shmem.c revision bda97eab0cc9c6385b9f26abdda6459f630f4513
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 * tiny-shmem: 18 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com> 19 * 20 * This file is released under the GPL. 21 */ 22 23#include <linux/fs.h> 24#include <linux/init.h> 25#include <linux/vfs.h> 26#include <linux/mount.h> 27#include <linux/pagemap.h> 28#include <linux/file.h> 29#include <linux/mm.h> 30#include <linux/module.h> 31#include <linux/swap.h> 32 33static struct vfsmount *shm_mnt; 34 35#ifdef CONFIG_SHMEM 36/* 37 * This virtual memory filesystem is heavily based on the ramfs. It 38 * extends ramfs by the ability to use swap and honor resource limits 39 * which makes it a completely usable filesystem. 40 */ 41 42#include <linux/xattr.h> 43#include <linux/exportfs.h> 44#include <linux/posix_acl.h> 45#include <linux/generic_acl.h> 46#include <linux/mman.h> 47#include <linux/string.h> 48#include <linux/slab.h> 49#include <linux/backing-dev.h> 50#include <linux/shmem_fs.h> 51#include <linux/writeback.h> 52#include <linux/blkdev.h> 53#include <linux/pagevec.h> 54#include <linux/percpu_counter.h> 55#include <linux/splice.h> 56#include <linux/security.h> 57#include <linux/swapops.h> 58#include <linux/mempolicy.h> 59#include <linux/namei.h> 60#include <linux/ctype.h> 61#include <linux/migrate.h> 62#include <linux/highmem.h> 63#include <linux/seq_file.h> 64#include <linux/magic.h> 65 66#include <asm/uaccess.h> 67#include <asm/pgtable.h> 68 69#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512) 70#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT) 71 72/* Pretend that each entry is of this size in directory's i_size */ 73#define BOGO_DIRENT_SIZE 20 74 75struct shmem_xattr { 76 struct list_head list; /* anchored by shmem_inode_info->xattr_list */ 77 char *name; /* xattr name */ 78 size_t size; 79 char value[0]; 80}; 81 82/* Flag allocation requirements to shmem_getpage */ 83enum sgp_type { 84 SGP_READ, /* don't exceed i_size, don't allocate page */ 85 SGP_CACHE, /* don't exceed i_size, may allocate page */ 86 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */ 87 SGP_WRITE, /* may exceed i_size, may allocate page */ 88}; 89 90#ifdef CONFIG_TMPFS 91static unsigned long shmem_default_max_blocks(void) 92{ 93 return totalram_pages / 2; 94} 95 96static unsigned long shmem_default_max_inodes(void) 97{ 98 return min(totalram_pages - totalhigh_pages, totalram_pages / 2); 99} 100#endif 101 102static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, 103 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type); 104 105static inline int shmem_getpage(struct inode *inode, pgoff_t index, 106 struct page **pagep, enum sgp_type sgp, int *fault_type) 107{ 108 return shmem_getpage_gfp(inode, index, pagep, sgp, 109 mapping_gfp_mask(inode->i_mapping), fault_type); 110} 111 112static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb) 113{ 114 return sb->s_fs_info; 115} 116 117/* 118 * shmem_file_setup pre-accounts the whole fixed size of a VM object, 119 * for shared memory and for shared anonymous (/dev/zero) mappings 120 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1), 121 * consistent with the pre-accounting of private mappings ... 122 */ 123static inline int shmem_acct_size(unsigned long flags, loff_t size) 124{ 125 return (flags & VM_NORESERVE) ? 126 0 : security_vm_enough_memory_kern(VM_ACCT(size)); 127} 128 129static inline void shmem_unacct_size(unsigned long flags, loff_t size) 130{ 131 if (!(flags & VM_NORESERVE)) 132 vm_unacct_memory(VM_ACCT(size)); 133} 134 135/* 136 * ... whereas tmpfs objects are accounted incrementally as 137 * pages are allocated, in order to allow huge sparse files. 138 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM, 139 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM. 140 */ 141static inline int shmem_acct_block(unsigned long flags) 142{ 143 return (flags & VM_NORESERVE) ? 144 security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE)) : 0; 145} 146 147static inline void shmem_unacct_blocks(unsigned long flags, long pages) 148{ 149 if (flags & VM_NORESERVE) 150 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE)); 151} 152 153static const struct super_operations shmem_ops; 154static const struct address_space_operations shmem_aops; 155static const struct file_operations shmem_file_operations; 156static const struct inode_operations shmem_inode_operations; 157static const struct inode_operations shmem_dir_inode_operations; 158static const struct inode_operations shmem_special_inode_operations; 159static const struct vm_operations_struct shmem_vm_ops; 160 161static struct backing_dev_info shmem_backing_dev_info __read_mostly = { 162 .ra_pages = 0, /* No readahead */ 163 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED, 164}; 165 166static LIST_HEAD(shmem_swaplist); 167static DEFINE_MUTEX(shmem_swaplist_mutex); 168 169static void shmem_free_blocks(struct inode *inode, long pages) 170{ 171 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb); 172 if (sbinfo->max_blocks) { 173 percpu_counter_add(&sbinfo->used_blocks, -pages); 174 inode->i_blocks -= pages*BLOCKS_PER_PAGE; 175 } 176} 177 178static int shmem_reserve_inode(struct super_block *sb) 179{ 180 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 181 if (sbinfo->max_inodes) { 182 spin_lock(&sbinfo->stat_lock); 183 if (!sbinfo->free_inodes) { 184 spin_unlock(&sbinfo->stat_lock); 185 return -ENOSPC; 186 } 187 sbinfo->free_inodes--; 188 spin_unlock(&sbinfo->stat_lock); 189 } 190 return 0; 191} 192 193static void shmem_free_inode(struct super_block *sb) 194{ 195 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 196 if (sbinfo->max_inodes) { 197 spin_lock(&sbinfo->stat_lock); 198 sbinfo->free_inodes++; 199 spin_unlock(&sbinfo->stat_lock); 200 } 201} 202 203/** 204 * shmem_recalc_inode - recalculate the block usage of an inode 205 * @inode: inode to recalc 206 * 207 * We have to calculate the free blocks since the mm can drop 208 * undirtied hole pages behind our back. 209 * 210 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped 211 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped) 212 * 213 * It has to be called with the spinlock held. 214 */ 215static void shmem_recalc_inode(struct inode *inode) 216{ 217 struct shmem_inode_info *info = SHMEM_I(inode); 218 long freed; 219 220 freed = info->alloced - info->swapped - inode->i_mapping->nrpages; 221 if (freed > 0) { 222 info->alloced -= freed; 223 shmem_unacct_blocks(info->flags, freed); 224 shmem_free_blocks(inode, freed); 225 } 226} 227 228static void shmem_put_swap(struct shmem_inode_info *info, pgoff_t index, 229 swp_entry_t swap) 230{ 231 if (index < SHMEM_NR_DIRECT) 232 info->i_direct[index] = swap; 233} 234 235static swp_entry_t shmem_get_swap(struct shmem_inode_info *info, pgoff_t index) 236{ 237 return (index < SHMEM_NR_DIRECT) ? 238 info->i_direct[index] : (swp_entry_t){0}; 239} 240 241void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) 242{ 243 struct address_space *mapping = inode->i_mapping; 244 struct shmem_inode_info *info = SHMEM_I(inode); 245 pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 246 unsigned partial = lstart & (PAGE_CACHE_SIZE - 1); 247 pgoff_t end = (lend >> PAGE_CACHE_SHIFT); 248 struct pagevec pvec; 249 pgoff_t index; 250 swp_entry_t swap; 251 int i; 252 253 BUG_ON((lend & (PAGE_CACHE_SIZE - 1)) != (PAGE_CACHE_SIZE - 1)); 254 255 pagevec_init(&pvec, 0); 256 index = start; 257 while (index <= end && pagevec_lookup(&pvec, mapping, index, 258 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) { 259 mem_cgroup_uncharge_start(); 260 for (i = 0; i < pagevec_count(&pvec); i++) { 261 struct page *page = pvec.pages[i]; 262 263 /* We rely upon deletion not changing page->index */ 264 index = page->index; 265 if (index > end) 266 break; 267 268 if (!trylock_page(page)) 269 continue; 270 WARN_ON(page->index != index); 271 if (PageWriteback(page)) { 272 unlock_page(page); 273 continue; 274 } 275 truncate_inode_page(mapping, page); 276 unlock_page(page); 277 } 278 pagevec_release(&pvec); 279 mem_cgroup_uncharge_end(); 280 cond_resched(); 281 index++; 282 } 283 284 if (partial) { 285 struct page *page = NULL; 286 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL); 287 if (page) { 288 zero_user_segment(page, partial, PAGE_CACHE_SIZE); 289 set_page_dirty(page); 290 unlock_page(page); 291 page_cache_release(page); 292 } 293 } 294 295 index = start; 296 for ( ; ; ) { 297 cond_resched(); 298 if (!pagevec_lookup(&pvec, mapping, index, 299 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) { 300 if (index == start) 301 break; 302 index = start; 303 continue; 304 } 305 if (index == start && pvec.pages[0]->index > end) { 306 pagevec_release(&pvec); 307 break; 308 } 309 mem_cgroup_uncharge_start(); 310 for (i = 0; i < pagevec_count(&pvec); i++) { 311 struct page *page = pvec.pages[i]; 312 313 /* We rely upon deletion not changing page->index */ 314 index = page->index; 315 if (index > end) 316 break; 317 318 lock_page(page); 319 WARN_ON(page->index != index); 320 wait_on_page_writeback(page); 321 truncate_inode_page(mapping, page); 322 unlock_page(page); 323 } 324 pagevec_release(&pvec); 325 mem_cgroup_uncharge_end(); 326 index++; 327 } 328 329 if (end > SHMEM_NR_DIRECT) 330 end = SHMEM_NR_DIRECT; 331 332 spin_lock(&info->lock); 333 for (index = start; index < end; index++) { 334 swap = shmem_get_swap(info, index); 335 if (swap.val) { 336 free_swap_and_cache(swap); 337 shmem_put_swap(info, index, (swp_entry_t){0}); 338 info->swapped--; 339 } 340 } 341 342 if (mapping->nrpages) { 343 spin_unlock(&info->lock); 344 /* 345 * A page may have meanwhile sneaked in from swap. 346 */ 347 truncate_inode_pages_range(mapping, lstart, lend); 348 spin_lock(&info->lock); 349 } 350 351 shmem_recalc_inode(inode); 352 spin_unlock(&info->lock); 353 354 inode->i_ctime = inode->i_mtime = CURRENT_TIME; 355} 356EXPORT_SYMBOL_GPL(shmem_truncate_range); 357 358static int shmem_setattr(struct dentry *dentry, struct iattr *attr) 359{ 360 struct inode *inode = dentry->d_inode; 361 int error; 362 363 error = inode_change_ok(inode, attr); 364 if (error) 365 return error; 366 367 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) { 368 loff_t oldsize = inode->i_size; 369 loff_t newsize = attr->ia_size; 370 371 if (newsize != oldsize) { 372 i_size_write(inode, newsize); 373 inode->i_ctime = inode->i_mtime = CURRENT_TIME; 374 } 375 if (newsize < oldsize) { 376 loff_t holebegin = round_up(newsize, PAGE_SIZE); 377 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1); 378 shmem_truncate_range(inode, newsize, (loff_t)-1); 379 /* unmap again to remove racily COWed private pages */ 380 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1); 381 } 382 } 383 384 setattr_copy(inode, attr); 385#ifdef CONFIG_TMPFS_POSIX_ACL 386 if (attr->ia_valid & ATTR_MODE) 387 error = generic_acl_chmod(inode); 388#endif 389 return error; 390} 391 392static void shmem_evict_inode(struct inode *inode) 393{ 394 struct shmem_inode_info *info = SHMEM_I(inode); 395 struct shmem_xattr *xattr, *nxattr; 396 397 if (inode->i_mapping->a_ops == &shmem_aops) { 398 shmem_unacct_size(info->flags, inode->i_size); 399 inode->i_size = 0; 400 shmem_truncate_range(inode, 0, (loff_t)-1); 401 if (!list_empty(&info->swaplist)) { 402 mutex_lock(&shmem_swaplist_mutex); 403 list_del_init(&info->swaplist); 404 mutex_unlock(&shmem_swaplist_mutex); 405 } 406 } 407 408 list_for_each_entry_safe(xattr, nxattr, &info->xattr_list, list) { 409 kfree(xattr->name); 410 kfree(xattr); 411 } 412 BUG_ON(inode->i_blocks); 413 shmem_free_inode(inode->i_sb); 414 end_writeback(inode); 415} 416 417static int shmem_unuse_inode(struct shmem_inode_info *info, 418 swp_entry_t swap, struct page *page) 419{ 420 struct address_space *mapping = info->vfs_inode.i_mapping; 421 pgoff_t index; 422 int error; 423 424 for (index = 0; index < SHMEM_NR_DIRECT; index++) 425 if (shmem_get_swap(info, index).val == swap.val) 426 goto found; 427 return 0; 428found: 429 spin_lock(&info->lock); 430 if (shmem_get_swap(info, index).val != swap.val) { 431 spin_unlock(&info->lock); 432 return 0; 433 } 434 435 /* 436 * Move _head_ to start search for next from here. 437 * But be careful: shmem_evict_inode checks list_empty without taking 438 * mutex, and there's an instant in list_move_tail when info->swaplist 439 * would appear empty, if it were the only one on shmem_swaplist. 440 */ 441 if (shmem_swaplist.next != &info->swaplist) 442 list_move_tail(&shmem_swaplist, &info->swaplist); 443 444 /* 445 * We rely on shmem_swaplist_mutex, not only to protect the swaplist, 446 * but also to hold up shmem_evict_inode(): so inode cannot be freed 447 * beneath us (pagelock doesn't help until the page is in pagecache). 448 */ 449 error = add_to_page_cache_locked(page, mapping, index, GFP_NOWAIT); 450 /* which does mem_cgroup_uncharge_cache_page on error */ 451 452 if (error != -ENOMEM) { 453 delete_from_swap_cache(page); 454 set_page_dirty(page); 455 shmem_put_swap(info, index, (swp_entry_t){0}); 456 info->swapped--; 457 swap_free(swap); 458 error = 1; /* not an error, but entry was found */ 459 } 460 spin_unlock(&info->lock); 461 return error; 462} 463 464/* 465 * shmem_unuse() search for an eventually swapped out shmem page. 466 */ 467int shmem_unuse(swp_entry_t swap, struct page *page) 468{ 469 struct list_head *this, *next; 470 struct shmem_inode_info *info; 471 int found = 0; 472 int error; 473 474 /* 475 * Charge page using GFP_KERNEL while we can wait, before taking 476 * the shmem_swaplist_mutex which might hold up shmem_writepage(). 477 * Charged back to the user (not to caller) when swap account is used. 478 * add_to_page_cache() will be called with GFP_NOWAIT. 479 */ 480 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL); 481 if (error) 482 goto out; 483 /* 484 * Try to preload while we can wait, to not make a habit of 485 * draining atomic reserves; but don't latch on to this cpu, 486 * it's okay if sometimes we get rescheduled after this. 487 */ 488 error = radix_tree_preload(GFP_KERNEL); 489 if (error) 490 goto uncharge; 491 radix_tree_preload_end(); 492 493 mutex_lock(&shmem_swaplist_mutex); 494 list_for_each_safe(this, next, &shmem_swaplist) { 495 info = list_entry(this, struct shmem_inode_info, swaplist); 496 if (!info->swapped) { 497 spin_lock(&info->lock); 498 if (!info->swapped) 499 list_del_init(&info->swaplist); 500 spin_unlock(&info->lock); 501 } 502 if (info->swapped) 503 found = shmem_unuse_inode(info, swap, page); 504 cond_resched(); 505 if (found) 506 break; 507 } 508 mutex_unlock(&shmem_swaplist_mutex); 509 510uncharge: 511 if (!found) 512 mem_cgroup_uncharge_cache_page(page); 513 if (found < 0) 514 error = found; 515out: 516 unlock_page(page); 517 page_cache_release(page); 518 return error; 519} 520 521/* 522 * Move the page from the page cache to the swap cache. 523 */ 524static int shmem_writepage(struct page *page, struct writeback_control *wbc) 525{ 526 struct shmem_inode_info *info; 527 swp_entry_t swap, oswap; 528 struct address_space *mapping; 529 pgoff_t index; 530 struct inode *inode; 531 532 BUG_ON(!PageLocked(page)); 533 mapping = page->mapping; 534 index = page->index; 535 inode = mapping->host; 536 info = SHMEM_I(inode); 537 if (info->flags & VM_LOCKED) 538 goto redirty; 539 if (!total_swap_pages) 540 goto redirty; 541 542 /* 543 * shmem_backing_dev_info's capabilities prevent regular writeback or 544 * sync from ever calling shmem_writepage; but a stacking filesystem 545 * might use ->writepage of its underlying filesystem, in which case 546 * tmpfs should write out to swap only in response to memory pressure, 547 * and not for the writeback threads or sync. 548 */ 549 if (!wbc->for_reclaim) { 550 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */ 551 goto redirty; 552 } 553 554 /* 555 * Just for this patch, we have a toy implementation, 556 * which can swap out only the first SHMEM_NR_DIRECT pages: 557 * for simple demonstration of where we need to think about swap. 558 */ 559 if (index >= SHMEM_NR_DIRECT) 560 goto redirty; 561 562 swap = get_swap_page(); 563 if (!swap.val) 564 goto redirty; 565 566 /* 567 * Add inode to shmem_unuse()'s list of swapped-out inodes, 568 * if it's not already there. Do it now because we cannot take 569 * mutex while holding spinlock, and must do so before the page 570 * is moved to swap cache, when its pagelock no longer protects 571 * the inode from eviction. But don't unlock the mutex until 572 * we've taken the spinlock, because shmem_unuse_inode() will 573 * prune a !swapped inode from the swaplist under both locks. 574 */ 575 mutex_lock(&shmem_swaplist_mutex); 576 if (list_empty(&info->swaplist)) 577 list_add_tail(&info->swaplist, &shmem_swaplist); 578 579 spin_lock(&info->lock); 580 mutex_unlock(&shmem_swaplist_mutex); 581 582 oswap = shmem_get_swap(info, index); 583 if (oswap.val) { 584 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */ 585 free_swap_and_cache(oswap); 586 shmem_put_swap(info, index, (swp_entry_t){0}); 587 info->swapped--; 588 } 589 shmem_recalc_inode(inode); 590 591 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) { 592 delete_from_page_cache(page); 593 shmem_put_swap(info, index, swap); 594 info->swapped++; 595 swap_shmem_alloc(swap); 596 spin_unlock(&info->lock); 597 BUG_ON(page_mapped(page)); 598 swap_writepage(page, wbc); 599 return 0; 600 } 601 602 spin_unlock(&info->lock); 603 swapcache_free(swap, NULL); 604redirty: 605 set_page_dirty(page); 606 if (wbc->for_reclaim) 607 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */ 608 unlock_page(page); 609 return 0; 610} 611 612#ifdef CONFIG_NUMA 613#ifdef CONFIG_TMPFS 614static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) 615{ 616 char buffer[64]; 617 618 if (!mpol || mpol->mode == MPOL_DEFAULT) 619 return; /* show nothing */ 620 621 mpol_to_str(buffer, sizeof(buffer), mpol, 1); 622 623 seq_printf(seq, ",mpol=%s", buffer); 624} 625 626static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) 627{ 628 struct mempolicy *mpol = NULL; 629 if (sbinfo->mpol) { 630 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */ 631 mpol = sbinfo->mpol; 632 mpol_get(mpol); 633 spin_unlock(&sbinfo->stat_lock); 634 } 635 return mpol; 636} 637#endif /* CONFIG_TMPFS */ 638 639static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, 640 struct shmem_inode_info *info, pgoff_t index) 641{ 642 struct mempolicy mpol, *spol; 643 struct vm_area_struct pvma; 644 645 spol = mpol_cond_copy(&mpol, 646 mpol_shared_policy_lookup(&info->policy, index)); 647 648 /* Create a pseudo vma that just contains the policy */ 649 pvma.vm_start = 0; 650 pvma.vm_pgoff = index; 651 pvma.vm_ops = NULL; 652 pvma.vm_policy = spol; 653 return swapin_readahead(swap, gfp, &pvma, 0); 654} 655 656static struct page *shmem_alloc_page(gfp_t gfp, 657 struct shmem_inode_info *info, pgoff_t index) 658{ 659 struct vm_area_struct pvma; 660 661 /* Create a pseudo vma that just contains the policy */ 662 pvma.vm_start = 0; 663 pvma.vm_pgoff = index; 664 pvma.vm_ops = NULL; 665 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index); 666 667 /* 668 * alloc_page_vma() will drop the shared policy reference 669 */ 670 return alloc_page_vma(gfp, &pvma, 0); 671} 672#else /* !CONFIG_NUMA */ 673#ifdef CONFIG_TMPFS 674static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol) 675{ 676} 677#endif /* CONFIG_TMPFS */ 678 679static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp, 680 struct shmem_inode_info *info, pgoff_t index) 681{ 682 return swapin_readahead(swap, gfp, NULL, 0); 683} 684 685static inline struct page *shmem_alloc_page(gfp_t gfp, 686 struct shmem_inode_info *info, pgoff_t index) 687{ 688 return alloc_page(gfp); 689} 690#endif /* CONFIG_NUMA */ 691 692#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS) 693static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo) 694{ 695 return NULL; 696} 697#endif 698 699/* 700 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate 701 * 702 * If we allocate a new one we do not mark it dirty. That's up to the 703 * vm. If we swap it in we mark it dirty since we also free the swap 704 * entry since a page cannot live in both the swap and page cache 705 */ 706static int shmem_getpage_gfp(struct inode *inode, pgoff_t index, 707 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type) 708{ 709 struct address_space *mapping = inode->i_mapping; 710 struct shmem_inode_info *info = SHMEM_I(inode); 711 struct shmem_sb_info *sbinfo; 712 struct page *page; 713 struct page *prealloc_page = NULL; 714 swp_entry_t swap; 715 int error; 716 717 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT)) 718 return -EFBIG; 719repeat: 720 page = find_lock_page(mapping, index); 721 if (page) { 722 /* 723 * Once we can get the page lock, it must be uptodate: 724 * if there were an error in reading back from swap, 725 * the page would not be inserted into the filecache. 726 */ 727 BUG_ON(!PageUptodate(page)); 728 goto done; 729 } 730 731 /* 732 * Try to preload while we can wait, to not make a habit of 733 * draining atomic reserves; but don't latch on to this cpu. 734 */ 735 error = radix_tree_preload(gfp & GFP_RECLAIM_MASK); 736 if (error) 737 goto out; 738 radix_tree_preload_end(); 739 740 if (sgp != SGP_READ && !prealloc_page) { 741 prealloc_page = shmem_alloc_page(gfp, info, index); 742 if (prealloc_page) { 743 SetPageSwapBacked(prealloc_page); 744 if (mem_cgroup_cache_charge(prealloc_page, 745 current->mm, GFP_KERNEL)) { 746 page_cache_release(prealloc_page); 747 prealloc_page = NULL; 748 } 749 } 750 } 751 752 spin_lock(&info->lock); 753 shmem_recalc_inode(inode); 754 swap = shmem_get_swap(info, index); 755 if (swap.val) { 756 /* Look it up and read it in.. */ 757 page = lookup_swap_cache(swap); 758 if (!page) { 759 spin_unlock(&info->lock); 760 /* here we actually do the io */ 761 if (fault_type) 762 *fault_type |= VM_FAULT_MAJOR; 763 page = shmem_swapin(swap, gfp, info, index); 764 if (!page) { 765 swp_entry_t nswap = shmem_get_swap(info, index); 766 if (nswap.val == swap.val) { 767 error = -ENOMEM; 768 goto out; 769 } 770 goto repeat; 771 } 772 wait_on_page_locked(page); 773 page_cache_release(page); 774 goto repeat; 775 } 776 777 /* We have to do this with page locked to prevent races */ 778 if (!trylock_page(page)) { 779 spin_unlock(&info->lock); 780 wait_on_page_locked(page); 781 page_cache_release(page); 782 goto repeat; 783 } 784 if (PageWriteback(page)) { 785 spin_unlock(&info->lock); 786 wait_on_page_writeback(page); 787 unlock_page(page); 788 page_cache_release(page); 789 goto repeat; 790 } 791 if (!PageUptodate(page)) { 792 spin_unlock(&info->lock); 793 unlock_page(page); 794 page_cache_release(page); 795 error = -EIO; 796 goto out; 797 } 798 799 error = add_to_page_cache_locked(page, mapping, 800 index, GFP_NOWAIT); 801 if (error) { 802 spin_unlock(&info->lock); 803 if (error == -ENOMEM) { 804 /* 805 * reclaim from proper memory cgroup and 806 * call memcg's OOM if needed. 807 */ 808 error = mem_cgroup_shmem_charge_fallback( 809 page, current->mm, gfp); 810 if (error) { 811 unlock_page(page); 812 page_cache_release(page); 813 goto out; 814 } 815 } 816 unlock_page(page); 817 page_cache_release(page); 818 goto repeat; 819 } 820 821 delete_from_swap_cache(page); 822 shmem_put_swap(info, index, (swp_entry_t){0}); 823 info->swapped--; 824 spin_unlock(&info->lock); 825 set_page_dirty(page); 826 swap_free(swap); 827 828 } else if (sgp == SGP_READ) { 829 page = find_get_page(mapping, index); 830 if (page && !trylock_page(page)) { 831 spin_unlock(&info->lock); 832 wait_on_page_locked(page); 833 page_cache_release(page); 834 goto repeat; 835 } 836 spin_unlock(&info->lock); 837 838 } else if (prealloc_page) { 839 sbinfo = SHMEM_SB(inode->i_sb); 840 if (sbinfo->max_blocks) { 841 if (percpu_counter_compare(&sbinfo->used_blocks, 842 sbinfo->max_blocks) >= 0 || 843 shmem_acct_block(info->flags)) 844 goto nospace; 845 percpu_counter_inc(&sbinfo->used_blocks); 846 inode->i_blocks += BLOCKS_PER_PAGE; 847 } else if (shmem_acct_block(info->flags)) 848 goto nospace; 849 850 page = prealloc_page; 851 prealloc_page = NULL; 852 853 swap = shmem_get_swap(info, index); 854 if (swap.val) 855 mem_cgroup_uncharge_cache_page(page); 856 else 857 error = add_to_page_cache_lru(page, mapping, 858 index, GFP_NOWAIT); 859 /* 860 * At add_to_page_cache_lru() failure, 861 * uncharge will be done automatically. 862 */ 863 if (swap.val || error) { 864 shmem_unacct_blocks(info->flags, 1); 865 shmem_free_blocks(inode, 1); 866 spin_unlock(&info->lock); 867 page_cache_release(page); 868 goto repeat; 869 } 870 871 info->alloced++; 872 spin_unlock(&info->lock); 873 clear_highpage(page); 874 flush_dcache_page(page); 875 SetPageUptodate(page); 876 if (sgp == SGP_DIRTY) 877 set_page_dirty(page); 878 879 } else { 880 spin_unlock(&info->lock); 881 error = -ENOMEM; 882 goto out; 883 } 884done: 885 *pagep = page; 886 error = 0; 887out: 888 if (prealloc_page) { 889 mem_cgroup_uncharge_cache_page(prealloc_page); 890 page_cache_release(prealloc_page); 891 } 892 return error; 893 894nospace: 895 /* 896 * Perhaps the page was brought in from swap between find_lock_page 897 * and taking info->lock? We allow for that at add_to_page_cache_lru, 898 * but must also avoid reporting a spurious ENOSPC while working on a 899 * full tmpfs. 900 */ 901 page = find_get_page(mapping, index); 902 spin_unlock(&info->lock); 903 if (page) { 904 page_cache_release(page); 905 goto repeat; 906 } 907 error = -ENOSPC; 908 goto out; 909} 910 911static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 912{ 913 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 914 int error; 915 int ret = VM_FAULT_LOCKED; 916 917 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 918 return VM_FAULT_SIGBUS; 919 920 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret); 921 if (error) 922 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS); 923 924 if (ret & VM_FAULT_MAJOR) { 925 count_vm_event(PGMAJFAULT); 926 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT); 927 } 928 return ret; 929} 930 931#ifdef CONFIG_NUMA 932static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol) 933{ 934 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 935 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol); 936} 937 938static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma, 939 unsigned long addr) 940{ 941 struct inode *inode = vma->vm_file->f_path.dentry->d_inode; 942 pgoff_t index; 943 944 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff; 945 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index); 946} 947#endif 948 949int shmem_lock(struct file *file, int lock, struct user_struct *user) 950{ 951 struct inode *inode = file->f_path.dentry->d_inode; 952 struct shmem_inode_info *info = SHMEM_I(inode); 953 int retval = -ENOMEM; 954 955 spin_lock(&info->lock); 956 if (lock && !(info->flags & VM_LOCKED)) { 957 if (!user_shm_lock(inode->i_size, user)) 958 goto out_nomem; 959 info->flags |= VM_LOCKED; 960 mapping_set_unevictable(file->f_mapping); 961 } 962 if (!lock && (info->flags & VM_LOCKED) && user) { 963 user_shm_unlock(inode->i_size, user); 964 info->flags &= ~VM_LOCKED; 965 mapping_clear_unevictable(file->f_mapping); 966 scan_mapping_unevictable_pages(file->f_mapping); 967 } 968 retval = 0; 969 970out_nomem: 971 spin_unlock(&info->lock); 972 return retval; 973} 974 975static int shmem_mmap(struct file *file, struct vm_area_struct *vma) 976{ 977 file_accessed(file); 978 vma->vm_ops = &shmem_vm_ops; 979 vma->vm_flags |= VM_CAN_NONLINEAR; 980 return 0; 981} 982 983static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir, 984 int mode, dev_t dev, unsigned long flags) 985{ 986 struct inode *inode; 987 struct shmem_inode_info *info; 988 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 989 990 if (shmem_reserve_inode(sb)) 991 return NULL; 992 993 inode = new_inode(sb); 994 if (inode) { 995 inode->i_ino = get_next_ino(); 996 inode_init_owner(inode, dir, mode); 997 inode->i_blocks = 0; 998 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info; 999 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; 1000 inode->i_generation = get_seconds(); 1001 info = SHMEM_I(inode); 1002 memset(info, 0, (char *)inode - (char *)info); 1003 spin_lock_init(&info->lock); 1004 info->flags = flags & VM_NORESERVE; 1005 INIT_LIST_HEAD(&info->swaplist); 1006 INIT_LIST_HEAD(&info->xattr_list); 1007 cache_no_acl(inode); 1008 1009 switch (mode & S_IFMT) { 1010 default: 1011 inode->i_op = &shmem_special_inode_operations; 1012 init_special_inode(inode, mode, dev); 1013 break; 1014 case S_IFREG: 1015 inode->i_mapping->a_ops = &shmem_aops; 1016 inode->i_op = &shmem_inode_operations; 1017 inode->i_fop = &shmem_file_operations; 1018 mpol_shared_policy_init(&info->policy, 1019 shmem_get_sbmpol(sbinfo)); 1020 break; 1021 case S_IFDIR: 1022 inc_nlink(inode); 1023 /* Some things misbehave if size == 0 on a directory */ 1024 inode->i_size = 2 * BOGO_DIRENT_SIZE; 1025 inode->i_op = &shmem_dir_inode_operations; 1026 inode->i_fop = &simple_dir_operations; 1027 break; 1028 case S_IFLNK: 1029 /* 1030 * Must not load anything in the rbtree, 1031 * mpol_free_shared_policy will not be called. 1032 */ 1033 mpol_shared_policy_init(&info->policy, NULL); 1034 break; 1035 } 1036 } else 1037 shmem_free_inode(sb); 1038 return inode; 1039} 1040 1041#ifdef CONFIG_TMPFS 1042static const struct inode_operations shmem_symlink_inode_operations; 1043static const struct inode_operations shmem_symlink_inline_operations; 1044 1045static int 1046shmem_write_begin(struct file *file, struct address_space *mapping, 1047 loff_t pos, unsigned len, unsigned flags, 1048 struct page **pagep, void **fsdata) 1049{ 1050 struct inode *inode = mapping->host; 1051 pgoff_t index = pos >> PAGE_CACHE_SHIFT; 1052 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL); 1053} 1054 1055static int 1056shmem_write_end(struct file *file, struct address_space *mapping, 1057 loff_t pos, unsigned len, unsigned copied, 1058 struct page *page, void *fsdata) 1059{ 1060 struct inode *inode = mapping->host; 1061 1062 if (pos + copied > inode->i_size) 1063 i_size_write(inode, pos + copied); 1064 1065 set_page_dirty(page); 1066 unlock_page(page); 1067 page_cache_release(page); 1068 1069 return copied; 1070} 1071 1072static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor) 1073{ 1074 struct inode *inode = filp->f_path.dentry->d_inode; 1075 struct address_space *mapping = inode->i_mapping; 1076 pgoff_t index; 1077 unsigned long offset; 1078 enum sgp_type sgp = SGP_READ; 1079 1080 /* 1081 * Might this read be for a stacking filesystem? Then when reading 1082 * holes of a sparse file, we actually need to allocate those pages, 1083 * and even mark them dirty, so it cannot exceed the max_blocks limit. 1084 */ 1085 if (segment_eq(get_fs(), KERNEL_DS)) 1086 sgp = SGP_DIRTY; 1087 1088 index = *ppos >> PAGE_CACHE_SHIFT; 1089 offset = *ppos & ~PAGE_CACHE_MASK; 1090 1091 for (;;) { 1092 struct page *page = NULL; 1093 pgoff_t end_index; 1094 unsigned long nr, ret; 1095 loff_t i_size = i_size_read(inode); 1096 1097 end_index = i_size >> PAGE_CACHE_SHIFT; 1098 if (index > end_index) 1099 break; 1100 if (index == end_index) { 1101 nr = i_size & ~PAGE_CACHE_MASK; 1102 if (nr <= offset) 1103 break; 1104 } 1105 1106 desc->error = shmem_getpage(inode, index, &page, sgp, NULL); 1107 if (desc->error) { 1108 if (desc->error == -EINVAL) 1109 desc->error = 0; 1110 break; 1111 } 1112 if (page) 1113 unlock_page(page); 1114 1115 /* 1116 * We must evaluate after, since reads (unlike writes) 1117 * are called without i_mutex protection against truncate 1118 */ 1119 nr = PAGE_CACHE_SIZE; 1120 i_size = i_size_read(inode); 1121 end_index = i_size >> PAGE_CACHE_SHIFT; 1122 if (index == end_index) { 1123 nr = i_size & ~PAGE_CACHE_MASK; 1124 if (nr <= offset) { 1125 if (page) 1126 page_cache_release(page); 1127 break; 1128 } 1129 } 1130 nr -= offset; 1131 1132 if (page) { 1133 /* 1134 * If users can be writing to this page using arbitrary 1135 * virtual addresses, take care about potential aliasing 1136 * before reading the page on the kernel side. 1137 */ 1138 if (mapping_writably_mapped(mapping)) 1139 flush_dcache_page(page); 1140 /* 1141 * Mark the page accessed if we read the beginning. 1142 */ 1143 if (!offset) 1144 mark_page_accessed(page); 1145 } else { 1146 page = ZERO_PAGE(0); 1147 page_cache_get(page); 1148 } 1149 1150 /* 1151 * Ok, we have the page, and it's up-to-date, so 1152 * now we can copy it to user space... 1153 * 1154 * The actor routine returns how many bytes were actually used.. 1155 * NOTE! This may not be the same as how much of a user buffer 1156 * we filled up (we may be padding etc), so we can only update 1157 * "pos" here (the actor routine has to update the user buffer 1158 * pointers and the remaining count). 1159 */ 1160 ret = actor(desc, page, offset, nr); 1161 offset += ret; 1162 index += offset >> PAGE_CACHE_SHIFT; 1163 offset &= ~PAGE_CACHE_MASK; 1164 1165 page_cache_release(page); 1166 if (ret != nr || !desc->count) 1167 break; 1168 1169 cond_resched(); 1170 } 1171 1172 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset; 1173 file_accessed(filp); 1174} 1175 1176static ssize_t shmem_file_aio_read(struct kiocb *iocb, 1177 const struct iovec *iov, unsigned long nr_segs, loff_t pos) 1178{ 1179 struct file *filp = iocb->ki_filp; 1180 ssize_t retval; 1181 unsigned long seg; 1182 size_t count; 1183 loff_t *ppos = &iocb->ki_pos; 1184 1185 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE); 1186 if (retval) 1187 return retval; 1188 1189 for (seg = 0; seg < nr_segs; seg++) { 1190 read_descriptor_t desc; 1191 1192 desc.written = 0; 1193 desc.arg.buf = iov[seg].iov_base; 1194 desc.count = iov[seg].iov_len; 1195 if (desc.count == 0) 1196 continue; 1197 desc.error = 0; 1198 do_shmem_file_read(filp, ppos, &desc, file_read_actor); 1199 retval += desc.written; 1200 if (desc.error) { 1201 retval = retval ?: desc.error; 1202 break; 1203 } 1204 if (desc.count > 0) 1205 break; 1206 } 1207 return retval; 1208} 1209 1210static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos, 1211 struct pipe_inode_info *pipe, size_t len, 1212 unsigned int flags) 1213{ 1214 struct address_space *mapping = in->f_mapping; 1215 struct inode *inode = mapping->host; 1216 unsigned int loff, nr_pages, req_pages; 1217 struct page *pages[PIPE_DEF_BUFFERS]; 1218 struct partial_page partial[PIPE_DEF_BUFFERS]; 1219 struct page *page; 1220 pgoff_t index, end_index; 1221 loff_t isize, left; 1222 int error, page_nr; 1223 struct splice_pipe_desc spd = { 1224 .pages = pages, 1225 .partial = partial, 1226 .flags = flags, 1227 .ops = &page_cache_pipe_buf_ops, 1228 .spd_release = spd_release_page, 1229 }; 1230 1231 isize = i_size_read(inode); 1232 if (unlikely(*ppos >= isize)) 1233 return 0; 1234 1235 left = isize - *ppos; 1236 if (unlikely(left < len)) 1237 len = left; 1238 1239 if (splice_grow_spd(pipe, &spd)) 1240 return -ENOMEM; 1241 1242 index = *ppos >> PAGE_CACHE_SHIFT; 1243 loff = *ppos & ~PAGE_CACHE_MASK; 1244 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT; 1245 nr_pages = min(req_pages, pipe->buffers); 1246 1247 spd.nr_pages = find_get_pages_contig(mapping, index, 1248 nr_pages, spd.pages); 1249 index += spd.nr_pages; 1250 error = 0; 1251 1252 while (spd.nr_pages < nr_pages) { 1253 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL); 1254 if (error) 1255 break; 1256 unlock_page(page); 1257 spd.pages[spd.nr_pages++] = page; 1258 index++; 1259 } 1260 1261 index = *ppos >> PAGE_CACHE_SHIFT; 1262 nr_pages = spd.nr_pages; 1263 spd.nr_pages = 0; 1264 1265 for (page_nr = 0; page_nr < nr_pages; page_nr++) { 1266 unsigned int this_len; 1267 1268 if (!len) 1269 break; 1270 1271 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff); 1272 page = spd.pages[page_nr]; 1273 1274 if (!PageUptodate(page) || page->mapping != mapping) { 1275 error = shmem_getpage(inode, index, &page, 1276 SGP_CACHE, NULL); 1277 if (error) 1278 break; 1279 unlock_page(page); 1280 page_cache_release(spd.pages[page_nr]); 1281 spd.pages[page_nr] = page; 1282 } 1283 1284 isize = i_size_read(inode); 1285 end_index = (isize - 1) >> PAGE_CACHE_SHIFT; 1286 if (unlikely(!isize || index > end_index)) 1287 break; 1288 1289 if (end_index == index) { 1290 unsigned int plen; 1291 1292 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1; 1293 if (plen <= loff) 1294 break; 1295 1296 this_len = min(this_len, plen - loff); 1297 len = this_len; 1298 } 1299 1300 spd.partial[page_nr].offset = loff; 1301 spd.partial[page_nr].len = this_len; 1302 len -= this_len; 1303 loff = 0; 1304 spd.nr_pages++; 1305 index++; 1306 } 1307 1308 while (page_nr < nr_pages) 1309 page_cache_release(spd.pages[page_nr++]); 1310 1311 if (spd.nr_pages) 1312 error = splice_to_pipe(pipe, &spd); 1313 1314 splice_shrink_spd(pipe, &spd); 1315 1316 if (error > 0) { 1317 *ppos += error; 1318 file_accessed(in); 1319 } 1320 return error; 1321} 1322 1323static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf) 1324{ 1325 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb); 1326 1327 buf->f_type = TMPFS_MAGIC; 1328 buf->f_bsize = PAGE_CACHE_SIZE; 1329 buf->f_namelen = NAME_MAX; 1330 if (sbinfo->max_blocks) { 1331 buf->f_blocks = sbinfo->max_blocks; 1332 buf->f_bavail = 1333 buf->f_bfree = sbinfo->max_blocks - 1334 percpu_counter_sum(&sbinfo->used_blocks); 1335 } 1336 if (sbinfo->max_inodes) { 1337 buf->f_files = sbinfo->max_inodes; 1338 buf->f_ffree = sbinfo->free_inodes; 1339 } 1340 /* else leave those fields 0 like simple_statfs */ 1341 return 0; 1342} 1343 1344/* 1345 * File creation. Allocate an inode, and we're done.. 1346 */ 1347static int 1348shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev) 1349{ 1350 struct inode *inode; 1351 int error = -ENOSPC; 1352 1353 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE); 1354 if (inode) { 1355 error = security_inode_init_security(inode, dir, 1356 &dentry->d_name, NULL, 1357 NULL, NULL); 1358 if (error) { 1359 if (error != -EOPNOTSUPP) { 1360 iput(inode); 1361 return error; 1362 } 1363 } 1364#ifdef CONFIG_TMPFS_POSIX_ACL 1365 error = generic_acl_init(inode, dir); 1366 if (error) { 1367 iput(inode); 1368 return error; 1369 } 1370#else 1371 error = 0; 1372#endif 1373 dir->i_size += BOGO_DIRENT_SIZE; 1374 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1375 d_instantiate(dentry, inode); 1376 dget(dentry); /* Extra count - pin the dentry in core */ 1377 } 1378 return error; 1379} 1380 1381static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode) 1382{ 1383 int error; 1384 1385 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0))) 1386 return error; 1387 inc_nlink(dir); 1388 return 0; 1389} 1390 1391static int shmem_create(struct inode *dir, struct dentry *dentry, int mode, 1392 struct nameidata *nd) 1393{ 1394 return shmem_mknod(dir, dentry, mode | S_IFREG, 0); 1395} 1396 1397/* 1398 * Link a file.. 1399 */ 1400static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry) 1401{ 1402 struct inode *inode = old_dentry->d_inode; 1403 int ret; 1404 1405 /* 1406 * No ordinary (disk based) filesystem counts links as inodes; 1407 * but each new link needs a new dentry, pinning lowmem, and 1408 * tmpfs dentries cannot be pruned until they are unlinked. 1409 */ 1410 ret = shmem_reserve_inode(inode->i_sb); 1411 if (ret) 1412 goto out; 1413 1414 dir->i_size += BOGO_DIRENT_SIZE; 1415 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1416 inc_nlink(inode); 1417 ihold(inode); /* New dentry reference */ 1418 dget(dentry); /* Extra pinning count for the created dentry */ 1419 d_instantiate(dentry, inode); 1420out: 1421 return ret; 1422} 1423 1424static int shmem_unlink(struct inode *dir, struct dentry *dentry) 1425{ 1426 struct inode *inode = dentry->d_inode; 1427 1428 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) 1429 shmem_free_inode(inode->i_sb); 1430 1431 dir->i_size -= BOGO_DIRENT_SIZE; 1432 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1433 drop_nlink(inode); 1434 dput(dentry); /* Undo the count from "create" - this does all the work */ 1435 return 0; 1436} 1437 1438static int shmem_rmdir(struct inode *dir, struct dentry *dentry) 1439{ 1440 if (!simple_empty(dentry)) 1441 return -ENOTEMPTY; 1442 1443 drop_nlink(dentry->d_inode); 1444 drop_nlink(dir); 1445 return shmem_unlink(dir, dentry); 1446} 1447 1448/* 1449 * The VFS layer already does all the dentry stuff for rename, 1450 * we just have to decrement the usage count for the target if 1451 * it exists so that the VFS layer correctly free's it when it 1452 * gets overwritten. 1453 */ 1454static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry) 1455{ 1456 struct inode *inode = old_dentry->d_inode; 1457 int they_are_dirs = S_ISDIR(inode->i_mode); 1458 1459 if (!simple_empty(new_dentry)) 1460 return -ENOTEMPTY; 1461 1462 if (new_dentry->d_inode) { 1463 (void) shmem_unlink(new_dir, new_dentry); 1464 if (they_are_dirs) 1465 drop_nlink(old_dir); 1466 } else if (they_are_dirs) { 1467 drop_nlink(old_dir); 1468 inc_nlink(new_dir); 1469 } 1470 1471 old_dir->i_size -= BOGO_DIRENT_SIZE; 1472 new_dir->i_size += BOGO_DIRENT_SIZE; 1473 old_dir->i_ctime = old_dir->i_mtime = 1474 new_dir->i_ctime = new_dir->i_mtime = 1475 inode->i_ctime = CURRENT_TIME; 1476 return 0; 1477} 1478 1479static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname) 1480{ 1481 int error; 1482 int len; 1483 struct inode *inode; 1484 struct page *page; 1485 char *kaddr; 1486 struct shmem_inode_info *info; 1487 1488 len = strlen(symname) + 1; 1489 if (len > PAGE_CACHE_SIZE) 1490 return -ENAMETOOLONG; 1491 1492 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE); 1493 if (!inode) 1494 return -ENOSPC; 1495 1496 error = security_inode_init_security(inode, dir, &dentry->d_name, NULL, 1497 NULL, NULL); 1498 if (error) { 1499 if (error != -EOPNOTSUPP) { 1500 iput(inode); 1501 return error; 1502 } 1503 error = 0; 1504 } 1505 1506 info = SHMEM_I(inode); 1507 inode->i_size = len-1; 1508 if (len <= SHMEM_SYMLINK_INLINE_LEN) { 1509 /* do it inline */ 1510 memcpy(info->inline_symlink, symname, len); 1511 inode->i_op = &shmem_symlink_inline_operations; 1512 } else { 1513 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL); 1514 if (error) { 1515 iput(inode); 1516 return error; 1517 } 1518 inode->i_mapping->a_ops = &shmem_aops; 1519 inode->i_op = &shmem_symlink_inode_operations; 1520 kaddr = kmap_atomic(page, KM_USER0); 1521 memcpy(kaddr, symname, len); 1522 kunmap_atomic(kaddr, KM_USER0); 1523 set_page_dirty(page); 1524 unlock_page(page); 1525 page_cache_release(page); 1526 } 1527 dir->i_size += BOGO_DIRENT_SIZE; 1528 dir->i_ctime = dir->i_mtime = CURRENT_TIME; 1529 d_instantiate(dentry, inode); 1530 dget(dentry); 1531 return 0; 1532} 1533 1534static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd) 1535{ 1536 nd_set_link(nd, SHMEM_I(dentry->d_inode)->inline_symlink); 1537 return NULL; 1538} 1539 1540static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd) 1541{ 1542 struct page *page = NULL; 1543 int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL); 1544 nd_set_link(nd, error ? ERR_PTR(error) : kmap(page)); 1545 if (page) 1546 unlock_page(page); 1547 return page; 1548} 1549 1550static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie) 1551{ 1552 if (!IS_ERR(nd_get_link(nd))) { 1553 struct page *page = cookie; 1554 kunmap(page); 1555 mark_page_accessed(page); 1556 page_cache_release(page); 1557 } 1558} 1559 1560#ifdef CONFIG_TMPFS_XATTR 1561/* 1562 * Superblocks without xattr inode operations may get some security.* xattr 1563 * support from the LSM "for free". As soon as we have any other xattrs 1564 * like ACLs, we also need to implement the security.* handlers at 1565 * filesystem level, though. 1566 */ 1567 1568static int shmem_xattr_get(struct dentry *dentry, const char *name, 1569 void *buffer, size_t size) 1570{ 1571 struct shmem_inode_info *info; 1572 struct shmem_xattr *xattr; 1573 int ret = -ENODATA; 1574 1575 info = SHMEM_I(dentry->d_inode); 1576 1577 spin_lock(&info->lock); 1578 list_for_each_entry(xattr, &info->xattr_list, list) { 1579 if (strcmp(name, xattr->name)) 1580 continue; 1581 1582 ret = xattr->size; 1583 if (buffer) { 1584 if (size < xattr->size) 1585 ret = -ERANGE; 1586 else 1587 memcpy(buffer, xattr->value, xattr->size); 1588 } 1589 break; 1590 } 1591 spin_unlock(&info->lock); 1592 return ret; 1593} 1594 1595static int shmem_xattr_set(struct dentry *dentry, const char *name, 1596 const void *value, size_t size, int flags) 1597{ 1598 struct inode *inode = dentry->d_inode; 1599 struct shmem_inode_info *info = SHMEM_I(inode); 1600 struct shmem_xattr *xattr; 1601 struct shmem_xattr *new_xattr = NULL; 1602 size_t len; 1603 int err = 0; 1604 1605 /* value == NULL means remove */ 1606 if (value) { 1607 /* wrap around? */ 1608 len = sizeof(*new_xattr) + size; 1609 if (len <= sizeof(*new_xattr)) 1610 return -ENOMEM; 1611 1612 new_xattr = kmalloc(len, GFP_KERNEL); 1613 if (!new_xattr) 1614 return -ENOMEM; 1615 1616 new_xattr->name = kstrdup(name, GFP_KERNEL); 1617 if (!new_xattr->name) { 1618 kfree(new_xattr); 1619 return -ENOMEM; 1620 } 1621 1622 new_xattr->size = size; 1623 memcpy(new_xattr->value, value, size); 1624 } 1625 1626 spin_lock(&info->lock); 1627 list_for_each_entry(xattr, &info->xattr_list, list) { 1628 if (!strcmp(name, xattr->name)) { 1629 if (flags & XATTR_CREATE) { 1630 xattr = new_xattr; 1631 err = -EEXIST; 1632 } else if (new_xattr) { 1633 list_replace(&xattr->list, &new_xattr->list); 1634 } else { 1635 list_del(&xattr->list); 1636 } 1637 goto out; 1638 } 1639 } 1640 if (flags & XATTR_REPLACE) { 1641 xattr = new_xattr; 1642 err = -ENODATA; 1643 } else { 1644 list_add(&new_xattr->list, &info->xattr_list); 1645 xattr = NULL; 1646 } 1647out: 1648 spin_unlock(&info->lock); 1649 if (xattr) 1650 kfree(xattr->name); 1651 kfree(xattr); 1652 return err; 1653} 1654 1655static const struct xattr_handler *shmem_xattr_handlers[] = { 1656#ifdef CONFIG_TMPFS_POSIX_ACL 1657 &generic_acl_access_handler, 1658 &generic_acl_default_handler, 1659#endif 1660 NULL 1661}; 1662 1663static int shmem_xattr_validate(const char *name) 1664{ 1665 struct { const char *prefix; size_t len; } arr[] = { 1666 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN }, 1667 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN } 1668 }; 1669 int i; 1670 1671 for (i = 0; i < ARRAY_SIZE(arr); i++) { 1672 size_t preflen = arr[i].len; 1673 if (strncmp(name, arr[i].prefix, preflen) == 0) { 1674 if (!name[preflen]) 1675 return -EINVAL; 1676 return 0; 1677 } 1678 } 1679 return -EOPNOTSUPP; 1680} 1681 1682static ssize_t shmem_getxattr(struct dentry *dentry, const char *name, 1683 void *buffer, size_t size) 1684{ 1685 int err; 1686 1687 /* 1688 * If this is a request for a synthetic attribute in the system.* 1689 * namespace use the generic infrastructure to resolve a handler 1690 * for it via sb->s_xattr. 1691 */ 1692 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN)) 1693 return generic_getxattr(dentry, name, buffer, size); 1694 1695 err = shmem_xattr_validate(name); 1696 if (err) 1697 return err; 1698 1699 return shmem_xattr_get(dentry, name, buffer, size); 1700} 1701 1702static int shmem_setxattr(struct dentry *dentry, const char *name, 1703 const void *value, size_t size, int flags) 1704{ 1705 int err; 1706 1707 /* 1708 * If this is a request for a synthetic attribute in the system.* 1709 * namespace use the generic infrastructure to resolve a handler 1710 * for it via sb->s_xattr. 1711 */ 1712 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN)) 1713 return generic_setxattr(dentry, name, value, size, flags); 1714 1715 err = shmem_xattr_validate(name); 1716 if (err) 1717 return err; 1718 1719 if (size == 0) 1720 value = ""; /* empty EA, do not remove */ 1721 1722 return shmem_xattr_set(dentry, name, value, size, flags); 1723 1724} 1725 1726static int shmem_removexattr(struct dentry *dentry, const char *name) 1727{ 1728 int err; 1729 1730 /* 1731 * If this is a request for a synthetic attribute in the system.* 1732 * namespace use the generic infrastructure to resolve a handler 1733 * for it via sb->s_xattr. 1734 */ 1735 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN)) 1736 return generic_removexattr(dentry, name); 1737 1738 err = shmem_xattr_validate(name); 1739 if (err) 1740 return err; 1741 1742 return shmem_xattr_set(dentry, name, NULL, 0, XATTR_REPLACE); 1743} 1744 1745static bool xattr_is_trusted(const char *name) 1746{ 1747 return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN); 1748} 1749 1750static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size) 1751{ 1752 bool trusted = capable(CAP_SYS_ADMIN); 1753 struct shmem_xattr *xattr; 1754 struct shmem_inode_info *info; 1755 size_t used = 0; 1756 1757 info = SHMEM_I(dentry->d_inode); 1758 1759 spin_lock(&info->lock); 1760 list_for_each_entry(xattr, &info->xattr_list, list) { 1761 size_t len; 1762 1763 /* skip "trusted." attributes for unprivileged callers */ 1764 if (!trusted && xattr_is_trusted(xattr->name)) 1765 continue; 1766 1767 len = strlen(xattr->name) + 1; 1768 used += len; 1769 if (buffer) { 1770 if (size < used) { 1771 used = -ERANGE; 1772 break; 1773 } 1774 memcpy(buffer, xattr->name, len); 1775 buffer += len; 1776 } 1777 } 1778 spin_unlock(&info->lock); 1779 1780 return used; 1781} 1782#endif /* CONFIG_TMPFS_XATTR */ 1783 1784static const struct inode_operations shmem_symlink_inline_operations = { 1785 .readlink = generic_readlink, 1786 .follow_link = shmem_follow_link_inline, 1787#ifdef CONFIG_TMPFS_XATTR 1788 .setxattr = shmem_setxattr, 1789 .getxattr = shmem_getxattr, 1790 .listxattr = shmem_listxattr, 1791 .removexattr = shmem_removexattr, 1792#endif 1793}; 1794 1795static const struct inode_operations shmem_symlink_inode_operations = { 1796 .readlink = generic_readlink, 1797 .follow_link = shmem_follow_link, 1798 .put_link = shmem_put_link, 1799#ifdef CONFIG_TMPFS_XATTR 1800 .setxattr = shmem_setxattr, 1801 .getxattr = shmem_getxattr, 1802 .listxattr = shmem_listxattr, 1803 .removexattr = shmem_removexattr, 1804#endif 1805}; 1806 1807static struct dentry *shmem_get_parent(struct dentry *child) 1808{ 1809 return ERR_PTR(-ESTALE); 1810} 1811 1812static int shmem_match(struct inode *ino, void *vfh) 1813{ 1814 __u32 *fh = vfh; 1815 __u64 inum = fh[2]; 1816 inum = (inum << 32) | fh[1]; 1817 return ino->i_ino == inum && fh[0] == ino->i_generation; 1818} 1819 1820static struct dentry *shmem_fh_to_dentry(struct super_block *sb, 1821 struct fid *fid, int fh_len, int fh_type) 1822{ 1823 struct inode *inode; 1824 struct dentry *dentry = NULL; 1825 u64 inum = fid->raw[2]; 1826 inum = (inum << 32) | fid->raw[1]; 1827 1828 if (fh_len < 3) 1829 return NULL; 1830 1831 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]), 1832 shmem_match, fid->raw); 1833 if (inode) { 1834 dentry = d_find_alias(inode); 1835 iput(inode); 1836 } 1837 1838 return dentry; 1839} 1840 1841static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len, 1842 int connectable) 1843{ 1844 struct inode *inode = dentry->d_inode; 1845 1846 if (*len < 3) { 1847 *len = 3; 1848 return 255; 1849 } 1850 1851 if (inode_unhashed(inode)) { 1852 /* Unfortunately insert_inode_hash is not idempotent, 1853 * so as we hash inodes here rather than at creation 1854 * time, we need a lock to ensure we only try 1855 * to do it once 1856 */ 1857 static DEFINE_SPINLOCK(lock); 1858 spin_lock(&lock); 1859 if (inode_unhashed(inode)) 1860 __insert_inode_hash(inode, 1861 inode->i_ino + inode->i_generation); 1862 spin_unlock(&lock); 1863 } 1864 1865 fh[0] = inode->i_generation; 1866 fh[1] = inode->i_ino; 1867 fh[2] = ((__u64)inode->i_ino) >> 32; 1868 1869 *len = 3; 1870 return 1; 1871} 1872 1873static const struct export_operations shmem_export_ops = { 1874 .get_parent = shmem_get_parent, 1875 .encode_fh = shmem_encode_fh, 1876 .fh_to_dentry = shmem_fh_to_dentry, 1877}; 1878 1879static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo, 1880 bool remount) 1881{ 1882 char *this_char, *value, *rest; 1883 1884 while (options != NULL) { 1885 this_char = options; 1886 for (;;) { 1887 /* 1888 * NUL-terminate this option: unfortunately, 1889 * mount options form a comma-separated list, 1890 * but mpol's nodelist may also contain commas. 1891 */ 1892 options = strchr(options, ','); 1893 if (options == NULL) 1894 break; 1895 options++; 1896 if (!isdigit(*options)) { 1897 options[-1] = '\0'; 1898 break; 1899 } 1900 } 1901 if (!*this_char) 1902 continue; 1903 if ((value = strchr(this_char,'=')) != NULL) { 1904 *value++ = 0; 1905 } else { 1906 printk(KERN_ERR 1907 "tmpfs: No value for mount option '%s'\n", 1908 this_char); 1909 return 1; 1910 } 1911 1912 if (!strcmp(this_char,"size")) { 1913 unsigned long long size; 1914 size = memparse(value,&rest); 1915 if (*rest == '%') { 1916 size <<= PAGE_SHIFT; 1917 size *= totalram_pages; 1918 do_div(size, 100); 1919 rest++; 1920 } 1921 if (*rest) 1922 goto bad_val; 1923 sbinfo->max_blocks = 1924 DIV_ROUND_UP(size, PAGE_CACHE_SIZE); 1925 } else if (!strcmp(this_char,"nr_blocks")) { 1926 sbinfo->max_blocks = memparse(value, &rest); 1927 if (*rest) 1928 goto bad_val; 1929 } else if (!strcmp(this_char,"nr_inodes")) { 1930 sbinfo->max_inodes = memparse(value, &rest); 1931 if (*rest) 1932 goto bad_val; 1933 } else if (!strcmp(this_char,"mode")) { 1934 if (remount) 1935 continue; 1936 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777; 1937 if (*rest) 1938 goto bad_val; 1939 } else if (!strcmp(this_char,"uid")) { 1940 if (remount) 1941 continue; 1942 sbinfo->uid = simple_strtoul(value, &rest, 0); 1943 if (*rest) 1944 goto bad_val; 1945 } else if (!strcmp(this_char,"gid")) { 1946 if (remount) 1947 continue; 1948 sbinfo->gid = simple_strtoul(value, &rest, 0); 1949 if (*rest) 1950 goto bad_val; 1951 } else if (!strcmp(this_char,"mpol")) { 1952 if (mpol_parse_str(value, &sbinfo->mpol, 1)) 1953 goto bad_val; 1954 } else { 1955 printk(KERN_ERR "tmpfs: Bad mount option %s\n", 1956 this_char); 1957 return 1; 1958 } 1959 } 1960 return 0; 1961 1962bad_val: 1963 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n", 1964 value, this_char); 1965 return 1; 1966 1967} 1968 1969static int shmem_remount_fs(struct super_block *sb, int *flags, char *data) 1970{ 1971 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 1972 struct shmem_sb_info config = *sbinfo; 1973 unsigned long inodes; 1974 int error = -EINVAL; 1975 1976 if (shmem_parse_options(data, &config, true)) 1977 return error; 1978 1979 spin_lock(&sbinfo->stat_lock); 1980 inodes = sbinfo->max_inodes - sbinfo->free_inodes; 1981 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0) 1982 goto out; 1983 if (config.max_inodes < inodes) 1984 goto out; 1985 /* 1986 * Those tests also disallow limited->unlimited while any are in 1987 * use, so i_blocks will always be zero when max_blocks is zero; 1988 * but we must separately disallow unlimited->limited, because 1989 * in that case we have no record of how much is already in use. 1990 */ 1991 if (config.max_blocks && !sbinfo->max_blocks) 1992 goto out; 1993 if (config.max_inodes && !sbinfo->max_inodes) 1994 goto out; 1995 1996 error = 0; 1997 sbinfo->max_blocks = config.max_blocks; 1998 sbinfo->max_inodes = config.max_inodes; 1999 sbinfo->free_inodes = config.max_inodes - inodes; 2000 2001 mpol_put(sbinfo->mpol); 2002 sbinfo->mpol = config.mpol; /* transfers initial ref */ 2003out: 2004 spin_unlock(&sbinfo->stat_lock); 2005 return error; 2006} 2007 2008static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs) 2009{ 2010 struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb); 2011 2012 if (sbinfo->max_blocks != shmem_default_max_blocks()) 2013 seq_printf(seq, ",size=%luk", 2014 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10)); 2015 if (sbinfo->max_inodes != shmem_default_max_inodes()) 2016 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes); 2017 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX)) 2018 seq_printf(seq, ",mode=%03o", sbinfo->mode); 2019 if (sbinfo->uid != 0) 2020 seq_printf(seq, ",uid=%u", sbinfo->uid); 2021 if (sbinfo->gid != 0) 2022 seq_printf(seq, ",gid=%u", sbinfo->gid); 2023 shmem_show_mpol(seq, sbinfo->mpol); 2024 return 0; 2025} 2026#endif /* CONFIG_TMPFS */ 2027 2028static void shmem_put_super(struct super_block *sb) 2029{ 2030 struct shmem_sb_info *sbinfo = SHMEM_SB(sb); 2031 2032 percpu_counter_destroy(&sbinfo->used_blocks); 2033 kfree(sbinfo); 2034 sb->s_fs_info = NULL; 2035} 2036 2037int shmem_fill_super(struct super_block *sb, void *data, int silent) 2038{ 2039 struct inode *inode; 2040 struct dentry *root; 2041 struct shmem_sb_info *sbinfo; 2042 int err = -ENOMEM; 2043 2044 /* Round up to L1_CACHE_BYTES to resist false sharing */ 2045 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info), 2046 L1_CACHE_BYTES), GFP_KERNEL); 2047 if (!sbinfo) 2048 return -ENOMEM; 2049 2050 sbinfo->mode = S_IRWXUGO | S_ISVTX; 2051 sbinfo->uid = current_fsuid(); 2052 sbinfo->gid = current_fsgid(); 2053 sb->s_fs_info = sbinfo; 2054 2055#ifdef CONFIG_TMPFS 2056 /* 2057 * Per default we only allow half of the physical ram per 2058 * tmpfs instance, limiting inodes to one per page of lowmem; 2059 * but the internal instance is left unlimited. 2060 */ 2061 if (!(sb->s_flags & MS_NOUSER)) { 2062 sbinfo->max_blocks = shmem_default_max_blocks(); 2063 sbinfo->max_inodes = shmem_default_max_inodes(); 2064 if (shmem_parse_options(data, sbinfo, false)) { 2065 err = -EINVAL; 2066 goto failed; 2067 } 2068 } 2069 sb->s_export_op = &shmem_export_ops; 2070#else 2071 sb->s_flags |= MS_NOUSER; 2072#endif 2073 2074 spin_lock_init(&sbinfo->stat_lock); 2075 if (percpu_counter_init(&sbinfo->used_blocks, 0)) 2076 goto failed; 2077 sbinfo->free_inodes = sbinfo->max_inodes; 2078 2079 sb->s_maxbytes = MAX_LFS_FILESIZE; 2080 sb->s_blocksize = PAGE_CACHE_SIZE; 2081 sb->s_blocksize_bits = PAGE_CACHE_SHIFT; 2082 sb->s_magic = TMPFS_MAGIC; 2083 sb->s_op = &shmem_ops; 2084 sb->s_time_gran = 1; 2085#ifdef CONFIG_TMPFS_XATTR 2086 sb->s_xattr = shmem_xattr_handlers; 2087#endif 2088#ifdef CONFIG_TMPFS_POSIX_ACL 2089 sb->s_flags |= MS_POSIXACL; 2090#endif 2091 2092 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE); 2093 if (!inode) 2094 goto failed; 2095 inode->i_uid = sbinfo->uid; 2096 inode->i_gid = sbinfo->gid; 2097 root = d_alloc_root(inode); 2098 if (!root) 2099 goto failed_iput; 2100 sb->s_root = root; 2101 return 0; 2102 2103failed_iput: 2104 iput(inode); 2105failed: 2106 shmem_put_super(sb); 2107 return err; 2108} 2109 2110static struct kmem_cache *shmem_inode_cachep; 2111 2112static struct inode *shmem_alloc_inode(struct super_block *sb) 2113{ 2114 struct shmem_inode_info *info; 2115 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL); 2116 if (!info) 2117 return NULL; 2118 return &info->vfs_inode; 2119} 2120 2121static void shmem_destroy_callback(struct rcu_head *head) 2122{ 2123 struct inode *inode = container_of(head, struct inode, i_rcu); 2124 INIT_LIST_HEAD(&inode->i_dentry); 2125 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode)); 2126} 2127 2128static void shmem_destroy_inode(struct inode *inode) 2129{ 2130 if ((inode->i_mode & S_IFMT) == S_IFREG) { 2131 /* only struct inode is valid if it's an inline symlink */ 2132 mpol_free_shared_policy(&SHMEM_I(inode)->policy); 2133 } 2134 call_rcu(&inode->i_rcu, shmem_destroy_callback); 2135} 2136 2137static void shmem_init_inode(void *foo) 2138{ 2139 struct shmem_inode_info *info = foo; 2140 inode_init_once(&info->vfs_inode); 2141} 2142 2143static int shmem_init_inodecache(void) 2144{ 2145 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache", 2146 sizeof(struct shmem_inode_info), 2147 0, SLAB_PANIC, shmem_init_inode); 2148 return 0; 2149} 2150 2151static void shmem_destroy_inodecache(void) 2152{ 2153 kmem_cache_destroy(shmem_inode_cachep); 2154} 2155 2156static const struct address_space_operations shmem_aops = { 2157 .writepage = shmem_writepage, 2158 .set_page_dirty = __set_page_dirty_no_writeback, 2159#ifdef CONFIG_TMPFS 2160 .write_begin = shmem_write_begin, 2161 .write_end = shmem_write_end, 2162#endif 2163 .migratepage = migrate_page, 2164 .error_remove_page = generic_error_remove_page, 2165}; 2166 2167static const struct file_operations shmem_file_operations = { 2168 .mmap = shmem_mmap, 2169#ifdef CONFIG_TMPFS 2170 .llseek = generic_file_llseek, 2171 .read = do_sync_read, 2172 .write = do_sync_write, 2173 .aio_read = shmem_file_aio_read, 2174 .aio_write = generic_file_aio_write, 2175 .fsync = noop_fsync, 2176 .splice_read = shmem_file_splice_read, 2177 .splice_write = generic_file_splice_write, 2178#endif 2179}; 2180 2181static const struct inode_operations shmem_inode_operations = { 2182 .setattr = shmem_setattr, 2183 .truncate_range = shmem_truncate_range, 2184#ifdef CONFIG_TMPFS_XATTR 2185 .setxattr = shmem_setxattr, 2186 .getxattr = shmem_getxattr, 2187 .listxattr = shmem_listxattr, 2188 .removexattr = shmem_removexattr, 2189#endif 2190}; 2191 2192static const 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#ifdef CONFIG_TMPFS_XATTR 2205 .setxattr = shmem_setxattr, 2206 .getxattr = shmem_getxattr, 2207 .listxattr = shmem_listxattr, 2208 .removexattr = shmem_removexattr, 2209#endif 2210#ifdef CONFIG_TMPFS_POSIX_ACL 2211 .setattr = shmem_setattr, 2212#endif 2213}; 2214 2215static const struct inode_operations shmem_special_inode_operations = { 2216#ifdef CONFIG_TMPFS_XATTR 2217 .setxattr = shmem_setxattr, 2218 .getxattr = shmem_getxattr, 2219 .listxattr = shmem_listxattr, 2220 .removexattr = shmem_removexattr, 2221#endif 2222#ifdef CONFIG_TMPFS_POSIX_ACL 2223 .setattr = shmem_setattr, 2224#endif 2225}; 2226 2227static const struct super_operations shmem_ops = { 2228 .alloc_inode = shmem_alloc_inode, 2229 .destroy_inode = shmem_destroy_inode, 2230#ifdef CONFIG_TMPFS 2231 .statfs = shmem_statfs, 2232 .remount_fs = shmem_remount_fs, 2233 .show_options = shmem_show_options, 2234#endif 2235 .evict_inode = shmem_evict_inode, 2236 .drop_inode = generic_delete_inode, 2237 .put_super = shmem_put_super, 2238}; 2239 2240static const struct vm_operations_struct shmem_vm_ops = { 2241 .fault = shmem_fault, 2242#ifdef CONFIG_NUMA 2243 .set_policy = shmem_set_policy, 2244 .get_policy = shmem_get_policy, 2245#endif 2246}; 2247 2248static struct dentry *shmem_mount(struct file_system_type *fs_type, 2249 int flags, const char *dev_name, void *data) 2250{ 2251 return mount_nodev(fs_type, flags, data, shmem_fill_super); 2252} 2253 2254static struct file_system_type shmem_fs_type = { 2255 .owner = THIS_MODULE, 2256 .name = "tmpfs", 2257 .mount = shmem_mount, 2258 .kill_sb = kill_litter_super, 2259}; 2260 2261int __init shmem_init(void) 2262{ 2263 int error; 2264 2265 error = bdi_init(&shmem_backing_dev_info); 2266 if (error) 2267 goto out4; 2268 2269 error = shmem_init_inodecache(); 2270 if (error) 2271 goto out3; 2272 2273 error = register_filesystem(&shmem_fs_type); 2274 if (error) { 2275 printk(KERN_ERR "Could not register tmpfs\n"); 2276 goto out2; 2277 } 2278 2279 shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER, 2280 shmem_fs_type.name, NULL); 2281 if (IS_ERR(shm_mnt)) { 2282 error = PTR_ERR(shm_mnt); 2283 printk(KERN_ERR "Could not kern_mount tmpfs\n"); 2284 goto out1; 2285 } 2286 return 0; 2287 2288out1: 2289 unregister_filesystem(&shmem_fs_type); 2290out2: 2291 shmem_destroy_inodecache(); 2292out3: 2293 bdi_destroy(&shmem_backing_dev_info); 2294out4: 2295 shm_mnt = ERR_PTR(error); 2296 return error; 2297} 2298 2299#ifdef CONFIG_CGROUP_MEM_RES_CTLR 2300/** 2301 * mem_cgroup_get_shmem_target - find page or swap assigned to the shmem file 2302 * @inode: the inode to be searched 2303 * @index: the page offset to be searched 2304 * @pagep: the pointer for the found page to be stored 2305 * @swapp: the pointer for the found swap entry to be stored 2306 * 2307 * If a page is found, refcount of it is incremented. Callers should handle 2308 * these refcount. 2309 */ 2310void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t index, 2311 struct page **pagep, swp_entry_t *swapp) 2312{ 2313 struct shmem_inode_info *info = SHMEM_I(inode); 2314 struct page *page = NULL; 2315 swp_entry_t swap = {0}; 2316 2317 if ((index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 2318 goto out; 2319 2320 spin_lock(&info->lock); 2321#ifdef CONFIG_SWAP 2322 swap = shmem_get_swap(info, index); 2323 if (swap.val) 2324 page = find_get_page(&swapper_space, swap.val); 2325 else 2326#endif 2327 page = find_get_page(inode->i_mapping, index); 2328 spin_unlock(&info->lock); 2329out: 2330 *pagep = page; 2331 *swapp = swap; 2332} 2333#endif 2334 2335#else /* !CONFIG_SHMEM */ 2336 2337/* 2338 * tiny-shmem: simple shmemfs and tmpfs using ramfs code 2339 * 2340 * This is intended for small system where the benefits of the full 2341 * shmem code (swap-backed and resource-limited) are outweighed by 2342 * their complexity. On systems without swap this code should be 2343 * effectively equivalent, but much lighter weight. 2344 */ 2345 2346#include <linux/ramfs.h> 2347 2348static struct file_system_type shmem_fs_type = { 2349 .name = "tmpfs", 2350 .mount = ramfs_mount, 2351 .kill_sb = kill_litter_super, 2352}; 2353 2354int __init shmem_init(void) 2355{ 2356 BUG_ON(register_filesystem(&shmem_fs_type) != 0); 2357 2358 shm_mnt = kern_mount(&shmem_fs_type); 2359 BUG_ON(IS_ERR(shm_mnt)); 2360 2361 return 0; 2362} 2363 2364int shmem_unuse(swp_entry_t swap, struct page *page) 2365{ 2366 return 0; 2367} 2368 2369int shmem_lock(struct file *file, int lock, struct user_struct *user) 2370{ 2371 return 0; 2372} 2373 2374void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend) 2375{ 2376 truncate_inode_pages_range(inode->i_mapping, lstart, lend); 2377} 2378EXPORT_SYMBOL_GPL(shmem_truncate_range); 2379 2380#ifdef CONFIG_CGROUP_MEM_RES_CTLR 2381/** 2382 * mem_cgroup_get_shmem_target - find page or swap assigned to the shmem file 2383 * @inode: the inode to be searched 2384 * @index: the page offset to be searched 2385 * @pagep: the pointer for the found page to be stored 2386 * @swapp: the pointer for the found swap entry to be stored 2387 * 2388 * If a page is found, refcount of it is incremented. Callers should handle 2389 * these refcount. 2390 */ 2391void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t index, 2392 struct page **pagep, swp_entry_t *swapp) 2393{ 2394 struct page *page = NULL; 2395 2396 if ((index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) 2397 goto out; 2398 page = find_get_page(inode->i_mapping, index); 2399out: 2400 *pagep = page; 2401 *swapp = (swp_entry_t){0}; 2402} 2403#endif 2404 2405#define shmem_vm_ops generic_file_vm_ops 2406#define shmem_file_operations ramfs_file_operations 2407#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev) 2408#define shmem_acct_size(flags, size) 0 2409#define shmem_unacct_size(flags, size) do {} while (0) 2410 2411#endif /* CONFIG_SHMEM */ 2412 2413/* common code */ 2414 2415/** 2416 * shmem_file_setup - get an unlinked file living in tmpfs 2417 * @name: name for dentry (to be seen in /proc/<pid>/maps 2418 * @size: size to be set for the file 2419 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size 2420 */ 2421struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags) 2422{ 2423 int error; 2424 struct file *file; 2425 struct inode *inode; 2426 struct path path; 2427 struct dentry *root; 2428 struct qstr this; 2429 2430 if (IS_ERR(shm_mnt)) 2431 return (void *)shm_mnt; 2432 2433 if (size < 0 || size > MAX_LFS_FILESIZE) 2434 return ERR_PTR(-EINVAL); 2435 2436 if (shmem_acct_size(flags, size)) 2437 return ERR_PTR(-ENOMEM); 2438 2439 error = -ENOMEM; 2440 this.name = name; 2441 this.len = strlen(name); 2442 this.hash = 0; /* will go */ 2443 root = shm_mnt->mnt_root; 2444 path.dentry = d_alloc(root, &this); 2445 if (!path.dentry) 2446 goto put_memory; 2447 path.mnt = mntget(shm_mnt); 2448 2449 error = -ENOSPC; 2450 inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags); 2451 if (!inode) 2452 goto put_dentry; 2453 2454 d_instantiate(path.dentry, inode); 2455 inode->i_size = size; 2456 inode->i_nlink = 0; /* It is unlinked */ 2457#ifndef CONFIG_MMU 2458 error = ramfs_nommu_expand_for_mapping(inode, size); 2459 if (error) 2460 goto put_dentry; 2461#endif 2462 2463 error = -ENFILE; 2464 file = alloc_file(&path, FMODE_WRITE | FMODE_READ, 2465 &shmem_file_operations); 2466 if (!file) 2467 goto put_dentry; 2468 2469 return file; 2470 2471put_dentry: 2472 path_put(&path); 2473put_memory: 2474 shmem_unacct_size(flags, size); 2475 return ERR_PTR(error); 2476} 2477EXPORT_SYMBOL_GPL(shmem_file_setup); 2478 2479/** 2480 * shmem_zero_setup - setup a shared anonymous mapping 2481 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff 2482 */ 2483int shmem_zero_setup(struct vm_area_struct *vma) 2484{ 2485 struct file *file; 2486 loff_t size = vma->vm_end - vma->vm_start; 2487 2488 file = shmem_file_setup("dev/zero", size, vma->vm_flags); 2489 if (IS_ERR(file)) 2490 return PTR_ERR(file); 2491 2492 if (vma->vm_file) 2493 fput(vma->vm_file); 2494 vma->vm_file = file; 2495 vma->vm_ops = &shmem_vm_ops; 2496 vma->vm_flags |= VM_CAN_NONLINEAR; 2497 return 0; 2498} 2499 2500/** 2501 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags. 2502 * @mapping: the page's address_space 2503 * @index: the page index 2504 * @gfp: the page allocator flags to use if allocating 2505 * 2506 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)", 2507 * with any new page allocations done using the specified allocation flags. 2508 * But read_cache_page_gfp() uses the ->readpage() method: which does not 2509 * suit tmpfs, since it may have pages in swapcache, and needs to find those 2510 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support. 2511 * 2512 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in 2513 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily. 2514 */ 2515struct page *shmem_read_mapping_page_gfp(struct address_space *mapping, 2516 pgoff_t index, gfp_t gfp) 2517{ 2518#ifdef CONFIG_SHMEM 2519 struct inode *inode = mapping->host; 2520 struct page *page; 2521 int error; 2522 2523 BUG_ON(mapping->a_ops != &shmem_aops); 2524 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL); 2525 if (error) 2526 page = ERR_PTR(error); 2527 else 2528 unlock_page(page); 2529 return page; 2530#else 2531 /* 2532 * The tiny !SHMEM case uses ramfs without swap 2533 */ 2534 return read_cache_page_gfp(mapping, index, gfp); 2535#endif 2536} 2537EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp); 2538