shmem.c revision 9fab5619bdd7f84cdd22cc760778f759f9819a33
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/file.h> 28#include <linux/mm.h> 29#include <linux/module.h> 30#include <linux/swap.h> 31#include <linux/ima.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/generic_acl.h> 45#include <linux/mman.h> 46#include <linux/pagemap.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/vfs.h> 53#include <linux/blkdev.h> 54#include <linux/security.h> 55#include <linux/swapops.h> 56#include <linux/mempolicy.h> 57#include <linux/namei.h> 58#include <linux/ctype.h> 59#include <linux/migrate.h> 60#include <linux/highmem.h> 61#include <linux/seq_file.h> 62#include <linux/magic.h> 63 64#include <asm/uaccess.h> 65#include <asm/div64.h> 66#include <asm/pgtable.h> 67 68#define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long)) 69#define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE) 70#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512) 71 72#define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1)) 73#define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT) 74 75#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT) 76 77/* info->flags needs VM_flags to handle pagein/truncate races efficiently */ 78#define SHMEM_PAGEIN VM_READ 79#define SHMEM_TRUNCATE VM_WRITE 80 81/* Definition to limit shmem_truncate's steps between cond_rescheds */ 82#define LATENCY_LIMIT 64 83 84/* Pretend that each entry is of this size in directory's i_size */ 85#define BOGO_DIRENT_SIZE 20 86 87/* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */ 88enum sgp_type { 89 SGP_READ, /* don't exceed i_size, don't allocate page */ 90 SGP_CACHE, /* don't exceed i_size, may allocate page */ 91 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */ 92 SGP_WRITE, /* may exceed i_size, may allocate page */ 93}; 94 95#ifdef CONFIG_TMPFS 96static unsigned long shmem_default_max_blocks(void) 97{ 98 return totalram_pages / 2; 99} 100 101static unsigned long shmem_default_max_inodes(void) 102{ 103 return min(totalram_pages - totalhigh_pages, totalram_pages / 2); 104} 105#endif 106 107static int shmem_getpage(struct inode *inode, unsigned long idx, 108 struct page **pagep, enum sgp_type sgp, int *type); 109 110static inline struct page *shmem_dir_alloc(gfp_t gfp_mask) 111{ 112 /* 113 * The above definition of ENTRIES_PER_PAGE, and the use of 114 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE: 115 * might be reconsidered if it ever diverges from PAGE_SIZE. 116 * 117 * Mobility flags are masked out as swap vectors cannot move 118 */ 119 return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO, 120 PAGE_CACHE_SHIFT-PAGE_SHIFT); 121} 122 123static inline void shmem_dir_free(struct page *page) 124{ 125 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT); 126} 127 128static struct page **shmem_dir_map(struct page *page) 129{ 130 return (struct page **)kmap_atomic(page, KM_USER0); 131} 132 133static inline void shmem_dir_unmap(struct page **dir) 134{ 135 kunmap_atomic(dir, KM_USER0); 136} 137 138static swp_entry_t *shmem_swp_map(struct page *page) 139{ 140 return (swp_entry_t *)kmap_atomic(page, KM_USER1); 141} 142 143static inline void shmem_swp_balance_unmap(void) 144{ 145 /* 146 * When passing a pointer to an i_direct entry, to code which 147 * also handles indirect entries and so will shmem_swp_unmap, 148 * we must arrange for the preempt count to remain in balance. 149 * What kmap_atomic of a lowmem page does depends on config 150 * and architecture, so pretend to kmap_atomic some lowmem page. 151 */ 152 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1); 153} 154 155static inline void shmem_swp_unmap(swp_entry_t *entry) 156{ 157 kunmap_atomic(entry, KM_USER1); 158} 159 160static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) 161{ 162 return sb->s_fs_info; 163} 164 165/* 166 * shmem_file_setup pre-accounts the whole fixed size of a VM object, 167 * for shared memory and for shared anonymous (/dev/zero) mappings 168 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), 169 * consistent with the pre-accounting of private mappings ... 170 */ 171static inline int shmem_acct_size(unsigned long flags, loff_t size) 172{ 173 return (flags & VM_NORESERVE) ? 174 0 : security_vm_enough_memory_kern(VM_ACCT(size)); 175} 176 177static inline void shmem_unacct_size(unsigned long flags, loff_t size) 178{ 179 if (!(flags & VM_NORESERVE)) 180 vm_unacct_memory(VM_ACCT(size)); 181} 182 183/* 184 * ... whereas tmpfs objects are accounted incrementally as 185 * pages are allocated, in order to allow huge sparse files. 186 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, 187 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. 188 */ 189static inline int shmem_acct_block(unsigned long flags) 190{ 191 return (flags & VM_NORESERVE) ? 192 security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE)) : 0; 193} 194 195static inline void shmem_unacct_blocks(unsigned long flags, long pages) 196{ 197 if (flags & VM_NORESERVE) 198 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE)); 199} 200 201static const struct super_operations shmem_ops; 202static const struct address_space_operations shmem_aops; 203static const struct file_operations shmem_file_operations; 204static const struct inode_operations shmem_inode_operations; 205static const struct inode_operations shmem_dir_inode_operations; 206static const struct inode_operations shmem_special_inode_operations; 207static struct vm_operations_struct shmem_vm_ops; 208 209static struct backing_dev_info shmem_backing_dev_info __read_mostly = { 210 .ra_pages = 0, /* No readahead */ 211 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED, 212 .unplug_io_fn = default_unplug_io_fn, 213}; 214 215static LIST_HEAD(shmem_swaplist); 216static DEFINE_MUTEX(shmem_swaplist_mutex); 217 218static void shmem_free_blocks(struct inode *inode, long pages) 219{ 220 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 221 if (sbinfo->max_blocks) { 222 spin_lock(&sbinfo->stat_lock); 223 sbinfo->free_blocks += pages; 224 inode->i_blocks -= pages*BLOCKS_PER_PAGE; 225 spin_unlock(&sbinfo->stat_lock); 226 } 227} 228 229static int shmem_reserve_inode(struct super_block *sb) 230{ 231 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 232 if (sbinfo->max_inodes) { 233 spin_lock(&sbinfo->stat_lock); 234 if (!sbinfo->free_inodes) { 235 spin_unlock(&sbinfo->stat_lock); 236 return -ENOSPC; 237 } 238 sbinfo->free_inodes--; 239 spin_unlock(&sbinfo->stat_lock); 240 } 241 return 0; 242} 243 244static void shmem_free_inode(struct super_block *sb) 245{ 246 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 247 if (sbinfo->max_inodes) { 248 spin_lock(&sbinfo->stat_lock); 249 sbinfo->free_inodes++; 250 spin_unlock(&sbinfo->stat_lock); 251 } 252} 253 254/** 255 * shmem_recalc_inode - recalculate the size of an inode 256 * @inode: inode to recalc 257 * 258 * We have to calculate the free blocks since the mm can drop 259 * undirtied hole pages behind our back. 260 * 261 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped 262 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) 263 * 264 * It has to be called with the spinlock held. 265 */ 266static void shmem_recalc_inode(struct inode *inode) 267{ 268 struct shmem_inode_info *info = SHMEM_I(inode); 269 long freed; 270 271 freed = info->alloced - info->swapped - inode->i_mapping->nrpages; 272 if (freed > 0) { 273 info->alloced -= freed; 274 shmem_unacct_blocks(info->flags, freed); 275 shmem_free_blocks(inode, freed); 276 } 277} 278 279/** 280 * shmem_swp_entry - find the swap vector position in the info structure 281 * @info: info structure for the inode 282 * @index: index of the page to find 283 * @page: optional page to add to the structure. Has to be preset to 284 * all zeros 285 * 286 * If there is no space allocated yet it will return NULL when 287 * page is NULL, else it will use the page for the needed block, 288 * setting it to NULL on return to indicate that it has been used. 289 * 290 * The swap vector is organized the following way: 291 * 292 * There are SHMEM_NR_DIRECT entries directly stored in the 293 * shmem_inode_info structure. So small files do not need an addional 294 * allocation. 295 * 296 * For pages with index > SHMEM_NR_DIRECT there is the pointer 297 * i_indirect which points to a page which holds in the first half 298 * doubly indirect blocks, in the second half triple indirect blocks: 299 * 300 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the 301 * following layout (for SHMEM_NR_DIRECT == 16): 302 * 303 * i_indirect -> dir --> 16-19 304 * | +-> 20-23 305 * | 306 * +-->dir2 --> 24-27 307 * | +-> 28-31 308 * | +-> 32-35 309 * | +-> 36-39 310 * | 311 * +-->dir3 --> 40-43 312 * +-> 44-47 313 * +-> 48-51 314 * +-> 52-55 315 */ 316static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page) 317{ 318 unsigned long offset; 319 struct page **dir; 320 struct page *subdir; 321 322 if (index < SHMEM_NR_DIRECT) { 323 shmem_swp_balance_unmap(); 324 return info->i_direct+index; 325 } 326 if (!info->i_indirect) { 327 if (page) { 328 info->i_indirect = *page; 329 *page = NULL; 330 } 331 return NULL; /* need another page */ 332 } 333 334 index -= SHMEM_NR_DIRECT; 335 offset = index % ENTRIES_PER_PAGE; 336 index /= ENTRIES_PER_PAGE; 337 dir = shmem_dir_map(info->i_indirect); 338 339 if (index >= ENTRIES_PER_PAGE/2) { 340 index -= ENTRIES_PER_PAGE/2; 341 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE; 342 index %= ENTRIES_PER_PAGE; 343 subdir = *dir; 344 if (!subdir) { 345 if (page) { 346 *dir = *page; 347 *page = NULL; 348 } 349 shmem_dir_unmap(dir); 350 return NULL; /* need another page */ 351 } 352 shmem_dir_unmap(dir); 353 dir = shmem_dir_map(subdir); 354 } 355 356 dir += index; 357 subdir = *dir; 358 if (!subdir) { 359 if (!page || !(subdir = *page)) { 360 shmem_dir_unmap(dir); 361 return NULL; /* need a page */ 362 } 363 *dir = subdir; 364 *page = NULL; 365 } 366 shmem_dir_unmap(dir); 367 return shmem_swp_map(subdir) + offset; 368} 369 370static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value) 371{ 372 long incdec = value? 1: -1; 373 374 entry->val = value; 375 info->swapped += incdec; 376 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) { 377 struct page *page = kmap_atomic_to_page(entry); 378 set_page_private(page, page_private(page) + incdec); 379 } 380} 381 382/** 383 * shmem_swp_alloc - get the position of the swap entry for the page. 384 * @info: info structure for the inode 385 * @index: index of the page to find 386 * @sgp: check and recheck i_size? skip allocation? 387 * 388 * If the entry does not exist, allocate it. 389 */ 390static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp) 391{ 392 struct inode *inode = &info->vfs_inode; 393 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 394 struct page *page = NULL; 395 swp_entry_t *entry; 396 397 if (sgp != SGP_WRITE && 398 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 399 return ERR_PTR(-EINVAL); 400 401 while (!(entry = shmem_swp_entry(info, index, &page))) { 402 if (sgp == SGP_READ) 403 return shmem_swp_map(ZERO_PAGE(0)); 404 /* 405 * Test free_blocks against 1 not 0, since we have 1 data 406 * page (and perhaps indirect index pages) yet to allocate: 407 * a waste to allocate index if we cannot allocate data. 408 */ 409 if (sbinfo->max_blocks) { 410 spin_lock(&sbinfo->stat_lock); 411 if (sbinfo->free_blocks <= 1) { 412 spin_unlock(&sbinfo->stat_lock); 413 return ERR_PTR(-ENOSPC); 414 } 415 sbinfo->free_blocks--; 416 inode->i_blocks += BLOCKS_PER_PAGE; 417 spin_unlock(&sbinfo->stat_lock); 418 } 419 420 spin_unlock(&info->lock); 421 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping)); 422 if (page) 423 set_page_private(page, 0); 424 spin_lock(&info->lock); 425 426 if (!page) { 427 shmem_free_blocks(inode, 1); 428 return ERR_PTR(-ENOMEM); 429 } 430 if (sgp != SGP_WRITE && 431 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { 432 entry = ERR_PTR(-EINVAL); 433 break; 434 } 435 if (info->next_index <= index) 436 info->next_index = index + 1; 437 } 438 if (page) { 439 /* another task gave its page, or truncated the file */ 440 shmem_free_blocks(inode, 1); 441 shmem_dir_free(page); 442 } 443 if (info->next_index <= index && !IS_ERR(entry)) 444 info->next_index = index + 1; 445 return entry; 446} 447 448/** 449 * shmem_free_swp - free some swap entries in a directory 450 * @dir: pointer to the directory 451 * @edir: pointer after last entry of the directory 452 * @punch_lock: pointer to spinlock when needed for the holepunch case 453 */ 454static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir, 455 spinlock_t *punch_lock) 456{ 457 spinlock_t *punch_unlock = NULL; 458 swp_entry_t *ptr; 459 int freed = 0; 460 461 for (ptr = dir; ptr < edir; ptr++) { 462 if (ptr->val) { 463 if (unlikely(punch_lock)) { 464 punch_unlock = punch_lock; 465 punch_lock = NULL; 466 spin_lock(punch_unlock); 467 if (!ptr->val) 468 continue; 469 } 470 free_swap_and_cache(*ptr); 471 *ptr = (swp_entry_t){0}; 472 freed++; 473 } 474 } 475 if (punch_unlock) 476 spin_unlock(punch_unlock); 477 return freed; 478} 479 480static int shmem_map_and_free_swp(struct page *subdir, int offset, 481 int limit, struct page ***dir, spinlock_t *punch_lock) 482{ 483 swp_entry_t *ptr; 484 int freed = 0; 485 486 ptr = shmem_swp_map(subdir); 487 for (; offset < limit; offset += LATENCY_LIMIT) { 488 int size = limit - offset; 489 if (size > LATENCY_LIMIT) 490 size = LATENCY_LIMIT; 491 freed += shmem_free_swp(ptr+offset, ptr+offset+size, 492 punch_lock); 493 if (need_resched()) { 494 shmem_swp_unmap(ptr); 495 if (*dir) { 496 shmem_dir_unmap(*dir); 497 *dir = NULL; 498 } 499 cond_resched(); 500 ptr = shmem_swp_map(subdir); 501 } 502 } 503 shmem_swp_unmap(ptr); 504 return freed; 505} 506 507static void shmem_free_pages(struct list_head *next) 508{ 509 struct page *page; 510 int freed = 0; 511 512 do { 513 page = container_of(next, struct page, lru); 514 next = next->next; 515 shmem_dir_free(page); 516 freed++; 517 if (freed >= LATENCY_LIMIT) { 518 cond_resched(); 519 freed = 0; 520 } 521 } while (next); 522} 523 524static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end) 525{ 526 struct shmem_inode_info *info = SHMEM_I(inode); 527 unsigned long idx; 528 unsigned long size; 529 unsigned long limit; 530 unsigned long stage; 531 unsigned long diroff; 532 struct page **dir; 533 struct page *topdir; 534 struct page *middir; 535 struct page *subdir; 536 swp_entry_t *ptr; 537 LIST_HEAD(pages_to_free); 538 long nr_pages_to_free = 0; 539 long nr_swaps_freed = 0; 540 int offset; 541 int freed; 542 int punch_hole; 543 spinlock_t *needs_lock; 544 spinlock_t *punch_lock; 545 unsigned long upper_limit; 546 547 inode->i_ctime = inode->i_mtime = CURRENT_TIME; 548 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 549 if (idx >= info->next_index) 550 return; 551 552 spin_lock(&info->lock); 553 info->flags |= SHMEM_TRUNCATE; 554 if (likely(end == (loff_t) -1)) { 555 limit = info->next_index; 556 upper_limit = SHMEM_MAX_INDEX; 557 info->next_index = idx; 558 needs_lock = NULL; 559 punch_hole = 0; 560 } else { 561 if (end + 1 >= inode->i_size) { /* we may free a little more */ 562 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >> 563 PAGE_CACHE_SHIFT; 564 upper_limit = SHMEM_MAX_INDEX; 565 } else { 566 limit = (end + 1) >> PAGE_CACHE_SHIFT; 567 upper_limit = limit; 568 } 569 needs_lock = &info->lock; 570 punch_hole = 1; 571 } 572 573 topdir = info->i_indirect; 574 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) { 575 info->i_indirect = NULL; 576 nr_pages_to_free++; 577 list_add(&topdir->lru, &pages_to_free); 578 } 579 spin_unlock(&info->lock); 580 581 if (info->swapped && idx < SHMEM_NR_DIRECT) { 582 ptr = info->i_direct; 583 size = limit; 584 if (size > SHMEM_NR_DIRECT) 585 size = SHMEM_NR_DIRECT; 586 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock); 587 } 588 589 /* 590 * If there are no indirect blocks or we are punching a hole 591 * below indirect blocks, nothing to be done. 592 */ 593 if (!topdir || limit <= SHMEM_NR_DIRECT) 594 goto done2; 595 596 /* 597 * The truncation case has already dropped info->lock, and we're safe 598 * because i_size and next_index have already been lowered, preventing 599 * access beyond. But in the punch_hole case, we still need to take 600 * the lock when updating the swap directory, because there might be 601 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or 602 * shmem_writepage. However, whenever we find we can remove a whole 603 * directory page (not at the misaligned start or end of the range), 604 * we first NULLify its pointer in the level above, and then have no 605 * need to take the lock when updating its contents: needs_lock and 606 * punch_lock (either pointing to info->lock or NULL) manage this. 607 */ 608 609 upper_limit -= SHMEM_NR_DIRECT; 610 limit -= SHMEM_NR_DIRECT; 611 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0; 612 offset = idx % ENTRIES_PER_PAGE; 613 idx -= offset; 614 615 dir = shmem_dir_map(topdir); 616 stage = ENTRIES_PER_PAGEPAGE/2; 617 if (idx < ENTRIES_PER_PAGEPAGE/2) { 618 middir = topdir; 619 diroff = idx/ENTRIES_PER_PAGE; 620 } else { 621 dir += ENTRIES_PER_PAGE/2; 622 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE; 623 while (stage <= idx) 624 stage += ENTRIES_PER_PAGEPAGE; 625 middir = *dir; 626 if (*dir) { 627 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) % 628 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE; 629 if (!diroff && !offset && upper_limit >= stage) { 630 if (needs_lock) { 631 spin_lock(needs_lock); 632 *dir = NULL; 633 spin_unlock(needs_lock); 634 needs_lock = NULL; 635 } else 636 *dir = NULL; 637 nr_pages_to_free++; 638 list_add(&middir->lru, &pages_to_free); 639 } 640 shmem_dir_unmap(dir); 641 dir = shmem_dir_map(middir); 642 } else { 643 diroff = 0; 644 offset = 0; 645 idx = stage; 646 } 647 } 648 649 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) { 650 if (unlikely(idx == stage)) { 651 shmem_dir_unmap(dir); 652 dir = shmem_dir_map(topdir) + 653 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 654 while (!*dir) { 655 dir++; 656 idx += ENTRIES_PER_PAGEPAGE; 657 if (idx >= limit) 658 goto done1; 659 } 660 stage = idx + ENTRIES_PER_PAGEPAGE; 661 middir = *dir; 662 if (punch_hole) 663 needs_lock = &info->lock; 664 if (upper_limit >= stage) { 665 if (needs_lock) { 666 spin_lock(needs_lock); 667 *dir = NULL; 668 spin_unlock(needs_lock); 669 needs_lock = NULL; 670 } else 671 *dir = NULL; 672 nr_pages_to_free++; 673 list_add(&middir->lru, &pages_to_free); 674 } 675 shmem_dir_unmap(dir); 676 cond_resched(); 677 dir = shmem_dir_map(middir); 678 diroff = 0; 679 } 680 punch_lock = needs_lock; 681 subdir = dir[diroff]; 682 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) { 683 if (needs_lock) { 684 spin_lock(needs_lock); 685 dir[diroff] = NULL; 686 spin_unlock(needs_lock); 687 punch_lock = NULL; 688 } else 689 dir[diroff] = NULL; 690 nr_pages_to_free++; 691 list_add(&subdir->lru, &pages_to_free); 692 } 693 if (subdir && page_private(subdir) /* has swap entries */) { 694 size = limit - idx; 695 if (size > ENTRIES_PER_PAGE) 696 size = ENTRIES_PER_PAGE; 697 freed = shmem_map_and_free_swp(subdir, 698 offset, size, &dir, punch_lock); 699 if (!dir) 700 dir = shmem_dir_map(middir); 701 nr_swaps_freed += freed; 702 if (offset || punch_lock) { 703 spin_lock(&info->lock); 704 set_page_private(subdir, 705 page_private(subdir) - freed); 706 spin_unlock(&info->lock); 707 } else 708 BUG_ON(page_private(subdir) != freed); 709 } 710 offset = 0; 711 } 712done1: 713 shmem_dir_unmap(dir); 714done2: 715 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) { 716 /* 717 * Call truncate_inode_pages again: racing shmem_unuse_inode 718 * may have swizzled a page in from swap since vmtruncate or 719 * generic_delete_inode did it, before we lowered next_index. 720 * Also, though shmem_getpage checks i_size before adding to 721 * cache, no recheck after: so fix the narrow window there too. 722 * 723 * Recalling truncate_inode_pages_range and unmap_mapping_range 724 * every time for punch_hole (which never got a chance to clear 725 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive, 726 * yet hardly ever necessary: try to optimize them out later. 727 */ 728 truncate_inode_pages_range(inode->i_mapping, start, end); 729 if (punch_hole) 730 unmap_mapping_range(inode->i_mapping, start, 731 end - start, 1); 732 } 733 734 spin_lock(&info->lock); 735 info->flags &= ~SHMEM_TRUNCATE; 736 info->swapped -= nr_swaps_freed; 737 if (nr_pages_to_free) 738 shmem_free_blocks(inode, nr_pages_to_free); 739 shmem_recalc_inode(inode); 740 spin_unlock(&info->lock); 741 742 /* 743 * Empty swap vector directory pages to be freed? 744 */ 745 if (!list_empty(&pages_to_free)) { 746 pages_to_free.prev->next = NULL; 747 shmem_free_pages(pages_to_free.next); 748 } 749} 750 751static void shmem_truncate(struct inode *inode) 752{ 753 shmem_truncate_range(inode, inode->i_size, (loff_t)-1); 754} 755 756static int shmem_notify_change(struct dentry *dentry, struct iattr *attr) 757{ 758 struct inode *inode = dentry->d_inode; 759 struct page *page = NULL; 760 int error; 761 762 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { 763 if (attr->ia_size < inode->i_size) { 764 /* 765 * If truncating down to a partial page, then 766 * if that page is already allocated, hold it 767 * in memory until the truncation is over, so 768 * truncate_partial_page cannnot miss it were 769 * it assigned to swap. 770 */ 771 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) { 772 (void) shmem_getpage(inode, 773 attr->ia_size>>PAGE_CACHE_SHIFT, 774 &page, SGP_READ, NULL); 775 if (page) 776 unlock_page(page); 777 } 778 /* 779 * Reset SHMEM_PAGEIN flag so that shmem_truncate can 780 * detect if any pages might have been added to cache 781 * after truncate_inode_pages. But we needn't bother 782 * if it's being fully truncated to zero-length: the 783 * nrpages check is efficient enough in that case. 784 */ 785 if (attr->ia_size) { 786 struct shmem_inode_info *info = SHMEM_I(inode); 787 spin_lock(&info->lock); 788 info->flags &= ~SHMEM_PAGEIN; 789 spin_unlock(&info->lock); 790 } 791 } 792 } 793 794 error = inode_change_ok(inode, attr); 795 if (!error) 796 error = inode_setattr(inode, attr); 797#ifdef CONFIG_TMPFS_POSIX_ACL 798 if (!error && (attr->ia_valid & ATTR_MODE)) 799 error = generic_acl_chmod(inode, &shmem_acl_ops); 800#endif 801 if (page) 802 page_cache_release(page); 803 return error; 804} 805 806static void shmem_delete_inode(struct inode *inode) 807{ 808 struct shmem_inode_info *info = SHMEM_I(inode); 809 810 if (inode->i_op->truncate == shmem_truncate) { 811 truncate_inode_pages(inode->i_mapping, 0); 812 shmem_unacct_size(info->flags, inode->i_size); 813 inode->i_size = 0; 814 shmem_truncate(inode); 815 if (!list_empty(&info->swaplist)) { 816 mutex_lock(&shmem_swaplist_mutex); 817 list_del_init(&info->swaplist); 818 mutex_unlock(&shmem_swaplist_mutex); 819 } 820 } 821 BUG_ON(inode->i_blocks); 822 shmem_free_inode(inode->i_sb); 823 clear_inode(inode); 824} 825 826static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir) 827{ 828 swp_entry_t *ptr; 829 830 for (ptr = dir; ptr < edir; ptr++) { 831 if (ptr->val == entry.val) 832 return ptr - dir; 833 } 834 return -1; 835} 836 837static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page) 838{ 839 struct inode *inode; 840 unsigned long idx; 841 unsigned long size; 842 unsigned long limit; 843 unsigned long stage; 844 struct page **dir; 845 struct page *subdir; 846 swp_entry_t *ptr; 847 int offset; 848 int error; 849 850 idx = 0; 851 ptr = info->i_direct; 852 spin_lock(&info->lock); 853 if (!info->swapped) { 854 list_del_init(&info->swaplist); 855 goto lost2; 856 } 857 limit = info->next_index; 858 size = limit; 859 if (size > SHMEM_NR_DIRECT) 860 size = SHMEM_NR_DIRECT; 861 offset = shmem_find_swp(entry, ptr, ptr+size); 862 if (offset >= 0) 863 goto found; 864 if (!info->i_indirect) 865 goto lost2; 866 867 dir = shmem_dir_map(info->i_indirect); 868 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2; 869 870 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) { 871 if (unlikely(idx == stage)) { 872 shmem_dir_unmap(dir-1); 873 if (cond_resched_lock(&info->lock)) { 874 /* check it has not been truncated */ 875 if (limit > info->next_index) { 876 limit = info->next_index; 877 if (idx >= limit) 878 goto lost2; 879 } 880 } 881 dir = shmem_dir_map(info->i_indirect) + 882 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 883 while (!*dir) { 884 dir++; 885 idx += ENTRIES_PER_PAGEPAGE; 886 if (idx >= limit) 887 goto lost1; 888 } 889 stage = idx + ENTRIES_PER_PAGEPAGE; 890 subdir = *dir; 891 shmem_dir_unmap(dir); 892 dir = shmem_dir_map(subdir); 893 } 894 subdir = *dir; 895 if (subdir && page_private(subdir)) { 896 ptr = shmem_swp_map(subdir); 897 size = limit - idx; 898 if (size > ENTRIES_PER_PAGE) 899 size = ENTRIES_PER_PAGE; 900 offset = shmem_find_swp(entry, ptr, ptr+size); 901 shmem_swp_unmap(ptr); 902 if (offset >= 0) { 903 shmem_dir_unmap(dir); 904 goto found; 905 } 906 } 907 } 908lost1: 909 shmem_dir_unmap(dir-1); 910lost2: 911 spin_unlock(&info->lock); 912 return 0; 913found: 914 idx += offset; 915 inode = igrab(&info->vfs_inode); 916 spin_unlock(&info->lock); 917 918 /* 919 * Move _head_ to start search for next from here. 920 * But be careful: shmem_delete_inode checks list_empty without taking 921 * mutex, and there's an instant in list_move_tail when info->swaplist 922 * would appear empty, if it were the only one on shmem_swaplist. We 923 * could avoid doing it if inode NULL; or use this minor optimization. 924 */ 925 if (shmem_swaplist.next != &info->swaplist) 926 list_move_tail(&shmem_swaplist, &info->swaplist); 927 mutex_unlock(&shmem_swaplist_mutex); 928 929 error = 1; 930 if (!inode) 931 goto out; 932 /* 933 * Charge page using GFP_KERNEL while we can wait. 934 * Charged back to the user(not to caller) when swap account is used. 935 * add_to_page_cache() will be called with GFP_NOWAIT. 936 */ 937 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL); 938 if (error) 939 goto out; 940 error = radix_tree_preload(GFP_KERNEL); 941 if (error) { 942 mem_cgroup_uncharge_cache_page(page); 943 goto out; 944 } 945 error = 1; 946 947 spin_lock(&info->lock); 948 ptr = shmem_swp_entry(info, idx, NULL); 949 if (ptr && ptr->val == entry.val) { 950 error = add_to_page_cache_locked(page, inode->i_mapping, 951 idx, GFP_NOWAIT); 952 /* does mem_cgroup_uncharge_cache_page on error */ 953 } else /* we must compensate for our precharge above */ 954 mem_cgroup_uncharge_cache_page(page); 955 956 if (error == -EEXIST) { 957 struct page *filepage = find_get_page(inode->i_mapping, idx); 958 error = 1; 959 if (filepage) { 960 /* 961 * There might be a more uptodate page coming down 962 * from a stacked writepage: forget our swappage if so. 963 */ 964 if (PageUptodate(filepage)) 965 error = 0; 966 page_cache_release(filepage); 967 } 968 } 969 if (!error) { 970 delete_from_swap_cache(page); 971 set_page_dirty(page); 972 info->flags |= SHMEM_PAGEIN; 973 shmem_swp_set(info, ptr, 0); 974 swap_free(entry); 975 error = 1; /* not an error, but entry was found */ 976 } 977 if (ptr) 978 shmem_swp_unmap(ptr); 979 spin_unlock(&info->lock); 980 radix_tree_preload_end(); 981out: 982 unlock_page(page); 983 page_cache_release(page); 984 iput(inode); /* allows for NULL */ 985 return error; 986} 987 988/* 989 * shmem_unuse() search for an eventually swapped out shmem page. 990 */ 991int shmem_unuse(swp_entry_t entry, struct page *page) 992{ 993 struct list_head *p, *next; 994 struct shmem_inode_info *info; 995 int found = 0; 996 997 mutex_lock(&shmem_swaplist_mutex); 998 list_for_each_safe(p, next, &shmem_swaplist) { 999 info = list_entry(p, struct shmem_inode_info, swaplist); 1000 found = shmem_unuse_inode(info, entry, page); 1001 cond_resched(); 1002 if (found) 1003 goto out; 1004 } 1005 mutex_unlock(&shmem_swaplist_mutex); 1006out: return found; /* 0 or 1 or -ENOMEM */ 1007} 1008 1009/* 1010 * Move the page from the page cache to the swap cache. 1011 */ 1012static int shmem_writepage(struct page *page, struct writeback_control *wbc) 1013{ 1014 struct shmem_inode_info *info; 1015 swp_entry_t *entry, swap; 1016 struct address_space *mapping; 1017 unsigned long index; 1018 struct inode *inode; 1019 1020 BUG_ON(!PageLocked(page)); 1021 mapping = page->mapping; 1022 index = page->index; 1023 inode = mapping->host; 1024 info = SHMEM_I(inode); 1025 if (info->flags & VM_LOCKED) 1026 goto redirty; 1027 if (!total_swap_pages) 1028 goto redirty; 1029 1030 /* 1031 * shmem_backing_dev_info's capabilities prevent regular writeback or 1032 * sync from ever calling shmem_writepage; but a stacking filesystem 1033 * may use the ->writepage of its underlying filesystem, in which case 1034 * tmpfs should write out to swap only in response to memory pressure, 1035 * and not for pdflush or sync. However, in those cases, we do still 1036 * want to check if there's a redundant swappage to be discarded. 1037 */ 1038 if (wbc->for_reclaim) 1039 swap = get_swap_page(); 1040 else 1041 swap.val = 0; 1042 1043 spin_lock(&info->lock); 1044 if (index >= info->next_index) { 1045 BUG_ON(!(info->flags & SHMEM_TRUNCATE)); 1046 goto unlock; 1047 } 1048 entry = shmem_swp_entry(info, index, NULL); 1049 if (entry->val) { 1050 /* 1051 * The more uptodate page coming down from a stacked 1052 * writepage should replace our old swappage. 1053 */ 1054 free_swap_and_cache(*entry); 1055 shmem_swp_set(info, entry, 0); 1056 } 1057 shmem_recalc_inode(inode); 1058 1059 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { 1060 remove_from_page_cache(page); 1061 shmem_swp_set(info, entry, swap.val); 1062 shmem_swp_unmap(entry); 1063 if (list_empty(&info->swaplist)) 1064 inode = igrab(inode); 1065 else 1066 inode = NULL; 1067 spin_unlock(&info->lock); 1068 swap_duplicate(swap); 1069 BUG_ON(page_mapped(page)); 1070 page_cache_release(page); /* pagecache ref */ 1071 swap_writepage(page, wbc); 1072 if (inode) { 1073 mutex_lock(&shmem_swaplist_mutex); 1074 /* move instead of add in case we're racing */ 1075 list_move_tail(&info->swaplist, &shmem_swaplist); 1076 mutex_unlock(&shmem_swaplist_mutex); 1077 iput(inode); 1078 } 1079 return 0; 1080 } 1081 1082 shmem_swp_unmap(entry); 1083unlock: 1084 spin_unlock(&info->lock); 1085 swap_free(swap); 1086redirty: 1087 set_page_dirty(page); 1088 if (wbc->for_reclaim) 1089 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ 1090 unlock_page(page); 1091 return 0; 1092} 1093 1094#ifdef CONFIG_NUMA 1095#ifdef CONFIG_TMPFS 1096static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) 1097{ 1098 char buffer[64]; 1099 1100 if (!mpol || mpol->mode == MPOL_DEFAULT) 1101 return; /* show nothing */ 1102 1103 mpol_to_str(buffer, sizeof(buffer), mpol, 1); 1104 1105 seq_printf(seq, ",mpol=%s", buffer); 1106} 1107 1108static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) 1109{ 1110 struct mempolicy *mpol = NULL; 1111 if (sbinfo->mpol) { 1112 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ 1113 mpol = sbinfo->mpol; 1114 mpol_get(mpol); 1115 spin_unlock(&sbinfo->stat_lock); 1116 } 1117 return mpol; 1118} 1119#endif /* CONFIG_TMPFS */ 1120 1121static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, 1122 struct shmem_inode_info *info, unsigned long idx) 1123{ 1124 struct mempolicy mpol, *spol; 1125 struct vm_area_struct pvma; 1126 struct page *page; 1127 1128 spol = mpol_cond_copy(&mpol, 1129 mpol_shared_policy_lookup(&info->policy, idx)); 1130 1131 /* Create a pseudo vma that just contains the policy */ 1132 pvma.vm_start = 0; 1133 pvma.vm_pgoff = idx; 1134 pvma.vm_ops = NULL; 1135 pvma.vm_policy = spol; 1136 page = swapin_readahead(entry, gfp, &pvma, 0); 1137 return page; 1138} 1139 1140static struct page *shmem_alloc_page(gfp_t gfp, 1141 struct shmem_inode_info *info, unsigned long idx) 1142{ 1143 struct vm_area_struct pvma; 1144 1145 /* Create a pseudo vma that just contains the policy */ 1146 pvma.vm_start = 0; 1147 pvma.vm_pgoff = idx; 1148 pvma.vm_ops = NULL; 1149 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx); 1150 1151 /* 1152 * alloc_page_vma() will drop the shared policy reference 1153 */ 1154 return alloc_page_vma(gfp, &pvma, 0); 1155} 1156#else /* !CONFIG_NUMA */ 1157#ifdef CONFIG_TMPFS 1158static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p) 1159{ 1160} 1161#endif /* CONFIG_TMPFS */ 1162 1163static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, 1164 struct shmem_inode_info *info, unsigned long idx) 1165{ 1166 return swapin_readahead(entry, gfp, NULL, 0); 1167} 1168 1169static inline struct page *shmem_alloc_page(gfp_t gfp, 1170 struct shmem_inode_info *info, unsigned long idx) 1171{ 1172 return alloc_page(gfp); 1173} 1174#endif /* CONFIG_NUMA */ 1175 1176#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS) 1177static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) 1178{ 1179 return NULL; 1180} 1181#endif 1182 1183/* 1184 * shmem_getpage - either get the page from swap or allocate a new one 1185 * 1186 * If we allocate a new one we do not mark it dirty. That's up to the 1187 * vm. If we swap it in we mark it dirty since we also free the swap 1188 * entry since a page cannot live in both the swap and page cache 1189 */ 1190static int shmem_getpage(struct inode *inode, unsigned long idx, 1191 struct page **pagep, enum sgp_type sgp, int *type) 1192{ 1193 struct address_space *mapping = inode->i_mapping; 1194 struct shmem_inode_info *info = SHMEM_I(inode); 1195 struct shmem_sb_info *sbinfo; 1196 struct page *filepage = *pagep; 1197 struct page *swappage; 1198 swp_entry_t *entry; 1199 swp_entry_t swap; 1200 gfp_t gfp; 1201 int error; 1202 1203 if (idx >= SHMEM_MAX_INDEX) 1204 return -EFBIG; 1205 1206 if (type) 1207 *type = 0; 1208 1209 /* 1210 * Normally, filepage is NULL on entry, and either found 1211 * uptodate immediately, or allocated and zeroed, or read 1212 * in under swappage, which is then assigned to filepage. 1213 * But shmem_readpage (required for splice) passes in a locked 1214 * filepage, which may be found not uptodate by other callers 1215 * too, and may need to be copied from the swappage read in. 1216 */ 1217repeat: 1218 if (!filepage) 1219 filepage = find_lock_page(mapping, idx); 1220 if (filepage && PageUptodate(filepage)) 1221 goto done; 1222 error = 0; 1223 gfp = mapping_gfp_mask(mapping); 1224 if (!filepage) { 1225 /* 1226 * Try to preload while we can wait, to not make a habit of 1227 * draining atomic reserves; but don't latch on to this cpu. 1228 */ 1229 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM); 1230 if (error) 1231 goto failed; 1232 radix_tree_preload_end(); 1233 } 1234 1235 spin_lock(&info->lock); 1236 shmem_recalc_inode(inode); 1237 entry = shmem_swp_alloc(info, idx, sgp); 1238 if (IS_ERR(entry)) { 1239 spin_unlock(&info->lock); 1240 error = PTR_ERR(entry); 1241 goto failed; 1242 } 1243 swap = *entry; 1244 1245 if (swap.val) { 1246 /* Look it up and read it in.. */ 1247 swappage = lookup_swap_cache(swap); 1248 if (!swappage) { 1249 shmem_swp_unmap(entry); 1250 /* here we actually do the io */ 1251 if (type && !(*type & VM_FAULT_MAJOR)) { 1252 __count_vm_event(PGMAJFAULT); 1253 *type |= VM_FAULT_MAJOR; 1254 } 1255 spin_unlock(&info->lock); 1256 swappage = shmem_swapin(swap, gfp, info, idx); 1257 if (!swappage) { 1258 spin_lock(&info->lock); 1259 entry = shmem_swp_alloc(info, idx, sgp); 1260 if (IS_ERR(entry)) 1261 error = PTR_ERR(entry); 1262 else { 1263 if (entry->val == swap.val) 1264 error = -ENOMEM; 1265 shmem_swp_unmap(entry); 1266 } 1267 spin_unlock(&info->lock); 1268 if (error) 1269 goto failed; 1270 goto repeat; 1271 } 1272 wait_on_page_locked(swappage); 1273 page_cache_release(swappage); 1274 goto repeat; 1275 } 1276 1277 /* We have to do this with page locked to prevent races */ 1278 if (!trylock_page(swappage)) { 1279 shmem_swp_unmap(entry); 1280 spin_unlock(&info->lock); 1281 wait_on_page_locked(swappage); 1282 page_cache_release(swappage); 1283 goto repeat; 1284 } 1285 if (PageWriteback(swappage)) { 1286 shmem_swp_unmap(entry); 1287 spin_unlock(&info->lock); 1288 wait_on_page_writeback(swappage); 1289 unlock_page(swappage); 1290 page_cache_release(swappage); 1291 goto repeat; 1292 } 1293 if (!PageUptodate(swappage)) { 1294 shmem_swp_unmap(entry); 1295 spin_unlock(&info->lock); 1296 unlock_page(swappage); 1297 page_cache_release(swappage); 1298 error = -EIO; 1299 goto failed; 1300 } 1301 1302 if (filepage) { 1303 shmem_swp_set(info, entry, 0); 1304 shmem_swp_unmap(entry); 1305 delete_from_swap_cache(swappage); 1306 spin_unlock(&info->lock); 1307 copy_highpage(filepage, swappage); 1308 unlock_page(swappage); 1309 page_cache_release(swappage); 1310 flush_dcache_page(filepage); 1311 SetPageUptodate(filepage); 1312 set_page_dirty(filepage); 1313 swap_free(swap); 1314 } else if (!(error = add_to_page_cache_locked(swappage, mapping, 1315 idx, GFP_NOWAIT))) { 1316 info->flags |= SHMEM_PAGEIN; 1317 shmem_swp_set(info, entry, 0); 1318 shmem_swp_unmap(entry); 1319 delete_from_swap_cache(swappage); 1320 spin_unlock(&info->lock); 1321 filepage = swappage; 1322 set_page_dirty(filepage); 1323 swap_free(swap); 1324 } else { 1325 shmem_swp_unmap(entry); 1326 spin_unlock(&info->lock); 1327 if (error == -ENOMEM) { 1328 /* allow reclaim from this memory cgroup */ 1329 error = mem_cgroup_shrink_usage(swappage, 1330 current->mm, 1331 gfp); 1332 if (error) { 1333 unlock_page(swappage); 1334 page_cache_release(swappage); 1335 goto failed; 1336 } 1337 } 1338 unlock_page(swappage); 1339 page_cache_release(swappage); 1340 goto repeat; 1341 } 1342 } else if (sgp == SGP_READ && !filepage) { 1343 shmem_swp_unmap(entry); 1344 filepage = find_get_page(mapping, idx); 1345 if (filepage && 1346 (!PageUptodate(filepage) || !trylock_page(filepage))) { 1347 spin_unlock(&info->lock); 1348 wait_on_page_locked(filepage); 1349 page_cache_release(filepage); 1350 filepage = NULL; 1351 goto repeat; 1352 } 1353 spin_unlock(&info->lock); 1354 } else { 1355 shmem_swp_unmap(entry); 1356 sbinfo = SHMEM_SB(inode->i_sb); 1357 if (sbinfo->max_blocks) { 1358 spin_lock(&sbinfo->stat_lock); 1359 if (sbinfo->free_blocks == 0 || 1360 shmem_acct_block(info->flags)) { 1361 spin_unlock(&sbinfo->stat_lock); 1362 spin_unlock(&info->lock); 1363 error = -ENOSPC; 1364 goto failed; 1365 } 1366 sbinfo->free_blocks--; 1367 inode->i_blocks += BLOCKS_PER_PAGE; 1368 spin_unlock(&sbinfo->stat_lock); 1369 } else if (shmem_acct_block(info->flags)) { 1370 spin_unlock(&info->lock); 1371 error = -ENOSPC; 1372 goto failed; 1373 } 1374 1375 if (!filepage) { 1376 int ret; 1377 1378 spin_unlock(&info->lock); 1379 filepage = shmem_alloc_page(gfp, info, idx); 1380 if (!filepage) { 1381 shmem_unacct_blocks(info->flags, 1); 1382 shmem_free_blocks(inode, 1); 1383 error = -ENOMEM; 1384 goto failed; 1385 } 1386 SetPageSwapBacked(filepage); 1387 1388 /* Precharge page while we can wait, compensate after */ 1389 error = mem_cgroup_cache_charge(filepage, current->mm, 1390 GFP_KERNEL); 1391 if (error) { 1392 page_cache_release(filepage); 1393 shmem_unacct_blocks(info->flags, 1); 1394 shmem_free_blocks(inode, 1); 1395 filepage = NULL; 1396 goto failed; 1397 } 1398 1399 spin_lock(&info->lock); 1400 entry = shmem_swp_alloc(info, idx, sgp); 1401 if (IS_ERR(entry)) 1402 error = PTR_ERR(entry); 1403 else { 1404 swap = *entry; 1405 shmem_swp_unmap(entry); 1406 } 1407 ret = error || swap.val; 1408 if (ret) 1409 mem_cgroup_uncharge_cache_page(filepage); 1410 else 1411 ret = add_to_page_cache_lru(filepage, mapping, 1412 idx, GFP_NOWAIT); 1413 /* 1414 * At add_to_page_cache_lru() failure, uncharge will 1415 * be done automatically. 1416 */ 1417 if (ret) { 1418 spin_unlock(&info->lock); 1419 page_cache_release(filepage); 1420 shmem_unacct_blocks(info->flags, 1); 1421 shmem_free_blocks(inode, 1); 1422 filepage = NULL; 1423 if (error) 1424 goto failed; 1425 goto repeat; 1426 } 1427 info->flags |= SHMEM_PAGEIN; 1428 } 1429 1430 info->alloced++; 1431 spin_unlock(&info->lock); 1432 clear_highpage(filepage); 1433 flush_dcache_page(filepage); 1434 SetPageUptodate(filepage); 1435 if (sgp == SGP_DIRTY) 1436 set_page_dirty(filepage); 1437 } 1438done: 1439 *pagep = filepage; 1440 return 0; 1441 1442failed: 1443 if (*pagep != filepage) { 1444 unlock_page(filepage); 1445 page_cache_release(filepage); 1446 } 1447 return error; 1448} 1449 1450static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1451{ 1452 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 1453 int error; 1454 int ret; 1455 1456 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 1457 return VM_FAULT_SIGBUS; 1458 1459 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret); 1460 if (error) 1461 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); 1462 1463 return ret | VM_FAULT_LOCKED; 1464} 1465 1466#ifdef CONFIG_NUMA 1467static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new) 1468{ 1469 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1470 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new); 1471} 1472 1473static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, 1474 unsigned long addr) 1475{ 1476 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1477 unsigned long idx; 1478 1479 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; 1480 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx); 1481} 1482#endif 1483 1484int shmem_lock(struct file *file, int lock, struct user_struct *user) 1485{ 1486 struct inode *inode = file->f_path.dentry->d_inode; 1487 struct shmem_inode_info *info = SHMEM_I(inode); 1488 int retval = -ENOMEM; 1489 1490 spin_lock(&info->lock); 1491 if (lock && !(info->flags & VM_LOCKED)) { 1492 if (!user_shm_lock(inode->i_size, user)) 1493 goto out_nomem; 1494 info->flags |= VM_LOCKED; 1495 mapping_set_unevictable(file->f_mapping); 1496 } 1497 if (!lock && (info->flags & VM_LOCKED) && user) { 1498 user_shm_unlock(inode->i_size, user); 1499 info->flags &= ~VM_LOCKED; 1500 mapping_clear_unevictable(file->f_mapping); 1501 scan_mapping_unevictable_pages(file->f_mapping); 1502 } 1503 retval = 0; 1504 1505out_nomem: 1506 spin_unlock(&info->lock); 1507 return retval; 1508} 1509 1510static int shmem_mmap(struct file *file, struct vm_area_struct *vma) 1511{ 1512 file_accessed(file); 1513 vma->vm_ops = &shmem_vm_ops; 1514 vma->vm_flags |= VM_CAN_NONLINEAR; 1515 return 0; 1516} 1517 1518static struct inode *shmem_get_inode(struct super_block *sb, int mode, 1519 dev_t dev, unsigned long flags) 1520{ 1521 struct inode *inode; 1522 struct shmem_inode_info *info; 1523 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 1524 1525 if (shmem_reserve_inode(sb)) 1526 return NULL; 1527 1528 inode = new_inode(sb); 1529 if (inode) { 1530 inode->i_mode = mode; 1531 inode->i_uid = current_fsuid(); 1532 inode->i_gid = current_fsgid(); 1533 inode->i_blocks = 0; 1534 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; 1535 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 1536 inode->i_generation = get_seconds(); 1537 info = SHMEM_I(inode); 1538 memset(info, 0, (char *)inode - (char *)info); 1539 spin_lock_init(&info->lock); 1540 info->flags = flags & VM_NORESERVE; 1541 INIT_LIST_HEAD(&info->swaplist); 1542 1543 switch (mode & S_IFMT) { 1544 default: 1545 inode->i_op = &shmem_special_inode_operations; 1546 init_special_inode(inode, mode, dev); 1547 break; 1548 case S_IFREG: 1549 inode->i_mapping->a_ops = &shmem_aops; 1550 inode->i_op = &shmem_inode_operations; 1551 inode->i_fop = &shmem_file_operations; 1552 mpol_shared_policy_init(&info->policy, 1553 shmem_get_sbmpol(sbinfo)); 1554 break; 1555 case S_IFDIR: 1556 inc_nlink(inode); 1557 /* Some things misbehave if size == 0 on a directory */ 1558 inode->i_size = 2 * BOGO_DIRENT_SIZE; 1559 inode->i_op = &shmem_dir_inode_operations; 1560 inode->i_fop = &simple_dir_operations; 1561 break; 1562 case S_IFLNK: 1563 /* 1564 * Must not load anything in the rbtree, 1565 * mpol_free_shared_policy will not be called. 1566 */ 1567 mpol_shared_policy_init(&info->policy, NULL); 1568 break; 1569 } 1570 } else 1571 shmem_free_inode(sb); 1572 return inode; 1573} 1574 1575#ifdef CONFIG_TMPFS 1576static const struct inode_operations shmem_symlink_inode_operations; 1577static const struct inode_operations shmem_symlink_inline_operations; 1578 1579/* 1580 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin; 1581 * but providing them allows a tmpfs file to be used for splice, sendfile, and 1582 * below the loop driver, in the generic fashion that many filesystems support. 1583 */ 1584static int shmem_readpage(struct file *file, struct page *page) 1585{ 1586 struct inode *inode = page->mapping->host; 1587 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL); 1588 unlock_page(page); 1589 return error; 1590} 1591 1592static int 1593shmem_write_begin(struct file *file, struct address_space *mapping, 1594 loff_t pos, unsigned len, unsigned flags, 1595 struct page **pagep, void **fsdata) 1596{ 1597 struct inode *inode = mapping->host; 1598 pgoff_t index = pos >> PAGE_CACHE_SHIFT; 1599 *pagep = NULL; 1600 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL); 1601} 1602 1603static int 1604shmem_write_end(struct file *file, struct address_space *mapping, 1605 loff_t pos, unsigned len, unsigned copied, 1606 struct page *page, void *fsdata) 1607{ 1608 struct inode *inode = mapping->host; 1609 1610 if (pos + copied > inode->i_size) 1611 i_size_write(inode, pos + copied); 1612 1613 unlock_page(page); 1614 set_page_dirty(page); 1615 page_cache_release(page); 1616 1617 return copied; 1618} 1619 1620static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor) 1621{ 1622 struct inode *inode = filp->f_path.dentry->d_inode; 1623 struct address_space *mapping = inode->i_mapping; 1624 unsigned long index, offset; 1625 enum sgp_type sgp = SGP_READ; 1626 1627 /* 1628 * Might this read be for a stacking filesystem? Then when reading 1629 * holes of a sparse file, we actually need to allocate those pages, 1630 * and even mark them dirty, so it cannot exceed the max_blocks limit. 1631 */ 1632 if (segment_eq(get_fs(), KERNEL_DS)) 1633 sgp = SGP_DIRTY; 1634 1635 index = *ppos >> PAGE_CACHE_SHIFT; 1636 offset = *ppos & ~PAGE_CACHE_MASK; 1637 1638 for (;;) { 1639 struct page *page = NULL; 1640 unsigned long end_index, nr, ret; 1641 loff_t i_size = i_size_read(inode); 1642 1643 end_index = i_size >> PAGE_CACHE_SHIFT; 1644 if (index > end_index) 1645 break; 1646 if (index == end_index) { 1647 nr = i_size & ~PAGE_CACHE_MASK; 1648 if (nr <= offset) 1649 break; 1650 } 1651 1652 desc->error = shmem_getpage(inode, index, &page, sgp, NULL); 1653 if (desc->error) { 1654 if (desc->error == -EINVAL) 1655 desc->error = 0; 1656 break; 1657 } 1658 if (page) 1659 unlock_page(page); 1660 1661 /* 1662 * We must evaluate after, since reads (unlike writes) 1663 * are called without i_mutex protection against truncate 1664 */ 1665 nr = PAGE_CACHE_SIZE; 1666 i_size = i_size_read(inode); 1667 end_index = i_size >> PAGE_CACHE_SHIFT; 1668 if (index == end_index) { 1669 nr = i_size & ~PAGE_CACHE_MASK; 1670 if (nr <= offset) { 1671 if (page) 1672 page_cache_release(page); 1673 break; 1674 } 1675 } 1676 nr -= offset; 1677 1678 if (page) { 1679 /* 1680 * If users can be writing to this page using arbitrary 1681 * virtual addresses, take care about potential aliasing 1682 * before reading the page on the kernel side. 1683 */ 1684 if (mapping_writably_mapped(mapping)) 1685 flush_dcache_page(page); 1686 /* 1687 * Mark the page accessed if we read the beginning. 1688 */ 1689 if (!offset) 1690 mark_page_accessed(page); 1691 } else { 1692 page = ZERO_PAGE(0); 1693 page_cache_get(page); 1694 } 1695 1696 /* 1697 * Ok, we have the page, and it's up-to-date, so 1698 * now we can copy it to user space... 1699 * 1700 * The actor routine returns how many bytes were actually used.. 1701 * NOTE! This may not be the same as how much of a user buffer 1702 * we filled up (we may be padding etc), so we can only update 1703 * "pos" here (the actor routine has to update the user buffer 1704 * pointers and the remaining count). 1705 */ 1706 ret = actor(desc, page, offset, nr); 1707 offset += ret; 1708 index += offset >> PAGE_CACHE_SHIFT; 1709 offset &= ~PAGE_CACHE_MASK; 1710 1711 page_cache_release(page); 1712 if (ret != nr || !desc->count) 1713 break; 1714 1715 cond_resched(); 1716 } 1717 1718 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; 1719 file_accessed(filp); 1720} 1721 1722static ssize_t shmem_file_aio_read(struct kiocb *iocb, 1723 const struct iovec *iov, unsigned long nr_segs, loff_t pos) 1724{ 1725 struct file *filp = iocb->ki_filp; 1726 ssize_t retval; 1727 unsigned long seg; 1728 size_t count; 1729 loff_t *ppos = &iocb->ki_pos; 1730 1731 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); 1732 if (retval) 1733 return retval; 1734 1735 for (seg = 0; seg < nr_segs; seg++) { 1736 read_descriptor_t desc; 1737 1738 desc.written = 0; 1739 desc.arg.buf = iov[seg].iov_base; 1740 desc.count = iov[seg].iov_len; 1741 if (desc.count == 0) 1742 continue; 1743 desc.error = 0; 1744 do_shmem_file_read(filp, ppos, &desc, file_read_actor); 1745 retval += desc.written; 1746 if (desc.error) { 1747 retval = retval ?: desc.error; 1748 break; 1749 } 1750 if (desc.count > 0) 1751 break; 1752 } 1753 return retval; 1754} 1755 1756static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) 1757{ 1758 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); 1759 1760 buf->f_type = TMPFS_MAGIC; 1761 buf->f_bsize = PAGE_CACHE_SIZE; 1762 buf->f_namelen = NAME_MAX; 1763 spin_lock(&sbinfo->stat_lock); 1764 if (sbinfo->max_blocks) { 1765 buf->f_blocks = sbinfo->max_blocks; 1766 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks; 1767 } 1768 if (sbinfo->max_inodes) { 1769 buf->f_files = sbinfo->max_inodes; 1770 buf->f_ffree = sbinfo->free_inodes; 1771 } 1772 /* else leave those fields 0 like simple_statfs */ 1773 spin_unlock(&sbinfo->stat_lock); 1774 return 0; 1775} 1776 1777/* 1778 * File creation. Allocate an inode, and we're done.. 1779 */ 1780static int 1781shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 1782{ 1783 struct inode *inode; 1784 int error = -ENOSPC; 1785 1786 inode = shmem_get_inode(dir->i_sb, mode, dev, VM_NORESERVE); 1787 if (inode) { 1788 error = security_inode_init_security(inode, dir, NULL, NULL, 1789 NULL); 1790 if (error) { 1791 if (error != -EOPNOTSUPP) { 1792 iput(inode); 1793 return error; 1794 } 1795 } 1796 error = shmem_acl_init(inode, dir); 1797 if (error) { 1798 iput(inode); 1799 return error; 1800 } 1801 if (dir->i_mode & S_ISGID) { 1802 inode->i_gid = dir->i_gid; 1803 if (S_ISDIR(mode)) 1804 inode->i_mode |= S_ISGID; 1805 } 1806 dir->i_size += BOGO_DIRENT_SIZE; 1807 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1808 d_instantiate(dentry, inode); 1809 dget(dentry); /* Extra count - pin the dentry in core */ 1810 } 1811 return error; 1812} 1813 1814static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1815{ 1816 int error; 1817 1818 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) 1819 return error; 1820 inc_nlink(dir); 1821 return 0; 1822} 1823 1824static int shmem_create(struct inode *dir, struct dentry *dentry, int mode, 1825 struct nameidata *nd) 1826{ 1827 return shmem_mknod(dir, dentry, mode | S_IFREG, 0); 1828} 1829 1830/* 1831 * Link a file.. 1832 */ 1833static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1834{ 1835 struct inode *inode = old_dentry->d_inode; 1836 int ret; 1837 1838 /* 1839 * No ordinary (disk based) filesystem counts links as inodes; 1840 * but each new link needs a new dentry, pinning lowmem, and 1841 * tmpfs dentries cannot be pruned until they are unlinked. 1842 */ 1843 ret = shmem_reserve_inode(inode->i_sb); 1844 if (ret) 1845 goto out; 1846 1847 dir->i_size += BOGO_DIRENT_SIZE; 1848 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1849 inc_nlink(inode); 1850 atomic_inc(&inode->i_count); /* New dentry reference */ 1851 dget(dentry); /* Extra pinning count for the created dentry */ 1852 d_instantiate(dentry, inode); 1853out: 1854 return ret; 1855} 1856 1857static int shmem_unlink(struct inode *dir, struct dentry *dentry) 1858{ 1859 struct inode *inode = dentry->d_inode; 1860 1861 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) 1862 shmem_free_inode(inode->i_sb); 1863 1864 dir->i_size -= BOGO_DIRENT_SIZE; 1865 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1866 drop_nlink(inode); 1867 dput(dentry); /* Undo the count from "create" - this does all the work */ 1868 return 0; 1869} 1870 1871static int shmem_rmdir(struct inode *dir, struct dentry *dentry) 1872{ 1873 if (!simple_empty(dentry)) 1874 return -ENOTEMPTY; 1875 1876 drop_nlink(dentry->d_inode); 1877 drop_nlink(dir); 1878 return shmem_unlink(dir, dentry); 1879} 1880 1881/* 1882 * The VFS layer already does all the dentry stuff for rename, 1883 * we just have to decrement the usage count for the target if 1884 * it exists so that the VFS layer correctly free's it when it 1885 * gets overwritten. 1886 */ 1887static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) 1888{ 1889 struct inode *inode = old_dentry->d_inode; 1890 int they_are_dirs = S_ISDIR(inode->i_mode); 1891 1892 if (!simple_empty(new_dentry)) 1893 return -ENOTEMPTY; 1894 1895 if (new_dentry->d_inode) { 1896 (void) shmem_unlink(new_dir, new_dentry); 1897 if (they_are_dirs) 1898 drop_nlink(old_dir); 1899 } else if (they_are_dirs) { 1900 drop_nlink(old_dir); 1901 inc_nlink(new_dir); 1902 } 1903 1904 old_dir->i_size -= BOGO_DIRENT_SIZE; 1905 new_dir->i_size += BOGO_DIRENT_SIZE; 1906 old_dir->i_ctime = old_dir->i_mtime = 1907 new_dir->i_ctime = new_dir->i_mtime = 1908 inode->i_ctime = CURRENT_TIME; 1909 return 0; 1910} 1911 1912static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1913{ 1914 int error; 1915 int len; 1916 struct inode *inode; 1917 struct page *page = NULL; 1918 char *kaddr; 1919 struct shmem_inode_info *info; 1920 1921 len = strlen(symname) + 1; 1922 if (len > PAGE_CACHE_SIZE) 1923 return -ENAMETOOLONG; 1924 1925 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE); 1926 if (!inode) 1927 return -ENOSPC; 1928 1929 error = security_inode_init_security(inode, dir, NULL, NULL, 1930 NULL); 1931 if (error) { 1932 if (error != -EOPNOTSUPP) { 1933 iput(inode); 1934 return error; 1935 } 1936 error = 0; 1937 } 1938 1939 info = SHMEM_I(inode); 1940 inode->i_size = len-1; 1941 if (len <= (char *)inode - (char *)info) { 1942 /* do it inline */ 1943 memcpy(info, symname, len); 1944 inode->i_op = &shmem_symlink_inline_operations; 1945 } else { 1946 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); 1947 if (error) { 1948 iput(inode); 1949 return error; 1950 } 1951 unlock_page(page); 1952 inode->i_mapping->a_ops = &shmem_aops; 1953 inode->i_op = &shmem_symlink_inode_operations; 1954 kaddr = kmap_atomic(page, KM_USER0); 1955 memcpy(kaddr, symname, len); 1956 kunmap_atomic(kaddr, KM_USER0); 1957 set_page_dirty(page); 1958 page_cache_release(page); 1959 } 1960 if (dir->i_mode & S_ISGID) 1961 inode->i_gid = dir->i_gid; 1962 dir->i_size += BOGO_DIRENT_SIZE; 1963 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1964 d_instantiate(dentry, inode); 1965 dget(dentry); 1966 return 0; 1967} 1968 1969static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd) 1970{ 1971 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode)); 1972 return NULL; 1973} 1974 1975static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) 1976{ 1977 struct page *page = NULL; 1978 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); 1979 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page)); 1980 if (page) 1981 unlock_page(page); 1982 return page; 1983} 1984 1985static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 1986{ 1987 if (!IS_ERR(nd_get_link(nd))) { 1988 struct page *page = cookie; 1989 kunmap(page); 1990 mark_page_accessed(page); 1991 page_cache_release(page); 1992 } 1993} 1994 1995static const struct inode_operations shmem_symlink_inline_operations = { 1996 .readlink = generic_readlink, 1997 .follow_link = shmem_follow_link_inline, 1998}; 1999 2000static const struct inode_operations shmem_symlink_inode_operations = { 2001 .truncate = shmem_truncate, 2002 .readlink = generic_readlink, 2003 .follow_link = shmem_follow_link, 2004 .put_link = shmem_put_link, 2005}; 2006 2007#ifdef CONFIG_TMPFS_POSIX_ACL 2008/* 2009 * Superblocks without xattr inode operations will get security.* xattr 2010 * support from the VFS "for free". As soon as we have any other xattrs 2011 * like ACLs, we also need to implement the security.* handlers at 2012 * filesystem level, though. 2013 */ 2014 2015static size_t shmem_xattr_security_list(struct inode *inode, char *list, 2016 size_t list_len, const char *name, 2017 size_t name_len) 2018{ 2019 return security_inode_listsecurity(inode, list, list_len); 2020} 2021 2022static int shmem_xattr_security_get(struct inode *inode, const char *name, 2023 void *buffer, size_t size) 2024{ 2025 if (strcmp(name, "") == 0) 2026 return -EINVAL; 2027 return xattr_getsecurity(inode, name, buffer, size); 2028} 2029 2030static int shmem_xattr_security_set(struct inode *inode, const char *name, 2031 const void *value, size_t size, int flags) 2032{ 2033 if (strcmp(name, "") == 0) 2034 return -EINVAL; 2035 return security_inode_setsecurity(inode, name, value, size, flags); 2036} 2037 2038static struct xattr_handler shmem_xattr_security_handler = { 2039 .prefix = XATTR_SECURITY_PREFIX, 2040 .list = shmem_xattr_security_list, 2041 .get = shmem_xattr_security_get, 2042 .set = shmem_xattr_security_set, 2043}; 2044 2045static struct xattr_handler *shmem_xattr_handlers[] = { 2046 &shmem_xattr_acl_access_handler, 2047 &shmem_xattr_acl_default_handler, 2048 &shmem_xattr_security_handler, 2049 NULL 2050}; 2051#endif 2052 2053static struct dentry *shmem_get_parent(struct dentry *child) 2054{ 2055 return ERR_PTR(-ESTALE); 2056} 2057 2058static int shmem_match(struct inode *ino, void *vfh) 2059{ 2060 __u32 *fh = vfh; 2061 __u64 inum = fh[2]; 2062 inum = (inum << 32) | fh[1]; 2063 return ino->i_ino == inum && fh[0] == ino->i_generation; 2064} 2065 2066static struct dentry *shmem_fh_to_dentry(struct super_block *sb, 2067 struct fid *fid, int fh_len, int fh_type) 2068{ 2069 struct inode *inode; 2070 struct dentry *dentry = NULL; 2071 u64 inum = fid->raw[2]; 2072 inum = (inum << 32) | fid->raw[1]; 2073 2074 if (fh_len < 3) 2075 return NULL; 2076 2077 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), 2078 shmem_match, fid->raw); 2079 if (inode) { 2080 dentry = d_find_alias(inode); 2081 iput(inode); 2082 } 2083 2084 return dentry; 2085} 2086 2087static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len, 2088 int connectable) 2089{ 2090 struct inode *inode = dentry->d_inode; 2091 2092 if (*len < 3) 2093 return 255; 2094 2095 if (hlist_unhashed(&inode->i_hash)) { 2096 /* Unfortunately insert_inode_hash is not idempotent, 2097 * so as we hash inodes here rather than at creation 2098 * time, we need a lock to ensure we only try 2099 * to do it once 2100 */ 2101 static DEFINE_SPINLOCK(lock); 2102 spin_lock(&lock); 2103 if (hlist_unhashed(&inode->i_hash)) 2104 __insert_inode_hash(inode, 2105 inode->i_ino + inode->i_generation); 2106 spin_unlock(&lock); 2107 } 2108 2109 fh[0] = inode->i_generation; 2110 fh[1] = inode->i_ino; 2111 fh[2] = ((__u64)inode->i_ino) >> 32; 2112 2113 *len = 3; 2114 return 1; 2115} 2116 2117static const struct export_operations shmem_export_ops = { 2118 .get_parent = shmem_get_parent, 2119 .encode_fh = shmem_encode_fh, 2120 .fh_to_dentry = shmem_fh_to_dentry, 2121}; 2122 2123static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, 2124 bool remount) 2125{ 2126 char *this_char, *value, *rest; 2127 2128 while (options != NULL) { 2129 this_char = options; 2130 for (;;) { 2131 /* 2132 * NUL-terminate this option: unfortunately, 2133 * mount options form a comma-separated list, 2134 * but mpol's nodelist may also contain commas. 2135 */ 2136 options = strchr(options, ','); 2137 if (options == NULL) 2138 break; 2139 options++; 2140 if (!isdigit(*options)) { 2141 options[-1] = '\0'; 2142 break; 2143 } 2144 } 2145 if (!*this_char) 2146 continue; 2147 if ((value = strchr(this_char,'=')) != NULL) { 2148 *value++ = 0; 2149 } else { 2150 printk(KERN_ERR 2151 "tmpfs: No value for mount option '%s'\n", 2152 this_char); 2153 return 1; 2154 } 2155 2156 if (!strcmp(this_char,"size")) { 2157 unsigned long long size; 2158 size = memparse(value,&rest); 2159 if (*rest == '%') { 2160 size <<= PAGE_SHIFT; 2161 size *= totalram_pages; 2162 do_div(size, 100); 2163 rest++; 2164 } 2165 if (*rest) 2166 goto bad_val; 2167 sbinfo->max_blocks = 2168 DIV_ROUND_UP(size, PAGE_CACHE_SIZE); 2169 } else if (!strcmp(this_char,"nr_blocks")) { 2170 sbinfo->max_blocks = memparse(value, &rest); 2171 if (*rest) 2172 goto bad_val; 2173 } else if (!strcmp(this_char,"nr_inodes")) { 2174 sbinfo->max_inodes = memparse(value, &rest); 2175 if (*rest) 2176 goto bad_val; 2177 } else if (!strcmp(this_char,"mode")) { 2178 if (remount) 2179 continue; 2180 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; 2181 if (*rest) 2182 goto bad_val; 2183 } else if (!strcmp(this_char,"uid")) { 2184 if (remount) 2185 continue; 2186 sbinfo->uid = simple_strtoul(value, &rest, 0); 2187 if (*rest) 2188 goto bad_val; 2189 } else if (!strcmp(this_char,"gid")) { 2190 if (remount) 2191 continue; 2192 sbinfo->gid = simple_strtoul(value, &rest, 0); 2193 if (*rest) 2194 goto bad_val; 2195 } else if (!strcmp(this_char,"mpol")) { 2196 if (mpol_parse_str(value, &sbinfo->mpol, 1)) 2197 goto bad_val; 2198 } else { 2199 printk(KERN_ERR "tmpfs: Bad mount option %s\n", 2200 this_char); 2201 return 1; 2202 } 2203 } 2204 return 0; 2205 2206bad_val: 2207 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", 2208 value, this_char); 2209 return 1; 2210 2211} 2212 2213static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) 2214{ 2215 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 2216 struct shmem_sb_info config = *sbinfo; 2217 unsigned long blocks; 2218 unsigned long inodes; 2219 int error = -EINVAL; 2220 2221 if (shmem_parse_options(data, &config, true)) 2222 return error; 2223 2224 spin_lock(&sbinfo->stat_lock); 2225 blocks = sbinfo->max_blocks - sbinfo->free_blocks; 2226 inodes = sbinfo->max_inodes - sbinfo->free_inodes; 2227 if (config.max_blocks < blocks) 2228 goto out; 2229 if (config.max_inodes < inodes) 2230 goto out; 2231 /* 2232 * Those tests also disallow limited->unlimited while any are in 2233 * use, so i_blocks will always be zero when max_blocks is zero; 2234 * but we must separately disallow unlimited->limited, because 2235 * in that case we have no record of how much is already in use. 2236 */ 2237 if (config.max_blocks && !sbinfo->max_blocks) 2238 goto out; 2239 if (config.max_inodes && !sbinfo->max_inodes) 2240 goto out; 2241 2242 error = 0; 2243 sbinfo->max_blocks = config.max_blocks; 2244 sbinfo->free_blocks = config.max_blocks - blocks; 2245 sbinfo->max_inodes = config.max_inodes; 2246 sbinfo->free_inodes = config.max_inodes - inodes; 2247 2248 mpol_put(sbinfo->mpol); 2249 sbinfo->mpol = config.mpol; /* transfers initial ref */ 2250out: 2251 spin_unlock(&sbinfo->stat_lock); 2252 return error; 2253} 2254 2255static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs) 2256{ 2257 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb); 2258 2259 if (sbinfo->max_blocks != shmem_default_max_blocks()) 2260 seq_printf(seq, ",size=%luk", 2261 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10)); 2262 if (sbinfo->max_inodes != shmem_default_max_inodes()) 2263 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); 2264 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) 2265 seq_printf(seq, ",mode=%03o", sbinfo->mode); 2266 if (sbinfo->uid != 0) 2267 seq_printf(seq, ",uid=%u", sbinfo->uid); 2268 if (sbinfo->gid != 0) 2269 seq_printf(seq, ",gid=%u", sbinfo->gid); 2270 shmem_show_mpol(seq, sbinfo->mpol); 2271 return 0; 2272} 2273#endif /* CONFIG_TMPFS */ 2274 2275static void shmem_put_super(struct super_block *sb) 2276{ 2277 kfree(sb->s_fs_info); 2278 sb->s_fs_info = NULL; 2279} 2280 2281static int shmem_fill_super(struct super_block *sb, 2282 void *data, int silent) 2283{ 2284 struct inode *inode; 2285 struct dentry *root; 2286 struct shmem_sb_info *sbinfo; 2287 int err = -ENOMEM; 2288 2289 /* Round up to L1_CACHE_BYTES to resist false sharing */ 2290 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info), 2291 L1_CACHE_BYTES), GFP_KERNEL); 2292 if (!sbinfo) 2293 return -ENOMEM; 2294 2295 sbinfo->max_blocks = 0; 2296 sbinfo->max_inodes = 0; 2297 sbinfo->mode = S_IRWXUGO | S_ISVTX; 2298 sbinfo->uid = current_fsuid(); 2299 sbinfo->gid = current_fsgid(); 2300 sbinfo->mpol = NULL; 2301 sb->s_fs_info = sbinfo; 2302 2303#ifdef CONFIG_TMPFS 2304 /* 2305 * Per default we only allow half of the physical ram per 2306 * tmpfs instance, limiting inodes to one per page of lowmem; 2307 * but the internal instance is left unlimited. 2308 */ 2309 if (!(sb->s_flags & MS_NOUSER)) { 2310 sbinfo->max_blocks = shmem_default_max_blocks(); 2311 sbinfo->max_inodes = shmem_default_max_inodes(); 2312 if (shmem_parse_options(data, sbinfo, false)) { 2313 err = -EINVAL; 2314 goto failed; 2315 } 2316 } 2317 sb->s_export_op = &shmem_export_ops; 2318#else 2319 sb->s_flags |= MS_NOUSER; 2320#endif 2321 2322 spin_lock_init(&sbinfo->stat_lock); 2323 sbinfo->free_blocks = sbinfo->max_blocks; 2324 sbinfo->free_inodes = sbinfo->max_inodes; 2325 2326 sb->s_maxbytes = SHMEM_MAX_BYTES; 2327 sb->s_blocksize = PAGE_CACHE_SIZE; 2328 sb->s_blocksize_bits = PAGE_CACHE_SHIFT; 2329 sb->s_magic = TMPFS_MAGIC; 2330 sb->s_op = &shmem_ops; 2331 sb->s_time_gran = 1; 2332#ifdef CONFIG_TMPFS_POSIX_ACL 2333 sb->s_xattr = shmem_xattr_handlers; 2334 sb->s_flags |= MS_POSIXACL; 2335#endif 2336 2337 inode = shmem_get_inode(sb, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); 2338 if (!inode) 2339 goto failed; 2340 inode->i_uid = sbinfo->uid; 2341 inode->i_gid = sbinfo->gid; 2342 root = d_alloc_root(inode); 2343 if (!root) 2344 goto failed_iput; 2345 sb->s_root = root; 2346 return 0; 2347 2348failed_iput: 2349 iput(inode); 2350failed: 2351 shmem_put_super(sb); 2352 return err; 2353} 2354 2355static struct kmem_cache *shmem_inode_cachep; 2356 2357static struct inode *shmem_alloc_inode(struct super_block *sb) 2358{ 2359 struct shmem_inode_info *p; 2360 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); 2361 if (!p) 2362 return NULL; 2363 return &p->vfs_inode; 2364} 2365 2366static void shmem_destroy_inode(struct inode *inode) 2367{ 2368 if ((inode->i_mode & S_IFMT) == S_IFREG) { 2369 /* only struct inode is valid if it's an inline symlink */ 2370 mpol_free_shared_policy(&SHMEM_I(inode)->policy); 2371 } 2372 shmem_acl_destroy_inode(inode); 2373 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); 2374} 2375 2376static void init_once(void *foo) 2377{ 2378 struct shmem_inode_info *p = (struct shmem_inode_info *) foo; 2379 2380 inode_init_once(&p->vfs_inode); 2381#ifdef CONFIG_TMPFS_POSIX_ACL 2382 p->i_acl = NULL; 2383 p->i_default_acl = NULL; 2384#endif 2385} 2386 2387static int init_inodecache(void) 2388{ 2389 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", 2390 sizeof(struct shmem_inode_info), 2391 0, SLAB_PANIC, init_once); 2392 return 0; 2393} 2394 2395static void destroy_inodecache(void) 2396{ 2397 kmem_cache_destroy(shmem_inode_cachep); 2398} 2399 2400static const struct address_space_operations shmem_aops = { 2401 .writepage = shmem_writepage, 2402 .set_page_dirty = __set_page_dirty_no_writeback, 2403#ifdef CONFIG_TMPFS 2404 .readpage = shmem_readpage, 2405 .write_begin = shmem_write_begin, 2406 .write_end = shmem_write_end, 2407#endif 2408 .migratepage = migrate_page, 2409}; 2410 2411static const struct file_operations shmem_file_operations = { 2412 .mmap = shmem_mmap, 2413#ifdef CONFIG_TMPFS 2414 .llseek = generic_file_llseek, 2415 .read = do_sync_read, 2416 .write = do_sync_write, 2417 .aio_read = shmem_file_aio_read, 2418 .aio_write = generic_file_aio_write, 2419 .fsync = simple_sync_file, 2420 .splice_read = generic_file_splice_read, 2421 .splice_write = generic_file_splice_write, 2422#endif 2423}; 2424 2425static const struct inode_operations shmem_inode_operations = { 2426 .truncate = shmem_truncate, 2427 .setattr = shmem_notify_change, 2428 .truncate_range = shmem_truncate_range, 2429#ifdef CONFIG_TMPFS_POSIX_ACL 2430 .setxattr = generic_setxattr, 2431 .getxattr = generic_getxattr, 2432 .listxattr = generic_listxattr, 2433 .removexattr = generic_removexattr, 2434 .permission = shmem_permission, 2435#endif 2436 2437}; 2438 2439static const struct inode_operations shmem_dir_inode_operations = { 2440#ifdef CONFIG_TMPFS 2441 .create = shmem_create, 2442 .lookup = simple_lookup, 2443 .link = shmem_link, 2444 .unlink = shmem_unlink, 2445 .symlink = shmem_symlink, 2446 .mkdir = shmem_mkdir, 2447 .rmdir = shmem_rmdir, 2448 .mknod = shmem_mknod, 2449 .rename = shmem_rename, 2450#endif 2451#ifdef CONFIG_TMPFS_POSIX_ACL 2452 .setattr = shmem_notify_change, 2453 .setxattr = generic_setxattr, 2454 .getxattr = generic_getxattr, 2455 .listxattr = generic_listxattr, 2456 .removexattr = generic_removexattr, 2457 .permission = shmem_permission, 2458#endif 2459}; 2460 2461static const struct inode_operations shmem_special_inode_operations = { 2462#ifdef CONFIG_TMPFS_POSIX_ACL 2463 .setattr = shmem_notify_change, 2464 .setxattr = generic_setxattr, 2465 .getxattr = generic_getxattr, 2466 .listxattr = generic_listxattr, 2467 .removexattr = generic_removexattr, 2468 .permission = shmem_permission, 2469#endif 2470}; 2471 2472static const struct super_operations shmem_ops = { 2473 .alloc_inode = shmem_alloc_inode, 2474 .destroy_inode = shmem_destroy_inode, 2475#ifdef CONFIG_TMPFS 2476 .statfs = shmem_statfs, 2477 .remount_fs = shmem_remount_fs, 2478 .show_options = shmem_show_options, 2479#endif 2480 .delete_inode = shmem_delete_inode, 2481 .drop_inode = generic_delete_inode, 2482 .put_super = shmem_put_super, 2483}; 2484 2485static struct vm_operations_struct shmem_vm_ops = { 2486 .fault = shmem_fault, 2487#ifdef CONFIG_NUMA 2488 .set_policy = shmem_set_policy, 2489 .get_policy = shmem_get_policy, 2490#endif 2491}; 2492 2493 2494static int shmem_get_sb(struct file_system_type *fs_type, 2495 int flags, const char *dev_name, void *data, struct vfsmount *mnt) 2496{ 2497 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt); 2498} 2499 2500static struct file_system_type tmpfs_fs_type = { 2501 .owner = THIS_MODULE, 2502 .name = "tmpfs", 2503 .get_sb = shmem_get_sb, 2504 .kill_sb = kill_litter_super, 2505}; 2506 2507static int __init init_tmpfs(void) 2508{ 2509 int error; 2510 2511 error = bdi_init(&shmem_backing_dev_info); 2512 if (error) 2513 goto out4; 2514 2515 error = init_inodecache(); 2516 if (error) 2517 goto out3; 2518 2519 error = register_filesystem(&tmpfs_fs_type); 2520 if (error) { 2521 printk(KERN_ERR "Could not register tmpfs\n"); 2522 goto out2; 2523 } 2524 2525 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER, 2526 tmpfs_fs_type.name, NULL); 2527 if (IS_ERR(shm_mnt)) { 2528 error = PTR_ERR(shm_mnt); 2529 printk(KERN_ERR "Could not kern_mount tmpfs\n"); 2530 goto out1; 2531 } 2532 return 0; 2533 2534out1: 2535 unregister_filesystem(&tmpfs_fs_type); 2536out2: 2537 destroy_inodecache(); 2538out3: 2539 bdi_destroy(&shmem_backing_dev_info); 2540out4: 2541 shm_mnt = ERR_PTR(error); 2542 return error; 2543} 2544 2545#else /* !CONFIG_SHMEM */ 2546 2547/* 2548 * tiny-shmem: simple shmemfs and tmpfs using ramfs code 2549 * 2550 * This is intended for small system where the benefits of the full 2551 * shmem code (swap-backed and resource-limited) are outweighed by 2552 * their complexity. On systems without swap this code should be 2553 * effectively equivalent, but much lighter weight. 2554 */ 2555 2556#include <linux/ramfs.h> 2557 2558static struct file_system_type tmpfs_fs_type = { 2559 .name = "tmpfs", 2560 .get_sb = ramfs_get_sb, 2561 .kill_sb = kill_litter_super, 2562}; 2563 2564static int __init init_tmpfs(void) 2565{ 2566 BUG_ON(register_filesystem(&tmpfs_fs_type) != 0); 2567 2568 shm_mnt = kern_mount(&tmpfs_fs_type); 2569 BUG_ON(IS_ERR(shm_mnt)); 2570 2571 return 0; 2572} 2573 2574int shmem_unuse(swp_entry_t entry, struct page *page) 2575{ 2576 return 0; 2577} 2578 2579#define shmem_vm_ops generic_file_vm_ops 2580#define shmem_file_operations ramfs_file_operations 2581#define shmem_get_inode(sb, mode, dev, flags) ramfs_get_inode(sb, mode, dev) 2582#define shmem_acct_size(flags, size) 0 2583#define shmem_unacct_size(flags, size) do {} while (0) 2584#define SHMEM_MAX_BYTES LLONG_MAX 2585 2586#endif /* CONFIG_SHMEM */ 2587 2588/* common code */ 2589 2590/** 2591 * shmem_file_setup - get an unlinked file living in tmpfs 2592 * @name: name for dentry (to be seen in /proc/<pid>/maps 2593 * @size: size to be set for the file 2594 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size 2595 */ 2596struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags) 2597{ 2598 int error; 2599 struct file *file; 2600 struct inode *inode; 2601 struct dentry *dentry, *root; 2602 struct qstr this; 2603 2604 if (IS_ERR(shm_mnt)) 2605 return (void *)shm_mnt; 2606 2607 if (size < 0 || size > SHMEM_MAX_BYTES) 2608 return ERR_PTR(-EINVAL); 2609 2610 if (shmem_acct_size(flags, size)) 2611 return ERR_PTR(-ENOMEM); 2612 2613 error = -ENOMEM; 2614 this.name = name; 2615 this.len = strlen(name); 2616 this.hash = 0; /* will go */ 2617 root = shm_mnt->mnt_root; 2618 dentry = d_alloc(root, &this); 2619 if (!dentry) 2620 goto put_memory; 2621 2622 error = -ENFILE; 2623 file = get_empty_filp(); 2624 if (!file) 2625 goto put_dentry; 2626 2627 error = -ENOSPC; 2628 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0, flags); 2629 if (!inode) 2630 goto close_file; 2631 2632 d_instantiate(dentry, inode); 2633 inode->i_size = size; 2634 inode->i_nlink = 0; /* It is unlinked */ 2635 init_file(file, shm_mnt, dentry, FMODE_WRITE | FMODE_READ, 2636 &shmem_file_operations); 2637 2638#ifndef CONFIG_MMU 2639 error = ramfs_nommu_expand_for_mapping(inode, size); 2640 if (error) 2641 goto close_file; 2642#endif 2643 return file; 2644 2645close_file: 2646 put_filp(file); 2647put_dentry: 2648 dput(dentry); 2649put_memory: 2650 shmem_unacct_size(flags, size); 2651 return ERR_PTR(error); 2652} 2653EXPORT_SYMBOL_GPL(shmem_file_setup); 2654 2655/** 2656 * shmem_zero_setup - setup a shared anonymous mapping 2657 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff 2658 */ 2659int shmem_zero_setup(struct vm_area_struct *vma) 2660{ 2661 struct file *file; 2662 loff_t size = vma->vm_end - vma->vm_start; 2663 2664 file = shmem_file_setup("dev/zero", size, vma->vm_flags); 2665 if (IS_ERR(file)) 2666 return PTR_ERR(file); 2667 2668 ima_shm_check(file); 2669 if (vma->vm_file) 2670 fput(vma->vm_file); 2671 vma->vm_file = file; 2672 vma->vm_ops = &shmem_vm_ops; 2673 return 0; 2674} 2675 2676module_init(init_tmpfs) 2677