shmem.c revision 454abafe9d8beb3fe7da06131b3d81d6d352a959
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/swap.h> 32 33static struct vfsmount *shm_mnt; 34 35#ifdef CONFIG_SHMEM 36/* 37 * This virtual memory filesystem is heavily based on the ramfs. It 38 * extends ramfs by the ability to use swap and honor resource limits 39 * which makes it a completely usable filesystem. 40 */ 41 42#include <linux/xattr.h> 43#include <linux/exportfs.h> 44#include <linux/posix_acl.h> 45#include <linux/generic_acl.h> 46#include <linux/mman.h> 47#include <linux/string.h> 48#include <linux/slab.h> 49#include <linux/backing-dev.h> 50#include <linux/shmem_fs.h> 51#include <linux/writeback.h> 52#include <linux/blkdev.h> 53#include <linux/security.h> 54#include <linux/swapops.h> 55#include <linux/mempolicy.h> 56#include <linux/namei.h> 57#include <linux/ctype.h> 58#include <linux/migrate.h> 59#include <linux/highmem.h> 60#include <linux/seq_file.h> 61#include <linux/magic.h> 62 63#include <asm/uaccess.h> 64#include <asm/div64.h> 65#include <asm/pgtable.h> 66 67/* 68 * The maximum size of a shmem/tmpfs file is limited by the maximum size of 69 * its triple-indirect swap vector - see illustration at shmem_swp_entry(). 70 * 71 * With 4kB page size, maximum file size is just over 2TB on a 32-bit kernel, 72 * but one eighth of that on a 64-bit kernel. With 8kB page size, maximum 73 * file size is just over 4TB on a 64-bit kernel, but 16TB on a 32-bit kernel, 74 * MAX_LFS_FILESIZE being then more restrictive than swap vector layout. 75 * 76 * We use / and * instead of shifts in the definitions below, so that the swap 77 * vector can be tested with small even values (e.g. 20) for ENTRIES_PER_PAGE. 78 */ 79#define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long)) 80#define ENTRIES_PER_PAGEPAGE ((unsigned long long)ENTRIES_PER_PAGE*ENTRIES_PER_PAGE) 81 82#define SHMSWP_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1)) 83#define SHMSWP_MAX_BYTES (SHMSWP_MAX_INDEX << PAGE_CACHE_SHIFT) 84 85#define SHMEM_MAX_BYTES min_t(unsigned long long, SHMSWP_MAX_BYTES, MAX_LFS_FILESIZE) 86#define SHMEM_MAX_INDEX ((unsigned long)((SHMEM_MAX_BYTES+1) >> PAGE_CACHE_SHIFT)) 87 88#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512) 89#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT) 90 91/* info->flags needs VM_flags to handle pagein/truncate races efficiently */ 92#define SHMEM_PAGEIN VM_READ 93#define SHMEM_TRUNCATE VM_WRITE 94 95/* Definition to limit shmem_truncate's steps between cond_rescheds */ 96#define LATENCY_LIMIT 64 97 98/* Pretend that each entry is of this size in directory's i_size */ 99#define BOGO_DIRENT_SIZE 20 100 101/* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */ 102enum sgp_type { 103 SGP_READ, /* don't exceed i_size, don't allocate page */ 104 SGP_CACHE, /* don't exceed i_size, may allocate page */ 105 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */ 106 SGP_WRITE, /* may exceed i_size, may allocate page */ 107}; 108 109#ifdef CONFIG_TMPFS 110static unsigned long shmem_default_max_blocks(void) 111{ 112 return totalram_pages / 2; 113} 114 115static unsigned long shmem_default_max_inodes(void) 116{ 117 return min(totalram_pages - totalhigh_pages, totalram_pages / 2); 118} 119#endif 120 121static int shmem_getpage(struct inode *inode, unsigned long idx, 122 struct page **pagep, enum sgp_type sgp, int *type); 123 124static inline struct page *shmem_dir_alloc(gfp_t gfp_mask) 125{ 126 /* 127 * The above definition of ENTRIES_PER_PAGE, and the use of 128 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE: 129 * might be reconsidered if it ever diverges from PAGE_SIZE. 130 * 131 * Mobility flags are masked out as swap vectors cannot move 132 */ 133 return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO, 134 PAGE_CACHE_SHIFT-PAGE_SHIFT); 135} 136 137static inline void shmem_dir_free(struct page *page) 138{ 139 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT); 140} 141 142static struct page **shmem_dir_map(struct page *page) 143{ 144 return (struct page **)kmap_atomic(page, KM_USER0); 145} 146 147static inline void shmem_dir_unmap(struct page **dir) 148{ 149 kunmap_atomic(dir, KM_USER0); 150} 151 152static swp_entry_t *shmem_swp_map(struct page *page) 153{ 154 return (swp_entry_t *)kmap_atomic(page, KM_USER1); 155} 156 157static inline void shmem_swp_balance_unmap(void) 158{ 159 /* 160 * When passing a pointer to an i_direct entry, to code which 161 * also handles indirect entries and so will shmem_swp_unmap, 162 * we must arrange for the preempt count to remain in balance. 163 * What kmap_atomic of a lowmem page does depends on config 164 * and architecture, so pretend to kmap_atomic some lowmem page. 165 */ 166 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1); 167} 168 169static inline void shmem_swp_unmap(swp_entry_t *entry) 170{ 171 kunmap_atomic(entry, KM_USER1); 172} 173 174static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) 175{ 176 return sb->s_fs_info; 177} 178 179/* 180 * shmem_file_setup pre-accounts the whole fixed size of a VM object, 181 * for shared memory and for shared anonymous (/dev/zero) mappings 182 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), 183 * consistent with the pre-accounting of private mappings ... 184 */ 185static inline int shmem_acct_size(unsigned long flags, loff_t size) 186{ 187 return (flags & VM_NORESERVE) ? 188 0 : security_vm_enough_memory_kern(VM_ACCT(size)); 189} 190 191static inline void shmem_unacct_size(unsigned long flags, loff_t size) 192{ 193 if (!(flags & VM_NORESERVE)) 194 vm_unacct_memory(VM_ACCT(size)); 195} 196 197/* 198 * ... whereas tmpfs objects are accounted incrementally as 199 * pages are allocated, in order to allow huge sparse files. 200 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, 201 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. 202 */ 203static inline int shmem_acct_block(unsigned long flags) 204{ 205 return (flags & VM_NORESERVE) ? 206 security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE)) : 0; 207} 208 209static inline void shmem_unacct_blocks(unsigned long flags, long pages) 210{ 211 if (flags & VM_NORESERVE) 212 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE)); 213} 214 215static const struct super_operations shmem_ops; 216static const struct address_space_operations shmem_aops; 217static const struct file_operations shmem_file_operations; 218static const struct inode_operations shmem_inode_operations; 219static const struct inode_operations shmem_dir_inode_operations; 220static const struct inode_operations shmem_special_inode_operations; 221static const struct vm_operations_struct shmem_vm_ops; 222 223static struct backing_dev_info shmem_backing_dev_info __read_mostly = { 224 .ra_pages = 0, /* No readahead */ 225 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED, 226 .unplug_io_fn = default_unplug_io_fn, 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 spin_lock(&sbinfo->stat_lock); 237 sbinfo->free_blocks += pages; 238 inode->i_blocks -= pages*BLOCKS_PER_PAGE; 239 spin_unlock(&sbinfo->stat_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 free_blocks against 1 not 0, 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 spin_lock(&sbinfo->stat_lock); 425 if (sbinfo->free_blocks <= 1) { 426 spin_unlock(&sbinfo->stat_lock); 427 return ERR_PTR(-ENOSPC); 428 } 429 sbinfo->free_blocks--; 430 inode->i_blocks += BLOCKS_PER_PAGE; 431 spin_unlock(&sbinfo->stat_lock); 432 } 433 434 spin_unlock(&info->lock); 435 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping)); 436 if (page) 437 set_page_private(page, 0); 438 spin_lock(&info->lock); 439 440 if (!page) { 441 shmem_free_blocks(inode, 1); 442 return ERR_PTR(-ENOMEM); 443 } 444 if (sgp != SGP_WRITE && 445 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { 446 entry = ERR_PTR(-EINVAL); 447 break; 448 } 449 if (info->next_index <= index) 450 info->next_index = index + 1; 451 } 452 if (page) { 453 /* another task gave its page, or truncated the file */ 454 shmem_free_blocks(inode, 1); 455 shmem_dir_free(page); 456 } 457 if (info->next_index <= index && !IS_ERR(entry)) 458 info->next_index = index + 1; 459 return entry; 460} 461 462/** 463 * shmem_free_swp - free some swap entries in a directory 464 * @dir: pointer to the directory 465 * @edir: pointer after last entry of the directory 466 * @punch_lock: pointer to spinlock when needed for the holepunch case 467 */ 468static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir, 469 spinlock_t *punch_lock) 470{ 471 spinlock_t *punch_unlock = NULL; 472 swp_entry_t *ptr; 473 int freed = 0; 474 475 for (ptr = dir; ptr < edir; ptr++) { 476 if (ptr->val) { 477 if (unlikely(punch_lock)) { 478 punch_unlock = punch_lock; 479 punch_lock = NULL; 480 spin_lock(punch_unlock); 481 if (!ptr->val) 482 continue; 483 } 484 free_swap_and_cache(*ptr); 485 *ptr = (swp_entry_t){0}; 486 freed++; 487 } 488 } 489 if (punch_unlock) 490 spin_unlock(punch_unlock); 491 return freed; 492} 493 494static int shmem_map_and_free_swp(struct page *subdir, int offset, 495 int limit, struct page ***dir, spinlock_t *punch_lock) 496{ 497 swp_entry_t *ptr; 498 int freed = 0; 499 500 ptr = shmem_swp_map(subdir); 501 for (; offset < limit; offset += LATENCY_LIMIT) { 502 int size = limit - offset; 503 if (size > LATENCY_LIMIT) 504 size = LATENCY_LIMIT; 505 freed += shmem_free_swp(ptr+offset, ptr+offset+size, 506 punch_lock); 507 if (need_resched()) { 508 shmem_swp_unmap(ptr); 509 if (*dir) { 510 shmem_dir_unmap(*dir); 511 *dir = NULL; 512 } 513 cond_resched(); 514 ptr = shmem_swp_map(subdir); 515 } 516 } 517 shmem_swp_unmap(ptr); 518 return freed; 519} 520 521static void shmem_free_pages(struct list_head *next) 522{ 523 struct page *page; 524 int freed = 0; 525 526 do { 527 page = container_of(next, struct page, lru); 528 next = next->next; 529 shmem_dir_free(page); 530 freed++; 531 if (freed >= LATENCY_LIMIT) { 532 cond_resched(); 533 freed = 0; 534 } 535 } while (next); 536} 537 538static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end) 539{ 540 struct shmem_inode_info *info = SHMEM_I(inode); 541 unsigned long idx; 542 unsigned long size; 543 unsigned long limit; 544 unsigned long stage; 545 unsigned long diroff; 546 struct page **dir; 547 struct page *topdir; 548 struct page *middir; 549 struct page *subdir; 550 swp_entry_t *ptr; 551 LIST_HEAD(pages_to_free); 552 long nr_pages_to_free = 0; 553 long nr_swaps_freed = 0; 554 int offset; 555 int freed; 556 int punch_hole; 557 spinlock_t *needs_lock; 558 spinlock_t *punch_lock; 559 unsigned long upper_limit; 560 561 inode->i_ctime = inode->i_mtime = CURRENT_TIME; 562 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 563 if (idx >= info->next_index) 564 return; 565 566 spin_lock(&info->lock); 567 info->flags |= SHMEM_TRUNCATE; 568 if (likely(end == (loff_t) -1)) { 569 limit = info->next_index; 570 upper_limit = SHMEM_MAX_INDEX; 571 info->next_index = idx; 572 needs_lock = NULL; 573 punch_hole = 0; 574 } else { 575 if (end + 1 >= inode->i_size) { /* we may free a little more */ 576 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >> 577 PAGE_CACHE_SHIFT; 578 upper_limit = SHMEM_MAX_INDEX; 579 } else { 580 limit = (end + 1) >> PAGE_CACHE_SHIFT; 581 upper_limit = limit; 582 } 583 needs_lock = &info->lock; 584 punch_hole = 1; 585 } 586 587 topdir = info->i_indirect; 588 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) { 589 info->i_indirect = NULL; 590 nr_pages_to_free++; 591 list_add(&topdir->lru, &pages_to_free); 592 } 593 spin_unlock(&info->lock); 594 595 if (info->swapped && idx < SHMEM_NR_DIRECT) { 596 ptr = info->i_direct; 597 size = limit; 598 if (size > SHMEM_NR_DIRECT) 599 size = SHMEM_NR_DIRECT; 600 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock); 601 } 602 603 /* 604 * If there are no indirect blocks or we are punching a hole 605 * below indirect blocks, nothing to be done. 606 */ 607 if (!topdir || limit <= SHMEM_NR_DIRECT) 608 goto done2; 609 610 /* 611 * The truncation case has already dropped info->lock, and we're safe 612 * because i_size and next_index have already been lowered, preventing 613 * access beyond. But in the punch_hole case, we still need to take 614 * the lock when updating the swap directory, because there might be 615 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or 616 * shmem_writepage. However, whenever we find we can remove a whole 617 * directory page (not at the misaligned start or end of the range), 618 * we first NULLify its pointer in the level above, and then have no 619 * need to take the lock when updating its contents: needs_lock and 620 * punch_lock (either pointing to info->lock or NULL) manage this. 621 */ 622 623 upper_limit -= SHMEM_NR_DIRECT; 624 limit -= SHMEM_NR_DIRECT; 625 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0; 626 offset = idx % ENTRIES_PER_PAGE; 627 idx -= offset; 628 629 dir = shmem_dir_map(topdir); 630 stage = ENTRIES_PER_PAGEPAGE/2; 631 if (idx < ENTRIES_PER_PAGEPAGE/2) { 632 middir = topdir; 633 diroff = idx/ENTRIES_PER_PAGE; 634 } else { 635 dir += ENTRIES_PER_PAGE/2; 636 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE; 637 while (stage <= idx) 638 stage += ENTRIES_PER_PAGEPAGE; 639 middir = *dir; 640 if (*dir) { 641 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) % 642 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE; 643 if (!diroff && !offset && upper_limit >= stage) { 644 if (needs_lock) { 645 spin_lock(needs_lock); 646 *dir = NULL; 647 spin_unlock(needs_lock); 648 needs_lock = NULL; 649 } else 650 *dir = NULL; 651 nr_pages_to_free++; 652 list_add(&middir->lru, &pages_to_free); 653 } 654 shmem_dir_unmap(dir); 655 dir = shmem_dir_map(middir); 656 } else { 657 diroff = 0; 658 offset = 0; 659 idx = stage; 660 } 661 } 662 663 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) { 664 if (unlikely(idx == stage)) { 665 shmem_dir_unmap(dir); 666 dir = shmem_dir_map(topdir) + 667 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 668 while (!*dir) { 669 dir++; 670 idx += ENTRIES_PER_PAGEPAGE; 671 if (idx >= limit) 672 goto done1; 673 } 674 stage = idx + ENTRIES_PER_PAGEPAGE; 675 middir = *dir; 676 if (punch_hole) 677 needs_lock = &info->lock; 678 if (upper_limit >= stage) { 679 if (needs_lock) { 680 spin_lock(needs_lock); 681 *dir = NULL; 682 spin_unlock(needs_lock); 683 needs_lock = NULL; 684 } else 685 *dir = NULL; 686 nr_pages_to_free++; 687 list_add(&middir->lru, &pages_to_free); 688 } 689 shmem_dir_unmap(dir); 690 cond_resched(); 691 dir = shmem_dir_map(middir); 692 diroff = 0; 693 } 694 punch_lock = needs_lock; 695 subdir = dir[diroff]; 696 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) { 697 if (needs_lock) { 698 spin_lock(needs_lock); 699 dir[diroff] = NULL; 700 spin_unlock(needs_lock); 701 punch_lock = NULL; 702 } else 703 dir[diroff] = NULL; 704 nr_pages_to_free++; 705 list_add(&subdir->lru, &pages_to_free); 706 } 707 if (subdir && page_private(subdir) /* has swap entries */) { 708 size = limit - idx; 709 if (size > ENTRIES_PER_PAGE) 710 size = ENTRIES_PER_PAGE; 711 freed = shmem_map_and_free_swp(subdir, 712 offset, size, &dir, punch_lock); 713 if (!dir) 714 dir = shmem_dir_map(middir); 715 nr_swaps_freed += freed; 716 if (offset || punch_lock) { 717 spin_lock(&info->lock); 718 set_page_private(subdir, 719 page_private(subdir) - freed); 720 spin_unlock(&info->lock); 721 } else 722 BUG_ON(page_private(subdir) != freed); 723 } 724 offset = 0; 725 } 726done1: 727 shmem_dir_unmap(dir); 728done2: 729 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) { 730 /* 731 * Call truncate_inode_pages again: racing shmem_unuse_inode 732 * may have swizzled a page in from swap since vmtruncate or 733 * generic_delete_inode did it, before we lowered next_index. 734 * Also, though shmem_getpage checks i_size before adding to 735 * cache, no recheck after: so fix the narrow window there too. 736 * 737 * Recalling truncate_inode_pages_range and unmap_mapping_range 738 * every time for punch_hole (which never got a chance to clear 739 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive, 740 * yet hardly ever necessary: try to optimize them out later. 741 */ 742 truncate_inode_pages_range(inode->i_mapping, start, end); 743 if (punch_hole) 744 unmap_mapping_range(inode->i_mapping, start, 745 end - start, 1); 746 } 747 748 spin_lock(&info->lock); 749 info->flags &= ~SHMEM_TRUNCATE; 750 info->swapped -= nr_swaps_freed; 751 if (nr_pages_to_free) 752 shmem_free_blocks(inode, nr_pages_to_free); 753 shmem_recalc_inode(inode); 754 spin_unlock(&info->lock); 755 756 /* 757 * Empty swap vector directory pages to be freed? 758 */ 759 if (!list_empty(&pages_to_free)) { 760 pages_to_free.prev->next = NULL; 761 shmem_free_pages(pages_to_free.next); 762 } 763} 764 765static void shmem_truncate(struct inode *inode) 766{ 767 shmem_truncate_range(inode, inode->i_size, (loff_t)-1); 768} 769 770static int shmem_notify_change(struct dentry *dentry, struct iattr *attr) 771{ 772 struct inode *inode = dentry->d_inode; 773 struct page *page = NULL; 774 int error; 775 776 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { 777 if (attr->ia_size < 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 cannnot miss it were 783 * it assigned to swap. 784 */ 785 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) { 786 (void) shmem_getpage(inode, 787 attr->ia_size>>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 (attr->ia_size) { 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 } 807 808 error = inode_change_ok(inode, attr); 809 if (!error) 810 error = inode_setattr(inode, attr); 811#ifdef CONFIG_TMPFS_POSIX_ACL 812 if (!error && (attr->ia_valid & ATTR_MODE)) 813 error = generic_acl_chmod(inode); 814#endif 815 if (page) 816 page_cache_release(page); 817 return error; 818} 819 820static void shmem_delete_inode(struct inode *inode) 821{ 822 struct shmem_inode_info *info = SHMEM_I(inode); 823 824 if (inode->i_op->truncate == shmem_truncate) { 825 truncate_inode_pages(inode->i_mapping, 0); 826 shmem_unacct_size(info->flags, inode->i_size); 827 inode->i_size = 0; 828 shmem_truncate(inode); 829 if (!list_empty(&info->swaplist)) { 830 mutex_lock(&shmem_swaplist_mutex); 831 list_del_init(&info->swaplist); 832 mutex_unlock(&shmem_swaplist_mutex); 833 } 834 } 835 BUG_ON(inode->i_blocks); 836 shmem_free_inode(inode->i_sb); 837 clear_inode(inode); 838} 839 840static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir) 841{ 842 swp_entry_t *ptr; 843 844 for (ptr = dir; ptr < edir; ptr++) { 845 if (ptr->val == entry.val) 846 return ptr - dir; 847 } 848 return -1; 849} 850 851static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page) 852{ 853 struct inode *inode; 854 unsigned long idx; 855 unsigned long size; 856 unsigned long limit; 857 unsigned long stage; 858 struct page **dir; 859 struct page *subdir; 860 swp_entry_t *ptr; 861 int offset; 862 int error; 863 864 idx = 0; 865 ptr = info->i_direct; 866 spin_lock(&info->lock); 867 if (!info->swapped) { 868 list_del_init(&info->swaplist); 869 goto lost2; 870 } 871 limit = info->next_index; 872 size = limit; 873 if (size > SHMEM_NR_DIRECT) 874 size = SHMEM_NR_DIRECT; 875 offset = shmem_find_swp(entry, ptr, ptr+size); 876 if (offset >= 0) 877 goto found; 878 if (!info->i_indirect) 879 goto lost2; 880 881 dir = shmem_dir_map(info->i_indirect); 882 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2; 883 884 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) { 885 if (unlikely(idx == stage)) { 886 shmem_dir_unmap(dir-1); 887 if (cond_resched_lock(&info->lock)) { 888 /* check it has not been truncated */ 889 if (limit > info->next_index) { 890 limit = info->next_index; 891 if (idx >= limit) 892 goto lost2; 893 } 894 } 895 dir = shmem_dir_map(info->i_indirect) + 896 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 897 while (!*dir) { 898 dir++; 899 idx += ENTRIES_PER_PAGEPAGE; 900 if (idx >= limit) 901 goto lost1; 902 } 903 stage = idx + ENTRIES_PER_PAGEPAGE; 904 subdir = *dir; 905 shmem_dir_unmap(dir); 906 dir = shmem_dir_map(subdir); 907 } 908 subdir = *dir; 909 if (subdir && page_private(subdir)) { 910 ptr = shmem_swp_map(subdir); 911 size = limit - idx; 912 if (size > ENTRIES_PER_PAGE) 913 size = ENTRIES_PER_PAGE; 914 offset = shmem_find_swp(entry, ptr, ptr+size); 915 shmem_swp_unmap(ptr); 916 if (offset >= 0) { 917 shmem_dir_unmap(dir); 918 goto found; 919 } 920 } 921 } 922lost1: 923 shmem_dir_unmap(dir-1); 924lost2: 925 spin_unlock(&info->lock); 926 return 0; 927found: 928 idx += offset; 929 inode = igrab(&info->vfs_inode); 930 spin_unlock(&info->lock); 931 932 /* 933 * Move _head_ to start search for next from here. 934 * But be careful: shmem_delete_inode checks list_empty without taking 935 * mutex, and there's an instant in list_move_tail when info->swaplist 936 * would appear empty, if it were the only one on shmem_swaplist. We 937 * could avoid doing it if inode NULL; or use this minor optimization. 938 */ 939 if (shmem_swaplist.next != &info->swaplist) 940 list_move_tail(&shmem_swaplist, &info->swaplist); 941 mutex_unlock(&shmem_swaplist_mutex); 942 943 error = 1; 944 if (!inode) 945 goto out; 946 /* 947 * Charge page using GFP_KERNEL while we can wait. 948 * Charged back to the user(not to caller) when swap account is used. 949 * add_to_page_cache() will be called with GFP_NOWAIT. 950 */ 951 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL); 952 if (error) 953 goto out; 954 error = radix_tree_preload(GFP_KERNEL); 955 if (error) { 956 mem_cgroup_uncharge_cache_page(page); 957 goto out; 958 } 959 error = 1; 960 961 spin_lock(&info->lock); 962 ptr = shmem_swp_entry(info, idx, NULL); 963 if (ptr && ptr->val == entry.val) { 964 error = add_to_page_cache_locked(page, inode->i_mapping, 965 idx, GFP_NOWAIT); 966 /* does mem_cgroup_uncharge_cache_page on error */ 967 } else /* we must compensate for our precharge above */ 968 mem_cgroup_uncharge_cache_page(page); 969 970 if (error == -EEXIST) { 971 struct page *filepage = find_get_page(inode->i_mapping, idx); 972 error = 1; 973 if (filepage) { 974 /* 975 * There might be a more uptodate page coming down 976 * from a stacked writepage: forget our swappage if so. 977 */ 978 if (PageUptodate(filepage)) 979 error = 0; 980 page_cache_release(filepage); 981 } 982 } 983 if (!error) { 984 delete_from_swap_cache(page); 985 set_page_dirty(page); 986 info->flags |= SHMEM_PAGEIN; 987 shmem_swp_set(info, ptr, 0); 988 swap_free(entry); 989 error = 1; /* not an error, but entry was found */ 990 } 991 if (ptr) 992 shmem_swp_unmap(ptr); 993 spin_unlock(&info->lock); 994 radix_tree_preload_end(); 995out: 996 unlock_page(page); 997 page_cache_release(page); 998 iput(inode); /* allows for NULL */ 999 return error; 1000} 1001 1002/* 1003 * shmem_unuse() search for an eventually swapped out shmem page. 1004 */ 1005int shmem_unuse(swp_entry_t entry, struct page *page) 1006{ 1007 struct list_head *p, *next; 1008 struct shmem_inode_info *info; 1009 int found = 0; 1010 1011 mutex_lock(&shmem_swaplist_mutex); 1012 list_for_each_safe(p, next, &shmem_swaplist) { 1013 info = list_entry(p, struct shmem_inode_info, swaplist); 1014 found = shmem_unuse_inode(info, entry, page); 1015 cond_resched(); 1016 if (found) 1017 goto out; 1018 } 1019 mutex_unlock(&shmem_swaplist_mutex); 1020 /* 1021 * Can some race bring us here? We've been holding page lock, 1022 * so I think not; but would rather try again later than BUG() 1023 */ 1024 unlock_page(page); 1025 page_cache_release(page); 1026out: 1027 return (found < 0) ? found : 0; 1028} 1029 1030/* 1031 * Move the page from the page cache to the swap cache. 1032 */ 1033static int shmem_writepage(struct page *page, struct writeback_control *wbc) 1034{ 1035 struct shmem_inode_info *info; 1036 swp_entry_t *entry, swap; 1037 struct address_space *mapping; 1038 unsigned long index; 1039 struct inode *inode; 1040 1041 BUG_ON(!PageLocked(page)); 1042 mapping = page->mapping; 1043 index = page->index; 1044 inode = mapping->host; 1045 info = SHMEM_I(inode); 1046 if (info->flags & VM_LOCKED) 1047 goto redirty; 1048 if (!total_swap_pages) 1049 goto redirty; 1050 1051 /* 1052 * shmem_backing_dev_info's capabilities prevent regular writeback or 1053 * sync from ever calling shmem_writepage; but a stacking filesystem 1054 * may use the ->writepage of its underlying filesystem, in which case 1055 * tmpfs should write out to swap only in response to memory pressure, 1056 * and not for the writeback threads or sync. However, in those cases, 1057 * we do still want to check if there's a redundant swappage to be 1058 * discarded. 1059 */ 1060 if (wbc->for_reclaim) 1061 swap = get_swap_page(); 1062 else 1063 swap.val = 0; 1064 1065 spin_lock(&info->lock); 1066 if (index >= info->next_index) { 1067 BUG_ON(!(info->flags & SHMEM_TRUNCATE)); 1068 goto unlock; 1069 } 1070 entry = shmem_swp_entry(info, index, NULL); 1071 if (entry->val) { 1072 /* 1073 * The more uptodate page coming down from a stacked 1074 * writepage should replace our old swappage. 1075 */ 1076 free_swap_and_cache(*entry); 1077 shmem_swp_set(info, entry, 0); 1078 } 1079 shmem_recalc_inode(inode); 1080 1081 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { 1082 remove_from_page_cache(page); 1083 shmem_swp_set(info, entry, swap.val); 1084 shmem_swp_unmap(entry); 1085 if (list_empty(&info->swaplist)) 1086 inode = igrab(inode); 1087 else 1088 inode = NULL; 1089 spin_unlock(&info->lock); 1090 swap_shmem_alloc(swap); 1091 BUG_ON(page_mapped(page)); 1092 page_cache_release(page); /* pagecache ref */ 1093 swap_writepage(page, wbc); 1094 if (inode) { 1095 mutex_lock(&shmem_swaplist_mutex); 1096 /* move instead of add in case we're racing */ 1097 list_move_tail(&info->swaplist, &shmem_swaplist); 1098 mutex_unlock(&shmem_swaplist_mutex); 1099 iput(inode); 1100 } 1101 return 0; 1102 } 1103 1104 shmem_swp_unmap(entry); 1105unlock: 1106 spin_unlock(&info->lock); 1107 /* 1108 * add_to_swap_cache() doesn't return -EEXIST, so we can safely 1109 * clear SWAP_HAS_CACHE flag. 1110 */ 1111 swapcache_free(swap, NULL); 1112redirty: 1113 set_page_dirty(page); 1114 if (wbc->for_reclaim) 1115 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ 1116 unlock_page(page); 1117 return 0; 1118} 1119 1120#ifdef CONFIG_NUMA 1121#ifdef CONFIG_TMPFS 1122static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) 1123{ 1124 char buffer[64]; 1125 1126 if (!mpol || mpol->mode == MPOL_DEFAULT) 1127 return; /* show nothing */ 1128 1129 mpol_to_str(buffer, sizeof(buffer), mpol, 1); 1130 1131 seq_printf(seq, ",mpol=%s", buffer); 1132} 1133 1134static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) 1135{ 1136 struct mempolicy *mpol = NULL; 1137 if (sbinfo->mpol) { 1138 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ 1139 mpol = sbinfo->mpol; 1140 mpol_get(mpol); 1141 spin_unlock(&sbinfo->stat_lock); 1142 } 1143 return mpol; 1144} 1145#endif /* CONFIG_TMPFS */ 1146 1147static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, 1148 struct shmem_inode_info *info, unsigned long idx) 1149{ 1150 struct mempolicy mpol, *spol; 1151 struct vm_area_struct pvma; 1152 struct page *page; 1153 1154 spol = mpol_cond_copy(&mpol, 1155 mpol_shared_policy_lookup(&info->policy, idx)); 1156 1157 /* Create a pseudo vma that just contains the policy */ 1158 pvma.vm_start = 0; 1159 pvma.vm_pgoff = idx; 1160 pvma.vm_ops = NULL; 1161 pvma.vm_policy = spol; 1162 page = swapin_readahead(entry, gfp, &pvma, 0); 1163 return page; 1164} 1165 1166static struct page *shmem_alloc_page(gfp_t gfp, 1167 struct shmem_inode_info *info, unsigned long idx) 1168{ 1169 struct vm_area_struct pvma; 1170 1171 /* Create a pseudo vma that just contains the policy */ 1172 pvma.vm_start = 0; 1173 pvma.vm_pgoff = idx; 1174 pvma.vm_ops = NULL; 1175 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx); 1176 1177 /* 1178 * alloc_page_vma() will drop the shared policy reference 1179 */ 1180 return alloc_page_vma(gfp, &pvma, 0); 1181} 1182#else /* !CONFIG_NUMA */ 1183#ifdef CONFIG_TMPFS 1184static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p) 1185{ 1186} 1187#endif /* CONFIG_TMPFS */ 1188 1189static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, 1190 struct shmem_inode_info *info, unsigned long idx) 1191{ 1192 return swapin_readahead(entry, gfp, NULL, 0); 1193} 1194 1195static inline struct page *shmem_alloc_page(gfp_t gfp, 1196 struct shmem_inode_info *info, unsigned long idx) 1197{ 1198 return alloc_page(gfp); 1199} 1200#endif /* CONFIG_NUMA */ 1201 1202#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS) 1203static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) 1204{ 1205 return NULL; 1206} 1207#endif 1208 1209/* 1210 * shmem_getpage - either get the page from swap or allocate a new one 1211 * 1212 * If we allocate a new one we do not mark it dirty. That's up to the 1213 * vm. If we swap it in we mark it dirty since we also free the swap 1214 * entry since a page cannot live in both the swap and page cache 1215 */ 1216static int shmem_getpage(struct inode *inode, unsigned long idx, 1217 struct page **pagep, enum sgp_type sgp, int *type) 1218{ 1219 struct address_space *mapping = inode->i_mapping; 1220 struct shmem_inode_info *info = SHMEM_I(inode); 1221 struct shmem_sb_info *sbinfo; 1222 struct page *filepage = *pagep; 1223 struct page *swappage; 1224 swp_entry_t *entry; 1225 swp_entry_t swap; 1226 gfp_t gfp; 1227 int error; 1228 1229 if (idx >= SHMEM_MAX_INDEX) 1230 return -EFBIG; 1231 1232 if (type) 1233 *type = 0; 1234 1235 /* 1236 * Normally, filepage is NULL on entry, and either found 1237 * uptodate immediately, or allocated and zeroed, or read 1238 * in under swappage, which is then assigned to filepage. 1239 * But shmem_readpage (required for splice) passes in a locked 1240 * filepage, which may be found not uptodate by other callers 1241 * too, and may need to be copied from the swappage read in. 1242 */ 1243repeat: 1244 if (!filepage) 1245 filepage = find_lock_page(mapping, idx); 1246 if (filepage && PageUptodate(filepage)) 1247 goto done; 1248 error = 0; 1249 gfp = mapping_gfp_mask(mapping); 1250 if (!filepage) { 1251 /* 1252 * Try to preload while we can wait, to not make a habit of 1253 * draining atomic reserves; but don't latch on to this cpu. 1254 */ 1255 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM); 1256 if (error) 1257 goto failed; 1258 radix_tree_preload_end(); 1259 } 1260 1261 spin_lock(&info->lock); 1262 shmem_recalc_inode(inode); 1263 entry = shmem_swp_alloc(info, idx, sgp); 1264 if (IS_ERR(entry)) { 1265 spin_unlock(&info->lock); 1266 error = PTR_ERR(entry); 1267 goto failed; 1268 } 1269 swap = *entry; 1270 1271 if (swap.val) { 1272 /* Look it up and read it in.. */ 1273 swappage = lookup_swap_cache(swap); 1274 if (!swappage) { 1275 shmem_swp_unmap(entry); 1276 /* here we actually do the io */ 1277 if (type && !(*type & VM_FAULT_MAJOR)) { 1278 __count_vm_event(PGMAJFAULT); 1279 *type |= VM_FAULT_MAJOR; 1280 } 1281 spin_unlock(&info->lock); 1282 swappage = shmem_swapin(swap, gfp, info, idx); 1283 if (!swappage) { 1284 spin_lock(&info->lock); 1285 entry = shmem_swp_alloc(info, idx, sgp); 1286 if (IS_ERR(entry)) 1287 error = PTR_ERR(entry); 1288 else { 1289 if (entry->val == swap.val) 1290 error = -ENOMEM; 1291 shmem_swp_unmap(entry); 1292 } 1293 spin_unlock(&info->lock); 1294 if (error) 1295 goto failed; 1296 goto repeat; 1297 } 1298 wait_on_page_locked(swappage); 1299 page_cache_release(swappage); 1300 goto repeat; 1301 } 1302 1303 /* We have to do this with page locked to prevent races */ 1304 if (!trylock_page(swappage)) { 1305 shmem_swp_unmap(entry); 1306 spin_unlock(&info->lock); 1307 wait_on_page_locked(swappage); 1308 page_cache_release(swappage); 1309 goto repeat; 1310 } 1311 if (PageWriteback(swappage)) { 1312 shmem_swp_unmap(entry); 1313 spin_unlock(&info->lock); 1314 wait_on_page_writeback(swappage); 1315 unlock_page(swappage); 1316 page_cache_release(swappage); 1317 goto repeat; 1318 } 1319 if (!PageUptodate(swappage)) { 1320 shmem_swp_unmap(entry); 1321 spin_unlock(&info->lock); 1322 unlock_page(swappage); 1323 page_cache_release(swappage); 1324 error = -EIO; 1325 goto failed; 1326 } 1327 1328 if (filepage) { 1329 shmem_swp_set(info, entry, 0); 1330 shmem_swp_unmap(entry); 1331 delete_from_swap_cache(swappage); 1332 spin_unlock(&info->lock); 1333 copy_highpage(filepage, swappage); 1334 unlock_page(swappage); 1335 page_cache_release(swappage); 1336 flush_dcache_page(filepage); 1337 SetPageUptodate(filepage); 1338 set_page_dirty(filepage); 1339 swap_free(swap); 1340 } else if (!(error = add_to_page_cache_locked(swappage, mapping, 1341 idx, GFP_NOWAIT))) { 1342 info->flags |= SHMEM_PAGEIN; 1343 shmem_swp_set(info, entry, 0); 1344 shmem_swp_unmap(entry); 1345 delete_from_swap_cache(swappage); 1346 spin_unlock(&info->lock); 1347 filepage = swappage; 1348 set_page_dirty(filepage); 1349 swap_free(swap); 1350 } else { 1351 shmem_swp_unmap(entry); 1352 spin_unlock(&info->lock); 1353 if (error == -ENOMEM) { 1354 /* 1355 * reclaim from proper memory cgroup and 1356 * call memcg's OOM if needed. 1357 */ 1358 error = mem_cgroup_shmem_charge_fallback( 1359 swappage, 1360 current->mm, 1361 gfp); 1362 if (error) { 1363 unlock_page(swappage); 1364 page_cache_release(swappage); 1365 goto failed; 1366 } 1367 } 1368 unlock_page(swappage); 1369 page_cache_release(swappage); 1370 goto repeat; 1371 } 1372 } else if (sgp == SGP_READ && !filepage) { 1373 shmem_swp_unmap(entry); 1374 filepage = find_get_page(mapping, idx); 1375 if (filepage && 1376 (!PageUptodate(filepage) || !trylock_page(filepage))) { 1377 spin_unlock(&info->lock); 1378 wait_on_page_locked(filepage); 1379 page_cache_release(filepage); 1380 filepage = NULL; 1381 goto repeat; 1382 } 1383 spin_unlock(&info->lock); 1384 } else { 1385 shmem_swp_unmap(entry); 1386 sbinfo = SHMEM_SB(inode->i_sb); 1387 if (sbinfo->max_blocks) { 1388 spin_lock(&sbinfo->stat_lock); 1389 if (sbinfo->free_blocks == 0 || 1390 shmem_acct_block(info->flags)) { 1391 spin_unlock(&sbinfo->stat_lock); 1392 spin_unlock(&info->lock); 1393 error = -ENOSPC; 1394 goto failed; 1395 } 1396 sbinfo->free_blocks--; 1397 inode->i_blocks += BLOCKS_PER_PAGE; 1398 spin_unlock(&sbinfo->stat_lock); 1399 } else if (shmem_acct_block(info->flags)) { 1400 spin_unlock(&info->lock); 1401 error = -ENOSPC; 1402 goto failed; 1403 } 1404 1405 if (!filepage) { 1406 int ret; 1407 1408 spin_unlock(&info->lock); 1409 filepage = shmem_alloc_page(gfp, info, idx); 1410 if (!filepage) { 1411 shmem_unacct_blocks(info->flags, 1); 1412 shmem_free_blocks(inode, 1); 1413 error = -ENOMEM; 1414 goto failed; 1415 } 1416 SetPageSwapBacked(filepage); 1417 1418 /* Precharge page while we can wait, compensate after */ 1419 error = mem_cgroup_cache_charge(filepage, current->mm, 1420 GFP_KERNEL); 1421 if (error) { 1422 page_cache_release(filepage); 1423 shmem_unacct_blocks(info->flags, 1); 1424 shmem_free_blocks(inode, 1); 1425 filepage = NULL; 1426 goto failed; 1427 } 1428 1429 spin_lock(&info->lock); 1430 entry = shmem_swp_alloc(info, idx, sgp); 1431 if (IS_ERR(entry)) 1432 error = PTR_ERR(entry); 1433 else { 1434 swap = *entry; 1435 shmem_swp_unmap(entry); 1436 } 1437 ret = error || swap.val; 1438 if (ret) 1439 mem_cgroup_uncharge_cache_page(filepage); 1440 else 1441 ret = add_to_page_cache_lru(filepage, mapping, 1442 idx, GFP_NOWAIT); 1443 /* 1444 * At add_to_page_cache_lru() failure, uncharge will 1445 * be done automatically. 1446 */ 1447 if (ret) { 1448 spin_unlock(&info->lock); 1449 page_cache_release(filepage); 1450 shmem_unacct_blocks(info->flags, 1); 1451 shmem_free_blocks(inode, 1); 1452 filepage = NULL; 1453 if (error) 1454 goto failed; 1455 goto repeat; 1456 } 1457 info->flags |= SHMEM_PAGEIN; 1458 } 1459 1460 info->alloced++; 1461 spin_unlock(&info->lock); 1462 clear_highpage(filepage); 1463 flush_dcache_page(filepage); 1464 SetPageUptodate(filepage); 1465 if (sgp == SGP_DIRTY) 1466 set_page_dirty(filepage); 1467 } 1468done: 1469 *pagep = filepage; 1470 return 0; 1471 1472failed: 1473 if (*pagep != filepage) { 1474 unlock_page(filepage); 1475 page_cache_release(filepage); 1476 } 1477 return error; 1478} 1479 1480static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1481{ 1482 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 1483 int error; 1484 int ret; 1485 1486 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 1487 return VM_FAULT_SIGBUS; 1488 1489 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret); 1490 if (error) 1491 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); 1492 1493 return ret | VM_FAULT_LOCKED; 1494} 1495 1496#ifdef CONFIG_NUMA 1497static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new) 1498{ 1499 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1500 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new); 1501} 1502 1503static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, 1504 unsigned long addr) 1505{ 1506 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1507 unsigned long idx; 1508 1509 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; 1510 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx); 1511} 1512#endif 1513 1514int shmem_lock(struct file *file, int lock, struct user_struct *user) 1515{ 1516 struct inode *inode = file->f_path.dentry->d_inode; 1517 struct shmem_inode_info *info = SHMEM_I(inode); 1518 int retval = -ENOMEM; 1519 1520 spin_lock(&info->lock); 1521 if (lock && !(info->flags & VM_LOCKED)) { 1522 if (!user_shm_lock(inode->i_size, user)) 1523 goto out_nomem; 1524 info->flags |= VM_LOCKED; 1525 mapping_set_unevictable(file->f_mapping); 1526 } 1527 if (!lock && (info->flags & VM_LOCKED) && user) { 1528 user_shm_unlock(inode->i_size, user); 1529 info->flags &= ~VM_LOCKED; 1530 mapping_clear_unevictable(file->f_mapping); 1531 scan_mapping_unevictable_pages(file->f_mapping); 1532 } 1533 retval = 0; 1534 1535out_nomem: 1536 spin_unlock(&info->lock); 1537 return retval; 1538} 1539 1540static int shmem_mmap(struct file *file, struct vm_area_struct *vma) 1541{ 1542 file_accessed(file); 1543 vma->vm_ops = &shmem_vm_ops; 1544 vma->vm_flags |= VM_CAN_NONLINEAR; 1545 return 0; 1546} 1547 1548static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir, 1549 int mode, dev_t dev, unsigned long flags) 1550{ 1551 struct inode *inode; 1552 struct shmem_inode_info *info; 1553 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 1554 1555 if (shmem_reserve_inode(sb)) 1556 return NULL; 1557 1558 inode = new_inode(sb); 1559 if (inode) { 1560 inode_init_owner(inode, dir, mode); 1561 inode->i_blocks = 0; 1562 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; 1563 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 1564 inode->i_generation = get_seconds(); 1565 info = SHMEM_I(inode); 1566 memset(info, 0, (char *)inode - (char *)info); 1567 spin_lock_init(&info->lock); 1568 info->flags = flags & VM_NORESERVE; 1569 INIT_LIST_HEAD(&info->swaplist); 1570 cache_no_acl(inode); 1571 1572 switch (mode & S_IFMT) { 1573 default: 1574 inode->i_op = &shmem_special_inode_operations; 1575 init_special_inode(inode, mode, dev); 1576 break; 1577 case S_IFREG: 1578 inode->i_mapping->a_ops = &shmem_aops; 1579 inode->i_op = &shmem_inode_operations; 1580 inode->i_fop = &shmem_file_operations; 1581 mpol_shared_policy_init(&info->policy, 1582 shmem_get_sbmpol(sbinfo)); 1583 break; 1584 case S_IFDIR: 1585 inc_nlink(inode); 1586 /* Some things misbehave if size == 0 on a directory */ 1587 inode->i_size = 2 * BOGO_DIRENT_SIZE; 1588 inode->i_op = &shmem_dir_inode_operations; 1589 inode->i_fop = &simple_dir_operations; 1590 break; 1591 case S_IFLNK: 1592 /* 1593 * Must not load anything in the rbtree, 1594 * mpol_free_shared_policy will not be called. 1595 */ 1596 mpol_shared_policy_init(&info->policy, NULL); 1597 break; 1598 } 1599 } else 1600 shmem_free_inode(sb); 1601 return inode; 1602} 1603 1604#ifdef CONFIG_TMPFS 1605static const struct inode_operations shmem_symlink_inode_operations; 1606static const struct inode_operations shmem_symlink_inline_operations; 1607 1608/* 1609 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin; 1610 * but providing them allows a tmpfs file to be used for splice, sendfile, and 1611 * below the loop driver, in the generic fashion that many filesystems support. 1612 */ 1613static int shmem_readpage(struct file *file, struct page *page) 1614{ 1615 struct inode *inode = page->mapping->host; 1616 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL); 1617 unlock_page(page); 1618 return error; 1619} 1620 1621static int 1622shmem_write_begin(struct file *file, struct address_space *mapping, 1623 loff_t pos, unsigned len, unsigned flags, 1624 struct page **pagep, void **fsdata) 1625{ 1626 struct inode *inode = mapping->host; 1627 pgoff_t index = pos >> PAGE_CACHE_SHIFT; 1628 *pagep = NULL; 1629 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL); 1630} 1631 1632static int 1633shmem_write_end(struct file *file, struct address_space *mapping, 1634 loff_t pos, unsigned len, unsigned copied, 1635 struct page *page, void *fsdata) 1636{ 1637 struct inode *inode = mapping->host; 1638 1639 if (pos + copied > inode->i_size) 1640 i_size_write(inode, pos + copied); 1641 1642 set_page_dirty(page); 1643 unlock_page(page); 1644 page_cache_release(page); 1645 1646 return copied; 1647} 1648 1649static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor) 1650{ 1651 struct inode *inode = filp->f_path.dentry->d_inode; 1652 struct address_space *mapping = inode->i_mapping; 1653 unsigned long index, offset; 1654 enum sgp_type sgp = SGP_READ; 1655 1656 /* 1657 * Might this read be for a stacking filesystem? Then when reading 1658 * holes of a sparse file, we actually need to allocate those pages, 1659 * and even mark them dirty, so it cannot exceed the max_blocks limit. 1660 */ 1661 if (segment_eq(get_fs(), KERNEL_DS)) 1662 sgp = SGP_DIRTY; 1663 1664 index = *ppos >> PAGE_CACHE_SHIFT; 1665 offset = *ppos & ~PAGE_CACHE_MASK; 1666 1667 for (;;) { 1668 struct page *page = NULL; 1669 unsigned long end_index, nr, ret; 1670 loff_t i_size = i_size_read(inode); 1671 1672 end_index = i_size >> PAGE_CACHE_SHIFT; 1673 if (index > end_index) 1674 break; 1675 if (index == end_index) { 1676 nr = i_size & ~PAGE_CACHE_MASK; 1677 if (nr <= offset) 1678 break; 1679 } 1680 1681 desc->error = shmem_getpage(inode, index, &page, sgp, NULL); 1682 if (desc->error) { 1683 if (desc->error == -EINVAL) 1684 desc->error = 0; 1685 break; 1686 } 1687 if (page) 1688 unlock_page(page); 1689 1690 /* 1691 * We must evaluate after, since reads (unlike writes) 1692 * are called without i_mutex protection against truncate 1693 */ 1694 nr = PAGE_CACHE_SIZE; 1695 i_size = i_size_read(inode); 1696 end_index = i_size >> PAGE_CACHE_SHIFT; 1697 if (index == end_index) { 1698 nr = i_size & ~PAGE_CACHE_MASK; 1699 if (nr <= offset) { 1700 if (page) 1701 page_cache_release(page); 1702 break; 1703 } 1704 } 1705 nr -= offset; 1706 1707 if (page) { 1708 /* 1709 * If users can be writing to this page using arbitrary 1710 * virtual addresses, take care about potential aliasing 1711 * before reading the page on the kernel side. 1712 */ 1713 if (mapping_writably_mapped(mapping)) 1714 flush_dcache_page(page); 1715 /* 1716 * Mark the page accessed if we read the beginning. 1717 */ 1718 if (!offset) 1719 mark_page_accessed(page); 1720 } else { 1721 page = ZERO_PAGE(0); 1722 page_cache_get(page); 1723 } 1724 1725 /* 1726 * Ok, we have the page, and it's up-to-date, so 1727 * now we can copy it to user space... 1728 * 1729 * The actor routine returns how many bytes were actually used.. 1730 * NOTE! This may not be the same as how much of a user buffer 1731 * we filled up (we may be padding etc), so we can only update 1732 * "pos" here (the actor routine has to update the user buffer 1733 * pointers and the remaining count). 1734 */ 1735 ret = actor(desc, page, offset, nr); 1736 offset += ret; 1737 index += offset >> PAGE_CACHE_SHIFT; 1738 offset &= ~PAGE_CACHE_MASK; 1739 1740 page_cache_release(page); 1741 if (ret != nr || !desc->count) 1742 break; 1743 1744 cond_resched(); 1745 } 1746 1747 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; 1748 file_accessed(filp); 1749} 1750 1751static ssize_t shmem_file_aio_read(struct kiocb *iocb, 1752 const struct iovec *iov, unsigned long nr_segs, loff_t pos) 1753{ 1754 struct file *filp = iocb->ki_filp; 1755 ssize_t retval; 1756 unsigned long seg; 1757 size_t count; 1758 loff_t *ppos = &iocb->ki_pos; 1759 1760 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); 1761 if (retval) 1762 return retval; 1763 1764 for (seg = 0; seg < nr_segs; seg++) { 1765 read_descriptor_t desc; 1766 1767 desc.written = 0; 1768 desc.arg.buf = iov[seg].iov_base; 1769 desc.count = iov[seg].iov_len; 1770 if (desc.count == 0) 1771 continue; 1772 desc.error = 0; 1773 do_shmem_file_read(filp, ppos, &desc, file_read_actor); 1774 retval += desc.written; 1775 if (desc.error) { 1776 retval = retval ?: desc.error; 1777 break; 1778 } 1779 if (desc.count > 0) 1780 break; 1781 } 1782 return retval; 1783} 1784 1785static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) 1786{ 1787 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); 1788 1789 buf->f_type = TMPFS_MAGIC; 1790 buf->f_bsize = PAGE_CACHE_SIZE; 1791 buf->f_namelen = NAME_MAX; 1792 spin_lock(&sbinfo->stat_lock); 1793 if (sbinfo->max_blocks) { 1794 buf->f_blocks = sbinfo->max_blocks; 1795 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks; 1796 } 1797 if (sbinfo->max_inodes) { 1798 buf->f_files = sbinfo->max_inodes; 1799 buf->f_ffree = sbinfo->free_inodes; 1800 } 1801 /* else leave those fields 0 like simple_statfs */ 1802 spin_unlock(&sbinfo->stat_lock); 1803 return 0; 1804} 1805 1806/* 1807 * File creation. Allocate an inode, and we're done.. 1808 */ 1809static int 1810shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 1811{ 1812 struct inode *inode; 1813 int error = -ENOSPC; 1814 1815 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE); 1816 if (inode) { 1817 error = security_inode_init_security(inode, dir, NULL, NULL, 1818 NULL); 1819 if (error) { 1820 if (error != -EOPNOTSUPP) { 1821 iput(inode); 1822 return error; 1823 } 1824 } 1825#ifdef CONFIG_TMPFS_POSIX_ACL 1826 error = generic_acl_init(inode, dir); 1827 if (error) { 1828 iput(inode); 1829 return error; 1830 } 1831#else 1832 error = 0; 1833#endif 1834 dir->i_size += BOGO_DIRENT_SIZE; 1835 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1836 d_instantiate(dentry, inode); 1837 dget(dentry); /* Extra count - pin the dentry in core */ 1838 } 1839 return error; 1840} 1841 1842static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1843{ 1844 int error; 1845 1846 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) 1847 return error; 1848 inc_nlink(dir); 1849 return 0; 1850} 1851 1852static int shmem_create(struct inode *dir, struct dentry *dentry, int mode, 1853 struct nameidata *nd) 1854{ 1855 return shmem_mknod(dir, dentry, mode | S_IFREG, 0); 1856} 1857 1858/* 1859 * Link a file.. 1860 */ 1861static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1862{ 1863 struct inode *inode = old_dentry->d_inode; 1864 int ret; 1865 1866 /* 1867 * No ordinary (disk based) filesystem counts links as inodes; 1868 * but each new link needs a new dentry, pinning lowmem, and 1869 * tmpfs dentries cannot be pruned until they are unlinked. 1870 */ 1871 ret = shmem_reserve_inode(inode->i_sb); 1872 if (ret) 1873 goto out; 1874 1875 dir->i_size += BOGO_DIRENT_SIZE; 1876 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1877 inc_nlink(inode); 1878 atomic_inc(&inode->i_count); /* New dentry reference */ 1879 dget(dentry); /* Extra pinning count for the created dentry */ 1880 d_instantiate(dentry, inode); 1881out: 1882 return ret; 1883} 1884 1885static int shmem_unlink(struct inode *dir, struct dentry *dentry) 1886{ 1887 struct inode *inode = dentry->d_inode; 1888 1889 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) 1890 shmem_free_inode(inode->i_sb); 1891 1892 dir->i_size -= BOGO_DIRENT_SIZE; 1893 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1894 drop_nlink(inode); 1895 dput(dentry); /* Undo the count from "create" - this does all the work */ 1896 return 0; 1897} 1898 1899static int shmem_rmdir(struct inode *dir, struct dentry *dentry) 1900{ 1901 if (!simple_empty(dentry)) 1902 return -ENOTEMPTY; 1903 1904 drop_nlink(dentry->d_inode); 1905 drop_nlink(dir); 1906 return shmem_unlink(dir, dentry); 1907} 1908 1909/* 1910 * The VFS layer already does all the dentry stuff for rename, 1911 * we just have to decrement the usage count for the target if 1912 * it exists so that the VFS layer correctly free's it when it 1913 * gets overwritten. 1914 */ 1915static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) 1916{ 1917 struct inode *inode = old_dentry->d_inode; 1918 int they_are_dirs = S_ISDIR(inode->i_mode); 1919 1920 if (!simple_empty(new_dentry)) 1921 return -ENOTEMPTY; 1922 1923 if (new_dentry->d_inode) { 1924 (void) shmem_unlink(new_dir, new_dentry); 1925 if (they_are_dirs) 1926 drop_nlink(old_dir); 1927 } else if (they_are_dirs) { 1928 drop_nlink(old_dir); 1929 inc_nlink(new_dir); 1930 } 1931 1932 old_dir->i_size -= BOGO_DIRENT_SIZE; 1933 new_dir->i_size += BOGO_DIRENT_SIZE; 1934 old_dir->i_ctime = old_dir->i_mtime = 1935 new_dir->i_ctime = new_dir->i_mtime = 1936 inode->i_ctime = CURRENT_TIME; 1937 return 0; 1938} 1939 1940static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1941{ 1942 int error; 1943 int len; 1944 struct inode *inode; 1945 struct page *page = NULL; 1946 char *kaddr; 1947 struct shmem_inode_info *info; 1948 1949 len = strlen(symname) + 1; 1950 if (len > PAGE_CACHE_SIZE) 1951 return -ENAMETOOLONG; 1952 1953 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE); 1954 if (!inode) 1955 return -ENOSPC; 1956 1957 error = security_inode_init_security(inode, dir, NULL, NULL, 1958 NULL); 1959 if (error) { 1960 if (error != -EOPNOTSUPP) { 1961 iput(inode); 1962 return error; 1963 } 1964 error = 0; 1965 } 1966 1967 info = SHMEM_I(inode); 1968 inode->i_size = len-1; 1969 if (len <= (char *)inode - (char *)info) { 1970 /* do it inline */ 1971 memcpy(info, symname, len); 1972 inode->i_op = &shmem_symlink_inline_operations; 1973 } else { 1974 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); 1975 if (error) { 1976 iput(inode); 1977 return error; 1978 } 1979 inode->i_mapping->a_ops = &shmem_aops; 1980 inode->i_op = &shmem_symlink_inode_operations; 1981 kaddr = kmap_atomic(page, KM_USER0); 1982 memcpy(kaddr, symname, len); 1983 kunmap_atomic(kaddr, KM_USER0); 1984 set_page_dirty(page); 1985 unlock_page(page); 1986 page_cache_release(page); 1987 } 1988 dir->i_size += BOGO_DIRENT_SIZE; 1989 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1990 d_instantiate(dentry, inode); 1991 dget(dentry); 1992 return 0; 1993} 1994 1995static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd) 1996{ 1997 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode)); 1998 return NULL; 1999} 2000 2001static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) 2002{ 2003 struct page *page = NULL; 2004 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); 2005 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page)); 2006 if (page) 2007 unlock_page(page); 2008 return page; 2009} 2010 2011static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 2012{ 2013 if (!IS_ERR(nd_get_link(nd))) { 2014 struct page *page = cookie; 2015 kunmap(page); 2016 mark_page_accessed(page); 2017 page_cache_release(page); 2018 } 2019} 2020 2021static const struct inode_operations shmem_symlink_inline_operations = { 2022 .readlink = generic_readlink, 2023 .follow_link = shmem_follow_link_inline, 2024}; 2025 2026static const struct inode_operations shmem_symlink_inode_operations = { 2027 .truncate = shmem_truncate, 2028 .readlink = generic_readlink, 2029 .follow_link = shmem_follow_link, 2030 .put_link = shmem_put_link, 2031}; 2032 2033#ifdef CONFIG_TMPFS_POSIX_ACL 2034/* 2035 * Superblocks without xattr inode operations will get security.* xattr 2036 * support from the VFS "for free". As soon as we have any other xattrs 2037 * like ACLs, we also need to implement the security.* handlers at 2038 * filesystem level, though. 2039 */ 2040 2041static size_t shmem_xattr_security_list(struct dentry *dentry, char *list, 2042 size_t list_len, const char *name, 2043 size_t name_len, int handler_flags) 2044{ 2045 return security_inode_listsecurity(dentry->d_inode, list, list_len); 2046} 2047 2048static int shmem_xattr_security_get(struct dentry *dentry, const char *name, 2049 void *buffer, size_t size, int handler_flags) 2050{ 2051 if (strcmp(name, "") == 0) 2052 return -EINVAL; 2053 return xattr_getsecurity(dentry->d_inode, name, buffer, size); 2054} 2055 2056static int shmem_xattr_security_set(struct dentry *dentry, const char *name, 2057 const void *value, size_t size, int flags, int handler_flags) 2058{ 2059 if (strcmp(name, "") == 0) 2060 return -EINVAL; 2061 return security_inode_setsecurity(dentry->d_inode, name, value, 2062 size, flags); 2063} 2064 2065static const struct xattr_handler shmem_xattr_security_handler = { 2066 .prefix = XATTR_SECURITY_PREFIX, 2067 .list = shmem_xattr_security_list, 2068 .get = shmem_xattr_security_get, 2069 .set = shmem_xattr_security_set, 2070}; 2071 2072static const struct xattr_handler *shmem_xattr_handlers[] = { 2073 &generic_acl_access_handler, 2074 &generic_acl_default_handler, 2075 &shmem_xattr_security_handler, 2076 NULL 2077}; 2078#endif 2079 2080static struct dentry *shmem_get_parent(struct dentry *child) 2081{ 2082 return ERR_PTR(-ESTALE); 2083} 2084 2085static int shmem_match(struct inode *ino, void *vfh) 2086{ 2087 __u32 *fh = vfh; 2088 __u64 inum = fh[2]; 2089 inum = (inum << 32) | fh[1]; 2090 return ino->i_ino == inum && fh[0] == ino->i_generation; 2091} 2092 2093static struct dentry *shmem_fh_to_dentry(struct super_block *sb, 2094 struct fid *fid, int fh_len, int fh_type) 2095{ 2096 struct inode *inode; 2097 struct dentry *dentry = NULL; 2098 u64 inum = fid->raw[2]; 2099 inum = (inum << 32) | fid->raw[1]; 2100 2101 if (fh_len < 3) 2102 return NULL; 2103 2104 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), 2105 shmem_match, fid->raw); 2106 if (inode) { 2107 dentry = d_find_alias(inode); 2108 iput(inode); 2109 } 2110 2111 return dentry; 2112} 2113 2114static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len, 2115 int connectable) 2116{ 2117 struct inode *inode = dentry->d_inode; 2118 2119 if (*len < 3) 2120 return 255; 2121 2122 if (hlist_unhashed(&inode->i_hash)) { 2123 /* Unfortunately insert_inode_hash is not idempotent, 2124 * so as we hash inodes here rather than at creation 2125 * time, we need a lock to ensure we only try 2126 * to do it once 2127 */ 2128 static DEFINE_SPINLOCK(lock); 2129 spin_lock(&lock); 2130 if (hlist_unhashed(&inode->i_hash)) 2131 __insert_inode_hash(inode, 2132 inode->i_ino + inode->i_generation); 2133 spin_unlock(&lock); 2134 } 2135 2136 fh[0] = inode->i_generation; 2137 fh[1] = inode->i_ino; 2138 fh[2] = ((__u64)inode->i_ino) >> 32; 2139 2140 *len = 3; 2141 return 1; 2142} 2143 2144static const struct export_operations shmem_export_ops = { 2145 .get_parent = shmem_get_parent, 2146 .encode_fh = shmem_encode_fh, 2147 .fh_to_dentry = shmem_fh_to_dentry, 2148}; 2149 2150static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, 2151 bool remount) 2152{ 2153 char *this_char, *value, *rest; 2154 2155 while (options != NULL) { 2156 this_char = options; 2157 for (;;) { 2158 /* 2159 * NUL-terminate this option: unfortunately, 2160 * mount options form a comma-separated list, 2161 * but mpol's nodelist may also contain commas. 2162 */ 2163 options = strchr(options, ','); 2164 if (options == NULL) 2165 break; 2166 options++; 2167 if (!isdigit(*options)) { 2168 options[-1] = '\0'; 2169 break; 2170 } 2171 } 2172 if (!*this_char) 2173 continue; 2174 if ((value = strchr(this_char,'=')) != NULL) { 2175 *value++ = 0; 2176 } else { 2177 printk(KERN_ERR 2178 "tmpfs: No value for mount option '%s'\n", 2179 this_char); 2180 return 1; 2181 } 2182 2183 if (!strcmp(this_char,"size")) { 2184 unsigned long long size; 2185 size = memparse(value,&rest); 2186 if (*rest == '%') { 2187 size <<= PAGE_SHIFT; 2188 size *= totalram_pages; 2189 do_div(size, 100); 2190 rest++; 2191 } 2192 if (*rest) 2193 goto bad_val; 2194 sbinfo->max_blocks = 2195 DIV_ROUND_UP(size, PAGE_CACHE_SIZE); 2196 } else if (!strcmp(this_char,"nr_blocks")) { 2197 sbinfo->max_blocks = memparse(value, &rest); 2198 if (*rest) 2199 goto bad_val; 2200 } else if (!strcmp(this_char,"nr_inodes")) { 2201 sbinfo->max_inodes = memparse(value, &rest); 2202 if (*rest) 2203 goto bad_val; 2204 } else if (!strcmp(this_char,"mode")) { 2205 if (remount) 2206 continue; 2207 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; 2208 if (*rest) 2209 goto bad_val; 2210 } else if (!strcmp(this_char,"uid")) { 2211 if (remount) 2212 continue; 2213 sbinfo->uid = simple_strtoul(value, &rest, 0); 2214 if (*rest) 2215 goto bad_val; 2216 } else if (!strcmp(this_char,"gid")) { 2217 if (remount) 2218 continue; 2219 sbinfo->gid = simple_strtoul(value, &rest, 0); 2220 if (*rest) 2221 goto bad_val; 2222 } else if (!strcmp(this_char,"mpol")) { 2223 if (mpol_parse_str(value, &sbinfo->mpol, 1)) 2224 goto bad_val; 2225 } else { 2226 printk(KERN_ERR "tmpfs: Bad mount option %s\n", 2227 this_char); 2228 return 1; 2229 } 2230 } 2231 return 0; 2232 2233bad_val: 2234 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", 2235 value, this_char); 2236 return 1; 2237 2238} 2239 2240static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) 2241{ 2242 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 2243 struct shmem_sb_info config = *sbinfo; 2244 unsigned long blocks; 2245 unsigned long inodes; 2246 int error = -EINVAL; 2247 2248 if (shmem_parse_options(data, &config, true)) 2249 return error; 2250 2251 spin_lock(&sbinfo->stat_lock); 2252 blocks = sbinfo->max_blocks - sbinfo->free_blocks; 2253 inodes = sbinfo->max_inodes - sbinfo->free_inodes; 2254 if (config.max_blocks < blocks) 2255 goto out; 2256 if (config.max_inodes < inodes) 2257 goto out; 2258 /* 2259 * Those tests also disallow limited->unlimited while any are in 2260 * use, so i_blocks will always be zero when max_blocks is zero; 2261 * but we must separately disallow unlimited->limited, because 2262 * in that case we have no record of how much is already in use. 2263 */ 2264 if (config.max_blocks && !sbinfo->max_blocks) 2265 goto out; 2266 if (config.max_inodes && !sbinfo->max_inodes) 2267 goto out; 2268 2269 error = 0; 2270 sbinfo->max_blocks = config.max_blocks; 2271 sbinfo->free_blocks = config.max_blocks - blocks; 2272 sbinfo->max_inodes = config.max_inodes; 2273 sbinfo->free_inodes = config.max_inodes - inodes; 2274 2275 mpol_put(sbinfo->mpol); 2276 sbinfo->mpol = config.mpol; /* transfers initial ref */ 2277out: 2278 spin_unlock(&sbinfo->stat_lock); 2279 return error; 2280} 2281 2282static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs) 2283{ 2284 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb); 2285 2286 if (sbinfo->max_blocks != shmem_default_max_blocks()) 2287 seq_printf(seq, ",size=%luk", 2288 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10)); 2289 if (sbinfo->max_inodes != shmem_default_max_inodes()) 2290 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); 2291 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) 2292 seq_printf(seq, ",mode=%03o", sbinfo->mode); 2293 if (sbinfo->uid != 0) 2294 seq_printf(seq, ",uid=%u", sbinfo->uid); 2295 if (sbinfo->gid != 0) 2296 seq_printf(seq, ",gid=%u", sbinfo->gid); 2297 shmem_show_mpol(seq, sbinfo->mpol); 2298 return 0; 2299} 2300#endif /* CONFIG_TMPFS */ 2301 2302static void shmem_put_super(struct super_block *sb) 2303{ 2304 kfree(sb->s_fs_info); 2305 sb->s_fs_info = NULL; 2306} 2307 2308int shmem_fill_super(struct super_block *sb, void *data, int silent) 2309{ 2310 struct inode *inode; 2311 struct dentry *root; 2312 struct shmem_sb_info *sbinfo; 2313 int err = -ENOMEM; 2314 2315 /* Round up to L1_CACHE_BYTES to resist false sharing */ 2316 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), 2317 L1_CACHE_BYTES), GFP_KERNEL); 2318 if (!sbinfo) 2319 return -ENOMEM; 2320 2321 sbinfo->mode = S_IRWXUGO | S_ISVTX; 2322 sbinfo->uid = current_fsuid(); 2323 sbinfo->gid = current_fsgid(); 2324 sb->s_fs_info = sbinfo; 2325 2326#ifdef CONFIG_TMPFS 2327 /* 2328 * Per default we only allow half of the physical ram per 2329 * tmpfs instance, limiting inodes to one per page of lowmem; 2330 * but the internal instance is left unlimited. 2331 */ 2332 if (!(sb->s_flags & MS_NOUSER)) { 2333 sbinfo->max_blocks = shmem_default_max_blocks(); 2334 sbinfo->max_inodes = shmem_default_max_inodes(); 2335 if (shmem_parse_options(data, sbinfo, false)) { 2336 err = -EINVAL; 2337 goto failed; 2338 } 2339 } 2340 sb->s_export_op = &shmem_export_ops; 2341#else 2342 sb->s_flags |= MS_NOUSER; 2343#endif 2344 2345 spin_lock_init(&sbinfo->stat_lock); 2346 sbinfo->free_blocks = sbinfo->max_blocks; 2347 sbinfo->free_inodes = sbinfo->max_inodes; 2348 2349 sb->s_maxbytes = SHMEM_MAX_BYTES; 2350 sb->s_blocksize = PAGE_CACHE_SIZE; 2351 sb->s_blocksize_bits = PAGE_CACHE_SHIFT; 2352 sb->s_magic = TMPFS_MAGIC; 2353 sb->s_op = &shmem_ops; 2354 sb->s_time_gran = 1; 2355#ifdef CONFIG_TMPFS_POSIX_ACL 2356 sb->s_xattr = shmem_xattr_handlers; 2357 sb->s_flags |= MS_POSIXACL; 2358#endif 2359 2360 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); 2361 if (!inode) 2362 goto failed; 2363 inode->i_uid = sbinfo->uid; 2364 inode->i_gid = sbinfo->gid; 2365 root = d_alloc_root(inode); 2366 if (!root) 2367 goto failed_iput; 2368 sb->s_root = root; 2369 return 0; 2370 2371failed_iput: 2372 iput(inode); 2373failed: 2374 shmem_put_super(sb); 2375 return err; 2376} 2377 2378static struct kmem_cache *shmem_inode_cachep; 2379 2380static struct inode *shmem_alloc_inode(struct super_block *sb) 2381{ 2382 struct shmem_inode_info *p; 2383 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); 2384 if (!p) 2385 return NULL; 2386 return &p->vfs_inode; 2387} 2388 2389static void shmem_destroy_inode(struct inode *inode) 2390{ 2391 if ((inode->i_mode & S_IFMT) == S_IFREG) { 2392 /* only struct inode is valid if it's an inline symlink */ 2393 mpol_free_shared_policy(&SHMEM_I(inode)->policy); 2394 } 2395 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); 2396} 2397 2398static void init_once(void *foo) 2399{ 2400 struct shmem_inode_info *p = (struct shmem_inode_info *) foo; 2401 2402 inode_init_once(&p->vfs_inode); 2403} 2404 2405static int init_inodecache(void) 2406{ 2407 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", 2408 sizeof(struct shmem_inode_info), 2409 0, SLAB_PANIC, init_once); 2410 return 0; 2411} 2412 2413static void destroy_inodecache(void) 2414{ 2415 kmem_cache_destroy(shmem_inode_cachep); 2416} 2417 2418static const struct address_space_operations shmem_aops = { 2419 .writepage = shmem_writepage, 2420 .set_page_dirty = __set_page_dirty_no_writeback, 2421#ifdef CONFIG_TMPFS 2422 .readpage = shmem_readpage, 2423 .write_begin = shmem_write_begin, 2424 .write_end = shmem_write_end, 2425#endif 2426 .migratepage = migrate_page, 2427 .error_remove_page = generic_error_remove_page, 2428}; 2429 2430static const struct file_operations shmem_file_operations = { 2431 .mmap = shmem_mmap, 2432#ifdef CONFIG_TMPFS 2433 .llseek = generic_file_llseek, 2434 .read = do_sync_read, 2435 .write = do_sync_write, 2436 .aio_read = shmem_file_aio_read, 2437 .aio_write = generic_file_aio_write, 2438 .fsync = simple_sync_file, 2439 .splice_read = generic_file_splice_read, 2440 .splice_write = generic_file_splice_write, 2441#endif 2442}; 2443 2444static const struct inode_operations shmem_inode_operations = { 2445 .truncate = shmem_truncate, 2446 .setattr = shmem_notify_change, 2447 .truncate_range = shmem_truncate_range, 2448#ifdef CONFIG_TMPFS_POSIX_ACL 2449 .setxattr = generic_setxattr, 2450 .getxattr = generic_getxattr, 2451 .listxattr = generic_listxattr, 2452 .removexattr = generic_removexattr, 2453 .check_acl = generic_check_acl, 2454#endif 2455 2456}; 2457 2458static const struct inode_operations shmem_dir_inode_operations = { 2459#ifdef CONFIG_TMPFS 2460 .create = shmem_create, 2461 .lookup = simple_lookup, 2462 .link = shmem_link, 2463 .unlink = shmem_unlink, 2464 .symlink = shmem_symlink, 2465 .mkdir = shmem_mkdir, 2466 .rmdir = shmem_rmdir, 2467 .mknod = shmem_mknod, 2468 .rename = shmem_rename, 2469#endif 2470#ifdef CONFIG_TMPFS_POSIX_ACL 2471 .setattr = shmem_notify_change, 2472 .setxattr = generic_setxattr, 2473 .getxattr = generic_getxattr, 2474 .listxattr = generic_listxattr, 2475 .removexattr = generic_removexattr, 2476 .check_acl = generic_check_acl, 2477#endif 2478}; 2479 2480static const struct inode_operations shmem_special_inode_operations = { 2481#ifdef CONFIG_TMPFS_POSIX_ACL 2482 .setattr = shmem_notify_change, 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 2491static const struct super_operations shmem_ops = { 2492 .alloc_inode = shmem_alloc_inode, 2493 .destroy_inode = shmem_destroy_inode, 2494#ifdef CONFIG_TMPFS 2495 .statfs = shmem_statfs, 2496 .remount_fs = shmem_remount_fs, 2497 .show_options = shmem_show_options, 2498#endif 2499 .delete_inode = shmem_delete_inode, 2500 .drop_inode = generic_delete_inode, 2501 .put_super = shmem_put_super, 2502}; 2503 2504static const struct vm_operations_struct shmem_vm_ops = { 2505 .fault = shmem_fault, 2506#ifdef CONFIG_NUMA 2507 .set_policy = shmem_set_policy, 2508 .get_policy = shmem_get_policy, 2509#endif 2510}; 2511 2512 2513static int shmem_get_sb(struct file_system_type *fs_type, 2514 int flags, const char *dev_name, void *data, struct vfsmount *mnt) 2515{ 2516 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt); 2517} 2518 2519static struct file_system_type tmpfs_fs_type = { 2520 .owner = THIS_MODULE, 2521 .name = "tmpfs", 2522 .get_sb = shmem_get_sb, 2523 .kill_sb = kill_litter_super, 2524}; 2525 2526int __init init_tmpfs(void) 2527{ 2528 int error; 2529 2530 error = bdi_init(&shmem_backing_dev_info); 2531 if (error) 2532 goto out4; 2533 2534 error = init_inodecache(); 2535 if (error) 2536 goto out3; 2537 2538 error = register_filesystem(&tmpfs_fs_type); 2539 if (error) { 2540 printk(KERN_ERR "Could not register tmpfs\n"); 2541 goto out2; 2542 } 2543 2544 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER, 2545 tmpfs_fs_type.name, NULL); 2546 if (IS_ERR(shm_mnt)) { 2547 error = PTR_ERR(shm_mnt); 2548 printk(KERN_ERR "Could not kern_mount tmpfs\n"); 2549 goto out1; 2550 } 2551 return 0; 2552 2553out1: 2554 unregister_filesystem(&tmpfs_fs_type); 2555out2: 2556 destroy_inodecache(); 2557out3: 2558 bdi_destroy(&shmem_backing_dev_info); 2559out4: 2560 shm_mnt = ERR_PTR(error); 2561 return error; 2562} 2563 2564#else /* !CONFIG_SHMEM */ 2565 2566/* 2567 * tiny-shmem: simple shmemfs and tmpfs using ramfs code 2568 * 2569 * This is intended for small system where the benefits of the full 2570 * shmem code (swap-backed and resource-limited) are outweighed by 2571 * their complexity. On systems without swap this code should be 2572 * effectively equivalent, but much lighter weight. 2573 */ 2574 2575#include <linux/ramfs.h> 2576 2577static struct file_system_type tmpfs_fs_type = { 2578 .name = "tmpfs", 2579 .get_sb = ramfs_get_sb, 2580 .kill_sb = kill_litter_super, 2581}; 2582 2583int __init init_tmpfs(void) 2584{ 2585 BUG_ON(register_filesystem(&tmpfs_fs_type) != 0); 2586 2587 shm_mnt = kern_mount(&tmpfs_fs_type); 2588 BUG_ON(IS_ERR(shm_mnt)); 2589 2590 return 0; 2591} 2592 2593int shmem_unuse(swp_entry_t entry, struct page *page) 2594{ 2595 return 0; 2596} 2597 2598int shmem_lock(struct file *file, int lock, struct user_struct *user) 2599{ 2600 return 0; 2601} 2602 2603#define shmem_vm_ops generic_file_vm_ops 2604#define shmem_file_operations ramfs_file_operations 2605#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev) 2606#define shmem_acct_size(flags, size) 0 2607#define shmem_unacct_size(flags, size) do {} while (0) 2608#define SHMEM_MAX_BYTES MAX_LFS_FILESIZE 2609 2610#endif /* CONFIG_SHMEM */ 2611 2612/* common code */ 2613 2614/** 2615 * shmem_file_setup - get an unlinked file living in tmpfs 2616 * @name: name for dentry (to be seen in /proc/<pid>/maps 2617 * @size: size to be set for the file 2618 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size 2619 */ 2620struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) 2621{ 2622 int error; 2623 struct file *file; 2624 struct inode *inode; 2625 struct path path; 2626 struct dentry *root; 2627 struct qstr this; 2628 2629 if (IS_ERR(shm_mnt)) 2630 return (void *)shm_mnt; 2631 2632 if (size < 0 || size > SHMEM_MAX_BYTES) 2633 return ERR_PTR(-EINVAL); 2634 2635 if (shmem_acct_size(flags, size)) 2636 return ERR_PTR(-ENOMEM); 2637 2638 error = -ENOMEM; 2639 this.name = name; 2640 this.len = strlen(name); 2641 this.hash = 0; /* will go */ 2642 root = shm_mnt->mnt_root; 2643 path.dentry = d_alloc(root, &this); 2644 if (!path.dentry) 2645 goto put_memory; 2646 path.mnt = mntget(shm_mnt); 2647 2648 error = -ENOSPC; 2649 inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags); 2650 if (!inode) 2651 goto put_dentry; 2652 2653 d_instantiate(path.dentry, inode); 2654 inode->i_size = size; 2655 inode->i_nlink = 0; /* It is unlinked */ 2656#ifndef CONFIG_MMU 2657 error = ramfs_nommu_expand_for_mapping(inode, size); 2658 if (error) 2659 goto put_dentry; 2660#endif 2661 2662 error = -ENFILE; 2663 file = alloc_file(&path, FMODE_WRITE | FMODE_READ, 2664 &shmem_file_operations); 2665 if (!file) 2666 goto put_dentry; 2667 2668 return file; 2669 2670put_dentry: 2671 path_put(&path); 2672put_memory: 2673 shmem_unacct_size(flags, size); 2674 return ERR_PTR(error); 2675} 2676EXPORT_SYMBOL_GPL(shmem_file_setup); 2677 2678/** 2679 * shmem_zero_setup - setup a shared anonymous mapping 2680 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff 2681 */ 2682int shmem_zero_setup(struct vm_area_struct *vma) 2683{ 2684 struct file *file; 2685 loff_t size = vma->vm_end - vma->vm_start; 2686 2687 file = shmem_file_setup("dev/zero", size, vma->vm_flags); 2688 if (IS_ERR(file)) 2689 return PTR_ERR(file); 2690 2691 if (vma->vm_file) 2692 fput(vma->vm_file); 2693 vma->vm_file = file; 2694 vma->vm_ops = &shmem_vm_ops; 2695 return 0; 2696} 2697