shmem.c revision f0be3d32b05d3fea2fcdbbb81a39dac2a7163169
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 * @inode: inode to recalc 250 * 251 * We have to calculate the free blocks since the mm can drop 252 * undirtied hole pages behind our back. 253 * 254 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped 255 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) 256 * 257 * It has to be called with the spinlock held. 258 */ 259static void shmem_recalc_inode(struct inode *inode) 260{ 261 struct shmem_inode_info *info = SHMEM_I(inode); 262 long freed; 263 264 freed = info->alloced - info->swapped - inode->i_mapping->nrpages; 265 if (freed > 0) { 266 info->alloced -= freed; 267 shmem_unacct_blocks(info->flags, freed); 268 shmem_free_blocks(inode, freed); 269 } 270} 271 272/** 273 * shmem_swp_entry - find the swap vector position in the info structure 274 * @info: info structure for the inode 275 * @index: index of the page to find 276 * @page: optional page to add to the structure. Has to be preset to 277 * all zeros 278 * 279 * If there is no space allocated yet it will return NULL when 280 * page is NULL, else it will use the page for the needed block, 281 * setting it to NULL on return to indicate that it has been used. 282 * 283 * The swap vector is organized the following way: 284 * 285 * There are SHMEM_NR_DIRECT entries directly stored in the 286 * shmem_inode_info structure. So small files do not need an addional 287 * allocation. 288 * 289 * For pages with index > SHMEM_NR_DIRECT there is the pointer 290 * i_indirect which points to a page which holds in the first half 291 * doubly indirect blocks, in the second half triple indirect blocks: 292 * 293 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the 294 * following layout (for SHMEM_NR_DIRECT == 16): 295 * 296 * i_indirect -> dir --> 16-19 297 * | +-> 20-23 298 * | 299 * +-->dir2 --> 24-27 300 * | +-> 28-31 301 * | +-> 32-35 302 * | +-> 36-39 303 * | 304 * +-->dir3 --> 40-43 305 * +-> 44-47 306 * +-> 48-51 307 * +-> 52-55 308 */ 309static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page) 310{ 311 unsigned long offset; 312 struct page **dir; 313 struct page *subdir; 314 315 if (index < SHMEM_NR_DIRECT) { 316 shmem_swp_balance_unmap(); 317 return info->i_direct+index; 318 } 319 if (!info->i_indirect) { 320 if (page) { 321 info->i_indirect = *page; 322 *page = NULL; 323 } 324 return NULL; /* need another page */ 325 } 326 327 index -= SHMEM_NR_DIRECT; 328 offset = index % ENTRIES_PER_PAGE; 329 index /= ENTRIES_PER_PAGE; 330 dir = shmem_dir_map(info->i_indirect); 331 332 if (index >= ENTRIES_PER_PAGE/2) { 333 index -= ENTRIES_PER_PAGE/2; 334 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE; 335 index %= ENTRIES_PER_PAGE; 336 subdir = *dir; 337 if (!subdir) { 338 if (page) { 339 *dir = *page; 340 *page = NULL; 341 } 342 shmem_dir_unmap(dir); 343 return NULL; /* need another page */ 344 } 345 shmem_dir_unmap(dir); 346 dir = shmem_dir_map(subdir); 347 } 348 349 dir += index; 350 subdir = *dir; 351 if (!subdir) { 352 if (!page || !(subdir = *page)) { 353 shmem_dir_unmap(dir); 354 return NULL; /* need a page */ 355 } 356 *dir = subdir; 357 *page = NULL; 358 } 359 shmem_dir_unmap(dir); 360 return shmem_swp_map(subdir) + offset; 361} 362 363static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value) 364{ 365 long incdec = value? 1: -1; 366 367 entry->val = value; 368 info->swapped += incdec; 369 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) { 370 struct page *page = kmap_atomic_to_page(entry); 371 set_page_private(page, page_private(page) + incdec); 372 } 373} 374 375/** 376 * shmem_swp_alloc - get the position of the swap entry for the page. 377 * @info: info structure for the inode 378 * @index: index of the page to find 379 * @sgp: check and recheck i_size? skip allocation? 380 * 381 * If the entry does not exist, allocate it. 382 */ 383static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp) 384{ 385 struct inode *inode = &info->vfs_inode; 386 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 387 struct page *page = NULL; 388 swp_entry_t *entry; 389 390 if (sgp != SGP_WRITE && 391 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 392 return ERR_PTR(-EINVAL); 393 394 while (!(entry = shmem_swp_entry(info, index, &page))) { 395 if (sgp == SGP_READ) 396 return shmem_swp_map(ZERO_PAGE(0)); 397 /* 398 * Test free_blocks against 1 not 0, since we have 1 data 399 * page (and perhaps indirect index pages) yet to allocate: 400 * a waste to allocate index if we cannot allocate data. 401 */ 402 if (sbinfo->max_blocks) { 403 spin_lock(&sbinfo->stat_lock); 404 if (sbinfo->free_blocks <= 1) { 405 spin_unlock(&sbinfo->stat_lock); 406 return ERR_PTR(-ENOSPC); 407 } 408 sbinfo->free_blocks--; 409 inode->i_blocks += BLOCKS_PER_PAGE; 410 spin_unlock(&sbinfo->stat_lock); 411 } 412 413 spin_unlock(&info->lock); 414 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping)); 415 if (page) 416 set_page_private(page, 0); 417 spin_lock(&info->lock); 418 419 if (!page) { 420 shmem_free_blocks(inode, 1); 421 return ERR_PTR(-ENOMEM); 422 } 423 if (sgp != SGP_WRITE && 424 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { 425 entry = ERR_PTR(-EINVAL); 426 break; 427 } 428 if (info->next_index <= index) 429 info->next_index = index + 1; 430 } 431 if (page) { 432 /* another task gave its page, or truncated the file */ 433 shmem_free_blocks(inode, 1); 434 shmem_dir_free(page); 435 } 436 if (info->next_index <= index && !IS_ERR(entry)) 437 info->next_index = index + 1; 438 return entry; 439} 440 441/** 442 * shmem_free_swp - free some swap entries in a directory 443 * @dir: pointer to the directory 444 * @edir: pointer after last entry of the directory 445 * @punch_lock: pointer to spinlock when needed for the holepunch case 446 */ 447static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir, 448 spinlock_t *punch_lock) 449{ 450 spinlock_t *punch_unlock = NULL; 451 swp_entry_t *ptr; 452 int freed = 0; 453 454 for (ptr = dir; ptr < edir; ptr++) { 455 if (ptr->val) { 456 if (unlikely(punch_lock)) { 457 punch_unlock = punch_lock; 458 punch_lock = NULL; 459 spin_lock(punch_unlock); 460 if (!ptr->val) 461 continue; 462 } 463 free_swap_and_cache(*ptr); 464 *ptr = (swp_entry_t){0}; 465 freed++; 466 } 467 } 468 if (punch_unlock) 469 spin_unlock(punch_unlock); 470 return freed; 471} 472 473static int shmem_map_and_free_swp(struct page *subdir, int offset, 474 int limit, struct page ***dir, spinlock_t *punch_lock) 475{ 476 swp_entry_t *ptr; 477 int freed = 0; 478 479 ptr = shmem_swp_map(subdir); 480 for (; offset < limit; offset += LATENCY_LIMIT) { 481 int size = limit - offset; 482 if (size > LATENCY_LIMIT) 483 size = LATENCY_LIMIT; 484 freed += shmem_free_swp(ptr+offset, ptr+offset+size, 485 punch_lock); 486 if (need_resched()) { 487 shmem_swp_unmap(ptr); 488 if (*dir) { 489 shmem_dir_unmap(*dir); 490 *dir = NULL; 491 } 492 cond_resched(); 493 ptr = shmem_swp_map(subdir); 494 } 495 } 496 shmem_swp_unmap(ptr); 497 return freed; 498} 499 500static void shmem_free_pages(struct list_head *next) 501{ 502 struct page *page; 503 int freed = 0; 504 505 do { 506 page = container_of(next, struct page, lru); 507 next = next->next; 508 shmem_dir_free(page); 509 freed++; 510 if (freed >= LATENCY_LIMIT) { 511 cond_resched(); 512 freed = 0; 513 } 514 } while (next); 515} 516 517static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end) 518{ 519 struct shmem_inode_info *info = SHMEM_I(inode); 520 unsigned long idx; 521 unsigned long size; 522 unsigned long limit; 523 unsigned long stage; 524 unsigned long diroff; 525 struct page **dir; 526 struct page *topdir; 527 struct page *middir; 528 struct page *subdir; 529 swp_entry_t *ptr; 530 LIST_HEAD(pages_to_free); 531 long nr_pages_to_free = 0; 532 long nr_swaps_freed = 0; 533 int offset; 534 int freed; 535 int punch_hole; 536 spinlock_t *needs_lock; 537 spinlock_t *punch_lock; 538 unsigned long upper_limit; 539 540 inode->i_ctime = inode->i_mtime = CURRENT_TIME; 541 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 542 if (idx >= info->next_index) 543 return; 544 545 spin_lock(&info->lock); 546 info->flags |= SHMEM_TRUNCATE; 547 if (likely(end == (loff_t) -1)) { 548 limit = info->next_index; 549 upper_limit = SHMEM_MAX_INDEX; 550 info->next_index = idx; 551 needs_lock = NULL; 552 punch_hole = 0; 553 } else { 554 if (end + 1 >= inode->i_size) { /* we may free a little more */ 555 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >> 556 PAGE_CACHE_SHIFT; 557 upper_limit = SHMEM_MAX_INDEX; 558 } else { 559 limit = (end + 1) >> PAGE_CACHE_SHIFT; 560 upper_limit = limit; 561 } 562 needs_lock = &info->lock; 563 punch_hole = 1; 564 } 565 566 topdir = info->i_indirect; 567 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) { 568 info->i_indirect = NULL; 569 nr_pages_to_free++; 570 list_add(&topdir->lru, &pages_to_free); 571 } 572 spin_unlock(&info->lock); 573 574 if (info->swapped && idx < SHMEM_NR_DIRECT) { 575 ptr = info->i_direct; 576 size = limit; 577 if (size > SHMEM_NR_DIRECT) 578 size = SHMEM_NR_DIRECT; 579 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock); 580 } 581 582 /* 583 * If there are no indirect blocks or we are punching a hole 584 * below indirect blocks, nothing to be done. 585 */ 586 if (!topdir || limit <= SHMEM_NR_DIRECT) 587 goto done2; 588 589 /* 590 * The truncation case has already dropped info->lock, and we're safe 591 * because i_size and next_index have already been lowered, preventing 592 * access beyond. But in the punch_hole case, we still need to take 593 * the lock when updating the swap directory, because there might be 594 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or 595 * shmem_writepage. However, whenever we find we can remove a whole 596 * directory page (not at the misaligned start or end of the range), 597 * we first NULLify its pointer in the level above, and then have no 598 * need to take the lock when updating its contents: needs_lock and 599 * punch_lock (either pointing to info->lock or NULL) manage this. 600 */ 601 602 upper_limit -= SHMEM_NR_DIRECT; 603 limit -= SHMEM_NR_DIRECT; 604 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0; 605 offset = idx % ENTRIES_PER_PAGE; 606 idx -= offset; 607 608 dir = shmem_dir_map(topdir); 609 stage = ENTRIES_PER_PAGEPAGE/2; 610 if (idx < ENTRIES_PER_PAGEPAGE/2) { 611 middir = topdir; 612 diroff = idx/ENTRIES_PER_PAGE; 613 } else { 614 dir += ENTRIES_PER_PAGE/2; 615 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE; 616 while (stage <= idx) 617 stage += ENTRIES_PER_PAGEPAGE; 618 middir = *dir; 619 if (*dir) { 620 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) % 621 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE; 622 if (!diroff && !offset && upper_limit >= stage) { 623 if (needs_lock) { 624 spin_lock(needs_lock); 625 *dir = NULL; 626 spin_unlock(needs_lock); 627 needs_lock = NULL; 628 } else 629 *dir = NULL; 630 nr_pages_to_free++; 631 list_add(&middir->lru, &pages_to_free); 632 } 633 shmem_dir_unmap(dir); 634 dir = shmem_dir_map(middir); 635 } else { 636 diroff = 0; 637 offset = 0; 638 idx = stage; 639 } 640 } 641 642 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) { 643 if (unlikely(idx == stage)) { 644 shmem_dir_unmap(dir); 645 dir = shmem_dir_map(topdir) + 646 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 647 while (!*dir) { 648 dir++; 649 idx += ENTRIES_PER_PAGEPAGE; 650 if (idx >= limit) 651 goto done1; 652 } 653 stage = idx + ENTRIES_PER_PAGEPAGE; 654 middir = *dir; 655 if (punch_hole) 656 needs_lock = &info->lock; 657 if (upper_limit >= stage) { 658 if (needs_lock) { 659 spin_lock(needs_lock); 660 *dir = NULL; 661 spin_unlock(needs_lock); 662 needs_lock = NULL; 663 } else 664 *dir = NULL; 665 nr_pages_to_free++; 666 list_add(&middir->lru, &pages_to_free); 667 } 668 shmem_dir_unmap(dir); 669 cond_resched(); 670 dir = shmem_dir_map(middir); 671 diroff = 0; 672 } 673 punch_lock = needs_lock; 674 subdir = dir[diroff]; 675 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) { 676 if (needs_lock) { 677 spin_lock(needs_lock); 678 dir[diroff] = NULL; 679 spin_unlock(needs_lock); 680 punch_lock = NULL; 681 } else 682 dir[diroff] = NULL; 683 nr_pages_to_free++; 684 list_add(&subdir->lru, &pages_to_free); 685 } 686 if (subdir && page_private(subdir) /* has swap entries */) { 687 size = limit - idx; 688 if (size > ENTRIES_PER_PAGE) 689 size = ENTRIES_PER_PAGE; 690 freed = shmem_map_and_free_swp(subdir, 691 offset, size, &dir, punch_lock); 692 if (!dir) 693 dir = shmem_dir_map(middir); 694 nr_swaps_freed += freed; 695 if (offset || punch_lock) { 696 spin_lock(&info->lock); 697 set_page_private(subdir, 698 page_private(subdir) - freed); 699 spin_unlock(&info->lock); 700 } else 701 BUG_ON(page_private(subdir) != freed); 702 } 703 offset = 0; 704 } 705done1: 706 shmem_dir_unmap(dir); 707done2: 708 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) { 709 /* 710 * Call truncate_inode_pages again: racing shmem_unuse_inode 711 * may have swizzled a page in from swap since vmtruncate or 712 * generic_delete_inode did it, before we lowered next_index. 713 * Also, though shmem_getpage checks i_size before adding to 714 * cache, no recheck after: so fix the narrow window there too. 715 * 716 * Recalling truncate_inode_pages_range and unmap_mapping_range 717 * every time for punch_hole (which never got a chance to clear 718 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive, 719 * yet hardly ever necessary: try to optimize them out later. 720 */ 721 truncate_inode_pages_range(inode->i_mapping, start, end); 722 if (punch_hole) 723 unmap_mapping_range(inode->i_mapping, start, 724 end - start, 1); 725 } 726 727 spin_lock(&info->lock); 728 info->flags &= ~SHMEM_TRUNCATE; 729 info->swapped -= nr_swaps_freed; 730 if (nr_pages_to_free) 731 shmem_free_blocks(inode, nr_pages_to_free); 732 shmem_recalc_inode(inode); 733 spin_unlock(&info->lock); 734 735 /* 736 * Empty swap vector directory pages to be freed? 737 */ 738 if (!list_empty(&pages_to_free)) { 739 pages_to_free.prev->next = NULL; 740 shmem_free_pages(pages_to_free.next); 741 } 742} 743 744static void shmem_truncate(struct inode *inode) 745{ 746 shmem_truncate_range(inode, inode->i_size, (loff_t)-1); 747} 748 749static int shmem_notify_change(struct dentry *dentry, struct iattr *attr) 750{ 751 struct inode *inode = dentry->d_inode; 752 struct page *page = NULL; 753 int error; 754 755 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { 756 if (attr->ia_size < inode->i_size) { 757 /* 758 * If truncating down to a partial page, then 759 * if that page is already allocated, hold it 760 * in memory until the truncation is over, so 761 * truncate_partial_page cannnot miss it were 762 * it assigned to swap. 763 */ 764 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) { 765 (void) shmem_getpage(inode, 766 attr->ia_size>>PAGE_CACHE_SHIFT, 767 &page, SGP_READ, NULL); 768 if (page) 769 unlock_page(page); 770 } 771 /* 772 * Reset SHMEM_PAGEIN flag so that shmem_truncate can 773 * detect if any pages might have been added to cache 774 * after truncate_inode_pages. But we needn't bother 775 * if it's being fully truncated to zero-length: the 776 * nrpages check is efficient enough in that case. 777 */ 778 if (attr->ia_size) { 779 struct shmem_inode_info *info = SHMEM_I(inode); 780 spin_lock(&info->lock); 781 info->flags &= ~SHMEM_PAGEIN; 782 spin_unlock(&info->lock); 783 } 784 } 785 } 786 787 error = inode_change_ok(inode, attr); 788 if (!error) 789 error = inode_setattr(inode, attr); 790#ifdef CONFIG_TMPFS_POSIX_ACL 791 if (!error && (attr->ia_valid & ATTR_MODE)) 792 error = generic_acl_chmod(inode, &shmem_acl_ops); 793#endif 794 if (page) 795 page_cache_release(page); 796 return error; 797} 798 799static void shmem_delete_inode(struct inode *inode) 800{ 801 struct shmem_inode_info *info = SHMEM_I(inode); 802 803 if (inode->i_op->truncate == shmem_truncate) { 804 truncate_inode_pages(inode->i_mapping, 0); 805 shmem_unacct_size(info->flags, inode->i_size); 806 inode->i_size = 0; 807 shmem_truncate(inode); 808 if (!list_empty(&info->swaplist)) { 809 mutex_lock(&shmem_swaplist_mutex); 810 list_del_init(&info->swaplist); 811 mutex_unlock(&shmem_swaplist_mutex); 812 } 813 } 814 BUG_ON(inode->i_blocks); 815 shmem_free_inode(inode->i_sb); 816 clear_inode(inode); 817} 818 819static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir) 820{ 821 swp_entry_t *ptr; 822 823 for (ptr = dir; ptr < edir; ptr++) { 824 if (ptr->val == entry.val) 825 return ptr - dir; 826 } 827 return -1; 828} 829 830static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page) 831{ 832 struct inode *inode; 833 unsigned long idx; 834 unsigned long size; 835 unsigned long limit; 836 unsigned long stage; 837 struct page **dir; 838 struct page *subdir; 839 swp_entry_t *ptr; 840 int offset; 841 int error; 842 843 idx = 0; 844 ptr = info->i_direct; 845 spin_lock(&info->lock); 846 if (!info->swapped) { 847 list_del_init(&info->swaplist); 848 goto lost2; 849 } 850 limit = info->next_index; 851 size = limit; 852 if (size > SHMEM_NR_DIRECT) 853 size = SHMEM_NR_DIRECT; 854 offset = shmem_find_swp(entry, ptr, ptr+size); 855 if (offset >= 0) 856 goto found; 857 if (!info->i_indirect) 858 goto lost2; 859 860 dir = shmem_dir_map(info->i_indirect); 861 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2; 862 863 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) { 864 if (unlikely(idx == stage)) { 865 shmem_dir_unmap(dir-1); 866 if (cond_resched_lock(&info->lock)) { 867 /* check it has not been truncated */ 868 if (limit > info->next_index) { 869 limit = info->next_index; 870 if (idx >= limit) 871 goto lost2; 872 } 873 } 874 dir = shmem_dir_map(info->i_indirect) + 875 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 876 while (!*dir) { 877 dir++; 878 idx += ENTRIES_PER_PAGEPAGE; 879 if (idx >= limit) 880 goto lost1; 881 } 882 stage = idx + ENTRIES_PER_PAGEPAGE; 883 subdir = *dir; 884 shmem_dir_unmap(dir); 885 dir = shmem_dir_map(subdir); 886 } 887 subdir = *dir; 888 if (subdir && page_private(subdir)) { 889 ptr = shmem_swp_map(subdir); 890 size = limit - idx; 891 if (size > ENTRIES_PER_PAGE) 892 size = ENTRIES_PER_PAGE; 893 offset = shmem_find_swp(entry, ptr, ptr+size); 894 shmem_swp_unmap(ptr); 895 if (offset >= 0) { 896 shmem_dir_unmap(dir); 897 goto found; 898 } 899 } 900 } 901lost1: 902 shmem_dir_unmap(dir-1); 903lost2: 904 spin_unlock(&info->lock); 905 return 0; 906found: 907 idx += offset; 908 inode = igrab(&info->vfs_inode); 909 spin_unlock(&info->lock); 910 911 /* 912 * Move _head_ to start search for next from here. 913 * But be careful: shmem_delete_inode checks list_empty without taking 914 * mutex, and there's an instant in list_move_tail when info->swaplist 915 * would appear empty, if it were the only one on shmem_swaplist. We 916 * could avoid doing it if inode NULL; or use this minor optimization. 917 */ 918 if (shmem_swaplist.next != &info->swaplist) 919 list_move_tail(&shmem_swaplist, &info->swaplist); 920 mutex_unlock(&shmem_swaplist_mutex); 921 922 error = 1; 923 if (!inode) 924 goto out; 925 /* Precharge page while we can wait, compensate afterwards */ 926 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL); 927 if (error) 928 goto out; 929 error = radix_tree_preload(GFP_KERNEL); 930 if (error) 931 goto uncharge; 932 error = 1; 933 934 spin_lock(&info->lock); 935 ptr = shmem_swp_entry(info, idx, NULL); 936 if (ptr && ptr->val == entry.val) 937 error = add_to_page_cache(page, inode->i_mapping, 938 idx, GFP_NOWAIT); 939 if (error == -EEXIST) { 940 struct page *filepage = find_get_page(inode->i_mapping, idx); 941 error = 1; 942 if (filepage) { 943 /* 944 * There might be a more uptodate page coming down 945 * from a stacked writepage: forget our swappage if so. 946 */ 947 if (PageUptodate(filepage)) 948 error = 0; 949 page_cache_release(filepage); 950 } 951 } 952 if (!error) { 953 delete_from_swap_cache(page); 954 set_page_dirty(page); 955 info->flags |= SHMEM_PAGEIN; 956 shmem_swp_set(info, ptr, 0); 957 swap_free(entry); 958 error = 1; /* not an error, but entry was found */ 959 } 960 if (ptr) 961 shmem_swp_unmap(ptr); 962 spin_unlock(&info->lock); 963 radix_tree_preload_end(); 964uncharge: 965 mem_cgroup_uncharge_page(page); 966out: 967 unlock_page(page); 968 page_cache_release(page); 969 iput(inode); /* allows for NULL */ 970 return error; 971} 972 973/* 974 * shmem_unuse() search for an eventually swapped out shmem page. 975 */ 976int shmem_unuse(swp_entry_t entry, struct page *page) 977{ 978 struct list_head *p, *next; 979 struct shmem_inode_info *info; 980 int found = 0; 981 982 mutex_lock(&shmem_swaplist_mutex); 983 list_for_each_safe(p, next, &shmem_swaplist) { 984 info = list_entry(p, struct shmem_inode_info, swaplist); 985 found = shmem_unuse_inode(info, entry, page); 986 cond_resched(); 987 if (found) 988 goto out; 989 } 990 mutex_unlock(&shmem_swaplist_mutex); 991out: return found; /* 0 or 1 or -ENOMEM */ 992} 993 994/* 995 * Move the page from the page cache to the swap cache. 996 */ 997static int shmem_writepage(struct page *page, struct writeback_control *wbc) 998{ 999 struct shmem_inode_info *info; 1000 swp_entry_t *entry, swap; 1001 struct address_space *mapping; 1002 unsigned long index; 1003 struct inode *inode; 1004 1005 BUG_ON(!PageLocked(page)); 1006 mapping = page->mapping; 1007 index = page->index; 1008 inode = mapping->host; 1009 info = SHMEM_I(inode); 1010 if (info->flags & VM_LOCKED) 1011 goto redirty; 1012 if (!total_swap_pages) 1013 goto redirty; 1014 1015 /* 1016 * shmem_backing_dev_info's capabilities prevent regular writeback or 1017 * sync from ever calling shmem_writepage; but a stacking filesystem 1018 * may use the ->writepage of its underlying filesystem, in which case 1019 * tmpfs should write out to swap only in response to memory pressure, 1020 * and not for pdflush or sync. However, in those cases, we do still 1021 * want to check if there's a redundant swappage to be discarded. 1022 */ 1023 if (wbc->for_reclaim) 1024 swap = get_swap_page(); 1025 else 1026 swap.val = 0; 1027 1028 spin_lock(&info->lock); 1029 if (index >= info->next_index) { 1030 BUG_ON(!(info->flags & SHMEM_TRUNCATE)); 1031 goto unlock; 1032 } 1033 entry = shmem_swp_entry(info, index, NULL); 1034 if (entry->val) { 1035 /* 1036 * The more uptodate page coming down from a stacked 1037 * writepage should replace our old swappage. 1038 */ 1039 free_swap_and_cache(*entry); 1040 shmem_swp_set(info, entry, 0); 1041 } 1042 shmem_recalc_inode(inode); 1043 1044 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { 1045 remove_from_page_cache(page); 1046 shmem_swp_set(info, entry, swap.val); 1047 shmem_swp_unmap(entry); 1048 if (list_empty(&info->swaplist)) 1049 inode = igrab(inode); 1050 else 1051 inode = NULL; 1052 spin_unlock(&info->lock); 1053 swap_duplicate(swap); 1054 BUG_ON(page_mapped(page)); 1055 page_cache_release(page); /* pagecache ref */ 1056 set_page_dirty(page); 1057 unlock_page(page); 1058 if (inode) { 1059 mutex_lock(&shmem_swaplist_mutex); 1060 /* move instead of add in case we're racing */ 1061 list_move_tail(&info->swaplist, &shmem_swaplist); 1062 mutex_unlock(&shmem_swaplist_mutex); 1063 iput(inode); 1064 } 1065 return 0; 1066 } 1067 1068 shmem_swp_unmap(entry); 1069unlock: 1070 spin_unlock(&info->lock); 1071 swap_free(swap); 1072redirty: 1073 set_page_dirty(page); 1074 if (wbc->for_reclaim) 1075 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ 1076 unlock_page(page); 1077 return 0; 1078} 1079 1080#ifdef CONFIG_NUMA 1081#ifdef CONFIG_TMPFS 1082static int shmem_parse_mpol(char *value, unsigned short *policy, 1083 unsigned short *mode_flags, nodemask_t *policy_nodes) 1084{ 1085 char *nodelist = strchr(value, ':'); 1086 char *flags = strchr(value, '='); 1087 int err = 1; 1088 1089 if (nodelist) { 1090 /* NUL-terminate policy string */ 1091 *nodelist++ = '\0'; 1092 if (nodelist_parse(nodelist, *policy_nodes)) 1093 goto out; 1094 if (!nodes_subset(*policy_nodes, node_states[N_HIGH_MEMORY])) 1095 goto out; 1096 } 1097 if (flags) 1098 *flags++ = '\0'; 1099 if (!strcmp(value, "default")) { 1100 *policy = MPOL_DEFAULT; 1101 /* Don't allow a nodelist */ 1102 if (!nodelist) 1103 err = 0; 1104 } else if (!strcmp(value, "prefer")) { 1105 *policy = MPOL_PREFERRED; 1106 /* Insist on a nodelist of one node only */ 1107 if (nodelist) { 1108 char *rest = nodelist; 1109 while (isdigit(*rest)) 1110 rest++; 1111 if (!*rest) 1112 err = 0; 1113 } 1114 } else if (!strcmp(value, "bind")) { 1115 *policy = MPOL_BIND; 1116 /* Insist on a nodelist */ 1117 if (nodelist) 1118 err = 0; 1119 } else if (!strcmp(value, "interleave")) { 1120 *policy = MPOL_INTERLEAVE; 1121 /* 1122 * Default to online nodes with memory if no nodelist 1123 */ 1124 if (!nodelist) 1125 *policy_nodes = node_states[N_HIGH_MEMORY]; 1126 err = 0; 1127 } 1128 1129 *mode_flags = 0; 1130 if (flags) { 1131 /* 1132 * Currently, we only support two mutually exclusive 1133 * mode flags. 1134 */ 1135 if (!strcmp(flags, "static")) 1136 *mode_flags |= MPOL_F_STATIC_NODES; 1137 else if (!strcmp(flags, "relative")) 1138 *mode_flags |= MPOL_F_RELATIVE_NODES; 1139 else 1140 err = 1; /* unrecognized flag */ 1141 } 1142out: 1143 /* Restore string for error message */ 1144 if (nodelist) 1145 *--nodelist = ':'; 1146 if (flags) 1147 *--flags = '='; 1148 return err; 1149} 1150 1151static void shmem_show_mpol(struct seq_file *seq, unsigned short policy, 1152 unsigned short flags, const nodemask_t policy_nodes) 1153{ 1154 char *policy_string; 1155 1156 switch (policy) { 1157 case MPOL_PREFERRED: 1158 policy_string = "prefer"; 1159 break; 1160 case MPOL_BIND: 1161 policy_string = "bind"; 1162 break; 1163 case MPOL_INTERLEAVE: 1164 policy_string = "interleave"; 1165 break; 1166 default: 1167 /* MPOL_DEFAULT */ 1168 return; 1169 } 1170 1171 seq_printf(seq, ",mpol=%s", policy_string); 1172 1173 if (policy != MPOL_INTERLEAVE || 1174 !nodes_equal(policy_nodes, node_states[N_HIGH_MEMORY])) { 1175 char buffer[64]; 1176 int len; 1177 1178 len = nodelist_scnprintf(buffer, sizeof(buffer), policy_nodes); 1179 if (len < sizeof(buffer)) 1180 seq_printf(seq, ":%s", buffer); 1181 else 1182 seq_printf(seq, ":?"); 1183 } 1184} 1185#endif /* CONFIG_TMPFS */ 1186 1187static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, 1188 struct shmem_inode_info *info, unsigned long idx) 1189{ 1190 struct vm_area_struct pvma; 1191 struct page *page; 1192 1193 /* Create a pseudo vma that just contains the policy */ 1194 pvma.vm_start = 0; 1195 pvma.vm_pgoff = idx; 1196 pvma.vm_ops = NULL; 1197 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx); 1198 page = swapin_readahead(entry, gfp, &pvma, 0); 1199 mpol_put(pvma.vm_policy); 1200 return page; 1201} 1202 1203static struct page *shmem_alloc_page(gfp_t gfp, 1204 struct shmem_inode_info *info, unsigned long idx) 1205{ 1206 struct vm_area_struct pvma; 1207 struct page *page; 1208 1209 /* Create a pseudo vma that just contains the policy */ 1210 pvma.vm_start = 0; 1211 pvma.vm_pgoff = idx; 1212 pvma.vm_ops = NULL; 1213 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx); 1214 page = alloc_page_vma(gfp, &pvma, 0); 1215 mpol_put(pvma.vm_policy); 1216 return page; 1217} 1218#else /* !CONFIG_NUMA */ 1219#ifdef CONFIG_TMPFS 1220static inline int shmem_parse_mpol(char *value, unsigned short *policy, 1221 unsigned short *mode_flags, nodemask_t *policy_nodes) 1222{ 1223 return 1; 1224} 1225 1226static inline void shmem_show_mpol(struct seq_file *seq, unsigned short policy, 1227 unsigned short flags, const nodemask_t policy_nodes) 1228{ 1229} 1230#endif /* CONFIG_TMPFS */ 1231 1232static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, 1233 struct shmem_inode_info *info, unsigned long idx) 1234{ 1235 return swapin_readahead(entry, gfp, NULL, 0); 1236} 1237 1238static inline struct page *shmem_alloc_page(gfp_t gfp, 1239 struct shmem_inode_info *info, unsigned long idx) 1240{ 1241 return alloc_page(gfp); 1242} 1243#endif /* CONFIG_NUMA */ 1244 1245/* 1246 * shmem_getpage - either get the page from swap or allocate a new one 1247 * 1248 * If we allocate a new one we do not mark it dirty. That's up to the 1249 * vm. If we swap it in we mark it dirty since we also free the swap 1250 * entry since a page cannot live in both the swap and page cache 1251 */ 1252static int shmem_getpage(struct inode *inode, unsigned long idx, 1253 struct page **pagep, enum sgp_type sgp, int *type) 1254{ 1255 struct address_space *mapping = inode->i_mapping; 1256 struct shmem_inode_info *info = SHMEM_I(inode); 1257 struct shmem_sb_info *sbinfo; 1258 struct page *filepage = *pagep; 1259 struct page *swappage; 1260 swp_entry_t *entry; 1261 swp_entry_t swap; 1262 gfp_t gfp; 1263 int error; 1264 1265 if (idx >= SHMEM_MAX_INDEX) 1266 return -EFBIG; 1267 1268 if (type) 1269 *type = 0; 1270 1271 /* 1272 * Normally, filepage is NULL on entry, and either found 1273 * uptodate immediately, or allocated and zeroed, or read 1274 * in under swappage, which is then assigned to filepage. 1275 * But shmem_readpage (required for splice) passes in a locked 1276 * filepage, which may be found not uptodate by other callers 1277 * too, and may need to be copied from the swappage read in. 1278 */ 1279repeat: 1280 if (!filepage) 1281 filepage = find_lock_page(mapping, idx); 1282 if (filepage && PageUptodate(filepage)) 1283 goto done; 1284 error = 0; 1285 gfp = mapping_gfp_mask(mapping); 1286 if (!filepage) { 1287 /* 1288 * Try to preload while we can wait, to not make a habit of 1289 * draining atomic reserves; but don't latch on to this cpu. 1290 */ 1291 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM); 1292 if (error) 1293 goto failed; 1294 radix_tree_preload_end(); 1295 } 1296 1297 spin_lock(&info->lock); 1298 shmem_recalc_inode(inode); 1299 entry = shmem_swp_alloc(info, idx, sgp); 1300 if (IS_ERR(entry)) { 1301 spin_unlock(&info->lock); 1302 error = PTR_ERR(entry); 1303 goto failed; 1304 } 1305 swap = *entry; 1306 1307 if (swap.val) { 1308 /* Look it up and read it in.. */ 1309 swappage = lookup_swap_cache(swap); 1310 if (!swappage) { 1311 shmem_swp_unmap(entry); 1312 /* here we actually do the io */ 1313 if (type && !(*type & VM_FAULT_MAJOR)) { 1314 __count_vm_event(PGMAJFAULT); 1315 *type |= VM_FAULT_MAJOR; 1316 } 1317 spin_unlock(&info->lock); 1318 swappage = shmem_swapin(swap, gfp, info, idx); 1319 if (!swappage) { 1320 spin_lock(&info->lock); 1321 entry = shmem_swp_alloc(info, idx, sgp); 1322 if (IS_ERR(entry)) 1323 error = PTR_ERR(entry); 1324 else { 1325 if (entry->val == swap.val) 1326 error = -ENOMEM; 1327 shmem_swp_unmap(entry); 1328 } 1329 spin_unlock(&info->lock); 1330 if (error) 1331 goto failed; 1332 goto repeat; 1333 } 1334 wait_on_page_locked(swappage); 1335 page_cache_release(swappage); 1336 goto repeat; 1337 } 1338 1339 /* We have to do this with page locked to prevent races */ 1340 if (TestSetPageLocked(swappage)) { 1341 shmem_swp_unmap(entry); 1342 spin_unlock(&info->lock); 1343 wait_on_page_locked(swappage); 1344 page_cache_release(swappage); 1345 goto repeat; 1346 } 1347 if (PageWriteback(swappage)) { 1348 shmem_swp_unmap(entry); 1349 spin_unlock(&info->lock); 1350 wait_on_page_writeback(swappage); 1351 unlock_page(swappage); 1352 page_cache_release(swappage); 1353 goto repeat; 1354 } 1355 if (!PageUptodate(swappage)) { 1356 shmem_swp_unmap(entry); 1357 spin_unlock(&info->lock); 1358 unlock_page(swappage); 1359 page_cache_release(swappage); 1360 error = -EIO; 1361 goto failed; 1362 } 1363 1364 if (filepage) { 1365 shmem_swp_set(info, entry, 0); 1366 shmem_swp_unmap(entry); 1367 delete_from_swap_cache(swappage); 1368 spin_unlock(&info->lock); 1369 copy_highpage(filepage, swappage); 1370 unlock_page(swappage); 1371 page_cache_release(swappage); 1372 flush_dcache_page(filepage); 1373 SetPageUptodate(filepage); 1374 set_page_dirty(filepage); 1375 swap_free(swap); 1376 } else if (!(error = add_to_page_cache( 1377 swappage, mapping, idx, GFP_NOWAIT))) { 1378 info->flags |= SHMEM_PAGEIN; 1379 shmem_swp_set(info, entry, 0); 1380 shmem_swp_unmap(entry); 1381 delete_from_swap_cache(swappage); 1382 spin_unlock(&info->lock); 1383 filepage = swappage; 1384 set_page_dirty(filepage); 1385 swap_free(swap); 1386 } else { 1387 shmem_swp_unmap(entry); 1388 spin_unlock(&info->lock); 1389 unlock_page(swappage); 1390 if (error == -ENOMEM) { 1391 /* allow reclaim from this memory cgroup */ 1392 error = mem_cgroup_cache_charge(swappage, 1393 current->mm, gfp & ~__GFP_HIGHMEM); 1394 if (error) { 1395 page_cache_release(swappage); 1396 goto failed; 1397 } 1398 mem_cgroup_uncharge_page(swappage); 1399 } 1400 page_cache_release(swappage); 1401 goto repeat; 1402 } 1403 } else if (sgp == SGP_READ && !filepage) { 1404 shmem_swp_unmap(entry); 1405 filepage = find_get_page(mapping, idx); 1406 if (filepage && 1407 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) { 1408 spin_unlock(&info->lock); 1409 wait_on_page_locked(filepage); 1410 page_cache_release(filepage); 1411 filepage = NULL; 1412 goto repeat; 1413 } 1414 spin_unlock(&info->lock); 1415 } else { 1416 shmem_swp_unmap(entry); 1417 sbinfo = SHMEM_SB(inode->i_sb); 1418 if (sbinfo->max_blocks) { 1419 spin_lock(&sbinfo->stat_lock); 1420 if (sbinfo->free_blocks == 0 || 1421 shmem_acct_block(info->flags)) { 1422 spin_unlock(&sbinfo->stat_lock); 1423 spin_unlock(&info->lock); 1424 error = -ENOSPC; 1425 goto failed; 1426 } 1427 sbinfo->free_blocks--; 1428 inode->i_blocks += BLOCKS_PER_PAGE; 1429 spin_unlock(&sbinfo->stat_lock); 1430 } else if (shmem_acct_block(info->flags)) { 1431 spin_unlock(&info->lock); 1432 error = -ENOSPC; 1433 goto failed; 1434 } 1435 1436 if (!filepage) { 1437 spin_unlock(&info->lock); 1438 filepage = shmem_alloc_page(gfp, info, idx); 1439 if (!filepage) { 1440 shmem_unacct_blocks(info->flags, 1); 1441 shmem_free_blocks(inode, 1); 1442 error = -ENOMEM; 1443 goto failed; 1444 } 1445 1446 /* Precharge page while we can wait, compensate after */ 1447 error = mem_cgroup_cache_charge(filepage, current->mm, 1448 gfp & ~__GFP_HIGHMEM); 1449 if (error) { 1450 page_cache_release(filepage); 1451 shmem_unacct_blocks(info->flags, 1); 1452 shmem_free_blocks(inode, 1); 1453 filepage = NULL; 1454 goto failed; 1455 } 1456 1457 spin_lock(&info->lock); 1458 entry = shmem_swp_alloc(info, idx, sgp); 1459 if (IS_ERR(entry)) 1460 error = PTR_ERR(entry); 1461 else { 1462 swap = *entry; 1463 shmem_swp_unmap(entry); 1464 } 1465 if (error || swap.val || 0 != add_to_page_cache_lru( 1466 filepage, mapping, idx, GFP_NOWAIT)) { 1467 spin_unlock(&info->lock); 1468 mem_cgroup_uncharge_page(filepage); 1469 page_cache_release(filepage); 1470 shmem_unacct_blocks(info->flags, 1); 1471 shmem_free_blocks(inode, 1); 1472 filepage = NULL; 1473 if (error) 1474 goto failed; 1475 goto repeat; 1476 } 1477 mem_cgroup_uncharge_page(filepage); 1478 info->flags |= SHMEM_PAGEIN; 1479 } 1480 1481 info->alloced++; 1482 spin_unlock(&info->lock); 1483 clear_highpage(filepage); 1484 flush_dcache_page(filepage); 1485 SetPageUptodate(filepage); 1486 if (sgp == SGP_DIRTY) 1487 set_page_dirty(filepage); 1488 } 1489done: 1490 *pagep = filepage; 1491 return 0; 1492 1493failed: 1494 if (*pagep != filepage) { 1495 unlock_page(filepage); 1496 page_cache_release(filepage); 1497 } 1498 return error; 1499} 1500 1501static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1502{ 1503 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 1504 int error; 1505 int ret; 1506 1507 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 1508 return VM_FAULT_SIGBUS; 1509 1510 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret); 1511 if (error) 1512 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); 1513 1514 mark_page_accessed(vmf->page); 1515 return ret | VM_FAULT_LOCKED; 1516} 1517 1518#ifdef CONFIG_NUMA 1519static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new) 1520{ 1521 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1522 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new); 1523} 1524 1525static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, 1526 unsigned long addr) 1527{ 1528 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1529 unsigned long idx; 1530 1531 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; 1532 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx); 1533} 1534#endif 1535 1536int shmem_lock(struct file *file, int lock, struct user_struct *user) 1537{ 1538 struct inode *inode = file->f_path.dentry->d_inode; 1539 struct shmem_inode_info *info = SHMEM_I(inode); 1540 int retval = -ENOMEM; 1541 1542 spin_lock(&info->lock); 1543 if (lock && !(info->flags & VM_LOCKED)) { 1544 if (!user_shm_lock(inode->i_size, user)) 1545 goto out_nomem; 1546 info->flags |= VM_LOCKED; 1547 } 1548 if (!lock && (info->flags & VM_LOCKED) && user) { 1549 user_shm_unlock(inode->i_size, user); 1550 info->flags &= ~VM_LOCKED; 1551 } 1552 retval = 0; 1553out_nomem: 1554 spin_unlock(&info->lock); 1555 return retval; 1556} 1557 1558static int shmem_mmap(struct file *file, struct vm_area_struct *vma) 1559{ 1560 file_accessed(file); 1561 vma->vm_ops = &shmem_vm_ops; 1562 vma->vm_flags |= VM_CAN_NONLINEAR; 1563 return 0; 1564} 1565 1566static struct inode * 1567shmem_get_inode(struct super_block *sb, int mode, dev_t dev) 1568{ 1569 struct inode *inode; 1570 struct shmem_inode_info *info; 1571 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 1572 1573 if (shmem_reserve_inode(sb)) 1574 return NULL; 1575 1576 inode = new_inode(sb); 1577 if (inode) { 1578 inode->i_mode = mode; 1579 inode->i_uid = current->fsuid; 1580 inode->i_gid = current->fsgid; 1581 inode->i_blocks = 0; 1582 inode->i_mapping->a_ops = &shmem_aops; 1583 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; 1584 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 1585 inode->i_generation = get_seconds(); 1586 info = SHMEM_I(inode); 1587 memset(info, 0, (char *)inode - (char *)info); 1588 spin_lock_init(&info->lock); 1589 INIT_LIST_HEAD(&info->swaplist); 1590 1591 switch (mode & S_IFMT) { 1592 default: 1593 inode->i_op = &shmem_special_inode_operations; 1594 init_special_inode(inode, mode, dev); 1595 break; 1596 case S_IFREG: 1597 inode->i_op = &shmem_inode_operations; 1598 inode->i_fop = &shmem_file_operations; 1599 mpol_shared_policy_init(&info->policy, sbinfo->policy, 1600 sbinfo->flags, &sbinfo->policy_nodes); 1601 break; 1602 case S_IFDIR: 1603 inc_nlink(inode); 1604 /* Some things misbehave if size == 0 on a directory */ 1605 inode->i_size = 2 * BOGO_DIRENT_SIZE; 1606 inode->i_op = &shmem_dir_inode_operations; 1607 inode->i_fop = &simple_dir_operations; 1608 break; 1609 case S_IFLNK: 1610 /* 1611 * Must not load anything in the rbtree, 1612 * mpol_free_shared_policy will not be called. 1613 */ 1614 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, 0, 1615 NULL); 1616 break; 1617 } 1618 } else 1619 shmem_free_inode(sb); 1620 return inode; 1621} 1622 1623#ifdef CONFIG_TMPFS 1624static const struct inode_operations shmem_symlink_inode_operations; 1625static const struct inode_operations shmem_symlink_inline_operations; 1626 1627/* 1628 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin; 1629 * but providing them allows a tmpfs file to be used for splice, sendfile, and 1630 * below the loop driver, in the generic fashion that many filesystems support. 1631 */ 1632static int shmem_readpage(struct file *file, struct page *page) 1633{ 1634 struct inode *inode = page->mapping->host; 1635 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL); 1636 unlock_page(page); 1637 return error; 1638} 1639 1640static int 1641shmem_write_begin(struct file *file, struct address_space *mapping, 1642 loff_t pos, unsigned len, unsigned flags, 1643 struct page **pagep, void **fsdata) 1644{ 1645 struct inode *inode = mapping->host; 1646 pgoff_t index = pos >> PAGE_CACHE_SHIFT; 1647 *pagep = NULL; 1648 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL); 1649} 1650 1651static int 1652shmem_write_end(struct file *file, struct address_space *mapping, 1653 loff_t pos, unsigned len, unsigned copied, 1654 struct page *page, void *fsdata) 1655{ 1656 struct inode *inode = mapping->host; 1657 1658 if (pos + copied > inode->i_size) 1659 i_size_write(inode, pos + copied); 1660 1661 unlock_page(page); 1662 set_page_dirty(page); 1663 page_cache_release(page); 1664 1665 return copied; 1666} 1667 1668static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor) 1669{ 1670 struct inode *inode = filp->f_path.dentry->d_inode; 1671 struct address_space *mapping = inode->i_mapping; 1672 unsigned long index, offset; 1673 enum sgp_type sgp = SGP_READ; 1674 1675 /* 1676 * Might this read be for a stacking filesystem? Then when reading 1677 * holes of a sparse file, we actually need to allocate those pages, 1678 * and even mark them dirty, so it cannot exceed the max_blocks limit. 1679 */ 1680 if (segment_eq(get_fs(), KERNEL_DS)) 1681 sgp = SGP_DIRTY; 1682 1683 index = *ppos >> PAGE_CACHE_SHIFT; 1684 offset = *ppos & ~PAGE_CACHE_MASK; 1685 1686 for (;;) { 1687 struct page *page = NULL; 1688 unsigned long end_index, nr, ret; 1689 loff_t i_size = i_size_read(inode); 1690 1691 end_index = i_size >> PAGE_CACHE_SHIFT; 1692 if (index > end_index) 1693 break; 1694 if (index == end_index) { 1695 nr = i_size & ~PAGE_CACHE_MASK; 1696 if (nr <= offset) 1697 break; 1698 } 1699 1700 desc->error = shmem_getpage(inode, index, &page, sgp, NULL); 1701 if (desc->error) { 1702 if (desc->error == -EINVAL) 1703 desc->error = 0; 1704 break; 1705 } 1706 if (page) 1707 unlock_page(page); 1708 1709 /* 1710 * We must evaluate after, since reads (unlike writes) 1711 * are called without i_mutex protection against truncate 1712 */ 1713 nr = PAGE_CACHE_SIZE; 1714 i_size = i_size_read(inode); 1715 end_index = i_size >> PAGE_CACHE_SHIFT; 1716 if (index == end_index) { 1717 nr = i_size & ~PAGE_CACHE_MASK; 1718 if (nr <= offset) { 1719 if (page) 1720 page_cache_release(page); 1721 break; 1722 } 1723 } 1724 nr -= offset; 1725 1726 if (page) { 1727 /* 1728 * If users can be writing to this page using arbitrary 1729 * virtual addresses, take care about potential aliasing 1730 * before reading the page on the kernel side. 1731 */ 1732 if (mapping_writably_mapped(mapping)) 1733 flush_dcache_page(page); 1734 /* 1735 * Mark the page accessed if we read the beginning. 1736 */ 1737 if (!offset) 1738 mark_page_accessed(page); 1739 } else { 1740 page = ZERO_PAGE(0); 1741 page_cache_get(page); 1742 } 1743 1744 /* 1745 * Ok, we have the page, and it's up-to-date, so 1746 * now we can copy it to user space... 1747 * 1748 * The actor routine returns how many bytes were actually used.. 1749 * NOTE! This may not be the same as how much of a user buffer 1750 * we filled up (we may be padding etc), so we can only update 1751 * "pos" here (the actor routine has to update the user buffer 1752 * pointers and the remaining count). 1753 */ 1754 ret = actor(desc, page, offset, nr); 1755 offset += ret; 1756 index += offset >> PAGE_CACHE_SHIFT; 1757 offset &= ~PAGE_CACHE_MASK; 1758 1759 page_cache_release(page); 1760 if (ret != nr || !desc->count) 1761 break; 1762 1763 cond_resched(); 1764 } 1765 1766 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; 1767 file_accessed(filp); 1768} 1769 1770static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos) 1771{ 1772 read_descriptor_t desc; 1773 1774 if ((ssize_t) count < 0) 1775 return -EINVAL; 1776 if (!access_ok(VERIFY_WRITE, buf, count)) 1777 return -EFAULT; 1778 if (!count) 1779 return 0; 1780 1781 desc.written = 0; 1782 desc.count = count; 1783 desc.arg.buf = buf; 1784 desc.error = 0; 1785 1786 do_shmem_file_read(filp, ppos, &desc, file_read_actor); 1787 if (desc.written) 1788 return desc.written; 1789 return desc.error; 1790} 1791 1792static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) 1793{ 1794 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); 1795 1796 buf->f_type = TMPFS_MAGIC; 1797 buf->f_bsize = PAGE_CACHE_SIZE; 1798 buf->f_namelen = NAME_MAX; 1799 spin_lock(&sbinfo->stat_lock); 1800 if (sbinfo->max_blocks) { 1801 buf->f_blocks = sbinfo->max_blocks; 1802 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks; 1803 } 1804 if (sbinfo->max_inodes) { 1805 buf->f_files = sbinfo->max_inodes; 1806 buf->f_ffree = sbinfo->free_inodes; 1807 } 1808 /* else leave those fields 0 like simple_statfs */ 1809 spin_unlock(&sbinfo->stat_lock); 1810 return 0; 1811} 1812 1813/* 1814 * File creation. Allocate an inode, and we're done.. 1815 */ 1816static int 1817shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 1818{ 1819 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev); 1820 int error = -ENOSPC; 1821 1822 if (inode) { 1823 error = security_inode_init_security(inode, dir, NULL, NULL, 1824 NULL); 1825 if (error) { 1826 if (error != -EOPNOTSUPP) { 1827 iput(inode); 1828 return error; 1829 } 1830 } 1831 error = shmem_acl_init(inode, dir); 1832 if (error) { 1833 iput(inode); 1834 return error; 1835 } 1836 if (dir->i_mode & S_ISGID) { 1837 inode->i_gid = dir->i_gid; 1838 if (S_ISDIR(mode)) 1839 inode->i_mode |= S_ISGID; 1840 } 1841 dir->i_size += BOGO_DIRENT_SIZE; 1842 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1843 d_instantiate(dentry, inode); 1844 dget(dentry); /* Extra count - pin the dentry in core */ 1845 } 1846 return error; 1847} 1848 1849static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1850{ 1851 int error; 1852 1853 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) 1854 return error; 1855 inc_nlink(dir); 1856 return 0; 1857} 1858 1859static int shmem_create(struct inode *dir, struct dentry *dentry, int mode, 1860 struct nameidata *nd) 1861{ 1862 return shmem_mknod(dir, dentry, mode | S_IFREG, 0); 1863} 1864 1865/* 1866 * Link a file.. 1867 */ 1868static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1869{ 1870 struct inode *inode = old_dentry->d_inode; 1871 int ret; 1872 1873 /* 1874 * No ordinary (disk based) filesystem counts links as inodes; 1875 * but each new link needs a new dentry, pinning lowmem, and 1876 * tmpfs dentries cannot be pruned until they are unlinked. 1877 */ 1878 ret = shmem_reserve_inode(inode->i_sb); 1879 if (ret) 1880 goto out; 1881 1882 dir->i_size += BOGO_DIRENT_SIZE; 1883 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1884 inc_nlink(inode); 1885 atomic_inc(&inode->i_count); /* New dentry reference */ 1886 dget(dentry); /* Extra pinning count for the created dentry */ 1887 d_instantiate(dentry, inode); 1888out: 1889 return ret; 1890} 1891 1892static int shmem_unlink(struct inode *dir, struct dentry *dentry) 1893{ 1894 struct inode *inode = dentry->d_inode; 1895 1896 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) 1897 shmem_free_inode(inode->i_sb); 1898 1899 dir->i_size -= BOGO_DIRENT_SIZE; 1900 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1901 drop_nlink(inode); 1902 dput(dentry); /* Undo the count from "create" - this does all the work */ 1903 return 0; 1904} 1905 1906static int shmem_rmdir(struct inode *dir, struct dentry *dentry) 1907{ 1908 if (!simple_empty(dentry)) 1909 return -ENOTEMPTY; 1910 1911 drop_nlink(dentry->d_inode); 1912 drop_nlink(dir); 1913 return shmem_unlink(dir, dentry); 1914} 1915 1916/* 1917 * The VFS layer already does all the dentry stuff for rename, 1918 * we just have to decrement the usage count for the target if 1919 * it exists so that the VFS layer correctly free's it when it 1920 * gets overwritten. 1921 */ 1922static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) 1923{ 1924 struct inode *inode = old_dentry->d_inode; 1925 int they_are_dirs = S_ISDIR(inode->i_mode); 1926 1927 if (!simple_empty(new_dentry)) 1928 return -ENOTEMPTY; 1929 1930 if (new_dentry->d_inode) { 1931 (void) shmem_unlink(new_dir, new_dentry); 1932 if (they_are_dirs) 1933 drop_nlink(old_dir); 1934 } else if (they_are_dirs) { 1935 drop_nlink(old_dir); 1936 inc_nlink(new_dir); 1937 } 1938 1939 old_dir->i_size -= BOGO_DIRENT_SIZE; 1940 new_dir->i_size += BOGO_DIRENT_SIZE; 1941 old_dir->i_ctime = old_dir->i_mtime = 1942 new_dir->i_ctime = new_dir->i_mtime = 1943 inode->i_ctime = CURRENT_TIME; 1944 return 0; 1945} 1946 1947static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1948{ 1949 int error; 1950 int len; 1951 struct inode *inode; 1952 struct page *page = NULL; 1953 char *kaddr; 1954 struct shmem_inode_info *info; 1955 1956 len = strlen(symname) + 1; 1957 if (len > PAGE_CACHE_SIZE) 1958 return -ENAMETOOLONG; 1959 1960 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0); 1961 if (!inode) 1962 return -ENOSPC; 1963 1964 error = security_inode_init_security(inode, dir, NULL, NULL, 1965 NULL); 1966 if (error) { 1967 if (error != -EOPNOTSUPP) { 1968 iput(inode); 1969 return error; 1970 } 1971 error = 0; 1972 } 1973 1974 info = SHMEM_I(inode); 1975 inode->i_size = len-1; 1976 if (len <= (char *)inode - (char *)info) { 1977 /* do it inline */ 1978 memcpy(info, symname, len); 1979 inode->i_op = &shmem_symlink_inline_operations; 1980 } else { 1981 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); 1982 if (error) { 1983 iput(inode); 1984 return error; 1985 } 1986 unlock_page(page); 1987 inode->i_op = &shmem_symlink_inode_operations; 1988 kaddr = kmap_atomic(page, KM_USER0); 1989 memcpy(kaddr, symname, len); 1990 kunmap_atomic(kaddr, KM_USER0); 1991 set_page_dirty(page); 1992 page_cache_release(page); 1993 } 1994 if (dir->i_mode & S_ISGID) 1995 inode->i_gid = dir->i_gid; 1996 dir->i_size += BOGO_DIRENT_SIZE; 1997 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1998 d_instantiate(dentry, inode); 1999 dget(dentry); 2000 return 0; 2001} 2002 2003static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd) 2004{ 2005 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode)); 2006 return NULL; 2007} 2008 2009static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) 2010{ 2011 struct page *page = NULL; 2012 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); 2013 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page)); 2014 if (page) 2015 unlock_page(page); 2016 return page; 2017} 2018 2019static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 2020{ 2021 if (!IS_ERR(nd_get_link(nd))) { 2022 struct page *page = cookie; 2023 kunmap(page); 2024 mark_page_accessed(page); 2025 page_cache_release(page); 2026 } 2027} 2028 2029static const struct inode_operations shmem_symlink_inline_operations = { 2030 .readlink = generic_readlink, 2031 .follow_link = shmem_follow_link_inline, 2032}; 2033 2034static const struct inode_operations shmem_symlink_inode_operations = { 2035 .truncate = shmem_truncate, 2036 .readlink = generic_readlink, 2037 .follow_link = shmem_follow_link, 2038 .put_link = shmem_put_link, 2039}; 2040 2041#ifdef CONFIG_TMPFS_POSIX_ACL 2042/* 2043 * Superblocks without xattr inode operations will get security.* xattr 2044 * support from the VFS "for free". As soon as we have any other xattrs 2045 * like ACLs, we also need to implement the security.* handlers at 2046 * filesystem level, though. 2047 */ 2048 2049static size_t shmem_xattr_security_list(struct inode *inode, char *list, 2050 size_t list_len, const char *name, 2051 size_t name_len) 2052{ 2053 return security_inode_listsecurity(inode, list, list_len); 2054} 2055 2056static int shmem_xattr_security_get(struct inode *inode, const char *name, 2057 void *buffer, size_t size) 2058{ 2059 if (strcmp(name, "") == 0) 2060 return -EINVAL; 2061 return xattr_getsecurity(inode, name, buffer, size); 2062} 2063 2064static int shmem_xattr_security_set(struct inode *inode, const char *name, 2065 const void *value, size_t size, int flags) 2066{ 2067 if (strcmp(name, "") == 0) 2068 return -EINVAL; 2069 return security_inode_setsecurity(inode, name, value, size, flags); 2070} 2071 2072static struct xattr_handler shmem_xattr_security_handler = { 2073 .prefix = XATTR_SECURITY_PREFIX, 2074 .list = shmem_xattr_security_list, 2075 .get = shmem_xattr_security_get, 2076 .set = shmem_xattr_security_set, 2077}; 2078 2079static struct xattr_handler *shmem_xattr_handlers[] = { 2080 &shmem_xattr_acl_access_handler, 2081 &shmem_xattr_acl_default_handler, 2082 &shmem_xattr_security_handler, 2083 NULL 2084}; 2085#endif 2086 2087static struct dentry *shmem_get_parent(struct dentry *child) 2088{ 2089 return ERR_PTR(-ESTALE); 2090} 2091 2092static int shmem_match(struct inode *ino, void *vfh) 2093{ 2094 __u32 *fh = vfh; 2095 __u64 inum = fh[2]; 2096 inum = (inum << 32) | fh[1]; 2097 return ino->i_ino == inum && fh[0] == ino->i_generation; 2098} 2099 2100static struct dentry *shmem_fh_to_dentry(struct super_block *sb, 2101 struct fid *fid, int fh_len, int fh_type) 2102{ 2103 struct inode *inode; 2104 struct dentry *dentry = NULL; 2105 u64 inum = fid->raw[2]; 2106 inum = (inum << 32) | fid->raw[1]; 2107 2108 if (fh_len < 3) 2109 return NULL; 2110 2111 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), 2112 shmem_match, fid->raw); 2113 if (inode) { 2114 dentry = d_find_alias(inode); 2115 iput(inode); 2116 } 2117 2118 return dentry; 2119} 2120 2121static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len, 2122 int connectable) 2123{ 2124 struct inode *inode = dentry->d_inode; 2125 2126 if (*len < 3) 2127 return 255; 2128 2129 if (hlist_unhashed(&inode->i_hash)) { 2130 /* Unfortunately insert_inode_hash is not idempotent, 2131 * so as we hash inodes here rather than at creation 2132 * time, we need a lock to ensure we only try 2133 * to do it once 2134 */ 2135 static DEFINE_SPINLOCK(lock); 2136 spin_lock(&lock); 2137 if (hlist_unhashed(&inode->i_hash)) 2138 __insert_inode_hash(inode, 2139 inode->i_ino + inode->i_generation); 2140 spin_unlock(&lock); 2141 } 2142 2143 fh[0] = inode->i_generation; 2144 fh[1] = inode->i_ino; 2145 fh[2] = ((__u64)inode->i_ino) >> 32; 2146 2147 *len = 3; 2148 return 1; 2149} 2150 2151static const struct export_operations shmem_export_ops = { 2152 .get_parent = shmem_get_parent, 2153 .encode_fh = shmem_encode_fh, 2154 .fh_to_dentry = shmem_fh_to_dentry, 2155}; 2156 2157static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, 2158 bool remount) 2159{ 2160 char *this_char, *value, *rest; 2161 2162 while (options != NULL) { 2163 this_char = options; 2164 for (;;) { 2165 /* 2166 * NUL-terminate this option: unfortunately, 2167 * mount options form a comma-separated list, 2168 * but mpol's nodelist may also contain commas. 2169 */ 2170 options = strchr(options, ','); 2171 if (options == NULL) 2172 break; 2173 options++; 2174 if (!isdigit(*options)) { 2175 options[-1] = '\0'; 2176 break; 2177 } 2178 } 2179 if (!*this_char) 2180 continue; 2181 if ((value = strchr(this_char,'=')) != NULL) { 2182 *value++ = 0; 2183 } else { 2184 printk(KERN_ERR 2185 "tmpfs: No value for mount option '%s'\n", 2186 this_char); 2187 return 1; 2188 } 2189 2190 if (!strcmp(this_char,"size")) { 2191 unsigned long long size; 2192 size = memparse(value,&rest); 2193 if (*rest == '%') { 2194 size <<= PAGE_SHIFT; 2195 size *= totalram_pages; 2196 do_div(size, 100); 2197 rest++; 2198 } 2199 if (*rest) 2200 goto bad_val; 2201 sbinfo->max_blocks = 2202 DIV_ROUND_UP(size, PAGE_CACHE_SIZE); 2203 } else if (!strcmp(this_char,"nr_blocks")) { 2204 sbinfo->max_blocks = memparse(value, &rest); 2205 if (*rest) 2206 goto bad_val; 2207 } else if (!strcmp(this_char,"nr_inodes")) { 2208 sbinfo->max_inodes = memparse(value, &rest); 2209 if (*rest) 2210 goto bad_val; 2211 } else if (!strcmp(this_char,"mode")) { 2212 if (remount) 2213 continue; 2214 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; 2215 if (*rest) 2216 goto bad_val; 2217 } else if (!strcmp(this_char,"uid")) { 2218 if (remount) 2219 continue; 2220 sbinfo->uid = simple_strtoul(value, &rest, 0); 2221 if (*rest) 2222 goto bad_val; 2223 } else if (!strcmp(this_char,"gid")) { 2224 if (remount) 2225 continue; 2226 sbinfo->gid = simple_strtoul(value, &rest, 0); 2227 if (*rest) 2228 goto bad_val; 2229 } else if (!strcmp(this_char,"mpol")) { 2230 if (shmem_parse_mpol(value, &sbinfo->policy, 2231 &sbinfo->flags, &sbinfo->policy_nodes)) 2232 goto bad_val; 2233 } else { 2234 printk(KERN_ERR "tmpfs: Bad mount option %s\n", 2235 this_char); 2236 return 1; 2237 } 2238 } 2239 return 0; 2240 2241bad_val: 2242 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", 2243 value, this_char); 2244 return 1; 2245 2246} 2247 2248static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) 2249{ 2250 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 2251 struct shmem_sb_info config = *sbinfo; 2252 unsigned long blocks; 2253 unsigned long inodes; 2254 int error = -EINVAL; 2255 2256 if (shmem_parse_options(data, &config, true)) 2257 return error; 2258 2259 spin_lock(&sbinfo->stat_lock); 2260 blocks = sbinfo->max_blocks - sbinfo->free_blocks; 2261 inodes = sbinfo->max_inodes - sbinfo->free_inodes; 2262 if (config.max_blocks < blocks) 2263 goto out; 2264 if (config.max_inodes < inodes) 2265 goto out; 2266 /* 2267 * Those tests also disallow limited->unlimited while any are in 2268 * use, so i_blocks will always be zero when max_blocks is zero; 2269 * but we must separately disallow unlimited->limited, because 2270 * in that case we have no record of how much is already in use. 2271 */ 2272 if (config.max_blocks && !sbinfo->max_blocks) 2273 goto out; 2274 if (config.max_inodes && !sbinfo->max_inodes) 2275 goto out; 2276 2277 error = 0; 2278 sbinfo->max_blocks = config.max_blocks; 2279 sbinfo->free_blocks = config.max_blocks - blocks; 2280 sbinfo->max_inodes = config.max_inodes; 2281 sbinfo->free_inodes = config.max_inodes - inodes; 2282 sbinfo->policy = config.policy; 2283 sbinfo->flags = config.flags; 2284 sbinfo->policy_nodes = config.policy_nodes; 2285out: 2286 spin_unlock(&sbinfo->stat_lock); 2287 return error; 2288} 2289 2290static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs) 2291{ 2292 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb); 2293 2294 if (sbinfo->max_blocks != shmem_default_max_blocks()) 2295 seq_printf(seq, ",size=%luk", 2296 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10)); 2297 if (sbinfo->max_inodes != shmem_default_max_inodes()) 2298 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); 2299 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) 2300 seq_printf(seq, ",mode=%03o", sbinfo->mode); 2301 if (sbinfo->uid != 0) 2302 seq_printf(seq, ",uid=%u", sbinfo->uid); 2303 if (sbinfo->gid != 0) 2304 seq_printf(seq, ",gid=%u", sbinfo->gid); 2305 shmem_show_mpol(seq, sbinfo->policy, sbinfo->flags, 2306 sbinfo->policy_nodes); 2307 return 0; 2308} 2309#endif /* CONFIG_TMPFS */ 2310 2311static void shmem_put_super(struct super_block *sb) 2312{ 2313 kfree(sb->s_fs_info); 2314 sb->s_fs_info = NULL; 2315} 2316 2317static int shmem_fill_super(struct super_block *sb, 2318 void *data, int silent) 2319{ 2320 struct inode *inode; 2321 struct dentry *root; 2322 struct shmem_sb_info *sbinfo; 2323 int err = -ENOMEM; 2324 2325 /* Round up to L1_CACHE_BYTES to resist false sharing */ 2326 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info), 2327 L1_CACHE_BYTES), GFP_KERNEL); 2328 if (!sbinfo) 2329 return -ENOMEM; 2330 2331 sbinfo->max_blocks = 0; 2332 sbinfo->max_inodes = 0; 2333 sbinfo->mode = S_IRWXUGO | S_ISVTX; 2334 sbinfo->uid = current->fsuid; 2335 sbinfo->gid = current->fsgid; 2336 sbinfo->policy = MPOL_DEFAULT; 2337 sbinfo->flags = 0; 2338 sbinfo->policy_nodes = node_states[N_HIGH_MEMORY]; 2339 sb->s_fs_info = sbinfo; 2340 2341#ifdef CONFIG_TMPFS 2342 /* 2343 * Per default we only allow half of the physical ram per 2344 * tmpfs instance, limiting inodes to one per page of lowmem; 2345 * but the internal instance is left unlimited. 2346 */ 2347 if (!(sb->s_flags & MS_NOUSER)) { 2348 sbinfo->max_blocks = shmem_default_max_blocks(); 2349 sbinfo->max_inodes = shmem_default_max_inodes(); 2350 if (shmem_parse_options(data, sbinfo, false)) { 2351 err = -EINVAL; 2352 goto failed; 2353 } 2354 } 2355 sb->s_export_op = &shmem_export_ops; 2356#else 2357 sb->s_flags |= MS_NOUSER; 2358#endif 2359 2360 spin_lock_init(&sbinfo->stat_lock); 2361 sbinfo->free_blocks = sbinfo->max_blocks; 2362 sbinfo->free_inodes = sbinfo->max_inodes; 2363 2364 sb->s_maxbytes = SHMEM_MAX_BYTES; 2365 sb->s_blocksize = PAGE_CACHE_SIZE; 2366 sb->s_blocksize_bits = PAGE_CACHE_SHIFT; 2367 sb->s_magic = TMPFS_MAGIC; 2368 sb->s_op = &shmem_ops; 2369 sb->s_time_gran = 1; 2370#ifdef CONFIG_TMPFS_POSIX_ACL 2371 sb->s_xattr = shmem_xattr_handlers; 2372 sb->s_flags |= MS_POSIXACL; 2373#endif 2374 2375 inode = shmem_get_inode(sb, S_IFDIR | sbinfo->mode, 0); 2376 if (!inode) 2377 goto failed; 2378 inode->i_uid = sbinfo->uid; 2379 inode->i_gid = sbinfo->gid; 2380 root = d_alloc_root(inode); 2381 if (!root) 2382 goto failed_iput; 2383 sb->s_root = root; 2384 return 0; 2385 2386failed_iput: 2387 iput(inode); 2388failed: 2389 shmem_put_super(sb); 2390 return err; 2391} 2392 2393static struct kmem_cache *shmem_inode_cachep; 2394 2395static struct inode *shmem_alloc_inode(struct super_block *sb) 2396{ 2397 struct shmem_inode_info *p; 2398 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); 2399 if (!p) 2400 return NULL; 2401 return &p->vfs_inode; 2402} 2403 2404static void shmem_destroy_inode(struct inode *inode) 2405{ 2406 if ((inode->i_mode & S_IFMT) == S_IFREG) { 2407 /* only struct inode is valid if it's an inline symlink */ 2408 mpol_free_shared_policy(&SHMEM_I(inode)->policy); 2409 } 2410 shmem_acl_destroy_inode(inode); 2411 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); 2412} 2413 2414static void init_once(struct kmem_cache *cachep, void *foo) 2415{ 2416 struct shmem_inode_info *p = (struct shmem_inode_info *) foo; 2417 2418 inode_init_once(&p->vfs_inode); 2419#ifdef CONFIG_TMPFS_POSIX_ACL 2420 p->i_acl = NULL; 2421 p->i_default_acl = NULL; 2422#endif 2423} 2424 2425static int init_inodecache(void) 2426{ 2427 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", 2428 sizeof(struct shmem_inode_info), 2429 0, SLAB_PANIC, init_once); 2430 return 0; 2431} 2432 2433static void destroy_inodecache(void) 2434{ 2435 kmem_cache_destroy(shmem_inode_cachep); 2436} 2437 2438static const struct address_space_operations shmem_aops = { 2439 .writepage = shmem_writepage, 2440 .set_page_dirty = __set_page_dirty_no_writeback, 2441#ifdef CONFIG_TMPFS 2442 .readpage = shmem_readpage, 2443 .write_begin = shmem_write_begin, 2444 .write_end = shmem_write_end, 2445#endif 2446 .migratepage = migrate_page, 2447}; 2448 2449static const struct file_operations shmem_file_operations = { 2450 .mmap = shmem_mmap, 2451#ifdef CONFIG_TMPFS 2452 .llseek = generic_file_llseek, 2453 .read = shmem_file_read, 2454 .write = do_sync_write, 2455 .aio_write = generic_file_aio_write, 2456 .fsync = simple_sync_file, 2457 .splice_read = generic_file_splice_read, 2458 .splice_write = generic_file_splice_write, 2459#endif 2460}; 2461 2462static const struct inode_operations shmem_inode_operations = { 2463 .truncate = shmem_truncate, 2464 .setattr = shmem_notify_change, 2465 .truncate_range = shmem_truncate_range, 2466#ifdef CONFIG_TMPFS_POSIX_ACL 2467 .setxattr = generic_setxattr, 2468 .getxattr = generic_getxattr, 2469 .listxattr = generic_listxattr, 2470 .removexattr = generic_removexattr, 2471 .permission = shmem_permission, 2472#endif 2473 2474}; 2475 2476static const struct inode_operations shmem_dir_inode_operations = { 2477#ifdef CONFIG_TMPFS 2478 .create = shmem_create, 2479 .lookup = simple_lookup, 2480 .link = shmem_link, 2481 .unlink = shmem_unlink, 2482 .symlink = shmem_symlink, 2483 .mkdir = shmem_mkdir, 2484 .rmdir = shmem_rmdir, 2485 .mknod = shmem_mknod, 2486 .rename = shmem_rename, 2487#endif 2488#ifdef CONFIG_TMPFS_POSIX_ACL 2489 .setattr = shmem_notify_change, 2490 .setxattr = generic_setxattr, 2491 .getxattr = generic_getxattr, 2492 .listxattr = generic_listxattr, 2493 .removexattr = generic_removexattr, 2494 .permission = shmem_permission, 2495#endif 2496}; 2497 2498static const struct inode_operations shmem_special_inode_operations = { 2499#ifdef CONFIG_TMPFS_POSIX_ACL 2500 .setattr = shmem_notify_change, 2501 .setxattr = generic_setxattr, 2502 .getxattr = generic_getxattr, 2503 .listxattr = generic_listxattr, 2504 .removexattr = generic_removexattr, 2505 .permission = shmem_permission, 2506#endif 2507}; 2508 2509static const struct super_operations shmem_ops = { 2510 .alloc_inode = shmem_alloc_inode, 2511 .destroy_inode = shmem_destroy_inode, 2512#ifdef CONFIG_TMPFS 2513 .statfs = shmem_statfs, 2514 .remount_fs = shmem_remount_fs, 2515 .show_options = shmem_show_options, 2516#endif 2517 .delete_inode = shmem_delete_inode, 2518 .drop_inode = generic_delete_inode, 2519 .put_super = shmem_put_super, 2520}; 2521 2522static struct vm_operations_struct shmem_vm_ops = { 2523 .fault = shmem_fault, 2524#ifdef CONFIG_NUMA 2525 .set_policy = shmem_set_policy, 2526 .get_policy = shmem_get_policy, 2527#endif 2528}; 2529 2530 2531static int shmem_get_sb(struct file_system_type *fs_type, 2532 int flags, const char *dev_name, void *data, struct vfsmount *mnt) 2533{ 2534 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt); 2535} 2536 2537static struct file_system_type tmpfs_fs_type = { 2538 .owner = THIS_MODULE, 2539 .name = "tmpfs", 2540 .get_sb = shmem_get_sb, 2541 .kill_sb = kill_litter_super, 2542}; 2543static struct vfsmount *shm_mnt; 2544 2545static int __init init_tmpfs(void) 2546{ 2547 int error; 2548 2549 error = bdi_init(&shmem_backing_dev_info); 2550 if (error) 2551 goto out4; 2552 2553 error = init_inodecache(); 2554 if (error) 2555 goto out3; 2556 2557 error = register_filesystem(&tmpfs_fs_type); 2558 if (error) { 2559 printk(KERN_ERR "Could not register tmpfs\n"); 2560 goto out2; 2561 } 2562 2563 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER, 2564 tmpfs_fs_type.name, NULL); 2565 if (IS_ERR(shm_mnt)) { 2566 error = PTR_ERR(shm_mnt); 2567 printk(KERN_ERR "Could not kern_mount tmpfs\n"); 2568 goto out1; 2569 } 2570 return 0; 2571 2572out1: 2573 unregister_filesystem(&tmpfs_fs_type); 2574out2: 2575 destroy_inodecache(); 2576out3: 2577 bdi_destroy(&shmem_backing_dev_info); 2578out4: 2579 shm_mnt = ERR_PTR(error); 2580 return error; 2581} 2582module_init(init_tmpfs) 2583 2584/** 2585 * shmem_file_setup - get an unlinked file living in tmpfs 2586 * @name: name for dentry (to be seen in /proc/<pid>/maps 2587 * @size: size to be set for the file 2588 * @flags: vm_flags 2589 */ 2590struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags) 2591{ 2592 int error; 2593 struct file *file; 2594 struct inode *inode; 2595 struct dentry *dentry, *root; 2596 struct qstr this; 2597 2598 if (IS_ERR(shm_mnt)) 2599 return (void *)shm_mnt; 2600 2601 if (size < 0 || size > SHMEM_MAX_BYTES) 2602 return ERR_PTR(-EINVAL); 2603 2604 if (shmem_acct_size(flags, size)) 2605 return ERR_PTR(-ENOMEM); 2606 2607 error = -ENOMEM; 2608 this.name = name; 2609 this.len = strlen(name); 2610 this.hash = 0; /* will go */ 2611 root = shm_mnt->mnt_root; 2612 dentry = d_alloc(root, &this); 2613 if (!dentry) 2614 goto put_memory; 2615 2616 error = -ENFILE; 2617 file = get_empty_filp(); 2618 if (!file) 2619 goto put_dentry; 2620 2621 error = -ENOSPC; 2622 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0); 2623 if (!inode) 2624 goto close_file; 2625 2626 SHMEM_I(inode)->flags = flags & VM_ACCOUNT; 2627 d_instantiate(dentry, inode); 2628 inode->i_size = size; 2629 inode->i_nlink = 0; /* It is unlinked */ 2630 init_file(file, shm_mnt, dentry, FMODE_WRITE | FMODE_READ, 2631 &shmem_file_operations); 2632 return file; 2633 2634close_file: 2635 put_filp(file); 2636put_dentry: 2637 dput(dentry); 2638put_memory: 2639 shmem_unacct_size(flags, size); 2640 return ERR_PTR(error); 2641} 2642 2643/** 2644 * shmem_zero_setup - setup a shared anonymous mapping 2645 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff 2646 */ 2647int shmem_zero_setup(struct vm_area_struct *vma) 2648{ 2649 struct file *file; 2650 loff_t size = vma->vm_end - vma->vm_start; 2651 2652 file = shmem_file_setup("dev/zero", size, vma->vm_flags); 2653 if (IS_ERR(file)) 2654 return PTR_ERR(file); 2655 2656 if (vma->vm_file) 2657 fput(vma->vm_file); 2658 vma->vm_file = file; 2659 vma->vm_ops = &shmem_vm_ops; 2660 return 0; 2661} 2662