shmem.c revision ae976416464b741913a13eea62eb6953ee065733
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/generic_acl.h> 31#include <linux/mm.h> 32#include <linux/mman.h> 33#include <linux/file.h> 34#include <linux/swap.h> 35#include <linux/pagemap.h> 36#include <linux/string.h> 37#include <linux/slab.h> 38#include <linux/backing-dev.h> 39#include <linux/shmem_fs.h> 40#include <linux/mount.h> 41#include <linux/writeback.h> 42#include <linux/vfs.h> 43#include <linux/blkdev.h> 44#include <linux/security.h> 45#include <linux/swapops.h> 46#include <linux/mempolicy.h> 47#include <linux/namei.h> 48#include <linux/ctype.h> 49#include <linux/migrate.h> 50#include <linux/highmem.h> 51#include <linux/backing-dev.h> 52 53#include <asm/uaccess.h> 54#include <asm/div64.h> 55#include <asm/pgtable.h> 56 57/* This magic number is used in glibc for posix shared memory */ 58#define TMPFS_MAGIC 0x01021994 59 60#define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long)) 61#define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE) 62#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512) 63 64#define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1)) 65#define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT) 66 67#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT) 68 69/* info->flags needs VM_flags to handle pagein/truncate races efficiently */ 70#define SHMEM_PAGEIN VM_READ 71#define SHMEM_TRUNCATE VM_WRITE 72 73/* Definition to limit shmem_truncate's steps between cond_rescheds */ 74#define LATENCY_LIMIT 64 75 76/* Pretend that each entry is of this size in directory's i_size */ 77#define BOGO_DIRENT_SIZE 20 78 79/* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */ 80enum sgp_type { 81 SGP_QUICK, /* don't try more than file page cache lookup */ 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_WRITE, /* may exceed i_size, may allocate page */ 85}; 86 87static int shmem_getpage(struct inode *inode, unsigned long idx, 88 struct page **pagep, enum sgp_type sgp, int *type); 89 90static inline struct page *shmem_dir_alloc(gfp_t gfp_mask) 91{ 92 /* 93 * The above definition of ENTRIES_PER_PAGE, and the use of 94 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE: 95 * might be reconsidered if it ever diverges from PAGE_SIZE. 96 */ 97 return alloc_pages(gfp_mask, PAGE_CACHE_SHIFT-PAGE_SHIFT); 98} 99 100static inline void shmem_dir_free(struct page *page) 101{ 102 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT); 103} 104 105static struct page **shmem_dir_map(struct page *page) 106{ 107 return (struct page **)kmap_atomic(page, KM_USER0); 108} 109 110static inline void shmem_dir_unmap(struct page **dir) 111{ 112 kunmap_atomic(dir, KM_USER0); 113} 114 115static swp_entry_t *shmem_swp_map(struct page *page) 116{ 117 return (swp_entry_t *)kmap_atomic(page, KM_USER1); 118} 119 120static inline void shmem_swp_balance_unmap(void) 121{ 122 /* 123 * When passing a pointer to an i_direct entry, to code which 124 * also handles indirect entries and so will shmem_swp_unmap, 125 * we must arrange for the preempt count to remain in balance. 126 * What kmap_atomic of a lowmem page does depends on config 127 * and architecture, so pretend to kmap_atomic some lowmem page. 128 */ 129 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1); 130} 131 132static inline void shmem_swp_unmap(swp_entry_t *entry) 133{ 134 kunmap_atomic(entry, KM_USER1); 135} 136 137static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) 138{ 139 return sb->s_fs_info; 140} 141 142/* 143 * shmem_file_setup pre-accounts the whole fixed size of a VM object, 144 * for shared memory and for shared anonymous (/dev/zero) mappings 145 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), 146 * consistent with the pre-accounting of private mappings ... 147 */ 148static inline int shmem_acct_size(unsigned long flags, loff_t size) 149{ 150 return (flags & VM_ACCOUNT)? 151 security_vm_enough_memory(VM_ACCT(size)): 0; 152} 153 154static inline void shmem_unacct_size(unsigned long flags, loff_t size) 155{ 156 if (flags & VM_ACCOUNT) 157 vm_unacct_memory(VM_ACCT(size)); 158} 159 160/* 161 * ... whereas tmpfs objects are accounted incrementally as 162 * pages are allocated, in order to allow huge sparse files. 163 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, 164 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. 165 */ 166static inline int shmem_acct_block(unsigned long flags) 167{ 168 return (flags & VM_ACCOUNT)? 169 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE)); 170} 171 172static inline void shmem_unacct_blocks(unsigned long flags, long pages) 173{ 174 if (!(flags & VM_ACCOUNT)) 175 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE)); 176} 177 178static const struct super_operations shmem_ops; 179static const struct address_space_operations shmem_aops; 180static const struct file_operations shmem_file_operations; 181static const struct inode_operations shmem_inode_operations; 182static const struct inode_operations shmem_dir_inode_operations; 183static const struct inode_operations shmem_special_inode_operations; 184static struct vm_operations_struct shmem_vm_ops; 185 186static struct backing_dev_info shmem_backing_dev_info __read_mostly = { 187 .ra_pages = 0, /* No readahead */ 188 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK, 189 .unplug_io_fn = default_unplug_io_fn, 190}; 191 192static LIST_HEAD(shmem_swaplist); 193static DEFINE_SPINLOCK(shmem_swaplist_lock); 194 195static void shmem_free_blocks(struct inode *inode, long pages) 196{ 197 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 198 if (sbinfo->max_blocks) { 199 spin_lock(&sbinfo->stat_lock); 200 sbinfo->free_blocks += pages; 201 inode->i_blocks -= pages*BLOCKS_PER_PAGE; 202 spin_unlock(&sbinfo->stat_lock); 203 } 204} 205 206/* 207 * shmem_recalc_inode - recalculate the size of an inode 208 * 209 * @inode: inode to recalc 210 * 211 * We have to calculate the free blocks since the mm can drop 212 * undirtied hole pages behind our back. 213 * 214 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped 215 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) 216 * 217 * It has to be called with the spinlock held. 218 */ 219static void shmem_recalc_inode(struct inode *inode) 220{ 221 struct shmem_inode_info *info = SHMEM_I(inode); 222 long freed; 223 224 freed = info->alloced - info->swapped - inode->i_mapping->nrpages; 225 if (freed > 0) { 226 info->alloced -= freed; 227 shmem_unacct_blocks(info->flags, freed); 228 shmem_free_blocks(inode, freed); 229 } 230} 231 232/* 233 * shmem_swp_entry - find the swap vector position in the info structure 234 * 235 * @info: info structure for the inode 236 * @index: index of the page to find 237 * @page: optional page to add to the structure. Has to be preset to 238 * all zeros 239 * 240 * If there is no space allocated yet it will return NULL when 241 * page is NULL, else it will use the page for the needed block, 242 * setting it to NULL on return to indicate that it has been used. 243 * 244 * The swap vector is organized the following way: 245 * 246 * There are SHMEM_NR_DIRECT entries directly stored in the 247 * shmem_inode_info structure. So small files do not need an addional 248 * allocation. 249 * 250 * For pages with index > SHMEM_NR_DIRECT there is the pointer 251 * i_indirect which points to a page which holds in the first half 252 * doubly indirect blocks, in the second half triple indirect blocks: 253 * 254 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the 255 * following layout (for SHMEM_NR_DIRECT == 16): 256 * 257 * i_indirect -> dir --> 16-19 258 * | +-> 20-23 259 * | 260 * +-->dir2 --> 24-27 261 * | +-> 28-31 262 * | +-> 32-35 263 * | +-> 36-39 264 * | 265 * +-->dir3 --> 40-43 266 * +-> 44-47 267 * +-> 48-51 268 * +-> 52-55 269 */ 270static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page) 271{ 272 unsigned long offset; 273 struct page **dir; 274 struct page *subdir; 275 276 if (index < SHMEM_NR_DIRECT) { 277 shmem_swp_balance_unmap(); 278 return info->i_direct+index; 279 } 280 if (!info->i_indirect) { 281 if (page) { 282 info->i_indirect = *page; 283 *page = NULL; 284 } 285 return NULL; /* need another page */ 286 } 287 288 index -= SHMEM_NR_DIRECT; 289 offset = index % ENTRIES_PER_PAGE; 290 index /= ENTRIES_PER_PAGE; 291 dir = shmem_dir_map(info->i_indirect); 292 293 if (index >= ENTRIES_PER_PAGE/2) { 294 index -= ENTRIES_PER_PAGE/2; 295 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE; 296 index %= ENTRIES_PER_PAGE; 297 subdir = *dir; 298 if (!subdir) { 299 if (page) { 300 *dir = *page; 301 *page = NULL; 302 } 303 shmem_dir_unmap(dir); 304 return NULL; /* need another page */ 305 } 306 shmem_dir_unmap(dir); 307 dir = shmem_dir_map(subdir); 308 } 309 310 dir += index; 311 subdir = *dir; 312 if (!subdir) { 313 if (!page || !(subdir = *page)) { 314 shmem_dir_unmap(dir); 315 return NULL; /* need a page */ 316 } 317 *dir = subdir; 318 *page = NULL; 319 } 320 shmem_dir_unmap(dir); 321 return shmem_swp_map(subdir) + offset; 322} 323 324static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value) 325{ 326 long incdec = value? 1: -1; 327 328 entry->val = value; 329 info->swapped += incdec; 330 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) { 331 struct page *page = kmap_atomic_to_page(entry); 332 set_page_private(page, page_private(page) + incdec); 333 } 334} 335 336/* 337 * shmem_swp_alloc - get the position of the swap entry for the page. 338 * If it does not exist allocate the entry. 339 * 340 * @info: info structure for the inode 341 * @index: index of the page to find 342 * @sgp: check and recheck i_size? skip allocation? 343 */ 344static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp) 345{ 346 struct inode *inode = &info->vfs_inode; 347 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 348 struct page *page = NULL; 349 swp_entry_t *entry; 350 351 if (sgp != SGP_WRITE && 352 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 353 return ERR_PTR(-EINVAL); 354 355 while (!(entry = shmem_swp_entry(info, index, &page))) { 356 if (sgp == SGP_READ) 357 return shmem_swp_map(ZERO_PAGE(0)); 358 /* 359 * Test free_blocks against 1 not 0, since we have 1 data 360 * page (and perhaps indirect index pages) yet to allocate: 361 * a waste to allocate index if we cannot allocate data. 362 */ 363 if (sbinfo->max_blocks) { 364 spin_lock(&sbinfo->stat_lock); 365 if (sbinfo->free_blocks <= 1) { 366 spin_unlock(&sbinfo->stat_lock); 367 return ERR_PTR(-ENOSPC); 368 } 369 sbinfo->free_blocks--; 370 inode->i_blocks += BLOCKS_PER_PAGE; 371 spin_unlock(&sbinfo->stat_lock); 372 } 373 374 spin_unlock(&info->lock); 375 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping) | __GFP_ZERO); 376 if (page) 377 set_page_private(page, 0); 378 spin_lock(&info->lock); 379 380 if (!page) { 381 shmem_free_blocks(inode, 1); 382 return ERR_PTR(-ENOMEM); 383 } 384 if (sgp != SGP_WRITE && 385 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { 386 entry = ERR_PTR(-EINVAL); 387 break; 388 } 389 if (info->next_index <= index) 390 info->next_index = index + 1; 391 } 392 if (page) { 393 /* another task gave its page, or truncated the file */ 394 shmem_free_blocks(inode, 1); 395 shmem_dir_free(page); 396 } 397 if (info->next_index <= index && !IS_ERR(entry)) 398 info->next_index = index + 1; 399 return entry; 400} 401 402/* 403 * shmem_free_swp - free some swap entries in a directory 404 * 405 * @dir: pointer to the directory 406 * @edir: pointer after last entry of the directory 407 * @punch_lock: pointer to spinlock when needed for the holepunch case 408 */ 409static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir, 410 spinlock_t *punch_lock) 411{ 412 spinlock_t *punch_unlock = NULL; 413 swp_entry_t *ptr; 414 int freed = 0; 415 416 for (ptr = dir; ptr < edir; ptr++) { 417 if (ptr->val) { 418 if (unlikely(punch_lock)) { 419 punch_unlock = punch_lock; 420 punch_lock = NULL; 421 spin_lock(punch_unlock); 422 if (!ptr->val) 423 continue; 424 } 425 free_swap_and_cache(*ptr); 426 *ptr = (swp_entry_t){0}; 427 freed++; 428 } 429 } 430 if (punch_unlock) 431 spin_unlock(punch_unlock); 432 return freed; 433} 434 435static int shmem_map_and_free_swp(struct page *subdir, int offset, 436 int limit, struct page ***dir, spinlock_t *punch_lock) 437{ 438 swp_entry_t *ptr; 439 int freed = 0; 440 441 ptr = shmem_swp_map(subdir); 442 for (; offset < limit; offset += LATENCY_LIMIT) { 443 int size = limit - offset; 444 if (size > LATENCY_LIMIT) 445 size = LATENCY_LIMIT; 446 freed += shmem_free_swp(ptr+offset, ptr+offset+size, 447 punch_lock); 448 if (need_resched()) { 449 shmem_swp_unmap(ptr); 450 if (*dir) { 451 shmem_dir_unmap(*dir); 452 *dir = NULL; 453 } 454 cond_resched(); 455 ptr = shmem_swp_map(subdir); 456 } 457 } 458 shmem_swp_unmap(ptr); 459 return freed; 460} 461 462static void shmem_free_pages(struct list_head *next) 463{ 464 struct page *page; 465 int freed = 0; 466 467 do { 468 page = container_of(next, struct page, lru); 469 next = next->next; 470 shmem_dir_free(page); 471 freed++; 472 if (freed >= LATENCY_LIMIT) { 473 cond_resched(); 474 freed = 0; 475 } 476 } while (next); 477} 478 479static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end) 480{ 481 struct shmem_inode_info *info = SHMEM_I(inode); 482 unsigned long idx; 483 unsigned long size; 484 unsigned long limit; 485 unsigned long stage; 486 unsigned long diroff; 487 struct page **dir; 488 struct page *topdir; 489 struct page *middir; 490 struct page *subdir; 491 swp_entry_t *ptr; 492 LIST_HEAD(pages_to_free); 493 long nr_pages_to_free = 0; 494 long nr_swaps_freed = 0; 495 int offset; 496 int freed; 497 int punch_hole; 498 spinlock_t *needs_lock; 499 spinlock_t *punch_lock; 500 unsigned long upper_limit; 501 502 inode->i_ctime = inode->i_mtime = CURRENT_TIME; 503 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 504 if (idx >= info->next_index) 505 return; 506 507 spin_lock(&info->lock); 508 info->flags |= SHMEM_TRUNCATE; 509 if (likely(end == (loff_t) -1)) { 510 limit = info->next_index; 511 upper_limit = SHMEM_MAX_INDEX; 512 info->next_index = idx; 513 needs_lock = NULL; 514 punch_hole = 0; 515 } else { 516 if (end + 1 >= inode->i_size) { /* we may free a little more */ 517 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >> 518 PAGE_CACHE_SHIFT; 519 upper_limit = SHMEM_MAX_INDEX; 520 } else { 521 limit = (end + 1) >> PAGE_CACHE_SHIFT; 522 upper_limit = limit; 523 } 524 needs_lock = &info->lock; 525 punch_hole = 1; 526 } 527 528 topdir = info->i_indirect; 529 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) { 530 info->i_indirect = NULL; 531 nr_pages_to_free++; 532 list_add(&topdir->lru, &pages_to_free); 533 } 534 spin_unlock(&info->lock); 535 536 if (info->swapped && idx < SHMEM_NR_DIRECT) { 537 ptr = info->i_direct; 538 size = limit; 539 if (size > SHMEM_NR_DIRECT) 540 size = SHMEM_NR_DIRECT; 541 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock); 542 } 543 544 /* 545 * If there are no indirect blocks or we are punching a hole 546 * below indirect blocks, nothing to be done. 547 */ 548 if (!topdir || limit <= SHMEM_NR_DIRECT) 549 goto done2; 550 551 /* 552 * The truncation case has already dropped info->lock, and we're safe 553 * because i_size and next_index have already been lowered, preventing 554 * access beyond. But in the punch_hole case, we still need to take 555 * the lock when updating the swap directory, because there might be 556 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or 557 * shmem_writepage. However, whenever we find we can remove a whole 558 * directory page (not at the misaligned start or end of the range), 559 * we first NULLify its pointer in the level above, and then have no 560 * need to take the lock when updating its contents: needs_lock and 561 * punch_lock (either pointing to info->lock or NULL) manage this. 562 */ 563 564 upper_limit -= SHMEM_NR_DIRECT; 565 limit -= SHMEM_NR_DIRECT; 566 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0; 567 offset = idx % ENTRIES_PER_PAGE; 568 idx -= offset; 569 570 dir = shmem_dir_map(topdir); 571 stage = ENTRIES_PER_PAGEPAGE/2; 572 if (idx < ENTRIES_PER_PAGEPAGE/2) { 573 middir = topdir; 574 diroff = idx/ENTRIES_PER_PAGE; 575 } else { 576 dir += ENTRIES_PER_PAGE/2; 577 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE; 578 while (stage <= idx) 579 stage += ENTRIES_PER_PAGEPAGE; 580 middir = *dir; 581 if (*dir) { 582 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) % 583 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE; 584 if (!diroff && !offset && upper_limit >= stage) { 585 if (needs_lock) { 586 spin_lock(needs_lock); 587 *dir = NULL; 588 spin_unlock(needs_lock); 589 needs_lock = NULL; 590 } else 591 *dir = NULL; 592 nr_pages_to_free++; 593 list_add(&middir->lru, &pages_to_free); 594 } 595 shmem_dir_unmap(dir); 596 dir = shmem_dir_map(middir); 597 } else { 598 diroff = 0; 599 offset = 0; 600 idx = stage; 601 } 602 } 603 604 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) { 605 if (unlikely(idx == stage)) { 606 shmem_dir_unmap(dir); 607 dir = shmem_dir_map(topdir) + 608 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 609 while (!*dir) { 610 dir++; 611 idx += ENTRIES_PER_PAGEPAGE; 612 if (idx >= limit) 613 goto done1; 614 } 615 stage = idx + ENTRIES_PER_PAGEPAGE; 616 middir = *dir; 617 if (punch_hole) 618 needs_lock = &info->lock; 619 if (upper_limit >= stage) { 620 if (needs_lock) { 621 spin_lock(needs_lock); 622 *dir = NULL; 623 spin_unlock(needs_lock); 624 needs_lock = NULL; 625 } else 626 *dir = NULL; 627 nr_pages_to_free++; 628 list_add(&middir->lru, &pages_to_free); 629 } 630 shmem_dir_unmap(dir); 631 cond_resched(); 632 dir = shmem_dir_map(middir); 633 diroff = 0; 634 } 635 punch_lock = needs_lock; 636 subdir = dir[diroff]; 637 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) { 638 if (needs_lock) { 639 spin_lock(needs_lock); 640 dir[diroff] = NULL; 641 spin_unlock(needs_lock); 642 punch_lock = NULL; 643 } else 644 dir[diroff] = NULL; 645 nr_pages_to_free++; 646 list_add(&subdir->lru, &pages_to_free); 647 } 648 if (subdir && page_private(subdir) /* has swap entries */) { 649 size = limit - idx; 650 if (size > ENTRIES_PER_PAGE) 651 size = ENTRIES_PER_PAGE; 652 freed = shmem_map_and_free_swp(subdir, 653 offset, size, &dir, punch_lock); 654 if (!dir) 655 dir = shmem_dir_map(middir); 656 nr_swaps_freed += freed; 657 if (offset || punch_lock) { 658 spin_lock(&info->lock); 659 set_page_private(subdir, 660 page_private(subdir) - freed); 661 spin_unlock(&info->lock); 662 } else 663 BUG_ON(page_private(subdir) != freed); 664 } 665 offset = 0; 666 } 667done1: 668 shmem_dir_unmap(dir); 669done2: 670 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) { 671 /* 672 * Call truncate_inode_pages again: racing shmem_unuse_inode 673 * may have swizzled a page in from swap since vmtruncate or 674 * generic_delete_inode did it, before we lowered next_index. 675 * Also, though shmem_getpage checks i_size before adding to 676 * cache, no recheck after: so fix the narrow window there too. 677 * 678 * Recalling truncate_inode_pages_range and unmap_mapping_range 679 * every time for punch_hole (which never got a chance to clear 680 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive, 681 * yet hardly ever necessary: try to optimize them out later. 682 */ 683 truncate_inode_pages_range(inode->i_mapping, start, end); 684 if (punch_hole) 685 unmap_mapping_range(inode->i_mapping, start, 686 end - start, 1); 687 } 688 689 spin_lock(&info->lock); 690 info->flags &= ~SHMEM_TRUNCATE; 691 info->swapped -= nr_swaps_freed; 692 if (nr_pages_to_free) 693 shmem_free_blocks(inode, nr_pages_to_free); 694 shmem_recalc_inode(inode); 695 spin_unlock(&info->lock); 696 697 /* 698 * Empty swap vector directory pages to be freed? 699 */ 700 if (!list_empty(&pages_to_free)) { 701 pages_to_free.prev->next = NULL; 702 shmem_free_pages(pages_to_free.next); 703 } 704} 705 706static void shmem_truncate(struct inode *inode) 707{ 708 shmem_truncate_range(inode, inode->i_size, (loff_t)-1); 709} 710 711static int shmem_notify_change(struct dentry *dentry, struct iattr *attr) 712{ 713 struct inode *inode = dentry->d_inode; 714 struct page *page = NULL; 715 int error; 716 717 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { 718 if (attr->ia_size < inode->i_size) { 719 /* 720 * If truncating down to a partial page, then 721 * if that page is already allocated, hold it 722 * in memory until the truncation is over, so 723 * truncate_partial_page cannnot miss it were 724 * it assigned to swap. 725 */ 726 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) { 727 (void) shmem_getpage(inode, 728 attr->ia_size>>PAGE_CACHE_SHIFT, 729 &page, SGP_READ, NULL); 730 } 731 /* 732 * Reset SHMEM_PAGEIN flag so that shmem_truncate can 733 * detect if any pages might have been added to cache 734 * after truncate_inode_pages. But we needn't bother 735 * if it's being fully truncated to zero-length: the 736 * nrpages check is efficient enough in that case. 737 */ 738 if (attr->ia_size) { 739 struct shmem_inode_info *info = SHMEM_I(inode); 740 spin_lock(&info->lock); 741 info->flags &= ~SHMEM_PAGEIN; 742 spin_unlock(&info->lock); 743 } 744 } 745 } 746 747 error = inode_change_ok(inode, attr); 748 if (!error) 749 error = inode_setattr(inode, attr); 750#ifdef CONFIG_TMPFS_POSIX_ACL 751 if (!error && (attr->ia_valid & ATTR_MODE)) 752 error = generic_acl_chmod(inode, &shmem_acl_ops); 753#endif 754 if (page) 755 page_cache_release(page); 756 return error; 757} 758 759static void shmem_delete_inode(struct inode *inode) 760{ 761 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 762 struct shmem_inode_info *info = SHMEM_I(inode); 763 764 if (inode->i_op->truncate == shmem_truncate) { 765 truncate_inode_pages(inode->i_mapping, 0); 766 shmem_unacct_size(info->flags, inode->i_size); 767 inode->i_size = 0; 768 shmem_truncate(inode); 769 if (!list_empty(&info->swaplist)) { 770 spin_lock(&shmem_swaplist_lock); 771 list_del_init(&info->swaplist); 772 spin_unlock(&shmem_swaplist_lock); 773 } 774 } 775 BUG_ON(inode->i_blocks); 776 if (sbinfo->max_inodes) { 777 spin_lock(&sbinfo->stat_lock); 778 sbinfo->free_inodes++; 779 spin_unlock(&sbinfo->stat_lock); 780 } 781 clear_inode(inode); 782} 783 784static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir) 785{ 786 swp_entry_t *ptr; 787 788 for (ptr = dir; ptr < edir; ptr++) { 789 if (ptr->val == entry.val) 790 return ptr - dir; 791 } 792 return -1; 793} 794 795static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page) 796{ 797 struct inode *inode; 798 unsigned long idx; 799 unsigned long size; 800 unsigned long limit; 801 unsigned long stage; 802 struct page **dir; 803 struct page *subdir; 804 swp_entry_t *ptr; 805 int offset; 806 807 idx = 0; 808 ptr = info->i_direct; 809 spin_lock(&info->lock); 810 limit = info->next_index; 811 size = limit; 812 if (size > SHMEM_NR_DIRECT) 813 size = SHMEM_NR_DIRECT; 814 offset = shmem_find_swp(entry, ptr, ptr+size); 815 if (offset >= 0) { 816 shmem_swp_balance_unmap(); 817 goto found; 818 } 819 if (!info->i_indirect) 820 goto lost2; 821 822 dir = shmem_dir_map(info->i_indirect); 823 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2; 824 825 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) { 826 if (unlikely(idx == stage)) { 827 shmem_dir_unmap(dir-1); 828 dir = shmem_dir_map(info->i_indirect) + 829 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 830 while (!*dir) { 831 dir++; 832 idx += ENTRIES_PER_PAGEPAGE; 833 if (idx >= limit) 834 goto lost1; 835 } 836 stage = idx + ENTRIES_PER_PAGEPAGE; 837 subdir = *dir; 838 shmem_dir_unmap(dir); 839 dir = shmem_dir_map(subdir); 840 } 841 subdir = *dir; 842 if (subdir && page_private(subdir)) { 843 ptr = shmem_swp_map(subdir); 844 size = limit - idx; 845 if (size > ENTRIES_PER_PAGE) 846 size = ENTRIES_PER_PAGE; 847 offset = shmem_find_swp(entry, ptr, ptr+size); 848 if (offset >= 0) { 849 shmem_dir_unmap(dir); 850 goto found; 851 } 852 shmem_swp_unmap(ptr); 853 } 854 } 855lost1: 856 shmem_dir_unmap(dir-1); 857lost2: 858 spin_unlock(&info->lock); 859 return 0; 860found: 861 idx += offset; 862 inode = &info->vfs_inode; 863 if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) { 864 info->flags |= SHMEM_PAGEIN; 865 shmem_swp_set(info, ptr + offset, 0); 866 } 867 shmem_swp_unmap(ptr); 868 spin_unlock(&info->lock); 869 /* 870 * Decrement swap count even when the entry is left behind: 871 * try_to_unuse will skip over mms, then reincrement count. 872 */ 873 swap_free(entry); 874 return 1; 875} 876 877/* 878 * shmem_unuse() search for an eventually swapped out shmem page. 879 */ 880int shmem_unuse(swp_entry_t entry, struct page *page) 881{ 882 struct list_head *p, *next; 883 struct shmem_inode_info *info; 884 int found = 0; 885 886 spin_lock(&shmem_swaplist_lock); 887 list_for_each_safe(p, next, &shmem_swaplist) { 888 info = list_entry(p, struct shmem_inode_info, swaplist); 889 if (!info->swapped) 890 list_del_init(&info->swaplist); 891 else if (shmem_unuse_inode(info, entry, page)) { 892 /* move head to start search for next from here */ 893 list_move_tail(&shmem_swaplist, &info->swaplist); 894 found = 1; 895 break; 896 } 897 } 898 spin_unlock(&shmem_swaplist_lock); 899 return found; 900} 901 902/* 903 * Move the page from the page cache to the swap cache. 904 */ 905static int shmem_writepage(struct page *page, struct writeback_control *wbc) 906{ 907 struct shmem_inode_info *info; 908 swp_entry_t *entry, swap; 909 struct address_space *mapping; 910 unsigned long index; 911 struct inode *inode; 912 913 BUG_ON(!PageLocked(page)); 914 BUG_ON(page_mapped(page)); 915 916 mapping = page->mapping; 917 index = page->index; 918 inode = mapping->host; 919 info = SHMEM_I(inode); 920 if (info->flags & VM_LOCKED) 921 goto redirty; 922 swap = get_swap_page(); 923 if (!swap.val) 924 goto redirty; 925 926 spin_lock(&info->lock); 927 shmem_recalc_inode(inode); 928 if (index >= info->next_index) { 929 BUG_ON(!(info->flags & SHMEM_TRUNCATE)); 930 goto unlock; 931 } 932 entry = shmem_swp_entry(info, index, NULL); 933 BUG_ON(!entry); 934 BUG_ON(entry->val); 935 936 if (move_to_swap_cache(page, swap) == 0) { 937 shmem_swp_set(info, entry, swap.val); 938 shmem_swp_unmap(entry); 939 spin_unlock(&info->lock); 940 if (list_empty(&info->swaplist)) { 941 spin_lock(&shmem_swaplist_lock); 942 /* move instead of add in case we're racing */ 943 list_move_tail(&info->swaplist, &shmem_swaplist); 944 spin_unlock(&shmem_swaplist_lock); 945 } 946 unlock_page(page); 947 return 0; 948 } 949 950 shmem_swp_unmap(entry); 951unlock: 952 spin_unlock(&info->lock); 953 swap_free(swap); 954redirty: 955 set_page_dirty(page); 956 return AOP_WRITEPAGE_ACTIVATE; /* Return with the page locked */ 957} 958 959#ifdef CONFIG_NUMA 960static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes) 961{ 962 char *nodelist = strchr(value, ':'); 963 int err = 1; 964 965 if (nodelist) { 966 /* NUL-terminate policy string */ 967 *nodelist++ = '\0'; 968 if (nodelist_parse(nodelist, *policy_nodes)) 969 goto out; 970 if (!nodes_subset(*policy_nodes, node_online_map)) 971 goto out; 972 } 973 if (!strcmp(value, "default")) { 974 *policy = MPOL_DEFAULT; 975 /* Don't allow a nodelist */ 976 if (!nodelist) 977 err = 0; 978 } else if (!strcmp(value, "prefer")) { 979 *policy = MPOL_PREFERRED; 980 /* Insist on a nodelist of one node only */ 981 if (nodelist) { 982 char *rest = nodelist; 983 while (isdigit(*rest)) 984 rest++; 985 if (!*rest) 986 err = 0; 987 } 988 } else if (!strcmp(value, "bind")) { 989 *policy = MPOL_BIND; 990 /* Insist on a nodelist */ 991 if (nodelist) 992 err = 0; 993 } else if (!strcmp(value, "interleave")) { 994 *policy = MPOL_INTERLEAVE; 995 /* Default to nodes online if no nodelist */ 996 if (!nodelist) 997 *policy_nodes = node_online_map; 998 err = 0; 999 } 1000out: 1001 /* Restore string for error message */ 1002 if (nodelist) 1003 *--nodelist = ':'; 1004 return err; 1005} 1006 1007static struct page *shmem_swapin_async(struct shared_policy *p, 1008 swp_entry_t entry, unsigned long idx) 1009{ 1010 struct page *page; 1011 struct vm_area_struct pvma; 1012 1013 /* Create a pseudo vma that just contains the policy */ 1014 memset(&pvma, 0, sizeof(struct vm_area_struct)); 1015 pvma.vm_end = PAGE_SIZE; 1016 pvma.vm_pgoff = idx; 1017 pvma.vm_policy = mpol_shared_policy_lookup(p, idx); 1018 page = read_swap_cache_async(entry, &pvma, 0); 1019 mpol_free(pvma.vm_policy); 1020 return page; 1021} 1022 1023struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry, 1024 unsigned long idx) 1025{ 1026 struct shared_policy *p = &info->policy; 1027 int i, num; 1028 struct page *page; 1029 unsigned long offset; 1030 1031 num = valid_swaphandles(entry, &offset); 1032 for (i = 0; i < num; offset++, i++) { 1033 page = shmem_swapin_async(p, 1034 swp_entry(swp_type(entry), offset), idx); 1035 if (!page) 1036 break; 1037 page_cache_release(page); 1038 } 1039 lru_add_drain(); /* Push any new pages onto the LRU now */ 1040 return shmem_swapin_async(p, entry, idx); 1041} 1042 1043static struct page * 1044shmem_alloc_page(gfp_t gfp, struct shmem_inode_info *info, 1045 unsigned long idx) 1046{ 1047 struct vm_area_struct pvma; 1048 struct page *page; 1049 1050 memset(&pvma, 0, sizeof(struct vm_area_struct)); 1051 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx); 1052 pvma.vm_pgoff = idx; 1053 pvma.vm_end = PAGE_SIZE; 1054 page = alloc_page_vma(gfp | __GFP_ZERO, &pvma, 0); 1055 mpol_free(pvma.vm_policy); 1056 return page; 1057} 1058#else 1059static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes) 1060{ 1061 return 1; 1062} 1063 1064static inline struct page * 1065shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx) 1066{ 1067 swapin_readahead(entry, 0, NULL); 1068 return read_swap_cache_async(entry, NULL, 0); 1069} 1070 1071static inline struct page * 1072shmem_alloc_page(gfp_t gfp,struct shmem_inode_info *info, unsigned long idx) 1073{ 1074 return alloc_page(gfp | __GFP_ZERO); 1075} 1076#endif 1077 1078/* 1079 * shmem_getpage - either get the page from swap or allocate a new one 1080 * 1081 * If we allocate a new one we do not mark it dirty. That's up to the 1082 * vm. If we swap it in we mark it dirty since we also free the swap 1083 * entry since a page cannot live in both the swap and page cache 1084 */ 1085static int shmem_getpage(struct inode *inode, unsigned long idx, 1086 struct page **pagep, enum sgp_type sgp, int *type) 1087{ 1088 struct address_space *mapping = inode->i_mapping; 1089 struct shmem_inode_info *info = SHMEM_I(inode); 1090 struct shmem_sb_info *sbinfo; 1091 struct page *filepage = *pagep; 1092 struct page *swappage; 1093 swp_entry_t *entry; 1094 swp_entry_t swap; 1095 int error; 1096 1097 if (idx >= SHMEM_MAX_INDEX) 1098 return -EFBIG; 1099 /* 1100 * Normally, filepage is NULL on entry, and either found 1101 * uptodate immediately, or allocated and zeroed, or read 1102 * in under swappage, which is then assigned to filepage. 1103 * But shmem_readpage and shmem_prepare_write pass in a locked 1104 * filepage, which may be found not uptodate by other callers 1105 * too, and may need to be copied from the swappage read in. 1106 */ 1107repeat: 1108 if (!filepage) 1109 filepage = find_lock_page(mapping, idx); 1110 if (filepage && PageUptodate(filepage)) 1111 goto done; 1112 error = 0; 1113 if (sgp == SGP_QUICK) 1114 goto failed; 1115 1116 spin_lock(&info->lock); 1117 shmem_recalc_inode(inode); 1118 entry = shmem_swp_alloc(info, idx, sgp); 1119 if (IS_ERR(entry)) { 1120 spin_unlock(&info->lock); 1121 error = PTR_ERR(entry); 1122 goto failed; 1123 } 1124 swap = *entry; 1125 1126 if (swap.val) { 1127 /* Look it up and read it in.. */ 1128 swappage = lookup_swap_cache(swap); 1129 if (!swappage) { 1130 shmem_swp_unmap(entry); 1131 /* here we actually do the io */ 1132 if (type && *type == VM_FAULT_MINOR) { 1133 __count_vm_event(PGMAJFAULT); 1134 *type = VM_FAULT_MAJOR; 1135 } 1136 spin_unlock(&info->lock); 1137 swappage = shmem_swapin(info, swap, idx); 1138 if (!swappage) { 1139 spin_lock(&info->lock); 1140 entry = shmem_swp_alloc(info, idx, sgp); 1141 if (IS_ERR(entry)) 1142 error = PTR_ERR(entry); 1143 else { 1144 if (entry->val == swap.val) 1145 error = -ENOMEM; 1146 shmem_swp_unmap(entry); 1147 } 1148 spin_unlock(&info->lock); 1149 if (error) 1150 goto failed; 1151 goto repeat; 1152 } 1153 wait_on_page_locked(swappage); 1154 page_cache_release(swappage); 1155 goto repeat; 1156 } 1157 1158 /* We have to do this with page locked to prevent races */ 1159 if (TestSetPageLocked(swappage)) { 1160 shmem_swp_unmap(entry); 1161 spin_unlock(&info->lock); 1162 wait_on_page_locked(swappage); 1163 page_cache_release(swappage); 1164 goto repeat; 1165 } 1166 if (PageWriteback(swappage)) { 1167 shmem_swp_unmap(entry); 1168 spin_unlock(&info->lock); 1169 wait_on_page_writeback(swappage); 1170 unlock_page(swappage); 1171 page_cache_release(swappage); 1172 goto repeat; 1173 } 1174 if (!PageUptodate(swappage)) { 1175 shmem_swp_unmap(entry); 1176 spin_unlock(&info->lock); 1177 unlock_page(swappage); 1178 page_cache_release(swappage); 1179 error = -EIO; 1180 goto failed; 1181 } 1182 1183 if (filepage) { 1184 shmem_swp_set(info, entry, 0); 1185 shmem_swp_unmap(entry); 1186 delete_from_swap_cache(swappage); 1187 spin_unlock(&info->lock); 1188 copy_highpage(filepage, swappage); 1189 unlock_page(swappage); 1190 page_cache_release(swappage); 1191 flush_dcache_page(filepage); 1192 SetPageUptodate(filepage); 1193 set_page_dirty(filepage); 1194 swap_free(swap); 1195 } else if (!(error = move_from_swap_cache( 1196 swappage, idx, mapping))) { 1197 info->flags |= SHMEM_PAGEIN; 1198 shmem_swp_set(info, entry, 0); 1199 shmem_swp_unmap(entry); 1200 spin_unlock(&info->lock); 1201 filepage = swappage; 1202 swap_free(swap); 1203 } else { 1204 shmem_swp_unmap(entry); 1205 spin_unlock(&info->lock); 1206 unlock_page(swappage); 1207 page_cache_release(swappage); 1208 if (error == -ENOMEM) { 1209 /* let kswapd refresh zone for GFP_ATOMICs */ 1210 congestion_wait(WRITE, HZ/50); 1211 } 1212 goto repeat; 1213 } 1214 } else if (sgp == SGP_READ && !filepage) { 1215 shmem_swp_unmap(entry); 1216 filepage = find_get_page(mapping, idx); 1217 if (filepage && 1218 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) { 1219 spin_unlock(&info->lock); 1220 wait_on_page_locked(filepage); 1221 page_cache_release(filepage); 1222 filepage = NULL; 1223 goto repeat; 1224 } 1225 spin_unlock(&info->lock); 1226 } else { 1227 shmem_swp_unmap(entry); 1228 sbinfo = SHMEM_SB(inode->i_sb); 1229 if (sbinfo->max_blocks) { 1230 spin_lock(&sbinfo->stat_lock); 1231 if (sbinfo->free_blocks == 0 || 1232 shmem_acct_block(info->flags)) { 1233 spin_unlock(&sbinfo->stat_lock); 1234 spin_unlock(&info->lock); 1235 error = -ENOSPC; 1236 goto failed; 1237 } 1238 sbinfo->free_blocks--; 1239 inode->i_blocks += BLOCKS_PER_PAGE; 1240 spin_unlock(&sbinfo->stat_lock); 1241 } else if (shmem_acct_block(info->flags)) { 1242 spin_unlock(&info->lock); 1243 error = -ENOSPC; 1244 goto failed; 1245 } 1246 1247 if (!filepage) { 1248 spin_unlock(&info->lock); 1249 filepage = shmem_alloc_page(mapping_gfp_mask(mapping), 1250 info, 1251 idx); 1252 if (!filepage) { 1253 shmem_unacct_blocks(info->flags, 1); 1254 shmem_free_blocks(inode, 1); 1255 error = -ENOMEM; 1256 goto failed; 1257 } 1258 1259 spin_lock(&info->lock); 1260 entry = shmem_swp_alloc(info, idx, sgp); 1261 if (IS_ERR(entry)) 1262 error = PTR_ERR(entry); 1263 else { 1264 swap = *entry; 1265 shmem_swp_unmap(entry); 1266 } 1267 if (error || swap.val || 0 != add_to_page_cache_lru( 1268 filepage, mapping, idx, GFP_ATOMIC)) { 1269 spin_unlock(&info->lock); 1270 page_cache_release(filepage); 1271 shmem_unacct_blocks(info->flags, 1); 1272 shmem_free_blocks(inode, 1); 1273 filepage = NULL; 1274 if (error) 1275 goto failed; 1276 goto repeat; 1277 } 1278 info->flags |= SHMEM_PAGEIN; 1279 } 1280 1281 info->alloced++; 1282 spin_unlock(&info->lock); 1283 flush_dcache_page(filepage); 1284 SetPageUptodate(filepage); 1285 } 1286done: 1287 if (*pagep != filepage) { 1288 unlock_page(filepage); 1289 *pagep = filepage; 1290 } 1291 return 0; 1292 1293failed: 1294 if (*pagep != filepage) { 1295 unlock_page(filepage); 1296 page_cache_release(filepage); 1297 } 1298 return error; 1299} 1300 1301static struct page *shmem_nopage(struct vm_area_struct *vma, 1302 unsigned long address, int *type) 1303{ 1304 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 1305 struct page *page = NULL; 1306 unsigned long idx; 1307 int error; 1308 1309 idx = (address - vma->vm_start) >> PAGE_SHIFT; 1310 idx += vma->vm_pgoff; 1311 idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT; 1312 if (((loff_t) idx << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 1313 return NOPAGE_SIGBUS; 1314 1315 error = shmem_getpage(inode, idx, &page, SGP_CACHE, type); 1316 if (error) 1317 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS; 1318 1319 mark_page_accessed(page); 1320 return page; 1321} 1322 1323static int shmem_populate(struct vm_area_struct *vma, 1324 unsigned long addr, unsigned long len, 1325 pgprot_t prot, unsigned long pgoff, int nonblock) 1326{ 1327 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 1328 struct mm_struct *mm = vma->vm_mm; 1329 enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE; 1330 unsigned long size; 1331 1332 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT; 1333 if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size) 1334 return -EINVAL; 1335 1336 while ((long) len > 0) { 1337 struct page *page = NULL; 1338 int err; 1339 /* 1340 * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE 1341 */ 1342 err = shmem_getpage(inode, pgoff, &page, sgp, NULL); 1343 if (err) 1344 return err; 1345 /* Page may still be null, but only if nonblock was set. */ 1346 if (page) { 1347 mark_page_accessed(page); 1348 err = install_page(mm, vma, addr, page, prot); 1349 if (err) { 1350 page_cache_release(page); 1351 return err; 1352 } 1353 } else if (vma->vm_flags & VM_NONLINEAR) { 1354 /* No page was found just because we can't read it in 1355 * now (being here implies nonblock != 0), but the page 1356 * may exist, so set the PTE to fault it in later. */ 1357 err = install_file_pte(mm, vma, addr, pgoff, prot); 1358 if (err) 1359 return err; 1360 } 1361 1362 len -= PAGE_SIZE; 1363 addr += PAGE_SIZE; 1364 pgoff++; 1365 } 1366 return 0; 1367} 1368 1369#ifdef CONFIG_NUMA 1370int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new) 1371{ 1372 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1373 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new); 1374} 1375 1376struct mempolicy * 1377shmem_get_policy(struct vm_area_struct *vma, unsigned long addr) 1378{ 1379 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1380 unsigned long idx; 1381 1382 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; 1383 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx); 1384} 1385#endif 1386 1387int shmem_lock(struct file *file, int lock, struct user_struct *user) 1388{ 1389 struct inode *inode = file->f_path.dentry->d_inode; 1390 struct shmem_inode_info *info = SHMEM_I(inode); 1391 int retval = -ENOMEM; 1392 1393 spin_lock(&info->lock); 1394 if (lock && !(info->flags & VM_LOCKED)) { 1395 if (!user_shm_lock(inode->i_size, user)) 1396 goto out_nomem; 1397 info->flags |= VM_LOCKED; 1398 } 1399 if (!lock && (info->flags & VM_LOCKED) && user) { 1400 user_shm_unlock(inode->i_size, user); 1401 info->flags &= ~VM_LOCKED; 1402 } 1403 retval = 0; 1404out_nomem: 1405 spin_unlock(&info->lock); 1406 return retval; 1407} 1408 1409static int shmem_mmap(struct file *file, struct vm_area_struct *vma) 1410{ 1411 file_accessed(file); 1412 vma->vm_ops = &shmem_vm_ops; 1413 return 0; 1414} 1415 1416static struct inode * 1417shmem_get_inode(struct super_block *sb, int mode, dev_t dev) 1418{ 1419 struct inode *inode; 1420 struct shmem_inode_info *info; 1421 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 1422 1423 if (sbinfo->max_inodes) { 1424 spin_lock(&sbinfo->stat_lock); 1425 if (!sbinfo->free_inodes) { 1426 spin_unlock(&sbinfo->stat_lock); 1427 return NULL; 1428 } 1429 sbinfo->free_inodes--; 1430 spin_unlock(&sbinfo->stat_lock); 1431 } 1432 1433 inode = new_inode(sb); 1434 if (inode) { 1435 inode->i_mode = mode; 1436 inode->i_uid = current->fsuid; 1437 inode->i_gid = current->fsgid; 1438 inode->i_blocks = 0; 1439 inode->i_mapping->a_ops = &shmem_aops; 1440 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; 1441 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 1442 inode->i_generation = get_seconds(); 1443 info = SHMEM_I(inode); 1444 memset(info, 0, (char *)inode - (char *)info); 1445 spin_lock_init(&info->lock); 1446 INIT_LIST_HEAD(&info->swaplist); 1447 1448 switch (mode & S_IFMT) { 1449 default: 1450 inode->i_op = &shmem_special_inode_operations; 1451 init_special_inode(inode, mode, dev); 1452 break; 1453 case S_IFREG: 1454 inode->i_op = &shmem_inode_operations; 1455 inode->i_fop = &shmem_file_operations; 1456 mpol_shared_policy_init(&info->policy, sbinfo->policy, 1457 &sbinfo->policy_nodes); 1458 break; 1459 case S_IFDIR: 1460 inc_nlink(inode); 1461 /* Some things misbehave if size == 0 on a directory */ 1462 inode->i_size = 2 * BOGO_DIRENT_SIZE; 1463 inode->i_op = &shmem_dir_inode_operations; 1464 inode->i_fop = &simple_dir_operations; 1465 break; 1466 case S_IFLNK: 1467 /* 1468 * Must not load anything in the rbtree, 1469 * mpol_free_shared_policy will not be called. 1470 */ 1471 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, 1472 NULL); 1473 break; 1474 } 1475 } else if (sbinfo->max_inodes) { 1476 spin_lock(&sbinfo->stat_lock); 1477 sbinfo->free_inodes++; 1478 spin_unlock(&sbinfo->stat_lock); 1479 } 1480 return inode; 1481} 1482 1483#ifdef CONFIG_TMPFS 1484static const struct inode_operations shmem_symlink_inode_operations; 1485static const struct inode_operations shmem_symlink_inline_operations; 1486 1487/* 1488 * Normally tmpfs avoids the use of shmem_readpage and shmem_prepare_write; 1489 * but providing them allows a tmpfs file to be used for splice, sendfile, and 1490 * below the loop driver, in the generic fashion that many filesystems support. 1491 */ 1492static int shmem_readpage(struct file *file, struct page *page) 1493{ 1494 struct inode *inode = page->mapping->host; 1495 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL); 1496 unlock_page(page); 1497 return error; 1498} 1499 1500static int 1501shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to) 1502{ 1503 struct inode *inode = page->mapping->host; 1504 return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL); 1505} 1506 1507static ssize_t 1508shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos) 1509{ 1510 struct inode *inode = file->f_path.dentry->d_inode; 1511 loff_t pos; 1512 unsigned long written; 1513 ssize_t err; 1514 1515 if ((ssize_t) count < 0) 1516 return -EINVAL; 1517 1518 if (!access_ok(VERIFY_READ, buf, count)) 1519 return -EFAULT; 1520 1521 mutex_lock(&inode->i_mutex); 1522 1523 pos = *ppos; 1524 written = 0; 1525 1526 err = generic_write_checks(file, &pos, &count, 0); 1527 if (err || !count) 1528 goto out; 1529 1530 err = remove_suid(file->f_path.dentry); 1531 if (err) 1532 goto out; 1533 1534 inode->i_ctime = inode->i_mtime = CURRENT_TIME; 1535 1536 do { 1537 struct page *page = NULL; 1538 unsigned long bytes, index, offset; 1539 char *kaddr; 1540 int left; 1541 1542 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */ 1543 index = pos >> PAGE_CACHE_SHIFT; 1544 bytes = PAGE_CACHE_SIZE - offset; 1545 if (bytes > count) 1546 bytes = count; 1547 1548 /* 1549 * We don't hold page lock across copy from user - 1550 * what would it guard against? - so no deadlock here. 1551 * But it still may be a good idea to prefault below. 1552 */ 1553 1554 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL); 1555 if (err) 1556 break; 1557 1558 left = bytes; 1559 if (PageHighMem(page)) { 1560 volatile unsigned char dummy; 1561 __get_user(dummy, buf); 1562 __get_user(dummy, buf + bytes - 1); 1563 1564 kaddr = kmap_atomic(page, KM_USER0); 1565 left = __copy_from_user_inatomic(kaddr + offset, 1566 buf, bytes); 1567 kunmap_atomic(kaddr, KM_USER0); 1568 } 1569 if (left) { 1570 kaddr = kmap(page); 1571 left = __copy_from_user(kaddr + offset, buf, bytes); 1572 kunmap(page); 1573 } 1574 1575 written += bytes; 1576 count -= bytes; 1577 pos += bytes; 1578 buf += bytes; 1579 if (pos > inode->i_size) 1580 i_size_write(inode, pos); 1581 1582 flush_dcache_page(page); 1583 set_page_dirty(page); 1584 mark_page_accessed(page); 1585 page_cache_release(page); 1586 1587 if (left) { 1588 pos -= left; 1589 written -= left; 1590 err = -EFAULT; 1591 break; 1592 } 1593 1594 /* 1595 * Our dirty pages are not counted in nr_dirty, 1596 * and we do not attempt to balance dirty pages. 1597 */ 1598 1599 cond_resched(); 1600 } while (count); 1601 1602 *ppos = pos; 1603 if (written) 1604 err = written; 1605out: 1606 mutex_unlock(&inode->i_mutex); 1607 return err; 1608} 1609 1610static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor) 1611{ 1612 struct inode *inode = filp->f_path.dentry->d_inode; 1613 struct address_space *mapping = inode->i_mapping; 1614 unsigned long index, offset; 1615 1616 index = *ppos >> PAGE_CACHE_SHIFT; 1617 offset = *ppos & ~PAGE_CACHE_MASK; 1618 1619 for (;;) { 1620 struct page *page = NULL; 1621 unsigned long end_index, nr, ret; 1622 loff_t i_size = i_size_read(inode); 1623 1624 end_index = i_size >> PAGE_CACHE_SHIFT; 1625 if (index > end_index) 1626 break; 1627 if (index == end_index) { 1628 nr = i_size & ~PAGE_CACHE_MASK; 1629 if (nr <= offset) 1630 break; 1631 } 1632 1633 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL); 1634 if (desc->error) { 1635 if (desc->error == -EINVAL) 1636 desc->error = 0; 1637 break; 1638 } 1639 1640 /* 1641 * We must evaluate after, since reads (unlike writes) 1642 * are called without i_mutex protection against truncate 1643 */ 1644 nr = PAGE_CACHE_SIZE; 1645 i_size = i_size_read(inode); 1646 end_index = i_size >> PAGE_CACHE_SHIFT; 1647 if (index == end_index) { 1648 nr = i_size & ~PAGE_CACHE_MASK; 1649 if (nr <= offset) { 1650 if (page) 1651 page_cache_release(page); 1652 break; 1653 } 1654 } 1655 nr -= offset; 1656 1657 if (page) { 1658 /* 1659 * If users can be writing to this page using arbitrary 1660 * virtual addresses, take care about potential aliasing 1661 * before reading the page on the kernel side. 1662 */ 1663 if (mapping_writably_mapped(mapping)) 1664 flush_dcache_page(page); 1665 /* 1666 * Mark the page accessed if we read the beginning. 1667 */ 1668 if (!offset) 1669 mark_page_accessed(page); 1670 } else { 1671 page = ZERO_PAGE(0); 1672 page_cache_get(page); 1673 } 1674 1675 /* 1676 * Ok, we have the page, and it's up-to-date, so 1677 * now we can copy it to user space... 1678 * 1679 * The actor routine returns how many bytes were actually used.. 1680 * NOTE! This may not be the same as how much of a user buffer 1681 * we filled up (we may be padding etc), so we can only update 1682 * "pos" here (the actor routine has to update the user buffer 1683 * pointers and the remaining count). 1684 */ 1685 ret = actor(desc, page, offset, nr); 1686 offset += ret; 1687 index += offset >> PAGE_CACHE_SHIFT; 1688 offset &= ~PAGE_CACHE_MASK; 1689 1690 page_cache_release(page); 1691 if (ret != nr || !desc->count) 1692 break; 1693 1694 cond_resched(); 1695 } 1696 1697 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; 1698 file_accessed(filp); 1699} 1700 1701static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos) 1702{ 1703 read_descriptor_t desc; 1704 1705 if ((ssize_t) count < 0) 1706 return -EINVAL; 1707 if (!access_ok(VERIFY_WRITE, buf, count)) 1708 return -EFAULT; 1709 if (!count) 1710 return 0; 1711 1712 desc.written = 0; 1713 desc.count = count; 1714 desc.arg.buf = buf; 1715 desc.error = 0; 1716 1717 do_shmem_file_read(filp, ppos, &desc, file_read_actor); 1718 if (desc.written) 1719 return desc.written; 1720 return desc.error; 1721} 1722 1723static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) 1724{ 1725 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); 1726 1727 buf->f_type = TMPFS_MAGIC; 1728 buf->f_bsize = PAGE_CACHE_SIZE; 1729 buf->f_namelen = NAME_MAX; 1730 spin_lock(&sbinfo->stat_lock); 1731 if (sbinfo->max_blocks) { 1732 buf->f_blocks = sbinfo->max_blocks; 1733 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks; 1734 } 1735 if (sbinfo->max_inodes) { 1736 buf->f_files = sbinfo->max_inodes; 1737 buf->f_ffree = sbinfo->free_inodes; 1738 } 1739 /* else leave those fields 0 like simple_statfs */ 1740 spin_unlock(&sbinfo->stat_lock); 1741 return 0; 1742} 1743 1744/* 1745 * File creation. Allocate an inode, and we're done.. 1746 */ 1747static int 1748shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 1749{ 1750 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev); 1751 int error = -ENOSPC; 1752 1753 if (inode) { 1754 error = security_inode_init_security(inode, dir, NULL, NULL, 1755 NULL); 1756 if (error) { 1757 if (error != -EOPNOTSUPP) { 1758 iput(inode); 1759 return error; 1760 } 1761 } 1762 error = shmem_acl_init(inode, dir); 1763 if (error) { 1764 iput(inode); 1765 return error; 1766 } 1767 if (dir->i_mode & S_ISGID) { 1768 inode->i_gid = dir->i_gid; 1769 if (S_ISDIR(mode)) 1770 inode->i_mode |= S_ISGID; 1771 } 1772 dir->i_size += BOGO_DIRENT_SIZE; 1773 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1774 d_instantiate(dentry, inode); 1775 dget(dentry); /* Extra count - pin the dentry in core */ 1776 } 1777 return error; 1778} 1779 1780static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1781{ 1782 int error; 1783 1784 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) 1785 return error; 1786 inc_nlink(dir); 1787 return 0; 1788} 1789 1790static int shmem_create(struct inode *dir, struct dentry *dentry, int mode, 1791 struct nameidata *nd) 1792{ 1793 return shmem_mknod(dir, dentry, mode | S_IFREG, 0); 1794} 1795 1796/* 1797 * Link a file.. 1798 */ 1799static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1800{ 1801 struct inode *inode = old_dentry->d_inode; 1802 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 1803 1804 /* 1805 * No ordinary (disk based) filesystem counts links as inodes; 1806 * but each new link needs a new dentry, pinning lowmem, and 1807 * tmpfs dentries cannot be pruned until they are unlinked. 1808 */ 1809 if (sbinfo->max_inodes) { 1810 spin_lock(&sbinfo->stat_lock); 1811 if (!sbinfo->free_inodes) { 1812 spin_unlock(&sbinfo->stat_lock); 1813 return -ENOSPC; 1814 } 1815 sbinfo->free_inodes--; 1816 spin_unlock(&sbinfo->stat_lock); 1817 } 1818 1819 dir->i_size += BOGO_DIRENT_SIZE; 1820 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1821 inc_nlink(inode); 1822 atomic_inc(&inode->i_count); /* New dentry reference */ 1823 dget(dentry); /* Extra pinning count for the created dentry */ 1824 d_instantiate(dentry, inode); 1825 return 0; 1826} 1827 1828static int shmem_unlink(struct inode *dir, struct dentry *dentry) 1829{ 1830 struct inode *inode = dentry->d_inode; 1831 1832 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) { 1833 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 1834 if (sbinfo->max_inodes) { 1835 spin_lock(&sbinfo->stat_lock); 1836 sbinfo->free_inodes++; 1837 spin_unlock(&sbinfo->stat_lock); 1838 } 1839 } 1840 1841 dir->i_size -= BOGO_DIRENT_SIZE; 1842 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1843 drop_nlink(inode); 1844 dput(dentry); /* Undo the count from "create" - this does all the work */ 1845 return 0; 1846} 1847 1848static int shmem_rmdir(struct inode *dir, struct dentry *dentry) 1849{ 1850 if (!simple_empty(dentry)) 1851 return -ENOTEMPTY; 1852 1853 drop_nlink(dentry->d_inode); 1854 drop_nlink(dir); 1855 return shmem_unlink(dir, dentry); 1856} 1857 1858/* 1859 * The VFS layer already does all the dentry stuff for rename, 1860 * we just have to decrement the usage count for the target if 1861 * it exists so that the VFS layer correctly free's it when it 1862 * gets overwritten. 1863 */ 1864static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) 1865{ 1866 struct inode *inode = old_dentry->d_inode; 1867 int they_are_dirs = S_ISDIR(inode->i_mode); 1868 1869 if (!simple_empty(new_dentry)) 1870 return -ENOTEMPTY; 1871 1872 if (new_dentry->d_inode) { 1873 (void) shmem_unlink(new_dir, new_dentry); 1874 if (they_are_dirs) 1875 drop_nlink(old_dir); 1876 } else if (they_are_dirs) { 1877 drop_nlink(old_dir); 1878 inc_nlink(new_dir); 1879 } 1880 1881 old_dir->i_size -= BOGO_DIRENT_SIZE; 1882 new_dir->i_size += BOGO_DIRENT_SIZE; 1883 old_dir->i_ctime = old_dir->i_mtime = 1884 new_dir->i_ctime = new_dir->i_mtime = 1885 inode->i_ctime = CURRENT_TIME; 1886 return 0; 1887} 1888 1889static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1890{ 1891 int error; 1892 int len; 1893 struct inode *inode; 1894 struct page *page = NULL; 1895 char *kaddr; 1896 struct shmem_inode_info *info; 1897 1898 len = strlen(symname) + 1; 1899 if (len > PAGE_CACHE_SIZE) 1900 return -ENAMETOOLONG; 1901 1902 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0); 1903 if (!inode) 1904 return -ENOSPC; 1905 1906 error = security_inode_init_security(inode, dir, NULL, NULL, 1907 NULL); 1908 if (error) { 1909 if (error != -EOPNOTSUPP) { 1910 iput(inode); 1911 return error; 1912 } 1913 error = 0; 1914 } 1915 1916 info = SHMEM_I(inode); 1917 inode->i_size = len-1; 1918 if (len <= (char *)inode - (char *)info) { 1919 /* do it inline */ 1920 memcpy(info, symname, len); 1921 inode->i_op = &shmem_symlink_inline_operations; 1922 } else { 1923 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); 1924 if (error) { 1925 iput(inode); 1926 return error; 1927 } 1928 inode->i_op = &shmem_symlink_inode_operations; 1929 kaddr = kmap_atomic(page, KM_USER0); 1930 memcpy(kaddr, symname, len); 1931 kunmap_atomic(kaddr, KM_USER0); 1932 set_page_dirty(page); 1933 page_cache_release(page); 1934 } 1935 if (dir->i_mode & S_ISGID) 1936 inode->i_gid = dir->i_gid; 1937 dir->i_size += BOGO_DIRENT_SIZE; 1938 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1939 d_instantiate(dentry, inode); 1940 dget(dentry); 1941 return 0; 1942} 1943 1944static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd) 1945{ 1946 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode)); 1947 return NULL; 1948} 1949 1950static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) 1951{ 1952 struct page *page = NULL; 1953 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); 1954 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page)); 1955 return page; 1956} 1957 1958static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 1959{ 1960 if (!IS_ERR(nd_get_link(nd))) { 1961 struct page *page = cookie; 1962 kunmap(page); 1963 mark_page_accessed(page); 1964 page_cache_release(page); 1965 } 1966} 1967 1968static const struct inode_operations shmem_symlink_inline_operations = { 1969 .readlink = generic_readlink, 1970 .follow_link = shmem_follow_link_inline, 1971}; 1972 1973static const struct inode_operations shmem_symlink_inode_operations = { 1974 .truncate = shmem_truncate, 1975 .readlink = generic_readlink, 1976 .follow_link = shmem_follow_link, 1977 .put_link = shmem_put_link, 1978}; 1979 1980#ifdef CONFIG_TMPFS_POSIX_ACL 1981/** 1982 * Superblocks without xattr inode operations will get security.* xattr 1983 * support from the VFS "for free". As soon as we have any other xattrs 1984 * like ACLs, we also need to implement the security.* handlers at 1985 * filesystem level, though. 1986 */ 1987 1988static size_t shmem_xattr_security_list(struct inode *inode, char *list, 1989 size_t list_len, const char *name, 1990 size_t name_len) 1991{ 1992 return security_inode_listsecurity(inode, list, list_len); 1993} 1994 1995static int shmem_xattr_security_get(struct inode *inode, const char *name, 1996 void *buffer, size_t size) 1997{ 1998 if (strcmp(name, "") == 0) 1999 return -EINVAL; 2000 return security_inode_getsecurity(inode, name, buffer, size, 2001 -EOPNOTSUPP); 2002} 2003 2004static int shmem_xattr_security_set(struct inode *inode, const char *name, 2005 const void *value, size_t size, int flags) 2006{ 2007 if (strcmp(name, "") == 0) 2008 return -EINVAL; 2009 return security_inode_setsecurity(inode, name, value, size, flags); 2010} 2011 2012static struct xattr_handler shmem_xattr_security_handler = { 2013 .prefix = XATTR_SECURITY_PREFIX, 2014 .list = shmem_xattr_security_list, 2015 .get = shmem_xattr_security_get, 2016 .set = shmem_xattr_security_set, 2017}; 2018 2019static struct xattr_handler *shmem_xattr_handlers[] = { 2020 &shmem_xattr_acl_access_handler, 2021 &shmem_xattr_acl_default_handler, 2022 &shmem_xattr_security_handler, 2023 NULL 2024}; 2025#endif 2026 2027static struct dentry *shmem_get_parent(struct dentry *child) 2028{ 2029 return ERR_PTR(-ESTALE); 2030} 2031 2032static int shmem_match(struct inode *ino, void *vfh) 2033{ 2034 __u32 *fh = vfh; 2035 __u64 inum = fh[2]; 2036 inum = (inum << 32) | fh[1]; 2037 return ino->i_ino == inum && fh[0] == ino->i_generation; 2038} 2039 2040static struct dentry *shmem_get_dentry(struct super_block *sb, void *vfh) 2041{ 2042 struct dentry *de = NULL; 2043 struct inode *inode; 2044 __u32 *fh = vfh; 2045 __u64 inum = fh[2]; 2046 inum = (inum << 32) | fh[1]; 2047 2048 inode = ilookup5(sb, (unsigned long)(inum+fh[0]), shmem_match, vfh); 2049 if (inode) { 2050 de = d_find_alias(inode); 2051 iput(inode); 2052 } 2053 2054 return de? de: ERR_PTR(-ESTALE); 2055} 2056 2057static struct dentry *shmem_decode_fh(struct super_block *sb, __u32 *fh, 2058 int len, int type, 2059 int (*acceptable)(void *context, struct dentry *de), 2060 void *context) 2061{ 2062 if (len < 3) 2063 return ERR_PTR(-ESTALE); 2064 2065 return sb->s_export_op->find_exported_dentry(sb, fh, NULL, acceptable, 2066 context); 2067} 2068 2069static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len, 2070 int connectable) 2071{ 2072 struct inode *inode = dentry->d_inode; 2073 2074 if (*len < 3) 2075 return 255; 2076 2077 if (hlist_unhashed(&inode->i_hash)) { 2078 /* Unfortunately insert_inode_hash is not idempotent, 2079 * so as we hash inodes here rather than at creation 2080 * time, we need a lock to ensure we only try 2081 * to do it once 2082 */ 2083 static DEFINE_SPINLOCK(lock); 2084 spin_lock(&lock); 2085 if (hlist_unhashed(&inode->i_hash)) 2086 __insert_inode_hash(inode, 2087 inode->i_ino + inode->i_generation); 2088 spin_unlock(&lock); 2089 } 2090 2091 fh[0] = inode->i_generation; 2092 fh[1] = inode->i_ino; 2093 fh[2] = ((__u64)inode->i_ino) >> 32; 2094 2095 *len = 3; 2096 return 1; 2097} 2098 2099static struct export_operations shmem_export_ops = { 2100 .get_parent = shmem_get_parent, 2101 .get_dentry = shmem_get_dentry, 2102 .encode_fh = shmem_encode_fh, 2103 .decode_fh = shmem_decode_fh, 2104}; 2105 2106static int shmem_parse_options(char *options, int *mode, uid_t *uid, 2107 gid_t *gid, unsigned long *blocks, unsigned long *inodes, 2108 int *policy, nodemask_t *policy_nodes) 2109{ 2110 char *this_char, *value, *rest; 2111 2112 while (options != NULL) { 2113 this_char = options; 2114 for (;;) { 2115 /* 2116 * NUL-terminate this option: unfortunately, 2117 * mount options form a comma-separated list, 2118 * but mpol's nodelist may also contain commas. 2119 */ 2120 options = strchr(options, ','); 2121 if (options == NULL) 2122 break; 2123 options++; 2124 if (!isdigit(*options)) { 2125 options[-1] = '\0'; 2126 break; 2127 } 2128 } 2129 if (!*this_char) 2130 continue; 2131 if ((value = strchr(this_char,'=')) != NULL) { 2132 *value++ = 0; 2133 } else { 2134 printk(KERN_ERR 2135 "tmpfs: No value for mount option '%s'\n", 2136 this_char); 2137 return 1; 2138 } 2139 2140 if (!strcmp(this_char,"size")) { 2141 unsigned long long size; 2142 size = memparse(value,&rest); 2143 if (*rest == '%') { 2144 size <<= PAGE_SHIFT; 2145 size *= totalram_pages; 2146 do_div(size, 100); 2147 rest++; 2148 } 2149 if (*rest) 2150 goto bad_val; 2151 *blocks = size >> PAGE_CACHE_SHIFT; 2152 } else if (!strcmp(this_char,"nr_blocks")) { 2153 *blocks = memparse(value,&rest); 2154 if (*rest) 2155 goto bad_val; 2156 } else if (!strcmp(this_char,"nr_inodes")) { 2157 *inodes = memparse(value,&rest); 2158 if (*rest) 2159 goto bad_val; 2160 } else if (!strcmp(this_char,"mode")) { 2161 if (!mode) 2162 continue; 2163 *mode = simple_strtoul(value,&rest,8); 2164 if (*rest) 2165 goto bad_val; 2166 } else if (!strcmp(this_char,"uid")) { 2167 if (!uid) 2168 continue; 2169 *uid = simple_strtoul(value,&rest,0); 2170 if (*rest) 2171 goto bad_val; 2172 } else if (!strcmp(this_char,"gid")) { 2173 if (!gid) 2174 continue; 2175 *gid = simple_strtoul(value,&rest,0); 2176 if (*rest) 2177 goto bad_val; 2178 } else if (!strcmp(this_char,"mpol")) { 2179 if (shmem_parse_mpol(value,policy,policy_nodes)) 2180 goto bad_val; 2181 } else { 2182 printk(KERN_ERR "tmpfs: Bad mount option %s\n", 2183 this_char); 2184 return 1; 2185 } 2186 } 2187 return 0; 2188 2189bad_val: 2190 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", 2191 value, this_char); 2192 return 1; 2193 2194} 2195 2196static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) 2197{ 2198 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 2199 unsigned long max_blocks = sbinfo->max_blocks; 2200 unsigned long max_inodes = sbinfo->max_inodes; 2201 int policy = sbinfo->policy; 2202 nodemask_t policy_nodes = sbinfo->policy_nodes; 2203 unsigned long blocks; 2204 unsigned long inodes; 2205 int error = -EINVAL; 2206 2207 if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks, 2208 &max_inodes, &policy, &policy_nodes)) 2209 return error; 2210 2211 spin_lock(&sbinfo->stat_lock); 2212 blocks = sbinfo->max_blocks - sbinfo->free_blocks; 2213 inodes = sbinfo->max_inodes - sbinfo->free_inodes; 2214 if (max_blocks < blocks) 2215 goto out; 2216 if (max_inodes < inodes) 2217 goto out; 2218 /* 2219 * Those tests also disallow limited->unlimited while any are in 2220 * use, so i_blocks will always be zero when max_blocks is zero; 2221 * but we must separately disallow unlimited->limited, because 2222 * in that case we have no record of how much is already in use. 2223 */ 2224 if (max_blocks && !sbinfo->max_blocks) 2225 goto out; 2226 if (max_inodes && !sbinfo->max_inodes) 2227 goto out; 2228 2229 error = 0; 2230 sbinfo->max_blocks = max_blocks; 2231 sbinfo->free_blocks = max_blocks - blocks; 2232 sbinfo->max_inodes = max_inodes; 2233 sbinfo->free_inodes = max_inodes - inodes; 2234 sbinfo->policy = policy; 2235 sbinfo->policy_nodes = policy_nodes; 2236out: 2237 spin_unlock(&sbinfo->stat_lock); 2238 return error; 2239} 2240#endif 2241 2242static void shmem_put_super(struct super_block *sb) 2243{ 2244 kfree(sb->s_fs_info); 2245 sb->s_fs_info = NULL; 2246} 2247 2248static int shmem_fill_super(struct super_block *sb, 2249 void *data, int silent) 2250{ 2251 struct inode *inode; 2252 struct dentry *root; 2253 int mode = S_IRWXUGO | S_ISVTX; 2254 uid_t uid = current->fsuid; 2255 gid_t gid = current->fsgid; 2256 int err = -ENOMEM; 2257 struct shmem_sb_info *sbinfo; 2258 unsigned long blocks = 0; 2259 unsigned long inodes = 0; 2260 int policy = MPOL_DEFAULT; 2261 nodemask_t policy_nodes = node_online_map; 2262 2263#ifdef CONFIG_TMPFS 2264 /* 2265 * Per default we only allow half of the physical ram per 2266 * tmpfs instance, limiting inodes to one per page of lowmem; 2267 * but the internal instance is left unlimited. 2268 */ 2269 if (!(sb->s_flags & MS_NOUSER)) { 2270 blocks = totalram_pages / 2; 2271 inodes = totalram_pages - totalhigh_pages; 2272 if (inodes > blocks) 2273 inodes = blocks; 2274 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks, 2275 &inodes, &policy, &policy_nodes)) 2276 return -EINVAL; 2277 } 2278 sb->s_export_op = &shmem_export_ops; 2279#else 2280 sb->s_flags |= MS_NOUSER; 2281#endif 2282 2283 /* Round up to L1_CACHE_BYTES to resist false sharing */ 2284 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info), 2285 L1_CACHE_BYTES), GFP_KERNEL); 2286 if (!sbinfo) 2287 return -ENOMEM; 2288 2289 spin_lock_init(&sbinfo->stat_lock); 2290 sbinfo->max_blocks = blocks; 2291 sbinfo->free_blocks = blocks; 2292 sbinfo->max_inodes = inodes; 2293 sbinfo->free_inodes = inodes; 2294 sbinfo->policy = policy; 2295 sbinfo->policy_nodes = policy_nodes; 2296 2297 sb->s_fs_info = sbinfo; 2298 sb->s_maxbytes = SHMEM_MAX_BYTES; 2299 sb->s_blocksize = PAGE_CACHE_SIZE; 2300 sb->s_blocksize_bits = PAGE_CACHE_SHIFT; 2301 sb->s_magic = TMPFS_MAGIC; 2302 sb->s_op = &shmem_ops; 2303 sb->s_time_gran = 1; 2304#ifdef CONFIG_TMPFS_POSIX_ACL 2305 sb->s_xattr = shmem_xattr_handlers; 2306 sb->s_flags |= MS_POSIXACL; 2307#endif 2308 2309 inode = shmem_get_inode(sb, S_IFDIR | mode, 0); 2310 if (!inode) 2311 goto failed; 2312 inode->i_uid = uid; 2313 inode->i_gid = gid; 2314 root = d_alloc_root(inode); 2315 if (!root) 2316 goto failed_iput; 2317 sb->s_root = root; 2318 return 0; 2319 2320failed_iput: 2321 iput(inode); 2322failed: 2323 shmem_put_super(sb); 2324 return err; 2325} 2326 2327static struct kmem_cache *shmem_inode_cachep; 2328 2329static struct inode *shmem_alloc_inode(struct super_block *sb) 2330{ 2331 struct shmem_inode_info *p; 2332 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); 2333 if (!p) 2334 return NULL; 2335 return &p->vfs_inode; 2336} 2337 2338static void shmem_destroy_inode(struct inode *inode) 2339{ 2340 if ((inode->i_mode & S_IFMT) == S_IFREG) { 2341 /* only struct inode is valid if it's an inline symlink */ 2342 mpol_free_shared_policy(&SHMEM_I(inode)->policy); 2343 } 2344 shmem_acl_destroy_inode(inode); 2345 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); 2346} 2347 2348static void init_once(void *foo, struct kmem_cache *cachep, 2349 unsigned long flags) 2350{ 2351 struct shmem_inode_info *p = (struct shmem_inode_info *) foo; 2352 2353 inode_init_once(&p->vfs_inode); 2354#ifdef CONFIG_TMPFS_POSIX_ACL 2355 p->i_acl = NULL; 2356 p->i_default_acl = NULL; 2357#endif 2358} 2359 2360static int init_inodecache(void) 2361{ 2362 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", 2363 sizeof(struct shmem_inode_info), 2364 0, 0, init_once, NULL); 2365 if (shmem_inode_cachep == NULL) 2366 return -ENOMEM; 2367 return 0; 2368} 2369 2370static void destroy_inodecache(void) 2371{ 2372 kmem_cache_destroy(shmem_inode_cachep); 2373} 2374 2375static const struct address_space_operations shmem_aops = { 2376 .writepage = shmem_writepage, 2377 .set_page_dirty = __set_page_dirty_no_writeback, 2378#ifdef CONFIG_TMPFS 2379 .readpage = shmem_readpage, 2380 .prepare_write = shmem_prepare_write, 2381 .commit_write = simple_commit_write, 2382#endif 2383 .migratepage = migrate_page, 2384}; 2385 2386static const struct file_operations shmem_file_operations = { 2387 .mmap = shmem_mmap, 2388#ifdef CONFIG_TMPFS 2389 .llseek = generic_file_llseek, 2390 .read = shmem_file_read, 2391 .write = shmem_file_write, 2392 .fsync = simple_sync_file, 2393 .splice_read = generic_file_splice_read, 2394 .splice_write = generic_file_splice_write, 2395#endif 2396}; 2397 2398static const struct inode_operations shmem_inode_operations = { 2399 .truncate = shmem_truncate, 2400 .setattr = shmem_notify_change, 2401 .truncate_range = shmem_truncate_range, 2402#ifdef CONFIG_TMPFS_POSIX_ACL 2403 .setxattr = generic_setxattr, 2404 .getxattr = generic_getxattr, 2405 .listxattr = generic_listxattr, 2406 .removexattr = generic_removexattr, 2407 .permission = shmem_permission, 2408#endif 2409 2410}; 2411 2412static const struct inode_operations shmem_dir_inode_operations = { 2413#ifdef CONFIG_TMPFS 2414 .create = shmem_create, 2415 .lookup = simple_lookup, 2416 .link = shmem_link, 2417 .unlink = shmem_unlink, 2418 .symlink = shmem_symlink, 2419 .mkdir = shmem_mkdir, 2420 .rmdir = shmem_rmdir, 2421 .mknod = shmem_mknod, 2422 .rename = shmem_rename, 2423#endif 2424#ifdef CONFIG_TMPFS_POSIX_ACL 2425 .setattr = shmem_notify_change, 2426 .setxattr = generic_setxattr, 2427 .getxattr = generic_getxattr, 2428 .listxattr = generic_listxattr, 2429 .removexattr = generic_removexattr, 2430 .permission = shmem_permission, 2431#endif 2432}; 2433 2434static const struct inode_operations shmem_special_inode_operations = { 2435#ifdef CONFIG_TMPFS_POSIX_ACL 2436 .setattr = shmem_notify_change, 2437 .setxattr = generic_setxattr, 2438 .getxattr = generic_getxattr, 2439 .listxattr = generic_listxattr, 2440 .removexattr = generic_removexattr, 2441 .permission = shmem_permission, 2442#endif 2443}; 2444 2445static const struct super_operations shmem_ops = { 2446 .alloc_inode = shmem_alloc_inode, 2447 .destroy_inode = shmem_destroy_inode, 2448#ifdef CONFIG_TMPFS 2449 .statfs = shmem_statfs, 2450 .remount_fs = shmem_remount_fs, 2451#endif 2452 .delete_inode = shmem_delete_inode, 2453 .drop_inode = generic_delete_inode, 2454 .put_super = shmem_put_super, 2455}; 2456 2457static struct vm_operations_struct shmem_vm_ops = { 2458 .nopage = shmem_nopage, 2459 .populate = shmem_populate, 2460#ifdef CONFIG_NUMA 2461 .set_policy = shmem_set_policy, 2462 .get_policy = shmem_get_policy, 2463#endif 2464}; 2465 2466 2467static int shmem_get_sb(struct file_system_type *fs_type, 2468 int flags, const char *dev_name, void *data, struct vfsmount *mnt) 2469{ 2470 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt); 2471} 2472 2473static struct file_system_type tmpfs_fs_type = { 2474 .owner = THIS_MODULE, 2475 .name = "tmpfs", 2476 .get_sb = shmem_get_sb, 2477 .kill_sb = kill_litter_super, 2478}; 2479static struct vfsmount *shm_mnt; 2480 2481static int __init init_tmpfs(void) 2482{ 2483 int error; 2484 2485 error = init_inodecache(); 2486 if (error) 2487 goto out3; 2488 2489 error = register_filesystem(&tmpfs_fs_type); 2490 if (error) { 2491 printk(KERN_ERR "Could not register tmpfs\n"); 2492 goto out2; 2493 } 2494 2495 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER, 2496 tmpfs_fs_type.name, NULL); 2497 if (IS_ERR(shm_mnt)) { 2498 error = PTR_ERR(shm_mnt); 2499 printk(KERN_ERR "Could not kern_mount tmpfs\n"); 2500 goto out1; 2501 } 2502 return 0; 2503 2504out1: 2505 unregister_filesystem(&tmpfs_fs_type); 2506out2: 2507 destroy_inodecache(); 2508out3: 2509 shm_mnt = ERR_PTR(error); 2510 return error; 2511} 2512module_init(init_tmpfs) 2513 2514/* 2515 * shmem_file_setup - get an unlinked file living in tmpfs 2516 * 2517 * @name: name for dentry (to be seen in /proc/<pid>/maps 2518 * @size: size to be set for the file 2519 * 2520 */ 2521struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags) 2522{ 2523 int error; 2524 struct file *file; 2525 struct inode *inode; 2526 struct dentry *dentry, *root; 2527 struct qstr this; 2528 2529 if (IS_ERR(shm_mnt)) 2530 return (void *)shm_mnt; 2531 2532 if (size < 0 || size > SHMEM_MAX_BYTES) 2533 return ERR_PTR(-EINVAL); 2534 2535 if (shmem_acct_size(flags, size)) 2536 return ERR_PTR(-ENOMEM); 2537 2538 error = -ENOMEM; 2539 this.name = name; 2540 this.len = strlen(name); 2541 this.hash = 0; /* will go */ 2542 root = shm_mnt->mnt_root; 2543 dentry = d_alloc(root, &this); 2544 if (!dentry) 2545 goto put_memory; 2546 2547 error = -ENFILE; 2548 file = get_empty_filp(); 2549 if (!file) 2550 goto put_dentry; 2551 2552 error = -ENOSPC; 2553 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0); 2554 if (!inode) 2555 goto close_file; 2556 2557 SHMEM_I(inode)->flags = flags & VM_ACCOUNT; 2558 d_instantiate(dentry, inode); 2559 inode->i_size = size; 2560 inode->i_nlink = 0; /* It is unlinked */ 2561 file->f_path.mnt = mntget(shm_mnt); 2562 file->f_path.dentry = dentry; 2563 file->f_mapping = inode->i_mapping; 2564 file->f_op = &shmem_file_operations; 2565 file->f_mode = FMODE_WRITE | FMODE_READ; 2566 return file; 2567 2568close_file: 2569 put_filp(file); 2570put_dentry: 2571 dput(dentry); 2572put_memory: 2573 shmem_unacct_size(flags, size); 2574 return ERR_PTR(error); 2575} 2576 2577/* 2578 * shmem_zero_setup - setup a shared anonymous mapping 2579 * 2580 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff 2581 */ 2582int shmem_zero_setup(struct vm_area_struct *vma) 2583{ 2584 struct file *file; 2585 loff_t size = vma->vm_end - vma->vm_start; 2586 2587 file = shmem_file_setup("dev/zero", size, vma->vm_flags); 2588 if (IS_ERR(file)) 2589 return PTR_ERR(file); 2590 2591 if (vma->vm_file) 2592 fput(vma->vm_file); 2593 vma->vm_file = file; 2594 vma->vm_ops = &shmem_vm_ops; 2595 return 0; 2596} 2597