blob: bd106361be4bf2d080e67f11252198c710745278 [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 Dickins6922c0c2011-08-03 16:21:25 -07009 * Copyright (C) 2002-2011 Hugh Dickins.
10 * Copyright (C) 2011 Google Inc.
Hugh Dickins0edd73b2005-06-21 17:15:04 -070011 * Copyright (C) 2002-2005 VERITAS Software Corporation.
Linus Torvalds1da177e2005-04-16 15:20:36 -070012 * Copyright (C) 2004 Andi Kleen, SuSE Labs
13 *
14 * Extended attribute support for tmpfs:
15 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
17 *
Matt Mackall853ac432009-01-06 14:40:20 -080018 * tiny-shmem:
19 * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
20 *
Linus Torvalds1da177e2005-04-16 15:20:36 -070021 * This file is released under the GPL.
22 */
23
Matt Mackall853ac432009-01-06 14:40:20 -080024#include <linux/fs.h>
25#include <linux/init.h>
26#include <linux/vfs.h>
27#include <linux/mount.h>
Hugh Dickinscaefba12009-04-13 14:40:12 -070028#include <linux/pagemap.h>
Matt Mackall853ac432009-01-06 14:40:20 -080029#include <linux/file.h>
30#include <linux/mm.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040031#include <linux/export.h>
Matt Mackall853ac432009-01-06 14:40:20 -080032#include <linux/swap.h>
33
34static struct vfsmount *shm_mnt;
35
36#ifdef CONFIG_SHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -070037/*
38 * This virtual memory filesystem is heavily based on the ramfs. It
39 * extends ramfs by the ability to use swap and honor resource limits
40 * which makes it a completely usable filesystem.
41 */
42
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070043#include <linux/xattr.h>
Christoph Hellwiga5694252007-07-17 04:04:28 -070044#include <linux/exportfs.h>
Christoph Hellwig1c7c4742009-11-03 16:44:44 +010045#include <linux/posix_acl.h>
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070046#include <linux/generic_acl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070047#include <linux/mman.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#include <linux/string.h>
49#include <linux/slab.h>
50#include <linux/backing-dev.h>
51#include <linux/shmem_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070053#include <linux/blkdev.h>
Hugh Dickinsbda97ea2011-08-03 16:21:21 -070054#include <linux/pagevec.h>
Hugh Dickins41ffe5d2011-08-03 16:21:21 -070055#include <linux/percpu_counter.h>
Hugh Dickins83e4fa92012-05-29 15:06:40 -070056#include <linux/falloc.h>
Hugh Dickins708e3502011-07-25 17:12:32 -070057#include <linux/splice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070058#include <linux/security.h>
59#include <linux/swapops.h>
60#include <linux/mempolicy.h>
61#include <linux/namei.h>
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +000062#include <linux/ctype.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070063#include <linux/migrate.h>
Christoph Lameterc1f60a52006-09-25 23:31:11 -070064#include <linux/highmem.h>
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -080065#include <linux/seq_file.h>
Mimi Zohar92562922008-10-07 14:00:12 -040066#include <linux/magic.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070067
Linus Torvalds1da177e2005-04-16 15:20:36 -070068#include <asm/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070069#include <asm/pgtable.h>
70
Linus Torvalds1da177e2005-04-16 15:20:36 -070071#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
Linus Torvalds1da177e2005-04-16 15:20:36 -070072#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
73
Linus Torvalds1da177e2005-04-16 15:20:36 -070074/* Pretend that each entry is of this size in directory's i_size */
75#define BOGO_DIRENT_SIZE 20
76
Hugh Dickins69f07ec2011-08-03 16:21:26 -070077/* Symlink up to this size is kmalloc'ed instead of using a swappable page */
78#define SHORT_SYMLINK_LEN 128
79
Eric Parisb09e0fa2011-05-24 17:12:39 -070080struct shmem_xattr {
81 struct list_head list; /* anchored by shmem_inode_info->xattr_list */
82 char *name; /* xattr name */
83 size_t size;
84 char value[0];
85};
86
Hugh Dickins1aac1402012-05-29 15:06:42 -070087/*
88 * shmem_fallocate and shmem_writepage communicate via inode->i_private
89 * (with i_mutex making sure that it has only one user at a time):
90 * we would prefer not to enlarge the shmem inode just for that.
91 */
92struct shmem_falloc {
93 pgoff_t start; /* start of range currently being fallocated */
94 pgoff_t next; /* the next page offset to be fallocated */
95 pgoff_t nr_falloced; /* how many new pages have been fallocated */
96 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
97};
98
Hugh Dickins285b2c42011-08-03 16:21:20 -070099/* Flag allocation requirements to shmem_getpage */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700100enum sgp_type {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101 SGP_READ, /* don't exceed i_size, don't allocate page */
102 SGP_CACHE, /* don't exceed i_size, may allocate page */
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -0800103 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700104 SGP_WRITE, /* may exceed i_size, may allocate !Uptodate page */
105 SGP_FALLOC, /* like SGP_WRITE, but make existing page Uptodate */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700106};
107
Andrew Mortonb76db732008-02-08 04:21:49 -0800108#ifdef CONFIG_TMPFS
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800109static unsigned long shmem_default_max_blocks(void)
110{
111 return totalram_pages / 2;
112}
113
114static unsigned long shmem_default_max_inodes(void)
115{
116 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
117}
Andrew Mortonb76db732008-02-08 04:21:49 -0800118#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800119
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700120static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
121static int shmem_replace_page(struct page **pagep, gfp_t gfp,
122 struct shmem_inode_info *info, pgoff_t index);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700123static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
124 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
125
126static inline int shmem_getpage(struct inode *inode, pgoff_t index,
127 struct page **pagep, enum sgp_type sgp, int *fault_type)
128{
129 return shmem_getpage_gfp(inode, index, pagep, sgp,
130 mapping_gfp_mask(inode->i_mapping), fault_type);
131}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
134{
135 return sb->s_fs_info;
136}
137
138/*
139 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
140 * for shared memory and for shared anonymous (/dev/zero) mappings
141 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
142 * consistent with the pre-accounting of private mappings ...
143 */
144static inline int shmem_acct_size(unsigned long flags, loff_t size)
145{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000146 return (flags & VM_NORESERVE) ?
Al Viro191c5422012-02-13 03:58:52 +0000147 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148}
149
150static inline void shmem_unacct_size(unsigned long flags, loff_t size)
151{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000152 if (!(flags & VM_NORESERVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 vm_unacct_memory(VM_ACCT(size));
154}
155
156/*
157 * ... whereas tmpfs objects are accounted incrementally as
158 * pages are allocated, in order to allow huge sparse files.
159 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
160 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
161 */
162static inline int shmem_acct_block(unsigned long flags)
163{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000164 return (flags & VM_NORESERVE) ?
Al Viro191c5422012-02-13 03:58:52 +0000165 security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166}
167
168static inline void shmem_unacct_blocks(unsigned long flags, long pages)
169{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000170 if (flags & VM_NORESERVE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
172}
173
Hugh Dickins759b9772007-03-05 00:30:28 -0800174static const struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700175static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800176static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800177static const struct inode_operations shmem_inode_operations;
178static const struct inode_operations shmem_dir_inode_operations;
179static const struct inode_operations shmem_special_inode_operations;
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400180static const struct vm_operations_struct shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700182static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183 .ra_pages = 0, /* No readahead */
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700184 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185};
186
187static LIST_HEAD(shmem_swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800188static DEFINE_MUTEX(shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800190static int shmem_reserve_inode(struct super_block *sb)
191{
192 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
193 if (sbinfo->max_inodes) {
194 spin_lock(&sbinfo->stat_lock);
195 if (!sbinfo->free_inodes) {
196 spin_unlock(&sbinfo->stat_lock);
197 return -ENOSPC;
198 }
199 sbinfo->free_inodes--;
200 spin_unlock(&sbinfo->stat_lock);
201 }
202 return 0;
203}
204
205static void shmem_free_inode(struct super_block *sb)
206{
207 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
208 if (sbinfo->max_inodes) {
209 spin_lock(&sbinfo->stat_lock);
210 sbinfo->free_inodes++;
211 spin_unlock(&sbinfo->stat_lock);
212 }
213}
214
Randy Dunlap46711812008-03-19 17:00:41 -0700215/**
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700216 * shmem_recalc_inode - recalculate the block usage of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700217 * @inode: inode to recalc
218 *
219 * We have to calculate the free blocks since the mm can drop
220 * undirtied hole pages behind our back.
221 *
222 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
223 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
224 *
225 * It has to be called with the spinlock held.
226 */
227static void shmem_recalc_inode(struct inode *inode)
228{
229 struct shmem_inode_info *info = SHMEM_I(inode);
230 long freed;
231
232 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
233 if (freed > 0) {
Hugh Dickins54af60422011-08-03 16:21:24 -0700234 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
235 if (sbinfo->max_blocks)
236 percpu_counter_add(&sbinfo->used_blocks, -freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 info->alloced -= freed;
Hugh Dickins54af60422011-08-03 16:21:24 -0700238 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239 shmem_unacct_blocks(info->flags, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240 }
241}
242
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700243/*
244 * Replace item expected in radix tree by a new item, while holding tree lock.
245 */
246static int shmem_radix_tree_replace(struct address_space *mapping,
247 pgoff_t index, void *expected, void *replacement)
248{
249 void **pslot;
250 void *item = NULL;
251
252 VM_BUG_ON(!expected);
253 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
254 if (pslot)
255 item = radix_tree_deref_slot_protected(pslot,
256 &mapping->tree_lock);
257 if (item != expected)
258 return -ENOENT;
259 if (replacement)
260 radix_tree_replace_slot(pslot, replacement);
261 else
262 radix_tree_delete(&mapping->page_tree, index);
263 return 0;
264}
265
266/*
Hugh Dickinsd1899222012-07-11 14:02:47 -0700267 * Sometimes, before we decide whether to proceed or to fail, we must check
268 * that an entry was not already brought back from swap by a racing thread.
269 *
270 * Checking page is not enough: by the time a SwapCache page is locked, it
271 * might be reused, and again be SwapCache, using the same swap as before.
272 */
273static bool shmem_confirm_swap(struct address_space *mapping,
274 pgoff_t index, swp_entry_t swap)
275{
276 void *item;
277
278 rcu_read_lock();
279 item = radix_tree_lookup(&mapping->page_tree, index);
280 rcu_read_unlock();
281 return item == swp_to_radix_entry(swap);
282}
283
284/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700285 * Like add_to_page_cache_locked, but error if expected item has gone.
286 */
287static int shmem_add_to_page_cache(struct page *page,
288 struct address_space *mapping,
289 pgoff_t index, gfp_t gfp, void *expected)
290{
Hugh Dickinsb065b432012-07-11 14:02:48 -0700291 int error;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700292
293 VM_BUG_ON(!PageLocked(page));
294 VM_BUG_ON(!PageSwapBacked(page));
295
Hugh Dickinsb065b432012-07-11 14:02:48 -0700296 page_cache_get(page);
297 page->mapping = mapping;
298 page->index = index;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700299
Hugh Dickinsb065b432012-07-11 14:02:48 -0700300 spin_lock_irq(&mapping->tree_lock);
301 if (!expected)
302 error = radix_tree_insert(&mapping->page_tree, index, page);
303 else
304 error = shmem_radix_tree_replace(mapping, index, expected,
305 page);
306 if (!error) {
307 mapping->nrpages++;
308 __inc_zone_page_state(page, NR_FILE_PAGES);
309 __inc_zone_page_state(page, NR_SHMEM);
310 spin_unlock_irq(&mapping->tree_lock);
311 } else {
312 page->mapping = NULL;
313 spin_unlock_irq(&mapping->tree_lock);
314 page_cache_release(page);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700315 }
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700316 return error;
317}
318
319/*
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700320 * Like delete_from_page_cache, but substitutes swap for page.
321 */
322static void shmem_delete_from_page_cache(struct page *page, void *radswap)
323{
324 struct address_space *mapping = page->mapping;
325 int error;
326
327 spin_lock_irq(&mapping->tree_lock);
328 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
329 page->mapping = NULL;
330 mapping->nrpages--;
331 __dec_zone_page_state(page, NR_FILE_PAGES);
332 __dec_zone_page_state(page, NR_SHMEM);
333 spin_unlock_irq(&mapping->tree_lock);
334 page_cache_release(page);
335 BUG_ON(error);
336}
337
338/*
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700339 * Like find_get_pages, but collecting swap entries as well as pages.
340 */
341static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping,
342 pgoff_t start, unsigned int nr_pages,
343 struct page **pages, pgoff_t *indices)
344{
345 unsigned int i;
346 unsigned int ret;
347 unsigned int nr_found;
348
349 rcu_read_lock();
350restart:
351 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
352 (void ***)pages, indices, start, nr_pages);
353 ret = 0;
354 for (i = 0; i < nr_found; i++) {
355 struct page *page;
356repeat:
357 page = radix_tree_deref_slot((void **)pages[i]);
358 if (unlikely(!page))
359 continue;
360 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700361 if (radix_tree_deref_retry(page))
362 goto restart;
363 /*
364 * Otherwise, we must be storing a swap entry
365 * here as an exceptional entry: so return it
366 * without attempting to raise page count.
367 */
368 goto export;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700369 }
370 if (!page_cache_get_speculative(page))
371 goto repeat;
372
373 /* Has the page moved? */
374 if (unlikely(page != *((void **)pages[i]))) {
375 page_cache_release(page);
376 goto repeat;
377 }
378export:
379 indices[ret] = indices[i];
380 pages[ret] = page;
381 ret++;
382 }
383 if (unlikely(!ret && nr_found))
384 goto restart;
385 rcu_read_unlock();
386 return ret;
387}
388
389/*
390 * Remove swap entry from radix tree, free the swap and its page cache.
391 */
392static int shmem_free_swap(struct address_space *mapping,
393 pgoff_t index, void *radswap)
394{
395 int error;
396
397 spin_lock_irq(&mapping->tree_lock);
398 error = shmem_radix_tree_replace(mapping, index, radswap, NULL);
399 spin_unlock_irq(&mapping->tree_lock);
400 if (!error)
401 free_swap_and_cache(radix_to_swp_entry(radswap));
402 return error;
403}
404
405/*
406 * Pagevec may contain swap entries, so shuffle up pages before releasing.
407 */
Hugh Dickins24513262012-01-20 14:34:21 -0800408static void shmem_deswap_pagevec(struct pagevec *pvec)
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700409{
410 int i, j;
411
412 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
413 struct page *page = pvec->pages[i];
414 if (!radix_tree_exceptional_entry(page))
415 pvec->pages[j++] = page;
416 }
417 pvec->nr = j;
Hugh Dickins24513262012-01-20 14:34:21 -0800418}
419
420/*
421 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
422 */
423void shmem_unlock_mapping(struct address_space *mapping)
424{
425 struct pagevec pvec;
426 pgoff_t indices[PAGEVEC_SIZE];
427 pgoff_t index = 0;
428
429 pagevec_init(&pvec, 0);
430 /*
431 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
432 */
433 while (!mapping_unevictable(mapping)) {
434 /*
435 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
436 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
437 */
438 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
439 PAGEVEC_SIZE, pvec.pages, indices);
440 if (!pvec.nr)
441 break;
442 index = indices[pvec.nr - 1] + 1;
443 shmem_deswap_pagevec(&pvec);
444 check_move_unevictable_pages(pvec.pages, pvec.nr);
445 pagevec_release(&pvec);
446 cond_resched();
447 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700448}
449
450/*
451 * Remove range of pages and swap entries from radix tree, and free them.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700452 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700453 */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700454static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
455 bool unfalloc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700457 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700459 pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700460 pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
461 unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
462 unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700463 struct pagevec pvec;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700464 pgoff_t indices[PAGEVEC_SIZE];
465 long nr_swaps_freed = 0;
Hugh Dickins285b2c42011-08-03 16:21:20 -0700466 pgoff_t index;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700467 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700469 if (lend == -1)
470 end = -1; /* unsigned, so actually very big */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700471
472 pagevec_init(&pvec, 0);
473 index = start;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700474 while (index < end) {
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700475 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700476 min(end - index, (pgoff_t)PAGEVEC_SIZE),
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700477 pvec.pages, indices);
478 if (!pvec.nr)
479 break;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700480 mem_cgroup_uncharge_start();
481 for (i = 0; i < pagevec_count(&pvec); i++) {
482 struct page *page = pvec.pages[i];
483
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700484 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700485 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700486 break;
487
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700488 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700489 if (unfalloc)
490 continue;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700491 nr_swaps_freed += !shmem_free_swap(mapping,
492 index, page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700493 continue;
494 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700495
496 if (!trylock_page(page))
497 continue;
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700498 if (!unfalloc || !PageUptodate(page)) {
499 if (page->mapping == mapping) {
500 VM_BUG_ON(PageWriteback(page));
501 truncate_inode_page(mapping, page);
502 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700503 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700504 unlock_page(page);
505 }
Hugh Dickins24513262012-01-20 14:34:21 -0800506 shmem_deswap_pagevec(&pvec);
507 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700508 mem_cgroup_uncharge_end();
509 cond_resched();
510 index++;
511 }
512
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700513 if (partial_start) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700514 struct page *page = NULL;
515 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
516 if (page) {
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700517 unsigned int top = PAGE_CACHE_SIZE;
518 if (start > end) {
519 top = partial_end;
520 partial_end = 0;
521 }
522 zero_user_segment(page, partial_start, top);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700523 set_page_dirty(page);
524 unlock_page(page);
525 page_cache_release(page);
526 }
527 }
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700528 if (partial_end) {
529 struct page *page = NULL;
530 shmem_getpage(inode, end, &page, SGP_READ, NULL);
531 if (page) {
532 zero_user_segment(page, 0, partial_end);
533 set_page_dirty(page);
534 unlock_page(page);
535 page_cache_release(page);
536 }
537 }
538 if (start >= end)
539 return;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700540
541 index = start;
542 for ( ; ; ) {
543 cond_resched();
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700544 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700545 min(end - index, (pgoff_t)PAGEVEC_SIZE),
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700546 pvec.pages, indices);
547 if (!pvec.nr) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700548 if (index == start || unfalloc)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700549 break;
550 index = start;
551 continue;
552 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700553 if ((index == start || unfalloc) && indices[0] >= end) {
Hugh Dickins24513262012-01-20 14:34:21 -0800554 shmem_deswap_pagevec(&pvec);
555 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700556 break;
557 }
558 mem_cgroup_uncharge_start();
559 for (i = 0; i < pagevec_count(&pvec); i++) {
560 struct page *page = pvec.pages[i];
561
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700562 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700563 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700564 break;
565
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700566 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700567 if (unfalloc)
568 continue;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700569 nr_swaps_freed += !shmem_free_swap(mapping,
570 index, page);
571 continue;
572 }
573
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700574 lock_page(page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700575 if (!unfalloc || !PageUptodate(page)) {
576 if (page->mapping == mapping) {
577 VM_BUG_ON(PageWriteback(page));
578 truncate_inode_page(mapping, page);
579 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700580 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700581 unlock_page(page);
582 }
Hugh Dickins24513262012-01-20 14:34:21 -0800583 shmem_deswap_pagevec(&pvec);
584 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700585 mem_cgroup_uncharge_end();
586 index++;
587 }
Hugh Dickins94c1e622011-06-27 16:18:03 -0700588
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 spin_lock(&info->lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700590 info->swapped -= nr_swaps_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 shmem_recalc_inode(inode);
592 spin_unlock(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700593}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700595void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
596{
597 shmem_undo_range(inode, lstart, lend, false);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700598 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599}
Hugh Dickins94c1e622011-06-27 16:18:03 -0700600EXPORT_SYMBOL_GPL(shmem_truncate_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601
Hugh Dickins94c1e622011-06-27 16:18:03 -0700602static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603{
604 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 int error;
606
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200607 error = inode_change_ok(inode, attr);
608 if (error)
609 return error;
610
Hugh Dickins94c1e622011-06-27 16:18:03 -0700611 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
612 loff_t oldsize = inode->i_size;
613 loff_t newsize = attr->ia_size;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000614
Hugh Dickins94c1e622011-06-27 16:18:03 -0700615 if (newsize != oldsize) {
616 i_size_write(inode, newsize);
617 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
618 }
619 if (newsize < oldsize) {
620 loff_t holebegin = round_up(newsize, PAGE_SIZE);
621 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
622 shmem_truncate_range(inode, newsize, (loff_t)-1);
623 /* unmap again to remove racily COWed private pages */
624 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
625 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 }
627
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200628 setattr_copy(inode, attr);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700629#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200630 if (attr->ia_valid & ATTR_MODE)
Christoph Hellwig1c7c4742009-11-03 16:44:44 +0100631 error = generic_acl_chmod(inode);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700632#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 return error;
634}
635
Al Viro1f895f72010-06-05 19:10:41 -0400636static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 struct shmem_inode_info *info = SHMEM_I(inode);
Eric Parisb09e0fa2011-05-24 17:12:39 -0700639 struct shmem_xattr *xattr, *nxattr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000641 if (inode->i_mapping->a_ops == &shmem_aops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 shmem_unacct_size(info->flags, inode->i_size);
643 inode->i_size = 0;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000644 shmem_truncate_range(inode, 0, (loff_t)-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800646 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800648 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 }
Hugh Dickins69f07ec2011-08-03 16:21:26 -0700650 } else
651 kfree(info->symlink);
Eric Parisb09e0fa2011-05-24 17:12:39 -0700652
653 list_for_each_entry_safe(xattr, nxattr, &info->xattr_list, list) {
654 kfree(xattr->name);
655 kfree(xattr);
656 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700657 BUG_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800658 shmem_free_inode(inode->i_sb);
Jan Karadbd57682012-05-03 14:48:02 +0200659 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700660}
661
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700662/*
663 * If swap found in inode, free it and move page from swapcache to filecache.
664 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700665static int shmem_unuse_inode(struct shmem_inode_info *info,
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700666 swp_entry_t swap, struct page **pagep)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700668 struct address_space *mapping = info->vfs_inode.i_mapping;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700669 void *radswap;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700670 pgoff_t index;
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700671 gfp_t gfp;
672 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700674 radswap = swp_to_radix_entry(swap);
Hugh Dickinse504f3f2011-08-03 16:21:27 -0700675 index = radix_tree_locate_item(&mapping->page_tree, radswap);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700676 if (index == -1)
Hugh Dickins285b2c42011-08-03 16:21:20 -0700677 return 0;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800678
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800679 /*
680 * Move _head_ to start search for next from here.
Al Viro1f895f72010-06-05 19:10:41 -0400681 * But be careful: shmem_evict_inode checks list_empty without taking
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800682 * mutex, and there's an instant in list_move_tail when info->swaplist
Hugh Dickins285b2c42011-08-03 16:21:20 -0700683 * would appear empty, if it were the only one on shmem_swaplist.
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800684 */
685 if (shmem_swaplist.next != &info->swaplist)
686 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800687
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700688 gfp = mapping_gfp_mask(mapping);
689 if (shmem_should_replace_page(*pagep, gfp)) {
690 mutex_unlock(&shmem_swaplist_mutex);
691 error = shmem_replace_page(pagep, gfp, info, index);
692 mutex_lock(&shmem_swaplist_mutex);
693 /*
694 * We needed to drop mutex to make that restrictive page
Hugh Dickins0142ef62012-06-07 14:21:09 -0700695 * allocation, but the inode might have been freed while we
696 * dropped it: although a racing shmem_evict_inode() cannot
697 * complete without emptying the radix_tree, our page lock
698 * on this swapcache page is not enough to prevent that -
699 * free_swap_and_cache() of our swap entry will only
700 * trylock_page(), removing swap from radix_tree whatever.
701 *
702 * We must not proceed to shmem_add_to_page_cache() if the
703 * inode has been freed, but of course we cannot rely on
704 * inode or mapping or info to check that. However, we can
705 * safely check if our swap entry is still in use (and here
706 * it can't have got reused for another page): if it's still
707 * in use, then the inode cannot have been freed yet, and we
708 * can safely proceed (if it's no longer in use, that tells
709 * nothing about the inode, but we don't need to unuse swap).
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700710 */
711 if (!page_swapcount(*pagep))
712 error = -ENOENT;
713 }
714
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800715 /*
Hugh Dickins778dd892011-05-11 15:13:37 -0700716 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
717 * but also to hold up shmem_evict_inode(): so inode cannot be freed
718 * beneath us (pagelock doesn't help until the page is in pagecache).
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800719 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700720 if (!error)
721 error = shmem_add_to_page_cache(*pagep, mapping, index,
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700722 GFP_NOWAIT, radswap);
Hugh Dickins48f170f2011-07-25 17:12:37 -0700723 if (error != -ENOMEM) {
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700724 /*
725 * Truncation and eviction use free_swap_and_cache(), which
726 * only does trylock page: if we raced, best clean up here.
727 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700728 delete_from_swap_cache(*pagep);
729 set_page_dirty(*pagep);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700730 if (!error) {
731 spin_lock(&info->lock);
732 info->swapped--;
733 spin_unlock(&info->lock);
734 swap_free(swap);
735 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800736 error = 1; /* not an error, but entry was found */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800738 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700739}
740
741/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700742 * Search through swapped inodes to find and replace swap by page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700744int shmem_unuse(swp_entry_t swap, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700746 struct list_head *this, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 struct shmem_inode_info *info;
748 int found = 0;
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700749 int error = 0;
750
751 /*
752 * There's a faint possibility that swap page was replaced before
Hugh Dickins0142ef62012-06-07 14:21:09 -0700753 * caller locked it: caller will come back later with the right page.
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700754 */
Hugh Dickins0142ef62012-06-07 14:21:09 -0700755 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700756 goto out;
Hugh Dickins778dd892011-05-11 15:13:37 -0700757
758 /*
759 * Charge page using GFP_KERNEL while we can wait, before taking
760 * the shmem_swaplist_mutex which might hold up shmem_writepage().
761 * Charged back to the user (not to caller) when swap account is used.
Hugh Dickins778dd892011-05-11 15:13:37 -0700762 */
763 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
764 if (error)
765 goto out;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700766 /* No radix_tree_preload: swap entry keeps a place for page in tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800768 mutex_lock(&shmem_swaplist_mutex);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700769 list_for_each_safe(this, next, &shmem_swaplist) {
770 info = list_entry(this, struct shmem_inode_info, swaplist);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700771 if (info->swapped)
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700772 found = shmem_unuse_inode(info, swap, &page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700773 else
774 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800775 cond_resched();
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800776 if (found)
Hugh Dickins778dd892011-05-11 15:13:37 -0700777 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800779 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins778dd892011-05-11 15:13:37 -0700780
Hugh Dickins778dd892011-05-11 15:13:37 -0700781 if (found < 0)
782 error = found;
783out:
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800784 unlock_page(page);
785 page_cache_release(page);
Hugh Dickins778dd892011-05-11 15:13:37 -0700786 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787}
788
789/*
790 * Move the page from the page cache to the swap cache.
791 */
792static int shmem_writepage(struct page *page, struct writeback_control *wbc)
793{
794 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796 struct inode *inode;
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700797 swp_entry_t swap;
798 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799
800 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 mapping = page->mapping;
802 index = page->index;
803 inode = mapping->host;
804 info = SHMEM_I(inode);
805 if (info->flags & VM_LOCKED)
806 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800807 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808 goto redirty;
809
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800810 /*
811 * shmem_backing_dev_info's capabilities prevent regular writeback or
812 * sync from ever calling shmem_writepage; but a stacking filesystem
Hugh Dickins48f170f2011-07-25 17:12:37 -0700813 * might use ->writepage of its underlying filesystem, in which case
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800814 * tmpfs should write out to swap only in response to memory pressure,
Hugh Dickins48f170f2011-07-25 17:12:37 -0700815 * and not for the writeback threads or sync.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800816 */
Hugh Dickins48f170f2011-07-25 17:12:37 -0700817 if (!wbc->for_reclaim) {
818 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
819 goto redirty;
820 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700821
822 /*
823 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
824 * value into swapfile.c, the only way we can correctly account for a
825 * fallocated page arriving here is now to initialize it and write it.
Hugh Dickins1aac1402012-05-29 15:06:42 -0700826 *
827 * That's okay for a page already fallocated earlier, but if we have
828 * not yet completed the fallocation, then (a) we want to keep track
829 * of this page in case we have to undo it, and (b) it may not be a
830 * good idea to continue anyway, once we're pushing into swap. So
831 * reactivate the page, and let shmem_fallocate() quit when too many.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700832 */
833 if (!PageUptodate(page)) {
Hugh Dickins1aac1402012-05-29 15:06:42 -0700834 if (inode->i_private) {
835 struct shmem_falloc *shmem_falloc;
836 spin_lock(&inode->i_lock);
837 shmem_falloc = inode->i_private;
838 if (shmem_falloc &&
839 index >= shmem_falloc->start &&
840 index < shmem_falloc->next)
841 shmem_falloc->nr_unswapped++;
842 else
843 shmem_falloc = NULL;
844 spin_unlock(&inode->i_lock);
845 if (shmem_falloc)
846 goto redirty;
847 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700848 clear_highpage(page);
849 flush_dcache_page(page);
850 SetPageUptodate(page);
851 }
852
Hugh Dickins48f170f2011-07-25 17:12:37 -0700853 swap = get_swap_page();
854 if (!swap.val)
855 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800856
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700857 /*
858 * Add inode to shmem_unuse()'s list of swapped-out inodes,
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700859 * if it's not already there. Do it now before the page is
860 * moved to swap cache, when its pagelock no longer protects
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700861 * the inode from eviction. But don't unlock the mutex until
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700862 * we've incremented swapped, because shmem_unuse_inode() will
863 * prune a !swapped inode from the swaplist under this mutex.
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700864 */
Hugh Dickins48f170f2011-07-25 17:12:37 -0700865 mutex_lock(&shmem_swaplist_mutex);
866 if (list_empty(&info->swaplist))
867 list_add_tail(&info->swaplist, &shmem_swaplist);
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700868
Hugh Dickins48f170f2011-07-25 17:12:37 -0700869 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800870 swap_shmem_alloc(swap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700871 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
872
873 spin_lock(&info->lock);
874 info->swapped++;
875 shmem_recalc_inode(inode);
Hugh Dickins826267c2011-05-28 13:14:09 -0700876 spin_unlock(&info->lock);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700877
878 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800879 BUG_ON(page_mapped(page));
Hugh Dickins9fab5612009-03-31 15:23:33 -0700880 swap_writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700881 return 0;
882 }
883
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700884 mutex_unlock(&shmem_swaplist_mutex);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700885 swapcache_free(swap, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886redirty:
887 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800888 if (wbc->for_reclaim)
889 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
890 unlock_page(page);
891 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700892}
893
894#ifdef CONFIG_NUMA
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800895#ifdef CONFIG_TMPFS
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700896static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800897{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700898 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800899
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700900 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700901 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800902
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700903 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800904
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700905 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800906}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700907
908static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
909{
910 struct mempolicy *mpol = NULL;
911 if (sbinfo->mpol) {
912 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
913 mpol = sbinfo->mpol;
914 mpol_get(mpol);
915 spin_unlock(&sbinfo->stat_lock);
916 }
917 return mpol;
918}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800919#endif /* CONFIG_TMPFS */
920
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700921static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
922 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923{
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700924 struct mempolicy mpol, *spol;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700925 struct vm_area_struct pvma;
926
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700927 spol = mpol_cond_copy(&mpol,
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700928 mpol_shared_policy_lookup(&info->policy, index));
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700929
Linus Torvalds1da177e2005-04-16 15:20:36 -0700930 /* Create a pseudo vma that just contains the policy */
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800931 pvma.vm_start = 0;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700932 pvma.vm_pgoff = index;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800933 pvma.vm_ops = NULL;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700934 pvma.vm_policy = spol;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700935 return swapin_readahead(swap, gfp, &pvma, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936}
937
Hugh Dickins02098fe2008-02-04 22:28:42 -0800938static struct page *shmem_alloc_page(gfp_t gfp,
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700939 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700940{
941 struct vm_area_struct pvma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700942
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800943 /* Create a pseudo vma that just contains the policy */
944 pvma.vm_start = 0;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700945 pvma.vm_pgoff = index;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800946 pvma.vm_ops = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700947 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700948
949 /*
950 * alloc_page_vma() will drop the shared policy reference
951 */
952 return alloc_page_vma(gfp, &pvma, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800954#else /* !CONFIG_NUMA */
955#ifdef CONFIG_TMPFS
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700956static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800957{
958}
959#endif /* CONFIG_TMPFS */
960
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700961static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
962 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700964 return swapin_readahead(swap, gfp, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965}
966
Hugh Dickins02098fe2008-02-04 22:28:42 -0800967static inline struct page *shmem_alloc_page(gfp_t gfp,
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700968 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969{
Hugh Dickinse84e2e12007-11-28 18:55:10 +0000970 return alloc_page(gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800972#endif /* CONFIG_NUMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700974#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
975static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
976{
977 return NULL;
978}
979#endif
980
Linus Torvalds1da177e2005-04-16 15:20:36 -0700981/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700982 * When a page is moved from swapcache to shmem filecache (either by the
983 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
984 * shmem_unuse_inode()), it may have been read in earlier from swap, in
985 * ignorance of the mapping it belongs to. If that mapping has special
986 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
987 * we may need to copy to a suitable page before moving to filecache.
988 *
989 * In a future release, this may well be extended to respect cpuset and
990 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
991 * but for now it is a simple matter of zone.
992 */
993static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
994{
995 return page_zonenum(page) > gfp_zone(gfp);
996}
997
998static int shmem_replace_page(struct page **pagep, gfp_t gfp,
999 struct shmem_inode_info *info, pgoff_t index)
1000{
1001 struct page *oldpage, *newpage;
1002 struct address_space *swap_mapping;
1003 pgoff_t swap_index;
1004 int error;
1005
1006 oldpage = *pagep;
1007 swap_index = page_private(oldpage);
1008 swap_mapping = page_mapping(oldpage);
1009
1010 /*
1011 * We have arrived here because our zones are constrained, so don't
1012 * limit chance of success by further cpuset and node constraints.
1013 */
1014 gfp &= ~GFP_CONSTRAINT_MASK;
1015 newpage = shmem_alloc_page(gfp, info, index);
1016 if (!newpage)
1017 return -ENOMEM;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001018
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001019 page_cache_get(newpage);
1020 copy_highpage(newpage, oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001021 flush_dcache_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001022
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001023 __set_page_locked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001024 SetPageUptodate(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001025 SetPageSwapBacked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001026 set_page_private(newpage, swap_index);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001027 SetPageSwapCache(newpage);
1028
1029 /*
1030 * Our caller will very soon move newpage out of swapcache, but it's
1031 * a nice clean interface for us to replace oldpage by newpage there.
1032 */
1033 spin_lock_irq(&swap_mapping->tree_lock);
1034 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1035 newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001036 if (!error) {
1037 __inc_zone_page_state(newpage, NR_FILE_PAGES);
1038 __dec_zone_page_state(oldpage, NR_FILE_PAGES);
1039 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001040 spin_unlock_irq(&swap_mapping->tree_lock);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001041
Hugh Dickins0142ef62012-06-07 14:21:09 -07001042 if (unlikely(error)) {
1043 /*
1044 * Is this possible? I think not, now that our callers check
1045 * both PageSwapCache and page_private after getting page lock;
1046 * but be defensive. Reverse old to newpage for clear and free.
1047 */
1048 oldpage = newpage;
1049 } else {
1050 mem_cgroup_replace_page_cache(oldpage, newpage);
1051 lru_cache_add_anon(newpage);
1052 *pagep = newpage;
1053 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001054
1055 ClearPageSwapCache(oldpage);
1056 set_page_private(oldpage, 0);
1057
1058 unlock_page(oldpage);
1059 page_cache_release(oldpage);
1060 page_cache_release(oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001061 return error;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001062}
1063
1064/*
Hugh Dickins68da9f02011-07-25 17:12:34 -07001065 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 *
1067 * If we allocate a new one we do not mark it dirty. That's up to the
1068 * vm. If we swap it in we mark it dirty since we also free the swap
1069 * entry since a page cannot live in both the swap and page cache
1070 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001071static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Hugh Dickins68da9f02011-07-25 17:12:34 -07001072 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073{
1074 struct address_space *mapping = inode->i_mapping;
Hugh Dickins54af60422011-08-03 16:21:24 -07001075 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 struct shmem_sb_info *sbinfo;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001077 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 swp_entry_t swap;
1079 int error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001080 int once = 0;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001081 int alloced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001083 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085repeat:
Hugh Dickins54af60422011-08-03 16:21:24 -07001086 swap.val = 0;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001087 page = find_lock_page(mapping, index);
Hugh Dickins54af60422011-08-03 16:21:24 -07001088 if (radix_tree_exceptional_entry(page)) {
1089 swap = radix_to_swp_entry(page);
1090 page = NULL;
1091 }
1092
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001093 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
Hugh Dickins54af60422011-08-03 16:21:24 -07001094 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1095 error = -EINVAL;
1096 goto failed;
1097 }
1098
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001099 /* fallocated page? */
1100 if (page && !PageUptodate(page)) {
1101 if (sgp != SGP_READ)
1102 goto clear;
1103 unlock_page(page);
1104 page_cache_release(page);
1105 page = NULL;
1106 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001107 if (page || (sgp == SGP_READ && !swap.val)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001108 *pagep = page;
1109 return 0;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001110 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111
1112 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001113 * Fast cache lookup did not find it:
1114 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001116 info = SHMEM_I(inode);
1117 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001118
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 if (swap.val) {
1120 /* Look it up and read it in.. */
Hugh Dickins27ab7002011-07-25 17:12:36 -07001121 page = lookup_swap_cache(swap);
1122 if (!page) {
Ying Han456f9982011-05-26 16:25:38 -07001123 /* here we actually do the io */
Hugh Dickins68da9f02011-07-25 17:12:34 -07001124 if (fault_type)
1125 *fault_type |= VM_FAULT_MAJOR;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001126 page = shmem_swapin(swap, gfp, info, index);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001127 if (!page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001128 error = -ENOMEM;
1129 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 }
1132
1133 /* We have to do this with page locked to prevent races */
Hugh Dickins54af60422011-08-03 16:21:24 -07001134 lock_page(page);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001135 if (!PageSwapCache(page) || page_private(page) != swap.val ||
Hugh Dickinsd1899222012-07-11 14:02:47 -07001136 !shmem_confirm_swap(mapping, index, swap)) {
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001137 error = -EEXIST; /* try again */
Hugh Dickinsd1899222012-07-11 14:02:47 -07001138 goto unlock;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001139 }
Hugh Dickins27ab7002011-07-25 17:12:36 -07001140 if (!PageUptodate(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141 error = -EIO;
Hugh Dickins54af60422011-08-03 16:21:24 -07001142 goto failed;
1143 }
1144 wait_on_page_writeback(page);
1145
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001146 if (shmem_should_replace_page(page, gfp)) {
1147 error = shmem_replace_page(&page, gfp, info, index);
1148 if (error)
1149 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150 }
1151
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001152 error = mem_cgroup_cache_charge(page, current->mm,
1153 gfp & GFP_RECLAIM_MASK);
Hugh Dickinsd1899222012-07-11 14:02:47 -07001154 if (!error) {
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001155 error = shmem_add_to_page_cache(page, mapping, index,
1156 gfp, swp_to_radix_entry(swap));
Hugh Dickinsd1899222012-07-11 14:02:47 -07001157 /* We already confirmed swap, and make no allocation */
1158 VM_BUG_ON(error);
1159 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001160 if (error)
1161 goto failed;
1162
1163 spin_lock(&info->lock);
1164 info->swapped--;
1165 shmem_recalc_inode(inode);
1166 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001167
Hugh Dickins27ab7002011-07-25 17:12:36 -07001168 delete_from_swap_cache(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001169 set_page_dirty(page);
1170 swap_free(swap);
1171
Hugh Dickins54af60422011-08-03 16:21:24 -07001172 } else {
1173 if (shmem_acct_block(info->flags)) {
1174 error = -ENOSPC;
1175 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001177 if (sbinfo->max_blocks) {
Hugh Dickinsfc5da222011-04-14 15:22:07 -07001178 if (percpu_counter_compare(&sbinfo->used_blocks,
Hugh Dickins54af60422011-08-03 16:21:24 -07001179 sbinfo->max_blocks) >= 0) {
1180 error = -ENOSPC;
1181 goto unacct;
1182 }
Tim Chen7e496292010-08-09 17:19:05 -07001183 percpu_counter_inc(&sbinfo->used_blocks);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001184 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001185
Hugh Dickins54af60422011-08-03 16:21:24 -07001186 page = shmem_alloc_page(gfp, info, index);
1187 if (!page) {
1188 error = -ENOMEM;
1189 goto decused;
1190 }
1191
1192 SetPageSwapBacked(page);
1193 __set_page_locked(page);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001194 error = mem_cgroup_cache_charge(page, current->mm,
1195 gfp & GFP_RECLAIM_MASK);
Hugh Dickins54af60422011-08-03 16:21:24 -07001196 if (error)
1197 goto decused;
Hugh Dickinsb065b432012-07-11 14:02:48 -07001198 error = radix_tree_preload(gfp & GFP_RECLAIM_MASK);
1199 if (!error) {
1200 error = shmem_add_to_page_cache(page, mapping, index,
1201 gfp, NULL);
1202 radix_tree_preload_end();
1203 }
1204 if (error) {
1205 mem_cgroup_uncharge_cache_page(page);
1206 goto decused;
1207 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001208 lru_cache_add_anon(page);
1209
1210 spin_lock(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 info->alloced++;
Hugh Dickins54af60422011-08-03 16:21:24 -07001212 inode->i_blocks += BLOCKS_PER_PAGE;
1213 shmem_recalc_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 spin_unlock(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001215 alloced = true;
Hugh Dickins54af60422011-08-03 16:21:24 -07001216
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001217 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001218 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1219 */
1220 if (sgp == SGP_FALLOC)
1221 sgp = SGP_WRITE;
1222clear:
1223 /*
1224 * Let SGP_WRITE caller clear ends if write does not fill page;
1225 * but SGP_FALLOC on a page fallocated earlier must initialize
1226 * it now, lest undo on failure cancel our earlier guarantee.
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001227 */
1228 if (sgp != SGP_WRITE) {
1229 clear_highpage(page);
1230 flush_dcache_page(page);
1231 SetPageUptodate(page);
1232 }
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001233 if (sgp == SGP_DIRTY)
Hugh Dickins27ab7002011-07-25 17:12:36 -07001234 set_page_dirty(page);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001235 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001236
Hugh Dickins54af60422011-08-03 16:21:24 -07001237 /* Perhaps the file has been truncated since we checked */
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001238 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
Hugh Dickins54af60422011-08-03 16:21:24 -07001239 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1240 error = -EINVAL;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001241 if (alloced)
1242 goto trunc;
1243 else
1244 goto failed;
Shaohua Liff36b8012010-08-09 17:19:06 -07001245 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001246 *pagep = page;
1247 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07001248
Nick Piggind0217ac2007-07-19 01:47:03 -07001249 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001250 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001251 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001252trunc:
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001253 info = SHMEM_I(inode);
Hugh Dickins54af60422011-08-03 16:21:24 -07001254 ClearPageDirty(page);
1255 delete_from_page_cache(page);
1256 spin_lock(&info->lock);
1257 info->alloced--;
1258 inode->i_blocks -= BLOCKS_PER_PAGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 spin_unlock(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001260decused:
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001261 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins54af60422011-08-03 16:21:24 -07001262 if (sbinfo->max_blocks)
1263 percpu_counter_add(&sbinfo->used_blocks, -1);
1264unacct:
1265 shmem_unacct_blocks(info->flags, 1);
1266failed:
Hugh Dickinsd1899222012-07-11 14:02:47 -07001267 if (swap.val && error != -EINVAL &&
1268 !shmem_confirm_swap(mapping, index, swap))
1269 error = -EEXIST;
1270unlock:
Hugh Dickins27ab7002011-07-25 17:12:36 -07001271 if (page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001272 unlock_page(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001273 page_cache_release(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001274 }
1275 if (error == -ENOSPC && !once++) {
1276 info = SHMEM_I(inode);
1277 spin_lock(&info->lock);
1278 shmem_recalc_inode(inode);
1279 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001280 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001281 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001282 if (error == -EEXIST) /* from above or from radix_tree_insert */
Hugh Dickins54af60422011-08-03 16:21:24 -07001283 goto repeat;
1284 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285}
1286
1287static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1288{
1289 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1290 int error;
Hugh Dickins68da9f02011-07-25 17:12:34 -07001291 int ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001292
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1294 if (error)
1295 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Hugh Dickins68da9f02011-07-25 17:12:34 -07001296
Ying Han456f9982011-05-26 16:25:38 -07001297 if (ret & VM_FAULT_MAJOR) {
1298 count_vm_event(PGMAJFAULT);
1299 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1300 }
Hugh Dickins68da9f02011-07-25 17:12:34 -07001301 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302}
1303
1304#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001305static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001306{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001307 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1308 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309}
1310
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001311static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1312 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001314 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1315 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001317 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1318 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319}
1320#endif
1321
1322int shmem_lock(struct file *file, int lock, struct user_struct *user)
1323{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001324 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 struct shmem_inode_info *info = SHMEM_I(inode);
1326 int retval = -ENOMEM;
1327
1328 spin_lock(&info->lock);
1329 if (lock && !(info->flags & VM_LOCKED)) {
1330 if (!user_shm_lock(inode->i_size, user))
1331 goto out_nomem;
1332 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001333 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 }
1335 if (!lock && (info->flags & VM_LOCKED) && user) {
1336 user_shm_unlock(inode->i_size, user);
1337 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001338 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 }
1340 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001341
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342out_nomem:
1343 spin_unlock(&info->lock);
1344 return retval;
1345}
1346
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08001347static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348{
1349 file_accessed(file);
1350 vma->vm_ops = &shmem_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001351 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 return 0;
1353}
1354
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001355static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04001356 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357{
1358 struct inode *inode;
1359 struct shmem_inode_info *info;
1360 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1361
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001362 if (shmem_reserve_inode(sb))
1363 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364
1365 inode = new_inode(sb);
1366 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -04001367 inode->i_ino = get_next_ino();
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001368 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 inode->i_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001370 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1371 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
David M. Grimes91828a42006-10-17 00:09:45 -07001372 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 info = SHMEM_I(inode);
1374 memset(info, 0, (char *)inode - (char *)info);
1375 spin_lock_init(&info->lock);
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001376 info->flags = flags & VM_NORESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 INIT_LIST_HEAD(&info->swaplist);
Eric Parisb09e0fa2011-05-24 17:12:39 -07001378 INIT_LIST_HEAD(&info->xattr_list);
Al Viro72c04902009-06-24 16:58:48 -04001379 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380
1381 switch (mode & S_IFMT) {
1382 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001383 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384 init_special_inode(inode, mode, dev);
1385 break;
1386 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07001387 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 inode->i_op = &shmem_inode_operations;
1389 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001390 mpol_shared_policy_init(&info->policy,
1391 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 break;
1393 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07001394 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 /* Some things misbehave if size == 0 on a directory */
1396 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1397 inode->i_op = &shmem_dir_inode_operations;
1398 inode->i_fop = &simple_dir_operations;
1399 break;
1400 case S_IFLNK:
1401 /*
1402 * Must not load anything in the rbtree,
1403 * mpol_free_shared_policy will not be called.
1404 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001405 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 break;
1407 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001408 } else
1409 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410 return inode;
1411}
1412
1413#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001414static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07001415static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001417#ifdef CONFIG_TMPFS_XATTR
1418static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1419#else
1420#define shmem_initxattrs NULL
1421#endif
1422
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423static int
Nick Piggin800d15a2007-10-16 01:25:03 -07001424shmem_write_begin(struct file *file, struct address_space *mapping,
1425 loff_t pos, unsigned len, unsigned flags,
1426 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427{
Nick Piggin800d15a2007-10-16 01:25:03 -07001428 struct inode *inode = mapping->host;
1429 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
Nick Piggin800d15a2007-10-16 01:25:03 -07001430 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1431}
1432
1433static int
1434shmem_write_end(struct file *file, struct address_space *mapping,
1435 loff_t pos, unsigned len, unsigned copied,
1436 struct page *page, void *fsdata)
1437{
1438 struct inode *inode = mapping->host;
1439
Hugh Dickinsd3602442008-02-04 22:28:44 -08001440 if (pos + copied > inode->i_size)
1441 i_size_write(inode, pos + copied);
1442
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001443 if (!PageUptodate(page)) {
1444 if (copied < PAGE_CACHE_SIZE) {
1445 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1446 zero_user_segments(page, 0, from,
1447 from + copied, PAGE_CACHE_SIZE);
1448 }
1449 SetPageUptodate(page);
1450 }
Nick Piggin800d15a2007-10-16 01:25:03 -07001451 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02001452 unlock_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07001453 page_cache_release(page);
1454
Nick Piggin800d15a2007-10-16 01:25:03 -07001455 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001456}
1457
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1459{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001460 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001462 pgoff_t index;
1463 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001464 enum sgp_type sgp = SGP_READ;
1465
1466 /*
1467 * Might this read be for a stacking filesystem? Then when reading
1468 * holes of a sparse file, we actually need to allocate those pages,
1469 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1470 */
1471 if (segment_eq(get_fs(), KERNEL_DS))
1472 sgp = SGP_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473
1474 index = *ppos >> PAGE_CACHE_SHIFT;
1475 offset = *ppos & ~PAGE_CACHE_MASK;
1476
1477 for (;;) {
1478 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001479 pgoff_t end_index;
1480 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 loff_t i_size = i_size_read(inode);
1482
1483 end_index = i_size >> PAGE_CACHE_SHIFT;
1484 if (index > end_index)
1485 break;
1486 if (index == end_index) {
1487 nr = i_size & ~PAGE_CACHE_MASK;
1488 if (nr <= offset)
1489 break;
1490 }
1491
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001492 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 if (desc->error) {
1494 if (desc->error == -EINVAL)
1495 desc->error = 0;
1496 break;
1497 }
Hugh Dickinsd3602442008-02-04 22:28:44 -08001498 if (page)
1499 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500
1501 /*
1502 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001503 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 */
1505 nr = PAGE_CACHE_SIZE;
1506 i_size = i_size_read(inode);
1507 end_index = i_size >> PAGE_CACHE_SHIFT;
1508 if (index == end_index) {
1509 nr = i_size & ~PAGE_CACHE_MASK;
1510 if (nr <= offset) {
1511 if (page)
1512 page_cache_release(page);
1513 break;
1514 }
1515 }
1516 nr -= offset;
1517
1518 if (page) {
1519 /*
1520 * If users can be writing to this page using arbitrary
1521 * virtual addresses, take care about potential aliasing
1522 * before reading the page on the kernel side.
1523 */
1524 if (mapping_writably_mapped(mapping))
1525 flush_dcache_page(page);
1526 /*
1527 * Mark the page accessed if we read the beginning.
1528 */
1529 if (!offset)
1530 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001531 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 page = ZERO_PAGE(0);
Nick Pigginb5810032005-10-29 18:16:12 -07001533 page_cache_get(page);
1534 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535
1536 /*
1537 * Ok, we have the page, and it's up-to-date, so
1538 * now we can copy it to user space...
1539 *
1540 * The actor routine returns how many bytes were actually used..
1541 * NOTE! This may not be the same as how much of a user buffer
1542 * we filled up (we may be padding etc), so we can only update
1543 * "pos" here (the actor routine has to update the user buffer
1544 * pointers and the remaining count).
1545 */
1546 ret = actor(desc, page, offset, nr);
1547 offset += ret;
1548 index += offset >> PAGE_CACHE_SHIFT;
1549 offset &= ~PAGE_CACHE_MASK;
1550
1551 page_cache_release(page);
1552 if (ret != nr || !desc->count)
1553 break;
1554
1555 cond_resched();
1556 }
1557
1558 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1559 file_accessed(filp);
1560}
1561
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001562static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1563 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564{
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001565 struct file *filp = iocb->ki_filp;
1566 ssize_t retval;
1567 unsigned long seg;
1568 size_t count;
1569 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001571 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1572 if (retval)
1573 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001575 for (seg = 0; seg < nr_segs; seg++) {
1576 read_descriptor_t desc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001578 desc.written = 0;
1579 desc.arg.buf = iov[seg].iov_base;
1580 desc.count = iov[seg].iov_len;
1581 if (desc.count == 0)
1582 continue;
1583 desc.error = 0;
1584 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1585 retval += desc.written;
1586 if (desc.error) {
1587 retval = retval ?: desc.error;
1588 break;
1589 }
1590 if (desc.count > 0)
1591 break;
1592 }
1593 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594}
1595
Hugh Dickins708e3502011-07-25 17:12:32 -07001596static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1597 struct pipe_inode_info *pipe, size_t len,
1598 unsigned int flags)
1599{
1600 struct address_space *mapping = in->f_mapping;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001601 struct inode *inode = mapping->host;
Hugh Dickins708e3502011-07-25 17:12:32 -07001602 unsigned int loff, nr_pages, req_pages;
1603 struct page *pages[PIPE_DEF_BUFFERS];
1604 struct partial_page partial[PIPE_DEF_BUFFERS];
1605 struct page *page;
1606 pgoff_t index, end_index;
1607 loff_t isize, left;
1608 int error, page_nr;
1609 struct splice_pipe_desc spd = {
1610 .pages = pages,
1611 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +02001612 .nr_pages_max = PIPE_DEF_BUFFERS,
Hugh Dickins708e3502011-07-25 17:12:32 -07001613 .flags = flags,
1614 .ops = &page_cache_pipe_buf_ops,
1615 .spd_release = spd_release_page,
1616 };
1617
Hugh Dickins71f0e072011-07-25 17:12:33 -07001618 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07001619 if (unlikely(*ppos >= isize))
1620 return 0;
1621
1622 left = isize - *ppos;
1623 if (unlikely(left < len))
1624 len = left;
1625
1626 if (splice_grow_spd(pipe, &spd))
1627 return -ENOMEM;
1628
1629 index = *ppos >> PAGE_CACHE_SHIFT;
1630 loff = *ppos & ~PAGE_CACHE_MASK;
1631 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1632 nr_pages = min(req_pages, pipe->buffers);
1633
Hugh Dickins708e3502011-07-25 17:12:32 -07001634 spd.nr_pages = find_get_pages_contig(mapping, index,
1635 nr_pages, spd.pages);
1636 index += spd.nr_pages;
Hugh Dickins708e3502011-07-25 17:12:32 -07001637 error = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001638
Hugh Dickins708e3502011-07-25 17:12:32 -07001639 while (spd.nr_pages < nr_pages) {
Hugh Dickins71f0e072011-07-25 17:12:33 -07001640 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1641 if (error)
1642 break;
1643 unlock_page(page);
Hugh Dickins708e3502011-07-25 17:12:32 -07001644 spd.pages[spd.nr_pages++] = page;
1645 index++;
1646 }
1647
Hugh Dickins708e3502011-07-25 17:12:32 -07001648 index = *ppos >> PAGE_CACHE_SHIFT;
1649 nr_pages = spd.nr_pages;
1650 spd.nr_pages = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001651
Hugh Dickins708e3502011-07-25 17:12:32 -07001652 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1653 unsigned int this_len;
1654
1655 if (!len)
1656 break;
1657
Hugh Dickins708e3502011-07-25 17:12:32 -07001658 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1659 page = spd.pages[page_nr];
1660
Hugh Dickins71f0e072011-07-25 17:12:33 -07001661 if (!PageUptodate(page) || page->mapping != mapping) {
Hugh Dickins71f0e072011-07-25 17:12:33 -07001662 error = shmem_getpage(inode, index, &page,
1663 SGP_CACHE, NULL);
1664 if (error)
Hugh Dickins708e3502011-07-25 17:12:32 -07001665 break;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001666 unlock_page(page);
1667 page_cache_release(spd.pages[page_nr]);
1668 spd.pages[page_nr] = page;
Hugh Dickins708e3502011-07-25 17:12:32 -07001669 }
Hugh Dickins71f0e072011-07-25 17:12:33 -07001670
1671 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07001672 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1673 if (unlikely(!isize || index > end_index))
1674 break;
1675
Hugh Dickins708e3502011-07-25 17:12:32 -07001676 if (end_index == index) {
1677 unsigned int plen;
1678
Hugh Dickins708e3502011-07-25 17:12:32 -07001679 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1680 if (plen <= loff)
1681 break;
1682
Hugh Dickins708e3502011-07-25 17:12:32 -07001683 this_len = min(this_len, plen - loff);
1684 len = this_len;
1685 }
1686
1687 spd.partial[page_nr].offset = loff;
1688 spd.partial[page_nr].len = this_len;
1689 len -= this_len;
1690 loff = 0;
1691 spd.nr_pages++;
1692 index++;
1693 }
1694
Hugh Dickins708e3502011-07-25 17:12:32 -07001695 while (page_nr < nr_pages)
1696 page_cache_release(spd.pages[page_nr++]);
Hugh Dickins708e3502011-07-25 17:12:32 -07001697
1698 if (spd.nr_pages)
1699 error = splice_to_pipe(pipe, &spd);
1700
Eric Dumazet047fe362012-06-12 15:24:40 +02001701 splice_shrink_spd(&spd);
Hugh Dickins708e3502011-07-25 17:12:32 -07001702
1703 if (error > 0) {
1704 *ppos += error;
1705 file_accessed(in);
1706 }
1707 return error;
1708}
1709
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001710static long shmem_fallocate(struct file *file, int mode, loff_t offset,
1711 loff_t len)
1712{
1713 struct inode *inode = file->f_path.dentry->d_inode;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001714 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins1aac1402012-05-29 15:06:42 -07001715 struct shmem_falloc shmem_falloc;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001716 pgoff_t start, index, end;
1717 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001718
1719 mutex_lock(&inode->i_mutex);
1720
1721 if (mode & FALLOC_FL_PUNCH_HOLE) {
1722 struct address_space *mapping = file->f_mapping;
1723 loff_t unmap_start = round_up(offset, PAGE_SIZE);
1724 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
1725
1726 if ((u64)unmap_end > (u64)unmap_start)
1727 unmap_mapping_range(mapping, unmap_start,
1728 1 + unmap_end - unmap_start, 0);
1729 shmem_truncate_range(inode, offset, offset + len - 1);
1730 /* No need to unmap again: hole-punching leaves COWed pages */
1731 error = 0;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001732 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001733 }
1734
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001735 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1736 error = inode_newsize_ok(inode, offset + len);
1737 if (error)
1738 goto out;
1739
1740 start = offset >> PAGE_CACHE_SHIFT;
1741 end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1742 /* Try to avoid a swapstorm if len is impossible to satisfy */
1743 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
1744 error = -ENOSPC;
1745 goto out;
1746 }
1747
Hugh Dickins1aac1402012-05-29 15:06:42 -07001748 shmem_falloc.start = start;
1749 shmem_falloc.next = start;
1750 shmem_falloc.nr_falloced = 0;
1751 shmem_falloc.nr_unswapped = 0;
1752 spin_lock(&inode->i_lock);
1753 inode->i_private = &shmem_falloc;
1754 spin_unlock(&inode->i_lock);
1755
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001756 for (index = start; index < end; index++) {
1757 struct page *page;
1758
1759 /*
1760 * Good, the fallocate(2) manpage permits EINTR: we may have
1761 * been interrupted because we are using up too much memory.
1762 */
1763 if (signal_pending(current))
1764 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07001765 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
1766 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001767 else
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001768 error = shmem_getpage(inode, index, &page, SGP_FALLOC,
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001769 NULL);
1770 if (error) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001771 /* Remove the !PageUptodate pages we added */
1772 shmem_undo_range(inode,
1773 (loff_t)start << PAGE_CACHE_SHIFT,
1774 (loff_t)index << PAGE_CACHE_SHIFT, true);
Hugh Dickins1aac1402012-05-29 15:06:42 -07001775 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001776 }
1777
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001778 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07001779 * Inform shmem_writepage() how far we have reached.
1780 * No need for lock or barrier: we have the page lock.
1781 */
1782 shmem_falloc.next++;
1783 if (!PageUptodate(page))
1784 shmem_falloc.nr_falloced++;
1785
1786 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001787 * If !PageUptodate, leave it that way so that freeable pages
1788 * can be recognized if we need to rollback on error later.
1789 * But set_page_dirty so that memory pressure will swap rather
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001790 * than free the pages we are allocating (and SGP_CACHE pages
1791 * might still be clean: we now need to mark those dirty too).
1792 */
1793 set_page_dirty(page);
1794 unlock_page(page);
1795 page_cache_release(page);
1796 cond_resched();
1797 }
1798
1799 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
1800 i_size_write(inode, offset + len);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001801 inode->i_ctime = CURRENT_TIME;
Hugh Dickins1aac1402012-05-29 15:06:42 -07001802undone:
1803 spin_lock(&inode->i_lock);
1804 inode->i_private = NULL;
1805 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001806out:
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001807 mutex_unlock(&inode->i_mutex);
1808 return error;
1809}
1810
David Howells726c3342006-06-23 02:02:58 -07001811static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812{
David Howells726c3342006-06-23 02:02:58 -07001813 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814
1815 buf->f_type = TMPFS_MAGIC;
1816 buf->f_bsize = PAGE_CACHE_SIZE;
1817 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001818 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001820 buf->f_bavail =
1821 buf->f_bfree = sbinfo->max_blocks -
1822 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001823 }
1824 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 buf->f_files = sbinfo->max_inodes;
1826 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 }
1828 /* else leave those fields 0 like simple_statfs */
1829 return 0;
1830}
1831
1832/*
1833 * File creation. Allocate an inode, and we're done..
1834 */
1835static int
Al Viro1a67aaf2011-07-26 01:52:52 -04001836shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001838 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 int error = -ENOSPC;
1840
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001841 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 if (inode) {
Eric Paris2a7dba32011-02-01 11:05:39 -05001843 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04001844 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001845 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001846 if (error) {
1847 if (error != -EOPNOTSUPP) {
1848 iput(inode);
1849 return error;
1850 }
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001851 }
Christoph Hellwig1c7c4742009-11-03 16:44:44 +01001852#ifdef CONFIG_TMPFS_POSIX_ACL
1853 error = generic_acl_init(inode, dir);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001854 if (error) {
1855 iput(inode);
1856 return error;
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001857 }
Al Viro718deb62009-12-16 19:35:36 -05001858#else
1859 error = 0;
Christoph Hellwig1c7c4742009-11-03 16:44:44 +01001860#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 dir->i_size += BOGO_DIRENT_SIZE;
1862 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1863 d_instantiate(dentry, inode);
1864 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 }
1866 return error;
1867}
1868
Al Viro18bb1db2011-07-26 01:41:39 -04001869static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870{
1871 int error;
1872
1873 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1874 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07001875 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876 return 0;
1877}
1878
Al Viro4acdaf22011-07-26 01:42:34 -04001879static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 struct nameidata *nd)
1881{
1882 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1883}
1884
1885/*
1886 * Link a file..
1887 */
1888static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1889{
1890 struct inode *inode = old_dentry->d_inode;
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001891 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892
1893 /*
1894 * No ordinary (disk based) filesystem counts links as inodes;
1895 * but each new link needs a new dentry, pinning lowmem, and
1896 * tmpfs dentries cannot be pruned until they are unlinked.
1897 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001898 ret = shmem_reserve_inode(inode->i_sb);
1899 if (ret)
1900 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901
1902 dir->i_size += BOGO_DIRENT_SIZE;
1903 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansend8c76e62006-09-30 23:29:04 -07001904 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04001905 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 dget(dentry); /* Extra pinning count for the created dentry */
1907 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001908out:
1909 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910}
1911
1912static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1913{
1914 struct inode *inode = dentry->d_inode;
1915
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001916 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1917 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918
1919 dir->i_size -= BOGO_DIRENT_SIZE;
1920 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001921 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 dput(dentry); /* Undo the count from "create" - this does all the work */
1923 return 0;
1924}
1925
1926static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1927{
1928 if (!simple_empty(dentry))
1929 return -ENOTEMPTY;
1930
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001931 drop_nlink(dentry->d_inode);
1932 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933 return shmem_unlink(dir, dentry);
1934}
1935
1936/*
1937 * The VFS layer already does all the dentry stuff for rename,
1938 * we just have to decrement the usage count for the target if
1939 * it exists so that the VFS layer correctly free's it when it
1940 * gets overwritten.
1941 */
1942static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1943{
1944 struct inode *inode = old_dentry->d_inode;
1945 int they_are_dirs = S_ISDIR(inode->i_mode);
1946
1947 if (!simple_empty(new_dentry))
1948 return -ENOTEMPTY;
1949
1950 if (new_dentry->d_inode) {
1951 (void) shmem_unlink(new_dir, new_dentry);
1952 if (they_are_dirs)
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001953 drop_nlink(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001955 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07001956 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 }
1958
1959 old_dir->i_size -= BOGO_DIRENT_SIZE;
1960 new_dir->i_size += BOGO_DIRENT_SIZE;
1961 old_dir->i_ctime = old_dir->i_mtime =
1962 new_dir->i_ctime = new_dir->i_mtime =
1963 inode->i_ctime = CURRENT_TIME;
1964 return 0;
1965}
1966
1967static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1968{
1969 int error;
1970 int len;
1971 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07001972 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973 char *kaddr;
1974 struct shmem_inode_info *info;
1975
1976 len = strlen(symname) + 1;
1977 if (len > PAGE_CACHE_SIZE)
1978 return -ENAMETOOLONG;
1979
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001980 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 if (!inode)
1982 return -ENOSPC;
1983
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04001984 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001985 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001986 if (error) {
1987 if (error != -EOPNOTSUPP) {
1988 iput(inode);
1989 return error;
1990 }
1991 error = 0;
1992 }
1993
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 info = SHMEM_I(inode);
1995 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07001996 if (len <= SHORT_SYMLINK_LEN) {
1997 info->symlink = kmemdup(symname, len, GFP_KERNEL);
1998 if (!info->symlink) {
1999 iput(inode);
2000 return -ENOMEM;
2001 }
2002 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002003 } else {
2004 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2005 if (error) {
2006 iput(inode);
2007 return error;
2008 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07002009 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002010 inode->i_op = &shmem_symlink_inode_operations;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002011 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 memcpy(kaddr, symname, len);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002013 kunmap_atomic(kaddr);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002014 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002015 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002016 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 page_cache_release(page);
2018 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 dir->i_size += BOGO_DIRENT_SIZE;
2020 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2021 d_instantiate(dentry, inode);
2022 dget(dentry);
2023 return 0;
2024}
2025
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002026static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027{
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002028 nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002029 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030}
2031
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002032static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033{
2034 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002035 int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2036 nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
Hugh Dickinsd3602442008-02-04 22:28:44 -08002037 if (page)
2038 unlock_page(page);
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002039 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040}
2041
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002042static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043{
2044 if (!IS_ERR(nd_get_link(nd))) {
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002045 struct page *page = cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 kunmap(page);
2047 mark_page_accessed(page);
2048 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 }
2050}
2051
Eric Parisb09e0fa2011-05-24 17:12:39 -07002052#ifdef CONFIG_TMPFS_XATTR
2053/*
2054 * Superblocks without xattr inode operations may get some security.* xattr
2055 * support from the LSM "for free". As soon as we have any other xattrs
2056 * like ACLs, we also need to implement the security.* handlers at
2057 * filesystem level, though.
2058 */
2059
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002060/*
2061 * Allocate new xattr and copy in the value; but leave the name to callers.
2062 */
2063static struct shmem_xattr *shmem_xattr_alloc(const void *value, size_t size)
2064{
2065 struct shmem_xattr *new_xattr;
2066 size_t len;
2067
2068 /* wrap around? */
2069 len = sizeof(*new_xattr) + size;
2070 if (len <= sizeof(*new_xattr))
2071 return NULL;
2072
2073 new_xattr = kmalloc(len, GFP_KERNEL);
2074 if (!new_xattr)
2075 return NULL;
2076
2077 new_xattr->size = size;
2078 memcpy(new_xattr->value, value, size);
2079 return new_xattr;
2080}
2081
2082/*
2083 * Callback for security_inode_init_security() for acquiring xattrs.
2084 */
2085static int shmem_initxattrs(struct inode *inode,
2086 const struct xattr *xattr_array,
2087 void *fs_info)
2088{
2089 struct shmem_inode_info *info = SHMEM_I(inode);
2090 const struct xattr *xattr;
2091 struct shmem_xattr *new_xattr;
2092 size_t len;
2093
2094 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
2095 new_xattr = shmem_xattr_alloc(xattr->value, xattr->value_len);
2096 if (!new_xattr)
2097 return -ENOMEM;
2098
2099 len = strlen(xattr->name) + 1;
2100 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2101 GFP_KERNEL);
2102 if (!new_xattr->name) {
2103 kfree(new_xattr);
2104 return -ENOMEM;
2105 }
2106
2107 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2108 XATTR_SECURITY_PREFIX_LEN);
2109 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2110 xattr->name, len);
2111
2112 spin_lock(&info->lock);
2113 list_add(&new_xattr->list, &info->xattr_list);
2114 spin_unlock(&info->lock);
2115 }
2116
2117 return 0;
2118}
2119
Eric Parisb09e0fa2011-05-24 17:12:39 -07002120static int shmem_xattr_get(struct dentry *dentry, const char *name,
2121 void *buffer, size_t size)
2122{
2123 struct shmem_inode_info *info;
2124 struct shmem_xattr *xattr;
2125 int ret = -ENODATA;
2126
2127 info = SHMEM_I(dentry->d_inode);
2128
2129 spin_lock(&info->lock);
2130 list_for_each_entry(xattr, &info->xattr_list, list) {
2131 if (strcmp(name, xattr->name))
2132 continue;
2133
2134 ret = xattr->size;
2135 if (buffer) {
2136 if (size < xattr->size)
2137 ret = -ERANGE;
2138 else
2139 memcpy(buffer, xattr->value, xattr->size);
2140 }
2141 break;
2142 }
2143 spin_unlock(&info->lock);
2144 return ret;
2145}
2146
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002147static int shmem_xattr_set(struct inode *inode, const char *name,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002148 const void *value, size_t size, int flags)
2149{
Eric Parisb09e0fa2011-05-24 17:12:39 -07002150 struct shmem_inode_info *info = SHMEM_I(inode);
2151 struct shmem_xattr *xattr;
2152 struct shmem_xattr *new_xattr = NULL;
Eric Parisb09e0fa2011-05-24 17:12:39 -07002153 int err = 0;
2154
2155 /* value == NULL means remove */
2156 if (value) {
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002157 new_xattr = shmem_xattr_alloc(value, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002158 if (!new_xattr)
2159 return -ENOMEM;
2160
2161 new_xattr->name = kstrdup(name, GFP_KERNEL);
2162 if (!new_xattr->name) {
2163 kfree(new_xattr);
2164 return -ENOMEM;
2165 }
Eric Parisb09e0fa2011-05-24 17:12:39 -07002166 }
2167
2168 spin_lock(&info->lock);
2169 list_for_each_entry(xattr, &info->xattr_list, list) {
2170 if (!strcmp(name, xattr->name)) {
2171 if (flags & XATTR_CREATE) {
2172 xattr = new_xattr;
2173 err = -EEXIST;
2174 } else if (new_xattr) {
2175 list_replace(&xattr->list, &new_xattr->list);
2176 } else {
2177 list_del(&xattr->list);
2178 }
2179 goto out;
2180 }
2181 }
2182 if (flags & XATTR_REPLACE) {
2183 xattr = new_xattr;
2184 err = -ENODATA;
2185 } else {
2186 list_add(&new_xattr->list, &info->xattr_list);
2187 xattr = NULL;
2188 }
2189out:
2190 spin_unlock(&info->lock);
2191 if (xattr)
2192 kfree(xattr->name);
2193 kfree(xattr);
2194 return err;
2195}
2196
Eric Parisb09e0fa2011-05-24 17:12:39 -07002197static const struct xattr_handler *shmem_xattr_handlers[] = {
2198#ifdef CONFIG_TMPFS_POSIX_ACL
2199 &generic_acl_access_handler,
2200 &generic_acl_default_handler,
2201#endif
2202 NULL
2203};
2204
2205static int shmem_xattr_validate(const char *name)
2206{
2207 struct { const char *prefix; size_t len; } arr[] = {
2208 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2209 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2210 };
2211 int i;
2212
2213 for (i = 0; i < ARRAY_SIZE(arr); i++) {
2214 size_t preflen = arr[i].len;
2215 if (strncmp(name, arr[i].prefix, preflen) == 0) {
2216 if (!name[preflen])
2217 return -EINVAL;
2218 return 0;
2219 }
2220 }
2221 return -EOPNOTSUPP;
2222}
2223
2224static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2225 void *buffer, size_t size)
2226{
2227 int err;
2228
2229 /*
2230 * If this is a request for a synthetic attribute in the system.*
2231 * namespace use the generic infrastructure to resolve a handler
2232 * for it via sb->s_xattr.
2233 */
2234 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2235 return generic_getxattr(dentry, name, buffer, size);
2236
2237 err = shmem_xattr_validate(name);
2238 if (err)
2239 return err;
2240
2241 return shmem_xattr_get(dentry, name, buffer, size);
2242}
2243
2244static int shmem_setxattr(struct dentry *dentry, const char *name,
2245 const void *value, size_t size, int flags)
2246{
2247 int err;
2248
2249 /*
2250 * If this is a request for a synthetic attribute in the system.*
2251 * namespace use the generic infrastructure to resolve a handler
2252 * for it via sb->s_xattr.
2253 */
2254 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2255 return generic_setxattr(dentry, name, value, size, flags);
2256
2257 err = shmem_xattr_validate(name);
2258 if (err)
2259 return err;
2260
2261 if (size == 0)
2262 value = ""; /* empty EA, do not remove */
2263
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002264 return shmem_xattr_set(dentry->d_inode, name, value, size, flags);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002265
2266}
2267
2268static int shmem_removexattr(struct dentry *dentry, const char *name)
2269{
2270 int err;
2271
2272 /*
2273 * If this is a request for a synthetic attribute in the system.*
2274 * namespace use the generic infrastructure to resolve a handler
2275 * for it via sb->s_xattr.
2276 */
2277 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2278 return generic_removexattr(dentry, name);
2279
2280 err = shmem_xattr_validate(name);
2281 if (err)
2282 return err;
2283
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002284 return shmem_xattr_set(dentry->d_inode, name, NULL, 0, XATTR_REPLACE);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002285}
2286
2287static bool xattr_is_trusted(const char *name)
2288{
2289 return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN);
2290}
2291
2292static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2293{
2294 bool trusted = capable(CAP_SYS_ADMIN);
2295 struct shmem_xattr *xattr;
2296 struct shmem_inode_info *info;
2297 size_t used = 0;
2298
2299 info = SHMEM_I(dentry->d_inode);
2300
2301 spin_lock(&info->lock);
2302 list_for_each_entry(xattr, &info->xattr_list, list) {
2303 size_t len;
2304
2305 /* skip "trusted." attributes for unprivileged callers */
2306 if (!trusted && xattr_is_trusted(xattr->name))
2307 continue;
2308
2309 len = strlen(xattr->name) + 1;
2310 used += len;
2311 if (buffer) {
2312 if (size < used) {
2313 used = -ERANGE;
2314 break;
2315 }
2316 memcpy(buffer, xattr->name, len);
2317 buffer += len;
2318 }
2319 }
2320 spin_unlock(&info->lock);
2321
2322 return used;
2323}
2324#endif /* CONFIG_TMPFS_XATTR */
2325
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002326static const struct inode_operations shmem_short_symlink_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002327 .readlink = generic_readlink,
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002328 .follow_link = shmem_follow_short_symlink,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002329#ifdef CONFIG_TMPFS_XATTR
2330 .setxattr = shmem_setxattr,
2331 .getxattr = shmem_getxattr,
2332 .listxattr = shmem_listxattr,
2333 .removexattr = shmem_removexattr,
2334#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335};
2336
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002337static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338 .readlink = generic_readlink,
2339 .follow_link = shmem_follow_link,
2340 .put_link = shmem_put_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002341#ifdef CONFIG_TMPFS_XATTR
2342 .setxattr = shmem_setxattr,
2343 .getxattr = shmem_getxattr,
2344 .listxattr = shmem_listxattr,
2345 .removexattr = shmem_removexattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002346#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07002347};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002348
David M. Grimes91828a42006-10-17 00:09:45 -07002349static struct dentry *shmem_get_parent(struct dentry *child)
2350{
2351 return ERR_PTR(-ESTALE);
2352}
2353
2354static int shmem_match(struct inode *ino, void *vfh)
2355{
2356 __u32 *fh = vfh;
2357 __u64 inum = fh[2];
2358 inum = (inum << 32) | fh[1];
2359 return ino->i_ino == inum && fh[0] == ino->i_generation;
2360}
2361
Christoph Hellwig480b1162007-10-21 16:42:13 -07002362static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2363 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07002364{
David M. Grimes91828a42006-10-17 00:09:45 -07002365 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07002366 struct dentry *dentry = NULL;
2367 u64 inum = fid->raw[2];
2368 inum = (inum << 32) | fid->raw[1];
David M. Grimes91828a42006-10-17 00:09:45 -07002369
Christoph Hellwig480b1162007-10-21 16:42:13 -07002370 if (fh_len < 3)
2371 return NULL;
2372
2373 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2374 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07002375 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07002376 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002377 iput(inode);
2378 }
2379
Christoph Hellwig480b1162007-10-21 16:42:13 -07002380 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07002381}
2382
Al Virob0b03822012-04-02 14:34:06 -04002383static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2384 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07002385{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302386 if (*len < 3) {
2387 *len = 3;
David M. Grimes91828a42006-10-17 00:09:45 -07002388 return 255;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302389 }
David M. Grimes91828a42006-10-17 00:09:45 -07002390
Al Viro1d3382cb2010-10-23 15:19:20 -04002391 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07002392 /* Unfortunately insert_inode_hash is not idempotent,
2393 * so as we hash inodes here rather than at creation
2394 * time, we need a lock to ensure we only try
2395 * to do it once
2396 */
2397 static DEFINE_SPINLOCK(lock);
2398 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04002399 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07002400 __insert_inode_hash(inode,
2401 inode->i_ino + inode->i_generation);
2402 spin_unlock(&lock);
2403 }
2404
2405 fh[0] = inode->i_generation;
2406 fh[1] = inode->i_ino;
2407 fh[2] = ((__u64)inode->i_ino) >> 32;
2408
2409 *len = 3;
2410 return 1;
2411}
2412
Christoph Hellwig39655162007-10-21 16:42:17 -07002413static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07002414 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07002415 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07002416 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07002417};
2418
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002419static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2420 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421{
2422 char *this_char, *value, *rest;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002423 uid_t uid;
2424 gid_t gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002426 while (options != NULL) {
2427 this_char = options;
2428 for (;;) {
2429 /*
2430 * NUL-terminate this option: unfortunately,
2431 * mount options form a comma-separated list,
2432 * but mpol's nodelist may also contain commas.
2433 */
2434 options = strchr(options, ',');
2435 if (options == NULL)
2436 break;
2437 options++;
2438 if (!isdigit(*options)) {
2439 options[-1] = '\0';
2440 break;
2441 }
2442 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 if (!*this_char)
2444 continue;
2445 if ((value = strchr(this_char,'=')) != NULL) {
2446 *value++ = 0;
2447 } else {
2448 printk(KERN_ERR
2449 "tmpfs: No value for mount option '%s'\n",
2450 this_char);
2451 return 1;
2452 }
2453
2454 if (!strcmp(this_char,"size")) {
2455 unsigned long long size;
2456 size = memparse(value,&rest);
2457 if (*rest == '%') {
2458 size <<= PAGE_SHIFT;
2459 size *= totalram_pages;
2460 do_div(size, 100);
2461 rest++;
2462 }
2463 if (*rest)
2464 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002465 sbinfo->max_blocks =
2466 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002468 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002469 if (*rest)
2470 goto bad_val;
2471 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002472 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473 if (*rest)
2474 goto bad_val;
2475 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002476 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002478 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479 if (*rest)
2480 goto bad_val;
2481 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002482 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002484 uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485 if (*rest)
2486 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002487 sbinfo->uid = make_kuid(current_user_ns(), uid);
2488 if (!uid_valid(sbinfo->uid))
2489 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002491 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002493 gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002494 if (*rest)
2495 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002496 sbinfo->gid = make_kgid(current_user_ns(), gid);
2497 if (!gid_valid(sbinfo->gid))
2498 goto bad_val;
Robin Holt7339ff82006-01-14 13:20:48 -08002499 } else if (!strcmp(this_char,"mpol")) {
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002500 if (mpol_parse_str(value, &sbinfo->mpol, 1))
Robin Holt7339ff82006-01-14 13:20:48 -08002501 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502 } else {
2503 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2504 this_char);
2505 return 1;
2506 }
2507 }
2508 return 0;
2509
2510bad_val:
2511 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2512 value, this_char);
2513 return 1;
2514
2515}
2516
2517static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2518{
2519 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002520 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002521 unsigned long inodes;
2522 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002524 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002525 return error;
2526
2527 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002528 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
Tim Chen7e496292010-08-09 17:19:05 -07002529 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002530 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002531 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002532 goto out;
2533 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07002534 * Those tests disallow limited->unlimited while any are in use;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002535 * but we must separately disallow unlimited->limited, because
2536 * in that case we have no record of how much is already in use.
2537 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002538 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002539 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002540 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002541 goto out;
2542
2543 error = 0;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002544 sbinfo->max_blocks = config.max_blocks;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002545 sbinfo->max_inodes = config.max_inodes;
2546 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002547
2548 mpol_put(sbinfo->mpol);
2549 sbinfo->mpol = config.mpol; /* transfers initial ref */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002550out:
2551 spin_unlock(&sbinfo->stat_lock);
2552 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002554
Al Viro34c80b12011-12-08 21:32:45 -05002555static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002556{
Al Viro34c80b12011-12-08 21:32:45 -05002557 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002558
2559 if (sbinfo->max_blocks != shmem_default_max_blocks())
2560 seq_printf(seq, ",size=%luk",
2561 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2562 if (sbinfo->max_inodes != shmem_default_max_inodes())
2563 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2564 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04002565 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08002566 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2567 seq_printf(seq, ",uid=%u",
2568 from_kuid_munged(&init_user_ns, sbinfo->uid));
2569 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2570 seq_printf(seq, ",gid=%u",
2571 from_kgid_munged(&init_user_ns, sbinfo->gid));
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002572 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002573 return 0;
2574}
2575#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002576
2577static void shmem_put_super(struct super_block *sb)
2578{
Hugh Dickins602586a2010-08-17 15:23:56 -07002579 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2580
2581 percpu_counter_destroy(&sbinfo->used_blocks);
2582 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 sb->s_fs_info = NULL;
2584}
2585
Kay Sievers2b2af542009-04-30 15:23:42 +02002586int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587{
2588 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002589 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002590 int err = -ENOMEM;
2591
2592 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07002593 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002594 L1_CACHE_BYTES), GFP_KERNEL);
2595 if (!sbinfo)
2596 return -ENOMEM;
2597
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002598 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11002599 sbinfo->uid = current_fsuid();
2600 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002601 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002603#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604 /*
2605 * Per default we only allow half of the physical ram per
2606 * tmpfs instance, limiting inodes to one per page of lowmem;
2607 * but the internal instance is left unlimited.
2608 */
2609 if (!(sb->s_flags & MS_NOUSER)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002610 sbinfo->max_blocks = shmem_default_max_blocks();
2611 sbinfo->max_inodes = shmem_default_max_inodes();
2612 if (shmem_parse_options(data, sbinfo, false)) {
2613 err = -EINVAL;
2614 goto failed;
2615 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616 }
David M. Grimes91828a42006-10-17 00:09:45 -07002617 sb->s_export_op = &shmem_export_ops;
Hugh Dickins2f6e38f2012-05-29 15:06:38 -07002618 sb->s_flags |= MS_NOSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619#else
2620 sb->s_flags |= MS_NOUSER;
2621#endif
2622
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002623 spin_lock_init(&sbinfo->stat_lock);
Hugh Dickins602586a2010-08-17 15:23:56 -07002624 if (percpu_counter_init(&sbinfo->used_blocks, 0))
2625 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002626 sbinfo->free_inodes = sbinfo->max_inodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002627
Hugh Dickins285b2c42011-08-03 16:21:20 -07002628 sb->s_maxbytes = MAX_LFS_FILESIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002629 sb->s_blocksize = PAGE_CACHE_SIZE;
2630 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2631 sb->s_magic = TMPFS_MAGIC;
2632 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01002633 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07002634#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002635 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07002636#endif
2637#ifdef CONFIG_TMPFS_POSIX_ACL
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002638 sb->s_flags |= MS_POSIXACL;
2639#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002640
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002641 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642 if (!inode)
2643 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002644 inode->i_uid = sbinfo->uid;
2645 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05002646 sb->s_root = d_make_root(inode);
2647 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05002648 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649 return 0;
2650
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651failed:
2652 shmem_put_super(sb);
2653 return err;
2654}
2655
Pekka Enbergfcc234f2006-03-22 00:08:13 -08002656static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657
2658static struct inode *shmem_alloc_inode(struct super_block *sb)
2659{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002660 struct shmem_inode_info *info;
2661 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2662 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002664 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665}
2666
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002667static void shmem_destroy_callback(struct rcu_head *head)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11002668{
2669 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11002670 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2671}
2672
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673static void shmem_destroy_inode(struct inode *inode)
2674{
Al Viro09208d12011-07-26 03:15:03 -04002675 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002677 call_rcu(&inode->i_rcu, shmem_destroy_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678}
2679
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002680static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002682 struct shmem_inode_info *info = foo;
2683 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002684}
2685
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002686static int shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687{
2688 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2689 sizeof(struct shmem_inode_info),
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002690 0, SLAB_PANIC, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691 return 0;
2692}
2693
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002694static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07002696 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697}
2698
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002699static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08002701 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07002703 .write_begin = shmem_write_begin,
2704 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705#endif
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07002706 .migratepage = migrate_page,
Andi Kleenaa261f52009-09-16 11:50:16 +02002707 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708};
2709
Helge Deller15ad7cd2006-12-06 20:40:36 -08002710static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711 .mmap = shmem_mmap,
2712#ifdef CONFIG_TMPFS
Hugh Dickinsf21f8062012-07-11 14:02:45 -07002713 .llseek = generic_file_llseek,
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07002714 .read = do_sync_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002715 .write = do_sync_write,
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07002716 .aio_read = shmem_file_aio_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002717 .aio_write = generic_file_aio_write,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02002718 .fsync = noop_fsync,
Hugh Dickins708e3502011-07-25 17:12:32 -07002719 .splice_read = shmem_file_splice_read,
Hugh Dickinsae9764162007-06-04 10:00:39 +02002720 .splice_write = generic_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002721 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722#endif
2723};
2724
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002725static const struct inode_operations shmem_inode_operations = {
Hugh Dickins94c1e622011-06-27 16:18:03 -07002726 .setattr = shmem_setattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002727#ifdef CONFIG_TMPFS_XATTR
2728 .setxattr = shmem_setxattr,
2729 .getxattr = shmem_getxattr,
2730 .listxattr = shmem_listxattr,
2731 .removexattr = shmem_removexattr,
2732#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733};
2734
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002735static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002736#ifdef CONFIG_TMPFS
2737 .create = shmem_create,
2738 .lookup = simple_lookup,
2739 .link = shmem_link,
2740 .unlink = shmem_unlink,
2741 .symlink = shmem_symlink,
2742 .mkdir = shmem_mkdir,
2743 .rmdir = shmem_rmdir,
2744 .mknod = shmem_mknod,
2745 .rename = shmem_rename,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07002747#ifdef CONFIG_TMPFS_XATTR
2748 .setxattr = shmem_setxattr,
2749 .getxattr = shmem_getxattr,
2750 .listxattr = shmem_listxattr,
2751 .removexattr = shmem_removexattr,
2752#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002753#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07002754 .setattr = shmem_setattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002755#endif
2756};
2757
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002758static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07002759#ifdef CONFIG_TMPFS_XATTR
2760 .setxattr = shmem_setxattr,
2761 .getxattr = shmem_getxattr,
2762 .listxattr = shmem_listxattr,
2763 .removexattr = shmem_removexattr,
2764#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002765#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07002766 .setattr = shmem_setattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002767#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768};
2769
Hugh Dickins759b9772007-03-05 00:30:28 -08002770static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771 .alloc_inode = shmem_alloc_inode,
2772 .destroy_inode = shmem_destroy_inode,
2773#ifdef CONFIG_TMPFS
2774 .statfs = shmem_statfs,
2775 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002776 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777#endif
Al Viro1f895f72010-06-05 19:10:41 -04002778 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779 .drop_inode = generic_delete_inode,
2780 .put_super = shmem_put_super,
2781};
2782
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002783static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002784 .fault = shmem_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785#ifdef CONFIG_NUMA
2786 .set_policy = shmem_set_policy,
2787 .get_policy = shmem_get_policy,
2788#endif
2789};
2790
Al Viro3c26ff62010-07-25 11:46:36 +04002791static struct dentry *shmem_mount(struct file_system_type *fs_type,
2792 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793{
Al Viro3c26ff62010-07-25 11:46:36 +04002794 return mount_nodev(fs_type, flags, data, shmem_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002795}
2796
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002797static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798 .owner = THIS_MODULE,
2799 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04002800 .mount = shmem_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801 .kill_sb = kill_litter_super,
2802};
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002804int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805{
2806 int error;
2807
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002808 error = bdi_init(&shmem_backing_dev_info);
2809 if (error)
2810 goto out4;
2811
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002812 error = shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002813 if (error)
2814 goto out3;
2815
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002816 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817 if (error) {
2818 printk(KERN_ERR "Could not register tmpfs\n");
2819 goto out2;
2820 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07002821
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002822 shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER,
2823 shmem_fs_type.name, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824 if (IS_ERR(shm_mnt)) {
2825 error = PTR_ERR(shm_mnt);
2826 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2827 goto out1;
2828 }
2829 return 0;
2830
2831out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002832 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002834 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835out3:
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002836 bdi_destroy(&shmem_backing_dev_info);
2837out4:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838 shm_mnt = ERR_PTR(error);
2839 return error;
2840}
Matt Mackall853ac432009-01-06 14:40:20 -08002841
2842#else /* !CONFIG_SHMEM */
2843
2844/*
2845 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2846 *
2847 * This is intended for small system where the benefits of the full
2848 * shmem code (swap-backed and resource-limited) are outweighed by
2849 * their complexity. On systems without swap this code should be
2850 * effectively equivalent, but much lighter weight.
2851 */
2852
2853#include <linux/ramfs.h>
2854
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002855static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08002856 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04002857 .mount = ramfs_mount,
Matt Mackall853ac432009-01-06 14:40:20 -08002858 .kill_sb = kill_litter_super,
2859};
2860
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002861int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08002862{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002863 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08002864
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002865 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08002866 BUG_ON(IS_ERR(shm_mnt));
2867
2868 return 0;
2869}
2870
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002871int shmem_unuse(swp_entry_t swap, struct page *page)
Matt Mackall853ac432009-01-06 14:40:20 -08002872{
2873 return 0;
2874}
2875
Hugh Dickins3f96b792009-09-21 17:03:37 -07002876int shmem_lock(struct file *file, int lock, struct user_struct *user)
2877{
2878 return 0;
2879}
2880
Hugh Dickins24513262012-01-20 14:34:21 -08002881void shmem_unlock_mapping(struct address_space *mapping)
2882{
2883}
2884
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002885void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07002886{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002887 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07002888}
2889EXPORT_SYMBOL_GPL(shmem_truncate_range);
2890
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002891#define shmem_vm_ops generic_file_vm_ops
2892#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002893#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002894#define shmem_acct_size(flags, size) 0
2895#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08002896
2897#endif /* CONFIG_SHMEM */
2898
2899/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002900
Randy Dunlap46711812008-03-19 17:00:41 -07002901/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002902 * shmem_file_setup - get an unlinked file living in tmpfs
Linus Torvalds1da177e2005-04-16 15:20:36 -07002903 * @name: name for dentry (to be seen in /proc/<pid>/maps
2904 * @size: size to be set for the file
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002905 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906 */
Sergei Trofimovich168f5ac2009-06-16 15:33:02 -07002907struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908{
2909 int error;
2910 struct file *file;
2911 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04002912 struct path path;
2913 struct dentry *root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002914 struct qstr this;
2915
2916 if (IS_ERR(shm_mnt))
2917 return (void *)shm_mnt;
2918
Hugh Dickins285b2c42011-08-03 16:21:20 -07002919 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002920 return ERR_PTR(-EINVAL);
2921
2922 if (shmem_acct_size(flags, size))
2923 return ERR_PTR(-ENOMEM);
2924
2925 error = -ENOMEM;
2926 this.name = name;
2927 this.len = strlen(name);
2928 this.hash = 0; /* will go */
2929 root = shm_mnt->mnt_root;
Al Viro2c48b9c2009-08-09 00:52:35 +04002930 path.dentry = d_alloc(root, &this);
2931 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932 goto put_memory;
Al Viro2c48b9c2009-08-09 00:52:35 +04002933 path.mnt = mntget(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934
Linus Torvalds1da177e2005-04-16 15:20:36 -07002935 error = -ENOSPC;
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002936 inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937 if (!inode)
Al Viro4b42af82009-08-05 18:25:56 +04002938 goto put_dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939
Al Viro2c48b9c2009-08-09 00:52:35 +04002940 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002941 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02002942 clear_nlink(inode); /* It is unlinked */
Matt Mackall853ac432009-01-06 14:40:20 -08002943#ifndef CONFIG_MMU
2944 error = ramfs_nommu_expand_for_mapping(inode, size);
2945 if (error)
Al Viro4b42af82009-08-05 18:25:56 +04002946 goto put_dentry;
Matt Mackall853ac432009-01-06 14:40:20 -08002947#endif
Al Viro4b42af82009-08-05 18:25:56 +04002948
2949 error = -ENFILE;
Al Viro2c48b9c2009-08-09 00:52:35 +04002950 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Al Viro4b42af82009-08-05 18:25:56 +04002951 &shmem_file_operations);
2952 if (!file)
2953 goto put_dentry;
2954
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955 return file;
2956
Linus Torvalds1da177e2005-04-16 15:20:36 -07002957put_dentry:
Al Viro2c48b9c2009-08-09 00:52:35 +04002958 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002959put_memory:
2960 shmem_unacct_size(flags, size);
2961 return ERR_PTR(error);
2962}
Keith Packard395e0dd2008-06-20 00:08:06 -07002963EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002964
Randy Dunlap46711812008-03-19 17:00:41 -07002965/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002966 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07002967 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2968 */
2969int shmem_zero_setup(struct vm_area_struct *vma)
2970{
2971 struct file *file;
2972 loff_t size = vma->vm_end - vma->vm_start;
2973
2974 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2975 if (IS_ERR(file))
2976 return PTR_ERR(file);
2977
2978 if (vma->vm_file)
2979 fput(vma->vm_file);
2980 vma->vm_file = file;
2981 vma->vm_ops = &shmem_vm_ops;
Hugh Dickinsbee4c36a2011-03-22 16:33:43 -07002982 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983 return 0;
2984}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07002985
2986/**
2987 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2988 * @mapping: the page's address_space
2989 * @index: the page index
2990 * @gfp: the page allocator flags to use if allocating
2991 *
2992 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2993 * with any new page allocations done using the specified allocation flags.
2994 * But read_cache_page_gfp() uses the ->readpage() method: which does not
2995 * suit tmpfs, since it may have pages in swapcache, and needs to find those
2996 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
2997 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07002998 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
2999 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003000 */
3001struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
3002 pgoff_t index, gfp_t gfp)
3003{
Hugh Dickins68da9f02011-07-25 17:12:34 -07003004#ifdef CONFIG_SHMEM
3005 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07003006 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07003007 int error;
3008
3009 BUG_ON(mapping->a_ops != &shmem_aops);
3010 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
3011 if (error)
3012 page = ERR_PTR(error);
3013 else
3014 unlock_page(page);
3015 return page;
3016#else
3017 /*
3018 * The tiny !SHMEM case uses ramfs without swap
3019 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003020 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07003021#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003022}
3023EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);