shmem.c revision 7eaceaccab5f40bbfda044629a6298616aeaed50
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, (sbinfo->max_blocks - 1)) > 0) 425 return ERR_PTR(-ENOSPC); 426 percpu_counter_inc(&sbinfo->used_blocks); 427 spin_lock(&inode->i_lock); 428 inode->i_blocks += BLOCKS_PER_PAGE; 429 spin_unlock(&inode->i_lock); 430 } 431 432 spin_unlock(&info->lock); 433 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping)); 434 spin_lock(&info->lock); 435 436 if (!page) { 437 shmem_free_blocks(inode, 1); 438 return ERR_PTR(-ENOMEM); 439 } 440 if (sgp != SGP_WRITE && 441 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { 442 entry = ERR_PTR(-EINVAL); 443 break; 444 } 445 if (info->next_index <= index) 446 info->next_index = index + 1; 447 } 448 if (page) { 449 /* another task gave its page, or truncated the file */ 450 shmem_free_blocks(inode, 1); 451 shmem_dir_free(page); 452 } 453 if (info->next_index <= index && !IS_ERR(entry)) 454 info->next_index = index + 1; 455 return entry; 456} 457 458/** 459 * shmem_free_swp - free some swap entries in a directory 460 * @dir: pointer to the directory 461 * @edir: pointer after last entry of the directory 462 * @punch_lock: pointer to spinlock when needed for the holepunch case 463 */ 464static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir, 465 spinlock_t *punch_lock) 466{ 467 spinlock_t *punch_unlock = NULL; 468 swp_entry_t *ptr; 469 int freed = 0; 470 471 for (ptr = dir; ptr < edir; ptr++) { 472 if (ptr->val) { 473 if (unlikely(punch_lock)) { 474 punch_unlock = punch_lock; 475 punch_lock = NULL; 476 spin_lock(punch_unlock); 477 if (!ptr->val) 478 continue; 479 } 480 free_swap_and_cache(*ptr); 481 *ptr = (swp_entry_t){0}; 482 freed++; 483 } 484 } 485 if (punch_unlock) 486 spin_unlock(punch_unlock); 487 return freed; 488} 489 490static int shmem_map_and_free_swp(struct page *subdir, int offset, 491 int limit, struct page ***dir, spinlock_t *punch_lock) 492{ 493 swp_entry_t *ptr; 494 int freed = 0; 495 496 ptr = shmem_swp_map(subdir); 497 for (; offset < limit; offset += LATENCY_LIMIT) { 498 int size = limit - offset; 499 if (size > LATENCY_LIMIT) 500 size = LATENCY_LIMIT; 501 freed += shmem_free_swp(ptr+offset, ptr+offset+size, 502 punch_lock); 503 if (need_resched()) { 504 shmem_swp_unmap(ptr); 505 if (*dir) { 506 shmem_dir_unmap(*dir); 507 *dir = NULL; 508 } 509 cond_resched(); 510 ptr = shmem_swp_map(subdir); 511 } 512 } 513 shmem_swp_unmap(ptr); 514 return freed; 515} 516 517static void shmem_free_pages(struct list_head *next) 518{ 519 struct page *page; 520 int freed = 0; 521 522 do { 523 page = container_of(next, struct page, lru); 524 next = next->next; 525 shmem_dir_free(page); 526 freed++; 527 if (freed >= LATENCY_LIMIT) { 528 cond_resched(); 529 freed = 0; 530 } 531 } while (next); 532} 533 534static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end) 535{ 536 struct shmem_inode_info *info = SHMEM_I(inode); 537 unsigned long idx; 538 unsigned long size; 539 unsigned long limit; 540 unsigned long stage; 541 unsigned long diroff; 542 struct page **dir; 543 struct page *topdir; 544 struct page *middir; 545 struct page *subdir; 546 swp_entry_t *ptr; 547 LIST_HEAD(pages_to_free); 548 long nr_pages_to_free = 0; 549 long nr_swaps_freed = 0; 550 int offset; 551 int freed; 552 int punch_hole; 553 spinlock_t *needs_lock; 554 spinlock_t *punch_lock; 555 unsigned long upper_limit; 556 557 inode->i_ctime = inode->i_mtime = CURRENT_TIME; 558 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 559 if (idx >= info->next_index) 560 return; 561 562 spin_lock(&info->lock); 563 info->flags |= SHMEM_TRUNCATE; 564 if (likely(end == (loff_t) -1)) { 565 limit = info->next_index; 566 upper_limit = SHMEM_MAX_INDEX; 567 info->next_index = idx; 568 needs_lock = NULL; 569 punch_hole = 0; 570 } else { 571 if (end + 1 >= inode->i_size) { /* we may free a little more */ 572 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >> 573 PAGE_CACHE_SHIFT; 574 upper_limit = SHMEM_MAX_INDEX; 575 } else { 576 limit = (end + 1) >> PAGE_CACHE_SHIFT; 577 upper_limit = limit; 578 } 579 needs_lock = &info->lock; 580 punch_hole = 1; 581 } 582 583 topdir = info->i_indirect; 584 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) { 585 info->i_indirect = NULL; 586 nr_pages_to_free++; 587 list_add(&topdir->lru, &pages_to_free); 588 } 589 spin_unlock(&info->lock); 590 591 if (info->swapped && idx < SHMEM_NR_DIRECT) { 592 ptr = info->i_direct; 593 size = limit; 594 if (size > SHMEM_NR_DIRECT) 595 size = SHMEM_NR_DIRECT; 596 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock); 597 } 598 599 /* 600 * If there are no indirect blocks or we are punching a hole 601 * below indirect blocks, nothing to be done. 602 */ 603 if (!topdir || limit <= SHMEM_NR_DIRECT) 604 goto done2; 605 606 /* 607 * The truncation case has already dropped info->lock, and we're safe 608 * because i_size and next_index have already been lowered, preventing 609 * access beyond. But in the punch_hole case, we still need to take 610 * the lock when updating the swap directory, because there might be 611 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or 612 * shmem_writepage. However, whenever we find we can remove a whole 613 * directory page (not at the misaligned start or end of the range), 614 * we first NULLify its pointer in the level above, and then have no 615 * need to take the lock when updating its contents: needs_lock and 616 * punch_lock (either pointing to info->lock or NULL) manage this. 617 */ 618 619 upper_limit -= SHMEM_NR_DIRECT; 620 limit -= SHMEM_NR_DIRECT; 621 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0; 622 offset = idx % ENTRIES_PER_PAGE; 623 idx -= offset; 624 625 dir = shmem_dir_map(topdir); 626 stage = ENTRIES_PER_PAGEPAGE/2; 627 if (idx < ENTRIES_PER_PAGEPAGE/2) { 628 middir = topdir; 629 diroff = idx/ENTRIES_PER_PAGE; 630 } else { 631 dir += ENTRIES_PER_PAGE/2; 632 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE; 633 while (stage <= idx) 634 stage += ENTRIES_PER_PAGEPAGE; 635 middir = *dir; 636 if (*dir) { 637 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) % 638 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE; 639 if (!diroff && !offset && upper_limit >= stage) { 640 if (needs_lock) { 641 spin_lock(needs_lock); 642 *dir = NULL; 643 spin_unlock(needs_lock); 644 needs_lock = NULL; 645 } else 646 *dir = NULL; 647 nr_pages_to_free++; 648 list_add(&middir->lru, &pages_to_free); 649 } 650 shmem_dir_unmap(dir); 651 dir = shmem_dir_map(middir); 652 } else { 653 diroff = 0; 654 offset = 0; 655 idx = stage; 656 } 657 } 658 659 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) { 660 if (unlikely(idx == stage)) { 661 shmem_dir_unmap(dir); 662 dir = shmem_dir_map(topdir) + 663 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 664 while (!*dir) { 665 dir++; 666 idx += ENTRIES_PER_PAGEPAGE; 667 if (idx >= limit) 668 goto done1; 669 } 670 stage = idx + ENTRIES_PER_PAGEPAGE; 671 middir = *dir; 672 if (punch_hole) 673 needs_lock = &info->lock; 674 if (upper_limit >= stage) { 675 if (needs_lock) { 676 spin_lock(needs_lock); 677 *dir = NULL; 678 spin_unlock(needs_lock); 679 needs_lock = NULL; 680 } else 681 *dir = NULL; 682 nr_pages_to_free++; 683 list_add(&middir->lru, &pages_to_free); 684 } 685 shmem_dir_unmap(dir); 686 cond_resched(); 687 dir = shmem_dir_map(middir); 688 diroff = 0; 689 } 690 punch_lock = needs_lock; 691 subdir = dir[diroff]; 692 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) { 693 if (needs_lock) { 694 spin_lock(needs_lock); 695 dir[diroff] = NULL; 696 spin_unlock(needs_lock); 697 punch_lock = NULL; 698 } else 699 dir[diroff] = NULL; 700 nr_pages_to_free++; 701 list_add(&subdir->lru, &pages_to_free); 702 } 703 if (subdir && page_private(subdir) /* has swap entries */) { 704 size = limit - idx; 705 if (size > ENTRIES_PER_PAGE) 706 size = ENTRIES_PER_PAGE; 707 freed = shmem_map_and_free_swp(subdir, 708 offset, size, &dir, punch_lock); 709 if (!dir) 710 dir = shmem_dir_map(middir); 711 nr_swaps_freed += freed; 712 if (offset || punch_lock) { 713 spin_lock(&info->lock); 714 set_page_private(subdir, 715 page_private(subdir) - freed); 716 spin_unlock(&info->lock); 717 } else 718 BUG_ON(page_private(subdir) != freed); 719 } 720 offset = 0; 721 } 722done1: 723 shmem_dir_unmap(dir); 724done2: 725 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) { 726 /* 727 * Call truncate_inode_pages again: racing shmem_unuse_inode 728 * may have swizzled a page in from swap since 729 * truncate_pagecache or generic_delete_inode did it, before we 730 * lowered next_index. Also, though shmem_getpage checks 731 * i_size before adding to cache, no recheck after: so fix the 732 * narrow window there too. 733 * 734 * Recalling truncate_inode_pages_range and unmap_mapping_range 735 * every time for punch_hole (which never got a chance to clear 736 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive, 737 * yet hardly ever necessary: try to optimize them out later. 738 */ 739 truncate_inode_pages_range(inode->i_mapping, start, end); 740 if (punch_hole) 741 unmap_mapping_range(inode->i_mapping, start, 742 end - start, 1); 743 } 744 745 spin_lock(&info->lock); 746 info->flags &= ~SHMEM_TRUNCATE; 747 info->swapped -= nr_swaps_freed; 748 if (nr_pages_to_free) 749 shmem_free_blocks(inode, nr_pages_to_free); 750 shmem_recalc_inode(inode); 751 spin_unlock(&info->lock); 752 753 /* 754 * Empty swap vector directory pages to be freed? 755 */ 756 if (!list_empty(&pages_to_free)) { 757 pages_to_free.prev->next = NULL; 758 shmem_free_pages(pages_to_free.next); 759 } 760} 761 762static int shmem_notify_change(struct dentry *dentry, struct iattr *attr) 763{ 764 struct inode *inode = dentry->d_inode; 765 loff_t newsize = attr->ia_size; 766 int error; 767 768 error = inode_change_ok(inode, attr); 769 if (error) 770 return error; 771 772 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE) 773 && newsize != inode->i_size) { 774 struct page *page = NULL; 775 776 if (newsize < inode->i_size) { 777 /* 778 * If truncating down to a partial page, then 779 * if that page is already allocated, hold it 780 * in memory until the truncation is over, so 781 * truncate_partial_page cannnot miss it were 782 * it assigned to swap. 783 */ 784 if (newsize & (PAGE_CACHE_SIZE-1)) { 785 (void) shmem_getpage(inode, 786 newsize >> PAGE_CACHE_SHIFT, 787 &page, SGP_READ, NULL); 788 if (page) 789 unlock_page(page); 790 } 791 /* 792 * Reset SHMEM_PAGEIN flag so that shmem_truncate can 793 * detect if any pages might have been added to cache 794 * after truncate_inode_pages. But we needn't bother 795 * if it's being fully truncated to zero-length: the 796 * nrpages check is efficient enough in that case. 797 */ 798 if (newsize) { 799 struct shmem_inode_info *info = SHMEM_I(inode); 800 spin_lock(&info->lock); 801 info->flags &= ~SHMEM_PAGEIN; 802 spin_unlock(&info->lock); 803 } 804 } 805 806 /* XXX(truncate): truncate_setsize should be called last */ 807 truncate_setsize(inode, newsize); 808 if (page) 809 page_cache_release(page); 810 shmem_truncate_range(inode, newsize, (loff_t)-1); 811 } 812 813 setattr_copy(inode, attr); 814#ifdef CONFIG_TMPFS_POSIX_ACL 815 if (attr->ia_valid & ATTR_MODE) 816 error = generic_acl_chmod(inode); 817#endif 818 return error; 819} 820 821static void shmem_evict_inode(struct inode *inode) 822{ 823 struct shmem_inode_info *info = SHMEM_I(inode); 824 825 if (inode->i_mapping->a_ops == &shmem_aops) { 826 truncate_inode_pages(inode->i_mapping, 0); 827 shmem_unacct_size(info->flags, inode->i_size); 828 inode->i_size = 0; 829 shmem_truncate_range(inode, 0, (loff_t)-1); 830 if (!list_empty(&info->swaplist)) { 831 mutex_lock(&shmem_swaplist_mutex); 832 list_del_init(&info->swaplist); 833 mutex_unlock(&shmem_swaplist_mutex); 834 } 835 } 836 BUG_ON(inode->i_blocks); 837 shmem_free_inode(inode->i_sb); 838 end_writeback(inode); 839} 840 841static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir) 842{ 843 swp_entry_t *ptr; 844 845 for (ptr = dir; ptr < edir; ptr++) { 846 if (ptr->val == entry.val) 847 return ptr - dir; 848 } 849 return -1; 850} 851 852static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page) 853{ 854 struct inode *inode; 855 unsigned long idx; 856 unsigned long size; 857 unsigned long limit; 858 unsigned long stage; 859 struct page **dir; 860 struct page *subdir; 861 swp_entry_t *ptr; 862 int offset; 863 int error; 864 865 idx = 0; 866 ptr = info->i_direct; 867 spin_lock(&info->lock); 868 if (!info->swapped) { 869 list_del_init(&info->swaplist); 870 goto lost2; 871 } 872 limit = info->next_index; 873 size = limit; 874 if (size > SHMEM_NR_DIRECT) 875 size = SHMEM_NR_DIRECT; 876 offset = shmem_find_swp(entry, ptr, ptr+size); 877 if (offset >= 0) 878 goto found; 879 if (!info->i_indirect) 880 goto lost2; 881 882 dir = shmem_dir_map(info->i_indirect); 883 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2; 884 885 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) { 886 if (unlikely(idx == stage)) { 887 shmem_dir_unmap(dir-1); 888 if (cond_resched_lock(&info->lock)) { 889 /* check it has not been truncated */ 890 if (limit > info->next_index) { 891 limit = info->next_index; 892 if (idx >= limit) 893 goto lost2; 894 } 895 } 896 dir = shmem_dir_map(info->i_indirect) + 897 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 898 while (!*dir) { 899 dir++; 900 idx += ENTRIES_PER_PAGEPAGE; 901 if (idx >= limit) 902 goto lost1; 903 } 904 stage = idx + ENTRIES_PER_PAGEPAGE; 905 subdir = *dir; 906 shmem_dir_unmap(dir); 907 dir = shmem_dir_map(subdir); 908 } 909 subdir = *dir; 910 if (subdir && page_private(subdir)) { 911 ptr = shmem_swp_map(subdir); 912 size = limit - idx; 913 if (size > ENTRIES_PER_PAGE) 914 size = ENTRIES_PER_PAGE; 915 offset = shmem_find_swp(entry, ptr, ptr+size); 916 shmem_swp_unmap(ptr); 917 if (offset >= 0) { 918 shmem_dir_unmap(dir); 919 goto found; 920 } 921 } 922 } 923lost1: 924 shmem_dir_unmap(dir-1); 925lost2: 926 spin_unlock(&info->lock); 927 return 0; 928found: 929 idx += offset; 930 inode = igrab(&info->vfs_inode); 931 spin_unlock(&info->lock); 932 933 /* 934 * Move _head_ to start search for next from here. 935 * But be careful: shmem_evict_inode checks list_empty without taking 936 * mutex, and there's an instant in list_move_tail when info->swaplist 937 * would appear empty, if it were the only one on shmem_swaplist. We 938 * could avoid doing it if inode NULL; or use this minor optimization. 939 */ 940 if (shmem_swaplist.next != &info->swaplist) 941 list_move_tail(&shmem_swaplist, &info->swaplist); 942 mutex_unlock(&shmem_swaplist_mutex); 943 944 error = 1; 945 if (!inode) 946 goto out; 947 /* 948 * Charge page using GFP_KERNEL while we can wait. 949 * Charged back to the user(not to caller) when swap account is used. 950 * add_to_page_cache() will be called with GFP_NOWAIT. 951 */ 952 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL); 953 if (error) 954 goto out; 955 error = radix_tree_preload(GFP_KERNEL); 956 if (error) { 957 mem_cgroup_uncharge_cache_page(page); 958 goto out; 959 } 960 error = 1; 961 962 spin_lock(&info->lock); 963 ptr = shmem_swp_entry(info, idx, NULL); 964 if (ptr && ptr->val == entry.val) { 965 error = add_to_page_cache_locked(page, inode->i_mapping, 966 idx, GFP_NOWAIT); 967 /* does mem_cgroup_uncharge_cache_page on error */ 968 } else /* we must compensate for our precharge above */ 969 mem_cgroup_uncharge_cache_page(page); 970 971 if (error == -EEXIST) { 972 struct page *filepage = find_get_page(inode->i_mapping, idx); 973 error = 1; 974 if (filepage) { 975 /* 976 * There might be a more uptodate page coming down 977 * from a stacked writepage: forget our swappage if so. 978 */ 979 if (PageUptodate(filepage)) 980 error = 0; 981 page_cache_release(filepage); 982 } 983 } 984 if (!error) { 985 delete_from_swap_cache(page); 986 set_page_dirty(page); 987 info->flags |= SHMEM_PAGEIN; 988 shmem_swp_set(info, ptr, 0); 989 swap_free(entry); 990 error = 1; /* not an error, but entry was found */ 991 } 992 if (ptr) 993 shmem_swp_unmap(ptr); 994 spin_unlock(&info->lock); 995 radix_tree_preload_end(); 996out: 997 unlock_page(page); 998 page_cache_release(page); 999 iput(inode); /* allows for NULL */ 1000 return error; 1001} 1002 1003/* 1004 * shmem_unuse() search for an eventually swapped out shmem page. 1005 */ 1006int shmem_unuse(swp_entry_t entry, struct page *page) 1007{ 1008 struct list_head *p, *next; 1009 struct shmem_inode_info *info; 1010 int found = 0; 1011 1012 mutex_lock(&shmem_swaplist_mutex); 1013 list_for_each_safe(p, next, &shmem_swaplist) { 1014 info = list_entry(p, struct shmem_inode_info, swaplist); 1015 found = shmem_unuse_inode(info, entry, page); 1016 cond_resched(); 1017 if (found) 1018 goto out; 1019 } 1020 mutex_unlock(&shmem_swaplist_mutex); 1021 /* 1022 * Can some race bring us here? We've been holding page lock, 1023 * so I think not; but would rather try again later than BUG() 1024 */ 1025 unlock_page(page); 1026 page_cache_release(page); 1027out: 1028 return (found < 0) ? found : 0; 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 spin_lock(&info->lock); 1067 if (index >= info->next_index) { 1068 BUG_ON(!(info->flags & SHMEM_TRUNCATE)); 1069 goto unlock; 1070 } 1071 entry = shmem_swp_entry(info, index, NULL); 1072 if (entry->val) { 1073 /* 1074 * The more uptodate page coming down from a stacked 1075 * writepage should replace our old swappage. 1076 */ 1077 free_swap_and_cache(*entry); 1078 shmem_swp_set(info, entry, 0); 1079 } 1080 shmem_recalc_inode(inode); 1081 1082 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { 1083 remove_from_page_cache(page); 1084 shmem_swp_set(info, entry, swap.val); 1085 shmem_swp_unmap(entry); 1086 if (list_empty(&info->swaplist)) 1087 inode = igrab(inode); 1088 else 1089 inode = NULL; 1090 spin_unlock(&info->lock); 1091 swap_shmem_alloc(swap); 1092 BUG_ON(page_mapped(page)); 1093 page_cache_release(page); /* pagecache ref */ 1094 swap_writepage(page, wbc); 1095 if (inode) { 1096 mutex_lock(&shmem_swaplist_mutex); 1097 /* move instead of add in case we're racing */ 1098 list_move_tail(&info->swaplist, &shmem_swaplist); 1099 mutex_unlock(&shmem_swaplist_mutex); 1100 iput(inode); 1101 } 1102 return 0; 1103 } 1104 1105 shmem_swp_unmap(entry); 1106unlock: 1107 spin_unlock(&info->lock); 1108 /* 1109 * add_to_swap_cache() doesn't return -EEXIST, so we can safely 1110 * clear SWAP_HAS_CACHE flag. 1111 */ 1112 swapcache_free(swap, NULL); 1113redirty: 1114 set_page_dirty(page); 1115 if (wbc->for_reclaim) 1116 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ 1117 unlock_page(page); 1118 return 0; 1119} 1120 1121#ifdef CONFIG_NUMA 1122#ifdef CONFIG_TMPFS 1123static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) 1124{ 1125 char buffer[64]; 1126 1127 if (!mpol || mpol->mode == MPOL_DEFAULT) 1128 return; /* show nothing */ 1129 1130 mpol_to_str(buffer, sizeof(buffer), mpol, 1); 1131 1132 seq_printf(seq, ",mpol=%s", buffer); 1133} 1134 1135static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) 1136{ 1137 struct mempolicy *mpol = NULL; 1138 if (sbinfo->mpol) { 1139 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ 1140 mpol = sbinfo->mpol; 1141 mpol_get(mpol); 1142 spin_unlock(&sbinfo->stat_lock); 1143 } 1144 return mpol; 1145} 1146#endif /* CONFIG_TMPFS */ 1147 1148static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, 1149 struct shmem_inode_info *info, unsigned long idx) 1150{ 1151 struct mempolicy mpol, *spol; 1152 struct vm_area_struct pvma; 1153 struct page *page; 1154 1155 spol = mpol_cond_copy(&mpol, 1156 mpol_shared_policy_lookup(&info->policy, idx)); 1157 1158 /* Create a pseudo vma that just contains the policy */ 1159 pvma.vm_start = 0; 1160 pvma.vm_pgoff = idx; 1161 pvma.vm_ops = NULL; 1162 pvma.vm_policy = spol; 1163 page = swapin_readahead(entry, gfp, &pvma, 0); 1164 return page; 1165} 1166 1167static struct page *shmem_alloc_page(gfp_t gfp, 1168 struct shmem_inode_info *info, unsigned long idx) 1169{ 1170 struct vm_area_struct pvma; 1171 1172 /* Create a pseudo vma that just contains the policy */ 1173 pvma.vm_start = 0; 1174 pvma.vm_pgoff = idx; 1175 pvma.vm_ops = NULL; 1176 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx); 1177 1178 /* 1179 * alloc_page_vma() will drop the shared policy reference 1180 */ 1181 return alloc_page_vma(gfp, &pvma, 0); 1182} 1183#else /* !CONFIG_NUMA */ 1184#ifdef CONFIG_TMPFS 1185static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p) 1186{ 1187} 1188#endif /* CONFIG_TMPFS */ 1189 1190static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, 1191 struct shmem_inode_info *info, unsigned long idx) 1192{ 1193 return swapin_readahead(entry, gfp, NULL, 0); 1194} 1195 1196static inline struct page *shmem_alloc_page(gfp_t gfp, 1197 struct shmem_inode_info *info, unsigned long idx) 1198{ 1199 return alloc_page(gfp); 1200} 1201#endif /* CONFIG_NUMA */ 1202 1203#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS) 1204static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) 1205{ 1206 return NULL; 1207} 1208#endif 1209 1210/* 1211 * shmem_getpage - either get the page from swap or allocate a new one 1212 * 1213 * If we allocate a new one we do not mark it dirty. That's up to the 1214 * vm. If we swap it in we mark it dirty since we also free the swap 1215 * entry since a page cannot live in both the swap and page cache 1216 */ 1217static int shmem_getpage(struct inode *inode, unsigned long idx, 1218 struct page **pagep, enum sgp_type sgp, int *type) 1219{ 1220 struct address_space *mapping = inode->i_mapping; 1221 struct shmem_inode_info *info = SHMEM_I(inode); 1222 struct shmem_sb_info *sbinfo; 1223 struct page *filepage = *pagep; 1224 struct page *swappage; 1225 struct page *prealloc_page = NULL; 1226 swp_entry_t *entry; 1227 swp_entry_t swap; 1228 gfp_t gfp; 1229 int error; 1230 1231 if (idx >= SHMEM_MAX_INDEX) 1232 return -EFBIG; 1233 1234 if (type) 1235 *type = 0; 1236 1237 /* 1238 * Normally, filepage is NULL on entry, and either found 1239 * uptodate immediately, or allocated and zeroed, or read 1240 * in under swappage, which is then assigned to filepage. 1241 * But shmem_readpage (required for splice) passes in a locked 1242 * filepage, which may be found not uptodate by other callers 1243 * too, and may need to be copied from the swappage read in. 1244 */ 1245repeat: 1246 if (!filepage) 1247 filepage = find_lock_page(mapping, idx); 1248 if (filepage && PageUptodate(filepage)) 1249 goto done; 1250 gfp = mapping_gfp_mask(mapping); 1251 if (!filepage) { 1252 /* 1253 * Try to preload while we can wait, to not make a habit of 1254 * draining atomic reserves; but don't latch on to this cpu. 1255 */ 1256 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM); 1257 if (error) 1258 goto failed; 1259 radix_tree_preload_end(); 1260 if (sgp != SGP_READ && !prealloc_page) { 1261 /* We don't care if this fails */ 1262 prealloc_page = shmem_alloc_page(gfp, info, idx); 1263 if (prealloc_page) { 1264 if (mem_cgroup_cache_charge(prealloc_page, 1265 current->mm, GFP_KERNEL)) { 1266 page_cache_release(prealloc_page); 1267 prealloc_page = NULL; 1268 } 1269 } 1270 } 1271 } 1272 error = 0; 1273 1274 spin_lock(&info->lock); 1275 shmem_recalc_inode(inode); 1276 entry = shmem_swp_alloc(info, idx, sgp); 1277 if (IS_ERR(entry)) { 1278 spin_unlock(&info->lock); 1279 error = PTR_ERR(entry); 1280 goto failed; 1281 } 1282 swap = *entry; 1283 1284 if (swap.val) { 1285 /* Look it up and read it in.. */ 1286 swappage = lookup_swap_cache(swap); 1287 if (!swappage) { 1288 shmem_swp_unmap(entry); 1289 /* here we actually do the io */ 1290 if (type && !(*type & VM_FAULT_MAJOR)) { 1291 __count_vm_event(PGMAJFAULT); 1292 *type |= VM_FAULT_MAJOR; 1293 } 1294 spin_unlock(&info->lock); 1295 swappage = shmem_swapin(swap, gfp, info, idx); 1296 if (!swappage) { 1297 spin_lock(&info->lock); 1298 entry = shmem_swp_alloc(info, idx, sgp); 1299 if (IS_ERR(entry)) 1300 error = PTR_ERR(entry); 1301 else { 1302 if (entry->val == swap.val) 1303 error = -ENOMEM; 1304 shmem_swp_unmap(entry); 1305 } 1306 spin_unlock(&info->lock); 1307 if (error) 1308 goto failed; 1309 goto repeat; 1310 } 1311 wait_on_page_locked(swappage); 1312 page_cache_release(swappage); 1313 goto repeat; 1314 } 1315 1316 /* We have to do this with page locked to prevent races */ 1317 if (!trylock_page(swappage)) { 1318 shmem_swp_unmap(entry); 1319 spin_unlock(&info->lock); 1320 wait_on_page_locked(swappage); 1321 page_cache_release(swappage); 1322 goto repeat; 1323 } 1324 if (PageWriteback(swappage)) { 1325 shmem_swp_unmap(entry); 1326 spin_unlock(&info->lock); 1327 wait_on_page_writeback(swappage); 1328 unlock_page(swappage); 1329 page_cache_release(swappage); 1330 goto repeat; 1331 } 1332 if (!PageUptodate(swappage)) { 1333 shmem_swp_unmap(entry); 1334 spin_unlock(&info->lock); 1335 unlock_page(swappage); 1336 page_cache_release(swappage); 1337 error = -EIO; 1338 goto failed; 1339 } 1340 1341 if (filepage) { 1342 shmem_swp_set(info, entry, 0); 1343 shmem_swp_unmap(entry); 1344 delete_from_swap_cache(swappage); 1345 spin_unlock(&info->lock); 1346 copy_highpage(filepage, swappage); 1347 unlock_page(swappage); 1348 page_cache_release(swappage); 1349 flush_dcache_page(filepage); 1350 SetPageUptodate(filepage); 1351 set_page_dirty(filepage); 1352 swap_free(swap); 1353 } else if (!(error = add_to_page_cache_locked(swappage, mapping, 1354 idx, GFP_NOWAIT))) { 1355 info->flags |= SHMEM_PAGEIN; 1356 shmem_swp_set(info, entry, 0); 1357 shmem_swp_unmap(entry); 1358 delete_from_swap_cache(swappage); 1359 spin_unlock(&info->lock); 1360 filepage = swappage; 1361 set_page_dirty(filepage); 1362 swap_free(swap); 1363 } else { 1364 shmem_swp_unmap(entry); 1365 spin_unlock(&info->lock); 1366 if (error == -ENOMEM) { 1367 /* 1368 * reclaim from proper memory cgroup and 1369 * call memcg's OOM if needed. 1370 */ 1371 error = mem_cgroup_shmem_charge_fallback( 1372 swappage, 1373 current->mm, 1374 gfp); 1375 if (error) { 1376 unlock_page(swappage); 1377 page_cache_release(swappage); 1378 goto failed; 1379 } 1380 } 1381 unlock_page(swappage); 1382 page_cache_release(swappage); 1383 goto repeat; 1384 } 1385 } else if (sgp == SGP_READ && !filepage) { 1386 shmem_swp_unmap(entry); 1387 filepage = find_get_page(mapping, idx); 1388 if (filepage && 1389 (!PageUptodate(filepage) || !trylock_page(filepage))) { 1390 spin_unlock(&info->lock); 1391 wait_on_page_locked(filepage); 1392 page_cache_release(filepage); 1393 filepage = NULL; 1394 goto repeat; 1395 } 1396 spin_unlock(&info->lock); 1397 } else { 1398 shmem_swp_unmap(entry); 1399 sbinfo = SHMEM_SB(inode->i_sb); 1400 if (sbinfo->max_blocks) { 1401 if ((percpu_counter_compare(&sbinfo->used_blocks, sbinfo->max_blocks) > 0) || 1402 shmem_acct_block(info->flags)) { 1403 spin_unlock(&info->lock); 1404 error = -ENOSPC; 1405 goto failed; 1406 } 1407 percpu_counter_inc(&sbinfo->used_blocks); 1408 spin_lock(&inode->i_lock); 1409 inode->i_blocks += BLOCKS_PER_PAGE; 1410 spin_unlock(&inode->i_lock); 1411 } else if (shmem_acct_block(info->flags)) { 1412 spin_unlock(&info->lock); 1413 error = -ENOSPC; 1414 goto failed; 1415 } 1416 1417 if (!filepage) { 1418 int ret; 1419 1420 if (!prealloc_page) { 1421 spin_unlock(&info->lock); 1422 filepage = shmem_alloc_page(gfp, info, idx); 1423 if (!filepage) { 1424 shmem_unacct_blocks(info->flags, 1); 1425 shmem_free_blocks(inode, 1); 1426 error = -ENOMEM; 1427 goto failed; 1428 } 1429 SetPageSwapBacked(filepage); 1430 1431 /* 1432 * Precharge page while we can wait, compensate 1433 * after 1434 */ 1435 error = mem_cgroup_cache_charge(filepage, 1436 current->mm, GFP_KERNEL); 1437 if (error) { 1438 page_cache_release(filepage); 1439 shmem_unacct_blocks(info->flags, 1); 1440 shmem_free_blocks(inode, 1); 1441 filepage = NULL; 1442 goto failed; 1443 } 1444 1445 spin_lock(&info->lock); 1446 } else { 1447 filepage = prealloc_page; 1448 prealloc_page = NULL; 1449 SetPageSwapBacked(filepage); 1450 } 1451 1452 entry = shmem_swp_alloc(info, idx, sgp); 1453 if (IS_ERR(entry)) 1454 error = PTR_ERR(entry); 1455 else { 1456 swap = *entry; 1457 shmem_swp_unmap(entry); 1458 } 1459 ret = error || swap.val; 1460 if (ret) 1461 mem_cgroup_uncharge_cache_page(filepage); 1462 else 1463 ret = add_to_page_cache_lru(filepage, mapping, 1464 idx, GFP_NOWAIT); 1465 /* 1466 * At add_to_page_cache_lru() failure, uncharge will 1467 * be done automatically. 1468 */ 1469 if (ret) { 1470 spin_unlock(&info->lock); 1471 page_cache_release(filepage); 1472 shmem_unacct_blocks(info->flags, 1); 1473 shmem_free_blocks(inode, 1); 1474 filepage = NULL; 1475 if (error) 1476 goto failed; 1477 goto repeat; 1478 } 1479 info->flags |= SHMEM_PAGEIN; 1480 } 1481 1482 info->alloced++; 1483 spin_unlock(&info->lock); 1484 clear_highpage(filepage); 1485 flush_dcache_page(filepage); 1486 SetPageUptodate(filepage); 1487 if (sgp == SGP_DIRTY) 1488 set_page_dirty(filepage); 1489 } 1490done: 1491 *pagep = filepage; 1492 error = 0; 1493 goto out; 1494 1495failed: 1496 if (*pagep != filepage) { 1497 unlock_page(filepage); 1498 page_cache_release(filepage); 1499 } 1500out: 1501 if (prealloc_page) { 1502 mem_cgroup_uncharge_cache_page(prealloc_page); 1503 page_cache_release(prealloc_page); 1504 } 1505 return error; 1506} 1507 1508static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1509{ 1510 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 1511 int error; 1512 int ret; 1513 1514 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 1515 return VM_FAULT_SIGBUS; 1516 1517 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret); 1518 if (error) 1519 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); 1520 1521 return ret | VM_FAULT_LOCKED; 1522} 1523 1524#ifdef CONFIG_NUMA 1525static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new) 1526{ 1527 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1528 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new); 1529} 1530 1531static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, 1532 unsigned long addr) 1533{ 1534 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1535 unsigned long idx; 1536 1537 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; 1538 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx); 1539} 1540#endif 1541 1542int shmem_lock(struct file *file, int lock, struct user_struct *user) 1543{ 1544 struct inode *inode = file->f_path.dentry->d_inode; 1545 struct shmem_inode_info *info = SHMEM_I(inode); 1546 int retval = -ENOMEM; 1547 1548 spin_lock(&info->lock); 1549 if (lock && !(info->flags & VM_LOCKED)) { 1550 if (!user_shm_lock(inode->i_size, user)) 1551 goto out_nomem; 1552 info->flags |= VM_LOCKED; 1553 mapping_set_unevictable(file->f_mapping); 1554 } 1555 if (!lock && (info->flags & VM_LOCKED) && user) { 1556 user_shm_unlock(inode->i_size, user); 1557 info->flags &= ~VM_LOCKED; 1558 mapping_clear_unevictable(file->f_mapping); 1559 scan_mapping_unevictable_pages(file->f_mapping); 1560 } 1561 retval = 0; 1562 1563out_nomem: 1564 spin_unlock(&info->lock); 1565 return retval; 1566} 1567 1568static int shmem_mmap(struct file *file, struct vm_area_struct *vma) 1569{ 1570 file_accessed(file); 1571 vma->vm_ops = &shmem_vm_ops; 1572 vma->vm_flags |= VM_CAN_NONLINEAR; 1573 return 0; 1574} 1575 1576static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir, 1577 int mode, dev_t dev, unsigned long flags) 1578{ 1579 struct inode *inode; 1580 struct shmem_inode_info *info; 1581 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 1582 1583 if (shmem_reserve_inode(sb)) 1584 return NULL; 1585 1586 inode = new_inode(sb); 1587 if (inode) { 1588 inode->i_ino = get_next_ino(); 1589 inode_init_owner(inode, dir, mode); 1590 inode->i_blocks = 0; 1591 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; 1592 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 1593 inode->i_generation = get_seconds(); 1594 info = SHMEM_I(inode); 1595 memset(info, 0, (char *)inode - (char *)info); 1596 spin_lock_init(&info->lock); 1597 info->flags = flags & VM_NORESERVE; 1598 INIT_LIST_HEAD(&info->swaplist); 1599 cache_no_acl(inode); 1600 1601 switch (mode & S_IFMT) { 1602 default: 1603 inode->i_op = &shmem_special_inode_operations; 1604 init_special_inode(inode, mode, dev); 1605 break; 1606 case S_IFREG: 1607 inode->i_mapping->a_ops = &shmem_aops; 1608 inode->i_op = &shmem_inode_operations; 1609 inode->i_fop = &shmem_file_operations; 1610 mpol_shared_policy_init(&info->policy, 1611 shmem_get_sbmpol(sbinfo)); 1612 break; 1613 case S_IFDIR: 1614 inc_nlink(inode); 1615 /* Some things misbehave if size == 0 on a directory */ 1616 inode->i_size = 2 * BOGO_DIRENT_SIZE; 1617 inode->i_op = &shmem_dir_inode_operations; 1618 inode->i_fop = &simple_dir_operations; 1619 break; 1620 case S_IFLNK: 1621 /* 1622 * Must not load anything in the rbtree, 1623 * mpol_free_shared_policy will not be called. 1624 */ 1625 mpol_shared_policy_init(&info->policy, NULL); 1626 break; 1627 } 1628 } else 1629 shmem_free_inode(sb); 1630 return inode; 1631} 1632 1633#ifdef CONFIG_TMPFS 1634static const struct inode_operations shmem_symlink_inode_operations; 1635static const struct inode_operations shmem_symlink_inline_operations; 1636 1637/* 1638 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin; 1639 * but providing them allows a tmpfs file to be used for splice, sendfile, and 1640 * below the loop driver, in the generic fashion that many filesystems support. 1641 */ 1642static int shmem_readpage(struct file *file, struct page *page) 1643{ 1644 struct inode *inode = page->mapping->host; 1645 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL); 1646 unlock_page(page); 1647 return error; 1648} 1649 1650static int 1651shmem_write_begin(struct file *file, struct address_space *mapping, 1652 loff_t pos, unsigned len, unsigned flags, 1653 struct page **pagep, void **fsdata) 1654{ 1655 struct inode *inode = mapping->host; 1656 pgoff_t index = pos >> PAGE_CACHE_SHIFT; 1657 *pagep = NULL; 1658 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL); 1659} 1660 1661static int 1662shmem_write_end(struct file *file, struct address_space *mapping, 1663 loff_t pos, unsigned len, unsigned copied, 1664 struct page *page, void *fsdata) 1665{ 1666 struct inode *inode = mapping->host; 1667 1668 if (pos + copied > inode->i_size) 1669 i_size_write(inode, pos + copied); 1670 1671 set_page_dirty(page); 1672 unlock_page(page); 1673 page_cache_release(page); 1674 1675 return copied; 1676} 1677 1678static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor) 1679{ 1680 struct inode *inode = filp->f_path.dentry->d_inode; 1681 struct address_space *mapping = inode->i_mapping; 1682 unsigned long index, offset; 1683 enum sgp_type sgp = SGP_READ; 1684 1685 /* 1686 * Might this read be for a stacking filesystem? Then when reading 1687 * holes of a sparse file, we actually need to allocate those pages, 1688 * and even mark them dirty, so it cannot exceed the max_blocks limit. 1689 */ 1690 if (segment_eq(get_fs(), KERNEL_DS)) 1691 sgp = SGP_DIRTY; 1692 1693 index = *ppos >> PAGE_CACHE_SHIFT; 1694 offset = *ppos & ~PAGE_CACHE_MASK; 1695 1696 for (;;) { 1697 struct page *page = NULL; 1698 unsigned long end_index, nr, ret; 1699 loff_t i_size = i_size_read(inode); 1700 1701 end_index = i_size >> PAGE_CACHE_SHIFT; 1702 if (index > end_index) 1703 break; 1704 if (index == end_index) { 1705 nr = i_size & ~PAGE_CACHE_MASK; 1706 if (nr <= offset) 1707 break; 1708 } 1709 1710 desc->error = shmem_getpage(inode, index, &page, sgp, NULL); 1711 if (desc->error) { 1712 if (desc->error == -EINVAL) 1713 desc->error = 0; 1714 break; 1715 } 1716 if (page) 1717 unlock_page(page); 1718 1719 /* 1720 * We must evaluate after, since reads (unlike writes) 1721 * are called without i_mutex protection against truncate 1722 */ 1723 nr = PAGE_CACHE_SIZE; 1724 i_size = i_size_read(inode); 1725 end_index = i_size >> PAGE_CACHE_SHIFT; 1726 if (index == end_index) { 1727 nr = i_size & ~PAGE_CACHE_MASK; 1728 if (nr <= offset) { 1729 if (page) 1730 page_cache_release(page); 1731 break; 1732 } 1733 } 1734 nr -= offset; 1735 1736 if (page) { 1737 /* 1738 * If users can be writing to this page using arbitrary 1739 * virtual addresses, take care about potential aliasing 1740 * before reading the page on the kernel side. 1741 */ 1742 if (mapping_writably_mapped(mapping)) 1743 flush_dcache_page(page); 1744 /* 1745 * Mark the page accessed if we read the beginning. 1746 */ 1747 if (!offset) 1748 mark_page_accessed(page); 1749 } else { 1750 page = ZERO_PAGE(0); 1751 page_cache_get(page); 1752 } 1753 1754 /* 1755 * Ok, we have the page, and it's up-to-date, so 1756 * now we can copy it to user space... 1757 * 1758 * The actor routine returns how many bytes were actually used.. 1759 * NOTE! This may not be the same as how much of a user buffer 1760 * we filled up (we may be padding etc), so we can only update 1761 * "pos" here (the actor routine has to update the user buffer 1762 * pointers and the remaining count). 1763 */ 1764 ret = actor(desc, page, offset, nr); 1765 offset += ret; 1766 index += offset >> PAGE_CACHE_SHIFT; 1767 offset &= ~PAGE_CACHE_MASK; 1768 1769 page_cache_release(page); 1770 if (ret != nr || !desc->count) 1771 break; 1772 1773 cond_resched(); 1774 } 1775 1776 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; 1777 file_accessed(filp); 1778} 1779 1780static ssize_t shmem_file_aio_read(struct kiocb *iocb, 1781 const struct iovec *iov, unsigned long nr_segs, loff_t pos) 1782{ 1783 struct file *filp = iocb->ki_filp; 1784 ssize_t retval; 1785 unsigned long seg; 1786 size_t count; 1787 loff_t *ppos = &iocb->ki_pos; 1788 1789 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); 1790 if (retval) 1791 return retval; 1792 1793 for (seg = 0; seg < nr_segs; seg++) { 1794 read_descriptor_t desc; 1795 1796 desc.written = 0; 1797 desc.arg.buf = iov[seg].iov_base; 1798 desc.count = iov[seg].iov_len; 1799 if (desc.count == 0) 1800 continue; 1801 desc.error = 0; 1802 do_shmem_file_read(filp, ppos, &desc, file_read_actor); 1803 retval += desc.written; 1804 if (desc.error) { 1805 retval = retval ?: desc.error; 1806 break; 1807 } 1808 if (desc.count > 0) 1809 break; 1810 } 1811 return retval; 1812} 1813 1814static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) 1815{ 1816 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); 1817 1818 buf->f_type = TMPFS_MAGIC; 1819 buf->f_bsize = PAGE_CACHE_SIZE; 1820 buf->f_namelen = NAME_MAX; 1821 if (sbinfo->max_blocks) { 1822 buf->f_blocks = sbinfo->max_blocks; 1823 buf->f_bavail = buf->f_bfree = 1824 sbinfo->max_blocks - percpu_counter_sum(&sbinfo->used_blocks); 1825 } 1826 if (sbinfo->max_inodes) { 1827 buf->f_files = sbinfo->max_inodes; 1828 buf->f_ffree = sbinfo->free_inodes; 1829 } 1830 /* else leave those fields 0 like simple_statfs */ 1831 return 0; 1832} 1833 1834/* 1835 * File creation. Allocate an inode, and we're done.. 1836 */ 1837static int 1838shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 1839{ 1840 struct inode *inode; 1841 int error = -ENOSPC; 1842 1843 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE); 1844 if (inode) { 1845 error = security_inode_init_security(inode, dir, NULL, NULL, 1846 NULL); 1847 if (error) { 1848 if (error != -EOPNOTSUPP) { 1849 iput(inode); 1850 return error; 1851 } 1852 } 1853#ifdef CONFIG_TMPFS_POSIX_ACL 1854 error = generic_acl_init(inode, dir); 1855 if (error) { 1856 iput(inode); 1857 return error; 1858 } 1859#else 1860 error = 0; 1861#endif 1862 dir->i_size += BOGO_DIRENT_SIZE; 1863 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1864 d_instantiate(dentry, inode); 1865 dget(dentry); /* Extra count - pin the dentry in core */ 1866 } 1867 return error; 1868} 1869 1870static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1871{ 1872 int error; 1873 1874 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) 1875 return error; 1876 inc_nlink(dir); 1877 return 0; 1878} 1879 1880static int shmem_create(struct inode *dir, struct dentry *dentry, int mode, 1881 struct nameidata *nd) 1882{ 1883 return shmem_mknod(dir, dentry, mode | S_IFREG, 0); 1884} 1885 1886/* 1887 * Link a file.. 1888 */ 1889static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1890{ 1891 struct inode *inode = old_dentry->d_inode; 1892 int ret; 1893 1894 /* 1895 * No ordinary (disk based) filesystem counts links as inodes; 1896 * but each new link needs a new dentry, pinning lowmem, and 1897 * tmpfs dentries cannot be pruned until they are unlinked. 1898 */ 1899 ret = shmem_reserve_inode(inode->i_sb); 1900 if (ret) 1901 goto out; 1902 1903 dir->i_size += BOGO_DIRENT_SIZE; 1904 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1905 inc_nlink(inode); 1906 ihold(inode); /* New dentry reference */ 1907 dget(dentry); /* Extra pinning count for the created dentry */ 1908 d_instantiate(dentry, inode); 1909out: 1910 return ret; 1911} 1912 1913static int shmem_unlink(struct inode *dir, struct dentry *dentry) 1914{ 1915 struct inode *inode = dentry->d_inode; 1916 1917 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) 1918 shmem_free_inode(inode->i_sb); 1919 1920 dir->i_size -= BOGO_DIRENT_SIZE; 1921 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1922 drop_nlink(inode); 1923 dput(dentry); /* Undo the count from "create" - this does all the work */ 1924 return 0; 1925} 1926 1927static int shmem_rmdir(struct inode *dir, struct dentry *dentry) 1928{ 1929 if (!simple_empty(dentry)) 1930 return -ENOTEMPTY; 1931 1932 drop_nlink(dentry->d_inode); 1933 drop_nlink(dir); 1934 return shmem_unlink(dir, dentry); 1935} 1936 1937/* 1938 * The VFS layer already does all the dentry stuff for rename, 1939 * we just have to decrement the usage count for the target if 1940 * it exists so that the VFS layer correctly free's it when it 1941 * gets overwritten. 1942 */ 1943static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) 1944{ 1945 struct inode *inode = old_dentry->d_inode; 1946 int they_are_dirs = S_ISDIR(inode->i_mode); 1947 1948 if (!simple_empty(new_dentry)) 1949 return -ENOTEMPTY; 1950 1951 if (new_dentry->d_inode) { 1952 (void) shmem_unlink(new_dir, new_dentry); 1953 if (they_are_dirs) 1954 drop_nlink(old_dir); 1955 } else if (they_are_dirs) { 1956 drop_nlink(old_dir); 1957 inc_nlink(new_dir); 1958 } 1959 1960 old_dir->i_size -= BOGO_DIRENT_SIZE; 1961 new_dir->i_size += BOGO_DIRENT_SIZE; 1962 old_dir->i_ctime = old_dir->i_mtime = 1963 new_dir->i_ctime = new_dir->i_mtime = 1964 inode->i_ctime = CURRENT_TIME; 1965 return 0; 1966} 1967 1968static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1969{ 1970 int error; 1971 int len; 1972 struct inode *inode; 1973 struct page *page = NULL; 1974 char *kaddr; 1975 struct shmem_inode_info *info; 1976 1977 len = strlen(symname) + 1; 1978 if (len > PAGE_CACHE_SIZE) 1979 return -ENAMETOOLONG; 1980 1981 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE); 1982 if (!inode) 1983 return -ENOSPC; 1984 1985 error = security_inode_init_security(inode, dir, NULL, NULL, 1986 NULL); 1987 if (error) { 1988 if (error != -EOPNOTSUPP) { 1989 iput(inode); 1990 return error; 1991 } 1992 error = 0; 1993 } 1994 1995 info = SHMEM_I(inode); 1996 inode->i_size = len-1; 1997 if (len <= (char *)inode - (char *)info) { 1998 /* do it inline */ 1999 memcpy(info, symname, len); 2000 inode->i_op = &shmem_symlink_inline_operations; 2001 } else { 2002 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); 2003 if (error) { 2004 iput(inode); 2005 return error; 2006 } 2007 inode->i_mapping->a_ops = &shmem_aops; 2008 inode->i_op = &shmem_symlink_inode_operations; 2009 kaddr = kmap_atomic(page, KM_USER0); 2010 memcpy(kaddr, symname, len); 2011 kunmap_atomic(kaddr, KM_USER0); 2012 set_page_dirty(page); 2013 unlock_page(page); 2014 page_cache_release(page); 2015 } 2016 dir->i_size += BOGO_DIRENT_SIZE; 2017 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 2018 d_instantiate(dentry, inode); 2019 dget(dentry); 2020 return 0; 2021} 2022 2023static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd) 2024{ 2025 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode)); 2026 return NULL; 2027} 2028 2029static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) 2030{ 2031 struct page *page = NULL; 2032 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); 2033 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page)); 2034 if (page) 2035 unlock_page(page); 2036 return page; 2037} 2038 2039static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 2040{ 2041 if (!IS_ERR(nd_get_link(nd))) { 2042 struct page *page = cookie; 2043 kunmap(page); 2044 mark_page_accessed(page); 2045 page_cache_release(page); 2046 } 2047} 2048 2049static const struct inode_operations shmem_symlink_inline_operations = { 2050 .readlink = generic_readlink, 2051 .follow_link = shmem_follow_link_inline, 2052}; 2053 2054static const struct inode_operations shmem_symlink_inode_operations = { 2055 .readlink = generic_readlink, 2056 .follow_link = shmem_follow_link, 2057 .put_link = shmem_put_link, 2058}; 2059 2060#ifdef CONFIG_TMPFS_POSIX_ACL 2061/* 2062 * Superblocks without xattr inode operations will get security.* xattr 2063 * support from the VFS "for free". As soon as we have any other xattrs 2064 * like ACLs, we also need to implement the security.* handlers at 2065 * filesystem level, though. 2066 */ 2067 2068static size_t shmem_xattr_security_list(struct dentry *dentry, char *list, 2069 size_t list_len, const char *name, 2070 size_t name_len, int handler_flags) 2071{ 2072 return security_inode_listsecurity(dentry->d_inode, list, list_len); 2073} 2074 2075static int shmem_xattr_security_get(struct dentry *dentry, const char *name, 2076 void *buffer, size_t size, int handler_flags) 2077{ 2078 if (strcmp(name, "") == 0) 2079 return -EINVAL; 2080 return xattr_getsecurity(dentry->d_inode, name, buffer, size); 2081} 2082 2083static int shmem_xattr_security_set(struct dentry *dentry, const char *name, 2084 const void *value, size_t size, int flags, int handler_flags) 2085{ 2086 if (strcmp(name, "") == 0) 2087 return -EINVAL; 2088 return security_inode_setsecurity(dentry->d_inode, name, value, 2089 size, flags); 2090} 2091 2092static const struct xattr_handler shmem_xattr_security_handler = { 2093 .prefix = XATTR_SECURITY_PREFIX, 2094 .list = shmem_xattr_security_list, 2095 .get = shmem_xattr_security_get, 2096 .set = shmem_xattr_security_set, 2097}; 2098 2099static const struct xattr_handler *shmem_xattr_handlers[] = { 2100 &generic_acl_access_handler, 2101 &generic_acl_default_handler, 2102 &shmem_xattr_security_handler, 2103 NULL 2104}; 2105#endif 2106 2107static struct dentry *shmem_get_parent(struct dentry *child) 2108{ 2109 return ERR_PTR(-ESTALE); 2110} 2111 2112static int shmem_match(struct inode *ino, void *vfh) 2113{ 2114 __u32 *fh = vfh; 2115 __u64 inum = fh[2]; 2116 inum = (inum << 32) | fh[1]; 2117 return ino->i_ino == inum && fh[0] == ino->i_generation; 2118} 2119 2120static struct dentry *shmem_fh_to_dentry(struct super_block *sb, 2121 struct fid *fid, int fh_len, int fh_type) 2122{ 2123 struct inode *inode; 2124 struct dentry *dentry = NULL; 2125 u64 inum = fid->raw[2]; 2126 inum = (inum << 32) | fid->raw[1]; 2127 2128 if (fh_len < 3) 2129 return NULL; 2130 2131 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), 2132 shmem_match, fid->raw); 2133 if (inode) { 2134 dentry = d_find_alias(inode); 2135 iput(inode); 2136 } 2137 2138 return dentry; 2139} 2140 2141static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len, 2142 int connectable) 2143{ 2144 struct inode *inode = dentry->d_inode; 2145 2146 if (*len < 3) 2147 return 255; 2148 2149 if (inode_unhashed(inode)) { 2150 /* Unfortunately insert_inode_hash is not idempotent, 2151 * so as we hash inodes here rather than at creation 2152 * time, we need a lock to ensure we only try 2153 * to do it once 2154 */ 2155 static DEFINE_SPINLOCK(lock); 2156 spin_lock(&lock); 2157 if (inode_unhashed(inode)) 2158 __insert_inode_hash(inode, 2159 inode->i_ino + inode->i_generation); 2160 spin_unlock(&lock); 2161 } 2162 2163 fh[0] = inode->i_generation; 2164 fh[1] = inode->i_ino; 2165 fh[2] = ((__u64)inode->i_ino) >> 32; 2166 2167 *len = 3; 2168 return 1; 2169} 2170 2171static const struct export_operations shmem_export_ops = { 2172 .get_parent = shmem_get_parent, 2173 .encode_fh = shmem_encode_fh, 2174 .fh_to_dentry = shmem_fh_to_dentry, 2175}; 2176 2177static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, 2178 bool remount) 2179{ 2180 char *this_char, *value, *rest; 2181 2182 while (options != NULL) { 2183 this_char = options; 2184 for (;;) { 2185 /* 2186 * NUL-terminate this option: unfortunately, 2187 * mount options form a comma-separated list, 2188 * but mpol's nodelist may also contain commas. 2189 */ 2190 options = strchr(options, ','); 2191 if (options == NULL) 2192 break; 2193 options++; 2194 if (!isdigit(*options)) { 2195 options[-1] = '\0'; 2196 break; 2197 } 2198 } 2199 if (!*this_char) 2200 continue; 2201 if ((value = strchr(this_char,'=')) != NULL) { 2202 *value++ = 0; 2203 } else { 2204 printk(KERN_ERR 2205 "tmpfs: No value for mount option '%s'\n", 2206 this_char); 2207 return 1; 2208 } 2209 2210 if (!strcmp(this_char,"size")) { 2211 unsigned long long size; 2212 size = memparse(value,&rest); 2213 if (*rest == '%') { 2214 size <<= PAGE_SHIFT; 2215 size *= totalram_pages; 2216 do_div(size, 100); 2217 rest++; 2218 } 2219 if (*rest) 2220 goto bad_val; 2221 sbinfo->max_blocks = 2222 DIV_ROUND_UP(size, PAGE_CACHE_SIZE); 2223 } else if (!strcmp(this_char,"nr_blocks")) { 2224 sbinfo->max_blocks = memparse(value, &rest); 2225 if (*rest) 2226 goto bad_val; 2227 } else if (!strcmp(this_char,"nr_inodes")) { 2228 sbinfo->max_inodes = memparse(value, &rest); 2229 if (*rest) 2230 goto bad_val; 2231 } else if (!strcmp(this_char,"mode")) { 2232 if (remount) 2233 continue; 2234 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; 2235 if (*rest) 2236 goto bad_val; 2237 } else if (!strcmp(this_char,"uid")) { 2238 if (remount) 2239 continue; 2240 sbinfo->uid = simple_strtoul(value, &rest, 0); 2241 if (*rest) 2242 goto bad_val; 2243 } else if (!strcmp(this_char,"gid")) { 2244 if (remount) 2245 continue; 2246 sbinfo->gid = simple_strtoul(value, &rest, 0); 2247 if (*rest) 2248 goto bad_val; 2249 } else if (!strcmp(this_char,"mpol")) { 2250 if (mpol_parse_str(value, &sbinfo->mpol, 1)) 2251 goto bad_val; 2252 } else { 2253 printk(KERN_ERR "tmpfs: Bad mount option %s\n", 2254 this_char); 2255 return 1; 2256 } 2257 } 2258 return 0; 2259 2260bad_val: 2261 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", 2262 value, this_char); 2263 return 1; 2264 2265} 2266 2267static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) 2268{ 2269 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 2270 struct shmem_sb_info config = *sbinfo; 2271 unsigned long inodes; 2272 int error = -EINVAL; 2273 2274 if (shmem_parse_options(data, &config, true)) 2275 return error; 2276 2277 spin_lock(&sbinfo->stat_lock); 2278 inodes = sbinfo->max_inodes - sbinfo->free_inodes; 2279 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0) 2280 goto out; 2281 if (config.max_inodes < inodes) 2282 goto out; 2283 /* 2284 * Those tests also disallow limited->unlimited while any are in 2285 * use, so i_blocks will always be zero when max_blocks is zero; 2286 * but we must separately disallow unlimited->limited, because 2287 * in that case we have no record of how much is already in use. 2288 */ 2289 if (config.max_blocks && !sbinfo->max_blocks) 2290 goto out; 2291 if (config.max_inodes && !sbinfo->max_inodes) 2292 goto out; 2293 2294 error = 0; 2295 sbinfo->max_blocks = config.max_blocks; 2296 sbinfo->max_inodes = config.max_inodes; 2297 sbinfo->free_inodes = config.max_inodes - inodes; 2298 2299 mpol_put(sbinfo->mpol); 2300 sbinfo->mpol = config.mpol; /* transfers initial ref */ 2301out: 2302 spin_unlock(&sbinfo->stat_lock); 2303 return error; 2304} 2305 2306static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs) 2307{ 2308 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb); 2309 2310 if (sbinfo->max_blocks != shmem_default_max_blocks()) 2311 seq_printf(seq, ",size=%luk", 2312 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10)); 2313 if (sbinfo->max_inodes != shmem_default_max_inodes()) 2314 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); 2315 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) 2316 seq_printf(seq, ",mode=%03o", sbinfo->mode); 2317 if (sbinfo->uid != 0) 2318 seq_printf(seq, ",uid=%u", sbinfo->uid); 2319 if (sbinfo->gid != 0) 2320 seq_printf(seq, ",gid=%u", sbinfo->gid); 2321 shmem_show_mpol(seq, sbinfo->mpol); 2322 return 0; 2323} 2324#endif /* CONFIG_TMPFS */ 2325 2326static void shmem_put_super(struct super_block *sb) 2327{ 2328 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 2329 2330 percpu_counter_destroy(&sbinfo->used_blocks); 2331 kfree(sbinfo); 2332 sb->s_fs_info = NULL; 2333} 2334 2335int shmem_fill_super(struct super_block *sb, void *data, int silent) 2336{ 2337 struct inode *inode; 2338 struct dentry *root; 2339 struct shmem_sb_info *sbinfo; 2340 int err = -ENOMEM; 2341 2342 /* Round up to L1_CACHE_BYTES to resist false sharing */ 2343 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), 2344 L1_CACHE_BYTES), GFP_KERNEL); 2345 if (!sbinfo) 2346 return -ENOMEM; 2347 2348 sbinfo->mode = S_IRWXUGO | S_ISVTX; 2349 sbinfo->uid = current_fsuid(); 2350 sbinfo->gid = current_fsgid(); 2351 sb->s_fs_info = sbinfo; 2352 2353#ifdef CONFIG_TMPFS 2354 /* 2355 * Per default we only allow half of the physical ram per 2356 * tmpfs instance, limiting inodes to one per page of lowmem; 2357 * but the internal instance is left unlimited. 2358 */ 2359 if (!(sb->s_flags & MS_NOUSER)) { 2360 sbinfo->max_blocks = shmem_default_max_blocks(); 2361 sbinfo->max_inodes = shmem_default_max_inodes(); 2362 if (shmem_parse_options(data, sbinfo, false)) { 2363 err = -EINVAL; 2364 goto failed; 2365 } 2366 } 2367 sb->s_export_op = &shmem_export_ops; 2368#else 2369 sb->s_flags |= MS_NOUSER; 2370#endif 2371 2372 spin_lock_init(&sbinfo->stat_lock); 2373 if (percpu_counter_init(&sbinfo->used_blocks, 0)) 2374 goto failed; 2375 sbinfo->free_inodes = sbinfo->max_inodes; 2376 2377 sb->s_maxbytes = SHMEM_MAX_BYTES; 2378 sb->s_blocksize = PAGE_CACHE_SIZE; 2379 sb->s_blocksize_bits = PAGE_CACHE_SHIFT; 2380 sb->s_magic = TMPFS_MAGIC; 2381 sb->s_op = &shmem_ops; 2382 sb->s_time_gran = 1; 2383#ifdef CONFIG_TMPFS_POSIX_ACL 2384 sb->s_xattr = shmem_xattr_handlers; 2385 sb->s_flags |= MS_POSIXACL; 2386#endif 2387 2388 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); 2389 if (!inode) 2390 goto failed; 2391 inode->i_uid = sbinfo->uid; 2392 inode->i_gid = sbinfo->gid; 2393 root = d_alloc_root(inode); 2394 if (!root) 2395 goto failed_iput; 2396 sb->s_root = root; 2397 return 0; 2398 2399failed_iput: 2400 iput(inode); 2401failed: 2402 shmem_put_super(sb); 2403 return err; 2404} 2405 2406static struct kmem_cache *shmem_inode_cachep; 2407 2408static struct inode *shmem_alloc_inode(struct super_block *sb) 2409{ 2410 struct shmem_inode_info *p; 2411 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); 2412 if (!p) 2413 return NULL; 2414 return &p->vfs_inode; 2415} 2416 2417static void shmem_i_callback(struct rcu_head *head) 2418{ 2419 struct inode *inode = container_of(head, struct inode, i_rcu); 2420 INIT_LIST_HEAD(&inode->i_dentry); 2421 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); 2422} 2423 2424static void shmem_destroy_inode(struct inode *inode) 2425{ 2426 if ((inode->i_mode & S_IFMT) == S_IFREG) { 2427 /* only struct inode is valid if it's an inline symlink */ 2428 mpol_free_shared_policy(&SHMEM_I(inode)->policy); 2429 } 2430 call_rcu(&inode->i_rcu, shmem_i_callback); 2431} 2432 2433static void init_once(void *foo) 2434{ 2435 struct shmem_inode_info *p = (struct shmem_inode_info *) foo; 2436 2437 inode_init_once(&p->vfs_inode); 2438} 2439 2440static int init_inodecache(void) 2441{ 2442 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", 2443 sizeof(struct shmem_inode_info), 2444 0, SLAB_PANIC, init_once); 2445 return 0; 2446} 2447 2448static void destroy_inodecache(void) 2449{ 2450 kmem_cache_destroy(shmem_inode_cachep); 2451} 2452 2453static const struct address_space_operations shmem_aops = { 2454 .writepage = shmem_writepage, 2455 .set_page_dirty = __set_page_dirty_no_writeback, 2456#ifdef CONFIG_TMPFS 2457 .readpage = shmem_readpage, 2458 .write_begin = shmem_write_begin, 2459 .write_end = shmem_write_end, 2460#endif 2461 .migratepage = migrate_page, 2462 .error_remove_page = generic_error_remove_page, 2463}; 2464 2465static const struct file_operations shmem_file_operations = { 2466 .mmap = shmem_mmap, 2467#ifdef CONFIG_TMPFS 2468 .llseek = generic_file_llseek, 2469 .read = do_sync_read, 2470 .write = do_sync_write, 2471 .aio_read = shmem_file_aio_read, 2472 .aio_write = generic_file_aio_write, 2473 .fsync = noop_fsync, 2474 .splice_read = generic_file_splice_read, 2475 .splice_write = generic_file_splice_write, 2476#endif 2477}; 2478 2479static const struct inode_operations shmem_inode_operations = { 2480 .setattr = shmem_notify_change, 2481 .truncate_range = shmem_truncate_range, 2482#ifdef CONFIG_TMPFS_POSIX_ACL 2483 .setxattr = generic_setxattr, 2484 .getxattr = generic_getxattr, 2485 .listxattr = generic_listxattr, 2486 .removexattr = generic_removexattr, 2487 .check_acl = generic_check_acl, 2488#endif 2489 2490}; 2491 2492static const struct inode_operations shmem_dir_inode_operations = { 2493#ifdef CONFIG_TMPFS 2494 .create = shmem_create, 2495 .lookup = simple_lookup, 2496 .link = shmem_link, 2497 .unlink = shmem_unlink, 2498 .symlink = shmem_symlink, 2499 .mkdir = shmem_mkdir, 2500 .rmdir = shmem_rmdir, 2501 .mknod = shmem_mknod, 2502 .rename = shmem_rename, 2503#endif 2504#ifdef CONFIG_TMPFS_POSIX_ACL 2505 .setattr = shmem_notify_change, 2506 .setxattr = generic_setxattr, 2507 .getxattr = generic_getxattr, 2508 .listxattr = generic_listxattr, 2509 .removexattr = generic_removexattr, 2510 .check_acl = generic_check_acl, 2511#endif 2512}; 2513 2514static const struct inode_operations shmem_special_inode_operations = { 2515#ifdef CONFIG_TMPFS_POSIX_ACL 2516 .setattr = shmem_notify_change, 2517 .setxattr = generic_setxattr, 2518 .getxattr = generic_getxattr, 2519 .listxattr = generic_listxattr, 2520 .removexattr = generic_removexattr, 2521 .check_acl = generic_check_acl, 2522#endif 2523}; 2524 2525static const struct super_operations shmem_ops = { 2526 .alloc_inode = shmem_alloc_inode, 2527 .destroy_inode = shmem_destroy_inode, 2528#ifdef CONFIG_TMPFS 2529 .statfs = shmem_statfs, 2530 .remount_fs = shmem_remount_fs, 2531 .show_options = shmem_show_options, 2532#endif 2533 .evict_inode = shmem_evict_inode, 2534 .drop_inode = generic_delete_inode, 2535 .put_super = shmem_put_super, 2536}; 2537 2538static const struct vm_operations_struct shmem_vm_ops = { 2539 .fault = shmem_fault, 2540#ifdef CONFIG_NUMA 2541 .set_policy = shmem_set_policy, 2542 .get_policy = shmem_get_policy, 2543#endif 2544}; 2545 2546 2547static struct dentry *shmem_mount(struct file_system_type *fs_type, 2548 int flags, const char *dev_name, void *data) 2549{ 2550 return mount_nodev(fs_type, flags, data, shmem_fill_super); 2551} 2552 2553static struct file_system_type tmpfs_fs_type = { 2554 .owner = THIS_MODULE, 2555 .name = "tmpfs", 2556 .mount = shmem_mount, 2557 .kill_sb = kill_litter_super, 2558}; 2559 2560int __init init_tmpfs(void) 2561{ 2562 int error; 2563 2564 error = bdi_init(&shmem_backing_dev_info); 2565 if (error) 2566 goto out4; 2567 2568 error = init_inodecache(); 2569 if (error) 2570 goto out3; 2571 2572 error = register_filesystem(&tmpfs_fs_type); 2573 if (error) { 2574 printk(KERN_ERR "Could not register tmpfs\n"); 2575 goto out2; 2576 } 2577 2578 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER, 2579 tmpfs_fs_type.name, NULL); 2580 if (IS_ERR(shm_mnt)) { 2581 error = PTR_ERR(shm_mnt); 2582 printk(KERN_ERR "Could not kern_mount tmpfs\n"); 2583 goto out1; 2584 } 2585 return 0; 2586 2587out1: 2588 unregister_filesystem(&tmpfs_fs_type); 2589out2: 2590 destroy_inodecache(); 2591out3: 2592 bdi_destroy(&shmem_backing_dev_info); 2593out4: 2594 shm_mnt = ERR_PTR(error); 2595 return error; 2596} 2597 2598#ifdef CONFIG_CGROUP_MEM_RES_CTLR 2599/** 2600 * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file 2601 * @inode: the inode to be searched 2602 * @pgoff: the offset to be searched 2603 * @pagep: the pointer for the found page to be stored 2604 * @ent: the pointer for the found swap entry to be stored 2605 * 2606 * If a page is found, refcount of it is incremented. Callers should handle 2607 * these refcount. 2608 */ 2609void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t pgoff, 2610 struct page **pagep, swp_entry_t *ent) 2611{ 2612 swp_entry_t entry = { .val = 0 }, *ptr; 2613 struct page *page = NULL; 2614 struct shmem_inode_info *info = SHMEM_I(inode); 2615 2616 if ((pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 2617 goto out; 2618 2619 spin_lock(&info->lock); 2620 ptr = shmem_swp_entry(info, pgoff, NULL); 2621#ifdef CONFIG_SWAP 2622 if (ptr && ptr->val) { 2623 entry.val = ptr->val; 2624 page = find_get_page(&swapper_space, entry.val); 2625 } else 2626#endif 2627 page = find_get_page(inode->i_mapping, pgoff); 2628 if (ptr) 2629 shmem_swp_unmap(ptr); 2630 spin_unlock(&info->lock); 2631out: 2632 *pagep = page; 2633 *ent = entry; 2634} 2635#endif 2636 2637#else /* !CONFIG_SHMEM */ 2638 2639/* 2640 * tiny-shmem: simple shmemfs and tmpfs using ramfs code 2641 * 2642 * This is intended for small system where the benefits of the full 2643 * shmem code (swap-backed and resource-limited) are outweighed by 2644 * their complexity. On systems without swap this code should be 2645 * effectively equivalent, but much lighter weight. 2646 */ 2647 2648#include <linux/ramfs.h> 2649 2650static struct file_system_type tmpfs_fs_type = { 2651 .name = "tmpfs", 2652 .mount = ramfs_mount, 2653 .kill_sb = kill_litter_super, 2654}; 2655 2656int __init init_tmpfs(void) 2657{ 2658 BUG_ON(register_filesystem(&tmpfs_fs_type) != 0); 2659 2660 shm_mnt = kern_mount(&tmpfs_fs_type); 2661 BUG_ON(IS_ERR(shm_mnt)); 2662 2663 return 0; 2664} 2665 2666int shmem_unuse(swp_entry_t entry, struct page *page) 2667{ 2668 return 0; 2669} 2670 2671int shmem_lock(struct file *file, int lock, struct user_struct *user) 2672{ 2673 return 0; 2674} 2675 2676#ifdef CONFIG_CGROUP_MEM_RES_CTLR 2677/** 2678 * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file 2679 * @inode: the inode to be searched 2680 * @pgoff: the offset to be searched 2681 * @pagep: the pointer for the found page to be stored 2682 * @ent: the pointer for the found swap entry to be stored 2683 * 2684 * If a page is found, refcount of it is incremented. Callers should handle 2685 * these refcount. 2686 */ 2687void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t pgoff, 2688 struct page **pagep, swp_entry_t *ent) 2689{ 2690 struct page *page = NULL; 2691 2692 if ((pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 2693 goto out; 2694 page = find_get_page(inode->i_mapping, pgoff); 2695out: 2696 *pagep = page; 2697 *ent = (swp_entry_t){ .val = 0 }; 2698} 2699#endif 2700 2701#define shmem_vm_ops generic_file_vm_ops 2702#define shmem_file_operations ramfs_file_operations 2703#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev) 2704#define shmem_acct_size(flags, size) 0 2705#define shmem_unacct_size(flags, size) do {} while (0) 2706#define SHMEM_MAX_BYTES MAX_LFS_FILESIZE 2707 2708#endif /* CONFIG_SHMEM */ 2709 2710/* common code */ 2711 2712/** 2713 * shmem_file_setup - get an unlinked file living in tmpfs 2714 * @name: name for dentry (to be seen in /proc/<pid>/maps 2715 * @size: size to be set for the file 2716 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size 2717 */ 2718struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) 2719{ 2720 int error; 2721 struct file *file; 2722 struct inode *inode; 2723 struct path path; 2724 struct dentry *root; 2725 struct qstr this; 2726 2727 if (IS_ERR(shm_mnt)) 2728 return (void *)shm_mnt; 2729 2730 if (size < 0 || size > SHMEM_MAX_BYTES) 2731 return ERR_PTR(-EINVAL); 2732 2733 if (shmem_acct_size(flags, size)) 2734 return ERR_PTR(-ENOMEM); 2735 2736 error = -ENOMEM; 2737 this.name = name; 2738 this.len = strlen(name); 2739 this.hash = 0; /* will go */ 2740 root = shm_mnt->mnt_root; 2741 path.dentry = d_alloc(root, &this); 2742 if (!path.dentry) 2743 goto put_memory; 2744 path.mnt = mntget(shm_mnt); 2745 2746 error = -ENOSPC; 2747 inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags); 2748 if (!inode) 2749 goto put_dentry; 2750 2751 d_instantiate(path.dentry, inode); 2752 inode->i_size = size; 2753 inode->i_nlink = 0; /* It is unlinked */ 2754#ifndef CONFIG_MMU 2755 error = ramfs_nommu_expand_for_mapping(inode, size); 2756 if (error) 2757 goto put_dentry; 2758#endif 2759 2760 error = -ENFILE; 2761 file = alloc_file(&path, FMODE_WRITE | FMODE_READ, 2762 &shmem_file_operations); 2763 if (!file) 2764 goto put_dentry; 2765 2766 return file; 2767 2768put_dentry: 2769 path_put(&path); 2770put_memory: 2771 shmem_unacct_size(flags, size); 2772 return ERR_PTR(error); 2773} 2774EXPORT_SYMBOL_GPL(shmem_file_setup); 2775 2776/** 2777 * shmem_zero_setup - setup a shared anonymous mapping 2778 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff 2779 */ 2780int shmem_zero_setup(struct vm_area_struct *vma) 2781{ 2782 struct file *file; 2783 loff_t size = vma->vm_end - vma->vm_start; 2784 2785 file = shmem_file_setup("dev/zero", size, vma->vm_flags); 2786 if (IS_ERR(file)) 2787 return PTR_ERR(file); 2788 2789 if (vma->vm_file) 2790 fput(vma->vm_file); 2791 vma->vm_file = file; 2792 vma->vm_ops = &shmem_vm_ops; 2793 return 0; 2794} 2795