shmem.c revision 680d794babebc74484c141448baa9b95b211cf5e
1/* 2 * Resizable virtual memory filesystem for Linux. 3 * 4 * Copyright (C) 2000 Linus Torvalds. 5 * 2000 Transmeta Corp. 6 * 2000-2001 Christoph Rohland 7 * 2000-2001 SAP AG 8 * 2002 Red Hat Inc. 9 * Copyright (C) 2002-2005 Hugh Dickins. 10 * Copyright (C) 2002-2005 VERITAS Software Corporation. 11 * Copyright (C) 2004 Andi Kleen, SuSE Labs 12 * 13 * Extended attribute support for tmpfs: 14 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net> 15 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com> 16 * 17 * This file is released under the GPL. 18 */ 19 20/* 21 * This virtual memory filesystem is heavily based on the ramfs. It 22 * extends ramfs by the ability to use swap and honor resource limits 23 * which makes it a completely usable filesystem. 24 */ 25 26#include <linux/module.h> 27#include <linux/init.h> 28#include <linux/fs.h> 29#include <linux/xattr.h> 30#include <linux/exportfs.h> 31#include <linux/generic_acl.h> 32#include <linux/mm.h> 33#include <linux/mman.h> 34#include <linux/file.h> 35#include <linux/swap.h> 36#include <linux/pagemap.h> 37#include <linux/string.h> 38#include <linux/slab.h> 39#include <linux/backing-dev.h> 40#include <linux/shmem_fs.h> 41#include <linux/mount.h> 42#include <linux/writeback.h> 43#include <linux/vfs.h> 44#include <linux/blkdev.h> 45#include <linux/security.h> 46#include <linux/swapops.h> 47#include <linux/mempolicy.h> 48#include <linux/namei.h> 49#include <linux/ctype.h> 50#include <linux/migrate.h> 51#include <linux/highmem.h> 52#include <linux/seq_file.h> 53 54#include <asm/uaccess.h> 55#include <asm/div64.h> 56#include <asm/pgtable.h> 57 58/* This magic number is used in glibc for posix shared memory */ 59#define TMPFS_MAGIC 0x01021994 60 61#define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long)) 62#define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE) 63#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512) 64 65#define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1)) 66#define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT) 67 68#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT) 69 70/* info->flags needs VM_flags to handle pagein/truncate races efficiently */ 71#define SHMEM_PAGEIN VM_READ 72#define SHMEM_TRUNCATE VM_WRITE 73 74/* Definition to limit shmem_truncate's steps between cond_rescheds */ 75#define LATENCY_LIMIT 64 76 77/* Pretend that each entry is of this size in directory's i_size */ 78#define BOGO_DIRENT_SIZE 20 79 80/* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */ 81enum sgp_type { 82 SGP_READ, /* don't exceed i_size, don't allocate page */ 83 SGP_CACHE, /* don't exceed i_size, may allocate page */ 84 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */ 85 SGP_WRITE, /* may exceed i_size, may allocate page */ 86}; 87 88static unsigned long shmem_default_max_blocks(void) 89{ 90 return totalram_pages / 2; 91} 92 93static unsigned long shmem_default_max_inodes(void) 94{ 95 return min(totalram_pages - totalhigh_pages, totalram_pages / 2); 96} 97 98static int shmem_getpage(struct inode *inode, unsigned long idx, 99 struct page **pagep, enum sgp_type sgp, int *type); 100 101static inline struct page *shmem_dir_alloc(gfp_t gfp_mask) 102{ 103 /* 104 * The above definition of ENTRIES_PER_PAGE, and the use of 105 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE: 106 * might be reconsidered if it ever diverges from PAGE_SIZE. 107 * 108 * Mobility flags are masked out as swap vectors cannot move 109 */ 110 return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO, 111 PAGE_CACHE_SHIFT-PAGE_SHIFT); 112} 113 114static inline void shmem_dir_free(struct page *page) 115{ 116 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT); 117} 118 119static struct page **shmem_dir_map(struct page *page) 120{ 121 return (struct page **)kmap_atomic(page, KM_USER0); 122} 123 124static inline void shmem_dir_unmap(struct page **dir) 125{ 126 kunmap_atomic(dir, KM_USER0); 127} 128 129static swp_entry_t *shmem_swp_map(struct page *page) 130{ 131 return (swp_entry_t *)kmap_atomic(page, KM_USER1); 132} 133 134static inline void shmem_swp_balance_unmap(void) 135{ 136 /* 137 * When passing a pointer to an i_direct entry, to code which 138 * also handles indirect entries and so will shmem_swp_unmap, 139 * we must arrange for the preempt count to remain in balance. 140 * What kmap_atomic of a lowmem page does depends on config 141 * and architecture, so pretend to kmap_atomic some lowmem page. 142 */ 143 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1); 144} 145 146static inline void shmem_swp_unmap(swp_entry_t *entry) 147{ 148 kunmap_atomic(entry, KM_USER1); 149} 150 151static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) 152{ 153 return sb->s_fs_info; 154} 155 156/* 157 * shmem_file_setup pre-accounts the whole fixed size of a VM object, 158 * for shared memory and for shared anonymous (/dev/zero) mappings 159 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), 160 * consistent with the pre-accounting of private mappings ... 161 */ 162static inline int shmem_acct_size(unsigned long flags, loff_t size) 163{ 164 return (flags & VM_ACCOUNT)? 165 security_vm_enough_memory(VM_ACCT(size)): 0; 166} 167 168static inline void shmem_unacct_size(unsigned long flags, loff_t size) 169{ 170 if (flags & VM_ACCOUNT) 171 vm_unacct_memory(VM_ACCT(size)); 172} 173 174/* 175 * ... whereas tmpfs objects are accounted incrementally as 176 * pages are allocated, in order to allow huge sparse files. 177 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, 178 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. 179 */ 180static inline int shmem_acct_block(unsigned long flags) 181{ 182 return (flags & VM_ACCOUNT)? 183 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE)); 184} 185 186static inline void shmem_unacct_blocks(unsigned long flags, long pages) 187{ 188 if (!(flags & VM_ACCOUNT)) 189 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE)); 190} 191 192static const struct super_operations shmem_ops; 193static const struct address_space_operations shmem_aops; 194static const struct file_operations shmem_file_operations; 195static const struct inode_operations shmem_inode_operations; 196static const struct inode_operations shmem_dir_inode_operations; 197static const struct inode_operations shmem_special_inode_operations; 198static struct vm_operations_struct shmem_vm_ops; 199 200static struct backing_dev_info shmem_backing_dev_info __read_mostly = { 201 .ra_pages = 0, /* No readahead */ 202 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK, 203 .unplug_io_fn = default_unplug_io_fn, 204}; 205 206static LIST_HEAD(shmem_swaplist); 207static DEFINE_MUTEX(shmem_swaplist_mutex); 208 209static void shmem_free_blocks(struct inode *inode, long pages) 210{ 211 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 212 if (sbinfo->max_blocks) { 213 spin_lock(&sbinfo->stat_lock); 214 sbinfo->free_blocks += pages; 215 inode->i_blocks -= pages*BLOCKS_PER_PAGE; 216 spin_unlock(&sbinfo->stat_lock); 217 } 218} 219 220static int shmem_reserve_inode(struct super_block *sb) 221{ 222 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 223 if (sbinfo->max_inodes) { 224 spin_lock(&sbinfo->stat_lock); 225 if (!sbinfo->free_inodes) { 226 spin_unlock(&sbinfo->stat_lock); 227 return -ENOSPC; 228 } 229 sbinfo->free_inodes--; 230 spin_unlock(&sbinfo->stat_lock); 231 } 232 return 0; 233} 234 235static void shmem_free_inode(struct super_block *sb) 236{ 237 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 238 if (sbinfo->max_inodes) { 239 spin_lock(&sbinfo->stat_lock); 240 sbinfo->free_inodes++; 241 spin_unlock(&sbinfo->stat_lock); 242 } 243} 244 245/* 246 * shmem_recalc_inode - recalculate the size of an inode 247 * 248 * @inode: inode to recalc 249 * 250 * We have to calculate the free blocks since the mm can drop 251 * undirtied hole pages behind our back. 252 * 253 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped 254 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) 255 * 256 * It has to be called with the spinlock held. 257 */ 258static void shmem_recalc_inode(struct inode *inode) 259{ 260 struct shmem_inode_info *info = SHMEM_I(inode); 261 long freed; 262 263 freed = info->alloced - info->swapped - inode->i_mapping->nrpages; 264 if (freed > 0) { 265 info->alloced -= freed; 266 shmem_unacct_blocks(info->flags, freed); 267 shmem_free_blocks(inode, freed); 268 } 269} 270 271/* 272 * shmem_swp_entry - find the swap vector position in the info structure 273 * 274 * @info: info structure for the inode 275 * @index: index of the page to find 276 * @page: optional page to add to the structure. Has to be preset to 277 * all zeros 278 * 279 * If there is no space allocated yet it will return NULL when 280 * page is NULL, else it will use the page for the needed block, 281 * setting it to NULL on return to indicate that it has been used. 282 * 283 * The swap vector is organized the following way: 284 * 285 * There are SHMEM_NR_DIRECT entries directly stored in the 286 * shmem_inode_info structure. So small files do not need an addional 287 * allocation. 288 * 289 * For pages with index > SHMEM_NR_DIRECT there is the pointer 290 * i_indirect which points to a page which holds in the first half 291 * doubly indirect blocks, in the second half triple indirect blocks: 292 * 293 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the 294 * following layout (for SHMEM_NR_DIRECT == 16): 295 * 296 * i_indirect -> dir --> 16-19 297 * | +-> 20-23 298 * | 299 * +-->dir2 --> 24-27 300 * | +-> 28-31 301 * | +-> 32-35 302 * | +-> 36-39 303 * | 304 * +-->dir3 --> 40-43 305 * +-> 44-47 306 * +-> 48-51 307 * +-> 52-55 308 */ 309static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page) 310{ 311 unsigned long offset; 312 struct page **dir; 313 struct page *subdir; 314 315 if (index < SHMEM_NR_DIRECT) { 316 shmem_swp_balance_unmap(); 317 return info->i_direct+index; 318 } 319 if (!info->i_indirect) { 320 if (page) { 321 info->i_indirect = *page; 322 *page = NULL; 323 } 324 return NULL; /* need another page */ 325 } 326 327 index -= SHMEM_NR_DIRECT; 328 offset = index % ENTRIES_PER_PAGE; 329 index /= ENTRIES_PER_PAGE; 330 dir = shmem_dir_map(info->i_indirect); 331 332 if (index >= ENTRIES_PER_PAGE/2) { 333 index -= ENTRIES_PER_PAGE/2; 334 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE; 335 index %= ENTRIES_PER_PAGE; 336 subdir = *dir; 337 if (!subdir) { 338 if (page) { 339 *dir = *page; 340 *page = NULL; 341 } 342 shmem_dir_unmap(dir); 343 return NULL; /* need another page */ 344 } 345 shmem_dir_unmap(dir); 346 dir = shmem_dir_map(subdir); 347 } 348 349 dir += index; 350 subdir = *dir; 351 if (!subdir) { 352 if (!page || !(subdir = *page)) { 353 shmem_dir_unmap(dir); 354 return NULL; /* need a page */ 355 } 356 *dir = subdir; 357 *page = NULL; 358 } 359 shmem_dir_unmap(dir); 360 return shmem_swp_map(subdir) + offset; 361} 362 363static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value) 364{ 365 long incdec = value? 1: -1; 366 367 entry->val = value; 368 info->swapped += incdec; 369 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) { 370 struct page *page = kmap_atomic_to_page(entry); 371 set_page_private(page, page_private(page) + incdec); 372 } 373} 374 375/* 376 * shmem_swp_alloc - get the position of the swap entry for the page. 377 * If it does not exist allocate the entry. 378 * 379 * @info: info structure for the inode 380 * @index: index of the page to find 381 * @sgp: check and recheck i_size? skip allocation? 382 */ 383static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp) 384{ 385 struct inode *inode = &info->vfs_inode; 386 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 387 struct page *page = NULL; 388 swp_entry_t *entry; 389 390 if (sgp != SGP_WRITE && 391 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 392 return ERR_PTR(-EINVAL); 393 394 while (!(entry = shmem_swp_entry(info, index, &page))) { 395 if (sgp == SGP_READ) 396 return shmem_swp_map(ZERO_PAGE(0)); 397 /* 398 * Test free_blocks against 1 not 0, since we have 1 data 399 * page (and perhaps indirect index pages) yet to allocate: 400 * a waste to allocate index if we cannot allocate data. 401 */ 402 if (sbinfo->max_blocks) { 403 spin_lock(&sbinfo->stat_lock); 404 if (sbinfo->free_blocks <= 1) { 405 spin_unlock(&sbinfo->stat_lock); 406 return ERR_PTR(-ENOSPC); 407 } 408 sbinfo->free_blocks--; 409 inode->i_blocks += BLOCKS_PER_PAGE; 410 spin_unlock(&sbinfo->stat_lock); 411 } 412 413 spin_unlock(&info->lock); 414 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping)); 415 if (page) 416 set_page_private(page, 0); 417 spin_lock(&info->lock); 418 419 if (!page) { 420 shmem_free_blocks(inode, 1); 421 return ERR_PTR(-ENOMEM); 422 } 423 if (sgp != SGP_WRITE && 424 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { 425 entry = ERR_PTR(-EINVAL); 426 break; 427 } 428 if (info->next_index <= index) 429 info->next_index = index + 1; 430 } 431 if (page) { 432 /* another task gave its page, or truncated the file */ 433 shmem_free_blocks(inode, 1); 434 shmem_dir_free(page); 435 } 436 if (info->next_index <= index && !IS_ERR(entry)) 437 info->next_index = index + 1; 438 return entry; 439} 440 441/* 442 * shmem_free_swp - free some swap entries in a directory 443 * 444 * @dir: pointer to the directory 445 * @edir: pointer after last entry of the directory 446 * @punch_lock: pointer to spinlock when needed for the holepunch case 447 */ 448static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir, 449 spinlock_t *punch_lock) 450{ 451 spinlock_t *punch_unlock = NULL; 452 swp_entry_t *ptr; 453 int freed = 0; 454 455 for (ptr = dir; ptr < edir; ptr++) { 456 if (ptr->val) { 457 if (unlikely(punch_lock)) { 458 punch_unlock = punch_lock; 459 punch_lock = NULL; 460 spin_lock(punch_unlock); 461 if (!ptr->val) 462 continue; 463 } 464 free_swap_and_cache(*ptr); 465 *ptr = (swp_entry_t){0}; 466 freed++; 467 } 468 } 469 if (punch_unlock) 470 spin_unlock(punch_unlock); 471 return freed; 472} 473 474static int shmem_map_and_free_swp(struct page *subdir, int offset, 475 int limit, struct page ***dir, spinlock_t *punch_lock) 476{ 477 swp_entry_t *ptr; 478 int freed = 0; 479 480 ptr = shmem_swp_map(subdir); 481 for (; offset < limit; offset += LATENCY_LIMIT) { 482 int size = limit - offset; 483 if (size > LATENCY_LIMIT) 484 size = LATENCY_LIMIT; 485 freed += shmem_free_swp(ptr+offset, ptr+offset+size, 486 punch_lock); 487 if (need_resched()) { 488 shmem_swp_unmap(ptr); 489 if (*dir) { 490 shmem_dir_unmap(*dir); 491 *dir = NULL; 492 } 493 cond_resched(); 494 ptr = shmem_swp_map(subdir); 495 } 496 } 497 shmem_swp_unmap(ptr); 498 return freed; 499} 500 501static void shmem_free_pages(struct list_head *next) 502{ 503 struct page *page; 504 int freed = 0; 505 506 do { 507 page = container_of(next, struct page, lru); 508 next = next->next; 509 shmem_dir_free(page); 510 freed++; 511 if (freed >= LATENCY_LIMIT) { 512 cond_resched(); 513 freed = 0; 514 } 515 } while (next); 516} 517 518static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end) 519{ 520 struct shmem_inode_info *info = SHMEM_I(inode); 521 unsigned long idx; 522 unsigned long size; 523 unsigned long limit; 524 unsigned long stage; 525 unsigned long diroff; 526 struct page **dir; 527 struct page *topdir; 528 struct page *middir; 529 struct page *subdir; 530 swp_entry_t *ptr; 531 LIST_HEAD(pages_to_free); 532 long nr_pages_to_free = 0; 533 long nr_swaps_freed = 0; 534 int offset; 535 int freed; 536 int punch_hole; 537 spinlock_t *needs_lock; 538 spinlock_t *punch_lock; 539 unsigned long upper_limit; 540 541 inode->i_ctime = inode->i_mtime = CURRENT_TIME; 542 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 543 if (idx >= info->next_index) 544 return; 545 546 spin_lock(&info->lock); 547 info->flags |= SHMEM_TRUNCATE; 548 if (likely(end == (loff_t) -1)) { 549 limit = info->next_index; 550 upper_limit = SHMEM_MAX_INDEX; 551 info->next_index = idx; 552 needs_lock = NULL; 553 punch_hole = 0; 554 } else { 555 if (end + 1 >= inode->i_size) { /* we may free a little more */ 556 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >> 557 PAGE_CACHE_SHIFT; 558 upper_limit = SHMEM_MAX_INDEX; 559 } else { 560 limit = (end + 1) >> PAGE_CACHE_SHIFT; 561 upper_limit = limit; 562 } 563 needs_lock = &info->lock; 564 punch_hole = 1; 565 } 566 567 topdir = info->i_indirect; 568 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) { 569 info->i_indirect = NULL; 570 nr_pages_to_free++; 571 list_add(&topdir->lru, &pages_to_free); 572 } 573 spin_unlock(&info->lock); 574 575 if (info->swapped && idx < SHMEM_NR_DIRECT) { 576 ptr = info->i_direct; 577 size = limit; 578 if (size > SHMEM_NR_DIRECT) 579 size = SHMEM_NR_DIRECT; 580 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock); 581 } 582 583 /* 584 * If there are no indirect blocks or we are punching a hole 585 * below indirect blocks, nothing to be done. 586 */ 587 if (!topdir || limit <= SHMEM_NR_DIRECT) 588 goto done2; 589 590 /* 591 * The truncation case has already dropped info->lock, and we're safe 592 * because i_size and next_index have already been lowered, preventing 593 * access beyond. But in the punch_hole case, we still need to take 594 * the lock when updating the swap directory, because there might be 595 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or 596 * shmem_writepage. However, whenever we find we can remove a whole 597 * directory page (not at the misaligned start or end of the range), 598 * we first NULLify its pointer in the level above, and then have no 599 * need to take the lock when updating its contents: needs_lock and 600 * punch_lock (either pointing to info->lock or NULL) manage this. 601 */ 602 603 upper_limit -= SHMEM_NR_DIRECT; 604 limit -= SHMEM_NR_DIRECT; 605 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0; 606 offset = idx % ENTRIES_PER_PAGE; 607 idx -= offset; 608 609 dir = shmem_dir_map(topdir); 610 stage = ENTRIES_PER_PAGEPAGE/2; 611 if (idx < ENTRIES_PER_PAGEPAGE/2) { 612 middir = topdir; 613 diroff = idx/ENTRIES_PER_PAGE; 614 } else { 615 dir += ENTRIES_PER_PAGE/2; 616 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE; 617 while (stage <= idx) 618 stage += ENTRIES_PER_PAGEPAGE; 619 middir = *dir; 620 if (*dir) { 621 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) % 622 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE; 623 if (!diroff && !offset && upper_limit >= stage) { 624 if (needs_lock) { 625 spin_lock(needs_lock); 626 *dir = NULL; 627 spin_unlock(needs_lock); 628 needs_lock = NULL; 629 } else 630 *dir = NULL; 631 nr_pages_to_free++; 632 list_add(&middir->lru, &pages_to_free); 633 } 634 shmem_dir_unmap(dir); 635 dir = shmem_dir_map(middir); 636 } else { 637 diroff = 0; 638 offset = 0; 639 idx = stage; 640 } 641 } 642 643 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) { 644 if (unlikely(idx == stage)) { 645 shmem_dir_unmap(dir); 646 dir = shmem_dir_map(topdir) + 647 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 648 while (!*dir) { 649 dir++; 650 idx += ENTRIES_PER_PAGEPAGE; 651 if (idx >= limit) 652 goto done1; 653 } 654 stage = idx + ENTRIES_PER_PAGEPAGE; 655 middir = *dir; 656 if (punch_hole) 657 needs_lock = &info->lock; 658 if (upper_limit >= stage) { 659 if (needs_lock) { 660 spin_lock(needs_lock); 661 *dir = NULL; 662 spin_unlock(needs_lock); 663 needs_lock = NULL; 664 } else 665 *dir = NULL; 666 nr_pages_to_free++; 667 list_add(&middir->lru, &pages_to_free); 668 } 669 shmem_dir_unmap(dir); 670 cond_resched(); 671 dir = shmem_dir_map(middir); 672 diroff = 0; 673 } 674 punch_lock = needs_lock; 675 subdir = dir[diroff]; 676 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) { 677 if (needs_lock) { 678 spin_lock(needs_lock); 679 dir[diroff] = NULL; 680 spin_unlock(needs_lock); 681 punch_lock = NULL; 682 } else 683 dir[diroff] = NULL; 684 nr_pages_to_free++; 685 list_add(&subdir->lru, &pages_to_free); 686 } 687 if (subdir && page_private(subdir) /* has swap entries */) { 688 size = limit - idx; 689 if (size > ENTRIES_PER_PAGE) 690 size = ENTRIES_PER_PAGE; 691 freed = shmem_map_and_free_swp(subdir, 692 offset, size, &dir, punch_lock); 693 if (!dir) 694 dir = shmem_dir_map(middir); 695 nr_swaps_freed += freed; 696 if (offset || punch_lock) { 697 spin_lock(&info->lock); 698 set_page_private(subdir, 699 page_private(subdir) - freed); 700 spin_unlock(&info->lock); 701 } else 702 BUG_ON(page_private(subdir) != freed); 703 } 704 offset = 0; 705 } 706done1: 707 shmem_dir_unmap(dir); 708done2: 709 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) { 710 /* 711 * Call truncate_inode_pages again: racing shmem_unuse_inode 712 * may have swizzled a page in from swap since vmtruncate or 713 * generic_delete_inode did it, before we lowered next_index. 714 * Also, though shmem_getpage checks i_size before adding to 715 * cache, no recheck after: so fix the narrow window there too. 716 * 717 * Recalling truncate_inode_pages_range and unmap_mapping_range 718 * every time for punch_hole (which never got a chance to clear 719 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive, 720 * yet hardly ever necessary: try to optimize them out later. 721 */ 722 truncate_inode_pages_range(inode->i_mapping, start, end); 723 if (punch_hole) 724 unmap_mapping_range(inode->i_mapping, start, 725 end - start, 1); 726 } 727 728 spin_lock(&info->lock); 729 info->flags &= ~SHMEM_TRUNCATE; 730 info->swapped -= nr_swaps_freed; 731 if (nr_pages_to_free) 732 shmem_free_blocks(inode, nr_pages_to_free); 733 shmem_recalc_inode(inode); 734 spin_unlock(&info->lock); 735 736 /* 737 * Empty swap vector directory pages to be freed? 738 */ 739 if (!list_empty(&pages_to_free)) { 740 pages_to_free.prev->next = NULL; 741 shmem_free_pages(pages_to_free.next); 742 } 743} 744 745static void shmem_truncate(struct inode *inode) 746{ 747 shmem_truncate_range(inode, inode->i_size, (loff_t)-1); 748} 749 750static int shmem_notify_change(struct dentry *dentry, struct iattr *attr) 751{ 752 struct inode *inode = dentry->d_inode; 753 struct page *page = NULL; 754 int error; 755 756 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { 757 if (attr->ia_size < inode->i_size) { 758 /* 759 * If truncating down to a partial page, then 760 * if that page is already allocated, hold it 761 * in memory until the truncation is over, so 762 * truncate_partial_page cannnot miss it were 763 * it assigned to swap. 764 */ 765 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) { 766 (void) shmem_getpage(inode, 767 attr->ia_size>>PAGE_CACHE_SHIFT, 768 &page, SGP_READ, NULL); 769 if (page) 770 unlock_page(page); 771 } 772 /* 773 * Reset SHMEM_PAGEIN flag so that shmem_truncate can 774 * detect if any pages might have been added to cache 775 * after truncate_inode_pages. But we needn't bother 776 * if it's being fully truncated to zero-length: the 777 * nrpages check is efficient enough in that case. 778 */ 779 if (attr->ia_size) { 780 struct shmem_inode_info *info = SHMEM_I(inode); 781 spin_lock(&info->lock); 782 info->flags &= ~SHMEM_PAGEIN; 783 spin_unlock(&info->lock); 784 } 785 } 786 } 787 788 error = inode_change_ok(inode, attr); 789 if (!error) 790 error = inode_setattr(inode, attr); 791#ifdef CONFIG_TMPFS_POSIX_ACL 792 if (!error && (attr->ia_valid & ATTR_MODE)) 793 error = generic_acl_chmod(inode, &shmem_acl_ops); 794#endif 795 if (page) 796 page_cache_release(page); 797 return error; 798} 799 800static void shmem_delete_inode(struct inode *inode) 801{ 802 struct shmem_inode_info *info = SHMEM_I(inode); 803 804 if (inode->i_op->truncate == shmem_truncate) { 805 truncate_inode_pages(inode->i_mapping, 0); 806 shmem_unacct_size(info->flags, inode->i_size); 807 inode->i_size = 0; 808 shmem_truncate(inode); 809 if (!list_empty(&info->swaplist)) { 810 mutex_lock(&shmem_swaplist_mutex); 811 list_del_init(&info->swaplist); 812 mutex_unlock(&shmem_swaplist_mutex); 813 } 814 } 815 BUG_ON(inode->i_blocks); 816 shmem_free_inode(inode->i_sb); 817 clear_inode(inode); 818} 819 820static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir) 821{ 822 swp_entry_t *ptr; 823 824 for (ptr = dir; ptr < edir; ptr++) { 825 if (ptr->val == entry.val) 826 return ptr - dir; 827 } 828 return -1; 829} 830 831static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page) 832{ 833 struct inode *inode; 834 unsigned long idx; 835 unsigned long size; 836 unsigned long limit; 837 unsigned long stage; 838 struct page **dir; 839 struct page *subdir; 840 swp_entry_t *ptr; 841 int offset; 842 int error; 843 844 idx = 0; 845 ptr = info->i_direct; 846 spin_lock(&info->lock); 847 if (!info->swapped) { 848 list_del_init(&info->swaplist); 849 goto lost2; 850 } 851 limit = info->next_index; 852 size = limit; 853 if (size > SHMEM_NR_DIRECT) 854 size = SHMEM_NR_DIRECT; 855 offset = shmem_find_swp(entry, ptr, ptr+size); 856 if (offset >= 0) 857 goto found; 858 if (!info->i_indirect) 859 goto lost2; 860 861 dir = shmem_dir_map(info->i_indirect); 862 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2; 863 864 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) { 865 if (unlikely(idx == stage)) { 866 shmem_dir_unmap(dir-1); 867 if (cond_resched_lock(&info->lock)) { 868 /* check it has not been truncated */ 869 if (limit > info->next_index) { 870 limit = info->next_index; 871 if (idx >= limit) 872 goto lost2; 873 } 874 } 875 dir = shmem_dir_map(info->i_indirect) + 876 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 877 while (!*dir) { 878 dir++; 879 idx += ENTRIES_PER_PAGEPAGE; 880 if (idx >= limit) 881 goto lost1; 882 } 883 stage = idx + ENTRIES_PER_PAGEPAGE; 884 subdir = *dir; 885 shmem_dir_unmap(dir); 886 dir = shmem_dir_map(subdir); 887 } 888 subdir = *dir; 889 if (subdir && page_private(subdir)) { 890 ptr = shmem_swp_map(subdir); 891 size = limit - idx; 892 if (size > ENTRIES_PER_PAGE) 893 size = ENTRIES_PER_PAGE; 894 offset = shmem_find_swp(entry, ptr, ptr+size); 895 shmem_swp_unmap(ptr); 896 if (offset >= 0) { 897 shmem_dir_unmap(dir); 898 goto found; 899 } 900 } 901 } 902lost1: 903 shmem_dir_unmap(dir-1); 904lost2: 905 spin_unlock(&info->lock); 906 return 0; 907found: 908 idx += offset; 909 inode = igrab(&info->vfs_inode); 910 spin_unlock(&info->lock); 911 912 /* 913 * Move _head_ to start search for next from here. 914 * But be careful: shmem_delete_inode checks list_empty without taking 915 * mutex, and there's an instant in list_move_tail when info->swaplist 916 * would appear empty, if it were the only one on shmem_swaplist. We 917 * could avoid doing it if inode NULL; or use this minor optimization. 918 */ 919 if (shmem_swaplist.next != &info->swaplist) 920 list_move_tail(&shmem_swaplist, &info->swaplist); 921 mutex_unlock(&shmem_swaplist_mutex); 922 923 error = 1; 924 if (!inode) 925 goto out; 926 /* Precharge page while we can wait, compensate afterwards */ 927 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL); 928 if (error) 929 goto out; 930 error = radix_tree_preload(GFP_KERNEL); 931 if (error) 932 goto uncharge; 933 error = 1; 934 935 spin_lock(&info->lock); 936 ptr = shmem_swp_entry(info, idx, NULL); 937 if (ptr && ptr->val == entry.val) 938 error = add_to_page_cache(page, inode->i_mapping, 939 idx, GFP_NOWAIT); 940 if (error == -EEXIST) { 941 struct page *filepage = find_get_page(inode->i_mapping, idx); 942 error = 1; 943 if (filepage) { 944 /* 945 * There might be a more uptodate page coming down 946 * from a stacked writepage: forget our swappage if so. 947 */ 948 if (PageUptodate(filepage)) 949 error = 0; 950 page_cache_release(filepage); 951 } 952 } 953 if (!error) { 954 delete_from_swap_cache(page); 955 set_page_dirty(page); 956 info->flags |= SHMEM_PAGEIN; 957 shmem_swp_set(info, ptr, 0); 958 swap_free(entry); 959 error = 1; /* not an error, but entry was found */ 960 } 961 if (ptr) 962 shmem_swp_unmap(ptr); 963 spin_unlock(&info->lock); 964 radix_tree_preload_end(); 965uncharge: 966 mem_cgroup_uncharge_page(page); 967out: 968 unlock_page(page); 969 page_cache_release(page); 970 iput(inode); /* allows for NULL */ 971 return error; 972} 973 974/* 975 * shmem_unuse() search for an eventually swapped out shmem page. 976 */ 977int shmem_unuse(swp_entry_t entry, struct page *page) 978{ 979 struct list_head *p, *next; 980 struct shmem_inode_info *info; 981 int found = 0; 982 983 mutex_lock(&shmem_swaplist_mutex); 984 list_for_each_safe(p, next, &shmem_swaplist) { 985 info = list_entry(p, struct shmem_inode_info, swaplist); 986 found = shmem_unuse_inode(info, entry, page); 987 cond_resched(); 988 if (found) 989 goto out; 990 } 991 mutex_unlock(&shmem_swaplist_mutex); 992out: return found; /* 0 or 1 or -ENOMEM */ 993} 994 995/* 996 * Move the page from the page cache to the swap cache. 997 */ 998static int shmem_writepage(struct page *page, struct writeback_control *wbc) 999{ 1000 struct shmem_inode_info *info; 1001 swp_entry_t *entry, swap; 1002 struct address_space *mapping; 1003 unsigned long index; 1004 struct inode *inode; 1005 1006 BUG_ON(!PageLocked(page)); 1007 mapping = page->mapping; 1008 index = page->index; 1009 inode = mapping->host; 1010 info = SHMEM_I(inode); 1011 if (info->flags & VM_LOCKED) 1012 goto redirty; 1013 if (!total_swap_pages) 1014 goto redirty; 1015 1016 /* 1017 * shmem_backing_dev_info's capabilities prevent regular writeback or 1018 * sync from ever calling shmem_writepage; but a stacking filesystem 1019 * may use the ->writepage of its underlying filesystem, in which case 1020 * tmpfs should write out to swap only in response to memory pressure, 1021 * and not for pdflush or sync. However, in those cases, we do still 1022 * want to check if there's a redundant swappage to be discarded. 1023 */ 1024 if (wbc->for_reclaim) 1025 swap = get_swap_page(); 1026 else 1027 swap.val = 0; 1028 1029 spin_lock(&info->lock); 1030 if (index >= info->next_index) { 1031 BUG_ON(!(info->flags & SHMEM_TRUNCATE)); 1032 goto unlock; 1033 } 1034 entry = shmem_swp_entry(info, index, NULL); 1035 if (entry->val) { 1036 /* 1037 * The more uptodate page coming down from a stacked 1038 * writepage should replace our old swappage. 1039 */ 1040 free_swap_and_cache(*entry); 1041 shmem_swp_set(info, entry, 0); 1042 } 1043 shmem_recalc_inode(inode); 1044 1045 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { 1046 remove_from_page_cache(page); 1047 shmem_swp_set(info, entry, swap.val); 1048 shmem_swp_unmap(entry); 1049 if (list_empty(&info->swaplist)) 1050 inode = igrab(inode); 1051 else 1052 inode = NULL; 1053 spin_unlock(&info->lock); 1054 swap_duplicate(swap); 1055 BUG_ON(page_mapped(page)); 1056 page_cache_release(page); /* pagecache ref */ 1057 set_page_dirty(page); 1058 unlock_page(page); 1059 if (inode) { 1060 mutex_lock(&shmem_swaplist_mutex); 1061 /* move instead of add in case we're racing */ 1062 list_move_tail(&info->swaplist, &shmem_swaplist); 1063 mutex_unlock(&shmem_swaplist_mutex); 1064 iput(inode); 1065 } 1066 return 0; 1067 } 1068 1069 shmem_swp_unmap(entry); 1070unlock: 1071 spin_unlock(&info->lock); 1072 swap_free(swap); 1073redirty: 1074 set_page_dirty(page); 1075 if (wbc->for_reclaim) 1076 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ 1077 unlock_page(page); 1078 return 0; 1079} 1080 1081#ifdef CONFIG_NUMA 1082#ifdef CONFIG_TMPFS 1083static int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes) 1084{ 1085 char *nodelist = strchr(value, ':'); 1086 int err = 1; 1087 1088 if (nodelist) { 1089 /* NUL-terminate policy string */ 1090 *nodelist++ = '\0'; 1091 if (nodelist_parse(nodelist, *policy_nodes)) 1092 goto out; 1093 if (!nodes_subset(*policy_nodes, node_states[N_HIGH_MEMORY])) 1094 goto out; 1095 } 1096 if (!strcmp(value, "default")) { 1097 *policy = MPOL_DEFAULT; 1098 /* Don't allow a nodelist */ 1099 if (!nodelist) 1100 err = 0; 1101 } else if (!strcmp(value, "prefer")) { 1102 *policy = MPOL_PREFERRED; 1103 /* Insist on a nodelist of one node only */ 1104 if (nodelist) { 1105 char *rest = nodelist; 1106 while (isdigit(*rest)) 1107 rest++; 1108 if (!*rest) 1109 err = 0; 1110 } 1111 } else if (!strcmp(value, "bind")) { 1112 *policy = MPOL_BIND; 1113 /* Insist on a nodelist */ 1114 if (nodelist) 1115 err = 0; 1116 } else if (!strcmp(value, "interleave")) { 1117 *policy = MPOL_INTERLEAVE; 1118 /* 1119 * Default to online nodes with memory if no nodelist 1120 */ 1121 if (!nodelist) 1122 *policy_nodes = node_states[N_HIGH_MEMORY]; 1123 err = 0; 1124 } 1125out: 1126 /* Restore string for error message */ 1127 if (nodelist) 1128 *--nodelist = ':'; 1129 return err; 1130} 1131 1132static void shmem_show_mpol(struct seq_file *seq, int policy, 1133 const nodemask_t policy_nodes) 1134{ 1135 char *policy_string; 1136 1137 switch (policy) { 1138 case MPOL_PREFERRED: 1139 policy_string = "prefer"; 1140 break; 1141 case MPOL_BIND: 1142 policy_string = "bind"; 1143 break; 1144 case MPOL_INTERLEAVE: 1145 policy_string = "interleave"; 1146 break; 1147 default: 1148 /* MPOL_DEFAULT */ 1149 return; 1150 } 1151 1152 seq_printf(seq, ",mpol=%s", policy_string); 1153 1154 if (policy != MPOL_INTERLEAVE || 1155 !nodes_equal(policy_nodes, node_states[N_HIGH_MEMORY])) { 1156 char buffer[64]; 1157 int len; 1158 1159 len = nodelist_scnprintf(buffer, sizeof(buffer), policy_nodes); 1160 if (len < sizeof(buffer)) 1161 seq_printf(seq, ":%s", buffer); 1162 else 1163 seq_printf(seq, ":?"); 1164 } 1165} 1166#endif /* CONFIG_TMPFS */ 1167 1168static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, 1169 struct shmem_inode_info *info, unsigned long idx) 1170{ 1171 struct vm_area_struct pvma; 1172 struct page *page; 1173 1174 /* Create a pseudo vma that just contains the policy */ 1175 pvma.vm_start = 0; 1176 pvma.vm_pgoff = idx; 1177 pvma.vm_ops = NULL; 1178 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx); 1179 page = swapin_readahead(entry, gfp, &pvma, 0); 1180 mpol_free(pvma.vm_policy); 1181 return page; 1182} 1183 1184static struct page *shmem_alloc_page(gfp_t gfp, 1185 struct shmem_inode_info *info, unsigned long idx) 1186{ 1187 struct vm_area_struct pvma; 1188 struct page *page; 1189 1190 /* Create a pseudo vma that just contains the policy */ 1191 pvma.vm_start = 0; 1192 pvma.vm_pgoff = idx; 1193 pvma.vm_ops = NULL; 1194 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx); 1195 page = alloc_page_vma(gfp, &pvma, 0); 1196 mpol_free(pvma.vm_policy); 1197 return page; 1198} 1199#else /* !CONFIG_NUMA */ 1200#ifdef CONFIG_TMPFS 1201static inline int shmem_parse_mpol(char *value, int *policy, 1202 nodemask_t *policy_nodes) 1203{ 1204 return 1; 1205} 1206 1207static inline void shmem_show_mpol(struct seq_file *seq, int policy, 1208 const nodemask_t policy_nodes) 1209{ 1210} 1211#endif /* CONFIG_TMPFS */ 1212 1213static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp, 1214 struct shmem_inode_info *info, unsigned long idx) 1215{ 1216 return swapin_readahead(entry, gfp, NULL, 0); 1217} 1218 1219static inline struct page *shmem_alloc_page(gfp_t gfp, 1220 struct shmem_inode_info *info, unsigned long idx) 1221{ 1222 return alloc_page(gfp); 1223} 1224#endif /* CONFIG_NUMA */ 1225 1226/* 1227 * shmem_getpage - either get the page from swap or allocate a new one 1228 * 1229 * If we allocate a new one we do not mark it dirty. That's up to the 1230 * vm. If we swap it in we mark it dirty since we also free the swap 1231 * entry since a page cannot live in both the swap and page cache 1232 */ 1233static int shmem_getpage(struct inode *inode, unsigned long idx, 1234 struct page **pagep, enum sgp_type sgp, int *type) 1235{ 1236 struct address_space *mapping = inode->i_mapping; 1237 struct shmem_inode_info *info = SHMEM_I(inode); 1238 struct shmem_sb_info *sbinfo; 1239 struct page *filepage = *pagep; 1240 struct page *swappage; 1241 swp_entry_t *entry; 1242 swp_entry_t swap; 1243 gfp_t gfp; 1244 int error; 1245 1246 if (idx >= SHMEM_MAX_INDEX) 1247 return -EFBIG; 1248 1249 if (type) 1250 *type = 0; 1251 1252 /* 1253 * Normally, filepage is NULL on entry, and either found 1254 * uptodate immediately, or allocated and zeroed, or read 1255 * in under swappage, which is then assigned to filepage. 1256 * But shmem_readpage (required for splice) passes in a locked 1257 * filepage, which may be found not uptodate by other callers 1258 * too, and may need to be copied from the swappage read in. 1259 */ 1260repeat: 1261 if (!filepage) 1262 filepage = find_lock_page(mapping, idx); 1263 if (filepage && PageUptodate(filepage)) 1264 goto done; 1265 error = 0; 1266 gfp = mapping_gfp_mask(mapping); 1267 if (!filepage) { 1268 /* 1269 * Try to preload while we can wait, to not make a habit of 1270 * draining atomic reserves; but don't latch on to this cpu. 1271 */ 1272 error = radix_tree_preload(gfp & ~__GFP_HIGHMEM); 1273 if (error) 1274 goto failed; 1275 radix_tree_preload_end(); 1276 } 1277 1278 spin_lock(&info->lock); 1279 shmem_recalc_inode(inode); 1280 entry = shmem_swp_alloc(info, idx, sgp); 1281 if (IS_ERR(entry)) { 1282 spin_unlock(&info->lock); 1283 error = PTR_ERR(entry); 1284 goto failed; 1285 } 1286 swap = *entry; 1287 1288 if (swap.val) { 1289 /* Look it up and read it in.. */ 1290 swappage = lookup_swap_cache(swap); 1291 if (!swappage) { 1292 shmem_swp_unmap(entry); 1293 /* here we actually do the io */ 1294 if (type && !(*type & VM_FAULT_MAJOR)) { 1295 __count_vm_event(PGMAJFAULT); 1296 *type |= VM_FAULT_MAJOR; 1297 } 1298 spin_unlock(&info->lock); 1299 swappage = shmem_swapin(swap, gfp, info, idx); 1300 if (!swappage) { 1301 spin_lock(&info->lock); 1302 entry = shmem_swp_alloc(info, idx, sgp); 1303 if (IS_ERR(entry)) 1304 error = PTR_ERR(entry); 1305 else { 1306 if (entry->val == swap.val) 1307 error = -ENOMEM; 1308 shmem_swp_unmap(entry); 1309 } 1310 spin_unlock(&info->lock); 1311 if (error) 1312 goto failed; 1313 goto repeat; 1314 } 1315 wait_on_page_locked(swappage); 1316 page_cache_release(swappage); 1317 goto repeat; 1318 } 1319 1320 /* We have to do this with page locked to prevent races */ 1321 if (TestSetPageLocked(swappage)) { 1322 shmem_swp_unmap(entry); 1323 spin_unlock(&info->lock); 1324 wait_on_page_locked(swappage); 1325 page_cache_release(swappage); 1326 goto repeat; 1327 } 1328 if (PageWriteback(swappage)) { 1329 shmem_swp_unmap(entry); 1330 spin_unlock(&info->lock); 1331 wait_on_page_writeback(swappage); 1332 unlock_page(swappage); 1333 page_cache_release(swappage); 1334 goto repeat; 1335 } 1336 if (!PageUptodate(swappage)) { 1337 shmem_swp_unmap(entry); 1338 spin_unlock(&info->lock); 1339 unlock_page(swappage); 1340 page_cache_release(swappage); 1341 error = -EIO; 1342 goto failed; 1343 } 1344 1345 if (filepage) { 1346 shmem_swp_set(info, entry, 0); 1347 shmem_swp_unmap(entry); 1348 delete_from_swap_cache(swappage); 1349 spin_unlock(&info->lock); 1350 copy_highpage(filepage, swappage); 1351 unlock_page(swappage); 1352 page_cache_release(swappage); 1353 flush_dcache_page(filepage); 1354 SetPageUptodate(filepage); 1355 set_page_dirty(filepage); 1356 swap_free(swap); 1357 } else if (!(error = add_to_page_cache( 1358 swappage, mapping, idx, GFP_NOWAIT))) { 1359 info->flags |= SHMEM_PAGEIN; 1360 shmem_swp_set(info, entry, 0); 1361 shmem_swp_unmap(entry); 1362 delete_from_swap_cache(swappage); 1363 spin_unlock(&info->lock); 1364 filepage = swappage; 1365 set_page_dirty(filepage); 1366 swap_free(swap); 1367 } else { 1368 shmem_swp_unmap(entry); 1369 spin_unlock(&info->lock); 1370 unlock_page(swappage); 1371 page_cache_release(swappage); 1372 if (error == -ENOMEM) { 1373 /* allow reclaim from this memory cgroup */ 1374 error = mem_cgroup_cache_charge(NULL, 1375 current->mm, gfp & ~__GFP_HIGHMEM); 1376 if (error) 1377 goto failed; 1378 } 1379 goto repeat; 1380 } 1381 } else if (sgp == SGP_READ && !filepage) { 1382 shmem_swp_unmap(entry); 1383 filepage = find_get_page(mapping, idx); 1384 if (filepage && 1385 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) { 1386 spin_unlock(&info->lock); 1387 wait_on_page_locked(filepage); 1388 page_cache_release(filepage); 1389 filepage = NULL; 1390 goto repeat; 1391 } 1392 spin_unlock(&info->lock); 1393 } else { 1394 shmem_swp_unmap(entry); 1395 sbinfo = SHMEM_SB(inode->i_sb); 1396 if (sbinfo->max_blocks) { 1397 spin_lock(&sbinfo->stat_lock); 1398 if (sbinfo->free_blocks == 0 || 1399 shmem_acct_block(info->flags)) { 1400 spin_unlock(&sbinfo->stat_lock); 1401 spin_unlock(&info->lock); 1402 error = -ENOSPC; 1403 goto failed; 1404 } 1405 sbinfo->free_blocks--; 1406 inode->i_blocks += BLOCKS_PER_PAGE; 1407 spin_unlock(&sbinfo->stat_lock); 1408 } else if (shmem_acct_block(info->flags)) { 1409 spin_unlock(&info->lock); 1410 error = -ENOSPC; 1411 goto failed; 1412 } 1413 1414 if (!filepage) { 1415 spin_unlock(&info->lock); 1416 filepage = shmem_alloc_page(gfp, info, idx); 1417 if (!filepage) { 1418 shmem_unacct_blocks(info->flags, 1); 1419 shmem_free_blocks(inode, 1); 1420 error = -ENOMEM; 1421 goto failed; 1422 } 1423 1424 /* Precharge page while we can wait, compensate after */ 1425 error = mem_cgroup_cache_charge(filepage, current->mm, 1426 gfp & ~__GFP_HIGHMEM); 1427 if (error) { 1428 page_cache_release(filepage); 1429 shmem_unacct_blocks(info->flags, 1); 1430 shmem_free_blocks(inode, 1); 1431 filepage = NULL; 1432 goto failed; 1433 } 1434 1435 spin_lock(&info->lock); 1436 entry = shmem_swp_alloc(info, idx, sgp); 1437 if (IS_ERR(entry)) 1438 error = PTR_ERR(entry); 1439 else { 1440 swap = *entry; 1441 shmem_swp_unmap(entry); 1442 } 1443 if (error || swap.val || 0 != add_to_page_cache_lru( 1444 filepage, mapping, idx, GFP_NOWAIT)) { 1445 spin_unlock(&info->lock); 1446 mem_cgroup_uncharge_page(filepage); 1447 page_cache_release(filepage); 1448 shmem_unacct_blocks(info->flags, 1); 1449 shmem_free_blocks(inode, 1); 1450 filepage = NULL; 1451 if (error) 1452 goto failed; 1453 goto repeat; 1454 } 1455 mem_cgroup_uncharge_page(filepage); 1456 info->flags |= SHMEM_PAGEIN; 1457 } 1458 1459 info->alloced++; 1460 spin_unlock(&info->lock); 1461 clear_highpage(filepage); 1462 flush_dcache_page(filepage); 1463 SetPageUptodate(filepage); 1464 if (sgp == SGP_DIRTY) 1465 set_page_dirty(filepage); 1466 } 1467done: 1468 *pagep = filepage; 1469 return 0; 1470 1471failed: 1472 if (*pagep != filepage) { 1473 unlock_page(filepage); 1474 page_cache_release(filepage); 1475 } 1476 return error; 1477} 1478 1479static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1480{ 1481 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 1482 int error; 1483 int ret; 1484 1485 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 1486 return VM_FAULT_SIGBUS; 1487 1488 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret); 1489 if (error) 1490 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); 1491 1492 mark_page_accessed(vmf->page); 1493 return ret | VM_FAULT_LOCKED; 1494} 1495 1496#ifdef CONFIG_NUMA 1497static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new) 1498{ 1499 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1500 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new); 1501} 1502 1503static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, 1504 unsigned long addr) 1505{ 1506 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1507 unsigned long idx; 1508 1509 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; 1510 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx); 1511} 1512#endif 1513 1514int shmem_lock(struct file *file, int lock, struct user_struct *user) 1515{ 1516 struct inode *inode = file->f_path.dentry->d_inode; 1517 struct shmem_inode_info *info = SHMEM_I(inode); 1518 int retval = -ENOMEM; 1519 1520 spin_lock(&info->lock); 1521 if (lock && !(info->flags & VM_LOCKED)) { 1522 if (!user_shm_lock(inode->i_size, user)) 1523 goto out_nomem; 1524 info->flags |= VM_LOCKED; 1525 } 1526 if (!lock && (info->flags & VM_LOCKED) && user) { 1527 user_shm_unlock(inode->i_size, user); 1528 info->flags &= ~VM_LOCKED; 1529 } 1530 retval = 0; 1531out_nomem: 1532 spin_unlock(&info->lock); 1533 return retval; 1534} 1535 1536static int shmem_mmap(struct file *file, struct vm_area_struct *vma) 1537{ 1538 file_accessed(file); 1539 vma->vm_ops = &shmem_vm_ops; 1540 vma->vm_flags |= VM_CAN_NONLINEAR; 1541 return 0; 1542} 1543 1544static struct inode * 1545shmem_get_inode(struct super_block *sb, int mode, dev_t dev) 1546{ 1547 struct inode *inode; 1548 struct shmem_inode_info *info; 1549 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 1550 1551 if (shmem_reserve_inode(sb)) 1552 return NULL; 1553 1554 inode = new_inode(sb); 1555 if (inode) { 1556 inode->i_mode = mode; 1557 inode->i_uid = current->fsuid; 1558 inode->i_gid = current->fsgid; 1559 inode->i_blocks = 0; 1560 inode->i_mapping->a_ops = &shmem_aops; 1561 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; 1562 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 1563 inode->i_generation = get_seconds(); 1564 info = SHMEM_I(inode); 1565 memset(info, 0, (char *)inode - (char *)info); 1566 spin_lock_init(&info->lock); 1567 INIT_LIST_HEAD(&info->swaplist); 1568 1569 switch (mode & S_IFMT) { 1570 default: 1571 inode->i_op = &shmem_special_inode_operations; 1572 init_special_inode(inode, mode, dev); 1573 break; 1574 case S_IFREG: 1575 inode->i_op = &shmem_inode_operations; 1576 inode->i_fop = &shmem_file_operations; 1577 mpol_shared_policy_init(&info->policy, sbinfo->policy, 1578 &sbinfo->policy_nodes); 1579 break; 1580 case S_IFDIR: 1581 inc_nlink(inode); 1582 /* Some things misbehave if size == 0 on a directory */ 1583 inode->i_size = 2 * BOGO_DIRENT_SIZE; 1584 inode->i_op = &shmem_dir_inode_operations; 1585 inode->i_fop = &simple_dir_operations; 1586 break; 1587 case S_IFLNK: 1588 /* 1589 * Must not load anything in the rbtree, 1590 * mpol_free_shared_policy will not be called. 1591 */ 1592 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, 1593 NULL); 1594 break; 1595 } 1596 } else 1597 shmem_free_inode(sb); 1598 return inode; 1599} 1600 1601#ifdef CONFIG_TMPFS 1602static const struct inode_operations shmem_symlink_inode_operations; 1603static const struct inode_operations shmem_symlink_inline_operations; 1604 1605/* 1606 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin; 1607 * but providing them allows a tmpfs file to be used for splice, sendfile, and 1608 * below the loop driver, in the generic fashion that many filesystems support. 1609 */ 1610static int shmem_readpage(struct file *file, struct page *page) 1611{ 1612 struct inode *inode = page->mapping->host; 1613 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL); 1614 unlock_page(page); 1615 return error; 1616} 1617 1618static int 1619shmem_write_begin(struct file *file, struct address_space *mapping, 1620 loff_t pos, unsigned len, unsigned flags, 1621 struct page **pagep, void **fsdata) 1622{ 1623 struct inode *inode = mapping->host; 1624 pgoff_t index = pos >> PAGE_CACHE_SHIFT; 1625 *pagep = NULL; 1626 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL); 1627} 1628 1629static int 1630shmem_write_end(struct file *file, struct address_space *mapping, 1631 loff_t pos, unsigned len, unsigned copied, 1632 struct page *page, void *fsdata) 1633{ 1634 struct inode *inode = mapping->host; 1635 1636 if (pos + copied > inode->i_size) 1637 i_size_write(inode, pos + copied); 1638 1639 unlock_page(page); 1640 set_page_dirty(page); 1641 page_cache_release(page); 1642 1643 return copied; 1644} 1645 1646static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor) 1647{ 1648 struct inode *inode = filp->f_path.dentry->d_inode; 1649 struct address_space *mapping = inode->i_mapping; 1650 unsigned long index, offset; 1651 enum sgp_type sgp = SGP_READ; 1652 1653 /* 1654 * Might this read be for a stacking filesystem? Then when reading 1655 * holes of a sparse file, we actually need to allocate those pages, 1656 * and even mark them dirty, so it cannot exceed the max_blocks limit. 1657 */ 1658 if (segment_eq(get_fs(), KERNEL_DS)) 1659 sgp = SGP_DIRTY; 1660 1661 index = *ppos >> PAGE_CACHE_SHIFT; 1662 offset = *ppos & ~PAGE_CACHE_MASK; 1663 1664 for (;;) { 1665 struct page *page = NULL; 1666 unsigned long end_index, nr, ret; 1667 loff_t i_size = i_size_read(inode); 1668 1669 end_index = i_size >> PAGE_CACHE_SHIFT; 1670 if (index > end_index) 1671 break; 1672 if (index == end_index) { 1673 nr = i_size & ~PAGE_CACHE_MASK; 1674 if (nr <= offset) 1675 break; 1676 } 1677 1678 desc->error = shmem_getpage(inode, index, &page, sgp, NULL); 1679 if (desc->error) { 1680 if (desc->error == -EINVAL) 1681 desc->error = 0; 1682 break; 1683 } 1684 if (page) 1685 unlock_page(page); 1686 1687 /* 1688 * We must evaluate after, since reads (unlike writes) 1689 * are called without i_mutex protection against truncate 1690 */ 1691 nr = PAGE_CACHE_SIZE; 1692 i_size = i_size_read(inode); 1693 end_index = i_size >> PAGE_CACHE_SHIFT; 1694 if (index == end_index) { 1695 nr = i_size & ~PAGE_CACHE_MASK; 1696 if (nr <= offset) { 1697 if (page) 1698 page_cache_release(page); 1699 break; 1700 } 1701 } 1702 nr -= offset; 1703 1704 if (page) { 1705 /* 1706 * If users can be writing to this page using arbitrary 1707 * virtual addresses, take care about potential aliasing 1708 * before reading the page on the kernel side. 1709 */ 1710 if (mapping_writably_mapped(mapping)) 1711 flush_dcache_page(page); 1712 /* 1713 * Mark the page accessed if we read the beginning. 1714 */ 1715 if (!offset) 1716 mark_page_accessed(page); 1717 } else { 1718 page = ZERO_PAGE(0); 1719 page_cache_get(page); 1720 } 1721 1722 /* 1723 * Ok, we have the page, and it's up-to-date, so 1724 * now we can copy it to user space... 1725 * 1726 * The actor routine returns how many bytes were actually used.. 1727 * NOTE! This may not be the same as how much of a user buffer 1728 * we filled up (we may be padding etc), so we can only update 1729 * "pos" here (the actor routine has to update the user buffer 1730 * pointers and the remaining count). 1731 */ 1732 ret = actor(desc, page, offset, nr); 1733 offset += ret; 1734 index += offset >> PAGE_CACHE_SHIFT; 1735 offset &= ~PAGE_CACHE_MASK; 1736 1737 page_cache_release(page); 1738 if (ret != nr || !desc->count) 1739 break; 1740 1741 cond_resched(); 1742 } 1743 1744 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; 1745 file_accessed(filp); 1746} 1747 1748static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos) 1749{ 1750 read_descriptor_t desc; 1751 1752 if ((ssize_t) count < 0) 1753 return -EINVAL; 1754 if (!access_ok(VERIFY_WRITE, buf, count)) 1755 return -EFAULT; 1756 if (!count) 1757 return 0; 1758 1759 desc.written = 0; 1760 desc.count = count; 1761 desc.arg.buf = buf; 1762 desc.error = 0; 1763 1764 do_shmem_file_read(filp, ppos, &desc, file_read_actor); 1765 if (desc.written) 1766 return desc.written; 1767 return desc.error; 1768} 1769 1770static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) 1771{ 1772 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); 1773 1774 buf->f_type = TMPFS_MAGIC; 1775 buf->f_bsize = PAGE_CACHE_SIZE; 1776 buf->f_namelen = NAME_MAX; 1777 spin_lock(&sbinfo->stat_lock); 1778 if (sbinfo->max_blocks) { 1779 buf->f_blocks = sbinfo->max_blocks; 1780 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks; 1781 } 1782 if (sbinfo->max_inodes) { 1783 buf->f_files = sbinfo->max_inodes; 1784 buf->f_ffree = sbinfo->free_inodes; 1785 } 1786 /* else leave those fields 0 like simple_statfs */ 1787 spin_unlock(&sbinfo->stat_lock); 1788 return 0; 1789} 1790 1791/* 1792 * File creation. Allocate an inode, and we're done.. 1793 */ 1794static int 1795shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 1796{ 1797 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev); 1798 int error = -ENOSPC; 1799 1800 if (inode) { 1801 error = security_inode_init_security(inode, dir, NULL, NULL, 1802 NULL); 1803 if (error) { 1804 if (error != -EOPNOTSUPP) { 1805 iput(inode); 1806 return error; 1807 } 1808 } 1809 error = shmem_acl_init(inode, dir); 1810 if (error) { 1811 iput(inode); 1812 return error; 1813 } 1814 if (dir->i_mode & S_ISGID) { 1815 inode->i_gid = dir->i_gid; 1816 if (S_ISDIR(mode)) 1817 inode->i_mode |= S_ISGID; 1818 } 1819 dir->i_size += BOGO_DIRENT_SIZE; 1820 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1821 d_instantiate(dentry, inode); 1822 dget(dentry); /* Extra count - pin the dentry in core */ 1823 } 1824 return error; 1825} 1826 1827static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1828{ 1829 int error; 1830 1831 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) 1832 return error; 1833 inc_nlink(dir); 1834 return 0; 1835} 1836 1837static int shmem_create(struct inode *dir, struct dentry *dentry, int mode, 1838 struct nameidata *nd) 1839{ 1840 return shmem_mknod(dir, dentry, mode | S_IFREG, 0); 1841} 1842 1843/* 1844 * Link a file.. 1845 */ 1846static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1847{ 1848 struct inode *inode = old_dentry->d_inode; 1849 int ret; 1850 1851 /* 1852 * No ordinary (disk based) filesystem counts links as inodes; 1853 * but each new link needs a new dentry, pinning lowmem, and 1854 * tmpfs dentries cannot be pruned until they are unlinked. 1855 */ 1856 ret = shmem_reserve_inode(inode->i_sb); 1857 if (ret) 1858 goto out; 1859 1860 dir->i_size += BOGO_DIRENT_SIZE; 1861 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1862 inc_nlink(inode); 1863 atomic_inc(&inode->i_count); /* New dentry reference */ 1864 dget(dentry); /* Extra pinning count for the created dentry */ 1865 d_instantiate(dentry, inode); 1866out: 1867 return ret; 1868} 1869 1870static int shmem_unlink(struct inode *dir, struct dentry *dentry) 1871{ 1872 struct inode *inode = dentry->d_inode; 1873 1874 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) 1875 shmem_free_inode(inode->i_sb); 1876 1877 dir->i_size -= BOGO_DIRENT_SIZE; 1878 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1879 drop_nlink(inode); 1880 dput(dentry); /* Undo the count from "create" - this does all the work */ 1881 return 0; 1882} 1883 1884static int shmem_rmdir(struct inode *dir, struct dentry *dentry) 1885{ 1886 if (!simple_empty(dentry)) 1887 return -ENOTEMPTY; 1888 1889 drop_nlink(dentry->d_inode); 1890 drop_nlink(dir); 1891 return shmem_unlink(dir, dentry); 1892} 1893 1894/* 1895 * The VFS layer already does all the dentry stuff for rename, 1896 * we just have to decrement the usage count for the target if 1897 * it exists so that the VFS layer correctly free's it when it 1898 * gets overwritten. 1899 */ 1900static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) 1901{ 1902 struct inode *inode = old_dentry->d_inode; 1903 int they_are_dirs = S_ISDIR(inode->i_mode); 1904 1905 if (!simple_empty(new_dentry)) 1906 return -ENOTEMPTY; 1907 1908 if (new_dentry->d_inode) { 1909 (void) shmem_unlink(new_dir, new_dentry); 1910 if (they_are_dirs) 1911 drop_nlink(old_dir); 1912 } else if (they_are_dirs) { 1913 drop_nlink(old_dir); 1914 inc_nlink(new_dir); 1915 } 1916 1917 old_dir->i_size -= BOGO_DIRENT_SIZE; 1918 new_dir->i_size += BOGO_DIRENT_SIZE; 1919 old_dir->i_ctime = old_dir->i_mtime = 1920 new_dir->i_ctime = new_dir->i_mtime = 1921 inode->i_ctime = CURRENT_TIME; 1922 return 0; 1923} 1924 1925static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1926{ 1927 int error; 1928 int len; 1929 struct inode *inode; 1930 struct page *page = NULL; 1931 char *kaddr; 1932 struct shmem_inode_info *info; 1933 1934 len = strlen(symname) + 1; 1935 if (len > PAGE_CACHE_SIZE) 1936 return -ENAMETOOLONG; 1937 1938 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0); 1939 if (!inode) 1940 return -ENOSPC; 1941 1942 error = security_inode_init_security(inode, dir, NULL, NULL, 1943 NULL); 1944 if (error) { 1945 if (error != -EOPNOTSUPP) { 1946 iput(inode); 1947 return error; 1948 } 1949 error = 0; 1950 } 1951 1952 info = SHMEM_I(inode); 1953 inode->i_size = len-1; 1954 if (len <= (char *)inode - (char *)info) { 1955 /* do it inline */ 1956 memcpy(info, symname, len); 1957 inode->i_op = &shmem_symlink_inline_operations; 1958 } else { 1959 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); 1960 if (error) { 1961 iput(inode); 1962 return error; 1963 } 1964 unlock_page(page); 1965 inode->i_op = &shmem_symlink_inode_operations; 1966 kaddr = kmap_atomic(page, KM_USER0); 1967 memcpy(kaddr, symname, len); 1968 kunmap_atomic(kaddr, KM_USER0); 1969 set_page_dirty(page); 1970 page_cache_release(page); 1971 } 1972 if (dir->i_mode & S_ISGID) 1973 inode->i_gid = dir->i_gid; 1974 dir->i_size += BOGO_DIRENT_SIZE; 1975 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1976 d_instantiate(dentry, inode); 1977 dget(dentry); 1978 return 0; 1979} 1980 1981static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd) 1982{ 1983 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode)); 1984 return NULL; 1985} 1986 1987static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) 1988{ 1989 struct page *page = NULL; 1990 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); 1991 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page)); 1992 if (page) 1993 unlock_page(page); 1994 return page; 1995} 1996 1997static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 1998{ 1999 if (!IS_ERR(nd_get_link(nd))) { 2000 struct page *page = cookie; 2001 kunmap(page); 2002 mark_page_accessed(page); 2003 page_cache_release(page); 2004 } 2005} 2006 2007static const struct inode_operations shmem_symlink_inline_operations = { 2008 .readlink = generic_readlink, 2009 .follow_link = shmem_follow_link_inline, 2010}; 2011 2012static const struct inode_operations shmem_symlink_inode_operations = { 2013 .truncate = shmem_truncate, 2014 .readlink = generic_readlink, 2015 .follow_link = shmem_follow_link, 2016 .put_link = shmem_put_link, 2017}; 2018 2019#ifdef CONFIG_TMPFS_POSIX_ACL 2020/** 2021 * Superblocks without xattr inode operations will get security.* xattr 2022 * support from the VFS "for free". As soon as we have any other xattrs 2023 * like ACLs, we also need to implement the security.* handlers at 2024 * filesystem level, though. 2025 */ 2026 2027static size_t shmem_xattr_security_list(struct inode *inode, char *list, 2028 size_t list_len, const char *name, 2029 size_t name_len) 2030{ 2031 return security_inode_listsecurity(inode, list, list_len); 2032} 2033 2034static int shmem_xattr_security_get(struct inode *inode, const char *name, 2035 void *buffer, size_t size) 2036{ 2037 if (strcmp(name, "") == 0) 2038 return -EINVAL; 2039 return xattr_getsecurity(inode, name, buffer, size); 2040} 2041 2042static int shmem_xattr_security_set(struct inode *inode, const char *name, 2043 const void *value, size_t size, int flags) 2044{ 2045 if (strcmp(name, "") == 0) 2046 return -EINVAL; 2047 return security_inode_setsecurity(inode, name, value, size, flags); 2048} 2049 2050static struct xattr_handler shmem_xattr_security_handler = { 2051 .prefix = XATTR_SECURITY_PREFIX, 2052 .list = shmem_xattr_security_list, 2053 .get = shmem_xattr_security_get, 2054 .set = shmem_xattr_security_set, 2055}; 2056 2057static struct xattr_handler *shmem_xattr_handlers[] = { 2058 &shmem_xattr_acl_access_handler, 2059 &shmem_xattr_acl_default_handler, 2060 &shmem_xattr_security_handler, 2061 NULL 2062}; 2063#endif 2064 2065static struct dentry *shmem_get_parent(struct dentry *child) 2066{ 2067 return ERR_PTR(-ESTALE); 2068} 2069 2070static int shmem_match(struct inode *ino, void *vfh) 2071{ 2072 __u32 *fh = vfh; 2073 __u64 inum = fh[2]; 2074 inum = (inum << 32) | fh[1]; 2075 return ino->i_ino == inum && fh[0] == ino->i_generation; 2076} 2077 2078static struct dentry *shmem_fh_to_dentry(struct super_block *sb, 2079 struct fid *fid, int fh_len, int fh_type) 2080{ 2081 struct inode *inode; 2082 struct dentry *dentry = NULL; 2083 u64 inum = fid->raw[2]; 2084 inum = (inum << 32) | fid->raw[1]; 2085 2086 if (fh_len < 3) 2087 return NULL; 2088 2089 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), 2090 shmem_match, fid->raw); 2091 if (inode) { 2092 dentry = d_find_alias(inode); 2093 iput(inode); 2094 } 2095 2096 return dentry; 2097} 2098 2099static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len, 2100 int connectable) 2101{ 2102 struct inode *inode = dentry->d_inode; 2103 2104 if (*len < 3) 2105 return 255; 2106 2107 if (hlist_unhashed(&inode->i_hash)) { 2108 /* Unfortunately insert_inode_hash is not idempotent, 2109 * so as we hash inodes here rather than at creation 2110 * time, we need a lock to ensure we only try 2111 * to do it once 2112 */ 2113 static DEFINE_SPINLOCK(lock); 2114 spin_lock(&lock); 2115 if (hlist_unhashed(&inode->i_hash)) 2116 __insert_inode_hash(inode, 2117 inode->i_ino + inode->i_generation); 2118 spin_unlock(&lock); 2119 } 2120 2121 fh[0] = inode->i_generation; 2122 fh[1] = inode->i_ino; 2123 fh[2] = ((__u64)inode->i_ino) >> 32; 2124 2125 *len = 3; 2126 return 1; 2127} 2128 2129static const struct export_operations shmem_export_ops = { 2130 .get_parent = shmem_get_parent, 2131 .encode_fh = shmem_encode_fh, 2132 .fh_to_dentry = shmem_fh_to_dentry, 2133}; 2134 2135static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, 2136 bool remount) 2137{ 2138 char *this_char, *value, *rest; 2139 2140 while (options != NULL) { 2141 this_char = options; 2142 for (;;) { 2143 /* 2144 * NUL-terminate this option: unfortunately, 2145 * mount options form a comma-separated list, 2146 * but mpol's nodelist may also contain commas. 2147 */ 2148 options = strchr(options, ','); 2149 if (options == NULL) 2150 break; 2151 options++; 2152 if (!isdigit(*options)) { 2153 options[-1] = '\0'; 2154 break; 2155 } 2156 } 2157 if (!*this_char) 2158 continue; 2159 if ((value = strchr(this_char,'=')) != NULL) { 2160 *value++ = 0; 2161 } else { 2162 printk(KERN_ERR 2163 "tmpfs: No value for mount option '%s'\n", 2164 this_char); 2165 return 1; 2166 } 2167 2168 if (!strcmp(this_char,"size")) { 2169 unsigned long long size; 2170 size = memparse(value,&rest); 2171 if (*rest == '%') { 2172 size <<= PAGE_SHIFT; 2173 size *= totalram_pages; 2174 do_div(size, 100); 2175 rest++; 2176 } 2177 if (*rest) 2178 goto bad_val; 2179 sbinfo->max_blocks = 2180 DIV_ROUND_UP(size, PAGE_CACHE_SIZE); 2181 } else if (!strcmp(this_char,"nr_blocks")) { 2182 sbinfo->max_blocks = memparse(value, &rest); 2183 if (*rest) 2184 goto bad_val; 2185 } else if (!strcmp(this_char,"nr_inodes")) { 2186 sbinfo->max_inodes = memparse(value, &rest); 2187 if (*rest) 2188 goto bad_val; 2189 } else if (!strcmp(this_char,"mode")) { 2190 if (remount) 2191 continue; 2192 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; 2193 if (*rest) 2194 goto bad_val; 2195 } else if (!strcmp(this_char,"uid")) { 2196 if (remount) 2197 continue; 2198 sbinfo->uid = simple_strtoul(value, &rest, 0); 2199 if (*rest) 2200 goto bad_val; 2201 } else if (!strcmp(this_char,"gid")) { 2202 if (remount) 2203 continue; 2204 sbinfo->gid = simple_strtoul(value, &rest, 0); 2205 if (*rest) 2206 goto bad_val; 2207 } else if (!strcmp(this_char,"mpol")) { 2208 if (shmem_parse_mpol(value, &sbinfo->policy, 2209 &sbinfo->policy_nodes)) 2210 goto bad_val; 2211 } else { 2212 printk(KERN_ERR "tmpfs: Bad mount option %s\n", 2213 this_char); 2214 return 1; 2215 } 2216 } 2217 return 0; 2218 2219bad_val: 2220 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", 2221 value, this_char); 2222 return 1; 2223 2224} 2225 2226static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) 2227{ 2228 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 2229 struct shmem_sb_info config = *sbinfo; 2230 unsigned long blocks; 2231 unsigned long inodes; 2232 int error = -EINVAL; 2233 2234 if (shmem_parse_options(data, &config, true)) 2235 return error; 2236 2237 spin_lock(&sbinfo->stat_lock); 2238 blocks = sbinfo->max_blocks - sbinfo->free_blocks; 2239 inodes = sbinfo->max_inodes - sbinfo->free_inodes; 2240 if (config.max_blocks < blocks) 2241 goto out; 2242 if (config.max_inodes < inodes) 2243 goto out; 2244 /* 2245 * Those tests also disallow limited->unlimited while any are in 2246 * use, so i_blocks will always be zero when max_blocks is zero; 2247 * but we must separately disallow unlimited->limited, because 2248 * in that case we have no record of how much is already in use. 2249 */ 2250 if (config.max_blocks && !sbinfo->max_blocks) 2251 goto out; 2252 if (config.max_inodes && !sbinfo->max_inodes) 2253 goto out; 2254 2255 error = 0; 2256 sbinfo->max_blocks = config.max_blocks; 2257 sbinfo->free_blocks = config.max_blocks - blocks; 2258 sbinfo->max_inodes = config.max_inodes; 2259 sbinfo->free_inodes = config.max_inodes - inodes; 2260 sbinfo->policy = config.policy; 2261 sbinfo->policy_nodes = config.policy_nodes; 2262out: 2263 spin_unlock(&sbinfo->stat_lock); 2264 return error; 2265} 2266 2267static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs) 2268{ 2269 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb); 2270 2271 if (sbinfo->max_blocks != shmem_default_max_blocks()) 2272 seq_printf(seq, ",size=%luk", 2273 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10)); 2274 if (sbinfo->max_inodes != shmem_default_max_inodes()) 2275 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); 2276 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) 2277 seq_printf(seq, ",mode=%03o", sbinfo->mode); 2278 if (sbinfo->uid != 0) 2279 seq_printf(seq, ",uid=%u", sbinfo->uid); 2280 if (sbinfo->gid != 0) 2281 seq_printf(seq, ",gid=%u", sbinfo->gid); 2282 shmem_show_mpol(seq, sbinfo->policy, sbinfo->policy_nodes); 2283 return 0; 2284} 2285#endif /* CONFIG_TMPFS */ 2286 2287static void shmem_put_super(struct super_block *sb) 2288{ 2289 kfree(sb->s_fs_info); 2290 sb->s_fs_info = NULL; 2291} 2292 2293static int shmem_fill_super(struct super_block *sb, 2294 void *data, int silent) 2295{ 2296 struct inode *inode; 2297 struct dentry *root; 2298 struct shmem_sb_info *sbinfo; 2299 int err = -ENOMEM; 2300 2301 /* Round up to L1_CACHE_BYTES to resist false sharing */ 2302 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info), 2303 L1_CACHE_BYTES), GFP_KERNEL); 2304 if (!sbinfo) 2305 return -ENOMEM; 2306 2307 sbinfo->max_blocks = 0; 2308 sbinfo->max_inodes = 0; 2309 sbinfo->mode = S_IRWXUGO | S_ISVTX; 2310 sbinfo->uid = current->fsuid; 2311 sbinfo->gid = current->fsgid; 2312 sbinfo->policy = MPOL_DEFAULT; 2313 sbinfo->policy_nodes = node_states[N_HIGH_MEMORY]; 2314 sb->s_fs_info = sbinfo; 2315 2316#ifdef CONFIG_TMPFS 2317 /* 2318 * Per default we only allow half of the physical ram per 2319 * tmpfs instance, limiting inodes to one per page of lowmem; 2320 * but the internal instance is left unlimited. 2321 */ 2322 if (!(sb->s_flags & MS_NOUSER)) { 2323 sbinfo->max_blocks = shmem_default_max_blocks(); 2324 sbinfo->max_inodes = shmem_default_max_inodes(); 2325 if (shmem_parse_options(data, sbinfo, false)) { 2326 err = -EINVAL; 2327 goto failed; 2328 } 2329 } 2330 sb->s_export_op = &shmem_export_ops; 2331#else 2332 sb->s_flags |= MS_NOUSER; 2333#endif 2334 2335 spin_lock_init(&sbinfo->stat_lock); 2336 sbinfo->free_blocks = sbinfo->max_blocks; 2337 sbinfo->free_inodes = sbinfo->max_inodes; 2338 2339 sb->s_maxbytes = SHMEM_MAX_BYTES; 2340 sb->s_blocksize = PAGE_CACHE_SIZE; 2341 sb->s_blocksize_bits = PAGE_CACHE_SHIFT; 2342 sb->s_magic = TMPFS_MAGIC; 2343 sb->s_op = &shmem_ops; 2344 sb->s_time_gran = 1; 2345#ifdef CONFIG_TMPFS_POSIX_ACL 2346 sb->s_xattr = shmem_xattr_handlers; 2347 sb->s_flags |= MS_POSIXACL; 2348#endif 2349 2350 inode = shmem_get_inode(sb, S_IFDIR | sbinfo->mode, 0); 2351 if (!inode) 2352 goto failed; 2353 inode->i_uid = sbinfo->uid; 2354 inode->i_gid = sbinfo->gid; 2355 root = d_alloc_root(inode); 2356 if (!root) 2357 goto failed_iput; 2358 sb->s_root = root; 2359 return 0; 2360 2361failed_iput: 2362 iput(inode); 2363failed: 2364 shmem_put_super(sb); 2365 return err; 2366} 2367 2368static struct kmem_cache *shmem_inode_cachep; 2369 2370static struct inode *shmem_alloc_inode(struct super_block *sb) 2371{ 2372 struct shmem_inode_info *p; 2373 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); 2374 if (!p) 2375 return NULL; 2376 return &p->vfs_inode; 2377} 2378 2379static void shmem_destroy_inode(struct inode *inode) 2380{ 2381 if ((inode->i_mode & S_IFMT) == S_IFREG) { 2382 /* only struct inode is valid if it's an inline symlink */ 2383 mpol_free_shared_policy(&SHMEM_I(inode)->policy); 2384 } 2385 shmem_acl_destroy_inode(inode); 2386 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); 2387} 2388 2389static void init_once(struct kmem_cache *cachep, void *foo) 2390{ 2391 struct shmem_inode_info *p = (struct shmem_inode_info *) foo; 2392 2393 inode_init_once(&p->vfs_inode); 2394#ifdef CONFIG_TMPFS_POSIX_ACL 2395 p->i_acl = NULL; 2396 p->i_default_acl = NULL; 2397#endif 2398} 2399 2400static int init_inodecache(void) 2401{ 2402 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", 2403 sizeof(struct shmem_inode_info), 2404 0, SLAB_PANIC, init_once); 2405 return 0; 2406} 2407 2408static void destroy_inodecache(void) 2409{ 2410 kmem_cache_destroy(shmem_inode_cachep); 2411} 2412 2413static const struct address_space_operations shmem_aops = { 2414 .writepage = shmem_writepage, 2415 .set_page_dirty = __set_page_dirty_no_writeback, 2416#ifdef CONFIG_TMPFS 2417 .readpage = shmem_readpage, 2418 .write_begin = shmem_write_begin, 2419 .write_end = shmem_write_end, 2420#endif 2421 .migratepage = migrate_page, 2422}; 2423 2424static const struct file_operations shmem_file_operations = { 2425 .mmap = shmem_mmap, 2426#ifdef CONFIG_TMPFS 2427 .llseek = generic_file_llseek, 2428 .read = shmem_file_read, 2429 .write = do_sync_write, 2430 .aio_write = generic_file_aio_write, 2431 .fsync = simple_sync_file, 2432 .splice_read = generic_file_splice_read, 2433 .splice_write = generic_file_splice_write, 2434#endif 2435}; 2436 2437static const struct inode_operations shmem_inode_operations = { 2438 .truncate = shmem_truncate, 2439 .setattr = shmem_notify_change, 2440 .truncate_range = shmem_truncate_range, 2441#ifdef CONFIG_TMPFS_POSIX_ACL 2442 .setxattr = generic_setxattr, 2443 .getxattr = generic_getxattr, 2444 .listxattr = generic_listxattr, 2445 .removexattr = generic_removexattr, 2446 .permission = shmem_permission, 2447#endif 2448 2449}; 2450 2451static const struct inode_operations shmem_dir_inode_operations = { 2452#ifdef CONFIG_TMPFS 2453 .create = shmem_create, 2454 .lookup = simple_lookup, 2455 .link = shmem_link, 2456 .unlink = shmem_unlink, 2457 .symlink = shmem_symlink, 2458 .mkdir = shmem_mkdir, 2459 .rmdir = shmem_rmdir, 2460 .mknod = shmem_mknod, 2461 .rename = shmem_rename, 2462#endif 2463#ifdef CONFIG_TMPFS_POSIX_ACL 2464 .setattr = shmem_notify_change, 2465 .setxattr = generic_setxattr, 2466 .getxattr = generic_getxattr, 2467 .listxattr = generic_listxattr, 2468 .removexattr = generic_removexattr, 2469 .permission = shmem_permission, 2470#endif 2471}; 2472 2473static const struct inode_operations shmem_special_inode_operations = { 2474#ifdef CONFIG_TMPFS_POSIX_ACL 2475 .setattr = shmem_notify_change, 2476 .setxattr = generic_setxattr, 2477 .getxattr = generic_getxattr, 2478 .listxattr = generic_listxattr, 2479 .removexattr = generic_removexattr, 2480 .permission = shmem_permission, 2481#endif 2482}; 2483 2484static const struct super_operations shmem_ops = { 2485 .alloc_inode = shmem_alloc_inode, 2486 .destroy_inode = shmem_destroy_inode, 2487#ifdef CONFIG_TMPFS 2488 .statfs = shmem_statfs, 2489 .remount_fs = shmem_remount_fs, 2490 .show_options = shmem_show_options, 2491#endif 2492 .delete_inode = shmem_delete_inode, 2493 .drop_inode = generic_delete_inode, 2494 .put_super = shmem_put_super, 2495}; 2496 2497static struct vm_operations_struct shmem_vm_ops = { 2498 .fault = shmem_fault, 2499#ifdef CONFIG_NUMA 2500 .set_policy = shmem_set_policy, 2501 .get_policy = shmem_get_policy, 2502#endif 2503}; 2504 2505 2506static int shmem_get_sb(struct file_system_type *fs_type, 2507 int flags, const char *dev_name, void *data, struct vfsmount *mnt) 2508{ 2509 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt); 2510} 2511 2512static struct file_system_type tmpfs_fs_type = { 2513 .owner = THIS_MODULE, 2514 .name = "tmpfs", 2515 .get_sb = shmem_get_sb, 2516 .kill_sb = kill_litter_super, 2517}; 2518static struct vfsmount *shm_mnt; 2519 2520static int __init init_tmpfs(void) 2521{ 2522 int error; 2523 2524 error = bdi_init(&shmem_backing_dev_info); 2525 if (error) 2526 goto out4; 2527 2528 error = init_inodecache(); 2529 if (error) 2530 goto out3; 2531 2532 error = register_filesystem(&tmpfs_fs_type); 2533 if (error) { 2534 printk(KERN_ERR "Could not register tmpfs\n"); 2535 goto out2; 2536 } 2537 2538 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER, 2539 tmpfs_fs_type.name, NULL); 2540 if (IS_ERR(shm_mnt)) { 2541 error = PTR_ERR(shm_mnt); 2542 printk(KERN_ERR "Could not kern_mount tmpfs\n"); 2543 goto out1; 2544 } 2545 return 0; 2546 2547out1: 2548 unregister_filesystem(&tmpfs_fs_type); 2549out2: 2550 destroy_inodecache(); 2551out3: 2552 bdi_destroy(&shmem_backing_dev_info); 2553out4: 2554 shm_mnt = ERR_PTR(error); 2555 return error; 2556} 2557module_init(init_tmpfs) 2558 2559/* 2560 * shmem_file_setup - get an unlinked file living in tmpfs 2561 * 2562 * @name: name for dentry (to be seen in /proc/<pid>/maps 2563 * @size: size to be set for the file 2564 * 2565 */ 2566struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags) 2567{ 2568 int error; 2569 struct file *file; 2570 struct inode *inode; 2571 struct dentry *dentry, *root; 2572 struct qstr this; 2573 2574 if (IS_ERR(shm_mnt)) 2575 return (void *)shm_mnt; 2576 2577 if (size < 0 || size > SHMEM_MAX_BYTES) 2578 return ERR_PTR(-EINVAL); 2579 2580 if (shmem_acct_size(flags, size)) 2581 return ERR_PTR(-ENOMEM); 2582 2583 error = -ENOMEM; 2584 this.name = name; 2585 this.len = strlen(name); 2586 this.hash = 0; /* will go */ 2587 root = shm_mnt->mnt_root; 2588 dentry = d_alloc(root, &this); 2589 if (!dentry) 2590 goto put_memory; 2591 2592 error = -ENFILE; 2593 file = get_empty_filp(); 2594 if (!file) 2595 goto put_dentry; 2596 2597 error = -ENOSPC; 2598 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0); 2599 if (!inode) 2600 goto close_file; 2601 2602 SHMEM_I(inode)->flags = flags & VM_ACCOUNT; 2603 d_instantiate(dentry, inode); 2604 inode->i_size = size; 2605 inode->i_nlink = 0; /* It is unlinked */ 2606 init_file(file, shm_mnt, dentry, FMODE_WRITE | FMODE_READ, 2607 &shmem_file_operations); 2608 return file; 2609 2610close_file: 2611 put_filp(file); 2612put_dentry: 2613 dput(dentry); 2614put_memory: 2615 shmem_unacct_size(flags, size); 2616 return ERR_PTR(error); 2617} 2618 2619/* 2620 * shmem_zero_setup - setup a shared anonymous mapping 2621 * 2622 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff 2623 */ 2624int shmem_zero_setup(struct vm_area_struct *vma) 2625{ 2626 struct file *file; 2627 loff_t size = vma->vm_end - vma->vm_start; 2628 2629 file = shmem_file_setup("dev/zero", size, vma->vm_flags); 2630 if (IS_ERR(file)) 2631 return PTR_ERR(file); 2632 2633 if (vma->vm_file) 2634 fput(vma->vm_file); 2635 vma->vm_file = file; 2636 vma->vm_ops = &shmem_vm_ops; 2637 return 0; 2638} 2639