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