shmem.c revision a2646d1e6c8d2239d8054a7d342eb9775a1d273a
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 */ 408static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir) 409{ 410 swp_entry_t *ptr; 411 int freed = 0; 412 413 for (ptr = dir; ptr < edir; ptr++) { 414 if (ptr->val) { 415 free_swap_and_cache(*ptr); 416 *ptr = (swp_entry_t){0}; 417 freed++; 418 } 419 } 420 return freed; 421} 422 423static int shmem_map_and_free_swp(struct page *subdir, 424 int offset, int limit, struct page ***dir) 425{ 426 swp_entry_t *ptr; 427 int freed = 0; 428 429 ptr = shmem_swp_map(subdir); 430 for (; offset < limit; offset += LATENCY_LIMIT) { 431 int size = limit - offset; 432 if (size > LATENCY_LIMIT) 433 size = LATENCY_LIMIT; 434 freed += shmem_free_swp(ptr+offset, ptr+offset+size); 435 if (need_resched()) { 436 shmem_swp_unmap(ptr); 437 if (*dir) { 438 shmem_dir_unmap(*dir); 439 *dir = NULL; 440 } 441 cond_resched(); 442 ptr = shmem_swp_map(subdir); 443 } 444 } 445 shmem_swp_unmap(ptr); 446 return freed; 447} 448 449static void shmem_free_pages(struct list_head *next) 450{ 451 struct page *page; 452 int freed = 0; 453 454 do { 455 page = container_of(next, struct page, lru); 456 next = next->next; 457 shmem_dir_free(page); 458 freed++; 459 if (freed >= LATENCY_LIMIT) { 460 cond_resched(); 461 freed = 0; 462 } 463 } while (next); 464} 465 466static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end) 467{ 468 struct shmem_inode_info *info = SHMEM_I(inode); 469 unsigned long idx; 470 unsigned long size; 471 unsigned long limit; 472 unsigned long stage; 473 unsigned long diroff; 474 struct page **dir; 475 struct page *topdir; 476 struct page *middir; 477 struct page *subdir; 478 swp_entry_t *ptr; 479 LIST_HEAD(pages_to_free); 480 long nr_pages_to_free = 0; 481 long nr_swaps_freed = 0; 482 int offset; 483 int freed; 484 int punch_hole; 485 unsigned long upper_limit; 486 487 inode->i_ctime = inode->i_mtime = CURRENT_TIME; 488 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 489 if (idx >= info->next_index) 490 return; 491 492 spin_lock(&info->lock); 493 info->flags |= SHMEM_TRUNCATE; 494 if (likely(end == (loff_t) -1)) { 495 limit = info->next_index; 496 upper_limit = SHMEM_MAX_INDEX; 497 info->next_index = idx; 498 punch_hole = 0; 499 } else { 500 if (end + 1 >= inode->i_size) { /* we may free a little more */ 501 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >> 502 PAGE_CACHE_SHIFT; 503 upper_limit = SHMEM_MAX_INDEX; 504 } else { 505 limit = (end + 1) >> PAGE_CACHE_SHIFT; 506 upper_limit = limit; 507 } 508 punch_hole = 1; 509 } 510 511 topdir = info->i_indirect; 512 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) { 513 info->i_indirect = NULL; 514 nr_pages_to_free++; 515 list_add(&topdir->lru, &pages_to_free); 516 } 517 spin_unlock(&info->lock); 518 519 if (info->swapped && idx < SHMEM_NR_DIRECT) { 520 ptr = info->i_direct; 521 size = limit; 522 if (size > SHMEM_NR_DIRECT) 523 size = SHMEM_NR_DIRECT; 524 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size); 525 } 526 527 /* 528 * If there are no indirect blocks or we are punching a hole 529 * below indirect blocks, nothing to be done. 530 */ 531 if (!topdir || limit <= SHMEM_NR_DIRECT) 532 goto done2; 533 534 upper_limit -= SHMEM_NR_DIRECT; 535 limit -= SHMEM_NR_DIRECT; 536 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0; 537 offset = idx % ENTRIES_PER_PAGE; 538 idx -= offset; 539 540 dir = shmem_dir_map(topdir); 541 stage = ENTRIES_PER_PAGEPAGE/2; 542 if (idx < ENTRIES_PER_PAGEPAGE/2) { 543 middir = topdir; 544 diroff = idx/ENTRIES_PER_PAGE; 545 } else { 546 dir += ENTRIES_PER_PAGE/2; 547 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE; 548 while (stage <= idx) 549 stage += ENTRIES_PER_PAGEPAGE; 550 middir = *dir; 551 if (*dir) { 552 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) % 553 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE; 554 if (!diroff && !offset && upper_limit >= stage) { 555 *dir = NULL; 556 nr_pages_to_free++; 557 list_add(&middir->lru, &pages_to_free); 558 } 559 shmem_dir_unmap(dir); 560 dir = shmem_dir_map(middir); 561 } else { 562 diroff = 0; 563 offset = 0; 564 idx = stage; 565 } 566 } 567 568 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) { 569 if (unlikely(idx == stage)) { 570 shmem_dir_unmap(dir); 571 dir = shmem_dir_map(topdir) + 572 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 573 while (!*dir) { 574 dir++; 575 idx += ENTRIES_PER_PAGEPAGE; 576 if (idx >= limit) 577 goto done1; 578 } 579 stage = idx + ENTRIES_PER_PAGEPAGE; 580 middir = *dir; 581 if (upper_limit >= stage) { 582 *dir = NULL; 583 nr_pages_to_free++; 584 list_add(&middir->lru, &pages_to_free); 585 } 586 shmem_dir_unmap(dir); 587 cond_resched(); 588 dir = shmem_dir_map(middir); 589 diroff = 0; 590 } 591 subdir = dir[diroff]; 592 if (subdir && page_private(subdir)) { 593 size = limit - idx; 594 if (size > ENTRIES_PER_PAGE) 595 size = ENTRIES_PER_PAGE; 596 freed = shmem_map_and_free_swp(subdir, 597 offset, size, &dir); 598 if (!dir) 599 dir = shmem_dir_map(middir); 600 nr_swaps_freed += freed; 601 if (offset) 602 spin_lock(&info->lock); 603 set_page_private(subdir, page_private(subdir) - freed); 604 if (offset) 605 spin_unlock(&info->lock); 606 if (!punch_hole) 607 BUG_ON(page_private(subdir) > offset); 608 } 609 if (offset) 610 offset = 0; 611 else if (subdir && upper_limit - idx >= ENTRIES_PER_PAGE) { 612 dir[diroff] = NULL; 613 nr_pages_to_free++; 614 list_add(&subdir->lru, &pages_to_free); 615 } 616 } 617done1: 618 shmem_dir_unmap(dir); 619done2: 620 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) { 621 /* 622 * Call truncate_inode_pages again: racing shmem_unuse_inode 623 * may have swizzled a page in from swap since vmtruncate or 624 * generic_delete_inode did it, before we lowered next_index. 625 * Also, though shmem_getpage checks i_size before adding to 626 * cache, no recheck after: so fix the narrow window there too. 627 */ 628 truncate_inode_pages_range(inode->i_mapping, start, end); 629 } 630 631 spin_lock(&info->lock); 632 info->flags &= ~SHMEM_TRUNCATE; 633 info->swapped -= nr_swaps_freed; 634 if (nr_pages_to_free) 635 shmem_free_blocks(inode, nr_pages_to_free); 636 shmem_recalc_inode(inode); 637 spin_unlock(&info->lock); 638 639 /* 640 * Empty swap vector directory pages to be freed? 641 */ 642 if (!list_empty(&pages_to_free)) { 643 pages_to_free.prev->next = NULL; 644 shmem_free_pages(pages_to_free.next); 645 } 646} 647 648static void shmem_truncate(struct inode *inode) 649{ 650 shmem_truncate_range(inode, inode->i_size, (loff_t)-1); 651} 652 653static int shmem_notify_change(struct dentry *dentry, struct iattr *attr) 654{ 655 struct inode *inode = dentry->d_inode; 656 struct page *page = NULL; 657 int error; 658 659 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { 660 if (attr->ia_size < inode->i_size) { 661 /* 662 * If truncating down to a partial page, then 663 * if that page is already allocated, hold it 664 * in memory until the truncation is over, so 665 * truncate_partial_page cannnot miss it were 666 * it assigned to swap. 667 */ 668 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) { 669 (void) shmem_getpage(inode, 670 attr->ia_size>>PAGE_CACHE_SHIFT, 671 &page, SGP_READ, NULL); 672 } 673 /* 674 * Reset SHMEM_PAGEIN flag so that shmem_truncate can 675 * detect if any pages might have been added to cache 676 * after truncate_inode_pages. But we needn't bother 677 * if it's being fully truncated to zero-length: the 678 * nrpages check is efficient enough in that case. 679 */ 680 if (attr->ia_size) { 681 struct shmem_inode_info *info = SHMEM_I(inode); 682 spin_lock(&info->lock); 683 info->flags &= ~SHMEM_PAGEIN; 684 spin_unlock(&info->lock); 685 } 686 } 687 } 688 689 error = inode_change_ok(inode, attr); 690 if (!error) 691 error = inode_setattr(inode, attr); 692#ifdef CONFIG_TMPFS_POSIX_ACL 693 if (!error && (attr->ia_valid & ATTR_MODE)) 694 error = generic_acl_chmod(inode, &shmem_acl_ops); 695#endif 696 if (page) 697 page_cache_release(page); 698 return error; 699} 700 701static void shmem_delete_inode(struct inode *inode) 702{ 703 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 704 struct shmem_inode_info *info = SHMEM_I(inode); 705 706 if (inode->i_op->truncate == shmem_truncate) { 707 truncate_inode_pages(inode->i_mapping, 0); 708 shmem_unacct_size(info->flags, inode->i_size); 709 inode->i_size = 0; 710 shmem_truncate(inode); 711 if (!list_empty(&info->swaplist)) { 712 spin_lock(&shmem_swaplist_lock); 713 list_del_init(&info->swaplist); 714 spin_unlock(&shmem_swaplist_lock); 715 } 716 } 717 BUG_ON(inode->i_blocks); 718 if (sbinfo->max_inodes) { 719 spin_lock(&sbinfo->stat_lock); 720 sbinfo->free_inodes++; 721 spin_unlock(&sbinfo->stat_lock); 722 } 723 clear_inode(inode); 724} 725 726static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir) 727{ 728 swp_entry_t *ptr; 729 730 for (ptr = dir; ptr < edir; ptr++) { 731 if (ptr->val == entry.val) 732 return ptr - dir; 733 } 734 return -1; 735} 736 737static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page) 738{ 739 struct inode *inode; 740 unsigned long idx; 741 unsigned long size; 742 unsigned long limit; 743 unsigned long stage; 744 struct page **dir; 745 struct page *subdir; 746 swp_entry_t *ptr; 747 int offset; 748 749 idx = 0; 750 ptr = info->i_direct; 751 spin_lock(&info->lock); 752 limit = info->next_index; 753 size = limit; 754 if (size > SHMEM_NR_DIRECT) 755 size = SHMEM_NR_DIRECT; 756 offset = shmem_find_swp(entry, ptr, ptr+size); 757 if (offset >= 0) { 758 shmem_swp_balance_unmap(); 759 goto found; 760 } 761 if (!info->i_indirect) 762 goto lost2; 763 764 dir = shmem_dir_map(info->i_indirect); 765 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2; 766 767 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) { 768 if (unlikely(idx == stage)) { 769 shmem_dir_unmap(dir-1); 770 dir = shmem_dir_map(info->i_indirect) + 771 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 772 while (!*dir) { 773 dir++; 774 idx += ENTRIES_PER_PAGEPAGE; 775 if (idx >= limit) 776 goto lost1; 777 } 778 stage = idx + ENTRIES_PER_PAGEPAGE; 779 subdir = *dir; 780 shmem_dir_unmap(dir); 781 dir = shmem_dir_map(subdir); 782 } 783 subdir = *dir; 784 if (subdir && page_private(subdir)) { 785 ptr = shmem_swp_map(subdir); 786 size = limit - idx; 787 if (size > ENTRIES_PER_PAGE) 788 size = ENTRIES_PER_PAGE; 789 offset = shmem_find_swp(entry, ptr, ptr+size); 790 if (offset >= 0) { 791 shmem_dir_unmap(dir); 792 goto found; 793 } 794 shmem_swp_unmap(ptr); 795 } 796 } 797lost1: 798 shmem_dir_unmap(dir-1); 799lost2: 800 spin_unlock(&info->lock); 801 return 0; 802found: 803 idx += offset; 804 inode = &info->vfs_inode; 805 if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) { 806 info->flags |= SHMEM_PAGEIN; 807 shmem_swp_set(info, ptr + offset, 0); 808 } 809 shmem_swp_unmap(ptr); 810 spin_unlock(&info->lock); 811 /* 812 * Decrement swap count even when the entry is left behind: 813 * try_to_unuse will skip over mms, then reincrement count. 814 */ 815 swap_free(entry); 816 return 1; 817} 818 819/* 820 * shmem_unuse() search for an eventually swapped out shmem page. 821 */ 822int shmem_unuse(swp_entry_t entry, struct page *page) 823{ 824 struct list_head *p, *next; 825 struct shmem_inode_info *info; 826 int found = 0; 827 828 spin_lock(&shmem_swaplist_lock); 829 list_for_each_safe(p, next, &shmem_swaplist) { 830 info = list_entry(p, struct shmem_inode_info, swaplist); 831 if (!info->swapped) 832 list_del_init(&info->swaplist); 833 else if (shmem_unuse_inode(info, entry, page)) { 834 /* move head to start search for next from here */ 835 list_move_tail(&shmem_swaplist, &info->swaplist); 836 found = 1; 837 break; 838 } 839 } 840 spin_unlock(&shmem_swaplist_lock); 841 return found; 842} 843 844/* 845 * Move the page from the page cache to the swap cache. 846 */ 847static int shmem_writepage(struct page *page, struct writeback_control *wbc) 848{ 849 struct shmem_inode_info *info; 850 swp_entry_t *entry, swap; 851 struct address_space *mapping; 852 unsigned long index; 853 struct inode *inode; 854 855 BUG_ON(!PageLocked(page)); 856 BUG_ON(page_mapped(page)); 857 858 mapping = page->mapping; 859 index = page->index; 860 inode = mapping->host; 861 info = SHMEM_I(inode); 862 if (info->flags & VM_LOCKED) 863 goto redirty; 864 swap = get_swap_page(); 865 if (!swap.val) 866 goto redirty; 867 868 spin_lock(&info->lock); 869 shmem_recalc_inode(inode); 870 if (index >= info->next_index) { 871 BUG_ON(!(info->flags & SHMEM_TRUNCATE)); 872 goto unlock; 873 } 874 entry = shmem_swp_entry(info, index, NULL); 875 BUG_ON(!entry); 876 BUG_ON(entry->val); 877 878 if (move_to_swap_cache(page, swap) == 0) { 879 shmem_swp_set(info, entry, swap.val); 880 shmem_swp_unmap(entry); 881 spin_unlock(&info->lock); 882 if (list_empty(&info->swaplist)) { 883 spin_lock(&shmem_swaplist_lock); 884 /* move instead of add in case we're racing */ 885 list_move_tail(&info->swaplist, &shmem_swaplist); 886 spin_unlock(&shmem_swaplist_lock); 887 } 888 unlock_page(page); 889 return 0; 890 } 891 892 shmem_swp_unmap(entry); 893unlock: 894 spin_unlock(&info->lock); 895 swap_free(swap); 896redirty: 897 set_page_dirty(page); 898 return AOP_WRITEPAGE_ACTIVATE; /* Return with the page locked */ 899} 900 901#ifdef CONFIG_NUMA 902static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes) 903{ 904 char *nodelist = strchr(value, ':'); 905 int err = 1; 906 907 if (nodelist) { 908 /* NUL-terminate policy string */ 909 *nodelist++ = '\0'; 910 if (nodelist_parse(nodelist, *policy_nodes)) 911 goto out; 912 } 913 if (!strcmp(value, "default")) { 914 *policy = MPOL_DEFAULT; 915 /* Don't allow a nodelist */ 916 if (!nodelist) 917 err = 0; 918 } else if (!strcmp(value, "prefer")) { 919 *policy = MPOL_PREFERRED; 920 /* Insist on a nodelist of one node only */ 921 if (nodelist) { 922 char *rest = nodelist; 923 while (isdigit(*rest)) 924 rest++; 925 if (!*rest) 926 err = 0; 927 } 928 } else if (!strcmp(value, "bind")) { 929 *policy = MPOL_BIND; 930 /* Insist on a nodelist */ 931 if (nodelist) 932 err = 0; 933 } else if (!strcmp(value, "interleave")) { 934 *policy = MPOL_INTERLEAVE; 935 /* Default to nodes online if no nodelist */ 936 if (!nodelist) 937 *policy_nodes = node_online_map; 938 err = 0; 939 } 940out: 941 /* Restore string for error message */ 942 if (nodelist) 943 *--nodelist = ':'; 944 return err; 945} 946 947static struct page *shmem_swapin_async(struct shared_policy *p, 948 swp_entry_t entry, unsigned long idx) 949{ 950 struct page *page; 951 struct vm_area_struct pvma; 952 953 /* Create a pseudo vma that just contains the policy */ 954 memset(&pvma, 0, sizeof(struct vm_area_struct)); 955 pvma.vm_end = PAGE_SIZE; 956 pvma.vm_pgoff = idx; 957 pvma.vm_policy = mpol_shared_policy_lookup(p, idx); 958 page = read_swap_cache_async(entry, &pvma, 0); 959 mpol_free(pvma.vm_policy); 960 return page; 961} 962 963struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry, 964 unsigned long idx) 965{ 966 struct shared_policy *p = &info->policy; 967 int i, num; 968 struct page *page; 969 unsigned long offset; 970 971 num = valid_swaphandles(entry, &offset); 972 for (i = 0; i < num; offset++, i++) { 973 page = shmem_swapin_async(p, 974 swp_entry(swp_type(entry), offset), idx); 975 if (!page) 976 break; 977 page_cache_release(page); 978 } 979 lru_add_drain(); /* Push any new pages onto the LRU now */ 980 return shmem_swapin_async(p, entry, idx); 981} 982 983static struct page * 984shmem_alloc_page(gfp_t gfp, struct shmem_inode_info *info, 985 unsigned long idx) 986{ 987 struct vm_area_struct pvma; 988 struct page *page; 989 990 memset(&pvma, 0, sizeof(struct vm_area_struct)); 991 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx); 992 pvma.vm_pgoff = idx; 993 pvma.vm_end = PAGE_SIZE; 994 page = alloc_page_vma(gfp | __GFP_ZERO, &pvma, 0); 995 mpol_free(pvma.vm_policy); 996 return page; 997} 998#else 999static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes) 1000{ 1001 return 1; 1002} 1003 1004static inline struct page * 1005shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx) 1006{ 1007 swapin_readahead(entry, 0, NULL); 1008 return read_swap_cache_async(entry, NULL, 0); 1009} 1010 1011static inline struct page * 1012shmem_alloc_page(gfp_t gfp,struct shmem_inode_info *info, unsigned long idx) 1013{ 1014 return alloc_page(gfp | __GFP_ZERO); 1015} 1016#endif 1017 1018/* 1019 * shmem_getpage - either get the page from swap or allocate a new one 1020 * 1021 * If we allocate a new one we do not mark it dirty. That's up to the 1022 * vm. If we swap it in we mark it dirty since we also free the swap 1023 * entry since a page cannot live in both the swap and page cache 1024 */ 1025static int shmem_getpage(struct inode *inode, unsigned long idx, 1026 struct page **pagep, enum sgp_type sgp, int *type) 1027{ 1028 struct address_space *mapping = inode->i_mapping; 1029 struct shmem_inode_info *info = SHMEM_I(inode); 1030 struct shmem_sb_info *sbinfo; 1031 struct page *filepage = *pagep; 1032 struct page *swappage; 1033 swp_entry_t *entry; 1034 swp_entry_t swap; 1035 int error; 1036 1037 if (idx >= SHMEM_MAX_INDEX) 1038 return -EFBIG; 1039 /* 1040 * Normally, filepage is NULL on entry, and either found 1041 * uptodate immediately, or allocated and zeroed, or read 1042 * in under swappage, which is then assigned to filepage. 1043 * But shmem_prepare_write passes in a locked filepage, 1044 * which may be found not uptodate by other callers too, 1045 * and may need to be copied from the swappage read in. 1046 */ 1047repeat: 1048 if (!filepage) 1049 filepage = find_lock_page(mapping, idx); 1050 if (filepage && PageUptodate(filepage)) 1051 goto done; 1052 error = 0; 1053 if (sgp == SGP_QUICK) 1054 goto failed; 1055 1056 spin_lock(&info->lock); 1057 shmem_recalc_inode(inode); 1058 entry = shmem_swp_alloc(info, idx, sgp); 1059 if (IS_ERR(entry)) { 1060 spin_unlock(&info->lock); 1061 error = PTR_ERR(entry); 1062 goto failed; 1063 } 1064 swap = *entry; 1065 1066 if (swap.val) { 1067 /* Look it up and read it in.. */ 1068 swappage = lookup_swap_cache(swap); 1069 if (!swappage) { 1070 shmem_swp_unmap(entry); 1071 /* here we actually do the io */ 1072 if (type && *type == VM_FAULT_MINOR) { 1073 __count_vm_event(PGMAJFAULT); 1074 *type = VM_FAULT_MAJOR; 1075 } 1076 spin_unlock(&info->lock); 1077 swappage = shmem_swapin(info, swap, idx); 1078 if (!swappage) { 1079 spin_lock(&info->lock); 1080 entry = shmem_swp_alloc(info, idx, sgp); 1081 if (IS_ERR(entry)) 1082 error = PTR_ERR(entry); 1083 else { 1084 if (entry->val == swap.val) 1085 error = -ENOMEM; 1086 shmem_swp_unmap(entry); 1087 } 1088 spin_unlock(&info->lock); 1089 if (error) 1090 goto failed; 1091 goto repeat; 1092 } 1093 wait_on_page_locked(swappage); 1094 page_cache_release(swappage); 1095 goto repeat; 1096 } 1097 1098 /* We have to do this with page locked to prevent races */ 1099 if (TestSetPageLocked(swappage)) { 1100 shmem_swp_unmap(entry); 1101 spin_unlock(&info->lock); 1102 wait_on_page_locked(swappage); 1103 page_cache_release(swappage); 1104 goto repeat; 1105 } 1106 if (PageWriteback(swappage)) { 1107 shmem_swp_unmap(entry); 1108 spin_unlock(&info->lock); 1109 wait_on_page_writeback(swappage); 1110 unlock_page(swappage); 1111 page_cache_release(swappage); 1112 goto repeat; 1113 } 1114 if (!PageUptodate(swappage)) { 1115 shmem_swp_unmap(entry); 1116 spin_unlock(&info->lock); 1117 unlock_page(swappage); 1118 page_cache_release(swappage); 1119 error = -EIO; 1120 goto failed; 1121 } 1122 1123 if (filepage) { 1124 shmem_swp_set(info, entry, 0); 1125 shmem_swp_unmap(entry); 1126 delete_from_swap_cache(swappage); 1127 spin_unlock(&info->lock); 1128 copy_highpage(filepage, swappage); 1129 unlock_page(swappage); 1130 page_cache_release(swappage); 1131 flush_dcache_page(filepage); 1132 SetPageUptodate(filepage); 1133 set_page_dirty(filepage); 1134 swap_free(swap); 1135 } else if (!(error = move_from_swap_cache( 1136 swappage, idx, mapping))) { 1137 info->flags |= SHMEM_PAGEIN; 1138 shmem_swp_set(info, entry, 0); 1139 shmem_swp_unmap(entry); 1140 spin_unlock(&info->lock); 1141 filepage = swappage; 1142 swap_free(swap); 1143 } else { 1144 shmem_swp_unmap(entry); 1145 spin_unlock(&info->lock); 1146 unlock_page(swappage); 1147 page_cache_release(swappage); 1148 if (error == -ENOMEM) { 1149 /* let kswapd refresh zone for GFP_ATOMICs */ 1150 congestion_wait(WRITE, HZ/50); 1151 } 1152 goto repeat; 1153 } 1154 } else if (sgp == SGP_READ && !filepage) { 1155 shmem_swp_unmap(entry); 1156 filepage = find_get_page(mapping, idx); 1157 if (filepage && 1158 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) { 1159 spin_unlock(&info->lock); 1160 wait_on_page_locked(filepage); 1161 page_cache_release(filepage); 1162 filepage = NULL; 1163 goto repeat; 1164 } 1165 spin_unlock(&info->lock); 1166 } else { 1167 shmem_swp_unmap(entry); 1168 sbinfo = SHMEM_SB(inode->i_sb); 1169 if (sbinfo->max_blocks) { 1170 spin_lock(&sbinfo->stat_lock); 1171 if (sbinfo->free_blocks == 0 || 1172 shmem_acct_block(info->flags)) { 1173 spin_unlock(&sbinfo->stat_lock); 1174 spin_unlock(&info->lock); 1175 error = -ENOSPC; 1176 goto failed; 1177 } 1178 sbinfo->free_blocks--; 1179 inode->i_blocks += BLOCKS_PER_PAGE; 1180 spin_unlock(&sbinfo->stat_lock); 1181 } else if (shmem_acct_block(info->flags)) { 1182 spin_unlock(&info->lock); 1183 error = -ENOSPC; 1184 goto failed; 1185 } 1186 1187 if (!filepage) { 1188 spin_unlock(&info->lock); 1189 filepage = shmem_alloc_page(mapping_gfp_mask(mapping), 1190 info, 1191 idx); 1192 if (!filepage) { 1193 shmem_unacct_blocks(info->flags, 1); 1194 shmem_free_blocks(inode, 1); 1195 error = -ENOMEM; 1196 goto failed; 1197 } 1198 1199 spin_lock(&info->lock); 1200 entry = shmem_swp_alloc(info, idx, sgp); 1201 if (IS_ERR(entry)) 1202 error = PTR_ERR(entry); 1203 else { 1204 swap = *entry; 1205 shmem_swp_unmap(entry); 1206 } 1207 if (error || swap.val || 0 != add_to_page_cache_lru( 1208 filepage, mapping, idx, GFP_ATOMIC)) { 1209 spin_unlock(&info->lock); 1210 page_cache_release(filepage); 1211 shmem_unacct_blocks(info->flags, 1); 1212 shmem_free_blocks(inode, 1); 1213 filepage = NULL; 1214 if (error) 1215 goto failed; 1216 goto repeat; 1217 } 1218 info->flags |= SHMEM_PAGEIN; 1219 } 1220 1221 info->alloced++; 1222 spin_unlock(&info->lock); 1223 flush_dcache_page(filepage); 1224 SetPageUptodate(filepage); 1225 } 1226done: 1227 if (*pagep != filepage) { 1228 unlock_page(filepage); 1229 *pagep = filepage; 1230 } 1231 return 0; 1232 1233failed: 1234 if (*pagep != filepage) { 1235 unlock_page(filepage); 1236 page_cache_release(filepage); 1237 } 1238 return error; 1239} 1240 1241static struct page *shmem_nopage(struct vm_area_struct *vma, 1242 unsigned long address, int *type) 1243{ 1244 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 1245 struct page *page = NULL; 1246 unsigned long idx; 1247 int error; 1248 1249 idx = (address - vma->vm_start) >> PAGE_SHIFT; 1250 idx += vma->vm_pgoff; 1251 idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT; 1252 if (((loff_t) idx << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 1253 return NOPAGE_SIGBUS; 1254 1255 error = shmem_getpage(inode, idx, &page, SGP_CACHE, type); 1256 if (error) 1257 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS; 1258 1259 mark_page_accessed(page); 1260 return page; 1261} 1262 1263static int shmem_populate(struct vm_area_struct *vma, 1264 unsigned long addr, unsigned long len, 1265 pgprot_t prot, unsigned long pgoff, int nonblock) 1266{ 1267 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 1268 struct mm_struct *mm = vma->vm_mm; 1269 enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE; 1270 unsigned long size; 1271 1272 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT; 1273 if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size) 1274 return -EINVAL; 1275 1276 while ((long) len > 0) { 1277 struct page *page = NULL; 1278 int err; 1279 /* 1280 * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE 1281 */ 1282 err = shmem_getpage(inode, pgoff, &page, sgp, NULL); 1283 if (err) 1284 return err; 1285 /* Page may still be null, but only if nonblock was set. */ 1286 if (page) { 1287 mark_page_accessed(page); 1288 err = install_page(mm, vma, addr, page, prot); 1289 if (err) { 1290 page_cache_release(page); 1291 return err; 1292 } 1293 } else if (vma->vm_flags & VM_NONLINEAR) { 1294 /* No page was found just because we can't read it in 1295 * now (being here implies nonblock != 0), but the page 1296 * may exist, so set the PTE to fault it in later. */ 1297 err = install_file_pte(mm, vma, addr, pgoff, prot); 1298 if (err) 1299 return err; 1300 } 1301 1302 len -= PAGE_SIZE; 1303 addr += PAGE_SIZE; 1304 pgoff++; 1305 } 1306 return 0; 1307} 1308 1309#ifdef CONFIG_NUMA 1310int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new) 1311{ 1312 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1313 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new); 1314} 1315 1316struct mempolicy * 1317shmem_get_policy(struct vm_area_struct *vma, unsigned long addr) 1318{ 1319 struct inode *i = vma->vm_file->f_path.dentry->d_inode; 1320 unsigned long idx; 1321 1322 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; 1323 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx); 1324} 1325#endif 1326 1327int shmem_lock(struct file *file, int lock, struct user_struct *user) 1328{ 1329 struct inode *inode = file->f_path.dentry->d_inode; 1330 struct shmem_inode_info *info = SHMEM_I(inode); 1331 int retval = -ENOMEM; 1332 1333 spin_lock(&info->lock); 1334 if (lock && !(info->flags & VM_LOCKED)) { 1335 if (!user_shm_lock(inode->i_size, user)) 1336 goto out_nomem; 1337 info->flags |= VM_LOCKED; 1338 } 1339 if (!lock && (info->flags & VM_LOCKED) && user) { 1340 user_shm_unlock(inode->i_size, user); 1341 info->flags &= ~VM_LOCKED; 1342 } 1343 retval = 0; 1344out_nomem: 1345 spin_unlock(&info->lock); 1346 return retval; 1347} 1348 1349static int shmem_mmap(struct file *file, struct vm_area_struct *vma) 1350{ 1351 file_accessed(file); 1352 vma->vm_ops = &shmem_vm_ops; 1353 return 0; 1354} 1355 1356static struct inode * 1357shmem_get_inode(struct super_block *sb, int mode, dev_t dev) 1358{ 1359 struct inode *inode; 1360 struct shmem_inode_info *info; 1361 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 1362 1363 if (sbinfo->max_inodes) { 1364 spin_lock(&sbinfo->stat_lock); 1365 if (!sbinfo->free_inodes) { 1366 spin_unlock(&sbinfo->stat_lock); 1367 return NULL; 1368 } 1369 sbinfo->free_inodes--; 1370 spin_unlock(&sbinfo->stat_lock); 1371 } 1372 1373 inode = new_inode(sb); 1374 if (inode) { 1375 inode->i_mode = mode; 1376 inode->i_uid = current->fsuid; 1377 inode->i_gid = current->fsgid; 1378 inode->i_blocks = 0; 1379 inode->i_mapping->a_ops = &shmem_aops; 1380 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; 1381 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 1382 inode->i_generation = get_seconds(); 1383 info = SHMEM_I(inode); 1384 memset(info, 0, (char *)inode - (char *)info); 1385 spin_lock_init(&info->lock); 1386 INIT_LIST_HEAD(&info->swaplist); 1387 1388 switch (mode & S_IFMT) { 1389 default: 1390 inode->i_op = &shmem_special_inode_operations; 1391 init_special_inode(inode, mode, dev); 1392 break; 1393 case S_IFREG: 1394 inode->i_op = &shmem_inode_operations; 1395 inode->i_fop = &shmem_file_operations; 1396 mpol_shared_policy_init(&info->policy, sbinfo->policy, 1397 &sbinfo->policy_nodes); 1398 break; 1399 case S_IFDIR: 1400 inc_nlink(inode); 1401 /* Some things misbehave if size == 0 on a directory */ 1402 inode->i_size = 2 * BOGO_DIRENT_SIZE; 1403 inode->i_op = &shmem_dir_inode_operations; 1404 inode->i_fop = &simple_dir_operations; 1405 break; 1406 case S_IFLNK: 1407 /* 1408 * Must not load anything in the rbtree, 1409 * mpol_free_shared_policy will not be called. 1410 */ 1411 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, 1412 NULL); 1413 break; 1414 } 1415 } else if (sbinfo->max_inodes) { 1416 spin_lock(&sbinfo->stat_lock); 1417 sbinfo->free_inodes++; 1418 spin_unlock(&sbinfo->stat_lock); 1419 } 1420 return inode; 1421} 1422 1423#ifdef CONFIG_TMPFS 1424static const struct inode_operations shmem_symlink_inode_operations; 1425static const struct inode_operations shmem_symlink_inline_operations; 1426 1427/* 1428 * Normally tmpfs makes no use of shmem_prepare_write, but it 1429 * lets a tmpfs file be used read-write below the loop driver. 1430 */ 1431static int 1432shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to) 1433{ 1434 struct inode *inode = page->mapping->host; 1435 return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL); 1436} 1437 1438static ssize_t 1439shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos) 1440{ 1441 struct inode *inode = file->f_path.dentry->d_inode; 1442 loff_t pos; 1443 unsigned long written; 1444 ssize_t err; 1445 1446 if ((ssize_t) count < 0) 1447 return -EINVAL; 1448 1449 if (!access_ok(VERIFY_READ, buf, count)) 1450 return -EFAULT; 1451 1452 mutex_lock(&inode->i_mutex); 1453 1454 pos = *ppos; 1455 written = 0; 1456 1457 err = generic_write_checks(file, &pos, &count, 0); 1458 if (err || !count) 1459 goto out; 1460 1461 err = remove_suid(file->f_path.dentry); 1462 if (err) 1463 goto out; 1464 1465 inode->i_ctime = inode->i_mtime = CURRENT_TIME; 1466 1467 do { 1468 struct page *page = NULL; 1469 unsigned long bytes, index, offset; 1470 char *kaddr; 1471 int left; 1472 1473 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */ 1474 index = pos >> PAGE_CACHE_SHIFT; 1475 bytes = PAGE_CACHE_SIZE - offset; 1476 if (bytes > count) 1477 bytes = count; 1478 1479 /* 1480 * We don't hold page lock across copy from user - 1481 * what would it guard against? - so no deadlock here. 1482 * But it still may be a good idea to prefault below. 1483 */ 1484 1485 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL); 1486 if (err) 1487 break; 1488 1489 left = bytes; 1490 if (PageHighMem(page)) { 1491 volatile unsigned char dummy; 1492 __get_user(dummy, buf); 1493 __get_user(dummy, buf + bytes - 1); 1494 1495 kaddr = kmap_atomic(page, KM_USER0); 1496 left = __copy_from_user_inatomic(kaddr + offset, 1497 buf, bytes); 1498 kunmap_atomic(kaddr, KM_USER0); 1499 } 1500 if (left) { 1501 kaddr = kmap(page); 1502 left = __copy_from_user(kaddr + offset, buf, bytes); 1503 kunmap(page); 1504 } 1505 1506 written += bytes; 1507 count -= bytes; 1508 pos += bytes; 1509 buf += bytes; 1510 if (pos > inode->i_size) 1511 i_size_write(inode, pos); 1512 1513 flush_dcache_page(page); 1514 set_page_dirty(page); 1515 mark_page_accessed(page); 1516 page_cache_release(page); 1517 1518 if (left) { 1519 pos -= left; 1520 written -= left; 1521 err = -EFAULT; 1522 break; 1523 } 1524 1525 /* 1526 * Our dirty pages are not counted in nr_dirty, 1527 * and we do not attempt to balance dirty pages. 1528 */ 1529 1530 cond_resched(); 1531 } while (count); 1532 1533 *ppos = pos; 1534 if (written) 1535 err = written; 1536out: 1537 mutex_unlock(&inode->i_mutex); 1538 return err; 1539} 1540 1541static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor) 1542{ 1543 struct inode *inode = filp->f_path.dentry->d_inode; 1544 struct address_space *mapping = inode->i_mapping; 1545 unsigned long index, offset; 1546 1547 index = *ppos >> PAGE_CACHE_SHIFT; 1548 offset = *ppos & ~PAGE_CACHE_MASK; 1549 1550 for (;;) { 1551 struct page *page = NULL; 1552 unsigned long end_index, nr, ret; 1553 loff_t i_size = i_size_read(inode); 1554 1555 end_index = i_size >> PAGE_CACHE_SHIFT; 1556 if (index > end_index) 1557 break; 1558 if (index == end_index) { 1559 nr = i_size & ~PAGE_CACHE_MASK; 1560 if (nr <= offset) 1561 break; 1562 } 1563 1564 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL); 1565 if (desc->error) { 1566 if (desc->error == -EINVAL) 1567 desc->error = 0; 1568 break; 1569 } 1570 1571 /* 1572 * We must evaluate after, since reads (unlike writes) 1573 * are called without i_mutex protection against truncate 1574 */ 1575 nr = PAGE_CACHE_SIZE; 1576 i_size = i_size_read(inode); 1577 end_index = i_size >> PAGE_CACHE_SHIFT; 1578 if (index == end_index) { 1579 nr = i_size & ~PAGE_CACHE_MASK; 1580 if (nr <= offset) { 1581 if (page) 1582 page_cache_release(page); 1583 break; 1584 } 1585 } 1586 nr -= offset; 1587 1588 if (page) { 1589 /* 1590 * If users can be writing to this page using arbitrary 1591 * virtual addresses, take care about potential aliasing 1592 * before reading the page on the kernel side. 1593 */ 1594 if (mapping_writably_mapped(mapping)) 1595 flush_dcache_page(page); 1596 /* 1597 * Mark the page accessed if we read the beginning. 1598 */ 1599 if (!offset) 1600 mark_page_accessed(page); 1601 } else { 1602 page = ZERO_PAGE(0); 1603 page_cache_get(page); 1604 } 1605 1606 /* 1607 * Ok, we have the page, and it's up-to-date, so 1608 * now we can copy it to user space... 1609 * 1610 * The actor routine returns how many bytes were actually used.. 1611 * NOTE! This may not be the same as how much of a user buffer 1612 * we filled up (we may be padding etc), so we can only update 1613 * "pos" here (the actor routine has to update the user buffer 1614 * pointers and the remaining count). 1615 */ 1616 ret = actor(desc, page, offset, nr); 1617 offset += ret; 1618 index += offset >> PAGE_CACHE_SHIFT; 1619 offset &= ~PAGE_CACHE_MASK; 1620 1621 page_cache_release(page); 1622 if (ret != nr || !desc->count) 1623 break; 1624 1625 cond_resched(); 1626 } 1627 1628 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; 1629 file_accessed(filp); 1630} 1631 1632static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos) 1633{ 1634 read_descriptor_t desc; 1635 1636 if ((ssize_t) count < 0) 1637 return -EINVAL; 1638 if (!access_ok(VERIFY_WRITE, buf, count)) 1639 return -EFAULT; 1640 if (!count) 1641 return 0; 1642 1643 desc.written = 0; 1644 desc.count = count; 1645 desc.arg.buf = buf; 1646 desc.error = 0; 1647 1648 do_shmem_file_read(filp, ppos, &desc, file_read_actor); 1649 if (desc.written) 1650 return desc.written; 1651 return desc.error; 1652} 1653 1654static ssize_t shmem_file_sendfile(struct file *in_file, loff_t *ppos, 1655 size_t count, read_actor_t actor, void *target) 1656{ 1657 read_descriptor_t desc; 1658 1659 if (!count) 1660 return 0; 1661 1662 desc.written = 0; 1663 desc.count = count; 1664 desc.arg.data = target; 1665 desc.error = 0; 1666 1667 do_shmem_file_read(in_file, ppos, &desc, actor); 1668 if (desc.written) 1669 return desc.written; 1670 return desc.error; 1671} 1672 1673static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) 1674{ 1675 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); 1676 1677 buf->f_type = TMPFS_MAGIC; 1678 buf->f_bsize = PAGE_CACHE_SIZE; 1679 buf->f_namelen = NAME_MAX; 1680 spin_lock(&sbinfo->stat_lock); 1681 if (sbinfo->max_blocks) { 1682 buf->f_blocks = sbinfo->max_blocks; 1683 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks; 1684 } 1685 if (sbinfo->max_inodes) { 1686 buf->f_files = sbinfo->max_inodes; 1687 buf->f_ffree = sbinfo->free_inodes; 1688 } 1689 /* else leave those fields 0 like simple_statfs */ 1690 spin_unlock(&sbinfo->stat_lock); 1691 return 0; 1692} 1693 1694/* 1695 * File creation. Allocate an inode, and we're done.. 1696 */ 1697static int 1698shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 1699{ 1700 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev); 1701 int error = -ENOSPC; 1702 1703 if (inode) { 1704 error = security_inode_init_security(inode, dir, NULL, NULL, 1705 NULL); 1706 if (error) { 1707 if (error != -EOPNOTSUPP) { 1708 iput(inode); 1709 return error; 1710 } 1711 } 1712 error = shmem_acl_init(inode, dir); 1713 if (error) { 1714 iput(inode); 1715 return error; 1716 } 1717 if (dir->i_mode & S_ISGID) { 1718 inode->i_gid = dir->i_gid; 1719 if (S_ISDIR(mode)) 1720 inode->i_mode |= S_ISGID; 1721 } 1722 dir->i_size += BOGO_DIRENT_SIZE; 1723 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1724 d_instantiate(dentry, inode); 1725 dget(dentry); /* Extra count - pin the dentry in core */ 1726 } 1727 return error; 1728} 1729 1730static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1731{ 1732 int error; 1733 1734 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) 1735 return error; 1736 inc_nlink(dir); 1737 return 0; 1738} 1739 1740static int shmem_create(struct inode *dir, struct dentry *dentry, int mode, 1741 struct nameidata *nd) 1742{ 1743 return shmem_mknod(dir, dentry, mode | S_IFREG, 0); 1744} 1745 1746/* 1747 * Link a file.. 1748 */ 1749static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1750{ 1751 struct inode *inode = old_dentry->d_inode; 1752 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 1753 1754 /* 1755 * No ordinary (disk based) filesystem counts links as inodes; 1756 * but each new link needs a new dentry, pinning lowmem, and 1757 * tmpfs dentries cannot be pruned until they are unlinked. 1758 */ 1759 if (sbinfo->max_inodes) { 1760 spin_lock(&sbinfo->stat_lock); 1761 if (!sbinfo->free_inodes) { 1762 spin_unlock(&sbinfo->stat_lock); 1763 return -ENOSPC; 1764 } 1765 sbinfo->free_inodes--; 1766 spin_unlock(&sbinfo->stat_lock); 1767 } 1768 1769 dir->i_size += BOGO_DIRENT_SIZE; 1770 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1771 inc_nlink(inode); 1772 atomic_inc(&inode->i_count); /* New dentry reference */ 1773 dget(dentry); /* Extra pinning count for the created dentry */ 1774 d_instantiate(dentry, inode); 1775 return 0; 1776} 1777 1778static int shmem_unlink(struct inode *dir, struct dentry *dentry) 1779{ 1780 struct inode *inode = dentry->d_inode; 1781 1782 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) { 1783 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 1784 if (sbinfo->max_inodes) { 1785 spin_lock(&sbinfo->stat_lock); 1786 sbinfo->free_inodes++; 1787 spin_unlock(&sbinfo->stat_lock); 1788 } 1789 } 1790 1791 dir->i_size -= BOGO_DIRENT_SIZE; 1792 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1793 drop_nlink(inode); 1794 dput(dentry); /* Undo the count from "create" - this does all the work */ 1795 return 0; 1796} 1797 1798static int shmem_rmdir(struct inode *dir, struct dentry *dentry) 1799{ 1800 if (!simple_empty(dentry)) 1801 return -ENOTEMPTY; 1802 1803 drop_nlink(dentry->d_inode); 1804 drop_nlink(dir); 1805 return shmem_unlink(dir, dentry); 1806} 1807 1808/* 1809 * The VFS layer already does all the dentry stuff for rename, 1810 * we just have to decrement the usage count for the target if 1811 * it exists so that the VFS layer correctly free's it when it 1812 * gets overwritten. 1813 */ 1814static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) 1815{ 1816 struct inode *inode = old_dentry->d_inode; 1817 int they_are_dirs = S_ISDIR(inode->i_mode); 1818 1819 if (!simple_empty(new_dentry)) 1820 return -ENOTEMPTY; 1821 1822 if (new_dentry->d_inode) { 1823 (void) shmem_unlink(new_dir, new_dentry); 1824 if (they_are_dirs) 1825 drop_nlink(old_dir); 1826 } else if (they_are_dirs) { 1827 drop_nlink(old_dir); 1828 inc_nlink(new_dir); 1829 } 1830 1831 old_dir->i_size -= BOGO_DIRENT_SIZE; 1832 new_dir->i_size += BOGO_DIRENT_SIZE; 1833 old_dir->i_ctime = old_dir->i_mtime = 1834 new_dir->i_ctime = new_dir->i_mtime = 1835 inode->i_ctime = CURRENT_TIME; 1836 return 0; 1837} 1838 1839static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1840{ 1841 int error; 1842 int len; 1843 struct inode *inode; 1844 struct page *page = NULL; 1845 char *kaddr; 1846 struct shmem_inode_info *info; 1847 1848 len = strlen(symname) + 1; 1849 if (len > PAGE_CACHE_SIZE) 1850 return -ENAMETOOLONG; 1851 1852 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0); 1853 if (!inode) 1854 return -ENOSPC; 1855 1856 error = security_inode_init_security(inode, dir, NULL, NULL, 1857 NULL); 1858 if (error) { 1859 if (error != -EOPNOTSUPP) { 1860 iput(inode); 1861 return error; 1862 } 1863 error = 0; 1864 } 1865 1866 info = SHMEM_I(inode); 1867 inode->i_size = len-1; 1868 if (len <= (char *)inode - (char *)info) { 1869 /* do it inline */ 1870 memcpy(info, symname, len); 1871 inode->i_op = &shmem_symlink_inline_operations; 1872 } else { 1873 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); 1874 if (error) { 1875 iput(inode); 1876 return error; 1877 } 1878 inode->i_op = &shmem_symlink_inode_operations; 1879 kaddr = kmap_atomic(page, KM_USER0); 1880 memcpy(kaddr, symname, len); 1881 kunmap_atomic(kaddr, KM_USER0); 1882 set_page_dirty(page); 1883 page_cache_release(page); 1884 } 1885 if (dir->i_mode & S_ISGID) 1886 inode->i_gid = dir->i_gid; 1887 dir->i_size += BOGO_DIRENT_SIZE; 1888 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1889 d_instantiate(dentry, inode); 1890 dget(dentry); 1891 return 0; 1892} 1893 1894static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd) 1895{ 1896 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode)); 1897 return NULL; 1898} 1899 1900static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) 1901{ 1902 struct page *page = NULL; 1903 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); 1904 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page)); 1905 return page; 1906} 1907 1908static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 1909{ 1910 if (!IS_ERR(nd_get_link(nd))) { 1911 struct page *page = cookie; 1912 kunmap(page); 1913 mark_page_accessed(page); 1914 page_cache_release(page); 1915 } 1916} 1917 1918static const struct inode_operations shmem_symlink_inline_operations = { 1919 .readlink = generic_readlink, 1920 .follow_link = shmem_follow_link_inline, 1921}; 1922 1923static const struct inode_operations shmem_symlink_inode_operations = { 1924 .truncate = shmem_truncate, 1925 .readlink = generic_readlink, 1926 .follow_link = shmem_follow_link, 1927 .put_link = shmem_put_link, 1928}; 1929 1930#ifdef CONFIG_TMPFS_POSIX_ACL 1931/** 1932 * Superblocks without xattr inode operations will get security.* xattr 1933 * support from the VFS "for free". As soon as we have any other xattrs 1934 * like ACLs, we also need to implement the security.* handlers at 1935 * filesystem level, though. 1936 */ 1937 1938static size_t shmem_xattr_security_list(struct inode *inode, char *list, 1939 size_t list_len, const char *name, 1940 size_t name_len) 1941{ 1942 return security_inode_listsecurity(inode, list, list_len); 1943} 1944 1945static int shmem_xattr_security_get(struct inode *inode, const char *name, 1946 void *buffer, size_t size) 1947{ 1948 if (strcmp(name, "") == 0) 1949 return -EINVAL; 1950 return security_inode_getsecurity(inode, name, buffer, size, 1951 -EOPNOTSUPP); 1952} 1953 1954static int shmem_xattr_security_set(struct inode *inode, const char *name, 1955 const void *value, size_t size, int flags) 1956{ 1957 if (strcmp(name, "") == 0) 1958 return -EINVAL; 1959 return security_inode_setsecurity(inode, name, value, size, flags); 1960} 1961 1962static struct xattr_handler shmem_xattr_security_handler = { 1963 .prefix = XATTR_SECURITY_PREFIX, 1964 .list = shmem_xattr_security_list, 1965 .get = shmem_xattr_security_get, 1966 .set = shmem_xattr_security_set, 1967}; 1968 1969static struct xattr_handler *shmem_xattr_handlers[] = { 1970 &shmem_xattr_acl_access_handler, 1971 &shmem_xattr_acl_default_handler, 1972 &shmem_xattr_security_handler, 1973 NULL 1974}; 1975#endif 1976 1977static struct dentry *shmem_get_parent(struct dentry *child) 1978{ 1979 return ERR_PTR(-ESTALE); 1980} 1981 1982static int shmem_match(struct inode *ino, void *vfh) 1983{ 1984 __u32 *fh = vfh; 1985 __u64 inum = fh[2]; 1986 inum = (inum << 32) | fh[1]; 1987 return ino->i_ino == inum && fh[0] == ino->i_generation; 1988} 1989 1990static struct dentry *shmem_get_dentry(struct super_block *sb, void *vfh) 1991{ 1992 struct dentry *de = NULL; 1993 struct inode *inode; 1994 __u32 *fh = vfh; 1995 __u64 inum = fh[2]; 1996 inum = (inum << 32) | fh[1]; 1997 1998 inode = ilookup5(sb, (unsigned long)(inum+fh[0]), shmem_match, vfh); 1999 if (inode) { 2000 de = d_find_alias(inode); 2001 iput(inode); 2002 } 2003 2004 return de? de: ERR_PTR(-ESTALE); 2005} 2006 2007static struct dentry *shmem_decode_fh(struct super_block *sb, __u32 *fh, 2008 int len, int type, 2009 int (*acceptable)(void *context, struct dentry *de), 2010 void *context) 2011{ 2012 if (len < 3) 2013 return ERR_PTR(-ESTALE); 2014 2015 return sb->s_export_op->find_exported_dentry(sb, fh, NULL, acceptable, 2016 context); 2017} 2018 2019static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len, 2020 int connectable) 2021{ 2022 struct inode *inode = dentry->d_inode; 2023 2024 if (*len < 3) 2025 return 255; 2026 2027 if (hlist_unhashed(&inode->i_hash)) { 2028 /* Unfortunately insert_inode_hash is not idempotent, 2029 * so as we hash inodes here rather than at creation 2030 * time, we need a lock to ensure we only try 2031 * to do it once 2032 */ 2033 static DEFINE_SPINLOCK(lock); 2034 spin_lock(&lock); 2035 if (hlist_unhashed(&inode->i_hash)) 2036 __insert_inode_hash(inode, 2037 inode->i_ino + inode->i_generation); 2038 spin_unlock(&lock); 2039 } 2040 2041 fh[0] = inode->i_generation; 2042 fh[1] = inode->i_ino; 2043 fh[2] = ((__u64)inode->i_ino) >> 32; 2044 2045 *len = 3; 2046 return 1; 2047} 2048 2049static struct export_operations shmem_export_ops = { 2050 .get_parent = shmem_get_parent, 2051 .get_dentry = shmem_get_dentry, 2052 .encode_fh = shmem_encode_fh, 2053 .decode_fh = shmem_decode_fh, 2054}; 2055 2056static int shmem_parse_options(char *options, int *mode, uid_t *uid, 2057 gid_t *gid, unsigned long *blocks, unsigned long *inodes, 2058 int *policy, nodemask_t *policy_nodes) 2059{ 2060 char *this_char, *value, *rest; 2061 2062 while (options != NULL) { 2063 this_char = options; 2064 for (;;) { 2065 /* 2066 * NUL-terminate this option: unfortunately, 2067 * mount options form a comma-separated list, 2068 * but mpol's nodelist may also contain commas. 2069 */ 2070 options = strchr(options, ','); 2071 if (options == NULL) 2072 break; 2073 options++; 2074 if (!isdigit(*options)) { 2075 options[-1] = '\0'; 2076 break; 2077 } 2078 } 2079 if (!*this_char) 2080 continue; 2081 if ((value = strchr(this_char,'=')) != NULL) { 2082 *value++ = 0; 2083 } else { 2084 printk(KERN_ERR 2085 "tmpfs: No value for mount option '%s'\n", 2086 this_char); 2087 return 1; 2088 } 2089 2090 if (!strcmp(this_char,"size")) { 2091 unsigned long long size; 2092 size = memparse(value,&rest); 2093 if (*rest == '%') { 2094 size <<= PAGE_SHIFT; 2095 size *= totalram_pages; 2096 do_div(size, 100); 2097 rest++; 2098 } 2099 if (*rest) 2100 goto bad_val; 2101 *blocks = size >> PAGE_CACHE_SHIFT; 2102 } else if (!strcmp(this_char,"nr_blocks")) { 2103 *blocks = memparse(value,&rest); 2104 if (*rest) 2105 goto bad_val; 2106 } else if (!strcmp(this_char,"nr_inodes")) { 2107 *inodes = memparse(value,&rest); 2108 if (*rest) 2109 goto bad_val; 2110 } else if (!strcmp(this_char,"mode")) { 2111 if (!mode) 2112 continue; 2113 *mode = simple_strtoul(value,&rest,8); 2114 if (*rest) 2115 goto bad_val; 2116 } else if (!strcmp(this_char,"uid")) { 2117 if (!uid) 2118 continue; 2119 *uid = simple_strtoul(value,&rest,0); 2120 if (*rest) 2121 goto bad_val; 2122 } else if (!strcmp(this_char,"gid")) { 2123 if (!gid) 2124 continue; 2125 *gid = simple_strtoul(value,&rest,0); 2126 if (*rest) 2127 goto bad_val; 2128 } else if (!strcmp(this_char,"mpol")) { 2129 if (shmem_parse_mpol(value,policy,policy_nodes)) 2130 goto bad_val; 2131 } else { 2132 printk(KERN_ERR "tmpfs: Bad mount option %s\n", 2133 this_char); 2134 return 1; 2135 } 2136 } 2137 return 0; 2138 2139bad_val: 2140 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", 2141 value, this_char); 2142 return 1; 2143 2144} 2145 2146static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) 2147{ 2148 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 2149 unsigned long max_blocks = sbinfo->max_blocks; 2150 unsigned long max_inodes = sbinfo->max_inodes; 2151 int policy = sbinfo->policy; 2152 nodemask_t policy_nodes = sbinfo->policy_nodes; 2153 unsigned long blocks; 2154 unsigned long inodes; 2155 int error = -EINVAL; 2156 2157 if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks, 2158 &max_inodes, &policy, &policy_nodes)) 2159 return error; 2160 2161 spin_lock(&sbinfo->stat_lock); 2162 blocks = sbinfo->max_blocks - sbinfo->free_blocks; 2163 inodes = sbinfo->max_inodes - sbinfo->free_inodes; 2164 if (max_blocks < blocks) 2165 goto out; 2166 if (max_inodes < inodes) 2167 goto out; 2168 /* 2169 * Those tests also disallow limited->unlimited while any are in 2170 * use, so i_blocks will always be zero when max_blocks is zero; 2171 * but we must separately disallow unlimited->limited, because 2172 * in that case we have no record of how much is already in use. 2173 */ 2174 if (max_blocks && !sbinfo->max_blocks) 2175 goto out; 2176 if (max_inodes && !sbinfo->max_inodes) 2177 goto out; 2178 2179 error = 0; 2180 sbinfo->max_blocks = max_blocks; 2181 sbinfo->free_blocks = max_blocks - blocks; 2182 sbinfo->max_inodes = max_inodes; 2183 sbinfo->free_inodes = max_inodes - inodes; 2184 sbinfo->policy = policy; 2185 sbinfo->policy_nodes = policy_nodes; 2186out: 2187 spin_unlock(&sbinfo->stat_lock); 2188 return error; 2189} 2190#endif 2191 2192static void shmem_put_super(struct super_block *sb) 2193{ 2194 kfree(sb->s_fs_info); 2195 sb->s_fs_info = NULL; 2196} 2197 2198static int shmem_fill_super(struct super_block *sb, 2199 void *data, int silent) 2200{ 2201 struct inode *inode; 2202 struct dentry *root; 2203 int mode = S_IRWXUGO | S_ISVTX; 2204 uid_t uid = current->fsuid; 2205 gid_t gid = current->fsgid; 2206 int err = -ENOMEM; 2207 struct shmem_sb_info *sbinfo; 2208 unsigned long blocks = 0; 2209 unsigned long inodes = 0; 2210 int policy = MPOL_DEFAULT; 2211 nodemask_t policy_nodes = node_online_map; 2212 2213#ifdef CONFIG_TMPFS 2214 /* 2215 * Per default we only allow half of the physical ram per 2216 * tmpfs instance, limiting inodes to one per page of lowmem; 2217 * but the internal instance is left unlimited. 2218 */ 2219 if (!(sb->s_flags & MS_NOUSER)) { 2220 blocks = totalram_pages / 2; 2221 inodes = totalram_pages - totalhigh_pages; 2222 if (inodes > blocks) 2223 inodes = blocks; 2224 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks, 2225 &inodes, &policy, &policy_nodes)) 2226 return -EINVAL; 2227 } 2228 sb->s_export_op = &shmem_export_ops; 2229#else 2230 sb->s_flags |= MS_NOUSER; 2231#endif 2232 2233 /* Round up to L1_CACHE_BYTES to resist false sharing */ 2234 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info), 2235 L1_CACHE_BYTES), GFP_KERNEL); 2236 if (!sbinfo) 2237 return -ENOMEM; 2238 2239 spin_lock_init(&sbinfo->stat_lock); 2240 sbinfo->max_blocks = blocks; 2241 sbinfo->free_blocks = blocks; 2242 sbinfo->max_inodes = inodes; 2243 sbinfo->free_inodes = inodes; 2244 sbinfo->policy = policy; 2245 sbinfo->policy_nodes = policy_nodes; 2246 2247 sb->s_fs_info = sbinfo; 2248 sb->s_maxbytes = SHMEM_MAX_BYTES; 2249 sb->s_blocksize = PAGE_CACHE_SIZE; 2250 sb->s_blocksize_bits = PAGE_CACHE_SHIFT; 2251 sb->s_magic = TMPFS_MAGIC; 2252 sb->s_op = &shmem_ops; 2253 sb->s_time_gran = 1; 2254#ifdef CONFIG_TMPFS_POSIX_ACL 2255 sb->s_xattr = shmem_xattr_handlers; 2256 sb->s_flags |= MS_POSIXACL; 2257#endif 2258 2259 inode = shmem_get_inode(sb, S_IFDIR | mode, 0); 2260 if (!inode) 2261 goto failed; 2262 inode->i_uid = uid; 2263 inode->i_gid = gid; 2264 root = d_alloc_root(inode); 2265 if (!root) 2266 goto failed_iput; 2267 sb->s_root = root; 2268 return 0; 2269 2270failed_iput: 2271 iput(inode); 2272failed: 2273 shmem_put_super(sb); 2274 return err; 2275} 2276 2277static struct kmem_cache *shmem_inode_cachep; 2278 2279static struct inode *shmem_alloc_inode(struct super_block *sb) 2280{ 2281 struct shmem_inode_info *p; 2282 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); 2283 if (!p) 2284 return NULL; 2285 return &p->vfs_inode; 2286} 2287 2288static void shmem_destroy_inode(struct inode *inode) 2289{ 2290 if ((inode->i_mode & S_IFMT) == S_IFREG) { 2291 /* only struct inode is valid if it's an inline symlink */ 2292 mpol_free_shared_policy(&SHMEM_I(inode)->policy); 2293 } 2294 shmem_acl_destroy_inode(inode); 2295 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); 2296} 2297 2298static void init_once(void *foo, struct kmem_cache *cachep, 2299 unsigned long flags) 2300{ 2301 struct shmem_inode_info *p = (struct shmem_inode_info *) foo; 2302 2303 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) == 2304 SLAB_CTOR_CONSTRUCTOR) { 2305 inode_init_once(&p->vfs_inode); 2306#ifdef CONFIG_TMPFS_POSIX_ACL 2307 p->i_acl = NULL; 2308 p->i_default_acl = NULL; 2309#endif 2310 } 2311} 2312 2313static int init_inodecache(void) 2314{ 2315 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", 2316 sizeof(struct shmem_inode_info), 2317 0, 0, init_once, NULL); 2318 if (shmem_inode_cachep == NULL) 2319 return -ENOMEM; 2320 return 0; 2321} 2322 2323static void destroy_inodecache(void) 2324{ 2325 kmem_cache_destroy(shmem_inode_cachep); 2326} 2327 2328static const struct address_space_operations shmem_aops = { 2329 .writepage = shmem_writepage, 2330 .set_page_dirty = __set_page_dirty_no_writeback, 2331#ifdef CONFIG_TMPFS 2332 .prepare_write = shmem_prepare_write, 2333 .commit_write = simple_commit_write, 2334#endif 2335 .migratepage = migrate_page, 2336}; 2337 2338static const struct file_operations shmem_file_operations = { 2339 .mmap = shmem_mmap, 2340#ifdef CONFIG_TMPFS 2341 .llseek = generic_file_llseek, 2342 .read = shmem_file_read, 2343 .write = shmem_file_write, 2344 .fsync = simple_sync_file, 2345 .sendfile = shmem_file_sendfile, 2346#endif 2347}; 2348 2349static const struct inode_operations shmem_inode_operations = { 2350 .truncate = shmem_truncate, 2351 .setattr = shmem_notify_change, 2352 .truncate_range = shmem_truncate_range, 2353#ifdef CONFIG_TMPFS_POSIX_ACL 2354 .setxattr = generic_setxattr, 2355 .getxattr = generic_getxattr, 2356 .listxattr = generic_listxattr, 2357 .removexattr = generic_removexattr, 2358 .permission = shmem_permission, 2359#endif 2360 2361}; 2362 2363static const struct inode_operations shmem_dir_inode_operations = { 2364#ifdef CONFIG_TMPFS 2365 .create = shmem_create, 2366 .lookup = simple_lookup, 2367 .link = shmem_link, 2368 .unlink = shmem_unlink, 2369 .symlink = shmem_symlink, 2370 .mkdir = shmem_mkdir, 2371 .rmdir = shmem_rmdir, 2372 .mknod = shmem_mknod, 2373 .rename = shmem_rename, 2374#endif 2375#ifdef CONFIG_TMPFS_POSIX_ACL 2376 .setattr = shmem_notify_change, 2377 .setxattr = generic_setxattr, 2378 .getxattr = generic_getxattr, 2379 .listxattr = generic_listxattr, 2380 .removexattr = generic_removexattr, 2381 .permission = shmem_permission, 2382#endif 2383}; 2384 2385static const struct inode_operations shmem_special_inode_operations = { 2386#ifdef CONFIG_TMPFS_POSIX_ACL 2387 .setattr = shmem_notify_change, 2388 .setxattr = generic_setxattr, 2389 .getxattr = generic_getxattr, 2390 .listxattr = generic_listxattr, 2391 .removexattr = generic_removexattr, 2392 .permission = shmem_permission, 2393#endif 2394}; 2395 2396static const struct super_operations shmem_ops = { 2397 .alloc_inode = shmem_alloc_inode, 2398 .destroy_inode = shmem_destroy_inode, 2399#ifdef CONFIG_TMPFS 2400 .statfs = shmem_statfs, 2401 .remount_fs = shmem_remount_fs, 2402#endif 2403 .delete_inode = shmem_delete_inode, 2404 .drop_inode = generic_delete_inode, 2405 .put_super = shmem_put_super, 2406}; 2407 2408static struct vm_operations_struct shmem_vm_ops = { 2409 .nopage = shmem_nopage, 2410 .populate = shmem_populate, 2411#ifdef CONFIG_NUMA 2412 .set_policy = shmem_set_policy, 2413 .get_policy = shmem_get_policy, 2414#endif 2415}; 2416 2417 2418static int shmem_get_sb(struct file_system_type *fs_type, 2419 int flags, const char *dev_name, void *data, struct vfsmount *mnt) 2420{ 2421 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt); 2422} 2423 2424static struct file_system_type tmpfs_fs_type = { 2425 .owner = THIS_MODULE, 2426 .name = "tmpfs", 2427 .get_sb = shmem_get_sb, 2428 .kill_sb = kill_litter_super, 2429}; 2430static struct vfsmount *shm_mnt; 2431 2432static int __init init_tmpfs(void) 2433{ 2434 int error; 2435 2436 error = init_inodecache(); 2437 if (error) 2438 goto out3; 2439 2440 error = register_filesystem(&tmpfs_fs_type); 2441 if (error) { 2442 printk(KERN_ERR "Could not register tmpfs\n"); 2443 goto out2; 2444 } 2445 2446 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER, 2447 tmpfs_fs_type.name, NULL); 2448 if (IS_ERR(shm_mnt)) { 2449 error = PTR_ERR(shm_mnt); 2450 printk(KERN_ERR "Could not kern_mount tmpfs\n"); 2451 goto out1; 2452 } 2453 return 0; 2454 2455out1: 2456 unregister_filesystem(&tmpfs_fs_type); 2457out2: 2458 destroy_inodecache(); 2459out3: 2460 shm_mnt = ERR_PTR(error); 2461 return error; 2462} 2463module_init(init_tmpfs) 2464 2465/* 2466 * shmem_file_setup - get an unlinked file living in tmpfs 2467 * 2468 * @name: name for dentry (to be seen in /proc/<pid>/maps 2469 * @size: size to be set for the file 2470 * 2471 */ 2472struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags) 2473{ 2474 int error; 2475 struct file *file; 2476 struct inode *inode; 2477 struct dentry *dentry, *root; 2478 struct qstr this; 2479 2480 if (IS_ERR(shm_mnt)) 2481 return (void *)shm_mnt; 2482 2483 if (size < 0 || size > SHMEM_MAX_BYTES) 2484 return ERR_PTR(-EINVAL); 2485 2486 if (shmem_acct_size(flags, size)) 2487 return ERR_PTR(-ENOMEM); 2488 2489 error = -ENOMEM; 2490 this.name = name; 2491 this.len = strlen(name); 2492 this.hash = 0; /* will go */ 2493 root = shm_mnt->mnt_root; 2494 dentry = d_alloc(root, &this); 2495 if (!dentry) 2496 goto put_memory; 2497 2498 error = -ENFILE; 2499 file = get_empty_filp(); 2500 if (!file) 2501 goto put_dentry; 2502 2503 error = -ENOSPC; 2504 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0); 2505 if (!inode) 2506 goto close_file; 2507 2508 SHMEM_I(inode)->flags = flags & VM_ACCOUNT; 2509 d_instantiate(dentry, inode); 2510 inode->i_size = size; 2511 inode->i_nlink = 0; /* It is unlinked */ 2512 file->f_path.mnt = mntget(shm_mnt); 2513 file->f_path.dentry = dentry; 2514 file->f_mapping = inode->i_mapping; 2515 file->f_op = &shmem_file_operations; 2516 file->f_mode = FMODE_WRITE | FMODE_READ; 2517 return file; 2518 2519close_file: 2520 put_filp(file); 2521put_dentry: 2522 dput(dentry); 2523put_memory: 2524 shmem_unacct_size(flags, size); 2525 return ERR_PTR(error); 2526} 2527 2528/* 2529 * shmem_zero_setup - setup a shared anonymous mapping 2530 * 2531 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff 2532 */ 2533int shmem_zero_setup(struct vm_area_struct *vma) 2534{ 2535 struct file *file; 2536 loff_t size = vma->vm_end - vma->vm_start; 2537 2538 file = shmem_file_setup("dev/zero", size, vma->vm_flags); 2539 if (IS_ERR(file)) 2540 return PTR_ERR(file); 2541 2542 if (vma->vm_file) 2543 fput(vma->vm_file); 2544 vma->vm_file = file; 2545 vma->vm_ops = &shmem_vm_ops; 2546 return 0; 2547} 2548