shmem.c revision 54cb8821de07f2ffcd28c380ce9b93d5784b40d7
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/backing-dev.h> 53 54#include <asm/uaccess.h> 55#include <asm/div64.h> 56#include <asm/pgtable.h> 57 58/* This magic number is used in glibc for posix shared memory */ 59#define TMPFS_MAGIC 0x01021994 60 61#define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long)) 62#define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE) 63#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512) 64 65#define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1)) 66#define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT) 67 68#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT) 69 70/* info->flags needs VM_flags to handle pagein/truncate races efficiently */ 71#define SHMEM_PAGEIN VM_READ 72#define SHMEM_TRUNCATE VM_WRITE 73 74/* Definition to limit shmem_truncate's steps between cond_rescheds */ 75#define LATENCY_LIMIT 64 76 77/* Pretend that each entry is of this size in directory's i_size */ 78#define BOGO_DIRENT_SIZE 20 79 80/* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */ 81enum sgp_type { 82 SGP_QUICK, /* don't try more than file page cache lookup */ 83 SGP_READ, /* don't exceed i_size, don't allocate page */ 84 SGP_CACHE, /* don't exceed i_size, may allocate page */ 85 SGP_WRITE, /* may exceed i_size, may allocate page */ 86 SGP_FAULT, /* same as SGP_CACHE, return with page locked */ 87}; 88 89static int shmem_getpage(struct inode *inode, unsigned long idx, 90 struct page **pagep, enum sgp_type sgp, int *type); 91 92static inline struct page *shmem_dir_alloc(gfp_t gfp_mask) 93{ 94 /* 95 * The above definition of ENTRIES_PER_PAGE, and the use of 96 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE: 97 * might be reconsidered if it ever diverges from PAGE_SIZE. 98 * 99 * __GFP_MOVABLE is masked out as swap vectors cannot move 100 */ 101 return alloc_pages((gfp_mask & ~__GFP_MOVABLE) | __GFP_ZERO, 102 PAGE_CACHE_SHIFT-PAGE_SHIFT); 103} 104 105static inline void shmem_dir_free(struct page *page) 106{ 107 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT); 108} 109 110static struct page **shmem_dir_map(struct page *page) 111{ 112 return (struct page **)kmap_atomic(page, KM_USER0); 113} 114 115static inline void shmem_dir_unmap(struct page **dir) 116{ 117 kunmap_atomic(dir, KM_USER0); 118} 119 120static swp_entry_t *shmem_swp_map(struct page *page) 121{ 122 return (swp_entry_t *)kmap_atomic(page, KM_USER1); 123} 124 125static inline void shmem_swp_balance_unmap(void) 126{ 127 /* 128 * When passing a pointer to an i_direct entry, to code which 129 * also handles indirect entries and so will shmem_swp_unmap, 130 * we must arrange for the preempt count to remain in balance. 131 * What kmap_atomic of a lowmem page does depends on config 132 * and architecture, so pretend to kmap_atomic some lowmem page. 133 */ 134 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1); 135} 136 137static inline void shmem_swp_unmap(swp_entry_t *entry) 138{ 139 kunmap_atomic(entry, KM_USER1); 140} 141 142static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) 143{ 144 return sb->s_fs_info; 145} 146 147/* 148 * shmem_file_setup pre-accounts the whole fixed size of a VM object, 149 * for shared memory and for shared anonymous (/dev/zero) mappings 150 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), 151 * consistent with the pre-accounting of private mappings ... 152 */ 153static inline int shmem_acct_size(unsigned long flags, loff_t size) 154{ 155 return (flags & VM_ACCOUNT)? 156 security_vm_enough_memory(VM_ACCT(size)): 0; 157} 158 159static inline void shmem_unacct_size(unsigned long flags, loff_t size) 160{ 161 if (flags & VM_ACCOUNT) 162 vm_unacct_memory(VM_ACCT(size)); 163} 164 165/* 166 * ... whereas tmpfs objects are accounted incrementally as 167 * pages are allocated, in order to allow huge sparse files. 168 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, 169 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. 170 */ 171static inline int shmem_acct_block(unsigned long flags) 172{ 173 return (flags & VM_ACCOUNT)? 174 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE)); 175} 176 177static inline void shmem_unacct_blocks(unsigned long flags, long pages) 178{ 179 if (!(flags & VM_ACCOUNT)) 180 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE)); 181} 182 183static const struct super_operations shmem_ops; 184static const struct address_space_operations shmem_aops; 185static const struct file_operations shmem_file_operations; 186static const struct inode_operations shmem_inode_operations; 187static const struct inode_operations shmem_dir_inode_operations; 188static const struct inode_operations shmem_special_inode_operations; 189static struct vm_operations_struct shmem_vm_ops; 190 191static struct backing_dev_info shmem_backing_dev_info __read_mostly = { 192 .ra_pages = 0, /* No readahead */ 193 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK, 194 .unplug_io_fn = default_unplug_io_fn, 195}; 196 197static LIST_HEAD(shmem_swaplist); 198static DEFINE_SPINLOCK(shmem_swaplist_lock); 199 200static void shmem_free_blocks(struct inode *inode, long pages) 201{ 202 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 203 if (sbinfo->max_blocks) { 204 spin_lock(&sbinfo->stat_lock); 205 sbinfo->free_blocks += pages; 206 inode->i_blocks -= pages*BLOCKS_PER_PAGE; 207 spin_unlock(&sbinfo->stat_lock); 208 } 209} 210 211/* 212 * shmem_recalc_inode - recalculate the size of an inode 213 * 214 * @inode: inode to recalc 215 * 216 * We have to calculate the free blocks since the mm can drop 217 * undirtied hole pages behind our back. 218 * 219 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped 220 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) 221 * 222 * It has to be called with the spinlock held. 223 */ 224static void shmem_recalc_inode(struct inode *inode) 225{ 226 struct shmem_inode_info *info = SHMEM_I(inode); 227 long freed; 228 229 freed = info->alloced - info->swapped - inode->i_mapping->nrpages; 230 if (freed > 0) { 231 info->alloced -= freed; 232 shmem_unacct_blocks(info->flags, freed); 233 shmem_free_blocks(inode, freed); 234 } 235} 236 237/* 238 * shmem_swp_entry - find the swap vector position in the info structure 239 * 240 * @info: info structure for the inode 241 * @index: index of the page to find 242 * @page: optional page to add to the structure. Has to be preset to 243 * all zeros 244 * 245 * If there is no space allocated yet it will return NULL when 246 * page is NULL, else it will use the page for the needed block, 247 * setting it to NULL on return to indicate that it has been used. 248 * 249 * The swap vector is organized the following way: 250 * 251 * There are SHMEM_NR_DIRECT entries directly stored in the 252 * shmem_inode_info structure. So small files do not need an addional 253 * allocation. 254 * 255 * For pages with index > SHMEM_NR_DIRECT there is the pointer 256 * i_indirect which points to a page which holds in the first half 257 * doubly indirect blocks, in the second half triple indirect blocks: 258 * 259 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the 260 * following layout (for SHMEM_NR_DIRECT == 16): 261 * 262 * i_indirect -> dir --> 16-19 263 * | +-> 20-23 264 * | 265 * +-->dir2 --> 24-27 266 * | +-> 28-31 267 * | +-> 32-35 268 * | +-> 36-39 269 * | 270 * +-->dir3 --> 40-43 271 * +-> 44-47 272 * +-> 48-51 273 * +-> 52-55 274 */ 275static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page) 276{ 277 unsigned long offset; 278 struct page **dir; 279 struct page *subdir; 280 281 if (index < SHMEM_NR_DIRECT) { 282 shmem_swp_balance_unmap(); 283 return info->i_direct+index; 284 } 285 if (!info->i_indirect) { 286 if (page) { 287 info->i_indirect = *page; 288 *page = NULL; 289 } 290 return NULL; /* need another page */ 291 } 292 293 index -= SHMEM_NR_DIRECT; 294 offset = index % ENTRIES_PER_PAGE; 295 index /= ENTRIES_PER_PAGE; 296 dir = shmem_dir_map(info->i_indirect); 297 298 if (index >= ENTRIES_PER_PAGE/2) { 299 index -= ENTRIES_PER_PAGE/2; 300 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE; 301 index %= ENTRIES_PER_PAGE; 302 subdir = *dir; 303 if (!subdir) { 304 if (page) { 305 *dir = *page; 306 *page = NULL; 307 } 308 shmem_dir_unmap(dir); 309 return NULL; /* need another page */ 310 } 311 shmem_dir_unmap(dir); 312 dir = shmem_dir_map(subdir); 313 } 314 315 dir += index; 316 subdir = *dir; 317 if (!subdir) { 318 if (!page || !(subdir = *page)) { 319 shmem_dir_unmap(dir); 320 return NULL; /* need a page */ 321 } 322 *dir = subdir; 323 *page = NULL; 324 } 325 shmem_dir_unmap(dir); 326 return shmem_swp_map(subdir) + offset; 327} 328 329static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value) 330{ 331 long incdec = value? 1: -1; 332 333 entry->val = value; 334 info->swapped += incdec; 335 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) { 336 struct page *page = kmap_atomic_to_page(entry); 337 set_page_private(page, page_private(page) + incdec); 338 } 339} 340 341/* 342 * shmem_swp_alloc - get the position of the swap entry for the page. 343 * If it does not exist allocate the entry. 344 * 345 * @info: info structure for the inode 346 * @index: index of the page to find 347 * @sgp: check and recheck i_size? skip allocation? 348 */ 349static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp) 350{ 351 struct inode *inode = &info->vfs_inode; 352 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 353 struct page *page = NULL; 354 swp_entry_t *entry; 355 356 if (sgp != SGP_WRITE && 357 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 358 return ERR_PTR(-EINVAL); 359 360 while (!(entry = shmem_swp_entry(info, index, &page))) { 361 if (sgp == SGP_READ) 362 return shmem_swp_map(ZERO_PAGE(0)); 363 /* 364 * Test free_blocks against 1 not 0, since we have 1 data 365 * page (and perhaps indirect index pages) yet to allocate: 366 * a waste to allocate index if we cannot allocate data. 367 */ 368 if (sbinfo->max_blocks) { 369 spin_lock(&sbinfo->stat_lock); 370 if (sbinfo->free_blocks <= 1) { 371 spin_unlock(&sbinfo->stat_lock); 372 return ERR_PTR(-ENOSPC); 373 } 374 sbinfo->free_blocks--; 375 inode->i_blocks += BLOCKS_PER_PAGE; 376 spin_unlock(&sbinfo->stat_lock); 377 } 378 379 spin_unlock(&info->lock); 380 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping)); 381 if (page) 382 set_page_private(page, 0); 383 spin_lock(&info->lock); 384 385 if (!page) { 386 shmem_free_blocks(inode, 1); 387 return ERR_PTR(-ENOMEM); 388 } 389 if (sgp != SGP_WRITE && 390 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { 391 entry = ERR_PTR(-EINVAL); 392 break; 393 } 394 if (info->next_index <= index) 395 info->next_index = index + 1; 396 } 397 if (page) { 398 /* another task gave its page, or truncated the file */ 399 shmem_free_blocks(inode, 1); 400 shmem_dir_free(page); 401 } 402 if (info->next_index <= index && !IS_ERR(entry)) 403 info->next_index = index + 1; 404 return entry; 405} 406 407/* 408 * shmem_free_swp - free some swap entries in a directory 409 * 410 * @dir: pointer to the directory 411 * @edir: pointer after last entry of the directory 412 * @punch_lock: pointer to spinlock when needed for the holepunch case 413 */ 414static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir, 415 spinlock_t *punch_lock) 416{ 417 spinlock_t *punch_unlock = NULL; 418 swp_entry_t *ptr; 419 int freed = 0; 420 421 for (ptr = dir; ptr < edir; ptr++) { 422 if (ptr->val) { 423 if (unlikely(punch_lock)) { 424 punch_unlock = punch_lock; 425 punch_lock = NULL; 426 spin_lock(punch_unlock); 427 if (!ptr->val) 428 continue; 429 } 430 free_swap_and_cache(*ptr); 431 *ptr = (swp_entry_t){0}; 432 freed++; 433 } 434 } 435 if (punch_unlock) 436 spin_unlock(punch_unlock); 437 return freed; 438} 439 440static int shmem_map_and_free_swp(struct page *subdir, int offset, 441 int limit, struct page ***dir, spinlock_t *punch_lock) 442{ 443 swp_entry_t *ptr; 444 int freed = 0; 445 446 ptr = shmem_swp_map(subdir); 447 for (; offset < limit; offset += LATENCY_LIMIT) { 448 int size = limit - offset; 449 if (size > LATENCY_LIMIT) 450 size = LATENCY_LIMIT; 451 freed += shmem_free_swp(ptr+offset, ptr+offset+size, 452 punch_lock); 453 if (need_resched()) { 454 shmem_swp_unmap(ptr); 455 if (*dir) { 456 shmem_dir_unmap(*dir); 457 *dir = NULL; 458 } 459 cond_resched(); 460 ptr = shmem_swp_map(subdir); 461 } 462 } 463 shmem_swp_unmap(ptr); 464 return freed; 465} 466 467static void shmem_free_pages(struct list_head *next) 468{ 469 struct page *page; 470 int freed = 0; 471 472 do { 473 page = container_of(next, struct page, lru); 474 next = next->next; 475 shmem_dir_free(page); 476 freed++; 477 if (freed >= LATENCY_LIMIT) { 478 cond_resched(); 479 freed = 0; 480 } 481 } while (next); 482} 483 484static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end) 485{ 486 struct shmem_inode_info *info = SHMEM_I(inode); 487 unsigned long idx; 488 unsigned long size; 489 unsigned long limit; 490 unsigned long stage; 491 unsigned long diroff; 492 struct page **dir; 493 struct page *topdir; 494 struct page *middir; 495 struct page *subdir; 496 swp_entry_t *ptr; 497 LIST_HEAD(pages_to_free); 498 long nr_pages_to_free = 0; 499 long nr_swaps_freed = 0; 500 int offset; 501 int freed; 502 int punch_hole; 503 spinlock_t *needs_lock; 504 spinlock_t *punch_lock; 505 unsigned long upper_limit; 506 507 inode->i_ctime = inode->i_mtime = CURRENT_TIME; 508 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 509 if (idx >= info->next_index) 510 return; 511 512 spin_lock(&info->lock); 513 info->flags |= SHMEM_TRUNCATE; 514 if (likely(end == (loff_t) -1)) { 515 limit = info->next_index; 516 upper_limit = SHMEM_MAX_INDEX; 517 info->next_index = idx; 518 needs_lock = NULL; 519 punch_hole = 0; 520 } else { 521 if (end + 1 >= inode->i_size) { /* we may free a little more */ 522 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >> 523 PAGE_CACHE_SHIFT; 524 upper_limit = SHMEM_MAX_INDEX; 525 } else { 526 limit = (end + 1) >> PAGE_CACHE_SHIFT; 527 upper_limit = limit; 528 } 529 needs_lock = &info->lock; 530 punch_hole = 1; 531 } 532 533 topdir = info->i_indirect; 534 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) { 535 info->i_indirect = NULL; 536 nr_pages_to_free++; 537 list_add(&topdir->lru, &pages_to_free); 538 } 539 spin_unlock(&info->lock); 540 541 if (info->swapped && idx < SHMEM_NR_DIRECT) { 542 ptr = info->i_direct; 543 size = limit; 544 if (size > SHMEM_NR_DIRECT) 545 size = SHMEM_NR_DIRECT; 546 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock); 547 } 548 549 /* 550 * If there are no indirect blocks or we are punching a hole 551 * below indirect blocks, nothing to be done. 552 */ 553 if (!topdir || limit <= SHMEM_NR_DIRECT) 554 goto done2; 555 556 /* 557 * The truncation case has already dropped info->lock, and we're safe 558 * because i_size and next_index have already been lowered, preventing 559 * access beyond. But in the punch_hole case, we still need to take 560 * the lock when updating the swap directory, because there might be 561 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or 562 * shmem_writepage. However, whenever we find we can remove a whole 563 * directory page (not at the misaligned start or end of the range), 564 * we first NULLify its pointer in the level above, and then have no 565 * need to take the lock when updating its contents: needs_lock and 566 * punch_lock (either pointing to info->lock or NULL) manage this. 567 */ 568 569 upper_limit -= SHMEM_NR_DIRECT; 570 limit -= SHMEM_NR_DIRECT; 571 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0; 572 offset = idx % ENTRIES_PER_PAGE; 573 idx -= offset; 574 575 dir = shmem_dir_map(topdir); 576 stage = ENTRIES_PER_PAGEPAGE/2; 577 if (idx < ENTRIES_PER_PAGEPAGE/2) { 578 middir = topdir; 579 diroff = idx/ENTRIES_PER_PAGE; 580 } else { 581 dir += ENTRIES_PER_PAGE/2; 582 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE; 583 while (stage <= idx) 584 stage += ENTRIES_PER_PAGEPAGE; 585 middir = *dir; 586 if (*dir) { 587 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) % 588 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE; 589 if (!diroff && !offset && upper_limit >= stage) { 590 if (needs_lock) { 591 spin_lock(needs_lock); 592 *dir = NULL; 593 spin_unlock(needs_lock); 594 needs_lock = NULL; 595 } else 596 *dir = NULL; 597 nr_pages_to_free++; 598 list_add(&middir->lru, &pages_to_free); 599 } 600 shmem_dir_unmap(dir); 601 dir = shmem_dir_map(middir); 602 } else { 603 diroff = 0; 604 offset = 0; 605 idx = stage; 606 } 607 } 608 609 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) { 610 if (unlikely(idx == stage)) { 611 shmem_dir_unmap(dir); 612 dir = shmem_dir_map(topdir) + 613 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 614 while (!*dir) { 615 dir++; 616 idx += ENTRIES_PER_PAGEPAGE; 617 if (idx >= limit) 618 goto done1; 619 } 620 stage = idx + ENTRIES_PER_PAGEPAGE; 621 middir = *dir; 622 if (punch_hole) 623 needs_lock = &info->lock; 624 if (upper_limit >= stage) { 625 if (needs_lock) { 626 spin_lock(needs_lock); 627 *dir = NULL; 628 spin_unlock(needs_lock); 629 needs_lock = NULL; 630 } else 631 *dir = NULL; 632 nr_pages_to_free++; 633 list_add(&middir->lru, &pages_to_free); 634 } 635 shmem_dir_unmap(dir); 636 cond_resched(); 637 dir = shmem_dir_map(middir); 638 diroff = 0; 639 } 640 punch_lock = needs_lock; 641 subdir = dir[diroff]; 642 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) { 643 if (needs_lock) { 644 spin_lock(needs_lock); 645 dir[diroff] = NULL; 646 spin_unlock(needs_lock); 647 punch_lock = NULL; 648 } else 649 dir[diroff] = NULL; 650 nr_pages_to_free++; 651 list_add(&subdir->lru, &pages_to_free); 652 } 653 if (subdir && page_private(subdir) /* has swap entries */) { 654 size = limit - idx; 655 if (size > ENTRIES_PER_PAGE) 656 size = ENTRIES_PER_PAGE; 657 freed = shmem_map_and_free_swp(subdir, 658 offset, size, &dir, punch_lock); 659 if (!dir) 660 dir = shmem_dir_map(middir); 661 nr_swaps_freed += freed; 662 if (offset || punch_lock) { 663 spin_lock(&info->lock); 664 set_page_private(subdir, 665 page_private(subdir) - freed); 666 spin_unlock(&info->lock); 667 } else 668 BUG_ON(page_private(subdir) != freed); 669 } 670 offset = 0; 671 } 672done1: 673 shmem_dir_unmap(dir); 674done2: 675 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) { 676 /* 677 * Call truncate_inode_pages again: racing shmem_unuse_inode 678 * may have swizzled a page in from swap since vmtruncate or 679 * generic_delete_inode did it, before we lowered next_index. 680 * Also, though shmem_getpage checks i_size before adding to 681 * cache, no recheck after: so fix the narrow window there too. 682 * 683 * Recalling truncate_inode_pages_range and unmap_mapping_range 684 * every time for punch_hole (which never got a chance to clear 685 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive, 686 * yet hardly ever necessary: try to optimize them out later. 687 */ 688 truncate_inode_pages_range(inode->i_mapping, start, end); 689 if (punch_hole) 690 unmap_mapping_range(inode->i_mapping, start, 691 end - start, 1); 692 } 693 694 spin_lock(&info->lock); 695 info->flags &= ~SHMEM_TRUNCATE; 696 info->swapped -= nr_swaps_freed; 697 if (nr_pages_to_free) 698 shmem_free_blocks(inode, nr_pages_to_free); 699 shmem_recalc_inode(inode); 700 spin_unlock(&info->lock); 701 702 /* 703 * Empty swap vector directory pages to be freed? 704 */ 705 if (!list_empty(&pages_to_free)) { 706 pages_to_free.prev->next = NULL; 707 shmem_free_pages(pages_to_free.next); 708 } 709} 710 711static void shmem_truncate(struct inode *inode) 712{ 713 shmem_truncate_range(inode, inode->i_size, (loff_t)-1); 714} 715 716static int shmem_notify_change(struct dentry *dentry, struct iattr *attr) 717{ 718 struct inode *inode = dentry->d_inode; 719 struct page *page = NULL; 720 int error; 721 722 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { 723 if (attr->ia_size < inode->i_size) { 724 /* 725 * If truncating down to a partial page, then 726 * if that page is already allocated, hold it 727 * in memory until the truncation is over, so 728 * truncate_partial_page cannnot miss it were 729 * it assigned to swap. 730 */ 731 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) { 732 (void) shmem_getpage(inode, 733 attr->ia_size>>PAGE_CACHE_SHIFT, 734 &page, SGP_READ, NULL); 735 } 736 /* 737 * Reset SHMEM_PAGEIN flag so that shmem_truncate can 738 * detect if any pages might have been added to cache 739 * after truncate_inode_pages. But we needn't bother 740 * if it's being fully truncated to zero-length: the 741 * nrpages check is efficient enough in that case. 742 */ 743 if (attr->ia_size) { 744 struct shmem_inode_info *info = SHMEM_I(inode); 745 spin_lock(&info->lock); 746 info->flags &= ~SHMEM_PAGEIN; 747 spin_unlock(&info->lock); 748 } 749 } 750 } 751 752 error = inode_change_ok(inode, attr); 753 if (!error) 754 error = inode_setattr(inode, attr); 755#ifdef CONFIG_TMPFS_POSIX_ACL 756 if (!error && (attr->ia_valid & ATTR_MODE)) 757 error = generic_acl_chmod(inode, &shmem_acl_ops); 758#endif 759 if (page) 760 page_cache_release(page); 761 return error; 762} 763 764static void shmem_delete_inode(struct inode *inode) 765{ 766 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 767 struct shmem_inode_info *info = SHMEM_I(inode); 768 769 if (inode->i_op->truncate == shmem_truncate) { 770 truncate_inode_pages(inode->i_mapping, 0); 771 shmem_unacct_size(info->flags, inode->i_size); 772 inode->i_size = 0; 773 shmem_truncate(inode); 774 if (!list_empty(&info->swaplist)) { 775 spin_lock(&shmem_swaplist_lock); 776 list_del_init(&info->swaplist); 777 spin_unlock(&shmem_swaplist_lock); 778 } 779 } 780 BUG_ON(inode->i_blocks); 781 if (sbinfo->max_inodes) { 782 spin_lock(&sbinfo->stat_lock); 783 sbinfo->free_inodes++; 784 spin_unlock(&sbinfo->stat_lock); 785 } 786 clear_inode(inode); 787} 788 789static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir) 790{ 791 swp_entry_t *ptr; 792 793 for (ptr = dir; ptr < edir; ptr++) { 794 if (ptr->val == entry.val) 795 return ptr - dir; 796 } 797 return -1; 798} 799 800static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page) 801{ 802 struct inode *inode; 803 unsigned long idx; 804 unsigned long size; 805 unsigned long limit; 806 unsigned long stage; 807 struct page **dir; 808 struct page *subdir; 809 swp_entry_t *ptr; 810 int offset; 811 812 idx = 0; 813 ptr = info->i_direct; 814 spin_lock(&info->lock); 815 limit = info->next_index; 816 size = limit; 817 if (size > SHMEM_NR_DIRECT) 818 size = SHMEM_NR_DIRECT; 819 offset = shmem_find_swp(entry, ptr, ptr+size); 820 if (offset >= 0) { 821 shmem_swp_balance_unmap(); 822 goto found; 823 } 824 if (!info->i_indirect) 825 goto lost2; 826 827 dir = shmem_dir_map(info->i_indirect); 828 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2; 829 830 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) { 831 if (unlikely(idx == stage)) { 832 shmem_dir_unmap(dir-1); 833 dir = shmem_dir_map(info->i_indirect) + 834 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 835 while (!*dir) { 836 dir++; 837 idx += ENTRIES_PER_PAGEPAGE; 838 if (idx >= limit) 839 goto lost1; 840 } 841 stage = idx + ENTRIES_PER_PAGEPAGE; 842 subdir = *dir; 843 shmem_dir_unmap(dir); 844 dir = shmem_dir_map(subdir); 845 } 846 subdir = *dir; 847 if (subdir && page_private(subdir)) { 848 ptr = shmem_swp_map(subdir); 849 size = limit - idx; 850 if (size > ENTRIES_PER_PAGE) 851 size = ENTRIES_PER_PAGE; 852 offset = shmem_find_swp(entry, ptr, ptr+size); 853 if (offset >= 0) { 854 shmem_dir_unmap(dir); 855 goto found; 856 } 857 shmem_swp_unmap(ptr); 858 } 859 } 860lost1: 861 shmem_dir_unmap(dir-1); 862lost2: 863 spin_unlock(&info->lock); 864 return 0; 865found: 866 idx += offset; 867 inode = &info->vfs_inode; 868 if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) { 869 info->flags |= SHMEM_PAGEIN; 870 shmem_swp_set(info, ptr + offset, 0); 871 } 872 shmem_swp_unmap(ptr); 873 spin_unlock(&info->lock); 874 /* 875 * Decrement swap count even when the entry is left behind: 876 * try_to_unuse will skip over mms, then reincrement count. 877 */ 878 swap_free(entry); 879 return 1; 880} 881 882/* 883 * shmem_unuse() search for an eventually swapped out shmem page. 884 */ 885int shmem_unuse(swp_entry_t entry, struct page *page) 886{ 887 struct list_head *p, *next; 888 struct shmem_inode_info *info; 889 int found = 0; 890 891 spin_lock(&shmem_swaplist_lock); 892 list_for_each_safe(p, next, &shmem_swaplist) { 893 info = list_entry(p, struct shmem_inode_info, swaplist); 894 if (!info->swapped) 895 list_del_init(&info->swaplist); 896 else if (shmem_unuse_inode(info, entry, page)) { 897 /* move head to start search for next from here */ 898 list_move_tail(&shmem_swaplist, &info->swaplist); 899 found = 1; 900 break; 901 } 902 } 903 spin_unlock(&shmem_swaplist_lock); 904 return found; 905} 906 907/* 908 * Move the page from the page cache to the swap cache. 909 */ 910static int shmem_writepage(struct page *page, struct writeback_control *wbc) 911{ 912 struct shmem_inode_info *info; 913 swp_entry_t *entry, swap; 914 struct address_space *mapping; 915 unsigned long index; 916 struct inode *inode; 917 918 BUG_ON(!PageLocked(page)); 919 BUG_ON(page_mapped(page)); 920 921 mapping = page->mapping; 922 index = page->index; 923 inode = mapping->host; 924 info = SHMEM_I(inode); 925 if (info->flags & VM_LOCKED) 926 goto redirty; 927 swap = get_swap_page(); 928 if (!swap.val) 929 goto redirty; 930 931 spin_lock(&info->lock); 932 shmem_recalc_inode(inode); 933 if (index >= info->next_index) { 934 BUG_ON(!(info->flags & SHMEM_TRUNCATE)); 935 goto unlock; 936 } 937 entry = shmem_swp_entry(info, index, NULL); 938 BUG_ON(!entry); 939 BUG_ON(entry->val); 940 941 if (move_to_swap_cache(page, swap) == 0) { 942 shmem_swp_set(info, entry, swap.val); 943 shmem_swp_unmap(entry); 944 spin_unlock(&info->lock); 945 if (list_empty(&info->swaplist)) { 946 spin_lock(&shmem_swaplist_lock); 947 /* move instead of add in case we're racing */ 948 list_move_tail(&info->swaplist, &shmem_swaplist); 949 spin_unlock(&shmem_swaplist_lock); 950 } 951 unlock_page(page); 952 return 0; 953 } 954 955 shmem_swp_unmap(entry); 956unlock: 957 spin_unlock(&info->lock); 958 swap_free(swap); 959redirty: 960 set_page_dirty(page); 961 return AOP_WRITEPAGE_ACTIVATE; /* Return with the page locked */ 962} 963 964#ifdef CONFIG_NUMA 965static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes) 966{ 967 char *nodelist = strchr(value, ':'); 968 int err = 1; 969 970 if (nodelist) { 971 /* NUL-terminate policy string */ 972 *nodelist++ = '\0'; 973 if (nodelist_parse(nodelist, *policy_nodes)) 974 goto out; 975 if (!nodes_subset(*policy_nodes, node_online_map)) 976 goto out; 977 } 978 if (!strcmp(value, "default")) { 979 *policy = MPOL_DEFAULT; 980 /* Don't allow a nodelist */ 981 if (!nodelist) 982 err = 0; 983 } else if (!strcmp(value, "prefer")) { 984 *policy = MPOL_PREFERRED; 985 /* Insist on a nodelist of one node only */ 986 if (nodelist) { 987 char *rest = nodelist; 988 while (isdigit(*rest)) 989 rest++; 990 if (!*rest) 991 err = 0; 992 } 993 } else if (!strcmp(value, "bind")) { 994 *policy = MPOL_BIND; 995 /* Insist on a nodelist */ 996 if (nodelist) 997 err = 0; 998 } else if (!strcmp(value, "interleave")) { 999 *policy = MPOL_INTERLEAVE; 1000 /* Default to nodes online if no nodelist */ 1001 if (!nodelist) 1002 *policy_nodes = node_online_map; 1003 err = 0; 1004 } 1005out: 1006 /* Restore string for error message */ 1007 if (nodelist) 1008 *--nodelist = ':'; 1009 return err; 1010} 1011 1012static struct page *shmem_swapin_async(struct shared_policy *p, 1013 swp_entry_t entry, unsigned long idx) 1014{ 1015 struct page *page; 1016 struct vm_area_struct pvma; 1017 1018 /* Create a pseudo vma that just contains the policy */ 1019 memset(&pvma, 0, sizeof(struct vm_area_struct)); 1020 pvma.vm_end = PAGE_SIZE; 1021 pvma.vm_pgoff = idx; 1022 pvma.vm_policy = mpol_shared_policy_lookup(p, idx); 1023 page = read_swap_cache_async(entry, &pvma, 0); 1024 mpol_free(pvma.vm_policy); 1025 return page; 1026} 1027 1028struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry, 1029 unsigned long idx) 1030{ 1031 struct shared_policy *p = &info->policy; 1032 int i, num; 1033 struct page *page; 1034 unsigned long offset; 1035 1036 num = valid_swaphandles(entry, &offset); 1037 for (i = 0; i < num; offset++, i++) { 1038 page = shmem_swapin_async(p, 1039 swp_entry(swp_type(entry), offset), idx); 1040 if (!page) 1041 break; 1042 page_cache_release(page); 1043 } 1044 lru_add_drain(); /* Push any new pages onto the LRU now */ 1045 return shmem_swapin_async(p, entry, idx); 1046} 1047 1048static struct page * 1049shmem_alloc_page(gfp_t gfp, struct shmem_inode_info *info, 1050 unsigned long idx) 1051{ 1052 struct vm_area_struct pvma; 1053 struct page *page; 1054 1055 memset(&pvma, 0, sizeof(struct vm_area_struct)); 1056 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx); 1057 pvma.vm_pgoff = idx; 1058 pvma.vm_end = PAGE_SIZE; 1059 page = alloc_page_vma(gfp | __GFP_ZERO, &pvma, 0); 1060 mpol_free(pvma.vm_policy); 1061 return page; 1062} 1063#else 1064static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes) 1065{ 1066 return 1; 1067} 1068 1069static inline struct page * 1070shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx) 1071{ 1072 swapin_readahead(entry, 0, NULL); 1073 return read_swap_cache_async(entry, NULL, 0); 1074} 1075 1076static inline struct page * 1077shmem_alloc_page(gfp_t gfp,struct shmem_inode_info *info, unsigned long idx) 1078{ 1079 return alloc_page(gfp | __GFP_ZERO); 1080} 1081#endif 1082 1083/* 1084 * shmem_getpage - either get the page from swap or allocate a new one 1085 * 1086 * If we allocate a new one we do not mark it dirty. That's up to the 1087 * vm. If we swap it in we mark it dirty since we also free the swap 1088 * entry since a page cannot live in both the swap and page cache 1089 */ 1090static int shmem_getpage(struct inode *inode, unsigned long idx, 1091 struct page **pagep, enum sgp_type sgp, int *type) 1092{ 1093 struct address_space *mapping = inode->i_mapping; 1094 struct shmem_inode_info *info = SHMEM_I(inode); 1095 struct shmem_sb_info *sbinfo; 1096 struct page *filepage = *pagep; 1097 struct page *swappage; 1098 swp_entry_t *entry; 1099 swp_entry_t swap; 1100 int error; 1101 1102 if (idx >= SHMEM_MAX_INDEX) 1103 return -EFBIG; 1104 1105 if (type) 1106 *type = VM_FAULT_MINOR; 1107 1108 /* 1109 * Normally, filepage is NULL on entry, and either found 1110 * uptodate immediately, or allocated and zeroed, or read 1111 * in under swappage, which is then assigned to filepage. 1112 * But shmem_readpage and shmem_prepare_write pass in a locked 1113 * filepage, which may be found not uptodate by other callers 1114 * too, and may need to be copied from the swappage read in. 1115 */ 1116repeat: 1117 if (!filepage) 1118 filepage = find_lock_page(mapping, idx); 1119 if (filepage && PageUptodate(filepage)) 1120 goto done; 1121 error = 0; 1122 if (sgp == SGP_QUICK) 1123 goto failed; 1124 1125 spin_lock(&info->lock); 1126 shmem_recalc_inode(inode); 1127 entry = shmem_swp_alloc(info, idx, sgp); 1128 if (IS_ERR(entry)) { 1129 spin_unlock(&info->lock); 1130 error = PTR_ERR(entry); 1131 goto failed; 1132 } 1133 swap = *entry; 1134 1135 if (swap.val) { 1136 /* Look it up and read it in.. */ 1137 swappage = lookup_swap_cache(swap); 1138 if (!swappage) { 1139 shmem_swp_unmap(entry); 1140 /* here we actually do the io */ 1141 if (type && *type == VM_FAULT_MINOR) { 1142 __count_vm_event(PGMAJFAULT); 1143 *type = VM_FAULT_MAJOR; 1144 } 1145 spin_unlock(&info->lock); 1146 swappage = shmem_swapin(info, swap, idx); 1147 if (!swappage) { 1148 spin_lock(&info->lock); 1149 entry = shmem_swp_alloc(info, idx, sgp); 1150 if (IS_ERR(entry)) 1151 error = PTR_ERR(entry); 1152 else { 1153 if (entry->val == swap.val) 1154 error = -ENOMEM; 1155 shmem_swp_unmap(entry); 1156 } 1157 spin_unlock(&info->lock); 1158 if (error) 1159 goto failed; 1160 goto repeat; 1161 } 1162 wait_on_page_locked(swappage); 1163 page_cache_release(swappage); 1164 goto repeat; 1165 } 1166 1167 /* We have to do this with page locked to prevent races */ 1168 if (TestSetPageLocked(swappage)) { 1169 shmem_swp_unmap(entry); 1170 spin_unlock(&info->lock); 1171 wait_on_page_locked(swappage); 1172 page_cache_release(swappage); 1173 goto repeat; 1174 } 1175 if (PageWriteback(swappage)) { 1176 shmem_swp_unmap(entry); 1177 spin_unlock(&info->lock); 1178 wait_on_page_writeback(swappage); 1179 unlock_page(swappage); 1180 page_cache_release(swappage); 1181 goto repeat; 1182 } 1183 if (!PageUptodate(swappage)) { 1184 shmem_swp_unmap(entry); 1185 spin_unlock(&info->lock); 1186 unlock_page(swappage); 1187 page_cache_release(swappage); 1188 error = -EIO; 1189 goto failed; 1190 } 1191 1192 if (filepage) { 1193 shmem_swp_set(info, entry, 0); 1194 shmem_swp_unmap(entry); 1195 delete_from_swap_cache(swappage); 1196 spin_unlock(&info->lock); 1197 copy_highpage(filepage, swappage); 1198 unlock_page(swappage); 1199 page_cache_release(swappage); 1200 flush_dcache_page(filepage); 1201 SetPageUptodate(filepage); 1202 set_page_dirty(filepage); 1203 swap_free(swap); 1204 } else if (!(error = move_from_swap_cache( 1205 swappage, idx, mapping))) { 1206 info->flags |= SHMEM_PAGEIN; 1207 shmem_swp_set(info, entry, 0); 1208 shmem_swp_unmap(entry); 1209 spin_unlock(&info->lock); 1210 filepage = swappage; 1211 swap_free(swap); 1212 } else { 1213 shmem_swp_unmap(entry); 1214 spin_unlock(&info->lock); 1215 unlock_page(swappage); 1216 page_cache_release(swappage); 1217 if (error == -ENOMEM) { 1218 /* let kswapd refresh zone for GFP_ATOMICs */ 1219 congestion_wait(WRITE, HZ/50); 1220 } 1221 goto repeat; 1222 } 1223 } else if (sgp == SGP_READ && !filepage) { 1224 shmem_swp_unmap(entry); 1225 filepage = find_get_page(mapping, idx); 1226 if (filepage && 1227 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) { 1228 spin_unlock(&info->lock); 1229 wait_on_page_locked(filepage); 1230 page_cache_release(filepage); 1231 filepage = NULL; 1232 goto repeat; 1233 } 1234 spin_unlock(&info->lock); 1235 } else { 1236 shmem_swp_unmap(entry); 1237 sbinfo = SHMEM_SB(inode->i_sb); 1238 if (sbinfo->max_blocks) { 1239 spin_lock(&sbinfo->stat_lock); 1240 if (sbinfo->free_blocks == 0 || 1241 shmem_acct_block(info->flags)) { 1242 spin_unlock(&sbinfo->stat_lock); 1243 spin_unlock(&info->lock); 1244 error = -ENOSPC; 1245 goto failed; 1246 } 1247 sbinfo->free_blocks--; 1248 inode->i_blocks += BLOCKS_PER_PAGE; 1249 spin_unlock(&sbinfo->stat_lock); 1250 } else if (shmem_acct_block(info->flags)) { 1251 spin_unlock(&info->lock); 1252 error = -ENOSPC; 1253 goto failed; 1254 } 1255 1256 if (!filepage) { 1257 spin_unlock(&info->lock); 1258 filepage = shmem_alloc_page(mapping_gfp_mask(mapping), 1259 info, 1260 idx); 1261 if (!filepage) { 1262 shmem_unacct_blocks(info->flags, 1); 1263 shmem_free_blocks(inode, 1); 1264 error = -ENOMEM; 1265 goto failed; 1266 } 1267 1268 spin_lock(&info->lock); 1269 entry = shmem_swp_alloc(info, idx, sgp); 1270 if (IS_ERR(entry)) 1271 error = PTR_ERR(entry); 1272 else { 1273 swap = *entry; 1274 shmem_swp_unmap(entry); 1275 } 1276 if (error || swap.val || 0 != add_to_page_cache_lru( 1277 filepage, mapping, idx, GFP_ATOMIC)) { 1278 spin_unlock(&info->lock); 1279 page_cache_release(filepage); 1280 shmem_unacct_blocks(info->flags, 1); 1281 shmem_free_blocks(inode, 1); 1282 filepage = NULL; 1283 if (error) 1284 goto failed; 1285 goto repeat; 1286 } 1287 info->flags |= SHMEM_PAGEIN; 1288 } 1289 1290 info->alloced++; 1291 spin_unlock(&info->lock); 1292 flush_dcache_page(filepage); 1293 SetPageUptodate(filepage); 1294 } 1295done: 1296 if (*pagep != filepage) { 1297 *pagep = filepage; 1298 if (sgp != SGP_FAULT) 1299 unlock_page(filepage); 1300 1301 } 1302 return 0; 1303 1304failed: 1305 if (*pagep != filepage) { 1306 unlock_page(filepage); 1307 page_cache_release(filepage); 1308 } 1309 return error; 1310} 1311 1312static struct page *shmem_fault(struct vm_area_struct *vma, 1313 struct fault_data *fdata) 1314{ 1315 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 1316 struct page *page = NULL; 1317 int error; 1318 1319 BUG_ON(!(vma->vm_flags & VM_CAN_INVALIDATE)); 1320 1321 if (((loff_t)fdata->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { 1322 fdata->type = VM_FAULT_SIGBUS; 1323 return NULL; 1324 } 1325 1326 error = shmem_getpage(inode, fdata->pgoff, &page, 1327 SGP_FAULT, &fdata->type); 1328 if (error) { 1329 fdata->type = ((error == -ENOMEM)?VM_FAULT_OOM:VM_FAULT_SIGBUS); 1330 return NULL; 1331 } 1332 1333 mark_page_accessed(page); 1334 return page; 1335} 1336 1337#ifdef CONFIG_NUMA 1338int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new) 1339{ 1340 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1341 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new); 1342} 1343 1344struct mempolicy * 1345shmem_get_policy(struct vm_area_struct *vma, unsigned long addr) 1346{ 1347 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1348 unsigned long idx; 1349 1350 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; 1351 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx); 1352} 1353#endif 1354 1355int shmem_lock(struct file *file, int lock, struct user_struct *user) 1356{ 1357 struct inode *inode = file->f_path.dentry->d_inode; 1358 struct shmem_inode_info *info = SHMEM_I(inode); 1359 int retval = -ENOMEM; 1360 1361 spin_lock(&info->lock); 1362 if (lock && !(info->flags & VM_LOCKED)) { 1363 if (!user_shm_lock(inode->i_size, user)) 1364 goto out_nomem; 1365 info->flags |= VM_LOCKED; 1366 } 1367 if (!lock && (info->flags & VM_LOCKED) && user) { 1368 user_shm_unlock(inode->i_size, user); 1369 info->flags &= ~VM_LOCKED; 1370 } 1371 retval = 0; 1372out_nomem: 1373 spin_unlock(&info->lock); 1374 return retval; 1375} 1376 1377static int shmem_mmap(struct file *file, struct vm_area_struct *vma) 1378{ 1379 file_accessed(file); 1380 vma->vm_ops = &shmem_vm_ops; 1381 vma->vm_flags |= VM_CAN_INVALIDATE | VM_CAN_NONLINEAR; 1382 return 0; 1383} 1384 1385static struct inode * 1386shmem_get_inode(struct super_block *sb, int mode, dev_t dev) 1387{ 1388 struct inode *inode; 1389 struct shmem_inode_info *info; 1390 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 1391 1392 if (sbinfo->max_inodes) { 1393 spin_lock(&sbinfo->stat_lock); 1394 if (!sbinfo->free_inodes) { 1395 spin_unlock(&sbinfo->stat_lock); 1396 return NULL; 1397 } 1398 sbinfo->free_inodes--; 1399 spin_unlock(&sbinfo->stat_lock); 1400 } 1401 1402 inode = new_inode(sb); 1403 if (inode) { 1404 inode->i_mode = mode; 1405 inode->i_uid = current->fsuid; 1406 inode->i_gid = current->fsgid; 1407 inode->i_blocks = 0; 1408 inode->i_mapping->a_ops = &shmem_aops; 1409 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; 1410 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 1411 inode->i_generation = get_seconds(); 1412 info = SHMEM_I(inode); 1413 memset(info, 0, (char *)inode - (char *)info); 1414 spin_lock_init(&info->lock); 1415 INIT_LIST_HEAD(&info->swaplist); 1416 1417 switch (mode & S_IFMT) { 1418 default: 1419 inode->i_op = &shmem_special_inode_operations; 1420 init_special_inode(inode, mode, dev); 1421 break; 1422 case S_IFREG: 1423 inode->i_op = &shmem_inode_operations; 1424 inode->i_fop = &shmem_file_operations; 1425 mpol_shared_policy_init(&info->policy, sbinfo->policy, 1426 &sbinfo->policy_nodes); 1427 break; 1428 case S_IFDIR: 1429 inc_nlink(inode); 1430 /* Some things misbehave if size == 0 on a directory */ 1431 inode->i_size = 2 * BOGO_DIRENT_SIZE; 1432 inode->i_op = &shmem_dir_inode_operations; 1433 inode->i_fop = &simple_dir_operations; 1434 break; 1435 case S_IFLNK: 1436 /* 1437 * Must not load anything in the rbtree, 1438 * mpol_free_shared_policy will not be called. 1439 */ 1440 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, 1441 NULL); 1442 break; 1443 } 1444 } else if (sbinfo->max_inodes) { 1445 spin_lock(&sbinfo->stat_lock); 1446 sbinfo->free_inodes++; 1447 spin_unlock(&sbinfo->stat_lock); 1448 } 1449 return inode; 1450} 1451 1452#ifdef CONFIG_TMPFS 1453static const struct inode_operations shmem_symlink_inode_operations; 1454static const struct inode_operations shmem_symlink_inline_operations; 1455 1456/* 1457 * Normally tmpfs avoids the use of shmem_readpage and shmem_prepare_write; 1458 * but providing them allows a tmpfs file to be used for splice, sendfile, and 1459 * below the loop driver, in the generic fashion that many filesystems support. 1460 */ 1461static int shmem_readpage(struct file *file, struct page *page) 1462{ 1463 struct inode *inode = page->mapping->host; 1464 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL); 1465 unlock_page(page); 1466 return error; 1467} 1468 1469static int 1470shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to) 1471{ 1472 struct inode *inode = page->mapping->host; 1473 return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL); 1474} 1475 1476static ssize_t 1477shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos) 1478{ 1479 struct inode *inode = file->f_path.dentry->d_inode; 1480 loff_t pos; 1481 unsigned long written; 1482 ssize_t err; 1483 1484 if ((ssize_t) count < 0) 1485 return -EINVAL; 1486 1487 if (!access_ok(VERIFY_READ, buf, count)) 1488 return -EFAULT; 1489 1490 mutex_lock(&inode->i_mutex); 1491 1492 pos = *ppos; 1493 written = 0; 1494 1495 err = generic_write_checks(file, &pos, &count, 0); 1496 if (err || !count) 1497 goto out; 1498 1499 err = remove_suid(file->f_path.dentry); 1500 if (err) 1501 goto out; 1502 1503 inode->i_ctime = inode->i_mtime = CURRENT_TIME; 1504 1505 do { 1506 struct page *page = NULL; 1507 unsigned long bytes, index, offset; 1508 char *kaddr; 1509 int left; 1510 1511 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */ 1512 index = pos >> PAGE_CACHE_SHIFT; 1513 bytes = PAGE_CACHE_SIZE - offset; 1514 if (bytes > count) 1515 bytes = count; 1516 1517 /* 1518 * We don't hold page lock across copy from user - 1519 * what would it guard against? - so no deadlock here. 1520 * But it still may be a good idea to prefault below. 1521 */ 1522 1523 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL); 1524 if (err) 1525 break; 1526 1527 left = bytes; 1528 if (PageHighMem(page)) { 1529 volatile unsigned char dummy; 1530 __get_user(dummy, buf); 1531 __get_user(dummy, buf + bytes - 1); 1532 1533 kaddr = kmap_atomic(page, KM_USER0); 1534 left = __copy_from_user_inatomic(kaddr + offset, 1535 buf, bytes); 1536 kunmap_atomic(kaddr, KM_USER0); 1537 } 1538 if (left) { 1539 kaddr = kmap(page); 1540 left = __copy_from_user(kaddr + offset, buf, bytes); 1541 kunmap(page); 1542 } 1543 1544 written += bytes; 1545 count -= bytes; 1546 pos += bytes; 1547 buf += bytes; 1548 if (pos > inode->i_size) 1549 i_size_write(inode, pos); 1550 1551 flush_dcache_page(page); 1552 set_page_dirty(page); 1553 mark_page_accessed(page); 1554 page_cache_release(page); 1555 1556 if (left) { 1557 pos -= left; 1558 written -= left; 1559 err = -EFAULT; 1560 break; 1561 } 1562 1563 /* 1564 * Our dirty pages are not counted in nr_dirty, 1565 * and we do not attempt to balance dirty pages. 1566 */ 1567 1568 cond_resched(); 1569 } while (count); 1570 1571 *ppos = pos; 1572 if (written) 1573 err = written; 1574out: 1575 mutex_unlock(&inode->i_mutex); 1576 return err; 1577} 1578 1579static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor) 1580{ 1581 struct inode *inode = filp->f_path.dentry->d_inode; 1582 struct address_space *mapping = inode->i_mapping; 1583 unsigned long index, offset; 1584 1585 index = *ppos >> PAGE_CACHE_SHIFT; 1586 offset = *ppos & ~PAGE_CACHE_MASK; 1587 1588 for (;;) { 1589 struct page *page = NULL; 1590 unsigned long end_index, nr, ret; 1591 loff_t i_size = i_size_read(inode); 1592 1593 end_index = i_size >> PAGE_CACHE_SHIFT; 1594 if (index > end_index) 1595 break; 1596 if (index == end_index) { 1597 nr = i_size & ~PAGE_CACHE_MASK; 1598 if (nr <= offset) 1599 break; 1600 } 1601 1602 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL); 1603 if (desc->error) { 1604 if (desc->error == -EINVAL) 1605 desc->error = 0; 1606 break; 1607 } 1608 1609 /* 1610 * We must evaluate after, since reads (unlike writes) 1611 * are called without i_mutex protection against truncate 1612 */ 1613 nr = PAGE_CACHE_SIZE; 1614 i_size = i_size_read(inode); 1615 end_index = i_size >> PAGE_CACHE_SHIFT; 1616 if (index == end_index) { 1617 nr = i_size & ~PAGE_CACHE_MASK; 1618 if (nr <= offset) { 1619 if (page) 1620 page_cache_release(page); 1621 break; 1622 } 1623 } 1624 nr -= offset; 1625 1626 if (page) { 1627 /* 1628 * If users can be writing to this page using arbitrary 1629 * virtual addresses, take care about potential aliasing 1630 * before reading the page on the kernel side. 1631 */ 1632 if (mapping_writably_mapped(mapping)) 1633 flush_dcache_page(page); 1634 /* 1635 * Mark the page accessed if we read the beginning. 1636 */ 1637 if (!offset) 1638 mark_page_accessed(page); 1639 } else { 1640 page = ZERO_PAGE(0); 1641 page_cache_get(page); 1642 } 1643 1644 /* 1645 * Ok, we have the page, and it's up-to-date, so 1646 * now we can copy it to user space... 1647 * 1648 * The actor routine returns how many bytes were actually used.. 1649 * NOTE! This may not be the same as how much of a user buffer 1650 * we filled up (we may be padding etc), so we can only update 1651 * "pos" here (the actor routine has to update the user buffer 1652 * pointers and the remaining count). 1653 */ 1654 ret = actor(desc, page, offset, nr); 1655 offset += ret; 1656 index += offset >> PAGE_CACHE_SHIFT; 1657 offset &= ~PAGE_CACHE_MASK; 1658 1659 page_cache_release(page); 1660 if (ret != nr || !desc->count) 1661 break; 1662 1663 cond_resched(); 1664 } 1665 1666 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; 1667 file_accessed(filp); 1668} 1669 1670static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos) 1671{ 1672 read_descriptor_t desc; 1673 1674 if ((ssize_t) count < 0) 1675 return -EINVAL; 1676 if (!access_ok(VERIFY_WRITE, buf, count)) 1677 return -EFAULT; 1678 if (!count) 1679 return 0; 1680 1681 desc.written = 0; 1682 desc.count = count; 1683 desc.arg.buf = buf; 1684 desc.error = 0; 1685 1686 do_shmem_file_read(filp, ppos, &desc, file_read_actor); 1687 if (desc.written) 1688 return desc.written; 1689 return desc.error; 1690} 1691 1692static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) 1693{ 1694 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); 1695 1696 buf->f_type = TMPFS_MAGIC; 1697 buf->f_bsize = PAGE_CACHE_SIZE; 1698 buf->f_namelen = NAME_MAX; 1699 spin_lock(&sbinfo->stat_lock); 1700 if (sbinfo->max_blocks) { 1701 buf->f_blocks = sbinfo->max_blocks; 1702 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks; 1703 } 1704 if (sbinfo->max_inodes) { 1705 buf->f_files = sbinfo->max_inodes; 1706 buf->f_ffree = sbinfo->free_inodes; 1707 } 1708 /* else leave those fields 0 like simple_statfs */ 1709 spin_unlock(&sbinfo->stat_lock); 1710 return 0; 1711} 1712 1713/* 1714 * File creation. Allocate an inode, and we're done.. 1715 */ 1716static int 1717shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 1718{ 1719 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev); 1720 int error = -ENOSPC; 1721 1722 if (inode) { 1723 error = security_inode_init_security(inode, dir, NULL, NULL, 1724 NULL); 1725 if (error) { 1726 if (error != -EOPNOTSUPP) { 1727 iput(inode); 1728 return error; 1729 } 1730 } 1731 error = shmem_acl_init(inode, dir); 1732 if (error) { 1733 iput(inode); 1734 return error; 1735 } 1736 if (dir->i_mode & S_ISGID) { 1737 inode->i_gid = dir->i_gid; 1738 if (S_ISDIR(mode)) 1739 inode->i_mode |= S_ISGID; 1740 } 1741 dir->i_size += BOGO_DIRENT_SIZE; 1742 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1743 d_instantiate(dentry, inode); 1744 dget(dentry); /* Extra count - pin the dentry in core */ 1745 } 1746 return error; 1747} 1748 1749static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1750{ 1751 int error; 1752 1753 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) 1754 return error; 1755 inc_nlink(dir); 1756 return 0; 1757} 1758 1759static int shmem_create(struct inode *dir, struct dentry *dentry, int mode, 1760 struct nameidata *nd) 1761{ 1762 return shmem_mknod(dir, dentry, mode | S_IFREG, 0); 1763} 1764 1765/* 1766 * Link a file.. 1767 */ 1768static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1769{ 1770 struct inode *inode = old_dentry->d_inode; 1771 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 1772 1773 /* 1774 * No ordinary (disk based) filesystem counts links as inodes; 1775 * but each new link needs a new dentry, pinning lowmem, and 1776 * tmpfs dentries cannot be pruned until they are unlinked. 1777 */ 1778 if (sbinfo->max_inodes) { 1779 spin_lock(&sbinfo->stat_lock); 1780 if (!sbinfo->free_inodes) { 1781 spin_unlock(&sbinfo->stat_lock); 1782 return -ENOSPC; 1783 } 1784 sbinfo->free_inodes--; 1785 spin_unlock(&sbinfo->stat_lock); 1786 } 1787 1788 dir->i_size += BOGO_DIRENT_SIZE; 1789 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1790 inc_nlink(inode); 1791 atomic_inc(&inode->i_count); /* New dentry reference */ 1792 dget(dentry); /* Extra pinning count for the created dentry */ 1793 d_instantiate(dentry, inode); 1794 return 0; 1795} 1796 1797static int shmem_unlink(struct inode *dir, struct dentry *dentry) 1798{ 1799 struct inode *inode = dentry->d_inode; 1800 1801 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) { 1802 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 1803 if (sbinfo->max_inodes) { 1804 spin_lock(&sbinfo->stat_lock); 1805 sbinfo->free_inodes++; 1806 spin_unlock(&sbinfo->stat_lock); 1807 } 1808 } 1809 1810 dir->i_size -= BOGO_DIRENT_SIZE; 1811 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1812 drop_nlink(inode); 1813 dput(dentry); /* Undo the count from "create" - this does all the work */ 1814 return 0; 1815} 1816 1817static int shmem_rmdir(struct inode *dir, struct dentry *dentry) 1818{ 1819 if (!simple_empty(dentry)) 1820 return -ENOTEMPTY; 1821 1822 drop_nlink(dentry->d_inode); 1823 drop_nlink(dir); 1824 return shmem_unlink(dir, dentry); 1825} 1826 1827/* 1828 * The VFS layer already does all the dentry stuff for rename, 1829 * we just have to decrement the usage count for the target if 1830 * it exists so that the VFS layer correctly free's it when it 1831 * gets overwritten. 1832 */ 1833static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) 1834{ 1835 struct inode *inode = old_dentry->d_inode; 1836 int they_are_dirs = S_ISDIR(inode->i_mode); 1837 1838 if (!simple_empty(new_dentry)) 1839 return -ENOTEMPTY; 1840 1841 if (new_dentry->d_inode) { 1842 (void) shmem_unlink(new_dir, new_dentry); 1843 if (they_are_dirs) 1844 drop_nlink(old_dir); 1845 } else if (they_are_dirs) { 1846 drop_nlink(old_dir); 1847 inc_nlink(new_dir); 1848 } 1849 1850 old_dir->i_size -= BOGO_DIRENT_SIZE; 1851 new_dir->i_size += BOGO_DIRENT_SIZE; 1852 old_dir->i_ctime = old_dir->i_mtime = 1853 new_dir->i_ctime = new_dir->i_mtime = 1854 inode->i_ctime = CURRENT_TIME; 1855 return 0; 1856} 1857 1858static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1859{ 1860 int error; 1861 int len; 1862 struct inode *inode; 1863 struct page *page = NULL; 1864 char *kaddr; 1865 struct shmem_inode_info *info; 1866 1867 len = strlen(symname) + 1; 1868 if (len > PAGE_CACHE_SIZE) 1869 return -ENAMETOOLONG; 1870 1871 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0); 1872 if (!inode) 1873 return -ENOSPC; 1874 1875 error = security_inode_init_security(inode, dir, NULL, NULL, 1876 NULL); 1877 if (error) { 1878 if (error != -EOPNOTSUPP) { 1879 iput(inode); 1880 return error; 1881 } 1882 error = 0; 1883 } 1884 1885 info = SHMEM_I(inode); 1886 inode->i_size = len-1; 1887 if (len <= (char *)inode - (char *)info) { 1888 /* do it inline */ 1889 memcpy(info, symname, len); 1890 inode->i_op = &shmem_symlink_inline_operations; 1891 } else { 1892 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); 1893 if (error) { 1894 iput(inode); 1895 return error; 1896 } 1897 inode->i_op = &shmem_symlink_inode_operations; 1898 kaddr = kmap_atomic(page, KM_USER0); 1899 memcpy(kaddr, symname, len); 1900 kunmap_atomic(kaddr, KM_USER0); 1901 set_page_dirty(page); 1902 page_cache_release(page); 1903 } 1904 if (dir->i_mode & S_ISGID) 1905 inode->i_gid = dir->i_gid; 1906 dir->i_size += BOGO_DIRENT_SIZE; 1907 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1908 d_instantiate(dentry, inode); 1909 dget(dentry); 1910 return 0; 1911} 1912 1913static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd) 1914{ 1915 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode)); 1916 return NULL; 1917} 1918 1919static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) 1920{ 1921 struct page *page = NULL; 1922 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); 1923 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page)); 1924 return page; 1925} 1926 1927static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 1928{ 1929 if (!IS_ERR(nd_get_link(nd))) { 1930 struct page *page = cookie; 1931 kunmap(page); 1932 mark_page_accessed(page); 1933 page_cache_release(page); 1934 } 1935} 1936 1937static const struct inode_operations shmem_symlink_inline_operations = { 1938 .readlink = generic_readlink, 1939 .follow_link = shmem_follow_link_inline, 1940}; 1941 1942static const struct inode_operations shmem_symlink_inode_operations = { 1943 .truncate = shmem_truncate, 1944 .readlink = generic_readlink, 1945 .follow_link = shmem_follow_link, 1946 .put_link = shmem_put_link, 1947}; 1948 1949#ifdef CONFIG_TMPFS_POSIX_ACL 1950/** 1951 * Superblocks without xattr inode operations will get security.* xattr 1952 * support from the VFS "for free". As soon as we have any other xattrs 1953 * like ACLs, we also need to implement the security.* handlers at 1954 * filesystem level, though. 1955 */ 1956 1957static size_t shmem_xattr_security_list(struct inode *inode, char *list, 1958 size_t list_len, const char *name, 1959 size_t name_len) 1960{ 1961 return security_inode_listsecurity(inode, list, list_len); 1962} 1963 1964static int shmem_xattr_security_get(struct inode *inode, const char *name, 1965 void *buffer, size_t size) 1966{ 1967 if (strcmp(name, "") == 0) 1968 return -EINVAL; 1969 return security_inode_getsecurity(inode, name, buffer, size, 1970 -EOPNOTSUPP); 1971} 1972 1973static int shmem_xattr_security_set(struct inode *inode, const char *name, 1974 const void *value, size_t size, int flags) 1975{ 1976 if (strcmp(name, "") == 0) 1977 return -EINVAL; 1978 return security_inode_setsecurity(inode, name, value, size, flags); 1979} 1980 1981static struct xattr_handler shmem_xattr_security_handler = { 1982 .prefix = XATTR_SECURITY_PREFIX, 1983 .list = shmem_xattr_security_list, 1984 .get = shmem_xattr_security_get, 1985 .set = shmem_xattr_security_set, 1986}; 1987 1988static struct xattr_handler *shmem_xattr_handlers[] = { 1989 &shmem_xattr_acl_access_handler, 1990 &shmem_xattr_acl_default_handler, 1991 &shmem_xattr_security_handler, 1992 NULL 1993}; 1994#endif 1995 1996static struct dentry *shmem_get_parent(struct dentry *child) 1997{ 1998 return ERR_PTR(-ESTALE); 1999} 2000 2001static int shmem_match(struct inode *ino, void *vfh) 2002{ 2003 __u32 *fh = vfh; 2004 __u64 inum = fh[2]; 2005 inum = (inum << 32) | fh[1]; 2006 return ino->i_ino == inum && fh[0] == ino->i_generation; 2007} 2008 2009static struct dentry *shmem_get_dentry(struct super_block *sb, void *vfh) 2010{ 2011 struct dentry *de = NULL; 2012 struct inode *inode; 2013 __u32 *fh = vfh; 2014 __u64 inum = fh[2]; 2015 inum = (inum << 32) | fh[1]; 2016 2017 inode = ilookup5(sb, (unsigned long)(inum+fh[0]), shmem_match, vfh); 2018 if (inode) { 2019 de = d_find_alias(inode); 2020 iput(inode); 2021 } 2022 2023 return de? de: ERR_PTR(-ESTALE); 2024} 2025 2026static struct dentry *shmem_decode_fh(struct super_block *sb, __u32 *fh, 2027 int len, int type, 2028 int (*acceptable)(void *context, struct dentry *de), 2029 void *context) 2030{ 2031 if (len < 3) 2032 return ERR_PTR(-ESTALE); 2033 2034 return sb->s_export_op->find_exported_dentry(sb, fh, NULL, acceptable, 2035 context); 2036} 2037 2038static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len, 2039 int connectable) 2040{ 2041 struct inode *inode = dentry->d_inode; 2042 2043 if (*len < 3) 2044 return 255; 2045 2046 if (hlist_unhashed(&inode->i_hash)) { 2047 /* Unfortunately insert_inode_hash is not idempotent, 2048 * so as we hash inodes here rather than at creation 2049 * time, we need a lock to ensure we only try 2050 * to do it once 2051 */ 2052 static DEFINE_SPINLOCK(lock); 2053 spin_lock(&lock); 2054 if (hlist_unhashed(&inode->i_hash)) 2055 __insert_inode_hash(inode, 2056 inode->i_ino + inode->i_generation); 2057 spin_unlock(&lock); 2058 } 2059 2060 fh[0] = inode->i_generation; 2061 fh[1] = inode->i_ino; 2062 fh[2] = ((__u64)inode->i_ino) >> 32; 2063 2064 *len = 3; 2065 return 1; 2066} 2067 2068static struct export_operations shmem_export_ops = { 2069 .get_parent = shmem_get_parent, 2070 .get_dentry = shmem_get_dentry, 2071 .encode_fh = shmem_encode_fh, 2072 .decode_fh = shmem_decode_fh, 2073}; 2074 2075static int shmem_parse_options(char *options, int *mode, uid_t *uid, 2076 gid_t *gid, unsigned long *blocks, unsigned long *inodes, 2077 int *policy, nodemask_t *policy_nodes) 2078{ 2079 char *this_char, *value, *rest; 2080 2081 while (options != NULL) { 2082 this_char = options; 2083 for (;;) { 2084 /* 2085 * NUL-terminate this option: unfortunately, 2086 * mount options form a comma-separated list, 2087 * but mpol's nodelist may also contain commas. 2088 */ 2089 options = strchr(options, ','); 2090 if (options == NULL) 2091 break; 2092 options++; 2093 if (!isdigit(*options)) { 2094 options[-1] = '\0'; 2095 break; 2096 } 2097 } 2098 if (!*this_char) 2099 continue; 2100 if ((value = strchr(this_char,'=')) != NULL) { 2101 *value++ = 0; 2102 } else { 2103 printk(KERN_ERR 2104 "tmpfs: No value for mount option '%s'\n", 2105 this_char); 2106 return 1; 2107 } 2108 2109 if (!strcmp(this_char,"size")) { 2110 unsigned long long size; 2111 size = memparse(value,&rest); 2112 if (*rest == '%') { 2113 size <<= PAGE_SHIFT; 2114 size *= totalram_pages; 2115 do_div(size, 100); 2116 rest++; 2117 } 2118 if (*rest) 2119 goto bad_val; 2120 *blocks = size >> PAGE_CACHE_SHIFT; 2121 } else if (!strcmp(this_char,"nr_blocks")) { 2122 *blocks = memparse(value,&rest); 2123 if (*rest) 2124 goto bad_val; 2125 } else if (!strcmp(this_char,"nr_inodes")) { 2126 *inodes = memparse(value,&rest); 2127 if (*rest) 2128 goto bad_val; 2129 } else if (!strcmp(this_char,"mode")) { 2130 if (!mode) 2131 continue; 2132 *mode = simple_strtoul(value,&rest,8); 2133 if (*rest) 2134 goto bad_val; 2135 } else if (!strcmp(this_char,"uid")) { 2136 if (!uid) 2137 continue; 2138 *uid = simple_strtoul(value,&rest,0); 2139 if (*rest) 2140 goto bad_val; 2141 } else if (!strcmp(this_char,"gid")) { 2142 if (!gid) 2143 continue; 2144 *gid = simple_strtoul(value,&rest,0); 2145 if (*rest) 2146 goto bad_val; 2147 } else if (!strcmp(this_char,"mpol")) { 2148 if (shmem_parse_mpol(value,policy,policy_nodes)) 2149 goto bad_val; 2150 } else { 2151 printk(KERN_ERR "tmpfs: Bad mount option %s\n", 2152 this_char); 2153 return 1; 2154 } 2155 } 2156 return 0; 2157 2158bad_val: 2159 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", 2160 value, this_char); 2161 return 1; 2162 2163} 2164 2165static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) 2166{ 2167 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 2168 unsigned long max_blocks = sbinfo->max_blocks; 2169 unsigned long max_inodes = sbinfo->max_inodes; 2170 int policy = sbinfo->policy; 2171 nodemask_t policy_nodes = sbinfo->policy_nodes; 2172 unsigned long blocks; 2173 unsigned long inodes; 2174 int error = -EINVAL; 2175 2176 if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks, 2177 &max_inodes, &policy, &policy_nodes)) 2178 return error; 2179 2180 spin_lock(&sbinfo->stat_lock); 2181 blocks = sbinfo->max_blocks - sbinfo->free_blocks; 2182 inodes = sbinfo->max_inodes - sbinfo->free_inodes; 2183 if (max_blocks < blocks) 2184 goto out; 2185 if (max_inodes < inodes) 2186 goto out; 2187 /* 2188 * Those tests also disallow limited->unlimited while any are in 2189 * use, so i_blocks will always be zero when max_blocks is zero; 2190 * but we must separately disallow unlimited->limited, because 2191 * in that case we have no record of how much is already in use. 2192 */ 2193 if (max_blocks && !sbinfo->max_blocks) 2194 goto out; 2195 if (max_inodes && !sbinfo->max_inodes) 2196 goto out; 2197 2198 error = 0; 2199 sbinfo->max_blocks = max_blocks; 2200 sbinfo->free_blocks = max_blocks - blocks; 2201 sbinfo->max_inodes = max_inodes; 2202 sbinfo->free_inodes = max_inodes - inodes; 2203 sbinfo->policy = policy; 2204 sbinfo->policy_nodes = policy_nodes; 2205out: 2206 spin_unlock(&sbinfo->stat_lock); 2207 return error; 2208} 2209#endif 2210 2211static void shmem_put_super(struct super_block *sb) 2212{ 2213 kfree(sb->s_fs_info); 2214 sb->s_fs_info = NULL; 2215} 2216 2217static int shmem_fill_super(struct super_block *sb, 2218 void *data, int silent) 2219{ 2220 struct inode *inode; 2221 struct dentry *root; 2222 int mode = S_IRWXUGO | S_ISVTX; 2223 uid_t uid = current->fsuid; 2224 gid_t gid = current->fsgid; 2225 int err = -ENOMEM; 2226 struct shmem_sb_info *sbinfo; 2227 unsigned long blocks = 0; 2228 unsigned long inodes = 0; 2229 int policy = MPOL_DEFAULT; 2230 nodemask_t policy_nodes = node_online_map; 2231 2232#ifdef CONFIG_TMPFS 2233 /* 2234 * Per default we only allow half of the physical ram per 2235 * tmpfs instance, limiting inodes to one per page of lowmem; 2236 * but the internal instance is left unlimited. 2237 */ 2238 if (!(sb->s_flags & MS_NOUSER)) { 2239 blocks = totalram_pages / 2; 2240 inodes = totalram_pages - totalhigh_pages; 2241 if (inodes > blocks) 2242 inodes = blocks; 2243 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks, 2244 &inodes, &policy, &policy_nodes)) 2245 return -EINVAL; 2246 } 2247 sb->s_export_op = &shmem_export_ops; 2248#else 2249 sb->s_flags |= MS_NOUSER; 2250#endif 2251 2252 /* Round up to L1_CACHE_BYTES to resist false sharing */ 2253 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info), 2254 L1_CACHE_BYTES), GFP_KERNEL); 2255 if (!sbinfo) 2256 return -ENOMEM; 2257 2258 spin_lock_init(&sbinfo->stat_lock); 2259 sbinfo->max_blocks = blocks; 2260 sbinfo->free_blocks = blocks; 2261 sbinfo->max_inodes = inodes; 2262 sbinfo->free_inodes = inodes; 2263 sbinfo->policy = policy; 2264 sbinfo->policy_nodes = policy_nodes; 2265 2266 sb->s_fs_info = sbinfo; 2267 sb->s_maxbytes = SHMEM_MAX_BYTES; 2268 sb->s_blocksize = PAGE_CACHE_SIZE; 2269 sb->s_blocksize_bits = PAGE_CACHE_SHIFT; 2270 sb->s_magic = TMPFS_MAGIC; 2271 sb->s_op = &shmem_ops; 2272 sb->s_time_gran = 1; 2273#ifdef CONFIG_TMPFS_POSIX_ACL 2274 sb->s_xattr = shmem_xattr_handlers; 2275 sb->s_flags |= MS_POSIXACL; 2276#endif 2277 2278 inode = shmem_get_inode(sb, S_IFDIR | mode, 0); 2279 if (!inode) 2280 goto failed; 2281 inode->i_uid = uid; 2282 inode->i_gid = gid; 2283 root = d_alloc_root(inode); 2284 if (!root) 2285 goto failed_iput; 2286 sb->s_root = root; 2287 return 0; 2288 2289failed_iput: 2290 iput(inode); 2291failed: 2292 shmem_put_super(sb); 2293 return err; 2294} 2295 2296static struct kmem_cache *shmem_inode_cachep; 2297 2298static struct inode *shmem_alloc_inode(struct super_block *sb) 2299{ 2300 struct shmem_inode_info *p; 2301 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); 2302 if (!p) 2303 return NULL; 2304 return &p->vfs_inode; 2305} 2306 2307static void shmem_destroy_inode(struct inode *inode) 2308{ 2309 if ((inode->i_mode & S_IFMT) == S_IFREG) { 2310 /* only struct inode is valid if it's an inline symlink */ 2311 mpol_free_shared_policy(&SHMEM_I(inode)->policy); 2312 } 2313 shmem_acl_destroy_inode(inode); 2314 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); 2315} 2316 2317static void init_once(void *foo, struct kmem_cache *cachep, 2318 unsigned long flags) 2319{ 2320 struct shmem_inode_info *p = (struct shmem_inode_info *) foo; 2321 2322 inode_init_once(&p->vfs_inode); 2323#ifdef CONFIG_TMPFS_POSIX_ACL 2324 p->i_acl = NULL; 2325 p->i_default_acl = NULL; 2326#endif 2327} 2328 2329static int init_inodecache(void) 2330{ 2331 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", 2332 sizeof(struct shmem_inode_info), 2333 0, 0, init_once, NULL); 2334 if (shmem_inode_cachep == NULL) 2335 return -ENOMEM; 2336 return 0; 2337} 2338 2339static void destroy_inodecache(void) 2340{ 2341 kmem_cache_destroy(shmem_inode_cachep); 2342} 2343 2344static const struct address_space_operations shmem_aops = { 2345 .writepage = shmem_writepage, 2346 .set_page_dirty = __set_page_dirty_no_writeback, 2347#ifdef CONFIG_TMPFS 2348 .readpage = shmem_readpage, 2349 .prepare_write = shmem_prepare_write, 2350 .commit_write = simple_commit_write, 2351#endif 2352 .migratepage = migrate_page, 2353}; 2354 2355static const struct file_operations shmem_file_operations = { 2356 .mmap = shmem_mmap, 2357#ifdef CONFIG_TMPFS 2358 .llseek = generic_file_llseek, 2359 .read = shmem_file_read, 2360 .write = shmem_file_write, 2361 .fsync = simple_sync_file, 2362 .splice_read = generic_file_splice_read, 2363 .splice_write = generic_file_splice_write, 2364#endif 2365}; 2366 2367static const struct inode_operations shmem_inode_operations = { 2368 .truncate = shmem_truncate, 2369 .setattr = shmem_notify_change, 2370 .truncate_range = shmem_truncate_range, 2371#ifdef CONFIG_TMPFS_POSIX_ACL 2372 .setxattr = generic_setxattr, 2373 .getxattr = generic_getxattr, 2374 .listxattr = generic_listxattr, 2375 .removexattr = generic_removexattr, 2376 .permission = shmem_permission, 2377#endif 2378 2379}; 2380 2381static const struct inode_operations shmem_dir_inode_operations = { 2382#ifdef CONFIG_TMPFS 2383 .create = shmem_create, 2384 .lookup = simple_lookup, 2385 .link = shmem_link, 2386 .unlink = shmem_unlink, 2387 .symlink = shmem_symlink, 2388 .mkdir = shmem_mkdir, 2389 .rmdir = shmem_rmdir, 2390 .mknod = shmem_mknod, 2391 .rename = shmem_rename, 2392#endif 2393#ifdef CONFIG_TMPFS_POSIX_ACL 2394 .setattr = shmem_notify_change, 2395 .setxattr = generic_setxattr, 2396 .getxattr = generic_getxattr, 2397 .listxattr = generic_listxattr, 2398 .removexattr = generic_removexattr, 2399 .permission = shmem_permission, 2400#endif 2401}; 2402 2403static const struct inode_operations shmem_special_inode_operations = { 2404#ifdef CONFIG_TMPFS_POSIX_ACL 2405 .setattr = shmem_notify_change, 2406 .setxattr = generic_setxattr, 2407 .getxattr = generic_getxattr, 2408 .listxattr = generic_listxattr, 2409 .removexattr = generic_removexattr, 2410 .permission = shmem_permission, 2411#endif 2412}; 2413 2414static const struct super_operations shmem_ops = { 2415 .alloc_inode = shmem_alloc_inode, 2416 .destroy_inode = shmem_destroy_inode, 2417#ifdef CONFIG_TMPFS 2418 .statfs = shmem_statfs, 2419 .remount_fs = shmem_remount_fs, 2420#endif 2421 .delete_inode = shmem_delete_inode, 2422 .drop_inode = generic_delete_inode, 2423 .put_super = shmem_put_super, 2424}; 2425 2426static struct vm_operations_struct shmem_vm_ops = { 2427 .fault = shmem_fault, 2428#ifdef CONFIG_NUMA 2429 .set_policy = shmem_set_policy, 2430 .get_policy = shmem_get_policy, 2431#endif 2432}; 2433 2434 2435static int shmem_get_sb(struct file_system_type *fs_type, 2436 int flags, const char *dev_name, void *data, struct vfsmount *mnt) 2437{ 2438 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt); 2439} 2440 2441static struct file_system_type tmpfs_fs_type = { 2442 .owner = THIS_MODULE, 2443 .name = "tmpfs", 2444 .get_sb = shmem_get_sb, 2445 .kill_sb = kill_litter_super, 2446}; 2447static struct vfsmount *shm_mnt; 2448 2449static int __init init_tmpfs(void) 2450{ 2451 int error; 2452 2453 error = init_inodecache(); 2454 if (error) 2455 goto out3; 2456 2457 error = register_filesystem(&tmpfs_fs_type); 2458 if (error) { 2459 printk(KERN_ERR "Could not register tmpfs\n"); 2460 goto out2; 2461 } 2462 2463 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER, 2464 tmpfs_fs_type.name, NULL); 2465 if (IS_ERR(shm_mnt)) { 2466 error = PTR_ERR(shm_mnt); 2467 printk(KERN_ERR "Could not kern_mount tmpfs\n"); 2468 goto out1; 2469 } 2470 return 0; 2471 2472out1: 2473 unregister_filesystem(&tmpfs_fs_type); 2474out2: 2475 destroy_inodecache(); 2476out3: 2477 shm_mnt = ERR_PTR(error); 2478 return error; 2479} 2480module_init(init_tmpfs) 2481 2482/* 2483 * shmem_file_setup - get an unlinked file living in tmpfs 2484 * 2485 * @name: name for dentry (to be seen in /proc/<pid>/maps 2486 * @size: size to be set for the file 2487 * 2488 */ 2489struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags) 2490{ 2491 int error; 2492 struct file *file; 2493 struct inode *inode; 2494 struct dentry *dentry, *root; 2495 struct qstr this; 2496 2497 if (IS_ERR(shm_mnt)) 2498 return (void *)shm_mnt; 2499 2500 if (size < 0 || size > SHMEM_MAX_BYTES) 2501 return ERR_PTR(-EINVAL); 2502 2503 if (shmem_acct_size(flags, size)) 2504 return ERR_PTR(-ENOMEM); 2505 2506 error = -ENOMEM; 2507 this.name = name; 2508 this.len = strlen(name); 2509 this.hash = 0; /* will go */ 2510 root = shm_mnt->mnt_root; 2511 dentry = d_alloc(root, &this); 2512 if (!dentry) 2513 goto put_memory; 2514 2515 error = -ENFILE; 2516 file = get_empty_filp(); 2517 if (!file) 2518 goto put_dentry; 2519 2520 error = -ENOSPC; 2521 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0); 2522 if (!inode) 2523 goto close_file; 2524 2525 SHMEM_I(inode)->flags = flags & VM_ACCOUNT; 2526 d_instantiate(dentry, inode); 2527 inode->i_size = size; 2528 inode->i_nlink = 0; /* It is unlinked */ 2529 file->f_path.mnt = mntget(shm_mnt); 2530 file->f_path.dentry = dentry; 2531 file->f_mapping = inode->i_mapping; 2532 file->f_op = &shmem_file_operations; 2533 file->f_mode = FMODE_WRITE | FMODE_READ; 2534 return file; 2535 2536close_file: 2537 put_filp(file); 2538put_dentry: 2539 dput(dentry); 2540put_memory: 2541 shmem_unacct_size(flags, size); 2542 return ERR_PTR(error); 2543} 2544 2545/* 2546 * shmem_zero_setup - setup a shared anonymous mapping 2547 * 2548 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff 2549 */ 2550int shmem_zero_setup(struct vm_area_struct *vma) 2551{ 2552 struct file *file; 2553 loff_t size = vma->vm_end - vma->vm_start; 2554 2555 file = shmem_file_setup("dev/zero", size, vma->vm_flags); 2556 if (IS_ERR(file)) 2557 return PTR_ERR(file); 2558 2559 if (vma->vm_file) 2560 fput(vma->vm_file); 2561 vma->vm_file = file; 2562 vma->vm_ops = &shmem_vm_ops; 2563 vma->vm_flags |= VM_CAN_INVALIDATE; 2564 return 0; 2565} 2566