shmem.c revision a569425512253992cc64ebf8b6d00a62f986db3e
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}; 87 88static int shmem_getpage(struct inode *inode, unsigned long idx, 89 struct page **pagep, enum sgp_type sgp, int *type); 90 91static inline struct page *shmem_dir_alloc(gfp_t gfp_mask) 92{ 93 /* 94 * The above definition of ENTRIES_PER_PAGE, and the use of 95 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE: 96 * might be reconsidered if it ever diverges from PAGE_SIZE. 97 * 98 * __GFP_MOVABLE is masked out as swap vectors cannot move 99 */ 100 return alloc_pages((gfp_mask & ~__GFP_MOVABLE) | __GFP_ZERO, 101 PAGE_CACHE_SHIFT-PAGE_SHIFT); 102} 103 104static inline void shmem_dir_free(struct page *page) 105{ 106 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT); 107} 108 109static struct page **shmem_dir_map(struct page *page) 110{ 111 return (struct page **)kmap_atomic(page, KM_USER0); 112} 113 114static inline void shmem_dir_unmap(struct page **dir) 115{ 116 kunmap_atomic(dir, KM_USER0); 117} 118 119static swp_entry_t *shmem_swp_map(struct page *page) 120{ 121 return (swp_entry_t *)kmap_atomic(page, KM_USER1); 122} 123 124static inline void shmem_swp_balance_unmap(void) 125{ 126 /* 127 * When passing a pointer to an i_direct entry, to code which 128 * also handles indirect entries and so will shmem_swp_unmap, 129 * we must arrange for the preempt count to remain in balance. 130 * What kmap_atomic of a lowmem page does depends on config 131 * and architecture, so pretend to kmap_atomic some lowmem page. 132 */ 133 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1); 134} 135 136static inline void shmem_swp_unmap(swp_entry_t *entry) 137{ 138 kunmap_atomic(entry, KM_USER1); 139} 140 141static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) 142{ 143 return sb->s_fs_info; 144} 145 146/* 147 * shmem_file_setup pre-accounts the whole fixed size of a VM object, 148 * for shared memory and for shared anonymous (/dev/zero) mappings 149 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), 150 * consistent with the pre-accounting of private mappings ... 151 */ 152static inline int shmem_acct_size(unsigned long flags, loff_t size) 153{ 154 return (flags & VM_ACCOUNT)? 155 security_vm_enough_memory(VM_ACCT(size)): 0; 156} 157 158static inline void shmem_unacct_size(unsigned long flags, loff_t size) 159{ 160 if (flags & VM_ACCOUNT) 161 vm_unacct_memory(VM_ACCT(size)); 162} 163 164/* 165 * ... whereas tmpfs objects are accounted incrementally as 166 * pages are allocated, in order to allow huge sparse files. 167 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, 168 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. 169 */ 170static inline int shmem_acct_block(unsigned long flags) 171{ 172 return (flags & VM_ACCOUNT)? 173 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE)); 174} 175 176static inline void shmem_unacct_blocks(unsigned long flags, long pages) 177{ 178 if (!(flags & VM_ACCOUNT)) 179 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE)); 180} 181 182static const struct super_operations shmem_ops; 183static const struct address_space_operations shmem_aops; 184static const struct file_operations shmem_file_operations; 185static const struct inode_operations shmem_inode_operations; 186static const struct inode_operations shmem_dir_inode_operations; 187static const struct inode_operations shmem_special_inode_operations; 188static struct vm_operations_struct shmem_vm_ops; 189 190static struct backing_dev_info shmem_backing_dev_info __read_mostly = { 191 .ra_pages = 0, /* No readahead */ 192 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK, 193 .unplug_io_fn = default_unplug_io_fn, 194}; 195 196static LIST_HEAD(shmem_swaplist); 197static DEFINE_SPINLOCK(shmem_swaplist_lock); 198 199static void shmem_free_blocks(struct inode *inode, long pages) 200{ 201 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 202 if (sbinfo->max_blocks) { 203 spin_lock(&sbinfo->stat_lock); 204 sbinfo->free_blocks += pages; 205 inode->i_blocks -= pages*BLOCKS_PER_PAGE; 206 spin_unlock(&sbinfo->stat_lock); 207 } 208} 209 210/* 211 * shmem_recalc_inode - recalculate the size of an inode 212 * 213 * @inode: inode to recalc 214 * 215 * We have to calculate the free blocks since the mm can drop 216 * undirtied hole pages behind our back. 217 * 218 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped 219 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) 220 * 221 * It has to be called with the spinlock held. 222 */ 223static void shmem_recalc_inode(struct inode *inode) 224{ 225 struct shmem_inode_info *info = SHMEM_I(inode); 226 long freed; 227 228 freed = info->alloced - info->swapped - inode->i_mapping->nrpages; 229 if (freed > 0) { 230 info->alloced -= freed; 231 shmem_unacct_blocks(info->flags, freed); 232 shmem_free_blocks(inode, freed); 233 } 234} 235 236/* 237 * shmem_swp_entry - find the swap vector position in the info structure 238 * 239 * @info: info structure for the inode 240 * @index: index of the page to find 241 * @page: optional page to add to the structure. Has to be preset to 242 * all zeros 243 * 244 * If there is no space allocated yet it will return NULL when 245 * page is NULL, else it will use the page for the needed block, 246 * setting it to NULL on return to indicate that it has been used. 247 * 248 * The swap vector is organized the following way: 249 * 250 * There are SHMEM_NR_DIRECT entries directly stored in the 251 * shmem_inode_info structure. So small files do not need an addional 252 * allocation. 253 * 254 * For pages with index > SHMEM_NR_DIRECT there is the pointer 255 * i_indirect which points to a page which holds in the first half 256 * doubly indirect blocks, in the second half triple indirect blocks: 257 * 258 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the 259 * following layout (for SHMEM_NR_DIRECT == 16): 260 * 261 * i_indirect -> dir --> 16-19 262 * | +-> 20-23 263 * | 264 * +-->dir2 --> 24-27 265 * | +-> 28-31 266 * | +-> 32-35 267 * | +-> 36-39 268 * | 269 * +-->dir3 --> 40-43 270 * +-> 44-47 271 * +-> 48-51 272 * +-> 52-55 273 */ 274static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page) 275{ 276 unsigned long offset; 277 struct page **dir; 278 struct page *subdir; 279 280 if (index < SHMEM_NR_DIRECT) { 281 shmem_swp_balance_unmap(); 282 return info->i_direct+index; 283 } 284 if (!info->i_indirect) { 285 if (page) { 286 info->i_indirect = *page; 287 *page = NULL; 288 } 289 return NULL; /* need another page */ 290 } 291 292 index -= SHMEM_NR_DIRECT; 293 offset = index % ENTRIES_PER_PAGE; 294 index /= ENTRIES_PER_PAGE; 295 dir = shmem_dir_map(info->i_indirect); 296 297 if (index >= ENTRIES_PER_PAGE/2) { 298 index -= ENTRIES_PER_PAGE/2; 299 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE; 300 index %= ENTRIES_PER_PAGE; 301 subdir = *dir; 302 if (!subdir) { 303 if (page) { 304 *dir = *page; 305 *page = NULL; 306 } 307 shmem_dir_unmap(dir); 308 return NULL; /* need another page */ 309 } 310 shmem_dir_unmap(dir); 311 dir = shmem_dir_map(subdir); 312 } 313 314 dir += index; 315 subdir = *dir; 316 if (!subdir) { 317 if (!page || !(subdir = *page)) { 318 shmem_dir_unmap(dir); 319 return NULL; /* need a page */ 320 } 321 *dir = subdir; 322 *page = NULL; 323 } 324 shmem_dir_unmap(dir); 325 return shmem_swp_map(subdir) + offset; 326} 327 328static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value) 329{ 330 long incdec = value? 1: -1; 331 332 entry->val = value; 333 info->swapped += incdec; 334 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) { 335 struct page *page = kmap_atomic_to_page(entry); 336 set_page_private(page, page_private(page) + incdec); 337 } 338} 339 340/* 341 * shmem_swp_alloc - get the position of the swap entry for the page. 342 * If it does not exist allocate the entry. 343 * 344 * @info: info structure for the inode 345 * @index: index of the page to find 346 * @sgp: check and recheck i_size? skip allocation? 347 */ 348static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp) 349{ 350 struct inode *inode = &info->vfs_inode; 351 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 352 struct page *page = NULL; 353 swp_entry_t *entry; 354 355 if (sgp != SGP_WRITE && 356 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 357 return ERR_PTR(-EINVAL); 358 359 while (!(entry = shmem_swp_entry(info, index, &page))) { 360 if (sgp == SGP_READ) 361 return shmem_swp_map(ZERO_PAGE(0)); 362 /* 363 * Test free_blocks against 1 not 0, since we have 1 data 364 * page (and perhaps indirect index pages) yet to allocate: 365 * a waste to allocate index if we cannot allocate data. 366 */ 367 if (sbinfo->max_blocks) { 368 spin_lock(&sbinfo->stat_lock); 369 if (sbinfo->free_blocks <= 1) { 370 spin_unlock(&sbinfo->stat_lock); 371 return ERR_PTR(-ENOSPC); 372 } 373 sbinfo->free_blocks--; 374 inode->i_blocks += BLOCKS_PER_PAGE; 375 spin_unlock(&sbinfo->stat_lock); 376 } 377 378 spin_unlock(&info->lock); 379 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping)); 380 if (page) 381 set_page_private(page, 0); 382 spin_lock(&info->lock); 383 384 if (!page) { 385 shmem_free_blocks(inode, 1); 386 return ERR_PTR(-ENOMEM); 387 } 388 if (sgp != SGP_WRITE && 389 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { 390 entry = ERR_PTR(-EINVAL); 391 break; 392 } 393 if (info->next_index <= index) 394 info->next_index = index + 1; 395 } 396 if (page) { 397 /* another task gave its page, or truncated the file */ 398 shmem_free_blocks(inode, 1); 399 shmem_dir_free(page); 400 } 401 if (info->next_index <= index && !IS_ERR(entry)) 402 info->next_index = index + 1; 403 return entry; 404} 405 406/* 407 * shmem_free_swp - free some swap entries in a directory 408 * 409 * @dir: pointer to the directory 410 * @edir: pointer after last entry of the directory 411 * @punch_lock: pointer to spinlock when needed for the holepunch case 412 */ 413static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir, 414 spinlock_t *punch_lock) 415{ 416 spinlock_t *punch_unlock = NULL; 417 swp_entry_t *ptr; 418 int freed = 0; 419 420 for (ptr = dir; ptr < edir; ptr++) { 421 if (ptr->val) { 422 if (unlikely(punch_lock)) { 423 punch_unlock = punch_lock; 424 punch_lock = NULL; 425 spin_lock(punch_unlock); 426 if (!ptr->val) 427 continue; 428 } 429 free_swap_and_cache(*ptr); 430 *ptr = (swp_entry_t){0}; 431 freed++; 432 } 433 } 434 if (punch_unlock) 435 spin_unlock(punch_unlock); 436 return freed; 437} 438 439static int shmem_map_and_free_swp(struct page *subdir, int offset, 440 int limit, struct page ***dir, spinlock_t *punch_lock) 441{ 442 swp_entry_t *ptr; 443 int freed = 0; 444 445 ptr = shmem_swp_map(subdir); 446 for (; offset < limit; offset += LATENCY_LIMIT) { 447 int size = limit - offset; 448 if (size > LATENCY_LIMIT) 449 size = LATENCY_LIMIT; 450 freed += shmem_free_swp(ptr+offset, ptr+offset+size, 451 punch_lock); 452 if (need_resched()) { 453 shmem_swp_unmap(ptr); 454 if (*dir) { 455 shmem_dir_unmap(*dir); 456 *dir = NULL; 457 } 458 cond_resched(); 459 ptr = shmem_swp_map(subdir); 460 } 461 } 462 shmem_swp_unmap(ptr); 463 return freed; 464} 465 466static void shmem_free_pages(struct list_head *next) 467{ 468 struct page *page; 469 int freed = 0; 470 471 do { 472 page = container_of(next, struct page, lru); 473 next = next->next; 474 shmem_dir_free(page); 475 freed++; 476 if (freed >= LATENCY_LIMIT) { 477 cond_resched(); 478 freed = 0; 479 } 480 } while (next); 481} 482 483static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end) 484{ 485 struct shmem_inode_info *info = SHMEM_I(inode); 486 unsigned long idx; 487 unsigned long size; 488 unsigned long limit; 489 unsigned long stage; 490 unsigned long diroff; 491 struct page **dir; 492 struct page *topdir; 493 struct page *middir; 494 struct page *subdir; 495 swp_entry_t *ptr; 496 LIST_HEAD(pages_to_free); 497 long nr_pages_to_free = 0; 498 long nr_swaps_freed = 0; 499 int offset; 500 int freed; 501 int punch_hole; 502 spinlock_t *needs_lock; 503 spinlock_t *punch_lock; 504 unsigned long upper_limit; 505 506 inode->i_ctime = inode->i_mtime = CURRENT_TIME; 507 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 508 if (idx >= info->next_index) 509 return; 510 511 spin_lock(&info->lock); 512 info->flags |= SHMEM_TRUNCATE; 513 if (likely(end == (loff_t) -1)) { 514 limit = info->next_index; 515 upper_limit = SHMEM_MAX_INDEX; 516 info->next_index = idx; 517 needs_lock = NULL; 518 punch_hole = 0; 519 } else { 520 if (end + 1 >= inode->i_size) { /* we may free a little more */ 521 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >> 522 PAGE_CACHE_SHIFT; 523 upper_limit = SHMEM_MAX_INDEX; 524 } else { 525 limit = (end + 1) >> PAGE_CACHE_SHIFT; 526 upper_limit = limit; 527 } 528 needs_lock = &info->lock; 529 punch_hole = 1; 530 } 531 532 topdir = info->i_indirect; 533 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) { 534 info->i_indirect = NULL; 535 nr_pages_to_free++; 536 list_add(&topdir->lru, &pages_to_free); 537 } 538 spin_unlock(&info->lock); 539 540 if (info->swapped && idx < SHMEM_NR_DIRECT) { 541 ptr = info->i_direct; 542 size = limit; 543 if (size > SHMEM_NR_DIRECT) 544 size = SHMEM_NR_DIRECT; 545 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock); 546 } 547 548 /* 549 * If there are no indirect blocks or we are punching a hole 550 * below indirect blocks, nothing to be done. 551 */ 552 if (!topdir || limit <= SHMEM_NR_DIRECT) 553 goto done2; 554 555 /* 556 * The truncation case has already dropped info->lock, and we're safe 557 * because i_size and next_index have already been lowered, preventing 558 * access beyond. But in the punch_hole case, we still need to take 559 * the lock when updating the swap directory, because there might be 560 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or 561 * shmem_writepage. However, whenever we find we can remove a whole 562 * directory page (not at the misaligned start or end of the range), 563 * we first NULLify its pointer in the level above, and then have no 564 * need to take the lock when updating its contents: needs_lock and 565 * punch_lock (either pointing to info->lock or NULL) manage this. 566 */ 567 568 upper_limit -= SHMEM_NR_DIRECT; 569 limit -= SHMEM_NR_DIRECT; 570 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0; 571 offset = idx % ENTRIES_PER_PAGE; 572 idx -= offset; 573 574 dir = shmem_dir_map(topdir); 575 stage = ENTRIES_PER_PAGEPAGE/2; 576 if (idx < ENTRIES_PER_PAGEPAGE/2) { 577 middir = topdir; 578 diroff = idx/ENTRIES_PER_PAGE; 579 } else { 580 dir += ENTRIES_PER_PAGE/2; 581 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE; 582 while (stage <= idx) 583 stage += ENTRIES_PER_PAGEPAGE; 584 middir = *dir; 585 if (*dir) { 586 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) % 587 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE; 588 if (!diroff && !offset && upper_limit >= stage) { 589 if (needs_lock) { 590 spin_lock(needs_lock); 591 *dir = NULL; 592 spin_unlock(needs_lock); 593 needs_lock = NULL; 594 } else 595 *dir = NULL; 596 nr_pages_to_free++; 597 list_add(&middir->lru, &pages_to_free); 598 } 599 shmem_dir_unmap(dir); 600 dir = shmem_dir_map(middir); 601 } else { 602 diroff = 0; 603 offset = 0; 604 idx = stage; 605 } 606 } 607 608 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) { 609 if (unlikely(idx == stage)) { 610 shmem_dir_unmap(dir); 611 dir = shmem_dir_map(topdir) + 612 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 613 while (!*dir) { 614 dir++; 615 idx += ENTRIES_PER_PAGEPAGE; 616 if (idx >= limit) 617 goto done1; 618 } 619 stage = idx + ENTRIES_PER_PAGEPAGE; 620 middir = *dir; 621 if (punch_hole) 622 needs_lock = &info->lock; 623 if (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 cond_resched(); 636 dir = shmem_dir_map(middir); 637 diroff = 0; 638 } 639 punch_lock = needs_lock; 640 subdir = dir[diroff]; 641 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) { 642 if (needs_lock) { 643 spin_lock(needs_lock); 644 dir[diroff] = NULL; 645 spin_unlock(needs_lock); 646 punch_lock = NULL; 647 } else 648 dir[diroff] = NULL; 649 nr_pages_to_free++; 650 list_add(&subdir->lru, &pages_to_free); 651 } 652 if (subdir && page_private(subdir) /* has swap entries */) { 653 size = limit - idx; 654 if (size > ENTRIES_PER_PAGE) 655 size = ENTRIES_PER_PAGE; 656 freed = shmem_map_and_free_swp(subdir, 657 offset, size, &dir, punch_lock); 658 if (!dir) 659 dir = shmem_dir_map(middir); 660 nr_swaps_freed += freed; 661 if (offset || punch_lock) { 662 spin_lock(&info->lock); 663 set_page_private(subdir, 664 page_private(subdir) - freed); 665 spin_unlock(&info->lock); 666 } else 667 BUG_ON(page_private(subdir) != freed); 668 } 669 offset = 0; 670 } 671done1: 672 shmem_dir_unmap(dir); 673done2: 674 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) { 675 /* 676 * Call truncate_inode_pages again: racing shmem_unuse_inode 677 * may have swizzled a page in from swap since vmtruncate or 678 * generic_delete_inode did it, before we lowered next_index. 679 * Also, though shmem_getpage checks i_size before adding to 680 * cache, no recheck after: so fix the narrow window there too. 681 * 682 * Recalling truncate_inode_pages_range and unmap_mapping_range 683 * every time for punch_hole (which never got a chance to clear 684 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive, 685 * yet hardly ever necessary: try to optimize them out later. 686 */ 687 truncate_inode_pages_range(inode->i_mapping, start, end); 688 if (punch_hole) 689 unmap_mapping_range(inode->i_mapping, start, 690 end - start, 1); 691 } 692 693 spin_lock(&info->lock); 694 info->flags &= ~SHMEM_TRUNCATE; 695 info->swapped -= nr_swaps_freed; 696 if (nr_pages_to_free) 697 shmem_free_blocks(inode, nr_pages_to_free); 698 shmem_recalc_inode(inode); 699 spin_unlock(&info->lock); 700 701 /* 702 * Empty swap vector directory pages to be freed? 703 */ 704 if (!list_empty(&pages_to_free)) { 705 pages_to_free.prev->next = NULL; 706 shmem_free_pages(pages_to_free.next); 707 } 708} 709 710static void shmem_truncate(struct inode *inode) 711{ 712 shmem_truncate_range(inode, inode->i_size, (loff_t)-1); 713} 714 715static int shmem_notify_change(struct dentry *dentry, struct iattr *attr) 716{ 717 struct inode *inode = dentry->d_inode; 718 struct page *page = NULL; 719 int error; 720 721 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { 722 if (attr->ia_size < inode->i_size) { 723 /* 724 * If truncating down to a partial page, then 725 * if that page is already allocated, hold it 726 * in memory until the truncation is over, so 727 * truncate_partial_page cannnot miss it were 728 * it assigned to swap. 729 */ 730 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) { 731 (void) shmem_getpage(inode, 732 attr->ia_size>>PAGE_CACHE_SHIFT, 733 &page, SGP_READ, NULL); 734 } 735 /* 736 * Reset SHMEM_PAGEIN flag so that shmem_truncate can 737 * detect if any pages might have been added to cache 738 * after truncate_inode_pages. But we needn't bother 739 * if it's being fully truncated to zero-length: the 740 * nrpages check is efficient enough in that case. 741 */ 742 if (attr->ia_size) { 743 struct shmem_inode_info *info = SHMEM_I(inode); 744 spin_lock(&info->lock); 745 info->flags &= ~SHMEM_PAGEIN; 746 spin_unlock(&info->lock); 747 } 748 } 749 } 750 751 error = inode_change_ok(inode, attr); 752 if (!error) 753 error = inode_setattr(inode, attr); 754#ifdef CONFIG_TMPFS_POSIX_ACL 755 if (!error && (attr->ia_valid & ATTR_MODE)) 756 error = generic_acl_chmod(inode, &shmem_acl_ops); 757#endif 758 if (page) 759 page_cache_release(page); 760 return error; 761} 762 763static void shmem_delete_inode(struct inode *inode) 764{ 765 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 766 struct shmem_inode_info *info = SHMEM_I(inode); 767 768 if (inode->i_op->truncate == shmem_truncate) { 769 truncate_inode_pages(inode->i_mapping, 0); 770 shmem_unacct_size(info->flags, inode->i_size); 771 inode->i_size = 0; 772 shmem_truncate(inode); 773 if (!list_empty(&info->swaplist)) { 774 spin_lock(&shmem_swaplist_lock); 775 list_del_init(&info->swaplist); 776 spin_unlock(&shmem_swaplist_lock); 777 } 778 } 779 BUG_ON(inode->i_blocks); 780 if (sbinfo->max_inodes) { 781 spin_lock(&sbinfo->stat_lock); 782 sbinfo->free_inodes++; 783 spin_unlock(&sbinfo->stat_lock); 784 } 785 clear_inode(inode); 786} 787 788static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir) 789{ 790 swp_entry_t *ptr; 791 792 for (ptr = dir; ptr < edir; ptr++) { 793 if (ptr->val == entry.val) 794 return ptr - dir; 795 } 796 return -1; 797} 798 799static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page) 800{ 801 struct inode *inode; 802 unsigned long idx; 803 unsigned long size; 804 unsigned long limit; 805 unsigned long stage; 806 struct page **dir; 807 struct page *subdir; 808 swp_entry_t *ptr; 809 int offset; 810 811 idx = 0; 812 ptr = info->i_direct; 813 spin_lock(&info->lock); 814 limit = info->next_index; 815 size = limit; 816 if (size > SHMEM_NR_DIRECT) 817 size = SHMEM_NR_DIRECT; 818 offset = shmem_find_swp(entry, ptr, ptr+size); 819 if (offset >= 0) { 820 shmem_swp_balance_unmap(); 821 goto found; 822 } 823 if (!info->i_indirect) 824 goto lost2; 825 826 dir = shmem_dir_map(info->i_indirect); 827 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2; 828 829 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) { 830 if (unlikely(idx == stage)) { 831 shmem_dir_unmap(dir-1); 832 dir = shmem_dir_map(info->i_indirect) + 833 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 834 while (!*dir) { 835 dir++; 836 idx += ENTRIES_PER_PAGEPAGE; 837 if (idx >= limit) 838 goto lost1; 839 } 840 stage = idx + ENTRIES_PER_PAGEPAGE; 841 subdir = *dir; 842 shmem_dir_unmap(dir); 843 dir = shmem_dir_map(subdir); 844 } 845 subdir = *dir; 846 if (subdir && page_private(subdir)) { 847 ptr = shmem_swp_map(subdir); 848 size = limit - idx; 849 if (size > ENTRIES_PER_PAGE) 850 size = ENTRIES_PER_PAGE; 851 offset = shmem_find_swp(entry, ptr, ptr+size); 852 if (offset >= 0) { 853 shmem_dir_unmap(dir); 854 goto found; 855 } 856 shmem_swp_unmap(ptr); 857 } 858 } 859lost1: 860 shmem_dir_unmap(dir-1); 861lost2: 862 spin_unlock(&info->lock); 863 return 0; 864found: 865 idx += offset; 866 inode = &info->vfs_inode; 867 if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) { 868 info->flags |= SHMEM_PAGEIN; 869 shmem_swp_set(info, ptr + offset, 0); 870 } 871 shmem_swp_unmap(ptr); 872 spin_unlock(&info->lock); 873 /* 874 * Decrement swap count even when the entry is left behind: 875 * try_to_unuse will skip over mms, then reincrement count. 876 */ 877 swap_free(entry); 878 return 1; 879} 880 881/* 882 * shmem_unuse() search for an eventually swapped out shmem page. 883 */ 884int shmem_unuse(swp_entry_t entry, struct page *page) 885{ 886 struct list_head *p, *next; 887 struct shmem_inode_info *info; 888 int found = 0; 889 890 spin_lock(&shmem_swaplist_lock); 891 list_for_each_safe(p, next, &shmem_swaplist) { 892 info = list_entry(p, struct shmem_inode_info, swaplist); 893 if (!info->swapped) 894 list_del_init(&info->swaplist); 895 else if (shmem_unuse_inode(info, entry, page)) { 896 /* move head to start search for next from here */ 897 list_move_tail(&shmem_swaplist, &info->swaplist); 898 found = 1; 899 break; 900 } 901 } 902 spin_unlock(&shmem_swaplist_lock); 903 return found; 904} 905 906/* 907 * Move the page from the page cache to the swap cache. 908 */ 909static int shmem_writepage(struct page *page, struct writeback_control *wbc) 910{ 911 struct shmem_inode_info *info; 912 swp_entry_t *entry, swap; 913 struct address_space *mapping; 914 unsigned long index; 915 struct inode *inode; 916 917 BUG_ON(!PageLocked(page)); 918 BUG_ON(page_mapped(page)); 919 920 mapping = page->mapping; 921 index = page->index; 922 inode = mapping->host; 923 info = SHMEM_I(inode); 924 if (info->flags & VM_LOCKED) 925 goto redirty; 926 swap = get_swap_page(); 927 if (!swap.val) 928 goto redirty; 929 930 spin_lock(&info->lock); 931 shmem_recalc_inode(inode); 932 if (index >= info->next_index) { 933 BUG_ON(!(info->flags & SHMEM_TRUNCATE)); 934 goto unlock; 935 } 936 entry = shmem_swp_entry(info, index, NULL); 937 BUG_ON(!entry); 938 BUG_ON(entry->val); 939 940 if (move_to_swap_cache(page, swap) == 0) { 941 shmem_swp_set(info, entry, swap.val); 942 shmem_swp_unmap(entry); 943 spin_unlock(&info->lock); 944 if (list_empty(&info->swaplist)) { 945 spin_lock(&shmem_swaplist_lock); 946 /* move instead of add in case we're racing */ 947 list_move_tail(&info->swaplist, &shmem_swaplist); 948 spin_unlock(&shmem_swaplist_lock); 949 } 950 unlock_page(page); 951 return 0; 952 } 953 954 shmem_swp_unmap(entry); 955unlock: 956 spin_unlock(&info->lock); 957 swap_free(swap); 958redirty: 959 set_page_dirty(page); 960 return AOP_WRITEPAGE_ACTIVATE; /* Return with the page locked */ 961} 962 963#ifdef CONFIG_NUMA 964static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes) 965{ 966 char *nodelist = strchr(value, ':'); 967 int err = 1; 968 969 if (nodelist) { 970 /* NUL-terminate policy string */ 971 *nodelist++ = '\0'; 972 if (nodelist_parse(nodelist, *policy_nodes)) 973 goto out; 974 if (!nodes_subset(*policy_nodes, node_online_map)) 975 goto out; 976 } 977 if (!strcmp(value, "default")) { 978 *policy = MPOL_DEFAULT; 979 /* Don't allow a nodelist */ 980 if (!nodelist) 981 err = 0; 982 } else if (!strcmp(value, "prefer")) { 983 *policy = MPOL_PREFERRED; 984 /* Insist on a nodelist of one node only */ 985 if (nodelist) { 986 char *rest = nodelist; 987 while (isdigit(*rest)) 988 rest++; 989 if (!*rest) 990 err = 0; 991 } 992 } else if (!strcmp(value, "bind")) { 993 *policy = MPOL_BIND; 994 /* Insist on a nodelist */ 995 if (nodelist) 996 err = 0; 997 } else if (!strcmp(value, "interleave")) { 998 *policy = MPOL_INTERLEAVE; 999 /* Default to nodes online if no nodelist */ 1000 if (!nodelist) 1001 *policy_nodes = node_online_map; 1002 err = 0; 1003 } 1004out: 1005 /* Restore string for error message */ 1006 if (nodelist) 1007 *--nodelist = ':'; 1008 return err; 1009} 1010 1011static struct page *shmem_swapin_async(struct shared_policy *p, 1012 swp_entry_t entry, unsigned long idx) 1013{ 1014 struct page *page; 1015 struct vm_area_struct pvma; 1016 1017 /* Create a pseudo vma that just contains the policy */ 1018 memset(&pvma, 0, sizeof(struct vm_area_struct)); 1019 pvma.vm_end = PAGE_SIZE; 1020 pvma.vm_pgoff = idx; 1021 pvma.vm_policy = mpol_shared_policy_lookup(p, idx); 1022 page = read_swap_cache_async(entry, &pvma, 0); 1023 mpol_free(pvma.vm_policy); 1024 return page; 1025} 1026 1027struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry, 1028 unsigned long idx) 1029{ 1030 struct shared_policy *p = &info->policy; 1031 int i, num; 1032 struct page *page; 1033 unsigned long offset; 1034 1035 num = valid_swaphandles(entry, &offset); 1036 for (i = 0; i < num; offset++, i++) { 1037 page = shmem_swapin_async(p, 1038 swp_entry(swp_type(entry), offset), idx); 1039 if (!page) 1040 break; 1041 page_cache_release(page); 1042 } 1043 lru_add_drain(); /* Push any new pages onto the LRU now */ 1044 return shmem_swapin_async(p, entry, idx); 1045} 1046 1047static struct page * 1048shmem_alloc_page(gfp_t gfp, struct shmem_inode_info *info, 1049 unsigned long idx) 1050{ 1051 struct vm_area_struct pvma; 1052 struct page *page; 1053 1054 memset(&pvma, 0, sizeof(struct vm_area_struct)); 1055 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx); 1056 pvma.vm_pgoff = idx; 1057 pvma.vm_end = PAGE_SIZE; 1058 page = alloc_page_vma(gfp | __GFP_ZERO, &pvma, 0); 1059 mpol_free(pvma.vm_policy); 1060 return page; 1061} 1062#else 1063static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes) 1064{ 1065 return 1; 1066} 1067 1068static inline struct page * 1069shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx) 1070{ 1071 swapin_readahead(entry, 0, NULL); 1072 return read_swap_cache_async(entry, NULL, 0); 1073} 1074 1075static inline struct page * 1076shmem_alloc_page(gfp_t gfp,struct shmem_inode_info *info, unsigned long idx) 1077{ 1078 return alloc_page(gfp | __GFP_ZERO); 1079} 1080#endif 1081 1082/* 1083 * shmem_getpage - either get the page from swap or allocate a new one 1084 * 1085 * If we allocate a new one we do not mark it dirty. That's up to the 1086 * vm. If we swap it in we mark it dirty since we also free the swap 1087 * entry since a page cannot live in both the swap and page cache 1088 */ 1089static int shmem_getpage(struct inode *inode, unsigned long idx, 1090 struct page **pagep, enum sgp_type sgp, int *type) 1091{ 1092 struct address_space *mapping = inode->i_mapping; 1093 struct shmem_inode_info *info = SHMEM_I(inode); 1094 struct shmem_sb_info *sbinfo; 1095 struct page *filepage = *pagep; 1096 struct page *swappage; 1097 swp_entry_t *entry; 1098 swp_entry_t swap; 1099 int error; 1100 1101 if (idx >= SHMEM_MAX_INDEX) 1102 return -EFBIG; 1103 /* 1104 * Normally, filepage is NULL on entry, and either found 1105 * uptodate immediately, or allocated and zeroed, or read 1106 * in under swappage, which is then assigned to filepage. 1107 * But shmem_readpage and shmem_prepare_write pass in a locked 1108 * filepage, which may be found not uptodate by other callers 1109 * too, and may need to be copied from the swappage read in. 1110 */ 1111repeat: 1112 if (!filepage) 1113 filepage = find_lock_page(mapping, idx); 1114 if (filepage && PageUptodate(filepage)) 1115 goto done; 1116 error = 0; 1117 if (sgp == SGP_QUICK) 1118 goto failed; 1119 1120 spin_lock(&info->lock); 1121 shmem_recalc_inode(inode); 1122 entry = shmem_swp_alloc(info, idx, sgp); 1123 if (IS_ERR(entry)) { 1124 spin_unlock(&info->lock); 1125 error = PTR_ERR(entry); 1126 goto failed; 1127 } 1128 swap = *entry; 1129 1130 if (swap.val) { 1131 /* Look it up and read it in.. */ 1132 swappage = lookup_swap_cache(swap); 1133 if (!swappage) { 1134 shmem_swp_unmap(entry); 1135 /* here we actually do the io */ 1136 if (type && *type == VM_FAULT_MINOR) { 1137 __count_vm_event(PGMAJFAULT); 1138 *type = VM_FAULT_MAJOR; 1139 } 1140 spin_unlock(&info->lock); 1141 swappage = shmem_swapin(info, swap, idx); 1142 if (!swappage) { 1143 spin_lock(&info->lock); 1144 entry = shmem_swp_alloc(info, idx, sgp); 1145 if (IS_ERR(entry)) 1146 error = PTR_ERR(entry); 1147 else { 1148 if (entry->val == swap.val) 1149 error = -ENOMEM; 1150 shmem_swp_unmap(entry); 1151 } 1152 spin_unlock(&info->lock); 1153 if (error) 1154 goto failed; 1155 goto repeat; 1156 } 1157 wait_on_page_locked(swappage); 1158 page_cache_release(swappage); 1159 goto repeat; 1160 } 1161 1162 /* We have to do this with page locked to prevent races */ 1163 if (TestSetPageLocked(swappage)) { 1164 shmem_swp_unmap(entry); 1165 spin_unlock(&info->lock); 1166 wait_on_page_locked(swappage); 1167 page_cache_release(swappage); 1168 goto repeat; 1169 } 1170 if (PageWriteback(swappage)) { 1171 shmem_swp_unmap(entry); 1172 spin_unlock(&info->lock); 1173 wait_on_page_writeback(swappage); 1174 unlock_page(swappage); 1175 page_cache_release(swappage); 1176 goto repeat; 1177 } 1178 if (!PageUptodate(swappage)) { 1179 shmem_swp_unmap(entry); 1180 spin_unlock(&info->lock); 1181 unlock_page(swappage); 1182 page_cache_release(swappage); 1183 error = -EIO; 1184 goto failed; 1185 } 1186 1187 if (filepage) { 1188 shmem_swp_set(info, entry, 0); 1189 shmem_swp_unmap(entry); 1190 delete_from_swap_cache(swappage); 1191 spin_unlock(&info->lock); 1192 copy_highpage(filepage, swappage); 1193 unlock_page(swappage); 1194 page_cache_release(swappage); 1195 flush_dcache_page(filepage); 1196 SetPageUptodate(filepage); 1197 set_page_dirty(filepage); 1198 swap_free(swap); 1199 } else if (!(error = move_from_swap_cache( 1200 swappage, idx, mapping))) { 1201 info->flags |= SHMEM_PAGEIN; 1202 shmem_swp_set(info, entry, 0); 1203 shmem_swp_unmap(entry); 1204 spin_unlock(&info->lock); 1205 filepage = swappage; 1206 swap_free(swap); 1207 } else { 1208 shmem_swp_unmap(entry); 1209 spin_unlock(&info->lock); 1210 unlock_page(swappage); 1211 page_cache_release(swappage); 1212 if (error == -ENOMEM) { 1213 /* let kswapd refresh zone for GFP_ATOMICs */ 1214 congestion_wait(WRITE, HZ/50); 1215 } 1216 goto repeat; 1217 } 1218 } else if (sgp == SGP_READ && !filepage) { 1219 shmem_swp_unmap(entry); 1220 filepage = find_get_page(mapping, idx); 1221 if (filepage && 1222 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) { 1223 spin_unlock(&info->lock); 1224 wait_on_page_locked(filepage); 1225 page_cache_release(filepage); 1226 filepage = NULL; 1227 goto repeat; 1228 } 1229 spin_unlock(&info->lock); 1230 } else { 1231 shmem_swp_unmap(entry); 1232 sbinfo = SHMEM_SB(inode->i_sb); 1233 if (sbinfo->max_blocks) { 1234 spin_lock(&sbinfo->stat_lock); 1235 if (sbinfo->free_blocks == 0 || 1236 shmem_acct_block(info->flags)) { 1237 spin_unlock(&sbinfo->stat_lock); 1238 spin_unlock(&info->lock); 1239 error = -ENOSPC; 1240 goto failed; 1241 } 1242 sbinfo->free_blocks--; 1243 inode->i_blocks += BLOCKS_PER_PAGE; 1244 spin_unlock(&sbinfo->stat_lock); 1245 } else if (shmem_acct_block(info->flags)) { 1246 spin_unlock(&info->lock); 1247 error = -ENOSPC; 1248 goto failed; 1249 } 1250 1251 if (!filepage) { 1252 spin_unlock(&info->lock); 1253 filepage = shmem_alloc_page(mapping_gfp_mask(mapping), 1254 info, 1255 idx); 1256 if (!filepage) { 1257 shmem_unacct_blocks(info->flags, 1); 1258 shmem_free_blocks(inode, 1); 1259 error = -ENOMEM; 1260 goto failed; 1261 } 1262 1263 spin_lock(&info->lock); 1264 entry = shmem_swp_alloc(info, idx, sgp); 1265 if (IS_ERR(entry)) 1266 error = PTR_ERR(entry); 1267 else { 1268 swap = *entry; 1269 shmem_swp_unmap(entry); 1270 } 1271 if (error || swap.val || 0 != add_to_page_cache_lru( 1272 filepage, mapping, idx, GFP_ATOMIC)) { 1273 spin_unlock(&info->lock); 1274 page_cache_release(filepage); 1275 shmem_unacct_blocks(info->flags, 1); 1276 shmem_free_blocks(inode, 1); 1277 filepage = NULL; 1278 if (error) 1279 goto failed; 1280 goto repeat; 1281 } 1282 info->flags |= SHMEM_PAGEIN; 1283 } 1284 1285 info->alloced++; 1286 spin_unlock(&info->lock); 1287 flush_dcache_page(filepage); 1288 SetPageUptodate(filepage); 1289 } 1290done: 1291 if (*pagep != filepage) { 1292 unlock_page(filepage); 1293 *pagep = filepage; 1294 } 1295 return 0; 1296 1297failed: 1298 if (*pagep != filepage) { 1299 unlock_page(filepage); 1300 page_cache_release(filepage); 1301 } 1302 return error; 1303} 1304 1305static struct page *shmem_nopage(struct vm_area_struct *vma, 1306 unsigned long address, int *type) 1307{ 1308 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 1309 struct page *page = NULL; 1310 unsigned long idx; 1311 int error; 1312 1313 idx = (address - vma->vm_start) >> PAGE_SHIFT; 1314 idx += vma->vm_pgoff; 1315 idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT; 1316 if (((loff_t) idx << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 1317 return NOPAGE_SIGBUS; 1318 1319 error = shmem_getpage(inode, idx, &page, SGP_CACHE, type); 1320 if (error) 1321 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS; 1322 1323 mark_page_accessed(page); 1324 return page; 1325} 1326 1327static int shmem_populate(struct vm_area_struct *vma, 1328 unsigned long addr, unsigned long len, 1329 pgprot_t prot, unsigned long pgoff, int nonblock) 1330{ 1331 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 1332 struct mm_struct *mm = vma->vm_mm; 1333 enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE; 1334 unsigned long size; 1335 1336 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT; 1337 if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size) 1338 return -EINVAL; 1339 1340 while ((long) len > 0) { 1341 struct page *page = NULL; 1342 int err; 1343 /* 1344 * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE 1345 */ 1346 err = shmem_getpage(inode, pgoff, &page, sgp, NULL); 1347 if (err) 1348 return err; 1349 /* Page may still be null, but only if nonblock was set. */ 1350 if (page) { 1351 mark_page_accessed(page); 1352 err = install_page(mm, vma, addr, page, prot); 1353 if (err) { 1354 page_cache_release(page); 1355 return err; 1356 } 1357 } else if (vma->vm_flags & VM_NONLINEAR) { 1358 /* No page was found just because we can't read it in 1359 * now (being here implies nonblock != 0), but the page 1360 * may exist, so set the PTE to fault it in later. */ 1361 err = install_file_pte(mm, vma, addr, pgoff, prot); 1362 if (err) 1363 return err; 1364 } 1365 1366 len -= PAGE_SIZE; 1367 addr += PAGE_SIZE; 1368 pgoff++; 1369 } 1370 return 0; 1371} 1372 1373#ifdef CONFIG_NUMA 1374int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new) 1375{ 1376 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1377 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new); 1378} 1379 1380struct mempolicy * 1381shmem_get_policy(struct vm_area_struct *vma, unsigned long addr) 1382{ 1383 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1384 unsigned long idx; 1385 1386 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; 1387 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx); 1388} 1389#endif 1390 1391int shmem_lock(struct file *file, int lock, struct user_struct *user) 1392{ 1393 struct inode *inode = file->f_path.dentry->d_inode; 1394 struct shmem_inode_info *info = SHMEM_I(inode); 1395 int retval = -ENOMEM; 1396 1397 spin_lock(&info->lock); 1398 if (lock && !(info->flags & VM_LOCKED)) { 1399 if (!user_shm_lock(inode->i_size, user)) 1400 goto out_nomem; 1401 info->flags |= VM_LOCKED; 1402 } 1403 if (!lock && (info->flags & VM_LOCKED) && user) { 1404 user_shm_unlock(inode->i_size, user); 1405 info->flags &= ~VM_LOCKED; 1406 } 1407 retval = 0; 1408out_nomem: 1409 spin_unlock(&info->lock); 1410 return retval; 1411} 1412 1413static int shmem_mmap(struct file *file, struct vm_area_struct *vma) 1414{ 1415 file_accessed(file); 1416 vma->vm_ops = &shmem_vm_ops; 1417 return 0; 1418} 1419 1420static struct inode * 1421shmem_get_inode(struct super_block *sb, int mode, dev_t dev) 1422{ 1423 struct inode *inode; 1424 struct shmem_inode_info *info; 1425 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 1426 1427 if (sbinfo->max_inodes) { 1428 spin_lock(&sbinfo->stat_lock); 1429 if (!sbinfo->free_inodes) { 1430 spin_unlock(&sbinfo->stat_lock); 1431 return NULL; 1432 } 1433 sbinfo->free_inodes--; 1434 spin_unlock(&sbinfo->stat_lock); 1435 } 1436 1437 inode = new_inode(sb); 1438 if (inode) { 1439 inode->i_mode = mode; 1440 inode->i_uid = current->fsuid; 1441 inode->i_gid = current->fsgid; 1442 inode->i_blocks = 0; 1443 inode->i_mapping->a_ops = &shmem_aops; 1444 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; 1445 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 1446 inode->i_generation = get_seconds(); 1447 info = SHMEM_I(inode); 1448 memset(info, 0, (char *)inode - (char *)info); 1449 spin_lock_init(&info->lock); 1450 INIT_LIST_HEAD(&info->swaplist); 1451 1452 switch (mode & S_IFMT) { 1453 default: 1454 inode->i_op = &shmem_special_inode_operations; 1455 init_special_inode(inode, mode, dev); 1456 break; 1457 case S_IFREG: 1458 inode->i_op = &shmem_inode_operations; 1459 inode->i_fop = &shmem_file_operations; 1460 mpol_shared_policy_init(&info->policy, sbinfo->policy, 1461 &sbinfo->policy_nodes); 1462 break; 1463 case S_IFDIR: 1464 inc_nlink(inode); 1465 /* Some things misbehave if size == 0 on a directory */ 1466 inode->i_size = 2 * BOGO_DIRENT_SIZE; 1467 inode->i_op = &shmem_dir_inode_operations; 1468 inode->i_fop = &simple_dir_operations; 1469 break; 1470 case S_IFLNK: 1471 /* 1472 * Must not load anything in the rbtree, 1473 * mpol_free_shared_policy will not be called. 1474 */ 1475 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, 1476 NULL); 1477 break; 1478 } 1479 } else if (sbinfo->max_inodes) { 1480 spin_lock(&sbinfo->stat_lock); 1481 sbinfo->free_inodes++; 1482 spin_unlock(&sbinfo->stat_lock); 1483 } 1484 return inode; 1485} 1486 1487#ifdef CONFIG_TMPFS 1488static const struct inode_operations shmem_symlink_inode_operations; 1489static const struct inode_operations shmem_symlink_inline_operations; 1490 1491/* 1492 * Normally tmpfs avoids the use of shmem_readpage and shmem_prepare_write; 1493 * but providing them allows a tmpfs file to be used for splice, sendfile, and 1494 * below the loop driver, in the generic fashion that many filesystems support. 1495 */ 1496static int shmem_readpage(struct file *file, struct page *page) 1497{ 1498 struct inode *inode = page->mapping->host; 1499 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL); 1500 unlock_page(page); 1501 return error; 1502} 1503 1504static int 1505shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to) 1506{ 1507 struct inode *inode = page->mapping->host; 1508 return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL); 1509} 1510 1511static ssize_t 1512shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos) 1513{ 1514 struct inode *inode = file->f_path.dentry->d_inode; 1515 loff_t pos; 1516 unsigned long written; 1517 ssize_t err; 1518 1519 if ((ssize_t) count < 0) 1520 return -EINVAL; 1521 1522 if (!access_ok(VERIFY_READ, buf, count)) 1523 return -EFAULT; 1524 1525 mutex_lock(&inode->i_mutex); 1526 1527 pos = *ppos; 1528 written = 0; 1529 1530 err = generic_write_checks(file, &pos, &count, 0); 1531 if (err || !count) 1532 goto out; 1533 1534 err = remove_suid(file->f_path.dentry); 1535 if (err) 1536 goto out; 1537 1538 inode->i_ctime = inode->i_mtime = CURRENT_TIME; 1539 1540 do { 1541 struct page *page = NULL; 1542 unsigned long bytes, index, offset; 1543 char *kaddr; 1544 int left; 1545 1546 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */ 1547 index = pos >> PAGE_CACHE_SHIFT; 1548 bytes = PAGE_CACHE_SIZE - offset; 1549 if (bytes > count) 1550 bytes = count; 1551 1552 /* 1553 * We don't hold page lock across copy from user - 1554 * what would it guard against? - so no deadlock here. 1555 * But it still may be a good idea to prefault below. 1556 */ 1557 1558 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL); 1559 if (err) 1560 break; 1561 1562 left = bytes; 1563 if (PageHighMem(page)) { 1564 volatile unsigned char dummy; 1565 __get_user(dummy, buf); 1566 __get_user(dummy, buf + bytes - 1); 1567 1568 kaddr = kmap_atomic(page, KM_USER0); 1569 left = __copy_from_user_inatomic(kaddr + offset, 1570 buf, bytes); 1571 kunmap_atomic(kaddr, KM_USER0); 1572 } 1573 if (left) { 1574 kaddr = kmap(page); 1575 left = __copy_from_user(kaddr + offset, buf, bytes); 1576 kunmap(page); 1577 } 1578 1579 written += bytes; 1580 count -= bytes; 1581 pos += bytes; 1582 buf += bytes; 1583 if (pos > inode->i_size) 1584 i_size_write(inode, pos); 1585 1586 flush_dcache_page(page); 1587 set_page_dirty(page); 1588 mark_page_accessed(page); 1589 page_cache_release(page); 1590 1591 if (left) { 1592 pos -= left; 1593 written -= left; 1594 err = -EFAULT; 1595 break; 1596 } 1597 1598 /* 1599 * Our dirty pages are not counted in nr_dirty, 1600 * and we do not attempt to balance dirty pages. 1601 */ 1602 1603 cond_resched(); 1604 } while (count); 1605 1606 *ppos = pos; 1607 if (written) 1608 err = written; 1609out: 1610 mutex_unlock(&inode->i_mutex); 1611 return err; 1612} 1613 1614static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor) 1615{ 1616 struct inode *inode = filp->f_path.dentry->d_inode; 1617 struct address_space *mapping = inode->i_mapping; 1618 unsigned long index, offset; 1619 1620 index = *ppos >> PAGE_CACHE_SHIFT; 1621 offset = *ppos & ~PAGE_CACHE_MASK; 1622 1623 for (;;) { 1624 struct page *page = NULL; 1625 unsigned long end_index, nr, ret; 1626 loff_t i_size = i_size_read(inode); 1627 1628 end_index = i_size >> PAGE_CACHE_SHIFT; 1629 if (index > end_index) 1630 break; 1631 if (index == end_index) { 1632 nr = i_size & ~PAGE_CACHE_MASK; 1633 if (nr <= offset) 1634 break; 1635 } 1636 1637 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL); 1638 if (desc->error) { 1639 if (desc->error == -EINVAL) 1640 desc->error = 0; 1641 break; 1642 } 1643 1644 /* 1645 * We must evaluate after, since reads (unlike writes) 1646 * are called without i_mutex protection against truncate 1647 */ 1648 nr = PAGE_CACHE_SIZE; 1649 i_size = i_size_read(inode); 1650 end_index = i_size >> PAGE_CACHE_SHIFT; 1651 if (index == end_index) { 1652 nr = i_size & ~PAGE_CACHE_MASK; 1653 if (nr <= offset) { 1654 if (page) 1655 page_cache_release(page); 1656 break; 1657 } 1658 } 1659 nr -= offset; 1660 1661 if (page) { 1662 /* 1663 * If users can be writing to this page using arbitrary 1664 * virtual addresses, take care about potential aliasing 1665 * before reading the page on the kernel side. 1666 */ 1667 if (mapping_writably_mapped(mapping)) 1668 flush_dcache_page(page); 1669 /* 1670 * Mark the page accessed if we read the beginning. 1671 */ 1672 if (!offset) 1673 mark_page_accessed(page); 1674 } else { 1675 page = ZERO_PAGE(0); 1676 page_cache_get(page); 1677 } 1678 1679 /* 1680 * Ok, we have the page, and it's up-to-date, so 1681 * now we can copy it to user space... 1682 * 1683 * The actor routine returns how many bytes were actually used.. 1684 * NOTE! This may not be the same as how much of a user buffer 1685 * we filled up (we may be padding etc), so we can only update 1686 * "pos" here (the actor routine has to update the user buffer 1687 * pointers and the remaining count). 1688 */ 1689 ret = actor(desc, page, offset, nr); 1690 offset += ret; 1691 index += offset >> PAGE_CACHE_SHIFT; 1692 offset &= ~PAGE_CACHE_MASK; 1693 1694 page_cache_release(page); 1695 if (ret != nr || !desc->count) 1696 break; 1697 1698 cond_resched(); 1699 } 1700 1701 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; 1702 file_accessed(filp); 1703} 1704 1705static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos) 1706{ 1707 read_descriptor_t desc; 1708 1709 if ((ssize_t) count < 0) 1710 return -EINVAL; 1711 if (!access_ok(VERIFY_WRITE, buf, count)) 1712 return -EFAULT; 1713 if (!count) 1714 return 0; 1715 1716 desc.written = 0; 1717 desc.count = count; 1718 desc.arg.buf = buf; 1719 desc.error = 0; 1720 1721 do_shmem_file_read(filp, ppos, &desc, file_read_actor); 1722 if (desc.written) 1723 return desc.written; 1724 return desc.error; 1725} 1726 1727static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) 1728{ 1729 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); 1730 1731 buf->f_type = TMPFS_MAGIC; 1732 buf->f_bsize = PAGE_CACHE_SIZE; 1733 buf->f_namelen = NAME_MAX; 1734 spin_lock(&sbinfo->stat_lock); 1735 if (sbinfo->max_blocks) { 1736 buf->f_blocks = sbinfo->max_blocks; 1737 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks; 1738 } 1739 if (sbinfo->max_inodes) { 1740 buf->f_files = sbinfo->max_inodes; 1741 buf->f_ffree = sbinfo->free_inodes; 1742 } 1743 /* else leave those fields 0 like simple_statfs */ 1744 spin_unlock(&sbinfo->stat_lock); 1745 return 0; 1746} 1747 1748/* 1749 * File creation. Allocate an inode, and we're done.. 1750 */ 1751static int 1752shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 1753{ 1754 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev); 1755 int error = -ENOSPC; 1756 1757 if (inode) { 1758 error = security_inode_init_security(inode, dir, NULL, NULL, 1759 NULL); 1760 if (error) { 1761 if (error != -EOPNOTSUPP) { 1762 iput(inode); 1763 return error; 1764 } 1765 } 1766 error = shmem_acl_init(inode, dir); 1767 if (error) { 1768 iput(inode); 1769 return error; 1770 } 1771 if (dir->i_mode & S_ISGID) { 1772 inode->i_gid = dir->i_gid; 1773 if (S_ISDIR(mode)) 1774 inode->i_mode |= S_ISGID; 1775 } 1776 dir->i_size += BOGO_DIRENT_SIZE; 1777 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1778 d_instantiate(dentry, inode); 1779 dget(dentry); /* Extra count - pin the dentry in core */ 1780 } 1781 return error; 1782} 1783 1784static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1785{ 1786 int error; 1787 1788 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) 1789 return error; 1790 inc_nlink(dir); 1791 return 0; 1792} 1793 1794static int shmem_create(struct inode *dir, struct dentry *dentry, int mode, 1795 struct nameidata *nd) 1796{ 1797 return shmem_mknod(dir, dentry, mode | S_IFREG, 0); 1798} 1799 1800/* 1801 * Link a file.. 1802 */ 1803static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1804{ 1805 struct inode *inode = old_dentry->d_inode; 1806 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 1807 1808 /* 1809 * No ordinary (disk based) filesystem counts links as inodes; 1810 * but each new link needs a new dentry, pinning lowmem, and 1811 * tmpfs dentries cannot be pruned until they are unlinked. 1812 */ 1813 if (sbinfo->max_inodes) { 1814 spin_lock(&sbinfo->stat_lock); 1815 if (!sbinfo->free_inodes) { 1816 spin_unlock(&sbinfo->stat_lock); 1817 return -ENOSPC; 1818 } 1819 sbinfo->free_inodes--; 1820 spin_unlock(&sbinfo->stat_lock); 1821 } 1822 1823 dir->i_size += BOGO_DIRENT_SIZE; 1824 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1825 inc_nlink(inode); 1826 atomic_inc(&inode->i_count); /* New dentry reference */ 1827 dget(dentry); /* Extra pinning count for the created dentry */ 1828 d_instantiate(dentry, inode); 1829 return 0; 1830} 1831 1832static int shmem_unlink(struct inode *dir, struct dentry *dentry) 1833{ 1834 struct inode *inode = dentry->d_inode; 1835 1836 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) { 1837 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 1838 if (sbinfo->max_inodes) { 1839 spin_lock(&sbinfo->stat_lock); 1840 sbinfo->free_inodes++; 1841 spin_unlock(&sbinfo->stat_lock); 1842 } 1843 } 1844 1845 dir->i_size -= BOGO_DIRENT_SIZE; 1846 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1847 drop_nlink(inode); 1848 dput(dentry); /* Undo the count from "create" - this does all the work */ 1849 return 0; 1850} 1851 1852static int shmem_rmdir(struct inode *dir, struct dentry *dentry) 1853{ 1854 if (!simple_empty(dentry)) 1855 return -ENOTEMPTY; 1856 1857 drop_nlink(dentry->d_inode); 1858 drop_nlink(dir); 1859 return shmem_unlink(dir, dentry); 1860} 1861 1862/* 1863 * The VFS layer already does all the dentry stuff for rename, 1864 * we just have to decrement the usage count for the target if 1865 * it exists so that the VFS layer correctly free's it when it 1866 * gets overwritten. 1867 */ 1868static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) 1869{ 1870 struct inode *inode = old_dentry->d_inode; 1871 int they_are_dirs = S_ISDIR(inode->i_mode); 1872 1873 if (!simple_empty(new_dentry)) 1874 return -ENOTEMPTY; 1875 1876 if (new_dentry->d_inode) { 1877 (void) shmem_unlink(new_dir, new_dentry); 1878 if (they_are_dirs) 1879 drop_nlink(old_dir); 1880 } else if (they_are_dirs) { 1881 drop_nlink(old_dir); 1882 inc_nlink(new_dir); 1883 } 1884 1885 old_dir->i_size -= BOGO_DIRENT_SIZE; 1886 new_dir->i_size += BOGO_DIRENT_SIZE; 1887 old_dir->i_ctime = old_dir->i_mtime = 1888 new_dir->i_ctime = new_dir->i_mtime = 1889 inode->i_ctime = CURRENT_TIME; 1890 return 0; 1891} 1892 1893static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1894{ 1895 int error; 1896 int len; 1897 struct inode *inode; 1898 struct page *page = NULL; 1899 char *kaddr; 1900 struct shmem_inode_info *info; 1901 1902 len = strlen(symname) + 1; 1903 if (len > PAGE_CACHE_SIZE) 1904 return -ENAMETOOLONG; 1905 1906 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0); 1907 if (!inode) 1908 return -ENOSPC; 1909 1910 error = security_inode_init_security(inode, dir, NULL, NULL, 1911 NULL); 1912 if (error) { 1913 if (error != -EOPNOTSUPP) { 1914 iput(inode); 1915 return error; 1916 } 1917 error = 0; 1918 } 1919 1920 info = SHMEM_I(inode); 1921 inode->i_size = len-1; 1922 if (len <= (char *)inode - (char *)info) { 1923 /* do it inline */ 1924 memcpy(info, symname, len); 1925 inode->i_op = &shmem_symlink_inline_operations; 1926 } else { 1927 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); 1928 if (error) { 1929 iput(inode); 1930 return error; 1931 } 1932 inode->i_op = &shmem_symlink_inode_operations; 1933 kaddr = kmap_atomic(page, KM_USER0); 1934 memcpy(kaddr, symname, len); 1935 kunmap_atomic(kaddr, KM_USER0); 1936 set_page_dirty(page); 1937 page_cache_release(page); 1938 } 1939 if (dir->i_mode & S_ISGID) 1940 inode->i_gid = dir->i_gid; 1941 dir->i_size += BOGO_DIRENT_SIZE; 1942 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1943 d_instantiate(dentry, inode); 1944 dget(dentry); 1945 return 0; 1946} 1947 1948static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd) 1949{ 1950 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode)); 1951 return NULL; 1952} 1953 1954static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) 1955{ 1956 struct page *page = NULL; 1957 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); 1958 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page)); 1959 return page; 1960} 1961 1962static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 1963{ 1964 if (!IS_ERR(nd_get_link(nd))) { 1965 struct page *page = cookie; 1966 kunmap(page); 1967 mark_page_accessed(page); 1968 page_cache_release(page); 1969 } 1970} 1971 1972static const struct inode_operations shmem_symlink_inline_operations = { 1973 .readlink = generic_readlink, 1974 .follow_link = shmem_follow_link_inline, 1975}; 1976 1977static const struct inode_operations shmem_symlink_inode_operations = { 1978 .truncate = shmem_truncate, 1979 .readlink = generic_readlink, 1980 .follow_link = shmem_follow_link, 1981 .put_link = shmem_put_link, 1982}; 1983 1984#ifdef CONFIG_TMPFS_POSIX_ACL 1985/** 1986 * Superblocks without xattr inode operations will get security.* xattr 1987 * support from the VFS "for free". As soon as we have any other xattrs 1988 * like ACLs, we also need to implement the security.* handlers at 1989 * filesystem level, though. 1990 */ 1991 1992static size_t shmem_xattr_security_list(struct inode *inode, char *list, 1993 size_t list_len, const char *name, 1994 size_t name_len) 1995{ 1996 return security_inode_listsecurity(inode, list, list_len); 1997} 1998 1999static int shmem_xattr_security_get(struct inode *inode, const char *name, 2000 void *buffer, size_t size) 2001{ 2002 if (strcmp(name, "") == 0) 2003 return -EINVAL; 2004 return security_inode_getsecurity(inode, name, buffer, size, 2005 -EOPNOTSUPP); 2006} 2007 2008static int shmem_xattr_security_set(struct inode *inode, const char *name, 2009 const void *value, size_t size, int flags) 2010{ 2011 if (strcmp(name, "") == 0) 2012 return -EINVAL; 2013 return security_inode_setsecurity(inode, name, value, size, flags); 2014} 2015 2016static struct xattr_handler shmem_xattr_security_handler = { 2017 .prefix = XATTR_SECURITY_PREFIX, 2018 .list = shmem_xattr_security_list, 2019 .get = shmem_xattr_security_get, 2020 .set = shmem_xattr_security_set, 2021}; 2022 2023static struct xattr_handler *shmem_xattr_handlers[] = { 2024 &shmem_xattr_acl_access_handler, 2025 &shmem_xattr_acl_default_handler, 2026 &shmem_xattr_security_handler, 2027 NULL 2028}; 2029#endif 2030 2031static struct dentry *shmem_get_parent(struct dentry *child) 2032{ 2033 return ERR_PTR(-ESTALE); 2034} 2035 2036static int shmem_match(struct inode *ino, void *vfh) 2037{ 2038 __u32 *fh = vfh; 2039 __u64 inum = fh[2]; 2040 inum = (inum << 32) | fh[1]; 2041 return ino->i_ino == inum && fh[0] == ino->i_generation; 2042} 2043 2044static struct dentry *shmem_get_dentry(struct super_block *sb, void *vfh) 2045{ 2046 struct dentry *de = NULL; 2047 struct inode *inode; 2048 __u32 *fh = vfh; 2049 __u64 inum = fh[2]; 2050 inum = (inum << 32) | fh[1]; 2051 2052 inode = ilookup5(sb, (unsigned long)(inum+fh[0]), shmem_match, vfh); 2053 if (inode) { 2054 de = d_find_alias(inode); 2055 iput(inode); 2056 } 2057 2058 return de? de: ERR_PTR(-ESTALE); 2059} 2060 2061static struct dentry *shmem_decode_fh(struct super_block *sb, __u32 *fh, 2062 int len, int type, 2063 int (*acceptable)(void *context, struct dentry *de), 2064 void *context) 2065{ 2066 if (len < 3) 2067 return ERR_PTR(-ESTALE); 2068 2069 return sb->s_export_op->find_exported_dentry(sb, fh, NULL, acceptable, 2070 context); 2071} 2072 2073static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len, 2074 int connectable) 2075{ 2076 struct inode *inode = dentry->d_inode; 2077 2078 if (*len < 3) 2079 return 255; 2080 2081 if (hlist_unhashed(&inode->i_hash)) { 2082 /* Unfortunately insert_inode_hash is not idempotent, 2083 * so as we hash inodes here rather than at creation 2084 * time, we need a lock to ensure we only try 2085 * to do it once 2086 */ 2087 static DEFINE_SPINLOCK(lock); 2088 spin_lock(&lock); 2089 if (hlist_unhashed(&inode->i_hash)) 2090 __insert_inode_hash(inode, 2091 inode->i_ino + inode->i_generation); 2092 spin_unlock(&lock); 2093 } 2094 2095 fh[0] = inode->i_generation; 2096 fh[1] = inode->i_ino; 2097 fh[2] = ((__u64)inode->i_ino) >> 32; 2098 2099 *len = 3; 2100 return 1; 2101} 2102 2103static struct export_operations shmem_export_ops = { 2104 .get_parent = shmem_get_parent, 2105 .get_dentry = shmem_get_dentry, 2106 .encode_fh = shmem_encode_fh, 2107 .decode_fh = shmem_decode_fh, 2108}; 2109 2110static int shmem_parse_options(char *options, int *mode, uid_t *uid, 2111 gid_t *gid, unsigned long *blocks, unsigned long *inodes, 2112 int *policy, nodemask_t *policy_nodes) 2113{ 2114 char *this_char, *value, *rest; 2115 2116 while (options != NULL) { 2117 this_char = options; 2118 for (;;) { 2119 /* 2120 * NUL-terminate this option: unfortunately, 2121 * mount options form a comma-separated list, 2122 * but mpol's nodelist may also contain commas. 2123 */ 2124 options = strchr(options, ','); 2125 if (options == NULL) 2126 break; 2127 options++; 2128 if (!isdigit(*options)) { 2129 options[-1] = '\0'; 2130 break; 2131 } 2132 } 2133 if (!*this_char) 2134 continue; 2135 if ((value = strchr(this_char,'=')) != NULL) { 2136 *value++ = 0; 2137 } else { 2138 printk(KERN_ERR 2139 "tmpfs: No value for mount option '%s'\n", 2140 this_char); 2141 return 1; 2142 } 2143 2144 if (!strcmp(this_char,"size")) { 2145 unsigned long long size; 2146 size = memparse(value,&rest); 2147 if (*rest == '%') { 2148 size <<= PAGE_SHIFT; 2149 size *= totalram_pages; 2150 do_div(size, 100); 2151 rest++; 2152 } 2153 if (*rest) 2154 goto bad_val; 2155 *blocks = size >> PAGE_CACHE_SHIFT; 2156 } else if (!strcmp(this_char,"nr_blocks")) { 2157 *blocks = memparse(value,&rest); 2158 if (*rest) 2159 goto bad_val; 2160 } else if (!strcmp(this_char,"nr_inodes")) { 2161 *inodes = memparse(value,&rest); 2162 if (*rest) 2163 goto bad_val; 2164 } else if (!strcmp(this_char,"mode")) { 2165 if (!mode) 2166 continue; 2167 *mode = simple_strtoul(value,&rest,8); 2168 if (*rest) 2169 goto bad_val; 2170 } else if (!strcmp(this_char,"uid")) { 2171 if (!uid) 2172 continue; 2173 *uid = simple_strtoul(value,&rest,0); 2174 if (*rest) 2175 goto bad_val; 2176 } else if (!strcmp(this_char,"gid")) { 2177 if (!gid) 2178 continue; 2179 *gid = simple_strtoul(value,&rest,0); 2180 if (*rest) 2181 goto bad_val; 2182 } else if (!strcmp(this_char,"mpol")) { 2183 if (shmem_parse_mpol(value,policy,policy_nodes)) 2184 goto bad_val; 2185 } else { 2186 printk(KERN_ERR "tmpfs: Bad mount option %s\n", 2187 this_char); 2188 return 1; 2189 } 2190 } 2191 return 0; 2192 2193bad_val: 2194 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", 2195 value, this_char); 2196 return 1; 2197 2198} 2199 2200static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) 2201{ 2202 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 2203 unsigned long max_blocks = sbinfo->max_blocks; 2204 unsigned long max_inodes = sbinfo->max_inodes; 2205 int policy = sbinfo->policy; 2206 nodemask_t policy_nodes = sbinfo->policy_nodes; 2207 unsigned long blocks; 2208 unsigned long inodes; 2209 int error = -EINVAL; 2210 2211 if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks, 2212 &max_inodes, &policy, &policy_nodes)) 2213 return error; 2214 2215 spin_lock(&sbinfo->stat_lock); 2216 blocks = sbinfo->max_blocks - sbinfo->free_blocks; 2217 inodes = sbinfo->max_inodes - sbinfo->free_inodes; 2218 if (max_blocks < blocks) 2219 goto out; 2220 if (max_inodes < inodes) 2221 goto out; 2222 /* 2223 * Those tests also disallow limited->unlimited while any are in 2224 * use, so i_blocks will always be zero when max_blocks is zero; 2225 * but we must separately disallow unlimited->limited, because 2226 * in that case we have no record of how much is already in use. 2227 */ 2228 if (max_blocks && !sbinfo->max_blocks) 2229 goto out; 2230 if (max_inodes && !sbinfo->max_inodes) 2231 goto out; 2232 2233 error = 0; 2234 sbinfo->max_blocks = max_blocks; 2235 sbinfo->free_blocks = max_blocks - blocks; 2236 sbinfo->max_inodes = max_inodes; 2237 sbinfo->free_inodes = max_inodes - inodes; 2238 sbinfo->policy = policy; 2239 sbinfo->policy_nodes = policy_nodes; 2240out: 2241 spin_unlock(&sbinfo->stat_lock); 2242 return error; 2243} 2244#endif 2245 2246static void shmem_put_super(struct super_block *sb) 2247{ 2248 kfree(sb->s_fs_info); 2249 sb->s_fs_info = NULL; 2250} 2251 2252static int shmem_fill_super(struct super_block *sb, 2253 void *data, int silent) 2254{ 2255 struct inode *inode; 2256 struct dentry *root; 2257 int mode = S_IRWXUGO | S_ISVTX; 2258 uid_t uid = current->fsuid; 2259 gid_t gid = current->fsgid; 2260 int err = -ENOMEM; 2261 struct shmem_sb_info *sbinfo; 2262 unsigned long blocks = 0; 2263 unsigned long inodes = 0; 2264 int policy = MPOL_DEFAULT; 2265 nodemask_t policy_nodes = node_online_map; 2266 2267#ifdef CONFIG_TMPFS 2268 /* 2269 * Per default we only allow half of the physical ram per 2270 * tmpfs instance, limiting inodes to one per page of lowmem; 2271 * but the internal instance is left unlimited. 2272 */ 2273 if (!(sb->s_flags & MS_NOUSER)) { 2274 blocks = totalram_pages / 2; 2275 inodes = totalram_pages - totalhigh_pages; 2276 if (inodes > blocks) 2277 inodes = blocks; 2278 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks, 2279 &inodes, &policy, &policy_nodes)) 2280 return -EINVAL; 2281 } 2282 sb->s_export_op = &shmem_export_ops; 2283#else 2284 sb->s_flags |= MS_NOUSER; 2285#endif 2286 2287 /* Round up to L1_CACHE_BYTES to resist false sharing */ 2288 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info), 2289 L1_CACHE_BYTES), GFP_KERNEL); 2290 if (!sbinfo) 2291 return -ENOMEM; 2292 2293 spin_lock_init(&sbinfo->stat_lock); 2294 sbinfo->max_blocks = blocks; 2295 sbinfo->free_blocks = blocks; 2296 sbinfo->max_inodes = inodes; 2297 sbinfo->free_inodes = inodes; 2298 sbinfo->policy = policy; 2299 sbinfo->policy_nodes = policy_nodes; 2300 2301 sb->s_fs_info = sbinfo; 2302 sb->s_maxbytes = SHMEM_MAX_BYTES; 2303 sb->s_blocksize = PAGE_CACHE_SIZE; 2304 sb->s_blocksize_bits = PAGE_CACHE_SHIFT; 2305 sb->s_magic = TMPFS_MAGIC; 2306 sb->s_op = &shmem_ops; 2307 sb->s_time_gran = 1; 2308#ifdef CONFIG_TMPFS_POSIX_ACL 2309 sb->s_xattr = shmem_xattr_handlers; 2310 sb->s_flags |= MS_POSIXACL; 2311#endif 2312 2313 inode = shmem_get_inode(sb, S_IFDIR | mode, 0); 2314 if (!inode) 2315 goto failed; 2316 inode->i_uid = uid; 2317 inode->i_gid = gid; 2318 root = d_alloc_root(inode); 2319 if (!root) 2320 goto failed_iput; 2321 sb->s_root = root; 2322 return 0; 2323 2324failed_iput: 2325 iput(inode); 2326failed: 2327 shmem_put_super(sb); 2328 return err; 2329} 2330 2331static struct kmem_cache *shmem_inode_cachep; 2332 2333static struct inode *shmem_alloc_inode(struct super_block *sb) 2334{ 2335 struct shmem_inode_info *p; 2336 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); 2337 if (!p) 2338 return NULL; 2339 return &p->vfs_inode; 2340} 2341 2342static void shmem_destroy_inode(struct inode *inode) 2343{ 2344 if ((inode->i_mode & S_IFMT) == S_IFREG) { 2345 /* only struct inode is valid if it's an inline symlink */ 2346 mpol_free_shared_policy(&SHMEM_I(inode)->policy); 2347 } 2348 shmem_acl_destroy_inode(inode); 2349 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); 2350} 2351 2352static void init_once(void *foo, struct kmem_cache *cachep, 2353 unsigned long flags) 2354{ 2355 struct shmem_inode_info *p = (struct shmem_inode_info *) foo; 2356 2357 inode_init_once(&p->vfs_inode); 2358#ifdef CONFIG_TMPFS_POSIX_ACL 2359 p->i_acl = NULL; 2360 p->i_default_acl = NULL; 2361#endif 2362} 2363 2364static int init_inodecache(void) 2365{ 2366 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", 2367 sizeof(struct shmem_inode_info), 2368 0, 0, init_once, NULL); 2369 if (shmem_inode_cachep == NULL) 2370 return -ENOMEM; 2371 return 0; 2372} 2373 2374static void destroy_inodecache(void) 2375{ 2376 kmem_cache_destroy(shmem_inode_cachep); 2377} 2378 2379static const struct address_space_operations shmem_aops = { 2380 .writepage = shmem_writepage, 2381 .set_page_dirty = __set_page_dirty_no_writeback, 2382#ifdef CONFIG_TMPFS 2383 .readpage = shmem_readpage, 2384 .prepare_write = shmem_prepare_write, 2385 .commit_write = simple_commit_write, 2386#endif 2387 .migratepage = migrate_page, 2388}; 2389 2390static const struct file_operations shmem_file_operations = { 2391 .mmap = shmem_mmap, 2392#ifdef CONFIG_TMPFS 2393 .llseek = generic_file_llseek, 2394 .read = shmem_file_read, 2395 .write = shmem_file_write, 2396 .fsync = simple_sync_file, 2397 .splice_read = generic_file_splice_read, 2398 .splice_write = generic_file_splice_write, 2399#endif 2400}; 2401 2402static const struct inode_operations shmem_inode_operations = { 2403 .truncate = shmem_truncate, 2404 .setattr = shmem_notify_change, 2405 .truncate_range = shmem_truncate_range, 2406#ifdef CONFIG_TMPFS_POSIX_ACL 2407 .setxattr = generic_setxattr, 2408 .getxattr = generic_getxattr, 2409 .listxattr = generic_listxattr, 2410 .removexattr = generic_removexattr, 2411 .permission = shmem_permission, 2412#endif 2413 2414}; 2415 2416static const struct inode_operations shmem_dir_inode_operations = { 2417#ifdef CONFIG_TMPFS 2418 .create = shmem_create, 2419 .lookup = simple_lookup, 2420 .link = shmem_link, 2421 .unlink = shmem_unlink, 2422 .symlink = shmem_symlink, 2423 .mkdir = shmem_mkdir, 2424 .rmdir = shmem_rmdir, 2425 .mknod = shmem_mknod, 2426 .rename = shmem_rename, 2427#endif 2428#ifdef CONFIG_TMPFS_POSIX_ACL 2429 .setattr = shmem_notify_change, 2430 .setxattr = generic_setxattr, 2431 .getxattr = generic_getxattr, 2432 .listxattr = generic_listxattr, 2433 .removexattr = generic_removexattr, 2434 .permission = shmem_permission, 2435#endif 2436}; 2437 2438static const struct inode_operations shmem_special_inode_operations = { 2439#ifdef CONFIG_TMPFS_POSIX_ACL 2440 .setattr = shmem_notify_change, 2441 .setxattr = generic_setxattr, 2442 .getxattr = generic_getxattr, 2443 .listxattr = generic_listxattr, 2444 .removexattr = generic_removexattr, 2445 .permission = shmem_permission, 2446#endif 2447}; 2448 2449static const struct super_operations shmem_ops = { 2450 .alloc_inode = shmem_alloc_inode, 2451 .destroy_inode = shmem_destroy_inode, 2452#ifdef CONFIG_TMPFS 2453 .statfs = shmem_statfs, 2454 .remount_fs = shmem_remount_fs, 2455#endif 2456 .delete_inode = shmem_delete_inode, 2457 .drop_inode = generic_delete_inode, 2458 .put_super = shmem_put_super, 2459}; 2460 2461static struct vm_operations_struct shmem_vm_ops = { 2462 .nopage = shmem_nopage, 2463 .populate = shmem_populate, 2464#ifdef CONFIG_NUMA 2465 .set_policy = shmem_set_policy, 2466 .get_policy = shmem_get_policy, 2467#endif 2468}; 2469 2470 2471static int shmem_get_sb(struct file_system_type *fs_type, 2472 int flags, const char *dev_name, void *data, struct vfsmount *mnt) 2473{ 2474 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt); 2475} 2476 2477static struct file_system_type tmpfs_fs_type = { 2478 .owner = THIS_MODULE, 2479 .name = "tmpfs", 2480 .get_sb = shmem_get_sb, 2481 .kill_sb = kill_litter_super, 2482}; 2483static struct vfsmount *shm_mnt; 2484 2485static int __init init_tmpfs(void) 2486{ 2487 int error; 2488 2489 error = init_inodecache(); 2490 if (error) 2491 goto out3; 2492 2493 error = register_filesystem(&tmpfs_fs_type); 2494 if (error) { 2495 printk(KERN_ERR "Could not register tmpfs\n"); 2496 goto out2; 2497 } 2498 2499 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER, 2500 tmpfs_fs_type.name, NULL); 2501 if (IS_ERR(shm_mnt)) { 2502 error = PTR_ERR(shm_mnt); 2503 printk(KERN_ERR "Could not kern_mount tmpfs\n"); 2504 goto out1; 2505 } 2506 return 0; 2507 2508out1: 2509 unregister_filesystem(&tmpfs_fs_type); 2510out2: 2511 destroy_inodecache(); 2512out3: 2513 shm_mnt = ERR_PTR(error); 2514 return error; 2515} 2516module_init(init_tmpfs) 2517 2518/* 2519 * shmem_file_setup - get an unlinked file living in tmpfs 2520 * 2521 * @name: name for dentry (to be seen in /proc/<pid>/maps 2522 * @size: size to be set for the file 2523 * 2524 */ 2525struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags) 2526{ 2527 int error; 2528 struct file *file; 2529 struct inode *inode; 2530 struct dentry *dentry, *root; 2531 struct qstr this; 2532 2533 if (IS_ERR(shm_mnt)) 2534 return (void *)shm_mnt; 2535 2536 if (size < 0 || size > SHMEM_MAX_BYTES) 2537 return ERR_PTR(-EINVAL); 2538 2539 if (shmem_acct_size(flags, size)) 2540 return ERR_PTR(-ENOMEM); 2541 2542 error = -ENOMEM; 2543 this.name = name; 2544 this.len = strlen(name); 2545 this.hash = 0; /* will go */ 2546 root = shm_mnt->mnt_root; 2547 dentry = d_alloc(root, &this); 2548 if (!dentry) 2549 goto put_memory; 2550 2551 error = -ENFILE; 2552 file = get_empty_filp(); 2553 if (!file) 2554 goto put_dentry; 2555 2556 error = -ENOSPC; 2557 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0); 2558 if (!inode) 2559 goto close_file; 2560 2561 SHMEM_I(inode)->flags = flags & VM_ACCOUNT; 2562 d_instantiate(dentry, inode); 2563 inode->i_size = size; 2564 inode->i_nlink = 0; /* It is unlinked */ 2565 file->f_path.mnt = mntget(shm_mnt); 2566 file->f_path.dentry = dentry; 2567 file->f_mapping = inode->i_mapping; 2568 file->f_op = &shmem_file_operations; 2569 file->f_mode = FMODE_WRITE | FMODE_READ; 2570 return file; 2571 2572close_file: 2573 put_filp(file); 2574put_dentry: 2575 dput(dentry); 2576put_memory: 2577 shmem_unacct_size(flags, size); 2578 return ERR_PTR(error); 2579} 2580 2581/* 2582 * shmem_zero_setup - setup a shared anonymous mapping 2583 * 2584 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff 2585 */ 2586int shmem_zero_setup(struct vm_area_struct *vma) 2587{ 2588 struct file *file; 2589 loff_t size = vma->vm_end - vma->vm_start; 2590 2591 file = shmem_file_setup("dev/zero", size, vma->vm_flags); 2592 if (IS_ERR(file)) 2593 return PTR_ERR(file); 2594 2595 if (vma->vm_file) 2596 fput(vma->vm_file); 2597 vma->vm_file = file; 2598 vma->vm_ops = &shmem_vm_ops; 2599 return 0; 2600} 2601