shmem.c revision 92562927826fceb2f8e69c89e28161b8c1e0b125
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#include <linux/magic.h> 54 55#include <asm/uaccess.h> 56#include <asm/div64.h> 57#include <asm/pgtable.h> 58 59#define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long)) 60#define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE) 61#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512) 62 63#define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1)) 64#define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT) 65 66#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT) 67 68/* info->flags needs VM_flags to handle pagein/truncate races efficiently */ 69#define SHMEM_PAGEIN VM_READ 70#define SHMEM_TRUNCATE VM_WRITE 71 72/* Definition to limit shmem_truncate's steps between cond_rescheds */ 73#define LATENCY_LIMIT 64 74 75/* Pretend that each entry is of this size in directory's i_size */ 76#define BOGO_DIRENT_SIZE 20 77 78/* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */ 79enum sgp_type { 80 SGP_READ, /* don't exceed i_size, don't allocate page */ 81 SGP_CACHE, /* don't exceed i_size, may allocate page */ 82 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */ 83 SGP_WRITE, /* may exceed i_size, may allocate page */ 84}; 85 86#ifdef CONFIG_TMPFS 87static unsigned long shmem_default_max_blocks(void) 88{ 89 return totalram_pages / 2; 90} 91 92static unsigned long shmem_default_max_inodes(void) 93{ 94 return min(totalram_pages - totalhigh_pages, totalram_pages / 2); 95} 96#endif 97 98static int shmem_getpage(struct inode *inode, unsigned long idx, 99 struct page **pagep, enum sgp_type sgp, int *type); 100 101static inline struct page *shmem_dir_alloc(gfp_t gfp_mask) 102{ 103 /* 104 * The above definition of ENTRIES_PER_PAGE, and the use of 105 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE: 106 * might be reconsidered if it ever diverges from PAGE_SIZE. 107 * 108 * Mobility flags are masked out as swap vectors cannot move 109 */ 110 return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO, 111 PAGE_CACHE_SHIFT-PAGE_SHIFT); 112} 113 114static inline void shmem_dir_free(struct page *page) 115{ 116 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT); 117} 118 119static struct page **shmem_dir_map(struct page *page) 120{ 121 return (struct page **)kmap_atomic(page, KM_USER0); 122} 123 124static inline void shmem_dir_unmap(struct page **dir) 125{ 126 kunmap_atomic(dir, KM_USER0); 127} 128 129static swp_entry_t *shmem_swp_map(struct page *page) 130{ 131 return (swp_entry_t *)kmap_atomic(page, KM_USER1); 132} 133 134static inline void shmem_swp_balance_unmap(void) 135{ 136 /* 137 * When passing a pointer to an i_direct entry, to code which 138 * also handles indirect entries and so will shmem_swp_unmap, 139 * we must arrange for the preempt count to remain in balance. 140 * What kmap_atomic of a lowmem page does depends on config 141 * and architecture, so pretend to kmap_atomic some lowmem page. 142 */ 143 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1); 144} 145 146static inline void shmem_swp_unmap(swp_entry_t *entry) 147{ 148 kunmap_atomic(entry, KM_USER1); 149} 150 151static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) 152{ 153 return sb->s_fs_info; 154} 155 156/* 157 * shmem_file_setup pre-accounts the whole fixed size of a VM object, 158 * for shared memory and for shared anonymous (/dev/zero) mappings 159 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), 160 * consistent with the pre-accounting of private mappings ... 161 */ 162static inline int shmem_acct_size(unsigned long flags, loff_t size) 163{ 164 return (flags & VM_ACCOUNT)? 165 security_vm_enough_memory(VM_ACCT(size)): 0; 166} 167 168static inline void shmem_unacct_size(unsigned long flags, loff_t size) 169{ 170 if (flags & VM_ACCOUNT) 171 vm_unacct_memory(VM_ACCT(size)); 172} 173 174/* 175 * ... whereas tmpfs objects are accounted incrementally as 176 * pages are allocated, in order to allow huge sparse files. 177 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, 178 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. 179 */ 180static inline int shmem_acct_block(unsigned long flags) 181{ 182 return (flags & VM_ACCOUNT)? 183 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE)); 184} 185 186static inline void shmem_unacct_blocks(unsigned long flags, long pages) 187{ 188 if (!(flags & VM_ACCOUNT)) 189 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE)); 190} 191 192static const struct super_operations shmem_ops; 193static const struct address_space_operations shmem_aops; 194static const struct file_operations shmem_file_operations; 195static const struct inode_operations shmem_inode_operations; 196static const struct inode_operations shmem_dir_inode_operations; 197static const struct inode_operations shmem_special_inode_operations; 198static struct vm_operations_struct shmem_vm_ops; 199 200static struct backing_dev_info shmem_backing_dev_info __read_mostly = { 201 .ra_pages = 0, /* No readahead */ 202 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, 203 .unplug_io_fn = default_unplug_io_fn, 204}; 205 206static LIST_HEAD(shmem_swaplist); 207static DEFINE_MUTEX(shmem_swaplist_mutex); 208 209static void shmem_free_blocks(struct inode *inode, long pages) 210{ 211 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 212 if (sbinfo->max_blocks) { 213 spin_lock(&sbinfo->stat_lock); 214 sbinfo->free_blocks += pages; 215 inode->i_blocks -= pages*BLOCKS_PER_PAGE; 216 spin_unlock(&sbinfo->stat_lock); 217 } 218} 219 220static int shmem_reserve_inode(struct super_block *sb) 221{ 222 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 223 if (sbinfo->max_inodes) { 224 spin_lock(&sbinfo->stat_lock); 225 if (!sbinfo->free_inodes) { 226 spin_unlock(&sbinfo->stat_lock); 227 return -ENOSPC; 228 } 229 sbinfo->free_inodes--; 230 spin_unlock(&sbinfo->stat_lock); 231 } 232 return 0; 233} 234 235static void shmem_free_inode(struct super_block *sb) 236{ 237 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 238 if (sbinfo->max_inodes) { 239 spin_lock(&sbinfo->stat_lock); 240 sbinfo->free_inodes++; 241 spin_unlock(&sbinfo->stat_lock); 242 } 243} 244 245/** 246 * shmem_recalc_inode - recalculate the size of an inode 247 * @inode: inode to recalc 248 * 249 * We have to calculate the free blocks since the mm can drop 250 * undirtied hole pages behind our back. 251 * 252 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped 253 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) 254 * 255 * It has to be called with the spinlock held. 256 */ 257static void shmem_recalc_inode(struct inode *inode) 258{ 259 struct shmem_inode_info *info = SHMEM_I(inode); 260 long freed; 261 262 freed = info->alloced - info->swapped - inode->i_mapping->nrpages; 263 if (freed > 0) { 264 info->alloced -= freed; 265 shmem_unacct_blocks(info->flags, freed); 266 shmem_free_blocks(inode, freed); 267 } 268} 269 270/** 271 * shmem_swp_entry - find the swap vector position in the info structure 272 * @info: info structure for the inode 273 * @index: index of the page to find 274 * @page: optional page to add to the structure. Has to be preset to 275 * all zeros 276 * 277 * If there is no space allocated yet it will return NULL when 278 * page is NULL, else it will use the page for the needed block, 279 * setting it to NULL on return to indicate that it has been used. 280 * 281 * The swap vector is organized the following way: 282 * 283 * There are SHMEM_NR_DIRECT entries directly stored in the 284 * shmem_inode_info structure. So small files do not need an addional 285 * allocation. 286 * 287 * For pages with index > SHMEM_NR_DIRECT there is the pointer 288 * i_indirect which points to a page which holds in the first half 289 * doubly indirect blocks, in the second half triple indirect blocks: 290 * 291 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the 292 * following layout (for SHMEM_NR_DIRECT == 16): 293 * 294 * i_indirect -> dir --> 16-19 295 * | +-> 20-23 296 * | 297 * +-->dir2 --> 24-27 298 * | +-> 28-31 299 * | +-> 32-35 300 * | +-> 36-39 301 * | 302 * +-->dir3 --> 40-43 303 * +-> 44-47 304 * +-> 48-51 305 * +-> 52-55 306 */ 307static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page) 308{ 309 unsigned long offset; 310 struct page **dir; 311 struct page *subdir; 312 313 if (index < SHMEM_NR_DIRECT) { 314 shmem_swp_balance_unmap(); 315 return info->i_direct+index; 316 } 317 if (!info->i_indirect) { 318 if (page) { 319 info->i_indirect = *page; 320 *page = NULL; 321 } 322 return NULL; /* need another page */ 323 } 324 325 index -= SHMEM_NR_DIRECT; 326 offset = index % ENTRIES_PER_PAGE; 327 index /= ENTRIES_PER_PAGE; 328 dir = shmem_dir_map(info->i_indirect); 329 330 if (index >= ENTRIES_PER_PAGE/2) { 331 index -= ENTRIES_PER_PAGE/2; 332 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE; 333 index %= ENTRIES_PER_PAGE; 334 subdir = *dir; 335 if (!subdir) { 336 if (page) { 337 *dir = *page; 338 *page = NULL; 339 } 340 shmem_dir_unmap(dir); 341 return NULL; /* need another page */ 342 } 343 shmem_dir_unmap(dir); 344 dir = shmem_dir_map(subdir); 345 } 346 347 dir += index; 348 subdir = *dir; 349 if (!subdir) { 350 if (!page || !(subdir = *page)) { 351 shmem_dir_unmap(dir); 352 return NULL; /* need a page */ 353 } 354 *dir = subdir; 355 *page = NULL; 356 } 357 shmem_dir_unmap(dir); 358 return shmem_swp_map(subdir) + offset; 359} 360 361static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value) 362{ 363 long incdec = value? 1: -1; 364 365 entry->val = value; 366 info->swapped += incdec; 367 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) { 368 struct page *page = kmap_atomic_to_page(entry); 369 set_page_private(page, page_private(page) + incdec); 370 } 371} 372 373/** 374 * shmem_swp_alloc - get the position of the swap entry for the page. 375 * @info: info structure for the inode 376 * @index: index of the page to find 377 * @sgp: check and recheck i_size? skip allocation? 378 * 379 * If the entry does not exist, allocate it. 380 */ 381static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp) 382{ 383 struct inode *inode = &info->vfs_inode; 384 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 385 struct page *page = NULL; 386 swp_entry_t *entry; 387 388 if (sgp != SGP_WRITE && 389 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 390 return ERR_PTR(-EINVAL); 391 392 while (!(entry = shmem_swp_entry(info, index, &page))) { 393 if (sgp == SGP_READ) 394 return shmem_swp_map(ZERO_PAGE(0)); 395 /* 396 * Test free_blocks against 1 not 0, since we have 1 data 397 * page (and perhaps indirect index pages) yet to allocate: 398 * a waste to allocate index if we cannot allocate data. 399 */ 400 if (sbinfo->max_blocks) { 401 spin_lock(&sbinfo->stat_lock); 402 if (sbinfo->free_blocks <= 1) { 403 spin_unlock(&sbinfo->stat_lock); 404 return ERR_PTR(-ENOSPC); 405 } 406 sbinfo->free_blocks--; 407 inode->i_blocks += BLOCKS_PER_PAGE; 408 spin_unlock(&sbinfo->stat_lock); 409 } 410 411 spin_unlock(&info->lock); 412 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping)); 413 if (page) 414 set_page_private(page, 0); 415 spin_lock(&info->lock); 416 417 if (!page) { 418 shmem_free_blocks(inode, 1); 419 return ERR_PTR(-ENOMEM); 420 } 421 if (sgp != SGP_WRITE && 422 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { 423 entry = ERR_PTR(-EINVAL); 424 break; 425 } 426 if (info->next_index <= index) 427 info->next_index = index + 1; 428 } 429 if (page) { 430 /* another task gave its page, or truncated the file */ 431 shmem_free_blocks(inode, 1); 432 shmem_dir_free(page); 433 } 434 if (info->next_index <= index && !IS_ERR(entry)) 435 info->next_index = index + 1; 436 return entry; 437} 438 439/** 440 * shmem_free_swp - free some swap entries in a directory 441 * @dir: pointer to the directory 442 * @edir: pointer after last entry of the directory 443 * @punch_lock: pointer to spinlock when needed for the holepunch case 444 */ 445static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir, 446 spinlock_t *punch_lock) 447{ 448 spinlock_t *punch_unlock = NULL; 449 swp_entry_t *ptr; 450 int freed = 0; 451 452 for (ptr = dir; ptr < edir; ptr++) { 453 if (ptr->val) { 454 if (unlikely(punch_lock)) { 455 punch_unlock = punch_lock; 456 punch_lock = NULL; 457 spin_lock(punch_unlock); 458 if (!ptr->val) 459 continue; 460 } 461 free_swap_and_cache(*ptr); 462 *ptr = (swp_entry_t){0}; 463 freed++; 464 } 465 } 466 if (punch_unlock) 467 spin_unlock(punch_unlock); 468 return freed; 469} 470 471static int shmem_map_and_free_swp(struct page *subdir, int offset, 472 int limit, struct page ***dir, spinlock_t *punch_lock) 473{ 474 swp_entry_t *ptr; 475 int freed = 0; 476 477 ptr = shmem_swp_map(subdir); 478 for (; offset < limit; offset += LATENCY_LIMIT) { 479 int size = limit - offset; 480 if (size > LATENCY_LIMIT) 481 size = LATENCY_LIMIT; 482 freed += shmem_free_swp(ptr+offset, ptr+offset+size, 483 punch_lock); 484 if (need_resched()) { 485 shmem_swp_unmap(ptr); 486 if (*dir) { 487 shmem_dir_unmap(*dir); 488 *dir = NULL; 489 } 490 cond_resched(); 491 ptr = shmem_swp_map(subdir); 492 } 493 } 494 shmem_swp_unmap(ptr); 495 return freed; 496} 497 498static void shmem_free_pages(struct list_head *next) 499{ 500 struct page *page; 501 int freed = 0; 502 503 do { 504 page = container_of(next, struct page, lru); 505 next = next->next; 506 shmem_dir_free(page); 507 freed++; 508 if (freed >= LATENCY_LIMIT) { 509 cond_resched(); 510 freed = 0; 511 } 512 } while (next); 513} 514 515static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end) 516{ 517 struct shmem_inode_info *info = SHMEM_I(inode); 518 unsigned long idx; 519 unsigned long size; 520 unsigned long limit; 521 unsigned long stage; 522 unsigned long diroff; 523 struct page **dir; 524 struct page *topdir; 525 struct page *middir; 526 struct page *subdir; 527 swp_entry_t *ptr; 528 LIST_HEAD(pages_to_free); 529 long nr_pages_to_free = 0; 530 long nr_swaps_freed = 0; 531 int offset; 532 int freed; 533 int punch_hole; 534 spinlock_t *needs_lock; 535 spinlock_t *punch_lock; 536 unsigned long upper_limit; 537 538 inode->i_ctime = inode->i_mtime = CURRENT_TIME; 539 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 540 if (idx >= info->next_index) 541 return; 542 543 spin_lock(&info->lock); 544 info->flags |= SHMEM_TRUNCATE; 545 if (likely(end == (loff_t) -1)) { 546 limit = info->next_index; 547 upper_limit = SHMEM_MAX_INDEX; 548 info->next_index = idx; 549 needs_lock = NULL; 550 punch_hole = 0; 551 } else { 552 if (end + 1 >= inode->i_size) { /* we may free a little more */ 553 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >> 554 PAGE_CACHE_SHIFT; 555 upper_limit = SHMEM_MAX_INDEX; 556 } else { 557 limit = (end + 1) >> PAGE_CACHE_SHIFT; 558 upper_limit = limit; 559 } 560 needs_lock = &info->lock; 561 punch_hole = 1; 562 } 563 564 topdir = info->i_indirect; 565 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) { 566 info->i_indirect = NULL; 567 nr_pages_to_free++; 568 list_add(&topdir->lru, &pages_to_free); 569 } 570 spin_unlock(&info->lock); 571 572 if (info->swapped && idx < SHMEM_NR_DIRECT) { 573 ptr = info->i_direct; 574 size = limit; 575 if (size > SHMEM_NR_DIRECT) 576 size = SHMEM_NR_DIRECT; 577 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock); 578 } 579 580 /* 581 * If there are no indirect blocks or we are punching a hole 582 * below indirect blocks, nothing to be done. 583 */ 584 if (!topdir || limit <= SHMEM_NR_DIRECT) 585 goto done2; 586 587 /* 588 * The truncation case has already dropped info->lock, and we're safe 589 * because i_size and next_index have already been lowered, preventing 590 * access beyond. But in the punch_hole case, we still need to take 591 * the lock when updating the swap directory, because there might be 592 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or 593 * shmem_writepage. However, whenever we find we can remove a whole 594 * directory page (not at the misaligned start or end of the range), 595 * we first NULLify its pointer in the level above, and then have no 596 * need to take the lock when updating its contents: needs_lock and 597 * punch_lock (either pointing to info->lock or NULL) manage this. 598 */ 599 600 upper_limit -= SHMEM_NR_DIRECT; 601 limit -= SHMEM_NR_DIRECT; 602 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0; 603 offset = idx % ENTRIES_PER_PAGE; 604 idx -= offset; 605 606 dir = shmem_dir_map(topdir); 607 stage = ENTRIES_PER_PAGEPAGE/2; 608 if (idx < ENTRIES_PER_PAGEPAGE/2) { 609 middir = topdir; 610 diroff = idx/ENTRIES_PER_PAGE; 611 } else { 612 dir += ENTRIES_PER_PAGE/2; 613 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE; 614 while (stage <= idx) 615 stage += ENTRIES_PER_PAGEPAGE; 616 middir = *dir; 617 if (*dir) { 618 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) % 619 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE; 620 if (!diroff && !offset && upper_limit >= stage) { 621 if (needs_lock) { 622 spin_lock(needs_lock); 623 *dir = NULL; 624 spin_unlock(needs_lock); 625 needs_lock = NULL; 626 } else 627 *dir = NULL; 628 nr_pages_to_free++; 629 list_add(&middir->lru, &pages_to_free); 630 } 631 shmem_dir_unmap(dir); 632 dir = shmem_dir_map(middir); 633 } else { 634 diroff = 0; 635 offset = 0; 636 idx = stage; 637 } 638 } 639 640 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) { 641 if (unlikely(idx == stage)) { 642 shmem_dir_unmap(dir); 643 dir = shmem_dir_map(topdir) + 644 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 645 while (!*dir) { 646 dir++; 647 idx += ENTRIES_PER_PAGEPAGE; 648 if (idx >= limit) 649 goto done1; 650 } 651 stage = idx + ENTRIES_PER_PAGEPAGE; 652 middir = *dir; 653 if (punch_hole) 654 needs_lock = &info->lock; 655 if (upper_limit >= stage) { 656 if (needs_lock) { 657 spin_lock(needs_lock); 658 *dir = NULL; 659 spin_unlock(needs_lock); 660 needs_lock = NULL; 661 } else 662 *dir = NULL; 663 nr_pages_to_free++; 664 list_add(&middir->lru, &pages_to_free); 665 } 666 shmem_dir_unmap(dir); 667 cond_resched(); 668 dir = shmem_dir_map(middir); 669 diroff = 0; 670 } 671 punch_lock = needs_lock; 672 subdir = dir[diroff]; 673 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) { 674 if (needs_lock) { 675 spin_lock(needs_lock); 676 dir[diroff] = NULL; 677 spin_unlock(needs_lock); 678 punch_lock = NULL; 679 } else 680 dir[diroff] = NULL; 681 nr_pages_to_free++; 682 list_add(&subdir->lru, &pages_to_free); 683 } 684 if (subdir && page_private(subdir) /* has swap entries */) { 685 size = limit - idx; 686 if (size > ENTRIES_PER_PAGE) 687 size = ENTRIES_PER_PAGE; 688 freed = shmem_map_and_free_swp(subdir, 689 offset, size, &dir, punch_lock); 690 if (!dir) 691 dir = shmem_dir_map(middir); 692 nr_swaps_freed += freed; 693 if (offset || punch_lock) { 694 spin_lock(&info->lock); 695 set_page_private(subdir, 696 page_private(subdir) - freed); 697 spin_unlock(&info->lock); 698 } else 699 BUG_ON(page_private(subdir) != freed); 700 } 701 offset = 0; 702 } 703done1: 704 shmem_dir_unmap(dir); 705done2: 706 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) { 707 /* 708 * Call truncate_inode_pages again: racing shmem_unuse_inode 709 * may have swizzled a page in from swap since vmtruncate or 710 * generic_delete_inode did it, before we lowered next_index. 711 * Also, though shmem_getpage checks i_size before adding to 712 * cache, no recheck after: so fix the narrow window there too. 713 * 714 * Recalling truncate_inode_pages_range and unmap_mapping_range 715 * every time for punch_hole (which never got a chance to clear 716 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive, 717 * yet hardly ever necessary: try to optimize them out later. 718 */ 719 truncate_inode_pages_range(inode->i_mapping, start, end); 720 if (punch_hole) 721 unmap_mapping_range(inode->i_mapping, start, 722 end - start, 1); 723 } 724 725 spin_lock(&info->lock); 726 info->flags &= ~SHMEM_TRUNCATE; 727 info->swapped -= nr_swaps_freed; 728 if (nr_pages_to_free) 729 shmem_free_blocks(inode, nr_pages_to_free); 730 shmem_recalc_inode(inode); 731 spin_unlock(&info->lock); 732 733 /* 734 * Empty swap vector directory pages to be freed? 735 */ 736 if (!list_empty(&pages_to_free)) { 737 pages_to_free.prev->next = NULL; 738 shmem_free_pages(pages_to_free.next); 739 } 740} 741 742static void shmem_truncate(struct inode *inode) 743{ 744 shmem_truncate_range(inode, inode->i_size, (loff_t)-1); 745} 746 747static int shmem_notify_change(struct dentry *dentry, struct iattr *attr) 748{ 749 struct inode *inode = dentry->d_inode; 750 struct page *page = NULL; 751 int error; 752 753 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { 754 if (attr->ia_size < inode->i_size) { 755 /* 756 * If truncating down to a partial page, then 757 * if that page is already allocated, hold it 758 * in memory until the truncation is over, so 759 * truncate_partial_page cannnot miss it were 760 * it assigned to swap. 761 */ 762 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) { 763 (void) shmem_getpage(inode, 764 attr->ia_size>>PAGE_CACHE_SHIFT, 765 &page, SGP_READ, NULL); 766 if (page) 767 unlock_page(page); 768 } 769 /* 770 * Reset SHMEM_PAGEIN flag so that shmem_truncate can 771 * detect if any pages might have been added to cache 772 * after truncate_inode_pages. But we needn't bother 773 * if it's being fully truncated to zero-length: the 774 * nrpages check is efficient enough in that case. 775 */ 776 if (attr->ia_size) { 777 struct shmem_inode_info *info = SHMEM_I(inode); 778 spin_lock(&info->lock); 779 info->flags &= ~SHMEM_PAGEIN; 780 spin_unlock(&info->lock); 781 } 782 } 783 } 784 785 error = inode_change_ok(inode, attr); 786 if (!error) 787 error = inode_setattr(inode, attr); 788#ifdef CONFIG_TMPFS_POSIX_ACL 789 if (!error && (attr->ia_valid & ATTR_MODE)) 790 error = generic_acl_chmod(inode, &shmem_acl_ops); 791#endif 792 if (page) 793 page_cache_release(page); 794 return error; 795} 796 797static void shmem_delete_inode(struct inode *inode) 798{ 799 struct shmem_inode_info *info = SHMEM_I(inode); 800 801 if (inode->i_op->truncate == shmem_truncate) { 802 truncate_inode_pages(inode->i_mapping, 0); 803 shmem_unacct_size(info->flags, inode->i_size); 804 inode->i_size = 0; 805 shmem_truncate(inode); 806 if (!list_empty(&info->swaplist)) { 807 mutex_lock(&shmem_swaplist_mutex); 808 list_del_init(&info->swaplist); 809 mutex_unlock(&shmem_swaplist_mutex); 810 } 811 } 812 BUG_ON(inode->i_blocks); 813 shmem_free_inode(inode->i_sb); 814 clear_inode(inode); 815} 816 817static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir) 818{ 819 swp_entry_t *ptr; 820 821 for (ptr = dir; ptr < edir; ptr++) { 822 if (ptr->val == entry.val) 823 return ptr - dir; 824 } 825 return -1; 826} 827 828static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page) 829{ 830 struct inode *inode; 831 unsigned long idx; 832 unsigned long size; 833 unsigned long limit; 834 unsigned long stage; 835 struct page **dir; 836 struct page *subdir; 837 swp_entry_t *ptr; 838 int offset; 839 int error; 840 841 idx = 0; 842 ptr = info->i_direct; 843 spin_lock(&info->lock); 844 if (!info->swapped) { 845 list_del_init(&info->swaplist); 846 goto lost2; 847 } 848 limit = info->next_index; 849 size = limit; 850 if (size > SHMEM_NR_DIRECT) 851 size = SHMEM_NR_DIRECT; 852 offset = shmem_find_swp(entry, ptr, ptr+size); 853 if (offset >= 0) 854 goto found; 855 if (!info->i_indirect) 856 goto lost2; 857 858 dir = shmem_dir_map(info->i_indirect); 859 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2; 860 861 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) { 862 if (unlikely(idx == stage)) { 863 shmem_dir_unmap(dir-1); 864 if (cond_resched_lock(&info->lock)) { 865 /* check it has not been truncated */ 866 if (limit > info->next_index) { 867 limit = info->next_index; 868 if (idx >= limit) 869 goto lost2; 870 } 871 } 872 dir = shmem_dir_map(info->i_indirect) + 873 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 874 while (!*dir) { 875 dir++; 876 idx += ENTRIES_PER_PAGEPAGE; 877 if (idx >= limit) 878 goto lost1; 879 } 880 stage = idx + ENTRIES_PER_PAGEPAGE; 881 subdir = *dir; 882 shmem_dir_unmap(dir); 883 dir = shmem_dir_map(subdir); 884 } 885 subdir = *dir; 886 if (subdir && page_private(subdir)) { 887 ptr = shmem_swp_map(subdir); 888 size = limit - idx; 889 if (size > ENTRIES_PER_PAGE) 890 size = ENTRIES_PER_PAGE; 891 offset = shmem_find_swp(entry, ptr, ptr+size); 892 shmem_swp_unmap(ptr); 893 if (offset >= 0) { 894 shmem_dir_unmap(dir); 895 goto found; 896 } 897 } 898 } 899lost1: 900 shmem_dir_unmap(dir-1); 901lost2: 902 spin_unlock(&info->lock); 903 return 0; 904found: 905 idx += offset; 906 inode = igrab(&info->vfs_inode); 907 spin_unlock(&info->lock); 908 909 /* 910 * Move _head_ to start search for next from here. 911 * But be careful: shmem_delete_inode checks list_empty without taking 912 * mutex, and there's an instant in list_move_tail when info->swaplist 913 * would appear empty, if it were the only one on shmem_swaplist. We 914 * could avoid doing it if inode NULL; or use this minor optimization. 915 */ 916 if (shmem_swaplist.next != &info->swaplist) 917 list_move_tail(&shmem_swaplist, &info->swaplist); 918 mutex_unlock(&shmem_swaplist_mutex); 919 920 error = 1; 921 if (!inode) 922 goto out; 923 /* Precharge page using GFP_KERNEL while we can wait */ 924 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL); 925 if (error) 926 goto out; 927 error = radix_tree_preload(GFP_KERNEL); 928 if (error) { 929 mem_cgroup_uncharge_cache_page(page); 930 goto out; 931 } 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_locked(page, inode->i_mapping, 938 idx, GFP_NOWAIT); 939 /* does mem_cgroup_uncharge_cache_page on error */ 940 } else /* we must compensate for our precharge above */ 941 mem_cgroup_uncharge_cache_page(page); 942 943 if (error == -EEXIST) { 944 struct page *filepage = find_get_page(inode->i_mapping, idx); 945 error = 1; 946 if (filepage) { 947 /* 948 * There might be a more uptodate page coming down 949 * from a stacked writepage: forget our swappage if so. 950 */ 951 if (PageUptodate(filepage)) 952 error = 0; 953 page_cache_release(filepage); 954 } 955 } 956 if (!error) { 957 delete_from_swap_cache(page); 958 set_page_dirty(page); 959 info->flags |= SHMEM_PAGEIN; 960 shmem_swp_set(info, ptr, 0); 961 swap_free(entry); 962 error = 1; /* not an error, but entry was found */ 963 } 964 if (ptr) 965 shmem_swp_unmap(ptr); 966 spin_unlock(&info->lock); 967 radix_tree_preload_end(); 968out: 969 unlock_page(page); 970 page_cache_release(page); 971 iput(inode); /* allows for NULL */ 972 return error; 973} 974 975/* 976 * shmem_unuse() search for an eventually swapped out shmem page. 977 */ 978int shmem_unuse(swp_entry_t entry, struct page *page) 979{ 980 struct list_head *p, *next; 981 struct shmem_inode_info *info; 982 int found = 0; 983 984 mutex_lock(&shmem_swaplist_mutex); 985 list_for_each_safe(p, next, &shmem_swaplist) { 986 info = list_entry(p, struct shmem_inode_info, swaplist); 987 found = shmem_unuse_inode(info, entry, page); 988 cond_resched(); 989 if (found) 990 goto out; 991 } 992 mutex_unlock(&shmem_swaplist_mutex); 993out: return found; /* 0 or 1 or -ENOMEM */ 994} 995 996/* 997 * Move the page from the page cache to the swap cache. 998 */ 999static int shmem_writepage(struct page *page, struct writeback_control *wbc) 1000{ 1001 struct shmem_inode_info *info; 1002 swp_entry_t *entry, swap; 1003 struct address_space *mapping; 1004 unsigned long index; 1005 struct inode *inode; 1006 1007 BUG_ON(!PageLocked(page)); 1008 mapping = page->mapping; 1009 index = page->index; 1010 inode = mapping->host; 1011 info = SHMEM_I(inode); 1012 if (info->flags & VM_LOCKED) 1013 goto redirty; 1014 if (!total_swap_pages) 1015 goto redirty; 1016 1017 /* 1018 * shmem_backing_dev_info's capabilities prevent regular writeback or 1019 * sync from ever calling shmem_writepage; but a stacking filesystem 1020 * may use the ->writepage of its underlying filesystem, in which case 1021 * tmpfs should write out to swap only in response to memory pressure, 1022 * and not for pdflush or sync. However, in those cases, we do still 1023 * want to check if there's a redundant swappage to be discarded. 1024 */ 1025 if (wbc->for_reclaim) 1026 swap = get_swap_page(); 1027 else 1028 swap.val = 0; 1029 1030 spin_lock(&info->lock); 1031 if (index >= info->next_index) { 1032 BUG_ON(!(info->flags & SHMEM_TRUNCATE)); 1033 goto unlock; 1034 } 1035 entry = shmem_swp_entry(info, index, NULL); 1036 if (entry->val) { 1037 /* 1038 * The more uptodate page coming down from a stacked 1039 * writepage should replace our old swappage. 1040 */ 1041 free_swap_and_cache(*entry); 1042 shmem_swp_set(info, entry, 0); 1043 } 1044 shmem_recalc_inode(inode); 1045 1046 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { 1047 remove_from_page_cache(page); 1048 shmem_swp_set(info, entry, swap.val); 1049 shmem_swp_unmap(entry); 1050 if (list_empty(&info->swaplist)) 1051 inode = igrab(inode); 1052 else 1053 inode = NULL; 1054 spin_unlock(&info->lock); 1055 swap_duplicate(swap); 1056 BUG_ON(page_mapped(page)); 1057 page_cache_release(page); /* pagecache ref */ 1058 set_page_dirty(page); 1059 unlock_page(page); 1060 if (inode) { 1061 mutex_lock(&shmem_swaplist_mutex); 1062 /* move instead of add in case we're racing */ 1063 list_move_tail(&info->swaplist, &shmem_swaplist); 1064 mutex_unlock(&shmem_swaplist_mutex); 1065 iput(inode); 1066 } 1067 return 0; 1068 } 1069 1070 shmem_swp_unmap(entry); 1071unlock: 1072 spin_unlock(&info->lock); 1073 swap_free(swap); 1074redirty: 1075 set_page_dirty(page); 1076 if (wbc->for_reclaim) 1077 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ 1078 unlock_page(page); 1079 return 0; 1080} 1081 1082#ifdef CONFIG_NUMA 1083#ifdef CONFIG_TMPFS 1084static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) 1085{ 1086 char buffer[64]; 1087 1088 if (!mpol || mpol->mode == MPOL_DEFAULT) 1089 return; /* show nothing */ 1090 1091 mpol_to_str(buffer, sizeof(buffer), mpol, 1); 1092 1093 seq_printf(seq, ",mpol=%s", buffer); 1094} 1095 1096static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) 1097{ 1098 struct mempolicy *mpol = NULL; 1099 if (sbinfo->mpol) { 1100 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ 1101 mpol = sbinfo->mpol; 1102 mpol_get(mpol); 1103 spin_unlock(&sbinfo->stat_lock); 1104 } 1105 return mpol; 1106} 1107#endif /* CONFIG_TMPFS */ 1108 1109static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, 1110 struct shmem_inode_info *info, unsigned long idx) 1111{ 1112 struct mempolicy mpol, *spol; 1113 struct vm_area_struct pvma; 1114 struct page *page; 1115 1116 spol = mpol_cond_copy(&mpol, 1117 mpol_shared_policy_lookup(&info->policy, idx)); 1118 1119 /* Create a pseudo vma that just contains the policy */ 1120 pvma.vm_start = 0; 1121 pvma.vm_pgoff = idx; 1122 pvma.vm_ops = NULL; 1123 pvma.vm_policy = spol; 1124 page = swapin_readahead(entry, gfp, &pvma, 0); 1125 return page; 1126} 1127 1128static struct page *shmem_alloc_page(gfp_t gfp, 1129 struct shmem_inode_info *info, unsigned long idx) 1130{ 1131 struct vm_area_struct pvma; 1132 1133 /* Create a pseudo vma that just contains the policy */ 1134 pvma.vm_start = 0; 1135 pvma.vm_pgoff = idx; 1136 pvma.vm_ops = NULL; 1137 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx); 1138 1139 /* 1140 * alloc_page_vma() will drop the shared policy reference 1141 */ 1142 return alloc_page_vma(gfp, &pvma, 0); 1143} 1144#else /* !CONFIG_NUMA */ 1145#ifdef CONFIG_TMPFS 1146static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p) 1147{ 1148} 1149#endif /* CONFIG_TMPFS */ 1150 1151static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, 1152 struct shmem_inode_info *info, unsigned long idx) 1153{ 1154 return swapin_readahead(entry, gfp, NULL, 0); 1155} 1156 1157static inline struct page *shmem_alloc_page(gfp_t gfp, 1158 struct shmem_inode_info *info, unsigned long idx) 1159{ 1160 return alloc_page(gfp); 1161} 1162#endif /* CONFIG_NUMA */ 1163 1164#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS) 1165static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) 1166{ 1167 return NULL; 1168} 1169#endif 1170 1171/* 1172 * shmem_getpage - either get the page from swap or allocate a new one 1173 * 1174 * If we allocate a new one we do not mark it dirty. That's up to the 1175 * vm. If we swap it in we mark it dirty since we also free the swap 1176 * entry since a page cannot live in both the swap and page cache 1177 */ 1178static int shmem_getpage(struct inode *inode, unsigned long idx, 1179 struct page **pagep, enum sgp_type sgp, int *type) 1180{ 1181 struct address_space *mapping = inode->i_mapping; 1182 struct shmem_inode_info *info = SHMEM_I(inode); 1183 struct shmem_sb_info *sbinfo; 1184 struct page *filepage = *pagep; 1185 struct page *swappage; 1186 swp_entry_t *entry; 1187 swp_entry_t swap; 1188 gfp_t gfp; 1189 int error; 1190 1191 if (idx >= SHMEM_MAX_INDEX) 1192 return -EFBIG; 1193 1194 if (type) 1195 *type = 0; 1196 1197 /* 1198 * Normally, filepage is NULL on entry, and either found 1199 * uptodate immediately, or allocated and zeroed, or read 1200 * in under swappage, which is then assigned to filepage. 1201 * But shmem_readpage (required for splice) passes in a locked 1202 * filepage, which may be found not uptodate by other callers 1203 * too, and may need to be copied from the swappage read in. 1204 */ 1205repeat: 1206 if (!filepage) 1207 filepage = find_lock_page(mapping, idx); 1208 if (filepage && PageUptodate(filepage)) 1209 goto done; 1210 error = 0; 1211 gfp = mapping_gfp_mask(mapping); 1212 if (!filepage) { 1213 /* 1214 * Try to preload while we can wait, to not make a habit of 1215 * draining atomic reserves; but don't latch on to this cpu. 1216 */ 1217 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM); 1218 if (error) 1219 goto failed; 1220 radix_tree_preload_end(); 1221 } 1222 1223 spin_lock(&info->lock); 1224 shmem_recalc_inode(inode); 1225 entry = shmem_swp_alloc(info, idx, sgp); 1226 if (IS_ERR(entry)) { 1227 spin_unlock(&info->lock); 1228 error = PTR_ERR(entry); 1229 goto failed; 1230 } 1231 swap = *entry; 1232 1233 if (swap.val) { 1234 /* Look it up and read it in.. */ 1235 swappage = lookup_swap_cache(swap); 1236 if (!swappage) { 1237 shmem_swp_unmap(entry); 1238 /* here we actually do the io */ 1239 if (type && !(*type & VM_FAULT_MAJOR)) { 1240 __count_vm_event(PGMAJFAULT); 1241 *type |= VM_FAULT_MAJOR; 1242 } 1243 spin_unlock(&info->lock); 1244 swappage = shmem_swapin(swap, gfp, info, idx); 1245 if (!swappage) { 1246 spin_lock(&info->lock); 1247 entry = shmem_swp_alloc(info, idx, sgp); 1248 if (IS_ERR(entry)) 1249 error = PTR_ERR(entry); 1250 else { 1251 if (entry->val == swap.val) 1252 error = -ENOMEM; 1253 shmem_swp_unmap(entry); 1254 } 1255 spin_unlock(&info->lock); 1256 if (error) 1257 goto failed; 1258 goto repeat; 1259 } 1260 wait_on_page_locked(swappage); 1261 page_cache_release(swappage); 1262 goto repeat; 1263 } 1264 1265 /* We have to do this with page locked to prevent races */ 1266 if (!trylock_page(swappage)) { 1267 shmem_swp_unmap(entry); 1268 spin_unlock(&info->lock); 1269 wait_on_page_locked(swappage); 1270 page_cache_release(swappage); 1271 goto repeat; 1272 } 1273 if (PageWriteback(swappage)) { 1274 shmem_swp_unmap(entry); 1275 spin_unlock(&info->lock); 1276 wait_on_page_writeback(swappage); 1277 unlock_page(swappage); 1278 page_cache_release(swappage); 1279 goto repeat; 1280 } 1281 if (!PageUptodate(swappage)) { 1282 shmem_swp_unmap(entry); 1283 spin_unlock(&info->lock); 1284 unlock_page(swappage); 1285 page_cache_release(swappage); 1286 error = -EIO; 1287 goto failed; 1288 } 1289 1290 if (filepage) { 1291 shmem_swp_set(info, entry, 0); 1292 shmem_swp_unmap(entry); 1293 delete_from_swap_cache(swappage); 1294 spin_unlock(&info->lock); 1295 copy_highpage(filepage, swappage); 1296 unlock_page(swappage); 1297 page_cache_release(swappage); 1298 flush_dcache_page(filepage); 1299 SetPageUptodate(filepage); 1300 set_page_dirty(filepage); 1301 swap_free(swap); 1302 } else if (!(error = add_to_page_cache_locked(swappage, mapping, 1303 idx, GFP_NOWAIT))) { 1304 info->flags |= SHMEM_PAGEIN; 1305 shmem_swp_set(info, entry, 0); 1306 shmem_swp_unmap(entry); 1307 delete_from_swap_cache(swappage); 1308 spin_unlock(&info->lock); 1309 filepage = swappage; 1310 set_page_dirty(filepage); 1311 swap_free(swap); 1312 } else { 1313 shmem_swp_unmap(entry); 1314 spin_unlock(&info->lock); 1315 unlock_page(swappage); 1316 page_cache_release(swappage); 1317 if (error == -ENOMEM) { 1318 /* allow reclaim from this memory cgroup */ 1319 error = mem_cgroup_shrink_usage(current->mm, 1320 gfp); 1321 if (error) 1322 goto failed; 1323 } 1324 goto repeat; 1325 } 1326 } else if (sgp == SGP_READ && !filepage) { 1327 shmem_swp_unmap(entry); 1328 filepage = find_get_page(mapping, idx); 1329 if (filepage && 1330 (!PageUptodate(filepage) || !trylock_page(filepage))) { 1331 spin_unlock(&info->lock); 1332 wait_on_page_locked(filepage); 1333 page_cache_release(filepage); 1334 filepage = NULL; 1335 goto repeat; 1336 } 1337 spin_unlock(&info->lock); 1338 } else { 1339 shmem_swp_unmap(entry); 1340 sbinfo = SHMEM_SB(inode->i_sb); 1341 if (sbinfo->max_blocks) { 1342 spin_lock(&sbinfo->stat_lock); 1343 if (sbinfo->free_blocks == 0 || 1344 shmem_acct_block(info->flags)) { 1345 spin_unlock(&sbinfo->stat_lock); 1346 spin_unlock(&info->lock); 1347 error = -ENOSPC; 1348 goto failed; 1349 } 1350 sbinfo->free_blocks--; 1351 inode->i_blocks += BLOCKS_PER_PAGE; 1352 spin_unlock(&sbinfo->stat_lock); 1353 } else if (shmem_acct_block(info->flags)) { 1354 spin_unlock(&info->lock); 1355 error = -ENOSPC; 1356 goto failed; 1357 } 1358 1359 if (!filepage) { 1360 int ret; 1361 1362 spin_unlock(&info->lock); 1363 filepage = shmem_alloc_page(gfp, info, idx); 1364 if (!filepage) { 1365 shmem_unacct_blocks(info->flags, 1); 1366 shmem_free_blocks(inode, 1); 1367 error = -ENOMEM; 1368 goto failed; 1369 } 1370 1371 /* Precharge page while we can wait, compensate after */ 1372 error = mem_cgroup_cache_charge(filepage, current->mm, 1373 gfp & ~__GFP_HIGHMEM); 1374 if (error) { 1375 page_cache_release(filepage); 1376 shmem_unacct_blocks(info->flags, 1); 1377 shmem_free_blocks(inode, 1); 1378 filepage = NULL; 1379 goto failed; 1380 } 1381 1382 spin_lock(&info->lock); 1383 entry = shmem_swp_alloc(info, idx, sgp); 1384 if (IS_ERR(entry)) 1385 error = PTR_ERR(entry); 1386 else { 1387 swap = *entry; 1388 shmem_swp_unmap(entry); 1389 } 1390 ret = error || swap.val; 1391 if (ret) 1392 mem_cgroup_uncharge_cache_page(filepage); 1393 else 1394 ret = add_to_page_cache_lru(filepage, mapping, 1395 idx, GFP_NOWAIT); 1396 /* 1397 * At add_to_page_cache_lru() failure, uncharge will 1398 * be done automatically. 1399 */ 1400 if (ret) { 1401 spin_unlock(&info->lock); 1402 page_cache_release(filepage); 1403 shmem_unacct_blocks(info->flags, 1); 1404 shmem_free_blocks(inode, 1); 1405 filepage = NULL; 1406 if (error) 1407 goto failed; 1408 goto repeat; 1409 } 1410 info->flags |= SHMEM_PAGEIN; 1411 } 1412 1413 info->alloced++; 1414 spin_unlock(&info->lock); 1415 clear_highpage(filepage); 1416 flush_dcache_page(filepage); 1417 SetPageUptodate(filepage); 1418 if (sgp == SGP_DIRTY) 1419 set_page_dirty(filepage); 1420 } 1421done: 1422 *pagep = filepage; 1423 return 0; 1424 1425failed: 1426 if (*pagep != filepage) { 1427 unlock_page(filepage); 1428 page_cache_release(filepage); 1429 } 1430 return error; 1431} 1432 1433static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1434{ 1435 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 1436 int error; 1437 int ret; 1438 1439 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 1440 return VM_FAULT_SIGBUS; 1441 1442 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret); 1443 if (error) 1444 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); 1445 1446 mark_page_accessed(vmf->page); 1447 return ret | VM_FAULT_LOCKED; 1448} 1449 1450#ifdef CONFIG_NUMA 1451static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new) 1452{ 1453 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1454 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new); 1455} 1456 1457static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, 1458 unsigned long addr) 1459{ 1460 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1461 unsigned long idx; 1462 1463 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; 1464 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx); 1465} 1466#endif 1467 1468int shmem_lock(struct file *file, int lock, struct user_struct *user) 1469{ 1470 struct inode *inode = file->f_path.dentry->d_inode; 1471 struct shmem_inode_info *info = SHMEM_I(inode); 1472 int retval = -ENOMEM; 1473 1474 spin_lock(&info->lock); 1475 if (lock && !(info->flags & VM_LOCKED)) { 1476 if (!user_shm_lock(inode->i_size, user)) 1477 goto out_nomem; 1478 info->flags |= VM_LOCKED; 1479 } 1480 if (!lock && (info->flags & VM_LOCKED) && user) { 1481 user_shm_unlock(inode->i_size, user); 1482 info->flags &= ~VM_LOCKED; 1483 } 1484 retval = 0; 1485out_nomem: 1486 spin_unlock(&info->lock); 1487 return retval; 1488} 1489 1490static int shmem_mmap(struct file *file, struct vm_area_struct *vma) 1491{ 1492 file_accessed(file); 1493 vma->vm_ops = &shmem_vm_ops; 1494 vma->vm_flags |= VM_CAN_NONLINEAR; 1495 return 0; 1496} 1497 1498static struct inode * 1499shmem_get_inode(struct super_block *sb, int mode, dev_t dev) 1500{ 1501 struct inode *inode; 1502 struct shmem_inode_info *info; 1503 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 1504 1505 if (shmem_reserve_inode(sb)) 1506 return NULL; 1507 1508 inode = new_inode(sb); 1509 if (inode) { 1510 inode->i_mode = mode; 1511 inode->i_uid = current->fsuid; 1512 inode->i_gid = current->fsgid; 1513 inode->i_blocks = 0; 1514 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; 1515 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 1516 inode->i_generation = get_seconds(); 1517 info = SHMEM_I(inode); 1518 memset(info, 0, (char *)inode - (char *)info); 1519 spin_lock_init(&info->lock); 1520 INIT_LIST_HEAD(&info->swaplist); 1521 1522 switch (mode & S_IFMT) { 1523 default: 1524 inode->i_op = &shmem_special_inode_operations; 1525 init_special_inode(inode, mode, dev); 1526 break; 1527 case S_IFREG: 1528 inode->i_mapping->a_ops = &shmem_aops; 1529 inode->i_op = &shmem_inode_operations; 1530 inode->i_fop = &shmem_file_operations; 1531 mpol_shared_policy_init(&info->policy, 1532 shmem_get_sbmpol(sbinfo)); 1533 break; 1534 case S_IFDIR: 1535 inc_nlink(inode); 1536 /* Some things misbehave if size == 0 on a directory */ 1537 inode->i_size = 2 * BOGO_DIRENT_SIZE; 1538 inode->i_op = &shmem_dir_inode_operations; 1539 inode->i_fop = &simple_dir_operations; 1540 break; 1541 case S_IFLNK: 1542 /* 1543 * Must not load anything in the rbtree, 1544 * mpol_free_shared_policy will not be called. 1545 */ 1546 mpol_shared_policy_init(&info->policy, NULL); 1547 break; 1548 } 1549 } else 1550 shmem_free_inode(sb); 1551 return inode; 1552} 1553 1554#ifdef CONFIG_TMPFS 1555static const struct inode_operations shmem_symlink_inode_operations; 1556static const struct inode_operations shmem_symlink_inline_operations; 1557 1558/* 1559 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin; 1560 * but providing them allows a tmpfs file to be used for splice, sendfile, and 1561 * below the loop driver, in the generic fashion that many filesystems support. 1562 */ 1563static int shmem_readpage(struct file *file, struct page *page) 1564{ 1565 struct inode *inode = page->mapping->host; 1566 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL); 1567 unlock_page(page); 1568 return error; 1569} 1570 1571static int 1572shmem_write_begin(struct file *file, struct address_space *mapping, 1573 loff_t pos, unsigned len, unsigned flags, 1574 struct page **pagep, void **fsdata) 1575{ 1576 struct inode *inode = mapping->host; 1577 pgoff_t index = pos >> PAGE_CACHE_SHIFT; 1578 *pagep = NULL; 1579 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL); 1580} 1581 1582static int 1583shmem_write_end(struct file *file, struct address_space *mapping, 1584 loff_t pos, unsigned len, unsigned copied, 1585 struct page *page, void *fsdata) 1586{ 1587 struct inode *inode = mapping->host; 1588 1589 if (pos + copied > inode->i_size) 1590 i_size_write(inode, pos + copied); 1591 1592 unlock_page(page); 1593 set_page_dirty(page); 1594 page_cache_release(page); 1595 1596 return copied; 1597} 1598 1599static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor) 1600{ 1601 struct inode *inode = filp->f_path.dentry->d_inode; 1602 struct address_space *mapping = inode->i_mapping; 1603 unsigned long index, offset; 1604 enum sgp_type sgp = SGP_READ; 1605 1606 /* 1607 * Might this read be for a stacking filesystem? Then when reading 1608 * holes of a sparse file, we actually need to allocate those pages, 1609 * and even mark them dirty, so it cannot exceed the max_blocks limit. 1610 */ 1611 if (segment_eq(get_fs(), KERNEL_DS)) 1612 sgp = SGP_DIRTY; 1613 1614 index = *ppos >> PAGE_CACHE_SHIFT; 1615 offset = *ppos & ~PAGE_CACHE_MASK; 1616 1617 for (;;) { 1618 struct page *page = NULL; 1619 unsigned long end_index, nr, ret; 1620 loff_t i_size = i_size_read(inode); 1621 1622 end_index = i_size >> PAGE_CACHE_SHIFT; 1623 if (index > end_index) 1624 break; 1625 if (index == end_index) { 1626 nr = i_size & ~PAGE_CACHE_MASK; 1627 if (nr <= offset) 1628 break; 1629 } 1630 1631 desc->error = shmem_getpage(inode, index, &page, sgp, NULL); 1632 if (desc->error) { 1633 if (desc->error == -EINVAL) 1634 desc->error = 0; 1635 break; 1636 } 1637 if (page) 1638 unlock_page(page); 1639 1640 /* 1641 * We must evaluate after, since reads (unlike writes) 1642 * are called without i_mutex protection against truncate 1643 */ 1644 nr = PAGE_CACHE_SIZE; 1645 i_size = i_size_read(inode); 1646 end_index = i_size >> PAGE_CACHE_SHIFT; 1647 if (index == end_index) { 1648 nr = i_size & ~PAGE_CACHE_MASK; 1649 if (nr <= offset) { 1650 if (page) 1651 page_cache_release(page); 1652 break; 1653 } 1654 } 1655 nr -= offset; 1656 1657 if (page) { 1658 /* 1659 * If users can be writing to this page using arbitrary 1660 * virtual addresses, take care about potential aliasing 1661 * before reading the page on the kernel side. 1662 */ 1663 if (mapping_writably_mapped(mapping)) 1664 flush_dcache_page(page); 1665 /* 1666 * Mark the page accessed if we read the beginning. 1667 */ 1668 if (!offset) 1669 mark_page_accessed(page); 1670 } else { 1671 page = ZERO_PAGE(0); 1672 page_cache_get(page); 1673 } 1674 1675 /* 1676 * Ok, we have the page, and it's up-to-date, so 1677 * now we can copy it to user space... 1678 * 1679 * The actor routine returns how many bytes were actually used.. 1680 * NOTE! This may not be the same as how much of a user buffer 1681 * we filled up (we may be padding etc), so we can only update 1682 * "pos" here (the actor routine has to update the user buffer 1683 * pointers and the remaining count). 1684 */ 1685 ret = actor(desc, page, offset, nr); 1686 offset += ret; 1687 index += offset >> PAGE_CACHE_SHIFT; 1688 offset &= ~PAGE_CACHE_MASK; 1689 1690 page_cache_release(page); 1691 if (ret != nr || !desc->count) 1692 break; 1693 1694 cond_resched(); 1695 } 1696 1697 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; 1698 file_accessed(filp); 1699} 1700 1701static ssize_t shmem_file_aio_read(struct kiocb *iocb, 1702 const struct iovec *iov, unsigned long nr_segs, loff_t pos) 1703{ 1704 struct file *filp = iocb->ki_filp; 1705 ssize_t retval; 1706 unsigned long seg; 1707 size_t count; 1708 loff_t *ppos = &iocb->ki_pos; 1709 1710 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); 1711 if (retval) 1712 return retval; 1713 1714 for (seg = 0; seg < nr_segs; seg++) { 1715 read_descriptor_t desc; 1716 1717 desc.written = 0; 1718 desc.arg.buf = iov[seg].iov_base; 1719 desc.count = iov[seg].iov_len; 1720 if (desc.count == 0) 1721 continue; 1722 desc.error = 0; 1723 do_shmem_file_read(filp, ppos, &desc, file_read_actor); 1724 retval += desc.written; 1725 if (desc.error) { 1726 retval = retval ?: desc.error; 1727 break; 1728 } 1729 if (desc.count > 0) 1730 break; 1731 } 1732 return retval; 1733} 1734 1735static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) 1736{ 1737 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); 1738 1739 buf->f_type = TMPFS_MAGIC; 1740 buf->f_bsize = PAGE_CACHE_SIZE; 1741 buf->f_namelen = NAME_MAX; 1742 spin_lock(&sbinfo->stat_lock); 1743 if (sbinfo->max_blocks) { 1744 buf->f_blocks = sbinfo->max_blocks; 1745 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks; 1746 } 1747 if (sbinfo->max_inodes) { 1748 buf->f_files = sbinfo->max_inodes; 1749 buf->f_ffree = sbinfo->free_inodes; 1750 } 1751 /* else leave those fields 0 like simple_statfs */ 1752 spin_unlock(&sbinfo->stat_lock); 1753 return 0; 1754} 1755 1756/* 1757 * File creation. Allocate an inode, and we're done.. 1758 */ 1759static int 1760shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 1761{ 1762 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev); 1763 int error = -ENOSPC; 1764 1765 if (inode) { 1766 error = security_inode_init_security(inode, dir, NULL, NULL, 1767 NULL); 1768 if (error) { 1769 if (error != -EOPNOTSUPP) { 1770 iput(inode); 1771 return error; 1772 } 1773 } 1774 error = shmem_acl_init(inode, dir); 1775 if (error) { 1776 iput(inode); 1777 return error; 1778 } 1779 if (dir->i_mode & S_ISGID) { 1780 inode->i_gid = dir->i_gid; 1781 if (S_ISDIR(mode)) 1782 inode->i_mode |= S_ISGID; 1783 } 1784 dir->i_size += BOGO_DIRENT_SIZE; 1785 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1786 d_instantiate(dentry, inode); 1787 dget(dentry); /* Extra count - pin the dentry in core */ 1788 } 1789 return error; 1790} 1791 1792static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1793{ 1794 int error; 1795 1796 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) 1797 return error; 1798 inc_nlink(dir); 1799 return 0; 1800} 1801 1802static int shmem_create(struct inode *dir, struct dentry *dentry, int mode, 1803 struct nameidata *nd) 1804{ 1805 return shmem_mknod(dir, dentry, mode | S_IFREG, 0); 1806} 1807 1808/* 1809 * Link a file.. 1810 */ 1811static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1812{ 1813 struct inode *inode = old_dentry->d_inode; 1814 int ret; 1815 1816 /* 1817 * No ordinary (disk based) filesystem counts links as inodes; 1818 * but each new link needs a new dentry, pinning lowmem, and 1819 * tmpfs dentries cannot be pruned until they are unlinked. 1820 */ 1821 ret = shmem_reserve_inode(inode->i_sb); 1822 if (ret) 1823 goto out; 1824 1825 dir->i_size += BOGO_DIRENT_SIZE; 1826 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1827 inc_nlink(inode); 1828 atomic_inc(&inode->i_count); /* New dentry reference */ 1829 dget(dentry); /* Extra pinning count for the created dentry */ 1830 d_instantiate(dentry, inode); 1831out: 1832 return ret; 1833} 1834 1835static int shmem_unlink(struct inode *dir, struct dentry *dentry) 1836{ 1837 struct inode *inode = dentry->d_inode; 1838 1839 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) 1840 shmem_free_inode(inode->i_sb); 1841 1842 dir->i_size -= BOGO_DIRENT_SIZE; 1843 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1844 drop_nlink(inode); 1845 dput(dentry); /* Undo the count from "create" - this does all the work */ 1846 return 0; 1847} 1848 1849static int shmem_rmdir(struct inode *dir, struct dentry *dentry) 1850{ 1851 if (!simple_empty(dentry)) 1852 return -ENOTEMPTY; 1853 1854 drop_nlink(dentry->d_inode); 1855 drop_nlink(dir); 1856 return shmem_unlink(dir, dentry); 1857} 1858 1859/* 1860 * The VFS layer already does all the dentry stuff for rename, 1861 * we just have to decrement the usage count for the target if 1862 * it exists so that the VFS layer correctly free's it when it 1863 * gets overwritten. 1864 */ 1865static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) 1866{ 1867 struct inode *inode = old_dentry->d_inode; 1868 int they_are_dirs = S_ISDIR(inode->i_mode); 1869 1870 if (!simple_empty(new_dentry)) 1871 return -ENOTEMPTY; 1872 1873 if (new_dentry->d_inode) { 1874 (void) shmem_unlink(new_dir, new_dentry); 1875 if (they_are_dirs) 1876 drop_nlink(old_dir); 1877 } else if (they_are_dirs) { 1878 drop_nlink(old_dir); 1879 inc_nlink(new_dir); 1880 } 1881 1882 old_dir->i_size -= BOGO_DIRENT_SIZE; 1883 new_dir->i_size += BOGO_DIRENT_SIZE; 1884 old_dir->i_ctime = old_dir->i_mtime = 1885 new_dir->i_ctime = new_dir->i_mtime = 1886 inode->i_ctime = CURRENT_TIME; 1887 return 0; 1888} 1889 1890static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1891{ 1892 int error; 1893 int len; 1894 struct inode *inode; 1895 struct page *page = NULL; 1896 char *kaddr; 1897 struct shmem_inode_info *info; 1898 1899 len = strlen(symname) + 1; 1900 if (len > PAGE_CACHE_SIZE) 1901 return -ENAMETOOLONG; 1902 1903 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0); 1904 if (!inode) 1905 return -ENOSPC; 1906 1907 error = security_inode_init_security(inode, dir, NULL, NULL, 1908 NULL); 1909 if (error) { 1910 if (error != -EOPNOTSUPP) { 1911 iput(inode); 1912 return error; 1913 } 1914 error = 0; 1915 } 1916 1917 info = SHMEM_I(inode); 1918 inode->i_size = len-1; 1919 if (len <= (char *)inode - (char *)info) { 1920 /* do it inline */ 1921 memcpy(info, symname, len); 1922 inode->i_op = &shmem_symlink_inline_operations; 1923 } else { 1924 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); 1925 if (error) { 1926 iput(inode); 1927 return error; 1928 } 1929 unlock_page(page); 1930 inode->i_mapping->a_ops = &shmem_aops; 1931 inode->i_op = &shmem_symlink_inode_operations; 1932 kaddr = kmap_atomic(page, KM_USER0); 1933 memcpy(kaddr, symname, len); 1934 kunmap_atomic(kaddr, KM_USER0); 1935 set_page_dirty(page); 1936 page_cache_release(page); 1937 } 1938 if (dir->i_mode & S_ISGID) 1939 inode->i_gid = dir->i_gid; 1940 dir->i_size += BOGO_DIRENT_SIZE; 1941 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1942 d_instantiate(dentry, inode); 1943 dget(dentry); 1944 return 0; 1945} 1946 1947static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd) 1948{ 1949 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode)); 1950 return NULL; 1951} 1952 1953static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) 1954{ 1955 struct page *page = NULL; 1956 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); 1957 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page)); 1958 if (page) 1959 unlock_page(page); 1960 return page; 1961} 1962 1963static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 1964{ 1965 if (!IS_ERR(nd_get_link(nd))) { 1966 struct page *page = cookie; 1967 kunmap(page); 1968 mark_page_accessed(page); 1969 page_cache_release(page); 1970 } 1971} 1972 1973static const struct inode_operations shmem_symlink_inline_operations = { 1974 .readlink = generic_readlink, 1975 .follow_link = shmem_follow_link_inline, 1976}; 1977 1978static const struct inode_operations shmem_symlink_inode_operations = { 1979 .truncate = shmem_truncate, 1980 .readlink = generic_readlink, 1981 .follow_link = shmem_follow_link, 1982 .put_link = shmem_put_link, 1983}; 1984 1985#ifdef CONFIG_TMPFS_POSIX_ACL 1986/* 1987 * Superblocks without xattr inode operations will get security.* xattr 1988 * support from the VFS "for free". As soon as we have any other xattrs 1989 * like ACLs, we also need to implement the security.* handlers at 1990 * filesystem level, though. 1991 */ 1992 1993static size_t shmem_xattr_security_list(struct inode *inode, char *list, 1994 size_t list_len, const char *name, 1995 size_t name_len) 1996{ 1997 return security_inode_listsecurity(inode, list, list_len); 1998} 1999 2000static int shmem_xattr_security_get(struct inode *inode, const char *name, 2001 void *buffer, size_t size) 2002{ 2003 if (strcmp(name, "") == 0) 2004 return -EINVAL; 2005 return xattr_getsecurity(inode, name, buffer, size); 2006} 2007 2008static int shmem_xattr_security_set(struct inode *inode, const char *name, 2009 const void *value, size_t size, int flags) 2010{ 2011 if (strcmp(name, "") == 0) 2012 return -EINVAL; 2013 return security_inode_setsecurity(inode, name, value, size, flags); 2014} 2015 2016static struct xattr_handler shmem_xattr_security_handler = { 2017 .prefix = XATTR_SECURITY_PREFIX, 2018 .list = shmem_xattr_security_list, 2019 .get = shmem_xattr_security_get, 2020 .set = shmem_xattr_security_set, 2021}; 2022 2023static struct xattr_handler *shmem_xattr_handlers[] = { 2024 &shmem_xattr_acl_access_handler, 2025 &shmem_xattr_acl_default_handler, 2026 &shmem_xattr_security_handler, 2027 NULL 2028}; 2029#endif 2030 2031static struct dentry *shmem_get_parent(struct dentry *child) 2032{ 2033 return ERR_PTR(-ESTALE); 2034} 2035 2036static int shmem_match(struct inode *ino, void *vfh) 2037{ 2038 __u32 *fh = vfh; 2039 __u64 inum = fh[2]; 2040 inum = (inum << 32) | fh[1]; 2041 return ino->i_ino == inum && fh[0] == ino->i_generation; 2042} 2043 2044static struct dentry *shmem_fh_to_dentry(struct super_block *sb, 2045 struct fid *fid, int fh_len, int fh_type) 2046{ 2047 struct inode *inode; 2048 struct dentry *dentry = NULL; 2049 u64 inum = fid->raw[2]; 2050 inum = (inum << 32) | fid->raw[1]; 2051 2052 if (fh_len < 3) 2053 return NULL; 2054 2055 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), 2056 shmem_match, fid->raw); 2057 if (inode) { 2058 dentry = d_find_alias(inode); 2059 iput(inode); 2060 } 2061 2062 return dentry; 2063} 2064 2065static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len, 2066 int connectable) 2067{ 2068 struct inode *inode = dentry->d_inode; 2069 2070 if (*len < 3) 2071 return 255; 2072 2073 if (hlist_unhashed(&inode->i_hash)) { 2074 /* Unfortunately insert_inode_hash is not idempotent, 2075 * so as we hash inodes here rather than at creation 2076 * time, we need a lock to ensure we only try 2077 * to do it once 2078 */ 2079 static DEFINE_SPINLOCK(lock); 2080 spin_lock(&lock); 2081 if (hlist_unhashed(&inode->i_hash)) 2082 __insert_inode_hash(inode, 2083 inode->i_ino + inode->i_generation); 2084 spin_unlock(&lock); 2085 } 2086 2087 fh[0] = inode->i_generation; 2088 fh[1] = inode->i_ino; 2089 fh[2] = ((__u64)inode->i_ino) >> 32; 2090 2091 *len = 3; 2092 return 1; 2093} 2094 2095static const struct export_operations shmem_export_ops = { 2096 .get_parent = shmem_get_parent, 2097 .encode_fh = shmem_encode_fh, 2098 .fh_to_dentry = shmem_fh_to_dentry, 2099}; 2100 2101static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, 2102 bool remount) 2103{ 2104 char *this_char, *value, *rest; 2105 2106 while (options != NULL) { 2107 this_char = options; 2108 for (;;) { 2109 /* 2110 * NUL-terminate this option: unfortunately, 2111 * mount options form a comma-separated list, 2112 * but mpol's nodelist may also contain commas. 2113 */ 2114 options = strchr(options, ','); 2115 if (options == NULL) 2116 break; 2117 options++; 2118 if (!isdigit(*options)) { 2119 options[-1] = '\0'; 2120 break; 2121 } 2122 } 2123 if (!*this_char) 2124 continue; 2125 if ((value = strchr(this_char,'=')) != NULL) { 2126 *value++ = 0; 2127 } else { 2128 printk(KERN_ERR 2129 "tmpfs: No value for mount option '%s'\n", 2130 this_char); 2131 return 1; 2132 } 2133 2134 if (!strcmp(this_char,"size")) { 2135 unsigned long long size; 2136 size = memparse(value,&rest); 2137 if (*rest == '%') { 2138 size <<= PAGE_SHIFT; 2139 size *= totalram_pages; 2140 do_div(size, 100); 2141 rest++; 2142 } 2143 if (*rest) 2144 goto bad_val; 2145 sbinfo->max_blocks = 2146 DIV_ROUND_UP(size, PAGE_CACHE_SIZE); 2147 } else if (!strcmp(this_char,"nr_blocks")) { 2148 sbinfo->max_blocks = memparse(value, &rest); 2149 if (*rest) 2150 goto bad_val; 2151 } else if (!strcmp(this_char,"nr_inodes")) { 2152 sbinfo->max_inodes = memparse(value, &rest); 2153 if (*rest) 2154 goto bad_val; 2155 } else if (!strcmp(this_char,"mode")) { 2156 if (remount) 2157 continue; 2158 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; 2159 if (*rest) 2160 goto bad_val; 2161 } else if (!strcmp(this_char,"uid")) { 2162 if (remount) 2163 continue; 2164 sbinfo->uid = simple_strtoul(value, &rest, 0); 2165 if (*rest) 2166 goto bad_val; 2167 } else if (!strcmp(this_char,"gid")) { 2168 if (remount) 2169 continue; 2170 sbinfo->gid = simple_strtoul(value, &rest, 0); 2171 if (*rest) 2172 goto bad_val; 2173 } else if (!strcmp(this_char,"mpol")) { 2174 if (mpol_parse_str(value, &sbinfo->mpol, 1)) 2175 goto bad_val; 2176 } else { 2177 printk(KERN_ERR "tmpfs: Bad mount option %s\n", 2178 this_char); 2179 return 1; 2180 } 2181 } 2182 return 0; 2183 2184bad_val: 2185 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", 2186 value, this_char); 2187 return 1; 2188 2189} 2190 2191static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) 2192{ 2193 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 2194 struct shmem_sb_info config = *sbinfo; 2195 unsigned long blocks; 2196 unsigned long inodes; 2197 int error = -EINVAL; 2198 2199 if (shmem_parse_options(data, &config, true)) 2200 return error; 2201 2202 spin_lock(&sbinfo->stat_lock); 2203 blocks = sbinfo->max_blocks - sbinfo->free_blocks; 2204 inodes = sbinfo->max_inodes - sbinfo->free_inodes; 2205 if (config.max_blocks < blocks) 2206 goto out; 2207 if (config.max_inodes < inodes) 2208 goto out; 2209 /* 2210 * Those tests also disallow limited->unlimited while any are in 2211 * use, so i_blocks will always be zero when max_blocks is zero; 2212 * but we must separately disallow unlimited->limited, because 2213 * in that case we have no record of how much is already in use. 2214 */ 2215 if (config.max_blocks && !sbinfo->max_blocks) 2216 goto out; 2217 if (config.max_inodes && !sbinfo->max_inodes) 2218 goto out; 2219 2220 error = 0; 2221 sbinfo->max_blocks = config.max_blocks; 2222 sbinfo->free_blocks = config.max_blocks - blocks; 2223 sbinfo->max_inodes = config.max_inodes; 2224 sbinfo->free_inodes = config.max_inodes - inodes; 2225 2226 mpol_put(sbinfo->mpol); 2227 sbinfo->mpol = config.mpol; /* transfers initial ref */ 2228out: 2229 spin_unlock(&sbinfo->stat_lock); 2230 return error; 2231} 2232 2233static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs) 2234{ 2235 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb); 2236 2237 if (sbinfo->max_blocks != shmem_default_max_blocks()) 2238 seq_printf(seq, ",size=%luk", 2239 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10)); 2240 if (sbinfo->max_inodes != shmem_default_max_inodes()) 2241 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); 2242 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) 2243 seq_printf(seq, ",mode=%03o", sbinfo->mode); 2244 if (sbinfo->uid != 0) 2245 seq_printf(seq, ",uid=%u", sbinfo->uid); 2246 if (sbinfo->gid != 0) 2247 seq_printf(seq, ",gid=%u", sbinfo->gid); 2248 shmem_show_mpol(seq, sbinfo->mpol); 2249 return 0; 2250} 2251#endif /* CONFIG_TMPFS */ 2252 2253static void shmem_put_super(struct super_block *sb) 2254{ 2255 kfree(sb->s_fs_info); 2256 sb->s_fs_info = NULL; 2257} 2258 2259static int shmem_fill_super(struct super_block *sb, 2260 void *data, int silent) 2261{ 2262 struct inode *inode; 2263 struct dentry *root; 2264 struct shmem_sb_info *sbinfo; 2265 int err = -ENOMEM; 2266 2267 /* Round up to L1_CACHE_BYTES to resist false sharing */ 2268 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info), 2269 L1_CACHE_BYTES), GFP_KERNEL); 2270 if (!sbinfo) 2271 return -ENOMEM; 2272 2273 sbinfo->max_blocks = 0; 2274 sbinfo->max_inodes = 0; 2275 sbinfo->mode = S_IRWXUGO | S_ISVTX; 2276 sbinfo->uid = current->fsuid; 2277 sbinfo->gid = current->fsgid; 2278 sbinfo->mpol = NULL; 2279 sb->s_fs_info = sbinfo; 2280 2281#ifdef CONFIG_TMPFS 2282 /* 2283 * Per default we only allow half of the physical ram per 2284 * tmpfs instance, limiting inodes to one per page of lowmem; 2285 * but the internal instance is left unlimited. 2286 */ 2287 if (!(sb->s_flags & MS_NOUSER)) { 2288 sbinfo->max_blocks = shmem_default_max_blocks(); 2289 sbinfo->max_inodes = shmem_default_max_inodes(); 2290 if (shmem_parse_options(data, sbinfo, false)) { 2291 err = -EINVAL; 2292 goto failed; 2293 } 2294 } 2295 sb->s_export_op = &shmem_export_ops; 2296#else 2297 sb->s_flags |= MS_NOUSER; 2298#endif 2299 2300 spin_lock_init(&sbinfo->stat_lock); 2301 sbinfo->free_blocks = sbinfo->max_blocks; 2302 sbinfo->free_inodes = sbinfo->max_inodes; 2303 2304 sb->s_maxbytes = SHMEM_MAX_BYTES; 2305 sb->s_blocksize = PAGE_CACHE_SIZE; 2306 sb->s_blocksize_bits = PAGE_CACHE_SHIFT; 2307 sb->s_magic = TMPFS_MAGIC; 2308 sb->s_op = &shmem_ops; 2309 sb->s_time_gran = 1; 2310#ifdef CONFIG_TMPFS_POSIX_ACL 2311 sb->s_xattr = shmem_xattr_handlers; 2312 sb->s_flags |= MS_POSIXACL; 2313#endif 2314 2315 inode = shmem_get_inode(sb, S_IFDIR | sbinfo->mode, 0); 2316 if (!inode) 2317 goto failed; 2318 inode->i_uid = sbinfo->uid; 2319 inode->i_gid = sbinfo->gid; 2320 root = d_alloc_root(inode); 2321 if (!root) 2322 goto failed_iput; 2323 sb->s_root = root; 2324 return 0; 2325 2326failed_iput: 2327 iput(inode); 2328failed: 2329 shmem_put_super(sb); 2330 return err; 2331} 2332 2333static struct kmem_cache *shmem_inode_cachep; 2334 2335static struct inode *shmem_alloc_inode(struct super_block *sb) 2336{ 2337 struct shmem_inode_info *p; 2338 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); 2339 if (!p) 2340 return NULL; 2341 return &p->vfs_inode; 2342} 2343 2344static void shmem_destroy_inode(struct inode *inode) 2345{ 2346 if ((inode->i_mode & S_IFMT) == S_IFREG) { 2347 /* only struct inode is valid if it's an inline symlink */ 2348 mpol_free_shared_policy(&SHMEM_I(inode)->policy); 2349 } 2350 shmem_acl_destroy_inode(inode); 2351 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); 2352} 2353 2354static void init_once(void *foo) 2355{ 2356 struct shmem_inode_info *p = (struct shmem_inode_info *) foo; 2357 2358 inode_init_once(&p->vfs_inode); 2359#ifdef CONFIG_TMPFS_POSIX_ACL 2360 p->i_acl = NULL; 2361 p->i_default_acl = NULL; 2362#endif 2363} 2364 2365static int init_inodecache(void) 2366{ 2367 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", 2368 sizeof(struct shmem_inode_info), 2369 0, SLAB_PANIC, init_once); 2370 return 0; 2371} 2372 2373static void destroy_inodecache(void) 2374{ 2375 kmem_cache_destroy(shmem_inode_cachep); 2376} 2377 2378static const struct address_space_operations shmem_aops = { 2379 .writepage = shmem_writepage, 2380 .set_page_dirty = __set_page_dirty_no_writeback, 2381#ifdef CONFIG_TMPFS 2382 .readpage = shmem_readpage, 2383 .write_begin = shmem_write_begin, 2384 .write_end = shmem_write_end, 2385#endif 2386 .migratepage = migrate_page, 2387}; 2388 2389static const struct file_operations shmem_file_operations = { 2390 .mmap = shmem_mmap, 2391#ifdef CONFIG_TMPFS 2392 .llseek = generic_file_llseek, 2393 .read = do_sync_read, 2394 .write = do_sync_write, 2395 .aio_read = shmem_file_aio_read, 2396 .aio_write = generic_file_aio_write, 2397 .fsync = simple_sync_file, 2398 .splice_read = generic_file_splice_read, 2399 .splice_write = generic_file_splice_write, 2400#endif 2401}; 2402 2403static const struct inode_operations shmem_inode_operations = { 2404 .truncate = shmem_truncate, 2405 .setattr = shmem_notify_change, 2406 .truncate_range = shmem_truncate_range, 2407#ifdef CONFIG_TMPFS_POSIX_ACL 2408 .setxattr = generic_setxattr, 2409 .getxattr = generic_getxattr, 2410 .listxattr = generic_listxattr, 2411 .removexattr = generic_removexattr, 2412 .permission = shmem_permission, 2413#endif 2414 2415}; 2416 2417static const struct inode_operations shmem_dir_inode_operations = { 2418#ifdef CONFIG_TMPFS 2419 .create = shmem_create, 2420 .lookup = simple_lookup, 2421 .link = shmem_link, 2422 .unlink = shmem_unlink, 2423 .symlink = shmem_symlink, 2424 .mkdir = shmem_mkdir, 2425 .rmdir = shmem_rmdir, 2426 .mknod = shmem_mknod, 2427 .rename = shmem_rename, 2428#endif 2429#ifdef CONFIG_TMPFS_POSIX_ACL 2430 .setattr = shmem_notify_change, 2431 .setxattr = generic_setxattr, 2432 .getxattr = generic_getxattr, 2433 .listxattr = generic_listxattr, 2434 .removexattr = generic_removexattr, 2435 .permission = shmem_permission, 2436#endif 2437}; 2438 2439static const struct inode_operations shmem_special_inode_operations = { 2440#ifdef CONFIG_TMPFS_POSIX_ACL 2441 .setattr = shmem_notify_change, 2442 .setxattr = generic_setxattr, 2443 .getxattr = generic_getxattr, 2444 .listxattr = generic_listxattr, 2445 .removexattr = generic_removexattr, 2446 .permission = shmem_permission, 2447#endif 2448}; 2449 2450static const struct super_operations shmem_ops = { 2451 .alloc_inode = shmem_alloc_inode, 2452 .destroy_inode = shmem_destroy_inode, 2453#ifdef CONFIG_TMPFS 2454 .statfs = shmem_statfs, 2455 .remount_fs = shmem_remount_fs, 2456 .show_options = shmem_show_options, 2457#endif 2458 .delete_inode = shmem_delete_inode, 2459 .drop_inode = generic_delete_inode, 2460 .put_super = shmem_put_super, 2461}; 2462 2463static struct vm_operations_struct shmem_vm_ops = { 2464 .fault = shmem_fault, 2465#ifdef CONFIG_NUMA 2466 .set_policy = shmem_set_policy, 2467 .get_policy = shmem_get_policy, 2468#endif 2469}; 2470 2471 2472static int shmem_get_sb(struct file_system_type *fs_type, 2473 int flags, const char *dev_name, void *data, struct vfsmount *mnt) 2474{ 2475 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt); 2476} 2477 2478static struct file_system_type tmpfs_fs_type = { 2479 .owner = THIS_MODULE, 2480 .name = "tmpfs", 2481 .get_sb = shmem_get_sb, 2482 .kill_sb = kill_litter_super, 2483}; 2484static struct vfsmount *shm_mnt; 2485 2486static int __init init_tmpfs(void) 2487{ 2488 int error; 2489 2490 error = bdi_init(&shmem_backing_dev_info); 2491 if (error) 2492 goto out4; 2493 2494 error = init_inodecache(); 2495 if (error) 2496 goto out3; 2497 2498 error = register_filesystem(&tmpfs_fs_type); 2499 if (error) { 2500 printk(KERN_ERR "Could not register tmpfs\n"); 2501 goto out2; 2502 } 2503 2504 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER, 2505 tmpfs_fs_type.name, NULL); 2506 if (IS_ERR(shm_mnt)) { 2507 error = PTR_ERR(shm_mnt); 2508 printk(KERN_ERR "Could not kern_mount tmpfs\n"); 2509 goto out1; 2510 } 2511 return 0; 2512 2513out1: 2514 unregister_filesystem(&tmpfs_fs_type); 2515out2: 2516 destroy_inodecache(); 2517out3: 2518 bdi_destroy(&shmem_backing_dev_info); 2519out4: 2520 shm_mnt = ERR_PTR(error); 2521 return error; 2522} 2523module_init(init_tmpfs) 2524 2525/** 2526 * shmem_file_setup - get an unlinked file living in tmpfs 2527 * @name: name for dentry (to be seen in /proc/<pid>/maps 2528 * @size: size to be set for the file 2529 * @flags: vm_flags 2530 */ 2531struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags) 2532{ 2533 int error; 2534 struct file *file; 2535 struct inode *inode; 2536 struct dentry *dentry, *root; 2537 struct qstr this; 2538 2539 if (IS_ERR(shm_mnt)) 2540 return (void *)shm_mnt; 2541 2542 if (size < 0 || size > SHMEM_MAX_BYTES) 2543 return ERR_PTR(-EINVAL); 2544 2545 if (shmem_acct_size(flags, size)) 2546 return ERR_PTR(-ENOMEM); 2547 2548 error = -ENOMEM; 2549 this.name = name; 2550 this.len = strlen(name); 2551 this.hash = 0; /* will go */ 2552 root = shm_mnt->mnt_root; 2553 dentry = d_alloc(root, &this); 2554 if (!dentry) 2555 goto put_memory; 2556 2557 error = -ENFILE; 2558 file = get_empty_filp(); 2559 if (!file) 2560 goto put_dentry; 2561 2562 error = -ENOSPC; 2563 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0); 2564 if (!inode) 2565 goto close_file; 2566 2567 SHMEM_I(inode)->flags = flags & VM_ACCOUNT; 2568 d_instantiate(dentry, inode); 2569 inode->i_size = size; 2570 inode->i_nlink = 0; /* It is unlinked */ 2571 init_file(file, shm_mnt, dentry, FMODE_WRITE | FMODE_READ, 2572 &shmem_file_operations); 2573 return file; 2574 2575close_file: 2576 put_filp(file); 2577put_dentry: 2578 dput(dentry); 2579put_memory: 2580 shmem_unacct_size(flags, size); 2581 return ERR_PTR(error); 2582} 2583 2584/** 2585 * shmem_zero_setup - setup a shared anonymous mapping 2586 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff 2587 */ 2588int shmem_zero_setup(struct vm_area_struct *vma) 2589{ 2590 struct file *file; 2591 loff_t size = vma->vm_end - vma->vm_start; 2592 2593 file = shmem_file_setup("dev/zero", size, vma->vm_flags); 2594 if (IS_ERR(file)) 2595 return PTR_ERR(file); 2596 2597 if (vma->vm_file) 2598 fput(vma->vm_file); 2599 vma->vm_file = file; 2600 vma->vm_ops = &shmem_vm_ops; 2601 return 0; 2602} 2603