shmem.c revision 095f1fc4ebf36c64fddf9b6db29b1ab5517378e6
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 void shmem_show_mpol(struct seq_file *seq, unsigned short mode, 1083 unsigned short flags, const nodemask_t policy_nodes) 1084{ 1085 struct mempolicy temp; 1086 char buffer[64]; 1087 1088 if (mode == MPOL_DEFAULT) 1089 return; /* show nothing */ 1090 1091 temp.mode = mode; 1092 temp.flags = flags; 1093 temp.v.nodes = policy_nodes; 1094 1095 mpol_to_str(buffer, sizeof(buffer), &temp); 1096 1097 seq_printf(seq, ",mpol=%s", buffer); 1098} 1099#endif /* CONFIG_TMPFS */ 1100 1101static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, 1102 struct shmem_inode_info *info, unsigned long idx) 1103{ 1104 struct mempolicy mpol, *spol; 1105 struct vm_area_struct pvma; 1106 struct page *page; 1107 1108 spol = mpol_cond_copy(&mpol, 1109 mpol_shared_policy_lookup(&info->policy, idx)); 1110 1111 /* Create a pseudo vma that just contains the policy */ 1112 pvma.vm_start = 0; 1113 pvma.vm_pgoff = idx; 1114 pvma.vm_ops = NULL; 1115 pvma.vm_policy = spol; 1116 page = swapin_readahead(entry, gfp, &pvma, 0); 1117 return page; 1118} 1119 1120static struct page *shmem_alloc_page(gfp_t gfp, 1121 struct shmem_inode_info *info, unsigned long idx) 1122{ 1123 struct vm_area_struct pvma; 1124 1125 /* Create a pseudo vma that just contains the policy */ 1126 pvma.vm_start = 0; 1127 pvma.vm_pgoff = idx; 1128 pvma.vm_ops = NULL; 1129 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx); 1130 1131 /* 1132 * alloc_page_vma() will drop the shared policy reference 1133 */ 1134 return alloc_page_vma(gfp, &pvma, 0); 1135} 1136#else /* !CONFIG_NUMA */ 1137#ifdef CONFIG_TMPFS 1138static inline void shmem_show_mpol(struct seq_file *seq, unsigned short policy, 1139 unsigned short flags, const nodemask_t policy_nodes) 1140{ 1141} 1142#endif /* CONFIG_TMPFS */ 1143 1144static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, 1145 struct shmem_inode_info *info, unsigned long idx) 1146{ 1147 return swapin_readahead(entry, gfp, NULL, 0); 1148} 1149 1150static inline struct page *shmem_alloc_page(gfp_t gfp, 1151 struct shmem_inode_info *info, unsigned long idx) 1152{ 1153 return alloc_page(gfp); 1154} 1155#endif /* CONFIG_NUMA */ 1156 1157/* 1158 * shmem_getpage - either get the page from swap or allocate a new one 1159 * 1160 * If we allocate a new one we do not mark it dirty. That's up to the 1161 * vm. If we swap it in we mark it dirty since we also free the swap 1162 * entry since a page cannot live in both the swap and page cache 1163 */ 1164static int shmem_getpage(struct inode *inode, unsigned long idx, 1165 struct page **pagep, enum sgp_type sgp, int *type) 1166{ 1167 struct address_space *mapping = inode->i_mapping; 1168 struct shmem_inode_info *info = SHMEM_I(inode); 1169 struct shmem_sb_info *sbinfo; 1170 struct page *filepage = *pagep; 1171 struct page *swappage; 1172 swp_entry_t *entry; 1173 swp_entry_t swap; 1174 gfp_t gfp; 1175 int error; 1176 1177 if (idx >= SHMEM_MAX_INDEX) 1178 return -EFBIG; 1179 1180 if (type) 1181 *type = 0; 1182 1183 /* 1184 * Normally, filepage is NULL on entry, and either found 1185 * uptodate immediately, or allocated and zeroed, or read 1186 * in under swappage, which is then assigned to filepage. 1187 * But shmem_readpage (required for splice) passes in a locked 1188 * filepage, which may be found not uptodate by other callers 1189 * too, and may need to be copied from the swappage read in. 1190 */ 1191repeat: 1192 if (!filepage) 1193 filepage = find_lock_page(mapping, idx); 1194 if (filepage && PageUptodate(filepage)) 1195 goto done; 1196 error = 0; 1197 gfp = mapping_gfp_mask(mapping); 1198 if (!filepage) { 1199 /* 1200 * Try to preload while we can wait, to not make a habit of 1201 * draining atomic reserves; but don't latch on to this cpu. 1202 */ 1203 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM); 1204 if (error) 1205 goto failed; 1206 radix_tree_preload_end(); 1207 } 1208 1209 spin_lock(&info->lock); 1210 shmem_recalc_inode(inode); 1211 entry = shmem_swp_alloc(info, idx, sgp); 1212 if (IS_ERR(entry)) { 1213 spin_unlock(&info->lock); 1214 error = PTR_ERR(entry); 1215 goto failed; 1216 } 1217 swap = *entry; 1218 1219 if (swap.val) { 1220 /* Look it up and read it in.. */ 1221 swappage = lookup_swap_cache(swap); 1222 if (!swappage) { 1223 shmem_swp_unmap(entry); 1224 /* here we actually do the io */ 1225 if (type && !(*type & VM_FAULT_MAJOR)) { 1226 __count_vm_event(PGMAJFAULT); 1227 *type |= VM_FAULT_MAJOR; 1228 } 1229 spin_unlock(&info->lock); 1230 swappage = shmem_swapin(swap, gfp, info, idx); 1231 if (!swappage) { 1232 spin_lock(&info->lock); 1233 entry = shmem_swp_alloc(info, idx, sgp); 1234 if (IS_ERR(entry)) 1235 error = PTR_ERR(entry); 1236 else { 1237 if (entry->val == swap.val) 1238 error = -ENOMEM; 1239 shmem_swp_unmap(entry); 1240 } 1241 spin_unlock(&info->lock); 1242 if (error) 1243 goto failed; 1244 goto repeat; 1245 } 1246 wait_on_page_locked(swappage); 1247 page_cache_release(swappage); 1248 goto repeat; 1249 } 1250 1251 /* We have to do this with page locked to prevent races */ 1252 if (TestSetPageLocked(swappage)) { 1253 shmem_swp_unmap(entry); 1254 spin_unlock(&info->lock); 1255 wait_on_page_locked(swappage); 1256 page_cache_release(swappage); 1257 goto repeat; 1258 } 1259 if (PageWriteback(swappage)) { 1260 shmem_swp_unmap(entry); 1261 spin_unlock(&info->lock); 1262 wait_on_page_writeback(swappage); 1263 unlock_page(swappage); 1264 page_cache_release(swappage); 1265 goto repeat; 1266 } 1267 if (!PageUptodate(swappage)) { 1268 shmem_swp_unmap(entry); 1269 spin_unlock(&info->lock); 1270 unlock_page(swappage); 1271 page_cache_release(swappage); 1272 error = -EIO; 1273 goto failed; 1274 } 1275 1276 if (filepage) { 1277 shmem_swp_set(info, entry, 0); 1278 shmem_swp_unmap(entry); 1279 delete_from_swap_cache(swappage); 1280 spin_unlock(&info->lock); 1281 copy_highpage(filepage, swappage); 1282 unlock_page(swappage); 1283 page_cache_release(swappage); 1284 flush_dcache_page(filepage); 1285 SetPageUptodate(filepage); 1286 set_page_dirty(filepage); 1287 swap_free(swap); 1288 } else if (!(error = add_to_page_cache( 1289 swappage, mapping, idx, GFP_NOWAIT))) { 1290 info->flags |= SHMEM_PAGEIN; 1291 shmem_swp_set(info, entry, 0); 1292 shmem_swp_unmap(entry); 1293 delete_from_swap_cache(swappage); 1294 spin_unlock(&info->lock); 1295 filepage = swappage; 1296 set_page_dirty(filepage); 1297 swap_free(swap); 1298 } else { 1299 shmem_swp_unmap(entry); 1300 spin_unlock(&info->lock); 1301 unlock_page(swappage); 1302 if (error == -ENOMEM) { 1303 /* allow reclaim from this memory cgroup */ 1304 error = mem_cgroup_cache_charge(swappage, 1305 current->mm, gfp & ~__GFP_HIGHMEM); 1306 if (error) { 1307 page_cache_release(swappage); 1308 goto failed; 1309 } 1310 mem_cgroup_uncharge_page(swappage); 1311 } 1312 page_cache_release(swappage); 1313 goto repeat; 1314 } 1315 } else if (sgp == SGP_READ && !filepage) { 1316 shmem_swp_unmap(entry); 1317 filepage = find_get_page(mapping, idx); 1318 if (filepage && 1319 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) { 1320 spin_unlock(&info->lock); 1321 wait_on_page_locked(filepage); 1322 page_cache_release(filepage); 1323 filepage = NULL; 1324 goto repeat; 1325 } 1326 spin_unlock(&info->lock); 1327 } else { 1328 shmem_swp_unmap(entry); 1329 sbinfo = SHMEM_SB(inode->i_sb); 1330 if (sbinfo->max_blocks) { 1331 spin_lock(&sbinfo->stat_lock); 1332 if (sbinfo->free_blocks == 0 || 1333 shmem_acct_block(info->flags)) { 1334 spin_unlock(&sbinfo->stat_lock); 1335 spin_unlock(&info->lock); 1336 error = -ENOSPC; 1337 goto failed; 1338 } 1339 sbinfo->free_blocks--; 1340 inode->i_blocks += BLOCKS_PER_PAGE; 1341 spin_unlock(&sbinfo->stat_lock); 1342 } else if (shmem_acct_block(info->flags)) { 1343 spin_unlock(&info->lock); 1344 error = -ENOSPC; 1345 goto failed; 1346 } 1347 1348 if (!filepage) { 1349 spin_unlock(&info->lock); 1350 filepage = shmem_alloc_page(gfp, info, idx); 1351 if (!filepage) { 1352 shmem_unacct_blocks(info->flags, 1); 1353 shmem_free_blocks(inode, 1); 1354 error = -ENOMEM; 1355 goto failed; 1356 } 1357 1358 /* Precharge page while we can wait, compensate after */ 1359 error = mem_cgroup_cache_charge(filepage, current->mm, 1360 gfp & ~__GFP_HIGHMEM); 1361 if (error) { 1362 page_cache_release(filepage); 1363 shmem_unacct_blocks(info->flags, 1); 1364 shmem_free_blocks(inode, 1); 1365 filepage = NULL; 1366 goto failed; 1367 } 1368 1369 spin_lock(&info->lock); 1370 entry = shmem_swp_alloc(info, idx, sgp); 1371 if (IS_ERR(entry)) 1372 error = PTR_ERR(entry); 1373 else { 1374 swap = *entry; 1375 shmem_swp_unmap(entry); 1376 } 1377 if (error || swap.val || 0 != add_to_page_cache_lru( 1378 filepage, mapping, idx, GFP_NOWAIT)) { 1379 spin_unlock(&info->lock); 1380 mem_cgroup_uncharge_page(filepage); 1381 page_cache_release(filepage); 1382 shmem_unacct_blocks(info->flags, 1); 1383 shmem_free_blocks(inode, 1); 1384 filepage = NULL; 1385 if (error) 1386 goto failed; 1387 goto repeat; 1388 } 1389 mem_cgroup_uncharge_page(filepage); 1390 info->flags |= SHMEM_PAGEIN; 1391 } 1392 1393 info->alloced++; 1394 spin_unlock(&info->lock); 1395 clear_highpage(filepage); 1396 flush_dcache_page(filepage); 1397 SetPageUptodate(filepage); 1398 if (sgp == SGP_DIRTY) 1399 set_page_dirty(filepage); 1400 } 1401done: 1402 *pagep = filepage; 1403 return 0; 1404 1405failed: 1406 if (*pagep != filepage) { 1407 unlock_page(filepage); 1408 page_cache_release(filepage); 1409 } 1410 return error; 1411} 1412 1413static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1414{ 1415 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 1416 int error; 1417 int ret; 1418 1419 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 1420 return VM_FAULT_SIGBUS; 1421 1422 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret); 1423 if (error) 1424 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); 1425 1426 mark_page_accessed(vmf->page); 1427 return ret | VM_FAULT_LOCKED; 1428} 1429 1430#ifdef CONFIG_NUMA 1431static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new) 1432{ 1433 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1434 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new); 1435} 1436 1437static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, 1438 unsigned long addr) 1439{ 1440 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1441 unsigned long idx; 1442 1443 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; 1444 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx); 1445} 1446#endif 1447 1448int shmem_lock(struct file *file, int lock, struct user_struct *user) 1449{ 1450 struct inode *inode = file->f_path.dentry->d_inode; 1451 struct shmem_inode_info *info = SHMEM_I(inode); 1452 int retval = -ENOMEM; 1453 1454 spin_lock(&info->lock); 1455 if (lock && !(info->flags & VM_LOCKED)) { 1456 if (!user_shm_lock(inode->i_size, user)) 1457 goto out_nomem; 1458 info->flags |= VM_LOCKED; 1459 } 1460 if (!lock && (info->flags & VM_LOCKED) && user) { 1461 user_shm_unlock(inode->i_size, user); 1462 info->flags &= ~VM_LOCKED; 1463 } 1464 retval = 0; 1465out_nomem: 1466 spin_unlock(&info->lock); 1467 return retval; 1468} 1469 1470static int shmem_mmap(struct file *file, struct vm_area_struct *vma) 1471{ 1472 file_accessed(file); 1473 vma->vm_ops = &shmem_vm_ops; 1474 vma->vm_flags |= VM_CAN_NONLINEAR; 1475 return 0; 1476} 1477 1478static struct inode * 1479shmem_get_inode(struct super_block *sb, int mode, dev_t dev) 1480{ 1481 struct inode *inode; 1482 struct shmem_inode_info *info; 1483 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 1484 1485 if (shmem_reserve_inode(sb)) 1486 return NULL; 1487 1488 inode = new_inode(sb); 1489 if (inode) { 1490 inode->i_mode = mode; 1491 inode->i_uid = current->fsuid; 1492 inode->i_gid = current->fsgid; 1493 inode->i_blocks = 0; 1494 inode->i_mapping->a_ops = &shmem_aops; 1495 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; 1496 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 1497 inode->i_generation = get_seconds(); 1498 info = SHMEM_I(inode); 1499 memset(info, 0, (char *)inode - (char *)info); 1500 spin_lock_init(&info->lock); 1501 INIT_LIST_HEAD(&info->swaplist); 1502 1503 switch (mode & S_IFMT) { 1504 default: 1505 inode->i_op = &shmem_special_inode_operations; 1506 init_special_inode(inode, mode, dev); 1507 break; 1508 case S_IFREG: 1509 inode->i_op = &shmem_inode_operations; 1510 inode->i_fop = &shmem_file_operations; 1511 mpol_shared_policy_init(&info->policy, sbinfo->policy, 1512 sbinfo->flags, &sbinfo->policy_nodes); 1513 break; 1514 case S_IFDIR: 1515 inc_nlink(inode); 1516 /* Some things misbehave if size == 0 on a directory */ 1517 inode->i_size = 2 * BOGO_DIRENT_SIZE; 1518 inode->i_op = &shmem_dir_inode_operations; 1519 inode->i_fop = &simple_dir_operations; 1520 break; 1521 case S_IFLNK: 1522 /* 1523 * Must not load anything in the rbtree, 1524 * mpol_free_shared_policy will not be called. 1525 */ 1526 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, 0, 1527 NULL); 1528 break; 1529 } 1530 } else 1531 shmem_free_inode(sb); 1532 return inode; 1533} 1534 1535#ifdef CONFIG_TMPFS 1536static const struct inode_operations shmem_symlink_inode_operations; 1537static const struct inode_operations shmem_symlink_inline_operations; 1538 1539/* 1540 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin; 1541 * but providing them allows a tmpfs file to be used for splice, sendfile, and 1542 * below the loop driver, in the generic fashion that many filesystems support. 1543 */ 1544static int shmem_readpage(struct file *file, struct page *page) 1545{ 1546 struct inode *inode = page->mapping->host; 1547 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL); 1548 unlock_page(page); 1549 return error; 1550} 1551 1552static int 1553shmem_write_begin(struct file *file, struct address_space *mapping, 1554 loff_t pos, unsigned len, unsigned flags, 1555 struct page **pagep, void **fsdata) 1556{ 1557 struct inode *inode = mapping->host; 1558 pgoff_t index = pos >> PAGE_CACHE_SHIFT; 1559 *pagep = NULL; 1560 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL); 1561} 1562 1563static int 1564shmem_write_end(struct file *file, struct address_space *mapping, 1565 loff_t pos, unsigned len, unsigned copied, 1566 struct page *page, void *fsdata) 1567{ 1568 struct inode *inode = mapping->host; 1569 1570 if (pos + copied > inode->i_size) 1571 i_size_write(inode, pos + copied); 1572 1573 unlock_page(page); 1574 set_page_dirty(page); 1575 page_cache_release(page); 1576 1577 return copied; 1578} 1579 1580static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor) 1581{ 1582 struct inode *inode = filp->f_path.dentry->d_inode; 1583 struct address_space *mapping = inode->i_mapping; 1584 unsigned long index, offset; 1585 enum sgp_type sgp = SGP_READ; 1586 1587 /* 1588 * Might this read be for a stacking filesystem? Then when reading 1589 * holes of a sparse file, we actually need to allocate those pages, 1590 * and even mark them dirty, so it cannot exceed the max_blocks limit. 1591 */ 1592 if (segment_eq(get_fs(), KERNEL_DS)) 1593 sgp = SGP_DIRTY; 1594 1595 index = *ppos >> PAGE_CACHE_SHIFT; 1596 offset = *ppos & ~PAGE_CACHE_MASK; 1597 1598 for (;;) { 1599 struct page *page = NULL; 1600 unsigned long end_index, nr, ret; 1601 loff_t i_size = i_size_read(inode); 1602 1603 end_index = i_size >> PAGE_CACHE_SHIFT; 1604 if (index > end_index) 1605 break; 1606 if (index == end_index) { 1607 nr = i_size & ~PAGE_CACHE_MASK; 1608 if (nr <= offset) 1609 break; 1610 } 1611 1612 desc->error = shmem_getpage(inode, index, &page, sgp, NULL); 1613 if (desc->error) { 1614 if (desc->error == -EINVAL) 1615 desc->error = 0; 1616 break; 1617 } 1618 if (page) 1619 unlock_page(page); 1620 1621 /* 1622 * We must evaluate after, since reads (unlike writes) 1623 * are called without i_mutex protection against truncate 1624 */ 1625 nr = PAGE_CACHE_SIZE; 1626 i_size = i_size_read(inode); 1627 end_index = i_size >> PAGE_CACHE_SHIFT; 1628 if (index == end_index) { 1629 nr = i_size & ~PAGE_CACHE_MASK; 1630 if (nr <= offset) { 1631 if (page) 1632 page_cache_release(page); 1633 break; 1634 } 1635 } 1636 nr -= offset; 1637 1638 if (page) { 1639 /* 1640 * If users can be writing to this page using arbitrary 1641 * virtual addresses, take care about potential aliasing 1642 * before reading the page on the kernel side. 1643 */ 1644 if (mapping_writably_mapped(mapping)) 1645 flush_dcache_page(page); 1646 /* 1647 * Mark the page accessed if we read the beginning. 1648 */ 1649 if (!offset) 1650 mark_page_accessed(page); 1651 } else { 1652 page = ZERO_PAGE(0); 1653 page_cache_get(page); 1654 } 1655 1656 /* 1657 * Ok, we have the page, and it's up-to-date, so 1658 * now we can copy it to user space... 1659 * 1660 * The actor routine returns how many bytes were actually used.. 1661 * NOTE! This may not be the same as how much of a user buffer 1662 * we filled up (we may be padding etc), so we can only update 1663 * "pos" here (the actor routine has to update the user buffer 1664 * pointers and the remaining count). 1665 */ 1666 ret = actor(desc, page, offset, nr); 1667 offset += ret; 1668 index += offset >> PAGE_CACHE_SHIFT; 1669 offset &= ~PAGE_CACHE_MASK; 1670 1671 page_cache_release(page); 1672 if (ret != nr || !desc->count) 1673 break; 1674 1675 cond_resched(); 1676 } 1677 1678 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; 1679 file_accessed(filp); 1680} 1681 1682static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos) 1683{ 1684 read_descriptor_t desc; 1685 1686 if ((ssize_t) count < 0) 1687 return -EINVAL; 1688 if (!access_ok(VERIFY_WRITE, buf, count)) 1689 return -EFAULT; 1690 if (!count) 1691 return 0; 1692 1693 desc.written = 0; 1694 desc.count = count; 1695 desc.arg.buf = buf; 1696 desc.error = 0; 1697 1698 do_shmem_file_read(filp, ppos, &desc, file_read_actor); 1699 if (desc.written) 1700 return desc.written; 1701 return desc.error; 1702} 1703 1704static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) 1705{ 1706 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); 1707 1708 buf->f_type = TMPFS_MAGIC; 1709 buf->f_bsize = PAGE_CACHE_SIZE; 1710 buf->f_namelen = NAME_MAX; 1711 spin_lock(&sbinfo->stat_lock); 1712 if (sbinfo->max_blocks) { 1713 buf->f_blocks = sbinfo->max_blocks; 1714 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks; 1715 } 1716 if (sbinfo->max_inodes) { 1717 buf->f_files = sbinfo->max_inodes; 1718 buf->f_ffree = sbinfo->free_inodes; 1719 } 1720 /* else leave those fields 0 like simple_statfs */ 1721 spin_unlock(&sbinfo->stat_lock); 1722 return 0; 1723} 1724 1725/* 1726 * File creation. Allocate an inode, and we're done.. 1727 */ 1728static int 1729shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 1730{ 1731 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev); 1732 int error = -ENOSPC; 1733 1734 if (inode) { 1735 error = security_inode_init_security(inode, dir, NULL, NULL, 1736 NULL); 1737 if (error) { 1738 if (error != -EOPNOTSUPP) { 1739 iput(inode); 1740 return error; 1741 } 1742 } 1743 error = shmem_acl_init(inode, dir); 1744 if (error) { 1745 iput(inode); 1746 return error; 1747 } 1748 if (dir->i_mode & S_ISGID) { 1749 inode->i_gid = dir->i_gid; 1750 if (S_ISDIR(mode)) 1751 inode->i_mode |= S_ISGID; 1752 } 1753 dir->i_size += BOGO_DIRENT_SIZE; 1754 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1755 d_instantiate(dentry, inode); 1756 dget(dentry); /* Extra count - pin the dentry in core */ 1757 } 1758 return error; 1759} 1760 1761static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1762{ 1763 int error; 1764 1765 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) 1766 return error; 1767 inc_nlink(dir); 1768 return 0; 1769} 1770 1771static int shmem_create(struct inode *dir, struct dentry *dentry, int mode, 1772 struct nameidata *nd) 1773{ 1774 return shmem_mknod(dir, dentry, mode | S_IFREG, 0); 1775} 1776 1777/* 1778 * Link a file.. 1779 */ 1780static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1781{ 1782 struct inode *inode = old_dentry->d_inode; 1783 int ret; 1784 1785 /* 1786 * No ordinary (disk based) filesystem counts links as inodes; 1787 * but each new link needs a new dentry, pinning lowmem, and 1788 * tmpfs dentries cannot be pruned until they are unlinked. 1789 */ 1790 ret = shmem_reserve_inode(inode->i_sb); 1791 if (ret) 1792 goto out; 1793 1794 dir->i_size += BOGO_DIRENT_SIZE; 1795 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1796 inc_nlink(inode); 1797 atomic_inc(&inode->i_count); /* New dentry reference */ 1798 dget(dentry); /* Extra pinning count for the created dentry */ 1799 d_instantiate(dentry, inode); 1800out: 1801 return ret; 1802} 1803 1804static int shmem_unlink(struct inode *dir, struct dentry *dentry) 1805{ 1806 struct inode *inode = dentry->d_inode; 1807 1808 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) 1809 shmem_free_inode(inode->i_sb); 1810 1811 dir->i_size -= BOGO_DIRENT_SIZE; 1812 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1813 drop_nlink(inode); 1814 dput(dentry); /* Undo the count from "create" - this does all the work */ 1815 return 0; 1816} 1817 1818static int shmem_rmdir(struct inode *dir, struct dentry *dentry) 1819{ 1820 if (!simple_empty(dentry)) 1821 return -ENOTEMPTY; 1822 1823 drop_nlink(dentry->d_inode); 1824 drop_nlink(dir); 1825 return shmem_unlink(dir, dentry); 1826} 1827 1828/* 1829 * The VFS layer already does all the dentry stuff for rename, 1830 * we just have to decrement the usage count for the target if 1831 * it exists so that the VFS layer correctly free's it when it 1832 * gets overwritten. 1833 */ 1834static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) 1835{ 1836 struct inode *inode = old_dentry->d_inode; 1837 int they_are_dirs = S_ISDIR(inode->i_mode); 1838 1839 if (!simple_empty(new_dentry)) 1840 return -ENOTEMPTY; 1841 1842 if (new_dentry->d_inode) { 1843 (void) shmem_unlink(new_dir, new_dentry); 1844 if (they_are_dirs) 1845 drop_nlink(old_dir); 1846 } else if (they_are_dirs) { 1847 drop_nlink(old_dir); 1848 inc_nlink(new_dir); 1849 } 1850 1851 old_dir->i_size -= BOGO_DIRENT_SIZE; 1852 new_dir->i_size += BOGO_DIRENT_SIZE; 1853 old_dir->i_ctime = old_dir->i_mtime = 1854 new_dir->i_ctime = new_dir->i_mtime = 1855 inode->i_ctime = CURRENT_TIME; 1856 return 0; 1857} 1858 1859static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1860{ 1861 int error; 1862 int len; 1863 struct inode *inode; 1864 struct page *page = NULL; 1865 char *kaddr; 1866 struct shmem_inode_info *info; 1867 1868 len = strlen(symname) + 1; 1869 if (len > PAGE_CACHE_SIZE) 1870 return -ENAMETOOLONG; 1871 1872 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0); 1873 if (!inode) 1874 return -ENOSPC; 1875 1876 error = security_inode_init_security(inode, dir, NULL, NULL, 1877 NULL); 1878 if (error) { 1879 if (error != -EOPNOTSUPP) { 1880 iput(inode); 1881 return error; 1882 } 1883 error = 0; 1884 } 1885 1886 info = SHMEM_I(inode); 1887 inode->i_size = len-1; 1888 if (len <= (char *)inode - (char *)info) { 1889 /* do it inline */ 1890 memcpy(info, symname, len); 1891 inode->i_op = &shmem_symlink_inline_operations; 1892 } else { 1893 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); 1894 if (error) { 1895 iput(inode); 1896 return error; 1897 } 1898 unlock_page(page); 1899 inode->i_op = &shmem_symlink_inode_operations; 1900 kaddr = kmap_atomic(page, KM_USER0); 1901 memcpy(kaddr, symname, len); 1902 kunmap_atomic(kaddr, KM_USER0); 1903 set_page_dirty(page); 1904 page_cache_release(page); 1905 } 1906 if (dir->i_mode & S_ISGID) 1907 inode->i_gid = dir->i_gid; 1908 dir->i_size += BOGO_DIRENT_SIZE; 1909 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1910 d_instantiate(dentry, inode); 1911 dget(dentry); 1912 return 0; 1913} 1914 1915static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd) 1916{ 1917 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode)); 1918 return NULL; 1919} 1920 1921static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) 1922{ 1923 struct page *page = NULL; 1924 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); 1925 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page)); 1926 if (page) 1927 unlock_page(page); 1928 return page; 1929} 1930 1931static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 1932{ 1933 if (!IS_ERR(nd_get_link(nd))) { 1934 struct page *page = cookie; 1935 kunmap(page); 1936 mark_page_accessed(page); 1937 page_cache_release(page); 1938 } 1939} 1940 1941static const struct inode_operations shmem_symlink_inline_operations = { 1942 .readlink = generic_readlink, 1943 .follow_link = shmem_follow_link_inline, 1944}; 1945 1946static const struct inode_operations shmem_symlink_inode_operations = { 1947 .truncate = shmem_truncate, 1948 .readlink = generic_readlink, 1949 .follow_link = shmem_follow_link, 1950 .put_link = shmem_put_link, 1951}; 1952 1953#ifdef CONFIG_TMPFS_POSIX_ACL 1954/* 1955 * Superblocks without xattr inode operations will get security.* xattr 1956 * support from the VFS "for free". As soon as we have any other xattrs 1957 * like ACLs, we also need to implement the security.* handlers at 1958 * filesystem level, though. 1959 */ 1960 1961static size_t shmem_xattr_security_list(struct inode *inode, char *list, 1962 size_t list_len, const char *name, 1963 size_t name_len) 1964{ 1965 return security_inode_listsecurity(inode, list, list_len); 1966} 1967 1968static int shmem_xattr_security_get(struct inode *inode, const char *name, 1969 void *buffer, size_t size) 1970{ 1971 if (strcmp(name, "") == 0) 1972 return -EINVAL; 1973 return xattr_getsecurity(inode, name, buffer, size); 1974} 1975 1976static int shmem_xattr_security_set(struct inode *inode, const char *name, 1977 const void *value, size_t size, int flags) 1978{ 1979 if (strcmp(name, "") == 0) 1980 return -EINVAL; 1981 return security_inode_setsecurity(inode, name, value, size, flags); 1982} 1983 1984static struct xattr_handler shmem_xattr_security_handler = { 1985 .prefix = XATTR_SECURITY_PREFIX, 1986 .list = shmem_xattr_security_list, 1987 .get = shmem_xattr_security_get, 1988 .set = shmem_xattr_security_set, 1989}; 1990 1991static struct xattr_handler *shmem_xattr_handlers[] = { 1992 &shmem_xattr_acl_access_handler, 1993 &shmem_xattr_acl_default_handler, 1994 &shmem_xattr_security_handler, 1995 NULL 1996}; 1997#endif 1998 1999static struct dentry *shmem_get_parent(struct dentry *child) 2000{ 2001 return ERR_PTR(-ESTALE); 2002} 2003 2004static int shmem_match(struct inode *ino, void *vfh) 2005{ 2006 __u32 *fh = vfh; 2007 __u64 inum = fh[2]; 2008 inum = (inum << 32) | fh[1]; 2009 return ino->i_ino == inum && fh[0] == ino->i_generation; 2010} 2011 2012static struct dentry *shmem_fh_to_dentry(struct super_block *sb, 2013 struct fid *fid, int fh_len, int fh_type) 2014{ 2015 struct inode *inode; 2016 struct dentry *dentry = NULL; 2017 u64 inum = fid->raw[2]; 2018 inum = (inum << 32) | fid->raw[1]; 2019 2020 if (fh_len < 3) 2021 return NULL; 2022 2023 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), 2024 shmem_match, fid->raw); 2025 if (inode) { 2026 dentry = d_find_alias(inode); 2027 iput(inode); 2028 } 2029 2030 return dentry; 2031} 2032 2033static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len, 2034 int connectable) 2035{ 2036 struct inode *inode = dentry->d_inode; 2037 2038 if (*len < 3) 2039 return 255; 2040 2041 if (hlist_unhashed(&inode->i_hash)) { 2042 /* Unfortunately insert_inode_hash is not idempotent, 2043 * so as we hash inodes here rather than at creation 2044 * time, we need a lock to ensure we only try 2045 * to do it once 2046 */ 2047 static DEFINE_SPINLOCK(lock); 2048 spin_lock(&lock); 2049 if (hlist_unhashed(&inode->i_hash)) 2050 __insert_inode_hash(inode, 2051 inode->i_ino + inode->i_generation); 2052 spin_unlock(&lock); 2053 } 2054 2055 fh[0] = inode->i_generation; 2056 fh[1] = inode->i_ino; 2057 fh[2] = ((__u64)inode->i_ino) >> 32; 2058 2059 *len = 3; 2060 return 1; 2061} 2062 2063static const struct export_operations shmem_export_ops = { 2064 .get_parent = shmem_get_parent, 2065 .encode_fh = shmem_encode_fh, 2066 .fh_to_dentry = shmem_fh_to_dentry, 2067}; 2068 2069static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, 2070 bool remount) 2071{ 2072 char *this_char, *value, *rest; 2073 2074 while (options != NULL) { 2075 this_char = options; 2076 for (;;) { 2077 /* 2078 * NUL-terminate this option: unfortunately, 2079 * mount options form a comma-separated list, 2080 * but mpol's nodelist may also contain commas. 2081 */ 2082 options = strchr(options, ','); 2083 if (options == NULL) 2084 break; 2085 options++; 2086 if (!isdigit(*options)) { 2087 options[-1] = '\0'; 2088 break; 2089 } 2090 } 2091 if (!*this_char) 2092 continue; 2093 if ((value = strchr(this_char,'=')) != NULL) { 2094 *value++ = 0; 2095 } else { 2096 printk(KERN_ERR 2097 "tmpfs: No value for mount option '%s'\n", 2098 this_char); 2099 return 1; 2100 } 2101 2102 if (!strcmp(this_char,"size")) { 2103 unsigned long long size; 2104 size = memparse(value,&rest); 2105 if (*rest == '%') { 2106 size <<= PAGE_SHIFT; 2107 size *= totalram_pages; 2108 do_div(size, 100); 2109 rest++; 2110 } 2111 if (*rest) 2112 goto bad_val; 2113 sbinfo->max_blocks = 2114 DIV_ROUND_UP(size, PAGE_CACHE_SIZE); 2115 } else if (!strcmp(this_char,"nr_blocks")) { 2116 sbinfo->max_blocks = memparse(value, &rest); 2117 if (*rest) 2118 goto bad_val; 2119 } else if (!strcmp(this_char,"nr_inodes")) { 2120 sbinfo->max_inodes = memparse(value, &rest); 2121 if (*rest) 2122 goto bad_val; 2123 } else if (!strcmp(this_char,"mode")) { 2124 if (remount) 2125 continue; 2126 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; 2127 if (*rest) 2128 goto bad_val; 2129 } else if (!strcmp(this_char,"uid")) { 2130 if (remount) 2131 continue; 2132 sbinfo->uid = simple_strtoul(value, &rest, 0); 2133 if (*rest) 2134 goto bad_val; 2135 } else if (!strcmp(this_char,"gid")) { 2136 if (remount) 2137 continue; 2138 sbinfo->gid = simple_strtoul(value, &rest, 0); 2139 if (*rest) 2140 goto bad_val; 2141 } else if (!strcmp(this_char,"mpol")) { 2142 if (mpol_parse_str(value, &sbinfo->policy, 2143 &sbinfo->flags, &sbinfo->policy_nodes)) 2144 goto bad_val; 2145 } else { 2146 printk(KERN_ERR "tmpfs: Bad mount option %s\n", 2147 this_char); 2148 return 1; 2149 } 2150 } 2151 return 0; 2152 2153bad_val: 2154 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", 2155 value, this_char); 2156 return 1; 2157 2158} 2159 2160static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) 2161{ 2162 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 2163 struct shmem_sb_info config = *sbinfo; 2164 unsigned long blocks; 2165 unsigned long inodes; 2166 int error = -EINVAL; 2167 2168 if (shmem_parse_options(data, &config, true)) 2169 return error; 2170 2171 spin_lock(&sbinfo->stat_lock); 2172 blocks = sbinfo->max_blocks - sbinfo->free_blocks; 2173 inodes = sbinfo->max_inodes - sbinfo->free_inodes; 2174 if (config.max_blocks < blocks) 2175 goto out; 2176 if (config.max_inodes < inodes) 2177 goto out; 2178 /* 2179 * Those tests also disallow limited->unlimited while any are in 2180 * use, so i_blocks will always be zero when max_blocks is zero; 2181 * but we must separately disallow unlimited->limited, because 2182 * in that case we have no record of how much is already in use. 2183 */ 2184 if (config.max_blocks && !sbinfo->max_blocks) 2185 goto out; 2186 if (config.max_inodes && !sbinfo->max_inodes) 2187 goto out; 2188 2189 error = 0; 2190 sbinfo->max_blocks = config.max_blocks; 2191 sbinfo->free_blocks = config.max_blocks - blocks; 2192 sbinfo->max_inodes = config.max_inodes; 2193 sbinfo->free_inodes = config.max_inodes - inodes; 2194 sbinfo->policy = config.policy; 2195 sbinfo->flags = config.flags; 2196 sbinfo->policy_nodes = config.policy_nodes; 2197out: 2198 spin_unlock(&sbinfo->stat_lock); 2199 return error; 2200} 2201 2202static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs) 2203{ 2204 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb); 2205 2206 if (sbinfo->max_blocks != shmem_default_max_blocks()) 2207 seq_printf(seq, ",size=%luk", 2208 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10)); 2209 if (sbinfo->max_inodes != shmem_default_max_inodes()) 2210 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); 2211 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) 2212 seq_printf(seq, ",mode=%03o", sbinfo->mode); 2213 if (sbinfo->uid != 0) 2214 seq_printf(seq, ",uid=%u", sbinfo->uid); 2215 if (sbinfo->gid != 0) 2216 seq_printf(seq, ",gid=%u", sbinfo->gid); 2217 shmem_show_mpol(seq, sbinfo->policy, sbinfo->flags, 2218 sbinfo->policy_nodes); 2219 return 0; 2220} 2221#endif /* CONFIG_TMPFS */ 2222 2223static void shmem_put_super(struct super_block *sb) 2224{ 2225 kfree(sb->s_fs_info); 2226 sb->s_fs_info = NULL; 2227} 2228 2229static int shmem_fill_super(struct super_block *sb, 2230 void *data, int silent) 2231{ 2232 struct inode *inode; 2233 struct dentry *root; 2234 struct shmem_sb_info *sbinfo; 2235 int err = -ENOMEM; 2236 2237 /* Round up to L1_CACHE_BYTES to resist false sharing */ 2238 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info), 2239 L1_CACHE_BYTES), GFP_KERNEL); 2240 if (!sbinfo) 2241 return -ENOMEM; 2242 2243 sbinfo->max_blocks = 0; 2244 sbinfo->max_inodes = 0; 2245 sbinfo->mode = S_IRWXUGO | S_ISVTX; 2246 sbinfo->uid = current->fsuid; 2247 sbinfo->gid = current->fsgid; 2248 sbinfo->policy = MPOL_DEFAULT; 2249 sbinfo->flags = 0; 2250 sbinfo->policy_nodes = node_states[N_HIGH_MEMORY]; 2251 sb->s_fs_info = sbinfo; 2252 2253#ifdef CONFIG_TMPFS 2254 /* 2255 * Per default we only allow half of the physical ram per 2256 * tmpfs instance, limiting inodes to one per page of lowmem; 2257 * but the internal instance is left unlimited. 2258 */ 2259 if (!(sb->s_flags & MS_NOUSER)) { 2260 sbinfo->max_blocks = shmem_default_max_blocks(); 2261 sbinfo->max_inodes = shmem_default_max_inodes(); 2262 if (shmem_parse_options(data, sbinfo, false)) { 2263 err = -EINVAL; 2264 goto failed; 2265 } 2266 } 2267 sb->s_export_op = &shmem_export_ops; 2268#else 2269 sb->s_flags |= MS_NOUSER; 2270#endif 2271 2272 spin_lock_init(&sbinfo->stat_lock); 2273 sbinfo->free_blocks = sbinfo->max_blocks; 2274 sbinfo->free_inodes = sbinfo->max_inodes; 2275 2276 sb->s_maxbytes = SHMEM_MAX_BYTES; 2277 sb->s_blocksize = PAGE_CACHE_SIZE; 2278 sb->s_blocksize_bits = PAGE_CACHE_SHIFT; 2279 sb->s_magic = TMPFS_MAGIC; 2280 sb->s_op = &shmem_ops; 2281 sb->s_time_gran = 1; 2282#ifdef CONFIG_TMPFS_POSIX_ACL 2283 sb->s_xattr = shmem_xattr_handlers; 2284 sb->s_flags |= MS_POSIXACL; 2285#endif 2286 2287 inode = shmem_get_inode(sb, S_IFDIR | sbinfo->mode, 0); 2288 if (!inode) 2289 goto failed; 2290 inode->i_uid = sbinfo->uid; 2291 inode->i_gid = sbinfo->gid; 2292 root = d_alloc_root(inode); 2293 if (!root) 2294 goto failed_iput; 2295 sb->s_root = root; 2296 return 0; 2297 2298failed_iput: 2299 iput(inode); 2300failed: 2301 shmem_put_super(sb); 2302 return err; 2303} 2304 2305static struct kmem_cache *shmem_inode_cachep; 2306 2307static struct inode *shmem_alloc_inode(struct super_block *sb) 2308{ 2309 struct shmem_inode_info *p; 2310 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); 2311 if (!p) 2312 return NULL; 2313 return &p->vfs_inode; 2314} 2315 2316static void shmem_destroy_inode(struct inode *inode) 2317{ 2318 if ((inode->i_mode & S_IFMT) == S_IFREG) { 2319 /* only struct inode is valid if it's an inline symlink */ 2320 mpol_free_shared_policy(&SHMEM_I(inode)->policy); 2321 } 2322 shmem_acl_destroy_inode(inode); 2323 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); 2324} 2325 2326static void init_once(struct kmem_cache *cachep, void *foo) 2327{ 2328 struct shmem_inode_info *p = (struct shmem_inode_info *) foo; 2329 2330 inode_init_once(&p->vfs_inode); 2331#ifdef CONFIG_TMPFS_POSIX_ACL 2332 p->i_acl = NULL; 2333 p->i_default_acl = NULL; 2334#endif 2335} 2336 2337static int init_inodecache(void) 2338{ 2339 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", 2340 sizeof(struct shmem_inode_info), 2341 0, SLAB_PANIC, init_once); 2342 return 0; 2343} 2344 2345static void destroy_inodecache(void) 2346{ 2347 kmem_cache_destroy(shmem_inode_cachep); 2348} 2349 2350static const struct address_space_operations shmem_aops = { 2351 .writepage = shmem_writepage, 2352 .set_page_dirty = __set_page_dirty_no_writeback, 2353#ifdef CONFIG_TMPFS 2354 .readpage = shmem_readpage, 2355 .write_begin = shmem_write_begin, 2356 .write_end = shmem_write_end, 2357#endif 2358 .migratepage = migrate_page, 2359}; 2360 2361static const struct file_operations shmem_file_operations = { 2362 .mmap = shmem_mmap, 2363#ifdef CONFIG_TMPFS 2364 .llseek = generic_file_llseek, 2365 .read = shmem_file_read, 2366 .write = do_sync_write, 2367 .aio_write = generic_file_aio_write, 2368 .fsync = simple_sync_file, 2369 .splice_read = generic_file_splice_read, 2370 .splice_write = generic_file_splice_write, 2371#endif 2372}; 2373 2374static const struct inode_operations shmem_inode_operations = { 2375 .truncate = shmem_truncate, 2376 .setattr = shmem_notify_change, 2377 .truncate_range = shmem_truncate_range, 2378#ifdef CONFIG_TMPFS_POSIX_ACL 2379 .setxattr = generic_setxattr, 2380 .getxattr = generic_getxattr, 2381 .listxattr = generic_listxattr, 2382 .removexattr = generic_removexattr, 2383 .permission = shmem_permission, 2384#endif 2385 2386}; 2387 2388static const struct inode_operations shmem_dir_inode_operations = { 2389#ifdef CONFIG_TMPFS 2390 .create = shmem_create, 2391 .lookup = simple_lookup, 2392 .link = shmem_link, 2393 .unlink = shmem_unlink, 2394 .symlink = shmem_symlink, 2395 .mkdir = shmem_mkdir, 2396 .rmdir = shmem_rmdir, 2397 .mknod = shmem_mknod, 2398 .rename = shmem_rename, 2399#endif 2400#ifdef CONFIG_TMPFS_POSIX_ACL 2401 .setattr = shmem_notify_change, 2402 .setxattr = generic_setxattr, 2403 .getxattr = generic_getxattr, 2404 .listxattr = generic_listxattr, 2405 .removexattr = generic_removexattr, 2406 .permission = shmem_permission, 2407#endif 2408}; 2409 2410static const struct inode_operations shmem_special_inode_operations = { 2411#ifdef CONFIG_TMPFS_POSIX_ACL 2412 .setattr = shmem_notify_change, 2413 .setxattr = generic_setxattr, 2414 .getxattr = generic_getxattr, 2415 .listxattr = generic_listxattr, 2416 .removexattr = generic_removexattr, 2417 .permission = shmem_permission, 2418#endif 2419}; 2420 2421static const struct super_operations shmem_ops = { 2422 .alloc_inode = shmem_alloc_inode, 2423 .destroy_inode = shmem_destroy_inode, 2424#ifdef CONFIG_TMPFS 2425 .statfs = shmem_statfs, 2426 .remount_fs = shmem_remount_fs, 2427 .show_options = shmem_show_options, 2428#endif 2429 .delete_inode = shmem_delete_inode, 2430 .drop_inode = generic_delete_inode, 2431 .put_super = shmem_put_super, 2432}; 2433 2434static struct vm_operations_struct shmem_vm_ops = { 2435 .fault = shmem_fault, 2436#ifdef CONFIG_NUMA 2437 .set_policy = shmem_set_policy, 2438 .get_policy = shmem_get_policy, 2439#endif 2440}; 2441 2442 2443static int shmem_get_sb(struct file_system_type *fs_type, 2444 int flags, const char *dev_name, void *data, struct vfsmount *mnt) 2445{ 2446 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt); 2447} 2448 2449static struct file_system_type tmpfs_fs_type = { 2450 .owner = THIS_MODULE, 2451 .name = "tmpfs", 2452 .get_sb = shmem_get_sb, 2453 .kill_sb = kill_litter_super, 2454}; 2455static struct vfsmount *shm_mnt; 2456 2457static int __init init_tmpfs(void) 2458{ 2459 int error; 2460 2461 error = bdi_init(&shmem_backing_dev_info); 2462 if (error) 2463 goto out4; 2464 2465 error = init_inodecache(); 2466 if (error) 2467 goto out3; 2468 2469 error = register_filesystem(&tmpfs_fs_type); 2470 if (error) { 2471 printk(KERN_ERR "Could not register tmpfs\n"); 2472 goto out2; 2473 } 2474 2475 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER, 2476 tmpfs_fs_type.name, NULL); 2477 if (IS_ERR(shm_mnt)) { 2478 error = PTR_ERR(shm_mnt); 2479 printk(KERN_ERR "Could not kern_mount tmpfs\n"); 2480 goto out1; 2481 } 2482 return 0; 2483 2484out1: 2485 unregister_filesystem(&tmpfs_fs_type); 2486out2: 2487 destroy_inodecache(); 2488out3: 2489 bdi_destroy(&shmem_backing_dev_info); 2490out4: 2491 shm_mnt = ERR_PTR(error); 2492 return error; 2493} 2494module_init(init_tmpfs) 2495 2496/** 2497 * shmem_file_setup - get an unlinked file living in tmpfs 2498 * @name: name for dentry (to be seen in /proc/<pid>/maps 2499 * @size: size to be set for the file 2500 * @flags: vm_flags 2501 */ 2502struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags) 2503{ 2504 int error; 2505 struct file *file; 2506 struct inode *inode; 2507 struct dentry *dentry, *root; 2508 struct qstr this; 2509 2510 if (IS_ERR(shm_mnt)) 2511 return (void *)shm_mnt; 2512 2513 if (size < 0 || size > SHMEM_MAX_BYTES) 2514 return ERR_PTR(-EINVAL); 2515 2516 if (shmem_acct_size(flags, size)) 2517 return ERR_PTR(-ENOMEM); 2518 2519 error = -ENOMEM; 2520 this.name = name; 2521 this.len = strlen(name); 2522 this.hash = 0; /* will go */ 2523 root = shm_mnt->mnt_root; 2524 dentry = d_alloc(root, &this); 2525 if (!dentry) 2526 goto put_memory; 2527 2528 error = -ENFILE; 2529 file = get_empty_filp(); 2530 if (!file) 2531 goto put_dentry; 2532 2533 error = -ENOSPC; 2534 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0); 2535 if (!inode) 2536 goto close_file; 2537 2538 SHMEM_I(inode)->flags = flags & VM_ACCOUNT; 2539 d_instantiate(dentry, inode); 2540 inode->i_size = size; 2541 inode->i_nlink = 0; /* It is unlinked */ 2542 init_file(file, shm_mnt, dentry, FMODE_WRITE | FMODE_READ, 2543 &shmem_file_operations); 2544 return file; 2545 2546close_file: 2547 put_filp(file); 2548put_dentry: 2549 dput(dentry); 2550put_memory: 2551 shmem_unacct_size(flags, size); 2552 return ERR_PTR(error); 2553} 2554 2555/** 2556 * shmem_zero_setup - setup a shared anonymous mapping 2557 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff 2558 */ 2559int shmem_zero_setup(struct vm_area_struct *vma) 2560{ 2561 struct file *file; 2562 loff_t size = vma->vm_end - vma->vm_start; 2563 2564 file = shmem_file_setup("dev/zero", size, vma->vm_flags); 2565 if (IS_ERR(file)) 2566 return PTR_ERR(file); 2567 2568 if (vma->vm_file) 2569 fput(vma->vm_file); 2570 vma->vm_file = file; 2571 vma->vm_ops = &shmem_vm_ops; 2572 return 0; 2573} 2574