shmem.c revision b76db735407a26c1036fdfef249ddc35eb969bc4
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 * This file is released under the GPL. 18 */ 19 20/* 21 * This virtual memory filesystem is heavily based on the ramfs. It 22 * extends ramfs by the ability to use swap and honor resource limits 23 * which makes it a completely usable filesystem. 24 */ 25 26#include <linux/module.h> 27#include <linux/init.h> 28#include <linux/fs.h> 29#include <linux/xattr.h> 30#include <linux/exportfs.h> 31#include <linux/generic_acl.h> 32#include <linux/mm.h> 33#include <linux/mman.h> 34#include <linux/file.h> 35#include <linux/swap.h> 36#include <linux/pagemap.h> 37#include <linux/string.h> 38#include <linux/slab.h> 39#include <linux/backing-dev.h> 40#include <linux/shmem_fs.h> 41#include <linux/mount.h> 42#include <linux/writeback.h> 43#include <linux/vfs.h> 44#include <linux/blkdev.h> 45#include <linux/security.h> 46#include <linux/swapops.h> 47#include <linux/mempolicy.h> 48#include <linux/namei.h> 49#include <linux/ctype.h> 50#include <linux/migrate.h> 51#include <linux/highmem.h> 52#include <linux/seq_file.h> 53 54#include <asm/uaccess.h> 55#include <asm/div64.h> 56#include <asm/pgtable.h> 57 58/* This magic number is used in glibc for posix shared memory */ 59#define TMPFS_MAGIC 0x01021994 60 61#define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long)) 62#define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE) 63#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512) 64 65#define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1)) 66#define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT) 67 68#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT) 69 70/* info->flags needs VM_flags to handle pagein/truncate races efficiently */ 71#define SHMEM_PAGEIN VM_READ 72#define SHMEM_TRUNCATE VM_WRITE 73 74/* Definition to limit shmem_truncate's steps between cond_rescheds */ 75#define LATENCY_LIMIT 64 76 77/* Pretend that each entry is of this size in directory's i_size */ 78#define BOGO_DIRENT_SIZE 20 79 80/* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */ 81enum sgp_type { 82 SGP_READ, /* don't exceed i_size, don't allocate page */ 83 SGP_CACHE, /* don't exceed i_size, may allocate page */ 84 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */ 85 SGP_WRITE, /* may exceed i_size, may allocate page */ 86}; 87 88#ifdef CONFIG_TMPFS 89static unsigned long shmem_default_max_blocks(void) 90{ 91 return totalram_pages / 2; 92} 93 94static unsigned long shmem_default_max_inodes(void) 95{ 96 return min(totalram_pages - totalhigh_pages, totalram_pages / 2); 97} 98#endif 99 100static int shmem_getpage(struct inode *inode, unsigned long idx, 101 struct page **pagep, enum sgp_type sgp, int *type); 102 103static inline struct page *shmem_dir_alloc(gfp_t gfp_mask) 104{ 105 /* 106 * The above definition of ENTRIES_PER_PAGE, and the use of 107 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE: 108 * might be reconsidered if it ever diverges from PAGE_SIZE. 109 * 110 * Mobility flags are masked out as swap vectors cannot move 111 */ 112 return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO, 113 PAGE_CACHE_SHIFT-PAGE_SHIFT); 114} 115 116static inline void shmem_dir_free(struct page *page) 117{ 118 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT); 119} 120 121static struct page **shmem_dir_map(struct page *page) 122{ 123 return (struct page **)kmap_atomic(page, KM_USER0); 124} 125 126static inline void shmem_dir_unmap(struct page **dir) 127{ 128 kunmap_atomic(dir, KM_USER0); 129} 130 131static swp_entry_t *shmem_swp_map(struct page *page) 132{ 133 return (swp_entry_t *)kmap_atomic(page, KM_USER1); 134} 135 136static inline void shmem_swp_balance_unmap(void) 137{ 138 /* 139 * When passing a pointer to an i_direct entry, to code which 140 * also handles indirect entries and so will shmem_swp_unmap, 141 * we must arrange for the preempt count to remain in balance. 142 * What kmap_atomic of a lowmem page does depends on config 143 * and architecture, so pretend to kmap_atomic some lowmem page. 144 */ 145 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1); 146} 147 148static inline void shmem_swp_unmap(swp_entry_t *entry) 149{ 150 kunmap_atomic(entry, KM_USER1); 151} 152 153static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) 154{ 155 return sb->s_fs_info; 156} 157 158/* 159 * shmem_file_setup pre-accounts the whole fixed size of a VM object, 160 * for shared memory and for shared anonymous (/dev/zero) mappings 161 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), 162 * consistent with the pre-accounting of private mappings ... 163 */ 164static inline int shmem_acct_size(unsigned long flags, loff_t size) 165{ 166 return (flags & VM_ACCOUNT)? 167 security_vm_enough_memory(VM_ACCT(size)): 0; 168} 169 170static inline void shmem_unacct_size(unsigned long flags, loff_t size) 171{ 172 if (flags & VM_ACCOUNT) 173 vm_unacct_memory(VM_ACCT(size)); 174} 175 176/* 177 * ... whereas tmpfs objects are accounted incrementally as 178 * pages are allocated, in order to allow huge sparse files. 179 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, 180 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. 181 */ 182static inline int shmem_acct_block(unsigned long flags) 183{ 184 return (flags & VM_ACCOUNT)? 185 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE)); 186} 187 188static inline void shmem_unacct_blocks(unsigned long flags, long pages) 189{ 190 if (!(flags & VM_ACCOUNT)) 191 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE)); 192} 193 194static const struct super_operations shmem_ops; 195static const struct address_space_operations shmem_aops; 196static const struct file_operations shmem_file_operations; 197static const struct inode_operations shmem_inode_operations; 198static const struct inode_operations shmem_dir_inode_operations; 199static const struct inode_operations shmem_special_inode_operations; 200static struct vm_operations_struct shmem_vm_ops; 201 202static struct backing_dev_info shmem_backing_dev_info __read_mostly = { 203 .ra_pages = 0, /* No readahead */ 204 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK, 205 .unplug_io_fn = default_unplug_io_fn, 206}; 207 208static LIST_HEAD(shmem_swaplist); 209static DEFINE_MUTEX(shmem_swaplist_mutex); 210 211static void shmem_free_blocks(struct inode *inode, long pages) 212{ 213 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 214 if (sbinfo->max_blocks) { 215 spin_lock(&sbinfo->stat_lock); 216 sbinfo->free_blocks += pages; 217 inode->i_blocks -= pages*BLOCKS_PER_PAGE; 218 spin_unlock(&sbinfo->stat_lock); 219 } 220} 221 222static int shmem_reserve_inode(struct super_block *sb) 223{ 224 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 225 if (sbinfo->max_inodes) { 226 spin_lock(&sbinfo->stat_lock); 227 if (!sbinfo->free_inodes) { 228 spin_unlock(&sbinfo->stat_lock); 229 return -ENOSPC; 230 } 231 sbinfo->free_inodes--; 232 spin_unlock(&sbinfo->stat_lock); 233 } 234 return 0; 235} 236 237static void shmem_free_inode(struct super_block *sb) 238{ 239 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 240 if (sbinfo->max_inodes) { 241 spin_lock(&sbinfo->stat_lock); 242 sbinfo->free_inodes++; 243 spin_unlock(&sbinfo->stat_lock); 244 } 245} 246 247/* 248 * shmem_recalc_inode - recalculate the size of an inode 249 * 250 * @inode: inode to recalc 251 * 252 * We have to calculate the free blocks since the mm can drop 253 * undirtied hole pages behind our back. 254 * 255 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped 256 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) 257 * 258 * It has to be called with the spinlock held. 259 */ 260static void shmem_recalc_inode(struct inode *inode) 261{ 262 struct shmem_inode_info *info = SHMEM_I(inode); 263 long freed; 264 265 freed = info->alloced - info->swapped - inode->i_mapping->nrpages; 266 if (freed > 0) { 267 info->alloced -= freed; 268 shmem_unacct_blocks(info->flags, freed); 269 shmem_free_blocks(inode, freed); 270 } 271} 272 273/* 274 * shmem_swp_entry - find the swap vector position in the info structure 275 * 276 * @info: info structure for the inode 277 * @index: index of the page to find 278 * @page: optional page to add to the structure. Has to be preset to 279 * all zeros 280 * 281 * If there is no space allocated yet it will return NULL when 282 * page is NULL, else it will use the page for the needed block, 283 * setting it to NULL on return to indicate that it has been used. 284 * 285 * The swap vector is organized the following way: 286 * 287 * There are SHMEM_NR_DIRECT entries directly stored in the 288 * shmem_inode_info structure. So small files do not need an addional 289 * allocation. 290 * 291 * For pages with index > SHMEM_NR_DIRECT there is the pointer 292 * i_indirect which points to a page which holds in the first half 293 * doubly indirect blocks, in the second half triple indirect blocks: 294 * 295 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the 296 * following layout (for SHMEM_NR_DIRECT == 16): 297 * 298 * i_indirect -> dir --> 16-19 299 * | +-> 20-23 300 * | 301 * +-->dir2 --> 24-27 302 * | +-> 28-31 303 * | +-> 32-35 304 * | +-> 36-39 305 * | 306 * +-->dir3 --> 40-43 307 * +-> 44-47 308 * +-> 48-51 309 * +-> 52-55 310 */ 311static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page) 312{ 313 unsigned long offset; 314 struct page **dir; 315 struct page *subdir; 316 317 if (index < SHMEM_NR_DIRECT) { 318 shmem_swp_balance_unmap(); 319 return info->i_direct+index; 320 } 321 if (!info->i_indirect) { 322 if (page) { 323 info->i_indirect = *page; 324 *page = NULL; 325 } 326 return NULL; /* need another page */ 327 } 328 329 index -= SHMEM_NR_DIRECT; 330 offset = index % ENTRIES_PER_PAGE; 331 index /= ENTRIES_PER_PAGE; 332 dir = shmem_dir_map(info->i_indirect); 333 334 if (index >= ENTRIES_PER_PAGE/2) { 335 index -= ENTRIES_PER_PAGE/2; 336 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE; 337 index %= ENTRIES_PER_PAGE; 338 subdir = *dir; 339 if (!subdir) { 340 if (page) { 341 *dir = *page; 342 *page = NULL; 343 } 344 shmem_dir_unmap(dir); 345 return NULL; /* need another page */ 346 } 347 shmem_dir_unmap(dir); 348 dir = shmem_dir_map(subdir); 349 } 350 351 dir += index; 352 subdir = *dir; 353 if (!subdir) { 354 if (!page || !(subdir = *page)) { 355 shmem_dir_unmap(dir); 356 return NULL; /* need a page */ 357 } 358 *dir = subdir; 359 *page = NULL; 360 } 361 shmem_dir_unmap(dir); 362 return shmem_swp_map(subdir) + offset; 363} 364 365static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value) 366{ 367 long incdec = value? 1: -1; 368 369 entry->val = value; 370 info->swapped += incdec; 371 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) { 372 struct page *page = kmap_atomic_to_page(entry); 373 set_page_private(page, page_private(page) + incdec); 374 } 375} 376 377/* 378 * shmem_swp_alloc - get the position of the swap entry for the page. 379 * If it does not exist allocate the entry. 380 * 381 * @info: info structure for the inode 382 * @index: index of the page to find 383 * @sgp: check and recheck i_size? skip allocation? 384 */ 385static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp) 386{ 387 struct inode *inode = &info->vfs_inode; 388 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 389 struct page *page = NULL; 390 swp_entry_t *entry; 391 392 if (sgp != SGP_WRITE && 393 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 394 return ERR_PTR(-EINVAL); 395 396 while (!(entry = shmem_swp_entry(info, index, &page))) { 397 if (sgp == SGP_READ) 398 return shmem_swp_map(ZERO_PAGE(0)); 399 /* 400 * Test free_blocks against 1 not 0, since we have 1 data 401 * page (and perhaps indirect index pages) yet to allocate: 402 * a waste to allocate index if we cannot allocate data. 403 */ 404 if (sbinfo->max_blocks) { 405 spin_lock(&sbinfo->stat_lock); 406 if (sbinfo->free_blocks <= 1) { 407 spin_unlock(&sbinfo->stat_lock); 408 return ERR_PTR(-ENOSPC); 409 } 410 sbinfo->free_blocks--; 411 inode->i_blocks += BLOCKS_PER_PAGE; 412 spin_unlock(&sbinfo->stat_lock); 413 } 414 415 spin_unlock(&info->lock); 416 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping)); 417 if (page) 418 set_page_private(page, 0); 419 spin_lock(&info->lock); 420 421 if (!page) { 422 shmem_free_blocks(inode, 1); 423 return ERR_PTR(-ENOMEM); 424 } 425 if (sgp != SGP_WRITE && 426 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { 427 entry = ERR_PTR(-EINVAL); 428 break; 429 } 430 if (info->next_index <= index) 431 info->next_index = index + 1; 432 } 433 if (page) { 434 /* another task gave its page, or truncated the file */ 435 shmem_free_blocks(inode, 1); 436 shmem_dir_free(page); 437 } 438 if (info->next_index <= index && !IS_ERR(entry)) 439 info->next_index = index + 1; 440 return entry; 441} 442 443/* 444 * shmem_free_swp - free some swap entries in a directory 445 * 446 * @dir: pointer to the directory 447 * @edir: pointer after last entry of the directory 448 * @punch_lock: pointer to spinlock when needed for the holepunch case 449 */ 450static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir, 451 spinlock_t *punch_lock) 452{ 453 spinlock_t *punch_unlock = NULL; 454 swp_entry_t *ptr; 455 int freed = 0; 456 457 for (ptr = dir; ptr < edir; ptr++) { 458 if (ptr->val) { 459 if (unlikely(punch_lock)) { 460 punch_unlock = punch_lock; 461 punch_lock = NULL; 462 spin_lock(punch_unlock); 463 if (!ptr->val) 464 continue; 465 } 466 free_swap_and_cache(*ptr); 467 *ptr = (swp_entry_t){0}; 468 freed++; 469 } 470 } 471 if (punch_unlock) 472 spin_unlock(punch_unlock); 473 return freed; 474} 475 476static int shmem_map_and_free_swp(struct page *subdir, int offset, 477 int limit, struct page ***dir, spinlock_t *punch_lock) 478{ 479 swp_entry_t *ptr; 480 int freed = 0; 481 482 ptr = shmem_swp_map(subdir); 483 for (; offset < limit; offset += LATENCY_LIMIT) { 484 int size = limit - offset; 485 if (size > LATENCY_LIMIT) 486 size = LATENCY_LIMIT; 487 freed += shmem_free_swp(ptr+offset, ptr+offset+size, 488 punch_lock); 489 if (need_resched()) { 490 shmem_swp_unmap(ptr); 491 if (*dir) { 492 shmem_dir_unmap(*dir); 493 *dir = NULL; 494 } 495 cond_resched(); 496 ptr = shmem_swp_map(subdir); 497 } 498 } 499 shmem_swp_unmap(ptr); 500 return freed; 501} 502 503static void shmem_free_pages(struct list_head *next) 504{ 505 struct page *page; 506 int freed = 0; 507 508 do { 509 page = container_of(next, struct page, lru); 510 next = next->next; 511 shmem_dir_free(page); 512 freed++; 513 if (freed >= LATENCY_LIMIT) { 514 cond_resched(); 515 freed = 0; 516 } 517 } while (next); 518} 519 520static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end) 521{ 522 struct shmem_inode_info *info = SHMEM_I(inode); 523 unsigned long idx; 524 unsigned long size; 525 unsigned long limit; 526 unsigned long stage; 527 unsigned long diroff; 528 struct page **dir; 529 struct page *topdir; 530 struct page *middir; 531 struct page *subdir; 532 swp_entry_t *ptr; 533 LIST_HEAD(pages_to_free); 534 long nr_pages_to_free = 0; 535 long nr_swaps_freed = 0; 536 int offset; 537 int freed; 538 int punch_hole; 539 spinlock_t *needs_lock; 540 spinlock_t *punch_lock; 541 unsigned long upper_limit; 542 543 inode->i_ctime = inode->i_mtime = CURRENT_TIME; 544 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 545 if (idx >= info->next_index) 546 return; 547 548 spin_lock(&info->lock); 549 info->flags |= SHMEM_TRUNCATE; 550 if (likely(end == (loff_t) -1)) { 551 limit = info->next_index; 552 upper_limit = SHMEM_MAX_INDEX; 553 info->next_index = idx; 554 needs_lock = NULL; 555 punch_hole = 0; 556 } else { 557 if (end + 1 >= inode->i_size) { /* we may free a little more */ 558 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >> 559 PAGE_CACHE_SHIFT; 560 upper_limit = SHMEM_MAX_INDEX; 561 } else { 562 limit = (end + 1) >> PAGE_CACHE_SHIFT; 563 upper_limit = limit; 564 } 565 needs_lock = &info->lock; 566 punch_hole = 1; 567 } 568 569 topdir = info->i_indirect; 570 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) { 571 info->i_indirect = NULL; 572 nr_pages_to_free++; 573 list_add(&topdir->lru, &pages_to_free); 574 } 575 spin_unlock(&info->lock); 576 577 if (info->swapped && idx < SHMEM_NR_DIRECT) { 578 ptr = info->i_direct; 579 size = limit; 580 if (size > SHMEM_NR_DIRECT) 581 size = SHMEM_NR_DIRECT; 582 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock); 583 } 584 585 /* 586 * If there are no indirect blocks or we are punching a hole 587 * below indirect blocks, nothing to be done. 588 */ 589 if (!topdir || limit <= SHMEM_NR_DIRECT) 590 goto done2; 591 592 /* 593 * The truncation case has already dropped info->lock, and we're safe 594 * because i_size and next_index have already been lowered, preventing 595 * access beyond. But in the punch_hole case, we still need to take 596 * the lock when updating the swap directory, because there might be 597 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or 598 * shmem_writepage. However, whenever we find we can remove a whole 599 * directory page (not at the misaligned start or end of the range), 600 * we first NULLify its pointer in the level above, and then have no 601 * need to take the lock when updating its contents: needs_lock and 602 * punch_lock (either pointing to info->lock or NULL) manage this. 603 */ 604 605 upper_limit -= SHMEM_NR_DIRECT; 606 limit -= SHMEM_NR_DIRECT; 607 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0; 608 offset = idx % ENTRIES_PER_PAGE; 609 idx -= offset; 610 611 dir = shmem_dir_map(topdir); 612 stage = ENTRIES_PER_PAGEPAGE/2; 613 if (idx < ENTRIES_PER_PAGEPAGE/2) { 614 middir = topdir; 615 diroff = idx/ENTRIES_PER_PAGE; 616 } else { 617 dir += ENTRIES_PER_PAGE/2; 618 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE; 619 while (stage <= idx) 620 stage += ENTRIES_PER_PAGEPAGE; 621 middir = *dir; 622 if (*dir) { 623 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) % 624 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE; 625 if (!diroff && !offset && upper_limit >= stage) { 626 if (needs_lock) { 627 spin_lock(needs_lock); 628 *dir = NULL; 629 spin_unlock(needs_lock); 630 needs_lock = NULL; 631 } else 632 *dir = NULL; 633 nr_pages_to_free++; 634 list_add(&middir->lru, &pages_to_free); 635 } 636 shmem_dir_unmap(dir); 637 dir = shmem_dir_map(middir); 638 } else { 639 diroff = 0; 640 offset = 0; 641 idx = stage; 642 } 643 } 644 645 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) { 646 if (unlikely(idx == stage)) { 647 shmem_dir_unmap(dir); 648 dir = shmem_dir_map(topdir) + 649 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 650 while (!*dir) { 651 dir++; 652 idx += ENTRIES_PER_PAGEPAGE; 653 if (idx >= limit) 654 goto done1; 655 } 656 stage = idx + ENTRIES_PER_PAGEPAGE; 657 middir = *dir; 658 if (punch_hole) 659 needs_lock = &info->lock; 660 if (upper_limit >= stage) { 661 if (needs_lock) { 662 spin_lock(needs_lock); 663 *dir = NULL; 664 spin_unlock(needs_lock); 665 needs_lock = NULL; 666 } else 667 *dir = NULL; 668 nr_pages_to_free++; 669 list_add(&middir->lru, &pages_to_free); 670 } 671 shmem_dir_unmap(dir); 672 cond_resched(); 673 dir = shmem_dir_map(middir); 674 diroff = 0; 675 } 676 punch_lock = needs_lock; 677 subdir = dir[diroff]; 678 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) { 679 if (needs_lock) { 680 spin_lock(needs_lock); 681 dir[diroff] = NULL; 682 spin_unlock(needs_lock); 683 punch_lock = NULL; 684 } else 685 dir[diroff] = NULL; 686 nr_pages_to_free++; 687 list_add(&subdir->lru, &pages_to_free); 688 } 689 if (subdir && page_private(subdir) /* has swap entries */) { 690 size = limit - idx; 691 if (size > ENTRIES_PER_PAGE) 692 size = ENTRIES_PER_PAGE; 693 freed = shmem_map_and_free_swp(subdir, 694 offset, size, &dir, punch_lock); 695 if (!dir) 696 dir = shmem_dir_map(middir); 697 nr_swaps_freed += freed; 698 if (offset || punch_lock) { 699 spin_lock(&info->lock); 700 set_page_private(subdir, 701 page_private(subdir) - freed); 702 spin_unlock(&info->lock); 703 } else 704 BUG_ON(page_private(subdir) != freed); 705 } 706 offset = 0; 707 } 708done1: 709 shmem_dir_unmap(dir); 710done2: 711 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) { 712 /* 713 * Call truncate_inode_pages again: racing shmem_unuse_inode 714 * may have swizzled a page in from swap since vmtruncate or 715 * generic_delete_inode did it, before we lowered next_index. 716 * Also, though shmem_getpage checks i_size before adding to 717 * cache, no recheck after: so fix the narrow window there too. 718 * 719 * Recalling truncate_inode_pages_range and unmap_mapping_range 720 * every time for punch_hole (which never got a chance to clear 721 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive, 722 * yet hardly ever necessary: try to optimize them out later. 723 */ 724 truncate_inode_pages_range(inode->i_mapping, start, end); 725 if (punch_hole) 726 unmap_mapping_range(inode->i_mapping, start, 727 end - start, 1); 728 } 729 730 spin_lock(&info->lock); 731 info->flags &= ~SHMEM_TRUNCATE; 732 info->swapped -= nr_swaps_freed; 733 if (nr_pages_to_free) 734 shmem_free_blocks(inode, nr_pages_to_free); 735 shmem_recalc_inode(inode); 736 spin_unlock(&info->lock); 737 738 /* 739 * Empty swap vector directory pages to be freed? 740 */ 741 if (!list_empty(&pages_to_free)) { 742 pages_to_free.prev->next = NULL; 743 shmem_free_pages(pages_to_free.next); 744 } 745} 746 747static void shmem_truncate(struct inode *inode) 748{ 749 shmem_truncate_range(inode, inode->i_size, (loff_t)-1); 750} 751 752static int shmem_notify_change(struct dentry *dentry, struct iattr *attr) 753{ 754 struct inode *inode = dentry->d_inode; 755 struct page *page = NULL; 756 int error; 757 758 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { 759 if (attr->ia_size < inode->i_size) { 760 /* 761 * If truncating down to a partial page, then 762 * if that page is already allocated, hold it 763 * in memory until the truncation is over, so 764 * truncate_partial_page cannnot miss it were 765 * it assigned to swap. 766 */ 767 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) { 768 (void) shmem_getpage(inode, 769 attr->ia_size>>PAGE_CACHE_SHIFT, 770 &page, SGP_READ, NULL); 771 if (page) 772 unlock_page(page); 773 } 774 /* 775 * Reset SHMEM_PAGEIN flag so that shmem_truncate can 776 * detect if any pages might have been added to cache 777 * after truncate_inode_pages. But we needn't bother 778 * if it's being fully truncated to zero-length: the 779 * nrpages check is efficient enough in that case. 780 */ 781 if (attr->ia_size) { 782 struct shmem_inode_info *info = SHMEM_I(inode); 783 spin_lock(&info->lock); 784 info->flags &= ~SHMEM_PAGEIN; 785 spin_unlock(&info->lock); 786 } 787 } 788 } 789 790 error = inode_change_ok(inode, attr); 791 if (!error) 792 error = inode_setattr(inode, attr); 793#ifdef CONFIG_TMPFS_POSIX_ACL 794 if (!error && (attr->ia_valid & ATTR_MODE)) 795 error = generic_acl_chmod(inode, &shmem_acl_ops); 796#endif 797 if (page) 798 page_cache_release(page); 799 return error; 800} 801 802static void shmem_delete_inode(struct inode *inode) 803{ 804 struct shmem_inode_info *info = SHMEM_I(inode); 805 806 if (inode->i_op->truncate == shmem_truncate) { 807 truncate_inode_pages(inode->i_mapping, 0); 808 shmem_unacct_size(info->flags, inode->i_size); 809 inode->i_size = 0; 810 shmem_truncate(inode); 811 if (!list_empty(&info->swaplist)) { 812 mutex_lock(&shmem_swaplist_mutex); 813 list_del_init(&info->swaplist); 814 mutex_unlock(&shmem_swaplist_mutex); 815 } 816 } 817 BUG_ON(inode->i_blocks); 818 shmem_free_inode(inode->i_sb); 819 clear_inode(inode); 820} 821 822static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir) 823{ 824 swp_entry_t *ptr; 825 826 for (ptr = dir; ptr < edir; ptr++) { 827 if (ptr->val == entry.val) 828 return ptr - dir; 829 } 830 return -1; 831} 832 833static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page) 834{ 835 struct inode *inode; 836 unsigned long idx; 837 unsigned long size; 838 unsigned long limit; 839 unsigned long stage; 840 struct page **dir; 841 struct page *subdir; 842 swp_entry_t *ptr; 843 int offset; 844 int error; 845 846 idx = 0; 847 ptr = info->i_direct; 848 spin_lock(&info->lock); 849 if (!info->swapped) { 850 list_del_init(&info->swaplist); 851 goto lost2; 852 } 853 limit = info->next_index; 854 size = limit; 855 if (size > SHMEM_NR_DIRECT) 856 size = SHMEM_NR_DIRECT; 857 offset = shmem_find_swp(entry, ptr, ptr+size); 858 if (offset >= 0) 859 goto found; 860 if (!info->i_indirect) 861 goto lost2; 862 863 dir = shmem_dir_map(info->i_indirect); 864 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2; 865 866 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) { 867 if (unlikely(idx == stage)) { 868 shmem_dir_unmap(dir-1); 869 if (cond_resched_lock(&info->lock)) { 870 /* check it has not been truncated */ 871 if (limit > info->next_index) { 872 limit = info->next_index; 873 if (idx >= limit) 874 goto lost2; 875 } 876 } 877 dir = shmem_dir_map(info->i_indirect) + 878 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 879 while (!*dir) { 880 dir++; 881 idx += ENTRIES_PER_PAGEPAGE; 882 if (idx >= limit) 883 goto lost1; 884 } 885 stage = idx + ENTRIES_PER_PAGEPAGE; 886 subdir = *dir; 887 shmem_dir_unmap(dir); 888 dir = shmem_dir_map(subdir); 889 } 890 subdir = *dir; 891 if (subdir && page_private(subdir)) { 892 ptr = shmem_swp_map(subdir); 893 size = limit - idx; 894 if (size > ENTRIES_PER_PAGE) 895 size = ENTRIES_PER_PAGE; 896 offset = shmem_find_swp(entry, ptr, ptr+size); 897 shmem_swp_unmap(ptr); 898 if (offset >= 0) { 899 shmem_dir_unmap(dir); 900 goto found; 901 } 902 } 903 } 904lost1: 905 shmem_dir_unmap(dir-1); 906lost2: 907 spin_unlock(&info->lock); 908 return 0; 909found: 910 idx += offset; 911 inode = igrab(&info->vfs_inode); 912 spin_unlock(&info->lock); 913 914 /* 915 * Move _head_ to start search for next from here. 916 * But be careful: shmem_delete_inode checks list_empty without taking 917 * mutex, and there's an instant in list_move_tail when info->swaplist 918 * would appear empty, if it were the only one on shmem_swaplist. We 919 * could avoid doing it if inode NULL; or use this minor optimization. 920 */ 921 if (shmem_swaplist.next != &info->swaplist) 922 list_move_tail(&shmem_swaplist, &info->swaplist); 923 mutex_unlock(&shmem_swaplist_mutex); 924 925 error = 1; 926 if (!inode) 927 goto out; 928 /* Precharge page while we can wait, compensate afterwards */ 929 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL); 930 if (error) 931 goto out; 932 error = radix_tree_preload(GFP_KERNEL); 933 if (error) 934 goto uncharge; 935 error = 1; 936 937 spin_lock(&info->lock); 938 ptr = shmem_swp_entry(info, idx, NULL); 939 if (ptr && ptr->val == entry.val) 940 error = add_to_page_cache(page, inode->i_mapping, 941 idx, GFP_NOWAIT); 942 if (error == -EEXIST) { 943 struct page *filepage = find_get_page(inode->i_mapping, idx); 944 error = 1; 945 if (filepage) { 946 /* 947 * There might be a more uptodate page coming down 948 * from a stacked writepage: forget our swappage if so. 949 */ 950 if (PageUptodate(filepage)) 951 error = 0; 952 page_cache_release(filepage); 953 } 954 } 955 if (!error) { 956 delete_from_swap_cache(page); 957 set_page_dirty(page); 958 info->flags |= SHMEM_PAGEIN; 959 shmem_swp_set(info, ptr, 0); 960 swap_free(entry); 961 error = 1; /* not an error, but entry was found */ 962 } 963 if (ptr) 964 shmem_swp_unmap(ptr); 965 spin_unlock(&info->lock); 966 radix_tree_preload_end(); 967uncharge: 968 mem_cgroup_uncharge_page(page); 969out: 970 unlock_page(page); 971 page_cache_release(page); 972 iput(inode); /* allows for NULL */ 973 return error; 974} 975 976/* 977 * shmem_unuse() search for an eventually swapped out shmem page. 978 */ 979int shmem_unuse(swp_entry_t entry, struct page *page) 980{ 981 struct list_head *p, *next; 982 struct shmem_inode_info *info; 983 int found = 0; 984 985 mutex_lock(&shmem_swaplist_mutex); 986 list_for_each_safe(p, next, &shmem_swaplist) { 987 info = list_entry(p, struct shmem_inode_info, swaplist); 988 found = shmem_unuse_inode(info, entry, page); 989 cond_resched(); 990 if (found) 991 goto out; 992 } 993 mutex_unlock(&shmem_swaplist_mutex); 994out: return found; /* 0 or 1 or -ENOMEM */ 995} 996 997/* 998 * Move the page from the page cache to the swap cache. 999 */ 1000static int shmem_writepage(struct page *page, struct writeback_control *wbc) 1001{ 1002 struct shmem_inode_info *info; 1003 swp_entry_t *entry, swap; 1004 struct address_space *mapping; 1005 unsigned long index; 1006 struct inode *inode; 1007 1008 BUG_ON(!PageLocked(page)); 1009 mapping = page->mapping; 1010 index = page->index; 1011 inode = mapping->host; 1012 info = SHMEM_I(inode); 1013 if (info->flags & VM_LOCKED) 1014 goto redirty; 1015 if (!total_swap_pages) 1016 goto redirty; 1017 1018 /* 1019 * shmem_backing_dev_info's capabilities prevent regular writeback or 1020 * sync from ever calling shmem_writepage; but a stacking filesystem 1021 * may use the ->writepage of its underlying filesystem, in which case 1022 * tmpfs should write out to swap only in response to memory pressure, 1023 * and not for pdflush or sync. However, in those cases, we do still 1024 * want to check if there's a redundant swappage to be discarded. 1025 */ 1026 if (wbc->for_reclaim) 1027 swap = get_swap_page(); 1028 else 1029 swap.val = 0; 1030 1031 spin_lock(&info->lock); 1032 if (index >= info->next_index) { 1033 BUG_ON(!(info->flags & SHMEM_TRUNCATE)); 1034 goto unlock; 1035 } 1036 entry = shmem_swp_entry(info, index, NULL); 1037 if (entry->val) { 1038 /* 1039 * The more uptodate page coming down from a stacked 1040 * writepage should replace our old swappage. 1041 */ 1042 free_swap_and_cache(*entry); 1043 shmem_swp_set(info, entry, 0); 1044 } 1045 shmem_recalc_inode(inode); 1046 1047 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { 1048 remove_from_page_cache(page); 1049 shmem_swp_set(info, entry, swap.val); 1050 shmem_swp_unmap(entry); 1051 if (list_empty(&info->swaplist)) 1052 inode = igrab(inode); 1053 else 1054 inode = NULL; 1055 spin_unlock(&info->lock); 1056 swap_duplicate(swap); 1057 BUG_ON(page_mapped(page)); 1058 page_cache_release(page); /* pagecache ref */ 1059 set_page_dirty(page); 1060 unlock_page(page); 1061 if (inode) { 1062 mutex_lock(&shmem_swaplist_mutex); 1063 /* move instead of add in case we're racing */ 1064 list_move_tail(&info->swaplist, &shmem_swaplist); 1065 mutex_unlock(&shmem_swaplist_mutex); 1066 iput(inode); 1067 } 1068 return 0; 1069 } 1070 1071 shmem_swp_unmap(entry); 1072unlock: 1073 spin_unlock(&info->lock); 1074 swap_free(swap); 1075redirty: 1076 set_page_dirty(page); 1077 if (wbc->for_reclaim) 1078 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ 1079 unlock_page(page); 1080 return 0; 1081} 1082 1083#ifdef CONFIG_NUMA 1084#ifdef CONFIG_TMPFS 1085static int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes) 1086{ 1087 char *nodelist = strchr(value, ':'); 1088 int err = 1; 1089 1090 if (nodelist) { 1091 /* NUL-terminate policy string */ 1092 *nodelist++ = '\0'; 1093 if (nodelist_parse(nodelist, *policy_nodes)) 1094 goto out; 1095 if (!nodes_subset(*policy_nodes, node_states[N_HIGH_MEMORY])) 1096 goto out; 1097 } 1098 if (!strcmp(value, "default")) { 1099 *policy = MPOL_DEFAULT; 1100 /* Don't allow a nodelist */ 1101 if (!nodelist) 1102 err = 0; 1103 } else if (!strcmp(value, "prefer")) { 1104 *policy = MPOL_PREFERRED; 1105 /* Insist on a nodelist of one node only */ 1106 if (nodelist) { 1107 char *rest = nodelist; 1108 while (isdigit(*rest)) 1109 rest++; 1110 if (!*rest) 1111 err = 0; 1112 } 1113 } else if (!strcmp(value, "bind")) { 1114 *policy = MPOL_BIND; 1115 /* Insist on a nodelist */ 1116 if (nodelist) 1117 err = 0; 1118 } else if (!strcmp(value, "interleave")) { 1119 *policy = MPOL_INTERLEAVE; 1120 /* 1121 * Default to online nodes with memory if no nodelist 1122 */ 1123 if (!nodelist) 1124 *policy_nodes = node_states[N_HIGH_MEMORY]; 1125 err = 0; 1126 } 1127out: 1128 /* Restore string for error message */ 1129 if (nodelist) 1130 *--nodelist = ':'; 1131 return err; 1132} 1133 1134static void shmem_show_mpol(struct seq_file *seq, int policy, 1135 const nodemask_t policy_nodes) 1136{ 1137 char *policy_string; 1138 1139 switch (policy) { 1140 case MPOL_PREFERRED: 1141 policy_string = "prefer"; 1142 break; 1143 case MPOL_BIND: 1144 policy_string = "bind"; 1145 break; 1146 case MPOL_INTERLEAVE: 1147 policy_string = "interleave"; 1148 break; 1149 default: 1150 /* MPOL_DEFAULT */ 1151 return; 1152 } 1153 1154 seq_printf(seq, ",mpol=%s", policy_string); 1155 1156 if (policy != MPOL_INTERLEAVE || 1157 !nodes_equal(policy_nodes, node_states[N_HIGH_MEMORY])) { 1158 char buffer[64]; 1159 int len; 1160 1161 len = nodelist_scnprintf(buffer, sizeof(buffer), policy_nodes); 1162 if (len < sizeof(buffer)) 1163 seq_printf(seq, ":%s", buffer); 1164 else 1165 seq_printf(seq, ":?"); 1166 } 1167} 1168#endif /* CONFIG_TMPFS */ 1169 1170static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, 1171 struct shmem_inode_info *info, unsigned long idx) 1172{ 1173 struct vm_area_struct pvma; 1174 struct page *page; 1175 1176 /* Create a pseudo vma that just contains the policy */ 1177 pvma.vm_start = 0; 1178 pvma.vm_pgoff = idx; 1179 pvma.vm_ops = NULL; 1180 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx); 1181 page = swapin_readahead(entry, gfp, &pvma, 0); 1182 mpol_free(pvma.vm_policy); 1183 return page; 1184} 1185 1186static struct page *shmem_alloc_page(gfp_t gfp, 1187 struct shmem_inode_info *info, unsigned long idx) 1188{ 1189 struct vm_area_struct pvma; 1190 struct page *page; 1191 1192 /* Create a pseudo vma that just contains the policy */ 1193 pvma.vm_start = 0; 1194 pvma.vm_pgoff = idx; 1195 pvma.vm_ops = NULL; 1196 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx); 1197 page = alloc_page_vma(gfp, &pvma, 0); 1198 mpol_free(pvma.vm_policy); 1199 return page; 1200} 1201#else /* !CONFIG_NUMA */ 1202#ifdef CONFIG_TMPFS 1203static inline int shmem_parse_mpol(char *value, int *policy, 1204 nodemask_t *policy_nodes) 1205{ 1206 return 1; 1207} 1208 1209static inline void shmem_show_mpol(struct seq_file *seq, int policy, 1210 const nodemask_t policy_nodes) 1211{ 1212} 1213#endif /* CONFIG_TMPFS */ 1214 1215static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, 1216 struct shmem_inode_info *info, unsigned long idx) 1217{ 1218 return swapin_readahead(entry, gfp, NULL, 0); 1219} 1220 1221static inline struct page *shmem_alloc_page(gfp_t gfp, 1222 struct shmem_inode_info *info, unsigned long idx) 1223{ 1224 return alloc_page(gfp); 1225} 1226#endif /* CONFIG_NUMA */ 1227 1228/* 1229 * shmem_getpage - either get the page from swap or allocate a new one 1230 * 1231 * If we allocate a new one we do not mark it dirty. That's up to the 1232 * vm. If we swap it in we mark it dirty since we also free the swap 1233 * entry since a page cannot live in both the swap and page cache 1234 */ 1235static int shmem_getpage(struct inode *inode, unsigned long idx, 1236 struct page **pagep, enum sgp_type sgp, int *type) 1237{ 1238 struct address_space *mapping = inode->i_mapping; 1239 struct shmem_inode_info *info = SHMEM_I(inode); 1240 struct shmem_sb_info *sbinfo; 1241 struct page *filepage = *pagep; 1242 struct page *swappage; 1243 swp_entry_t *entry; 1244 swp_entry_t swap; 1245 gfp_t gfp; 1246 int error; 1247 1248 if (idx >= SHMEM_MAX_INDEX) 1249 return -EFBIG; 1250 1251 if (type) 1252 *type = 0; 1253 1254 /* 1255 * Normally, filepage is NULL on entry, and either found 1256 * uptodate immediately, or allocated and zeroed, or read 1257 * in under swappage, which is then assigned to filepage. 1258 * But shmem_readpage (required for splice) passes in a locked 1259 * filepage, which may be found not uptodate by other callers 1260 * too, and may need to be copied from the swappage read in. 1261 */ 1262repeat: 1263 if (!filepage) 1264 filepage = find_lock_page(mapping, idx); 1265 if (filepage && PageUptodate(filepage)) 1266 goto done; 1267 error = 0; 1268 gfp = mapping_gfp_mask(mapping); 1269 if (!filepage) { 1270 /* 1271 * Try to preload while we can wait, to not make a habit of 1272 * draining atomic reserves; but don't latch on to this cpu. 1273 */ 1274 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM); 1275 if (error) 1276 goto failed; 1277 radix_tree_preload_end(); 1278 } 1279 1280 spin_lock(&info->lock); 1281 shmem_recalc_inode(inode); 1282 entry = shmem_swp_alloc(info, idx, sgp); 1283 if (IS_ERR(entry)) { 1284 spin_unlock(&info->lock); 1285 error = PTR_ERR(entry); 1286 goto failed; 1287 } 1288 swap = *entry; 1289 1290 if (swap.val) { 1291 /* Look it up and read it in.. */ 1292 swappage = lookup_swap_cache(swap); 1293 if (!swappage) { 1294 shmem_swp_unmap(entry); 1295 /* here we actually do the io */ 1296 if (type && !(*type & VM_FAULT_MAJOR)) { 1297 __count_vm_event(PGMAJFAULT); 1298 *type |= VM_FAULT_MAJOR; 1299 } 1300 spin_unlock(&info->lock); 1301 swappage = shmem_swapin(swap, gfp, info, idx); 1302 if (!swappage) { 1303 spin_lock(&info->lock); 1304 entry = shmem_swp_alloc(info, idx, sgp); 1305 if (IS_ERR(entry)) 1306 error = PTR_ERR(entry); 1307 else { 1308 if (entry->val == swap.val) 1309 error = -ENOMEM; 1310 shmem_swp_unmap(entry); 1311 } 1312 spin_unlock(&info->lock); 1313 if (error) 1314 goto failed; 1315 goto repeat; 1316 } 1317 wait_on_page_locked(swappage); 1318 page_cache_release(swappage); 1319 goto repeat; 1320 } 1321 1322 /* We have to do this with page locked to prevent races */ 1323 if (TestSetPageLocked(swappage)) { 1324 shmem_swp_unmap(entry); 1325 spin_unlock(&info->lock); 1326 wait_on_page_locked(swappage); 1327 page_cache_release(swappage); 1328 goto repeat; 1329 } 1330 if (PageWriteback(swappage)) { 1331 shmem_swp_unmap(entry); 1332 spin_unlock(&info->lock); 1333 wait_on_page_writeback(swappage); 1334 unlock_page(swappage); 1335 page_cache_release(swappage); 1336 goto repeat; 1337 } 1338 if (!PageUptodate(swappage)) { 1339 shmem_swp_unmap(entry); 1340 spin_unlock(&info->lock); 1341 unlock_page(swappage); 1342 page_cache_release(swappage); 1343 error = -EIO; 1344 goto failed; 1345 } 1346 1347 if (filepage) { 1348 shmem_swp_set(info, entry, 0); 1349 shmem_swp_unmap(entry); 1350 delete_from_swap_cache(swappage); 1351 spin_unlock(&info->lock); 1352 copy_highpage(filepage, swappage); 1353 unlock_page(swappage); 1354 page_cache_release(swappage); 1355 flush_dcache_page(filepage); 1356 SetPageUptodate(filepage); 1357 set_page_dirty(filepage); 1358 swap_free(swap); 1359 } else if (!(error = add_to_page_cache( 1360 swappage, mapping, idx, GFP_NOWAIT))) { 1361 info->flags |= SHMEM_PAGEIN; 1362 shmem_swp_set(info, entry, 0); 1363 shmem_swp_unmap(entry); 1364 delete_from_swap_cache(swappage); 1365 spin_unlock(&info->lock); 1366 filepage = swappage; 1367 set_page_dirty(filepage); 1368 swap_free(swap); 1369 } else { 1370 shmem_swp_unmap(entry); 1371 spin_unlock(&info->lock); 1372 unlock_page(swappage); 1373 page_cache_release(swappage); 1374 if (error == -ENOMEM) { 1375 /* allow reclaim from this memory cgroup */ 1376 error = mem_cgroup_cache_charge(NULL, 1377 current->mm, gfp & ~__GFP_HIGHMEM); 1378 if (error) 1379 goto failed; 1380 } 1381 goto repeat; 1382 } 1383 } else if (sgp == SGP_READ && !filepage) { 1384 shmem_swp_unmap(entry); 1385 filepage = find_get_page(mapping, idx); 1386 if (filepage && 1387 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) { 1388 spin_unlock(&info->lock); 1389 wait_on_page_locked(filepage); 1390 page_cache_release(filepage); 1391 filepage = NULL; 1392 goto repeat; 1393 } 1394 spin_unlock(&info->lock); 1395 } else { 1396 shmem_swp_unmap(entry); 1397 sbinfo = SHMEM_SB(inode->i_sb); 1398 if (sbinfo->max_blocks) { 1399 spin_lock(&sbinfo->stat_lock); 1400 if (sbinfo->free_blocks == 0 || 1401 shmem_acct_block(info->flags)) { 1402 spin_unlock(&sbinfo->stat_lock); 1403 spin_unlock(&info->lock); 1404 error = -ENOSPC; 1405 goto failed; 1406 } 1407 sbinfo->free_blocks--; 1408 inode->i_blocks += BLOCKS_PER_PAGE; 1409 spin_unlock(&sbinfo->stat_lock); 1410 } else if (shmem_acct_block(info->flags)) { 1411 spin_unlock(&info->lock); 1412 error = -ENOSPC; 1413 goto failed; 1414 } 1415 1416 if (!filepage) { 1417 spin_unlock(&info->lock); 1418 filepage = shmem_alloc_page(gfp, info, idx); 1419 if (!filepage) { 1420 shmem_unacct_blocks(info->flags, 1); 1421 shmem_free_blocks(inode, 1); 1422 error = -ENOMEM; 1423 goto failed; 1424 } 1425 1426 /* Precharge page while we can wait, compensate after */ 1427 error = mem_cgroup_cache_charge(filepage, current->mm, 1428 gfp & ~__GFP_HIGHMEM); 1429 if (error) { 1430 page_cache_release(filepage); 1431 shmem_unacct_blocks(info->flags, 1); 1432 shmem_free_blocks(inode, 1); 1433 filepage = NULL; 1434 goto failed; 1435 } 1436 1437 spin_lock(&info->lock); 1438 entry = shmem_swp_alloc(info, idx, sgp); 1439 if (IS_ERR(entry)) 1440 error = PTR_ERR(entry); 1441 else { 1442 swap = *entry; 1443 shmem_swp_unmap(entry); 1444 } 1445 if (error || swap.val || 0 != add_to_page_cache_lru( 1446 filepage, mapping, idx, GFP_NOWAIT)) { 1447 spin_unlock(&info->lock); 1448 mem_cgroup_uncharge_page(filepage); 1449 page_cache_release(filepage); 1450 shmem_unacct_blocks(info->flags, 1); 1451 shmem_free_blocks(inode, 1); 1452 filepage = NULL; 1453 if (error) 1454 goto failed; 1455 goto repeat; 1456 } 1457 mem_cgroup_uncharge_page(filepage); 1458 info->flags |= SHMEM_PAGEIN; 1459 } 1460 1461 info->alloced++; 1462 spin_unlock(&info->lock); 1463 clear_highpage(filepage); 1464 flush_dcache_page(filepage); 1465 SetPageUptodate(filepage); 1466 if (sgp == SGP_DIRTY) 1467 set_page_dirty(filepage); 1468 } 1469done: 1470 *pagep = filepage; 1471 return 0; 1472 1473failed: 1474 if (*pagep != filepage) { 1475 unlock_page(filepage); 1476 page_cache_release(filepage); 1477 } 1478 return error; 1479} 1480 1481static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1482{ 1483 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 1484 int error; 1485 int ret; 1486 1487 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 1488 return VM_FAULT_SIGBUS; 1489 1490 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret); 1491 if (error) 1492 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); 1493 1494 mark_page_accessed(vmf->page); 1495 return ret | VM_FAULT_LOCKED; 1496} 1497 1498#ifdef CONFIG_NUMA 1499static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new) 1500{ 1501 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1502 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new); 1503} 1504 1505static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, 1506 unsigned long addr) 1507{ 1508 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1509 unsigned long idx; 1510 1511 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; 1512 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx); 1513} 1514#endif 1515 1516int shmem_lock(struct file *file, int lock, struct user_struct *user) 1517{ 1518 struct inode *inode = file->f_path.dentry->d_inode; 1519 struct shmem_inode_info *info = SHMEM_I(inode); 1520 int retval = -ENOMEM; 1521 1522 spin_lock(&info->lock); 1523 if (lock && !(info->flags & VM_LOCKED)) { 1524 if (!user_shm_lock(inode->i_size, user)) 1525 goto out_nomem; 1526 info->flags |= VM_LOCKED; 1527 } 1528 if (!lock && (info->flags & VM_LOCKED) && user) { 1529 user_shm_unlock(inode->i_size, user); 1530 info->flags &= ~VM_LOCKED; 1531 } 1532 retval = 0; 1533out_nomem: 1534 spin_unlock(&info->lock); 1535 return retval; 1536} 1537 1538static int shmem_mmap(struct file *file, struct vm_area_struct *vma) 1539{ 1540 file_accessed(file); 1541 vma->vm_ops = &shmem_vm_ops; 1542 vma->vm_flags |= VM_CAN_NONLINEAR; 1543 return 0; 1544} 1545 1546static struct inode * 1547shmem_get_inode(struct super_block *sb, int mode, dev_t dev) 1548{ 1549 struct inode *inode; 1550 struct shmem_inode_info *info; 1551 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 1552 1553 if (shmem_reserve_inode(sb)) 1554 return NULL; 1555 1556 inode = new_inode(sb); 1557 if (inode) { 1558 inode->i_mode = mode; 1559 inode->i_uid = current->fsuid; 1560 inode->i_gid = current->fsgid; 1561 inode->i_blocks = 0; 1562 inode->i_mapping->a_ops = &shmem_aops; 1563 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; 1564 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 1565 inode->i_generation = get_seconds(); 1566 info = SHMEM_I(inode); 1567 memset(info, 0, (char *)inode - (char *)info); 1568 spin_lock_init(&info->lock); 1569 INIT_LIST_HEAD(&info->swaplist); 1570 1571 switch (mode & S_IFMT) { 1572 default: 1573 inode->i_op = &shmem_special_inode_operations; 1574 init_special_inode(inode, mode, dev); 1575 break; 1576 case S_IFREG: 1577 inode->i_op = &shmem_inode_operations; 1578 inode->i_fop = &shmem_file_operations; 1579 mpol_shared_policy_init(&info->policy, sbinfo->policy, 1580 &sbinfo->policy_nodes); 1581 break; 1582 case S_IFDIR: 1583 inc_nlink(inode); 1584 /* Some things misbehave if size == 0 on a directory */ 1585 inode->i_size = 2 * BOGO_DIRENT_SIZE; 1586 inode->i_op = &shmem_dir_inode_operations; 1587 inode->i_fop = &simple_dir_operations; 1588 break; 1589 case S_IFLNK: 1590 /* 1591 * Must not load anything in the rbtree, 1592 * mpol_free_shared_policy will not be called. 1593 */ 1594 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, 1595 NULL); 1596 break; 1597 } 1598 } else 1599 shmem_free_inode(sb); 1600 return inode; 1601} 1602 1603#ifdef CONFIG_TMPFS 1604static const struct inode_operations shmem_symlink_inode_operations; 1605static const struct inode_operations shmem_symlink_inline_operations; 1606 1607/* 1608 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin; 1609 * but providing them allows a tmpfs file to be used for splice, sendfile, and 1610 * below the loop driver, in the generic fashion that many filesystems support. 1611 */ 1612static int shmem_readpage(struct file *file, struct page *page) 1613{ 1614 struct inode *inode = page->mapping->host; 1615 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL); 1616 unlock_page(page); 1617 return error; 1618} 1619 1620static int 1621shmem_write_begin(struct file *file, struct address_space *mapping, 1622 loff_t pos, unsigned len, unsigned flags, 1623 struct page **pagep, void **fsdata) 1624{ 1625 struct inode *inode = mapping->host; 1626 pgoff_t index = pos >> PAGE_CACHE_SHIFT; 1627 *pagep = NULL; 1628 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL); 1629} 1630 1631static int 1632shmem_write_end(struct file *file, struct address_space *mapping, 1633 loff_t pos, unsigned len, unsigned copied, 1634 struct page *page, void *fsdata) 1635{ 1636 struct inode *inode = mapping->host; 1637 1638 if (pos + copied > inode->i_size) 1639 i_size_write(inode, pos + copied); 1640 1641 unlock_page(page); 1642 set_page_dirty(page); 1643 page_cache_release(page); 1644 1645 return copied; 1646} 1647 1648static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor) 1649{ 1650 struct inode *inode = filp->f_path.dentry->d_inode; 1651 struct address_space *mapping = inode->i_mapping; 1652 unsigned long index, offset; 1653 enum sgp_type sgp = SGP_READ; 1654 1655 /* 1656 * Might this read be for a stacking filesystem? Then when reading 1657 * holes of a sparse file, we actually need to allocate those pages, 1658 * and even mark them dirty, so it cannot exceed the max_blocks limit. 1659 */ 1660 if (segment_eq(get_fs(), KERNEL_DS)) 1661 sgp = SGP_DIRTY; 1662 1663 index = *ppos >> PAGE_CACHE_SHIFT; 1664 offset = *ppos & ~PAGE_CACHE_MASK; 1665 1666 for (;;) { 1667 struct page *page = NULL; 1668 unsigned long end_index, nr, ret; 1669 loff_t i_size = i_size_read(inode); 1670 1671 end_index = i_size >> PAGE_CACHE_SHIFT; 1672 if (index > end_index) 1673 break; 1674 if (index == end_index) { 1675 nr = i_size & ~PAGE_CACHE_MASK; 1676 if (nr <= offset) 1677 break; 1678 } 1679 1680 desc->error = shmem_getpage(inode, index, &page, sgp, NULL); 1681 if (desc->error) { 1682 if (desc->error == -EINVAL) 1683 desc->error = 0; 1684 break; 1685 } 1686 if (page) 1687 unlock_page(page); 1688 1689 /* 1690 * We must evaluate after, since reads (unlike writes) 1691 * are called without i_mutex protection against truncate 1692 */ 1693 nr = PAGE_CACHE_SIZE; 1694 i_size = i_size_read(inode); 1695 end_index = i_size >> PAGE_CACHE_SHIFT; 1696 if (index == end_index) { 1697 nr = i_size & ~PAGE_CACHE_MASK; 1698 if (nr <= offset) { 1699 if (page) 1700 page_cache_release(page); 1701 break; 1702 } 1703 } 1704 nr -= offset; 1705 1706 if (page) { 1707 /* 1708 * If users can be writing to this page using arbitrary 1709 * virtual addresses, take care about potential aliasing 1710 * before reading the page on the kernel side. 1711 */ 1712 if (mapping_writably_mapped(mapping)) 1713 flush_dcache_page(page); 1714 /* 1715 * Mark the page accessed if we read the beginning. 1716 */ 1717 if (!offset) 1718 mark_page_accessed(page); 1719 } else { 1720 page = ZERO_PAGE(0); 1721 page_cache_get(page); 1722 } 1723 1724 /* 1725 * Ok, we have the page, and it's up-to-date, so 1726 * now we can copy it to user space... 1727 * 1728 * The actor routine returns how many bytes were actually used.. 1729 * NOTE! This may not be the same as how much of a user buffer 1730 * we filled up (we may be padding etc), so we can only update 1731 * "pos" here (the actor routine has to update the user buffer 1732 * pointers and the remaining count). 1733 */ 1734 ret = actor(desc, page, offset, nr); 1735 offset += ret; 1736 index += offset >> PAGE_CACHE_SHIFT; 1737 offset &= ~PAGE_CACHE_MASK; 1738 1739 page_cache_release(page); 1740 if (ret != nr || !desc->count) 1741 break; 1742 1743 cond_resched(); 1744 } 1745 1746 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; 1747 file_accessed(filp); 1748} 1749 1750static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos) 1751{ 1752 read_descriptor_t desc; 1753 1754 if ((ssize_t) count < 0) 1755 return -EINVAL; 1756 if (!access_ok(VERIFY_WRITE, buf, count)) 1757 return -EFAULT; 1758 if (!count) 1759 return 0; 1760 1761 desc.written = 0; 1762 desc.count = count; 1763 desc.arg.buf = buf; 1764 desc.error = 0; 1765 1766 do_shmem_file_read(filp, ppos, &desc, file_read_actor); 1767 if (desc.written) 1768 return desc.written; 1769 return desc.error; 1770} 1771 1772static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) 1773{ 1774 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); 1775 1776 buf->f_type = TMPFS_MAGIC; 1777 buf->f_bsize = PAGE_CACHE_SIZE; 1778 buf->f_namelen = NAME_MAX; 1779 spin_lock(&sbinfo->stat_lock); 1780 if (sbinfo->max_blocks) { 1781 buf->f_blocks = sbinfo->max_blocks; 1782 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks; 1783 } 1784 if (sbinfo->max_inodes) { 1785 buf->f_files = sbinfo->max_inodes; 1786 buf->f_ffree = sbinfo->free_inodes; 1787 } 1788 /* else leave those fields 0 like simple_statfs */ 1789 spin_unlock(&sbinfo->stat_lock); 1790 return 0; 1791} 1792 1793/* 1794 * File creation. Allocate an inode, and we're done.. 1795 */ 1796static int 1797shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 1798{ 1799 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev); 1800 int error = -ENOSPC; 1801 1802 if (inode) { 1803 error = security_inode_init_security(inode, dir, NULL, NULL, 1804 NULL); 1805 if (error) { 1806 if (error != -EOPNOTSUPP) { 1807 iput(inode); 1808 return error; 1809 } 1810 } 1811 error = shmem_acl_init(inode, dir); 1812 if (error) { 1813 iput(inode); 1814 return error; 1815 } 1816 if (dir->i_mode & S_ISGID) { 1817 inode->i_gid = dir->i_gid; 1818 if (S_ISDIR(mode)) 1819 inode->i_mode |= S_ISGID; 1820 } 1821 dir->i_size += BOGO_DIRENT_SIZE; 1822 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1823 d_instantiate(dentry, inode); 1824 dget(dentry); /* Extra count - pin the dentry in core */ 1825 } 1826 return error; 1827} 1828 1829static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1830{ 1831 int error; 1832 1833 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) 1834 return error; 1835 inc_nlink(dir); 1836 return 0; 1837} 1838 1839static int shmem_create(struct inode *dir, struct dentry *dentry, int mode, 1840 struct nameidata *nd) 1841{ 1842 return shmem_mknod(dir, dentry, mode | S_IFREG, 0); 1843} 1844 1845/* 1846 * Link a file.. 1847 */ 1848static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1849{ 1850 struct inode *inode = old_dentry->d_inode; 1851 int ret; 1852 1853 /* 1854 * No ordinary (disk based) filesystem counts links as inodes; 1855 * but each new link needs a new dentry, pinning lowmem, and 1856 * tmpfs dentries cannot be pruned until they are unlinked. 1857 */ 1858 ret = shmem_reserve_inode(inode->i_sb); 1859 if (ret) 1860 goto out; 1861 1862 dir->i_size += BOGO_DIRENT_SIZE; 1863 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1864 inc_nlink(inode); 1865 atomic_inc(&inode->i_count); /* New dentry reference */ 1866 dget(dentry); /* Extra pinning count for the created dentry */ 1867 d_instantiate(dentry, inode); 1868out: 1869 return ret; 1870} 1871 1872static int shmem_unlink(struct inode *dir, struct dentry *dentry) 1873{ 1874 struct inode *inode = dentry->d_inode; 1875 1876 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) 1877 shmem_free_inode(inode->i_sb); 1878 1879 dir->i_size -= BOGO_DIRENT_SIZE; 1880 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1881 drop_nlink(inode); 1882 dput(dentry); /* Undo the count from "create" - this does all the work */ 1883 return 0; 1884} 1885 1886static int shmem_rmdir(struct inode *dir, struct dentry *dentry) 1887{ 1888 if (!simple_empty(dentry)) 1889 return -ENOTEMPTY; 1890 1891 drop_nlink(dentry->d_inode); 1892 drop_nlink(dir); 1893 return shmem_unlink(dir, dentry); 1894} 1895 1896/* 1897 * The VFS layer already does all the dentry stuff for rename, 1898 * we just have to decrement the usage count for the target if 1899 * it exists so that the VFS layer correctly free's it when it 1900 * gets overwritten. 1901 */ 1902static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) 1903{ 1904 struct inode *inode = old_dentry->d_inode; 1905 int they_are_dirs = S_ISDIR(inode->i_mode); 1906 1907 if (!simple_empty(new_dentry)) 1908 return -ENOTEMPTY; 1909 1910 if (new_dentry->d_inode) { 1911 (void) shmem_unlink(new_dir, new_dentry); 1912 if (they_are_dirs) 1913 drop_nlink(old_dir); 1914 } else if (they_are_dirs) { 1915 drop_nlink(old_dir); 1916 inc_nlink(new_dir); 1917 } 1918 1919 old_dir->i_size -= BOGO_DIRENT_SIZE; 1920 new_dir->i_size += BOGO_DIRENT_SIZE; 1921 old_dir->i_ctime = old_dir->i_mtime = 1922 new_dir->i_ctime = new_dir->i_mtime = 1923 inode->i_ctime = CURRENT_TIME; 1924 return 0; 1925} 1926 1927static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1928{ 1929 int error; 1930 int len; 1931 struct inode *inode; 1932 struct page *page = NULL; 1933 char *kaddr; 1934 struct shmem_inode_info *info; 1935 1936 len = strlen(symname) + 1; 1937 if (len > PAGE_CACHE_SIZE) 1938 return -ENAMETOOLONG; 1939 1940 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0); 1941 if (!inode) 1942 return -ENOSPC; 1943 1944 error = security_inode_init_security(inode, dir, NULL, NULL, 1945 NULL); 1946 if (error) { 1947 if (error != -EOPNOTSUPP) { 1948 iput(inode); 1949 return error; 1950 } 1951 error = 0; 1952 } 1953 1954 info = SHMEM_I(inode); 1955 inode->i_size = len-1; 1956 if (len <= (char *)inode - (char *)info) { 1957 /* do it inline */ 1958 memcpy(info, symname, len); 1959 inode->i_op = &shmem_symlink_inline_operations; 1960 } else { 1961 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); 1962 if (error) { 1963 iput(inode); 1964 return error; 1965 } 1966 unlock_page(page); 1967 inode->i_op = &shmem_symlink_inode_operations; 1968 kaddr = kmap_atomic(page, KM_USER0); 1969 memcpy(kaddr, symname, len); 1970 kunmap_atomic(kaddr, KM_USER0); 1971 set_page_dirty(page); 1972 page_cache_release(page); 1973 } 1974 if (dir->i_mode & S_ISGID) 1975 inode->i_gid = dir->i_gid; 1976 dir->i_size += BOGO_DIRENT_SIZE; 1977 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1978 d_instantiate(dentry, inode); 1979 dget(dentry); 1980 return 0; 1981} 1982 1983static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd) 1984{ 1985 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode)); 1986 return NULL; 1987} 1988 1989static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) 1990{ 1991 struct page *page = NULL; 1992 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); 1993 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page)); 1994 if (page) 1995 unlock_page(page); 1996 return page; 1997} 1998 1999static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 2000{ 2001 if (!IS_ERR(nd_get_link(nd))) { 2002 struct page *page = cookie; 2003 kunmap(page); 2004 mark_page_accessed(page); 2005 page_cache_release(page); 2006 } 2007} 2008 2009static const struct inode_operations shmem_symlink_inline_operations = { 2010 .readlink = generic_readlink, 2011 .follow_link = shmem_follow_link_inline, 2012}; 2013 2014static const struct inode_operations shmem_symlink_inode_operations = { 2015 .truncate = shmem_truncate, 2016 .readlink = generic_readlink, 2017 .follow_link = shmem_follow_link, 2018 .put_link = shmem_put_link, 2019}; 2020 2021#ifdef CONFIG_TMPFS_POSIX_ACL 2022/** 2023 * Superblocks without xattr inode operations will get security.* xattr 2024 * support from the VFS "for free". As soon as we have any other xattrs 2025 * like ACLs, we also need to implement the security.* handlers at 2026 * filesystem level, though. 2027 */ 2028 2029static size_t shmem_xattr_security_list(struct inode *inode, char *list, 2030 size_t list_len, const char *name, 2031 size_t name_len) 2032{ 2033 return security_inode_listsecurity(inode, list, list_len); 2034} 2035 2036static int shmem_xattr_security_get(struct inode *inode, const char *name, 2037 void *buffer, size_t size) 2038{ 2039 if (strcmp(name, "") == 0) 2040 return -EINVAL; 2041 return xattr_getsecurity(inode, name, buffer, size); 2042} 2043 2044static int shmem_xattr_security_set(struct inode *inode, const char *name, 2045 const void *value, size_t size, int flags) 2046{ 2047 if (strcmp(name, "") == 0) 2048 return -EINVAL; 2049 return security_inode_setsecurity(inode, name, value, size, flags); 2050} 2051 2052static struct xattr_handler shmem_xattr_security_handler = { 2053 .prefix = XATTR_SECURITY_PREFIX, 2054 .list = shmem_xattr_security_list, 2055 .get = shmem_xattr_security_get, 2056 .set = shmem_xattr_security_set, 2057}; 2058 2059static struct xattr_handler *shmem_xattr_handlers[] = { 2060 &shmem_xattr_acl_access_handler, 2061 &shmem_xattr_acl_default_handler, 2062 &shmem_xattr_security_handler, 2063 NULL 2064}; 2065#endif 2066 2067static struct dentry *shmem_get_parent(struct dentry *child) 2068{ 2069 return ERR_PTR(-ESTALE); 2070} 2071 2072static int shmem_match(struct inode *ino, void *vfh) 2073{ 2074 __u32 *fh = vfh; 2075 __u64 inum = fh[2]; 2076 inum = (inum << 32) | fh[1]; 2077 return ino->i_ino == inum && fh[0] == ino->i_generation; 2078} 2079 2080static struct dentry *shmem_fh_to_dentry(struct super_block *sb, 2081 struct fid *fid, int fh_len, int fh_type) 2082{ 2083 struct inode *inode; 2084 struct dentry *dentry = NULL; 2085 u64 inum = fid->raw[2]; 2086 inum = (inum << 32) | fid->raw[1]; 2087 2088 if (fh_len < 3) 2089 return NULL; 2090 2091 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), 2092 shmem_match, fid->raw); 2093 if (inode) { 2094 dentry = d_find_alias(inode); 2095 iput(inode); 2096 } 2097 2098 return dentry; 2099} 2100 2101static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len, 2102 int connectable) 2103{ 2104 struct inode *inode = dentry->d_inode; 2105 2106 if (*len < 3) 2107 return 255; 2108 2109 if (hlist_unhashed(&inode->i_hash)) { 2110 /* Unfortunately insert_inode_hash is not idempotent, 2111 * so as we hash inodes here rather than at creation 2112 * time, we need a lock to ensure we only try 2113 * to do it once 2114 */ 2115 static DEFINE_SPINLOCK(lock); 2116 spin_lock(&lock); 2117 if (hlist_unhashed(&inode->i_hash)) 2118 __insert_inode_hash(inode, 2119 inode->i_ino + inode->i_generation); 2120 spin_unlock(&lock); 2121 } 2122 2123 fh[0] = inode->i_generation; 2124 fh[1] = inode->i_ino; 2125 fh[2] = ((__u64)inode->i_ino) >> 32; 2126 2127 *len = 3; 2128 return 1; 2129} 2130 2131static const struct export_operations shmem_export_ops = { 2132 .get_parent = shmem_get_parent, 2133 .encode_fh = shmem_encode_fh, 2134 .fh_to_dentry = shmem_fh_to_dentry, 2135}; 2136 2137static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, 2138 bool remount) 2139{ 2140 char *this_char, *value, *rest; 2141 2142 while (options != NULL) { 2143 this_char = options; 2144 for (;;) { 2145 /* 2146 * NUL-terminate this option: unfortunately, 2147 * mount options form a comma-separated list, 2148 * but mpol's nodelist may also contain commas. 2149 */ 2150 options = strchr(options, ','); 2151 if (options == NULL) 2152 break; 2153 options++; 2154 if (!isdigit(*options)) { 2155 options[-1] = '\0'; 2156 break; 2157 } 2158 } 2159 if (!*this_char) 2160 continue; 2161 if ((value = strchr(this_char,'=')) != NULL) { 2162 *value++ = 0; 2163 } else { 2164 printk(KERN_ERR 2165 "tmpfs: No value for mount option '%s'\n", 2166 this_char); 2167 return 1; 2168 } 2169 2170 if (!strcmp(this_char,"size")) { 2171 unsigned long long size; 2172 size = memparse(value,&rest); 2173 if (*rest == '%') { 2174 size <<= PAGE_SHIFT; 2175 size *= totalram_pages; 2176 do_div(size, 100); 2177 rest++; 2178 } 2179 if (*rest) 2180 goto bad_val; 2181 sbinfo->max_blocks = 2182 DIV_ROUND_UP(size, PAGE_CACHE_SIZE); 2183 } else if (!strcmp(this_char,"nr_blocks")) { 2184 sbinfo->max_blocks = memparse(value, &rest); 2185 if (*rest) 2186 goto bad_val; 2187 } else if (!strcmp(this_char,"nr_inodes")) { 2188 sbinfo->max_inodes = memparse(value, &rest); 2189 if (*rest) 2190 goto bad_val; 2191 } else if (!strcmp(this_char,"mode")) { 2192 if (remount) 2193 continue; 2194 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; 2195 if (*rest) 2196 goto bad_val; 2197 } else if (!strcmp(this_char,"uid")) { 2198 if (remount) 2199 continue; 2200 sbinfo->uid = simple_strtoul(value, &rest, 0); 2201 if (*rest) 2202 goto bad_val; 2203 } else if (!strcmp(this_char,"gid")) { 2204 if (remount) 2205 continue; 2206 sbinfo->gid = simple_strtoul(value, &rest, 0); 2207 if (*rest) 2208 goto bad_val; 2209 } else if (!strcmp(this_char,"mpol")) { 2210 if (shmem_parse_mpol(value, &sbinfo->policy, 2211 &sbinfo->policy_nodes)) 2212 goto bad_val; 2213 } else { 2214 printk(KERN_ERR "tmpfs: Bad mount option %s\n", 2215 this_char); 2216 return 1; 2217 } 2218 } 2219 return 0; 2220 2221bad_val: 2222 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", 2223 value, this_char); 2224 return 1; 2225 2226} 2227 2228static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) 2229{ 2230 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 2231 struct shmem_sb_info config = *sbinfo; 2232 unsigned long blocks; 2233 unsigned long inodes; 2234 int error = -EINVAL; 2235 2236 if (shmem_parse_options(data, &config, true)) 2237 return error; 2238 2239 spin_lock(&sbinfo->stat_lock); 2240 blocks = sbinfo->max_blocks - sbinfo->free_blocks; 2241 inodes = sbinfo->max_inodes - sbinfo->free_inodes; 2242 if (config.max_blocks < blocks) 2243 goto out; 2244 if (config.max_inodes < inodes) 2245 goto out; 2246 /* 2247 * Those tests also disallow limited->unlimited while any are in 2248 * use, so i_blocks will always be zero when max_blocks is zero; 2249 * but we must separately disallow unlimited->limited, because 2250 * in that case we have no record of how much is already in use. 2251 */ 2252 if (config.max_blocks && !sbinfo->max_blocks) 2253 goto out; 2254 if (config.max_inodes && !sbinfo->max_inodes) 2255 goto out; 2256 2257 error = 0; 2258 sbinfo->max_blocks = config.max_blocks; 2259 sbinfo->free_blocks = config.max_blocks - blocks; 2260 sbinfo->max_inodes = config.max_inodes; 2261 sbinfo->free_inodes = config.max_inodes - inodes; 2262 sbinfo->policy = config.policy; 2263 sbinfo->policy_nodes = config.policy_nodes; 2264out: 2265 spin_unlock(&sbinfo->stat_lock); 2266 return error; 2267} 2268 2269static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs) 2270{ 2271 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb); 2272 2273 if (sbinfo->max_blocks != shmem_default_max_blocks()) 2274 seq_printf(seq, ",size=%luk", 2275 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10)); 2276 if (sbinfo->max_inodes != shmem_default_max_inodes()) 2277 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); 2278 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) 2279 seq_printf(seq, ",mode=%03o", sbinfo->mode); 2280 if (sbinfo->uid != 0) 2281 seq_printf(seq, ",uid=%u", sbinfo->uid); 2282 if (sbinfo->gid != 0) 2283 seq_printf(seq, ",gid=%u", sbinfo->gid); 2284 shmem_show_mpol(seq, sbinfo->policy, sbinfo->policy_nodes); 2285 return 0; 2286} 2287#endif /* CONFIG_TMPFS */ 2288 2289static void shmem_put_super(struct super_block *sb) 2290{ 2291 kfree(sb->s_fs_info); 2292 sb->s_fs_info = NULL; 2293} 2294 2295static int shmem_fill_super(struct super_block *sb, 2296 void *data, int silent) 2297{ 2298 struct inode *inode; 2299 struct dentry *root; 2300 struct shmem_sb_info *sbinfo; 2301 int err = -ENOMEM; 2302 2303 /* Round up to L1_CACHE_BYTES to resist false sharing */ 2304 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info), 2305 L1_CACHE_BYTES), GFP_KERNEL); 2306 if (!sbinfo) 2307 return -ENOMEM; 2308 2309 sbinfo->max_blocks = 0; 2310 sbinfo->max_inodes = 0; 2311 sbinfo->mode = S_IRWXUGO | S_ISVTX; 2312 sbinfo->uid = current->fsuid; 2313 sbinfo->gid = current->fsgid; 2314 sbinfo->policy = MPOL_DEFAULT; 2315 sbinfo->policy_nodes = node_states[N_HIGH_MEMORY]; 2316 sb->s_fs_info = sbinfo; 2317 2318#ifdef CONFIG_TMPFS 2319 /* 2320 * Per default we only allow half of the physical ram per 2321 * tmpfs instance, limiting inodes to one per page of lowmem; 2322 * but the internal instance is left unlimited. 2323 */ 2324 if (!(sb->s_flags & MS_NOUSER)) { 2325 sbinfo->max_blocks = shmem_default_max_blocks(); 2326 sbinfo->max_inodes = shmem_default_max_inodes(); 2327 if (shmem_parse_options(data, sbinfo, false)) { 2328 err = -EINVAL; 2329 goto failed; 2330 } 2331 } 2332 sb->s_export_op = &shmem_export_ops; 2333#else 2334 sb->s_flags |= MS_NOUSER; 2335#endif 2336 2337 spin_lock_init(&sbinfo->stat_lock); 2338 sbinfo->free_blocks = sbinfo->max_blocks; 2339 sbinfo->free_inodes = sbinfo->max_inodes; 2340 2341 sb->s_maxbytes = SHMEM_MAX_BYTES; 2342 sb->s_blocksize = PAGE_CACHE_SIZE; 2343 sb->s_blocksize_bits = PAGE_CACHE_SHIFT; 2344 sb->s_magic = TMPFS_MAGIC; 2345 sb->s_op = &shmem_ops; 2346 sb->s_time_gran = 1; 2347#ifdef CONFIG_TMPFS_POSIX_ACL 2348 sb->s_xattr = shmem_xattr_handlers; 2349 sb->s_flags |= MS_POSIXACL; 2350#endif 2351 2352 inode = shmem_get_inode(sb, S_IFDIR | sbinfo->mode, 0); 2353 if (!inode) 2354 goto failed; 2355 inode->i_uid = sbinfo->uid; 2356 inode->i_gid = sbinfo->gid; 2357 root = d_alloc_root(inode); 2358 if (!root) 2359 goto failed_iput; 2360 sb->s_root = root; 2361 return 0; 2362 2363failed_iput: 2364 iput(inode); 2365failed: 2366 shmem_put_super(sb); 2367 return err; 2368} 2369 2370static struct kmem_cache *shmem_inode_cachep; 2371 2372static struct inode *shmem_alloc_inode(struct super_block *sb) 2373{ 2374 struct shmem_inode_info *p; 2375 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); 2376 if (!p) 2377 return NULL; 2378 return &p->vfs_inode; 2379} 2380 2381static void shmem_destroy_inode(struct inode *inode) 2382{ 2383 if ((inode->i_mode & S_IFMT) == S_IFREG) { 2384 /* only struct inode is valid if it's an inline symlink */ 2385 mpol_free_shared_policy(&SHMEM_I(inode)->policy); 2386 } 2387 shmem_acl_destroy_inode(inode); 2388 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); 2389} 2390 2391static void init_once(struct kmem_cache *cachep, void *foo) 2392{ 2393 struct shmem_inode_info *p = (struct shmem_inode_info *) foo; 2394 2395 inode_init_once(&p->vfs_inode); 2396#ifdef CONFIG_TMPFS_POSIX_ACL 2397 p->i_acl = NULL; 2398 p->i_default_acl = NULL; 2399#endif 2400} 2401 2402static int init_inodecache(void) 2403{ 2404 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", 2405 sizeof(struct shmem_inode_info), 2406 0, SLAB_PANIC, init_once); 2407 return 0; 2408} 2409 2410static void destroy_inodecache(void) 2411{ 2412 kmem_cache_destroy(shmem_inode_cachep); 2413} 2414 2415static const struct address_space_operations shmem_aops = { 2416 .writepage = shmem_writepage, 2417 .set_page_dirty = __set_page_dirty_no_writeback, 2418#ifdef CONFIG_TMPFS 2419 .readpage = shmem_readpage, 2420 .write_begin = shmem_write_begin, 2421 .write_end = shmem_write_end, 2422#endif 2423 .migratepage = migrate_page, 2424}; 2425 2426static const struct file_operations shmem_file_operations = { 2427 .mmap = shmem_mmap, 2428#ifdef CONFIG_TMPFS 2429 .llseek = generic_file_llseek, 2430 .read = shmem_file_read, 2431 .write = do_sync_write, 2432 .aio_write = generic_file_aio_write, 2433 .fsync = simple_sync_file, 2434 .splice_read = generic_file_splice_read, 2435 .splice_write = generic_file_splice_write, 2436#endif 2437}; 2438 2439static const struct inode_operations shmem_inode_operations = { 2440 .truncate = shmem_truncate, 2441 .setattr = shmem_notify_change, 2442 .truncate_range = shmem_truncate_range, 2443#ifdef CONFIG_TMPFS_POSIX_ACL 2444 .setxattr = generic_setxattr, 2445 .getxattr = generic_getxattr, 2446 .listxattr = generic_listxattr, 2447 .removexattr = generic_removexattr, 2448 .permission = shmem_permission, 2449#endif 2450 2451}; 2452 2453static const struct inode_operations shmem_dir_inode_operations = { 2454#ifdef CONFIG_TMPFS 2455 .create = shmem_create, 2456 .lookup = simple_lookup, 2457 .link = shmem_link, 2458 .unlink = shmem_unlink, 2459 .symlink = shmem_symlink, 2460 .mkdir = shmem_mkdir, 2461 .rmdir = shmem_rmdir, 2462 .mknod = shmem_mknod, 2463 .rename = shmem_rename, 2464#endif 2465#ifdef CONFIG_TMPFS_POSIX_ACL 2466 .setattr = shmem_notify_change, 2467 .setxattr = generic_setxattr, 2468 .getxattr = generic_getxattr, 2469 .listxattr = generic_listxattr, 2470 .removexattr = generic_removexattr, 2471 .permission = shmem_permission, 2472#endif 2473}; 2474 2475static const struct inode_operations shmem_special_inode_operations = { 2476#ifdef CONFIG_TMPFS_POSIX_ACL 2477 .setattr = shmem_notify_change, 2478 .setxattr = generic_setxattr, 2479 .getxattr = generic_getxattr, 2480 .listxattr = generic_listxattr, 2481 .removexattr = generic_removexattr, 2482 .permission = shmem_permission, 2483#endif 2484}; 2485 2486static const struct super_operations shmem_ops = { 2487 .alloc_inode = shmem_alloc_inode, 2488 .destroy_inode = shmem_destroy_inode, 2489#ifdef CONFIG_TMPFS 2490 .statfs = shmem_statfs, 2491 .remount_fs = shmem_remount_fs, 2492 .show_options = shmem_show_options, 2493#endif 2494 .delete_inode = shmem_delete_inode, 2495 .drop_inode = generic_delete_inode, 2496 .put_super = shmem_put_super, 2497}; 2498 2499static struct vm_operations_struct shmem_vm_ops = { 2500 .fault = shmem_fault, 2501#ifdef CONFIG_NUMA 2502 .set_policy = shmem_set_policy, 2503 .get_policy = shmem_get_policy, 2504#endif 2505}; 2506 2507 2508static int shmem_get_sb(struct file_system_type *fs_type, 2509 int flags, const char *dev_name, void *data, struct vfsmount *mnt) 2510{ 2511 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt); 2512} 2513 2514static struct file_system_type tmpfs_fs_type = { 2515 .owner = THIS_MODULE, 2516 .name = "tmpfs", 2517 .get_sb = shmem_get_sb, 2518 .kill_sb = kill_litter_super, 2519}; 2520static struct vfsmount *shm_mnt; 2521 2522static int __init init_tmpfs(void) 2523{ 2524 int error; 2525 2526 error = bdi_init(&shmem_backing_dev_info); 2527 if (error) 2528 goto out4; 2529 2530 error = init_inodecache(); 2531 if (error) 2532 goto out3; 2533 2534 error = register_filesystem(&tmpfs_fs_type); 2535 if (error) { 2536 printk(KERN_ERR "Could not register tmpfs\n"); 2537 goto out2; 2538 } 2539 2540 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER, 2541 tmpfs_fs_type.name, NULL); 2542 if (IS_ERR(shm_mnt)) { 2543 error = PTR_ERR(shm_mnt); 2544 printk(KERN_ERR "Could not kern_mount tmpfs\n"); 2545 goto out1; 2546 } 2547 return 0; 2548 2549out1: 2550 unregister_filesystem(&tmpfs_fs_type); 2551out2: 2552 destroy_inodecache(); 2553out3: 2554 bdi_destroy(&shmem_backing_dev_info); 2555out4: 2556 shm_mnt = ERR_PTR(error); 2557 return error; 2558} 2559module_init(init_tmpfs) 2560 2561/* 2562 * shmem_file_setup - get an unlinked file living in tmpfs 2563 * 2564 * @name: name for dentry (to be seen in /proc/<pid>/maps 2565 * @size: size to be set for the file 2566 * 2567 */ 2568struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags) 2569{ 2570 int error; 2571 struct file *file; 2572 struct inode *inode; 2573 struct dentry *dentry, *root; 2574 struct qstr this; 2575 2576 if (IS_ERR(shm_mnt)) 2577 return (void *)shm_mnt; 2578 2579 if (size < 0 || size > SHMEM_MAX_BYTES) 2580 return ERR_PTR(-EINVAL); 2581 2582 if (shmem_acct_size(flags, size)) 2583 return ERR_PTR(-ENOMEM); 2584 2585 error = -ENOMEM; 2586 this.name = name; 2587 this.len = strlen(name); 2588 this.hash = 0; /* will go */ 2589 root = shm_mnt->mnt_root; 2590 dentry = d_alloc(root, &this); 2591 if (!dentry) 2592 goto put_memory; 2593 2594 error = -ENFILE; 2595 file = get_empty_filp(); 2596 if (!file) 2597 goto put_dentry; 2598 2599 error = -ENOSPC; 2600 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0); 2601 if (!inode) 2602 goto close_file; 2603 2604 SHMEM_I(inode)->flags = flags & VM_ACCOUNT; 2605 d_instantiate(dentry, inode); 2606 inode->i_size = size; 2607 inode->i_nlink = 0; /* It is unlinked */ 2608 init_file(file, shm_mnt, dentry, FMODE_WRITE | FMODE_READ, 2609 &shmem_file_operations); 2610 return file; 2611 2612close_file: 2613 put_filp(file); 2614put_dentry: 2615 dput(dentry); 2616put_memory: 2617 shmem_unacct_size(flags, size); 2618 return ERR_PTR(error); 2619} 2620 2621/* 2622 * shmem_zero_setup - setup a shared anonymous mapping 2623 * 2624 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff 2625 */ 2626int shmem_zero_setup(struct vm_area_struct *vma) 2627{ 2628 struct file *file; 2629 loff_t size = vma->vm_end - vma->vm_start; 2630 2631 file = shmem_file_setup("dev/zero", size, vma->vm_flags); 2632 if (IS_ERR(file)) 2633 return PTR_ERR(file); 2634 2635 if (vma->vm_file) 2636 fput(vma->vm_file); 2637 vma->vm_file = file; 2638 vma->vm_ops = &shmem_vm_ops; 2639 return 0; 2640} 2641