shmem.c revision f5e54d6e53a20cef45af7499e86164f0e0d16bb2
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/config.h> 27#include <linux/module.h> 28#include <linux/init.h> 29#include <linux/devfs_fs_kernel.h> 30#include <linux/fs.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 51#include <asm/uaccess.h> 52#include <asm/div64.h> 53#include <asm/pgtable.h> 54 55/* This magic number is used in glibc for posix shared memory */ 56#define TMPFS_MAGIC 0x01021994 57 58#define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long)) 59#define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE) 60#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512) 61 62#define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1)) 63#define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT) 64 65#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT) 66 67/* info->flags needs VM_flags to handle pagein/truncate races efficiently */ 68#define SHMEM_PAGEIN VM_READ 69#define SHMEM_TRUNCATE VM_WRITE 70 71/* Definition to limit shmem_truncate's steps between cond_rescheds */ 72#define LATENCY_LIMIT 64 73 74/* Pretend that each entry is of this size in directory's i_size */ 75#define BOGO_DIRENT_SIZE 20 76 77/* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */ 78enum sgp_type { 79 SGP_QUICK, /* don't try more than file page cache lookup */ 80 SGP_READ, /* don't exceed i_size, don't allocate page */ 81 SGP_CACHE, /* don't exceed i_size, may allocate page */ 82 SGP_WRITE, /* may exceed i_size, may allocate page */ 83}; 84 85static int shmem_getpage(struct inode *inode, unsigned long idx, 86 struct page **pagep, enum sgp_type sgp, int *type); 87 88static inline struct page *shmem_dir_alloc(gfp_t gfp_mask) 89{ 90 /* 91 * The above definition of ENTRIES_PER_PAGE, and the use of 92 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE: 93 * might be reconsidered if it ever diverges from PAGE_SIZE. 94 */ 95 return alloc_pages(gfp_mask, PAGE_CACHE_SHIFT-PAGE_SHIFT); 96} 97 98static inline void shmem_dir_free(struct page *page) 99{ 100 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT); 101} 102 103static struct page **shmem_dir_map(struct page *page) 104{ 105 return (struct page **)kmap_atomic(page, KM_USER0); 106} 107 108static inline void shmem_dir_unmap(struct page **dir) 109{ 110 kunmap_atomic(dir, KM_USER0); 111} 112 113static swp_entry_t *shmem_swp_map(struct page *page) 114{ 115 return (swp_entry_t *)kmap_atomic(page, KM_USER1); 116} 117 118static inline void shmem_swp_balance_unmap(void) 119{ 120 /* 121 * When passing a pointer to an i_direct entry, to code which 122 * also handles indirect entries and so will shmem_swp_unmap, 123 * we must arrange for the preempt count to remain in balance. 124 * What kmap_atomic of a lowmem page does depends on config 125 * and architecture, so pretend to kmap_atomic some lowmem page. 126 */ 127 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1); 128} 129 130static inline void shmem_swp_unmap(swp_entry_t *entry) 131{ 132 kunmap_atomic(entry, KM_USER1); 133} 134 135static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) 136{ 137 return sb->s_fs_info; 138} 139 140/* 141 * shmem_file_setup pre-accounts the whole fixed size of a VM object, 142 * for shared memory and for shared anonymous (/dev/zero) mappings 143 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), 144 * consistent with the pre-accounting of private mappings ... 145 */ 146static inline int shmem_acct_size(unsigned long flags, loff_t size) 147{ 148 return (flags & VM_ACCOUNT)? 149 security_vm_enough_memory(VM_ACCT(size)): 0; 150} 151 152static inline void shmem_unacct_size(unsigned long flags, loff_t size) 153{ 154 if (flags & VM_ACCOUNT) 155 vm_unacct_memory(VM_ACCT(size)); 156} 157 158/* 159 * ... whereas tmpfs objects are accounted incrementally as 160 * pages are allocated, in order to allow huge sparse files. 161 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, 162 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. 163 */ 164static inline int shmem_acct_block(unsigned long flags) 165{ 166 return (flags & VM_ACCOUNT)? 167 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE)); 168} 169 170static inline void shmem_unacct_blocks(unsigned long flags, long pages) 171{ 172 if (!(flags & VM_ACCOUNT)) 173 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE)); 174} 175 176static struct super_operations shmem_ops; 177static const struct address_space_operations shmem_aops; 178static struct file_operations shmem_file_operations; 179static struct inode_operations shmem_inode_operations; 180static struct inode_operations shmem_dir_inode_operations; 181static struct vm_operations_struct shmem_vm_ops; 182 183static struct backing_dev_info shmem_backing_dev_info __read_mostly = { 184 .ra_pages = 0, /* No readahead */ 185 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK, 186 .unplug_io_fn = default_unplug_io_fn, 187}; 188 189static LIST_HEAD(shmem_swaplist); 190static DEFINE_SPINLOCK(shmem_swaplist_lock); 191 192static void shmem_free_blocks(struct inode *inode, long pages) 193{ 194 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 195 if (sbinfo->max_blocks) { 196 spin_lock(&sbinfo->stat_lock); 197 sbinfo->free_blocks += pages; 198 inode->i_blocks -= pages*BLOCKS_PER_PAGE; 199 spin_unlock(&sbinfo->stat_lock); 200 } 201} 202 203/* 204 * shmem_recalc_inode - recalculate the size of an inode 205 * 206 * @inode: inode to recalc 207 * 208 * We have to calculate the free blocks since the mm can drop 209 * undirtied hole pages behind our back. 210 * 211 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped 212 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) 213 * 214 * It has to be called with the spinlock held. 215 */ 216static void shmem_recalc_inode(struct inode *inode) 217{ 218 struct shmem_inode_info *info = SHMEM_I(inode); 219 long freed; 220 221 freed = info->alloced - info->swapped - inode->i_mapping->nrpages; 222 if (freed > 0) { 223 info->alloced -= freed; 224 shmem_unacct_blocks(info->flags, freed); 225 shmem_free_blocks(inode, freed); 226 } 227} 228 229/* 230 * shmem_swp_entry - find the swap vector position in the info structure 231 * 232 * @info: info structure for the inode 233 * @index: index of the page to find 234 * @page: optional page to add to the structure. Has to be preset to 235 * all zeros 236 * 237 * If there is no space allocated yet it will return NULL when 238 * page is NULL, else it will use the page for the needed block, 239 * setting it to NULL on return to indicate that it has been used. 240 * 241 * The swap vector is organized the following way: 242 * 243 * There are SHMEM_NR_DIRECT entries directly stored in the 244 * shmem_inode_info structure. So small files do not need an addional 245 * allocation. 246 * 247 * For pages with index > SHMEM_NR_DIRECT there is the pointer 248 * i_indirect which points to a page which holds in the first half 249 * doubly indirect blocks, in the second half triple indirect blocks: 250 * 251 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the 252 * following layout (for SHMEM_NR_DIRECT == 16): 253 * 254 * i_indirect -> dir --> 16-19 255 * | +-> 20-23 256 * | 257 * +-->dir2 --> 24-27 258 * | +-> 28-31 259 * | +-> 32-35 260 * | +-> 36-39 261 * | 262 * +-->dir3 --> 40-43 263 * +-> 44-47 264 * +-> 48-51 265 * +-> 52-55 266 */ 267static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page) 268{ 269 unsigned long offset; 270 struct page **dir; 271 struct page *subdir; 272 273 if (index < SHMEM_NR_DIRECT) { 274 shmem_swp_balance_unmap(); 275 return info->i_direct+index; 276 } 277 if (!info->i_indirect) { 278 if (page) { 279 info->i_indirect = *page; 280 *page = NULL; 281 } 282 return NULL; /* need another page */ 283 } 284 285 index -= SHMEM_NR_DIRECT; 286 offset = index % ENTRIES_PER_PAGE; 287 index /= ENTRIES_PER_PAGE; 288 dir = shmem_dir_map(info->i_indirect); 289 290 if (index >= ENTRIES_PER_PAGE/2) { 291 index -= ENTRIES_PER_PAGE/2; 292 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE; 293 index %= ENTRIES_PER_PAGE; 294 subdir = *dir; 295 if (!subdir) { 296 if (page) { 297 *dir = *page; 298 *page = NULL; 299 } 300 shmem_dir_unmap(dir); 301 return NULL; /* need another page */ 302 } 303 shmem_dir_unmap(dir); 304 dir = shmem_dir_map(subdir); 305 } 306 307 dir += index; 308 subdir = *dir; 309 if (!subdir) { 310 if (!page || !(subdir = *page)) { 311 shmem_dir_unmap(dir); 312 return NULL; /* need a page */ 313 } 314 *dir = subdir; 315 *page = NULL; 316 } 317 shmem_dir_unmap(dir); 318 return shmem_swp_map(subdir) + offset; 319} 320 321static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value) 322{ 323 long incdec = value? 1: -1; 324 325 entry->val = value; 326 info->swapped += incdec; 327 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) { 328 struct page *page = kmap_atomic_to_page(entry); 329 set_page_private(page, page_private(page) + incdec); 330 } 331} 332 333/* 334 * shmem_swp_alloc - get the position of the swap entry for the page. 335 * If it does not exist allocate the entry. 336 * 337 * @info: info structure for the inode 338 * @index: index of the page to find 339 * @sgp: check and recheck i_size? skip allocation? 340 */ 341static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp) 342{ 343 struct inode *inode = &info->vfs_inode; 344 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 345 struct page *page = NULL; 346 swp_entry_t *entry; 347 348 if (sgp != SGP_WRITE && 349 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 350 return ERR_PTR(-EINVAL); 351 352 while (!(entry = shmem_swp_entry(info, index, &page))) { 353 if (sgp == SGP_READ) 354 return shmem_swp_map(ZERO_PAGE(0)); 355 /* 356 * Test free_blocks against 1 not 0, since we have 1 data 357 * page (and perhaps indirect index pages) yet to allocate: 358 * a waste to allocate index if we cannot allocate data. 359 */ 360 if (sbinfo->max_blocks) { 361 spin_lock(&sbinfo->stat_lock); 362 if (sbinfo->free_blocks <= 1) { 363 spin_unlock(&sbinfo->stat_lock); 364 return ERR_PTR(-ENOSPC); 365 } 366 sbinfo->free_blocks--; 367 inode->i_blocks += BLOCKS_PER_PAGE; 368 spin_unlock(&sbinfo->stat_lock); 369 } 370 371 spin_unlock(&info->lock); 372 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping) | __GFP_ZERO); 373 if (page) 374 set_page_private(page, 0); 375 spin_lock(&info->lock); 376 377 if (!page) { 378 shmem_free_blocks(inode, 1); 379 return ERR_PTR(-ENOMEM); 380 } 381 if (sgp != SGP_WRITE && 382 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) { 383 entry = ERR_PTR(-EINVAL); 384 break; 385 } 386 if (info->next_index <= index) 387 info->next_index = index + 1; 388 } 389 if (page) { 390 /* another task gave its page, or truncated the file */ 391 shmem_free_blocks(inode, 1); 392 shmem_dir_free(page); 393 } 394 if (info->next_index <= index && !IS_ERR(entry)) 395 info->next_index = index + 1; 396 return entry; 397} 398 399/* 400 * shmem_free_swp - free some swap entries in a directory 401 * 402 * @dir: pointer to the directory 403 * @edir: pointer after last entry of the directory 404 */ 405static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir) 406{ 407 swp_entry_t *ptr; 408 int freed = 0; 409 410 for (ptr = dir; ptr < edir; ptr++) { 411 if (ptr->val) { 412 free_swap_and_cache(*ptr); 413 *ptr = (swp_entry_t){0}; 414 freed++; 415 } 416 } 417 return freed; 418} 419 420static int shmem_map_and_free_swp(struct page *subdir, 421 int offset, int limit, struct page ***dir) 422{ 423 swp_entry_t *ptr; 424 int freed = 0; 425 426 ptr = shmem_swp_map(subdir); 427 for (; offset < limit; offset += LATENCY_LIMIT) { 428 int size = limit - offset; 429 if (size > LATENCY_LIMIT) 430 size = LATENCY_LIMIT; 431 freed += shmem_free_swp(ptr+offset, ptr+offset+size); 432 if (need_resched()) { 433 shmem_swp_unmap(ptr); 434 if (*dir) { 435 shmem_dir_unmap(*dir); 436 *dir = NULL; 437 } 438 cond_resched(); 439 ptr = shmem_swp_map(subdir); 440 } 441 } 442 shmem_swp_unmap(ptr); 443 return freed; 444} 445 446static void shmem_free_pages(struct list_head *next) 447{ 448 struct page *page; 449 int freed = 0; 450 451 do { 452 page = container_of(next, struct page, lru); 453 next = next->next; 454 shmem_dir_free(page); 455 freed++; 456 if (freed >= LATENCY_LIMIT) { 457 cond_resched(); 458 freed = 0; 459 } 460 } while (next); 461} 462 463static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end) 464{ 465 struct shmem_inode_info *info = SHMEM_I(inode); 466 unsigned long idx; 467 unsigned long size; 468 unsigned long limit; 469 unsigned long stage; 470 unsigned long diroff; 471 struct page **dir; 472 struct page *topdir; 473 struct page *middir; 474 struct page *subdir; 475 swp_entry_t *ptr; 476 LIST_HEAD(pages_to_free); 477 long nr_pages_to_free = 0; 478 long nr_swaps_freed = 0; 479 int offset; 480 int freed; 481 int punch_hole = 0; 482 483 inode->i_ctime = inode->i_mtime = CURRENT_TIME; 484 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 485 if (idx >= info->next_index) 486 return; 487 488 spin_lock(&info->lock); 489 info->flags |= SHMEM_TRUNCATE; 490 if (likely(end == (loff_t) -1)) { 491 limit = info->next_index; 492 info->next_index = idx; 493 } else { 494 limit = (end + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 495 if (limit > info->next_index) 496 limit = info->next_index; 497 punch_hole = 1; 498 } 499 500 topdir = info->i_indirect; 501 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) { 502 info->i_indirect = NULL; 503 nr_pages_to_free++; 504 list_add(&topdir->lru, &pages_to_free); 505 } 506 spin_unlock(&info->lock); 507 508 if (info->swapped && idx < SHMEM_NR_DIRECT) { 509 ptr = info->i_direct; 510 size = limit; 511 if (size > SHMEM_NR_DIRECT) 512 size = SHMEM_NR_DIRECT; 513 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size); 514 } 515 if (!topdir) 516 goto done2; 517 518 BUG_ON(limit <= SHMEM_NR_DIRECT); 519 limit -= SHMEM_NR_DIRECT; 520 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0; 521 offset = idx % ENTRIES_PER_PAGE; 522 idx -= offset; 523 524 dir = shmem_dir_map(topdir); 525 stage = ENTRIES_PER_PAGEPAGE/2; 526 if (idx < ENTRIES_PER_PAGEPAGE/2) { 527 middir = topdir; 528 diroff = idx/ENTRIES_PER_PAGE; 529 } else { 530 dir += ENTRIES_PER_PAGE/2; 531 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE; 532 while (stage <= idx) 533 stage += ENTRIES_PER_PAGEPAGE; 534 middir = *dir; 535 if (*dir) { 536 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) % 537 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE; 538 if (!diroff && !offset) { 539 *dir = NULL; 540 nr_pages_to_free++; 541 list_add(&middir->lru, &pages_to_free); 542 } 543 shmem_dir_unmap(dir); 544 dir = shmem_dir_map(middir); 545 } else { 546 diroff = 0; 547 offset = 0; 548 idx = stage; 549 } 550 } 551 552 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) { 553 if (unlikely(idx == stage)) { 554 shmem_dir_unmap(dir); 555 dir = shmem_dir_map(topdir) + 556 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 557 while (!*dir) { 558 dir++; 559 idx += ENTRIES_PER_PAGEPAGE; 560 if (idx >= limit) 561 goto done1; 562 } 563 stage = idx + ENTRIES_PER_PAGEPAGE; 564 middir = *dir; 565 *dir = NULL; 566 nr_pages_to_free++; 567 list_add(&middir->lru, &pages_to_free); 568 shmem_dir_unmap(dir); 569 cond_resched(); 570 dir = shmem_dir_map(middir); 571 diroff = 0; 572 } 573 subdir = dir[diroff]; 574 if (subdir && page_private(subdir)) { 575 size = limit - idx; 576 if (size > ENTRIES_PER_PAGE) 577 size = ENTRIES_PER_PAGE; 578 freed = shmem_map_and_free_swp(subdir, 579 offset, size, &dir); 580 if (!dir) 581 dir = shmem_dir_map(middir); 582 nr_swaps_freed += freed; 583 if (offset) 584 spin_lock(&info->lock); 585 set_page_private(subdir, page_private(subdir) - freed); 586 if (offset) 587 spin_unlock(&info->lock); 588 if (!punch_hole) 589 BUG_ON(page_private(subdir) > offset); 590 } 591 if (offset) 592 offset = 0; 593 else if (subdir && !page_private(subdir)) { 594 dir[diroff] = NULL; 595 nr_pages_to_free++; 596 list_add(&subdir->lru, &pages_to_free); 597 } 598 } 599done1: 600 shmem_dir_unmap(dir); 601done2: 602 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) { 603 /* 604 * Call truncate_inode_pages again: racing shmem_unuse_inode 605 * may have swizzled a page in from swap since vmtruncate or 606 * generic_delete_inode did it, before we lowered next_index. 607 * Also, though shmem_getpage checks i_size before adding to 608 * cache, no recheck after: so fix the narrow window there too. 609 */ 610 truncate_inode_pages_range(inode->i_mapping, start, end); 611 } 612 613 spin_lock(&info->lock); 614 info->flags &= ~SHMEM_TRUNCATE; 615 info->swapped -= nr_swaps_freed; 616 if (nr_pages_to_free) 617 shmem_free_blocks(inode, nr_pages_to_free); 618 shmem_recalc_inode(inode); 619 spin_unlock(&info->lock); 620 621 /* 622 * Empty swap vector directory pages to be freed? 623 */ 624 if (!list_empty(&pages_to_free)) { 625 pages_to_free.prev->next = NULL; 626 shmem_free_pages(pages_to_free.next); 627 } 628} 629 630static void shmem_truncate(struct inode *inode) 631{ 632 shmem_truncate_range(inode, inode->i_size, (loff_t)-1); 633} 634 635static int shmem_notify_change(struct dentry *dentry, struct iattr *attr) 636{ 637 struct inode *inode = dentry->d_inode; 638 struct page *page = NULL; 639 int error; 640 641 if (attr->ia_valid & ATTR_SIZE) { 642 if (attr->ia_size < inode->i_size) { 643 /* 644 * If truncating down to a partial page, then 645 * if that page is already allocated, hold it 646 * in memory until the truncation is over, so 647 * truncate_partial_page cannnot miss it were 648 * it assigned to swap. 649 */ 650 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) { 651 (void) shmem_getpage(inode, 652 attr->ia_size>>PAGE_CACHE_SHIFT, 653 &page, SGP_READ, NULL); 654 } 655 /* 656 * Reset SHMEM_PAGEIN flag so that shmem_truncate can 657 * detect if any pages might have been added to cache 658 * after truncate_inode_pages. But we needn't bother 659 * if it's being fully truncated to zero-length: the 660 * nrpages check is efficient enough in that case. 661 */ 662 if (attr->ia_size) { 663 struct shmem_inode_info *info = SHMEM_I(inode); 664 spin_lock(&info->lock); 665 info->flags &= ~SHMEM_PAGEIN; 666 spin_unlock(&info->lock); 667 } 668 } 669 } 670 671 error = inode_change_ok(inode, attr); 672 if (!error) 673 error = inode_setattr(inode, attr); 674 if (page) 675 page_cache_release(page); 676 return error; 677} 678 679static void shmem_delete_inode(struct inode *inode) 680{ 681 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 682 struct shmem_inode_info *info = SHMEM_I(inode); 683 684 if (inode->i_op->truncate == shmem_truncate) { 685 truncate_inode_pages(inode->i_mapping, 0); 686 shmem_unacct_size(info->flags, inode->i_size); 687 inode->i_size = 0; 688 shmem_truncate(inode); 689 if (!list_empty(&info->swaplist)) { 690 spin_lock(&shmem_swaplist_lock); 691 list_del_init(&info->swaplist); 692 spin_unlock(&shmem_swaplist_lock); 693 } 694 } 695 BUG_ON(inode->i_blocks); 696 if (sbinfo->max_inodes) { 697 spin_lock(&sbinfo->stat_lock); 698 sbinfo->free_inodes++; 699 spin_unlock(&sbinfo->stat_lock); 700 } 701 clear_inode(inode); 702} 703 704static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir) 705{ 706 swp_entry_t *ptr; 707 708 for (ptr = dir; ptr < edir; ptr++) { 709 if (ptr->val == entry.val) 710 return ptr - dir; 711 } 712 return -1; 713} 714 715static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page) 716{ 717 struct inode *inode; 718 unsigned long idx; 719 unsigned long size; 720 unsigned long limit; 721 unsigned long stage; 722 struct page **dir; 723 struct page *subdir; 724 swp_entry_t *ptr; 725 int offset; 726 727 idx = 0; 728 ptr = info->i_direct; 729 spin_lock(&info->lock); 730 limit = info->next_index; 731 size = limit; 732 if (size > SHMEM_NR_DIRECT) 733 size = SHMEM_NR_DIRECT; 734 offset = shmem_find_swp(entry, ptr, ptr+size); 735 if (offset >= 0) { 736 shmem_swp_balance_unmap(); 737 goto found; 738 } 739 if (!info->i_indirect) 740 goto lost2; 741 742 dir = shmem_dir_map(info->i_indirect); 743 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2; 744 745 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) { 746 if (unlikely(idx == stage)) { 747 shmem_dir_unmap(dir-1); 748 dir = shmem_dir_map(info->i_indirect) + 749 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE; 750 while (!*dir) { 751 dir++; 752 idx += ENTRIES_PER_PAGEPAGE; 753 if (idx >= limit) 754 goto lost1; 755 } 756 stage = idx + ENTRIES_PER_PAGEPAGE; 757 subdir = *dir; 758 shmem_dir_unmap(dir); 759 dir = shmem_dir_map(subdir); 760 } 761 subdir = *dir; 762 if (subdir && page_private(subdir)) { 763 ptr = shmem_swp_map(subdir); 764 size = limit - idx; 765 if (size > ENTRIES_PER_PAGE) 766 size = ENTRIES_PER_PAGE; 767 offset = shmem_find_swp(entry, ptr, ptr+size); 768 if (offset >= 0) { 769 shmem_dir_unmap(dir); 770 goto found; 771 } 772 shmem_swp_unmap(ptr); 773 } 774 } 775lost1: 776 shmem_dir_unmap(dir-1); 777lost2: 778 spin_unlock(&info->lock); 779 return 0; 780found: 781 idx += offset; 782 inode = &info->vfs_inode; 783 if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) { 784 info->flags |= SHMEM_PAGEIN; 785 shmem_swp_set(info, ptr + offset, 0); 786 } 787 shmem_swp_unmap(ptr); 788 spin_unlock(&info->lock); 789 /* 790 * Decrement swap count even when the entry is left behind: 791 * try_to_unuse will skip over mms, then reincrement count. 792 */ 793 swap_free(entry); 794 return 1; 795} 796 797/* 798 * shmem_unuse() search for an eventually swapped out shmem page. 799 */ 800int shmem_unuse(swp_entry_t entry, struct page *page) 801{ 802 struct list_head *p, *next; 803 struct shmem_inode_info *info; 804 int found = 0; 805 806 spin_lock(&shmem_swaplist_lock); 807 list_for_each_safe(p, next, &shmem_swaplist) { 808 info = list_entry(p, struct shmem_inode_info, swaplist); 809 if (!info->swapped) 810 list_del_init(&info->swaplist); 811 else if (shmem_unuse_inode(info, entry, page)) { 812 /* move head to start search for next from here */ 813 list_move_tail(&shmem_swaplist, &info->swaplist); 814 found = 1; 815 break; 816 } 817 } 818 spin_unlock(&shmem_swaplist_lock); 819 return found; 820} 821 822/* 823 * Move the page from the page cache to the swap cache. 824 */ 825static int shmem_writepage(struct page *page, struct writeback_control *wbc) 826{ 827 struct shmem_inode_info *info; 828 swp_entry_t *entry, swap; 829 struct address_space *mapping; 830 unsigned long index; 831 struct inode *inode; 832 833 BUG_ON(!PageLocked(page)); 834 BUG_ON(page_mapped(page)); 835 836 mapping = page->mapping; 837 index = page->index; 838 inode = mapping->host; 839 info = SHMEM_I(inode); 840 if (info->flags & VM_LOCKED) 841 goto redirty; 842 swap = get_swap_page(); 843 if (!swap.val) 844 goto redirty; 845 846 spin_lock(&info->lock); 847 shmem_recalc_inode(inode); 848 if (index >= info->next_index) { 849 BUG_ON(!(info->flags & SHMEM_TRUNCATE)); 850 goto unlock; 851 } 852 entry = shmem_swp_entry(info, index, NULL); 853 BUG_ON(!entry); 854 BUG_ON(entry->val); 855 856 if (move_to_swap_cache(page, swap) == 0) { 857 shmem_swp_set(info, entry, swap.val); 858 shmem_swp_unmap(entry); 859 spin_unlock(&info->lock); 860 if (list_empty(&info->swaplist)) { 861 spin_lock(&shmem_swaplist_lock); 862 /* move instead of add in case we're racing */ 863 list_move_tail(&info->swaplist, &shmem_swaplist); 864 spin_unlock(&shmem_swaplist_lock); 865 } 866 unlock_page(page); 867 return 0; 868 } 869 870 shmem_swp_unmap(entry); 871unlock: 872 spin_unlock(&info->lock); 873 swap_free(swap); 874redirty: 875 set_page_dirty(page); 876 return AOP_WRITEPAGE_ACTIVATE; /* Return with the page locked */ 877} 878 879#ifdef CONFIG_NUMA 880static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes) 881{ 882 char *nodelist = strchr(value, ':'); 883 int err = 1; 884 885 if (nodelist) { 886 /* NUL-terminate policy string */ 887 *nodelist++ = '\0'; 888 if (nodelist_parse(nodelist, *policy_nodes)) 889 goto out; 890 } 891 if (!strcmp(value, "default")) { 892 *policy = MPOL_DEFAULT; 893 /* Don't allow a nodelist */ 894 if (!nodelist) 895 err = 0; 896 } else if (!strcmp(value, "prefer")) { 897 *policy = MPOL_PREFERRED; 898 /* Insist on a nodelist of one node only */ 899 if (nodelist) { 900 char *rest = nodelist; 901 while (isdigit(*rest)) 902 rest++; 903 if (!*rest) 904 err = 0; 905 } 906 } else if (!strcmp(value, "bind")) { 907 *policy = MPOL_BIND; 908 /* Insist on a nodelist */ 909 if (nodelist) 910 err = 0; 911 } else if (!strcmp(value, "interleave")) { 912 *policy = MPOL_INTERLEAVE; 913 /* Default to nodes online if no nodelist */ 914 if (!nodelist) 915 *policy_nodes = node_online_map; 916 err = 0; 917 } 918out: 919 /* Restore string for error message */ 920 if (nodelist) 921 *--nodelist = ':'; 922 return err; 923} 924 925static struct page *shmem_swapin_async(struct shared_policy *p, 926 swp_entry_t entry, unsigned long idx) 927{ 928 struct page *page; 929 struct vm_area_struct pvma; 930 931 /* Create a pseudo vma that just contains the policy */ 932 memset(&pvma, 0, sizeof(struct vm_area_struct)); 933 pvma.vm_end = PAGE_SIZE; 934 pvma.vm_pgoff = idx; 935 pvma.vm_policy = mpol_shared_policy_lookup(p, idx); 936 page = read_swap_cache_async(entry, &pvma, 0); 937 mpol_free(pvma.vm_policy); 938 return page; 939} 940 941struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry, 942 unsigned long idx) 943{ 944 struct shared_policy *p = &info->policy; 945 int i, num; 946 struct page *page; 947 unsigned long offset; 948 949 num = valid_swaphandles(entry, &offset); 950 for (i = 0; i < num; offset++, i++) { 951 page = shmem_swapin_async(p, 952 swp_entry(swp_type(entry), offset), idx); 953 if (!page) 954 break; 955 page_cache_release(page); 956 } 957 lru_add_drain(); /* Push any new pages onto the LRU now */ 958 return shmem_swapin_async(p, entry, idx); 959} 960 961static struct page * 962shmem_alloc_page(gfp_t gfp, struct shmem_inode_info *info, 963 unsigned long idx) 964{ 965 struct vm_area_struct pvma; 966 struct page *page; 967 968 memset(&pvma, 0, sizeof(struct vm_area_struct)); 969 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx); 970 pvma.vm_pgoff = idx; 971 pvma.vm_end = PAGE_SIZE; 972 page = alloc_page_vma(gfp | __GFP_ZERO, &pvma, 0); 973 mpol_free(pvma.vm_policy); 974 return page; 975} 976#else 977static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes) 978{ 979 return 1; 980} 981 982static inline struct page * 983shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx) 984{ 985 swapin_readahead(entry, 0, NULL); 986 return read_swap_cache_async(entry, NULL, 0); 987} 988 989static inline struct page * 990shmem_alloc_page(gfp_t gfp,struct shmem_inode_info *info, unsigned long idx) 991{ 992 return alloc_page(gfp | __GFP_ZERO); 993} 994#endif 995 996/* 997 * shmem_getpage - either get the page from swap or allocate a new one 998 * 999 * If we allocate a new one we do not mark it dirty. That's up to the 1000 * vm. If we swap it in we mark it dirty since we also free the swap 1001 * entry since a page cannot live in both the swap and page cache 1002 */ 1003static int shmem_getpage(struct inode *inode, unsigned long idx, 1004 struct page **pagep, enum sgp_type sgp, int *type) 1005{ 1006 struct address_space *mapping = inode->i_mapping; 1007 struct shmem_inode_info *info = SHMEM_I(inode); 1008 struct shmem_sb_info *sbinfo; 1009 struct page *filepage = *pagep; 1010 struct page *swappage; 1011 swp_entry_t *entry; 1012 swp_entry_t swap; 1013 int error; 1014 1015 if (idx >= SHMEM_MAX_INDEX) 1016 return -EFBIG; 1017 /* 1018 * Normally, filepage is NULL on entry, and either found 1019 * uptodate immediately, or allocated and zeroed, or read 1020 * in under swappage, which is then assigned to filepage. 1021 * But shmem_prepare_write passes in a locked filepage, 1022 * which may be found not uptodate by other callers too, 1023 * and may need to be copied from the swappage read in. 1024 */ 1025repeat: 1026 if (!filepage) 1027 filepage = find_lock_page(mapping, idx); 1028 if (filepage && PageUptodate(filepage)) 1029 goto done; 1030 error = 0; 1031 if (sgp == SGP_QUICK) 1032 goto failed; 1033 1034 spin_lock(&info->lock); 1035 shmem_recalc_inode(inode); 1036 entry = shmem_swp_alloc(info, idx, sgp); 1037 if (IS_ERR(entry)) { 1038 spin_unlock(&info->lock); 1039 error = PTR_ERR(entry); 1040 goto failed; 1041 } 1042 swap = *entry; 1043 1044 if (swap.val) { 1045 /* Look it up and read it in.. */ 1046 swappage = lookup_swap_cache(swap); 1047 if (!swappage) { 1048 shmem_swp_unmap(entry); 1049 spin_unlock(&info->lock); 1050 /* here we actually do the io */ 1051 if (type && *type == VM_FAULT_MINOR) { 1052 inc_page_state(pgmajfault); 1053 *type = VM_FAULT_MAJOR; 1054 } 1055 swappage = shmem_swapin(info, swap, idx); 1056 if (!swappage) { 1057 spin_lock(&info->lock); 1058 entry = shmem_swp_alloc(info, idx, sgp); 1059 if (IS_ERR(entry)) 1060 error = PTR_ERR(entry); 1061 else { 1062 if (entry->val == swap.val) 1063 error = -ENOMEM; 1064 shmem_swp_unmap(entry); 1065 } 1066 spin_unlock(&info->lock); 1067 if (error) 1068 goto failed; 1069 goto repeat; 1070 } 1071 wait_on_page_locked(swappage); 1072 page_cache_release(swappage); 1073 goto repeat; 1074 } 1075 1076 /* We have to do this with page locked to prevent races */ 1077 if (TestSetPageLocked(swappage)) { 1078 shmem_swp_unmap(entry); 1079 spin_unlock(&info->lock); 1080 wait_on_page_locked(swappage); 1081 page_cache_release(swappage); 1082 goto repeat; 1083 } 1084 if (PageWriteback(swappage)) { 1085 shmem_swp_unmap(entry); 1086 spin_unlock(&info->lock); 1087 wait_on_page_writeback(swappage); 1088 unlock_page(swappage); 1089 page_cache_release(swappage); 1090 goto repeat; 1091 } 1092 if (!PageUptodate(swappage)) { 1093 shmem_swp_unmap(entry); 1094 spin_unlock(&info->lock); 1095 unlock_page(swappage); 1096 page_cache_release(swappage); 1097 error = -EIO; 1098 goto failed; 1099 } 1100 1101 if (filepage) { 1102 shmem_swp_set(info, entry, 0); 1103 shmem_swp_unmap(entry); 1104 delete_from_swap_cache(swappage); 1105 spin_unlock(&info->lock); 1106 copy_highpage(filepage, swappage); 1107 unlock_page(swappage); 1108 page_cache_release(swappage); 1109 flush_dcache_page(filepage); 1110 SetPageUptodate(filepage); 1111 set_page_dirty(filepage); 1112 swap_free(swap); 1113 } else if (!(error = move_from_swap_cache( 1114 swappage, idx, mapping))) { 1115 info->flags |= SHMEM_PAGEIN; 1116 shmem_swp_set(info, entry, 0); 1117 shmem_swp_unmap(entry); 1118 spin_unlock(&info->lock); 1119 filepage = swappage; 1120 swap_free(swap); 1121 } else { 1122 shmem_swp_unmap(entry); 1123 spin_unlock(&info->lock); 1124 unlock_page(swappage); 1125 page_cache_release(swappage); 1126 if (error == -ENOMEM) { 1127 /* let kswapd refresh zone for GFP_ATOMICs */ 1128 blk_congestion_wait(WRITE, HZ/50); 1129 } 1130 goto repeat; 1131 } 1132 } else if (sgp == SGP_READ && !filepage) { 1133 shmem_swp_unmap(entry); 1134 filepage = find_get_page(mapping, idx); 1135 if (filepage && 1136 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) { 1137 spin_unlock(&info->lock); 1138 wait_on_page_locked(filepage); 1139 page_cache_release(filepage); 1140 filepage = NULL; 1141 goto repeat; 1142 } 1143 spin_unlock(&info->lock); 1144 } else { 1145 shmem_swp_unmap(entry); 1146 sbinfo = SHMEM_SB(inode->i_sb); 1147 if (sbinfo->max_blocks) { 1148 spin_lock(&sbinfo->stat_lock); 1149 if (sbinfo->free_blocks == 0 || 1150 shmem_acct_block(info->flags)) { 1151 spin_unlock(&sbinfo->stat_lock); 1152 spin_unlock(&info->lock); 1153 error = -ENOSPC; 1154 goto failed; 1155 } 1156 sbinfo->free_blocks--; 1157 inode->i_blocks += BLOCKS_PER_PAGE; 1158 spin_unlock(&sbinfo->stat_lock); 1159 } else if (shmem_acct_block(info->flags)) { 1160 spin_unlock(&info->lock); 1161 error = -ENOSPC; 1162 goto failed; 1163 } 1164 1165 if (!filepage) { 1166 spin_unlock(&info->lock); 1167 filepage = shmem_alloc_page(mapping_gfp_mask(mapping), 1168 info, 1169 idx); 1170 if (!filepage) { 1171 shmem_unacct_blocks(info->flags, 1); 1172 shmem_free_blocks(inode, 1); 1173 error = -ENOMEM; 1174 goto failed; 1175 } 1176 1177 spin_lock(&info->lock); 1178 entry = shmem_swp_alloc(info, idx, sgp); 1179 if (IS_ERR(entry)) 1180 error = PTR_ERR(entry); 1181 else { 1182 swap = *entry; 1183 shmem_swp_unmap(entry); 1184 } 1185 if (error || swap.val || 0 != add_to_page_cache_lru( 1186 filepage, mapping, idx, GFP_ATOMIC)) { 1187 spin_unlock(&info->lock); 1188 page_cache_release(filepage); 1189 shmem_unacct_blocks(info->flags, 1); 1190 shmem_free_blocks(inode, 1); 1191 filepage = NULL; 1192 if (error) 1193 goto failed; 1194 goto repeat; 1195 } 1196 info->flags |= SHMEM_PAGEIN; 1197 } 1198 1199 info->alloced++; 1200 spin_unlock(&info->lock); 1201 flush_dcache_page(filepage); 1202 SetPageUptodate(filepage); 1203 } 1204done: 1205 if (*pagep != filepage) { 1206 unlock_page(filepage); 1207 *pagep = filepage; 1208 } 1209 return 0; 1210 1211failed: 1212 if (*pagep != filepage) { 1213 unlock_page(filepage); 1214 page_cache_release(filepage); 1215 } 1216 return error; 1217} 1218 1219struct page *shmem_nopage(struct vm_area_struct *vma, unsigned long address, int *type) 1220{ 1221 struct inode *inode = vma->vm_file->f_dentry->d_inode; 1222 struct page *page = NULL; 1223 unsigned long idx; 1224 int error; 1225 1226 idx = (address - vma->vm_start) >> PAGE_SHIFT; 1227 idx += vma->vm_pgoff; 1228 idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT; 1229 if (((loff_t) idx << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 1230 return NOPAGE_SIGBUS; 1231 1232 error = shmem_getpage(inode, idx, &page, SGP_CACHE, type); 1233 if (error) 1234 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS; 1235 1236 mark_page_accessed(page); 1237 return page; 1238} 1239 1240static int shmem_populate(struct vm_area_struct *vma, 1241 unsigned long addr, unsigned long len, 1242 pgprot_t prot, unsigned long pgoff, int nonblock) 1243{ 1244 struct inode *inode = vma->vm_file->f_dentry->d_inode; 1245 struct mm_struct *mm = vma->vm_mm; 1246 enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE; 1247 unsigned long size; 1248 1249 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT; 1250 if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size) 1251 return -EINVAL; 1252 1253 while ((long) len > 0) { 1254 struct page *page = NULL; 1255 int err; 1256 /* 1257 * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE 1258 */ 1259 err = shmem_getpage(inode, pgoff, &page, sgp, NULL); 1260 if (err) 1261 return err; 1262 /* Page may still be null, but only if nonblock was set. */ 1263 if (page) { 1264 mark_page_accessed(page); 1265 err = install_page(mm, vma, addr, page, prot); 1266 if (err) { 1267 page_cache_release(page); 1268 return err; 1269 } 1270 } else if (vma->vm_flags & VM_NONLINEAR) { 1271 /* No page was found just because we can't read it in 1272 * now (being here implies nonblock != 0), but the page 1273 * may exist, so set the PTE to fault it in later. */ 1274 err = install_file_pte(mm, vma, addr, pgoff, prot); 1275 if (err) 1276 return err; 1277 } 1278 1279 len -= PAGE_SIZE; 1280 addr += PAGE_SIZE; 1281 pgoff++; 1282 } 1283 return 0; 1284} 1285 1286#ifdef CONFIG_NUMA 1287int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new) 1288{ 1289 struct inode *i = vma->vm_file->f_dentry->d_inode; 1290 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new); 1291} 1292 1293struct mempolicy * 1294shmem_get_policy(struct vm_area_struct *vma, unsigned long addr) 1295{ 1296 struct inode *i = vma->vm_file->f_dentry->d_inode; 1297 unsigned long idx; 1298 1299 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; 1300 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx); 1301} 1302#endif 1303 1304int shmem_lock(struct file *file, int lock, struct user_struct *user) 1305{ 1306 struct inode *inode = file->f_dentry->d_inode; 1307 struct shmem_inode_info *info = SHMEM_I(inode); 1308 int retval = -ENOMEM; 1309 1310 spin_lock(&info->lock); 1311 if (lock && !(info->flags & VM_LOCKED)) { 1312 if (!user_shm_lock(inode->i_size, user)) 1313 goto out_nomem; 1314 info->flags |= VM_LOCKED; 1315 } 1316 if (!lock && (info->flags & VM_LOCKED) && user) { 1317 user_shm_unlock(inode->i_size, user); 1318 info->flags &= ~VM_LOCKED; 1319 } 1320 retval = 0; 1321out_nomem: 1322 spin_unlock(&info->lock); 1323 return retval; 1324} 1325 1326int shmem_mmap(struct file *file, struct vm_area_struct *vma) 1327{ 1328 file_accessed(file); 1329 vma->vm_ops = &shmem_vm_ops; 1330 return 0; 1331} 1332 1333static struct inode * 1334shmem_get_inode(struct super_block *sb, int mode, dev_t dev) 1335{ 1336 struct inode *inode; 1337 struct shmem_inode_info *info; 1338 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 1339 1340 if (sbinfo->max_inodes) { 1341 spin_lock(&sbinfo->stat_lock); 1342 if (!sbinfo->free_inodes) { 1343 spin_unlock(&sbinfo->stat_lock); 1344 return NULL; 1345 } 1346 sbinfo->free_inodes--; 1347 spin_unlock(&sbinfo->stat_lock); 1348 } 1349 1350 inode = new_inode(sb); 1351 if (inode) { 1352 inode->i_mode = mode; 1353 inode->i_uid = current->fsuid; 1354 inode->i_gid = current->fsgid; 1355 inode->i_blksize = PAGE_CACHE_SIZE; 1356 inode->i_blocks = 0; 1357 inode->i_mapping->a_ops = &shmem_aops; 1358 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; 1359 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 1360 info = SHMEM_I(inode); 1361 memset(info, 0, (char *)inode - (char *)info); 1362 spin_lock_init(&info->lock); 1363 INIT_LIST_HEAD(&info->swaplist); 1364 1365 switch (mode & S_IFMT) { 1366 default: 1367 init_special_inode(inode, mode, dev); 1368 break; 1369 case S_IFREG: 1370 inode->i_op = &shmem_inode_operations; 1371 inode->i_fop = &shmem_file_operations; 1372 mpol_shared_policy_init(&info->policy, sbinfo->policy, 1373 &sbinfo->policy_nodes); 1374 break; 1375 case S_IFDIR: 1376 inode->i_nlink++; 1377 /* Some things misbehave if size == 0 on a directory */ 1378 inode->i_size = 2 * BOGO_DIRENT_SIZE; 1379 inode->i_op = &shmem_dir_inode_operations; 1380 inode->i_fop = &simple_dir_operations; 1381 break; 1382 case S_IFLNK: 1383 /* 1384 * Must not load anything in the rbtree, 1385 * mpol_free_shared_policy will not be called. 1386 */ 1387 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT, 1388 NULL); 1389 break; 1390 } 1391 } else if (sbinfo->max_inodes) { 1392 spin_lock(&sbinfo->stat_lock); 1393 sbinfo->free_inodes++; 1394 spin_unlock(&sbinfo->stat_lock); 1395 } 1396 return inode; 1397} 1398 1399#ifdef CONFIG_TMPFS 1400static struct inode_operations shmem_symlink_inode_operations; 1401static struct inode_operations shmem_symlink_inline_operations; 1402 1403/* 1404 * Normally tmpfs makes no use of shmem_prepare_write, but it 1405 * lets a tmpfs file be used read-write below the loop driver. 1406 */ 1407static int 1408shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to) 1409{ 1410 struct inode *inode = page->mapping->host; 1411 return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL); 1412} 1413 1414static ssize_t 1415shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos) 1416{ 1417 struct inode *inode = file->f_dentry->d_inode; 1418 loff_t pos; 1419 unsigned long written; 1420 ssize_t err; 1421 1422 if ((ssize_t) count < 0) 1423 return -EINVAL; 1424 1425 if (!access_ok(VERIFY_READ, buf, count)) 1426 return -EFAULT; 1427 1428 mutex_lock(&inode->i_mutex); 1429 1430 pos = *ppos; 1431 written = 0; 1432 1433 err = generic_write_checks(file, &pos, &count, 0); 1434 if (err || !count) 1435 goto out; 1436 1437 err = remove_suid(file->f_dentry); 1438 if (err) 1439 goto out; 1440 1441 inode->i_ctime = inode->i_mtime = CURRENT_TIME; 1442 1443 do { 1444 struct page *page = NULL; 1445 unsigned long bytes, index, offset; 1446 char *kaddr; 1447 int left; 1448 1449 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */ 1450 index = pos >> PAGE_CACHE_SHIFT; 1451 bytes = PAGE_CACHE_SIZE - offset; 1452 if (bytes > count) 1453 bytes = count; 1454 1455 /* 1456 * We don't hold page lock across copy from user - 1457 * what would it guard against? - so no deadlock here. 1458 * But it still may be a good idea to prefault below. 1459 */ 1460 1461 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL); 1462 if (err) 1463 break; 1464 1465 left = bytes; 1466 if (PageHighMem(page)) { 1467 volatile unsigned char dummy; 1468 __get_user(dummy, buf); 1469 __get_user(dummy, buf + bytes - 1); 1470 1471 kaddr = kmap_atomic(page, KM_USER0); 1472 left = __copy_from_user_inatomic(kaddr + offset, 1473 buf, bytes); 1474 kunmap_atomic(kaddr, KM_USER0); 1475 } 1476 if (left) { 1477 kaddr = kmap(page); 1478 left = __copy_from_user(kaddr + offset, buf, bytes); 1479 kunmap(page); 1480 } 1481 1482 written += bytes; 1483 count -= bytes; 1484 pos += bytes; 1485 buf += bytes; 1486 if (pos > inode->i_size) 1487 i_size_write(inode, pos); 1488 1489 flush_dcache_page(page); 1490 set_page_dirty(page); 1491 mark_page_accessed(page); 1492 page_cache_release(page); 1493 1494 if (left) { 1495 pos -= left; 1496 written -= left; 1497 err = -EFAULT; 1498 break; 1499 } 1500 1501 /* 1502 * Our dirty pages are not counted in nr_dirty, 1503 * and we do not attempt to balance dirty pages. 1504 */ 1505 1506 cond_resched(); 1507 } while (count); 1508 1509 *ppos = pos; 1510 if (written) 1511 err = written; 1512out: 1513 mutex_unlock(&inode->i_mutex); 1514 return err; 1515} 1516 1517static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor) 1518{ 1519 struct inode *inode = filp->f_dentry->d_inode; 1520 struct address_space *mapping = inode->i_mapping; 1521 unsigned long index, offset; 1522 1523 index = *ppos >> PAGE_CACHE_SHIFT; 1524 offset = *ppos & ~PAGE_CACHE_MASK; 1525 1526 for (;;) { 1527 struct page *page = NULL; 1528 unsigned long end_index, nr, ret; 1529 loff_t i_size = i_size_read(inode); 1530 1531 end_index = i_size >> PAGE_CACHE_SHIFT; 1532 if (index > end_index) 1533 break; 1534 if (index == end_index) { 1535 nr = i_size & ~PAGE_CACHE_MASK; 1536 if (nr <= offset) 1537 break; 1538 } 1539 1540 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL); 1541 if (desc->error) { 1542 if (desc->error == -EINVAL) 1543 desc->error = 0; 1544 break; 1545 } 1546 1547 /* 1548 * We must evaluate after, since reads (unlike writes) 1549 * are called without i_mutex protection against truncate 1550 */ 1551 nr = PAGE_CACHE_SIZE; 1552 i_size = i_size_read(inode); 1553 end_index = i_size >> PAGE_CACHE_SHIFT; 1554 if (index == end_index) { 1555 nr = i_size & ~PAGE_CACHE_MASK; 1556 if (nr <= offset) { 1557 if (page) 1558 page_cache_release(page); 1559 break; 1560 } 1561 } 1562 nr -= offset; 1563 1564 if (page) { 1565 /* 1566 * If users can be writing to this page using arbitrary 1567 * virtual addresses, take care about potential aliasing 1568 * before reading the page on the kernel side. 1569 */ 1570 if (mapping_writably_mapped(mapping)) 1571 flush_dcache_page(page); 1572 /* 1573 * Mark the page accessed if we read the beginning. 1574 */ 1575 if (!offset) 1576 mark_page_accessed(page); 1577 } else { 1578 page = ZERO_PAGE(0); 1579 page_cache_get(page); 1580 } 1581 1582 /* 1583 * Ok, we have the page, and it's up-to-date, so 1584 * now we can copy it to user space... 1585 * 1586 * The actor routine returns how many bytes were actually used.. 1587 * NOTE! This may not be the same as how much of a user buffer 1588 * we filled up (we may be padding etc), so we can only update 1589 * "pos" here (the actor routine has to update the user buffer 1590 * pointers and the remaining count). 1591 */ 1592 ret = actor(desc, page, offset, nr); 1593 offset += ret; 1594 index += offset >> PAGE_CACHE_SHIFT; 1595 offset &= ~PAGE_CACHE_MASK; 1596 1597 page_cache_release(page); 1598 if (ret != nr || !desc->count) 1599 break; 1600 1601 cond_resched(); 1602 } 1603 1604 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; 1605 file_accessed(filp); 1606} 1607 1608static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos) 1609{ 1610 read_descriptor_t desc; 1611 1612 if ((ssize_t) count < 0) 1613 return -EINVAL; 1614 if (!access_ok(VERIFY_WRITE, buf, count)) 1615 return -EFAULT; 1616 if (!count) 1617 return 0; 1618 1619 desc.written = 0; 1620 desc.count = count; 1621 desc.arg.buf = buf; 1622 desc.error = 0; 1623 1624 do_shmem_file_read(filp, ppos, &desc, file_read_actor); 1625 if (desc.written) 1626 return desc.written; 1627 return desc.error; 1628} 1629 1630static ssize_t shmem_file_sendfile(struct file *in_file, loff_t *ppos, 1631 size_t count, read_actor_t actor, void *target) 1632{ 1633 read_descriptor_t desc; 1634 1635 if (!count) 1636 return 0; 1637 1638 desc.written = 0; 1639 desc.count = count; 1640 desc.arg.data = target; 1641 desc.error = 0; 1642 1643 do_shmem_file_read(in_file, ppos, &desc, actor); 1644 if (desc.written) 1645 return desc.written; 1646 return desc.error; 1647} 1648 1649static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) 1650{ 1651 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); 1652 1653 buf->f_type = TMPFS_MAGIC; 1654 buf->f_bsize = PAGE_CACHE_SIZE; 1655 buf->f_namelen = NAME_MAX; 1656 spin_lock(&sbinfo->stat_lock); 1657 if (sbinfo->max_blocks) { 1658 buf->f_blocks = sbinfo->max_blocks; 1659 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks; 1660 } 1661 if (sbinfo->max_inodes) { 1662 buf->f_files = sbinfo->max_inodes; 1663 buf->f_ffree = sbinfo->free_inodes; 1664 } 1665 /* else leave those fields 0 like simple_statfs */ 1666 spin_unlock(&sbinfo->stat_lock); 1667 return 0; 1668} 1669 1670/* 1671 * File creation. Allocate an inode, and we're done.. 1672 */ 1673static int 1674shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 1675{ 1676 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev); 1677 int error = -ENOSPC; 1678 1679 if (inode) { 1680 error = security_inode_init_security(inode, dir, NULL, NULL, 1681 NULL); 1682 if (error) { 1683 if (error != -EOPNOTSUPP) { 1684 iput(inode); 1685 return error; 1686 } 1687 error = 0; 1688 } 1689 if (dir->i_mode & S_ISGID) { 1690 inode->i_gid = dir->i_gid; 1691 if (S_ISDIR(mode)) 1692 inode->i_mode |= S_ISGID; 1693 } 1694 dir->i_size += BOGO_DIRENT_SIZE; 1695 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1696 d_instantiate(dentry, inode); 1697 dget(dentry); /* Extra count - pin the dentry in core */ 1698 } 1699 return error; 1700} 1701 1702static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1703{ 1704 int error; 1705 1706 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) 1707 return error; 1708 dir->i_nlink++; 1709 return 0; 1710} 1711 1712static int shmem_create(struct inode *dir, struct dentry *dentry, int mode, 1713 struct nameidata *nd) 1714{ 1715 return shmem_mknod(dir, dentry, mode | S_IFREG, 0); 1716} 1717 1718/* 1719 * Link a file.. 1720 */ 1721static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1722{ 1723 struct inode *inode = old_dentry->d_inode; 1724 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 1725 1726 /* 1727 * No ordinary (disk based) filesystem counts links as inodes; 1728 * but each new link needs a new dentry, pinning lowmem, and 1729 * tmpfs dentries cannot be pruned until they are unlinked. 1730 */ 1731 if (sbinfo->max_inodes) { 1732 spin_lock(&sbinfo->stat_lock); 1733 if (!sbinfo->free_inodes) { 1734 spin_unlock(&sbinfo->stat_lock); 1735 return -ENOSPC; 1736 } 1737 sbinfo->free_inodes--; 1738 spin_unlock(&sbinfo->stat_lock); 1739 } 1740 1741 dir->i_size += BOGO_DIRENT_SIZE; 1742 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1743 inode->i_nlink++; 1744 atomic_inc(&inode->i_count); /* New dentry reference */ 1745 dget(dentry); /* Extra pinning count for the created dentry */ 1746 d_instantiate(dentry, inode); 1747 return 0; 1748} 1749 1750static int shmem_unlink(struct inode *dir, struct dentry *dentry) 1751{ 1752 struct inode *inode = dentry->d_inode; 1753 1754 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) { 1755 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 1756 if (sbinfo->max_inodes) { 1757 spin_lock(&sbinfo->stat_lock); 1758 sbinfo->free_inodes++; 1759 spin_unlock(&sbinfo->stat_lock); 1760 } 1761 } 1762 1763 dir->i_size -= BOGO_DIRENT_SIZE; 1764 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1765 inode->i_nlink--; 1766 dput(dentry); /* Undo the count from "create" - this does all the work */ 1767 return 0; 1768} 1769 1770static int shmem_rmdir(struct inode *dir, struct dentry *dentry) 1771{ 1772 if (!simple_empty(dentry)) 1773 return -ENOTEMPTY; 1774 1775 dentry->d_inode->i_nlink--; 1776 dir->i_nlink--; 1777 return shmem_unlink(dir, dentry); 1778} 1779 1780/* 1781 * The VFS layer already does all the dentry stuff for rename, 1782 * we just have to decrement the usage count for the target if 1783 * it exists so that the VFS layer correctly free's it when it 1784 * gets overwritten. 1785 */ 1786static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) 1787{ 1788 struct inode *inode = old_dentry->d_inode; 1789 int they_are_dirs = S_ISDIR(inode->i_mode); 1790 1791 if (!simple_empty(new_dentry)) 1792 return -ENOTEMPTY; 1793 1794 if (new_dentry->d_inode) { 1795 (void) shmem_unlink(new_dir, new_dentry); 1796 if (they_are_dirs) 1797 old_dir->i_nlink--; 1798 } else if (they_are_dirs) { 1799 old_dir->i_nlink--; 1800 new_dir->i_nlink++; 1801 } 1802 1803 old_dir->i_size -= BOGO_DIRENT_SIZE; 1804 new_dir->i_size += BOGO_DIRENT_SIZE; 1805 old_dir->i_ctime = old_dir->i_mtime = 1806 new_dir->i_ctime = new_dir->i_mtime = 1807 inode->i_ctime = CURRENT_TIME; 1808 return 0; 1809} 1810 1811static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1812{ 1813 int error; 1814 int len; 1815 struct inode *inode; 1816 struct page *page = NULL; 1817 char *kaddr; 1818 struct shmem_inode_info *info; 1819 1820 len = strlen(symname) + 1; 1821 if (len > PAGE_CACHE_SIZE) 1822 return -ENAMETOOLONG; 1823 1824 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0); 1825 if (!inode) 1826 return -ENOSPC; 1827 1828 error = security_inode_init_security(inode, dir, NULL, NULL, 1829 NULL); 1830 if (error) { 1831 if (error != -EOPNOTSUPP) { 1832 iput(inode); 1833 return error; 1834 } 1835 error = 0; 1836 } 1837 1838 info = SHMEM_I(inode); 1839 inode->i_size = len-1; 1840 if (len <= (char *)inode - (char *)info) { 1841 /* do it inline */ 1842 memcpy(info, symname, len); 1843 inode->i_op = &shmem_symlink_inline_operations; 1844 } else { 1845 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); 1846 if (error) { 1847 iput(inode); 1848 return error; 1849 } 1850 inode->i_op = &shmem_symlink_inode_operations; 1851 kaddr = kmap_atomic(page, KM_USER0); 1852 memcpy(kaddr, symname, len); 1853 kunmap_atomic(kaddr, KM_USER0); 1854 set_page_dirty(page); 1855 page_cache_release(page); 1856 } 1857 if (dir->i_mode & S_ISGID) 1858 inode->i_gid = dir->i_gid; 1859 dir->i_size += BOGO_DIRENT_SIZE; 1860 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1861 d_instantiate(dentry, inode); 1862 dget(dentry); 1863 return 0; 1864} 1865 1866static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd) 1867{ 1868 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode)); 1869 return NULL; 1870} 1871 1872static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) 1873{ 1874 struct page *page = NULL; 1875 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); 1876 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page)); 1877 return page; 1878} 1879 1880static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 1881{ 1882 if (!IS_ERR(nd_get_link(nd))) { 1883 struct page *page = cookie; 1884 kunmap(page); 1885 mark_page_accessed(page); 1886 page_cache_release(page); 1887 } 1888} 1889 1890static struct inode_operations shmem_symlink_inline_operations = { 1891 .readlink = generic_readlink, 1892 .follow_link = shmem_follow_link_inline, 1893}; 1894 1895static struct inode_operations shmem_symlink_inode_operations = { 1896 .truncate = shmem_truncate, 1897 .readlink = generic_readlink, 1898 .follow_link = shmem_follow_link, 1899 .put_link = shmem_put_link, 1900}; 1901 1902static int shmem_parse_options(char *options, int *mode, uid_t *uid, 1903 gid_t *gid, unsigned long *blocks, unsigned long *inodes, 1904 int *policy, nodemask_t *policy_nodes) 1905{ 1906 char *this_char, *value, *rest; 1907 1908 while (options != NULL) { 1909 this_char = options; 1910 for (;;) { 1911 /* 1912 * NUL-terminate this option: unfortunately, 1913 * mount options form a comma-separated list, 1914 * but mpol's nodelist may also contain commas. 1915 */ 1916 options = strchr(options, ','); 1917 if (options == NULL) 1918 break; 1919 options++; 1920 if (!isdigit(*options)) { 1921 options[-1] = '\0'; 1922 break; 1923 } 1924 } 1925 if (!*this_char) 1926 continue; 1927 if ((value = strchr(this_char,'=')) != NULL) { 1928 *value++ = 0; 1929 } else { 1930 printk(KERN_ERR 1931 "tmpfs: No value for mount option '%s'\n", 1932 this_char); 1933 return 1; 1934 } 1935 1936 if (!strcmp(this_char,"size")) { 1937 unsigned long long size; 1938 size = memparse(value,&rest); 1939 if (*rest == '%') { 1940 size <<= PAGE_SHIFT; 1941 size *= totalram_pages; 1942 do_div(size, 100); 1943 rest++; 1944 } 1945 if (*rest) 1946 goto bad_val; 1947 *blocks = size >> PAGE_CACHE_SHIFT; 1948 } else if (!strcmp(this_char,"nr_blocks")) { 1949 *blocks = memparse(value,&rest); 1950 if (*rest) 1951 goto bad_val; 1952 } else if (!strcmp(this_char,"nr_inodes")) { 1953 *inodes = memparse(value,&rest); 1954 if (*rest) 1955 goto bad_val; 1956 } else if (!strcmp(this_char,"mode")) { 1957 if (!mode) 1958 continue; 1959 *mode = simple_strtoul(value,&rest,8); 1960 if (*rest) 1961 goto bad_val; 1962 } else if (!strcmp(this_char,"uid")) { 1963 if (!uid) 1964 continue; 1965 *uid = simple_strtoul(value,&rest,0); 1966 if (*rest) 1967 goto bad_val; 1968 } else if (!strcmp(this_char,"gid")) { 1969 if (!gid) 1970 continue; 1971 *gid = simple_strtoul(value,&rest,0); 1972 if (*rest) 1973 goto bad_val; 1974 } else if (!strcmp(this_char,"mpol")) { 1975 if (shmem_parse_mpol(value,policy,policy_nodes)) 1976 goto bad_val; 1977 } else { 1978 printk(KERN_ERR "tmpfs: Bad mount option %s\n", 1979 this_char); 1980 return 1; 1981 } 1982 } 1983 return 0; 1984 1985bad_val: 1986 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", 1987 value, this_char); 1988 return 1; 1989 1990} 1991 1992static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) 1993{ 1994 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 1995 unsigned long max_blocks = sbinfo->max_blocks; 1996 unsigned long max_inodes = sbinfo->max_inodes; 1997 int policy = sbinfo->policy; 1998 nodemask_t policy_nodes = sbinfo->policy_nodes; 1999 unsigned long blocks; 2000 unsigned long inodes; 2001 int error = -EINVAL; 2002 2003 if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks, 2004 &max_inodes, &policy, &policy_nodes)) 2005 return error; 2006 2007 spin_lock(&sbinfo->stat_lock); 2008 blocks = sbinfo->max_blocks - sbinfo->free_blocks; 2009 inodes = sbinfo->max_inodes - sbinfo->free_inodes; 2010 if (max_blocks < blocks) 2011 goto out; 2012 if (max_inodes < inodes) 2013 goto out; 2014 /* 2015 * Those tests also disallow limited->unlimited while any are in 2016 * use, so i_blocks will always be zero when max_blocks is zero; 2017 * but we must separately disallow unlimited->limited, because 2018 * in that case we have no record of how much is already in use. 2019 */ 2020 if (max_blocks && !sbinfo->max_blocks) 2021 goto out; 2022 if (max_inodes && !sbinfo->max_inodes) 2023 goto out; 2024 2025 error = 0; 2026 sbinfo->max_blocks = max_blocks; 2027 sbinfo->free_blocks = max_blocks - blocks; 2028 sbinfo->max_inodes = max_inodes; 2029 sbinfo->free_inodes = max_inodes - inodes; 2030 sbinfo->policy = policy; 2031 sbinfo->policy_nodes = policy_nodes; 2032out: 2033 spin_unlock(&sbinfo->stat_lock); 2034 return error; 2035} 2036#endif 2037 2038static void shmem_put_super(struct super_block *sb) 2039{ 2040 kfree(sb->s_fs_info); 2041 sb->s_fs_info = NULL; 2042} 2043 2044static int shmem_fill_super(struct super_block *sb, 2045 void *data, int silent) 2046{ 2047 struct inode *inode; 2048 struct dentry *root; 2049 int mode = S_IRWXUGO | S_ISVTX; 2050 uid_t uid = current->fsuid; 2051 gid_t gid = current->fsgid; 2052 int err = -ENOMEM; 2053 struct shmem_sb_info *sbinfo; 2054 unsigned long blocks = 0; 2055 unsigned long inodes = 0; 2056 int policy = MPOL_DEFAULT; 2057 nodemask_t policy_nodes = node_online_map; 2058 2059#ifdef CONFIG_TMPFS 2060 /* 2061 * Per default we only allow half of the physical ram per 2062 * tmpfs instance, limiting inodes to one per page of lowmem; 2063 * but the internal instance is left unlimited. 2064 */ 2065 if (!(sb->s_flags & MS_NOUSER)) { 2066 blocks = totalram_pages / 2; 2067 inodes = totalram_pages - totalhigh_pages; 2068 if (inodes > blocks) 2069 inodes = blocks; 2070 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks, 2071 &inodes, &policy, &policy_nodes)) 2072 return -EINVAL; 2073 } 2074#else 2075 sb->s_flags |= MS_NOUSER; 2076#endif 2077 2078 /* Round up to L1_CACHE_BYTES to resist false sharing */ 2079 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info), 2080 L1_CACHE_BYTES), GFP_KERNEL); 2081 if (!sbinfo) 2082 return -ENOMEM; 2083 2084 spin_lock_init(&sbinfo->stat_lock); 2085 sbinfo->max_blocks = blocks; 2086 sbinfo->free_blocks = blocks; 2087 sbinfo->max_inodes = inodes; 2088 sbinfo->free_inodes = inodes; 2089 sbinfo->policy = policy; 2090 sbinfo->policy_nodes = policy_nodes; 2091 2092 sb->s_fs_info = sbinfo; 2093 sb->s_maxbytes = SHMEM_MAX_BYTES; 2094 sb->s_blocksize = PAGE_CACHE_SIZE; 2095 sb->s_blocksize_bits = PAGE_CACHE_SHIFT; 2096 sb->s_magic = TMPFS_MAGIC; 2097 sb->s_op = &shmem_ops; 2098 sb->s_time_gran = 1; 2099 2100 inode = shmem_get_inode(sb, S_IFDIR | mode, 0); 2101 if (!inode) 2102 goto failed; 2103 inode->i_uid = uid; 2104 inode->i_gid = gid; 2105 root = d_alloc_root(inode); 2106 if (!root) 2107 goto failed_iput; 2108 sb->s_root = root; 2109 return 0; 2110 2111failed_iput: 2112 iput(inode); 2113failed: 2114 shmem_put_super(sb); 2115 return err; 2116} 2117 2118static struct kmem_cache *shmem_inode_cachep; 2119 2120static struct inode *shmem_alloc_inode(struct super_block *sb) 2121{ 2122 struct shmem_inode_info *p; 2123 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, SLAB_KERNEL); 2124 if (!p) 2125 return NULL; 2126 return &p->vfs_inode; 2127} 2128 2129static void shmem_destroy_inode(struct inode *inode) 2130{ 2131 if ((inode->i_mode & S_IFMT) == S_IFREG) { 2132 /* only struct inode is valid if it's an inline symlink */ 2133 mpol_free_shared_policy(&SHMEM_I(inode)->policy); 2134 } 2135 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); 2136} 2137 2138static void init_once(void *foo, struct kmem_cache *cachep, 2139 unsigned long flags) 2140{ 2141 struct shmem_inode_info *p = (struct shmem_inode_info *) foo; 2142 2143 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) == 2144 SLAB_CTOR_CONSTRUCTOR) { 2145 inode_init_once(&p->vfs_inode); 2146 } 2147} 2148 2149static int init_inodecache(void) 2150{ 2151 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", 2152 sizeof(struct shmem_inode_info), 2153 0, 0, init_once, NULL); 2154 if (shmem_inode_cachep == NULL) 2155 return -ENOMEM; 2156 return 0; 2157} 2158 2159static void destroy_inodecache(void) 2160{ 2161 if (kmem_cache_destroy(shmem_inode_cachep)) 2162 printk(KERN_INFO "shmem_inode_cache: not all structures were freed\n"); 2163} 2164 2165static const struct address_space_operations shmem_aops = { 2166 .writepage = shmem_writepage, 2167 .set_page_dirty = __set_page_dirty_nobuffers, 2168#ifdef CONFIG_TMPFS 2169 .prepare_write = shmem_prepare_write, 2170 .commit_write = simple_commit_write, 2171#endif 2172 .migratepage = migrate_page, 2173}; 2174 2175static struct file_operations shmem_file_operations = { 2176 .mmap = shmem_mmap, 2177#ifdef CONFIG_TMPFS 2178 .llseek = generic_file_llseek, 2179 .read = shmem_file_read, 2180 .write = shmem_file_write, 2181 .fsync = simple_sync_file, 2182 .sendfile = shmem_file_sendfile, 2183#endif 2184}; 2185 2186static struct inode_operations shmem_inode_operations = { 2187 .truncate = shmem_truncate, 2188 .setattr = shmem_notify_change, 2189 .truncate_range = shmem_truncate_range, 2190}; 2191 2192static struct inode_operations shmem_dir_inode_operations = { 2193#ifdef CONFIG_TMPFS 2194 .create = shmem_create, 2195 .lookup = simple_lookup, 2196 .link = shmem_link, 2197 .unlink = shmem_unlink, 2198 .symlink = shmem_symlink, 2199 .mkdir = shmem_mkdir, 2200 .rmdir = shmem_rmdir, 2201 .mknod = shmem_mknod, 2202 .rename = shmem_rename, 2203#endif 2204}; 2205 2206static struct super_operations shmem_ops = { 2207 .alloc_inode = shmem_alloc_inode, 2208 .destroy_inode = shmem_destroy_inode, 2209#ifdef CONFIG_TMPFS 2210 .statfs = shmem_statfs, 2211 .remount_fs = shmem_remount_fs, 2212#endif 2213 .delete_inode = shmem_delete_inode, 2214 .drop_inode = generic_delete_inode, 2215 .put_super = shmem_put_super, 2216}; 2217 2218static struct vm_operations_struct shmem_vm_ops = { 2219 .nopage = shmem_nopage, 2220 .populate = shmem_populate, 2221#ifdef CONFIG_NUMA 2222 .set_policy = shmem_set_policy, 2223 .get_policy = shmem_get_policy, 2224#endif 2225}; 2226 2227 2228static int shmem_get_sb(struct file_system_type *fs_type, 2229 int flags, const char *dev_name, void *data, struct vfsmount *mnt) 2230{ 2231 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt); 2232} 2233 2234static struct file_system_type tmpfs_fs_type = { 2235 .owner = THIS_MODULE, 2236 .name = "tmpfs", 2237 .get_sb = shmem_get_sb, 2238 .kill_sb = kill_litter_super, 2239}; 2240static struct vfsmount *shm_mnt; 2241 2242static int __init init_tmpfs(void) 2243{ 2244 int error; 2245 2246 error = init_inodecache(); 2247 if (error) 2248 goto out3; 2249 2250 error = register_filesystem(&tmpfs_fs_type); 2251 if (error) { 2252 printk(KERN_ERR "Could not register tmpfs\n"); 2253 goto out2; 2254 } 2255#ifdef CONFIG_TMPFS 2256 devfs_mk_dir("shm"); 2257#endif 2258 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER, 2259 tmpfs_fs_type.name, NULL); 2260 if (IS_ERR(shm_mnt)) { 2261 error = PTR_ERR(shm_mnt); 2262 printk(KERN_ERR "Could not kern_mount tmpfs\n"); 2263 goto out1; 2264 } 2265 return 0; 2266 2267out1: 2268 unregister_filesystem(&tmpfs_fs_type); 2269out2: 2270 destroy_inodecache(); 2271out3: 2272 shm_mnt = ERR_PTR(error); 2273 return error; 2274} 2275module_init(init_tmpfs) 2276 2277/* 2278 * shmem_file_setup - get an unlinked file living in tmpfs 2279 * 2280 * @name: name for dentry (to be seen in /proc/<pid>/maps 2281 * @size: size to be set for the file 2282 * 2283 */ 2284struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags) 2285{ 2286 int error; 2287 struct file *file; 2288 struct inode *inode; 2289 struct dentry *dentry, *root; 2290 struct qstr this; 2291 2292 if (IS_ERR(shm_mnt)) 2293 return (void *)shm_mnt; 2294 2295 if (size < 0 || size > SHMEM_MAX_BYTES) 2296 return ERR_PTR(-EINVAL); 2297 2298 if (shmem_acct_size(flags, size)) 2299 return ERR_PTR(-ENOMEM); 2300 2301 error = -ENOMEM; 2302 this.name = name; 2303 this.len = strlen(name); 2304 this.hash = 0; /* will go */ 2305 root = shm_mnt->mnt_root; 2306 dentry = d_alloc(root, &this); 2307 if (!dentry) 2308 goto put_memory; 2309 2310 error = -ENFILE; 2311 file = get_empty_filp(); 2312 if (!file) 2313 goto put_dentry; 2314 2315 error = -ENOSPC; 2316 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0); 2317 if (!inode) 2318 goto close_file; 2319 2320 SHMEM_I(inode)->flags = flags & VM_ACCOUNT; 2321 d_instantiate(dentry, inode); 2322 inode->i_size = size; 2323 inode->i_nlink = 0; /* It is unlinked */ 2324 file->f_vfsmnt = mntget(shm_mnt); 2325 file->f_dentry = dentry; 2326 file->f_mapping = inode->i_mapping; 2327 file->f_op = &shmem_file_operations; 2328 file->f_mode = FMODE_WRITE | FMODE_READ; 2329 return file; 2330 2331close_file: 2332 put_filp(file); 2333put_dentry: 2334 dput(dentry); 2335put_memory: 2336 shmem_unacct_size(flags, size); 2337 return ERR_PTR(error); 2338} 2339 2340/* 2341 * shmem_zero_setup - setup a shared anonymous mapping 2342 * 2343 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff 2344 */ 2345int shmem_zero_setup(struct vm_area_struct *vma) 2346{ 2347 struct file *file; 2348 loff_t size = vma->vm_end - vma->vm_start; 2349 2350 file = shmem_file_setup("dev/zero", size, vma->vm_flags); 2351 if (IS_ERR(file)) 2352 return PTR_ERR(file); 2353 2354 if (vma->vm_file) 2355 fput(vma->vm_file); 2356 vma->vm_file = file; 2357 vma->vm_ops = &shmem_vm_ops; 2358 return 0; 2359} 2360