shmem.c revision cff397e6b3c5bd67e26946792e81ab064c9acc1f
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#include <linux/ima.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/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 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, &shmem_acl_ops); 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); 1020out: return found; /* 0 or 1 or -ENOMEM */ 1021} 1022 1023/* 1024 * Move the page from the page cache to the swap cache. 1025 */ 1026static int shmem_writepage(struct page *page, struct writeback_control *wbc) 1027{ 1028 struct shmem_inode_info *info; 1029 swp_entry_t *entry, swap; 1030 struct address_space *mapping; 1031 unsigned long index; 1032 struct inode *inode; 1033 1034 BUG_ON(!PageLocked(page)); 1035 mapping = page->mapping; 1036 index = page->index; 1037 inode = mapping->host; 1038 info = SHMEM_I(inode); 1039 if (info->flags & VM_LOCKED) 1040 goto redirty; 1041 if (!total_swap_pages) 1042 goto redirty; 1043 1044 /* 1045 * shmem_backing_dev_info's capabilities prevent regular writeback or 1046 * sync from ever calling shmem_writepage; but a stacking filesystem 1047 * may use the ->writepage of its underlying filesystem, in which case 1048 * tmpfs should write out to swap only in response to memory pressure, 1049 * and not for pdflush or sync. However, in those cases, we do still 1050 * want to check if there's a redundant swappage to be discarded. 1051 */ 1052 if (wbc->for_reclaim) 1053 swap = get_swap_page(); 1054 else 1055 swap.val = 0; 1056 1057 spin_lock(&info->lock); 1058 if (index >= info->next_index) { 1059 BUG_ON(!(info->flags & SHMEM_TRUNCATE)); 1060 goto unlock; 1061 } 1062 entry = shmem_swp_entry(info, index, NULL); 1063 if (entry->val) { 1064 /* 1065 * The more uptodate page coming down from a stacked 1066 * writepage should replace our old swappage. 1067 */ 1068 free_swap_and_cache(*entry); 1069 shmem_swp_set(info, entry, 0); 1070 } 1071 shmem_recalc_inode(inode); 1072 1073 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { 1074 remove_from_page_cache(page); 1075 shmem_swp_set(info, entry, swap.val); 1076 shmem_swp_unmap(entry); 1077 if (list_empty(&info->swaplist)) 1078 inode = igrab(inode); 1079 else 1080 inode = NULL; 1081 spin_unlock(&info->lock); 1082 swap_duplicate(swap); 1083 BUG_ON(page_mapped(page)); 1084 page_cache_release(page); /* pagecache ref */ 1085 swap_writepage(page, wbc); 1086 if (inode) { 1087 mutex_lock(&shmem_swaplist_mutex); 1088 /* move instead of add in case we're racing */ 1089 list_move_tail(&info->swaplist, &shmem_swaplist); 1090 mutex_unlock(&shmem_swaplist_mutex); 1091 iput(inode); 1092 } 1093 return 0; 1094 } 1095 1096 shmem_swp_unmap(entry); 1097unlock: 1098 spin_unlock(&info->lock); 1099 /* 1100 * add_to_swap_cache() doesn't return -EEXIST, so we can safely 1101 * clear SWAP_HAS_CACHE flag. 1102 */ 1103 swapcache_free(swap, NULL); 1104redirty: 1105 set_page_dirty(page); 1106 if (wbc->for_reclaim) 1107 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ 1108 unlock_page(page); 1109 return 0; 1110} 1111 1112#ifdef CONFIG_NUMA 1113#ifdef CONFIG_TMPFS 1114static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) 1115{ 1116 char buffer[64]; 1117 1118 if (!mpol || mpol->mode == MPOL_DEFAULT) 1119 return; /* show nothing */ 1120 1121 mpol_to_str(buffer, sizeof(buffer), mpol, 1); 1122 1123 seq_printf(seq, ",mpol=%s", buffer); 1124} 1125 1126static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) 1127{ 1128 struct mempolicy *mpol = NULL; 1129 if (sbinfo->mpol) { 1130 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ 1131 mpol = sbinfo->mpol; 1132 mpol_get(mpol); 1133 spin_unlock(&sbinfo->stat_lock); 1134 } 1135 return mpol; 1136} 1137#endif /* CONFIG_TMPFS */ 1138 1139static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, 1140 struct shmem_inode_info *info, unsigned long idx) 1141{ 1142 struct mempolicy mpol, *spol; 1143 struct vm_area_struct pvma; 1144 struct page *page; 1145 1146 spol = mpol_cond_copy(&mpol, 1147 mpol_shared_policy_lookup(&info->policy, idx)); 1148 1149 /* Create a pseudo vma that just contains the policy */ 1150 pvma.vm_start = 0; 1151 pvma.vm_pgoff = idx; 1152 pvma.vm_ops = NULL; 1153 pvma.vm_policy = spol; 1154 page = swapin_readahead(entry, gfp, &pvma, 0); 1155 return page; 1156} 1157 1158static struct page *shmem_alloc_page(gfp_t gfp, 1159 struct shmem_inode_info *info, unsigned long idx) 1160{ 1161 struct vm_area_struct pvma; 1162 1163 /* Create a pseudo vma that just contains the policy */ 1164 pvma.vm_start = 0; 1165 pvma.vm_pgoff = idx; 1166 pvma.vm_ops = NULL; 1167 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx); 1168 1169 /* 1170 * alloc_page_vma() will drop the shared policy reference 1171 */ 1172 return alloc_page_vma(gfp, &pvma, 0); 1173} 1174#else /* !CONFIG_NUMA */ 1175#ifdef CONFIG_TMPFS 1176static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p) 1177{ 1178} 1179#endif /* CONFIG_TMPFS */ 1180 1181static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, 1182 struct shmem_inode_info *info, unsigned long idx) 1183{ 1184 return swapin_readahead(entry, gfp, NULL, 0); 1185} 1186 1187static inline struct page *shmem_alloc_page(gfp_t gfp, 1188 struct shmem_inode_info *info, unsigned long idx) 1189{ 1190 return alloc_page(gfp); 1191} 1192#endif /* CONFIG_NUMA */ 1193 1194#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS) 1195static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) 1196{ 1197 return NULL; 1198} 1199#endif 1200 1201/* 1202 * shmem_getpage - either get the page from swap or allocate a new one 1203 * 1204 * If we allocate a new one we do not mark it dirty. That's up to the 1205 * vm. If we swap it in we mark it dirty since we also free the swap 1206 * entry since a page cannot live in both the swap and page cache 1207 */ 1208static int shmem_getpage(struct inode *inode, unsigned long idx, 1209 struct page **pagep, enum sgp_type sgp, int *type) 1210{ 1211 struct address_space *mapping = inode->i_mapping; 1212 struct shmem_inode_info *info = SHMEM_I(inode); 1213 struct shmem_sb_info *sbinfo; 1214 struct page *filepage = *pagep; 1215 struct page *swappage; 1216 swp_entry_t *entry; 1217 swp_entry_t swap; 1218 gfp_t gfp; 1219 int error; 1220 1221 if (idx >= SHMEM_MAX_INDEX) 1222 return -EFBIG; 1223 1224 if (type) 1225 *type = 0; 1226 1227 /* 1228 * Normally, filepage is NULL on entry, and either found 1229 * uptodate immediately, or allocated and zeroed, or read 1230 * in under swappage, which is then assigned to filepage. 1231 * But shmem_readpage (required for splice) passes in a locked 1232 * filepage, which may be found not uptodate by other callers 1233 * too, and may need to be copied from the swappage read in. 1234 */ 1235repeat: 1236 if (!filepage) 1237 filepage = find_lock_page(mapping, idx); 1238 if (filepage && PageUptodate(filepage)) 1239 goto done; 1240 error = 0; 1241 gfp = mapping_gfp_mask(mapping); 1242 if (!filepage) { 1243 /* 1244 * Try to preload while we can wait, to not make a habit of 1245 * draining atomic reserves; but don't latch on to this cpu. 1246 */ 1247 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM); 1248 if (error) 1249 goto failed; 1250 radix_tree_preload_end(); 1251 } 1252 1253 spin_lock(&info->lock); 1254 shmem_recalc_inode(inode); 1255 entry = shmem_swp_alloc(info, idx, sgp); 1256 if (IS_ERR(entry)) { 1257 spin_unlock(&info->lock); 1258 error = PTR_ERR(entry); 1259 goto failed; 1260 } 1261 swap = *entry; 1262 1263 if (swap.val) { 1264 /* Look it up and read it in.. */ 1265 swappage = lookup_swap_cache(swap); 1266 if (!swappage) { 1267 shmem_swp_unmap(entry); 1268 /* here we actually do the io */ 1269 if (type && !(*type & VM_FAULT_MAJOR)) { 1270 __count_vm_event(PGMAJFAULT); 1271 *type |= VM_FAULT_MAJOR; 1272 } 1273 spin_unlock(&info->lock); 1274 swappage = shmem_swapin(swap, gfp, info, idx); 1275 if (!swappage) { 1276 spin_lock(&info->lock); 1277 entry = shmem_swp_alloc(info, idx, sgp); 1278 if (IS_ERR(entry)) 1279 error = PTR_ERR(entry); 1280 else { 1281 if (entry->val == swap.val) 1282 error = -ENOMEM; 1283 shmem_swp_unmap(entry); 1284 } 1285 spin_unlock(&info->lock); 1286 if (error) 1287 goto failed; 1288 goto repeat; 1289 } 1290 wait_on_page_locked(swappage); 1291 page_cache_release(swappage); 1292 goto repeat; 1293 } 1294 1295 /* We have to do this with page locked to prevent races */ 1296 if (!trylock_page(swappage)) { 1297 shmem_swp_unmap(entry); 1298 spin_unlock(&info->lock); 1299 wait_on_page_locked(swappage); 1300 page_cache_release(swappage); 1301 goto repeat; 1302 } 1303 if (PageWriteback(swappage)) { 1304 shmem_swp_unmap(entry); 1305 spin_unlock(&info->lock); 1306 wait_on_page_writeback(swappage); 1307 unlock_page(swappage); 1308 page_cache_release(swappage); 1309 goto repeat; 1310 } 1311 if (!PageUptodate(swappage)) { 1312 shmem_swp_unmap(entry); 1313 spin_unlock(&info->lock); 1314 unlock_page(swappage); 1315 page_cache_release(swappage); 1316 error = -EIO; 1317 goto failed; 1318 } 1319 1320 if (filepage) { 1321 shmem_swp_set(info, entry, 0); 1322 shmem_swp_unmap(entry); 1323 delete_from_swap_cache(swappage); 1324 spin_unlock(&info->lock); 1325 copy_highpage(filepage, swappage); 1326 unlock_page(swappage); 1327 page_cache_release(swappage); 1328 flush_dcache_page(filepage); 1329 SetPageUptodate(filepage); 1330 set_page_dirty(filepage); 1331 swap_free(swap); 1332 } else if (!(error = add_to_page_cache_locked(swappage, mapping, 1333 idx, GFP_NOWAIT))) { 1334 info->flags |= SHMEM_PAGEIN; 1335 shmem_swp_set(info, entry, 0); 1336 shmem_swp_unmap(entry); 1337 delete_from_swap_cache(swappage); 1338 spin_unlock(&info->lock); 1339 filepage = swappage; 1340 set_page_dirty(filepage); 1341 swap_free(swap); 1342 } else { 1343 shmem_swp_unmap(entry); 1344 spin_unlock(&info->lock); 1345 if (error == -ENOMEM) { 1346 /* 1347 * reclaim from proper memory cgroup and 1348 * call memcg's OOM if needed. 1349 */ 1350 error = mem_cgroup_shmem_charge_fallback( 1351 swappage, 1352 current->mm, 1353 gfp); 1354 if (error) { 1355 unlock_page(swappage); 1356 page_cache_release(swappage); 1357 goto failed; 1358 } 1359 } 1360 unlock_page(swappage); 1361 page_cache_release(swappage); 1362 goto repeat; 1363 } 1364 } else if (sgp == SGP_READ && !filepage) { 1365 shmem_swp_unmap(entry); 1366 filepage = find_get_page(mapping, idx); 1367 if (filepage && 1368 (!PageUptodate(filepage) || !trylock_page(filepage))) { 1369 spin_unlock(&info->lock); 1370 wait_on_page_locked(filepage); 1371 page_cache_release(filepage); 1372 filepage = NULL; 1373 goto repeat; 1374 } 1375 spin_unlock(&info->lock); 1376 } else { 1377 shmem_swp_unmap(entry); 1378 sbinfo = SHMEM_SB(inode->i_sb); 1379 if (sbinfo->max_blocks) { 1380 spin_lock(&sbinfo->stat_lock); 1381 if (sbinfo->free_blocks == 0 || 1382 shmem_acct_block(info->flags)) { 1383 spin_unlock(&sbinfo->stat_lock); 1384 spin_unlock(&info->lock); 1385 error = -ENOSPC; 1386 goto failed; 1387 } 1388 sbinfo->free_blocks--; 1389 inode->i_blocks += BLOCKS_PER_PAGE; 1390 spin_unlock(&sbinfo->stat_lock); 1391 } else if (shmem_acct_block(info->flags)) { 1392 spin_unlock(&info->lock); 1393 error = -ENOSPC; 1394 goto failed; 1395 } 1396 1397 if (!filepage) { 1398 int ret; 1399 1400 spin_unlock(&info->lock); 1401 filepage = shmem_alloc_page(gfp, info, idx); 1402 if (!filepage) { 1403 shmem_unacct_blocks(info->flags, 1); 1404 shmem_free_blocks(inode, 1); 1405 error = -ENOMEM; 1406 goto failed; 1407 } 1408 SetPageSwapBacked(filepage); 1409 1410 /* Precharge page while we can wait, compensate after */ 1411 error = mem_cgroup_cache_charge(filepage, current->mm, 1412 GFP_KERNEL); 1413 if (error) { 1414 page_cache_release(filepage); 1415 shmem_unacct_blocks(info->flags, 1); 1416 shmem_free_blocks(inode, 1); 1417 filepage = NULL; 1418 goto failed; 1419 } 1420 1421 spin_lock(&info->lock); 1422 entry = shmem_swp_alloc(info, idx, sgp); 1423 if (IS_ERR(entry)) 1424 error = PTR_ERR(entry); 1425 else { 1426 swap = *entry; 1427 shmem_swp_unmap(entry); 1428 } 1429 ret = error || swap.val; 1430 if (ret) 1431 mem_cgroup_uncharge_cache_page(filepage); 1432 else 1433 ret = add_to_page_cache_lru(filepage, mapping, 1434 idx, GFP_NOWAIT); 1435 /* 1436 * At add_to_page_cache_lru() failure, uncharge will 1437 * be done automatically. 1438 */ 1439 if (ret) { 1440 spin_unlock(&info->lock); 1441 page_cache_release(filepage); 1442 shmem_unacct_blocks(info->flags, 1); 1443 shmem_free_blocks(inode, 1); 1444 filepage = NULL; 1445 if (error) 1446 goto failed; 1447 goto repeat; 1448 } 1449 info->flags |= SHMEM_PAGEIN; 1450 } 1451 1452 info->alloced++; 1453 spin_unlock(&info->lock); 1454 clear_highpage(filepage); 1455 flush_dcache_page(filepage); 1456 SetPageUptodate(filepage); 1457 if (sgp == SGP_DIRTY) 1458 set_page_dirty(filepage); 1459 } 1460done: 1461 *pagep = filepage; 1462 return 0; 1463 1464failed: 1465 if (*pagep != filepage) { 1466 unlock_page(filepage); 1467 page_cache_release(filepage); 1468 } 1469 return error; 1470} 1471 1472static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1473{ 1474 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 1475 int error; 1476 int ret; 1477 1478 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 1479 return VM_FAULT_SIGBUS; 1480 1481 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret); 1482 if (error) 1483 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); 1484 1485 return ret | VM_FAULT_LOCKED; 1486} 1487 1488#ifdef CONFIG_NUMA 1489static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new) 1490{ 1491 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1492 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new); 1493} 1494 1495static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, 1496 unsigned long addr) 1497{ 1498 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1499 unsigned long idx; 1500 1501 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; 1502 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx); 1503} 1504#endif 1505 1506int shmem_lock(struct file *file, int lock, struct user_struct *user) 1507{ 1508 struct inode *inode = file->f_path.dentry->d_inode; 1509 struct shmem_inode_info *info = SHMEM_I(inode); 1510 int retval = -ENOMEM; 1511 1512 spin_lock(&info->lock); 1513 if (lock && !(info->flags & VM_LOCKED)) { 1514 if (!user_shm_lock(inode->i_size, user)) 1515 goto out_nomem; 1516 info->flags |= VM_LOCKED; 1517 mapping_set_unevictable(file->f_mapping); 1518 } 1519 if (!lock && (info->flags & VM_LOCKED) && user) { 1520 user_shm_unlock(inode->i_size, user); 1521 info->flags &= ~VM_LOCKED; 1522 mapping_clear_unevictable(file->f_mapping); 1523 scan_mapping_unevictable_pages(file->f_mapping); 1524 } 1525 retval = 0; 1526 1527out_nomem: 1528 spin_unlock(&info->lock); 1529 return retval; 1530} 1531 1532static int shmem_mmap(struct file *file, struct vm_area_struct *vma) 1533{ 1534 file_accessed(file); 1535 vma->vm_ops = &shmem_vm_ops; 1536 vma->vm_flags |= VM_CAN_NONLINEAR; 1537 return 0; 1538} 1539 1540static struct inode *shmem_get_inode(struct super_block *sb, int mode, 1541 dev_t dev, unsigned long flags) 1542{ 1543 struct inode *inode; 1544 struct shmem_inode_info *info; 1545 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 1546 1547 if (shmem_reserve_inode(sb)) 1548 return NULL; 1549 1550 inode = new_inode(sb); 1551 if (inode) { 1552 inode->i_mode = mode; 1553 inode->i_uid = current_fsuid(); 1554 inode->i_gid = current_fsgid(); 1555 inode->i_blocks = 0; 1556 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; 1557 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 1558 inode->i_generation = get_seconds(); 1559 info = SHMEM_I(inode); 1560 memset(info, 0, (char *)inode - (char *)info); 1561 spin_lock_init(&info->lock); 1562 info->flags = flags & VM_NORESERVE; 1563 INIT_LIST_HEAD(&info->swaplist); 1564 cache_no_acl(inode); 1565 1566 switch (mode & S_IFMT) { 1567 default: 1568 inode->i_op = &shmem_special_inode_operations; 1569 init_special_inode(inode, mode, dev); 1570 break; 1571 case S_IFREG: 1572 inode->i_mapping->a_ops = &shmem_aops; 1573 inode->i_op = &shmem_inode_operations; 1574 inode->i_fop = &shmem_file_operations; 1575 mpol_shared_policy_init(&info->policy, 1576 shmem_get_sbmpol(sbinfo)); 1577 break; 1578 case S_IFDIR: 1579 inc_nlink(inode); 1580 /* Some things misbehave if size == 0 on a directory */ 1581 inode->i_size = 2 * BOGO_DIRENT_SIZE; 1582 inode->i_op = &shmem_dir_inode_operations; 1583 inode->i_fop = &simple_dir_operations; 1584 break; 1585 case S_IFLNK: 1586 /* 1587 * Must not load anything in the rbtree, 1588 * mpol_free_shared_policy will not be called. 1589 */ 1590 mpol_shared_policy_init(&info->policy, NULL); 1591 break; 1592 } 1593 } else 1594 shmem_free_inode(sb); 1595 return inode; 1596} 1597 1598#ifdef CONFIG_TMPFS 1599static const struct inode_operations shmem_symlink_inode_operations; 1600static const struct inode_operations shmem_symlink_inline_operations; 1601 1602/* 1603 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin; 1604 * but providing them allows a tmpfs file to be used for splice, sendfile, and 1605 * below the loop driver, in the generic fashion that many filesystems support. 1606 */ 1607static int shmem_readpage(struct file *file, struct page *page) 1608{ 1609 struct inode *inode = page->mapping->host; 1610 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL); 1611 unlock_page(page); 1612 return error; 1613} 1614 1615static int 1616shmem_write_begin(struct file *file, struct address_space *mapping, 1617 loff_t pos, unsigned len, unsigned flags, 1618 struct page **pagep, void **fsdata) 1619{ 1620 struct inode *inode = mapping->host; 1621 pgoff_t index = pos >> PAGE_CACHE_SHIFT; 1622 *pagep = NULL; 1623 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL); 1624} 1625 1626static int 1627shmem_write_end(struct file *file, struct address_space *mapping, 1628 loff_t pos, unsigned len, unsigned copied, 1629 struct page *page, void *fsdata) 1630{ 1631 struct inode *inode = mapping->host; 1632 1633 if (pos + copied > inode->i_size) 1634 i_size_write(inode, pos + copied); 1635 1636 unlock_page(page); 1637 set_page_dirty(page); 1638 page_cache_release(page); 1639 1640 return copied; 1641} 1642 1643static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor) 1644{ 1645 struct inode *inode = filp->f_path.dentry->d_inode; 1646 struct address_space *mapping = inode->i_mapping; 1647 unsigned long index, offset; 1648 enum sgp_type sgp = SGP_READ; 1649 1650 /* 1651 * Might this read be for a stacking filesystem? Then when reading 1652 * holes of a sparse file, we actually need to allocate those pages, 1653 * and even mark them dirty, so it cannot exceed the max_blocks limit. 1654 */ 1655 if (segment_eq(get_fs(), KERNEL_DS)) 1656 sgp = SGP_DIRTY; 1657 1658 index = *ppos >> PAGE_CACHE_SHIFT; 1659 offset = *ppos & ~PAGE_CACHE_MASK; 1660 1661 for (;;) { 1662 struct page *page = NULL; 1663 unsigned long end_index, nr, ret; 1664 loff_t i_size = i_size_read(inode); 1665 1666 end_index = i_size >> PAGE_CACHE_SHIFT; 1667 if (index > end_index) 1668 break; 1669 if (index == end_index) { 1670 nr = i_size & ~PAGE_CACHE_MASK; 1671 if (nr <= offset) 1672 break; 1673 } 1674 1675 desc->error = shmem_getpage(inode, index, &page, sgp, NULL); 1676 if (desc->error) { 1677 if (desc->error == -EINVAL) 1678 desc->error = 0; 1679 break; 1680 } 1681 if (page) 1682 unlock_page(page); 1683 1684 /* 1685 * We must evaluate after, since reads (unlike writes) 1686 * are called without i_mutex protection against truncate 1687 */ 1688 nr = PAGE_CACHE_SIZE; 1689 i_size = i_size_read(inode); 1690 end_index = i_size >> PAGE_CACHE_SHIFT; 1691 if (index == end_index) { 1692 nr = i_size & ~PAGE_CACHE_MASK; 1693 if (nr <= offset) { 1694 if (page) 1695 page_cache_release(page); 1696 break; 1697 } 1698 } 1699 nr -= offset; 1700 1701 if (page) { 1702 /* 1703 * If users can be writing to this page using arbitrary 1704 * virtual addresses, take care about potential aliasing 1705 * before reading the page on the kernel side. 1706 */ 1707 if (mapping_writably_mapped(mapping)) 1708 flush_dcache_page(page); 1709 /* 1710 * Mark the page accessed if we read the beginning. 1711 */ 1712 if (!offset) 1713 mark_page_accessed(page); 1714 } else { 1715 page = ZERO_PAGE(0); 1716 page_cache_get(page); 1717 } 1718 1719 /* 1720 * Ok, we have the page, and it's up-to-date, so 1721 * now we can copy it to user space... 1722 * 1723 * The actor routine returns how many bytes were actually used.. 1724 * NOTE! This may not be the same as how much of a user buffer 1725 * we filled up (we may be padding etc), so we can only update 1726 * "pos" here (the actor routine has to update the user buffer 1727 * pointers and the remaining count). 1728 */ 1729 ret = actor(desc, page, offset, nr); 1730 offset += ret; 1731 index += offset >> PAGE_CACHE_SHIFT; 1732 offset &= ~PAGE_CACHE_MASK; 1733 1734 page_cache_release(page); 1735 if (ret != nr || !desc->count) 1736 break; 1737 1738 cond_resched(); 1739 } 1740 1741 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; 1742 file_accessed(filp); 1743} 1744 1745static ssize_t shmem_file_aio_read(struct kiocb *iocb, 1746 const struct iovec *iov, unsigned long nr_segs, loff_t pos) 1747{ 1748 struct file *filp = iocb->ki_filp; 1749 ssize_t retval; 1750 unsigned long seg; 1751 size_t count; 1752 loff_t *ppos = &iocb->ki_pos; 1753 1754 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); 1755 if (retval) 1756 return retval; 1757 1758 for (seg = 0; seg < nr_segs; seg++) { 1759 read_descriptor_t desc; 1760 1761 desc.written = 0; 1762 desc.arg.buf = iov[seg].iov_base; 1763 desc.count = iov[seg].iov_len; 1764 if (desc.count == 0) 1765 continue; 1766 desc.error = 0; 1767 do_shmem_file_read(filp, ppos, &desc, file_read_actor); 1768 retval += desc.written; 1769 if (desc.error) { 1770 retval = retval ?: desc.error; 1771 break; 1772 } 1773 if (desc.count > 0) 1774 break; 1775 } 1776 return retval; 1777} 1778 1779static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) 1780{ 1781 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); 1782 1783 buf->f_type = TMPFS_MAGIC; 1784 buf->f_bsize = PAGE_CACHE_SIZE; 1785 buf->f_namelen = NAME_MAX; 1786 spin_lock(&sbinfo->stat_lock); 1787 if (sbinfo->max_blocks) { 1788 buf->f_blocks = sbinfo->max_blocks; 1789 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks; 1790 } 1791 if (sbinfo->max_inodes) { 1792 buf->f_files = sbinfo->max_inodes; 1793 buf->f_ffree = sbinfo->free_inodes; 1794 } 1795 /* else leave those fields 0 like simple_statfs */ 1796 spin_unlock(&sbinfo->stat_lock); 1797 return 0; 1798} 1799 1800/* 1801 * File creation. Allocate an inode, and we're done.. 1802 */ 1803static int 1804shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 1805{ 1806 struct inode *inode; 1807 int error = -ENOSPC; 1808 1809 inode = shmem_get_inode(dir->i_sb, mode, dev, VM_NORESERVE); 1810 if (inode) { 1811 error = security_inode_init_security(inode, dir, NULL, NULL, 1812 NULL); 1813 if (error) { 1814 if (error != -EOPNOTSUPP) { 1815 iput(inode); 1816 return error; 1817 } 1818 } 1819 error = shmem_acl_init(inode, dir); 1820 if (error) { 1821 iput(inode); 1822 return error; 1823 } 1824 if (dir->i_mode & S_ISGID) { 1825 inode->i_gid = dir->i_gid; 1826 if (S_ISDIR(mode)) 1827 inode->i_mode |= S_ISGID; 1828 } 1829 dir->i_size += BOGO_DIRENT_SIZE; 1830 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1831 d_instantiate(dentry, inode); 1832 dget(dentry); /* Extra count - pin the dentry in core */ 1833 } 1834 return error; 1835} 1836 1837static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1838{ 1839 int error; 1840 1841 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) 1842 return error; 1843 inc_nlink(dir); 1844 return 0; 1845} 1846 1847static int shmem_create(struct inode *dir, struct dentry *dentry, int mode, 1848 struct nameidata *nd) 1849{ 1850 return shmem_mknod(dir, dentry, mode | S_IFREG, 0); 1851} 1852 1853/* 1854 * Link a file.. 1855 */ 1856static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1857{ 1858 struct inode *inode = old_dentry->d_inode; 1859 int ret; 1860 1861 /* 1862 * No ordinary (disk based) filesystem counts links as inodes; 1863 * but each new link needs a new dentry, pinning lowmem, and 1864 * tmpfs dentries cannot be pruned until they are unlinked. 1865 */ 1866 ret = shmem_reserve_inode(inode->i_sb); 1867 if (ret) 1868 goto out; 1869 1870 dir->i_size += BOGO_DIRENT_SIZE; 1871 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1872 inc_nlink(inode); 1873 atomic_inc(&inode->i_count); /* New dentry reference */ 1874 dget(dentry); /* Extra pinning count for the created dentry */ 1875 d_instantiate(dentry, inode); 1876out: 1877 return ret; 1878} 1879 1880static int shmem_unlink(struct inode *dir, struct dentry *dentry) 1881{ 1882 struct inode *inode = dentry->d_inode; 1883 1884 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) 1885 shmem_free_inode(inode->i_sb); 1886 1887 dir->i_size -= BOGO_DIRENT_SIZE; 1888 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1889 drop_nlink(inode); 1890 dput(dentry); /* Undo the count from "create" - this does all the work */ 1891 return 0; 1892} 1893 1894static int shmem_rmdir(struct inode *dir, struct dentry *dentry) 1895{ 1896 if (!simple_empty(dentry)) 1897 return -ENOTEMPTY; 1898 1899 drop_nlink(dentry->d_inode); 1900 drop_nlink(dir); 1901 return shmem_unlink(dir, dentry); 1902} 1903 1904/* 1905 * The VFS layer already does all the dentry stuff for rename, 1906 * we just have to decrement the usage count for the target if 1907 * it exists so that the VFS layer correctly free's it when it 1908 * gets overwritten. 1909 */ 1910static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) 1911{ 1912 struct inode *inode = old_dentry->d_inode; 1913 int they_are_dirs = S_ISDIR(inode->i_mode); 1914 1915 if (!simple_empty(new_dentry)) 1916 return -ENOTEMPTY; 1917 1918 if (new_dentry->d_inode) { 1919 (void) shmem_unlink(new_dir, new_dentry); 1920 if (they_are_dirs) 1921 drop_nlink(old_dir); 1922 } else if (they_are_dirs) { 1923 drop_nlink(old_dir); 1924 inc_nlink(new_dir); 1925 } 1926 1927 old_dir->i_size -= BOGO_DIRENT_SIZE; 1928 new_dir->i_size += BOGO_DIRENT_SIZE; 1929 old_dir->i_ctime = old_dir->i_mtime = 1930 new_dir->i_ctime = new_dir->i_mtime = 1931 inode->i_ctime = CURRENT_TIME; 1932 return 0; 1933} 1934 1935static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1936{ 1937 int error; 1938 int len; 1939 struct inode *inode; 1940 struct page *page = NULL; 1941 char *kaddr; 1942 struct shmem_inode_info *info; 1943 1944 len = strlen(symname) + 1; 1945 if (len > PAGE_CACHE_SIZE) 1946 return -ENAMETOOLONG; 1947 1948 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE); 1949 if (!inode) 1950 return -ENOSPC; 1951 1952 error = security_inode_init_security(inode, dir, NULL, NULL, 1953 NULL); 1954 if (error) { 1955 if (error != -EOPNOTSUPP) { 1956 iput(inode); 1957 return error; 1958 } 1959 error = 0; 1960 } 1961 1962 info = SHMEM_I(inode); 1963 inode->i_size = len-1; 1964 if (len <= (char *)inode - (char *)info) { 1965 /* do it inline */ 1966 memcpy(info, symname, len); 1967 inode->i_op = &shmem_symlink_inline_operations; 1968 } else { 1969 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); 1970 if (error) { 1971 iput(inode); 1972 return error; 1973 } 1974 unlock_page(page); 1975 inode->i_mapping->a_ops = &shmem_aops; 1976 inode->i_op = &shmem_symlink_inode_operations; 1977 kaddr = kmap_atomic(page, KM_USER0); 1978 memcpy(kaddr, symname, len); 1979 kunmap_atomic(kaddr, KM_USER0); 1980 set_page_dirty(page); 1981 page_cache_release(page); 1982 } 1983 if (dir->i_mode & S_ISGID) 1984 inode->i_gid = dir->i_gid; 1985 dir->i_size += BOGO_DIRENT_SIZE; 1986 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1987 d_instantiate(dentry, inode); 1988 dget(dentry); 1989 return 0; 1990} 1991 1992static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd) 1993{ 1994 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode)); 1995 return NULL; 1996} 1997 1998static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) 1999{ 2000 struct page *page = NULL; 2001 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); 2002 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page)); 2003 if (page) 2004 unlock_page(page); 2005 return page; 2006} 2007 2008static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 2009{ 2010 if (!IS_ERR(nd_get_link(nd))) { 2011 struct page *page = cookie; 2012 kunmap(page); 2013 mark_page_accessed(page); 2014 page_cache_release(page); 2015 } 2016} 2017 2018static const struct inode_operations shmem_symlink_inline_operations = { 2019 .readlink = generic_readlink, 2020 .follow_link = shmem_follow_link_inline, 2021}; 2022 2023static const struct inode_operations shmem_symlink_inode_operations = { 2024 .truncate = shmem_truncate, 2025 .readlink = generic_readlink, 2026 .follow_link = shmem_follow_link, 2027 .put_link = shmem_put_link, 2028}; 2029 2030#ifdef CONFIG_TMPFS_POSIX_ACL 2031/* 2032 * Superblocks without xattr inode operations will get security.* xattr 2033 * support from the VFS "for free". As soon as we have any other xattrs 2034 * like ACLs, we also need to implement the security.* handlers at 2035 * filesystem level, though. 2036 */ 2037 2038static size_t shmem_xattr_security_list(struct inode *inode, char *list, 2039 size_t list_len, const char *name, 2040 size_t name_len) 2041{ 2042 return security_inode_listsecurity(inode, list, list_len); 2043} 2044 2045static int shmem_xattr_security_get(struct inode *inode, const char *name, 2046 void *buffer, size_t size) 2047{ 2048 if (strcmp(name, "") == 0) 2049 return -EINVAL; 2050 return xattr_getsecurity(inode, name, buffer, size); 2051} 2052 2053static int shmem_xattr_security_set(struct inode *inode, const char *name, 2054 const void *value, size_t size, int flags) 2055{ 2056 if (strcmp(name, "") == 0) 2057 return -EINVAL; 2058 return security_inode_setsecurity(inode, name, value, size, flags); 2059} 2060 2061static struct xattr_handler shmem_xattr_security_handler = { 2062 .prefix = XATTR_SECURITY_PREFIX, 2063 .list = shmem_xattr_security_list, 2064 .get = shmem_xattr_security_get, 2065 .set = shmem_xattr_security_set, 2066}; 2067 2068static struct xattr_handler *shmem_xattr_handlers[] = { 2069 &shmem_xattr_acl_access_handler, 2070 &shmem_xattr_acl_default_handler, 2071 &shmem_xattr_security_handler, 2072 NULL 2073}; 2074#endif 2075 2076static struct dentry *shmem_get_parent(struct dentry *child) 2077{ 2078 return ERR_PTR(-ESTALE); 2079} 2080 2081static int shmem_match(struct inode *ino, void *vfh) 2082{ 2083 __u32 *fh = vfh; 2084 __u64 inum = fh[2]; 2085 inum = (inum << 32) | fh[1]; 2086 return ino->i_ino == inum && fh[0] == ino->i_generation; 2087} 2088 2089static struct dentry *shmem_fh_to_dentry(struct super_block *sb, 2090 struct fid *fid, int fh_len, int fh_type) 2091{ 2092 struct inode *inode; 2093 struct dentry *dentry = NULL; 2094 u64 inum = fid->raw[2]; 2095 inum = (inum << 32) | fid->raw[1]; 2096 2097 if (fh_len < 3) 2098 return NULL; 2099 2100 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), 2101 shmem_match, fid->raw); 2102 if (inode) { 2103 dentry = d_find_alias(inode); 2104 iput(inode); 2105 } 2106 2107 return dentry; 2108} 2109 2110static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len, 2111 int connectable) 2112{ 2113 struct inode *inode = dentry->d_inode; 2114 2115 if (*len < 3) 2116 return 255; 2117 2118 if (hlist_unhashed(&inode->i_hash)) { 2119 /* Unfortunately insert_inode_hash is not idempotent, 2120 * so as we hash inodes here rather than at creation 2121 * time, we need a lock to ensure we only try 2122 * to do it once 2123 */ 2124 static DEFINE_SPINLOCK(lock); 2125 spin_lock(&lock); 2126 if (hlist_unhashed(&inode->i_hash)) 2127 __insert_inode_hash(inode, 2128 inode->i_ino + inode->i_generation); 2129 spin_unlock(&lock); 2130 } 2131 2132 fh[0] = inode->i_generation; 2133 fh[1] = inode->i_ino; 2134 fh[2] = ((__u64)inode->i_ino) >> 32; 2135 2136 *len = 3; 2137 return 1; 2138} 2139 2140static const struct export_operations shmem_export_ops = { 2141 .get_parent = shmem_get_parent, 2142 .encode_fh = shmem_encode_fh, 2143 .fh_to_dentry = shmem_fh_to_dentry, 2144}; 2145 2146static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, 2147 bool remount) 2148{ 2149 char *this_char, *value, *rest; 2150 2151 while (options != NULL) { 2152 this_char = options; 2153 for (;;) { 2154 /* 2155 * NUL-terminate this option: unfortunately, 2156 * mount options form a comma-separated list, 2157 * but mpol's nodelist may also contain commas. 2158 */ 2159 options = strchr(options, ','); 2160 if (options == NULL) 2161 break; 2162 options++; 2163 if (!isdigit(*options)) { 2164 options[-1] = '\0'; 2165 break; 2166 } 2167 } 2168 if (!*this_char) 2169 continue; 2170 if ((value = strchr(this_char,'=')) != NULL) { 2171 *value++ = 0; 2172 } else { 2173 printk(KERN_ERR 2174 "tmpfs: No value for mount option '%s'\n", 2175 this_char); 2176 return 1; 2177 } 2178 2179 if (!strcmp(this_char,"size")) { 2180 unsigned long long size; 2181 size = memparse(value,&rest); 2182 if (*rest == '%') { 2183 size <<= PAGE_SHIFT; 2184 size *= totalram_pages; 2185 do_div(size, 100); 2186 rest++; 2187 } 2188 if (*rest) 2189 goto bad_val; 2190 sbinfo->max_blocks = 2191 DIV_ROUND_UP(size, PAGE_CACHE_SIZE); 2192 } else if (!strcmp(this_char,"nr_blocks")) { 2193 sbinfo->max_blocks = memparse(value, &rest); 2194 if (*rest) 2195 goto bad_val; 2196 } else if (!strcmp(this_char,"nr_inodes")) { 2197 sbinfo->max_inodes = memparse(value, &rest); 2198 if (*rest) 2199 goto bad_val; 2200 } else if (!strcmp(this_char,"mode")) { 2201 if (remount) 2202 continue; 2203 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; 2204 if (*rest) 2205 goto bad_val; 2206 } else if (!strcmp(this_char,"uid")) { 2207 if (remount) 2208 continue; 2209 sbinfo->uid = simple_strtoul(value, &rest, 0); 2210 if (*rest) 2211 goto bad_val; 2212 } else if (!strcmp(this_char,"gid")) { 2213 if (remount) 2214 continue; 2215 sbinfo->gid = simple_strtoul(value, &rest, 0); 2216 if (*rest) 2217 goto bad_val; 2218 } else if (!strcmp(this_char,"mpol")) { 2219 if (mpol_parse_str(value, &sbinfo->mpol, 1)) 2220 goto bad_val; 2221 } else { 2222 printk(KERN_ERR "tmpfs: Bad mount option %s\n", 2223 this_char); 2224 return 1; 2225 } 2226 } 2227 return 0; 2228 2229bad_val: 2230 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", 2231 value, this_char); 2232 return 1; 2233 2234} 2235 2236static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) 2237{ 2238 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 2239 struct shmem_sb_info config = *sbinfo; 2240 unsigned long blocks; 2241 unsigned long inodes; 2242 int error = -EINVAL; 2243 2244 if (shmem_parse_options(data, &config, true)) 2245 return error; 2246 2247 spin_lock(&sbinfo->stat_lock); 2248 blocks = sbinfo->max_blocks - sbinfo->free_blocks; 2249 inodes = sbinfo->max_inodes - sbinfo->free_inodes; 2250 if (config.max_blocks < blocks) 2251 goto out; 2252 if (config.max_inodes < inodes) 2253 goto out; 2254 /* 2255 * Those tests also disallow limited->unlimited while any are in 2256 * use, so i_blocks will always be zero when max_blocks is zero; 2257 * but we must separately disallow unlimited->limited, because 2258 * in that case we have no record of how much is already in use. 2259 */ 2260 if (config.max_blocks && !sbinfo->max_blocks) 2261 goto out; 2262 if (config.max_inodes && !sbinfo->max_inodes) 2263 goto out; 2264 2265 error = 0; 2266 sbinfo->max_blocks = config.max_blocks; 2267 sbinfo->free_blocks = config.max_blocks - blocks; 2268 sbinfo->max_inodes = config.max_inodes; 2269 sbinfo->free_inodes = config.max_inodes - inodes; 2270 2271 mpol_put(sbinfo->mpol); 2272 sbinfo->mpol = config.mpol; /* transfers initial ref */ 2273out: 2274 spin_unlock(&sbinfo->stat_lock); 2275 return error; 2276} 2277 2278static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs) 2279{ 2280 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb); 2281 2282 if (sbinfo->max_blocks != shmem_default_max_blocks()) 2283 seq_printf(seq, ",size=%luk", 2284 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10)); 2285 if (sbinfo->max_inodes != shmem_default_max_inodes()) 2286 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); 2287 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) 2288 seq_printf(seq, ",mode=%03o", sbinfo->mode); 2289 if (sbinfo->uid != 0) 2290 seq_printf(seq, ",uid=%u", sbinfo->uid); 2291 if (sbinfo->gid != 0) 2292 seq_printf(seq, ",gid=%u", sbinfo->gid); 2293 shmem_show_mpol(seq, sbinfo->mpol); 2294 return 0; 2295} 2296#endif /* CONFIG_TMPFS */ 2297 2298static void shmem_put_super(struct super_block *sb) 2299{ 2300 kfree(sb->s_fs_info); 2301 sb->s_fs_info = NULL; 2302} 2303 2304int shmem_fill_super(struct super_block *sb, void *data, int silent) 2305{ 2306 struct inode *inode; 2307 struct dentry *root; 2308 struct shmem_sb_info *sbinfo; 2309 int err = -ENOMEM; 2310 2311 /* Round up to L1_CACHE_BYTES to resist false sharing */ 2312 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info), 2313 L1_CACHE_BYTES), GFP_KERNEL); 2314 if (!sbinfo) 2315 return -ENOMEM; 2316 2317 sbinfo->max_blocks = 0; 2318 sbinfo->max_inodes = 0; 2319 sbinfo->mode = S_IRWXUGO | S_ISVTX; 2320 sbinfo->uid = current_fsuid(); 2321 sbinfo->gid = current_fsgid(); 2322 sbinfo->mpol = NULL; 2323 sb->s_fs_info = sbinfo; 2324 2325#ifdef CONFIG_TMPFS 2326 /* 2327 * Per default we only allow half of the physical ram per 2328 * tmpfs instance, limiting inodes to one per page of lowmem; 2329 * but the internal instance is left unlimited. 2330 */ 2331 if (!(sb->s_flags & MS_NOUSER)) { 2332 sbinfo->max_blocks = shmem_default_max_blocks(); 2333 sbinfo->max_inodes = shmem_default_max_inodes(); 2334 if (shmem_parse_options(data, sbinfo, false)) { 2335 err = -EINVAL; 2336 goto failed; 2337 } 2338 } 2339 sb->s_export_op = &shmem_export_ops; 2340#else 2341 sb->s_flags |= MS_NOUSER; 2342#endif 2343 2344 spin_lock_init(&sbinfo->stat_lock); 2345 sbinfo->free_blocks = sbinfo->max_blocks; 2346 sbinfo->free_inodes = sbinfo->max_inodes; 2347 2348 sb->s_maxbytes = SHMEM_MAX_BYTES; 2349 sb->s_blocksize = PAGE_CACHE_SIZE; 2350 sb->s_blocksize_bits = PAGE_CACHE_SHIFT; 2351 sb->s_magic = TMPFS_MAGIC; 2352 sb->s_op = &shmem_ops; 2353 sb->s_time_gran = 1; 2354#ifdef CONFIG_TMPFS_POSIX_ACL 2355 sb->s_xattr = shmem_xattr_handlers; 2356 sb->s_flags |= MS_POSIXACL; 2357#endif 2358 2359 inode = shmem_get_inode(sb, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); 2360 if (!inode) 2361 goto failed; 2362 inode->i_uid = sbinfo->uid; 2363 inode->i_gid = sbinfo->gid; 2364 root = d_alloc_root(inode); 2365 if (!root) 2366 goto failed_iput; 2367 sb->s_root = root; 2368 return 0; 2369 2370failed_iput: 2371 iput(inode); 2372failed: 2373 shmem_put_super(sb); 2374 return err; 2375} 2376 2377static struct kmem_cache *shmem_inode_cachep; 2378 2379static struct inode *shmem_alloc_inode(struct super_block *sb) 2380{ 2381 struct shmem_inode_info *p; 2382 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); 2383 if (!p) 2384 return NULL; 2385 return &p->vfs_inode; 2386} 2387 2388static void shmem_destroy_inode(struct inode *inode) 2389{ 2390 if ((inode->i_mode & S_IFMT) == S_IFREG) { 2391 /* only struct inode is valid if it's an inline symlink */ 2392 mpol_free_shared_policy(&SHMEM_I(inode)->policy); 2393 } 2394 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); 2395} 2396 2397static void init_once(void *foo) 2398{ 2399 struct shmem_inode_info *p = (struct shmem_inode_info *) foo; 2400 2401 inode_init_once(&p->vfs_inode); 2402} 2403 2404static int init_inodecache(void) 2405{ 2406 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", 2407 sizeof(struct shmem_inode_info), 2408 0, SLAB_PANIC, init_once); 2409 return 0; 2410} 2411 2412static void destroy_inodecache(void) 2413{ 2414 kmem_cache_destroy(shmem_inode_cachep); 2415} 2416 2417static const struct address_space_operations shmem_aops = { 2418 .writepage = shmem_writepage, 2419 .set_page_dirty = __set_page_dirty_no_writeback, 2420#ifdef CONFIG_TMPFS 2421 .readpage = shmem_readpage, 2422 .write_begin = shmem_write_begin, 2423 .write_end = shmem_write_end, 2424#endif 2425 .migratepage = migrate_page, 2426}; 2427 2428static const struct file_operations shmem_file_operations = { 2429 .mmap = shmem_mmap, 2430#ifdef CONFIG_TMPFS 2431 .llseek = generic_file_llseek, 2432 .read = do_sync_read, 2433 .write = do_sync_write, 2434 .aio_read = shmem_file_aio_read, 2435 .aio_write = generic_file_aio_write, 2436 .fsync = simple_sync_file, 2437 .splice_read = generic_file_splice_read, 2438 .splice_write = generic_file_splice_write, 2439#endif 2440}; 2441 2442static const struct inode_operations shmem_inode_operations = { 2443 .truncate = shmem_truncate, 2444 .setattr = shmem_notify_change, 2445 .truncate_range = shmem_truncate_range, 2446#ifdef CONFIG_TMPFS_POSIX_ACL 2447 .setxattr = generic_setxattr, 2448 .getxattr = generic_getxattr, 2449 .listxattr = generic_listxattr, 2450 .removexattr = generic_removexattr, 2451 .check_acl = shmem_check_acl, 2452#endif 2453 2454}; 2455 2456static const struct inode_operations shmem_dir_inode_operations = { 2457#ifdef CONFIG_TMPFS 2458 .create = shmem_create, 2459 .lookup = simple_lookup, 2460 .link = shmem_link, 2461 .unlink = shmem_unlink, 2462 .symlink = shmem_symlink, 2463 .mkdir = shmem_mkdir, 2464 .rmdir = shmem_rmdir, 2465 .mknod = shmem_mknod, 2466 .rename = shmem_rename, 2467#endif 2468#ifdef CONFIG_TMPFS_POSIX_ACL 2469 .setattr = shmem_notify_change, 2470 .setxattr = generic_setxattr, 2471 .getxattr = generic_getxattr, 2472 .listxattr = generic_listxattr, 2473 .removexattr = generic_removexattr, 2474 .check_acl = shmem_check_acl, 2475#endif 2476}; 2477 2478static const struct inode_operations shmem_special_inode_operations = { 2479#ifdef CONFIG_TMPFS_POSIX_ACL 2480 .setattr = shmem_notify_change, 2481 .setxattr = generic_setxattr, 2482 .getxattr = generic_getxattr, 2483 .listxattr = generic_listxattr, 2484 .removexattr = generic_removexattr, 2485 .check_acl = shmem_check_acl, 2486#endif 2487}; 2488 2489static const struct super_operations shmem_ops = { 2490 .alloc_inode = shmem_alloc_inode, 2491 .destroy_inode = shmem_destroy_inode, 2492#ifdef CONFIG_TMPFS 2493 .statfs = shmem_statfs, 2494 .remount_fs = shmem_remount_fs, 2495 .show_options = shmem_show_options, 2496#endif 2497 .delete_inode = shmem_delete_inode, 2498 .drop_inode = generic_delete_inode, 2499 .put_super = shmem_put_super, 2500}; 2501 2502static struct vm_operations_struct shmem_vm_ops = { 2503 .fault = shmem_fault, 2504#ifdef CONFIG_NUMA 2505 .set_policy = shmem_set_policy, 2506 .get_policy = shmem_get_policy, 2507#endif 2508}; 2509 2510 2511static int shmem_get_sb(struct file_system_type *fs_type, 2512 int flags, const char *dev_name, void *data, struct vfsmount *mnt) 2513{ 2514 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt); 2515} 2516 2517static struct file_system_type tmpfs_fs_type = { 2518 .owner = THIS_MODULE, 2519 .name = "tmpfs", 2520 .get_sb = shmem_get_sb, 2521 .kill_sb = kill_litter_super, 2522}; 2523 2524int __init init_tmpfs(void) 2525{ 2526 int error; 2527 2528 error = bdi_init(&shmem_backing_dev_info); 2529 if (error) 2530 goto out4; 2531 2532 error = init_inodecache(); 2533 if (error) 2534 goto out3; 2535 2536 error = register_filesystem(&tmpfs_fs_type); 2537 if (error) { 2538 printk(KERN_ERR "Could not register tmpfs\n"); 2539 goto out2; 2540 } 2541 2542 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER, 2543 tmpfs_fs_type.name, NULL); 2544 if (IS_ERR(shm_mnt)) { 2545 error = PTR_ERR(shm_mnt); 2546 printk(KERN_ERR "Could not kern_mount tmpfs\n"); 2547 goto out1; 2548 } 2549 return 0; 2550 2551out1: 2552 unregister_filesystem(&tmpfs_fs_type); 2553out2: 2554 destroy_inodecache(); 2555out3: 2556 bdi_destroy(&shmem_backing_dev_info); 2557out4: 2558 shm_mnt = ERR_PTR(error); 2559 return error; 2560} 2561 2562#else /* !CONFIG_SHMEM */ 2563 2564/* 2565 * tiny-shmem: simple shmemfs and tmpfs using ramfs code 2566 * 2567 * This is intended for small system where the benefits of the full 2568 * shmem code (swap-backed and resource-limited) are outweighed by 2569 * their complexity. On systems without swap this code should be 2570 * effectively equivalent, but much lighter weight. 2571 */ 2572 2573#include <linux/ramfs.h> 2574 2575static struct file_system_type tmpfs_fs_type = { 2576 .name = "tmpfs", 2577 .get_sb = ramfs_get_sb, 2578 .kill_sb = kill_litter_super, 2579}; 2580 2581int __init init_tmpfs(void) 2582{ 2583 BUG_ON(register_filesystem(&tmpfs_fs_type) != 0); 2584 2585 shm_mnt = kern_mount(&tmpfs_fs_type); 2586 BUG_ON(IS_ERR(shm_mnt)); 2587 2588 return 0; 2589} 2590 2591int shmem_unuse(swp_entry_t entry, struct page *page) 2592{ 2593 return 0; 2594} 2595 2596#define shmem_vm_ops generic_file_vm_ops 2597#define shmem_file_operations ramfs_file_operations 2598#define shmem_get_inode(sb, mode, dev, flags) ramfs_get_inode(sb, mode, dev) 2599#define shmem_acct_size(flags, size) 0 2600#define shmem_unacct_size(flags, size) do {} while (0) 2601#define SHMEM_MAX_BYTES MAX_LFS_FILESIZE 2602 2603#endif /* CONFIG_SHMEM */ 2604 2605/* common code */ 2606 2607/** 2608 * shmem_file_setup - get an unlinked file living in tmpfs 2609 * @name: name for dentry (to be seen in /proc/<pid>/maps 2610 * @size: size to be set for the file 2611 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size 2612 */ 2613struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) 2614{ 2615 int error; 2616 struct file *file; 2617 struct inode *inode; 2618 struct dentry *dentry, *root; 2619 struct qstr this; 2620 2621 if (IS_ERR(shm_mnt)) 2622 return (void *)shm_mnt; 2623 2624 if (size < 0 || size > SHMEM_MAX_BYTES) 2625 return ERR_PTR(-EINVAL); 2626 2627 if (shmem_acct_size(flags, size)) 2628 return ERR_PTR(-ENOMEM); 2629 2630 error = -ENOMEM; 2631 this.name = name; 2632 this.len = strlen(name); 2633 this.hash = 0; /* will go */ 2634 root = shm_mnt->mnt_root; 2635 dentry = d_alloc(root, &this); 2636 if (!dentry) 2637 goto put_memory; 2638 2639 error = -ENFILE; 2640 file = get_empty_filp(); 2641 if (!file) 2642 goto put_dentry; 2643 2644 error = -ENOSPC; 2645 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0, flags); 2646 if (!inode) 2647 goto close_file; 2648 2649 d_instantiate(dentry, inode); 2650 inode->i_size = size; 2651 inode->i_nlink = 0; /* It is unlinked */ 2652 init_file(file, shm_mnt, dentry, FMODE_WRITE | FMODE_READ, 2653 &shmem_file_operations); 2654 2655#ifndef CONFIG_MMU 2656 error = ramfs_nommu_expand_for_mapping(inode, size); 2657 if (error) 2658 goto close_file; 2659#endif 2660 ima_counts_get(file); 2661 return file; 2662 2663close_file: 2664 put_filp(file); 2665put_dentry: 2666 dput(dentry); 2667put_memory: 2668 shmem_unacct_size(flags, size); 2669 return ERR_PTR(error); 2670} 2671EXPORT_SYMBOL_GPL(shmem_file_setup); 2672 2673/** 2674 * shmem_zero_setup - setup a shared anonymous mapping 2675 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff 2676 */ 2677int shmem_zero_setup(struct vm_area_struct *vma) 2678{ 2679 struct file *file; 2680 loff_t size = vma->vm_end - vma->vm_start; 2681 2682 file = shmem_file_setup("dev/zero", size, vma->vm_flags); 2683 if (IS_ERR(file)) 2684 return PTR_ERR(file); 2685 2686 if (vma->vm_file) 2687 fput(vma->vm_file); 2688 vma->vm_file = file; 2689 vma->vm_ops = &shmem_vm_ops; 2690 return 0; 2691} 2692