pagelock no longer protects
 738          * the inode from eviction.  But don't unlock the mutex until
 739          * we've incremented swapped, because shmem_unuse_inode() will
 740          * prune a !swapped inode from the swaplist under this mutex.
 741          */
 742         mutex_lock(&shmem_swaplist_mutex);
 743         if (list_empty(&info->swaplist))
 744                 list_add_tail(&info->swaplist, &shmem_swaplist);
 745
 746         if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
 747                 swap_shmem_alloc(swap);
 748                 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
 749
 750                 spin_lock(&info->lock);
 751                 info->swapped++;
 752                 shmem_recalc_inode(inode);
 753                 spin_unlock(&info->lock);
 754
 755                 mutex_unlock(&shmem_swaplist_mutex);
 756                 BUG_ON(page_mapped(page));
 757                 swap_writepage(page, wbc);
 758                 return 0;
 759         }
 760
 761         mutex_unlock(&shmem_swaplist_mutex);
 762         swapcache_free(swap, NULL);
 763 redirty:
 764         set_page_dirty(page);
 765         if (wbc->for_reclaim)
 766                 return AOP_WRITEPAGE_ACTIVATE;  /* Return with page locked */
 767         unlock_page(page);
 768         return 0;
 769 }
 770
 771 #ifdef CONFIG_NUMA
 772 #ifdef CONFIG_TMPFS
 773 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
 774 {
 775         char buffer[64];
 776
 777         if (!mpol || mpol->mode == MPOL_DEFAULT)
 778                 return;         /* show nothing */
 779
 780         mpol_to_str(buffer, sizeof(buffer), mpol, 1);
 781
 782         seq_printf(seq, ",mpol=%s", buffer);
 783 }
 784
 785 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
 786 {
 787         struct mempolicy *mpol = NULL;
 788         if (sbinfo->mpol) {
 789                 spin_lock(&sbinfo->stat_lock);  /* prevent replace/use races */
 790                 mpol = sbinfo->mpol;
 791                 mpol_get(mpol);
 792                 spin_unlock(&sbinfo->stat_lock);
 793         }
 794         return mpol;
 795 }
 796 #endif /* CONFIG_TMPFS */
 797
 798 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
 799                         struct shmem_inode_info *info, pgoff_t index)
 800 {
 801         struct mempolicy mpol, *spol;
 802         struct vm_area_struct pvma;
 803
 804         spol = mpol_cond_copy(&mpol,
 805                         mpol_shared_policy_lookup(&info->policy, index));
 806
 807         /* Create a pseudo vma that just contains the policy */
 808         pvma.vm_start = 0;
 809         pvma.vm_pgoff = index;
 810         pvma.vm_ops = NULL;
 811         pvma.vm_policy = spol;
 812         return swapin_readahead(swap, gfp, &pvma, 0);
 813 }
 814
 815 static struct page *shmem_alloc_page(gfp_t gfp,
 816                         struct shmem_inode_info *info, pgoff_t index)
 817 {
 818         struct vm_area_struct pvma;
 819
 820         /* Create a pseudo vma that just contains the policy */
 821         pvma.vm_start = 0;
 822         pvma.vm_pgoff = index;
 823         pvma.vm_ops = NULL;
 824         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
 825
 826         /*
 827          * alloc_page_vma() will drop the shared policy reference
 828          */
 829         return alloc_page_vma(gfp, &pvma, 0);
 830 }
 831 #else /* !CONFIG_NUMA */
 832 #ifdef CONFIG_TMPFS
 833 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
 834 {
 835 }
 836 #endif /* CONFIG_TMPFS */
 837
 838 static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
 839                         struct shmem_inode_info *info, pgoff_t index)
 840 {
 841         return swapin_readahead(swap, gfp, NULL, 0);
 842 }
 843
 844 static inline struct page *shmem_alloc_page(gfp_t gfp,
 845                         struct shmem_inode_info *info, pgoff_t index)
 846 {
 847         return alloc_page(gfp);
 848 }
 849 #endif /* CONFIG_NUMA */
 850
 851 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
 852 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
 853 {
 854         return NULL;
 855 }
 856 #endif
 857
 858 /*
 859  * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
 860  *
 861  * If we allocate a new one we do not mark it dirty. That's up to the
 862  * vm. If we swap it in we mark it dirty since we also free the swap
 863  * entry since a page cannot live in both the swap and page cache
 864  */
 865 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
 866         struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
 867 {
 868         struct address_space *mapping = inode->i_mapping;
 869         struct shmem_inode_info *info;
 870         struct shmem_sb_info *sbinfo;
 871         struct page *page;
 872         swp_entry_t swap;
 873         int error;
 874         int once = 0;
 875
 876         if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
 877                 return -EFBIG;
 878 repeat:
 879         swap.val = 0;
 880         page = find_lock_page(mapping, index);
 881         if (radix_tree_exceptional_entry(page)) {
 882                 swap = radix_to_swp_entry(page);
 883                 page = NULL;
 884         }
 885
 886         if (sgp != SGP_WRITE &&
 887             ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
 888                 error = -EINVAL;
 889                 goto failed;
 890         }
 891
 892         if (page || (sgp == SGP_READ && !swap.val)) {
 893                 /*
 894                  * Once we can get the page lock, it must be uptodate:
 895                  * if there were an error in reading back from swap,
 896                  * the page would not be inserted into the filecache.
 897                  */
 898                 BUG_ON(page && !PageUptodate(page));
 899                 *pagep = page;
 900                 return 0;
 901         }
 902
 903         /*
 904          * Fast cache lookup did not find it:
 905          * bring it back from swap or allocate.
 906          */
 907         info = SHMEM_I(inode);
 908         sbinfo = SHMEM_SB(inode->i_sb);
 909
 910         if (swap.val) {
 911                 /* Look it up and read it in.. */
 912                 page = lookup_swap_cache(swap);
 913                 if (!page) {
 914                         /* here we actually do the io */
 915                         if (fault_type)
 916                                 *fault_type |= VM_FAULT_MAJOR;
 917                         page = shmem_swapin(swap, gfp, info, index);
 918                         if (!page) {
 919                                 error = -ENOMEM;
 920                                 goto failed;
 921                         }
 922                 }
 923
 924                 /* We have to do this with page locked to prevent races */
 925                 lock_page(page);
 926                 if (!PageUptodate(page)) {
 927                         error = -EIO;
 928                         goto failed;
 929                 }
 930                 wait_on_page_writeback(page);
 931
 932                 /* Someone may have already done it for us */
 933                 if (page->mapping) {
 934                         if (page->mapping == mapping &&
 935                             page->index == index)
 936                                 goto done;
 937                         error = -EEXIST;
 938                         goto failed;
 939                 }
 940
 941                 error = mem_cgroup_cache_charge(page, current->mm,
 942                                                 gfp & GFP_RECLAIM_MASK);
 943                 if (!error)
 944                         error = shmem_add_to_page_cache(page, mapping, index,
 945                                                 gfp, swp_to_radix_entry(swap));
 946                 if (error)
 947                         goto failed;
 948
 949                 spin_lock(&info->lock);
 950                 info->swapped--;
 951                 shmem_recalc_inode(inode);
 952                 spin_unlock(&info->lock);
 953
 954                 delete_from_swap_cache(page);
 955                 set_page_dirty(page);
 956                 swap_free(swap);
 957
 958         } else {
 959                 if (shmem_acct_block(info->flags)) {
 960                         error = -ENOSPC;
 961                         goto failed;
 962                 }
 963                 if (sbinfo->max_blocks) {
 964                         if (percpu_counter_compare(&sbinfo->used_blocks,
 965                                                 sbinfo->max_blocks) >= 0) {
 966                                 error = -ENOSPC;
 967                                 goto unacct;
 968                         }
 969                         percpu_counter_inc(&sbinfo->used_blocks);
 970                 }
 971
 972                 page = shmem_alloc_page(gfp, info, index);
 973                 if (!page) {
 974                         error = -ENOMEM;
 975                         goto decused;
 976                 }
 977
 978                 SetPageSwapBacked(page);
 979                 __set_page_locked(page);
 980                 error = mem_cgroup_cache_charge(page, current->mm,
 981                                                 gfp & GFP_RECLAIM_MASK);
 982                 if (!error)
 983                         error = shmem_add_to_page_cache(page, mapping, index,
 984                                                 gfp, NULL);
 985                 if (error)
 986                         goto decused;
 987                 lru_cache_add_anon(page);
 988
 989                 spin_lock(&info->lock);
 990                 info->alloced++;
 991                 inode->i_blocks += BLOCKS_PER_PAGE;
 992                 shmem_recalc_inode(inode);
 993                 spin_unlock(&info->lock);
 994
 995                 clear_highpage(page);
 996                 flush_dcache_page(page);
 997                 SetPageUptodate(page);
 998                 if (sgp == SGP_DIRTY)
 999                         set_page_dirty(page);
1000         }
1001 done:
1002         /* Perhaps the file has been truncated since we checked */
1003         if (sgp != SGP_WRITE &&
1004             ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1005                 error = -EINVAL;
1006                 goto trunc;
1007         }
1008         *pagep = page;
1009         return 0;
1010
1011         /*
1012          * Error recovery.
1013          */
1014 trunc:
1015         ClearPageDirty(page);
1016         delete_from_page_cache(page);
1017         spin_lock(&info->lock);
1018         info->alloced--;
1019         inode->i_blocks -= BLOCKS_PER_PAGE;
1020         spin_unlock(&info->lock);
1021 decused:
1022         if (sbinfo->max_blocks)
1023                 percpu_counter_add(&sbinfo->used_blocks, -1);
1024 unacct:
1025         shmem_unacct_blocks(info->flags, 1);
1026 failed:
1027         if (swap.val && error != -EINVAL) {
1028                 struct page *test = find_get_page(mapping, index);
1029                 if (test && !radix_tree_exceptional_entry(test))
1030                         page_cache_release(test);
1031                 /* Have another try if the entry has changed */
1032                 if (test != swp_to_radix_entry(swap))
1033                         error = -EEXIST;
1034         }
1035         if (page) {
1036                 unlock_page(page);
1037                 page_cache_release(page);
1038         }
1039         if (error == -ENOSPC && !once++) {
1040                 info = SHMEM_I(inode);
1041                 spin_lock(&info->lock);
1042                 shmem_recalc_inode(inode);
1043                 spin_unlock(&info->lock);
1044                 goto repeat;
1045         }
1046         if (error == -EEXIST)
1047                 goto repeat;
1048         return error;
1049 }
1050
1051 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1052 {
1053         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1054         int error;
1055         int ret = VM_FAULT_LOCKED;
1056
1057         error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1058         if (error)
1059                 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1060
1061         if (ret & VM_FAULT_MAJOR) {
1062                 count_vm_event(PGMAJFAULT);
1063                 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1064         }
1065         return ret;
1066 }
1067
1068 #ifdef CONFIG_NUMA
1069 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
1070 {
1071         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1072         return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
1073 }
1074
1075 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1076                                           unsigned long addr)
1077 {
1078         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1079         pgoff_t index;
1080
1081         index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1082         return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
1083 }
1084 #endif
1085
1086 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1087 {
1088         struct inode *inode = file->f_path.dentry->d_inode;
1089         struct shmem_inode_info *info = SHMEM_I(inode);
1090         int retval = -ENOMEM;
1091
1092         spin_lock(&info->lock);
1093         if (lock && !(info->flags & VM_LOCKED)) {
1094                 if (!user_shm_lock(inode->i_size, user))
1095                         goto out_nomem;
1096                 info->flags |= VM_LOCKED;
1097                 mapping_set_unevictable(file->f_mapping);
1098         }
1099         if (!lock && (info->flags & VM_LOCKED) && user) {
1100                 user_shm_unlock(inode->i_size, user);
1101                 info->flags &= ~VM_LOCKED;
1102                 mapping_clear_unevictable(file->f_mapping);
1103         }
1104         retval = 0;
1105
1106 out_nomem:
1107         spin_unlock(&info->lock);
1108         return retval;
1109 }
1110
1111 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1112 {
1113         file_accessed(file);
1114         vma->vm_ops = &shmem_vm_ops;
1115         vma->vm_flags |= VM_CAN_NONLINEAR;
1116         return 0;
1117 }
1118
1119 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1120                                      umode_t mode, dev_t dev, unsigned long flags)
1121 {
1122         struct inode *inode;
1123         struct shmem_inode_info *info;
1124         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1125
1126         if (shmem_reserve_inode(sb))
1127                 return NULL;
1128
1129         inode = new_inode(sb);
1130         if (inode) {
1131                 inode->i_ino = get_next_ino();
1132                 inode_init_owner(inode, dir, mode);
1133                 inode->i_blocks = 0;
1134                 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1135                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1136                 inode->i_generation = get_seconds();
1137                 info = SHMEM_I(inode);
1138                 memset(info, 0, (char *)inode - (char *)info);
1139                 spin_lock_init(&info->lock);
1140                 info->flags = flags & VM_NORESERVE;
1141                 INIT_LIST_HEAD(&info->swaplist);
1142                 INIT_LIST_HEAD(&info->xattr_list);
1143                 cache_no_acl(inode);
1144
1145                 switch (mode & S_IFMT) {
1146                 default:
1147                         inode->i_op = &shmem_special_inode_operations;
1148                         init_special_inode(inode, mode, dev);
1149                         break;
1150                 case S_IFREG:
1151                         inode->i_mapping->a_ops = &shmem_aops;
1152                         inode->i_op = &shmem_inode_operations;
1153                         inode->i_fop = &shmem_file_operations;
1154                         mpol_shared_policy_init(&info->policy,
1155                                                  shmem_get_sbmpol(sbinfo));
1156                         break;
1157                 case S_IFDIR:
1158                         inc_nlink(inode);
1159                         /* Some things misbehave if size == 0 on a directory */
1160                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
1161                         inode->i_op = &shmem_dir_inode_operations;
1162                         inode->i_fop = &simple_dir_operations;
1163                         break;
1164                 case S_IFLNK:
1165                         /*
1166                          * Must not load anything in the rbtree,
1167                          * mpol_free_shared_policy will not be called.
1168                          */
1169                         mpol_shared_policy_init(&info->policy, NULL);
1170                         break;
1171                 }
1172         } else
1173                 shmem_free_inode(sb);
1174         return inode;
1175 }
1176
1177 #ifdef CONFIG_TMPFS
1178 static const struct inode_operations shmem_symlink_inode_operations;
1179 static const struct inode_operations shmem_short_symlink_operations;
1180
1181 static int
1182 shmem_write_begin(struct file *file, struct address_space *mapping,
1183                         loff_t pos, unsigned len, unsigned flags,
1184                         struct page **pagep, void **fsdata)
1185 {
1186         struct inode *inode = mapping->host;
1187         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1188         return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1189 }
1190
1191 static int
1192 shmem_write_end(struct file *file, struct address_space *mapping,
1193                         loff_t pos, unsigned len, unsigned copied,
1194                         struct page *page, void *fsdata)
1195 {
1196         struct inode *inode = mapping->host;
1197
1198         if (pos + copied > inode->i_size)
1199                 i_size_write(inode, pos + copied);
1200
1201         set_page_dirty(page);
1202         unlock_page(page);
1203         page_cache_release(page);
1204
1205         return copied;
1206 }
1207
1208 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1209 {
1210         struct inode *inode = filp->f_path.dentry->d_inode;
1211         struct address_space *mapping = inode->i_mapping;
1212         pgoff_t index;
1213         unsigned long offset;
1214         enum sgp_type sgp = SGP_READ;
1215
1216         /*
1217          * Might this read be for a stacking filesystem?  Then when reading
1218          * holes of a sparse file, we actually need to allocate those pages,
1219          * and even mark them dirty, so it cannot exceed the max_blocks limit.
1220          */
1221         if (segment_eq(get_fs(), KERNEL_DS))
1222                 sgp = SGP_DIRTY;
1223
1224         index = *ppos >> PAGE_CACHE_SHIFT;
1225         offset = *ppos & ~PAGE_CACHE_MASK;
1226
1227         for (;;) {
1228                 struct page *page = NULL;
1229                 pgoff_t end_index;
1230                 unsigned long nr, ret;
1231                 loff_t i_size = i_size_read(inode);
1232
1233                 end_index = i_size >> PAGE_CACHE_SHIFT;
1234                 if (index > end_index)
1235                         break;
1236                 if (index == end_index) {
1237                         nr = i_size & ~PAGE_CACHE_MASK;
1238                         if (nr <= offset)
1239                                 break;
1240                 }
1241
1242                 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1243                 if (desc->error) {
1244                         if (desc->error == -EINVAL)
1245                                 desc->error = 0;
1246                         break;
1247                 }
1248                 if (page)
1249                         unlock_page(page);
1250
1251                 /*
1252                  * We must evaluate after, since reads (unlike writes)
1253                  * are called without i_mutex protection against truncate
1254                  */
1255                 nr = PAGE_CACHE_SIZE;
1256                 i_size = i_size_read(inode);
1257                 end_index = i_size >> PAGE_CACHE_SHIFT;
1258                 if (index == end_index) {
1259                         nr = i_size & ~PAGE_CACHE_MASK;
1260                         if (nr <= offset) {
1261                                 if (page)
1262                                         page_cache_release(page);
1263                                 break;
1264                         }
1265                 }
1266                 nr -= offset;
1267
1268                 if (page) {
1269                         /*
1270                          * If users can be writing to this page using arbitrary
1271                          * virtual addresses, take care about potential aliasing
1272                          * before reading the page on the kernel side.
1273                          */
1274                         if (mapping_writably_mapped(mapping))
1275                                 flush_dcache_page(page);
1276                         /*
1277                          * Mark the page accessed if we read the beginning.
1278                          */
1279                         if (!offset)
1280                                 mark_page_accessed(page);
1281                 } else {
1282                         page = ZERO_PAGE(0);
1283                         page_cache_get(page);
1284                 }
1285
1286                 /*
1287                  * Ok, we have the page, and it's up-to-date, so
1288                  * now we can copy it to user space...
1289                  *
1290                  * The actor routine returns how many bytes were actually used..
1291                  * NOTE! This may not be the same as how much of a user buffer
1292                  * we filled up (we may be padding etc), so we can only update
1293                  * "pos" here (the actor routine has to update the user buffer
1294                  * pointers and the remaining count).
1295                  */
1296                 ret = actor(desc, page, offset, nr);
1297                 offset += ret;
1298                 index += offset >> PAGE_CACHE_SHIFT;
1299                 offset &= ~PAGE_CACHE_MASK;
1300
1301                 page_cache_release(page);
1302                 if (ret != nr || !desc->count)
1303                         break;
1304
1305                 cond_resched();
1306         }
1307
1308         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1309         file_accessed(filp);
1310 }
1311
1312 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1313                 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1314 {
1315         struct file *filp = iocb->ki_filp;
1316         ssize_t retval;
1317         unsigned long seg;
1318         size_t count;
1319         loff_t *ppos = &iocb->ki_pos;
1320
1321         retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1322         if (retval)
1323                 return retval;
1324
1325         for (seg = 0; seg < nr_segs; seg++) {
1326                 read_descriptor_t desc;
1327
1328                 desc.written = 0;
1329                 desc.arg.buf = iov[seg].iov_base;
1330                 desc.count = iov[seg].iov_len;
1331                 if (desc.count == 0)
1332                         continue;
1333                 desc.error = 0;
1334                 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1335                 retval += desc.written;
1336                 if (desc.error) {
1337                         retval = retval ?: desc.error;
1338                         break;
1339                 }
1340                 if (desc.count > 0)
1341                         break;
1342         }
1343         return retval;
1344 }
1345
1346 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1347                                 struct pipe_inode_info *pipe, size_t len,
1348                                 unsigned int flags)
1349 {
1350         struct address_space *mapping = in->f_mapping;
1351         struct inode *inode = mapping->host;
1352         unsigned int loff, nr_pages, req_pages;
1353         struct page *pages[PIPE_DEF_BUFFERS];
1354         struct partial_page partial[PIPE_DEF_BUFFERS];
1355         struct page *page;
1356         pgoff_t index, end_index;
1357         loff_t isize, left;
1358         int error, page_nr;
1359         struct splice_pipe_desc spd = {
1360                 .pages = pages,
1361                 .partial = partial,
1362                 .flags = flags,
1363                 .ops = &page_cache_pipe_buf_ops,
1364                 .spd_release = spd_release_page,
1365         };
1366
1367         isize = i_size_read(inode);
1368         if (unlikely(*ppos >= isize))
1369                 return 0;
1370
1371         left = isize - *ppos;
1372         if (unlikely(left < len))
1373                 len = left;
1374
1375         if (splice_grow_spd(pipe, &spd))
1376                 return -ENOMEM;
1377
1378         index = *ppos >> PAGE_CACHE_SHIFT;
1379         loff = *ppos & ~PAGE_CACHE_MASK;
1380         req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1381         nr_pages = min(req_pages, pipe->buffers);
1382
1383         spd.nr_pages = find_get_pages_contig(mapping, index,
1384                                                 nr_pages, spd.pages);
1385         index += spd.nr_pages;
1386         error = 0;
1387
1388         while (spd.nr_pages < nr_pages) {
1389                 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1390                 if (error)
1391                         break;
1392                 unlock_page(page);
1393                 spd.pages[spd.nr_pages++] = page;
1394                 index++;
1395         }
1396
1397         index = *ppos >> PAGE_CACHE_SHIFT;
1398         nr_pages = spd.nr_pages;
1399         spd.nr_pages = 0;
1400
1401         for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1402                 unsigned int this_len;
1403
1404                 if (!len)
1405                         break;
1406
1407                 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1408                 page = spd.pages[page_nr];
1409
1410                 if (!PageUptodate(page) || page->mapping != mapping) {
1411                         error = shmem_getpage(inode, index, &page,
1412                                                         SGP_CACHE, NULL);
1413                         if (error)
1414                                 break;
1415                         unlock_page(page);
1416                         page_cache_release(spd.pages[page_nr]);
1417                         spd.pages[page_nr] = page;
1418                 }
1419
1420                 isize = i_size_read(inode);
1421                 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1422                 if (unlikely(!isize || index > end_index))
1423                         break;
1424
1425                 if (end_index == index) {
1426                         unsigned int plen;
1427
1428                         plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1429                         if (plen <= loff)
1430                                 break;
1431
1432                         this_len = min(this_len, plen - loff);
1433                         len = this_len;
1434                 }
1435
1436                 spd.partial[page_nr].offset = loff;
1437                 spd.partial[page_nr].len = this_len;
1438                 len -= this_len;
1439                 loff = 0;
1440                 spd.nr_pages++;
1441                 index++;
1442         }
1443
1444         while (page_nr < nr_pages)
1445                 page_cache_release(spd.pages[page_nr++]);
1446
1447         if (spd.nr_pages)
1448                 error = splice_to_pipe(pipe, &spd);
1449
1450         splice_shrink_spd(pipe, &spd);
1451
1452         if (error > 0) {
1453                 *ppos += error;
1454                 file_accessed(in);
1455         }
1456         return error;
1457 }
1458
1459 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1460 {
1461         struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1462
1463         buf->f_type = TMPFS_MAGIC;
1464         buf->f_bsize = PAGE_CACHE_SIZE;
1465         buf->f_namelen = NAME_MAX;
1466         if (sbinfo->max_blocks) {
1467                 buf->f_blocks = sbinfo->max_blocks;
1468                 buf->f_bavail =
1469                 buf->f_bfree  = sbinfo->max_blocks -
1470                                 percpu_counter_sum(&sbinfo->used_blocks);
1471         }
1472         if (sbinfo->max_inodes) {
1473                 buf->f_files = sbinfo->max_inodes;
1474                 buf->f_ffree = sbinfo->free_inodes;
1475         }
1476         /* else leave those fields 0 like simple_statfs */
1477         return 0;
1478 }
1479
1480 /*
1481  * File creation. Allocate an inode, and we're done..
1482  */
1483 static int
1484 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
1485 {
1486         struct inode *inode;
1487         int error = -ENOSPC;
1488
1489         inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1490         if (inode) {
1491                 error = security_inode_init_security(inode, dir,
1492                                                      &dentry->d_name,
1493                                                      NULL, NULL);
1494                 if (error) {
1495                         if (error != -EOPNOTSUPP) {
1496                                 iput(inode);
1497                                 return error;
1498                         }
1499                 }
1500 #ifdef CONFIG_TMPFS_POSIX_ACL
1501                 error = generic_acl_init(inode, dir);
1502                 if (error) {
1503                         iput(inode);
1504                         return error;
1505                 }
1506 #else
1507                 error = 0;
1508 #endif
1509                 dir->i_size += BOGO_DIRENT_SIZE;
1510                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1511                 d_instantiate(dentry, inode);
1512                 dget(dentry); /* Extra count - pin the dentry in core */
1513         }
1514         return error;
1515 }
1516
1517 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1518 {
1519         int error;
1520
1521         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1522                 return error;
1523         inc_nlink(dir);
1524         return 0;
1525 }
1526
1527 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1528                 struct nameidata *nd)
1529 {
1530         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1531 }
1532
1533 /*
1534  * Link a file..
1535  */
1536 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1537 {
1538         struct inode *inode = old_dentry->d_inode;
1539         int ret;
1540
1541         /*
1542          * No ordinary (disk based) filesystem counts links as inodes;
1543          * but each new link needs a new dentry, pinning lowmem, and
1544          * tmpfs dentries cannot be pruned until they are unlinked.
1545          */
1546         ret = shmem_reserve_inode(inode->i_sb);
1547         if (ret)
1548                 goto out;
1549
1550         dir->i_size += BOGO_DIRENT_SIZE;
1551         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1552         inc_nlink(inode);
1553         ihold(inode);   /* New dentry reference */
1554         dget(dentry);           /* Extra pinning count for the created dentry */
1555         d_instantiate(dentry, inode);
1556 out:
1557         return ret;
1558 }
1559
1560 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1561 {
1562         struct inode *inode = dentry->d_inode;
1563
1564         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1565                 shmem_free_inode(inode->i_sb);
1566
1567         dir->i_size -= BOGO_DIRENT_SIZE;
1568         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1569         drop_nlink(inode);
1570         dput(dentry);   /* Undo the count from "create" - this does all the work */
1571         return 0;
1572 }
1573
1574 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1575 {
1576         if (!simple_empty(dentry))
1577                 return -ENOTEMPTY;
1578
1579         drop_nlink(dentry->d_inode);
1580         drop_nlink(dir);
1581         return shmem_unlink(dir, dentry);
1582 }
1583
1584 /*
1585  * The VFS layer already does all the dentry stuff for rename,
1586  * we just have to decrement the usage count for the target if
1587  * it exists so that the VFS layer correctly free's it when it
1588  * gets overwritten.
1589  */
1590 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1591 {
1592         struct inode *inode = old_dentry->d_inode;
1593         int they_are_dirs = S_ISDIR(inode->i_mode);
1594
1595         if (!simple_empty(new_dentry))
1596                 return -ENOTEMPTY;
1597
1598         if (new_dentry->d_inode) {
1599                 (void) shmem_unlink(new_dir, new_dentry);
1600                 if (they_are_dirs)
1601                         drop_nlink(old_dir);
1602         } else if (they_are_dirs) {
1603                 drop_nlink(old_dir);
1604                 inc_nlink(new_dir);
1605         }
1606
1607         old_dir->i_size -= BOGO_DIRENT_SIZE;
1608         new_dir->i_size += BOGO_DIRENT_SIZE;
1609         old_dir->i_ctime = old_dir->i_mtime =
1610         new_dir->i_ctime = new_dir->i_mtime =
1611         inode->i_ctime = CURRENT_TIME;
1612         return 0;
1613 }
1614
1615 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1616 {
1617         int error;
1618         int len;
1619         struct inode *inode;
1620         struct page *page;
1621         char *kaddr;
1622         struct shmem_inode_info *info;
1623
1624         len = strlen(symname) + 1;
1625         if (len > PAGE_CACHE_SIZE)
1626                 return -ENAMETOOLONG;
1627
1628         inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
1629         if (!inode)
1630                 return -ENOSPC;
1631
1632         error = security_inode_init_security(inode, dir, &dentry->d_name,
1633                                              NULL, NULL);
1634         if (error) {
1635                 if (error != -EOPNOTSUPP) {
1636                         iput(inode);
1637                         return error;
1638                 }
1639                 error = 0;
1640         }
1641
1642         info = SHMEM_I(inode);
1643         inode->i_size = len-1;
1644         if (len <= SHORT_SYMLINK_LEN) {
1645                 info->symlink = kmemdup(symname, len, GFP_KERNEL);
1646                 if (!info->symlink) {
1647                         iput(inode);
1648                         return -ENOMEM;
1649                 }
1650                 inode->i_op = &shmem_short_symlink_operations;
1651         } else {
1652                 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1653                 if (error) {
1654                         iput(inode);
1655                         return error;
1656                 }
1657                 inode->i_mapping->a_ops = &shmem_aops;
1658                 inode->i_op = &shmem_symlink_inode_operations;
1659                 kaddr = kmap_atomic(page);
1660                 memcpy(kaddr, symname, len);
1661                 kunmap_atomic(kaddr);
1662                 set_page_dirty(page);
1663                 unlock_page(page);
1664                 page_cache_release(page);
1665         }
1666         dir->i_size += BOGO_DIRENT_SIZE;
1667         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1668         d_instantiate(dentry, inode);
1669         dget(dentry);
1670         return 0;
1671 }
1672
1673 static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
1674 {
1675         nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
1676         return NULL;
1677 }
1678
1679 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1680 {
1681         struct page *page = NULL;
1682         int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1683         nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
1684         if (page)
1685                 unlock_page(page);
1686         return page;
1687 }
1688
1689 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1690 {
1691         if (!IS_ERR(nd_get_link(nd))) {
1692                 struct page *page = cookie;
1693                 kunmap(page);
1694                 mark_page_accessed(page);
1695                 page_cache_release(page);
1696         }
1697 }
1698
1699 #ifdef CONFIG_TMPFS_XATTR
1700 /*
1701  * Superblocks without xattr inode operations may get some security.* xattr
1702  * support from the LSM "for free". As soon as we have any other xattrs
1703  * like ACLs, we also need to implement the security.* handlers at
1704  * filesystem level, though.
1705  */
1706
1707 static int shmem_xattr_get(struct dentry *dentry, const char *name,
1708                            void *buffer, size_t size)
1709 {
1710         struct shmem_inode_info *info;
1711         struct shmem_xattr *xattr;
1712         int ret = -ENODATA;
1713
1714         info = SHMEM_I(dentry->d_inode);
1715
1716         spin_lock(&info->lock);
1717         list_for_each_entry(xattr, &info->xattr_list, list) {
1718                 if (strcmp(name, xattr->name))
1719                         continue;
1720
1721                 ret = xattr->size;
1722                 if (buffer) {
1723                         if (size < xattr->size)
1724                                 ret = -ERANGE;
1725                         else
1726                                 memcpy(buffer, xattr->value, xattr->size);
1727                 }
1728                 break;
1729         }
1730         spin_unlock(&info->lock);
1731         return ret;
1732 }
1733
1734 static int shmem_xattr_set(struct dentry *dentry, const char *name,
1735                            const void *value, size_t size, int flags)
1736 {
1737         struct inode *inode = dentry->d_inode;
1738         struct shmem_inode_info *info = SHMEM_I(inode);
1739         struct shmem_xattr *xattr;
1740         struct shmem_xattr *new_xattr = NULL;
1741         size_t len;
1742         int err = 0;
1743
1744         /* value == NULL means remove */
1745         if (value) {
1746                 /* wrap around? */
1747                 len = sizeof(*new_xattr) + size;
1748                 if (len <= sizeof(*new_xattr))
1749                         return -ENOMEM;
1750
1751                 new_xattr = kmalloc(len, GFP_KERNEL);
1752                 if (!new_xattr)
1753                         return -ENOMEM;
1754
1755                 new_xattr->name = kstrdup(name, GFP_KERNEL);
1756                 if (!new_xattr->name) {
1757                         kfree(new_xattr);
1758                         return -ENOMEM;
1759                 }
1760
1761                 new_xattr->size = size;
1762                 memcpy(new_xattr->value, value, size);
1763         }
1764
1765         spin_lock(&info->lock);
1766         list_for_each_entry(xattr, &info->xattr_list, list) {
1767                 if (!strcmp(name, xattr->name)) {
1768                         if (flags & XATTR_CREATE) {
1769                                 xattr = new_xattr;
1770                                 err = -EEXIST;
1771                         } else if (new_xattr) {
1772                                 list_replace(&xattr->list, &new_xattr->list);
1773                         } else {
1774                                 list_del(&xattr->list);
1775                         }
1776                         goto out;
1777                 }
1778         }
1779         if (flags & XATTR_REPLACE) {
1780                 xattr = new_xattr;
1781                 err = -ENODATA;
1782         } else {
1783                 list_add(&new_xattr->list, &info->xattr_list);
1784                 xattr = NULL;
1785         }
1786 out:
1787         spin_unlock(&info->lock);
1788         if (xattr)
1789                 kfree(xattr->name);
1790         kfree(xattr);
1791         return err;
1792 }
1793
1794 static const struct xattr_handler *shmem_xattr_handlers[] = {
1795 #ifdef CONFIG_TMPFS_POSIX_ACL
1796         &generic_acl_access_handler,
1797         &generic_acl_default_handler,
1798 #endif
1799         NULL
1800 };
1801
1802 static int shmem_xattr_validate(const char *name)
1803 {
1804         struct { const char *prefix; size_t len; } arr[] = {
1805                 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
1806                 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
1807         };
1808         int i;
1809
1810         for (i = 0; i < ARRAY_SIZE(arr); i++) {
1811                 size_t preflen = arr[i].len;
1812                 if (strncmp(name, arr[i].prefix, preflen) == 0) {
1813                         if (!name[preflen])
1814                                 return -EINVAL;
1815                         return 0;
1816                 }
1817         }
1818         return -EOPNOTSUPP;
1819 }
1820
1821 static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
1822                               void *buffer, size_t size)
1823 {
1824         int err;
1825
1826         /*
1827          * If this is a request for a synthetic attribute in the system.*
1828          * namespace use the generic infrastructure to resolve a handler
1829          * for it via sb->s_xattr.
1830          */
1831         if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1832                 return generic_getxattr(dentry, name, buffer, size);
1833
1834         err = shmem_xattr_validate(name);
1835         if (err)
1836                 return err;
1837
1838         return shmem_xattr_get(dentry, name, buffer, size);
1839 }
1840
1841 static int shmem_setxattr(struct dentry *dentry, const char *name,
1842                           const void *value, size_t size, int flags)
1843 {
1844         int err;
1845
1846         /*
1847          * If this is a request for a synthetic attribute in the system.*
1848          * namespace use the generic infrastructure to resolve a handler
1849          * for it via sb->s_xattr.
1850          */
1851         if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1852                 return generic_setxattr(dentry, name, value, size, flags);
1853
1854         err = shmem_xattr_validate(name);
1855         if (err)
1856                 return err;
1857
1858         if (size == 0)
1859                 value = "";  /* empty EA, do not remove */
1860
1861         return shmem_xattr_set(dentry, name, value, size, flags);
1862
1863 }
1864
1865 static int shmem_removexattr(struct dentry *dentry, const char *name)
1866 {
1867         int err;
1868
1869         /*
1870          * If this is a request for a synthetic attribute in the system.*
1871          * namespace use the generic infrastructure to resolve a handler
1872          * for it via sb->s_xattr.
1873          */
1874         if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1875                 return generic_removexattr(dentry, name);
1876
1877         err = shmem_xattr_validate(name);
1878         if (err)
1879                 return err;
1880
1881         return shmem_xattr_set(dentry, name, NULL, 0, XATTR_REPLACE);
1882 }
1883
1884 static bool xattr_is_trusted(const char *name)
1885 {
1886         return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN);
1887 }
1888
1889 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
1890 {
1891         bool trusted = capable(CAP_SYS_ADMIN);
1892         struct shmem_xattr *xattr;
1893         struct shmem_inode_info *info;
1894         size_t used = 0;
1895
1896         info = SHMEM_I(dentry->d_inode);
1897
1898         spin_lock(&info->lock);
1899         list_for_each_entry(xattr, &info->xattr_list, list) {
1900                 size_t len;
1901
1902                 /* skip "trusted." attributes for unprivileged callers */
1903                 if (!trusted && xattr_is_trusted(xattr->name))
1904                         continue;
1905
1906                 len = strlen(xattr->name) + 1;
1907                 used += len;
1908                 if (buffer) {
1909                         if (size < used) {
1910                                 used = -ERANGE;
1911                                 break;
1912                         }
1913                         memcpy(buffer, xattr->name, len);
1914                         buffer += len;
1915                 }
1916         }
1917         spin_unlock(&info->lock);
1918
1919         return used;
1920 }
1921 #endif /* CONFIG_TMPFS_XATTR */
1922
1923 static const struct inode_operations shmem_short_symlink_operations = {
1924         .readlink       = generic_readlink,
1925         .follow_link    = shmem_follow_short_symlink,
1926 #ifdef CONFIG_TMPFS_XATTR
1927         .setxattr       = shmem_setxattr,
1928         .getxattr       = shmem_getxattr,
1929         .listxattr      = shmem_listxattr,
1930         .removexattr    = shmem_removexattr,
1931 #endif
1932 };
1933
1934 static const struct inode_operations shmem_symlink_inode_operations = {
1935         .readlink       = generic_readlink,
1936         .follow_link    = shmem_follow_link,
1937         .put_link       = shmem_put_link,
1938 #ifdef CONFIG_TMPFS_XATTR
1939         .setxattr       = shmem_setxattr,
1940         .getxattr       = shmem_getxattr,
1941         .listxattr      = shmem_listxattr,
1942         .removexattr    = shmem_removexattr,
1943 #endif
1944 };
1945
1946 static struct dentry *shmem_get_parent(struct dentry *child)
1947 {
1948         return ERR_PTR(-ESTALE);
1949 }
1950
1951 static int shmem_match(struct inode *ino, void *vfh)
1952 {
1953         __u32 *fh = vfh;
1954         __u64 inum = fh[2];
1955         inum = (inum << 32) | fh[1];
1956         return ino->i_ino == inum && fh[0] == ino->i_generation;
1957 }
1958
1959 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
1960                 struct fid *fid, int fh_len, int fh_type)
1961 {
1962         struct inode *inode;
1963         struct dentry *dentry = NULL;
1964         u64 inum = fid->raw[2];
1965         inum = (inum << 32) | fid->raw[1];
1966
1967         if (fh_len < 3)
1968                 return NULL;
1969
1970         inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
1971                         shmem_match, fid->raw);
1972         if (inode) {
1973                 dentry = d_find_alias(inode);
1974                 iput(inode);
1975         }
1976
1977         return dentry;
1978 }
1979
1980 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
1981                                 int connectable)
1982 {
1983         struct inode *inode = dentry->d_inode;
1984
1985         if (*len < 3) {
1986                 *len = 3;
1987                 return 255;
1988         }
1989
1990         if (inode_unhashed(inode)) {
1991                 /* Unfortunately insert_inode_hash is not idempotent,
1992                  * so as we hash inodes here rather than at creation
1993                  * time, we need a lock to ensure we only try
1994                  * to do it once
1995                  */
1996                 static DEFINE_SPINLOCK(lock);
1997                 spin_lock(&lock);
1998                 if (inode_unhashed(inode))
1999                         __insert_inode_hash(inode,
2000                                             inode->i_ino + inode->i_generation);
2001                 spin_unlock(&lock);
2002         }
2003
2004         fh[0] = inode->i_generation;
2005         fh[1] = inode->i_ino;
2006         fh[2] = ((__u64)inode->i_ino) >> 32;
2007
2008         *len = 3;
2009         return 1;
2010 }
2011
2012 static const struct export_operations shmem_export_ops = {
2013         .get_parent     = shmem_get_parent,
2014         .encode_fh      = shmem_encode_fh,
2015         .fh_to_dentry   = shmem_fh_to_dentry,
2016 };
2017
2018 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2019                                bool remount)
2020 {
2021         char *this_char, *value, *rest;
2022
2023         while (options != NULL) {
2024                 this_char = options;
2025                 for (;;) {
2026                         /*
2027                          * NUL-terminate this option: unfortunately,
2028                          * mount options form a comma-separated list,
2029                          * but mpol's nodelist may also contain commas.
2030                          */
2031                         options = strchr(options, ',');
2032                         if (options == NULL)
2033                                 break;
2034                         options++;
2035                         if (!isdigit(*options)) {
2036                                 options[-1] = '\0';
2037                                 break;
2038                         }
2039                 }
2040                 if (!*this_char)
2041                         continue;
2042                 if ((value = strchr(this_char,'=')) != NULL) {
2043                         *value++ = 0;
2044                 } else {
2045                         printk(KERN_ERR
2046                             "tmpfs: No value for mount option '%s'\n",
2047                             this_char);
2048                         return 1;
2049                 }
2050
2051                 if (!strcmp(this_char,"size")) {
2052                         unsigned long long size;
2053                         size = memparse(value,&rest);
2054                         if (*rest == '%') {
2055                                 size <<= PAGE_SHIFT;
2056                                 size *= totalram_pages;
2057                                 do_div(size, 100);
2058                                 rest++;
2059                         }
2060                         if (*rest)
2061                                 goto bad_val;
2062                         sbinfo->max_blocks =
2063                                 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2064                 } else if (!strcmp(this_char,"nr_blocks")) {
2065                         sbinfo->max_blocks = memparse(value, &rest);
2066                         if (*rest)
2067                                 goto bad_val;
2068                 } else if (!strcmp(this_char,"nr_inodes")) {
2069                         sbinfo->max_inodes = memparse(value, &rest);
2070                         if (*rest)
2071                                 goto bad_val;
2072                 } else if (!strcmp(this_char,"mode")) {
2073                         if (remount)
2074                                 continue;
2075                         sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2076                         if (*rest)
2077                                 goto bad_val;
2078                 } else if (!strcmp(this_char,"uid")) {
2079                         if (remount)
2080                                 continue;
2081                         sbinfo->uid = simple_strtoul(value, &rest, 0);
2082                         if (*rest)
2083                                 goto bad_val;
2084                 } else if (!strcmp(this_char,"gid")) {
2085                         if (remount)
2086                                 continue;
2087                         sbinfo->gid = simple_strtoul(value, &rest, 0);
2088                         if (*rest)
2089                                 goto bad_val;
2090                 } else if (!strcmp(this_char,"mpol")) {
2091                         if (mpol_parse_str(value, &sbinfo->mpol, 1))
2092                                 goto bad_val;
2093                 } else {
2094                         printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2095                                this_char);
2096                         return 1;
2097                 }
2098         }
2099         return 0;
2100
2101 bad_val:
2102         printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2103                value, this_char);
2104         return 1;
2105
2106 }
2107
2108 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2109 {
2110         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2111         struct shmem_sb_info config = *sbinfo;
2112         unsigned long inodes;
2113         int error = -EINVAL;
2114
2115         if (shmem_parse_options(data, &config, true))
2116                 return error;
2117
2118         spin_lock(&sbinfo->stat_lock);
2119         inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2120         if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2121                 goto out;
2122         if (config.max_inodes < inodes)
2123                 goto out;
2124         /*
2125          * Those tests disallow limited->unlimited while any are in use;
2126          * but we must separately disallow unlimited->limited, because
2127          * in that case we have no record of how much is already in use.
2128          */
2129         if (config.max_blocks && !sbinfo->max_blocks)
2130                 goto out;
2131         if (config.max_inodes && !sbinfo->max_inodes)
2132                 goto out;
2133
2134         error = 0;
2135         sbinfo->max_blocks  = config.max_blocks;
2136         sbinfo->max_inodes  = config.max_inodes;
2137         sbinfo->free_inodes = config.max_inodes - inodes;
2138
2139         mpol_put(sbinfo->mpol);
2140         sbinfo->mpol        = config.mpol;      /* transfers initial ref */
2141 out:
2142         spin_unlock(&sbinfo->stat_lock);
2143         return error;
2144 }
2145
2146 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
2147 {
2148         struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
2149
2150         if (sbinfo->max_blocks != shmem_default_max_blocks())
2151                 seq_printf(seq, ",size=%luk",
2152                         sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2153         if (sbinfo->max_inodes != shmem_default_max_inodes())
2154                 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2155         if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2156                 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
2157         if (sbinfo->uid != 0)
2158                 seq_printf(seq, ",uid=%u", sbinfo->uid);
2159         if (sbinfo->gid != 0)
2160                 seq_printf(seq, ",gid=%u", sbinfo->gid);
2161         shmem_show_mpol(seq, sbinfo->mpol);
2162         return 0;
2163 }
2164 #endif /* CONFIG_TMPFS */
2165
2166 static void shmem_put_super(struct super_block *sb)
2167 {
2168         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2169
2170         percpu_counter_destroy(&sbinfo->used_blocks);
2171         kfree(sbinfo);
2172         sb->s_fs_info = NULL;
2173 }
2174
2175 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2176 {
2177         struct inode *inode;
2178         struct shmem_sb_info *sbinfo;
2179         int err = -ENOMEM;
2180
2181         /* Round up to L1_CACHE_BYTES to resist false sharing */
2182         sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2183                                 L1_CACHE_BYTES), GFP_KERNEL);
2184         if (!sbinfo)
2185                 return -ENOMEM;
2186
2187         sbinfo->mode = S_IRWXUGO | S_ISVTX;
2188         sbinfo->uid = current_fsuid();
2189         sbinfo->gid = current_fsgid();
2190         sb->s_fs_info = sbinfo;
2191
2192 #ifdef CONFIG_TMPFS
2193         /*
2194          * Per default we only allow half of the physical ram per
2195          * tmpfs instance, limiting inodes to one per page of lowmem;
2196          * but the internal instance is left unlimited.
2197          */
2198         if (!(sb->s_flags & MS_NOUSER)) {
2199                 sbinfo->max_blocks = shmem_default_max_blocks();
2200                 sbinfo->max_inodes = shmem_default_max_inodes();
2201                 if (shmem_parse_options(data, sbinfo, false)) {
2202                         err = -EINVAL;
2203                         goto failed;
2204                 }
2205         }
2206         sb->s_export_op = &shmem_export_ops;
2207 #else
2208         sb->s_flags |= MS_NOUSER;
2209 #endif
2210
2211         spin_lock_init(&sbinfo->stat_lock);
2212         if (percpu_counter_init(&sbinfo->used_blocks, 0))
2213                 goto failed;
2214         sbinfo->free_inodes = sbinfo->max_inodes;
2215
2216         sb->s_maxbytes = MAX_LFS_FILESIZE;
2217         sb->s_blocksize = PAGE_CACHE_SIZE;
2218         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2219         sb->s_magic = TMPFS_MAGIC;
2220         sb->s_op = &shmem_ops;
2221         sb->s_time_gran = 1;
2222 #ifdef CONFIG_TMPFS_XATTR
2223         sb->s_xattr = shmem_xattr_handlers;
2224 #endif
2225 #ifdef CONFIG_TMPFS_POSIX_ACL
2226         sb->s_flags |= MS_POSIXACL;
2227 #endif
2228
2229         inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2230         if (!inode)
2231                 goto failed;
2232         inode->i_uid = sbinfo->uid;
2233         inode->i_gid = sbinfo->gid;
2234         sb->s_root = d_make_root(inode);
2235         if (!sb->s_root)
2236                 goto failed;
2237         return 0;
2238
2239 failed:
2240         shmem_put_super(sb);
2241         return err;
2242 }
2243
2244 static struct kmem_cache *shmem_inode_cachep;
2245
2246 static struct inode *shmem_alloc_inode(struct super_block *sb)
2247 {
2248         struct shmem_inode_info *info;
2249         info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2250         if (!info)
2251                 return NULL;
2252         return &info->vfs_inode;
2253 }
2254
2255 static void shmem_destroy_callback(struct rcu_head *head)
2256 {
2257         struct inode *inode = container_of(head, struct inode, i_rcu);
2258         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2259 }
2260
2261 static void shmem_destroy_inode(struct inode *inode)
2262 {
2263         if (S_ISREG(inode->i_mode))
2264                 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2265         call_rcu(&inode->i_rcu, shmem_destroy_callback);
2266 }
2267
2268 static void shmem_init_inode(void *foo)
2269 {
2270         struct shmem_inode_info *info = foo;
2271         inode_init_once(&info->vfs_inode);
2272 }
2273
2274 static int shmem_init_inodecache(void)
2275 {
2276         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2277                                 sizeof(struct shmem_inode_info),
2278                                 0, SLAB_PANIC, shmem_init_inode);
2279         return 0;
2280 }
2281
2282 static void shmem_destroy_inodecache(void)
2283 {
2284         kmem_cache_destroy(shmem_inode_cachep);
2285 }
2286
2287 static const struct address_space_operations shmem_aops = {
2288         .writepage      = shmem_writepage,
2289         .set_page_dirty = __set_page_dirty_no_writeback,
2290 #ifdef CONFIG_TMPFS
2291         .write_begin    = shmem_write_begin,
2292         .write_end      = shmem_write_end,
2293 #endif
2294         .migratepage    = migrate_page,
2295         .error_remove_page = generic_error_remove_page,
2296 };
2297
2298 static const struct file_operations shmem_file_operations = {
2299         .mmap           = shmem_mmap,
2300 #ifdef CONFIG_TMPFS
2301         .llseek         = generic_file_llseek,
2302         .read           = do_sync_read,
2303         .write          = do_sync_write,
2304         .aio_read       = shmem_file_aio_read,
2305         .aio_write      = generic_file_aio_write,
2306         .fsync          = noop_fsync,
2307         .splice_read    = shmem_file_splice_read,
2308         .splice_write   = generic_file_splice_write,
2309 #endif
2310 };
2311
2312 static const struct inode_operations shmem_inode_operations = {
2313         .setattr        = shmem_setattr,
2314         .truncate_range = shmem_truncate_range,
2315 #ifdef CONFIG_TMPFS_XATTR
2316         .setxattr       = shmem_setxattr,
2317         .getxattr       = shmem_getxattr,
2318         .listxattr      = shmem_listxattr,
2319         .removexattr    = shmem_removexattr,
2320 #endif
2321 };
2322
2323 static const struct inode_operations shmem_dir_inode_operations = {
2324 #ifdef CONFIG_TMPFS
2325         .create         = shmem_create,
2326         .lookup         = simple_lookup,
2327         .link           = shmem_link,
2328         .unlink         = shmem_unlink,
2329         .symlink        = shmem_symlink,
2330         .mkdir          = shmem_mkdir,
2331         .rmdir          = shmem_rmdir,
2332         .mknod          = shmem_mknod,
2333         .rename         = shmem_rename,
2334 #endif
2335 #ifdef CONFIG_TMPFS_XATTR
2336         .setxattr       = shmem_setxattr,
2337         .getxattr       = shmem_getxattr,
2338         .listxattr      = shmem_listxattr,
2339         .removexattr    = shmem_removexattr,
2340 #endif
2341 #ifdef CONFIG_TMPFS_POSIX_ACL
2342         .setattr        = shmem_setattr,
2343 #endif
2344 };
2345
2346 static const struct inode_operations shmem_special_inode_operations = {
2347 #ifdef CONFIG_TMPFS_XATTR
2348         .setxattr       = shmem_setxattr,
2349         .getxattr       = shmem_getxattr,
2350         .listxattr      = shmem_listxattr,
2351         .removexattr    = shmem_removexattr,
2352 #endif
2353 #ifdef CONFIG_TMPFS_POSIX_ACL
2354         .setattr        = shmem_setattr,
2355 #endif
2356 };
2357
2358 static const struct super_operations shmem_ops = {
2359         .alloc_inode    = shmem_alloc_inode,
2360         .destroy_inode  = shmem_destroy_inode,
2361 #ifdef CONFIG_TMPFS
2362         .statfs         = shmem_statfs,
2363         .remount_fs     = shmem_remount_fs,
2364         .show_options   = shmem_show_options,
2365 #endif
2366         .evict_inode    = shmem_evict_inode,
2367         .drop_inode     = generic_delete_inode,
2368         .put_super      = shmem_put_super,
2369 };
2370
2371 static const struct vm_operations_struct shmem_vm_ops = {
2372         .fault          = shmem_fault,
2373 #ifdef CONFIG_NUMA
2374         .set_policy     = shmem_set_policy,
2375         .get_policy     = shmem_get_policy,
2376 #endif
2377 };
2378
2379 static struct dentry *shmem_mount(struct file_system_type *fs_type,
2380         int flags, const char *dev_name, void *data)
2381 {
2382         return mount_nodev(fs_type, flags, data, shmem_fill_super);
2383 }
2384
2385 static struct file_system_type shmem_fs_type = {
2386         .owner          = THIS_MODULE,
2387         .name           = "tmpfs",
2388         .mount          = shmem_mount,
2389         .kill_sb        = kill_litter_super,
2390 };
2391
2392 int __init shmem_init(void)
2393 {
2394         int error;
2395
2396         error = bdi_init(&shmem_backing_dev_info);
2397         if (error)
2398                 goto out4;
2399
2400         error = shmem_init_inodecache();
2401         if (error)
2402                 goto out3;
2403
2404         error = register_filesystem(&shmem_fs_type);
2405         if (error) {
2406                 printk(KERN_ERR "Could not register tmpfs\n");
2407                 goto out2;
2408         }
2409
2410         shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER,
2411                                  shmem_fs_type.name, NULL);
2412         if (IS_ERR(shm_mnt)) {
2413                 error = PTR_ERR(shm_mnt);
2414                 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2415                 goto out1;
2416         }
2417         return 0;
2418
2419 out1:
2420         unregister_filesystem(&shmem_fs_type);
2421 out2:
2422         shmem_destroy_inodecache();
2423 out3:
2424         bdi_destroy(&shmem_backing_dev_info);
2425 out4:
2426         shm_mnt = ERR_PTR(error);
2427         return error;
2428 }
2429
2430 #else /* !CONFIG_SHMEM */
2431
2432 /*
2433  * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2434  *
2435  * This is intended for small system where the benefits of the full
2436  * shmem code (swap-backed and resource-limited) are outweighed by
2437  * their complexity. On systems without swap this code should be
2438  * effectively equivalent, but much lighter weight.
2439  */
2440
2441 #include <linux/ramfs.h>
2442
2443 static struct file_system_type shmem_fs_type = {
2444         .name           = "tmpfs",
2445         .mount          = ramfs_mount,
2446         .kill_sb        = kill_litter_super,
2447 };
2448
2449 int __init shmem_init(void)
2450 {
2451         BUG_ON(register_filesystem(&shmem_fs_type) != 0);
2452
2453         shm_mnt = kern_mount(&shmem_fs_type);
2454         BUG_ON(IS_ERR(shm_mnt));
2455
2456         return 0;
2457 }
2458
2459 int shmem_unuse(swp_entry_t swap, struct page *page)
2460 {
2461         return 0;
2462 }
2463
2464 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2465 {
2466         return 0;
2467 }
2468
2469 void shmem_unlock_mapping(struct address_space *mapping)
2470 {
2471 }
2472
2473 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
2474 {
2475         truncate_inode_pages_range(inode->i_mapping, lstart, lend);
2476 }
2477 EXPORT_SYMBOL_GPL(shmem_truncate_range);
2478
2479 #define shmem_vm_ops                            generic_file_vm_ops
2480 #define shmem_file_operations                   ramfs_file_operations
2481 #define shmem_get_inode(sb, dir, mode, dev, flags)      ramfs_get_inode(sb, dir, mode, dev)
2482 #define shmem_acct_size(flags, size)            0
2483 #define shmem_unacct_size(flags, size)          do {} while (0)
2484
2485 #endif /* CONFIG_SHMEM */
2486
2487 /* common code */
2488
2489 /**
2490  * shmem_file_setup - get an unlinked file living in tmpfs
2491  * @name: name for dentry (to be seen in /proc/<pid>/maps
2492  * @size: size to be set for the file
2493  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2494  */
2495 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2496 {
2497         int error;
2498         struct file *file;
2499         struct inode *inode;
2500         struct path path;
2501         struct dentry *root;
2502         struct qstr this;
2503
2504         if (IS_ERR(shm_mnt))
2505                 return (void *)shm_mnt;
2506
2507         if (size < 0 || size > MAX_LFS_FILESIZE)
2508                 return ERR_PTR(-EINVAL);
2509
2510         if (shmem_acct_size(flags, size))
2511                 return ERR_PTR(-ENOMEM);
2512
2513         error = -ENOMEM;
2514         this.name = name;
2515         this.len = strlen(name);
2516         this.hash = 0; /* will go */
2517         root = shm_mnt->mnt_root;
2518         path.dentry = d_alloc(root, &this);
2519         if (!path.dentry)
2520                 goto put_memory;
2521         path.mnt = mntget(shm_mnt);
2522
2523         error = -ENOSPC;
2524         inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2525         if (!inode)
2526                 goto put_dentry;
2527
2528         d_instantiate(path.dentry, inode);
2529         inode->i_size = size;
2530         clear_nlink(inode);     /* It is unlinked */
2531 #ifndef CONFIG_MMU
2532         error = ramfs_nommu_expand_for_mapping(inode, size);
2533         if (error)
2534                 goto put_dentry;
2535 #endif
2536
2537         error = -ENFILE;
2538         file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2539                   &shmem_file_operations);
2540         if (!file)
2541                 goto put_dentry;
2542
2543         return file;
2544
2545 put_dentry:
2546         path_put(&path);
2547 put_memory:
2548         shmem_unacct_size(flags, size);
2549         return ERR_PTR(error);
2550 }
2551 EXPORT_SYMBOL_GPL(shmem_file_setup);
2552
2553 /**
2554  * shmem_zero_setup - setup a shared anonymous mapping
2555  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2556  */
2557 int shmem_zero_setup(struct vm_area_struct *vma)
2558 {
2559         struct file *file;
2560         loff_t size = vma->vm_end - vma->vm_start;
2561
2562         file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2563         if (IS_ERR(file))
2564                 return PTR_ERR(file);
2565
2566         if (vma->vm_file)
2567                 fput(vma->vm_file);
2568         vma->vm_file = file;
2569         vma->vm_ops = &shmem_vm_ops;
2570         vma->vm_flags |= VM_CAN_NONLINEAR;
2571         return 0;
2572 }
2573
2574 /**
2575  * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2576  * @mapping:    the page's address_space
2577  * @index:      the page index
2578  * @gfp:        the page allocator flags to use if allocating
2579  *
2580  * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2581  * with any new page allocations done using the specified allocation flags.
2582  * But read_cache_page_gfp() uses the ->readpage() method: which does not
2583  * suit tmpfs, since it may have pages in swapcache, and needs to find those
2584  * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
2585  *
2586  * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
2587  * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
2588  */
2589 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
2590                                          pgoff_t index, gfp_t gfp)
2591 {
2592 #ifdef CONFIG_SHMEM
2593         struct inode *inode = mapping->host;
2594         struct page *page;
2595         int error;
2596
2597         BUG_ON(mapping->a_ops != &shmem_aops);
2598         error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
2599         if (error)
2600                 page = ERR_PTR(error);
2601         else
2602                 unlock_page(page);
2603         return page;
2604 #else
2605         /*
2606          * The tiny !SHMEM case uses ramfs without swap
2607          */
2608         return read_cache_page_gfp(mapping, index, gfp);
2609 #endif
2610 }
2611 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);