blob: a0126c4371054edf7120aaea97fd4d2785e4fdbf [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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.
Hugh Dickins0edd73b2005-06-21 17:15:04 -07009 * Copyright (C) 2002-2005 Hugh Dickins.
10 * Copyright (C) 2002-2005 VERITAS Software Corporation.
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * 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
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/module.h>
27#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/fs.h>
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070029#include <linux/xattr.h>
Christoph Hellwiga5694252007-07-17 04:04:28 -070030#include <linux/exportfs.h>
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070031#include <linux/generic_acl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#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>
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +000049#include <linux/ctype.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070050#include <linux/migrate.h>
Christoph Lameterc1f60a52006-09-25 23:31:11 -070051#include <linux/highmem.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070052
Linus Torvalds1da177e2005-04-16 15:20:36 -070053#include <asm/uaccess.h>
54#include <asm/div64.h>
55#include <asm/pgtable.h>
56
57/* This magic number is used in glibc for posix shared memory */
58#define TMPFS_MAGIC 0x01021994
59
60#define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
61#define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
62#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
63
64#define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
65#define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
66
67#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
68
69/* info->flags needs VM_flags to handle pagein/truncate races efficiently */
70#define SHMEM_PAGEIN VM_READ
71#define SHMEM_TRUNCATE VM_WRITE
72
73/* Definition to limit shmem_truncate's steps between cond_rescheds */
74#define LATENCY_LIMIT 64
75
76/* Pretend that each entry is of this size in directory's i_size */
77#define BOGO_DIRENT_SIZE 20
78
Linus Torvalds1da177e2005-04-16 15:20:36 -070079/* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
80enum sgp_type {
Linus Torvalds1da177e2005-04-16 15:20:36 -070081 SGP_READ, /* don't exceed i_size, don't allocate page */
82 SGP_CACHE, /* don't exceed i_size, may allocate page */
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -080083 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
Linus Torvalds1da177e2005-04-16 15:20:36 -070084 SGP_WRITE, /* may exceed i_size, may allocate page */
85};
86
87static int shmem_getpage(struct inode *inode, unsigned long idx,
88 struct page **pagep, enum sgp_type sgp, int *type);
89
Al Viro6daa0e22005-10-21 03:18:50 -040090static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -070091{
92 /*
93 * The above definition of ENTRIES_PER_PAGE, and the use of
94 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
95 * might be reconsidered if it ever diverges from PAGE_SIZE.
Mel Gorman769848c2007-07-17 04:03:05 -070096 *
Mel Gormane12ba742007-10-16 01:25:52 -070097 * Mobility flags are masked out as swap vectors cannot move
Linus Torvalds1da177e2005-04-16 15:20:36 -070098 */
Mel Gormane12ba742007-10-16 01:25:52 -070099 return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO,
Mel Gorman769848c2007-07-17 04:03:05 -0700100 PAGE_CACHE_SHIFT-PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101}
102
103static inline void shmem_dir_free(struct page *page)
104{
105 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
106}
107
108static struct page **shmem_dir_map(struct page *page)
109{
110 return (struct page **)kmap_atomic(page, KM_USER0);
111}
112
113static inline void shmem_dir_unmap(struct page **dir)
114{
115 kunmap_atomic(dir, KM_USER0);
116}
117
118static swp_entry_t *shmem_swp_map(struct page *page)
119{
120 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
121}
122
123static inline void shmem_swp_balance_unmap(void)
124{
125 /*
126 * When passing a pointer to an i_direct entry, to code which
127 * also handles indirect entries and so will shmem_swp_unmap,
128 * we must arrange for the preempt count to remain in balance.
129 * What kmap_atomic of a lowmem page does depends on config
130 * and architecture, so pretend to kmap_atomic some lowmem page.
131 */
132 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
133}
134
135static inline void shmem_swp_unmap(swp_entry_t *entry)
136{
137 kunmap_atomic(entry, KM_USER1);
138}
139
140static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
141{
142 return sb->s_fs_info;
143}
144
145/*
146 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
147 * for shared memory and for shared anonymous (/dev/zero) mappings
148 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
149 * consistent with the pre-accounting of private mappings ...
150 */
151static inline int shmem_acct_size(unsigned long flags, loff_t size)
152{
153 return (flags & VM_ACCOUNT)?
154 security_vm_enough_memory(VM_ACCT(size)): 0;
155}
156
157static inline void shmem_unacct_size(unsigned long flags, loff_t size)
158{
159 if (flags & VM_ACCOUNT)
160 vm_unacct_memory(VM_ACCT(size));
161}
162
163/*
164 * ... whereas tmpfs objects are accounted incrementally as
165 * pages are allocated, in order to allow huge sparse files.
166 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
167 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
168 */
169static inline int shmem_acct_block(unsigned long flags)
170{
171 return (flags & VM_ACCOUNT)?
172 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
173}
174
175static inline void shmem_unacct_blocks(unsigned long flags, long pages)
176{
177 if (!(flags & VM_ACCOUNT))
178 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
179}
180
Hugh Dickins759b9772007-03-05 00:30:28 -0800181static const struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700182static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800183static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800184static const struct inode_operations shmem_inode_operations;
185static const struct inode_operations shmem_dir_inode_operations;
186static const struct inode_operations shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187static struct vm_operations_struct shmem_vm_ops;
188
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700189static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 .ra_pages = 0, /* No readahead */
191 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
192 .unplug_io_fn = default_unplug_io_fn,
193};
194
195static LIST_HEAD(shmem_swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800196static DEFINE_MUTEX(shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700197
198static void shmem_free_blocks(struct inode *inode, long pages)
199{
200 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700201 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202 spin_lock(&sbinfo->stat_lock);
203 sbinfo->free_blocks += pages;
204 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
205 spin_unlock(&sbinfo->stat_lock);
206 }
207}
208
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800209static int shmem_reserve_inode(struct super_block *sb)
210{
211 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
212 if (sbinfo->max_inodes) {
213 spin_lock(&sbinfo->stat_lock);
214 if (!sbinfo->free_inodes) {
215 spin_unlock(&sbinfo->stat_lock);
216 return -ENOSPC;
217 }
218 sbinfo->free_inodes--;
219 spin_unlock(&sbinfo->stat_lock);
220 }
221 return 0;
222}
223
224static void shmem_free_inode(struct super_block *sb)
225{
226 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
227 if (sbinfo->max_inodes) {
228 spin_lock(&sbinfo->stat_lock);
229 sbinfo->free_inodes++;
230 spin_unlock(&sbinfo->stat_lock);
231 }
232}
233
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234/*
235 * shmem_recalc_inode - recalculate the size of an inode
236 *
237 * @inode: inode to recalc
238 *
239 * We have to calculate the free blocks since the mm can drop
240 * undirtied hole pages behind our back.
241 *
242 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
243 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
244 *
245 * It has to be called with the spinlock held.
246 */
247static void shmem_recalc_inode(struct inode *inode)
248{
249 struct shmem_inode_info *info = SHMEM_I(inode);
250 long freed;
251
252 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
253 if (freed > 0) {
254 info->alloced -= freed;
255 shmem_unacct_blocks(info->flags, freed);
256 shmem_free_blocks(inode, freed);
257 }
258}
259
260/*
261 * shmem_swp_entry - find the swap vector position in the info structure
262 *
263 * @info: info structure for the inode
264 * @index: index of the page to find
265 * @page: optional page to add to the structure. Has to be preset to
266 * all zeros
267 *
268 * If there is no space allocated yet it will return NULL when
269 * page is NULL, else it will use the page for the needed block,
270 * setting it to NULL on return to indicate that it has been used.
271 *
272 * The swap vector is organized the following way:
273 *
274 * There are SHMEM_NR_DIRECT entries directly stored in the
275 * shmem_inode_info structure. So small files do not need an addional
276 * allocation.
277 *
278 * For pages with index > SHMEM_NR_DIRECT there is the pointer
279 * i_indirect which points to a page which holds in the first half
280 * doubly indirect blocks, in the second half triple indirect blocks:
281 *
282 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
283 * following layout (for SHMEM_NR_DIRECT == 16):
284 *
285 * i_indirect -> dir --> 16-19
286 * | +-> 20-23
287 * |
288 * +-->dir2 --> 24-27
289 * | +-> 28-31
290 * | +-> 32-35
291 * | +-> 36-39
292 * |
293 * +-->dir3 --> 40-43
294 * +-> 44-47
295 * +-> 48-51
296 * +-> 52-55
297 */
298static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
299{
300 unsigned long offset;
301 struct page **dir;
302 struct page *subdir;
303
304 if (index < SHMEM_NR_DIRECT) {
305 shmem_swp_balance_unmap();
306 return info->i_direct+index;
307 }
308 if (!info->i_indirect) {
309 if (page) {
310 info->i_indirect = *page;
311 *page = NULL;
312 }
313 return NULL; /* need another page */
314 }
315
316 index -= SHMEM_NR_DIRECT;
317 offset = index % ENTRIES_PER_PAGE;
318 index /= ENTRIES_PER_PAGE;
319 dir = shmem_dir_map(info->i_indirect);
320
321 if (index >= ENTRIES_PER_PAGE/2) {
322 index -= ENTRIES_PER_PAGE/2;
323 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
324 index %= ENTRIES_PER_PAGE;
325 subdir = *dir;
326 if (!subdir) {
327 if (page) {
328 *dir = *page;
329 *page = NULL;
330 }
331 shmem_dir_unmap(dir);
332 return NULL; /* need another page */
333 }
334 shmem_dir_unmap(dir);
335 dir = shmem_dir_map(subdir);
336 }
337
338 dir += index;
339 subdir = *dir;
340 if (!subdir) {
341 if (!page || !(subdir = *page)) {
342 shmem_dir_unmap(dir);
343 return NULL; /* need a page */
344 }
345 *dir = subdir;
346 *page = NULL;
347 }
348 shmem_dir_unmap(dir);
349 return shmem_swp_map(subdir) + offset;
350}
351
352static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
353{
354 long incdec = value? 1: -1;
355
356 entry->val = value;
357 info->swapped += incdec;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700358 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
359 struct page *page = kmap_atomic_to_page(entry);
360 set_page_private(page, page_private(page) + incdec);
361 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700362}
363
364/*
365 * shmem_swp_alloc - get the position of the swap entry for the page.
366 * If it does not exist allocate the entry.
367 *
368 * @info: info structure for the inode
369 * @index: index of the page to find
370 * @sgp: check and recheck i_size? skip allocation?
371 */
372static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
373{
374 struct inode *inode = &info->vfs_inode;
375 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
376 struct page *page = NULL;
377 swp_entry_t *entry;
378
379 if (sgp != SGP_WRITE &&
380 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
381 return ERR_PTR(-EINVAL);
382
383 while (!(entry = shmem_swp_entry(info, index, &page))) {
384 if (sgp == SGP_READ)
385 return shmem_swp_map(ZERO_PAGE(0));
386 /*
387 * Test free_blocks against 1 not 0, since we have 1 data
388 * page (and perhaps indirect index pages) yet to allocate:
389 * a waste to allocate index if we cannot allocate data.
390 */
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700391 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700392 spin_lock(&sbinfo->stat_lock);
393 if (sbinfo->free_blocks <= 1) {
394 spin_unlock(&sbinfo->stat_lock);
395 return ERR_PTR(-ENOSPC);
396 }
397 sbinfo->free_blocks--;
398 inode->i_blocks += BLOCKS_PER_PAGE;
399 spin_unlock(&sbinfo->stat_lock);
400 }
401
402 spin_unlock(&info->lock);
Mel Gorman769848c2007-07-17 04:03:05 -0700403 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700404 if (page)
405 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 spin_lock(&info->lock);
407
408 if (!page) {
409 shmem_free_blocks(inode, 1);
410 return ERR_PTR(-ENOMEM);
411 }
412 if (sgp != SGP_WRITE &&
413 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
414 entry = ERR_PTR(-EINVAL);
415 break;
416 }
417 if (info->next_index <= index)
418 info->next_index = index + 1;
419 }
420 if (page) {
421 /* another task gave its page, or truncated the file */
422 shmem_free_blocks(inode, 1);
423 shmem_dir_free(page);
424 }
425 if (info->next_index <= index && !IS_ERR(entry))
426 info->next_index = index + 1;
427 return entry;
428}
429
430/*
431 * shmem_free_swp - free some swap entries in a directory
432 *
Hugh Dickins1ae70002007-03-29 01:20:36 -0700433 * @dir: pointer to the directory
434 * @edir: pointer after last entry of the directory
435 * @punch_lock: pointer to spinlock when needed for the holepunch case
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 */
Hugh Dickins1ae70002007-03-29 01:20:36 -0700437static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
438 spinlock_t *punch_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439{
Hugh Dickins1ae70002007-03-29 01:20:36 -0700440 spinlock_t *punch_unlock = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 swp_entry_t *ptr;
442 int freed = 0;
443
444 for (ptr = dir; ptr < edir; ptr++) {
445 if (ptr->val) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700446 if (unlikely(punch_lock)) {
447 punch_unlock = punch_lock;
448 punch_lock = NULL;
449 spin_lock(punch_unlock);
450 if (!ptr->val)
451 continue;
452 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453 free_swap_and_cache(*ptr);
454 *ptr = (swp_entry_t){0};
455 freed++;
456 }
457 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700458 if (punch_unlock)
459 spin_unlock(punch_unlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 return freed;
461}
462
Hugh Dickins1ae70002007-03-29 01:20:36 -0700463static int shmem_map_and_free_swp(struct page *subdir, int offset,
464 int limit, struct page ***dir, spinlock_t *punch_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465{
466 swp_entry_t *ptr;
467 int freed = 0;
468
469 ptr = shmem_swp_map(subdir);
470 for (; offset < limit; offset += LATENCY_LIMIT) {
471 int size = limit - offset;
472 if (size > LATENCY_LIMIT)
473 size = LATENCY_LIMIT;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700474 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
475 punch_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476 if (need_resched()) {
477 shmem_swp_unmap(ptr);
478 if (*dir) {
479 shmem_dir_unmap(*dir);
480 *dir = NULL;
481 }
482 cond_resched();
483 ptr = shmem_swp_map(subdir);
484 }
485 }
486 shmem_swp_unmap(ptr);
487 return freed;
488}
489
490static void shmem_free_pages(struct list_head *next)
491{
492 struct page *page;
493 int freed = 0;
494
495 do {
496 page = container_of(next, struct page, lru);
497 next = next->next;
498 shmem_dir_free(page);
499 freed++;
500 if (freed >= LATENCY_LIMIT) {
501 cond_resched();
502 freed = 0;
503 }
504 } while (next);
505}
506
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800507static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508{
509 struct shmem_inode_info *info = SHMEM_I(inode);
510 unsigned long idx;
511 unsigned long size;
512 unsigned long limit;
513 unsigned long stage;
514 unsigned long diroff;
515 struct page **dir;
516 struct page *topdir;
517 struct page *middir;
518 struct page *subdir;
519 swp_entry_t *ptr;
520 LIST_HEAD(pages_to_free);
521 long nr_pages_to_free = 0;
522 long nr_swaps_freed = 0;
523 int offset;
524 int freed;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700525 int punch_hole;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700526 spinlock_t *needs_lock;
527 spinlock_t *punch_lock;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700528 unsigned long upper_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700529
530 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800531 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 if (idx >= info->next_index)
533 return;
534
535 spin_lock(&info->lock);
536 info->flags |= SHMEM_TRUNCATE;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800537 if (likely(end == (loff_t) -1)) {
538 limit = info->next_index;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700539 upper_limit = SHMEM_MAX_INDEX;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800540 info->next_index = idx;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700541 needs_lock = NULL;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700542 punch_hole = 0;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800543 } else {
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700544 if (end + 1 >= inode->i_size) { /* we may free a little more */
545 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
546 PAGE_CACHE_SHIFT;
547 upper_limit = SHMEM_MAX_INDEX;
548 } else {
549 limit = (end + 1) >> PAGE_CACHE_SHIFT;
550 upper_limit = limit;
551 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700552 needs_lock = &info->lock;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800553 punch_hole = 1;
554 }
555
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556 topdir = info->i_indirect;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800557 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558 info->i_indirect = NULL;
559 nr_pages_to_free++;
560 list_add(&topdir->lru, &pages_to_free);
561 }
562 spin_unlock(&info->lock);
563
564 if (info->swapped && idx < SHMEM_NR_DIRECT) {
565 ptr = info->i_direct;
566 size = limit;
567 if (size > SHMEM_NR_DIRECT)
568 size = SHMEM_NR_DIRECT;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700569 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 }
Badari Pulavarty92a3d032006-12-22 01:06:23 -0800571
572 /*
573 * If there are no indirect blocks or we are punching a hole
574 * below indirect blocks, nothing to be done.
575 */
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700576 if (!topdir || limit <= SHMEM_NR_DIRECT)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 goto done2;
578
Hugh Dickins1ae70002007-03-29 01:20:36 -0700579 /*
580 * The truncation case has already dropped info->lock, and we're safe
581 * because i_size and next_index have already been lowered, preventing
582 * access beyond. But in the punch_hole case, we still need to take
583 * the lock when updating the swap directory, because there might be
584 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
585 * shmem_writepage. However, whenever we find we can remove a whole
586 * directory page (not at the misaligned start or end of the range),
587 * we first NULLify its pointer in the level above, and then have no
588 * need to take the lock when updating its contents: needs_lock and
589 * punch_lock (either pointing to info->lock or NULL) manage this.
590 */
591
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700592 upper_limit -= SHMEM_NR_DIRECT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 limit -= SHMEM_NR_DIRECT;
594 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
595 offset = idx % ENTRIES_PER_PAGE;
596 idx -= offset;
597
598 dir = shmem_dir_map(topdir);
599 stage = ENTRIES_PER_PAGEPAGE/2;
600 if (idx < ENTRIES_PER_PAGEPAGE/2) {
601 middir = topdir;
602 diroff = idx/ENTRIES_PER_PAGE;
603 } else {
604 dir += ENTRIES_PER_PAGE/2;
605 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
606 while (stage <= idx)
607 stage += ENTRIES_PER_PAGEPAGE;
608 middir = *dir;
609 if (*dir) {
610 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
611 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700612 if (!diroff && !offset && upper_limit >= stage) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700613 if (needs_lock) {
614 spin_lock(needs_lock);
615 *dir = NULL;
616 spin_unlock(needs_lock);
617 needs_lock = NULL;
618 } else
619 *dir = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 nr_pages_to_free++;
621 list_add(&middir->lru, &pages_to_free);
622 }
623 shmem_dir_unmap(dir);
624 dir = shmem_dir_map(middir);
625 } else {
626 diroff = 0;
627 offset = 0;
628 idx = stage;
629 }
630 }
631
632 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
633 if (unlikely(idx == stage)) {
634 shmem_dir_unmap(dir);
635 dir = shmem_dir_map(topdir) +
636 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
637 while (!*dir) {
638 dir++;
639 idx += ENTRIES_PER_PAGEPAGE;
640 if (idx >= limit)
641 goto done1;
642 }
643 stage = idx + ENTRIES_PER_PAGEPAGE;
644 middir = *dir;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700645 if (punch_hole)
646 needs_lock = &info->lock;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700647 if (upper_limit >= stage) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700648 if (needs_lock) {
649 spin_lock(needs_lock);
650 *dir = NULL;
651 spin_unlock(needs_lock);
652 needs_lock = NULL;
653 } else
654 *dir = NULL;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700655 nr_pages_to_free++;
656 list_add(&middir->lru, &pages_to_free);
657 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 shmem_dir_unmap(dir);
659 cond_resched();
660 dir = shmem_dir_map(middir);
661 diroff = 0;
662 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700663 punch_lock = needs_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 subdir = dir[diroff];
Hugh Dickins1ae70002007-03-29 01:20:36 -0700665 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
666 if (needs_lock) {
667 spin_lock(needs_lock);
668 dir[diroff] = NULL;
669 spin_unlock(needs_lock);
670 punch_lock = NULL;
671 } else
672 dir[diroff] = NULL;
673 nr_pages_to_free++;
674 list_add(&subdir->lru, &pages_to_free);
675 }
676 if (subdir && page_private(subdir) /* has swap entries */) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677 size = limit - idx;
678 if (size > ENTRIES_PER_PAGE)
679 size = ENTRIES_PER_PAGE;
680 freed = shmem_map_and_free_swp(subdir,
Hugh Dickins1ae70002007-03-29 01:20:36 -0700681 offset, size, &dir, punch_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 if (!dir)
683 dir = shmem_dir_map(middir);
684 nr_swaps_freed += freed;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700685 if (offset || punch_lock) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 spin_lock(&info->lock);
Hugh Dickins1ae70002007-03-29 01:20:36 -0700687 set_page_private(subdir,
688 page_private(subdir) - freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 spin_unlock(&info->lock);
Hugh Dickins1ae70002007-03-29 01:20:36 -0700690 } else
691 BUG_ON(page_private(subdir) != freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700693 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 }
695done1:
696 shmem_dir_unmap(dir);
697done2:
698 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
699 /*
700 * Call truncate_inode_pages again: racing shmem_unuse_inode
701 * may have swizzled a page in from swap since vmtruncate or
702 * generic_delete_inode did it, before we lowered next_index.
703 * Also, though shmem_getpage checks i_size before adding to
704 * cache, no recheck after: so fix the narrow window there too.
Hugh Dickins16a10012007-03-29 01:20:37 -0700705 *
706 * Recalling truncate_inode_pages_range and unmap_mapping_range
707 * every time for punch_hole (which never got a chance to clear
708 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
709 * yet hardly ever necessary: try to optimize them out later.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 */
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800711 truncate_inode_pages_range(inode->i_mapping, start, end);
Hugh Dickins16a10012007-03-29 01:20:37 -0700712 if (punch_hole)
713 unmap_mapping_range(inode->i_mapping, start,
714 end - start, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715 }
716
717 spin_lock(&info->lock);
718 info->flags &= ~SHMEM_TRUNCATE;
719 info->swapped -= nr_swaps_freed;
720 if (nr_pages_to_free)
721 shmem_free_blocks(inode, nr_pages_to_free);
722 shmem_recalc_inode(inode);
723 spin_unlock(&info->lock);
724
725 /*
726 * Empty swap vector directory pages to be freed?
727 */
728 if (!list_empty(&pages_to_free)) {
729 pages_to_free.prev->next = NULL;
730 shmem_free_pages(pages_to_free.next);
731 }
732}
733
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800734static void shmem_truncate(struct inode *inode)
735{
736 shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
737}
738
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
740{
741 struct inode *inode = dentry->d_inode;
742 struct page *page = NULL;
743 int error;
744
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700745 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 if (attr->ia_size < inode->i_size) {
747 /*
748 * If truncating down to a partial page, then
749 * if that page is already allocated, hold it
750 * in memory until the truncation is over, so
751 * truncate_partial_page cannnot miss it were
752 * it assigned to swap.
753 */
754 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
755 (void) shmem_getpage(inode,
756 attr->ia_size>>PAGE_CACHE_SHIFT,
757 &page, SGP_READ, NULL);
Hugh Dickinsd3602442008-02-04 22:28:44 -0800758 if (page)
759 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 }
761 /*
762 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
763 * detect if any pages might have been added to cache
764 * after truncate_inode_pages. But we needn't bother
765 * if it's being fully truncated to zero-length: the
766 * nrpages check is efficient enough in that case.
767 */
768 if (attr->ia_size) {
769 struct shmem_inode_info *info = SHMEM_I(inode);
770 spin_lock(&info->lock);
771 info->flags &= ~SHMEM_PAGEIN;
772 spin_unlock(&info->lock);
773 }
774 }
775 }
776
777 error = inode_change_ok(inode, attr);
778 if (!error)
779 error = inode_setattr(inode, attr);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700780#ifdef CONFIG_TMPFS_POSIX_ACL
781 if (!error && (attr->ia_valid & ATTR_MODE))
782 error = generic_acl_chmod(inode, &shmem_acl_ops);
783#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 if (page)
785 page_cache_release(page);
786 return error;
787}
788
789static void shmem_delete_inode(struct inode *inode)
790{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 struct shmem_inode_info *info = SHMEM_I(inode);
792
793 if (inode->i_op->truncate == shmem_truncate) {
Mark Fashehfef26652005-09-09 13:01:31 -0700794 truncate_inode_pages(inode->i_mapping, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 shmem_unacct_size(info->flags, inode->i_size);
796 inode->i_size = 0;
797 shmem_truncate(inode);
798 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800799 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700800 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800801 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700802 }
803 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700804 BUG_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800805 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700806 clear_inode(inode);
807}
808
809static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
810{
811 swp_entry_t *ptr;
812
813 for (ptr = dir; ptr < edir; ptr++) {
814 if (ptr->val == entry.val)
815 return ptr - dir;
816 }
817 return -1;
818}
819
820static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
821{
822 struct inode *inode;
823 unsigned long idx;
824 unsigned long size;
825 unsigned long limit;
826 unsigned long stage;
827 struct page **dir;
828 struct page *subdir;
829 swp_entry_t *ptr;
830 int offset;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800831 int error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832
833 idx = 0;
834 ptr = info->i_direct;
835 spin_lock(&info->lock);
836 limit = info->next_index;
837 size = limit;
838 if (size > SHMEM_NR_DIRECT)
839 size = SHMEM_NR_DIRECT;
840 offset = shmem_find_swp(entry, ptr, ptr+size);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800841 if (offset >= 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842 goto found;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 if (!info->i_indirect)
844 goto lost2;
845
846 dir = shmem_dir_map(info->i_indirect);
847 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
848
849 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
850 if (unlikely(idx == stage)) {
851 shmem_dir_unmap(dir-1);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800852 if (cond_resched_lock(&info->lock)) {
853 /* check it has not been truncated */
854 if (limit > info->next_index) {
855 limit = info->next_index;
856 if (idx >= limit)
857 goto lost2;
858 }
859 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 dir = shmem_dir_map(info->i_indirect) +
861 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
862 while (!*dir) {
863 dir++;
864 idx += ENTRIES_PER_PAGEPAGE;
865 if (idx >= limit)
866 goto lost1;
867 }
868 stage = idx + ENTRIES_PER_PAGEPAGE;
869 subdir = *dir;
870 shmem_dir_unmap(dir);
871 dir = shmem_dir_map(subdir);
872 }
873 subdir = *dir;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700874 if (subdir && page_private(subdir)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 ptr = shmem_swp_map(subdir);
876 size = limit - idx;
877 if (size > ENTRIES_PER_PAGE)
878 size = ENTRIES_PER_PAGE;
879 offset = shmem_find_swp(entry, ptr, ptr+size);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800880 shmem_swp_unmap(ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 if (offset >= 0) {
882 shmem_dir_unmap(dir);
883 goto found;
884 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700885 }
886 }
887lost1:
888 shmem_dir_unmap(dir-1);
889lost2:
890 spin_unlock(&info->lock);
891 return 0;
892found:
893 idx += offset;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800894 inode = igrab(&info->vfs_inode);
895 spin_unlock(&info->lock);
896
897 /* move head to start search for next from here */
898 list_move_tail(&shmem_swaplist, &info->swaplist);
899 mutex_unlock(&shmem_swaplist_mutex);
900
901 error = 1;
902 if (!inode)
903 goto out;
904
905 spin_lock(&info->lock);
906 ptr = shmem_swp_entry(info, idx, NULL);
907 if (ptr && ptr->val == entry.val)
908 error = add_to_page_cache(page, inode->i_mapping,
909 idx, GFP_ATOMIC);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800910 if (error == -EEXIST) {
911 struct page *filepage = find_get_page(inode->i_mapping, idx);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800912 error = 1;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800913 if (filepage) {
914 /*
915 * There might be a more uptodate page coming down
916 * from a stacked writepage: forget our swappage if so.
917 */
918 if (PageUptodate(filepage))
919 error = 0;
920 page_cache_release(filepage);
921 }
922 }
923 if (!error) {
Hugh Dickins73b12622008-02-04 22:28:50 -0800924 delete_from_swap_cache(page);
925 set_page_dirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926 info->flags |= SHMEM_PAGEIN;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800927 shmem_swp_set(info, ptr, 0);
928 swap_free(entry);
929 error = 1; /* not an error, but entry was found */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800931 if (ptr)
932 shmem_swp_unmap(ptr);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 spin_unlock(&info->lock);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800934out:
935 unlock_page(page);
936 page_cache_release(page);
937 iput(inode); /* allows for NULL */
938 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700939}
940
941/*
942 * shmem_unuse() search for an eventually swapped out shmem page.
943 */
944int shmem_unuse(swp_entry_t entry, struct page *page)
945{
946 struct list_head *p, *next;
947 struct shmem_inode_info *info;
948 int found = 0;
949
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800950 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951 list_for_each_safe(p, next, &shmem_swaplist) {
952 info = list_entry(p, struct shmem_inode_info, swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800953 if (info->swapped)
954 found = shmem_unuse_inode(info, entry, page);
955 else
Linus Torvalds1da177e2005-04-16 15:20:36 -0700956 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800957 cond_resched();
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800958 if (found)
959 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800961 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800962out: return found; /* 0 or 1 or -ENOMEM */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963}
964
965/*
966 * Move the page from the page cache to the swap cache.
967 */
968static int shmem_writepage(struct page *page, struct writeback_control *wbc)
969{
970 struct shmem_inode_info *info;
971 swp_entry_t *entry, swap;
972 struct address_space *mapping;
973 unsigned long index;
974 struct inode *inode;
975
976 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977 mapping = page->mapping;
978 index = page->index;
979 inode = mapping->host;
980 info = SHMEM_I(inode);
981 if (info->flags & VM_LOCKED)
982 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800983 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 goto redirty;
985
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800986 /*
987 * shmem_backing_dev_info's capabilities prevent regular writeback or
988 * sync from ever calling shmem_writepage; but a stacking filesystem
989 * may use the ->writepage of its underlying filesystem, in which case
990 * tmpfs should write out to swap only in response to memory pressure,
991 * and not for pdflush or sync. However, in those cases, we do still
992 * want to check if there's a redundant swappage to be discarded.
993 */
994 if (wbc->for_reclaim)
995 swap = get_swap_page();
996 else
997 swap.val = 0;
998
Linus Torvalds1da177e2005-04-16 15:20:36 -0700999 spin_lock(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000 if (index >= info->next_index) {
1001 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
1002 goto unlock;
1003 }
1004 entry = shmem_swp_entry(info, index, NULL);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001005 if (entry->val) {
1006 /*
1007 * The more uptodate page coming down from a stacked
1008 * writepage should replace our old swappage.
1009 */
1010 free_swap_and_cache(*entry);
1011 shmem_swp_set(info, entry, 0);
1012 }
1013 shmem_recalc_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001015 if (swap.val && add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Hugh Dickins73b12622008-02-04 22:28:50 -08001016 remove_from_page_cache(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 shmem_swp_set(info, entry, swap.val);
1018 shmem_swp_unmap(entry);
1019 spin_unlock(&info->lock);
1020 if (list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001021 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001022 /* move instead of add in case we're racing */
1023 list_move_tail(&info->swaplist, &shmem_swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001024 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 }
Hugh Dickins73b12622008-02-04 22:28:50 -08001026 swap_duplicate(swap);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001027 BUG_ON(page_mapped(page));
Hugh Dickins73b12622008-02-04 22:28:50 -08001028 page_cache_release(page); /* pagecache ref */
1029 set_page_dirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001030 unlock_page(page);
1031 return 0;
1032 }
1033
1034 shmem_swp_unmap(entry);
1035unlock:
1036 spin_unlock(&info->lock);
1037 swap_free(swap);
1038redirty:
1039 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001040 if (wbc->for_reclaim)
1041 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1042 unlock_page(page);
1043 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044}
1045
1046#ifdef CONFIG_NUMA
Hugh Dickinsd15c0232006-03-22 00:08:46 -08001047static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00001048{
1049 char *nodelist = strchr(value, ':');
1050 int err = 1;
1051
1052 if (nodelist) {
1053 /* NUL-terminate policy string */
1054 *nodelist++ = '\0';
1055 if (nodelist_parse(nodelist, *policy_nodes))
1056 goto out;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001057 if (!nodes_subset(*policy_nodes, node_states[N_HIGH_MEMORY]))
Hugh Dickinsa2109062007-06-08 13:46:46 -07001058 goto out;
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00001059 }
1060 if (!strcmp(value, "default")) {
1061 *policy = MPOL_DEFAULT;
1062 /* Don't allow a nodelist */
1063 if (!nodelist)
1064 err = 0;
1065 } else if (!strcmp(value, "prefer")) {
1066 *policy = MPOL_PREFERRED;
1067 /* Insist on a nodelist of one node only */
1068 if (nodelist) {
1069 char *rest = nodelist;
1070 while (isdigit(*rest))
1071 rest++;
1072 if (!*rest)
1073 err = 0;
1074 }
1075 } else if (!strcmp(value, "bind")) {
1076 *policy = MPOL_BIND;
1077 /* Insist on a nodelist */
1078 if (nodelist)
1079 err = 0;
1080 } else if (!strcmp(value, "interleave")) {
1081 *policy = MPOL_INTERLEAVE;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001082 /*
1083 * Default to online nodes with memory if no nodelist
1084 */
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00001085 if (!nodelist)
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001086 *policy_nodes = node_states[N_HIGH_MEMORY];
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00001087 err = 0;
1088 }
1089out:
1090 /* Restore string for error message */
1091 if (nodelist)
1092 *--nodelist = ':';
1093 return err;
1094}
1095
Hugh Dickins02098fe2008-02-04 22:28:42 -08001096static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1097 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099 struct vm_area_struct pvma;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001100 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101
1102 /* Create a pseudo vma that just contains the policy */
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001103 pvma.vm_start = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 pvma.vm_pgoff = idx;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001105 pvma.vm_ops = NULL;
1106 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001107 page = swapin_readahead(entry, gfp, &pvma, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108 mpol_free(pvma.vm_policy);
1109 return page;
1110}
1111
Hugh Dickins02098fe2008-02-04 22:28:42 -08001112static struct page *shmem_alloc_page(gfp_t gfp,
1113 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114{
1115 struct vm_area_struct pvma;
1116 struct page *page;
1117
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001118 /* Create a pseudo vma that just contains the policy */
1119 pvma.vm_start = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 pvma.vm_pgoff = idx;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -08001121 pvma.vm_ops = NULL;
1122 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
Hugh Dickinse84e2e12007-11-28 18:55:10 +00001123 page = alloc_page_vma(gfp, &pvma, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 mpol_free(pvma.vm_policy);
1125 return page;
1126}
1127#else
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001128static inline int shmem_parse_mpol(char *value, int *policy,
1129 nodemask_t *policy_nodes)
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00001130{
1131 return 1;
1132}
1133
Hugh Dickins02098fe2008-02-04 22:28:42 -08001134static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
1135 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136{
Hugh Dickins02098fe2008-02-04 22:28:42 -08001137 return swapin_readahead(entry, gfp, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138}
1139
Hugh Dickins02098fe2008-02-04 22:28:42 -08001140static inline struct page *shmem_alloc_page(gfp_t gfp,
1141 struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142{
Hugh Dickinse84e2e12007-11-28 18:55:10 +00001143 return alloc_page(gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144}
1145#endif
1146
1147/*
1148 * shmem_getpage - either get the page from swap or allocate a new one
1149 *
1150 * If we allocate a new one we do not mark it dirty. That's up to the
1151 * vm. If we swap it in we mark it dirty since we also free the swap
1152 * entry since a page cannot live in both the swap and page cache
1153 */
1154static int shmem_getpage(struct inode *inode, unsigned long idx,
1155 struct page **pagep, enum sgp_type sgp, int *type)
1156{
1157 struct address_space *mapping = inode->i_mapping;
1158 struct shmem_inode_info *info = SHMEM_I(inode);
1159 struct shmem_sb_info *sbinfo;
1160 struct page *filepage = *pagep;
1161 struct page *swappage;
1162 swp_entry_t *entry;
1163 swp_entry_t swap;
Hugh Dickins02098fe2008-02-04 22:28:42 -08001164 gfp_t gfp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 int error;
1166
1167 if (idx >= SHMEM_MAX_INDEX)
1168 return -EFBIG;
Nick Piggin54cb8822007-07-19 01:46:59 -07001169
1170 if (type)
Nick Piggin83c54072007-07-19 01:47:05 -07001171 *type = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07001172
Linus Torvalds1da177e2005-04-16 15:20:36 -07001173 /*
1174 * Normally, filepage is NULL on entry, and either found
1175 * uptodate immediately, or allocated and zeroed, or read
1176 * in under swappage, which is then assigned to filepage.
Hugh Dickins5402b972008-02-04 22:28:44 -08001177 * But shmem_readpage (required for splice) passes in a locked
Hugh Dickinsae976412007-06-04 10:00:39 +02001178 * filepage, which may be found not uptodate by other callers
1179 * too, and may need to be copied from the swappage read in.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 */
1181repeat:
1182 if (!filepage)
1183 filepage = find_lock_page(mapping, idx);
1184 if (filepage && PageUptodate(filepage))
1185 goto done;
1186 error = 0;
Hugh Dickins02098fe2008-02-04 22:28:42 -08001187 gfp = mapping_gfp_mask(mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188
1189 spin_lock(&info->lock);
1190 shmem_recalc_inode(inode);
1191 entry = shmem_swp_alloc(info, idx, sgp);
1192 if (IS_ERR(entry)) {
1193 spin_unlock(&info->lock);
1194 error = PTR_ERR(entry);
1195 goto failed;
1196 }
1197 swap = *entry;
1198
1199 if (swap.val) {
1200 /* Look it up and read it in.. */
1201 swappage = lookup_swap_cache(swap);
1202 if (!swappage) {
1203 shmem_swp_unmap(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001204 /* here we actually do the io */
Nick Piggin83c54072007-07-19 01:47:05 -07001205 if (type && !(*type & VM_FAULT_MAJOR)) {
Christoph Lameterf8891e52006-06-30 01:55:45 -07001206 __count_vm_event(PGMAJFAULT);
Nick Piggin83c54072007-07-19 01:47:05 -07001207 *type |= VM_FAULT_MAJOR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 }
Christoph Lameterf8891e52006-06-30 01:55:45 -07001209 spin_unlock(&info->lock);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001210 swappage = shmem_swapin(swap, gfp, info, idx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 if (!swappage) {
1212 spin_lock(&info->lock);
1213 entry = shmem_swp_alloc(info, idx, sgp);
1214 if (IS_ERR(entry))
1215 error = PTR_ERR(entry);
1216 else {
1217 if (entry->val == swap.val)
1218 error = -ENOMEM;
1219 shmem_swp_unmap(entry);
1220 }
1221 spin_unlock(&info->lock);
1222 if (error)
1223 goto failed;
1224 goto repeat;
1225 }
1226 wait_on_page_locked(swappage);
1227 page_cache_release(swappage);
1228 goto repeat;
1229 }
1230
1231 /* We have to do this with page locked to prevent races */
1232 if (TestSetPageLocked(swappage)) {
1233 shmem_swp_unmap(entry);
1234 spin_unlock(&info->lock);
1235 wait_on_page_locked(swappage);
1236 page_cache_release(swappage);
1237 goto repeat;
1238 }
1239 if (PageWriteback(swappage)) {
1240 shmem_swp_unmap(entry);
1241 spin_unlock(&info->lock);
1242 wait_on_page_writeback(swappage);
1243 unlock_page(swappage);
1244 page_cache_release(swappage);
1245 goto repeat;
1246 }
1247 if (!PageUptodate(swappage)) {
1248 shmem_swp_unmap(entry);
1249 spin_unlock(&info->lock);
1250 unlock_page(swappage);
1251 page_cache_release(swappage);
1252 error = -EIO;
1253 goto failed;
1254 }
1255
1256 if (filepage) {
1257 shmem_swp_set(info, entry, 0);
1258 shmem_swp_unmap(entry);
1259 delete_from_swap_cache(swappage);
1260 spin_unlock(&info->lock);
1261 copy_highpage(filepage, swappage);
1262 unlock_page(swappage);
1263 page_cache_release(swappage);
1264 flush_dcache_page(filepage);
1265 SetPageUptodate(filepage);
1266 set_page_dirty(filepage);
1267 swap_free(swap);
Hugh Dickins73b12622008-02-04 22:28:50 -08001268 } else if (!(error = add_to_page_cache(
1269 swappage, mapping, idx, GFP_ATOMIC))) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 info->flags |= SHMEM_PAGEIN;
1271 shmem_swp_set(info, entry, 0);
1272 shmem_swp_unmap(entry);
Hugh Dickins73b12622008-02-04 22:28:50 -08001273 delete_from_swap_cache(swappage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 spin_unlock(&info->lock);
1275 filepage = swappage;
Hugh Dickins73b12622008-02-04 22:28:50 -08001276 set_page_dirty(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 swap_free(swap);
1278 } else {
1279 shmem_swp_unmap(entry);
1280 spin_unlock(&info->lock);
1281 unlock_page(swappage);
1282 page_cache_release(swappage);
1283 if (error == -ENOMEM) {
1284 /* let kswapd refresh zone for GFP_ATOMICs */
Andrew Morton3fcfab12006-10-19 23:28:16 -07001285 congestion_wait(WRITE, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 }
1287 goto repeat;
1288 }
1289 } else if (sgp == SGP_READ && !filepage) {
1290 shmem_swp_unmap(entry);
1291 filepage = find_get_page(mapping, idx);
1292 if (filepage &&
1293 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1294 spin_unlock(&info->lock);
1295 wait_on_page_locked(filepage);
1296 page_cache_release(filepage);
1297 filepage = NULL;
1298 goto repeat;
1299 }
1300 spin_unlock(&info->lock);
1301 } else {
1302 shmem_swp_unmap(entry);
1303 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001304 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 spin_lock(&sbinfo->stat_lock);
1306 if (sbinfo->free_blocks == 0 ||
1307 shmem_acct_block(info->flags)) {
1308 spin_unlock(&sbinfo->stat_lock);
1309 spin_unlock(&info->lock);
1310 error = -ENOSPC;
1311 goto failed;
1312 }
1313 sbinfo->free_blocks--;
1314 inode->i_blocks += BLOCKS_PER_PAGE;
1315 spin_unlock(&sbinfo->stat_lock);
1316 } else if (shmem_acct_block(info->flags)) {
1317 spin_unlock(&info->lock);
1318 error = -ENOSPC;
1319 goto failed;
1320 }
1321
1322 if (!filepage) {
1323 spin_unlock(&info->lock);
Hugh Dickins02098fe2008-02-04 22:28:42 -08001324 filepage = shmem_alloc_page(gfp, info, idx);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 if (!filepage) {
1326 shmem_unacct_blocks(info->flags, 1);
1327 shmem_free_blocks(inode, 1);
1328 error = -ENOMEM;
1329 goto failed;
1330 }
1331
1332 spin_lock(&info->lock);
1333 entry = shmem_swp_alloc(info, idx, sgp);
1334 if (IS_ERR(entry))
1335 error = PTR_ERR(entry);
1336 else {
1337 swap = *entry;
1338 shmem_swp_unmap(entry);
1339 }
1340 if (error || swap.val || 0 != add_to_page_cache_lru(
1341 filepage, mapping, idx, GFP_ATOMIC)) {
1342 spin_unlock(&info->lock);
1343 page_cache_release(filepage);
1344 shmem_unacct_blocks(info->flags, 1);
1345 shmem_free_blocks(inode, 1);
1346 filepage = NULL;
1347 if (error)
1348 goto failed;
1349 goto repeat;
1350 }
1351 info->flags |= SHMEM_PAGEIN;
1352 }
1353
1354 info->alloced++;
1355 spin_unlock(&info->lock);
Hugh Dickinse84e2e12007-11-28 18:55:10 +00001356 clear_highpage(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 flush_dcache_page(filepage);
1358 SetPageUptodate(filepage);
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001359 if (sgp == SGP_DIRTY)
1360 set_page_dirty(filepage);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 }
1362done:
Hugh Dickinsd3602442008-02-04 22:28:44 -08001363 *pagep = filepage;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 return 0;
1365
1366failed:
1367 if (*pagep != filepage) {
1368 unlock_page(filepage);
1369 page_cache_release(filepage);
1370 }
1371 return error;
1372}
1373
Nick Piggind0217ac2007-07-19 01:47:03 -07001374static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001376 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 int error;
Nick Piggind0217ac2007-07-19 01:47:03 -07001378 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379
Nick Piggind0217ac2007-07-19 01:47:03 -07001380 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1381 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001382
Hugh Dickins27d54b32008-02-04 22:28:43 -08001383 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
Nick Piggind0217ac2007-07-19 01:47:03 -07001384 if (error)
1385 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386
Nick Piggind0217ac2007-07-19 01:47:03 -07001387 mark_page_accessed(vmf->page);
Nick Piggin83c54072007-07-19 01:47:05 -07001388 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389}
1390
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391#ifdef CONFIG_NUMA
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001392static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001394 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1396}
1397
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001398static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1399 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001401 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 unsigned long idx;
1403
1404 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1405 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1406}
1407#endif
1408
1409int shmem_lock(struct file *file, int lock, struct user_struct *user)
1410{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001411 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 struct shmem_inode_info *info = SHMEM_I(inode);
1413 int retval = -ENOMEM;
1414
1415 spin_lock(&info->lock);
1416 if (lock && !(info->flags & VM_LOCKED)) {
1417 if (!user_shm_lock(inode->i_size, user))
1418 goto out_nomem;
1419 info->flags |= VM_LOCKED;
1420 }
1421 if (!lock && (info->flags & VM_LOCKED) && user) {
1422 user_shm_unlock(inode->i_size, user);
1423 info->flags &= ~VM_LOCKED;
1424 }
1425 retval = 0;
1426out_nomem:
1427 spin_unlock(&info->lock);
1428 return retval;
1429}
1430
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08001431static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432{
1433 file_accessed(file);
1434 vma->vm_ops = &shmem_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001435 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 return 0;
1437}
1438
1439static struct inode *
1440shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1441{
1442 struct inode *inode;
1443 struct shmem_inode_info *info;
1444 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1445
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001446 if (shmem_reserve_inode(sb))
1447 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448
1449 inode = new_inode(sb);
1450 if (inode) {
1451 inode->i_mode = mode;
1452 inode->i_uid = current->fsuid;
1453 inode->i_gid = current->fsgid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 inode->i_blocks = 0;
1455 inode->i_mapping->a_ops = &shmem_aops;
1456 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1457 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
David M. Grimes91828a42006-10-17 00:09:45 -07001458 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459 info = SHMEM_I(inode);
1460 memset(info, 0, (char *)inode - (char *)info);
1461 spin_lock_init(&info->lock);
1462 INIT_LIST_HEAD(&info->swaplist);
1463
1464 switch (mode & S_IFMT) {
1465 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001466 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467 init_special_inode(inode, mode, dev);
1468 break;
1469 case S_IFREG:
1470 inode->i_op = &shmem_inode_operations;
1471 inode->i_fop = &shmem_file_operations;
Robin Holt7339ff82006-01-14 13:20:48 -08001472 mpol_shared_policy_init(&info->policy, sbinfo->policy,
1473 &sbinfo->policy_nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474 break;
1475 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07001476 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 /* Some things misbehave if size == 0 on a directory */
1478 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1479 inode->i_op = &shmem_dir_inode_operations;
1480 inode->i_fop = &simple_dir_operations;
1481 break;
1482 case S_IFLNK:
1483 /*
1484 * Must not load anything in the rbtree,
1485 * mpol_free_shared_policy will not be called.
1486 */
Robin Holt7339ff82006-01-14 13:20:48 -08001487 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT,
1488 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 break;
1490 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001491 } else
1492 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 return inode;
1494}
1495
1496#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001497static const struct inode_operations shmem_symlink_inode_operations;
1498static const struct inode_operations shmem_symlink_inline_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499
1500/*
Nick Piggin800d15a2007-10-16 01:25:03 -07001501 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
Hugh Dickinsae976412007-06-04 10:00:39 +02001502 * but providing them allows a tmpfs file to be used for splice, sendfile, and
1503 * below the loop driver, in the generic fashion that many filesystems support.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 */
Hugh Dickinsae976412007-06-04 10:00:39 +02001505static int shmem_readpage(struct file *file, struct page *page)
1506{
1507 struct inode *inode = page->mapping->host;
1508 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1509 unlock_page(page);
1510 return error;
1511}
1512
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513static int
Nick Piggin800d15a2007-10-16 01:25:03 -07001514shmem_write_begin(struct file *file, struct address_space *mapping,
1515 loff_t pos, unsigned len, unsigned flags,
1516 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001517{
Nick Piggin800d15a2007-10-16 01:25:03 -07001518 struct inode *inode = mapping->host;
1519 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1520 *pagep = NULL;
1521 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1522}
1523
1524static int
1525shmem_write_end(struct file *file, struct address_space *mapping,
1526 loff_t pos, unsigned len, unsigned copied,
1527 struct page *page, void *fsdata)
1528{
1529 struct inode *inode = mapping->host;
1530
Hugh Dickinsd3602442008-02-04 22:28:44 -08001531 if (pos + copied > inode->i_size)
1532 i_size_write(inode, pos + copied);
1533
1534 unlock_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07001535 set_page_dirty(page);
1536 page_cache_release(page);
1537
Nick Piggin800d15a2007-10-16 01:25:03 -07001538 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539}
1540
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1542{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001543 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 struct address_space *mapping = inode->i_mapping;
1545 unsigned long index, offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001546 enum sgp_type sgp = SGP_READ;
1547
1548 /*
1549 * Might this read be for a stacking filesystem? Then when reading
1550 * holes of a sparse file, we actually need to allocate those pages,
1551 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1552 */
1553 if (segment_eq(get_fs(), KERNEL_DS))
1554 sgp = SGP_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555
1556 index = *ppos >> PAGE_CACHE_SHIFT;
1557 offset = *ppos & ~PAGE_CACHE_MASK;
1558
1559 for (;;) {
1560 struct page *page = NULL;
1561 unsigned long end_index, nr, ret;
1562 loff_t i_size = i_size_read(inode);
1563
1564 end_index = i_size >> PAGE_CACHE_SHIFT;
1565 if (index > end_index)
1566 break;
1567 if (index == end_index) {
1568 nr = i_size & ~PAGE_CACHE_MASK;
1569 if (nr <= offset)
1570 break;
1571 }
1572
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001573 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574 if (desc->error) {
1575 if (desc->error == -EINVAL)
1576 desc->error = 0;
1577 break;
1578 }
Hugh Dickinsd3602442008-02-04 22:28:44 -08001579 if (page)
1580 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581
1582 /*
1583 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001584 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 */
1586 nr = PAGE_CACHE_SIZE;
1587 i_size = i_size_read(inode);
1588 end_index = i_size >> PAGE_CACHE_SHIFT;
1589 if (index == end_index) {
1590 nr = i_size & ~PAGE_CACHE_MASK;
1591 if (nr <= offset) {
1592 if (page)
1593 page_cache_release(page);
1594 break;
1595 }
1596 }
1597 nr -= offset;
1598
1599 if (page) {
1600 /*
1601 * If users can be writing to this page using arbitrary
1602 * virtual addresses, take care about potential aliasing
1603 * before reading the page on the kernel side.
1604 */
1605 if (mapping_writably_mapped(mapping))
1606 flush_dcache_page(page);
1607 /*
1608 * Mark the page accessed if we read the beginning.
1609 */
1610 if (!offset)
1611 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001612 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 page = ZERO_PAGE(0);
Nick Pigginb5810032005-10-29 18:16:12 -07001614 page_cache_get(page);
1615 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616
1617 /*
1618 * Ok, we have the page, and it's up-to-date, so
1619 * now we can copy it to user space...
1620 *
1621 * The actor routine returns how many bytes were actually used..
1622 * NOTE! This may not be the same as how much of a user buffer
1623 * we filled up (we may be padding etc), so we can only update
1624 * "pos" here (the actor routine has to update the user buffer
1625 * pointers and the remaining count).
1626 */
1627 ret = actor(desc, page, offset, nr);
1628 offset += ret;
1629 index += offset >> PAGE_CACHE_SHIFT;
1630 offset &= ~PAGE_CACHE_MASK;
1631
1632 page_cache_release(page);
1633 if (ret != nr || !desc->count)
1634 break;
1635
1636 cond_resched();
1637 }
1638
1639 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1640 file_accessed(filp);
1641}
1642
1643static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1644{
1645 read_descriptor_t desc;
1646
1647 if ((ssize_t) count < 0)
1648 return -EINVAL;
1649 if (!access_ok(VERIFY_WRITE, buf, count))
1650 return -EFAULT;
1651 if (!count)
1652 return 0;
1653
1654 desc.written = 0;
1655 desc.count = count;
1656 desc.arg.buf = buf;
1657 desc.error = 0;
1658
1659 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1660 if (desc.written)
1661 return desc.written;
1662 return desc.error;
1663}
1664
David Howells726c3342006-06-23 02:02:58 -07001665static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666{
David Howells726c3342006-06-23 02:02:58 -07001667 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668
1669 buf->f_type = TMPFS_MAGIC;
1670 buf->f_bsize = PAGE_CACHE_SIZE;
1671 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001672 spin_lock(&sbinfo->stat_lock);
1673 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 buf->f_blocks = sbinfo->max_blocks;
1675 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001676 }
1677 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 buf->f_files = sbinfo->max_inodes;
1679 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680 }
1681 /* else leave those fields 0 like simple_statfs */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001682 spin_unlock(&sbinfo->stat_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 return 0;
1684}
1685
1686/*
1687 * File creation. Allocate an inode, and we're done..
1688 */
1689static int
1690shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1691{
1692 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1693 int error = -ENOSPC;
1694
1695 if (inode) {
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001696 error = security_inode_init_security(inode, dir, NULL, NULL,
1697 NULL);
1698 if (error) {
1699 if (error != -EOPNOTSUPP) {
1700 iput(inode);
1701 return error;
1702 }
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001703 }
1704 error = shmem_acl_init(inode, dir);
1705 if (error) {
1706 iput(inode);
1707 return error;
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001708 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 if (dir->i_mode & S_ISGID) {
1710 inode->i_gid = dir->i_gid;
1711 if (S_ISDIR(mode))
1712 inode->i_mode |= S_ISGID;
1713 }
1714 dir->i_size += BOGO_DIRENT_SIZE;
1715 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1716 d_instantiate(dentry, inode);
1717 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 }
1719 return error;
1720}
1721
1722static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1723{
1724 int error;
1725
1726 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1727 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07001728 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 return 0;
1730}
1731
1732static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1733 struct nameidata *nd)
1734{
1735 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1736}
1737
1738/*
1739 * Link a file..
1740 */
1741static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1742{
1743 struct inode *inode = old_dentry->d_inode;
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001744 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745
1746 /*
1747 * No ordinary (disk based) filesystem counts links as inodes;
1748 * but each new link needs a new dentry, pinning lowmem, and
1749 * tmpfs dentries cannot be pruned until they are unlinked.
1750 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001751 ret = shmem_reserve_inode(inode->i_sb);
1752 if (ret)
1753 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001754
1755 dir->i_size += BOGO_DIRENT_SIZE;
1756 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansend8c76e62006-09-30 23:29:04 -07001757 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 atomic_inc(&inode->i_count); /* New dentry reference */
1759 dget(dentry); /* Extra pinning count for the created dentry */
1760 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001761out:
1762 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763}
1764
1765static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1766{
1767 struct inode *inode = dentry->d_inode;
1768
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001769 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1770 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771
1772 dir->i_size -= BOGO_DIRENT_SIZE;
1773 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001774 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 dput(dentry); /* Undo the count from "create" - this does all the work */
1776 return 0;
1777}
1778
1779static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1780{
1781 if (!simple_empty(dentry))
1782 return -ENOTEMPTY;
1783
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001784 drop_nlink(dentry->d_inode);
1785 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 return shmem_unlink(dir, dentry);
1787}
1788
1789/*
1790 * The VFS layer already does all the dentry stuff for rename,
1791 * we just have to decrement the usage count for the target if
1792 * it exists so that the VFS layer correctly free's it when it
1793 * gets overwritten.
1794 */
1795static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1796{
1797 struct inode *inode = old_dentry->d_inode;
1798 int they_are_dirs = S_ISDIR(inode->i_mode);
1799
1800 if (!simple_empty(new_dentry))
1801 return -ENOTEMPTY;
1802
1803 if (new_dentry->d_inode) {
1804 (void) shmem_unlink(new_dir, new_dentry);
1805 if (they_are_dirs)
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001806 drop_nlink(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001808 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07001809 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 }
1811
1812 old_dir->i_size -= BOGO_DIRENT_SIZE;
1813 new_dir->i_size += BOGO_DIRENT_SIZE;
1814 old_dir->i_ctime = old_dir->i_mtime =
1815 new_dir->i_ctime = new_dir->i_mtime =
1816 inode->i_ctime = CURRENT_TIME;
1817 return 0;
1818}
1819
1820static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1821{
1822 int error;
1823 int len;
1824 struct inode *inode;
1825 struct page *page = NULL;
1826 char *kaddr;
1827 struct shmem_inode_info *info;
1828
1829 len = strlen(symname) + 1;
1830 if (len > PAGE_CACHE_SIZE)
1831 return -ENAMETOOLONG;
1832
1833 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1834 if (!inode)
1835 return -ENOSPC;
1836
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001837 error = security_inode_init_security(inode, dir, NULL, NULL,
1838 NULL);
1839 if (error) {
1840 if (error != -EOPNOTSUPP) {
1841 iput(inode);
1842 return error;
1843 }
1844 error = 0;
1845 }
1846
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 info = SHMEM_I(inode);
1848 inode->i_size = len-1;
1849 if (len <= (char *)inode - (char *)info) {
1850 /* do it inline */
1851 memcpy(info, symname, len);
1852 inode->i_op = &shmem_symlink_inline_operations;
1853 } else {
1854 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1855 if (error) {
1856 iput(inode);
1857 return error;
1858 }
Hugh Dickinsd3602442008-02-04 22:28:44 -08001859 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 inode->i_op = &shmem_symlink_inode_operations;
1861 kaddr = kmap_atomic(page, KM_USER0);
1862 memcpy(kaddr, symname, len);
1863 kunmap_atomic(kaddr, KM_USER0);
1864 set_page_dirty(page);
1865 page_cache_release(page);
1866 }
1867 if (dir->i_mode & S_ISGID)
1868 inode->i_gid = dir->i_gid;
1869 dir->i_size += BOGO_DIRENT_SIZE;
1870 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1871 d_instantiate(dentry, inode);
1872 dget(dentry);
1873 return 0;
1874}
1875
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001876static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877{
1878 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001879 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880}
1881
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001882static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883{
1884 struct page *page = NULL;
1885 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1886 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
Hugh Dickinsd3602442008-02-04 22:28:44 -08001887 if (page)
1888 unlock_page(page);
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001889 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890}
1891
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001892static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893{
1894 if (!IS_ERR(nd_get_link(nd))) {
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001895 struct page *page = cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 kunmap(page);
1897 mark_page_accessed(page);
1898 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 }
1900}
1901
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001902static const struct inode_operations shmem_symlink_inline_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 .readlink = generic_readlink,
1904 .follow_link = shmem_follow_link_inline,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001905};
1906
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001907static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 .truncate = shmem_truncate,
1909 .readlink = generic_readlink,
1910 .follow_link = shmem_follow_link,
1911 .put_link = shmem_put_link,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912};
1913
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001914#ifdef CONFIG_TMPFS_POSIX_ACL
1915/**
1916 * Superblocks without xattr inode operations will get security.* xattr
1917 * support from the VFS "for free". As soon as we have any other xattrs
1918 * like ACLs, we also need to implement the security.* handlers at
1919 * filesystem level, though.
1920 */
1921
1922static size_t shmem_xattr_security_list(struct inode *inode, char *list,
1923 size_t list_len, const char *name,
1924 size_t name_len)
1925{
1926 return security_inode_listsecurity(inode, list, list_len);
1927}
1928
1929static int shmem_xattr_security_get(struct inode *inode, const char *name,
1930 void *buffer, size_t size)
1931{
1932 if (strcmp(name, "") == 0)
1933 return -EINVAL;
1934 return security_inode_getsecurity(inode, name, buffer, size,
1935 -EOPNOTSUPP);
1936}
1937
1938static int shmem_xattr_security_set(struct inode *inode, const char *name,
1939 const void *value, size_t size, int flags)
1940{
1941 if (strcmp(name, "") == 0)
1942 return -EINVAL;
1943 return security_inode_setsecurity(inode, name, value, size, flags);
1944}
1945
Adrian Bunk1f370a22006-12-06 20:38:21 -08001946static struct xattr_handler shmem_xattr_security_handler = {
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001947 .prefix = XATTR_SECURITY_PREFIX,
1948 .list = shmem_xattr_security_list,
1949 .get = shmem_xattr_security_get,
1950 .set = shmem_xattr_security_set,
1951};
1952
1953static struct xattr_handler *shmem_xattr_handlers[] = {
1954 &shmem_xattr_acl_access_handler,
1955 &shmem_xattr_acl_default_handler,
1956 &shmem_xattr_security_handler,
1957 NULL
1958};
1959#endif
1960
David M. Grimes91828a42006-10-17 00:09:45 -07001961static struct dentry *shmem_get_parent(struct dentry *child)
1962{
1963 return ERR_PTR(-ESTALE);
1964}
1965
1966static int shmem_match(struct inode *ino, void *vfh)
1967{
1968 __u32 *fh = vfh;
1969 __u64 inum = fh[2];
1970 inum = (inum << 32) | fh[1];
1971 return ino->i_ino == inum && fh[0] == ino->i_generation;
1972}
1973
Christoph Hellwig480b1162007-10-21 16:42:13 -07001974static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
1975 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07001976{
David M. Grimes91828a42006-10-17 00:09:45 -07001977 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07001978 struct dentry *dentry = NULL;
1979 u64 inum = fid->raw[2];
1980 inum = (inum << 32) | fid->raw[1];
David M. Grimes91828a42006-10-17 00:09:45 -07001981
Christoph Hellwig480b1162007-10-21 16:42:13 -07001982 if (fh_len < 3)
1983 return NULL;
1984
1985 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
1986 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07001987 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07001988 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07001989 iput(inode);
1990 }
1991
Christoph Hellwig480b1162007-10-21 16:42:13 -07001992 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07001993}
1994
1995static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
1996 int connectable)
1997{
1998 struct inode *inode = dentry->d_inode;
1999
2000 if (*len < 3)
2001 return 255;
2002
2003 if (hlist_unhashed(&inode->i_hash)) {
2004 /* Unfortunately insert_inode_hash is not idempotent,
2005 * so as we hash inodes here rather than at creation
2006 * time, we need a lock to ensure we only try
2007 * to do it once
2008 */
2009 static DEFINE_SPINLOCK(lock);
2010 spin_lock(&lock);
2011 if (hlist_unhashed(&inode->i_hash))
2012 __insert_inode_hash(inode,
2013 inode->i_ino + inode->i_generation);
2014 spin_unlock(&lock);
2015 }
2016
2017 fh[0] = inode->i_generation;
2018 fh[1] = inode->i_ino;
2019 fh[2] = ((__u64)inode->i_ino) >> 32;
2020
2021 *len = 3;
2022 return 1;
2023}
2024
Christoph Hellwig39655162007-10-21 16:42:17 -07002025static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07002026 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07002027 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07002028 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07002029};
2030
Robin Holt7339ff82006-01-14 13:20:48 -08002031static int shmem_parse_options(char *options, int *mode, uid_t *uid,
2032 gid_t *gid, unsigned long *blocks, unsigned long *inodes,
2033 int *policy, nodemask_t *policy_nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034{
2035 char *this_char, *value, *rest;
2036
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002037 while (options != NULL) {
2038 this_char = options;
2039 for (;;) {
2040 /*
2041 * NUL-terminate this option: unfortunately,
2042 * mount options form a comma-separated list,
2043 * but mpol's nodelist may also contain commas.
2044 */
2045 options = strchr(options, ',');
2046 if (options == NULL)
2047 break;
2048 options++;
2049 if (!isdigit(*options)) {
2050 options[-1] = '\0';
2051 break;
2052 }
2053 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054 if (!*this_char)
2055 continue;
2056 if ((value = strchr(this_char,'=')) != NULL) {
2057 *value++ = 0;
2058 } else {
2059 printk(KERN_ERR
2060 "tmpfs: No value for mount option '%s'\n",
2061 this_char);
2062 return 1;
2063 }
2064
2065 if (!strcmp(this_char,"size")) {
2066 unsigned long long size;
2067 size = memparse(value,&rest);
2068 if (*rest == '%') {
2069 size <<= PAGE_SHIFT;
2070 size *= totalram_pages;
2071 do_div(size, 100);
2072 rest++;
2073 }
2074 if (*rest)
2075 goto bad_val;
Michael Marineau818db352008-02-04 22:28:48 -08002076 *blocks = DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002077 } else if (!strcmp(this_char,"nr_blocks")) {
2078 *blocks = memparse(value,&rest);
2079 if (*rest)
2080 goto bad_val;
2081 } else if (!strcmp(this_char,"nr_inodes")) {
2082 *inodes = memparse(value,&rest);
2083 if (*rest)
2084 goto bad_val;
2085 } else if (!strcmp(this_char,"mode")) {
2086 if (!mode)
2087 continue;
2088 *mode = simple_strtoul(value,&rest,8);
2089 if (*rest)
2090 goto bad_val;
2091 } else if (!strcmp(this_char,"uid")) {
2092 if (!uid)
2093 continue;
2094 *uid = simple_strtoul(value,&rest,0);
2095 if (*rest)
2096 goto bad_val;
2097 } else if (!strcmp(this_char,"gid")) {
2098 if (!gid)
2099 continue;
2100 *gid = simple_strtoul(value,&rest,0);
2101 if (*rest)
2102 goto bad_val;
Robin Holt7339ff82006-01-14 13:20:48 -08002103 } else if (!strcmp(this_char,"mpol")) {
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002104 if (shmem_parse_mpol(value,policy,policy_nodes))
Robin Holt7339ff82006-01-14 13:20:48 -08002105 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 } else {
2107 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2108 this_char);
2109 return 1;
2110 }
2111 }
2112 return 0;
2113
2114bad_val:
2115 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2116 value, this_char);
2117 return 1;
2118
2119}
2120
2121static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2122{
2123 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002124 unsigned long max_blocks = sbinfo->max_blocks;
2125 unsigned long max_inodes = sbinfo->max_inodes;
Robin Holt7339ff82006-01-14 13:20:48 -08002126 int policy = sbinfo->policy;
2127 nodemask_t policy_nodes = sbinfo->policy_nodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002128 unsigned long blocks;
2129 unsigned long inodes;
2130 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131
Robin Holt7339ff82006-01-14 13:20:48 -08002132 if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks,
2133 &max_inodes, &policy, &policy_nodes))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002134 return error;
2135
2136 spin_lock(&sbinfo->stat_lock);
2137 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2138 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2139 if (max_blocks < blocks)
2140 goto out;
2141 if (max_inodes < inodes)
2142 goto out;
2143 /*
2144 * Those tests also disallow limited->unlimited while any are in
2145 * use, so i_blocks will always be zero when max_blocks is zero;
2146 * but we must separately disallow unlimited->limited, because
2147 * in that case we have no record of how much is already in use.
2148 */
2149 if (max_blocks && !sbinfo->max_blocks)
2150 goto out;
2151 if (max_inodes && !sbinfo->max_inodes)
2152 goto out;
2153
2154 error = 0;
2155 sbinfo->max_blocks = max_blocks;
2156 sbinfo->free_blocks = max_blocks - blocks;
2157 sbinfo->max_inodes = max_inodes;
2158 sbinfo->free_inodes = max_inodes - inodes;
Robin Holt7339ff82006-01-14 13:20:48 -08002159 sbinfo->policy = policy;
2160 sbinfo->policy_nodes = policy_nodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002161out:
2162 spin_unlock(&sbinfo->stat_lock);
2163 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164}
2165#endif
2166
2167static void shmem_put_super(struct super_block *sb)
2168{
2169 kfree(sb->s_fs_info);
2170 sb->s_fs_info = NULL;
2171}
2172
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173static int shmem_fill_super(struct super_block *sb,
2174 void *data, int silent)
2175{
2176 struct inode *inode;
2177 struct dentry *root;
2178 int mode = S_IRWXUGO | S_ISVTX;
2179 uid_t uid = current->fsuid;
2180 gid_t gid = current->fsgid;
2181 int err = -ENOMEM;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002182 struct shmem_sb_info *sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183 unsigned long blocks = 0;
2184 unsigned long inodes = 0;
Robin Holt7339ff82006-01-14 13:20:48 -08002185 int policy = MPOL_DEFAULT;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002186 nodemask_t policy_nodes = node_states[N_HIGH_MEMORY];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002188#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189 /*
2190 * Per default we only allow half of the physical ram per
2191 * tmpfs instance, limiting inodes to one per page of lowmem;
2192 * but the internal instance is left unlimited.
2193 */
2194 if (!(sb->s_flags & MS_NOUSER)) {
2195 blocks = totalram_pages / 2;
2196 inodes = totalram_pages - totalhigh_pages;
2197 if (inodes > blocks)
2198 inodes = blocks;
Robin Holt7339ff82006-01-14 13:20:48 -08002199 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks,
2200 &inodes, &policy, &policy_nodes))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 return -EINVAL;
2202 }
David M. Grimes91828a42006-10-17 00:09:45 -07002203 sb->s_export_op = &shmem_export_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204#else
2205 sb->s_flags |= MS_NOUSER;
2206#endif
2207
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002208 /* Round up to L1_CACHE_BYTES to resist false sharing */
2209 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
2210 L1_CACHE_BYTES), GFP_KERNEL);
2211 if (!sbinfo)
2212 return -ENOMEM;
2213
2214 spin_lock_init(&sbinfo->stat_lock);
2215 sbinfo->max_blocks = blocks;
2216 sbinfo->free_blocks = blocks;
2217 sbinfo->max_inodes = inodes;
2218 sbinfo->free_inodes = inodes;
Robin Holt7339ff82006-01-14 13:20:48 -08002219 sbinfo->policy = policy;
2220 sbinfo->policy_nodes = policy_nodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002221
2222 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 sb->s_maxbytes = SHMEM_MAX_BYTES;
2224 sb->s_blocksize = PAGE_CACHE_SIZE;
2225 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2226 sb->s_magic = TMPFS_MAGIC;
2227 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01002228 sb->s_time_gran = 1;
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002229#ifdef CONFIG_TMPFS_POSIX_ACL
2230 sb->s_xattr = shmem_xattr_handlers;
2231 sb->s_flags |= MS_POSIXACL;
2232#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002233
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
2235 if (!inode)
2236 goto failed;
2237 inode->i_uid = uid;
2238 inode->i_gid = gid;
2239 root = d_alloc_root(inode);
2240 if (!root)
2241 goto failed_iput;
2242 sb->s_root = root;
2243 return 0;
2244
2245failed_iput:
2246 iput(inode);
2247failed:
2248 shmem_put_super(sb);
2249 return err;
2250}
2251
Pekka Enbergfcc234f2006-03-22 00:08:13 -08002252static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253
2254static struct inode *shmem_alloc_inode(struct super_block *sb)
2255{
2256 struct shmem_inode_info *p;
Christoph Lametere94b1762006-12-06 20:33:17 -08002257 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258 if (!p)
2259 return NULL;
2260 return &p->vfs_inode;
2261}
2262
2263static void shmem_destroy_inode(struct inode *inode)
2264{
2265 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2266 /* only struct inode is valid if it's an inline symlink */
2267 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2268 }
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002269 shmem_acl_destroy_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2271}
2272
Christoph Lameter4ba9b9d2007-10-16 23:25:51 -07002273static void init_once(struct kmem_cache *cachep, void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274{
2275 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2276
Christoph Lametera35afb82007-05-16 22:10:57 -07002277 inode_init_once(&p->vfs_inode);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002278#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Lametera35afb82007-05-16 22:10:57 -07002279 p->i_acl = NULL;
2280 p->i_default_acl = NULL;
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002281#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282}
2283
2284static int init_inodecache(void)
2285{
2286 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2287 sizeof(struct shmem_inode_info),
Alexey Dobriyan040b5c62007-10-16 23:26:10 -07002288 0, SLAB_PANIC, init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289 return 0;
2290}
2291
2292static void destroy_inodecache(void)
2293{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07002294 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295}
2296
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002297static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08002299 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300#ifdef CONFIG_TMPFS
Hugh Dickinsae976412007-06-04 10:00:39 +02002301 .readpage = shmem_readpage,
Nick Piggin800d15a2007-10-16 01:25:03 -07002302 .write_begin = shmem_write_begin,
2303 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304#endif
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07002305 .migratepage = migrate_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306};
2307
Helge Deller15ad7cd2006-12-06 20:40:36 -08002308static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002309 .mmap = shmem_mmap,
2310#ifdef CONFIG_TMPFS
2311 .llseek = generic_file_llseek,
2312 .read = shmem_file_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002313 .write = do_sync_write,
2314 .aio_write = generic_file_aio_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 .fsync = simple_sync_file,
Hugh Dickinsae976412007-06-04 10:00:39 +02002316 .splice_read = generic_file_splice_read,
2317 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318#endif
2319};
2320
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002321static const struct inode_operations shmem_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 .truncate = shmem_truncate,
2323 .setattr = shmem_notify_change,
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002324 .truncate_range = shmem_truncate_range,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002325#ifdef CONFIG_TMPFS_POSIX_ACL
2326 .setxattr = generic_setxattr,
2327 .getxattr = generic_getxattr,
2328 .listxattr = generic_listxattr,
2329 .removexattr = generic_removexattr,
2330 .permission = shmem_permission,
2331#endif
2332
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333};
2334
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002335static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336#ifdef CONFIG_TMPFS
2337 .create = shmem_create,
2338 .lookup = simple_lookup,
2339 .link = shmem_link,
2340 .unlink = shmem_unlink,
2341 .symlink = shmem_symlink,
2342 .mkdir = shmem_mkdir,
2343 .rmdir = shmem_rmdir,
2344 .mknod = shmem_mknod,
2345 .rename = shmem_rename,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002347#ifdef CONFIG_TMPFS_POSIX_ACL
2348 .setattr = shmem_notify_change,
2349 .setxattr = generic_setxattr,
2350 .getxattr = generic_getxattr,
2351 .listxattr = generic_listxattr,
2352 .removexattr = generic_removexattr,
2353 .permission = shmem_permission,
2354#endif
2355};
2356
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002357static const struct inode_operations shmem_special_inode_operations = {
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002358#ifdef CONFIG_TMPFS_POSIX_ACL
2359 .setattr = shmem_notify_change,
2360 .setxattr = generic_setxattr,
2361 .getxattr = generic_getxattr,
2362 .listxattr = generic_listxattr,
2363 .removexattr = generic_removexattr,
2364 .permission = shmem_permission,
2365#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366};
2367
Hugh Dickins759b9772007-03-05 00:30:28 -08002368static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 .alloc_inode = shmem_alloc_inode,
2370 .destroy_inode = shmem_destroy_inode,
2371#ifdef CONFIG_TMPFS
2372 .statfs = shmem_statfs,
2373 .remount_fs = shmem_remount_fs,
2374#endif
2375 .delete_inode = shmem_delete_inode,
2376 .drop_inode = generic_delete_inode,
2377 .put_super = shmem_put_super,
2378};
2379
2380static struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002381 .fault = shmem_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382#ifdef CONFIG_NUMA
2383 .set_policy = shmem_set_policy,
2384 .get_policy = shmem_get_policy,
2385#endif
2386};
2387
2388
David Howells454e2392006-06-23 02:02:57 -07002389static int shmem_get_sb(struct file_system_type *fs_type,
2390 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391{
David Howells454e2392006-06-23 02:02:57 -07002392 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393}
2394
2395static struct file_system_type tmpfs_fs_type = {
2396 .owner = THIS_MODULE,
2397 .name = "tmpfs",
2398 .get_sb = shmem_get_sb,
2399 .kill_sb = kill_litter_super,
2400};
2401static struct vfsmount *shm_mnt;
2402
2403static int __init init_tmpfs(void)
2404{
2405 int error;
2406
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002407 error = bdi_init(&shmem_backing_dev_info);
2408 if (error)
2409 goto out4;
2410
Linus Torvalds1da177e2005-04-16 15:20:36 -07002411 error = init_inodecache();
2412 if (error)
2413 goto out3;
2414
2415 error = register_filesystem(&tmpfs_fs_type);
2416 if (error) {
2417 printk(KERN_ERR "Could not register tmpfs\n");
2418 goto out2;
2419 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07002420
Trond Myklebust1f5ce9e2006-06-09 09:34:16 -04002421 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422 tmpfs_fs_type.name, NULL);
2423 if (IS_ERR(shm_mnt)) {
2424 error = PTR_ERR(shm_mnt);
2425 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2426 goto out1;
2427 }
2428 return 0;
2429
2430out1:
2431 unregister_filesystem(&tmpfs_fs_type);
2432out2:
2433 destroy_inodecache();
2434out3:
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002435 bdi_destroy(&shmem_backing_dev_info);
2436out4:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 shm_mnt = ERR_PTR(error);
2438 return error;
2439}
2440module_init(init_tmpfs)
2441
2442/*
2443 * shmem_file_setup - get an unlinked file living in tmpfs
2444 *
2445 * @name: name for dentry (to be seen in /proc/<pid>/maps
2446 * @size: size to be set for the file
2447 *
2448 */
2449struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2450{
2451 int error;
2452 struct file *file;
2453 struct inode *inode;
2454 struct dentry *dentry, *root;
2455 struct qstr this;
2456
2457 if (IS_ERR(shm_mnt))
2458 return (void *)shm_mnt;
2459
2460 if (size < 0 || size > SHMEM_MAX_BYTES)
2461 return ERR_PTR(-EINVAL);
2462
2463 if (shmem_acct_size(flags, size))
2464 return ERR_PTR(-ENOMEM);
2465
2466 error = -ENOMEM;
2467 this.name = name;
2468 this.len = strlen(name);
2469 this.hash = 0; /* will go */
2470 root = shm_mnt->mnt_root;
2471 dentry = d_alloc(root, &this);
2472 if (!dentry)
2473 goto put_memory;
2474
2475 error = -ENFILE;
2476 file = get_empty_filp();
2477 if (!file)
2478 goto put_dentry;
2479
2480 error = -ENOSPC;
2481 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2482 if (!inode)
2483 goto close_file;
2484
2485 SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2486 d_instantiate(dentry, inode);
2487 inode->i_size = size;
2488 inode->i_nlink = 0; /* It is unlinked */
Dave Hansence8d2cd2007-10-16 23:31:13 -07002489 init_file(file, shm_mnt, dentry, FMODE_WRITE | FMODE_READ,
2490 &shmem_file_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491 return file;
2492
2493close_file:
2494 put_filp(file);
2495put_dentry:
2496 dput(dentry);
2497put_memory:
2498 shmem_unacct_size(flags, size);
2499 return ERR_PTR(error);
2500}
2501
2502/*
2503 * shmem_zero_setup - setup a shared anonymous mapping
2504 *
2505 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2506 */
2507int shmem_zero_setup(struct vm_area_struct *vma)
2508{
2509 struct file *file;
2510 loff_t size = vma->vm_end - vma->vm_start;
2511
2512 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2513 if (IS_ERR(file))
2514 return PTR_ERR(file);
2515
2516 if (vma->vm_file)
2517 fput(vma->vm_file);
2518 vma->vm_file = file;
2519 vma->vm_ops = &shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520 return 0;
2521}