shmem.c revision e6c9366b2adb52cba64b359b3050200743c7568c
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 * tiny-shmem: 18 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com> 19 * 20 * This file is released under the GPL. 21 */ 22 23#include <linux/fs.h> 24#include <linux/init.h> 25#include <linux/vfs.h> 26#include <linux/mount.h> 27#include <linux/pagemap.h> 28#include <linux/file.h> 29#include <linux/mm.h> 30#include <linux/module.h> 31#include <linux/percpu_counter.h> 32#include <linux/swap.h> 33 34static struct vfsmount *shm_mnt; 35 36#ifdef CONFIG_SHMEM 37/* 38 * This virtual memory filesystem is heavily based on the ramfs. It 39 * extends ramfs by the ability to use swap and honor resource limits 40 * which makes it a completely usable filesystem. 41 */ 42 43#include <linux/xattr.h> 44#include <linux/exportfs.h> 45#include <linux/posix_acl.h> 46#include <linux/generic_acl.h> 47#include <linux/mman.h> 48#include <linux/string.h> 49#include <linux/slab.h> 50#include <linux/backing-dev.h> 51#include <linux/shmem_fs.h> 52#include <linux/writeback.h> 53#include <linux/blkdev.h> 54#include <linux/security.h> 55#include <linux/swapops.h> 56#include <linux/mempolicy.h> 57#include <linux/namei.h> 58#include <linux/ctype.h> 59#include <linux/migrate.h> 60#include <linux/highmem.h> 61#include <linux/seq_file.h> 62#include <linux/magic.h> 63 64#include <asm/uaccess.h> 65#include <asm/div64.h> 66#include <asm/pgtable.h> 67 68/* 69 * The maximum size of a shmem/tmpfs file is limited by the maximum size of 70 * its triple-indirect swap vector - see illustration at shmem_swp_entry(). 71 * 72 * With 4kB page size, maximum file size is just over 2TB on a 32-bit kernel, 73 * but one eighth of that on a 64-bit kernel. With 8kB page size, maximum 74 * file size is just over 4TB on a 64-bit kernel, but 16TB on a 32-bit kernel, 75 * MAX_LFS_FILESIZE being then more restrictive than swap vector layout. 76 * 77 * We use / and * instead of shifts in the definitions below, so that the swap 78 * vector can be tested with small even values (e.g. 20) for ENTRIES_PER_PAGE. 79 */ 80#define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long)) 81#define ENTRIES_PER_PAGEPAGE ((unsigned long long)ENTRIES_PER_PAGE*ENTRIES_PER_PAGE) 82 83#define SHMSWP_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1)) 84#define SHMSWP_MAX_BYTES (SHMSWP_MAX_INDEX << PAGE_CACHE_SHIFT) 85 86#define SHMEM_MAX_BYTES min_t(unsigned long long, SHMSWP_MAX_BYTES, MAX_LFS_FILESIZE) 87#define SHMEM_MAX_INDEX ((unsigned long)((SHMEM_MAX_BYTES+1) >> PAGE_CACHE_SHIFT)) 88 89#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512) 90#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT) 91 92/* info->flags needs VM_flags to handle pagein/truncate races efficiently */ 93#define SHMEM_PAGEIN VM_READ 94#define SHMEM_TRUNCATE VM_WRITE 95 96/* Definition to limit shmem_truncate's steps between cond_rescheds */ 97#define LATENCY_LIMIT 64 98 99/* Pretend that each entry is of this size in directory's i_size */ 100#define BOGO_DIRENT_SIZE 20 101 102/* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */ 103enum sgp_type { 104 SGP_READ, /* don't exceed i_size, don't allocate page */ 105 SGP_CACHE, /* don't exceed i_size, may allocate page */ 106 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */ 107 SGP_WRITE, /* may exceed i_size, may allocate page */ 108}; 109 110#ifdef CONFIG_TMPFS 111static unsigned long shmem_default_max_blocks(void) 112{ 113 return totalram_pages / 2; 114} 115 116static unsigned long shmem_default_max_inodes(void) 117{ 118 return min(totalram_pages - totalhigh_pages, totalram_pages / 2); 119} 120#endif 121 122static int shmem_getpage(struct inode *inode, unsigned long idx, 123 struct page **pagep, enum sgp_type sgp, int *type); 124 125static inline struct page *shmem_dir_alloc(gfp_t gfp_mask) 126{ 127 /* 128 * The above definition of ENTRIES_PER_PAGE, and the use of 129 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE: 130 * might be reconsidered if it ever diverges from PAGE_SIZE. 131 * 132 * Mobility flags are masked out as swap vectors cannot move 133 */ 134 return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO, 135 PAGE_CACHE_SHIFT-PAGE_SHIFT); 136} 137 138static inline void shmem_dir_free(struct page *page) 139{ 140 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT); 141} 142 143static struct page **shmem_dir_map(struct page *page) 144{ 145 return (struct page **)kmap_atomic(page, KM_USER0); 146} 147 148static inline void shmem_dir_unmap(struct page **dir) 149{ 150 kunmap_atomic(dir, KM_USER0); 151} 152 153static swp_entry_t *shmem_swp_map(struct page *page) 154{ 155 return (swp_entry_t *)kmap_atomic(page, KM_USER1); 156} 157 158static inline void shmem_swp_balance_unmap(void) 159{ 160 /* 161 * When passing a pointer to an i_direct entry, to code which 162 * also handles indirect entries and so will shmem_swp_unmap, 163 * we must arrange for the preempt count to remain in balance. 164 * What kmap_atomic of a lowmem page does depends on config 165 * and architecture, so pretend to kmap_atomic some lowmem page. 166 */ 167 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1); 168} 169 170static inline void shmem_swp_unmap(swp_entry_t *entry) 171{ 172 kunmap_atomic(entry, KM_USER1); 173} 174 175static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) 176{ 177 return sb->s_fs_info; 178} 179 180/* 181 * shmem_file_setup pre-accounts the whole fixed size of a VM object, 182 * for shared memory and for shared anonymous (/dev/zero) mappings 183 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), 184 * consistent with the pre-accounting of private mappings ... 185 */ 186static inline int shmem_acct_size(unsigned long flags, loff_t size) 187{ 188 return (flags & VM_NORESERVE) ? 189 0 : security_vm_enough_memory_kern(VM_ACCT(size)); 190} 191 192static inline void shmem_unacct_size(unsigned long flags, loff_t size) 193{ 194 if (!(flags & VM_NORESERVE)) 195 vm_unacct_memory(VM_ACCT(size)); 196} 197 198/* 199 * ... whereas tmpfs objects are accounted incrementally as 200 * pages are allocated, in order to allow huge sparse files. 201 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, 202 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. 203 */ 204static inline int shmem_acct_block(unsigned long flags) 205{ 206 return (flags & VM_NORESERVE) ? 207 security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE)) : 0; 208} 209 210static inline void shmem_unacct_blocks(unsigned long flags, long pages) 211{ 212 if (flags & VM_NORESERVE) 213 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE)); 214} 215 216static const struct super_operations shmem_ops; 217static const struct address_space_operations shmem_aops; 218static const struct file_operations shmem_file_operations; 219static const struct inode_operations shmem_inode_operations; 220static const struct inode_operations shmem_dir_inode_operations; 221static const struct inode_operations shmem_special_inode_operations; 222static const struct vm_operations_struct shmem_vm_ops; 223 224static struct backing_dev_info shmem_backing_dev_info __read_mostly = { 225 .ra_pages = 0, /* No readahead */ 226 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED, 227}; 228 229static LIST_HEAD(shmem_swaplist); 230static DEFINE_MUTEX(shmem_swaplist_mutex); 231 232static void shmem_free_blocks(struct inode *inode, long pages) 233{ 234 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 235 if (sbinfo->max_blocks) { 236 percpu_counter_add(&sbinfo->used_blocks, -pages); 237 spin_lock(&inode->i_lock); 238 inode->i_blocks -= pages*BLOCKS_PER_PAGE; 239 spin_unlock(&inode->i_lock); 240 } 241} 242 243static int shmem_reserve_inode(struct super_block *sb) 244{ 245 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 246 if (sbinfo->max_inodes) { 247 spin_lock(&sbinfo->stat_lock); 248 if (!sbinfo->free_inodes) { 249 spin_unlock(&sbinfo->stat_lock); 250 return -ENOSPC; 251 } 252 sbinfo->free_inodes--; 253 spin_unlock(&sbinfo->stat_lock); 254 } 255 return 0; 256} 257 258static void shmem_free_inode(struct super_block *sb) 259{ 260 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 261 if (sbinfo->max_inodes) { 262 spin_lock(&sbinfo->stat_lock); 263 sbinfo->free_inodes++; 264 spin_unlock(&sbinfo->stat_lock); 265 } 266} 267 268/** 269 * shmem_recalc_inode - recalculate the size of an inode 270 * @inode: inode to recalc 271 * 272 * We have to calculate the free blocks since the mm can drop 273 * undirtied hole pages behind our back. 274 * 275 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped 276 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) 277 * 278 * It has to be called with the spinlock held. 279 */ 280static void shmem_recalc_inode(struct inode *inode) 281{ 282 struct shmem_inode_info *info = SHMEM_I(inode); 283 long freed; 284 285 freed = info->alloced - info->swapped - inode->i_mapping->nrpages; 286 if (freed > 0) { 287 info->alloced -= freed; 288 shmem_unacct_blocks(info->flags, freed); 289 shmem_free_blocks(inode, freed); 290 } 291} 292 293/** 294 * shmem_swp_entry - find the swap vector position in the info structure 295 * @info: info structure for the inode 296 * @index: index of the page to find 297 * @page: optional page to add to the structure. Has to be preset to 298 * all zeros 299 * 300 * If there is no space allocated yet it will return NULL when 301 * page is NULL, else it will use the page for the needed block, 302 * setting it to NULL on return to indicate that it has been used. 303 * 304 * The swap vector is organized the following way: 305 * 306 * There are SHMEM_NR_DIRECT entries directly stored in the 307 * shmem_inode_info structure. So small files do not need an addional 308 * allocation. 309 * 310 * For pages with index > SHMEM_NR_DIRECT there is the pointer 311 * i_indirect which points to a page which holds in the first half 312 * doubly indirect blocks, in the second half triple indirect blocks: 313 * 314 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the 315 * following layout (for SHMEM_NR_DIRECT == 16): 316 * 317 * i_indirect -> dir --> 16-19 318 * | +-> 20-23 319 * | 320 * +-->dir2 --> 24-27 321 * | +-> 28-31 322 * | +-> 32-35 323 * | +-> 36-39 324 * | 325 * +-->dir3 --> 40-43 326 * +-> 44-47 327 * +-> 48-51 328 * +-> 52-55 329 */ 330static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page) 331{ 332 unsigned long offset; 333 struct page **dir; 334 struct page *subdir; 335 336 if (index < SHMEM_NR_DIRECT) { 337 shmem_swp_balance_unmap(); 338 return info->i_direct+index; 339 } 340 if (!info->i_indirect) { 341 if (page) { 342 info->i_indirect = *page; 343 *page = NULL; 344 } 345 return NULL; /* need another page */ 346 } 347 348 index -= SHMEM_NR_DIRECT; 349 offset = index % ENTRIES_PER_PAGE; 350 index /= ENTRIES_PER_PAGE; 351 dir = shmem_dir_map(info->i_indirect); 352 353 if (index >= ENTRIES_PER_PAGE/2) { 354 index -= ENTRIES_PER_PAGE/2; 355 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE; 356 index %= ENTRIES_PER_PAGE; 357 subdir = *dir; 358 if (!subdir) { 359 if (page) { 360 *dir = *page; 361 *page = NULL; 362 } 363 shmem_dir_unmap(dir); 364 return NULL; /* need another page */ 365 } 366 shmem_dir_unmap(dir); 367 dir = shmem_dir_map(subdir); 368 } 369 370 dir += index; 371 subdir = *dir; 372 if (!subdir) { 373 if (!page || !(subdir = *page)) { 374 shmem_dir_unmap(dir); 375 return NULL; /* need a page */ 376 } 377 *dir = subdir; 378 *page = NULL; 379 } 380 shmem_dir_unmap(dir); 381 return shmem_swp_map(subdir) + offset; 382} 383 384static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value) 385{ 386 long incdec = value? 1: -1; 387 388 entry->val = value; 389 info->swapped += incdec; 390 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) { 391 struct page *page = kmap_atomic_to_page(entry); 392 set_page_private(page, page_private(page) + incdec); 393 } 394} 395 396/** 397 * shmem_swp_alloc - get the position of the swap entry for the page. 398 * @info: info structure for the inode 399 * @index: index of the page to find 400 * @sgp: check and recheck i_size? skip allocation? 401 * 402 * If the entry does not exist, allocate it. 403 */ 404static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp) 405{ 406 struct inode *inode = &info->vfs_inode; 407 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 408 struct page *page = NULL; 409 swp_entry_t *entry; 410 411 if (sgp != SGP_WRITE && 412 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 413 return ERR_PTR(-EINVAL); 414 415 while (!(entry = shmem_swp_entry(info, index, &page))) { 416 if (sgp == SGP_READ) 417 return shmem_swp_map(ZERO_PAGE(0)); 418 /* 419 * Test used_blocks against 1 less max_blocks, since we have 1 data 420 * page (and perhaps indirect index pages) yet to allocate: 421 * a waste to allocate index if we cannot allocate data. 422 */ 423 if (sbinfo->max_blocks) { 424 if (percpu_counter_compare(&sbinfo->used_blocks, 425 sbinfo->max_blocks - 1) >= 0) 426 return ERR_PTR(-ENOSPC); 427 percpu_counter_inc(&sbinfo->used_blocks); 428 spin_lock(&inode->i_lock); 429 inode->i_blocks += BLOCKS_PER_PAGE; 430 spin_unlock(&inode->i_lock); 431 } 432 433 spin_unlock(&info->lock); 434 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping)); 435 spin_lock(&info->lock); 436 437 if (!page) { 438 shmem_free_blocks(inode, 1); 439 return ERR_PTR(-ENOMEM); 440 } 441 if (sgp != SGP_WRITE && 442 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { 443 entry = ERR_PTR(-EINVAL); 444 break; 445 } 446 if (info->next_index <= index) 447 info->next_index = index + 1; 448 } 449 if (page) { 450 /* another task gave its page, or truncated the file */ 451 shmem_free_blocks(inode, 1); 452 shmem_dir_free(page); 453 } 454 if (info->next_index <= index && !IS_ERR(entry)) 455 info->next_index = index + 1; 456 return entry; 457} 458 459/** 460 * shmem_free_swp - free some swap entries in a directory 461 * @dir: pointer to the directory 462 * @edir: pointer after last entry of the directory 463 * @punch_lock: pointer to spinlock when needed for the holepunch case 464 */ 465static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir, 466 spinlock_t *punch_lock) 467{ 468 spinlock_t *punch_unlock = NULL; 469 swp_entry_t *ptr; 470 int freed = 0; 471 472 for (ptr = dir; ptr < edir; ptr++) { 473 if (ptr->val) { 474 if (unlikely(punch_lock)) { 475 punch_unlock = punch_lock; 476 punch_lock = NULL; 477 spin_lock(punch_unlock); 478 if (!ptr->val) 479 continue; 480 } 481 free_swap_and_cache(*ptr); 482 *ptr = (swp_entry_t){0}; 483 freed++; 484 } 485 } 486 if (punch_unlock) 487 spin_unlock(punch_unlock); 488 return freed; 489} 490 491static int shmem_map_and_free_swp(struct page *subdir, int offset, 492 int limit, struct page ***dir, spinlock_t *punch_lock) 493{ 494 swp_entry_t *ptr; 495 int freed = 0; 496 497 ptr = shmem_swp_map(subdir); 498 for (; offset < limit; offset += LATENCY_LIMIT) { 499 int size = limit - offset; 500 if (size > LATENCY_LIMIT) 501 size = LATENCY_LIMIT; 502 freed += shmem_free_swp(ptr+offset, ptr+offset+size, 503 punch_lock); 504 if (need_resched()) { 505 shmem_swp_unmap(ptr); 506 if (*dir) { 507 shmem_dir_unmap(*dir); 508 *dir = NULL; 509 } 510 cond_resched(); 511 ptr = shmem_swp_map(subdir); 512 } 513 } 514 shmem_swp_unmap(ptr); 515 return freed; 516} 517 518static void shmem_free_pages(struct list_head *next) 519{ 520 struct page *page; 521 int freed = 0; 522 523 do { 524 page = container_of(next, struct page, lru); 525 next = next->next; 526 shmem_dir_free(page); 527 freed++; 528 if (freed >= LATENCY_LIMIT) { 529 cond_resched(); 530 freed = 0; 531 } 532 } while (next); 533} 534 535static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end) 536{ 537 struct shmem_inode_info *info = SHMEM_I(inode); 538 unsigned long idx; 539 unsigned long size; 540 unsigned long limit; 541 unsigned long stage; 542 unsigned long diroff; 543 struct page **dir; 544 struct page *topdir; 545 struct page *middir; 546 struct page *subdir; 547 swp_entry_t *ptr; 548 LIST_HEAD(pages_to_free); 549 long nr_pages_to_free = 0; 550 long nr_swaps_freed = 0; 551 int offset; 552 int freed; 553 int punch_hole; 554 spinlock_t *needs_lock; 555 spinlock_t *punch_lock; 556 unsigned long upper_limit; 557 558 inode->i_ctime = inode->i_mtime = CURRENT_TIME; 559 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 560 if (idx >= info->next_index) 561 return; 562 563 spin_lock(&info->lock); 564 info->flags |= SHMEM_TRUNCATE; 565 if (likely(end == (loff_t) -1)) { 566 limit = info->next_index; 567 upper_limit = SHMEM_MAX_INDEX; 568 info->next_index = idx; 569 needs_lock = NULL; 570 punch_hole = 0; 571 } else { 572 if (end + 1 >= inode->i_size) { /* we may free a little more */ 573 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >> 574 PAGE_CACHE_SHIFT; 575 upper_limit = SHMEM_MAX_INDEX; 576 } else { 577 limit = (end + 1) >> PAGE_CACHE_SHIFT; 578 upper_limit = limit; 579 } 580 needs_lock = &info->lock; 581 punch_hole = 1; 582 } 583 584 topdir = info->i_indirect; 585 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) { 586 info->i_indirect = NULL; 587 nr_pages_to_free++; 588 list_add(&topdir->lru, &pages_to_free); 589 } 590 spin_unlock(&info->lock); 591 592 if (info->swapped && idx < SHMEM_NR_DIRECT) { 593 ptr = info->i_direct; 594 size = limit; 595 if (size > SHMEM_NR_DIRECT) 596 size = SHMEM_NR_DIRECT; 597 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock); 598 } 599 600 /* 601 * If there are no indirect blocks or we are punching a hole 602 * below indirect blocks, nothing to be done. 603 */ 604 if (!topdir || limit <= SHMEM_NR_DIRECT) 605 goto done2; 606 607 /* 608 * The truncation case has already dropped info->lock, and we're safe 609 * because i_size and next_index have already been lowered, preventing 610 * access beyond. But in the punch_hole case, we still need to take 611 * the lock when updating the swap directory, because there might be 612 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or 613 * shmem_writepage. However, whenever we find we can remove a whole 614 * directory page (not at the misaligned start or end of the range), 615 * we first NULLify its pointer in the level above, and then have no 616 * need to take the lock when updating its contents: needs_lock and 617 * punch_lock (either pointing to info->lock or NULL) manage this. 618 */ 619 620 upper_limit -= SHMEM_NR_DIRECT; 621 limit -= SHMEM_NR_DIRECT; 622 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0; 623 offset = idx % ENTRIES_PER_PAGE; 624 idx -= offset; 625 626 dir = shmem_dir_map(topdir); 627 stage = ENTRIES_PER_PAGEPAGE/2; 628 if (idx < ENTRIES_PER_PAGEPAGE/2) { 629 middir = topdir; 630 diroff = idx/ENTRIES_PER_PAGE; 631 } else { 632 dir += ENTRIES_PER_PAGE/2; 633 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE; 634 while (stage <= idx) 635 stage += ENTRIES_PER_PAGEPAGE; 636 middir = *dir; 637 if (*dir) { 638 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) % 639 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE; 640 if (!diroff && !offset && upper_limit >= stage) { 641 if (needs_lock) { 642 spin_lock(needs_lock); 643 *dir = NULL; 644 spin_unlock(needs_lock); 645 needs_lock = NULL; 646 } else 647 *dir = NULL; 648 nr_pages_to_free++; 649 list_add(&middir->lru, &pages_to_free); 650 } 651 shmem_dir_unmap(dir); 652 dir = shmem_dir_map(middir); 653 } else { 654 diroff = 0; 655 offset = 0; 656 idx = stage; 657 } 658 } 659 660 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) { 661 if (unlikely(idx == stage)) { 662 shmem_dir_unmap(dir); 663 dir = shmem_dir_map(topdir) + 664 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 665 while (!*dir) { 666 dir++; 667 idx += ENTRIES_PER_PAGEPAGE; 668 if (idx >= limit) 669 goto done1; 670 } 671 stage = idx + ENTRIES_PER_PAGEPAGE; 672 middir = *dir; 673 if (punch_hole) 674 needs_lock = &info->lock; 675 if (upper_limit >= stage) { 676 if (needs_lock) { 677 spin_lock(needs_lock); 678 *dir = NULL; 679 spin_unlock(needs_lock); 680 needs_lock = NULL; 681 } else 682 *dir = NULL; 683 nr_pages_to_free++; 684 list_add(&middir->lru, &pages_to_free); 685 } 686 shmem_dir_unmap(dir); 687 cond_resched(); 688 dir = shmem_dir_map(middir); 689 diroff = 0; 690 } 691 punch_lock = needs_lock; 692 subdir = dir[diroff]; 693 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) { 694 if (needs_lock) { 695 spin_lock(needs_lock); 696 dir[diroff] = NULL; 697 spin_unlock(needs_lock); 698 punch_lock = NULL; 699 } else 700 dir[diroff] = NULL; 701 nr_pages_to_free++; 702 list_add(&subdir->lru, &pages_to_free); 703 } 704 if (subdir && page_private(subdir) /* has swap entries */) { 705 size = limit - idx; 706 if (size > ENTRIES_PER_PAGE) 707 size = ENTRIES_PER_PAGE; 708 freed = shmem_map_and_free_swp(subdir, 709 offset, size, &dir, punch_lock); 710 if (!dir) 711 dir = shmem_dir_map(middir); 712 nr_swaps_freed += freed; 713 if (offset || punch_lock) { 714 spin_lock(&info->lock); 715 set_page_private(subdir, 716 page_private(subdir) - freed); 717 spin_unlock(&info->lock); 718 } else 719 BUG_ON(page_private(subdir) != freed); 720 } 721 offset = 0; 722 } 723done1: 724 shmem_dir_unmap(dir); 725done2: 726 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) { 727 /* 728 * Call truncate_inode_pages again: racing shmem_unuse_inode 729 * may have swizzled a page in from swap since 730 * truncate_pagecache or generic_delete_inode did it, before we 731 * lowered next_index. Also, though shmem_getpage checks 732 * i_size before adding to cache, no recheck after: so fix the 733 * narrow window there too. 734 * 735 * Recalling truncate_inode_pages_range and unmap_mapping_range 736 * every time for punch_hole (which never got a chance to clear 737 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive, 738 * yet hardly ever necessary: try to optimize them out later. 739 */ 740 truncate_inode_pages_range(inode->i_mapping, start, end); 741 if (punch_hole) 742 unmap_mapping_range(inode->i_mapping, start, 743 end - start, 1); 744 } 745 746 spin_lock(&info->lock); 747 info->flags &= ~SHMEM_TRUNCATE; 748 info->swapped -= nr_swaps_freed; 749 if (nr_pages_to_free) 750 shmem_free_blocks(inode, nr_pages_to_free); 751 shmem_recalc_inode(inode); 752 spin_unlock(&info->lock); 753 754 /* 755 * Empty swap vector directory pages to be freed? 756 */ 757 if (!list_empty(&pages_to_free)) { 758 pages_to_free.prev->next = NULL; 759 shmem_free_pages(pages_to_free.next); 760 } 761} 762 763static int shmem_notify_change(struct dentry *dentry, struct iattr *attr) 764{ 765 struct inode *inode = dentry->d_inode; 766 loff_t newsize = attr->ia_size; 767 int error; 768 769 error = inode_change_ok(inode, attr); 770 if (error) 771 return error; 772 773 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE) 774 && newsize != inode->i_size) { 775 struct page *page = NULL; 776 777 if (newsize < inode->i_size) { 778 /* 779 * If truncating down to a partial page, then 780 * if that page is already allocated, hold it 781 * in memory until the truncation is over, so 782 * truncate_partial_page cannot miss it were 783 * it assigned to swap. 784 */ 785 if (newsize & (PAGE_CACHE_SIZE-1)) { 786 (void) shmem_getpage(inode, 787 newsize >> PAGE_CACHE_SHIFT, 788 &page, SGP_READ, NULL); 789 if (page) 790 unlock_page(page); 791 } 792 /* 793 * Reset SHMEM_PAGEIN flag so that shmem_truncate can 794 * detect if any pages might have been added to cache 795 * after truncate_inode_pages. But we needn't bother 796 * if it's being fully truncated to zero-length: the 797 * nrpages check is efficient enough in that case. 798 */ 799 if (newsize) { 800 struct shmem_inode_info *info = SHMEM_I(inode); 801 spin_lock(&info->lock); 802 info->flags &= ~SHMEM_PAGEIN; 803 spin_unlock(&info->lock); 804 } 805 } 806 807 /* XXX(truncate): truncate_setsize should be called last */ 808 truncate_setsize(inode, newsize); 809 if (page) 810 page_cache_release(page); 811 shmem_truncate_range(inode, newsize, (loff_t)-1); 812 } 813 814 setattr_copy(inode, attr); 815#ifdef CONFIG_TMPFS_POSIX_ACL 816 if (attr->ia_valid & ATTR_MODE) 817 error = generic_acl_chmod(inode); 818#endif 819 return error; 820} 821 822static void shmem_evict_inode(struct inode *inode) 823{ 824 struct shmem_inode_info *info = SHMEM_I(inode); 825 826 if (inode->i_mapping->a_ops == &shmem_aops) { 827 truncate_inode_pages(inode->i_mapping, 0); 828 shmem_unacct_size(info->flags, inode->i_size); 829 inode->i_size = 0; 830 shmem_truncate_range(inode, 0, (loff_t)-1); 831 if (!list_empty(&info->swaplist)) { 832 mutex_lock(&shmem_swaplist_mutex); 833 list_del_init(&info->swaplist); 834 mutex_unlock(&shmem_swaplist_mutex); 835 } 836 } 837 BUG_ON(inode->i_blocks); 838 shmem_free_inode(inode->i_sb); 839 end_writeback(inode); 840} 841 842static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir) 843{ 844 swp_entry_t *ptr; 845 846 for (ptr = dir; ptr < edir; ptr++) { 847 if (ptr->val == entry.val) 848 return ptr - dir; 849 } 850 return -1; 851} 852 853static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page) 854{ 855 struct address_space *mapping; 856 unsigned long idx; 857 unsigned long size; 858 unsigned long limit; 859 unsigned long stage; 860 struct page **dir; 861 struct page *subdir; 862 swp_entry_t *ptr; 863 int offset; 864 int error; 865 866 idx = 0; 867 ptr = info->i_direct; 868 spin_lock(&info->lock); 869 if (!info->swapped) { 870 list_del_init(&info->swaplist); 871 goto lost2; 872 } 873 limit = info->next_index; 874 size = limit; 875 if (size > SHMEM_NR_DIRECT) 876 size = SHMEM_NR_DIRECT; 877 offset = shmem_find_swp(entry, ptr, ptr+size); 878 if (offset >= 0) { 879 shmem_swp_balance_unmap(); 880 goto found; 881 } 882 if (!info->i_indirect) 883 goto lost2; 884 885 dir = shmem_dir_map(info->i_indirect); 886 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2; 887 888 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) { 889 if (unlikely(idx == stage)) { 890 shmem_dir_unmap(dir-1); 891 if (cond_resched_lock(&info->lock)) { 892 /* check it has not been truncated */ 893 if (limit > info->next_index) { 894 limit = info->next_index; 895 if (idx >= limit) 896 goto lost2; 897 } 898 } 899 dir = shmem_dir_map(info->i_indirect) + 900 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 901 while (!*dir) { 902 dir++; 903 idx += ENTRIES_PER_PAGEPAGE; 904 if (idx >= limit) 905 goto lost1; 906 } 907 stage = idx + ENTRIES_PER_PAGEPAGE; 908 subdir = *dir; 909 shmem_dir_unmap(dir); 910 dir = shmem_dir_map(subdir); 911 } 912 subdir = *dir; 913 if (subdir && page_private(subdir)) { 914 ptr = shmem_swp_map(subdir); 915 size = limit - idx; 916 if (size > ENTRIES_PER_PAGE) 917 size = ENTRIES_PER_PAGE; 918 offset = shmem_find_swp(entry, ptr, ptr+size); 919 shmem_swp_unmap(ptr); 920 if (offset >= 0) { 921 shmem_dir_unmap(dir); 922 ptr = shmem_swp_map(subdir); 923 goto found; 924 } 925 } 926 } 927lost1: 928 shmem_dir_unmap(dir-1); 929lost2: 930 spin_unlock(&info->lock); 931 return 0; 932found: 933 idx += offset; 934 ptr += offset; 935 936 /* 937 * Move _head_ to start search for next from here. 938 * But be careful: shmem_evict_inode checks list_empty without taking 939 * mutex, and there's an instant in list_move_tail when info->swaplist 940 * would appear empty, if it were the only one on shmem_swaplist. We 941 * could avoid doing it if inode NULL; or use this minor optimization. 942 */ 943 if (shmem_swaplist.next != &info->swaplist) 944 list_move_tail(&shmem_swaplist, &info->swaplist); 945 946 /* 947 * We rely on shmem_swaplist_mutex, not only to protect the swaplist, 948 * but also to hold up shmem_evict_inode(): so inode cannot be freed 949 * beneath us (pagelock doesn't help until the page is in pagecache). 950 */ 951 mapping = info->vfs_inode.i_mapping; 952 error = add_to_page_cache_locked(page, mapping, idx, GFP_NOWAIT); 953 /* which does mem_cgroup_uncharge_cache_page on error */ 954 955 if (error == -EEXIST) { 956 struct page *filepage = find_get_page(mapping, idx); 957 error = 1; 958 if (filepage) { 959 /* 960 * There might be a more uptodate page coming down 961 * from a stacked writepage: forget our swappage if so. 962 */ 963 if (PageUptodate(filepage)) 964 error = 0; 965 page_cache_release(filepage); 966 } 967 } 968 if (!error) { 969 delete_from_swap_cache(page); 970 set_page_dirty(page); 971 info->flags |= SHMEM_PAGEIN; 972 shmem_swp_set(info, ptr, 0); 973 swap_free(entry); 974 error = 1; /* not an error, but entry was found */ 975 } 976 shmem_swp_unmap(ptr); 977 spin_unlock(&info->lock); 978 return error; 979} 980 981/* 982 * shmem_unuse() search for an eventually swapped out shmem page. 983 */ 984int shmem_unuse(swp_entry_t entry, struct page *page) 985{ 986 struct list_head *p, *next; 987 struct shmem_inode_info *info; 988 int found = 0; 989 int error; 990 991 /* 992 * Charge page using GFP_KERNEL while we can wait, before taking 993 * the shmem_swaplist_mutex which might hold up shmem_writepage(). 994 * Charged back to the user (not to caller) when swap account is used. 995 * add_to_page_cache() will be called with GFP_NOWAIT. 996 */ 997 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL); 998 if (error) 999 goto out; 1000 /* 1001 * Try to preload while we can wait, to not make a habit of 1002 * draining atomic reserves; but don't latch on to this cpu, 1003 * it's okay if sometimes we get rescheduled after this. 1004 */ 1005 error = radix_tree_preload(GFP_KERNEL); 1006 if (error) 1007 goto uncharge; 1008 radix_tree_preload_end(); 1009 1010 mutex_lock(&shmem_swaplist_mutex); 1011 list_for_each_safe(p, next, &shmem_swaplist) { 1012 info = list_entry(p, struct shmem_inode_info, swaplist); 1013 found = shmem_unuse_inode(info, entry, page); 1014 cond_resched(); 1015 if (found) 1016 break; 1017 } 1018 mutex_unlock(&shmem_swaplist_mutex); 1019 1020uncharge: 1021 if (!found) 1022 mem_cgroup_uncharge_cache_page(page); 1023 if (found < 0) 1024 error = found; 1025out: 1026 unlock_page(page); 1027 page_cache_release(page); 1028 return error; 1029} 1030 1031/* 1032 * Move the page from the page cache to the swap cache. 1033 */ 1034static int shmem_writepage(struct page *page, struct writeback_control *wbc) 1035{ 1036 struct shmem_inode_info *info; 1037 swp_entry_t *entry, swap; 1038 struct address_space *mapping; 1039 unsigned long index; 1040 struct inode *inode; 1041 1042 BUG_ON(!PageLocked(page)); 1043 mapping = page->mapping; 1044 index = page->index; 1045 inode = mapping->host; 1046 info = SHMEM_I(inode); 1047 if (info->flags & VM_LOCKED) 1048 goto redirty; 1049 if (!total_swap_pages) 1050 goto redirty; 1051 1052 /* 1053 * shmem_backing_dev_info's capabilities prevent regular writeback or 1054 * sync from ever calling shmem_writepage; but a stacking filesystem 1055 * may use the ->writepage of its underlying filesystem, in which case 1056 * tmpfs should write out to swap only in response to memory pressure, 1057 * and not for the writeback threads or sync. However, in those cases, 1058 * we do still want to check if there's a redundant swappage to be 1059 * discarded. 1060 */ 1061 if (wbc->for_reclaim) 1062 swap = get_swap_page(); 1063 else 1064 swap.val = 0; 1065 1066 /* 1067 * Add inode to shmem_unuse()'s list of swapped-out inodes, 1068 * if it's not already there. Do it now because we cannot take 1069 * mutex while holding spinlock, and must do so before the page 1070 * is moved to swap cache, when its pagelock no longer protects 1071 * the inode from eviction. But don't unlock the mutex until 1072 * we've taken the spinlock, because shmem_unuse_inode() will 1073 * prune a !swapped inode from the swaplist under both locks. 1074 */ 1075 if (swap.val) { 1076 mutex_lock(&shmem_swaplist_mutex); 1077 if (list_empty(&info->swaplist)) 1078 list_add_tail(&info->swaplist, &shmem_swaplist); 1079 } 1080 1081 spin_lock(&info->lock); 1082 if (swap.val) 1083 mutex_unlock(&shmem_swaplist_mutex); 1084 1085 if (index >= info->next_index) { 1086 BUG_ON(!(info->flags & SHMEM_TRUNCATE)); 1087 goto unlock; 1088 } 1089 entry = shmem_swp_entry(info, index, NULL); 1090 if (entry->val) { 1091 /* 1092 * The more uptodate page coming down from a stacked 1093 * writepage should replace our old swappage. 1094 */ 1095 free_swap_and_cache(*entry); 1096 shmem_swp_set(info, entry, 0); 1097 } 1098 shmem_recalc_inode(inode); 1099 1100 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { 1101 delete_from_page_cache(page); 1102 shmem_swp_set(info, entry, swap.val); 1103 shmem_swp_unmap(entry); 1104 spin_unlock(&info->lock); 1105 swap_shmem_alloc(swap); 1106 BUG_ON(page_mapped(page)); 1107 swap_writepage(page, wbc); 1108 return 0; 1109 } 1110 1111 shmem_swp_unmap(entry); 1112unlock: 1113 spin_unlock(&info->lock); 1114 /* 1115 * add_to_swap_cache() doesn't return -EEXIST, so we can safely 1116 * clear SWAP_HAS_CACHE flag. 1117 */ 1118 swapcache_free(swap, NULL); 1119redirty: 1120 set_page_dirty(page); 1121 if (wbc->for_reclaim) 1122 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ 1123 unlock_page(page); 1124 return 0; 1125} 1126 1127#ifdef CONFIG_NUMA 1128#ifdef CONFIG_TMPFS 1129static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) 1130{ 1131 char buffer[64]; 1132 1133 if (!mpol || mpol->mode == MPOL_DEFAULT) 1134 return; /* show nothing */ 1135 1136 mpol_to_str(buffer, sizeof(buffer), mpol, 1); 1137 1138 seq_printf(seq, ",mpol=%s", buffer); 1139} 1140 1141static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) 1142{ 1143 struct mempolicy *mpol = NULL; 1144 if (sbinfo->mpol) { 1145 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ 1146 mpol = sbinfo->mpol; 1147 mpol_get(mpol); 1148 spin_unlock(&sbinfo->stat_lock); 1149 } 1150 return mpol; 1151} 1152#endif /* CONFIG_TMPFS */ 1153 1154static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, 1155 struct shmem_inode_info *info, unsigned long idx) 1156{ 1157 struct mempolicy mpol, *spol; 1158 struct vm_area_struct pvma; 1159 struct page *page; 1160 1161 spol = mpol_cond_copy(&mpol, 1162 mpol_shared_policy_lookup(&info->policy, idx)); 1163 1164 /* Create a pseudo vma that just contains the policy */ 1165 pvma.vm_start = 0; 1166 pvma.vm_pgoff = idx; 1167 pvma.vm_ops = NULL; 1168 pvma.vm_policy = spol; 1169 page = swapin_readahead(entry, gfp, &pvma, 0); 1170 return page; 1171} 1172 1173static struct page *shmem_alloc_page(gfp_t gfp, 1174 struct shmem_inode_info *info, unsigned long idx) 1175{ 1176 struct vm_area_struct pvma; 1177 1178 /* Create a pseudo vma that just contains the policy */ 1179 pvma.vm_start = 0; 1180 pvma.vm_pgoff = idx; 1181 pvma.vm_ops = NULL; 1182 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx); 1183 1184 /* 1185 * alloc_page_vma() will drop the shared policy reference 1186 */ 1187 return alloc_page_vma(gfp, &pvma, 0); 1188} 1189#else /* !CONFIG_NUMA */ 1190#ifdef CONFIG_TMPFS 1191static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p) 1192{ 1193} 1194#endif /* CONFIG_TMPFS */ 1195 1196static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, 1197 struct shmem_inode_info *info, unsigned long idx) 1198{ 1199 return swapin_readahead(entry, gfp, NULL, 0); 1200} 1201 1202static inline struct page *shmem_alloc_page(gfp_t gfp, 1203 struct shmem_inode_info *info, unsigned long idx) 1204{ 1205 return alloc_page(gfp); 1206} 1207#endif /* CONFIG_NUMA */ 1208 1209#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS) 1210static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) 1211{ 1212 return NULL; 1213} 1214#endif 1215 1216/* 1217 * shmem_getpage - either get the page from swap or allocate a new one 1218 * 1219 * If we allocate a new one we do not mark it dirty. That's up to the 1220 * vm. If we swap it in we mark it dirty since we also free the swap 1221 * entry since a page cannot live in both the swap and page cache 1222 */ 1223static int shmem_getpage(struct inode *inode, unsigned long idx, 1224 struct page **pagep, enum sgp_type sgp, int *type) 1225{ 1226 struct address_space *mapping = inode->i_mapping; 1227 struct shmem_inode_info *info = SHMEM_I(inode); 1228 struct shmem_sb_info *sbinfo; 1229 struct page *filepage = *pagep; 1230 struct page *swappage; 1231 struct page *prealloc_page = NULL; 1232 swp_entry_t *entry; 1233 swp_entry_t swap; 1234 gfp_t gfp; 1235 int error; 1236 1237 if (idx >= SHMEM_MAX_INDEX) 1238 return -EFBIG; 1239 1240 if (type) 1241 *type = 0; 1242 1243 /* 1244 * Normally, filepage is NULL on entry, and either found 1245 * uptodate immediately, or allocated and zeroed, or read 1246 * in under swappage, which is then assigned to filepage. 1247 * But shmem_readpage (required for splice) passes in a locked 1248 * filepage, which may be found not uptodate by other callers 1249 * too, and may need to be copied from the swappage read in. 1250 */ 1251repeat: 1252 if (!filepage) 1253 filepage = find_lock_page(mapping, idx); 1254 if (filepage && PageUptodate(filepage)) 1255 goto done; 1256 gfp = mapping_gfp_mask(mapping); 1257 if (!filepage) { 1258 /* 1259 * Try to preload while we can wait, to not make a habit of 1260 * draining atomic reserves; but don't latch on to this cpu. 1261 */ 1262 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM); 1263 if (error) 1264 goto failed; 1265 radix_tree_preload_end(); 1266 if (sgp != SGP_READ && !prealloc_page) { 1267 /* We don't care if this fails */ 1268 prealloc_page = shmem_alloc_page(gfp, info, idx); 1269 if (prealloc_page) { 1270 if (mem_cgroup_cache_charge(prealloc_page, 1271 current->mm, GFP_KERNEL)) { 1272 page_cache_release(prealloc_page); 1273 prealloc_page = NULL; 1274 } 1275 } 1276 } 1277 } 1278 error = 0; 1279 1280 spin_lock(&info->lock); 1281 shmem_recalc_inode(inode); 1282 entry = shmem_swp_alloc(info, idx, sgp); 1283 if (IS_ERR(entry)) { 1284 spin_unlock(&info->lock); 1285 error = PTR_ERR(entry); 1286 goto failed; 1287 } 1288 swap = *entry; 1289 1290 if (swap.val) { 1291 /* Look it up and read it in.. */ 1292 swappage = lookup_swap_cache(swap); 1293 if (!swappage) { 1294 shmem_swp_unmap(entry); 1295 /* here we actually do the io */ 1296 if (type && !(*type & VM_FAULT_MAJOR)) { 1297 __count_vm_event(PGMAJFAULT); 1298 *type |= VM_FAULT_MAJOR; 1299 } 1300 spin_unlock(&info->lock); 1301 swappage = shmem_swapin(swap, gfp, info, idx); 1302 if (!swappage) { 1303 spin_lock(&info->lock); 1304 entry = shmem_swp_alloc(info, idx, sgp); 1305 if (IS_ERR(entry)) 1306 error = PTR_ERR(entry); 1307 else { 1308 if (entry->val == swap.val) 1309 error = -ENOMEM; 1310 shmem_swp_unmap(entry); 1311 } 1312 spin_unlock(&info->lock); 1313 if (error) 1314 goto failed; 1315 goto repeat; 1316 } 1317 wait_on_page_locked(swappage); 1318 page_cache_release(swappage); 1319 goto repeat; 1320 } 1321 1322 /* We have to do this with page locked to prevent races */ 1323 if (!trylock_page(swappage)) { 1324 shmem_swp_unmap(entry); 1325 spin_unlock(&info->lock); 1326 wait_on_page_locked(swappage); 1327 page_cache_release(swappage); 1328 goto repeat; 1329 } 1330 if (PageWriteback(swappage)) { 1331 shmem_swp_unmap(entry); 1332 spin_unlock(&info->lock); 1333 wait_on_page_writeback(swappage); 1334 unlock_page(swappage); 1335 page_cache_release(swappage); 1336 goto repeat; 1337 } 1338 if (!PageUptodate(swappage)) { 1339 shmem_swp_unmap(entry); 1340 spin_unlock(&info->lock); 1341 unlock_page(swappage); 1342 page_cache_release(swappage); 1343 error = -EIO; 1344 goto failed; 1345 } 1346 1347 if (filepage) { 1348 shmem_swp_set(info, entry, 0); 1349 shmem_swp_unmap(entry); 1350 delete_from_swap_cache(swappage); 1351 spin_unlock(&info->lock); 1352 copy_highpage(filepage, swappage); 1353 unlock_page(swappage); 1354 page_cache_release(swappage); 1355 flush_dcache_page(filepage); 1356 SetPageUptodate(filepage); 1357 set_page_dirty(filepage); 1358 swap_free(swap); 1359 } else if (!(error = add_to_page_cache_locked(swappage, mapping, 1360 idx, GFP_NOWAIT))) { 1361 info->flags |= SHMEM_PAGEIN; 1362 shmem_swp_set(info, entry, 0); 1363 shmem_swp_unmap(entry); 1364 delete_from_swap_cache(swappage); 1365 spin_unlock(&info->lock); 1366 filepage = swappage; 1367 set_page_dirty(filepage); 1368 swap_free(swap); 1369 } else { 1370 shmem_swp_unmap(entry); 1371 spin_unlock(&info->lock); 1372 if (error == -ENOMEM) { 1373 /* 1374 * reclaim from proper memory cgroup and 1375 * call memcg's OOM if needed. 1376 */ 1377 error = mem_cgroup_shmem_charge_fallback( 1378 swappage, 1379 current->mm, 1380 gfp); 1381 if (error) { 1382 unlock_page(swappage); 1383 page_cache_release(swappage); 1384 goto failed; 1385 } 1386 } 1387 unlock_page(swappage); 1388 page_cache_release(swappage); 1389 goto repeat; 1390 } 1391 } else if (sgp == SGP_READ && !filepage) { 1392 shmem_swp_unmap(entry); 1393 filepage = find_get_page(mapping, idx); 1394 if (filepage && 1395 (!PageUptodate(filepage) || !trylock_page(filepage))) { 1396 spin_unlock(&info->lock); 1397 wait_on_page_locked(filepage); 1398 page_cache_release(filepage); 1399 filepage = NULL; 1400 goto repeat; 1401 } 1402 spin_unlock(&info->lock); 1403 } else { 1404 shmem_swp_unmap(entry); 1405 sbinfo = SHMEM_SB(inode->i_sb); 1406 if (sbinfo->max_blocks) { 1407 if (percpu_counter_compare(&sbinfo->used_blocks, 1408 sbinfo->max_blocks) >= 0 || 1409 shmem_acct_block(info->flags)) 1410 goto nospace; 1411 percpu_counter_inc(&sbinfo->used_blocks); 1412 spin_lock(&inode->i_lock); 1413 inode->i_blocks += BLOCKS_PER_PAGE; 1414 spin_unlock(&inode->i_lock); 1415 } else if (shmem_acct_block(info->flags)) 1416 goto nospace; 1417 1418 if (!filepage) { 1419 int ret; 1420 1421 if (!prealloc_page) { 1422 spin_unlock(&info->lock); 1423 filepage = shmem_alloc_page(gfp, info, idx); 1424 if (!filepage) { 1425 shmem_unacct_blocks(info->flags, 1); 1426 shmem_free_blocks(inode, 1); 1427 error = -ENOMEM; 1428 goto failed; 1429 } 1430 SetPageSwapBacked(filepage); 1431 1432 /* 1433 * Precharge page while we can wait, compensate 1434 * after 1435 */ 1436 error = mem_cgroup_cache_charge(filepage, 1437 current->mm, GFP_KERNEL); 1438 if (error) { 1439 page_cache_release(filepage); 1440 shmem_unacct_blocks(info->flags, 1); 1441 shmem_free_blocks(inode, 1); 1442 filepage = NULL; 1443 goto failed; 1444 } 1445 1446 spin_lock(&info->lock); 1447 } else { 1448 filepage = prealloc_page; 1449 prealloc_page = NULL; 1450 SetPageSwapBacked(filepage); 1451 } 1452 1453 entry = shmem_swp_alloc(info, idx, sgp); 1454 if (IS_ERR(entry)) 1455 error = PTR_ERR(entry); 1456 else { 1457 swap = *entry; 1458 shmem_swp_unmap(entry); 1459 } 1460 ret = error || swap.val; 1461 if (ret) 1462 mem_cgroup_uncharge_cache_page(filepage); 1463 else 1464 ret = add_to_page_cache_lru(filepage, mapping, 1465 idx, GFP_NOWAIT); 1466 /* 1467 * At add_to_page_cache_lru() failure, uncharge will 1468 * be done automatically. 1469 */ 1470 if (ret) { 1471 spin_unlock(&info->lock); 1472 page_cache_release(filepage); 1473 shmem_unacct_blocks(info->flags, 1); 1474 shmem_free_blocks(inode, 1); 1475 filepage = NULL; 1476 if (error) 1477 goto failed; 1478 goto repeat; 1479 } 1480 info->flags |= SHMEM_PAGEIN; 1481 } 1482 1483 info->alloced++; 1484 spin_unlock(&info->lock); 1485 clear_highpage(filepage); 1486 flush_dcache_page(filepage); 1487 SetPageUptodate(filepage); 1488 if (sgp == SGP_DIRTY) 1489 set_page_dirty(filepage); 1490 } 1491done: 1492 *pagep = filepage; 1493 error = 0; 1494 goto out; 1495 1496nospace: 1497 /* 1498 * Perhaps the page was brought in from swap between find_lock_page 1499 * and taking info->lock? We allow for that at add_to_page_cache_lru, 1500 * but must also avoid reporting a spurious ENOSPC while working on a 1501 * full tmpfs. (When filepage has been passed in to shmem_getpage, it 1502 * is already in page cache, which prevents this race from occurring.) 1503 */ 1504 if (!filepage) { 1505 struct page *page = find_get_page(mapping, idx); 1506 if (page) { 1507 spin_unlock(&info->lock); 1508 page_cache_release(page); 1509 goto repeat; 1510 } 1511 } 1512 spin_unlock(&info->lock); 1513 error = -ENOSPC; 1514failed: 1515 if (*pagep != filepage) { 1516 unlock_page(filepage); 1517 page_cache_release(filepage); 1518 } 1519out: 1520 if (prealloc_page) { 1521 mem_cgroup_uncharge_cache_page(prealloc_page); 1522 page_cache_release(prealloc_page); 1523 } 1524 return error; 1525} 1526 1527static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1528{ 1529 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 1530 int error; 1531 int ret; 1532 1533 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 1534 return VM_FAULT_SIGBUS; 1535 1536 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret); 1537 if (error) 1538 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); 1539 1540 return ret | VM_FAULT_LOCKED; 1541} 1542 1543#ifdef CONFIG_NUMA 1544static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new) 1545{ 1546 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1547 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new); 1548} 1549 1550static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, 1551 unsigned long addr) 1552{ 1553 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1554 unsigned long idx; 1555 1556 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; 1557 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx); 1558} 1559#endif 1560 1561int shmem_lock(struct file *file, int lock, struct user_struct *user) 1562{ 1563 struct inode *inode = file->f_path.dentry->d_inode; 1564 struct shmem_inode_info *info = SHMEM_I(inode); 1565 int retval = -ENOMEM; 1566 1567 spin_lock(&info->lock); 1568 if (lock && !(info->flags & VM_LOCKED)) { 1569 if (!user_shm_lock(inode->i_size, user)) 1570 goto out_nomem; 1571 info->flags |= VM_LOCKED; 1572 mapping_set_unevictable(file->f_mapping); 1573 } 1574 if (!lock && (info->flags & VM_LOCKED) && user) { 1575 user_shm_unlock(inode->i_size, user); 1576 info->flags &= ~VM_LOCKED; 1577 mapping_clear_unevictable(file->f_mapping); 1578 scan_mapping_unevictable_pages(file->f_mapping); 1579 } 1580 retval = 0; 1581 1582out_nomem: 1583 spin_unlock(&info->lock); 1584 return retval; 1585} 1586 1587static int shmem_mmap(struct file *file, struct vm_area_struct *vma) 1588{ 1589 file_accessed(file); 1590 vma->vm_ops = &shmem_vm_ops; 1591 vma->vm_flags |= VM_CAN_NONLINEAR; 1592 return 0; 1593} 1594 1595static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir, 1596 int mode, dev_t dev, unsigned long flags) 1597{ 1598 struct inode *inode; 1599 struct shmem_inode_info *info; 1600 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 1601 1602 if (shmem_reserve_inode(sb)) 1603 return NULL; 1604 1605 inode = new_inode(sb); 1606 if (inode) { 1607 inode->i_ino = get_next_ino(); 1608 inode_init_owner(inode, dir, mode); 1609 inode->i_blocks = 0; 1610 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; 1611 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 1612 inode->i_generation = get_seconds(); 1613 info = SHMEM_I(inode); 1614 memset(info, 0, (char *)inode - (char *)info); 1615 spin_lock_init(&info->lock); 1616 info->flags = flags & VM_NORESERVE; 1617 INIT_LIST_HEAD(&info->swaplist); 1618 cache_no_acl(inode); 1619 1620 switch (mode & S_IFMT) { 1621 default: 1622 inode->i_op = &shmem_special_inode_operations; 1623 init_special_inode(inode, mode, dev); 1624 break; 1625 case S_IFREG: 1626 inode->i_mapping->a_ops = &shmem_aops; 1627 inode->i_op = &shmem_inode_operations; 1628 inode->i_fop = &shmem_file_operations; 1629 mpol_shared_policy_init(&info->policy, 1630 shmem_get_sbmpol(sbinfo)); 1631 break; 1632 case S_IFDIR: 1633 inc_nlink(inode); 1634 /* Some things misbehave if size == 0 on a directory */ 1635 inode->i_size = 2 * BOGO_DIRENT_SIZE; 1636 inode->i_op = &shmem_dir_inode_operations; 1637 inode->i_fop = &simple_dir_operations; 1638 break; 1639 case S_IFLNK: 1640 /* 1641 * Must not load anything in the rbtree, 1642 * mpol_free_shared_policy will not be called. 1643 */ 1644 mpol_shared_policy_init(&info->policy, NULL); 1645 break; 1646 } 1647 } else 1648 shmem_free_inode(sb); 1649 return inode; 1650} 1651 1652#ifdef CONFIG_TMPFS 1653static const struct inode_operations shmem_symlink_inode_operations; 1654static const struct inode_operations shmem_symlink_inline_operations; 1655 1656/* 1657 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin; 1658 * but providing them allows a tmpfs file to be used for splice, sendfile, and 1659 * below the loop driver, in the generic fashion that many filesystems support. 1660 */ 1661static int shmem_readpage(struct file *file, struct page *page) 1662{ 1663 struct inode *inode = page->mapping->host; 1664 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL); 1665 unlock_page(page); 1666 return error; 1667} 1668 1669static int 1670shmem_write_begin(struct file *file, struct address_space *mapping, 1671 loff_t pos, unsigned len, unsigned flags, 1672 struct page **pagep, void **fsdata) 1673{ 1674 struct inode *inode = mapping->host; 1675 pgoff_t index = pos >> PAGE_CACHE_SHIFT; 1676 *pagep = NULL; 1677 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL); 1678} 1679 1680static int 1681shmem_write_end(struct file *file, struct address_space *mapping, 1682 loff_t pos, unsigned len, unsigned copied, 1683 struct page *page, void *fsdata) 1684{ 1685 struct inode *inode = mapping->host; 1686 1687 if (pos + copied > inode->i_size) 1688 i_size_write(inode, pos + copied); 1689 1690 set_page_dirty(page); 1691 unlock_page(page); 1692 page_cache_release(page); 1693 1694 return copied; 1695} 1696 1697static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor) 1698{ 1699 struct inode *inode = filp->f_path.dentry->d_inode; 1700 struct address_space *mapping = inode->i_mapping; 1701 unsigned long index, offset; 1702 enum sgp_type sgp = SGP_READ; 1703 1704 /* 1705 * Might this read be for a stacking filesystem? Then when reading 1706 * holes of a sparse file, we actually need to allocate those pages, 1707 * and even mark them dirty, so it cannot exceed the max_blocks limit. 1708 */ 1709 if (segment_eq(get_fs(), KERNEL_DS)) 1710 sgp = SGP_DIRTY; 1711 1712 index = *ppos >> PAGE_CACHE_SHIFT; 1713 offset = *ppos & ~PAGE_CACHE_MASK; 1714 1715 for (;;) { 1716 struct page *page = NULL; 1717 unsigned long end_index, nr, ret; 1718 loff_t i_size = i_size_read(inode); 1719 1720 end_index = i_size >> PAGE_CACHE_SHIFT; 1721 if (index > end_index) 1722 break; 1723 if (index == end_index) { 1724 nr = i_size & ~PAGE_CACHE_MASK; 1725 if (nr <= offset) 1726 break; 1727 } 1728 1729 desc->error = shmem_getpage(inode, index, &page, sgp, NULL); 1730 if (desc->error) { 1731 if (desc->error == -EINVAL) 1732 desc->error = 0; 1733 break; 1734 } 1735 if (page) 1736 unlock_page(page); 1737 1738 /* 1739 * We must evaluate after, since reads (unlike writes) 1740 * are called without i_mutex protection against truncate 1741 */ 1742 nr = PAGE_CACHE_SIZE; 1743 i_size = i_size_read(inode); 1744 end_index = i_size >> PAGE_CACHE_SHIFT; 1745 if (index == end_index) { 1746 nr = i_size & ~PAGE_CACHE_MASK; 1747 if (nr <= offset) { 1748 if (page) 1749 page_cache_release(page); 1750 break; 1751 } 1752 } 1753 nr -= offset; 1754 1755 if (page) { 1756 /* 1757 * If users can be writing to this page using arbitrary 1758 * virtual addresses, take care about potential aliasing 1759 * before reading the page on the kernel side. 1760 */ 1761 if (mapping_writably_mapped(mapping)) 1762 flush_dcache_page(page); 1763 /* 1764 * Mark the page accessed if we read the beginning. 1765 */ 1766 if (!offset) 1767 mark_page_accessed(page); 1768 } else { 1769 page = ZERO_PAGE(0); 1770 page_cache_get(page); 1771 } 1772 1773 /* 1774 * Ok, we have the page, and it's up-to-date, so 1775 * now we can copy it to user space... 1776 * 1777 * The actor routine returns how many bytes were actually used.. 1778 * NOTE! This may not be the same as how much of a user buffer 1779 * we filled up (we may be padding etc), so we can only update 1780 * "pos" here (the actor routine has to update the user buffer 1781 * pointers and the remaining count). 1782 */ 1783 ret = actor(desc, page, offset, nr); 1784 offset += ret; 1785 index += offset >> PAGE_CACHE_SHIFT; 1786 offset &= ~PAGE_CACHE_MASK; 1787 1788 page_cache_release(page); 1789 if (ret != nr || !desc->count) 1790 break; 1791 1792 cond_resched(); 1793 } 1794 1795 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; 1796 file_accessed(filp); 1797} 1798 1799static ssize_t shmem_file_aio_read(struct kiocb *iocb, 1800 const struct iovec *iov, unsigned long nr_segs, loff_t pos) 1801{ 1802 struct file *filp = iocb->ki_filp; 1803 ssize_t retval; 1804 unsigned long seg; 1805 size_t count; 1806 loff_t *ppos = &iocb->ki_pos; 1807 1808 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); 1809 if (retval) 1810 return retval; 1811 1812 for (seg = 0; seg < nr_segs; seg++) { 1813 read_descriptor_t desc; 1814 1815 desc.written = 0; 1816 desc.arg.buf = iov[seg].iov_base; 1817 desc.count = iov[seg].iov_len; 1818 if (desc.count == 0) 1819 continue; 1820 desc.error = 0; 1821 do_shmem_file_read(filp, ppos, &desc, file_read_actor); 1822 retval += desc.written; 1823 if (desc.error) { 1824 retval = retval ?: desc.error; 1825 break; 1826 } 1827 if (desc.count > 0) 1828 break; 1829 } 1830 return retval; 1831} 1832 1833static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) 1834{ 1835 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); 1836 1837 buf->f_type = TMPFS_MAGIC; 1838 buf->f_bsize = PAGE_CACHE_SIZE; 1839 buf->f_namelen = NAME_MAX; 1840 if (sbinfo->max_blocks) { 1841 buf->f_blocks = sbinfo->max_blocks; 1842 buf->f_bavail = buf->f_bfree = 1843 sbinfo->max_blocks - percpu_counter_sum(&sbinfo->used_blocks); 1844 } 1845 if (sbinfo->max_inodes) { 1846 buf->f_files = sbinfo->max_inodes; 1847 buf->f_ffree = sbinfo->free_inodes; 1848 } 1849 /* else leave those fields 0 like simple_statfs */ 1850 return 0; 1851} 1852 1853/* 1854 * File creation. Allocate an inode, and we're done.. 1855 */ 1856static int 1857shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 1858{ 1859 struct inode *inode; 1860 int error = -ENOSPC; 1861 1862 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE); 1863 if (inode) { 1864 error = security_inode_init_security(inode, dir, 1865 &dentry->d_name, NULL, 1866 NULL, NULL); 1867 if (error) { 1868 if (error != -EOPNOTSUPP) { 1869 iput(inode); 1870 return error; 1871 } 1872 } 1873#ifdef CONFIG_TMPFS_POSIX_ACL 1874 error = generic_acl_init(inode, dir); 1875 if (error) { 1876 iput(inode); 1877 return error; 1878 } 1879#else 1880 error = 0; 1881#endif 1882 dir->i_size += BOGO_DIRENT_SIZE; 1883 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1884 d_instantiate(dentry, inode); 1885 dget(dentry); /* Extra count - pin the dentry in core */ 1886 } 1887 return error; 1888} 1889 1890static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1891{ 1892 int error; 1893 1894 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) 1895 return error; 1896 inc_nlink(dir); 1897 return 0; 1898} 1899 1900static int shmem_create(struct inode *dir, struct dentry *dentry, int mode, 1901 struct nameidata *nd) 1902{ 1903 return shmem_mknod(dir, dentry, mode | S_IFREG, 0); 1904} 1905 1906/* 1907 * Link a file.. 1908 */ 1909static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1910{ 1911 struct inode *inode = old_dentry->d_inode; 1912 int ret; 1913 1914 /* 1915 * No ordinary (disk based) filesystem counts links as inodes; 1916 * but each new link needs a new dentry, pinning lowmem, and 1917 * tmpfs dentries cannot be pruned until they are unlinked. 1918 */ 1919 ret = shmem_reserve_inode(inode->i_sb); 1920 if (ret) 1921 goto out; 1922 1923 dir->i_size += BOGO_DIRENT_SIZE; 1924 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1925 inc_nlink(inode); 1926 ihold(inode); /* New dentry reference */ 1927 dget(dentry); /* Extra pinning count for the created dentry */ 1928 d_instantiate(dentry, inode); 1929out: 1930 return ret; 1931} 1932 1933static int shmem_unlink(struct inode *dir, struct dentry *dentry) 1934{ 1935 struct inode *inode = dentry->d_inode; 1936 1937 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) 1938 shmem_free_inode(inode->i_sb); 1939 1940 dir->i_size -= BOGO_DIRENT_SIZE; 1941 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1942 drop_nlink(inode); 1943 dput(dentry); /* Undo the count from "create" - this does all the work */ 1944 return 0; 1945} 1946 1947static int shmem_rmdir(struct inode *dir, struct dentry *dentry) 1948{ 1949 if (!simple_empty(dentry)) 1950 return -ENOTEMPTY; 1951 1952 drop_nlink(dentry->d_inode); 1953 drop_nlink(dir); 1954 return shmem_unlink(dir, dentry); 1955} 1956 1957/* 1958 * The VFS layer already does all the dentry stuff for rename, 1959 * we just have to decrement the usage count for the target if 1960 * it exists so that the VFS layer correctly free's it when it 1961 * gets overwritten. 1962 */ 1963static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) 1964{ 1965 struct inode *inode = old_dentry->d_inode; 1966 int they_are_dirs = S_ISDIR(inode->i_mode); 1967 1968 if (!simple_empty(new_dentry)) 1969 return -ENOTEMPTY; 1970 1971 if (new_dentry->d_inode) { 1972 (void) shmem_unlink(new_dir, new_dentry); 1973 if (they_are_dirs) 1974 drop_nlink(old_dir); 1975 } else if (they_are_dirs) { 1976 drop_nlink(old_dir); 1977 inc_nlink(new_dir); 1978 } 1979 1980 old_dir->i_size -= BOGO_DIRENT_SIZE; 1981 new_dir->i_size += BOGO_DIRENT_SIZE; 1982 old_dir->i_ctime = old_dir->i_mtime = 1983 new_dir->i_ctime = new_dir->i_mtime = 1984 inode->i_ctime = CURRENT_TIME; 1985 return 0; 1986} 1987 1988static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1989{ 1990 int error; 1991 int len; 1992 struct inode *inode; 1993 struct page *page = NULL; 1994 char *kaddr; 1995 struct shmem_inode_info *info; 1996 1997 len = strlen(symname) + 1; 1998 if (len > PAGE_CACHE_SIZE) 1999 return -ENAMETOOLONG; 2000 2001 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE); 2002 if (!inode) 2003 return -ENOSPC; 2004 2005 error = security_inode_init_security(inode, dir, &dentry->d_name, NULL, 2006 NULL, NULL); 2007 if (error) { 2008 if (error != -EOPNOTSUPP) { 2009 iput(inode); 2010 return error; 2011 } 2012 error = 0; 2013 } 2014 2015 info = SHMEM_I(inode); 2016 inode->i_size = len-1; 2017 if (len <= (char *)inode - (char *)info) { 2018 /* do it inline */ 2019 memcpy(info, symname, len); 2020 inode->i_op = &shmem_symlink_inline_operations; 2021 } else { 2022 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); 2023 if (error) { 2024 iput(inode); 2025 return error; 2026 } 2027 inode->i_mapping->a_ops = &shmem_aops; 2028 inode->i_op = &shmem_symlink_inode_operations; 2029 kaddr = kmap_atomic(page, KM_USER0); 2030 memcpy(kaddr, symname, len); 2031 kunmap_atomic(kaddr, KM_USER0); 2032 set_page_dirty(page); 2033 unlock_page(page); 2034 page_cache_release(page); 2035 } 2036 dir->i_size += BOGO_DIRENT_SIZE; 2037 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 2038 d_instantiate(dentry, inode); 2039 dget(dentry); 2040 return 0; 2041} 2042 2043static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd) 2044{ 2045 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode)); 2046 return NULL; 2047} 2048 2049static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) 2050{ 2051 struct page *page = NULL; 2052 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); 2053 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page)); 2054 if (page) 2055 unlock_page(page); 2056 return page; 2057} 2058 2059static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 2060{ 2061 if (!IS_ERR(nd_get_link(nd))) { 2062 struct page *page = cookie; 2063 kunmap(page); 2064 mark_page_accessed(page); 2065 page_cache_release(page); 2066 } 2067} 2068 2069static const struct inode_operations shmem_symlink_inline_operations = { 2070 .readlink = generic_readlink, 2071 .follow_link = shmem_follow_link_inline, 2072}; 2073 2074static const struct inode_operations shmem_symlink_inode_operations = { 2075 .readlink = generic_readlink, 2076 .follow_link = shmem_follow_link, 2077 .put_link = shmem_put_link, 2078}; 2079 2080#ifdef CONFIG_TMPFS_POSIX_ACL 2081/* 2082 * Superblocks without xattr inode operations will get security.* xattr 2083 * support from the VFS "for free". As soon as we have any other xattrs 2084 * like ACLs, we also need to implement the security.* handlers at 2085 * filesystem level, though. 2086 */ 2087 2088static size_t shmem_xattr_security_list(struct dentry *dentry, char *list, 2089 size_t list_len, const char *name, 2090 size_t name_len, int handler_flags) 2091{ 2092 return security_inode_listsecurity(dentry->d_inode, list, list_len); 2093} 2094 2095static int shmem_xattr_security_get(struct dentry *dentry, const char *name, 2096 void *buffer, size_t size, int handler_flags) 2097{ 2098 if (strcmp(name, "") == 0) 2099 return -EINVAL; 2100 return xattr_getsecurity(dentry->d_inode, name, buffer, size); 2101} 2102 2103static int shmem_xattr_security_set(struct dentry *dentry, const char *name, 2104 const void *value, size_t size, int flags, int handler_flags) 2105{ 2106 if (strcmp(name, "") == 0) 2107 return -EINVAL; 2108 return security_inode_setsecurity(dentry->d_inode, name, value, 2109 size, flags); 2110} 2111 2112static const struct xattr_handler shmem_xattr_security_handler = { 2113 .prefix = XATTR_SECURITY_PREFIX, 2114 .list = shmem_xattr_security_list, 2115 .get = shmem_xattr_security_get, 2116 .set = shmem_xattr_security_set, 2117}; 2118 2119static const struct xattr_handler *shmem_xattr_handlers[] = { 2120 &generic_acl_access_handler, 2121 &generic_acl_default_handler, 2122 &shmem_xattr_security_handler, 2123 NULL 2124}; 2125#endif 2126 2127static struct dentry *shmem_get_parent(struct dentry *child) 2128{ 2129 return ERR_PTR(-ESTALE); 2130} 2131 2132static int shmem_match(struct inode *ino, void *vfh) 2133{ 2134 __u32 *fh = vfh; 2135 __u64 inum = fh[2]; 2136 inum = (inum << 32) | fh[1]; 2137 return ino->i_ino == inum && fh[0] == ino->i_generation; 2138} 2139 2140static struct dentry *shmem_fh_to_dentry(struct super_block *sb, 2141 struct fid *fid, int fh_len, int fh_type) 2142{ 2143 struct inode *inode; 2144 struct dentry *dentry = NULL; 2145 u64 inum = fid->raw[2]; 2146 inum = (inum << 32) | fid->raw[1]; 2147 2148 if (fh_len < 3) 2149 return NULL; 2150 2151 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), 2152 shmem_match, fid->raw); 2153 if (inode) { 2154 dentry = d_find_alias(inode); 2155 iput(inode); 2156 } 2157 2158 return dentry; 2159} 2160 2161static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len, 2162 int connectable) 2163{ 2164 struct inode *inode = dentry->d_inode; 2165 2166 if (*len < 3) { 2167 *len = 3; 2168 return 255; 2169 } 2170 2171 if (inode_unhashed(inode)) { 2172 /* Unfortunately insert_inode_hash is not idempotent, 2173 * so as we hash inodes here rather than at creation 2174 * time, we need a lock to ensure we only try 2175 * to do it once 2176 */ 2177 static DEFINE_SPINLOCK(lock); 2178 spin_lock(&lock); 2179 if (inode_unhashed(inode)) 2180 __insert_inode_hash(inode, 2181 inode->i_ino + inode->i_generation); 2182 spin_unlock(&lock); 2183 } 2184 2185 fh[0] = inode->i_generation; 2186 fh[1] = inode->i_ino; 2187 fh[2] = ((__u64)inode->i_ino) >> 32; 2188 2189 *len = 3; 2190 return 1; 2191} 2192 2193static const struct export_operations shmem_export_ops = { 2194 .get_parent = shmem_get_parent, 2195 .encode_fh = shmem_encode_fh, 2196 .fh_to_dentry = shmem_fh_to_dentry, 2197}; 2198 2199static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, 2200 bool remount) 2201{ 2202 char *this_char, *value, *rest; 2203 2204 while (options != NULL) { 2205 this_char = options; 2206 for (;;) { 2207 /* 2208 * NUL-terminate this option: unfortunately, 2209 * mount options form a comma-separated list, 2210 * but mpol's nodelist may also contain commas. 2211 */ 2212 options = strchr(options, ','); 2213 if (options == NULL) 2214 break; 2215 options++; 2216 if (!isdigit(*options)) { 2217 options[-1] = '\0'; 2218 break; 2219 } 2220 } 2221 if (!*this_char) 2222 continue; 2223 if ((value = strchr(this_char,'=')) != NULL) { 2224 *value++ = 0; 2225 } else { 2226 printk(KERN_ERR 2227 "tmpfs: No value for mount option '%s'\n", 2228 this_char); 2229 return 1; 2230 } 2231 2232 if (!strcmp(this_char,"size")) { 2233 unsigned long long size; 2234 size = memparse(value,&rest); 2235 if (*rest == '%') { 2236 size <<= PAGE_SHIFT; 2237 size *= totalram_pages; 2238 do_div(size, 100); 2239 rest++; 2240 } 2241 if (*rest) 2242 goto bad_val; 2243 sbinfo->max_blocks = 2244 DIV_ROUND_UP(size, PAGE_CACHE_SIZE); 2245 } else if (!strcmp(this_char,"nr_blocks")) { 2246 sbinfo->max_blocks = memparse(value, &rest); 2247 if (*rest) 2248 goto bad_val; 2249 } else if (!strcmp(this_char,"nr_inodes")) { 2250 sbinfo->max_inodes = memparse(value, &rest); 2251 if (*rest) 2252 goto bad_val; 2253 } else if (!strcmp(this_char,"mode")) { 2254 if (remount) 2255 continue; 2256 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; 2257 if (*rest) 2258 goto bad_val; 2259 } else if (!strcmp(this_char,"uid")) { 2260 if (remount) 2261 continue; 2262 sbinfo->uid = simple_strtoul(value, &rest, 0); 2263 if (*rest) 2264 goto bad_val; 2265 } else if (!strcmp(this_char,"gid")) { 2266 if (remount) 2267 continue; 2268 sbinfo->gid = simple_strtoul(value, &rest, 0); 2269 if (*rest) 2270 goto bad_val; 2271 } else if (!strcmp(this_char,"mpol")) { 2272 if (mpol_parse_str(value, &sbinfo->mpol, 1)) 2273 goto bad_val; 2274 } else { 2275 printk(KERN_ERR "tmpfs: Bad mount option %s\n", 2276 this_char); 2277 return 1; 2278 } 2279 } 2280 return 0; 2281 2282bad_val: 2283 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", 2284 value, this_char); 2285 return 1; 2286 2287} 2288 2289static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) 2290{ 2291 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 2292 struct shmem_sb_info config = *sbinfo; 2293 unsigned long inodes; 2294 int error = -EINVAL; 2295 2296 if (shmem_parse_options(data, &config, true)) 2297 return error; 2298 2299 spin_lock(&sbinfo->stat_lock); 2300 inodes = sbinfo->max_inodes - sbinfo->free_inodes; 2301 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0) 2302 goto out; 2303 if (config.max_inodes < inodes) 2304 goto out; 2305 /* 2306 * Those tests also disallow limited->unlimited while any are in 2307 * use, so i_blocks will always be zero when max_blocks is zero; 2308 * but we must separately disallow unlimited->limited, because 2309 * in that case we have no record of how much is already in use. 2310 */ 2311 if (config.max_blocks && !sbinfo->max_blocks) 2312 goto out; 2313 if (config.max_inodes && !sbinfo->max_inodes) 2314 goto out; 2315 2316 error = 0; 2317 sbinfo->max_blocks = config.max_blocks; 2318 sbinfo->max_inodes = config.max_inodes; 2319 sbinfo->free_inodes = config.max_inodes - inodes; 2320 2321 mpol_put(sbinfo->mpol); 2322 sbinfo->mpol = config.mpol; /* transfers initial ref */ 2323out: 2324 spin_unlock(&sbinfo->stat_lock); 2325 return error; 2326} 2327 2328static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs) 2329{ 2330 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb); 2331 2332 if (sbinfo->max_blocks != shmem_default_max_blocks()) 2333 seq_printf(seq, ",size=%luk", 2334 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10)); 2335 if (sbinfo->max_inodes != shmem_default_max_inodes()) 2336 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); 2337 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) 2338 seq_printf(seq, ",mode=%03o", sbinfo->mode); 2339 if (sbinfo->uid != 0) 2340 seq_printf(seq, ",uid=%u", sbinfo->uid); 2341 if (sbinfo->gid != 0) 2342 seq_printf(seq, ",gid=%u", sbinfo->gid); 2343 shmem_show_mpol(seq, sbinfo->mpol); 2344 return 0; 2345} 2346#endif /* CONFIG_TMPFS */ 2347 2348static void shmem_put_super(struct super_block *sb) 2349{ 2350 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 2351 2352 percpu_counter_destroy(&sbinfo->used_blocks); 2353 kfree(sbinfo); 2354 sb->s_fs_info = NULL; 2355} 2356 2357int shmem_fill_super(struct super_block *sb, void *data, int silent) 2358{ 2359 struct inode *inode; 2360 struct dentry *root; 2361 struct shmem_sb_info *sbinfo; 2362 int err = -ENOMEM; 2363 2364 /* Round up to L1_CACHE_BYTES to resist false sharing */ 2365 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), 2366 L1_CACHE_BYTES), GFP_KERNEL); 2367 if (!sbinfo) 2368 return -ENOMEM; 2369 2370 sbinfo->mode = S_IRWXUGO | S_ISVTX; 2371 sbinfo->uid = current_fsuid(); 2372 sbinfo->gid = current_fsgid(); 2373 sb->s_fs_info = sbinfo; 2374 2375#ifdef CONFIG_TMPFS 2376 /* 2377 * Per default we only allow half of the physical ram per 2378 * tmpfs instance, limiting inodes to one per page of lowmem; 2379 * but the internal instance is left unlimited. 2380 */ 2381 if (!(sb->s_flags & MS_NOUSER)) { 2382 sbinfo->max_blocks = shmem_default_max_blocks(); 2383 sbinfo->max_inodes = shmem_default_max_inodes(); 2384 if (shmem_parse_options(data, sbinfo, false)) { 2385 err = -EINVAL; 2386 goto failed; 2387 } 2388 } 2389 sb->s_export_op = &shmem_export_ops; 2390#else 2391 sb->s_flags |= MS_NOUSER; 2392#endif 2393 2394 spin_lock_init(&sbinfo->stat_lock); 2395 if (percpu_counter_init(&sbinfo->used_blocks, 0)) 2396 goto failed; 2397 sbinfo->free_inodes = sbinfo->max_inodes; 2398 2399 sb->s_maxbytes = SHMEM_MAX_BYTES; 2400 sb->s_blocksize = PAGE_CACHE_SIZE; 2401 sb->s_blocksize_bits = PAGE_CACHE_SHIFT; 2402 sb->s_magic = TMPFS_MAGIC; 2403 sb->s_op = &shmem_ops; 2404 sb->s_time_gran = 1; 2405#ifdef CONFIG_TMPFS_POSIX_ACL 2406 sb->s_xattr = shmem_xattr_handlers; 2407 sb->s_flags |= MS_POSIXACL; 2408#endif 2409 2410 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); 2411 if (!inode) 2412 goto failed; 2413 inode->i_uid = sbinfo->uid; 2414 inode->i_gid = sbinfo->gid; 2415 root = d_alloc_root(inode); 2416 if (!root) 2417 goto failed_iput; 2418 sb->s_root = root; 2419 return 0; 2420 2421failed_iput: 2422 iput(inode); 2423failed: 2424 shmem_put_super(sb); 2425 return err; 2426} 2427 2428static struct kmem_cache *shmem_inode_cachep; 2429 2430static struct inode *shmem_alloc_inode(struct super_block *sb) 2431{ 2432 struct shmem_inode_info *p; 2433 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); 2434 if (!p) 2435 return NULL; 2436 return &p->vfs_inode; 2437} 2438 2439static void shmem_i_callback(struct rcu_head *head) 2440{ 2441 struct inode *inode = container_of(head, struct inode, i_rcu); 2442 INIT_LIST_HEAD(&inode->i_dentry); 2443 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); 2444} 2445 2446static void shmem_destroy_inode(struct inode *inode) 2447{ 2448 if ((inode->i_mode & S_IFMT) == S_IFREG) { 2449 /* only struct inode is valid if it's an inline symlink */ 2450 mpol_free_shared_policy(&SHMEM_I(inode)->policy); 2451 } 2452 call_rcu(&inode->i_rcu, shmem_i_callback); 2453} 2454 2455static void init_once(void *foo) 2456{ 2457 struct shmem_inode_info *p = (struct shmem_inode_info *) foo; 2458 2459 inode_init_once(&p->vfs_inode); 2460} 2461 2462static int init_inodecache(void) 2463{ 2464 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", 2465 sizeof(struct shmem_inode_info), 2466 0, SLAB_PANIC, init_once); 2467 return 0; 2468} 2469 2470static void destroy_inodecache(void) 2471{ 2472 kmem_cache_destroy(shmem_inode_cachep); 2473} 2474 2475static const struct address_space_operations shmem_aops = { 2476 .writepage = shmem_writepage, 2477 .set_page_dirty = __set_page_dirty_no_writeback, 2478#ifdef CONFIG_TMPFS 2479 .readpage = shmem_readpage, 2480 .write_begin = shmem_write_begin, 2481 .write_end = shmem_write_end, 2482#endif 2483 .migratepage = migrate_page, 2484 .error_remove_page = generic_error_remove_page, 2485}; 2486 2487static const struct file_operations shmem_file_operations = { 2488 .mmap = shmem_mmap, 2489#ifdef CONFIG_TMPFS 2490 .llseek = generic_file_llseek, 2491 .read = do_sync_read, 2492 .write = do_sync_write, 2493 .aio_read = shmem_file_aio_read, 2494 .aio_write = generic_file_aio_write, 2495 .fsync = noop_fsync, 2496 .splice_read = generic_file_splice_read, 2497 .splice_write = generic_file_splice_write, 2498#endif 2499}; 2500 2501static const struct inode_operations shmem_inode_operations = { 2502 .setattr = shmem_notify_change, 2503 .truncate_range = shmem_truncate_range, 2504#ifdef CONFIG_TMPFS_POSIX_ACL 2505 .setxattr = generic_setxattr, 2506 .getxattr = generic_getxattr, 2507 .listxattr = generic_listxattr, 2508 .removexattr = generic_removexattr, 2509 .check_acl = generic_check_acl, 2510#endif 2511 2512}; 2513 2514static const struct inode_operations shmem_dir_inode_operations = { 2515#ifdef CONFIG_TMPFS 2516 .create = shmem_create, 2517 .lookup = simple_lookup, 2518 .link = shmem_link, 2519 .unlink = shmem_unlink, 2520 .symlink = shmem_symlink, 2521 .mkdir = shmem_mkdir, 2522 .rmdir = shmem_rmdir, 2523 .mknod = shmem_mknod, 2524 .rename = shmem_rename, 2525#endif 2526#ifdef CONFIG_TMPFS_POSIX_ACL 2527 .setattr = shmem_notify_change, 2528 .setxattr = generic_setxattr, 2529 .getxattr = generic_getxattr, 2530 .listxattr = generic_listxattr, 2531 .removexattr = generic_removexattr, 2532 .check_acl = generic_check_acl, 2533#endif 2534}; 2535 2536static const struct inode_operations shmem_special_inode_operations = { 2537#ifdef CONFIG_TMPFS_POSIX_ACL 2538 .setattr = shmem_notify_change, 2539 .setxattr = generic_setxattr, 2540 .getxattr = generic_getxattr, 2541 .listxattr = generic_listxattr, 2542 .removexattr = generic_removexattr, 2543 .check_acl = generic_check_acl, 2544#endif 2545}; 2546 2547static const struct super_operations shmem_ops = { 2548 .alloc_inode = shmem_alloc_inode, 2549 .destroy_inode = shmem_destroy_inode, 2550#ifdef CONFIG_TMPFS 2551 .statfs = shmem_statfs, 2552 .remount_fs = shmem_remount_fs, 2553 .show_options = shmem_show_options, 2554#endif 2555 .evict_inode = shmem_evict_inode, 2556 .drop_inode = generic_delete_inode, 2557 .put_super = shmem_put_super, 2558}; 2559 2560static const struct vm_operations_struct shmem_vm_ops = { 2561 .fault = shmem_fault, 2562#ifdef CONFIG_NUMA 2563 .set_policy = shmem_set_policy, 2564 .get_policy = shmem_get_policy, 2565#endif 2566}; 2567 2568 2569static struct dentry *shmem_mount(struct file_system_type *fs_type, 2570 int flags, const char *dev_name, void *data) 2571{ 2572 return mount_nodev(fs_type, flags, data, shmem_fill_super); 2573} 2574 2575static struct file_system_type tmpfs_fs_type = { 2576 .owner = THIS_MODULE, 2577 .name = "tmpfs", 2578 .mount = shmem_mount, 2579 .kill_sb = kill_litter_super, 2580}; 2581 2582int __init init_tmpfs(void) 2583{ 2584 int error; 2585 2586 error = bdi_init(&shmem_backing_dev_info); 2587 if (error) 2588 goto out4; 2589 2590 error = init_inodecache(); 2591 if (error) 2592 goto out3; 2593 2594 error = register_filesystem(&tmpfs_fs_type); 2595 if (error) { 2596 printk(KERN_ERR "Could not register tmpfs\n"); 2597 goto out2; 2598 } 2599 2600 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER, 2601 tmpfs_fs_type.name, NULL); 2602 if (IS_ERR(shm_mnt)) { 2603 error = PTR_ERR(shm_mnt); 2604 printk(KERN_ERR "Could not kern_mount tmpfs\n"); 2605 goto out1; 2606 } 2607 return 0; 2608 2609out1: 2610 unregister_filesystem(&tmpfs_fs_type); 2611out2: 2612 destroy_inodecache(); 2613out3: 2614 bdi_destroy(&shmem_backing_dev_info); 2615out4: 2616 shm_mnt = ERR_PTR(error); 2617 return error; 2618} 2619 2620#ifdef CONFIG_CGROUP_MEM_RES_CTLR 2621/** 2622 * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file 2623 * @inode: the inode to be searched 2624 * @pgoff: the offset to be searched 2625 * @pagep: the pointer for the found page to be stored 2626 * @ent: the pointer for the found swap entry to be stored 2627 * 2628 * If a page is found, refcount of it is incremented. Callers should handle 2629 * these refcount. 2630 */ 2631void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t pgoff, 2632 struct page **pagep, swp_entry_t *ent) 2633{ 2634 swp_entry_t entry = { .val = 0 }, *ptr; 2635 struct page *page = NULL; 2636 struct shmem_inode_info *info = SHMEM_I(inode); 2637 2638 if ((pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 2639 goto out; 2640 2641 spin_lock(&info->lock); 2642 ptr = shmem_swp_entry(info, pgoff, NULL); 2643#ifdef CONFIG_SWAP 2644 if (ptr && ptr->val) { 2645 entry.val = ptr->val; 2646 page = find_get_page(&swapper_space, entry.val); 2647 } else 2648#endif 2649 page = find_get_page(inode->i_mapping, pgoff); 2650 if (ptr) 2651 shmem_swp_unmap(ptr); 2652 spin_unlock(&info->lock); 2653out: 2654 *pagep = page; 2655 *ent = entry; 2656} 2657#endif 2658 2659#else /* !CONFIG_SHMEM */ 2660 2661/* 2662 * tiny-shmem: simple shmemfs and tmpfs using ramfs code 2663 * 2664 * This is intended for small system where the benefits of the full 2665 * shmem code (swap-backed and resource-limited) are outweighed by 2666 * their complexity. On systems without swap this code should be 2667 * effectively equivalent, but much lighter weight. 2668 */ 2669 2670#include <linux/ramfs.h> 2671 2672static struct file_system_type tmpfs_fs_type = { 2673 .name = "tmpfs", 2674 .mount = ramfs_mount, 2675 .kill_sb = kill_litter_super, 2676}; 2677 2678int __init init_tmpfs(void) 2679{ 2680 BUG_ON(register_filesystem(&tmpfs_fs_type) != 0); 2681 2682 shm_mnt = kern_mount(&tmpfs_fs_type); 2683 BUG_ON(IS_ERR(shm_mnt)); 2684 2685 return 0; 2686} 2687 2688int shmem_unuse(swp_entry_t entry, struct page *page) 2689{ 2690 return 0; 2691} 2692 2693int shmem_lock(struct file *file, int lock, struct user_struct *user) 2694{ 2695 return 0; 2696} 2697 2698#ifdef CONFIG_CGROUP_MEM_RES_CTLR 2699/** 2700 * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file 2701 * @inode: the inode to be searched 2702 * @pgoff: the offset to be searched 2703 * @pagep: the pointer for the found page to be stored 2704 * @ent: the pointer for the found swap entry to be stored 2705 * 2706 * If a page is found, refcount of it is incremented. Callers should handle 2707 * these refcount. 2708 */ 2709void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t pgoff, 2710 struct page **pagep, swp_entry_t *ent) 2711{ 2712 struct page *page = NULL; 2713 2714 if ((pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 2715 goto out; 2716 page = find_get_page(inode->i_mapping, pgoff); 2717out: 2718 *pagep = page; 2719 *ent = (swp_entry_t){ .val = 0 }; 2720} 2721#endif 2722 2723#define shmem_vm_ops generic_file_vm_ops 2724#define shmem_file_operations ramfs_file_operations 2725#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev) 2726#define shmem_acct_size(flags, size) 0 2727#define shmem_unacct_size(flags, size) do {} while (0) 2728#define SHMEM_MAX_BYTES MAX_LFS_FILESIZE 2729 2730#endif /* CONFIG_SHMEM */ 2731 2732/* common code */ 2733 2734/** 2735 * shmem_file_setup - get an unlinked file living in tmpfs 2736 * @name: name for dentry (to be seen in /proc/<pid>/maps 2737 * @size: size to be set for the file 2738 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size 2739 */ 2740struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) 2741{ 2742 int error; 2743 struct file *file; 2744 struct inode *inode; 2745 struct path path; 2746 struct dentry *root; 2747 struct qstr this; 2748 2749 if (IS_ERR(shm_mnt)) 2750 return (void *)shm_mnt; 2751 2752 if (size < 0 || size > SHMEM_MAX_BYTES) 2753 return ERR_PTR(-EINVAL); 2754 2755 if (shmem_acct_size(flags, size)) 2756 return ERR_PTR(-ENOMEM); 2757 2758 error = -ENOMEM; 2759 this.name = name; 2760 this.len = strlen(name); 2761 this.hash = 0; /* will go */ 2762 root = shm_mnt->mnt_root; 2763 path.dentry = d_alloc(root, &this); 2764 if (!path.dentry) 2765 goto put_memory; 2766 path.mnt = mntget(shm_mnt); 2767 2768 error = -ENOSPC; 2769 inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags); 2770 if (!inode) 2771 goto put_dentry; 2772 2773 d_instantiate(path.dentry, inode); 2774 inode->i_size = size; 2775 inode->i_nlink = 0; /* It is unlinked */ 2776#ifndef CONFIG_MMU 2777 error = ramfs_nommu_expand_for_mapping(inode, size); 2778 if (error) 2779 goto put_dentry; 2780#endif 2781 2782 error = -ENFILE; 2783 file = alloc_file(&path, FMODE_WRITE | FMODE_READ, 2784 &shmem_file_operations); 2785 if (!file) 2786 goto put_dentry; 2787 2788 return file; 2789 2790put_dentry: 2791 path_put(&path); 2792put_memory: 2793 shmem_unacct_size(flags, size); 2794 return ERR_PTR(error); 2795} 2796EXPORT_SYMBOL_GPL(shmem_file_setup); 2797 2798/** 2799 * shmem_zero_setup - setup a shared anonymous mapping 2800 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff 2801 */ 2802int shmem_zero_setup(struct vm_area_struct *vma) 2803{ 2804 struct file *file; 2805 loff_t size = vma->vm_end - vma->vm_start; 2806 2807 file = shmem_file_setup("dev/zero", size, vma->vm_flags); 2808 if (IS_ERR(file)) 2809 return PTR_ERR(file); 2810 2811 if (vma->vm_file) 2812 fput(vma->vm_file); 2813 vma->vm_file = file; 2814 vma->vm_ops = &shmem_vm_ops; 2815 vma->vm_flags |= VM_CAN_NONLINEAR; 2816 return 0; 2817} 2818