blob: 585bd220a21ee4e5eefaec2b9c46dc9ae68fde98 [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 Dickins46f65ec2011-08-03 16:21:23 -0700267 * Like add_to_page_cache_locked, but error if expected item has gone.
268 */
269static int shmem_add_to_page_cache(struct page *page,
270 struct address_space *mapping,
271 pgoff_t index, gfp_t gfp, void *expected)
272{
Hugh Dickinsaa3b1892011-08-03 16:21:24 -0700273 int error = 0;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700274
275 VM_BUG_ON(!PageLocked(page));
276 VM_BUG_ON(!PageSwapBacked(page));
277
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700278 if (!expected)
279 error = radix_tree_preload(gfp & GFP_RECLAIM_MASK);
280 if (!error) {
281 page_cache_get(page);
282 page->mapping = mapping;
283 page->index = index;
284
285 spin_lock_irq(&mapping->tree_lock);
286 if (!expected)
287 error = radix_tree_insert(&mapping->page_tree,
288 index, page);
289 else
290 error = shmem_radix_tree_replace(mapping, index,
291 expected, page);
292 if (!error) {
293 mapping->nrpages++;
294 __inc_zone_page_state(page, NR_FILE_PAGES);
295 __inc_zone_page_state(page, NR_SHMEM);
296 spin_unlock_irq(&mapping->tree_lock);
297 } else {
298 page->mapping = NULL;
299 spin_unlock_irq(&mapping->tree_lock);
300 page_cache_release(page);
301 }
302 if (!expected)
303 radix_tree_preload_end();
304 }
305 if (error)
306 mem_cgroup_uncharge_cache_page(page);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700307 return error;
308}
309
310/*
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700311 * Like delete_from_page_cache, but substitutes swap for page.
312 */
313static void shmem_delete_from_page_cache(struct page *page, void *radswap)
314{
315 struct address_space *mapping = page->mapping;
316 int error;
317
318 spin_lock_irq(&mapping->tree_lock);
319 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
320 page->mapping = NULL;
321 mapping->nrpages--;
322 __dec_zone_page_state(page, NR_FILE_PAGES);
323 __dec_zone_page_state(page, NR_SHMEM);
324 spin_unlock_irq(&mapping->tree_lock);
325 page_cache_release(page);
326 BUG_ON(error);
327}
328
329/*
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700330 * Like find_get_pages, but collecting swap entries as well as pages.
331 */
332static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping,
333 pgoff_t start, unsigned int nr_pages,
334 struct page **pages, pgoff_t *indices)
335{
336 unsigned int i;
337 unsigned int ret;
338 unsigned int nr_found;
339
340 rcu_read_lock();
341restart:
342 nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
343 (void ***)pages, indices, start, nr_pages);
344 ret = 0;
345 for (i = 0; i < nr_found; i++) {
346 struct page *page;
347repeat:
348 page = radix_tree_deref_slot((void **)pages[i]);
349 if (unlikely(!page))
350 continue;
351 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700352 if (radix_tree_deref_retry(page))
353 goto restart;
354 /*
355 * Otherwise, we must be storing a swap entry
356 * here as an exceptional entry: so return it
357 * without attempting to raise page count.
358 */
359 goto export;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700360 }
361 if (!page_cache_get_speculative(page))
362 goto repeat;
363
364 /* Has the page moved? */
365 if (unlikely(page != *((void **)pages[i]))) {
366 page_cache_release(page);
367 goto repeat;
368 }
369export:
370 indices[ret] = indices[i];
371 pages[ret] = page;
372 ret++;
373 }
374 if (unlikely(!ret && nr_found))
375 goto restart;
376 rcu_read_unlock();
377 return ret;
378}
379
380/*
381 * Remove swap entry from radix tree, free the swap and its page cache.
382 */
383static int shmem_free_swap(struct address_space *mapping,
384 pgoff_t index, void *radswap)
385{
386 int error;
387
388 spin_lock_irq(&mapping->tree_lock);
389 error = shmem_radix_tree_replace(mapping, index, radswap, NULL);
390 spin_unlock_irq(&mapping->tree_lock);
391 if (!error)
392 free_swap_and_cache(radix_to_swp_entry(radswap));
393 return error;
394}
395
396/*
397 * Pagevec may contain swap entries, so shuffle up pages before releasing.
398 */
Hugh Dickins24513262012-01-20 14:34:21 -0800399static void shmem_deswap_pagevec(struct pagevec *pvec)
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700400{
401 int i, j;
402
403 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
404 struct page *page = pvec->pages[i];
405 if (!radix_tree_exceptional_entry(page))
406 pvec->pages[j++] = page;
407 }
408 pvec->nr = j;
Hugh Dickins24513262012-01-20 14:34:21 -0800409}
410
411/*
412 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
413 */
414void shmem_unlock_mapping(struct address_space *mapping)
415{
416 struct pagevec pvec;
417 pgoff_t indices[PAGEVEC_SIZE];
418 pgoff_t index = 0;
419
420 pagevec_init(&pvec, 0);
421 /*
422 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
423 */
424 while (!mapping_unevictable(mapping)) {
425 /*
426 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
427 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
428 */
429 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
430 PAGEVEC_SIZE, pvec.pages, indices);
431 if (!pvec.nr)
432 break;
433 index = indices[pvec.nr - 1] + 1;
434 shmem_deswap_pagevec(&pvec);
435 check_move_unevictable_pages(pvec.pages, pvec.nr);
436 pagevec_release(&pvec);
437 cond_resched();
438 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700439}
440
441/*
442 * Remove range of pages and swap entries from radix tree, and free them.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700443 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700444 */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700445static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
446 bool unfalloc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700448 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700449 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700450 pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700451 pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
452 unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
453 unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700454 struct pagevec pvec;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700455 pgoff_t indices[PAGEVEC_SIZE];
456 long nr_swaps_freed = 0;
Hugh Dickins285b2c42011-08-03 16:21:20 -0700457 pgoff_t index;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700458 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700459
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700460 if (lend == -1)
461 end = -1; /* unsigned, so actually very big */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700462
463 pagevec_init(&pvec, 0);
464 index = start;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700465 while (index < end) {
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700466 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700467 min(end - index, (pgoff_t)PAGEVEC_SIZE),
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700468 pvec.pages, indices);
469 if (!pvec.nr)
470 break;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700471 mem_cgroup_uncharge_start();
472 for (i = 0; i < pagevec_count(&pvec); i++) {
473 struct page *page = pvec.pages[i];
474
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700475 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700476 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700477 break;
478
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700479 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700480 if (unfalloc)
481 continue;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700482 nr_swaps_freed += !shmem_free_swap(mapping,
483 index, page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700484 continue;
485 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700486
487 if (!trylock_page(page))
488 continue;
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700489 if (!unfalloc || !PageUptodate(page)) {
490 if (page->mapping == mapping) {
491 VM_BUG_ON(PageWriteback(page));
492 truncate_inode_page(mapping, page);
493 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700494 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700495 unlock_page(page);
496 }
Hugh Dickins24513262012-01-20 14:34:21 -0800497 shmem_deswap_pagevec(&pvec);
498 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700499 mem_cgroup_uncharge_end();
500 cond_resched();
501 index++;
502 }
503
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700504 if (partial_start) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700505 struct page *page = NULL;
506 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
507 if (page) {
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700508 unsigned int top = PAGE_CACHE_SIZE;
509 if (start > end) {
510 top = partial_end;
511 partial_end = 0;
512 }
513 zero_user_segment(page, partial_start, top);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700514 set_page_dirty(page);
515 unlock_page(page);
516 page_cache_release(page);
517 }
518 }
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700519 if (partial_end) {
520 struct page *page = NULL;
521 shmem_getpage(inode, end, &page, SGP_READ, NULL);
522 if (page) {
523 zero_user_segment(page, 0, partial_end);
524 set_page_dirty(page);
525 unlock_page(page);
526 page_cache_release(page);
527 }
528 }
529 if (start >= end)
530 return;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700531
532 index = start;
533 for ( ; ; ) {
534 cond_resched();
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700535 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700536 min(end - index, (pgoff_t)PAGEVEC_SIZE),
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700537 pvec.pages, indices);
538 if (!pvec.nr) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700539 if (index == start || unfalloc)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700540 break;
541 index = start;
542 continue;
543 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700544 if ((index == start || unfalloc) && indices[0] >= end) {
Hugh Dickins24513262012-01-20 14:34:21 -0800545 shmem_deswap_pagevec(&pvec);
546 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700547 break;
548 }
549 mem_cgroup_uncharge_start();
550 for (i = 0; i < pagevec_count(&pvec); i++) {
551 struct page *page = pvec.pages[i];
552
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700553 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700554 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700555 break;
556
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700557 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700558 if (unfalloc)
559 continue;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700560 nr_swaps_freed += !shmem_free_swap(mapping,
561 index, page);
562 continue;
563 }
564
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700565 lock_page(page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700566 if (!unfalloc || !PageUptodate(page)) {
567 if (page->mapping == mapping) {
568 VM_BUG_ON(PageWriteback(page));
569 truncate_inode_page(mapping, page);
570 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700571 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700572 unlock_page(page);
573 }
Hugh Dickins24513262012-01-20 14:34:21 -0800574 shmem_deswap_pagevec(&pvec);
575 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700576 mem_cgroup_uncharge_end();
577 index++;
578 }
Hugh Dickins94c1e622011-06-27 16:18:03 -0700579
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 spin_lock(&info->lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700581 info->swapped -= nr_swaps_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 shmem_recalc_inode(inode);
583 spin_unlock(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700584}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700586void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
587{
588 shmem_undo_range(inode, lstart, lend, false);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700589 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590}
Hugh Dickins94c1e622011-06-27 16:18:03 -0700591EXPORT_SYMBOL_GPL(shmem_truncate_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592
Hugh Dickins94c1e622011-06-27 16:18:03 -0700593static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594{
595 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 int error;
597
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200598 error = inode_change_ok(inode, attr);
599 if (error)
600 return error;
601
Hugh Dickins94c1e622011-06-27 16:18:03 -0700602 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
603 loff_t oldsize = inode->i_size;
604 loff_t newsize = attr->ia_size;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000605
Hugh Dickins94c1e622011-06-27 16:18:03 -0700606 if (newsize != oldsize) {
607 i_size_write(inode, newsize);
608 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
609 }
610 if (newsize < oldsize) {
611 loff_t holebegin = round_up(newsize, PAGE_SIZE);
612 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
613 shmem_truncate_range(inode, newsize, (loff_t)-1);
614 /* unmap again to remove racily COWed private pages */
615 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
616 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617 }
618
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200619 setattr_copy(inode, attr);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700620#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200621 if (attr->ia_valid & ATTR_MODE)
Christoph Hellwig1c7c4742009-11-03 16:44:44 +0100622 error = generic_acl_chmod(inode);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700623#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 return error;
625}
626
Al Viro1f895f72010-06-05 19:10:41 -0400627static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 struct shmem_inode_info *info = SHMEM_I(inode);
Eric Parisb09e0fa2011-05-24 17:12:39 -0700630 struct shmem_xattr *xattr, *nxattr;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000632 if (inode->i_mapping->a_ops == &shmem_aops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633 shmem_unacct_size(info->flags, inode->i_size);
634 inode->i_size = 0;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000635 shmem_truncate_range(inode, 0, (loff_t)-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800637 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800639 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 }
Hugh Dickins69f07ec2011-08-03 16:21:26 -0700641 } else
642 kfree(info->symlink);
Eric Parisb09e0fa2011-05-24 17:12:39 -0700643
644 list_for_each_entry_safe(xattr, nxattr, &info->xattr_list, list) {
645 kfree(xattr->name);
646 kfree(xattr);
647 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700648 BUG_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800649 shmem_free_inode(inode->i_sb);
Jan Karadbd57682012-05-03 14:48:02 +0200650 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651}
652
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700653/*
654 * If swap found in inode, free it and move page from swapcache to filecache.
655 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700656static int shmem_unuse_inode(struct shmem_inode_info *info,
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700657 swp_entry_t swap, struct page **pagep)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700659 struct address_space *mapping = info->vfs_inode.i_mapping;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700660 void *radswap;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700661 pgoff_t index;
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700662 gfp_t gfp;
663 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700665 radswap = swp_to_radix_entry(swap);
Hugh Dickinse504f3f2011-08-03 16:21:27 -0700666 index = radix_tree_locate_item(&mapping->page_tree, radswap);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700667 if (index == -1)
Hugh Dickins285b2c42011-08-03 16:21:20 -0700668 return 0;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800669
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800670 /*
671 * Move _head_ to start search for next from here.
Al Viro1f895f72010-06-05 19:10:41 -0400672 * But be careful: shmem_evict_inode checks list_empty without taking
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800673 * mutex, and there's an instant in list_move_tail when info->swaplist
Hugh Dickins285b2c42011-08-03 16:21:20 -0700674 * would appear empty, if it were the only one on shmem_swaplist.
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800675 */
676 if (shmem_swaplist.next != &info->swaplist)
677 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800678
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700679 gfp = mapping_gfp_mask(mapping);
680 if (shmem_should_replace_page(*pagep, gfp)) {
681 mutex_unlock(&shmem_swaplist_mutex);
682 error = shmem_replace_page(pagep, gfp, info, index);
683 mutex_lock(&shmem_swaplist_mutex);
684 /*
685 * We needed to drop mutex to make that restrictive page
686 * allocation; but the inode might already be freed by now,
687 * and we cannot refer to inode or mapping or info to check.
688 * However, we do hold page lock on the PageSwapCache page,
689 * so can check if that still has our reference remaining.
690 */
691 if (!page_swapcount(*pagep))
692 error = -ENOENT;
693 }
694
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800695 /*
Hugh Dickins778dd892011-05-11 15:13:37 -0700696 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
697 * but also to hold up shmem_evict_inode(): so inode cannot be freed
698 * beneath us (pagelock doesn't help until the page is in pagecache).
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800699 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700700 if (!error)
701 error = shmem_add_to_page_cache(*pagep, mapping, index,
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700702 GFP_NOWAIT, radswap);
Hugh Dickins48f170f2011-07-25 17:12:37 -0700703 if (error != -ENOMEM) {
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700704 /*
705 * Truncation and eviction use free_swap_and_cache(), which
706 * only does trylock page: if we raced, best clean up here.
707 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700708 delete_from_swap_cache(*pagep);
709 set_page_dirty(*pagep);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700710 if (!error) {
711 spin_lock(&info->lock);
712 info->swapped--;
713 spin_unlock(&info->lock);
714 swap_free(swap);
715 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800716 error = 1; /* not an error, but entry was found */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700717 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800718 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719}
720
721/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700722 * Search through swapped inodes to find and replace swap by page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700724int shmem_unuse(swp_entry_t swap, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700726 struct list_head *this, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 struct shmem_inode_info *info;
728 int found = 0;
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700729 int error = 0;
730
731 /*
732 * There's a faint possibility that swap page was replaced before
733 * caller locked it: it will come back later with the right page.
734 */
735 if (unlikely(!PageSwapCache(page)))
736 goto out;
Hugh Dickins778dd892011-05-11 15:13:37 -0700737
738 /*
739 * Charge page using GFP_KERNEL while we can wait, before taking
740 * the shmem_swaplist_mutex which might hold up shmem_writepage().
741 * Charged back to the user (not to caller) when swap account is used.
Hugh Dickins778dd892011-05-11 15:13:37 -0700742 */
743 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
744 if (error)
745 goto out;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700746 /* No radix_tree_preload: swap entry keeps a place for page in tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800748 mutex_lock(&shmem_swaplist_mutex);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700749 list_for_each_safe(this, next, &shmem_swaplist) {
750 info = list_entry(this, struct shmem_inode_info, swaplist);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700751 if (info->swapped)
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700752 found = shmem_unuse_inode(info, swap, &page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700753 else
754 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800755 cond_resched();
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800756 if (found)
Hugh Dickins778dd892011-05-11 15:13:37 -0700757 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800759 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins778dd892011-05-11 15:13:37 -0700760
Hugh Dickins778dd892011-05-11 15:13:37 -0700761 if (found < 0)
762 error = found;
763out:
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800764 unlock_page(page);
765 page_cache_release(page);
Hugh Dickins778dd892011-05-11 15:13:37 -0700766 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767}
768
769/*
770 * Move the page from the page cache to the swap cache.
771 */
772static int shmem_writepage(struct page *page, struct writeback_control *wbc)
773{
774 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 struct inode *inode;
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700777 swp_entry_t swap;
778 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700779
780 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 mapping = page->mapping;
782 index = page->index;
783 inode = mapping->host;
784 info = SHMEM_I(inode);
785 if (info->flags & VM_LOCKED)
786 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800787 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 goto redirty;
789
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800790 /*
791 * shmem_backing_dev_info's capabilities prevent regular writeback or
792 * sync from ever calling shmem_writepage; but a stacking filesystem
Hugh Dickins48f170f2011-07-25 17:12:37 -0700793 * might use ->writepage of its underlying filesystem, in which case
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800794 * tmpfs should write out to swap only in response to memory pressure,
Hugh Dickins48f170f2011-07-25 17:12:37 -0700795 * and not for the writeback threads or sync.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800796 */
Hugh Dickins48f170f2011-07-25 17:12:37 -0700797 if (!wbc->for_reclaim) {
798 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
799 goto redirty;
800 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700801
802 /*
803 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
804 * value into swapfile.c, the only way we can correctly account for a
805 * fallocated page arriving here is now to initialize it and write it.
Hugh Dickins1aac1402012-05-29 15:06:42 -0700806 *
807 * That's okay for a page already fallocated earlier, but if we have
808 * not yet completed the fallocation, then (a) we want to keep track
809 * of this page in case we have to undo it, and (b) it may not be a
810 * good idea to continue anyway, once we're pushing into swap. So
811 * reactivate the page, and let shmem_fallocate() quit when too many.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700812 */
813 if (!PageUptodate(page)) {
Hugh Dickins1aac1402012-05-29 15:06:42 -0700814 if (inode->i_private) {
815 struct shmem_falloc *shmem_falloc;
816 spin_lock(&inode->i_lock);
817 shmem_falloc = inode->i_private;
818 if (shmem_falloc &&
819 index >= shmem_falloc->start &&
820 index < shmem_falloc->next)
821 shmem_falloc->nr_unswapped++;
822 else
823 shmem_falloc = NULL;
824 spin_unlock(&inode->i_lock);
825 if (shmem_falloc)
826 goto redirty;
827 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700828 clear_highpage(page);
829 flush_dcache_page(page);
830 SetPageUptodate(page);
831 }
832
Hugh Dickins48f170f2011-07-25 17:12:37 -0700833 swap = get_swap_page();
834 if (!swap.val)
835 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800836
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700837 /*
838 * Add inode to shmem_unuse()'s list of swapped-out inodes,
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700839 * if it's not already there. Do it now before the page is
840 * moved to swap cache, when its pagelock no longer protects
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700841 * the inode from eviction. But don't unlock the mutex until
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700842 * we've incremented swapped, because shmem_unuse_inode() will
843 * prune a !swapped inode from the swaplist under this mutex.
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700844 */
Hugh Dickins48f170f2011-07-25 17:12:37 -0700845 mutex_lock(&shmem_swaplist_mutex);
846 if (list_empty(&info->swaplist))
847 list_add_tail(&info->swaplist, &shmem_swaplist);
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700848
Hugh Dickins48f170f2011-07-25 17:12:37 -0700849 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800850 swap_shmem_alloc(swap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700851 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
852
853 spin_lock(&info->lock);
854 info->swapped++;
855 shmem_recalc_inode(inode);
Hugh Dickins826267c2011-05-28 13:14:09 -0700856 spin_unlock(&info->lock);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700857
858 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800859 BUG_ON(page_mapped(page));
Hugh Dickins9fab5612009-03-31 15:23:33 -0700860 swap_writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 return 0;
862 }
863
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700864 mutex_unlock(&shmem_swaplist_mutex);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700865 swapcache_free(swap, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700866redirty:
867 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800868 if (wbc->for_reclaim)
869 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
870 unlock_page(page);
871 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700872}
873
874#ifdef CONFIG_NUMA
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800875#ifdef CONFIG_TMPFS
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700876static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800877{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700878 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800879
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700880 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700881 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800882
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700883 mpol_to_str(buffer, sizeof(buffer), mpol, 1);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800884
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700885 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800886}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700887
888static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
889{
890 struct mempolicy *mpol = NULL;
891 if (sbinfo->mpol) {
892 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
893 mpol = sbinfo->mpol;
894 mpol_get(mpol);
895 spin_unlock(&sbinfo->stat_lock);
896 }
897 return mpol;
898}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800899#endif /* CONFIG_TMPFS */
900
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700901static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
902 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903{
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700904 struct mempolicy mpol, *spol;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700905 struct vm_area_struct pvma;
906
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700907 spol = mpol_cond_copy(&mpol,
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700908 mpol_shared_policy_lookup(&info->policy, index));
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700909
Linus Torvalds1da177e2005-04-16 15:20:36 -0700910 /* Create a pseudo vma that just contains the policy */
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800911 pvma.vm_start = 0;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700912 pvma.vm_pgoff = index;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800913 pvma.vm_ops = NULL;
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700914 pvma.vm_policy = spol;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700915 return swapin_readahead(swap, gfp, &pvma, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916}
917
Hugh Dickins02098fe2008-02-04 22:28:42 -0800918static struct page *shmem_alloc_page(gfp_t gfp,
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700919 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700920{
921 struct vm_area_struct pvma;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800923 /* Create a pseudo vma that just contains the policy */
924 pvma.vm_start = 0;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700925 pvma.vm_pgoff = index;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800926 pvma.vm_ops = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700927 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700928
929 /*
930 * alloc_page_vma() will drop the shared policy reference
931 */
932 return alloc_page_vma(gfp, &pvma, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800934#else /* !CONFIG_NUMA */
935#ifdef CONFIG_TMPFS
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700936static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800937{
938}
939#endif /* CONFIG_TMPFS */
940
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700941static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
942 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700944 return swapin_readahead(swap, gfp, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700945}
946
Hugh Dickins02098fe2008-02-04 22:28:42 -0800947static inline struct page *shmem_alloc_page(gfp_t gfp,
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700948 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949{
Hugh Dickinse84e2e12007-11-28 18:55:10 +0000950 return alloc_page(gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800952#endif /* CONFIG_NUMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700953
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700954#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
955static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
956{
957 return NULL;
958}
959#endif
960
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700962 * When a page is moved from swapcache to shmem filecache (either by the
963 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
964 * shmem_unuse_inode()), it may have been read in earlier from swap, in
965 * ignorance of the mapping it belongs to. If that mapping has special
966 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
967 * we may need to copy to a suitable page before moving to filecache.
968 *
969 * In a future release, this may well be extended to respect cpuset and
970 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
971 * but for now it is a simple matter of zone.
972 */
973static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
974{
975 return page_zonenum(page) > gfp_zone(gfp);
976}
977
978static int shmem_replace_page(struct page **pagep, gfp_t gfp,
979 struct shmem_inode_info *info, pgoff_t index)
980{
981 struct page *oldpage, *newpage;
982 struct address_space *swap_mapping;
983 pgoff_t swap_index;
984 int error;
985
986 oldpage = *pagep;
987 swap_index = page_private(oldpage);
988 swap_mapping = page_mapping(oldpage);
989
990 /*
991 * We have arrived here because our zones are constrained, so don't
992 * limit chance of success by further cpuset and node constraints.
993 */
994 gfp &= ~GFP_CONSTRAINT_MASK;
995 newpage = shmem_alloc_page(gfp, info, index);
996 if (!newpage)
997 return -ENOMEM;
998 VM_BUG_ON(shmem_should_replace_page(newpage, gfp));
999
1000 *pagep = newpage;
1001 page_cache_get(newpage);
1002 copy_highpage(newpage, oldpage);
1003
1004 VM_BUG_ON(!PageLocked(oldpage));
1005 __set_page_locked(newpage);
1006 VM_BUG_ON(!PageUptodate(oldpage));
1007 SetPageUptodate(newpage);
1008 VM_BUG_ON(!PageSwapBacked(oldpage));
1009 SetPageSwapBacked(newpage);
1010 VM_BUG_ON(!swap_index);
1011 set_page_private(newpage, swap_index);
1012 VM_BUG_ON(!PageSwapCache(oldpage));
1013 SetPageSwapCache(newpage);
1014
1015 /*
1016 * Our caller will very soon move newpage out of swapcache, but it's
1017 * a nice clean interface for us to replace oldpage by newpage there.
1018 */
1019 spin_lock_irq(&swap_mapping->tree_lock);
1020 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1021 newpage);
1022 __inc_zone_page_state(newpage, NR_FILE_PAGES);
1023 __dec_zone_page_state(oldpage, NR_FILE_PAGES);
1024 spin_unlock_irq(&swap_mapping->tree_lock);
1025 BUG_ON(error);
1026
1027 mem_cgroup_replace_page_cache(oldpage, newpage);
1028 lru_cache_add_anon(newpage);
1029
1030 ClearPageSwapCache(oldpage);
1031 set_page_private(oldpage, 0);
1032
1033 unlock_page(oldpage);
1034 page_cache_release(oldpage);
1035 page_cache_release(oldpage);
1036 return 0;
1037}
1038
1039/*
Hugh Dickins68da9f02011-07-25 17:12:34 -07001040 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 *
1042 * If we allocate a new one we do not mark it dirty. That's up to the
1043 * vm. If we swap it in we mark it dirty since we also free the swap
1044 * entry since a page cannot live in both the swap and page cache
1045 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001046static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Hugh Dickins68da9f02011-07-25 17:12:34 -07001047 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001048{
1049 struct address_space *mapping = inode->i_mapping;
Hugh Dickins54af60422011-08-03 16:21:24 -07001050 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 struct shmem_sb_info *sbinfo;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001052 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 swp_entry_t swap;
1054 int error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001055 int once = 0;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001056 int alloced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001058 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001059 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060repeat:
Hugh Dickins54af60422011-08-03 16:21:24 -07001061 swap.val = 0;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001062 page = find_lock_page(mapping, index);
Hugh Dickins54af60422011-08-03 16:21:24 -07001063 if (radix_tree_exceptional_entry(page)) {
1064 swap = radix_to_swp_entry(page);
1065 page = NULL;
1066 }
1067
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001068 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
Hugh Dickins54af60422011-08-03 16:21:24 -07001069 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1070 error = -EINVAL;
1071 goto failed;
1072 }
1073
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001074 /* fallocated page? */
1075 if (page && !PageUptodate(page)) {
1076 if (sgp != SGP_READ)
1077 goto clear;
1078 unlock_page(page);
1079 page_cache_release(page);
1080 page = NULL;
1081 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001082 if (page || (sgp == SGP_READ && !swap.val)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001083 *pagep = page;
1084 return 0;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001085 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086
1087 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001088 * Fast cache lookup did not find it:
1089 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001091 info = SHMEM_I(inode);
1092 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001093
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 if (swap.val) {
1095 /* Look it up and read it in.. */
Hugh Dickins27ab7002011-07-25 17:12:36 -07001096 page = lookup_swap_cache(swap);
1097 if (!page) {
Ying Han456f9982011-05-26 16:25:38 -07001098 /* here we actually do the io */
Hugh Dickins68da9f02011-07-25 17:12:34 -07001099 if (fault_type)
1100 *fault_type |= VM_FAULT_MAJOR;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001101 page = shmem_swapin(swap, gfp, info, index);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001102 if (!page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001103 error = -ENOMEM;
1104 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 }
1107
1108 /* We have to do this with page locked to prevent races */
Hugh Dickins54af60422011-08-03 16:21:24 -07001109 lock_page(page);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001110 if (!PageSwapCache(page) || page->mapping) {
1111 error = -EEXIST; /* try again */
1112 goto failed;
1113 }
Hugh Dickins27ab7002011-07-25 17:12:36 -07001114 if (!PageUptodate(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 error = -EIO;
Hugh Dickins54af60422011-08-03 16:21:24 -07001116 goto failed;
1117 }
1118 wait_on_page_writeback(page);
1119
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001120 if (shmem_should_replace_page(page, gfp)) {
1121 error = shmem_replace_page(&page, gfp, info, index);
1122 if (error)
1123 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001124 }
1125
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001126 error = mem_cgroup_cache_charge(page, current->mm,
1127 gfp & GFP_RECLAIM_MASK);
1128 if (!error)
1129 error = shmem_add_to_page_cache(page, mapping, index,
1130 gfp, swp_to_radix_entry(swap));
Hugh Dickins54af60422011-08-03 16:21:24 -07001131 if (error)
1132 goto failed;
1133
1134 spin_lock(&info->lock);
1135 info->swapped--;
1136 shmem_recalc_inode(inode);
1137 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001138
Hugh Dickins27ab7002011-07-25 17:12:36 -07001139 delete_from_swap_cache(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001140 set_page_dirty(page);
1141 swap_free(swap);
1142
Hugh Dickins54af60422011-08-03 16:21:24 -07001143 } else {
1144 if (shmem_acct_block(info->flags)) {
1145 error = -ENOSPC;
1146 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001148 if (sbinfo->max_blocks) {
Hugh Dickinsfc5da222011-04-14 15:22:07 -07001149 if (percpu_counter_compare(&sbinfo->used_blocks,
Hugh Dickins54af60422011-08-03 16:21:24 -07001150 sbinfo->max_blocks) >= 0) {
1151 error = -ENOSPC;
1152 goto unacct;
1153 }
Tim Chen7e496292010-08-09 17:19:05 -07001154 percpu_counter_inc(&sbinfo->used_blocks);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001155 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156
Hugh Dickins54af60422011-08-03 16:21:24 -07001157 page = shmem_alloc_page(gfp, info, index);
1158 if (!page) {
1159 error = -ENOMEM;
1160 goto decused;
1161 }
1162
1163 SetPageSwapBacked(page);
1164 __set_page_locked(page);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001165 error = mem_cgroup_cache_charge(page, current->mm,
1166 gfp & GFP_RECLAIM_MASK);
1167 if (!error)
1168 error = shmem_add_to_page_cache(page, mapping, index,
1169 gfp, NULL);
Hugh Dickins54af60422011-08-03 16:21:24 -07001170 if (error)
1171 goto decused;
1172 lru_cache_add_anon(page);
1173
1174 spin_lock(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 info->alloced++;
Hugh Dickins54af60422011-08-03 16:21:24 -07001176 inode->i_blocks += BLOCKS_PER_PAGE;
1177 shmem_recalc_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 spin_unlock(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001179 alloced = true;
Hugh Dickins54af60422011-08-03 16:21:24 -07001180
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001181 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001182 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1183 */
1184 if (sgp == SGP_FALLOC)
1185 sgp = SGP_WRITE;
1186clear:
1187 /*
1188 * Let SGP_WRITE caller clear ends if write does not fill page;
1189 * but SGP_FALLOC on a page fallocated earlier must initialize
1190 * it now, lest undo on failure cancel our earlier guarantee.
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001191 */
1192 if (sgp != SGP_WRITE) {
1193 clear_highpage(page);
1194 flush_dcache_page(page);
1195 SetPageUptodate(page);
1196 }
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001197 if (sgp == SGP_DIRTY)
Hugh Dickins27ab7002011-07-25 17:12:36 -07001198 set_page_dirty(page);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001199 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001200
Hugh Dickins54af60422011-08-03 16:21:24 -07001201 /* Perhaps the file has been truncated since we checked */
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001202 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
Hugh Dickins54af60422011-08-03 16:21:24 -07001203 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1204 error = -EINVAL;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001205 if (alloced)
1206 goto trunc;
1207 else
1208 goto failed;
Shaohua Liff36b8012010-08-09 17:19:06 -07001209 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001210 *pagep = page;
1211 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07001212
Nick Piggind0217ac2007-07-19 01:47:03 -07001213 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001214 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001216trunc:
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001217 info = SHMEM_I(inode);
Hugh Dickins54af60422011-08-03 16:21:24 -07001218 ClearPageDirty(page);
1219 delete_from_page_cache(page);
1220 spin_lock(&info->lock);
1221 info->alloced--;
1222 inode->i_blocks -= BLOCKS_PER_PAGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 spin_unlock(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001224decused:
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001225 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins54af60422011-08-03 16:21:24 -07001226 if (sbinfo->max_blocks)
1227 percpu_counter_add(&sbinfo->used_blocks, -1);
1228unacct:
1229 shmem_unacct_blocks(info->flags, 1);
1230failed:
1231 if (swap.val && error != -EINVAL) {
1232 struct page *test = find_get_page(mapping, index);
1233 if (test && !radix_tree_exceptional_entry(test))
1234 page_cache_release(test);
1235 /* Have another try if the entry has changed */
1236 if (test != swp_to_radix_entry(swap))
1237 error = -EEXIST;
1238 }
Hugh Dickins27ab7002011-07-25 17:12:36 -07001239 if (page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001240 unlock_page(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001241 page_cache_release(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001242 }
1243 if (error == -ENOSPC && !once++) {
1244 info = SHMEM_I(inode);
1245 spin_lock(&info->lock);
1246 shmem_recalc_inode(inode);
1247 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001248 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001249 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001250 if (error == -EEXIST)
1251 goto repeat;
1252 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253}
1254
1255static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1256{
1257 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1258 int error;
Hugh Dickins68da9f02011-07-25 17:12:34 -07001259 int ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1262 if (error)
1263 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Hugh Dickins68da9f02011-07-25 17:12:34 -07001264
Ying Han456f9982011-05-26 16:25:38 -07001265 if (ret & VM_FAULT_MAJOR) {
1266 count_vm_event(PGMAJFAULT);
1267 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1268 }
Hugh Dickins68da9f02011-07-25 17:12:34 -07001269 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270}
1271
1272#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001273static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001275 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1276 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277}
1278
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001279static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1280 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001282 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1283 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001285 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1286 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287}
1288#endif
1289
1290int shmem_lock(struct file *file, int lock, struct user_struct *user)
1291{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001292 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293 struct shmem_inode_info *info = SHMEM_I(inode);
1294 int retval = -ENOMEM;
1295
1296 spin_lock(&info->lock);
1297 if (lock && !(info->flags & VM_LOCKED)) {
1298 if (!user_shm_lock(inode->i_size, user))
1299 goto out_nomem;
1300 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001301 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302 }
1303 if (!lock && (info->flags & VM_LOCKED) && user) {
1304 user_shm_unlock(inode->i_size, user);
1305 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001306 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 }
1308 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001309
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310out_nomem:
1311 spin_unlock(&info->lock);
1312 return retval;
1313}
1314
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08001315static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316{
1317 file_accessed(file);
1318 vma->vm_ops = &shmem_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001319 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320 return 0;
1321}
1322
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001323static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04001324 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325{
1326 struct inode *inode;
1327 struct shmem_inode_info *info;
1328 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1329
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001330 if (shmem_reserve_inode(sb))
1331 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332
1333 inode = new_inode(sb);
1334 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -04001335 inode->i_ino = get_next_ino();
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001336 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 inode->i_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1339 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
David M. Grimes91828a42006-10-17 00:09:45 -07001340 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341 info = SHMEM_I(inode);
1342 memset(info, 0, (char *)inode - (char *)info);
1343 spin_lock_init(&info->lock);
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001344 info->flags = flags & VM_NORESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 INIT_LIST_HEAD(&info->swaplist);
Eric Parisb09e0fa2011-05-24 17:12:39 -07001346 INIT_LIST_HEAD(&info->xattr_list);
Al Viro72c04902009-06-24 16:58:48 -04001347 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348
1349 switch (mode & S_IFMT) {
1350 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001351 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 init_special_inode(inode, mode, dev);
1353 break;
1354 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07001355 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 inode->i_op = &shmem_inode_operations;
1357 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001358 mpol_shared_policy_init(&info->policy,
1359 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 break;
1361 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07001362 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 /* Some things misbehave if size == 0 on a directory */
1364 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1365 inode->i_op = &shmem_dir_inode_operations;
1366 inode->i_fop = &simple_dir_operations;
1367 break;
1368 case S_IFLNK:
1369 /*
1370 * Must not load anything in the rbtree,
1371 * mpol_free_shared_policy will not be called.
1372 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001373 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 break;
1375 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001376 } else
1377 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 return inode;
1379}
1380
1381#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001382static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07001383static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001384
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001385#ifdef CONFIG_TMPFS_XATTR
1386static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1387#else
1388#define shmem_initxattrs NULL
1389#endif
1390
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391static int
Nick Piggin800d15a2007-10-16 01:25:03 -07001392shmem_write_begin(struct file *file, struct address_space *mapping,
1393 loff_t pos, unsigned len, unsigned flags,
1394 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395{
Nick Piggin800d15a2007-10-16 01:25:03 -07001396 struct inode *inode = mapping->host;
1397 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
Nick Piggin800d15a2007-10-16 01:25:03 -07001398 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1399}
1400
1401static int
1402shmem_write_end(struct file *file, struct address_space *mapping,
1403 loff_t pos, unsigned len, unsigned copied,
1404 struct page *page, void *fsdata)
1405{
1406 struct inode *inode = mapping->host;
1407
Hugh Dickinsd3602442008-02-04 22:28:44 -08001408 if (pos + copied > inode->i_size)
1409 i_size_write(inode, pos + copied);
1410
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001411 if (!PageUptodate(page)) {
1412 if (copied < PAGE_CACHE_SIZE) {
1413 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1414 zero_user_segments(page, 0, from,
1415 from + copied, PAGE_CACHE_SIZE);
1416 }
1417 SetPageUptodate(page);
1418 }
Nick Piggin800d15a2007-10-16 01:25:03 -07001419 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02001420 unlock_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07001421 page_cache_release(page);
1422
Nick Piggin800d15a2007-10-16 01:25:03 -07001423 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424}
1425
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1427{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001428 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001430 pgoff_t index;
1431 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001432 enum sgp_type sgp = SGP_READ;
1433
1434 /*
1435 * Might this read be for a stacking filesystem? Then when reading
1436 * holes of a sparse file, we actually need to allocate those pages,
1437 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1438 */
1439 if (segment_eq(get_fs(), KERNEL_DS))
1440 sgp = SGP_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441
1442 index = *ppos >> PAGE_CACHE_SHIFT;
1443 offset = *ppos & ~PAGE_CACHE_MASK;
1444
1445 for (;;) {
1446 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001447 pgoff_t end_index;
1448 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449 loff_t i_size = i_size_read(inode);
1450
1451 end_index = i_size >> PAGE_CACHE_SHIFT;
1452 if (index > end_index)
1453 break;
1454 if (index == end_index) {
1455 nr = i_size & ~PAGE_CACHE_MASK;
1456 if (nr <= offset)
1457 break;
1458 }
1459
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001460 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 if (desc->error) {
1462 if (desc->error == -EINVAL)
1463 desc->error = 0;
1464 break;
1465 }
Hugh Dickinsd3602442008-02-04 22:28:44 -08001466 if (page)
1467 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468
1469 /*
1470 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001471 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 */
1473 nr = PAGE_CACHE_SIZE;
1474 i_size = i_size_read(inode);
1475 end_index = i_size >> PAGE_CACHE_SHIFT;
1476 if (index == end_index) {
1477 nr = i_size & ~PAGE_CACHE_MASK;
1478 if (nr <= offset) {
1479 if (page)
1480 page_cache_release(page);
1481 break;
1482 }
1483 }
1484 nr -= offset;
1485
1486 if (page) {
1487 /*
1488 * If users can be writing to this page using arbitrary
1489 * virtual addresses, take care about potential aliasing
1490 * before reading the page on the kernel side.
1491 */
1492 if (mapping_writably_mapped(mapping))
1493 flush_dcache_page(page);
1494 /*
1495 * Mark the page accessed if we read the beginning.
1496 */
1497 if (!offset)
1498 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001499 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 page = ZERO_PAGE(0);
Nick Pigginb5810032005-10-29 18:16:12 -07001501 page_cache_get(page);
1502 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503
1504 /*
1505 * Ok, we have the page, and it's up-to-date, so
1506 * now we can copy it to user space...
1507 *
1508 * The actor routine returns how many bytes were actually used..
1509 * NOTE! This may not be the same as how much of a user buffer
1510 * we filled up (we may be padding etc), so we can only update
1511 * "pos" here (the actor routine has to update the user buffer
1512 * pointers and the remaining count).
1513 */
1514 ret = actor(desc, page, offset, nr);
1515 offset += ret;
1516 index += offset >> PAGE_CACHE_SHIFT;
1517 offset &= ~PAGE_CACHE_MASK;
1518
1519 page_cache_release(page);
1520 if (ret != nr || !desc->count)
1521 break;
1522
1523 cond_resched();
1524 }
1525
1526 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1527 file_accessed(filp);
1528}
1529
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001530static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1531 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532{
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001533 struct file *filp = iocb->ki_filp;
1534 ssize_t retval;
1535 unsigned long seg;
1536 size_t count;
1537 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001539 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1540 if (retval)
1541 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001543 for (seg = 0; seg < nr_segs; seg++) {
1544 read_descriptor_t desc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001546 desc.written = 0;
1547 desc.arg.buf = iov[seg].iov_base;
1548 desc.count = iov[seg].iov_len;
1549 if (desc.count == 0)
1550 continue;
1551 desc.error = 0;
1552 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1553 retval += desc.written;
1554 if (desc.error) {
1555 retval = retval ?: desc.error;
1556 break;
1557 }
1558 if (desc.count > 0)
1559 break;
1560 }
1561 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562}
1563
Hugh Dickins708e3502011-07-25 17:12:32 -07001564static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1565 struct pipe_inode_info *pipe, size_t len,
1566 unsigned int flags)
1567{
1568 struct address_space *mapping = in->f_mapping;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001569 struct inode *inode = mapping->host;
Hugh Dickins708e3502011-07-25 17:12:32 -07001570 unsigned int loff, nr_pages, req_pages;
1571 struct page *pages[PIPE_DEF_BUFFERS];
1572 struct partial_page partial[PIPE_DEF_BUFFERS];
1573 struct page *page;
1574 pgoff_t index, end_index;
1575 loff_t isize, left;
1576 int error, page_nr;
1577 struct splice_pipe_desc spd = {
1578 .pages = pages,
1579 .partial = partial,
1580 .flags = flags,
1581 .ops = &page_cache_pipe_buf_ops,
1582 .spd_release = spd_release_page,
1583 };
1584
Hugh Dickins71f0e072011-07-25 17:12:33 -07001585 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07001586 if (unlikely(*ppos >= isize))
1587 return 0;
1588
1589 left = isize - *ppos;
1590 if (unlikely(left < len))
1591 len = left;
1592
1593 if (splice_grow_spd(pipe, &spd))
1594 return -ENOMEM;
1595
1596 index = *ppos >> PAGE_CACHE_SHIFT;
1597 loff = *ppos & ~PAGE_CACHE_MASK;
1598 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1599 nr_pages = min(req_pages, pipe->buffers);
1600
Hugh Dickins708e3502011-07-25 17:12:32 -07001601 spd.nr_pages = find_get_pages_contig(mapping, index,
1602 nr_pages, spd.pages);
1603 index += spd.nr_pages;
Hugh Dickins708e3502011-07-25 17:12:32 -07001604 error = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001605
Hugh Dickins708e3502011-07-25 17:12:32 -07001606 while (spd.nr_pages < nr_pages) {
Hugh Dickins71f0e072011-07-25 17:12:33 -07001607 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1608 if (error)
1609 break;
1610 unlock_page(page);
Hugh Dickins708e3502011-07-25 17:12:32 -07001611 spd.pages[spd.nr_pages++] = page;
1612 index++;
1613 }
1614
Hugh Dickins708e3502011-07-25 17:12:32 -07001615 index = *ppos >> PAGE_CACHE_SHIFT;
1616 nr_pages = spd.nr_pages;
1617 spd.nr_pages = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001618
Hugh Dickins708e3502011-07-25 17:12:32 -07001619 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1620 unsigned int this_len;
1621
1622 if (!len)
1623 break;
1624
Hugh Dickins708e3502011-07-25 17:12:32 -07001625 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1626 page = spd.pages[page_nr];
1627
Hugh Dickins71f0e072011-07-25 17:12:33 -07001628 if (!PageUptodate(page) || page->mapping != mapping) {
Hugh Dickins71f0e072011-07-25 17:12:33 -07001629 error = shmem_getpage(inode, index, &page,
1630 SGP_CACHE, NULL);
1631 if (error)
Hugh Dickins708e3502011-07-25 17:12:32 -07001632 break;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001633 unlock_page(page);
1634 page_cache_release(spd.pages[page_nr]);
1635 spd.pages[page_nr] = page;
Hugh Dickins708e3502011-07-25 17:12:32 -07001636 }
Hugh Dickins71f0e072011-07-25 17:12:33 -07001637
1638 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07001639 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1640 if (unlikely(!isize || index > end_index))
1641 break;
1642
Hugh Dickins708e3502011-07-25 17:12:32 -07001643 if (end_index == index) {
1644 unsigned int plen;
1645
Hugh Dickins708e3502011-07-25 17:12:32 -07001646 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1647 if (plen <= loff)
1648 break;
1649
Hugh Dickins708e3502011-07-25 17:12:32 -07001650 this_len = min(this_len, plen - loff);
1651 len = this_len;
1652 }
1653
1654 spd.partial[page_nr].offset = loff;
1655 spd.partial[page_nr].len = this_len;
1656 len -= this_len;
1657 loff = 0;
1658 spd.nr_pages++;
1659 index++;
1660 }
1661
Hugh Dickins708e3502011-07-25 17:12:32 -07001662 while (page_nr < nr_pages)
1663 page_cache_release(spd.pages[page_nr++]);
Hugh Dickins708e3502011-07-25 17:12:32 -07001664
1665 if (spd.nr_pages)
1666 error = splice_to_pipe(pipe, &spd);
1667
1668 splice_shrink_spd(pipe, &spd);
1669
1670 if (error > 0) {
1671 *ppos += error;
1672 file_accessed(in);
1673 }
1674 return error;
1675}
1676
Hugh Dickins4fb5ef02012-05-29 15:06:43 -07001677/*
1678 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1679 */
1680static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
1681 pgoff_t index, pgoff_t end, int origin)
1682{
1683 struct page *page;
1684 struct pagevec pvec;
1685 pgoff_t indices[PAGEVEC_SIZE];
1686 bool done = false;
1687 int i;
1688
1689 pagevec_init(&pvec, 0);
1690 pvec.nr = 1; /* start small: we may be there already */
1691 while (!done) {
1692 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
1693 pvec.nr, pvec.pages, indices);
1694 if (!pvec.nr) {
1695 if (origin == SEEK_DATA)
1696 index = end;
1697 break;
1698 }
1699 for (i = 0; i < pvec.nr; i++, index++) {
1700 if (index < indices[i]) {
1701 if (origin == SEEK_HOLE) {
1702 done = true;
1703 break;
1704 }
1705 index = indices[i];
1706 }
1707 page = pvec.pages[i];
1708 if (page && !radix_tree_exceptional_entry(page)) {
1709 if (!PageUptodate(page))
1710 page = NULL;
1711 }
1712 if (index >= end ||
1713 (page && origin == SEEK_DATA) ||
1714 (!page && origin == SEEK_HOLE)) {
1715 done = true;
1716 break;
1717 }
1718 }
1719 shmem_deswap_pagevec(&pvec);
1720 pagevec_release(&pvec);
1721 pvec.nr = PAGEVEC_SIZE;
1722 cond_resched();
1723 }
1724 return index;
1725}
1726
1727static loff_t shmem_file_llseek(struct file *file, loff_t offset, int origin)
1728{
1729 struct address_space *mapping;
1730 struct inode *inode;
1731 pgoff_t start, end;
1732 loff_t new_offset;
1733
1734 if (origin != SEEK_DATA && origin != SEEK_HOLE)
1735 return generic_file_llseek_size(file, offset, origin,
1736 MAX_LFS_FILESIZE);
1737 mapping = file->f_mapping;
1738 inode = mapping->host;
1739 mutex_lock(&inode->i_mutex);
1740 /* We're holding i_mutex so we can access i_size directly */
1741
1742 if (offset < 0)
1743 offset = -EINVAL;
1744 else if (offset >= inode->i_size)
1745 offset = -ENXIO;
1746 else {
1747 start = offset >> PAGE_CACHE_SHIFT;
1748 end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1749 new_offset = shmem_seek_hole_data(mapping, start, end, origin);
1750 new_offset <<= PAGE_CACHE_SHIFT;
1751 if (new_offset > offset) {
1752 if (new_offset < inode->i_size)
1753 offset = new_offset;
1754 else if (origin == SEEK_DATA)
1755 offset = -ENXIO;
1756 else
1757 offset = inode->i_size;
1758 }
1759 }
1760
1761 if (offset >= 0 && offset != file->f_pos) {
1762 file->f_pos = offset;
1763 file->f_version = 0;
1764 }
1765 mutex_unlock(&inode->i_mutex);
1766 return offset;
1767}
1768
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001769static long shmem_fallocate(struct file *file, int mode, loff_t offset,
1770 loff_t len)
1771{
1772 struct inode *inode = file->f_path.dentry->d_inode;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001773 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins1aac1402012-05-29 15:06:42 -07001774 struct shmem_falloc shmem_falloc;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001775 pgoff_t start, index, end;
1776 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001777
1778 mutex_lock(&inode->i_mutex);
1779
1780 if (mode & FALLOC_FL_PUNCH_HOLE) {
1781 struct address_space *mapping = file->f_mapping;
1782 loff_t unmap_start = round_up(offset, PAGE_SIZE);
1783 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
1784
1785 if ((u64)unmap_end > (u64)unmap_start)
1786 unmap_mapping_range(mapping, unmap_start,
1787 1 + unmap_end - unmap_start, 0);
1788 shmem_truncate_range(inode, offset, offset + len - 1);
1789 /* No need to unmap again: hole-punching leaves COWed pages */
1790 error = 0;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001791 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001792 }
1793
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001794 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1795 error = inode_newsize_ok(inode, offset + len);
1796 if (error)
1797 goto out;
1798
1799 start = offset >> PAGE_CACHE_SHIFT;
1800 end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1801 /* Try to avoid a swapstorm if len is impossible to satisfy */
1802 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
1803 error = -ENOSPC;
1804 goto out;
1805 }
1806
Hugh Dickins1aac1402012-05-29 15:06:42 -07001807 shmem_falloc.start = start;
1808 shmem_falloc.next = start;
1809 shmem_falloc.nr_falloced = 0;
1810 shmem_falloc.nr_unswapped = 0;
1811 spin_lock(&inode->i_lock);
1812 inode->i_private = &shmem_falloc;
1813 spin_unlock(&inode->i_lock);
1814
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001815 for (index = start; index < end; index++) {
1816 struct page *page;
1817
1818 /*
1819 * Good, the fallocate(2) manpage permits EINTR: we may have
1820 * been interrupted because we are using up too much memory.
1821 */
1822 if (signal_pending(current))
1823 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07001824 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
1825 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001826 else
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001827 error = shmem_getpage(inode, index, &page, SGP_FALLOC,
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001828 NULL);
1829 if (error) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001830 /* Remove the !PageUptodate pages we added */
1831 shmem_undo_range(inode,
1832 (loff_t)start << PAGE_CACHE_SHIFT,
1833 (loff_t)index << PAGE_CACHE_SHIFT, true);
Hugh Dickins1aac1402012-05-29 15:06:42 -07001834 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001835 }
1836
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001837 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07001838 * Inform shmem_writepage() how far we have reached.
1839 * No need for lock or barrier: we have the page lock.
1840 */
1841 shmem_falloc.next++;
1842 if (!PageUptodate(page))
1843 shmem_falloc.nr_falloced++;
1844
1845 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001846 * If !PageUptodate, leave it that way so that freeable pages
1847 * can be recognized if we need to rollback on error later.
1848 * But set_page_dirty so that memory pressure will swap rather
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001849 * than free the pages we are allocating (and SGP_CACHE pages
1850 * might still be clean: we now need to mark those dirty too).
1851 */
1852 set_page_dirty(page);
1853 unlock_page(page);
1854 page_cache_release(page);
1855 cond_resched();
1856 }
1857
1858 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
1859 i_size_write(inode, offset + len);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001860 inode->i_ctime = CURRENT_TIME;
Hugh Dickins1aac1402012-05-29 15:06:42 -07001861undone:
1862 spin_lock(&inode->i_lock);
1863 inode->i_private = NULL;
1864 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001865out:
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001866 mutex_unlock(&inode->i_mutex);
1867 return error;
1868}
1869
David Howells726c3342006-06-23 02:02:58 -07001870static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871{
David Howells726c3342006-06-23 02:02:58 -07001872 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873
1874 buf->f_type = TMPFS_MAGIC;
1875 buf->f_bsize = PAGE_CACHE_SIZE;
1876 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001877 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001879 buf->f_bavail =
1880 buf->f_bfree = sbinfo->max_blocks -
1881 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001882 }
1883 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 buf->f_files = sbinfo->max_inodes;
1885 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 }
1887 /* else leave those fields 0 like simple_statfs */
1888 return 0;
1889}
1890
1891/*
1892 * File creation. Allocate an inode, and we're done..
1893 */
1894static int
Al Viro1a67aaf2011-07-26 01:52:52 -04001895shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001897 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898 int error = -ENOSPC;
1899
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001900 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 if (inode) {
Eric Paris2a7dba32011-02-01 11:05:39 -05001902 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04001903 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001904 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001905 if (error) {
1906 if (error != -EOPNOTSUPP) {
1907 iput(inode);
1908 return error;
1909 }
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001910 }
Christoph Hellwig1c7c4742009-11-03 16:44:44 +01001911#ifdef CONFIG_TMPFS_POSIX_ACL
1912 error = generic_acl_init(inode, dir);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001913 if (error) {
1914 iput(inode);
1915 return error;
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001916 }
Al Viro718deb62009-12-16 19:35:36 -05001917#else
1918 error = 0;
Christoph Hellwig1c7c4742009-11-03 16:44:44 +01001919#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07001920 dir->i_size += BOGO_DIRENT_SIZE;
1921 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1922 d_instantiate(dentry, inode);
1923 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 }
1925 return error;
1926}
1927
Al Viro18bb1db2011-07-26 01:41:39 -04001928static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929{
1930 int error;
1931
1932 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1933 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07001934 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 return 0;
1936}
1937
Al Viro4acdaf22011-07-26 01:42:34 -04001938static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 struct nameidata *nd)
1940{
1941 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1942}
1943
1944/*
1945 * Link a file..
1946 */
1947static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1948{
1949 struct inode *inode = old_dentry->d_inode;
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001950 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951
1952 /*
1953 * No ordinary (disk based) filesystem counts links as inodes;
1954 * but each new link needs a new dentry, pinning lowmem, and
1955 * tmpfs dentries cannot be pruned until they are unlinked.
1956 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001957 ret = shmem_reserve_inode(inode->i_sb);
1958 if (ret)
1959 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960
1961 dir->i_size += BOGO_DIRENT_SIZE;
1962 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansend8c76e62006-09-30 23:29:04 -07001963 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04001964 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 dget(dentry); /* Extra pinning count for the created dentry */
1966 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001967out:
1968 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969}
1970
1971static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1972{
1973 struct inode *inode = dentry->d_inode;
1974
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001975 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1976 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001977
1978 dir->i_size -= BOGO_DIRENT_SIZE;
1979 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001980 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 dput(dentry); /* Undo the count from "create" - this does all the work */
1982 return 0;
1983}
1984
1985static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1986{
1987 if (!simple_empty(dentry))
1988 return -ENOTEMPTY;
1989
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001990 drop_nlink(dentry->d_inode);
1991 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 return shmem_unlink(dir, dentry);
1993}
1994
1995/*
1996 * The VFS layer already does all the dentry stuff for rename,
1997 * we just have to decrement the usage count for the target if
1998 * it exists so that the VFS layer correctly free's it when it
1999 * gets overwritten.
2000 */
2001static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2002{
2003 struct inode *inode = old_dentry->d_inode;
2004 int they_are_dirs = S_ISDIR(inode->i_mode);
2005
2006 if (!simple_empty(new_dentry))
2007 return -ENOTEMPTY;
2008
2009 if (new_dentry->d_inode) {
2010 (void) shmem_unlink(new_dir, new_dentry);
2011 if (they_are_dirs)
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002012 drop_nlink(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002014 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07002015 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 }
2017
2018 old_dir->i_size -= BOGO_DIRENT_SIZE;
2019 new_dir->i_size += BOGO_DIRENT_SIZE;
2020 old_dir->i_ctime = old_dir->i_mtime =
2021 new_dir->i_ctime = new_dir->i_mtime =
2022 inode->i_ctime = CURRENT_TIME;
2023 return 0;
2024}
2025
2026static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2027{
2028 int error;
2029 int len;
2030 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07002031 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002032 char *kaddr;
2033 struct shmem_inode_info *info;
2034
2035 len = strlen(symname) + 1;
2036 if (len > PAGE_CACHE_SIZE)
2037 return -ENAMETOOLONG;
2038
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002039 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 if (!inode)
2041 return -ENOSPC;
2042
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002043 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002044 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07002045 if (error) {
2046 if (error != -EOPNOTSUPP) {
2047 iput(inode);
2048 return error;
2049 }
2050 error = 0;
2051 }
2052
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053 info = SHMEM_I(inode);
2054 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002055 if (len <= SHORT_SYMLINK_LEN) {
2056 info->symlink = kmemdup(symname, len, GFP_KERNEL);
2057 if (!info->symlink) {
2058 iput(inode);
2059 return -ENOMEM;
2060 }
2061 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062 } else {
2063 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2064 if (error) {
2065 iput(inode);
2066 return error;
2067 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07002068 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069 inode->i_op = &shmem_symlink_inode_operations;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002070 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 memcpy(kaddr, symname, len);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002072 kunmap_atomic(kaddr);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002073 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002075 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076 page_cache_release(page);
2077 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078 dir->i_size += BOGO_DIRENT_SIZE;
2079 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2080 d_instantiate(dentry, inode);
2081 dget(dentry);
2082 return 0;
2083}
2084
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002085static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086{
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002087 nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002088 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089}
2090
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002091static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092{
2093 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002094 int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2095 nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
Hugh Dickinsd3602442008-02-04 22:28:44 -08002096 if (page)
2097 unlock_page(page);
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002098 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099}
2100
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002101static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102{
2103 if (!IS_ERR(nd_get_link(nd))) {
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002104 struct page *page = cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105 kunmap(page);
2106 mark_page_accessed(page);
2107 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 }
2109}
2110
Eric Parisb09e0fa2011-05-24 17:12:39 -07002111#ifdef CONFIG_TMPFS_XATTR
2112/*
2113 * Superblocks without xattr inode operations may get some security.* xattr
2114 * support from the LSM "for free". As soon as we have any other xattrs
2115 * like ACLs, we also need to implement the security.* handlers at
2116 * filesystem level, though.
2117 */
2118
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002119/*
2120 * Allocate new xattr and copy in the value; but leave the name to callers.
2121 */
2122static struct shmem_xattr *shmem_xattr_alloc(const void *value, size_t size)
2123{
2124 struct shmem_xattr *new_xattr;
2125 size_t len;
2126
2127 /* wrap around? */
2128 len = sizeof(*new_xattr) + size;
2129 if (len <= sizeof(*new_xattr))
2130 return NULL;
2131
2132 new_xattr = kmalloc(len, GFP_KERNEL);
2133 if (!new_xattr)
2134 return NULL;
2135
2136 new_xattr->size = size;
2137 memcpy(new_xattr->value, value, size);
2138 return new_xattr;
2139}
2140
2141/*
2142 * Callback for security_inode_init_security() for acquiring xattrs.
2143 */
2144static int shmem_initxattrs(struct inode *inode,
2145 const struct xattr *xattr_array,
2146 void *fs_info)
2147{
2148 struct shmem_inode_info *info = SHMEM_I(inode);
2149 const struct xattr *xattr;
2150 struct shmem_xattr *new_xattr;
2151 size_t len;
2152
2153 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
2154 new_xattr = shmem_xattr_alloc(xattr->value, xattr->value_len);
2155 if (!new_xattr)
2156 return -ENOMEM;
2157
2158 len = strlen(xattr->name) + 1;
2159 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2160 GFP_KERNEL);
2161 if (!new_xattr->name) {
2162 kfree(new_xattr);
2163 return -ENOMEM;
2164 }
2165
2166 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2167 XATTR_SECURITY_PREFIX_LEN);
2168 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2169 xattr->name, len);
2170
2171 spin_lock(&info->lock);
2172 list_add(&new_xattr->list, &info->xattr_list);
2173 spin_unlock(&info->lock);
2174 }
2175
2176 return 0;
2177}
2178
Eric Parisb09e0fa2011-05-24 17:12:39 -07002179static int shmem_xattr_get(struct dentry *dentry, const char *name,
2180 void *buffer, size_t size)
2181{
2182 struct shmem_inode_info *info;
2183 struct shmem_xattr *xattr;
2184 int ret = -ENODATA;
2185
2186 info = SHMEM_I(dentry->d_inode);
2187
2188 spin_lock(&info->lock);
2189 list_for_each_entry(xattr, &info->xattr_list, list) {
2190 if (strcmp(name, xattr->name))
2191 continue;
2192
2193 ret = xattr->size;
2194 if (buffer) {
2195 if (size < xattr->size)
2196 ret = -ERANGE;
2197 else
2198 memcpy(buffer, xattr->value, xattr->size);
2199 }
2200 break;
2201 }
2202 spin_unlock(&info->lock);
2203 return ret;
2204}
2205
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002206static int shmem_xattr_set(struct inode *inode, const char *name,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002207 const void *value, size_t size, int flags)
2208{
Eric Parisb09e0fa2011-05-24 17:12:39 -07002209 struct shmem_inode_info *info = SHMEM_I(inode);
2210 struct shmem_xattr *xattr;
2211 struct shmem_xattr *new_xattr = NULL;
Eric Parisb09e0fa2011-05-24 17:12:39 -07002212 int err = 0;
2213
2214 /* value == NULL means remove */
2215 if (value) {
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002216 new_xattr = shmem_xattr_alloc(value, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002217 if (!new_xattr)
2218 return -ENOMEM;
2219
2220 new_xattr->name = kstrdup(name, GFP_KERNEL);
2221 if (!new_xattr->name) {
2222 kfree(new_xattr);
2223 return -ENOMEM;
2224 }
Eric Parisb09e0fa2011-05-24 17:12:39 -07002225 }
2226
2227 spin_lock(&info->lock);
2228 list_for_each_entry(xattr, &info->xattr_list, list) {
2229 if (!strcmp(name, xattr->name)) {
2230 if (flags & XATTR_CREATE) {
2231 xattr = new_xattr;
2232 err = -EEXIST;
2233 } else if (new_xattr) {
2234 list_replace(&xattr->list, &new_xattr->list);
2235 } else {
2236 list_del(&xattr->list);
2237 }
2238 goto out;
2239 }
2240 }
2241 if (flags & XATTR_REPLACE) {
2242 xattr = new_xattr;
2243 err = -ENODATA;
2244 } else {
2245 list_add(&new_xattr->list, &info->xattr_list);
2246 xattr = NULL;
2247 }
2248out:
2249 spin_unlock(&info->lock);
2250 if (xattr)
2251 kfree(xattr->name);
2252 kfree(xattr);
2253 return err;
2254}
2255
Eric Parisb09e0fa2011-05-24 17:12:39 -07002256static const struct xattr_handler *shmem_xattr_handlers[] = {
2257#ifdef CONFIG_TMPFS_POSIX_ACL
2258 &generic_acl_access_handler,
2259 &generic_acl_default_handler,
2260#endif
2261 NULL
2262};
2263
2264static int shmem_xattr_validate(const char *name)
2265{
2266 struct { const char *prefix; size_t len; } arr[] = {
2267 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2268 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2269 };
2270 int i;
2271
2272 for (i = 0; i < ARRAY_SIZE(arr); i++) {
2273 size_t preflen = arr[i].len;
2274 if (strncmp(name, arr[i].prefix, preflen) == 0) {
2275 if (!name[preflen])
2276 return -EINVAL;
2277 return 0;
2278 }
2279 }
2280 return -EOPNOTSUPP;
2281}
2282
2283static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2284 void *buffer, size_t size)
2285{
2286 int err;
2287
2288 /*
2289 * If this is a request for a synthetic attribute in the system.*
2290 * namespace use the generic infrastructure to resolve a handler
2291 * for it via sb->s_xattr.
2292 */
2293 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2294 return generic_getxattr(dentry, name, buffer, size);
2295
2296 err = shmem_xattr_validate(name);
2297 if (err)
2298 return err;
2299
2300 return shmem_xattr_get(dentry, name, buffer, size);
2301}
2302
2303static int shmem_setxattr(struct dentry *dentry, const char *name,
2304 const void *value, size_t size, int flags)
2305{
2306 int err;
2307
2308 /*
2309 * If this is a request for a synthetic attribute in the system.*
2310 * namespace use the generic infrastructure to resolve a handler
2311 * for it via sb->s_xattr.
2312 */
2313 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2314 return generic_setxattr(dentry, name, value, size, flags);
2315
2316 err = shmem_xattr_validate(name);
2317 if (err)
2318 return err;
2319
2320 if (size == 0)
2321 value = ""; /* empty EA, do not remove */
2322
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002323 return shmem_xattr_set(dentry->d_inode, name, value, size, flags);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002324
2325}
2326
2327static int shmem_removexattr(struct dentry *dentry, const char *name)
2328{
2329 int err;
2330
2331 /*
2332 * If this is a request for a synthetic attribute in the system.*
2333 * namespace use the generic infrastructure to resolve a handler
2334 * for it via sb->s_xattr.
2335 */
2336 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2337 return generic_removexattr(dentry, name);
2338
2339 err = shmem_xattr_validate(name);
2340 if (err)
2341 return err;
2342
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002343 return shmem_xattr_set(dentry->d_inode, name, NULL, 0, XATTR_REPLACE);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002344}
2345
2346static bool xattr_is_trusted(const char *name)
2347{
2348 return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN);
2349}
2350
2351static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2352{
2353 bool trusted = capable(CAP_SYS_ADMIN);
2354 struct shmem_xattr *xattr;
2355 struct shmem_inode_info *info;
2356 size_t used = 0;
2357
2358 info = SHMEM_I(dentry->d_inode);
2359
2360 spin_lock(&info->lock);
2361 list_for_each_entry(xattr, &info->xattr_list, list) {
2362 size_t len;
2363
2364 /* skip "trusted." attributes for unprivileged callers */
2365 if (!trusted && xattr_is_trusted(xattr->name))
2366 continue;
2367
2368 len = strlen(xattr->name) + 1;
2369 used += len;
2370 if (buffer) {
2371 if (size < used) {
2372 used = -ERANGE;
2373 break;
2374 }
2375 memcpy(buffer, xattr->name, len);
2376 buffer += len;
2377 }
2378 }
2379 spin_unlock(&info->lock);
2380
2381 return used;
2382}
2383#endif /* CONFIG_TMPFS_XATTR */
2384
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002385static const struct inode_operations shmem_short_symlink_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386 .readlink = generic_readlink,
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002387 .follow_link = shmem_follow_short_symlink,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002388#ifdef CONFIG_TMPFS_XATTR
2389 .setxattr = shmem_setxattr,
2390 .getxattr = shmem_getxattr,
2391 .listxattr = shmem_listxattr,
2392 .removexattr = shmem_removexattr,
2393#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394};
2395
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002396static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397 .readlink = generic_readlink,
2398 .follow_link = shmem_follow_link,
2399 .put_link = shmem_put_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002400#ifdef CONFIG_TMPFS_XATTR
2401 .setxattr = shmem_setxattr,
2402 .getxattr = shmem_getxattr,
2403 .listxattr = shmem_listxattr,
2404 .removexattr = shmem_removexattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002405#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07002406};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002407
David M. Grimes91828a42006-10-17 00:09:45 -07002408static struct dentry *shmem_get_parent(struct dentry *child)
2409{
2410 return ERR_PTR(-ESTALE);
2411}
2412
2413static int shmem_match(struct inode *ino, void *vfh)
2414{
2415 __u32 *fh = vfh;
2416 __u64 inum = fh[2];
2417 inum = (inum << 32) | fh[1];
2418 return ino->i_ino == inum && fh[0] == ino->i_generation;
2419}
2420
Christoph Hellwig480b1162007-10-21 16:42:13 -07002421static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2422 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07002423{
David M. Grimes91828a42006-10-17 00:09:45 -07002424 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07002425 struct dentry *dentry = NULL;
2426 u64 inum = fid->raw[2];
2427 inum = (inum << 32) | fid->raw[1];
David M. Grimes91828a42006-10-17 00:09:45 -07002428
Christoph Hellwig480b1162007-10-21 16:42:13 -07002429 if (fh_len < 3)
2430 return NULL;
2431
2432 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2433 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07002434 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07002435 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002436 iput(inode);
2437 }
2438
Christoph Hellwig480b1162007-10-21 16:42:13 -07002439 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07002440}
2441
Al Virob0b03822012-04-02 14:34:06 -04002442static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2443 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07002444{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302445 if (*len < 3) {
2446 *len = 3;
David M. Grimes91828a42006-10-17 00:09:45 -07002447 return 255;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302448 }
David M. Grimes91828a42006-10-17 00:09:45 -07002449
Al Viro1d3382cb2010-10-23 15:19:20 -04002450 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07002451 /* Unfortunately insert_inode_hash is not idempotent,
2452 * so as we hash inodes here rather than at creation
2453 * time, we need a lock to ensure we only try
2454 * to do it once
2455 */
2456 static DEFINE_SPINLOCK(lock);
2457 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04002458 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07002459 __insert_inode_hash(inode,
2460 inode->i_ino + inode->i_generation);
2461 spin_unlock(&lock);
2462 }
2463
2464 fh[0] = inode->i_generation;
2465 fh[1] = inode->i_ino;
2466 fh[2] = ((__u64)inode->i_ino) >> 32;
2467
2468 *len = 3;
2469 return 1;
2470}
2471
Christoph Hellwig39655162007-10-21 16:42:17 -07002472static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07002473 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07002474 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07002475 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07002476};
2477
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002478static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2479 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480{
2481 char *this_char, *value, *rest;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002482 uid_t uid;
2483 gid_t gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002485 while (options != NULL) {
2486 this_char = options;
2487 for (;;) {
2488 /*
2489 * NUL-terminate this option: unfortunately,
2490 * mount options form a comma-separated list,
2491 * but mpol's nodelist may also contain commas.
2492 */
2493 options = strchr(options, ',');
2494 if (options == NULL)
2495 break;
2496 options++;
2497 if (!isdigit(*options)) {
2498 options[-1] = '\0';
2499 break;
2500 }
2501 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502 if (!*this_char)
2503 continue;
2504 if ((value = strchr(this_char,'=')) != NULL) {
2505 *value++ = 0;
2506 } else {
2507 printk(KERN_ERR
2508 "tmpfs: No value for mount option '%s'\n",
2509 this_char);
2510 return 1;
2511 }
2512
2513 if (!strcmp(this_char,"size")) {
2514 unsigned long long size;
2515 size = memparse(value,&rest);
2516 if (*rest == '%') {
2517 size <<= PAGE_SHIFT;
2518 size *= totalram_pages;
2519 do_div(size, 100);
2520 rest++;
2521 }
2522 if (*rest)
2523 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002524 sbinfo->max_blocks =
2525 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002527 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 if (*rest)
2529 goto bad_val;
2530 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002531 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002532 if (*rest)
2533 goto bad_val;
2534 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002535 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002537 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 if (*rest)
2539 goto bad_val;
2540 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002541 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002542 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002543 uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544 if (*rest)
2545 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002546 sbinfo->uid = make_kuid(current_user_ns(), uid);
2547 if (!uid_valid(sbinfo->uid))
2548 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002549 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002550 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002552 gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553 if (*rest)
2554 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002555 sbinfo->gid = make_kgid(current_user_ns(), gid);
2556 if (!gid_valid(sbinfo->gid))
2557 goto bad_val;
Robin Holt7339ff82006-01-14 13:20:48 -08002558 } else if (!strcmp(this_char,"mpol")) {
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002559 if (mpol_parse_str(value, &sbinfo->mpol, 1))
Robin Holt7339ff82006-01-14 13:20:48 -08002560 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561 } else {
2562 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2563 this_char);
2564 return 1;
2565 }
2566 }
2567 return 0;
2568
2569bad_val:
2570 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2571 value, this_char);
2572 return 1;
2573
2574}
2575
2576static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2577{
2578 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002579 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002580 unsigned long inodes;
2581 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002583 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002584 return error;
2585
2586 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002587 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
Tim Chen7e496292010-08-09 17:19:05 -07002588 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002589 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002590 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002591 goto out;
2592 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07002593 * Those tests disallow limited->unlimited while any are in use;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002594 * but we must separately disallow unlimited->limited, because
2595 * in that case we have no record of how much is already in use.
2596 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002597 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002598 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002599 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002600 goto out;
2601
2602 error = 0;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002603 sbinfo->max_blocks = config.max_blocks;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002604 sbinfo->max_inodes = config.max_inodes;
2605 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002606
2607 mpol_put(sbinfo->mpol);
2608 sbinfo->mpol = config.mpol; /* transfers initial ref */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002609out:
2610 spin_unlock(&sbinfo->stat_lock);
2611 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002613
Al Viro34c80b12011-12-08 21:32:45 -05002614static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002615{
Al Viro34c80b12011-12-08 21:32:45 -05002616 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002617
2618 if (sbinfo->max_blocks != shmem_default_max_blocks())
2619 seq_printf(seq, ",size=%luk",
2620 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2621 if (sbinfo->max_inodes != shmem_default_max_inodes())
2622 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2623 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04002624 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08002625 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2626 seq_printf(seq, ",uid=%u",
2627 from_kuid_munged(&init_user_ns, sbinfo->uid));
2628 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2629 seq_printf(seq, ",gid=%u",
2630 from_kgid_munged(&init_user_ns, sbinfo->gid));
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002631 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002632 return 0;
2633}
2634#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635
2636static void shmem_put_super(struct super_block *sb)
2637{
Hugh Dickins602586a2010-08-17 15:23:56 -07002638 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2639
2640 percpu_counter_destroy(&sbinfo->used_blocks);
2641 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642 sb->s_fs_info = NULL;
2643}
2644
Kay Sievers2b2af542009-04-30 15:23:42 +02002645int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646{
2647 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002648 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002649 int err = -ENOMEM;
2650
2651 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07002652 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002653 L1_CACHE_BYTES), GFP_KERNEL);
2654 if (!sbinfo)
2655 return -ENOMEM;
2656
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002657 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11002658 sbinfo->uid = current_fsuid();
2659 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002660 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002662#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663 /*
2664 * Per default we only allow half of the physical ram per
2665 * tmpfs instance, limiting inodes to one per page of lowmem;
2666 * but the internal instance is left unlimited.
2667 */
2668 if (!(sb->s_flags & MS_NOUSER)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002669 sbinfo->max_blocks = shmem_default_max_blocks();
2670 sbinfo->max_inodes = shmem_default_max_inodes();
2671 if (shmem_parse_options(data, sbinfo, false)) {
2672 err = -EINVAL;
2673 goto failed;
2674 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002675 }
David M. Grimes91828a42006-10-17 00:09:45 -07002676 sb->s_export_op = &shmem_export_ops;
Hugh Dickins2f6e38f2012-05-29 15:06:38 -07002677 sb->s_flags |= MS_NOSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678#else
2679 sb->s_flags |= MS_NOUSER;
2680#endif
2681
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002682 spin_lock_init(&sbinfo->stat_lock);
Hugh Dickins602586a2010-08-17 15:23:56 -07002683 if (percpu_counter_init(&sbinfo->used_blocks, 0))
2684 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002685 sbinfo->free_inodes = sbinfo->max_inodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002686
Hugh Dickins285b2c42011-08-03 16:21:20 -07002687 sb->s_maxbytes = MAX_LFS_FILESIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 sb->s_blocksize = PAGE_CACHE_SIZE;
2689 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2690 sb->s_magic = TMPFS_MAGIC;
2691 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01002692 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07002693#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002694 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07002695#endif
2696#ifdef CONFIG_TMPFS_POSIX_ACL
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002697 sb->s_flags |= MS_POSIXACL;
2698#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002699
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002700 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701 if (!inode)
2702 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002703 inode->i_uid = sbinfo->uid;
2704 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05002705 sb->s_root = d_make_root(inode);
2706 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05002707 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708 return 0;
2709
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710failed:
2711 shmem_put_super(sb);
2712 return err;
2713}
2714
Pekka Enbergfcc234f2006-03-22 00:08:13 -08002715static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716
2717static struct inode *shmem_alloc_inode(struct super_block *sb)
2718{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002719 struct shmem_inode_info *info;
2720 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2721 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002723 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002724}
2725
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002726static void shmem_destroy_callback(struct rcu_head *head)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11002727{
2728 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11002729 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2730}
2731
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732static void shmem_destroy_inode(struct inode *inode)
2733{
Al Viro09208d12011-07-26 03:15:03 -04002734 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002736 call_rcu(&inode->i_rcu, shmem_destroy_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737}
2738
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002739static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002741 struct shmem_inode_info *info = foo;
2742 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743}
2744
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002745static int shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746{
2747 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2748 sizeof(struct shmem_inode_info),
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002749 0, SLAB_PANIC, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750 return 0;
2751}
2752
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002753static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07002755 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002756}
2757
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002758static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08002760 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07002762 .write_begin = shmem_write_begin,
2763 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002764#endif
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07002765 .migratepage = migrate_page,
Andi Kleenaa261f52009-09-16 11:50:16 +02002766 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002767};
2768
Helge Deller15ad7cd2006-12-06 20:40:36 -08002769static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002770 .mmap = shmem_mmap,
2771#ifdef CONFIG_TMPFS
Hugh Dickins4fb5ef02012-05-29 15:06:43 -07002772 .llseek = shmem_file_llseek,
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07002773 .read = do_sync_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002774 .write = do_sync_write,
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07002775 .aio_read = shmem_file_aio_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002776 .aio_write = generic_file_aio_write,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02002777 .fsync = noop_fsync,
Hugh Dickins708e3502011-07-25 17:12:32 -07002778 .splice_read = shmem_file_splice_read,
Hugh Dickinsae9764162007-06-04 10:00:39 +02002779 .splice_write = generic_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002780 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781#endif
2782};
2783
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002784static const struct inode_operations shmem_inode_operations = {
Hugh Dickins94c1e622011-06-27 16:18:03 -07002785 .setattr = shmem_setattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002786#ifdef CONFIG_TMPFS_XATTR
2787 .setxattr = shmem_setxattr,
2788 .getxattr = shmem_getxattr,
2789 .listxattr = shmem_listxattr,
2790 .removexattr = shmem_removexattr,
2791#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792};
2793
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002794static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002795#ifdef CONFIG_TMPFS
2796 .create = shmem_create,
2797 .lookup = simple_lookup,
2798 .link = shmem_link,
2799 .unlink = shmem_unlink,
2800 .symlink = shmem_symlink,
2801 .mkdir = shmem_mkdir,
2802 .rmdir = shmem_rmdir,
2803 .mknod = shmem_mknod,
2804 .rename = shmem_rename,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07002806#ifdef CONFIG_TMPFS_XATTR
2807 .setxattr = shmem_setxattr,
2808 .getxattr = shmem_getxattr,
2809 .listxattr = shmem_listxattr,
2810 .removexattr = shmem_removexattr,
2811#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002812#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07002813 .setattr = shmem_setattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002814#endif
2815};
2816
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002817static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07002818#ifdef CONFIG_TMPFS_XATTR
2819 .setxattr = shmem_setxattr,
2820 .getxattr = shmem_getxattr,
2821 .listxattr = shmem_listxattr,
2822 .removexattr = shmem_removexattr,
2823#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002824#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07002825 .setattr = shmem_setattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002826#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827};
2828
Hugh Dickins759b9772007-03-05 00:30:28 -08002829static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830 .alloc_inode = shmem_alloc_inode,
2831 .destroy_inode = shmem_destroy_inode,
2832#ifdef CONFIG_TMPFS
2833 .statfs = shmem_statfs,
2834 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002835 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836#endif
Al Viro1f895f72010-06-05 19:10:41 -04002837 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838 .drop_inode = generic_delete_inode,
2839 .put_super = shmem_put_super,
2840};
2841
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002842static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002843 .fault = shmem_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844#ifdef CONFIG_NUMA
2845 .set_policy = shmem_set_policy,
2846 .get_policy = shmem_get_policy,
2847#endif
2848};
2849
Al Viro3c26ff62010-07-25 11:46:36 +04002850static struct dentry *shmem_mount(struct file_system_type *fs_type,
2851 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852{
Al Viro3c26ff62010-07-25 11:46:36 +04002853 return mount_nodev(fs_type, flags, data, shmem_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854}
2855
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002856static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857 .owner = THIS_MODULE,
2858 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04002859 .mount = shmem_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860 .kill_sb = kill_litter_super,
2861};
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002863int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864{
2865 int error;
2866
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002867 error = bdi_init(&shmem_backing_dev_info);
2868 if (error)
2869 goto out4;
2870
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002871 error = shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872 if (error)
2873 goto out3;
2874
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002875 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876 if (error) {
2877 printk(KERN_ERR "Could not register tmpfs\n");
2878 goto out2;
2879 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07002880
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002881 shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER,
2882 shmem_fs_type.name, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883 if (IS_ERR(shm_mnt)) {
2884 error = PTR_ERR(shm_mnt);
2885 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2886 goto out1;
2887 }
2888 return 0;
2889
2890out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002891 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002893 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002894out3:
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002895 bdi_destroy(&shmem_backing_dev_info);
2896out4:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002897 shm_mnt = ERR_PTR(error);
2898 return error;
2899}
Matt Mackall853ac432009-01-06 14:40:20 -08002900
2901#else /* !CONFIG_SHMEM */
2902
2903/*
2904 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2905 *
2906 * This is intended for small system where the benefits of the full
2907 * shmem code (swap-backed and resource-limited) are outweighed by
2908 * their complexity. On systems without swap this code should be
2909 * effectively equivalent, but much lighter weight.
2910 */
2911
2912#include <linux/ramfs.h>
2913
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002914static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08002915 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04002916 .mount = ramfs_mount,
Matt Mackall853ac432009-01-06 14:40:20 -08002917 .kill_sb = kill_litter_super,
2918};
2919
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002920int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08002921{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002922 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08002923
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002924 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08002925 BUG_ON(IS_ERR(shm_mnt));
2926
2927 return 0;
2928}
2929
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002930int shmem_unuse(swp_entry_t swap, struct page *page)
Matt Mackall853ac432009-01-06 14:40:20 -08002931{
2932 return 0;
2933}
2934
Hugh Dickins3f96b792009-09-21 17:03:37 -07002935int shmem_lock(struct file *file, int lock, struct user_struct *user)
2936{
2937 return 0;
2938}
2939
Hugh Dickins24513262012-01-20 14:34:21 -08002940void shmem_unlock_mapping(struct address_space *mapping)
2941{
2942}
2943
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002944void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07002945{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002946 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07002947}
2948EXPORT_SYMBOL_GPL(shmem_truncate_range);
2949
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002950#define shmem_vm_ops generic_file_vm_ops
2951#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002952#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002953#define shmem_acct_size(flags, size) 0
2954#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08002955
2956#endif /* CONFIG_SHMEM */
2957
2958/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002959
Randy Dunlap46711812008-03-19 17:00:41 -07002960/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961 * shmem_file_setup - get an unlinked file living in tmpfs
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962 * @name: name for dentry (to be seen in /proc/<pid>/maps
2963 * @size: size to be set for the file
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002964 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
Linus Torvalds1da177e2005-04-16 15:20:36 -07002965 */
Sergei Trofimovich168f5ac2009-06-16 15:33:02 -07002966struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002967{
2968 int error;
2969 struct file *file;
2970 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04002971 struct path path;
2972 struct dentry *root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002973 struct qstr this;
2974
2975 if (IS_ERR(shm_mnt))
2976 return (void *)shm_mnt;
2977
Hugh Dickins285b2c42011-08-03 16:21:20 -07002978 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979 return ERR_PTR(-EINVAL);
2980
2981 if (shmem_acct_size(flags, size))
2982 return ERR_PTR(-ENOMEM);
2983
2984 error = -ENOMEM;
2985 this.name = name;
2986 this.len = strlen(name);
2987 this.hash = 0; /* will go */
2988 root = shm_mnt->mnt_root;
Al Viro2c48b9c2009-08-09 00:52:35 +04002989 path.dentry = d_alloc(root, &this);
2990 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991 goto put_memory;
Al Viro2c48b9c2009-08-09 00:52:35 +04002992 path.mnt = mntget(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002993
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994 error = -ENOSPC;
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002995 inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002996 if (!inode)
Al Viro4b42af82009-08-05 18:25:56 +04002997 goto put_dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998
Al Viro2c48b9c2009-08-09 00:52:35 +04002999 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02003001 clear_nlink(inode); /* It is unlinked */
Matt Mackall853ac432009-01-06 14:40:20 -08003002#ifndef CONFIG_MMU
3003 error = ramfs_nommu_expand_for_mapping(inode, size);
3004 if (error)
Al Viro4b42af82009-08-05 18:25:56 +04003005 goto put_dentry;
Matt Mackall853ac432009-01-06 14:40:20 -08003006#endif
Al Viro4b42af82009-08-05 18:25:56 +04003007
3008 error = -ENFILE;
Al Viro2c48b9c2009-08-09 00:52:35 +04003009 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Al Viro4b42af82009-08-05 18:25:56 +04003010 &shmem_file_operations);
3011 if (!file)
3012 goto put_dentry;
3013
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014 return file;
3015
Linus Torvalds1da177e2005-04-16 15:20:36 -07003016put_dentry:
Al Viro2c48b9c2009-08-09 00:52:35 +04003017 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003018put_memory:
3019 shmem_unacct_size(flags, size);
3020 return ERR_PTR(error);
3021}
Keith Packard395e0dd2008-06-20 00:08:06 -07003022EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003023
Randy Dunlap46711812008-03-19 17:00:41 -07003024/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07003025 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07003026 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3027 */
3028int shmem_zero_setup(struct vm_area_struct *vma)
3029{
3030 struct file *file;
3031 loff_t size = vma->vm_end - vma->vm_start;
3032
3033 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
3034 if (IS_ERR(file))
3035 return PTR_ERR(file);
3036
3037 if (vma->vm_file)
3038 fput(vma->vm_file);
3039 vma->vm_file = file;
3040 vma->vm_ops = &shmem_vm_ops;
Hugh Dickinsbee4c36a2011-03-22 16:33:43 -07003041 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042 return 0;
3043}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003044
3045/**
3046 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3047 * @mapping: the page's address_space
3048 * @index: the page index
3049 * @gfp: the page allocator flags to use if allocating
3050 *
3051 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3052 * with any new page allocations done using the specified allocation flags.
3053 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3054 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3055 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3056 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07003057 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3058 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003059 */
3060struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
3061 pgoff_t index, gfp_t gfp)
3062{
Hugh Dickins68da9f02011-07-25 17:12:34 -07003063#ifdef CONFIG_SHMEM
3064 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07003065 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07003066 int error;
3067
3068 BUG_ON(mapping->a_ops != &shmem_aops);
3069 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
3070 if (error)
3071 page = ERR_PTR(error);
3072 else
3073 unlock_page(page);
3074 return page;
3075#else
3076 /*
3077 * The tiny !SHMEM case uses ramfs without swap
3078 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003079 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07003080#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003081}
3082EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);