shmem.c revision f5b087b52f1710eb0bf15a2d2b030c51a6a1ca9e
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 if (flags) { 1129 if (!strcmp(flags, "static")) 1130 *mode_flags |= MPOL_F_STATIC_NODES; 1131 } 1132out: 1133 /* Restore string for error message */ 1134 if (nodelist) 1135 *--nodelist = ':'; 1136 return err; 1137} 1138 1139static void shmem_show_mpol(struct seq_file *seq, unsigned short policy, 1140 unsigned short flags, const nodemask_t policy_nodes) 1141{ 1142 char *policy_string; 1143 1144 switch (policy) { 1145 case MPOL_PREFERRED: 1146 policy_string = "prefer"; 1147 break; 1148 case MPOL_BIND: 1149 policy_string = "bind"; 1150 break; 1151 case MPOL_INTERLEAVE: 1152 policy_string = "interleave"; 1153 break; 1154 default: 1155 /* MPOL_DEFAULT */ 1156 return; 1157 } 1158 1159 seq_printf(seq, ",mpol=%s", policy_string); 1160 1161 if (policy != MPOL_INTERLEAVE || 1162 !nodes_equal(policy_nodes, node_states[N_HIGH_MEMORY])) { 1163 char buffer[64]; 1164 int len; 1165 1166 len = nodelist_scnprintf(buffer, sizeof(buffer), policy_nodes); 1167 if (len < sizeof(buffer)) 1168 seq_printf(seq, ":%s", buffer); 1169 else 1170 seq_printf(seq, ":?"); 1171 } 1172} 1173#endif /* CONFIG_TMPFS */ 1174 1175static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, 1176 struct shmem_inode_info *info, unsigned long idx) 1177{ 1178 struct vm_area_struct pvma; 1179 struct page *page; 1180 1181 /* Create a pseudo vma that just contains the policy */ 1182 pvma.vm_start = 0; 1183 pvma.vm_pgoff = idx; 1184 pvma.vm_ops = NULL; 1185 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx); 1186 page = swapin_readahead(entry, gfp, &pvma, 0); 1187 mpol_free(pvma.vm_policy); 1188 return page; 1189} 1190 1191static struct page *shmem_alloc_page(gfp_t gfp, 1192 struct shmem_inode_info *info, unsigned long idx) 1193{ 1194 struct vm_area_struct pvma; 1195 struct page *page; 1196 1197 /* Create a pseudo vma that just contains the policy */ 1198 pvma.vm_start = 0; 1199 pvma.vm_pgoff = idx; 1200 pvma.vm_ops = NULL; 1201 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx); 1202 page = alloc_page_vma(gfp, &pvma, 0); 1203 mpol_free(pvma.vm_policy); 1204 return page; 1205} 1206#else /* !CONFIG_NUMA */ 1207#ifdef CONFIG_TMPFS 1208static inline int shmem_parse_mpol(char *value, unsigned short *policy, 1209 unsigned short *mode_flags, nodemask_t *policy_nodes) 1210{ 1211 return 1; 1212} 1213 1214static inline void shmem_show_mpol(struct seq_file *seq, unsigned short policy, 1215 unsigned short flags, const nodemask_t policy_nodes) 1216{ 1217} 1218#endif /* CONFIG_TMPFS */ 1219 1220static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, 1221 struct shmem_inode_info *info, unsigned long idx) 1222{ 1223 return swapin_readahead(entry, gfp, NULL, 0); 1224} 1225 1226static inline struct page *shmem_alloc_page(gfp_t gfp, 1227 struct shmem_inode_info *info, unsigned long idx) 1228{ 1229 return alloc_page(gfp); 1230} 1231#endif /* CONFIG_NUMA */ 1232 1233/* 1234 * shmem_getpage - either get the page from swap or allocate a new one 1235 * 1236 * If we allocate a new one we do not mark it dirty. That's up to the 1237 * vm. If we swap it in we mark it dirty since we also free the swap 1238 * entry since a page cannot live in both the swap and page cache 1239 */ 1240static int shmem_getpage(struct inode *inode, unsigned long idx, 1241 struct page **pagep, enum sgp_type sgp, int *type) 1242{ 1243 struct address_space *mapping = inode->i_mapping; 1244 struct shmem_inode_info *info = SHMEM_I(inode); 1245 struct shmem_sb_info *sbinfo; 1246 struct page *filepage = *pagep; 1247 struct page *swappage; 1248 swp_entry_t *entry; 1249 swp_entry_t swap; 1250 gfp_t gfp; 1251 int error; 1252 1253 if (idx >= SHMEM_MAX_INDEX) 1254 return -EFBIG; 1255 1256 if (type) 1257 *type = 0; 1258 1259 /* 1260 * Normally, filepage is NULL on entry, and either found 1261 * uptodate immediately, or allocated and zeroed, or read 1262 * in under swappage, which is then assigned to filepage. 1263 * But shmem_readpage (required for splice) passes in a locked 1264 * filepage, which may be found not uptodate by other callers 1265 * too, and may need to be copied from the swappage read in. 1266 */ 1267repeat: 1268 if (!filepage) 1269 filepage = find_lock_page(mapping, idx); 1270 if (filepage && PageUptodate(filepage)) 1271 goto done; 1272 error = 0; 1273 gfp = mapping_gfp_mask(mapping); 1274 if (!filepage) { 1275 /* 1276 * Try to preload while we can wait, to not make a habit of 1277 * draining atomic reserves; but don't latch on to this cpu. 1278 */ 1279 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM); 1280 if (error) 1281 goto failed; 1282 radix_tree_preload_end(); 1283 } 1284 1285 spin_lock(&info->lock); 1286 shmem_recalc_inode(inode); 1287 entry = shmem_swp_alloc(info, idx, sgp); 1288 if (IS_ERR(entry)) { 1289 spin_unlock(&info->lock); 1290 error = PTR_ERR(entry); 1291 goto failed; 1292 } 1293 swap = *entry; 1294 1295 if (swap.val) { 1296 /* Look it up and read it in.. */ 1297 swappage = lookup_swap_cache(swap); 1298 if (!swappage) { 1299 shmem_swp_unmap(entry); 1300 /* here we actually do the io */ 1301 if (type && !(*type & VM_FAULT_MAJOR)) { 1302 __count_vm_event(PGMAJFAULT); 1303 *type |= VM_FAULT_MAJOR; 1304 } 1305 spin_unlock(&info->lock); 1306 swappage = shmem_swapin(swap, gfp, info, idx); 1307 if (!swappage) { 1308 spin_lock(&info->lock); 1309 entry = shmem_swp_alloc(info, idx, sgp); 1310 if (IS_ERR(entry)) 1311 error = PTR_ERR(entry); 1312 else { 1313 if (entry->val == swap.val) 1314 error = -ENOMEM; 1315 shmem_swp_unmap(entry); 1316 } 1317 spin_unlock(&info->lock); 1318 if (error) 1319 goto failed; 1320 goto repeat; 1321 } 1322 wait_on_page_locked(swappage); 1323 page_cache_release(swappage); 1324 goto repeat; 1325 } 1326 1327 /* We have to do this with page locked to prevent races */ 1328 if (TestSetPageLocked(swappage)) { 1329 shmem_swp_unmap(entry); 1330 spin_unlock(&info->lock); 1331 wait_on_page_locked(swappage); 1332 page_cache_release(swappage); 1333 goto repeat; 1334 } 1335 if (PageWriteback(swappage)) { 1336 shmem_swp_unmap(entry); 1337 spin_unlock(&info->lock); 1338 wait_on_page_writeback(swappage); 1339 unlock_page(swappage); 1340 page_cache_release(swappage); 1341 goto repeat; 1342 } 1343 if (!PageUptodate(swappage)) { 1344 shmem_swp_unmap(entry); 1345 spin_unlock(&info->lock); 1346 unlock_page(swappage); 1347 page_cache_release(swappage); 1348 error = -EIO; 1349 goto failed; 1350 } 1351 1352 if (filepage) { 1353 shmem_swp_set(info, entry, 0); 1354 shmem_swp_unmap(entry); 1355 delete_from_swap_cache(swappage); 1356 spin_unlock(&info->lock); 1357 copy_highpage(filepage, swappage); 1358 unlock_page(swappage); 1359 page_cache_release(swappage); 1360 flush_dcache_page(filepage); 1361 SetPageUptodate(filepage); 1362 set_page_dirty(filepage); 1363 swap_free(swap); 1364 } else if (!(error = add_to_page_cache( 1365 swappage, mapping, idx, GFP_NOWAIT))) { 1366 info->flags |= SHMEM_PAGEIN; 1367 shmem_swp_set(info, entry, 0); 1368 shmem_swp_unmap(entry); 1369 delete_from_swap_cache(swappage); 1370 spin_unlock(&info->lock); 1371 filepage = swappage; 1372 set_page_dirty(filepage); 1373 swap_free(swap); 1374 } else { 1375 shmem_swp_unmap(entry); 1376 spin_unlock(&info->lock); 1377 unlock_page(swappage); 1378 if (error == -ENOMEM) { 1379 /* allow reclaim from this memory cgroup */ 1380 error = mem_cgroup_cache_charge(swappage, 1381 current->mm, gfp & ~__GFP_HIGHMEM); 1382 if (error) { 1383 page_cache_release(swappage); 1384 goto failed; 1385 } 1386 mem_cgroup_uncharge_page(swappage); 1387 } 1388 page_cache_release(swappage); 1389 goto repeat; 1390 } 1391 } else if (sgp == SGP_READ && !filepage) { 1392 shmem_swp_unmap(entry); 1393 filepage = find_get_page(mapping, idx); 1394 if (filepage && 1395 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) { 1396 spin_unlock(&info->lock); 1397 wait_on_page_locked(filepage); 1398 page_cache_release(filepage); 1399 filepage = NULL; 1400 goto repeat; 1401 } 1402 spin_unlock(&info->lock); 1403 } else { 1404 shmem_swp_unmap(entry); 1405 sbinfo = SHMEM_SB(inode->i_sb); 1406 if (sbinfo->max_blocks) { 1407 spin_lock(&sbinfo->stat_lock); 1408 if (sbinfo->free_blocks == 0 || 1409 shmem_acct_block(info->flags)) { 1410 spin_unlock(&sbinfo->stat_lock); 1411 spin_unlock(&info->lock); 1412 error = -ENOSPC; 1413 goto failed; 1414 } 1415 sbinfo->free_blocks--; 1416 inode->i_blocks += BLOCKS_PER_PAGE; 1417 spin_unlock(&sbinfo->stat_lock); 1418 } else if (shmem_acct_block(info->flags)) { 1419 spin_unlock(&info->lock); 1420 error = -ENOSPC; 1421 goto failed; 1422 } 1423 1424 if (!filepage) { 1425 spin_unlock(&info->lock); 1426 filepage = shmem_alloc_page(gfp, info, idx); 1427 if (!filepage) { 1428 shmem_unacct_blocks(info->flags, 1); 1429 shmem_free_blocks(inode, 1); 1430 error = -ENOMEM; 1431 goto failed; 1432 } 1433 1434 /* Precharge page while we can wait, compensate after */ 1435 error = mem_cgroup_cache_charge(filepage, current->mm, 1436 gfp & ~__GFP_HIGHMEM); 1437 if (error) { 1438 page_cache_release(filepage); 1439 shmem_unacct_blocks(info->flags, 1); 1440 shmem_free_blocks(inode, 1); 1441 filepage = NULL; 1442 goto failed; 1443 } 1444 1445 spin_lock(&info->lock); 1446 entry = shmem_swp_alloc(info, idx, sgp); 1447 if (IS_ERR(entry)) 1448 error = PTR_ERR(entry); 1449 else { 1450 swap = *entry; 1451 shmem_swp_unmap(entry); 1452 } 1453 if (error || swap.val || 0 != add_to_page_cache_lru( 1454 filepage, mapping, idx, GFP_NOWAIT)) { 1455 spin_unlock(&info->lock); 1456 mem_cgroup_uncharge_page(filepage); 1457 page_cache_release(filepage); 1458 shmem_unacct_blocks(info->flags, 1); 1459 shmem_free_blocks(inode, 1); 1460 filepage = NULL; 1461 if (error) 1462 goto failed; 1463 goto repeat; 1464 } 1465 mem_cgroup_uncharge_page(filepage); 1466 info->flags |= SHMEM_PAGEIN; 1467 } 1468 1469 info->alloced++; 1470 spin_unlock(&info->lock); 1471 clear_highpage(filepage); 1472 flush_dcache_page(filepage); 1473 SetPageUptodate(filepage); 1474 if (sgp == SGP_DIRTY) 1475 set_page_dirty(filepage); 1476 } 1477done: 1478 *pagep = filepage; 1479 return 0; 1480 1481failed: 1482 if (*pagep != filepage) { 1483 unlock_page(filepage); 1484 page_cache_release(filepage); 1485 } 1486 return error; 1487} 1488 1489static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1490{ 1491 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 1492 int error; 1493 int ret; 1494 1495 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 1496 return VM_FAULT_SIGBUS; 1497 1498 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret); 1499 if (error) 1500 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); 1501 1502 mark_page_accessed(vmf->page); 1503 return ret | VM_FAULT_LOCKED; 1504} 1505 1506#ifdef CONFIG_NUMA 1507static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new) 1508{ 1509 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1510 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new); 1511} 1512 1513static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, 1514 unsigned long addr) 1515{ 1516 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1517 unsigned long idx; 1518 1519 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; 1520 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx); 1521} 1522#endif 1523 1524int shmem_lock(struct file *file, int lock, struct user_struct *user) 1525{ 1526 struct inode *inode = file->f_path.dentry->d_inode; 1527 struct shmem_inode_info *info = SHMEM_I(inode); 1528 int retval = -ENOMEM; 1529 1530 spin_lock(&info->lock); 1531 if (lock && !(info->flags & VM_LOCKED)) { 1532 if (!user_shm_lock(inode->i_size, user)) 1533 goto out_nomem; 1534 info->flags |= VM_LOCKED; 1535 } 1536 if (!lock && (info->flags & VM_LOCKED) && user) { 1537 user_shm_unlock(inode->i_size, user); 1538 info->flags &= ~VM_LOCKED; 1539 } 1540 retval = 0; 1541out_nomem: 1542 spin_unlock(&info->lock); 1543 return retval; 1544} 1545 1546static int shmem_mmap(struct file *file, struct vm_area_struct *vma) 1547{ 1548 file_accessed(file); 1549 vma->vm_ops = &shmem_vm_ops; 1550 vma->vm_flags |= VM_CAN_NONLINEAR; 1551 return 0; 1552} 1553 1554static struct inode * 1555shmem_get_inode(struct super_block *sb, int mode, dev_t dev) 1556{ 1557 struct inode *inode; 1558 struct shmem_inode_info *info; 1559 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 1560 1561 if (shmem_reserve_inode(sb)) 1562 return NULL; 1563 1564 inode = new_inode(sb); 1565 if (inode) { 1566 inode->i_mode = mode; 1567 inode->i_uid = current->fsuid; 1568 inode->i_gid = current->fsgid; 1569 inode->i_blocks = 0; 1570 inode->i_mapping->a_ops = &shmem_aops; 1571 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; 1572 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 1573 inode->i_generation = get_seconds(); 1574 info = SHMEM_I(inode); 1575 memset(info, 0, (char *)inode - (char *)info); 1576 spin_lock_init(&info->lock); 1577 INIT_LIST_HEAD(&info->swaplist); 1578 1579 switch (mode & S_IFMT) { 1580 default: 1581 inode->i_op = &shmem_special_inode_operations; 1582 init_special_inode(inode, mode, dev); 1583 break; 1584 case S_IFREG: 1585 inode->i_op = &shmem_inode_operations; 1586 inode->i_fop = &shmem_file_operations; 1587 mpol_shared_policy_init(&info->policy, sbinfo->policy, 1588 sbinfo->flags, &sbinfo->policy_nodes); 1589 break; 1590 case S_IFDIR: 1591 inc_nlink(inode); 1592 /* Some things misbehave if size == 0 on a directory */ 1593 inode->i_size = 2 * BOGO_DIRENT_SIZE; 1594 inode->i_op = &shmem_dir_inode_operations; 1595 inode->i_fop = &simple_dir_operations; 1596 break; 1597 case S_IFLNK: 1598 /* 1599 * Must not load anything in the rbtree, 1600 * mpol_free_shared_policy will not be called. 1601 */ 1602 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, 0, 1603 NULL); 1604 break; 1605 } 1606 } else 1607 shmem_free_inode(sb); 1608 return inode; 1609} 1610 1611#ifdef CONFIG_TMPFS 1612static const struct inode_operations shmem_symlink_inode_operations; 1613static const struct inode_operations shmem_symlink_inline_operations; 1614 1615/* 1616 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin; 1617 * but providing them allows a tmpfs file to be used for splice, sendfile, and 1618 * below the loop driver, in the generic fashion that many filesystems support. 1619 */ 1620static int shmem_readpage(struct file *file, struct page *page) 1621{ 1622 struct inode *inode = page->mapping->host; 1623 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL); 1624 unlock_page(page); 1625 return error; 1626} 1627 1628static int 1629shmem_write_begin(struct file *file, struct address_space *mapping, 1630 loff_t pos, unsigned len, unsigned flags, 1631 struct page **pagep, void **fsdata) 1632{ 1633 struct inode *inode = mapping->host; 1634 pgoff_t index = pos >> PAGE_CACHE_SHIFT; 1635 *pagep = NULL; 1636 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL); 1637} 1638 1639static int 1640shmem_write_end(struct file *file, struct address_space *mapping, 1641 loff_t pos, unsigned len, unsigned copied, 1642 struct page *page, void *fsdata) 1643{ 1644 struct inode *inode = mapping->host; 1645 1646 if (pos + copied > inode->i_size) 1647 i_size_write(inode, pos + copied); 1648 1649 unlock_page(page); 1650 set_page_dirty(page); 1651 page_cache_release(page); 1652 1653 return copied; 1654} 1655 1656static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor) 1657{ 1658 struct inode *inode = filp->f_path.dentry->d_inode; 1659 struct address_space *mapping = inode->i_mapping; 1660 unsigned long index, offset; 1661 enum sgp_type sgp = SGP_READ; 1662 1663 /* 1664 * Might this read be for a stacking filesystem? Then when reading 1665 * holes of a sparse file, we actually need to allocate those pages, 1666 * and even mark them dirty, so it cannot exceed the max_blocks limit. 1667 */ 1668 if (segment_eq(get_fs(), KERNEL_DS)) 1669 sgp = SGP_DIRTY; 1670 1671 index = *ppos >> PAGE_CACHE_SHIFT; 1672 offset = *ppos & ~PAGE_CACHE_MASK; 1673 1674 for (;;) { 1675 struct page *page = NULL; 1676 unsigned long end_index, nr, ret; 1677 loff_t i_size = i_size_read(inode); 1678 1679 end_index = i_size >> PAGE_CACHE_SHIFT; 1680 if (index > end_index) 1681 break; 1682 if (index == end_index) { 1683 nr = i_size & ~PAGE_CACHE_MASK; 1684 if (nr <= offset) 1685 break; 1686 } 1687 1688 desc->error = shmem_getpage(inode, index, &page, sgp, NULL); 1689 if (desc->error) { 1690 if (desc->error == -EINVAL) 1691 desc->error = 0; 1692 break; 1693 } 1694 if (page) 1695 unlock_page(page); 1696 1697 /* 1698 * We must evaluate after, since reads (unlike writes) 1699 * are called without i_mutex protection against truncate 1700 */ 1701 nr = PAGE_CACHE_SIZE; 1702 i_size = i_size_read(inode); 1703 end_index = i_size >> PAGE_CACHE_SHIFT; 1704 if (index == end_index) { 1705 nr = i_size & ~PAGE_CACHE_MASK; 1706 if (nr <= offset) { 1707 if (page) 1708 page_cache_release(page); 1709 break; 1710 } 1711 } 1712 nr -= offset; 1713 1714 if (page) { 1715 /* 1716 * If users can be writing to this page using arbitrary 1717 * virtual addresses, take care about potential aliasing 1718 * before reading the page on the kernel side. 1719 */ 1720 if (mapping_writably_mapped(mapping)) 1721 flush_dcache_page(page); 1722 /* 1723 * Mark the page accessed if we read the beginning. 1724 */ 1725 if (!offset) 1726 mark_page_accessed(page); 1727 } else { 1728 page = ZERO_PAGE(0); 1729 page_cache_get(page); 1730 } 1731 1732 /* 1733 * Ok, we have the page, and it's up-to-date, so 1734 * now we can copy it to user space... 1735 * 1736 * The actor routine returns how many bytes were actually used.. 1737 * NOTE! This may not be the same as how much of a user buffer 1738 * we filled up (we may be padding etc), so we can only update 1739 * "pos" here (the actor routine has to update the user buffer 1740 * pointers and the remaining count). 1741 */ 1742 ret = actor(desc, page, offset, nr); 1743 offset += ret; 1744 index += offset >> PAGE_CACHE_SHIFT; 1745 offset &= ~PAGE_CACHE_MASK; 1746 1747 page_cache_release(page); 1748 if (ret != nr || !desc->count) 1749 break; 1750 1751 cond_resched(); 1752 } 1753 1754 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; 1755 file_accessed(filp); 1756} 1757 1758static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos) 1759{ 1760 read_descriptor_t desc; 1761 1762 if ((ssize_t) count < 0) 1763 return -EINVAL; 1764 if (!access_ok(VERIFY_WRITE, buf, count)) 1765 return -EFAULT; 1766 if (!count) 1767 return 0; 1768 1769 desc.written = 0; 1770 desc.count = count; 1771 desc.arg.buf = buf; 1772 desc.error = 0; 1773 1774 do_shmem_file_read(filp, ppos, &desc, file_read_actor); 1775 if (desc.written) 1776 return desc.written; 1777 return desc.error; 1778} 1779 1780static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) 1781{ 1782 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); 1783 1784 buf->f_type = TMPFS_MAGIC; 1785 buf->f_bsize = PAGE_CACHE_SIZE; 1786 buf->f_namelen = NAME_MAX; 1787 spin_lock(&sbinfo->stat_lock); 1788 if (sbinfo->max_blocks) { 1789 buf->f_blocks = sbinfo->max_blocks; 1790 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks; 1791 } 1792 if (sbinfo->max_inodes) { 1793 buf->f_files = sbinfo->max_inodes; 1794 buf->f_ffree = sbinfo->free_inodes; 1795 } 1796 /* else leave those fields 0 like simple_statfs */ 1797 spin_unlock(&sbinfo->stat_lock); 1798 return 0; 1799} 1800 1801/* 1802 * File creation. Allocate an inode, and we're done.. 1803 */ 1804static int 1805shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 1806{ 1807 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev); 1808 int error = -ENOSPC; 1809 1810 if (inode) { 1811 error = security_inode_init_security(inode, dir, NULL, NULL, 1812 NULL); 1813 if (error) { 1814 if (error != -EOPNOTSUPP) { 1815 iput(inode); 1816 return error; 1817 } 1818 } 1819 error = shmem_acl_init(inode, dir); 1820 if (error) { 1821 iput(inode); 1822 return error; 1823 } 1824 if (dir->i_mode & S_ISGID) { 1825 inode->i_gid = dir->i_gid; 1826 if (S_ISDIR(mode)) 1827 inode->i_mode |= S_ISGID; 1828 } 1829 dir->i_size += BOGO_DIRENT_SIZE; 1830 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1831 d_instantiate(dentry, inode); 1832 dget(dentry); /* Extra count - pin the dentry in core */ 1833 } 1834 return error; 1835} 1836 1837static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1838{ 1839 int error; 1840 1841 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) 1842 return error; 1843 inc_nlink(dir); 1844 return 0; 1845} 1846 1847static int shmem_create(struct inode *dir, struct dentry *dentry, int mode, 1848 struct nameidata *nd) 1849{ 1850 return shmem_mknod(dir, dentry, mode | S_IFREG, 0); 1851} 1852 1853/* 1854 * Link a file.. 1855 */ 1856static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1857{ 1858 struct inode *inode = old_dentry->d_inode; 1859 int ret; 1860 1861 /* 1862 * No ordinary (disk based) filesystem counts links as inodes; 1863 * but each new link needs a new dentry, pinning lowmem, and 1864 * tmpfs dentries cannot be pruned until they are unlinked. 1865 */ 1866 ret = shmem_reserve_inode(inode->i_sb); 1867 if (ret) 1868 goto out; 1869 1870 dir->i_size += BOGO_DIRENT_SIZE; 1871 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1872 inc_nlink(inode); 1873 atomic_inc(&inode->i_count); /* New dentry reference */ 1874 dget(dentry); /* Extra pinning count for the created dentry */ 1875 d_instantiate(dentry, inode); 1876out: 1877 return ret; 1878} 1879 1880static int shmem_unlink(struct inode *dir, struct dentry *dentry) 1881{ 1882 struct inode *inode = dentry->d_inode; 1883 1884 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) 1885 shmem_free_inode(inode->i_sb); 1886 1887 dir->i_size -= BOGO_DIRENT_SIZE; 1888 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1889 drop_nlink(inode); 1890 dput(dentry); /* Undo the count from "create" - this does all the work */ 1891 return 0; 1892} 1893 1894static int shmem_rmdir(struct inode *dir, struct dentry *dentry) 1895{ 1896 if (!simple_empty(dentry)) 1897 return -ENOTEMPTY; 1898 1899 drop_nlink(dentry->d_inode); 1900 drop_nlink(dir); 1901 return shmem_unlink(dir, dentry); 1902} 1903 1904/* 1905 * The VFS layer already does all the dentry stuff for rename, 1906 * we just have to decrement the usage count for the target if 1907 * it exists so that the VFS layer correctly free's it when it 1908 * gets overwritten. 1909 */ 1910static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) 1911{ 1912 struct inode *inode = old_dentry->d_inode; 1913 int they_are_dirs = S_ISDIR(inode->i_mode); 1914 1915 if (!simple_empty(new_dentry)) 1916 return -ENOTEMPTY; 1917 1918 if (new_dentry->d_inode) { 1919 (void) shmem_unlink(new_dir, new_dentry); 1920 if (they_are_dirs) 1921 drop_nlink(old_dir); 1922 } else if (they_are_dirs) { 1923 drop_nlink(old_dir); 1924 inc_nlink(new_dir); 1925 } 1926 1927 old_dir->i_size -= BOGO_DIRENT_SIZE; 1928 new_dir->i_size += BOGO_DIRENT_SIZE; 1929 old_dir->i_ctime = old_dir->i_mtime = 1930 new_dir->i_ctime = new_dir->i_mtime = 1931 inode->i_ctime = CURRENT_TIME; 1932 return 0; 1933} 1934 1935static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1936{ 1937 int error; 1938 int len; 1939 struct inode *inode; 1940 struct page *page = NULL; 1941 char *kaddr; 1942 struct shmem_inode_info *info; 1943 1944 len = strlen(symname) + 1; 1945 if (len > PAGE_CACHE_SIZE) 1946 return -ENAMETOOLONG; 1947 1948 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0); 1949 if (!inode) 1950 return -ENOSPC; 1951 1952 error = security_inode_init_security(inode, dir, NULL, NULL, 1953 NULL); 1954 if (error) { 1955 if (error != -EOPNOTSUPP) { 1956 iput(inode); 1957 return error; 1958 } 1959 error = 0; 1960 } 1961 1962 info = SHMEM_I(inode); 1963 inode->i_size = len-1; 1964 if (len <= (char *)inode - (char *)info) { 1965 /* do it inline */ 1966 memcpy(info, symname, len); 1967 inode->i_op = &shmem_symlink_inline_operations; 1968 } else { 1969 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); 1970 if (error) { 1971 iput(inode); 1972 return error; 1973 } 1974 unlock_page(page); 1975 inode->i_op = &shmem_symlink_inode_operations; 1976 kaddr = kmap_atomic(page, KM_USER0); 1977 memcpy(kaddr, symname, len); 1978 kunmap_atomic(kaddr, KM_USER0); 1979 set_page_dirty(page); 1980 page_cache_release(page); 1981 } 1982 if (dir->i_mode & S_ISGID) 1983 inode->i_gid = dir->i_gid; 1984 dir->i_size += BOGO_DIRENT_SIZE; 1985 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1986 d_instantiate(dentry, inode); 1987 dget(dentry); 1988 return 0; 1989} 1990 1991static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd) 1992{ 1993 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode)); 1994 return NULL; 1995} 1996 1997static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) 1998{ 1999 struct page *page = NULL; 2000 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); 2001 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page)); 2002 if (page) 2003 unlock_page(page); 2004 return page; 2005} 2006 2007static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 2008{ 2009 if (!IS_ERR(nd_get_link(nd))) { 2010 struct page *page = cookie; 2011 kunmap(page); 2012 mark_page_accessed(page); 2013 page_cache_release(page); 2014 } 2015} 2016 2017static const struct inode_operations shmem_symlink_inline_operations = { 2018 .readlink = generic_readlink, 2019 .follow_link = shmem_follow_link_inline, 2020}; 2021 2022static const struct inode_operations shmem_symlink_inode_operations = { 2023 .truncate = shmem_truncate, 2024 .readlink = generic_readlink, 2025 .follow_link = shmem_follow_link, 2026 .put_link = shmem_put_link, 2027}; 2028 2029#ifdef CONFIG_TMPFS_POSIX_ACL 2030/* 2031 * Superblocks without xattr inode operations will get security.* xattr 2032 * support from the VFS "for free". As soon as we have any other xattrs 2033 * like ACLs, we also need to implement the security.* handlers at 2034 * filesystem level, though. 2035 */ 2036 2037static size_t shmem_xattr_security_list(struct inode *inode, char *list, 2038 size_t list_len, const char *name, 2039 size_t name_len) 2040{ 2041 return security_inode_listsecurity(inode, list, list_len); 2042} 2043 2044static int shmem_xattr_security_get(struct inode *inode, const char *name, 2045 void *buffer, size_t size) 2046{ 2047 if (strcmp(name, "") == 0) 2048 return -EINVAL; 2049 return xattr_getsecurity(inode, name, buffer, size); 2050} 2051 2052static int shmem_xattr_security_set(struct inode *inode, const char *name, 2053 const void *value, size_t size, int flags) 2054{ 2055 if (strcmp(name, "") == 0) 2056 return -EINVAL; 2057 return security_inode_setsecurity(inode, name, value, size, flags); 2058} 2059 2060static struct xattr_handler shmem_xattr_security_handler = { 2061 .prefix = XATTR_SECURITY_PREFIX, 2062 .list = shmem_xattr_security_list, 2063 .get = shmem_xattr_security_get, 2064 .set = shmem_xattr_security_set, 2065}; 2066 2067static struct xattr_handler *shmem_xattr_handlers[] = { 2068 &shmem_xattr_acl_access_handler, 2069 &shmem_xattr_acl_default_handler, 2070 &shmem_xattr_security_handler, 2071 NULL 2072}; 2073#endif 2074 2075static struct dentry *shmem_get_parent(struct dentry *child) 2076{ 2077 return ERR_PTR(-ESTALE); 2078} 2079 2080static int shmem_match(struct inode *ino, void *vfh) 2081{ 2082 __u32 *fh = vfh; 2083 __u64 inum = fh[2]; 2084 inum = (inum << 32) | fh[1]; 2085 return ino->i_ino == inum && fh[0] == ino->i_generation; 2086} 2087 2088static struct dentry *shmem_fh_to_dentry(struct super_block *sb, 2089 struct fid *fid, int fh_len, int fh_type) 2090{ 2091 struct inode *inode; 2092 struct dentry *dentry = NULL; 2093 u64 inum = fid->raw[2]; 2094 inum = (inum << 32) | fid->raw[1]; 2095 2096 if (fh_len < 3) 2097 return NULL; 2098 2099 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), 2100 shmem_match, fid->raw); 2101 if (inode) { 2102 dentry = d_find_alias(inode); 2103 iput(inode); 2104 } 2105 2106 return dentry; 2107} 2108 2109static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len, 2110 int connectable) 2111{ 2112 struct inode *inode = dentry->d_inode; 2113 2114 if (*len < 3) 2115 return 255; 2116 2117 if (hlist_unhashed(&inode->i_hash)) { 2118 /* Unfortunately insert_inode_hash is not idempotent, 2119 * so as we hash inodes here rather than at creation 2120 * time, we need a lock to ensure we only try 2121 * to do it once 2122 */ 2123 static DEFINE_SPINLOCK(lock); 2124 spin_lock(&lock); 2125 if (hlist_unhashed(&inode->i_hash)) 2126 __insert_inode_hash(inode, 2127 inode->i_ino + inode->i_generation); 2128 spin_unlock(&lock); 2129 } 2130 2131 fh[0] = inode->i_generation; 2132 fh[1] = inode->i_ino; 2133 fh[2] = ((__u64)inode->i_ino) >> 32; 2134 2135 *len = 3; 2136 return 1; 2137} 2138 2139static const struct export_operations shmem_export_ops = { 2140 .get_parent = shmem_get_parent, 2141 .encode_fh = shmem_encode_fh, 2142 .fh_to_dentry = shmem_fh_to_dentry, 2143}; 2144 2145static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, 2146 bool remount) 2147{ 2148 char *this_char, *value, *rest; 2149 2150 while (options != NULL) { 2151 this_char = options; 2152 for (;;) { 2153 /* 2154 * NUL-terminate this option: unfortunately, 2155 * mount options form a comma-separated list, 2156 * but mpol's nodelist may also contain commas. 2157 */ 2158 options = strchr(options, ','); 2159 if (options == NULL) 2160 break; 2161 options++; 2162 if (!isdigit(*options)) { 2163 options[-1] = '\0'; 2164 break; 2165 } 2166 } 2167 if (!*this_char) 2168 continue; 2169 if ((value = strchr(this_char,'=')) != NULL) { 2170 *value++ = 0; 2171 } else { 2172 printk(KERN_ERR 2173 "tmpfs: No value for mount option '%s'\n", 2174 this_char); 2175 return 1; 2176 } 2177 2178 if (!strcmp(this_char,"size")) { 2179 unsigned long long size; 2180 size = memparse(value,&rest); 2181 if (*rest == '%') { 2182 size <<= PAGE_SHIFT; 2183 size *= totalram_pages; 2184 do_div(size, 100); 2185 rest++; 2186 } 2187 if (*rest) 2188 goto bad_val; 2189 sbinfo->max_blocks = 2190 DIV_ROUND_UP(size, PAGE_CACHE_SIZE); 2191 } else if (!strcmp(this_char,"nr_blocks")) { 2192 sbinfo->max_blocks = memparse(value, &rest); 2193 if (*rest) 2194 goto bad_val; 2195 } else if (!strcmp(this_char,"nr_inodes")) { 2196 sbinfo->max_inodes = memparse(value, &rest); 2197 if (*rest) 2198 goto bad_val; 2199 } else if (!strcmp(this_char,"mode")) { 2200 if (remount) 2201 continue; 2202 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; 2203 if (*rest) 2204 goto bad_val; 2205 } else if (!strcmp(this_char,"uid")) { 2206 if (remount) 2207 continue; 2208 sbinfo->uid = simple_strtoul(value, &rest, 0); 2209 if (*rest) 2210 goto bad_val; 2211 } else if (!strcmp(this_char,"gid")) { 2212 if (remount) 2213 continue; 2214 sbinfo->gid = simple_strtoul(value, &rest, 0); 2215 if (*rest) 2216 goto bad_val; 2217 } else if (!strcmp(this_char,"mpol")) { 2218 if (shmem_parse_mpol(value, &sbinfo->policy, 2219 &sbinfo->flags, &sbinfo->policy_nodes)) 2220 goto bad_val; 2221 } else { 2222 printk(KERN_ERR "tmpfs: Bad mount option %s\n", 2223 this_char); 2224 return 1; 2225 } 2226 } 2227 return 0; 2228 2229bad_val: 2230 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", 2231 value, this_char); 2232 return 1; 2233 2234} 2235 2236static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) 2237{ 2238 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 2239 struct shmem_sb_info config = *sbinfo; 2240 unsigned long blocks; 2241 unsigned long inodes; 2242 int error = -EINVAL; 2243 2244 if (shmem_parse_options(data, &config, true)) 2245 return error; 2246 2247 spin_lock(&sbinfo->stat_lock); 2248 blocks = sbinfo->max_blocks - sbinfo->free_blocks; 2249 inodes = sbinfo->max_inodes - sbinfo->free_inodes; 2250 if (config.max_blocks < blocks) 2251 goto out; 2252 if (config.max_inodes < inodes) 2253 goto out; 2254 /* 2255 * Those tests also disallow limited->unlimited while any are in 2256 * use, so i_blocks will always be zero when max_blocks is zero; 2257 * but we must separately disallow unlimited->limited, because 2258 * in that case we have no record of how much is already in use. 2259 */ 2260 if (config.max_blocks && !sbinfo->max_blocks) 2261 goto out; 2262 if (config.max_inodes && !sbinfo->max_inodes) 2263 goto out; 2264 2265 error = 0; 2266 sbinfo->max_blocks = config.max_blocks; 2267 sbinfo->free_blocks = config.max_blocks - blocks; 2268 sbinfo->max_inodes = config.max_inodes; 2269 sbinfo->free_inodes = config.max_inodes - inodes; 2270 sbinfo->policy = config.policy; 2271 sbinfo->flags = config.flags; 2272 sbinfo->policy_nodes = config.policy_nodes; 2273out: 2274 spin_unlock(&sbinfo->stat_lock); 2275 return error; 2276} 2277 2278static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs) 2279{ 2280 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb); 2281 2282 if (sbinfo->max_blocks != shmem_default_max_blocks()) 2283 seq_printf(seq, ",size=%luk", 2284 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10)); 2285 if (sbinfo->max_inodes != shmem_default_max_inodes()) 2286 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); 2287 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) 2288 seq_printf(seq, ",mode=%03o", sbinfo->mode); 2289 if (sbinfo->uid != 0) 2290 seq_printf(seq, ",uid=%u", sbinfo->uid); 2291 if (sbinfo->gid != 0) 2292 seq_printf(seq, ",gid=%u", sbinfo->gid); 2293 shmem_show_mpol(seq, sbinfo->policy, sbinfo->flags, 2294 sbinfo->policy_nodes); 2295 return 0; 2296} 2297#endif /* CONFIG_TMPFS */ 2298 2299static void shmem_put_super(struct super_block *sb) 2300{ 2301 kfree(sb->s_fs_info); 2302 sb->s_fs_info = NULL; 2303} 2304 2305static int shmem_fill_super(struct super_block *sb, 2306 void *data, int silent) 2307{ 2308 struct inode *inode; 2309 struct dentry *root; 2310 struct shmem_sb_info *sbinfo; 2311 int err = -ENOMEM; 2312 2313 /* Round up to L1_CACHE_BYTES to resist false sharing */ 2314 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info), 2315 L1_CACHE_BYTES), GFP_KERNEL); 2316 if (!sbinfo) 2317 return -ENOMEM; 2318 2319 sbinfo->max_blocks = 0; 2320 sbinfo->max_inodes = 0; 2321 sbinfo->mode = S_IRWXUGO | S_ISVTX; 2322 sbinfo->uid = current->fsuid; 2323 sbinfo->gid = current->fsgid; 2324 sbinfo->policy = MPOL_DEFAULT; 2325 sbinfo->flags = 0; 2326 sbinfo->policy_nodes = node_states[N_HIGH_MEMORY]; 2327 sb->s_fs_info = sbinfo; 2328 2329#ifdef CONFIG_TMPFS 2330 /* 2331 * Per default we only allow half of the physical ram per 2332 * tmpfs instance, limiting inodes to one per page of lowmem; 2333 * but the internal instance is left unlimited. 2334 */ 2335 if (!(sb->s_flags & MS_NOUSER)) { 2336 sbinfo->max_blocks = shmem_default_max_blocks(); 2337 sbinfo->max_inodes = shmem_default_max_inodes(); 2338 if (shmem_parse_options(data, sbinfo, false)) { 2339 err = -EINVAL; 2340 goto failed; 2341 } 2342 } 2343 sb->s_export_op = &shmem_export_ops; 2344#else 2345 sb->s_flags |= MS_NOUSER; 2346#endif 2347 2348 spin_lock_init(&sbinfo->stat_lock); 2349 sbinfo->free_blocks = sbinfo->max_blocks; 2350 sbinfo->free_inodes = sbinfo->max_inodes; 2351 2352 sb->s_maxbytes = SHMEM_MAX_BYTES; 2353 sb->s_blocksize = PAGE_CACHE_SIZE; 2354 sb->s_blocksize_bits = PAGE_CACHE_SHIFT; 2355 sb->s_magic = TMPFS_MAGIC; 2356 sb->s_op = &shmem_ops; 2357 sb->s_time_gran = 1; 2358#ifdef CONFIG_TMPFS_POSIX_ACL 2359 sb->s_xattr = shmem_xattr_handlers; 2360 sb->s_flags |= MS_POSIXACL; 2361#endif 2362 2363 inode = shmem_get_inode(sb, S_IFDIR | sbinfo->mode, 0); 2364 if (!inode) 2365 goto failed; 2366 inode->i_uid = sbinfo->uid; 2367 inode->i_gid = sbinfo->gid; 2368 root = d_alloc_root(inode); 2369 if (!root) 2370 goto failed_iput; 2371 sb->s_root = root; 2372 return 0; 2373 2374failed_iput: 2375 iput(inode); 2376failed: 2377 shmem_put_super(sb); 2378 return err; 2379} 2380 2381static struct kmem_cache *shmem_inode_cachep; 2382 2383static struct inode *shmem_alloc_inode(struct super_block *sb) 2384{ 2385 struct shmem_inode_info *p; 2386 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); 2387 if (!p) 2388 return NULL; 2389 return &p->vfs_inode; 2390} 2391 2392static void shmem_destroy_inode(struct inode *inode) 2393{ 2394 if ((inode->i_mode & S_IFMT) == S_IFREG) { 2395 /* only struct inode is valid if it's an inline symlink */ 2396 mpol_free_shared_policy(&SHMEM_I(inode)->policy); 2397 } 2398 shmem_acl_destroy_inode(inode); 2399 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); 2400} 2401 2402static void init_once(struct kmem_cache *cachep, void *foo) 2403{ 2404 struct shmem_inode_info *p = (struct shmem_inode_info *) foo; 2405 2406 inode_init_once(&p->vfs_inode); 2407#ifdef CONFIG_TMPFS_POSIX_ACL 2408 p->i_acl = NULL; 2409 p->i_default_acl = NULL; 2410#endif 2411} 2412 2413static int init_inodecache(void) 2414{ 2415 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", 2416 sizeof(struct shmem_inode_info), 2417 0, SLAB_PANIC, init_once); 2418 return 0; 2419} 2420 2421static void destroy_inodecache(void) 2422{ 2423 kmem_cache_destroy(shmem_inode_cachep); 2424} 2425 2426static const struct address_space_operations shmem_aops = { 2427 .writepage = shmem_writepage, 2428 .set_page_dirty = __set_page_dirty_no_writeback, 2429#ifdef CONFIG_TMPFS 2430 .readpage = shmem_readpage, 2431 .write_begin = shmem_write_begin, 2432 .write_end = shmem_write_end, 2433#endif 2434 .migratepage = migrate_page, 2435}; 2436 2437static const struct file_operations shmem_file_operations = { 2438 .mmap = shmem_mmap, 2439#ifdef CONFIG_TMPFS 2440 .llseek = generic_file_llseek, 2441 .read = shmem_file_read, 2442 .write = do_sync_write, 2443 .aio_write = generic_file_aio_write, 2444 .fsync = simple_sync_file, 2445 .splice_read = generic_file_splice_read, 2446 .splice_write = generic_file_splice_write, 2447#endif 2448}; 2449 2450static const struct inode_operations shmem_inode_operations = { 2451 .truncate = shmem_truncate, 2452 .setattr = shmem_notify_change, 2453 .truncate_range = shmem_truncate_range, 2454#ifdef CONFIG_TMPFS_POSIX_ACL 2455 .setxattr = generic_setxattr, 2456 .getxattr = generic_getxattr, 2457 .listxattr = generic_listxattr, 2458 .removexattr = generic_removexattr, 2459 .permission = shmem_permission, 2460#endif 2461 2462}; 2463 2464static const struct inode_operations shmem_dir_inode_operations = { 2465#ifdef CONFIG_TMPFS 2466 .create = shmem_create, 2467 .lookup = simple_lookup, 2468 .link = shmem_link, 2469 .unlink = shmem_unlink, 2470 .symlink = shmem_symlink, 2471 .mkdir = shmem_mkdir, 2472 .rmdir = shmem_rmdir, 2473 .mknod = shmem_mknod, 2474 .rename = shmem_rename, 2475#endif 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 inode_operations shmem_special_inode_operations = { 2487#ifdef CONFIG_TMPFS_POSIX_ACL 2488 .setattr = shmem_notify_change, 2489 .setxattr = generic_setxattr, 2490 .getxattr = generic_getxattr, 2491 .listxattr = generic_listxattr, 2492 .removexattr = generic_removexattr, 2493 .permission = shmem_permission, 2494#endif 2495}; 2496 2497static const struct super_operations shmem_ops = { 2498 .alloc_inode = shmem_alloc_inode, 2499 .destroy_inode = shmem_destroy_inode, 2500#ifdef CONFIG_TMPFS 2501 .statfs = shmem_statfs, 2502 .remount_fs = shmem_remount_fs, 2503 .show_options = shmem_show_options, 2504#endif 2505 .delete_inode = shmem_delete_inode, 2506 .drop_inode = generic_delete_inode, 2507 .put_super = shmem_put_super, 2508}; 2509 2510static struct vm_operations_struct shmem_vm_ops = { 2511 .fault = shmem_fault, 2512#ifdef CONFIG_NUMA 2513 .set_policy = shmem_set_policy, 2514 .get_policy = shmem_get_policy, 2515#endif 2516}; 2517 2518 2519static int shmem_get_sb(struct file_system_type *fs_type, 2520 int flags, const char *dev_name, void *data, struct vfsmount *mnt) 2521{ 2522 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt); 2523} 2524 2525static struct file_system_type tmpfs_fs_type = { 2526 .owner = THIS_MODULE, 2527 .name = "tmpfs", 2528 .get_sb = shmem_get_sb, 2529 .kill_sb = kill_litter_super, 2530}; 2531static struct vfsmount *shm_mnt; 2532 2533static int __init init_tmpfs(void) 2534{ 2535 int error; 2536 2537 error = bdi_init(&shmem_backing_dev_info); 2538 if (error) 2539 goto out4; 2540 2541 error = init_inodecache(); 2542 if (error) 2543 goto out3; 2544 2545 error = register_filesystem(&tmpfs_fs_type); 2546 if (error) { 2547 printk(KERN_ERR "Could not register tmpfs\n"); 2548 goto out2; 2549 } 2550 2551 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER, 2552 tmpfs_fs_type.name, NULL); 2553 if (IS_ERR(shm_mnt)) { 2554 error = PTR_ERR(shm_mnt); 2555 printk(KERN_ERR "Could not kern_mount tmpfs\n"); 2556 goto out1; 2557 } 2558 return 0; 2559 2560out1: 2561 unregister_filesystem(&tmpfs_fs_type); 2562out2: 2563 destroy_inodecache(); 2564out3: 2565 bdi_destroy(&shmem_backing_dev_info); 2566out4: 2567 shm_mnt = ERR_PTR(error); 2568 return error; 2569} 2570module_init(init_tmpfs) 2571 2572/** 2573 * shmem_file_setup - get an unlinked file living in tmpfs 2574 * @name: name for dentry (to be seen in /proc/<pid>/maps 2575 * @size: size to be set for the file 2576 * @flags: vm_flags 2577 */ 2578struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags) 2579{ 2580 int error; 2581 struct file *file; 2582 struct inode *inode; 2583 struct dentry *dentry, *root; 2584 struct qstr this; 2585 2586 if (IS_ERR(shm_mnt)) 2587 return (void *)shm_mnt; 2588 2589 if (size < 0 || size > SHMEM_MAX_BYTES) 2590 return ERR_PTR(-EINVAL); 2591 2592 if (shmem_acct_size(flags, size)) 2593 return ERR_PTR(-ENOMEM); 2594 2595 error = -ENOMEM; 2596 this.name = name; 2597 this.len = strlen(name); 2598 this.hash = 0; /* will go */ 2599 root = shm_mnt->mnt_root; 2600 dentry = d_alloc(root, &this); 2601 if (!dentry) 2602 goto put_memory; 2603 2604 error = -ENFILE; 2605 file = get_empty_filp(); 2606 if (!file) 2607 goto put_dentry; 2608 2609 error = -ENOSPC; 2610 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0); 2611 if (!inode) 2612 goto close_file; 2613 2614 SHMEM_I(inode)->flags = flags & VM_ACCOUNT; 2615 d_instantiate(dentry, inode); 2616 inode->i_size = size; 2617 inode->i_nlink = 0; /* It is unlinked */ 2618 init_file(file, shm_mnt, dentry, FMODE_WRITE | FMODE_READ, 2619 &shmem_file_operations); 2620 return file; 2621 2622close_file: 2623 put_filp(file); 2624put_dentry: 2625 dput(dentry); 2626put_memory: 2627 shmem_unacct_size(flags, size); 2628 return ERR_PTR(error); 2629} 2630 2631/** 2632 * shmem_zero_setup - setup a shared anonymous mapping 2633 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff 2634 */ 2635int shmem_zero_setup(struct vm_area_struct *vma) 2636{ 2637 struct file *file; 2638 loff_t size = vma->vm_end - vma->vm_start; 2639 2640 file = shmem_file_setup("dev/zero", size, vma->vm_flags); 2641 if (IS_ERR(file)) 2642 return PTR_ERR(file); 2643 2644 if (vma->vm_file) 2645 fput(vma->vm_file); 2646 vma->vm_file = file; 2647 vma->vm_ops = &shmem_vm_ops; 2648 return 0; 2649} 2650