shmem.c revision 818db35992c249dc32c1d86daf7d533fb0952f5d
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 53#include <asm/uaccess.h> 54#include <asm/div64.h> 55#include <asm/pgtable.h> 56 57/* This magic number is used in glibc for posix shared memory */ 58#define TMPFS_MAGIC 0x01021994 59 60#define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long)) 61#define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE) 62#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512) 63 64#define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1)) 65#define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT) 66 67#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT) 68 69/* info->flags needs VM_flags to handle pagein/truncate races efficiently */ 70#define SHMEM_PAGEIN VM_READ 71#define SHMEM_TRUNCATE VM_WRITE 72 73/* Definition to limit shmem_truncate's steps between cond_rescheds */ 74#define LATENCY_LIMIT 64 75 76/* Pretend that each entry is of this size in directory's i_size */ 77#define BOGO_DIRENT_SIZE 20 78 79/* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */ 80enum sgp_type { 81 SGP_READ, /* don't exceed i_size, don't allocate page */ 82 SGP_CACHE, /* don't exceed i_size, may allocate page */ 83 SGP_WRITE, /* may exceed i_size, may allocate page */ 84}; 85 86static int shmem_getpage(struct inode *inode, unsigned long idx, 87 struct page **pagep, enum sgp_type sgp, int *type); 88 89static inline struct page *shmem_dir_alloc(gfp_t gfp_mask) 90{ 91 /* 92 * The above definition of ENTRIES_PER_PAGE, and the use of 93 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE: 94 * might be reconsidered if it ever diverges from PAGE_SIZE. 95 * 96 * Mobility flags are masked out as swap vectors cannot move 97 */ 98 return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO, 99 PAGE_CACHE_SHIFT-PAGE_SHIFT); 100} 101 102static inline void shmem_dir_free(struct page *page) 103{ 104 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT); 105} 106 107static struct page **shmem_dir_map(struct page *page) 108{ 109 return (struct page **)kmap_atomic(page, KM_USER0); 110} 111 112static inline void shmem_dir_unmap(struct page **dir) 113{ 114 kunmap_atomic(dir, KM_USER0); 115} 116 117static swp_entry_t *shmem_swp_map(struct page *page) 118{ 119 return (swp_entry_t *)kmap_atomic(page, KM_USER1); 120} 121 122static inline void shmem_swp_balance_unmap(void) 123{ 124 /* 125 * When passing a pointer to an i_direct entry, to code which 126 * also handles indirect entries and so will shmem_swp_unmap, 127 * we must arrange for the preempt count to remain in balance. 128 * What kmap_atomic of a lowmem page does depends on config 129 * and architecture, so pretend to kmap_atomic some lowmem page. 130 */ 131 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1); 132} 133 134static inline void shmem_swp_unmap(swp_entry_t *entry) 135{ 136 kunmap_atomic(entry, KM_USER1); 137} 138 139static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) 140{ 141 return sb->s_fs_info; 142} 143 144/* 145 * shmem_file_setup pre-accounts the whole fixed size of a VM object, 146 * for shared memory and for shared anonymous (/dev/zero) mappings 147 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), 148 * consistent with the pre-accounting of private mappings ... 149 */ 150static inline int shmem_acct_size(unsigned long flags, loff_t size) 151{ 152 return (flags & VM_ACCOUNT)? 153 security_vm_enough_memory(VM_ACCT(size)): 0; 154} 155 156static inline void shmem_unacct_size(unsigned long flags, loff_t size) 157{ 158 if (flags & VM_ACCOUNT) 159 vm_unacct_memory(VM_ACCT(size)); 160} 161 162/* 163 * ... whereas tmpfs objects are accounted incrementally as 164 * pages are allocated, in order to allow huge sparse files. 165 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, 166 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. 167 */ 168static inline int shmem_acct_block(unsigned long flags) 169{ 170 return (flags & VM_ACCOUNT)? 171 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE)); 172} 173 174static inline void shmem_unacct_blocks(unsigned long flags, long pages) 175{ 176 if (!(flags & VM_ACCOUNT)) 177 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE)); 178} 179 180static const struct super_operations shmem_ops; 181static const struct address_space_operations shmem_aops; 182static const struct file_operations shmem_file_operations; 183static const struct inode_operations shmem_inode_operations; 184static const struct inode_operations shmem_dir_inode_operations; 185static const struct inode_operations shmem_special_inode_operations; 186static struct vm_operations_struct shmem_vm_ops; 187 188static struct backing_dev_info shmem_backing_dev_info __read_mostly = { 189 .ra_pages = 0, /* No readahead */ 190 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK, 191 .unplug_io_fn = default_unplug_io_fn, 192}; 193 194static LIST_HEAD(shmem_swaplist); 195static DEFINE_SPINLOCK(shmem_swaplist_lock); 196 197static void shmem_free_blocks(struct inode *inode, long pages) 198{ 199 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 200 if (sbinfo->max_blocks) { 201 spin_lock(&sbinfo->stat_lock); 202 sbinfo->free_blocks += pages; 203 inode->i_blocks -= pages*BLOCKS_PER_PAGE; 204 spin_unlock(&sbinfo->stat_lock); 205 } 206} 207 208static int shmem_reserve_inode(struct super_block *sb) 209{ 210 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 211 if (sbinfo->max_inodes) { 212 spin_lock(&sbinfo->stat_lock); 213 if (!sbinfo->free_inodes) { 214 spin_unlock(&sbinfo->stat_lock); 215 return -ENOSPC; 216 } 217 sbinfo->free_inodes--; 218 spin_unlock(&sbinfo->stat_lock); 219 } 220 return 0; 221} 222 223static void shmem_free_inode(struct super_block *sb) 224{ 225 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 226 if (sbinfo->max_inodes) { 227 spin_lock(&sbinfo->stat_lock); 228 sbinfo->free_inodes++; 229 spin_unlock(&sbinfo->stat_lock); 230 } 231} 232 233/* 234 * shmem_recalc_inode - recalculate the size of an inode 235 * 236 * @inode: inode to recalc 237 * 238 * We have to calculate the free blocks since the mm can drop 239 * undirtied hole pages behind our back. 240 * 241 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped 242 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) 243 * 244 * It has to be called with the spinlock held. 245 */ 246static void shmem_recalc_inode(struct inode *inode) 247{ 248 struct shmem_inode_info *info = SHMEM_I(inode); 249 long freed; 250 251 freed = info->alloced - info->swapped - inode->i_mapping->nrpages; 252 if (freed > 0) { 253 info->alloced -= freed; 254 shmem_unacct_blocks(info->flags, freed); 255 shmem_free_blocks(inode, freed); 256 } 257} 258 259/* 260 * shmem_swp_entry - find the swap vector position in the info structure 261 * 262 * @info: info structure for the inode 263 * @index: index of the page to find 264 * @page: optional page to add to the structure. Has to be preset to 265 * all zeros 266 * 267 * If there is no space allocated yet it will return NULL when 268 * page is NULL, else it will use the page for the needed block, 269 * setting it to NULL on return to indicate that it has been used. 270 * 271 * The swap vector is organized the following way: 272 * 273 * There are SHMEM_NR_DIRECT entries directly stored in the 274 * shmem_inode_info structure. So small files do not need an addional 275 * allocation. 276 * 277 * For pages with index > SHMEM_NR_DIRECT there is the pointer 278 * i_indirect which points to a page which holds in the first half 279 * doubly indirect blocks, in the second half triple indirect blocks: 280 * 281 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the 282 * following layout (for SHMEM_NR_DIRECT == 16): 283 * 284 * i_indirect -> dir --> 16-19 285 * | +-> 20-23 286 * | 287 * +-->dir2 --> 24-27 288 * | +-> 28-31 289 * | +-> 32-35 290 * | +-> 36-39 291 * | 292 * +-->dir3 --> 40-43 293 * +-> 44-47 294 * +-> 48-51 295 * +-> 52-55 296 */ 297static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page) 298{ 299 unsigned long offset; 300 struct page **dir; 301 struct page *subdir; 302 303 if (index < SHMEM_NR_DIRECT) { 304 shmem_swp_balance_unmap(); 305 return info->i_direct+index; 306 } 307 if (!info->i_indirect) { 308 if (page) { 309 info->i_indirect = *page; 310 *page = NULL; 311 } 312 return NULL; /* need another page */ 313 } 314 315 index -= SHMEM_NR_DIRECT; 316 offset = index % ENTRIES_PER_PAGE; 317 index /= ENTRIES_PER_PAGE; 318 dir = shmem_dir_map(info->i_indirect); 319 320 if (index >= ENTRIES_PER_PAGE/2) { 321 index -= ENTRIES_PER_PAGE/2; 322 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE; 323 index %= ENTRIES_PER_PAGE; 324 subdir = *dir; 325 if (!subdir) { 326 if (page) { 327 *dir = *page; 328 *page = NULL; 329 } 330 shmem_dir_unmap(dir); 331 return NULL; /* need another page */ 332 } 333 shmem_dir_unmap(dir); 334 dir = shmem_dir_map(subdir); 335 } 336 337 dir += index; 338 subdir = *dir; 339 if (!subdir) { 340 if (!page || !(subdir = *page)) { 341 shmem_dir_unmap(dir); 342 return NULL; /* need a page */ 343 } 344 *dir = subdir; 345 *page = NULL; 346 } 347 shmem_dir_unmap(dir); 348 return shmem_swp_map(subdir) + offset; 349} 350 351static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value) 352{ 353 long incdec = value? 1: -1; 354 355 entry->val = value; 356 info->swapped += incdec; 357 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) { 358 struct page *page = kmap_atomic_to_page(entry); 359 set_page_private(page, page_private(page) + incdec); 360 } 361} 362 363/* 364 * shmem_swp_alloc - get the position of the swap entry for the page. 365 * If it does not exist allocate the entry. 366 * 367 * @info: info structure for the inode 368 * @index: index of the page to find 369 * @sgp: check and recheck i_size? skip allocation? 370 */ 371static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp) 372{ 373 struct inode *inode = &info->vfs_inode; 374 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 375 struct page *page = NULL; 376 swp_entry_t *entry; 377 378 if (sgp != SGP_WRITE && 379 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 380 return ERR_PTR(-EINVAL); 381 382 while (!(entry = shmem_swp_entry(info, index, &page))) { 383 if (sgp == SGP_READ) 384 return shmem_swp_map(ZERO_PAGE(0)); 385 /* 386 * Test free_blocks against 1 not 0, since we have 1 data 387 * page (and perhaps indirect index pages) yet to allocate: 388 * a waste to allocate index if we cannot allocate data. 389 */ 390 if (sbinfo->max_blocks) { 391 spin_lock(&sbinfo->stat_lock); 392 if (sbinfo->free_blocks <= 1) { 393 spin_unlock(&sbinfo->stat_lock); 394 return ERR_PTR(-ENOSPC); 395 } 396 sbinfo->free_blocks--; 397 inode->i_blocks += BLOCKS_PER_PAGE; 398 spin_unlock(&sbinfo->stat_lock); 399 } 400 401 spin_unlock(&info->lock); 402 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping)); 403 if (page) 404 set_page_private(page, 0); 405 spin_lock(&info->lock); 406 407 if (!page) { 408 shmem_free_blocks(inode, 1); 409 return ERR_PTR(-ENOMEM); 410 } 411 if (sgp != SGP_WRITE && 412 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { 413 entry = ERR_PTR(-EINVAL); 414 break; 415 } 416 if (info->next_index <= index) 417 info->next_index = index + 1; 418 } 419 if (page) { 420 /* another task gave its page, or truncated the file */ 421 shmem_free_blocks(inode, 1); 422 shmem_dir_free(page); 423 } 424 if (info->next_index <= index && !IS_ERR(entry)) 425 info->next_index = index + 1; 426 return entry; 427} 428 429/* 430 * shmem_free_swp - free some swap entries in a directory 431 * 432 * @dir: pointer to the directory 433 * @edir: pointer after last entry of the directory 434 * @punch_lock: pointer to spinlock when needed for the holepunch case 435 */ 436static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir, 437 spinlock_t *punch_lock) 438{ 439 spinlock_t *punch_unlock = NULL; 440 swp_entry_t *ptr; 441 int freed = 0; 442 443 for (ptr = dir; ptr < edir; ptr++) { 444 if (ptr->val) { 445 if (unlikely(punch_lock)) { 446 punch_unlock = punch_lock; 447 punch_lock = NULL; 448 spin_lock(punch_unlock); 449 if (!ptr->val) 450 continue; 451 } 452 free_swap_and_cache(*ptr); 453 *ptr = (swp_entry_t){0}; 454 freed++; 455 } 456 } 457 if (punch_unlock) 458 spin_unlock(punch_unlock); 459 return freed; 460} 461 462static int shmem_map_and_free_swp(struct page *subdir, int offset, 463 int limit, struct page ***dir, spinlock_t *punch_lock) 464{ 465 swp_entry_t *ptr; 466 int freed = 0; 467 468 ptr = shmem_swp_map(subdir); 469 for (; offset < limit; offset += LATENCY_LIMIT) { 470 int size = limit - offset; 471 if (size > LATENCY_LIMIT) 472 size = LATENCY_LIMIT; 473 freed += shmem_free_swp(ptr+offset, ptr+offset+size, 474 punch_lock); 475 if (need_resched()) { 476 shmem_swp_unmap(ptr); 477 if (*dir) { 478 shmem_dir_unmap(*dir); 479 *dir = NULL; 480 } 481 cond_resched(); 482 ptr = shmem_swp_map(subdir); 483 } 484 } 485 shmem_swp_unmap(ptr); 486 return freed; 487} 488 489static void shmem_free_pages(struct list_head *next) 490{ 491 struct page *page; 492 int freed = 0; 493 494 do { 495 page = container_of(next, struct page, lru); 496 next = next->next; 497 shmem_dir_free(page); 498 freed++; 499 if (freed >= LATENCY_LIMIT) { 500 cond_resched(); 501 freed = 0; 502 } 503 } while (next); 504} 505 506static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end) 507{ 508 struct shmem_inode_info *info = SHMEM_I(inode); 509 unsigned long idx; 510 unsigned long size; 511 unsigned long limit; 512 unsigned long stage; 513 unsigned long diroff; 514 struct page **dir; 515 struct page *topdir; 516 struct page *middir; 517 struct page *subdir; 518 swp_entry_t *ptr; 519 LIST_HEAD(pages_to_free); 520 long nr_pages_to_free = 0; 521 long nr_swaps_freed = 0; 522 int offset; 523 int freed; 524 int punch_hole; 525 spinlock_t *needs_lock; 526 spinlock_t *punch_lock; 527 unsigned long upper_limit; 528 529 inode->i_ctime = inode->i_mtime = CURRENT_TIME; 530 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 531 if (idx >= info->next_index) 532 return; 533 534 spin_lock(&info->lock); 535 info->flags |= SHMEM_TRUNCATE; 536 if (likely(end == (loff_t) -1)) { 537 limit = info->next_index; 538 upper_limit = SHMEM_MAX_INDEX; 539 info->next_index = idx; 540 needs_lock = NULL; 541 punch_hole = 0; 542 } else { 543 if (end + 1 >= inode->i_size) { /* we may free a little more */ 544 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >> 545 PAGE_CACHE_SHIFT; 546 upper_limit = SHMEM_MAX_INDEX; 547 } else { 548 limit = (end + 1) >> PAGE_CACHE_SHIFT; 549 upper_limit = limit; 550 } 551 needs_lock = &info->lock; 552 punch_hole = 1; 553 } 554 555 topdir = info->i_indirect; 556 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) { 557 info->i_indirect = NULL; 558 nr_pages_to_free++; 559 list_add(&topdir->lru, &pages_to_free); 560 } 561 spin_unlock(&info->lock); 562 563 if (info->swapped && idx < SHMEM_NR_DIRECT) { 564 ptr = info->i_direct; 565 size = limit; 566 if (size > SHMEM_NR_DIRECT) 567 size = SHMEM_NR_DIRECT; 568 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock); 569 } 570 571 /* 572 * If there are no indirect blocks or we are punching a hole 573 * below indirect blocks, nothing to be done. 574 */ 575 if (!topdir || limit <= SHMEM_NR_DIRECT) 576 goto done2; 577 578 /* 579 * The truncation case has already dropped info->lock, and we're safe 580 * because i_size and next_index have already been lowered, preventing 581 * access beyond. But in the punch_hole case, we still need to take 582 * the lock when updating the swap directory, because there might be 583 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or 584 * shmem_writepage. However, whenever we find we can remove a whole 585 * directory page (not at the misaligned start or end of the range), 586 * we first NULLify its pointer in the level above, and then have no 587 * need to take the lock when updating its contents: needs_lock and 588 * punch_lock (either pointing to info->lock or NULL) manage this. 589 */ 590 591 upper_limit -= SHMEM_NR_DIRECT; 592 limit -= SHMEM_NR_DIRECT; 593 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0; 594 offset = idx % ENTRIES_PER_PAGE; 595 idx -= offset; 596 597 dir = shmem_dir_map(topdir); 598 stage = ENTRIES_PER_PAGEPAGE/2; 599 if (idx < ENTRIES_PER_PAGEPAGE/2) { 600 middir = topdir; 601 diroff = idx/ENTRIES_PER_PAGE; 602 } else { 603 dir += ENTRIES_PER_PAGE/2; 604 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE; 605 while (stage <= idx) 606 stage += ENTRIES_PER_PAGEPAGE; 607 middir = *dir; 608 if (*dir) { 609 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) % 610 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE; 611 if (!diroff && !offset && upper_limit >= stage) { 612 if (needs_lock) { 613 spin_lock(needs_lock); 614 *dir = NULL; 615 spin_unlock(needs_lock); 616 needs_lock = NULL; 617 } else 618 *dir = NULL; 619 nr_pages_to_free++; 620 list_add(&middir->lru, &pages_to_free); 621 } 622 shmem_dir_unmap(dir); 623 dir = shmem_dir_map(middir); 624 } else { 625 diroff = 0; 626 offset = 0; 627 idx = stage; 628 } 629 } 630 631 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) { 632 if (unlikely(idx == stage)) { 633 shmem_dir_unmap(dir); 634 dir = shmem_dir_map(topdir) + 635 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 636 while (!*dir) { 637 dir++; 638 idx += ENTRIES_PER_PAGEPAGE; 639 if (idx >= limit) 640 goto done1; 641 } 642 stage = idx + ENTRIES_PER_PAGEPAGE; 643 middir = *dir; 644 if (punch_hole) 645 needs_lock = &info->lock; 646 if (upper_limit >= stage) { 647 if (needs_lock) { 648 spin_lock(needs_lock); 649 *dir = NULL; 650 spin_unlock(needs_lock); 651 needs_lock = NULL; 652 } else 653 *dir = NULL; 654 nr_pages_to_free++; 655 list_add(&middir->lru, &pages_to_free); 656 } 657 shmem_dir_unmap(dir); 658 cond_resched(); 659 dir = shmem_dir_map(middir); 660 diroff = 0; 661 } 662 punch_lock = needs_lock; 663 subdir = dir[diroff]; 664 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) { 665 if (needs_lock) { 666 spin_lock(needs_lock); 667 dir[diroff] = NULL; 668 spin_unlock(needs_lock); 669 punch_lock = NULL; 670 } else 671 dir[diroff] = NULL; 672 nr_pages_to_free++; 673 list_add(&subdir->lru, &pages_to_free); 674 } 675 if (subdir && page_private(subdir) /* has swap entries */) { 676 size = limit - idx; 677 if (size > ENTRIES_PER_PAGE) 678 size = ENTRIES_PER_PAGE; 679 freed = shmem_map_and_free_swp(subdir, 680 offset, size, &dir, punch_lock); 681 if (!dir) 682 dir = shmem_dir_map(middir); 683 nr_swaps_freed += freed; 684 if (offset || punch_lock) { 685 spin_lock(&info->lock); 686 set_page_private(subdir, 687 page_private(subdir) - freed); 688 spin_unlock(&info->lock); 689 } else 690 BUG_ON(page_private(subdir) != freed); 691 } 692 offset = 0; 693 } 694done1: 695 shmem_dir_unmap(dir); 696done2: 697 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) { 698 /* 699 * Call truncate_inode_pages again: racing shmem_unuse_inode 700 * may have swizzled a page in from swap since vmtruncate or 701 * generic_delete_inode did it, before we lowered next_index. 702 * Also, though shmem_getpage checks i_size before adding to 703 * cache, no recheck after: so fix the narrow window there too. 704 * 705 * Recalling truncate_inode_pages_range and unmap_mapping_range 706 * every time for punch_hole (which never got a chance to clear 707 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive, 708 * yet hardly ever necessary: try to optimize them out later. 709 */ 710 truncate_inode_pages_range(inode->i_mapping, start, end); 711 if (punch_hole) 712 unmap_mapping_range(inode->i_mapping, start, 713 end - start, 1); 714 } 715 716 spin_lock(&info->lock); 717 info->flags &= ~SHMEM_TRUNCATE; 718 info->swapped -= nr_swaps_freed; 719 if (nr_pages_to_free) 720 shmem_free_blocks(inode, nr_pages_to_free); 721 shmem_recalc_inode(inode); 722 spin_unlock(&info->lock); 723 724 /* 725 * Empty swap vector directory pages to be freed? 726 */ 727 if (!list_empty(&pages_to_free)) { 728 pages_to_free.prev->next = NULL; 729 shmem_free_pages(pages_to_free.next); 730 } 731} 732 733static void shmem_truncate(struct inode *inode) 734{ 735 shmem_truncate_range(inode, inode->i_size, (loff_t)-1); 736} 737 738static int shmem_notify_change(struct dentry *dentry, struct iattr *attr) 739{ 740 struct inode *inode = dentry->d_inode; 741 struct page *page = NULL; 742 int error; 743 744 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { 745 if (attr->ia_size < inode->i_size) { 746 /* 747 * If truncating down to a partial page, then 748 * if that page is already allocated, hold it 749 * in memory until the truncation is over, so 750 * truncate_partial_page cannnot miss it were 751 * it assigned to swap. 752 */ 753 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) { 754 (void) shmem_getpage(inode, 755 attr->ia_size>>PAGE_CACHE_SHIFT, 756 &page, SGP_READ, NULL); 757 if (page) 758 unlock_page(page); 759 } 760 /* 761 * Reset SHMEM_PAGEIN flag so that shmem_truncate can 762 * detect if any pages might have been added to cache 763 * after truncate_inode_pages. But we needn't bother 764 * if it's being fully truncated to zero-length: the 765 * nrpages check is efficient enough in that case. 766 */ 767 if (attr->ia_size) { 768 struct shmem_inode_info *info = SHMEM_I(inode); 769 spin_lock(&info->lock); 770 info->flags &= ~SHMEM_PAGEIN; 771 spin_unlock(&info->lock); 772 } 773 } 774 } 775 776 error = inode_change_ok(inode, attr); 777 if (!error) 778 error = inode_setattr(inode, attr); 779#ifdef CONFIG_TMPFS_POSIX_ACL 780 if (!error && (attr->ia_valid & ATTR_MODE)) 781 error = generic_acl_chmod(inode, &shmem_acl_ops); 782#endif 783 if (page) 784 page_cache_release(page); 785 return error; 786} 787 788static void shmem_delete_inode(struct inode *inode) 789{ 790 struct shmem_inode_info *info = SHMEM_I(inode); 791 792 if (inode->i_op->truncate == shmem_truncate) { 793 truncate_inode_pages(inode->i_mapping, 0); 794 shmem_unacct_size(info->flags, inode->i_size); 795 inode->i_size = 0; 796 shmem_truncate(inode); 797 if (!list_empty(&info->swaplist)) { 798 spin_lock(&shmem_swaplist_lock); 799 list_del_init(&info->swaplist); 800 spin_unlock(&shmem_swaplist_lock); 801 } 802 } 803 BUG_ON(inode->i_blocks); 804 shmem_free_inode(inode->i_sb); 805 clear_inode(inode); 806} 807 808static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir) 809{ 810 swp_entry_t *ptr; 811 812 for (ptr = dir; ptr < edir; ptr++) { 813 if (ptr->val == entry.val) 814 return ptr - dir; 815 } 816 return -1; 817} 818 819static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page) 820{ 821 struct inode *inode; 822 unsigned long idx; 823 unsigned long size; 824 unsigned long limit; 825 unsigned long stage; 826 struct page **dir; 827 struct page *subdir; 828 swp_entry_t *ptr; 829 int offset; 830 831 idx = 0; 832 ptr = info->i_direct; 833 spin_lock(&info->lock); 834 limit = info->next_index; 835 size = limit; 836 if (size > SHMEM_NR_DIRECT) 837 size = SHMEM_NR_DIRECT; 838 offset = shmem_find_swp(entry, ptr, ptr+size); 839 if (offset >= 0) { 840 shmem_swp_balance_unmap(); 841 goto found; 842 } 843 if (!info->i_indirect) 844 goto lost2; 845 846 dir = shmem_dir_map(info->i_indirect); 847 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2; 848 849 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) { 850 if (unlikely(idx == stage)) { 851 shmem_dir_unmap(dir-1); 852 dir = shmem_dir_map(info->i_indirect) + 853 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 854 while (!*dir) { 855 dir++; 856 idx += ENTRIES_PER_PAGEPAGE; 857 if (idx >= limit) 858 goto lost1; 859 } 860 stage = idx + ENTRIES_PER_PAGEPAGE; 861 subdir = *dir; 862 shmem_dir_unmap(dir); 863 dir = shmem_dir_map(subdir); 864 } 865 subdir = *dir; 866 if (subdir && page_private(subdir)) { 867 ptr = shmem_swp_map(subdir); 868 size = limit - idx; 869 if (size > ENTRIES_PER_PAGE) 870 size = ENTRIES_PER_PAGE; 871 offset = shmem_find_swp(entry, ptr, ptr+size); 872 if (offset >= 0) { 873 shmem_dir_unmap(dir); 874 goto found; 875 } 876 shmem_swp_unmap(ptr); 877 } 878 } 879lost1: 880 shmem_dir_unmap(dir-1); 881lost2: 882 spin_unlock(&info->lock); 883 return 0; 884found: 885 idx += offset; 886 inode = &info->vfs_inode; 887 if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) { 888 info->flags |= SHMEM_PAGEIN; 889 shmem_swp_set(info, ptr + offset, 0); 890 } 891 shmem_swp_unmap(ptr); 892 spin_unlock(&info->lock); 893 /* 894 * Decrement swap count even when the entry is left behind: 895 * try_to_unuse will skip over mms, then reincrement count. 896 */ 897 swap_free(entry); 898 return 1; 899} 900 901/* 902 * shmem_unuse() search for an eventually swapped out shmem page. 903 */ 904int shmem_unuse(swp_entry_t entry, struct page *page) 905{ 906 struct list_head *p, *next; 907 struct shmem_inode_info *info; 908 int found = 0; 909 910 spin_lock(&shmem_swaplist_lock); 911 list_for_each_safe(p, next, &shmem_swaplist) { 912 info = list_entry(p, struct shmem_inode_info, swaplist); 913 if (!info->swapped) 914 list_del_init(&info->swaplist); 915 else if (shmem_unuse_inode(info, entry, page)) { 916 /* move head to start search for next from here */ 917 list_move_tail(&shmem_swaplist, &info->swaplist); 918 found = 1; 919 break; 920 } 921 } 922 spin_unlock(&shmem_swaplist_lock); 923 return found; 924} 925 926/* 927 * Move the page from the page cache to the swap cache. 928 */ 929static int shmem_writepage(struct page *page, struct writeback_control *wbc) 930{ 931 struct shmem_inode_info *info; 932 swp_entry_t *entry, swap; 933 struct address_space *mapping; 934 unsigned long index; 935 struct inode *inode; 936 937 BUG_ON(!PageLocked(page)); 938 /* 939 * shmem_backing_dev_info's capabilities prevent regular writeback or 940 * sync from ever calling shmem_writepage; but a stacking filesystem 941 * may use the ->writepage of its underlying filesystem, in which case 942 * we want to do nothing when that underlying filesystem is tmpfs 943 * (writing out to swap is useful as a response to memory pressure, but 944 * of no use to stabilize the data) - just redirty the page, unlock it 945 * and claim success in this case. AOP_WRITEPAGE_ACTIVATE, and the 946 * page_mapped check below, must be avoided unless we're in reclaim. 947 */ 948 if (!wbc->for_reclaim) { 949 set_page_dirty(page); 950 unlock_page(page); 951 return 0; 952 } 953 BUG_ON(page_mapped(page)); 954 955 mapping = page->mapping; 956 index = page->index; 957 inode = mapping->host; 958 info = SHMEM_I(inode); 959 if (info->flags & VM_LOCKED) 960 goto redirty; 961 swap = get_swap_page(); 962 if (!swap.val) 963 goto redirty; 964 965 spin_lock(&info->lock); 966 shmem_recalc_inode(inode); 967 if (index >= info->next_index) { 968 BUG_ON(!(info->flags & SHMEM_TRUNCATE)); 969 goto unlock; 970 } 971 entry = shmem_swp_entry(info, index, NULL); 972 BUG_ON(!entry); 973 BUG_ON(entry->val); 974 975 if (move_to_swap_cache(page, swap) == 0) { 976 shmem_swp_set(info, entry, swap.val); 977 shmem_swp_unmap(entry); 978 spin_unlock(&info->lock); 979 if (list_empty(&info->swaplist)) { 980 spin_lock(&shmem_swaplist_lock); 981 /* move instead of add in case we're racing */ 982 list_move_tail(&info->swaplist, &shmem_swaplist); 983 spin_unlock(&shmem_swaplist_lock); 984 } 985 unlock_page(page); 986 return 0; 987 } 988 989 shmem_swp_unmap(entry); 990unlock: 991 spin_unlock(&info->lock); 992 swap_free(swap); 993redirty: 994 set_page_dirty(page); 995 return AOP_WRITEPAGE_ACTIVATE; /* Return with the page locked */ 996} 997 998#ifdef CONFIG_NUMA 999static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes) 1000{ 1001 char *nodelist = strchr(value, ':'); 1002 int err = 1; 1003 1004 if (nodelist) { 1005 /* NUL-terminate policy string */ 1006 *nodelist++ = '\0'; 1007 if (nodelist_parse(nodelist, *policy_nodes)) 1008 goto out; 1009 if (!nodes_subset(*policy_nodes, node_states[N_HIGH_MEMORY])) 1010 goto out; 1011 } 1012 if (!strcmp(value, "default")) { 1013 *policy = MPOL_DEFAULT; 1014 /* Don't allow a nodelist */ 1015 if (!nodelist) 1016 err = 0; 1017 } else if (!strcmp(value, "prefer")) { 1018 *policy = MPOL_PREFERRED; 1019 /* Insist on a nodelist of one node only */ 1020 if (nodelist) { 1021 char *rest = nodelist; 1022 while (isdigit(*rest)) 1023 rest++; 1024 if (!*rest) 1025 err = 0; 1026 } 1027 } else if (!strcmp(value, "bind")) { 1028 *policy = MPOL_BIND; 1029 /* Insist on a nodelist */ 1030 if (nodelist) 1031 err = 0; 1032 } else if (!strcmp(value, "interleave")) { 1033 *policy = MPOL_INTERLEAVE; 1034 /* 1035 * Default to online nodes with memory if no nodelist 1036 */ 1037 if (!nodelist) 1038 *policy_nodes = node_states[N_HIGH_MEMORY]; 1039 err = 0; 1040 } 1041out: 1042 /* Restore string for error message */ 1043 if (nodelist) 1044 *--nodelist = ':'; 1045 return err; 1046} 1047 1048static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, 1049 struct shmem_inode_info *info, unsigned long idx) 1050{ 1051 struct vm_area_struct pvma; 1052 struct page *page; 1053 1054 /* Create a pseudo vma that just contains the policy */ 1055 pvma.vm_start = 0; 1056 pvma.vm_pgoff = idx; 1057 pvma.vm_ops = NULL; 1058 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx); 1059 page = swapin_readahead(entry, gfp, &pvma, 0); 1060 mpol_free(pvma.vm_policy); 1061 return page; 1062} 1063 1064static struct page *shmem_alloc_page(gfp_t gfp, 1065 struct shmem_inode_info *info, unsigned long idx) 1066{ 1067 struct vm_area_struct pvma; 1068 struct page *page; 1069 1070 /* Create a pseudo vma that just contains the policy */ 1071 pvma.vm_start = 0; 1072 pvma.vm_pgoff = idx; 1073 pvma.vm_ops = NULL; 1074 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx); 1075 page = alloc_page_vma(gfp, &pvma, 0); 1076 mpol_free(pvma.vm_policy); 1077 return page; 1078} 1079#else 1080static inline int shmem_parse_mpol(char *value, int *policy, 1081 nodemask_t *policy_nodes) 1082{ 1083 return 1; 1084} 1085 1086static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, 1087 struct shmem_inode_info *info, unsigned long idx) 1088{ 1089 return swapin_readahead(entry, gfp, NULL, 0); 1090} 1091 1092static inline struct page *shmem_alloc_page(gfp_t gfp, 1093 struct shmem_inode_info *info, unsigned long idx) 1094{ 1095 return alloc_page(gfp); 1096} 1097#endif 1098 1099/* 1100 * shmem_getpage - either get the page from swap or allocate a new one 1101 * 1102 * If we allocate a new one we do not mark it dirty. That's up to the 1103 * vm. If we swap it in we mark it dirty since we also free the swap 1104 * entry since a page cannot live in both the swap and page cache 1105 */ 1106static int shmem_getpage(struct inode *inode, unsigned long idx, 1107 struct page **pagep, enum sgp_type sgp, int *type) 1108{ 1109 struct address_space *mapping = inode->i_mapping; 1110 struct shmem_inode_info *info = SHMEM_I(inode); 1111 struct shmem_sb_info *sbinfo; 1112 struct page *filepage = *pagep; 1113 struct page *swappage; 1114 swp_entry_t *entry; 1115 swp_entry_t swap; 1116 gfp_t gfp; 1117 int error; 1118 1119 if (idx >= SHMEM_MAX_INDEX) 1120 return -EFBIG; 1121 1122 if (type) 1123 *type = 0; 1124 1125 /* 1126 * Normally, filepage is NULL on entry, and either found 1127 * uptodate immediately, or allocated and zeroed, or read 1128 * in under swappage, which is then assigned to filepage. 1129 * But shmem_readpage (required for splice) passes in a locked 1130 * filepage, which may be found not uptodate by other callers 1131 * too, and may need to be copied from the swappage read in. 1132 */ 1133repeat: 1134 if (!filepage) 1135 filepage = find_lock_page(mapping, idx); 1136 if (filepage && PageUptodate(filepage)) 1137 goto done; 1138 error = 0; 1139 gfp = mapping_gfp_mask(mapping); 1140 1141 spin_lock(&info->lock); 1142 shmem_recalc_inode(inode); 1143 entry = shmem_swp_alloc(info, idx, sgp); 1144 if (IS_ERR(entry)) { 1145 spin_unlock(&info->lock); 1146 error = PTR_ERR(entry); 1147 goto failed; 1148 } 1149 swap = *entry; 1150 1151 if (swap.val) { 1152 /* Look it up and read it in.. */ 1153 swappage = lookup_swap_cache(swap); 1154 if (!swappage) { 1155 shmem_swp_unmap(entry); 1156 /* here we actually do the io */ 1157 if (type && !(*type & VM_FAULT_MAJOR)) { 1158 __count_vm_event(PGMAJFAULT); 1159 *type |= VM_FAULT_MAJOR; 1160 } 1161 spin_unlock(&info->lock); 1162 swappage = shmem_swapin(swap, gfp, info, idx); 1163 if (!swappage) { 1164 spin_lock(&info->lock); 1165 entry = shmem_swp_alloc(info, idx, sgp); 1166 if (IS_ERR(entry)) 1167 error = PTR_ERR(entry); 1168 else { 1169 if (entry->val == swap.val) 1170 error = -ENOMEM; 1171 shmem_swp_unmap(entry); 1172 } 1173 spin_unlock(&info->lock); 1174 if (error) 1175 goto failed; 1176 goto repeat; 1177 } 1178 wait_on_page_locked(swappage); 1179 page_cache_release(swappage); 1180 goto repeat; 1181 } 1182 1183 /* We have to do this with page locked to prevent races */ 1184 if (TestSetPageLocked(swappage)) { 1185 shmem_swp_unmap(entry); 1186 spin_unlock(&info->lock); 1187 wait_on_page_locked(swappage); 1188 page_cache_release(swappage); 1189 goto repeat; 1190 } 1191 if (PageWriteback(swappage)) { 1192 shmem_swp_unmap(entry); 1193 spin_unlock(&info->lock); 1194 wait_on_page_writeback(swappage); 1195 unlock_page(swappage); 1196 page_cache_release(swappage); 1197 goto repeat; 1198 } 1199 if (!PageUptodate(swappage)) { 1200 shmem_swp_unmap(entry); 1201 spin_unlock(&info->lock); 1202 unlock_page(swappage); 1203 page_cache_release(swappage); 1204 error = -EIO; 1205 goto failed; 1206 } 1207 1208 if (filepage) { 1209 shmem_swp_set(info, entry, 0); 1210 shmem_swp_unmap(entry); 1211 delete_from_swap_cache(swappage); 1212 spin_unlock(&info->lock); 1213 copy_highpage(filepage, swappage); 1214 unlock_page(swappage); 1215 page_cache_release(swappage); 1216 flush_dcache_page(filepage); 1217 SetPageUptodate(filepage); 1218 set_page_dirty(filepage); 1219 swap_free(swap); 1220 } else if (!(error = move_from_swap_cache( 1221 swappage, idx, mapping))) { 1222 info->flags |= SHMEM_PAGEIN; 1223 shmem_swp_set(info, entry, 0); 1224 shmem_swp_unmap(entry); 1225 spin_unlock(&info->lock); 1226 filepage = swappage; 1227 swap_free(swap); 1228 } else { 1229 shmem_swp_unmap(entry); 1230 spin_unlock(&info->lock); 1231 unlock_page(swappage); 1232 page_cache_release(swappage); 1233 if (error == -ENOMEM) { 1234 /* let kswapd refresh zone for GFP_ATOMICs */ 1235 congestion_wait(WRITE, HZ/50); 1236 } 1237 goto repeat; 1238 } 1239 } else if (sgp == SGP_READ && !filepage) { 1240 shmem_swp_unmap(entry); 1241 filepage = find_get_page(mapping, idx); 1242 if (filepage && 1243 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) { 1244 spin_unlock(&info->lock); 1245 wait_on_page_locked(filepage); 1246 page_cache_release(filepage); 1247 filepage = NULL; 1248 goto repeat; 1249 } 1250 spin_unlock(&info->lock); 1251 } else { 1252 shmem_swp_unmap(entry); 1253 sbinfo = SHMEM_SB(inode->i_sb); 1254 if (sbinfo->max_blocks) { 1255 spin_lock(&sbinfo->stat_lock); 1256 if (sbinfo->free_blocks == 0 || 1257 shmem_acct_block(info->flags)) { 1258 spin_unlock(&sbinfo->stat_lock); 1259 spin_unlock(&info->lock); 1260 error = -ENOSPC; 1261 goto failed; 1262 } 1263 sbinfo->free_blocks--; 1264 inode->i_blocks += BLOCKS_PER_PAGE; 1265 spin_unlock(&sbinfo->stat_lock); 1266 } else if (shmem_acct_block(info->flags)) { 1267 spin_unlock(&info->lock); 1268 error = -ENOSPC; 1269 goto failed; 1270 } 1271 1272 if (!filepage) { 1273 spin_unlock(&info->lock); 1274 filepage = shmem_alloc_page(gfp, info, idx); 1275 if (!filepage) { 1276 shmem_unacct_blocks(info->flags, 1); 1277 shmem_free_blocks(inode, 1); 1278 error = -ENOMEM; 1279 goto failed; 1280 } 1281 1282 spin_lock(&info->lock); 1283 entry = shmem_swp_alloc(info, idx, sgp); 1284 if (IS_ERR(entry)) 1285 error = PTR_ERR(entry); 1286 else { 1287 swap = *entry; 1288 shmem_swp_unmap(entry); 1289 } 1290 if (error || swap.val || 0 != add_to_page_cache_lru( 1291 filepage, mapping, idx, GFP_ATOMIC)) { 1292 spin_unlock(&info->lock); 1293 page_cache_release(filepage); 1294 shmem_unacct_blocks(info->flags, 1); 1295 shmem_free_blocks(inode, 1); 1296 filepage = NULL; 1297 if (error) 1298 goto failed; 1299 goto repeat; 1300 } 1301 info->flags |= SHMEM_PAGEIN; 1302 } 1303 1304 info->alloced++; 1305 spin_unlock(&info->lock); 1306 clear_highpage(filepage); 1307 flush_dcache_page(filepage); 1308 SetPageUptodate(filepage); 1309 } 1310done: 1311 *pagep = filepage; 1312 return 0; 1313 1314failed: 1315 if (*pagep != filepage) { 1316 unlock_page(filepage); 1317 page_cache_release(filepage); 1318 } 1319 return error; 1320} 1321 1322static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1323{ 1324 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 1325 int error; 1326 int ret; 1327 1328 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 1329 return VM_FAULT_SIGBUS; 1330 1331 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret); 1332 if (error) 1333 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); 1334 1335 mark_page_accessed(vmf->page); 1336 return ret | VM_FAULT_LOCKED; 1337} 1338 1339#ifdef CONFIG_NUMA 1340static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new) 1341{ 1342 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1343 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new); 1344} 1345 1346static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, 1347 unsigned long addr) 1348{ 1349 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1350 unsigned long idx; 1351 1352 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; 1353 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx); 1354} 1355#endif 1356 1357int shmem_lock(struct file *file, int lock, struct user_struct *user) 1358{ 1359 struct inode *inode = file->f_path.dentry->d_inode; 1360 struct shmem_inode_info *info = SHMEM_I(inode); 1361 int retval = -ENOMEM; 1362 1363 spin_lock(&info->lock); 1364 if (lock && !(info->flags & VM_LOCKED)) { 1365 if (!user_shm_lock(inode->i_size, user)) 1366 goto out_nomem; 1367 info->flags |= VM_LOCKED; 1368 } 1369 if (!lock && (info->flags & VM_LOCKED) && user) { 1370 user_shm_unlock(inode->i_size, user); 1371 info->flags &= ~VM_LOCKED; 1372 } 1373 retval = 0; 1374out_nomem: 1375 spin_unlock(&info->lock); 1376 return retval; 1377} 1378 1379static int shmem_mmap(struct file *file, struct vm_area_struct *vma) 1380{ 1381 file_accessed(file); 1382 vma->vm_ops = &shmem_vm_ops; 1383 vma->vm_flags |= VM_CAN_NONLINEAR; 1384 return 0; 1385} 1386 1387static struct inode * 1388shmem_get_inode(struct super_block *sb, int mode, dev_t dev) 1389{ 1390 struct inode *inode; 1391 struct shmem_inode_info *info; 1392 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 1393 1394 if (shmem_reserve_inode(sb)) 1395 return NULL; 1396 1397 inode = new_inode(sb); 1398 if (inode) { 1399 inode->i_mode = mode; 1400 inode->i_uid = current->fsuid; 1401 inode->i_gid = current->fsgid; 1402 inode->i_blocks = 0; 1403 inode->i_mapping->a_ops = &shmem_aops; 1404 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; 1405 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 1406 inode->i_generation = get_seconds(); 1407 info = SHMEM_I(inode); 1408 memset(info, 0, (char *)inode - (char *)info); 1409 spin_lock_init(&info->lock); 1410 INIT_LIST_HEAD(&info->swaplist); 1411 1412 switch (mode & S_IFMT) { 1413 default: 1414 inode->i_op = &shmem_special_inode_operations; 1415 init_special_inode(inode, mode, dev); 1416 break; 1417 case S_IFREG: 1418 inode->i_op = &shmem_inode_operations; 1419 inode->i_fop = &shmem_file_operations; 1420 mpol_shared_policy_init(&info->policy, sbinfo->policy, 1421 &sbinfo->policy_nodes); 1422 break; 1423 case S_IFDIR: 1424 inc_nlink(inode); 1425 /* Some things misbehave if size == 0 on a directory */ 1426 inode->i_size = 2 * BOGO_DIRENT_SIZE; 1427 inode->i_op = &shmem_dir_inode_operations; 1428 inode->i_fop = &simple_dir_operations; 1429 break; 1430 case S_IFLNK: 1431 /* 1432 * Must not load anything in the rbtree, 1433 * mpol_free_shared_policy will not be called. 1434 */ 1435 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, 1436 NULL); 1437 break; 1438 } 1439 } else 1440 shmem_free_inode(sb); 1441 return inode; 1442} 1443 1444#ifdef CONFIG_TMPFS 1445static const struct inode_operations shmem_symlink_inode_operations; 1446static const struct inode_operations shmem_symlink_inline_operations; 1447 1448/* 1449 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin; 1450 * but providing them allows a tmpfs file to be used for splice, sendfile, and 1451 * below the loop driver, in the generic fashion that many filesystems support. 1452 */ 1453static int shmem_readpage(struct file *file, struct page *page) 1454{ 1455 struct inode *inode = page->mapping->host; 1456 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL); 1457 unlock_page(page); 1458 return error; 1459} 1460 1461static int 1462shmem_write_begin(struct file *file, struct address_space *mapping, 1463 loff_t pos, unsigned len, unsigned flags, 1464 struct page **pagep, void **fsdata) 1465{ 1466 struct inode *inode = mapping->host; 1467 pgoff_t index = pos >> PAGE_CACHE_SHIFT; 1468 *pagep = NULL; 1469 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL); 1470} 1471 1472static int 1473shmem_write_end(struct file *file, struct address_space *mapping, 1474 loff_t pos, unsigned len, unsigned copied, 1475 struct page *page, void *fsdata) 1476{ 1477 struct inode *inode = mapping->host; 1478 1479 if (pos + copied > inode->i_size) 1480 i_size_write(inode, pos + copied); 1481 1482 unlock_page(page); 1483 set_page_dirty(page); 1484 page_cache_release(page); 1485 1486 return copied; 1487} 1488 1489static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor) 1490{ 1491 struct inode *inode = filp->f_path.dentry->d_inode; 1492 struct address_space *mapping = inode->i_mapping; 1493 unsigned long index, offset; 1494 1495 index = *ppos >> PAGE_CACHE_SHIFT; 1496 offset = *ppos & ~PAGE_CACHE_MASK; 1497 1498 for (;;) { 1499 struct page *page = NULL; 1500 unsigned long end_index, nr, ret; 1501 loff_t i_size = i_size_read(inode); 1502 1503 end_index = i_size >> PAGE_CACHE_SHIFT; 1504 if (index > end_index) 1505 break; 1506 if (index == end_index) { 1507 nr = i_size & ~PAGE_CACHE_MASK; 1508 if (nr <= offset) 1509 break; 1510 } 1511 1512 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL); 1513 if (desc->error) { 1514 if (desc->error == -EINVAL) 1515 desc->error = 0; 1516 break; 1517 } 1518 if (page) 1519 unlock_page(page); 1520 1521 /* 1522 * We must evaluate after, since reads (unlike writes) 1523 * are called without i_mutex protection against truncate 1524 */ 1525 nr = PAGE_CACHE_SIZE; 1526 i_size = i_size_read(inode); 1527 end_index = i_size >> PAGE_CACHE_SHIFT; 1528 if (index == end_index) { 1529 nr = i_size & ~PAGE_CACHE_MASK; 1530 if (nr <= offset) { 1531 if (page) 1532 page_cache_release(page); 1533 break; 1534 } 1535 } 1536 nr -= offset; 1537 1538 if (page) { 1539 /* 1540 * If users can be writing to this page using arbitrary 1541 * virtual addresses, take care about potential aliasing 1542 * before reading the page on the kernel side. 1543 */ 1544 if (mapping_writably_mapped(mapping)) 1545 flush_dcache_page(page); 1546 /* 1547 * Mark the page accessed if we read the beginning. 1548 */ 1549 if (!offset) 1550 mark_page_accessed(page); 1551 } else { 1552 page = ZERO_PAGE(0); 1553 page_cache_get(page); 1554 } 1555 1556 /* 1557 * Ok, we have the page, and it's up-to-date, so 1558 * now we can copy it to user space... 1559 * 1560 * The actor routine returns how many bytes were actually used.. 1561 * NOTE! This may not be the same as how much of a user buffer 1562 * we filled up (we may be padding etc), so we can only update 1563 * "pos" here (the actor routine has to update the user buffer 1564 * pointers and the remaining count). 1565 */ 1566 ret = actor(desc, page, offset, nr); 1567 offset += ret; 1568 index += offset >> PAGE_CACHE_SHIFT; 1569 offset &= ~PAGE_CACHE_MASK; 1570 1571 page_cache_release(page); 1572 if (ret != nr || !desc->count) 1573 break; 1574 1575 cond_resched(); 1576 } 1577 1578 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; 1579 file_accessed(filp); 1580} 1581 1582static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos) 1583{ 1584 read_descriptor_t desc; 1585 1586 if ((ssize_t) count < 0) 1587 return -EINVAL; 1588 if (!access_ok(VERIFY_WRITE, buf, count)) 1589 return -EFAULT; 1590 if (!count) 1591 return 0; 1592 1593 desc.written = 0; 1594 desc.count = count; 1595 desc.arg.buf = buf; 1596 desc.error = 0; 1597 1598 do_shmem_file_read(filp, ppos, &desc, file_read_actor); 1599 if (desc.written) 1600 return desc.written; 1601 return desc.error; 1602} 1603 1604static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) 1605{ 1606 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); 1607 1608 buf->f_type = TMPFS_MAGIC; 1609 buf->f_bsize = PAGE_CACHE_SIZE; 1610 buf->f_namelen = NAME_MAX; 1611 spin_lock(&sbinfo->stat_lock); 1612 if (sbinfo->max_blocks) { 1613 buf->f_blocks = sbinfo->max_blocks; 1614 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks; 1615 } 1616 if (sbinfo->max_inodes) { 1617 buf->f_files = sbinfo->max_inodes; 1618 buf->f_ffree = sbinfo->free_inodes; 1619 } 1620 /* else leave those fields 0 like simple_statfs */ 1621 spin_unlock(&sbinfo->stat_lock); 1622 return 0; 1623} 1624 1625/* 1626 * File creation. Allocate an inode, and we're done.. 1627 */ 1628static int 1629shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 1630{ 1631 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev); 1632 int error = -ENOSPC; 1633 1634 if (inode) { 1635 error = security_inode_init_security(inode, dir, NULL, NULL, 1636 NULL); 1637 if (error) { 1638 if (error != -EOPNOTSUPP) { 1639 iput(inode); 1640 return error; 1641 } 1642 } 1643 error = shmem_acl_init(inode, dir); 1644 if (error) { 1645 iput(inode); 1646 return error; 1647 } 1648 if (dir->i_mode & S_ISGID) { 1649 inode->i_gid = dir->i_gid; 1650 if (S_ISDIR(mode)) 1651 inode->i_mode |= S_ISGID; 1652 } 1653 dir->i_size += BOGO_DIRENT_SIZE; 1654 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1655 d_instantiate(dentry, inode); 1656 dget(dentry); /* Extra count - pin the dentry in core */ 1657 } 1658 return error; 1659} 1660 1661static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1662{ 1663 int error; 1664 1665 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) 1666 return error; 1667 inc_nlink(dir); 1668 return 0; 1669} 1670 1671static int shmem_create(struct inode *dir, struct dentry *dentry, int mode, 1672 struct nameidata *nd) 1673{ 1674 return shmem_mknod(dir, dentry, mode | S_IFREG, 0); 1675} 1676 1677/* 1678 * Link a file.. 1679 */ 1680static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1681{ 1682 struct inode *inode = old_dentry->d_inode; 1683 int ret; 1684 1685 /* 1686 * No ordinary (disk based) filesystem counts links as inodes; 1687 * but each new link needs a new dentry, pinning lowmem, and 1688 * tmpfs dentries cannot be pruned until they are unlinked. 1689 */ 1690 ret = shmem_reserve_inode(inode->i_sb); 1691 if (ret) 1692 goto out; 1693 1694 dir->i_size += BOGO_DIRENT_SIZE; 1695 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1696 inc_nlink(inode); 1697 atomic_inc(&inode->i_count); /* New dentry reference */ 1698 dget(dentry); /* Extra pinning count for the created dentry */ 1699 d_instantiate(dentry, inode); 1700out: 1701 return ret; 1702} 1703 1704static int shmem_unlink(struct inode *dir, struct dentry *dentry) 1705{ 1706 struct inode *inode = dentry->d_inode; 1707 1708 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) 1709 shmem_free_inode(inode->i_sb); 1710 1711 dir->i_size -= BOGO_DIRENT_SIZE; 1712 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1713 drop_nlink(inode); 1714 dput(dentry); /* Undo the count from "create" - this does all the work */ 1715 return 0; 1716} 1717 1718static int shmem_rmdir(struct inode *dir, struct dentry *dentry) 1719{ 1720 if (!simple_empty(dentry)) 1721 return -ENOTEMPTY; 1722 1723 drop_nlink(dentry->d_inode); 1724 drop_nlink(dir); 1725 return shmem_unlink(dir, dentry); 1726} 1727 1728/* 1729 * The VFS layer already does all the dentry stuff for rename, 1730 * we just have to decrement the usage count for the target if 1731 * it exists so that the VFS layer correctly free's it when it 1732 * gets overwritten. 1733 */ 1734static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) 1735{ 1736 struct inode *inode = old_dentry->d_inode; 1737 int they_are_dirs = S_ISDIR(inode->i_mode); 1738 1739 if (!simple_empty(new_dentry)) 1740 return -ENOTEMPTY; 1741 1742 if (new_dentry->d_inode) { 1743 (void) shmem_unlink(new_dir, new_dentry); 1744 if (they_are_dirs) 1745 drop_nlink(old_dir); 1746 } else if (they_are_dirs) { 1747 drop_nlink(old_dir); 1748 inc_nlink(new_dir); 1749 } 1750 1751 old_dir->i_size -= BOGO_DIRENT_SIZE; 1752 new_dir->i_size += BOGO_DIRENT_SIZE; 1753 old_dir->i_ctime = old_dir->i_mtime = 1754 new_dir->i_ctime = new_dir->i_mtime = 1755 inode->i_ctime = CURRENT_TIME; 1756 return 0; 1757} 1758 1759static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1760{ 1761 int error; 1762 int len; 1763 struct inode *inode; 1764 struct page *page = NULL; 1765 char *kaddr; 1766 struct shmem_inode_info *info; 1767 1768 len = strlen(symname) + 1; 1769 if (len > PAGE_CACHE_SIZE) 1770 return -ENAMETOOLONG; 1771 1772 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0); 1773 if (!inode) 1774 return -ENOSPC; 1775 1776 error = security_inode_init_security(inode, dir, NULL, NULL, 1777 NULL); 1778 if (error) { 1779 if (error != -EOPNOTSUPP) { 1780 iput(inode); 1781 return error; 1782 } 1783 error = 0; 1784 } 1785 1786 info = SHMEM_I(inode); 1787 inode->i_size = len-1; 1788 if (len <= (char *)inode - (char *)info) { 1789 /* do it inline */ 1790 memcpy(info, symname, len); 1791 inode->i_op = &shmem_symlink_inline_operations; 1792 } else { 1793 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); 1794 if (error) { 1795 iput(inode); 1796 return error; 1797 } 1798 unlock_page(page); 1799 inode->i_op = &shmem_symlink_inode_operations; 1800 kaddr = kmap_atomic(page, KM_USER0); 1801 memcpy(kaddr, symname, len); 1802 kunmap_atomic(kaddr, KM_USER0); 1803 set_page_dirty(page); 1804 page_cache_release(page); 1805 } 1806 if (dir->i_mode & S_ISGID) 1807 inode->i_gid = dir->i_gid; 1808 dir->i_size += BOGO_DIRENT_SIZE; 1809 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1810 d_instantiate(dentry, inode); 1811 dget(dentry); 1812 return 0; 1813} 1814 1815static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd) 1816{ 1817 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode)); 1818 return NULL; 1819} 1820 1821static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) 1822{ 1823 struct page *page = NULL; 1824 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); 1825 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page)); 1826 if (page) 1827 unlock_page(page); 1828 return page; 1829} 1830 1831static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 1832{ 1833 if (!IS_ERR(nd_get_link(nd))) { 1834 struct page *page = cookie; 1835 kunmap(page); 1836 mark_page_accessed(page); 1837 page_cache_release(page); 1838 } 1839} 1840 1841static const struct inode_operations shmem_symlink_inline_operations = { 1842 .readlink = generic_readlink, 1843 .follow_link = shmem_follow_link_inline, 1844}; 1845 1846static const struct inode_operations shmem_symlink_inode_operations = { 1847 .truncate = shmem_truncate, 1848 .readlink = generic_readlink, 1849 .follow_link = shmem_follow_link, 1850 .put_link = shmem_put_link, 1851}; 1852 1853#ifdef CONFIG_TMPFS_POSIX_ACL 1854/** 1855 * Superblocks without xattr inode operations will get security.* xattr 1856 * support from the VFS "for free". As soon as we have any other xattrs 1857 * like ACLs, we also need to implement the security.* handlers at 1858 * filesystem level, though. 1859 */ 1860 1861static size_t shmem_xattr_security_list(struct inode *inode, char *list, 1862 size_t list_len, const char *name, 1863 size_t name_len) 1864{ 1865 return security_inode_listsecurity(inode, list, list_len); 1866} 1867 1868static int shmem_xattr_security_get(struct inode *inode, const char *name, 1869 void *buffer, size_t size) 1870{ 1871 if (strcmp(name, "") == 0) 1872 return -EINVAL; 1873 return security_inode_getsecurity(inode, name, buffer, size, 1874 -EOPNOTSUPP); 1875} 1876 1877static int shmem_xattr_security_set(struct inode *inode, const char *name, 1878 const void *value, size_t size, int flags) 1879{ 1880 if (strcmp(name, "") == 0) 1881 return -EINVAL; 1882 return security_inode_setsecurity(inode, name, value, size, flags); 1883} 1884 1885static struct xattr_handler shmem_xattr_security_handler = { 1886 .prefix = XATTR_SECURITY_PREFIX, 1887 .list = shmem_xattr_security_list, 1888 .get = shmem_xattr_security_get, 1889 .set = shmem_xattr_security_set, 1890}; 1891 1892static struct xattr_handler *shmem_xattr_handlers[] = { 1893 &shmem_xattr_acl_access_handler, 1894 &shmem_xattr_acl_default_handler, 1895 &shmem_xattr_security_handler, 1896 NULL 1897}; 1898#endif 1899 1900static struct dentry *shmem_get_parent(struct dentry *child) 1901{ 1902 return ERR_PTR(-ESTALE); 1903} 1904 1905static int shmem_match(struct inode *ino, void *vfh) 1906{ 1907 __u32 *fh = vfh; 1908 __u64 inum = fh[2]; 1909 inum = (inum << 32) | fh[1]; 1910 return ino->i_ino == inum && fh[0] == ino->i_generation; 1911} 1912 1913static struct dentry *shmem_fh_to_dentry(struct super_block *sb, 1914 struct fid *fid, int fh_len, int fh_type) 1915{ 1916 struct inode *inode; 1917 struct dentry *dentry = NULL; 1918 u64 inum = fid->raw[2]; 1919 inum = (inum << 32) | fid->raw[1]; 1920 1921 if (fh_len < 3) 1922 return NULL; 1923 1924 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), 1925 shmem_match, fid->raw); 1926 if (inode) { 1927 dentry = d_find_alias(inode); 1928 iput(inode); 1929 } 1930 1931 return dentry; 1932} 1933 1934static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len, 1935 int connectable) 1936{ 1937 struct inode *inode = dentry->d_inode; 1938 1939 if (*len < 3) 1940 return 255; 1941 1942 if (hlist_unhashed(&inode->i_hash)) { 1943 /* Unfortunately insert_inode_hash is not idempotent, 1944 * so as we hash inodes here rather than at creation 1945 * time, we need a lock to ensure we only try 1946 * to do it once 1947 */ 1948 static DEFINE_SPINLOCK(lock); 1949 spin_lock(&lock); 1950 if (hlist_unhashed(&inode->i_hash)) 1951 __insert_inode_hash(inode, 1952 inode->i_ino + inode->i_generation); 1953 spin_unlock(&lock); 1954 } 1955 1956 fh[0] = inode->i_generation; 1957 fh[1] = inode->i_ino; 1958 fh[2] = ((__u64)inode->i_ino) >> 32; 1959 1960 *len = 3; 1961 return 1; 1962} 1963 1964static const struct export_operations shmem_export_ops = { 1965 .get_parent = shmem_get_parent, 1966 .encode_fh = shmem_encode_fh, 1967 .fh_to_dentry = shmem_fh_to_dentry, 1968}; 1969 1970static int shmem_parse_options(char *options, int *mode, uid_t *uid, 1971 gid_t *gid, unsigned long *blocks, unsigned long *inodes, 1972 int *policy, nodemask_t *policy_nodes) 1973{ 1974 char *this_char, *value, *rest; 1975 1976 while (options != NULL) { 1977 this_char = options; 1978 for (;;) { 1979 /* 1980 * NUL-terminate this option: unfortunately, 1981 * mount options form a comma-separated list, 1982 * but mpol's nodelist may also contain commas. 1983 */ 1984 options = strchr(options, ','); 1985 if (options == NULL) 1986 break; 1987 options++; 1988 if (!isdigit(*options)) { 1989 options[-1] = '\0'; 1990 break; 1991 } 1992 } 1993 if (!*this_char) 1994 continue; 1995 if ((value = strchr(this_char,'=')) != NULL) { 1996 *value++ = 0; 1997 } else { 1998 printk(KERN_ERR 1999 "tmpfs: No value for mount option '%s'\n", 2000 this_char); 2001 return 1; 2002 } 2003 2004 if (!strcmp(this_char,"size")) { 2005 unsigned long long size; 2006 size = memparse(value,&rest); 2007 if (*rest == '%') { 2008 size <<= PAGE_SHIFT; 2009 size *= totalram_pages; 2010 do_div(size, 100); 2011 rest++; 2012 } 2013 if (*rest) 2014 goto bad_val; 2015 *blocks = DIV_ROUND_UP(size, PAGE_CACHE_SIZE); 2016 } else if (!strcmp(this_char,"nr_blocks")) { 2017 *blocks = memparse(value,&rest); 2018 if (*rest) 2019 goto bad_val; 2020 } else if (!strcmp(this_char,"nr_inodes")) { 2021 *inodes = memparse(value,&rest); 2022 if (*rest) 2023 goto bad_val; 2024 } else if (!strcmp(this_char,"mode")) { 2025 if (!mode) 2026 continue; 2027 *mode = simple_strtoul(value,&rest,8); 2028 if (*rest) 2029 goto bad_val; 2030 } else if (!strcmp(this_char,"uid")) { 2031 if (!uid) 2032 continue; 2033 *uid = simple_strtoul(value,&rest,0); 2034 if (*rest) 2035 goto bad_val; 2036 } else if (!strcmp(this_char,"gid")) { 2037 if (!gid) 2038 continue; 2039 *gid = simple_strtoul(value,&rest,0); 2040 if (*rest) 2041 goto bad_val; 2042 } else if (!strcmp(this_char,"mpol")) { 2043 if (shmem_parse_mpol(value,policy,policy_nodes)) 2044 goto bad_val; 2045 } else { 2046 printk(KERN_ERR "tmpfs: Bad mount option %s\n", 2047 this_char); 2048 return 1; 2049 } 2050 } 2051 return 0; 2052 2053bad_val: 2054 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", 2055 value, this_char); 2056 return 1; 2057 2058} 2059 2060static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) 2061{ 2062 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 2063 unsigned long max_blocks = sbinfo->max_blocks; 2064 unsigned long max_inodes = sbinfo->max_inodes; 2065 int policy = sbinfo->policy; 2066 nodemask_t policy_nodes = sbinfo->policy_nodes; 2067 unsigned long blocks; 2068 unsigned long inodes; 2069 int error = -EINVAL; 2070 2071 if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks, 2072 &max_inodes, &policy, &policy_nodes)) 2073 return error; 2074 2075 spin_lock(&sbinfo->stat_lock); 2076 blocks = sbinfo->max_blocks - sbinfo->free_blocks; 2077 inodes = sbinfo->max_inodes - sbinfo->free_inodes; 2078 if (max_blocks < blocks) 2079 goto out; 2080 if (max_inodes < inodes) 2081 goto out; 2082 /* 2083 * Those tests also disallow limited->unlimited while any are in 2084 * use, so i_blocks will always be zero when max_blocks is zero; 2085 * but we must separately disallow unlimited->limited, because 2086 * in that case we have no record of how much is already in use. 2087 */ 2088 if (max_blocks && !sbinfo->max_blocks) 2089 goto out; 2090 if (max_inodes && !sbinfo->max_inodes) 2091 goto out; 2092 2093 error = 0; 2094 sbinfo->max_blocks = max_blocks; 2095 sbinfo->free_blocks = max_blocks - blocks; 2096 sbinfo->max_inodes = max_inodes; 2097 sbinfo->free_inodes = max_inodes - inodes; 2098 sbinfo->policy = policy; 2099 sbinfo->policy_nodes = policy_nodes; 2100out: 2101 spin_unlock(&sbinfo->stat_lock); 2102 return error; 2103} 2104#endif 2105 2106static void shmem_put_super(struct super_block *sb) 2107{ 2108 kfree(sb->s_fs_info); 2109 sb->s_fs_info = NULL; 2110} 2111 2112static int shmem_fill_super(struct super_block *sb, 2113 void *data, int silent) 2114{ 2115 struct inode *inode; 2116 struct dentry *root; 2117 int mode = S_IRWXUGO | S_ISVTX; 2118 uid_t uid = current->fsuid; 2119 gid_t gid = current->fsgid; 2120 int err = -ENOMEM; 2121 struct shmem_sb_info *sbinfo; 2122 unsigned long blocks = 0; 2123 unsigned long inodes = 0; 2124 int policy = MPOL_DEFAULT; 2125 nodemask_t policy_nodes = node_states[N_HIGH_MEMORY]; 2126 2127#ifdef CONFIG_TMPFS 2128 /* 2129 * Per default we only allow half of the physical ram per 2130 * tmpfs instance, limiting inodes to one per page of lowmem; 2131 * but the internal instance is left unlimited. 2132 */ 2133 if (!(sb->s_flags & MS_NOUSER)) { 2134 blocks = totalram_pages / 2; 2135 inodes = totalram_pages - totalhigh_pages; 2136 if (inodes > blocks) 2137 inodes = blocks; 2138 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks, 2139 &inodes, &policy, &policy_nodes)) 2140 return -EINVAL; 2141 } 2142 sb->s_export_op = &shmem_export_ops; 2143#else 2144 sb->s_flags |= MS_NOUSER; 2145#endif 2146 2147 /* Round up to L1_CACHE_BYTES to resist false sharing */ 2148 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info), 2149 L1_CACHE_BYTES), GFP_KERNEL); 2150 if (!sbinfo) 2151 return -ENOMEM; 2152 2153 spin_lock_init(&sbinfo->stat_lock); 2154 sbinfo->max_blocks = blocks; 2155 sbinfo->free_blocks = blocks; 2156 sbinfo->max_inodes = inodes; 2157 sbinfo->free_inodes = inodes; 2158 sbinfo->policy = policy; 2159 sbinfo->policy_nodes = policy_nodes; 2160 2161 sb->s_fs_info = sbinfo; 2162 sb->s_maxbytes = SHMEM_MAX_BYTES; 2163 sb->s_blocksize = PAGE_CACHE_SIZE; 2164 sb->s_blocksize_bits = PAGE_CACHE_SHIFT; 2165 sb->s_magic = TMPFS_MAGIC; 2166 sb->s_op = &shmem_ops; 2167 sb->s_time_gran = 1; 2168#ifdef CONFIG_TMPFS_POSIX_ACL 2169 sb->s_xattr = shmem_xattr_handlers; 2170 sb->s_flags |= MS_POSIXACL; 2171#endif 2172 2173 inode = shmem_get_inode(sb, S_IFDIR | mode, 0); 2174 if (!inode) 2175 goto failed; 2176 inode->i_uid = uid; 2177 inode->i_gid = gid; 2178 root = d_alloc_root(inode); 2179 if (!root) 2180 goto failed_iput; 2181 sb->s_root = root; 2182 return 0; 2183 2184failed_iput: 2185 iput(inode); 2186failed: 2187 shmem_put_super(sb); 2188 return err; 2189} 2190 2191static struct kmem_cache *shmem_inode_cachep; 2192 2193static struct inode *shmem_alloc_inode(struct super_block *sb) 2194{ 2195 struct shmem_inode_info *p; 2196 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); 2197 if (!p) 2198 return NULL; 2199 return &p->vfs_inode; 2200} 2201 2202static void shmem_destroy_inode(struct inode *inode) 2203{ 2204 if ((inode->i_mode & S_IFMT) == S_IFREG) { 2205 /* only struct inode is valid if it's an inline symlink */ 2206 mpol_free_shared_policy(&SHMEM_I(inode)->policy); 2207 } 2208 shmem_acl_destroy_inode(inode); 2209 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); 2210} 2211 2212static void init_once(struct kmem_cache *cachep, void *foo) 2213{ 2214 struct shmem_inode_info *p = (struct shmem_inode_info *) foo; 2215 2216 inode_init_once(&p->vfs_inode); 2217#ifdef CONFIG_TMPFS_POSIX_ACL 2218 p->i_acl = NULL; 2219 p->i_default_acl = NULL; 2220#endif 2221} 2222 2223static int init_inodecache(void) 2224{ 2225 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", 2226 sizeof(struct shmem_inode_info), 2227 0, SLAB_PANIC, init_once); 2228 return 0; 2229} 2230 2231static void destroy_inodecache(void) 2232{ 2233 kmem_cache_destroy(shmem_inode_cachep); 2234} 2235 2236static const struct address_space_operations shmem_aops = { 2237 .writepage = shmem_writepage, 2238 .set_page_dirty = __set_page_dirty_no_writeback, 2239#ifdef CONFIG_TMPFS 2240 .readpage = shmem_readpage, 2241 .write_begin = shmem_write_begin, 2242 .write_end = shmem_write_end, 2243#endif 2244 .migratepage = migrate_page, 2245}; 2246 2247static const struct file_operations shmem_file_operations = { 2248 .mmap = shmem_mmap, 2249#ifdef CONFIG_TMPFS 2250 .llseek = generic_file_llseek, 2251 .read = shmem_file_read, 2252 .write = do_sync_write, 2253 .aio_write = generic_file_aio_write, 2254 .fsync = simple_sync_file, 2255 .splice_read = generic_file_splice_read, 2256 .splice_write = generic_file_splice_write, 2257#endif 2258}; 2259 2260static const struct inode_operations shmem_inode_operations = { 2261 .truncate = shmem_truncate, 2262 .setattr = shmem_notify_change, 2263 .truncate_range = shmem_truncate_range, 2264#ifdef CONFIG_TMPFS_POSIX_ACL 2265 .setxattr = generic_setxattr, 2266 .getxattr = generic_getxattr, 2267 .listxattr = generic_listxattr, 2268 .removexattr = generic_removexattr, 2269 .permission = shmem_permission, 2270#endif 2271 2272}; 2273 2274static const struct inode_operations shmem_dir_inode_operations = { 2275#ifdef CONFIG_TMPFS 2276 .create = shmem_create, 2277 .lookup = simple_lookup, 2278 .link = shmem_link, 2279 .unlink = shmem_unlink, 2280 .symlink = shmem_symlink, 2281 .mkdir = shmem_mkdir, 2282 .rmdir = shmem_rmdir, 2283 .mknod = shmem_mknod, 2284 .rename = shmem_rename, 2285#endif 2286#ifdef CONFIG_TMPFS_POSIX_ACL 2287 .setattr = shmem_notify_change, 2288 .setxattr = generic_setxattr, 2289 .getxattr = generic_getxattr, 2290 .listxattr = generic_listxattr, 2291 .removexattr = generic_removexattr, 2292 .permission = shmem_permission, 2293#endif 2294}; 2295 2296static const struct inode_operations shmem_special_inode_operations = { 2297#ifdef CONFIG_TMPFS_POSIX_ACL 2298 .setattr = shmem_notify_change, 2299 .setxattr = generic_setxattr, 2300 .getxattr = generic_getxattr, 2301 .listxattr = generic_listxattr, 2302 .removexattr = generic_removexattr, 2303 .permission = shmem_permission, 2304#endif 2305}; 2306 2307static const struct super_operations shmem_ops = { 2308 .alloc_inode = shmem_alloc_inode, 2309 .destroy_inode = shmem_destroy_inode, 2310#ifdef CONFIG_TMPFS 2311 .statfs = shmem_statfs, 2312 .remount_fs = shmem_remount_fs, 2313#endif 2314 .delete_inode = shmem_delete_inode, 2315 .drop_inode = generic_delete_inode, 2316 .put_super = shmem_put_super, 2317}; 2318 2319static struct vm_operations_struct shmem_vm_ops = { 2320 .fault = shmem_fault, 2321#ifdef CONFIG_NUMA 2322 .set_policy = shmem_set_policy, 2323 .get_policy = shmem_get_policy, 2324#endif 2325}; 2326 2327 2328static int shmem_get_sb(struct file_system_type *fs_type, 2329 int flags, const char *dev_name, void *data, struct vfsmount *mnt) 2330{ 2331 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt); 2332} 2333 2334static struct file_system_type tmpfs_fs_type = { 2335 .owner = THIS_MODULE, 2336 .name = "tmpfs", 2337 .get_sb = shmem_get_sb, 2338 .kill_sb = kill_litter_super, 2339}; 2340static struct vfsmount *shm_mnt; 2341 2342static int __init init_tmpfs(void) 2343{ 2344 int error; 2345 2346 error = bdi_init(&shmem_backing_dev_info); 2347 if (error) 2348 goto out4; 2349 2350 error = init_inodecache(); 2351 if (error) 2352 goto out3; 2353 2354 error = register_filesystem(&tmpfs_fs_type); 2355 if (error) { 2356 printk(KERN_ERR "Could not register tmpfs\n"); 2357 goto out2; 2358 } 2359 2360 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER, 2361 tmpfs_fs_type.name, NULL); 2362 if (IS_ERR(shm_mnt)) { 2363 error = PTR_ERR(shm_mnt); 2364 printk(KERN_ERR "Could not kern_mount tmpfs\n"); 2365 goto out1; 2366 } 2367 return 0; 2368 2369out1: 2370 unregister_filesystem(&tmpfs_fs_type); 2371out2: 2372 destroy_inodecache(); 2373out3: 2374 bdi_destroy(&shmem_backing_dev_info); 2375out4: 2376 shm_mnt = ERR_PTR(error); 2377 return error; 2378} 2379module_init(init_tmpfs) 2380 2381/* 2382 * shmem_file_setup - get an unlinked file living in tmpfs 2383 * 2384 * @name: name for dentry (to be seen in /proc/<pid>/maps 2385 * @size: size to be set for the file 2386 * 2387 */ 2388struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags) 2389{ 2390 int error; 2391 struct file *file; 2392 struct inode *inode; 2393 struct dentry *dentry, *root; 2394 struct qstr this; 2395 2396 if (IS_ERR(shm_mnt)) 2397 return (void *)shm_mnt; 2398 2399 if (size < 0 || size > SHMEM_MAX_BYTES) 2400 return ERR_PTR(-EINVAL); 2401 2402 if (shmem_acct_size(flags, size)) 2403 return ERR_PTR(-ENOMEM); 2404 2405 error = -ENOMEM; 2406 this.name = name; 2407 this.len = strlen(name); 2408 this.hash = 0; /* will go */ 2409 root = shm_mnt->mnt_root; 2410 dentry = d_alloc(root, &this); 2411 if (!dentry) 2412 goto put_memory; 2413 2414 error = -ENFILE; 2415 file = get_empty_filp(); 2416 if (!file) 2417 goto put_dentry; 2418 2419 error = -ENOSPC; 2420 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0); 2421 if (!inode) 2422 goto close_file; 2423 2424 SHMEM_I(inode)->flags = flags & VM_ACCOUNT; 2425 d_instantiate(dentry, inode); 2426 inode->i_size = size; 2427 inode->i_nlink = 0; /* It is unlinked */ 2428 init_file(file, shm_mnt, dentry, FMODE_WRITE | FMODE_READ, 2429 &shmem_file_operations); 2430 return file; 2431 2432close_file: 2433 put_filp(file); 2434put_dentry: 2435 dput(dentry); 2436put_memory: 2437 shmem_unacct_size(flags, size); 2438 return ERR_PTR(error); 2439} 2440 2441/* 2442 * shmem_zero_setup - setup a shared anonymous mapping 2443 * 2444 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff 2445 */ 2446int shmem_zero_setup(struct vm_area_struct *vma) 2447{ 2448 struct file *file; 2449 loff_t size = vma->vm_end - vma->vm_start; 2450 2451 file = shmem_file_setup("dev/zero", size, vma->vm_flags); 2452 if (IS_ERR(file)) 2453 return PTR_ERR(file); 2454 2455 if (vma->vm_file) 2456 fput(vma->vm_file); 2457 vma->vm_file = file; 2458 vma->vm_ops = &shmem_vm_ops; 2459 return 0; 2460} 2461