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) 2002-2005 Hugh Dickins.
  10  * Copyright (C) 2002-2005 VERITAS Software Corporation.
  11  * Copyright (C) 2004 Andi Kleen, SuSE Labs
  12  *
  13  * Extended attribute support for tmpfs:
  14  * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
  15  * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
  16  *
  17  * This file is released under the GPL.
  18  */
  19
  20 /*
  21  * This virtual memory filesystem is heavily based on the ramfs. It
  22  * extends ramfs by the ability to use swap and honor resource limits
  23  * which makes it a completely usable filesystem.
  24  */
  25
  26 #include <linux/config.h>
  27 #include <linux/module.h>
  28 #include <linux/init.h>
  29 #include <linux/devfs_fs_kernel.h>
  30 #include <linux/fs.h>
  31 #include <linux/mm.h>
  32 #include <linux/mman.h>
  33 #include <linux/file.h>
  34 #include <linux/swap.h>
  35 #include <linux/pagemap.h>
  36 #include <linux/string.h>
  37 #include <linux/slab.h>
  38 #include <linux/backing-dev.h>
  39 #include <linux/shmem_fs.h>
  40 #include <linux/mount.h>
  41 #include <linux/writeback.h>
  42 #include <linux/vfs.h>
  43 #include <linux/blkdev.h>
  44 #include <linux/security.h>
  45 #include <linux/swapops.h>
  46 #include <linux/mempolicy.h>
  47 #include <linux/namei.h>
  48 #include <linux/ctype.h>
  49 #include <linux/migrate.h>
  50
  51 #include <asm/uaccess.h>
  52 #include <asm/div64.h>
  53 #include <asm/pgtable.h>
  54
  55 /* This magic number is used in glibc for posix shared memory */
  56
  57 #define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
  58 #define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
  59 #define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
  60
  61 #define SHMEM_MAX_INDEX  (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
  62 #define SHMEM_MAX_BYTES  ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
  63
  64 #define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
  65
  66 /* info->flags needs VM_flags to handle pagein/truncate races efficiently */
  67 #define SHMEM_PAGEIN     VM_READ
  68 #define SHMEM_TRUNCATE   VM_WRITE
  69
  70 /* Definition to limit shmem_truncate's steps between cond_rescheds */
  71 #define LATENCY_LIMIT    64
  72
  73 /* Pretend that each entry is of this size in directory's i_size */
  74 #define BOGO_DIRENT_SIZE 20
  75
  76 /* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
  77 enum sgp_type {
  78         SGP_QUICK,      /* don't try more than file page cache lookup */
  79         SGP_READ,       /* don't exceed i_size, don't allocate page */
  80         SGP_CACHE,      /* don't exceed i_size, may allocate page */
  81         SGP_WRITE,      /* may exceed i_size, may allocate page */
  82 };
  83
  84 static int shmem_getpage(struct inode *inode, unsigned long idx,
  85                          struct page **pagep, enum sgp_type sgp, int *type);
  86
  87 static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
  88 {
  89         /*
  90          * The above definition of ENTRIES_PER_PAGE, and the use of
  91          * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
  92          * might be reconsidered if it ever diverges from PAGE_SIZE.
  93          */
  94         return alloc_pages(gfp_mask, PAGE_CACHE_SHIFT-PAGE_SHIFT);
  95 }
  96
  97 static inline void shmem_dir_free(struct page *page)
  98 {
  99         __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
 100 }
 101
 102 static struct page **shmem_dir_map(struct page *page)
 103 {
 104         return (struct page **)kmap_atomic(page, KM_USER0);
 105 }
 106
 107 static inline void shmem_dir_unmap(struct page **dir)
 108 {
 109         kunmap_atomic(dir, KM_USER0);
 110 }
 111
 112 static swp_entry_t *shmem_swp_map(struct page *page)
 113 {
 114         return (swp_entry_t *)kmap_atomic(page, KM_USER1);
 115 }
 116
 117 static inline void shmem_swp_balance_unmap(void)
 118 {
 119         /*
 120          * When passing a pointer to an i_direct entry, to code which
 121          * also handles indirect entries and so will shmem_swp_unmap,
 122          * we must arrange for the preempt count to remain in balance.
 123          * What kmap_atomic of a lowmem page does depends on config
 124          * and architecture, so pretend to kmap_atomic some lowmem page.
 125          */
 126         (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
 127 }
 128
 129 static inline void shmem_swp_unmap(swp_entry_t *entry)
 130 {
 131         kunmap_atomic(entry, KM_USER1);
 132 }
 133
 134 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
 135 {
 136         return sb->s_fs_info;
 137 }
 138
 139 /*
 140  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
 141  * for shared memory and for shared anonymous (/dev/zero) mappings
 142  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
 143  * consistent with the pre-accounting of private mappings ...
 144  */
 145 static inline int shmem_acct_size(unsigned long flags, loff_t size)
 146 {
 147         return (flags & VM_ACCOUNT)?
 148                 security_vm_enough_memory(VM_ACCT(size)): 0;
 149 }
 150
 151 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
 152 {
 153         if (flags & VM_ACCOUNT)
 154                 vm_unacct_memory(VM_ACCT(size));
 155 }
 156
 157 /*
 158  * ... whereas tmpfs objects are accounted incrementally as
 159  * pages are allocated, in order to allow huge sparse files.
 160  * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
 161  * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
 162  */
 163 static inline int shmem_acct_block(unsigned long flags)
 164 {
 165         return (flags & VM_ACCOUNT)?
 166                 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
 167 }
 168
 169 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
 170 {
 171         if (!(flags & VM_ACCOUNT))
 172                 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
 173 }
 174
 175 static struct super_operations shmem_ops;
 176 static struct address_space_operations shmem_aops;
 177 static struct file_operations shmem_file_operations;
 178 static struct inode_operations shmem_inode_operations;
 179 static struct inode_operations shmem_dir_inode_operations;
 180 static struct vm_operations_struct shmem_vm_ops;
 181
 182 static struct backing_dev_info shmem_backing_dev_info  __read_mostly = {
 183         .ra_pages       = 0,    /* No readahead */
 184         .capabilities   = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
 185         .unplug_io_fn   = default_unplug_io_fn,
 186 };
 187
 188 static LIST_HEAD(shmem_swaplist);
 189 static DEFINE_SPINLOCK(shmem_swaplist_lock);
 190
 191 static void shmem_free_blocks(struct inode *inode, long pages)
 192 {
 193         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 194         if (sbinfo->max_blocks) {
 195                 spin_lock(&sbinfo->stat_lock);
 196                 sbinfo->free_blocks += pages;
 197                 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
 198                 spin_unlock(&sbinfo->stat_lock);
 199         }
 200 }
 201
 202 /*
 203  * shmem_recalc_inode - recalculate the size of an inode
 204  *
 205  * @inode: inode to recalc
 206  *
 207  * We have to calculate the free blocks since the mm can drop
 208  * undirtied hole pages behind our back.
 209  *
 210  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
 211  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
 212  *
 213  * It has to be called with the spinlock held.
 214  */
 215 static void shmem_recalc_inode(struct inode *inode)
 216 {
 217         struct shmem_inode_info *info = SHMEM_I(inode);
 218         long freed;
 219
 220         freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
 221         if (freed > 0) {
 222                 info->alloced -= freed;
 223                 shmem_unacct_blocks(info->flags, freed);
 224                 shmem_free_blocks(inode, freed);
 225         }
 226 }
 227
 228 /*
 229  * shmem_swp_entry - find the swap vector position in the info structure
 230  *
 231  * @info:  info structure for the inode
 232  * @index: index of the page to find
 233  * @page:  optional page to add to the structure. Has to be preset to
 234  *         all zeros
 235  *
 236  * If there is no space allocated yet it will return NULL when
 237  * page is NULL, else it will use the page for the needed block,
 238  * setting it to NULL on return to indicate that it has been used.
 239  *
 240  * The swap vector is organized the following way:
 241  *
 242  * There are SHMEM_NR_DIRECT entries directly stored in the
 243  * shmem_inode_info structure. So small files do not need an addional
 244  * allocation.
 245  *
 246  * For pages with index > SHMEM_NR_DIRECT there is the pointer
 247  * i_indirect which points to a page which holds in the first half
 248  * doubly indirect blocks, in the second half triple indirect blocks:
 249  *
 250  * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
 251  * following layout (for SHMEM_NR_DIRECT == 16):
 252  *
 253  * i_indirect -> dir --> 16-19
 254  *            |      +-> 20-23
 255  *            |
 256  *            +-->dir2 --> 24-27
 257  *            |        +-> 28-31
 258  *            |        +-> 32-35
 259  *            |        +-> 36-39
 260  *            |
 261  *            +-->dir3 --> 40-43
 262  *                     +-> 44-47
 263  *                     +-> 48-51
 264  *                     +-> 52-55
 265  */
 266 static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
 267 {
 268         unsigned long offset;
 269         struct page **dir;
 270         struct page *subdir;
 271
 272         if (index < SHMEM_NR_DIRECT) {
 273                 shmem_swp_balance_unmap();
 274                 return info->i_direct+index;
 275         }
 276         if (!info->i_indirect) {
 277                 if (page) {
 278                         info->i_indirect = *page;
 279                         *page = NULL;
 280                 }
 281                 return NULL;                    /* need another page */
 282         }
 283
 284         index -= SHMEM_NR_DIRECT;
 285         offset = index % ENTRIES_PER_PAGE;
 286         index /= ENTRIES_PER_PAGE;
 287         dir = shmem_dir_map(info->i_indirect);
 288
 289         if (index >= ENTRIES_PER_PAGE/2) {
 290                 index -= ENTRIES_PER_PAGE/2;
 291                 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
 292                 index %= ENTRIES_PER_PAGE;
 293                 subdir = *dir;
 294                 if (!subdir) {
 295                         if (page) {
 296                                 *dir = *page;
 297                                 *page = NULL;
 298                         }
 299                         shmem_dir_unmap(dir);
 300                         return NULL;            /* need another page */
 301                 }
 302                 shmem_dir_unmap(dir);
 303                 dir = shmem_dir_map(subdir);
 304         }
 305
 306         dir += index;
 307         subdir = *dir;
 308         if (!subdir) {
 309                 if (!page || !(subdir = *page)) {
 310                         shmem_dir_unmap(dir);
 311                         return NULL;            /* need a page */
 312                 }
 313                 *dir = subdir;
 314                 *page = NULL;
 315         }
 316         shmem_dir_unmap(dir);
 317         return shmem_swp_map(subdir) + offset;
 318 }
 319
 320 static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
 321 {
 322         long incdec = value? 1: -1;
 323
 324         entry->val = value;
 325         info->swapped += incdec;
 326         if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
 327                 struct page *page = kmap_atomic_to_page(entry);
 328                 set_page_private(page, page_private(page) + incdec);
 329         }
 330 }
 331
 332 /*
 333  * shmem_swp_alloc - get the position of the swap entry for the page.
 334  *                   If it does not exist allocate the entry.
 335  *
 336  * @info:       info structure for the inode
 337  * @index:      index of the page to find
 338  * @sgp:        check and recheck i_size? skip allocation?
 339  */
 340 static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
 341 {
 342         struct inode *inode = &info->vfs_inode;
 343         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 344         struct page *page = NULL;
 345         swp_entry_t *entry;
 346
 347         if (sgp != SGP_WRITE &&
 348             ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
 349                 return ERR_PTR(-EINVAL);
 350
 351         while (!(entry = shmem_swp_entry(info, index, &page))) {
 352                 if (sgp == SGP_READ)
 353                         return shmem_swp_map(ZERO_PAGE(0));
 354                 /*
 355                  * Test free_blocks against 1 not 0, since we have 1 data
 356                  * page (and perhaps indirect index pages) yet to allocate:
 357                  * a waste to allocate index if we cannot allocate data.
 358                  */
 359                 if (sbinfo->max_blocks) {
 360                         spin_lock(&sbinfo->stat_lock);
 361                         if (sbinfo->free_blocks <= 1) {
 362                                 spin_unlock(&sbinfo->stat_lock);
 363                                 return ERR_PTR(-ENOSPC);
 364                         }
 365                         sbinfo->free_blocks--;
 366                         inode->i_blocks += BLOCKS_PER_PAGE;
 367                         spin_unlock(&sbinfo->stat_lock);
 368                 }
 369
 370                 spin_unlock(&info->lock);
 371                 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping) | __GFP_ZERO);
 372                 if (page)
 373                         set_page_private(page, 0);
 374                 spin_lock(&info->lock);
 375
 376                 if (!page) {
 377                         shmem_free_blocks(inode, 1);
 378                         return ERR_PTR(-ENOMEM);
 379                 }
 380                 if (sgp != SGP_WRITE &&
 381                     ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
 382                         entry = ERR_PTR(-EINVAL);
 383                         break;
 384                 }
 385                 if (info->next_index <= index)
 386                         info->next_index = index + 1;
 387         }
 388         if (page) {
 389                 /* another task gave its page, or truncated the file */
 390                 shmem_free_blocks(inode, 1);
 391                 shmem_dir_free(page);
 392         }
 393         if (info->next_index <= index && !IS_ERR(entry))
 394                 info->next_index = index + 1;
 395         return entry;
 396 }
 397
 398 /*
 399  * shmem_free_swp - free some swap entries in a directory
 400  *
 401  * @dir:   pointer to the directory
 402  * @edir:  pointer after last entry of the directory
 403  */
 404 static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir)
 405 {
 406         swp_entry_t *ptr;
 407         int freed = 0;
 408
 409         for (ptr = dir; ptr < edir; ptr++) {
 410                 if (ptr->val) {
 411                         free_swap_and_cache(*ptr);
 412                         *ptr = (swp_entry_t){0};
 413                         freed++;
 414                 }
 415         }
 416         return freed;
 417 }
 418
 419 static int shmem_map_and_free_swp(struct page *subdir,
 420                 int offset, int limit, struct page ***dir)
 421 {
 422         swp_entry_t *ptr;
 423         int freed = 0;
 424
 425         ptr = shmem_swp_map(subdir);
 426         for (; offset < limit; offset += LATENCY_LIMIT) {
 427                 int size = limit - offset;
 428                 if (size > LATENCY_LIMIT)
 429                         size = LATENCY_LIMIT;
 430                 freed += shmem_free_swp(ptr+offset, ptr+offset+size);
 431                 if (need_resched()) {
 432                         shmem_swp_unmap(ptr);
 433                         if (*dir) {
 434                                 shmem_dir_unmap(*dir);
 435                                 *dir = NULL;
 436                         }
 437                         cond_resched();
 438                         ptr = shmem_swp_map(subdir);
 439                 }
 440         }
 441         shmem_swp_unmap(ptr);
 442         return freed;
 443 }
 444
 445 static void shmem_free_pages(struct list_head *next)
 446 {
 447         struct page *page;
 448         int freed = 0;
 449
 450         do {
 451                 page = container_of(next, struct page, lru);
 452                 next = next->next;
 453                 shmem_dir_free(page);
 454                 freed++;
 455                 if (freed >= LATENCY_LIMIT) {
 456                         cond_resched();
 457                         freed = 0;
 458                 }
 459         } while (next);
 460 }
 461
 462 static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
 463 {
 464         struct shmem_inode_info *info = SHMEM_I(inode);
 465         unsigned long idx;
 466         unsigned long size;
 467         unsigned long limit;
 468         unsigned long stage;
 469         unsigned long diroff;
 470         struct page **dir;
 471         struct page *topdir;
 472         struct page *middir;
 473         struct page *subdir;
 474         swp_entry_t *ptr;
 475         LIST_HEAD(pages_to_free);
 476         long nr_pages_to_free = 0;
 477         long nr_swaps_freed = 0;
 478         int offset;
 479         int freed;
 480         int punch_hole = 0;
 481
 482         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
 483         idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 484         if (idx >= info->next_index)
 485                 return;
 486
 487         spin_lock(&info->lock);
 488         info->flags |= SHMEM_TRUNCATE;
 489         if (likely(end == (loff_t) -1)) {
 490                 limit = info->next_index;
 491                 info->next_index = idx;
 492         } else {
 493                 limit = (end + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 494                 if (limit > info->next_index)
 495                         limit = info->next_index;
 496                 punch_hole = 1;
 497         }
 498
 499         topdir = info->i_indirect;
 500         if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
 501                 info->i_indirect = NULL;
 502                 nr_pages_to_free++;
 503                 list_add(&topdir->lru, &pages_to_free);
 504         }
 505         spin_unlock(&info->lock);
 506
 507         if (info->swapped && idx < SHMEM_NR_DIRECT) {
 508                 ptr = info->i_direct;
 509                 size = limit;
 510                 if (size > SHMEM_NR_DIRECT)
 511                         size = SHMEM_NR_DIRECT;
 512                 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size);
 513         }
 514         if (!topdir)
 515                 goto done2;
 516
 517         BUG_ON(limit <= SHMEM_NR_DIRECT);
 518         limit -= SHMEM_NR_DIRECT;
 519         idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
 520         offset = idx % ENTRIES_PER_PAGE;
 521         idx -= offset;
 522
 523         dir = shmem_dir_map(topdir);
 524         stage = ENTRIES_PER_PAGEPAGE/2;
 525         if (idx < ENTRIES_PER_PAGEPAGE/2) {
 526                 middir = topdir;
 527                 diroff = idx/ENTRIES_PER_PAGE;
 528         } else {
 529                 dir += ENTRIES_PER_PAGE/2;
 530                 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
 531                 while (stage <= idx)
 532                         stage += ENTRIES_PER_PAGEPAGE;
 533                 middir = *dir;
 534                 if (*dir) {
 535                         diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
 536                                 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
 537                         if (!diroff && !offset) {
 538                                 *dir = NULL;
 539                                 nr_pages_to_free++;
 540                                 list_add(&middir->lru, &pages_to_free);
 541                         }
 542                         shmem_dir_unmap(dir);
 543                         dir = shmem_dir_map(middir);
 544                 } else {
 545                         diroff = 0;
 546                         offset = 0;
 547                         idx = stage;
 548                 }
 549         }
 550
 551         for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
 552                 if (unlikely(idx == stage)) {
 553                         shmem_dir_unmap(dir);
 554                         dir = shmem_dir_map(topdir) +
 555                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
 556                         while (!*dir) {
 557                                 dir++;
 558                                 idx += ENTRIES_PER_PAGEPAGE;
 559                                 if (idx >= limit)
 560                                         goto done1;
 561                         }
 562                         stage = idx + ENTRIES_PER_PAGEPAGE;
 563                         middir = *dir;
 564                         *dir = NULL;
 565                         nr_pages_to_free++;
 566                         list_add(&middir->lru, &pages_to_free);
 567                         shmem_dir_unmap(dir);
 568                         cond_resched();
 569                         dir = shmem_dir_map(middir);
 570                         diroff = 0;
 571                 }
 572                 subdir = dir[diroff];
 573                 if (subdir && page_private(subdir)) {
 574                         size = limit - idx;
 575                         if (size > ENTRIES_PER_PAGE)
 576                                 size = ENTRIES_PER_PAGE;
 577                         freed = shmem_map_and_free_swp(subdir,
 578                                                 offset, size, &dir);
 579                         if (!dir)
 580                                 dir = shmem_dir_map(middir);
 581                         nr_swaps_freed += freed;
 582                         if (offset)
 583                                 spin_lock(&info->lock);
 584                         set_page_private(subdir, page_private(subdir) - freed);
 585                         if (offset)
 586                                 spin_unlock(&info->lock);
 587                         if (!punch_hole)
 588                                 BUG_ON(page_private(subdir) > offset);
 589                 }
 590                 if (offset)
 591                         offset = 0;
 592                 else if (subdir && !page_private(subdir)) {
 593                         dir[diroff] = NULL;
 594                         nr_pages_to_free++;
 595                         list_add(&subdir->lru, &pages_to_free);
 596                 }
 597         }
 598 done1:
 599         shmem_dir_unmap(dir);
 600 done2:
 601         if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
 602                 /*
 603                  * Call truncate_inode_pages again: racing shmem_unuse_inode
 604                  * may have swizzled a page in from swap since vmtruncate or
 605                  * generic_delete_inode did it, before we lowered next_index.
 606                  * Also, though shmem_getpage checks i_size before adding to
 607                  * cache, no recheck after: so fix the narrow window there too.
 608                  */
 609                 truncate_inode_pages_range(inode->i_mapping, start, end);
 610         }
 611
 612         spin_lock(&info->lock);
 613         info->flags &= ~SHMEM_TRUNCATE;
 614         info->swapped -= nr_swaps_freed;
 615         if (nr_pages_to_free)
 616                 shmem_free_blocks(inode, nr_pages_to_free);
 617         shmem_recalc_inode(inode);
 618         spin_unlock(&info->lock);
 619
 620         /*
 621          * Empty swap vector directory pages to be freed?
 622          */
 623         if (!list_empty(&pages_to_free)) {
 624                 pages_to_free.prev->next = NULL;
 625                 shmem_free_pages(pages_to_free.next);
 626         }
 627 }
 628
 629 static void shmem_truncate(struct inode *inode)
 630 {
 631         shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
 632 }
 633
 634 static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
 635 {
 636         struct inode *inode = dentry->d_inode;
 637         struct page *page = NULL;
 638         int error;
 639
 640         if (attr->ia_valid & ATTR_SIZE) {
 641                 if (attr->ia_size < inode->i_size) {
 642                         /*
 643                          * If truncating down to a partial page, then
 644                          * if that page is already allocated, hold it
 645                          * in memory until the truncation is over, so
 646                          * truncate_partial_page cannnot miss it were
 647                          * it assigned to swap.
 648                          */
 649                         if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
 650                                 (void) shmem_getpage(inode,
 651                                         attr->ia_size>>PAGE_CACHE_SHIFT,
 652                                                 &page, SGP_READ, NULL);
 653                         }
 654                         /*
 655                          * Reset SHMEM_PAGEIN flag so that shmem_truncate can
 656                          * detect if any pages might have been added to cache
 657                          * after truncate_inode_pages.  But we needn't bother
 658                          * if it's being fully truncated to zero-length: the
 659                          * nrpages check is efficient enough in that case.
 660                          */
 661                         if (attr->ia_size) {
 662                                 struct shmem_inode_info *info = SHMEM_I(inode);
 663                                 spin_lock(&info->lock);
 664                                 info->flags &= ~SHMEM_PAGEIN;
 665                                 spin_unlock(&info->lock);
 666                         }
 667                 }
 668         }
 669
 670         error = inode_change_ok(inode, attr);
 671         if (!error)
 672                 error = inode_setattr(inode, attr);
 673         if (page)
 674                 page_cache_release(page);
 675         return error;
 676 }
 677
 678 static void shmem_delete_inode(struct inode *inode)
 679 {
 680         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
 681         struct shmem_inode_info *info = SHMEM_I(inode);
 682
 683         if (inode->i_op->truncate == shmem_truncate) {
 684                 truncate_inode_pages(inode->i_mapping, 0);
 685                 shmem_unacct_size(info->flags, inode->i_size);
 686                 inode->i_size = 0;
 687                 shmem_truncate(inode);
 688                 if (!list_empty(&info->swaplist)) {
 689                         spin_lock(&shmem_swaplist_lock);
 690                         list_del_init(&info->swaplist);
 691                         spin_unlock(&shmem_swaplist_lock);
 692                 }
 693         }
 694         BUG_ON(inode->i_blocks);
 695         if (sbinfo->max_inodes) {
 696                 spin_lock(&sbinfo->stat_lock);
 697                 sbinfo->free_inodes++;
 698                 spin_unlock(&sbinfo->stat_lock);
 699         }
 700         clear_inode(inode);
 701 }
 702
 703 static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
 704 {
 705         swp_entry_t *ptr;
 706
 707         for (ptr = dir; ptr < edir; ptr++) {
 708                 if (ptr->val == entry.val)
 709                         return ptr - dir;
 710         }
 711         return -1;
 712 }
 713
 714 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
 715 {
 716         struct inode *inode;
 717         unsigned long idx;
 718         unsigned long size;
 719         unsigned long limit;
 720         unsigned long stage;
 721         struct page **dir;
 722         struct page *subdir;
 723         swp_entry_t *ptr;
 724         int offset;
 725
 726         idx = 0;
 727         ptr = info->i_direct;
 728         spin_lock(&info->lock);
 729         limit = info->next_index;
 730         size = limit;
 731         if (size > SHMEM_NR_DIRECT)
 732                 size = SHMEM_NR_DIRECT;
 733         offset = shmem_find_swp(entry, ptr, ptr+size);
 734         if (offset >= 0) {
 735                 shmem_swp_balance_unmap();
 736                 goto found;
 737         }
 738         if (!info->i_indirect)
 739                 goto lost2;
 740
 741         dir = shmem_dir_map(info->i_indirect);
 742         stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
 743
 744         for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
 745                 if (unlikely(idx == stage)) {
 746                         shmem_dir_unmap(dir-1);
 747                         dir = shmem_dir_map(info->i_indirect) +
 748                             ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
 749                         while (!*dir) {
 750                                 dir++;
 751                                 idx += ENTRIES_PER_PAGEPAGE;
 752                                 if (idx >= limit)
 753                                         goto lost1;
 754                         }
 755                         stage = idx + ENTRIES_PER_PAGEPAGE;
 756                         subdir = *dir;
 757                         shmem_dir_unmap(dir);
 758                         dir = shmem_dir_map(subdir);
 759                 }
 760                 subdir = *dir;
 761                 if (subdir && page_private(subdir)) {
 762                         ptr = shmem_swp_map(subdir);
 763                         size = limit - idx;
 764                         if (size > ENTRIES_PER_PAGE)
 765                                 size = ENTRIES_PER_PAGE;
 766                         offset = shmem_find_swp(entry, ptr, ptr+size);
 767                         if (offset >= 0) {
 768                                 shmem_dir_unmap(dir);
 769                                 goto found;
 770                         }
 771                         shmem_swp_unmap(ptr);
 772                 }
 773         }
 774 lost1:
 775         shmem_dir_unmap(dir-1);
 776 lost2:
 777         spin_unlock(&info->lock);
 778         return 0;
 779 found:
 780         idx += offset;
 781         inode = &info->vfs_inode;
 782         if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
 783                 info->flags |= SHMEM_PAGEIN;
 784                 shmem_swp_set(info, ptr + offset, 0);
 785         }
 786         shmem_swp_unmap(ptr);
 787         spin_unlock(&info->lock);
 788         /*
 789          * Decrement swap count even when the entry is left behind:
 790          * try_to_unuse will skip over mms, then reincrement count.
 791          */
 792         swap_free(entry);
 793         return 1;
 794 }
 795
 796 /*
 797  * shmem_unuse() search for an eventually swapped out shmem page.
 798  */
 799 int shmem_unuse(swp_entry_t entry, struct page *page)
 800 {
 801         struct list_head *p, *next;
 802         struct shmem_inode_info *info;
 803         int found = 0;
 804
 805         spin_lock(&shmem_swaplist_lock);
 806         list_for_each_safe(p, next, &shmem_swaplist) {
 807                 info = list_entry(p, struct shmem_inode_info, swaplist);
 808                 if (!info->swapped)
 809                         list_del_init(&info->swaplist);
 810                 else if (shmem_unuse_inode(info, entry, page)) {
 811                         /* move head to start search for next from here */
 812                         list_move_tail(&shmem_swaplist, &info->swaplist);
 813                         found = 1;
 814                         break;
 815                 }
 816         }
 817         spin_unlock(&shmem_swaplist_lock);
 818         return found;
 819 }
 820
 821 /*
 822  * Move the page from the page cache to the swap cache.
 823  */
 824 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
 825 {
 826         struct shmem_inode_info *info;
 827         swp_entry_t *entry, swap;
 828         struct address_space *mapping;
 829         unsigned long index;
 830         struct inode *inode;
 831
 832         BUG_ON(!PageLocked(page));
 833         BUG_ON(page_mapped(page));
 834
 835         mapping = page->mapping;
 836         index = page->index;
 837         inode = mapping->host;
 838         info = SHMEM_I(inode);
 839         if (info->flags & VM_LOCKED)
 840                 goto redirty;
 841         swap = get_swap_page();
 842         if (!swap.val)
 843                 goto redirty;
 844
 845         spin_lock(&info->lock);
 846         shmem_recalc_inode(inode);
 847         if (index >= info->next_index) {
 848                 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
 849                 goto unlock;
 850         }
 851         entry = shmem_swp_entry(info, index, NULL);
 852         BUG_ON(!entry);
 853         BUG_ON(entry->val);
 854
 855         if (move_to_swap_cache(page, swap) == 0) {
 856                 shmem_swp_set(info, entry, swap.val);
 857                 shmem_swp_unmap(entry);
 858                 spin_unlock(&info->lock);
 859                 if (list_empty(&info->swaplist)) {
 860                         spin_lock(&shmem_swaplist_lock);
 861                         /* move instead of add in case we're racing */
 862                         list_move_tail(&info->swaplist, &shmem_swaplist);
 863                         spin_unlock(&shmem_swaplist_lock);
 864                 }
 865                 unlock_page(page);
 866                 return 0;
 867         }
 868
 869         shmem_swp_unmap(entry);
 870 unlock:
 871         spin_unlock(&info->lock);
 872         swap_free(swap);
 873 redirty:
 874         set_page_dirty(page);
 875         return AOP_WRITEPAGE_ACTIVATE;  /* Return with the page locked */
 876 }
 877
 878 #ifdef CONFIG_NUMA
 879 static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
 880 {
 881         char *nodelist = strchr(value, ':');
 882         int err = 1;
 883
 884         if (nodelist) {
 885                 /* NUL-terminate policy string */
 886                 *nodelist++ = '\0';
 887                 if (nodelist_parse(nodelist, *policy_nodes))
 888                         goto out;
 889         }
 890         if (!strcmp(value, "default")) {
 891                 *policy = MPOL_DEFAULT;
 892                 /* Don't allow a nodelist */
 893                 if (!nodelist)
 894                         err = 0;
 895         } else if (!strcmp(value, "prefer")) {
 896                 *policy = MPOL_PREFERRED;
 897                 /* Insist on a nodelist of one node only */
 898                 if (nodelist) {
 899                         char *rest = nodelist;
 900                         while (isdigit(*rest))
 901                                 rest++;
 902                         if (!*rest)
 903                                 err = 0;
 904                 }
 905         } else if (!strcmp(value, "bind")) {
 906                 *policy = MPOL_BIND;
 907                 /* Insist on a nodelist */
 908                 if (nodelist)
 909                         err = 0;
 910         } else if (!strcmp(value, "interleave")) {
 911                 *policy = MPOL_INTERLEAVE;
 912                 /* Default to nodes online if no nodelist */
 913                 if (!nodelist)
 914                         *policy_nodes = node_online_map;
 915                 err = 0;
 916         }
 917 out:
 918         /* Restore string for error message */
 919         if (nodelist)
 920                 *--nodelist = ':';
 921         return err;
 922 }
 923
 924 static struct page *shmem_swapin_async(struct shared_policy *p,
 925                                        swp_entry_t entry, unsigned long idx)
 926 {
 927         struct page *page;
 928         struct vm_area_struct pvma;
 929
 930         /* Create a pseudo vma that just contains the policy */
 931         memset(&pvma, 0, sizeof(struct vm_area_struct));
 932         pvma.vm_end = PAGE_SIZE;
 933         pvma.vm_pgoff = idx;
 934         pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
 935         page = read_swap_cache_async(entry, &pvma, 0);
 936         mpol_free(pvma.vm_policy);
 937         return page;
 938 }
 939
 940 struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry,
 941                           unsigned long idx)
 942 {
 943         struct shared_policy *p = &info->policy;
 944         int i, num;
 945         struct page *page;
 946         unsigned long offset;
 947
 948         num = valid_swaphandles(entry, &offset);
 949         for (i = 0; i < num; offset++, i++) {
 950                 page = shmem_swapin_async(p,
 951                                 swp_entry(swp_type(entry), offset), idx);
 952                 if (!page)
 953                         break;
 954                 page_cache_release(page);
 955         }
 956         lru_add_drain();        /* Push any new pages onto the LRU now */
 957         return shmem_swapin_async(p, entry, idx);
 958 }
 959
 960 static struct page *
 961 shmem_alloc_page(gfp_t gfp, struct shmem_inode_info *info,
 962                  unsigned long idx)
 963 {
 964         struct vm_area_struct pvma;
 965         struct page *page;
 966
 967         memset(&pvma, 0, sizeof(struct vm_area_struct));
 968         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
 969         pvma.vm_pgoff = idx;
 970         pvma.vm_end = PAGE_SIZE;
 971         page = alloc_page_vma(gfp | __GFP_ZERO, &pvma, 0);
 972         mpol_free(pvma.vm_policy);
 973         return page;
 974 }
 975 #else
 976 static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
 977 {
 978         return 1;
 979 }
 980
 981 static inline struct page *
 982 shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
 983 {
 984         swapin_readahead(entry, 0, NULL);
 985         return read_swap_cache_async(entry, NULL, 0);
 986 }
 987
 988 static inline struct page *
 989 shmem_alloc_page(gfp_t gfp,struct shmem_inode_info *info, unsigned long idx)
 990 {
 991         return alloc_page(gfp | __GFP_ZERO);
 992 }
 993 #endif
 994
 995 /*
 996  * shmem_getpage - either get the page from swap or allocate a new one
 997  *
 998  * If we allocate a new one we do not mark it dirty. That's up to the
 999  * vm. If we swap it in we mark it dirty since we also free the swap
1000  * entry since a page cannot live in both the swap and page cache
1001  */
1002 static int shmem_getpage(struct inode *inode, unsigned long idx,
1003                         struct page **pagep, enum sgp_type sgp, int *type)
1004 {
1005         struct address_space *mapping = inode->i_mapping;
1006         struct shmem_inode_info *info = SHMEM_I(inode);
1007         struct shmem_sb_info *sbinfo;
1008         struct page *filepage = *pagep;
1009         struct page *swappage;
1010         swp_entry_t *entry;
1011         swp_entry_t swap;
1012         int error;
1013
1014         if (idx >= SHMEM_MAX_INDEX)
1015                 return -EFBIG;
1016         /*
1017          * Normally, filepage is NULL on entry, and either found
1018          * uptodate immediately, or allocated and zeroed, or read
1019          * in under swappage, which is then assigned to filepage.
1020          * But shmem_prepare_write passes in a locked filepage,
1021          * which may be found not uptodate by other callers too,
1022          * and may need to be copied from the swappage read in.
1023          */
1024 repeat:
1025         if (!filepage)
1026                 filepage = find_lock_page(mapping, idx);
1027         if (filepage && PageUptodate(filepage))
1028                 goto done;
1029         error = 0;
1030         if (sgp == SGP_QUICK)
1031                 goto failed;
1032
1033         spin_lock(&info->lock);
1034         shmem_recalc_inode(inode);
1035         entry = shmem_swp_alloc(info, idx, sgp);
1036         if (IS_ERR(entry)) {
1037                 spin_unlock(&info->lock);
1038                 error = PTR_ERR(entry);
1039                 goto failed;
1040         }
1041         swap = *entry;
1042
1043         if (swap.val) {
1044                 /* Look it up and read it in.. */
1045                 swappage = lookup_swap_cache(swap);
1046                 if (!swappage) {
1047                         shmem_swp_unmap(entry);
1048                         spin_unlock(&info->lock);
1049                         /* here we actually do the io */
1050                         if (type && *type == VM_FAULT_MINOR) {
1051                                 inc_page_state(pgmajfault);
1052                                 *type = VM_FAULT_MAJOR;
1053                         }
1054                         swappage = shmem_swapin(info, swap, idx);
1055                         if (!swappage) {
1056                                 spin_lock(&info->lock);
1057                                 entry = shmem_swp_alloc(info, idx, sgp);
1058                                 if (IS_ERR(entry))
1059                                         error = PTR_ERR(entry);
1060                                 else {
1061                                         if (entry->val == swap.val)
1062                                                 error = -ENOMEM;
1063                                         shmem_swp_unmap(entry);
1064                                 }
1065                                 spin_unlock(&info->lock);
1066                                 if (error)
1067                                         goto failed;
1068                                 goto repeat;
1069                         }
1070                         wait_on_page_locked(swappage);
1071                         page_cache_release(swappage);
1072                         goto repeat;
1073                 }
1074
1075                 /* We have to do this with page locked to prevent races */
1076                 if (TestSetPageLocked(swappage)) {
1077                         shmem_swp_unmap(entry);
1078                         spin_unlock(&info->lock);
1079                         wait_on_page_locked(swappage);
1080                         page_cache_release(swappage);
1081                         goto repeat;
1082                 }
1083                 if (!PageSwapCache(swappage)) {
1084                         /* Page migration has occured */
1085                         shmem_swp_unmap(entry);
1086                         spin_unlock(&info->lock);
1087                         unlock_page(swappage);
1088                         page_cache_release(swappage);
1089                         goto repeat;
1090                 }
1091                 if (PageWriteback(swappage)) {
1092                         shmem_swp_unmap(entry);
1093                         spin_unlock(&info->lock);
1094                         wait_on_page_writeback(swappage);
1095                         unlock_page(swappage);
1096                         page_cache_release(swappage);
1097                         goto repeat;
1098                 }
1099                 if (!PageUptodate(swappage)) {
1100                         shmem_swp_unmap(entry);
1101                         spin_unlock(&info->lock);
1102                         unlock_page(swappage);
1103                         page_cache_release(swappage);
1104                         error = -EIO;
1105                         goto failed;
1106                 }
1107
1108                 if (filepage) {
1109                         shmem_swp_set(info, entry, 0);
1110                         shmem_swp_unmap(entry);
1111                         delete_from_swap_cache(swappage);
1112                         spin_unlock(&info->lock);
1113                         copy_highpage(filepage, swappage);
1114                         unlock_page(swappage);
1115                         page_cache_release(swappage);
1116                         flush_dcache_page(filepage);
1117                         SetPageUptodate(filepage);
1118                         set_page_dirty(filepage);
1119                         swap_free(swap);
1120                 } else if (!(error = move_from_swap_cache(
1121                                 swappage, idx, mapping))) {
1122                         info->flags |= SHMEM_PAGEIN;
1123                         shmem_swp_set(info, entry, 0);
1124                         shmem_swp_unmap(entry);
1125                         spin_unlock(&info->lock);
1126                         filepage = swappage;
1127                         swap_free(swap);
1128                 } else {
1129                         shmem_swp_unmap(entry);
1130                         spin_unlock(&info->lock);
1131                         unlock_page(swappage);
1132                         page_cache_release(swappage);
1133                         if (error == -ENOMEM) {
1134                                 /* let kswapd refresh zone for GFP_ATOMICs */
1135                                 blk_congestion_wait(WRITE, HZ/50);
1136                         }
1137                         goto repeat;
1138                 }
1139         } else if (sgp == SGP_READ && !filepage) {
1140                 shmem_swp_unmap(entry);
1141                 filepage = find_get_page(mapping, idx);
1142                 if (filepage &&
1143                     (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1144                         spin_unlock(&info->lock);
1145                         wait_on_page_locked(filepage);
1146                         page_cache_release(filepage);
1147                         filepage = NULL;
1148                         goto repeat;
1149                 }
1150                 spin_unlock(&info->lock);
1151         } else {
1152                 shmem_swp_unmap(entry);
1153                 sbinfo = SHMEM_SB(inode->i_sb);
1154                 if (sbinfo->max_blocks) {
1155                         spin_lock(&sbinfo->stat_lock);
1156                         if (sbinfo->free_blocks == 0 ||
1157                             shmem_acct_block(info->flags)) {
1158                                 spin_unlock(&sbinfo->stat_lock);
1159                                 spin_unlock(&info->lock);
1160                                 error = -ENOSPC;
1161                                 goto failed;
1162                         }
1163                         sbinfo->free_blocks--;
1164                         inode->i_blocks += BLOCKS_PER_PAGE;
1165                         spin_unlock(&sbinfo->stat_lock);
1166                 } else if (shmem_acct_block(info->flags)) {
1167                         spin_unlock(&info->lock);
1168                         error = -ENOSPC;
1169                         goto failed;
1170                 }
1171
1172                 if (!filepage) {
1173                         spin_unlock(&info->lock);
1174                         filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
1175                                                     info,
1176                                                     idx);
1177                         if (!filepage) {
1178                                 shmem_unacct_blocks(info->flags, 1);
1179                                 shmem_free_blocks(inode, 1);
1180                                 error = -ENOMEM;
1181                                 goto failed;
1182                         }
1183
1184                         spin_lock(&info->lock);
1185                         entry = shmem_swp_alloc(info, idx, sgp);
1186                         if (IS_ERR(entry))
1187                                 error = PTR_ERR(entry);
1188                         else {
1189                                 swap = *entry;
1190                                 shmem_swp_unmap(entry);
1191                         }
1192                         if (error || swap.val || 0 != add_to_page_cache_lru(
1193                                         filepage, mapping, idx, GFP_ATOMIC)) {
1194                                 spin_unlock(&info->lock);
1195                                 page_cache_release(filepage);
1196                                 shmem_unacct_blocks(info->flags, 1);
1197                                 shmem_free_blocks(inode, 1);
1198                                 filepage = NULL;
1199                                 if (error)
1200                                         goto failed;
1201                                 goto repeat;
1202                         }
1203                         info->flags |= SHMEM_PAGEIN;
1204                 }
1205
1206                 info->alloced++;
1207                 spin_unlock(&info->lock);
1208                 flush_dcache_page(filepage);
1209                 SetPageUptodate(filepage);
1210         }
1211 done:
1212         if (*pagep != filepage) {
1213                 unlock_page(filepage);
1214                 *pagep = filepage;
1215         }
1216         return 0;
1217
1218 failed:
1219         if (*pagep != filepage) {
1220                 unlock_page(filepage);
1221                 page_cache_release(filepage);
1222         }
1223         return error;
1224 }
1225
1226 struct page *shmem_nopage(struct vm_area_struct *vma, unsigned long address, int *type)
1227 {
1228         struct inode *inode = vma->vm_file->f_dentry->d_inode;
1229         struct page *page = NULL;
1230         unsigned long idx;
1231         int error;
1232
1233         idx = (address - vma->vm_start) >> PAGE_SHIFT;
1234         idx += vma->vm_pgoff;
1235         idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT;
1236         if (((loff_t) idx << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1237                 return NOPAGE_SIGBUS;
1238
1239         error = shmem_getpage(inode, idx, &page, SGP_CACHE, type);
1240         if (error)
1241                 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS;
1242
1243         mark_page_accessed(page);
1244         return page;
1245 }
1246
1247 static int shmem_populate(struct vm_area_struct *vma,
1248         unsigned long addr, unsigned long len,
1249         pgprot_t prot, unsigned long pgoff, int nonblock)
1250 {
1251         struct inode *inode = vma->vm_file->f_dentry->d_inode;
1252         struct mm_struct *mm = vma->vm_mm;
1253         enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE;
1254         unsigned long size;
1255
1256         size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1257         if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size)
1258                 return -EINVAL;
1259
1260         while ((long) len > 0) {
1261                 struct page *page = NULL;
1262                 int err;
1263                 /*
1264                  * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE
1265                  */
1266                 err = shmem_getpage(inode, pgoff, &page, sgp, NULL);
1267                 if (err)
1268                         return err;
1269                 /* Page may still be null, but only if nonblock was set. */
1270                 if (page) {
1271                         mark_page_accessed(page);
1272                         err = install_page(mm, vma, addr, page, prot);
1273                         if (err) {
1274                                 page_cache_release(page);
1275                                 return err;
1276                         }
1277                 } else if (vma->vm_flags & VM_NONLINEAR) {
1278                         /* No page was found just because we can't read it in
1279                          * now (being here implies nonblock != 0), but the page
1280                          * may exist, so set the PTE to fault it in later. */
1281                         err = install_file_pte(mm, vma, addr, pgoff, prot);
1282                         if (err)
1283                                 return err;
1284                 }
1285
1286                 len -= PAGE_SIZE;
1287                 addr += PAGE_SIZE;
1288                 pgoff++;
1289         }
1290         return 0;
1291 }
1292
1293 #ifdef CONFIG_NUMA
1294 int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1295 {
1296         struct inode *i = vma->vm_file->f_dentry->d_inode;
1297         return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1298 }
1299
1300 struct mempolicy *
1301 shmem_get_policy(struct vm_area_struct *vma, unsigned long addr)
1302 {
1303         struct inode *i = vma->vm_file->f_dentry->d_inode;
1304         unsigned long idx;
1305
1306         idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1307         return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1308 }
1309 #endif
1310
1311 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1312 {
1313         struct inode *inode = file->f_dentry->d_inode;
1314         struct shmem_inode_info *info = SHMEM_I(inode);
1315         int retval = -ENOMEM;
1316
1317         spin_lock(&info->lock);
1318         if (lock && !(info->flags & VM_LOCKED)) {
1319                 if (!user_shm_lock(inode->i_size, user))
1320                         goto out_nomem;
1321                 info->flags |= VM_LOCKED;
1322         }
1323         if (!lock && (info->flags & VM_LOCKED) && user) {
1324                 user_shm_unlock(inode->i_size, user);
1325                 info->flags &= ~VM_LOCKED;
1326         }
1327         retval = 0;
1328 out_nomem:
1329         spin_unlock(&info->lock);
1330         return retval;
1331 }
1332
1333 int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1334 {
1335         file_accessed(file);
1336         vma->vm_ops = &shmem_vm_ops;
1337         return 0;
1338 }
1339
1340 static struct inode *
1341 shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1342 {
1343         struct inode *inode;
1344         struct shmem_inode_info *info;
1345         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1346
1347         if (sbinfo->max_inodes) {
1348                 spin_lock(&sbinfo->stat_lock);
1349                 if (!sbinfo->free_inodes) {
1350                         spin_unlock(&sbinfo->stat_lock);
1351                         return NULL;
1352                 }
1353                 sbinfo->free_inodes--;
1354                 spin_unlock(&sbinfo->stat_lock);
1355         }
1356
1357         inode = new_inode(sb);
1358         if (inode) {
1359                 inode->i_mode = mode;
1360                 inode->i_uid = current->fsuid;
1361                 inode->i_gid = current->fsgid;
1362                 inode->i_blksize = PAGE_CACHE_SIZE;
1363                 inode->i_blocks = 0;
1364                 inode->i_mapping->a_ops = &shmem_aops;
1365                 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1366                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1367                 info = SHMEM_I(inode);
1368                 memset(info, 0, (char *)inode - (char *)info);
1369                 spin_lock_init(&info->lock);
1370                 INIT_LIST_HEAD(&info->swaplist);
1371
1372                 switch (mode & S_IFMT) {
1373                 default:
1374                         init_special_inode(inode, mode, dev);
1375                         break;
1376                 case S_IFREG:
1377                         inode->i_op = &shmem_inode_operations;
1378                         inode->i_fop = &shmem_file_operations;
1379                         mpol_shared_policy_init(&info->policy, sbinfo->policy,
1380                                                         &sbinfo->policy_nodes);
1381                         break;
1382                 case S_IFDIR:
1383                         inode->i_nlink++;
1384                         /* Some things misbehave if size == 0 on a directory */
1385                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
1386                         inode->i_op = &shmem_dir_inode_operations;
1387                         inode->i_fop = &simple_dir_operations;
1388                         break;
1389                 case S_IFLNK:
1390                         /*
1391                          * Must not load anything in the rbtree,
1392                          * mpol_free_shared_policy will not be called.
1393                          */
1394                         mpol_shared_policy_init(&info->policy, MPOL_DEFAULT,
1395                                                 NULL);
1396                         break;
1397                 }
1398         } else if (sbinfo->max_inodes) {
1399                 spin_lock(&sbinfo->stat_lock);
1400                 sbinfo->free_inodes++;
1401                 spin_unlock(&sbinfo->stat_lock);
1402         }
1403         return inode;
1404 }
1405
1406 #ifdef CONFIG_TMPFS
1407 static struct inode_operations shmem_symlink_inode_operations;
1408 static struct inode_operations shmem_symlink_inline_operations;
1409
1410 /*
1411  * Normally tmpfs makes no use of shmem_prepare_write, but it
1412  * lets a tmpfs file be used read-write below the loop driver.
1413  */
1414 static int
1415 shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1416 {
1417         struct inode *inode = page->mapping->host;
1418         return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL);
1419 }
1420
1421 static ssize_t
1422 shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1423 {
1424         struct inode    *inode = file->f_dentry->d_inode;
1425         loff_t          pos;
1426         unsigned long   written;
1427         ssize_t         err;
1428
1429         if ((ssize_t) count < 0)
1430                 return -EINVAL;
1431
1432         if (!access_ok(VERIFY_READ, buf, count))
1433                 return -EFAULT;
1434
1435         mutex_lock(&inode->i_mutex);
1436
1437         pos = *ppos;
1438         written = 0;
1439
1440         err = generic_write_checks(file, &pos, &count, 0);
1441         if (err || !count)
1442                 goto out;
1443
1444         err = remove_suid(file->f_dentry);
1445         if (err)
1446                 goto out;
1447
1448         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1449
1450         do {
1451                 struct page *page = NULL;
1452                 unsigned long bytes, index, offset;
1453                 char *kaddr;
1454                 int left;
1455
1456                 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1457                 index = pos >> PAGE_CACHE_SHIFT;
1458                 bytes = PAGE_CACHE_SIZE - offset;
1459                 if (bytes > count)
1460                         bytes = count;
1461
1462                 /*
1463                  * We don't hold page lock across copy from user -
1464                  * what would it guard against? - so no deadlock here.
1465                  * But it still may be a good idea to prefault below.
1466                  */
1467
1468                 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1469                 if (err)
1470                         break;
1471
1472                 left = bytes;
1473                 if (PageHighMem(page)) {
1474                         volatile unsigned char dummy;
1475                         __get_user(dummy, buf);
1476                         __get_user(dummy, buf + bytes - 1);
1477
1478                         kaddr = kmap_atomic(page, KM_USER0);
1479                         left = __copy_from_user_inatomic(kaddr + offset,
1480                                                         buf, bytes);
1481                         kunmap_atomic(kaddr, KM_USER0);
1482                 }
1483                 if (left) {
1484                         kaddr = kmap(page);
1485                         left = __copy_from_user(kaddr + offset, buf, bytes);
1486                         kunmap(page);
1487                 }
1488
1489                 written += bytes;
1490                 count -= bytes;
1491                 pos += bytes;
1492                 buf += bytes;
1493                 if (pos > inode->i_size)
1494                         i_size_write(inode, pos);
1495
1496                 flush_dcache_page(page);
1497                 set_page_dirty(page);
1498                 mark_page_accessed(page);
1499                 page_cache_release(page);
1500
1501                 if (left) {
1502                         pos -= left;
1503                         written -= left;
1504                         err = -EFAULT;
1505                         break;
1506                 }
1507
1508                 /*
1509                  * Our dirty pages are not counted in nr_dirty,
1510                  * and we do not attempt to balance dirty pages.
1511                  */
1512
1513                 cond_resched();
1514         } while (count);
1515
1516         *ppos = pos;
1517         if (written)
1518                 err = written;
1519 out:
1520         mutex_unlock(&inode->i_mutex);
1521         return err;
1522 }
1523
1524 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1525 {
1526         struct inode *inode = filp->f_dentry->d_inode;
1527         struct address_space *mapping = inode->i_mapping;
1528         unsigned long index, offset;
1529
1530         index = *ppos >> PAGE_CACHE_SHIFT;
1531         offset = *ppos & ~PAGE_CACHE_MASK;
1532
1533         for (;;) {
1534                 struct page *page = NULL;
1535                 unsigned long end_index, nr, ret;
1536                 loff_t i_size = i_size_read(inode);
1537
1538                 end_index = i_size >> PAGE_CACHE_SHIFT;
1539                 if (index > end_index)
1540                         break;
1541                 if (index == end_index) {
1542                         nr = i_size & ~PAGE_CACHE_MASK;
1543                         if (nr <= offset)
1544                                 break;
1545                 }
1546
1547                 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1548                 if (desc->error) {
1549                         if (desc->error == -EINVAL)
1550                                 desc->error = 0;
1551                         break;
1552                 }
1553
1554                 /*
1555                  * We must evaluate after, since reads (unlike writes)
1556                  * are called without i_mutex protection against truncate
1557                  */
1558                 nr = PAGE_CACHE_SIZE;
1559                 i_size = i_size_read(inode);
1560                 end_index = i_size >> PAGE_CACHE_SHIFT;
1561                 if (index == end_index) {
1562                         nr = i_size & ~PAGE_CACHE_MASK;
1563                         if (nr <= offset) {
1564                                 if (page)
1565                                         page_cache_release(page);
1566                                 break;
1567                         }
1568                 }
1569                 nr -= offset;
1570
1571                 if (page) {
1572                         /*
1573                          * If users can be writing to this page using arbitrary
1574                          * virtual addresses, take care about potential aliasing
1575                          * before reading the page on the kernel side.
1576                          */
1577                         if (mapping_writably_mapped(mapping))
1578                                 flush_dcache_page(page);
1579                         /*
1580                          * Mark the page accessed if we read the beginning.
1581                          */
1582                         if (!offset)
1583                                 mark_page_accessed(page);
1584                 } else {
1585                         page = ZERO_PAGE(0);
1586                         page_cache_get(page);
1587                 }
1588
1589                 /*
1590                  * Ok, we have the page, and it's up-to-date, so
1591                  * now we can copy it to user space...
1592                  *
1593                  * The actor routine returns how many bytes were actually used..
1594                  * NOTE! This may not be the same as how much of a user buffer
1595                  * we filled up (we may be padding etc), so we can only update
1596                  * "pos" here (the actor routine has to update the user buffer
1597                  * pointers and the remaining count).
1598                  */
1599                 ret = actor(desc, page, offset, nr);
1600                 offset += ret;
1601                 index += offset >> PAGE_CACHE_SHIFT;
1602                 offset &= ~PAGE_CACHE_MASK;
1603
1604                 page_cache_release(page);
1605                 if (ret != nr || !desc->count)
1606                         break;
1607
1608                 cond_resched();
1609         }
1610
1611         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1612         file_accessed(filp);
1613 }
1614
1615 static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1616 {
1617         read_descriptor_t desc;
1618
1619         if ((ssize_t) count < 0)
1620                 return -EINVAL;
1621         if (!access_ok(VERIFY_WRITE, buf, count))
1622                 return -EFAULT;
1623         if (!count)
1624                 return 0;
1625
1626         desc.written = 0;
1627         desc.count = count;
1628         desc.arg.buf = buf;
1629         desc.error = 0;
1630
1631         do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1632         if (desc.written)
1633                 return desc.written;
1634         return desc.error;
1635 }
1636
1637 static ssize_t shmem_file_sendfile(struct file *in_file, loff_t *ppos,
1638                          size_t count, read_actor_t actor, void *target)
1639 {
1640         read_descriptor_t desc;
1641
1642         if (!count)
1643                 return 0;
1644
1645         desc.written = 0;
1646         desc.count = count;
1647         desc.arg.data = target;
1648         desc.error = 0;
1649
1650         do_shmem_file_read(in_file, ppos, &desc, actor);
1651         if (desc.written)
1652                 return desc.written;
1653         return desc.error;
1654 }
1655
1656 static int shmem_statfs(struct super_block *sb, struct kstatfs *buf)
1657 {
1658         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1659
1660         buf->f_type = TMPFS_SUPER_MAGIC;
1661         buf->f_bsize = PAGE_CACHE_SIZE;
1662         buf->f_namelen = NAME_MAX;
1663         spin_lock(&sbinfo->stat_lock);
1664         if (sbinfo->max_blocks) {
1665                 buf->f_blocks = sbinfo->max_blocks;
1666                 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
1667         }
1668         if (sbinfo->max_inodes) {
1669                 buf->f_files = sbinfo->max_inodes;
1670                 buf->f_ffree = sbinfo->free_inodes;
1671         }
1672         /* else leave those fields 0 like simple_statfs */
1673         spin_unlock(&sbinfo->stat_lock);
1674         return 0;
1675 }
1676
1677 /*
1678  * File creation. Allocate an inode, and we're done..
1679  */
1680 static int
1681 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1682 {
1683         struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1684         int error = -ENOSPC;
1685
1686         if (inode) {
1687                 error = security_inode_init_security(inode, dir, NULL, NULL,
1688                                                      NULL);
1689                 if (error) {
1690                         if (error != -EOPNOTSUPP) {
1691                                 iput(inode);
1692                                 return error;
1693                         }
1694                         error = 0;
1695                 }
1696                 if (dir->i_mode & S_ISGID) {
1697                         inode->i_gid = dir->i_gid;
1698                         if (S_ISDIR(mode))
1699                                 inode->i_mode |= S_ISGID;
1700                 }
1701                 dir->i_size += BOGO_DIRENT_SIZE;
1702                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1703                 d_instantiate(dentry, inode);
1704                 dget(dentry); /* Extra count - pin the dentry in core */
1705         }
1706         return error;
1707 }
1708
1709 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1710 {
1711         int error;
1712
1713         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1714                 return error;
1715         dir->i_nlink++;
1716         return 0;
1717 }
1718
1719 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1720                 struct nameidata *nd)
1721 {
1722         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1723 }
1724
1725 /*
1726  * Link a file..
1727  */
1728 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1729 {
1730         struct inode *inode = old_dentry->d_inode;
1731         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1732
1733         /*
1734          * No ordinary (disk based) filesystem counts links as inodes;
1735          * but each new link needs a new dentry, pinning lowmem, and
1736          * tmpfs dentries cannot be pruned until they are unlinked.
1737          */
1738         if (sbinfo->max_inodes) {
1739                 spin_lock(&sbinfo->stat_lock);
1740                 if (!sbinfo->free_inodes) {
1741                         spin_unlock(&sbinfo->stat_lock);
1742                         return -ENOSPC;
1743                 }
1744                 sbinfo->free_inodes--;
1745                 spin_unlock(&sbinfo->stat_lock);
1746         }
1747
1748         dir->i_size += BOGO_DIRENT_SIZE;
1749         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1750         inode->i_nlink++;
1751         atomic_inc(&inode->i_count);    /* New dentry reference */
1752         dget(dentry);           /* Extra pinning count for the created dentry */
1753         d_instantiate(dentry, inode);
1754         return 0;
1755 }
1756
1757 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1758 {
1759         struct inode *inode = dentry->d_inode;
1760
1761         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) {
1762                 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1763                 if (sbinfo->max_inodes) {
1764                         spin_lock(&sbinfo->stat_lock);
1765                         sbinfo->free_inodes++;
1766                         spin_unlock(&sbinfo->stat_lock);
1767                 }
1768         }
1769
1770         dir->i_size -= BOGO_DIRENT_SIZE;
1771         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1772         inode->i_nlink--;
1773         dput(dentry);   /* Undo the count from "create" - this does all the work */
1774         return 0;
1775 }
1776
1777 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1778 {
1779         if (!simple_empty(dentry))
1780                 return -ENOTEMPTY;
1781
1782         dentry->d_inode->i_nlink--;
1783         dir->i_nlink--;
1784         return shmem_unlink(dir, dentry);
1785 }
1786
1787 /*
1788  * The VFS layer already does all the dentry stuff for rename,
1789  * we just have to decrement the usage count for the target if
1790  * it exists so that the VFS layer correctly free's it when it
1791  * gets overwritten.
1792  */
1793 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1794 {
1795         struct inode *inode = old_dentry->d_inode;
1796         int they_are_dirs = S_ISDIR(inode->i_mode);
1797
1798         if (!simple_empty(new_dentry))
1799                 return -ENOTEMPTY;
1800
1801         if (new_dentry->d_inode) {
1802                 (void) shmem_unlink(new_dir, new_dentry);
1803                 if (they_are_dirs)
1804                         old_dir->i_nlink--;
1805         } else if (they_are_dirs) {
1806                 old_dir->i_nlink--;
1807                 new_dir->i_nlink++;
1808         }
1809
1810         old_dir->i_size -= BOGO_DIRENT_SIZE;
1811         new_dir->i_size += BOGO_DIRENT_SIZE;
1812         old_dir->i_ctime = old_dir->i_mtime =
1813         new_dir->i_ctime = new_dir->i_mtime =
1814         inode->i_ctime = CURRENT_TIME;
1815         return 0;
1816 }
1817
1818 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1819 {
1820         int error;
1821         int len;
1822         struct inode *inode;
1823         struct page *page = NULL;
1824         char *kaddr;
1825         struct shmem_inode_info *info;
1826
1827         len = strlen(symname) + 1;
1828         if (len > PAGE_CACHE_SIZE)
1829                 return -ENAMETOOLONG;
1830
1831         inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1832         if (!inode)
1833                 return -ENOSPC;
1834
1835         error = security_inode_init_security(inode, dir, NULL, NULL,
1836                                              NULL);
1837         if (error) {
1838                 if (error != -EOPNOTSUPP) {
1839                         iput(inode);
1840                         return error;
1841                 }
1842                 error = 0;
1843         }
1844
1845         info = SHMEM_I(inode);
1846         inode->i_size = len-1;
1847         if (len <= (char *)inode - (char *)info) {
1848                 /* do it inline */
1849                 memcpy(info, symname, len);
1850                 inode->i_op = &shmem_symlink_inline_operations;
1851         } else {
1852                 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1853                 if (error) {
1854                         iput(inode);
1855                         return error;
1856                 }
1857                 inode->i_op = &shmem_symlink_inode_operations;
1858                 kaddr = kmap_atomic(page, KM_USER0);
1859                 memcpy(kaddr, symname, len);
1860                 kunmap_atomic(kaddr, KM_USER0);
1861                 set_page_dirty(page);
1862                 page_cache_release(page);
1863         }
1864         if (dir->i_mode & S_ISGID)
1865                 inode->i_gid = dir->i_gid;
1866         dir->i_size += BOGO_DIRENT_SIZE;
1867         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1868         d_instantiate(dentry, inode);
1869         dget(dentry);
1870         return 0;
1871 }
1872
1873 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1874 {
1875         nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
1876         return NULL;
1877 }
1878
1879 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1880 {
1881         struct page *page = NULL;
1882         int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1883         nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1884         return page;
1885 }
1886
1887 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1888 {
1889         if (!IS_ERR(nd_get_link(nd))) {
1890                 struct page *page = cookie;
1891                 kunmap(page);
1892                 mark_page_accessed(page);
1893                 page_cache_release(page);
1894         }
1895 }
1896
1897 static struct inode_operations shmem_symlink_inline_operations = {
1898         .readlink       = generic_readlink,
1899         .follow_link    = shmem_follow_link_inline,
1900 };
1901
1902 static struct inode_operations shmem_symlink_inode_operations = {
1903         .truncate       = shmem_truncate,
1904         .readlink       = generic_readlink,
1905         .follow_link    = shmem_follow_link,
1906         .put_link       = shmem_put_link,
1907 };
1908
1909 static int shmem_parse_options(char *options, int *mode, uid_t *uid,
1910         gid_t *gid, unsigned long *blocks, unsigned long *inodes,
1911         int *policy, nodemask_t *policy_nodes)
1912 {
1913         char *this_char, *value, *rest;
1914
1915         while (options != NULL) {
1916                 this_char = options;
1917                 for (;;) {
1918                         /*
1919                          * NUL-terminate this option: unfortunately,
1920                          * mount options form a comma-separated list,
1921                          * but mpol's nodelist may also contain commas.
1922                          */
1923                         options = strchr(options, ',');
1924                         if (options == NULL)
1925                                 break;
1926                         options++;
1927                         if (!isdigit(*options)) {
1928                                 options[-1] = '\0';
1929                                 break;
1930                         }
1931                 }
1932                 if (!*this_char)
1933                         continue;
1934                 if ((value = strchr(this_char,'=')) != NULL) {
1935                         *value++ = 0;
1936                 } else {
1937                         printk(KERN_ERR
1938                             "tmpfs: No value for mount option '%s'\n",
1939                             this_char);
1940                         return 1;
1941                 }
1942
1943                 if (!strcmp(this_char,"size")) {
1944                         unsigned long long size;
1945                         size = memparse(value,&rest);
1946                         if (*rest == '%') {
1947                                 size <<= PAGE_SHIFT;
1948                                 size *= totalram_pages;
1949                                 do_div(size, 100);
1950                                 rest++;
1951                         }
1952                         if (*rest)
1953                                 goto bad_val;
1954                         *blocks = size >> PAGE_CACHE_SHIFT;
1955                 } else if (!strcmp(this_char,"nr_blocks")) {
1956                         *blocks = memparse(value,&rest);
1957                         if (*rest)
1958                                 goto bad_val;
1959                 } else if (!strcmp(this_char,"nr_inodes")) {
1960                         *inodes = memparse(value,&rest);
1961                         if (*rest)
1962                                 goto bad_val;
1963                 } else if (!strcmp(this_char,"mode")) {
1964                         if (!mode)
1965                                 continue;
1966                         *mode = simple_strtoul(value,&rest,8);
1967                         if (*rest)
1968                                 goto bad_val;
1969                 } else if (!strcmp(this_char,"uid")) {
1970                         if (!uid)
1971                                 continue;
1972                         *uid = simple_strtoul(value,&rest,0);
1973                         if (*rest)
1974                                 goto bad_val;
1975                 } else if (!strcmp(this_char,"gid")) {
1976                         if (!gid)
1977                                 continue;
1978                         *gid = simple_strtoul(value,&rest,0);
1979                         if (*rest)
1980                                 goto bad_val;
1981                 } else if (!strcmp(this_char,"mpol")) {
1982                         if (shmem_parse_mpol(value,policy,policy_nodes))
1983                                 goto bad_val;
1984                 } else {
1985                         printk(KERN_ERR "tmpfs: Bad mount option %s\n",
1986                                this_char);
1987                         return 1;
1988                 }
1989         }
1990         return 0;
1991
1992 bad_val:
1993         printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
1994                value, this_char);
1995         return 1;
1996
1997 }
1998
1999 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2000 {
2001         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2002         unsigned long max_blocks = sbinfo->max_blocks;
2003         unsigned long max_inodes = sbinfo->max_inodes;
2004         int policy = sbinfo->policy;
2005         nodemask_t policy_nodes = sbinfo->policy_nodes;
2006         unsigned long blocks;
2007         unsigned long inodes;
2008         int error = -EINVAL;
2009
2010         if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks,
2011                                 &max_inodes, &policy, &policy_nodes))
2012                 return error;
2013
2014         spin_lock(&sbinfo->stat_lock);
2015         blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2016         inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2017         if (max_blocks < blocks)
2018                 goto out;
2019         if (max_inodes < inodes)
2020                 goto out;
2021         /*
2022          * Those tests also disallow limited->unlimited while any are in
2023          * use, so i_blocks will always be zero when max_blocks is zero;
2024          * but we must separately disallow unlimited->limited, because
2025          * in that case we have no record of how much is already in use.
2026          */
2027         if (max_blocks && !sbinfo->max_blocks)
2028                 goto out;
2029         if (max_inodes && !sbinfo->max_inodes)
2030                 goto out;
2031
2032         error = 0;
2033         sbinfo->max_blocks  = max_blocks;
2034         sbinfo->free_blocks = max_blocks - blocks;
2035         sbinfo->max_inodes  = max_inodes;
2036         sbinfo->free_inodes = max_inodes - inodes;
2037         sbinfo->policy = policy;
2038         sbinfo->policy_nodes = policy_nodes;
2039 out:
2040         spin_unlock(&sbinfo->stat_lock);
2041         return error;
2042 }
2043 #endif
2044
2045 static void shmem_put_super(struct super_block *sb)
2046 {
2047         kfree(sb->s_fs_info);
2048         sb->s_fs_info = NULL;
2049 }
2050
2051 static int shmem_fill_super(struct super_block *sb,
2052                             void *data, int silent)
2053 {
2054         struct inode *inode;
2055         struct dentry *root;
2056         int mode   = S_IRWXUGO | S_ISVTX;
2057         uid_t uid = current->fsuid;
2058         gid_t gid = current->fsgid;
2059         int err = -ENOMEM;
2060         struct shmem_sb_info *sbinfo;
2061         unsigned long blocks = 0;
2062         unsigned long inodes = 0;
2063         int policy = MPOL_DEFAULT;
2064         nodemask_t policy_nodes = node_online_map;
2065
2066 #ifdef CONFIG_TMPFS
2067         /*
2068          * Per default we only allow half of the physical ram per
2069          * tmpfs instance, limiting inodes to one per page of lowmem;
2070          * but the internal instance is left unlimited.
2071          */
2072         if (!(sb->s_flags & MS_NOUSER)) {
2073                 blocks = totalram_pages / 2;
2074                 inodes = totalram_pages - totalhigh_pages;
2075                 if (inodes > blocks)
2076                         inodes = blocks;
2077                 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks,
2078                                         &inodes, &policy, &policy_nodes))
2079                         return -EINVAL;
2080         }
2081 #else
2082         sb->s_flags |= MS_NOUSER;
2083 #endif
2084
2085         /* Round up to L1_CACHE_BYTES to resist false sharing */
2086         sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
2087                                 L1_CACHE_BYTES), GFP_KERNEL);
2088         if (!sbinfo)
2089                 return -ENOMEM;
2090
2091         spin_lock_init(&sbinfo->stat_lock);
2092         sbinfo->max_blocks = blocks;
2093         sbinfo->free_blocks = blocks;
2094         sbinfo->max_inodes = inodes;
2095         sbinfo->free_inodes = inodes;
2096         sbinfo->policy = policy;
2097         sbinfo->policy_nodes = policy_nodes;
2098
2099         sb->s_fs_info = sbinfo;
2100         sb->s_maxbytes = SHMEM_MAX_BYTES;
2101         sb->s_blocksize = PAGE_CACHE_SIZE;
2102         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2103         sb->s_magic = TMPFS_SUPER_MAGIC;
2104         sb->s_op = &shmem_ops;
2105         sb->s_time_gran = 1;
2106
2107         inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
2108         if (!inode)
2109                 goto failed;
2110         inode->i_uid = uid;
2111         inode->i_gid = gid;
2112         root = d_alloc_root(inode);
2113         if (!root)
2114                 goto failed_iput;
2115         sb->s_root = root;
2116         return 0;
2117
2118 failed_iput:
2119         iput(inode);
2120 failed:
2121         shmem_put_super(sb);
2122         return err;
2123 }
2124
2125 static struct kmem_cache *shmem_inode_cachep;
2126
2127 static struct inode *shmem_alloc_inode(struct super_block *sb)
2128 {
2129         struct shmem_inode_info *p;
2130         p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, SLAB_KERNEL);
2131         if (!p)
2132                 return NULL;
2133         return &p->vfs_inode;
2134 }
2135
2136 static void shmem_destroy_inode(struct inode *inode)
2137 {
2138         if ((inode->i_mode & S_IFMT) == S_IFREG) {
2139                 /* only struct inode is valid if it's an inline symlink */
2140                 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2141         }
2142         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2143 }
2144
2145 static void init_once(void *foo, struct kmem_cache *cachep,
2146                       unsigned long flags)
2147 {
2148         struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2149
2150         if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2151             SLAB_CTOR_CONSTRUCTOR) {
2152                 inode_init_once(&p->vfs_inode);
2153         }
2154 }
2155
2156 static int init_inodecache(void)
2157 {
2158         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2159                                 sizeof(struct shmem_inode_info),
2160                                 0, 0, init_once, NULL);
2161         if (shmem_inode_cachep == NULL)
2162                 return -ENOMEM;
2163         return 0;
2164 }
2165
2166 static void destroy_inodecache(void)
2167 {
2168         if (kmem_cache_destroy(shmem_inode_cachep))
2169                 printk(KERN_INFO "shmem_inode_cache: not all structures were freed\n");
2170 }
2171
2172 static struct address_space_operations shmem_aops = {
2173         .writepage      = shmem_writepage,
2174         .set_page_dirty = __set_page_dirty_nobuffers,
2175 #ifdef CONFIG_TMPFS
2176         .prepare_write  = shmem_prepare_write,
2177         .commit_write   = simple_commit_write,
2178 #endif
2179         .migratepage    = migrate_page,
2180 };
2181
2182 static struct file_operations shmem_file_operations = {
2183         .mmap           = shmem_mmap,
2184 #ifdef CONFIG_TMPFS
2185         .llseek         = generic_file_llseek,
2186         .read           = shmem_file_read,
2187         .write          = shmem_file_write,
2188         .fsync          = simple_sync_file,
2189         .sendfile       = shmem_file_sendfile,
2190 #endif
2191 };
2192
2193 static struct inode_operations shmem_inode_operations = {
2194         .truncate       = shmem_truncate,
2195         .setattr        = shmem_notify_change,
2196         .truncate_range = shmem_truncate_range,
2197 };
2198
2199 static struct inode_operations shmem_dir_inode_operations = {
2200 #ifdef CONFIG_TMPFS
2201         .create         = shmem_create,
2202         .lookup         = simple_lookup,
2203         .link           = shmem_link,
2204         .unlink         = shmem_unlink,
2205         .symlink        = shmem_symlink,
2206         .mkdir          = shmem_mkdir,
2207         .rmdir          = shmem_rmdir,
2208         .mknod          = shmem_mknod,
2209         .rename         = shmem_rename,
2210 #endif
2211 };
2212
2213 static struct super_operations shmem_ops = {
2214         .alloc_inode    = shmem_alloc_inode,
2215         .destroy_inode  = shmem_destroy_inode,
2216 #ifdef CONFIG_TMPFS
2217         .statfs         = shmem_statfs,
2218         .remount_fs     = shmem_remount_fs,
2219 #endif
2220         .delete_inode   = shmem_delete_inode,
2221         .drop_inode     = generic_delete_inode,
2222         .put_super      = shmem_put_super,
2223 };
2224
2225 static struct vm_operations_struct shmem_vm_ops = {
2226         .nopage         = shmem_nopage,
2227         .populate       = shmem_populate,
2228 #ifdef CONFIG_NUMA
2229         .set_policy     = shmem_set_policy,
2230         .get_policy     = shmem_get_policy,
2231 #endif
2232 };
2233
2234
2235 static struct super_block *shmem_get_sb(struct file_system_type *fs_type,
2236         int flags, const char *dev_name, void *data)
2237 {
2238         return get_sb_nodev(fs_type, flags, data, shmem_fill_super);
2239 }
2240
2241 static struct file_system_type tmpfs_fs_type = {
2242         .owner          = THIS_MODULE,
2243         .name           = "tmpfs",
2244         .get_sb         = shmem_get_sb,
2245         .kill_sb        = kill_litter_super,
2246 };
2247 static struct vfsmount *shm_mnt;
2248
2249 static int __init init_tmpfs(void)
2250 {
2251         int error;
2252
2253         error = init_inodecache();
2254         if (error)
2255                 goto out3;
2256
2257         error = register_filesystem(&tmpfs_fs_type);
2258         if (error) {
2259                 printk(KERN_ERR "Could not register tmpfs\n");
2260                 goto out2;
2261         }
2262 #ifdef CONFIG_TMPFS
2263         devfs_mk_dir("shm");
2264 #endif
2265         shm_mnt = do_kern_mount(tmpfs_fs_type.name, MS_NOUSER,
2266                                 tmpfs_fs_type.name, NULL);
2267         if (IS_ERR(shm_mnt)) {
2268                 error = PTR_ERR(shm_mnt);
2269                 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2270                 goto out1;
2271         }
2272         return 0;
2273
2274 out1:
2275         unregister_filesystem(&tmpfs_fs_type);
2276 out2:
2277         destroy_inodecache();
2278 out3:
2279         shm_mnt = ERR_PTR(error);
2280         return error;
2281 }
2282 module_init(init_tmpfs)
2283
2284 /*
2285  * shmem_file_setup - get an unlinked file living in tmpfs
2286  *
2287  * @name: name for dentry (to be seen in /proc/<pid>/maps
2288  * @size: size to be set for the file
2289  *
2290  */
2291 struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2292 {
2293         int error;
2294         struct file *file;
2295         struct inode *inode;
2296         struct dentry *dentry, *root;
2297         struct qstr this;
2298
2299         if (IS_ERR(shm_mnt))
2300                 return (void *)shm_mnt;
2301
2302         if (size < 0 || size > SHMEM_MAX_BYTES)
2303                 return ERR_PTR(-EINVAL);
2304
2305         if (shmem_acct_size(flags, size))
2306                 return ERR_PTR(-ENOMEM);
2307
2308         error = -ENOMEM;
2309         this.name = name;
2310         this.len = strlen(name);
2311         this.hash = 0; /* will go */
2312         root = shm_mnt->mnt_root;
2313         dentry = d_alloc(root, &this);
2314         if (!dentry)
2315                 goto put_memory;
2316
2317         error = -ENFILE;
2318         file = get_empty_filp();
2319         if (!file)
2320                 goto put_dentry;
2321
2322         error = -ENOSPC;
2323         inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2324         if (!inode)
2325                 goto close_file;
2326
2327         SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2328         d_instantiate(dentry, inode);
2329         inode->i_size = size;
2330         inode->i_nlink = 0;     /* It is unlinked */
2331         file->f_vfsmnt = mntget(shm_mnt);
2332         file->f_dentry = dentry;
2333         file->f_mapping = inode->i_mapping;
2334         file->f_op = &shmem_file_operations;
2335         file->f_mode = FMODE_WRITE | FMODE_READ;
2336         return file;
2337
2338 close_file:
2339         put_filp(file);
2340 put_dentry:
2341         dput(dentry);
2342 put_memory:
2343         shmem_unacct_size(flags, size);
2344         return ERR_PTR(error);
2345 }
2346
2347 /*
2348  * shmem_zero_setup - setup a shared anonymous mapping
2349  *
2350  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2351  */
2352 int shmem_zero_setup(struct vm_area_struct *vma)
2353 {
2354         struct file *file;
2355         loff_t size = vma->vm_end - vma->vm_start;
2356
2357         file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2358         if (IS_ERR(file))
2359                 return PTR_ERR(file);
2360
2361         if (vma->vm_file)
2362                 fput(vma->vm_file);
2363         vma->vm_file = file;
2364         vma->vm_ops = &shmem_vm_ops;
2365         return 0;
2366 }