blob: 902a14842b74a6efe5a0de23f965e8a2a9dad820 [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>
Andrew Morton250297e2013-04-29 15:06:12 -070028#include <linux/ramfs.h>
Hugh Dickinscaefba12009-04-13 14:40:12 -070029#include <linux/pagemap.h>
Matt Mackall853ac432009-01-06 14:40:20 -080030#include <linux/file.h>
31#include <linux/mm.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040032#include <linux/export.h>
Matt Mackall853ac432009-01-06 14:40:20 -080033#include <linux/swap.h>
Kent Overstreeta27bb332013-05-07 16:19:08 -070034#include <linux/aio.h>
Matt Mackall853ac432009-01-06 14:40:20 -080035
36static struct vfsmount *shm_mnt;
37
38#ifdef CONFIG_SHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -070039/*
40 * This virtual memory filesystem is heavily based on the ramfs. It
41 * extends ramfs by the ability to use swap and honor resource limits
42 * which makes it a completely usable filesystem.
43 */
44
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070045#include <linux/xattr.h>
Christoph Hellwiga5694252007-07-17 04:04:28 -070046#include <linux/exportfs.h>
Christoph Hellwig1c7c4742009-11-03 16:44:44 +010047#include <linux/posix_acl.h>
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070048#include <linux/generic_acl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include <linux/mman.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070050#include <linux/string.h>
51#include <linux/slab.h>
52#include <linux/backing-dev.h>
53#include <linux/shmem_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070054#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070055#include <linux/blkdev.h>
Hugh Dickinsbda97ea2011-08-03 16:21:21 -070056#include <linux/pagevec.h>
Hugh Dickins41ffe5d2011-08-03 16:21:21 -070057#include <linux/percpu_counter.h>
Hugh Dickins83e4fa92012-05-29 15:06:40 -070058#include <linux/falloc.h>
Hugh Dickins708e3502011-07-25 17:12:32 -070059#include <linux/splice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070060#include <linux/security.h>
61#include <linux/swapops.h>
62#include <linux/mempolicy.h>
63#include <linux/namei.h>
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +000064#include <linux/ctype.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070065#include <linux/migrate.h>
Christoph Lameterc1f60a52006-09-25 23:31:11 -070066#include <linux/highmem.h>
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -080067#include <linux/seq_file.h>
Mimi Zohar92562922008-10-07 14:00:12 -040068#include <linux/magic.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070069
Linus Torvalds1da177e2005-04-16 15:20:36 -070070#include <asm/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070071#include <asm/pgtable.h>
72
Linus Torvalds1da177e2005-04-16 15:20:36 -070073#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
Linus Torvalds1da177e2005-04-16 15:20:36 -070074#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
75
Linus Torvalds1da177e2005-04-16 15:20:36 -070076/* Pretend that each entry is of this size in directory's i_size */
77#define BOGO_DIRENT_SIZE 20
78
Hugh Dickins69f07ec2011-08-03 16:21:26 -070079/* Symlink up to this size is kmalloc'ed instead of using a swappable page */
80#define SHORT_SYMLINK_LEN 128
81
Hugh Dickins1aac1402012-05-29 15:06:42 -070082/*
83 * shmem_fallocate and shmem_writepage communicate via inode->i_private
84 * (with i_mutex making sure that it has only one user at a time):
85 * we would prefer not to enlarge the shmem inode just for that.
86 */
87struct shmem_falloc {
88 pgoff_t start; /* start of range currently being fallocated */
89 pgoff_t next; /* the next page offset to be fallocated */
90 pgoff_t nr_falloced; /* how many new pages have been fallocated */
91 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
92};
93
Hugh Dickins285b2c42011-08-03 16:21:20 -070094/* Flag allocation requirements to shmem_getpage */
Linus Torvalds1da177e2005-04-16 15:20:36 -070095enum sgp_type {
Linus Torvalds1da177e2005-04-16 15:20:36 -070096 SGP_READ, /* don't exceed i_size, don't allocate page */
97 SGP_CACHE, /* don't exceed i_size, may allocate page */
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -080098 SGP_DIRTY, /* like SGP_CACHE, but set new page dirty */
Hugh Dickins1635f6a2012-05-29 15:06:42 -070099 SGP_WRITE, /* may exceed i_size, may allocate !Uptodate page */
100 SGP_FALLOC, /* like SGP_WRITE, but make existing page Uptodate */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700101};
102
Andrew Mortonb76db732008-02-08 04:21:49 -0800103#ifdef CONFIG_TMPFS
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800104static unsigned long shmem_default_max_blocks(void)
105{
106 return totalram_pages / 2;
107}
108
109static unsigned long shmem_default_max_inodes(void)
110{
111 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
112}
Andrew Mortonb76db732008-02-08 04:21:49 -0800113#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800114
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700115static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
116static int shmem_replace_page(struct page **pagep, gfp_t gfp,
117 struct shmem_inode_info *info, pgoff_t index);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700118static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
119 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
120
121static inline int shmem_getpage(struct inode *inode, pgoff_t index,
122 struct page **pagep, enum sgp_type sgp, int *fault_type)
123{
124 return shmem_getpage_gfp(inode, index, pagep, sgp,
125 mapping_gfp_mask(inode->i_mapping), fault_type);
126}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
129{
130 return sb->s_fs_info;
131}
132
133/*
134 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
135 * for shared memory and for shared anonymous (/dev/zero) mappings
136 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
137 * consistent with the pre-accounting of private mappings ...
138 */
139static inline int shmem_acct_size(unsigned long flags, loff_t size)
140{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000141 return (flags & VM_NORESERVE) ?
Al Viro191c5422012-02-13 03:58:52 +0000142 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143}
144
145static inline void shmem_unacct_size(unsigned long flags, loff_t size)
146{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000147 if (!(flags & VM_NORESERVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 vm_unacct_memory(VM_ACCT(size));
149}
150
151/*
152 * ... whereas tmpfs objects are accounted incrementally as
153 * pages are allocated, in order to allow huge sparse files.
154 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
155 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
156 */
157static inline int shmem_acct_block(unsigned long flags)
158{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000159 return (flags & VM_NORESERVE) ?
Al Viro191c5422012-02-13 03:58:52 +0000160 security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161}
162
163static inline void shmem_unacct_blocks(unsigned long flags, long pages)
164{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000165 if (flags & VM_NORESERVE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
167}
168
Hugh Dickins759b9772007-03-05 00:30:28 -0800169static const struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700170static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800171static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800172static const struct inode_operations shmem_inode_operations;
173static const struct inode_operations shmem_dir_inode_operations;
174static const struct inode_operations shmem_special_inode_operations;
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +0400175static const struct vm_operations_struct shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700177static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 .ra_pages = 0, /* No readahead */
Rik van Riel4f98a2f2008-10-18 20:26:32 -0700179 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180};
181
182static LIST_HEAD(shmem_swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800183static DEFINE_MUTEX(shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800185static int shmem_reserve_inode(struct super_block *sb)
186{
187 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
188 if (sbinfo->max_inodes) {
189 spin_lock(&sbinfo->stat_lock);
190 if (!sbinfo->free_inodes) {
191 spin_unlock(&sbinfo->stat_lock);
192 return -ENOSPC;
193 }
194 sbinfo->free_inodes--;
195 spin_unlock(&sbinfo->stat_lock);
196 }
197 return 0;
198}
199
200static void shmem_free_inode(struct super_block *sb)
201{
202 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
203 if (sbinfo->max_inodes) {
204 spin_lock(&sbinfo->stat_lock);
205 sbinfo->free_inodes++;
206 spin_unlock(&sbinfo->stat_lock);
207 }
208}
209
Randy Dunlap46711812008-03-19 17:00:41 -0700210/**
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700211 * shmem_recalc_inode - recalculate the block usage of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 * @inode: inode to recalc
213 *
214 * We have to calculate the free blocks since the mm can drop
215 * undirtied hole pages behind our back.
216 *
217 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
218 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
219 *
220 * It has to be called with the spinlock held.
221 */
222static void shmem_recalc_inode(struct inode *inode)
223{
224 struct shmem_inode_info *info = SHMEM_I(inode);
225 long freed;
226
227 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
228 if (freed > 0) {
Hugh Dickins54af60422011-08-03 16:21:24 -0700229 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
230 if (sbinfo->max_blocks)
231 percpu_counter_add(&sbinfo->used_blocks, -freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232 info->alloced -= freed;
Hugh Dickins54af60422011-08-03 16:21:24 -0700233 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 shmem_unacct_blocks(info->flags, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 }
236}
237
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700238/*
239 * Replace item expected in radix tree by a new item, while holding tree lock.
240 */
241static int shmem_radix_tree_replace(struct address_space *mapping,
242 pgoff_t index, void *expected, void *replacement)
243{
244 void **pslot;
245 void *item = NULL;
246
247 VM_BUG_ON(!expected);
248 pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
249 if (pslot)
250 item = radix_tree_deref_slot_protected(pslot,
251 &mapping->tree_lock);
252 if (item != expected)
253 return -ENOENT;
254 if (replacement)
255 radix_tree_replace_slot(pslot, replacement);
256 else
257 radix_tree_delete(&mapping->page_tree, index);
258 return 0;
259}
260
261/*
Hugh Dickinsd1899222012-07-11 14:02:47 -0700262 * Sometimes, before we decide whether to proceed or to fail, we must check
263 * that an entry was not already brought back from swap by a racing thread.
264 *
265 * Checking page is not enough: by the time a SwapCache page is locked, it
266 * might be reused, and again be SwapCache, using the same swap as before.
267 */
268static bool shmem_confirm_swap(struct address_space *mapping,
269 pgoff_t index, swp_entry_t swap)
270{
271 void *item;
272
273 rcu_read_lock();
274 item = radix_tree_lookup(&mapping->page_tree, index);
275 rcu_read_unlock();
276 return item == swp_to_radix_entry(swap);
277}
278
279/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700280 * Like add_to_page_cache_locked, but error if expected item has gone.
281 */
282static int shmem_add_to_page_cache(struct page *page,
283 struct address_space *mapping,
284 pgoff_t index, gfp_t gfp, void *expected)
285{
Hugh Dickinsb065b432012-07-11 14:02:48 -0700286 int error;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700287
288 VM_BUG_ON(!PageLocked(page));
289 VM_BUG_ON(!PageSwapBacked(page));
290
Hugh Dickinsb065b432012-07-11 14:02:48 -0700291 page_cache_get(page);
292 page->mapping = mapping;
293 page->index = index;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700294
Hugh Dickinsb065b432012-07-11 14:02:48 -0700295 spin_lock_irq(&mapping->tree_lock);
296 if (!expected)
297 error = radix_tree_insert(&mapping->page_tree, index, page);
298 else
299 error = shmem_radix_tree_replace(mapping, index, expected,
300 page);
301 if (!error) {
302 mapping->nrpages++;
303 __inc_zone_page_state(page, NR_FILE_PAGES);
304 __inc_zone_page_state(page, NR_SHMEM);
305 spin_unlock_irq(&mapping->tree_lock);
306 } else {
307 page->mapping = NULL;
308 spin_unlock_irq(&mapping->tree_lock);
309 page_cache_release(page);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700310 }
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700311 return error;
312}
313
314/*
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700315 * Like delete_from_page_cache, but substitutes swap for page.
316 */
317static void shmem_delete_from_page_cache(struct page *page, void *radswap)
318{
319 struct address_space *mapping = page->mapping;
320 int error;
321
322 spin_lock_irq(&mapping->tree_lock);
323 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
324 page->mapping = NULL;
325 mapping->nrpages--;
326 __dec_zone_page_state(page, NR_FILE_PAGES);
327 __dec_zone_page_state(page, NR_SHMEM);
328 spin_unlock_irq(&mapping->tree_lock);
329 page_cache_release(page);
330 BUG_ON(error);
331}
332
333/*
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700334 * Like find_get_pages, but collecting swap entries as well as pages.
335 */
336static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping,
337 pgoff_t start, unsigned int nr_pages,
338 struct page **pages, pgoff_t *indices)
339{
Johannes Weiner860f2752013-02-22 16:35:17 -0800340 void **slot;
341 unsigned int ret = 0;
342 struct radix_tree_iter iter;
343
344 if (!nr_pages)
345 return 0;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700346
347 rcu_read_lock();
348restart:
Johannes Weiner860f2752013-02-22 16:35:17 -0800349 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700350 struct page *page;
351repeat:
Johannes Weiner860f2752013-02-22 16:35:17 -0800352 page = radix_tree_deref_slot(slot);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700353 if (unlikely(!page))
354 continue;
355 if (radix_tree_exception(page)) {
Hugh Dickins8079b1c2011-08-03 16:21:28 -0700356 if (radix_tree_deref_retry(page))
357 goto restart;
358 /*
359 * Otherwise, we must be storing a swap entry
360 * here as an exceptional entry: so return it
361 * without attempting to raise page count.
362 */
363 goto export;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700364 }
365 if (!page_cache_get_speculative(page))
366 goto repeat;
367
368 /* Has the page moved? */
Johannes Weiner860f2752013-02-22 16:35:17 -0800369 if (unlikely(page != *slot)) {
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700370 page_cache_release(page);
371 goto repeat;
372 }
373export:
Johannes Weiner860f2752013-02-22 16:35:17 -0800374 indices[ret] = iter.index;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700375 pages[ret] = page;
Johannes Weiner860f2752013-02-22 16:35:17 -0800376 if (++ret == nr_pages)
377 break;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700378 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700379 rcu_read_unlock();
380 return ret;
381}
382
383/*
384 * Remove swap entry from radix tree, free the swap and its page cache.
385 */
386static int shmem_free_swap(struct address_space *mapping,
387 pgoff_t index, void *radswap)
388{
389 int error;
390
391 spin_lock_irq(&mapping->tree_lock);
392 error = shmem_radix_tree_replace(mapping, index, radswap, NULL);
393 spin_unlock_irq(&mapping->tree_lock);
394 if (!error)
395 free_swap_and_cache(radix_to_swp_entry(radswap));
396 return error;
397}
398
399/*
400 * Pagevec may contain swap entries, so shuffle up pages before releasing.
401 */
Hugh Dickins24513262012-01-20 14:34:21 -0800402static void shmem_deswap_pagevec(struct pagevec *pvec)
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700403{
404 int i, j;
405
406 for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
407 struct page *page = pvec->pages[i];
408 if (!radix_tree_exceptional_entry(page))
409 pvec->pages[j++] = page;
410 }
411 pvec->nr = j;
Hugh Dickins24513262012-01-20 14:34:21 -0800412}
413
414/*
415 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
416 */
417void shmem_unlock_mapping(struct address_space *mapping)
418{
419 struct pagevec pvec;
420 pgoff_t indices[PAGEVEC_SIZE];
421 pgoff_t index = 0;
422
423 pagevec_init(&pvec, 0);
424 /*
425 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
426 */
427 while (!mapping_unevictable(mapping)) {
428 /*
429 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
430 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
431 */
432 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
433 PAGEVEC_SIZE, pvec.pages, indices);
434 if (!pvec.nr)
435 break;
436 index = indices[pvec.nr - 1] + 1;
437 shmem_deswap_pagevec(&pvec);
438 check_move_unevictable_pages(pvec.pages, pvec.nr);
439 pagevec_release(&pvec);
440 cond_resched();
441 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700442}
443
444/*
445 * Remove range of pages and swap entries from radix tree, and free them.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700446 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700447 */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700448static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
449 bool unfalloc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700451 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700453 pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700454 pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
455 unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
456 unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700457 struct pagevec pvec;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700458 pgoff_t indices[PAGEVEC_SIZE];
459 long nr_swaps_freed = 0;
Hugh Dickins285b2c42011-08-03 16:21:20 -0700460 pgoff_t index;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700461 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700463 if (lend == -1)
464 end = -1; /* unsigned, so actually very big */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700465
466 pagevec_init(&pvec, 0);
467 index = start;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700468 while (index < end) {
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700469 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700470 min(end - index, (pgoff_t)PAGEVEC_SIZE),
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700471 pvec.pages, indices);
472 if (!pvec.nr)
473 break;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700474 mem_cgroup_uncharge_start();
475 for (i = 0; i < pagevec_count(&pvec); i++) {
476 struct page *page = pvec.pages[i];
477
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700478 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700479 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700480 break;
481
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700482 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700483 if (unfalloc)
484 continue;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700485 nr_swaps_freed += !shmem_free_swap(mapping,
486 index, page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700487 continue;
488 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700489
490 if (!trylock_page(page))
491 continue;
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700492 if (!unfalloc || !PageUptodate(page)) {
493 if (page->mapping == mapping) {
494 VM_BUG_ON(PageWriteback(page));
495 truncate_inode_page(mapping, page);
496 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700497 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700498 unlock_page(page);
499 }
Hugh Dickins24513262012-01-20 14:34:21 -0800500 shmem_deswap_pagevec(&pvec);
501 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700502 mem_cgroup_uncharge_end();
503 cond_resched();
504 index++;
505 }
506
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700507 if (partial_start) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700508 struct page *page = NULL;
509 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
510 if (page) {
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700511 unsigned int top = PAGE_CACHE_SIZE;
512 if (start > end) {
513 top = partial_end;
514 partial_end = 0;
515 }
516 zero_user_segment(page, partial_start, top);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700517 set_page_dirty(page);
518 unlock_page(page);
519 page_cache_release(page);
520 }
521 }
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700522 if (partial_end) {
523 struct page *page = NULL;
524 shmem_getpage(inode, end, &page, SGP_READ, NULL);
525 if (page) {
526 zero_user_segment(page, 0, partial_end);
527 set_page_dirty(page);
528 unlock_page(page);
529 page_cache_release(page);
530 }
531 }
532 if (start >= end)
533 return;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700534
535 index = start;
536 for ( ; ; ) {
537 cond_resched();
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700538 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700539 min(end - index, (pgoff_t)PAGEVEC_SIZE),
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700540 pvec.pages, indices);
541 if (!pvec.nr) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700542 if (index == start || unfalloc)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700543 break;
544 index = start;
545 continue;
546 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700547 if ((index == start || unfalloc) && indices[0] >= end) {
Hugh Dickins24513262012-01-20 14:34:21 -0800548 shmem_deswap_pagevec(&pvec);
549 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700550 break;
551 }
552 mem_cgroup_uncharge_start();
553 for (i = 0; i < pagevec_count(&pvec); i++) {
554 struct page *page = pvec.pages[i];
555
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700556 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700557 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700558 break;
559
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700560 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700561 if (unfalloc)
562 continue;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700563 nr_swaps_freed += !shmem_free_swap(mapping,
564 index, page);
565 continue;
566 }
567
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700568 lock_page(page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700569 if (!unfalloc || !PageUptodate(page)) {
570 if (page->mapping == mapping) {
571 VM_BUG_ON(PageWriteback(page));
572 truncate_inode_page(mapping, page);
573 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700574 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700575 unlock_page(page);
576 }
Hugh Dickins24513262012-01-20 14:34:21 -0800577 shmem_deswap_pagevec(&pvec);
578 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700579 mem_cgroup_uncharge_end();
580 index++;
581 }
Hugh Dickins94c1e622011-06-27 16:18:03 -0700582
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 spin_lock(&info->lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700584 info->swapped -= nr_swaps_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 shmem_recalc_inode(inode);
586 spin_unlock(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700587}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700589void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
590{
591 shmem_undo_range(inode, lstart, lend, false);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700592 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593}
Hugh Dickins94c1e622011-06-27 16:18:03 -0700594EXPORT_SYMBOL_GPL(shmem_truncate_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595
Hugh Dickins94c1e622011-06-27 16:18:03 -0700596static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597{
598 struct inode *inode = dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599 int error;
600
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200601 error = inode_change_ok(inode, attr);
602 if (error)
603 return error;
604
Hugh Dickins94c1e622011-06-27 16:18:03 -0700605 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
606 loff_t oldsize = inode->i_size;
607 loff_t newsize = attr->ia_size;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000608
Hugh Dickins94c1e622011-06-27 16:18:03 -0700609 if (newsize != oldsize) {
610 i_size_write(inode, newsize);
611 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
612 }
613 if (newsize < oldsize) {
614 loff_t holebegin = round_up(newsize, PAGE_SIZE);
615 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
616 shmem_truncate_range(inode, newsize, (loff_t)-1);
617 /* unmap again to remove racily COWed private pages */
618 unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
619 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 }
621
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200622 setattr_copy(inode, attr);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700623#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200624 if (attr->ia_valid & ATTR_MODE)
Christoph Hellwig1c7c4742009-11-03 16:44:44 +0100625 error = generic_acl_chmod(inode);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700626#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 return error;
628}
629
Al Viro1f895f72010-06-05 19:10:41 -0400630static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632 struct shmem_inode_info *info = SHMEM_I(inode);
633
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000634 if (inode->i_mapping->a_ops == &shmem_aops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700635 shmem_unacct_size(info->flags, inode->i_size);
636 inode->i_size = 0;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000637 shmem_truncate_range(inode, 0, (loff_t)-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700638 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800639 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800641 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 }
Hugh Dickins69f07ec2011-08-03 16:21:26 -0700643 } else
644 kfree(info->symlink);
Eric Parisb09e0fa2011-05-24 17:12:39 -0700645
Aristeu Rozanski38f38652012-08-23 16:53:28 -0400646 simple_xattrs_free(&info->xattrs);
Hugh Dickins0f3c42f2012-11-16 14:15:04 -0800647 WARN_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800648 shmem_free_inode(inode->i_sb);
Jan Karadbd57682012-05-03 14:48:02 +0200649 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650}
651
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700652/*
653 * If swap found in inode, free it and move page from swapcache to filecache.
654 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700655static int shmem_unuse_inode(struct shmem_inode_info *info,
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700656 swp_entry_t swap, struct page **pagep)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700657{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700658 struct address_space *mapping = info->vfs_inode.i_mapping;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700659 void *radswap;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700660 pgoff_t index;
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700661 gfp_t gfp;
662 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700663
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700664 radswap = swp_to_radix_entry(swap);
Hugh Dickinse504f3f2011-08-03 16:21:27 -0700665 index = radix_tree_locate_item(&mapping->page_tree, radswap);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700666 if (index == -1)
Hugh Dickins285b2c42011-08-03 16:21:20 -0700667 return 0;
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800668
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800669 /*
670 * Move _head_ to start search for next from here.
Al Viro1f895f72010-06-05 19:10:41 -0400671 * But be careful: shmem_evict_inode checks list_empty without taking
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800672 * mutex, and there's an instant in list_move_tail when info->swaplist
Hugh Dickins285b2c42011-08-03 16:21:20 -0700673 * would appear empty, if it were the only one on shmem_swaplist.
Hugh Dickins1b1b32f2008-02-04 22:28:55 -0800674 */
675 if (shmem_swaplist.next != &info->swaplist)
676 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800677
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700678 gfp = mapping_gfp_mask(mapping);
679 if (shmem_should_replace_page(*pagep, gfp)) {
680 mutex_unlock(&shmem_swaplist_mutex);
681 error = shmem_replace_page(pagep, gfp, info, index);
682 mutex_lock(&shmem_swaplist_mutex);
683 /*
684 * We needed to drop mutex to make that restrictive page
Hugh Dickins0142ef62012-06-07 14:21:09 -0700685 * allocation, but the inode might have been freed while we
686 * dropped it: although a racing shmem_evict_inode() cannot
687 * complete without emptying the radix_tree, our page lock
688 * on this swapcache page is not enough to prevent that -
689 * free_swap_and_cache() of our swap entry will only
690 * trylock_page(), removing swap from radix_tree whatever.
691 *
692 * We must not proceed to shmem_add_to_page_cache() if the
693 * inode has been freed, but of course we cannot rely on
694 * inode or mapping or info to check that. However, we can
695 * safely check if our swap entry is still in use (and here
696 * it can't have got reused for another page): if it's still
697 * in use, then the inode cannot have been freed yet, and we
698 * can safely proceed (if it's no longer in use, that tells
699 * nothing about the inode, but we don't need to unuse swap).
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700700 */
701 if (!page_swapcount(*pagep))
702 error = -ENOENT;
703 }
704
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800705 /*
Hugh Dickins778dd892011-05-11 15:13:37 -0700706 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
707 * but also to hold up shmem_evict_inode(): so inode cannot be freed
708 * beneath us (pagelock doesn't help until the page is in pagecache).
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -0800709 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700710 if (!error)
711 error = shmem_add_to_page_cache(*pagep, mapping, index,
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700712 GFP_NOWAIT, radswap);
Hugh Dickins48f170f2011-07-25 17:12:37 -0700713 if (error != -ENOMEM) {
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700714 /*
715 * Truncation and eviction use free_swap_and_cache(), which
716 * only does trylock page: if we raced, best clean up here.
717 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700718 delete_from_swap_cache(*pagep);
719 set_page_dirty(*pagep);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700720 if (!error) {
721 spin_lock(&info->lock);
722 info->swapped--;
723 spin_unlock(&info->lock);
724 swap_free(swap);
725 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800726 error = 1; /* not an error, but entry was found */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800728 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729}
730
731/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700732 * Search through swapped inodes to find and replace swap by page.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700734int shmem_unuse(swp_entry_t swap, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700736 struct list_head *this, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 struct shmem_inode_info *info;
738 int found = 0;
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700739 int error = 0;
740
741 /*
742 * There's a faint possibility that swap page was replaced before
Hugh Dickins0142ef62012-06-07 14:21:09 -0700743 * caller locked it: caller will come back later with the right page.
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700744 */
Hugh Dickins0142ef62012-06-07 14:21:09 -0700745 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700746 goto out;
Hugh Dickins778dd892011-05-11 15:13:37 -0700747
748 /*
749 * Charge page using GFP_KERNEL while we can wait, before taking
750 * the shmem_swaplist_mutex which might hold up shmem_writepage().
751 * Charged back to the user (not to caller) when swap account is used.
Hugh Dickins778dd892011-05-11 15:13:37 -0700752 */
753 error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
754 if (error)
755 goto out;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700756 /* No radix_tree_preload: swap entry keeps a place for page in tree */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800758 mutex_lock(&shmem_swaplist_mutex);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700759 list_for_each_safe(this, next, &shmem_swaplist) {
760 info = list_entry(this, struct shmem_inode_info, swaplist);
Hugh Dickins285b2c42011-08-03 16:21:20 -0700761 if (info->swapped)
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700762 found = shmem_unuse_inode(info, swap, &page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700763 else
764 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800765 cond_resched();
Hugh Dickins2e0e26c2008-02-04 22:28:53 -0800766 if (found)
Hugh Dickins778dd892011-05-11 15:13:37 -0700767 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700768 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800769 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins778dd892011-05-11 15:13:37 -0700770
Hugh Dickins778dd892011-05-11 15:13:37 -0700771 if (found < 0)
772 error = found;
773out:
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800774 unlock_page(page);
775 page_cache_release(page);
Hugh Dickins778dd892011-05-11 15:13:37 -0700776 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777}
778
779/*
780 * Move the page from the page cache to the swap cache.
781 */
782static int shmem_writepage(struct page *page, struct writeback_control *wbc)
783{
784 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 struct inode *inode;
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700787 swp_entry_t swap;
788 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789
790 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 mapping = page->mapping;
792 index = page->index;
793 inode = mapping->host;
794 info = SHMEM_I(inode);
795 if (info->flags & VM_LOCKED)
796 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800797 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798 goto redirty;
799
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800800 /*
801 * shmem_backing_dev_info's capabilities prevent regular writeback or
802 * sync from ever calling shmem_writepage; but a stacking filesystem
Hugh Dickins48f170f2011-07-25 17:12:37 -0700803 * might use ->writepage of its underlying filesystem, in which case
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800804 * tmpfs should write out to swap only in response to memory pressure,
Hugh Dickins48f170f2011-07-25 17:12:37 -0700805 * and not for the writeback threads or sync.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800806 */
Hugh Dickins48f170f2011-07-25 17:12:37 -0700807 if (!wbc->for_reclaim) {
808 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
809 goto redirty;
810 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700811
812 /*
813 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
814 * value into swapfile.c, the only way we can correctly account for a
815 * fallocated page arriving here is now to initialize it and write it.
Hugh Dickins1aac1402012-05-29 15:06:42 -0700816 *
817 * That's okay for a page already fallocated earlier, but if we have
818 * not yet completed the fallocation, then (a) we want to keep track
819 * of this page in case we have to undo it, and (b) it may not be a
820 * good idea to continue anyway, once we're pushing into swap. So
821 * reactivate the page, and let shmem_fallocate() quit when too many.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700822 */
823 if (!PageUptodate(page)) {
Hugh Dickins1aac1402012-05-29 15:06:42 -0700824 if (inode->i_private) {
825 struct shmem_falloc *shmem_falloc;
826 spin_lock(&inode->i_lock);
827 shmem_falloc = inode->i_private;
828 if (shmem_falloc &&
829 index >= shmem_falloc->start &&
830 index < shmem_falloc->next)
831 shmem_falloc->nr_unswapped++;
832 else
833 shmem_falloc = NULL;
834 spin_unlock(&inode->i_lock);
835 if (shmem_falloc)
836 goto redirty;
837 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700838 clear_highpage(page);
839 flush_dcache_page(page);
840 SetPageUptodate(page);
841 }
842
Hugh Dickins48f170f2011-07-25 17:12:37 -0700843 swap = get_swap_page();
844 if (!swap.val)
845 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800846
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700847 /*
848 * Add inode to shmem_unuse()'s list of swapped-out inodes,
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700849 * if it's not already there. Do it now before the page is
850 * moved to swap cache, when its pagelock no longer protects
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700851 * the inode from eviction. But don't unlock the mutex until
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700852 * we've incremented swapped, because shmem_unuse_inode() will
853 * prune a !swapped inode from the swaplist under this mutex.
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700854 */
Hugh Dickins48f170f2011-07-25 17:12:37 -0700855 mutex_lock(&shmem_swaplist_mutex);
856 if (list_empty(&info->swaplist))
857 list_add_tail(&info->swaplist, &shmem_swaplist);
Hugh Dickinsb1dea802011-05-11 15:13:36 -0700858
Hugh Dickins48f170f2011-07-25 17:12:37 -0700859 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Hugh Dickinsaaa46862009-12-14 17:58:47 -0800860 swap_shmem_alloc(swap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700861 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
862
863 spin_lock(&info->lock);
864 info->swapped++;
865 shmem_recalc_inode(inode);
Hugh Dickins826267c2011-05-28 13:14:09 -0700866 spin_unlock(&info->lock);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700867
868 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800869 BUG_ON(page_mapped(page));
Hugh Dickins9fab5612009-03-31 15:23:33 -0700870 swap_writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 return 0;
872 }
873
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700874 mutex_unlock(&shmem_swaplist_mutex);
KAMEZAWA Hiroyukicb4b86b2009-06-16 15:32:52 -0700875 swapcache_free(swap, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700876redirty:
877 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -0800878 if (wbc->for_reclaim)
879 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
880 unlock_page(page);
881 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882}
883
884#ifdef CONFIG_NUMA
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800885#ifdef CONFIG_TMPFS
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700886static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800887{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700888 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800889
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700890 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700891 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800892
Hugh Dickinsa7a88b22013-01-02 02:04:23 -0800893 mpol_to_str(buffer, sizeof(buffer), mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800894
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -0700895 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800896}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700897
898static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
899{
900 struct mempolicy *mpol = NULL;
901 if (sbinfo->mpol) {
902 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
903 mpol = sbinfo->mpol;
904 mpol_get(mpol);
905 spin_unlock(&sbinfo->stat_lock);
906 }
907 return mpol;
908}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800909#endif /* CONFIG_TMPFS */
910
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700911static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
912 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700913{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700914 struct vm_area_struct pvma;
Mel Gorman18a2f372012-12-05 14:01:41 -0800915 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700916
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800917 /* Create a pseudo vma that just contains the policy */
918 pvma.vm_start = 0;
Nathan Zimmer09c231c2012-07-31 16:46:17 -0700919 /* Bias interleave by inode number to distribute better across nodes */
920 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
Hugh Dickinsc4cc6d02008-02-04 22:28:40 -0800921 pvma.vm_ops = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700922 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
Lee Schermerhorn52cd3b02008-04-28 02:13:16 -0700923
Mel Gorman18a2f372012-12-05 14:01:41 -0800924 page = swapin_readahead(swap, gfp, &pvma, 0);
925
926 /* Drop reference taken by mpol_shared_policy_lookup() */
927 mpol_cond_put(pvma.vm_policy);
928
929 return page;
930}
931
932static struct page *shmem_alloc_page(gfp_t gfp,
933 struct shmem_inode_info *info, pgoff_t index)
934{
935 struct vm_area_struct pvma;
936 struct page *page;
937
938 /* Create a pseudo vma that just contains the policy */
939 pvma.vm_start = 0;
940 /* Bias interleave by inode number to distribute better across nodes */
941 pvma.vm_pgoff = index + info->vfs_inode.i_ino;
942 pvma.vm_ops = NULL;
943 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
944
945 page = alloc_page_vma(gfp, &pvma, 0);
946
947 /* Drop reference taken by mpol_shared_policy_lookup() */
948 mpol_cond_put(pvma.vm_policy);
949
950 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800952#else /* !CONFIG_NUMA */
953#ifdef CONFIG_TMPFS
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700954static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800955{
956}
957#endif /* CONFIG_TMPFS */
958
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700959static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
960 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700962 return swapin_readahead(swap, gfp, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963}
964
Hugh Dickins02098fe2008-02-04 22:28:42 -0800965static inline struct page *shmem_alloc_page(gfp_t gfp,
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700966 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967{
Hugh Dickinse84e2e12007-11-28 18:55:10 +0000968 return alloc_page(gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800970#endif /* CONFIG_NUMA */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971
Lee Schermerhorn71fe8042008-04-28 02:13:26 -0700972#if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
973static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
974{
975 return NULL;
976}
977#endif
978
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700980 * When a page is moved from swapcache to shmem filecache (either by the
981 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
982 * shmem_unuse_inode()), it may have been read in earlier from swap, in
983 * ignorance of the mapping it belongs to. If that mapping has special
984 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
985 * we may need to copy to a suitable page before moving to filecache.
986 *
987 * In a future release, this may well be extended to respect cpuset and
988 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
989 * but for now it is a simple matter of zone.
990 */
991static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
992{
993 return page_zonenum(page) > gfp_zone(gfp);
994}
995
996static int shmem_replace_page(struct page **pagep, gfp_t gfp,
997 struct shmem_inode_info *info, pgoff_t index)
998{
999 struct page *oldpage, *newpage;
1000 struct address_space *swap_mapping;
1001 pgoff_t swap_index;
1002 int error;
1003
1004 oldpage = *pagep;
1005 swap_index = page_private(oldpage);
1006 swap_mapping = page_mapping(oldpage);
1007
1008 /*
1009 * We have arrived here because our zones are constrained, so don't
1010 * limit chance of success by further cpuset and node constraints.
1011 */
1012 gfp &= ~GFP_CONSTRAINT_MASK;
1013 newpage = shmem_alloc_page(gfp, info, index);
1014 if (!newpage)
1015 return -ENOMEM;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001016
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001017 page_cache_get(newpage);
1018 copy_highpage(newpage, oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001019 flush_dcache_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001020
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001021 __set_page_locked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001022 SetPageUptodate(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001023 SetPageSwapBacked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001024 set_page_private(newpage, swap_index);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001025 SetPageSwapCache(newpage);
1026
1027 /*
1028 * Our caller will very soon move newpage out of swapcache, but it's
1029 * a nice clean interface for us to replace oldpage by newpage there.
1030 */
1031 spin_lock_irq(&swap_mapping->tree_lock);
1032 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1033 newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001034 if (!error) {
1035 __inc_zone_page_state(newpage, NR_FILE_PAGES);
1036 __dec_zone_page_state(oldpage, NR_FILE_PAGES);
1037 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001038 spin_unlock_irq(&swap_mapping->tree_lock);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001039
Hugh Dickins0142ef62012-06-07 14:21:09 -07001040 if (unlikely(error)) {
1041 /*
1042 * Is this possible? I think not, now that our callers check
1043 * both PageSwapCache and page_private after getting page lock;
1044 * but be defensive. Reverse old to newpage for clear and free.
1045 */
1046 oldpage = newpage;
1047 } else {
1048 mem_cgroup_replace_page_cache(oldpage, newpage);
1049 lru_cache_add_anon(newpage);
1050 *pagep = newpage;
1051 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001052
1053 ClearPageSwapCache(oldpage);
1054 set_page_private(oldpage, 0);
1055
1056 unlock_page(oldpage);
1057 page_cache_release(oldpage);
1058 page_cache_release(oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001059 return error;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001060}
1061
1062/*
Hugh Dickins68da9f02011-07-25 17:12:34 -07001063 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 *
1065 * If we allocate a new one we do not mark it dirty. That's up to the
1066 * vm. If we swap it in we mark it dirty since we also free the swap
1067 * entry since a page cannot live in both the swap and page cache
1068 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001069static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Hugh Dickins68da9f02011-07-25 17:12:34 -07001070 struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071{
1072 struct address_space *mapping = inode->i_mapping;
Hugh Dickins54af60422011-08-03 16:21:24 -07001073 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 struct shmem_sb_info *sbinfo;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001075 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 swp_entry_t swap;
1077 int error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001078 int once = 0;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001079 int alloced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001081 if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001082 return -EFBIG;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083repeat:
Hugh Dickins54af60422011-08-03 16:21:24 -07001084 swap.val = 0;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001085 page = find_lock_page(mapping, index);
Hugh Dickins54af60422011-08-03 16:21:24 -07001086 if (radix_tree_exceptional_entry(page)) {
1087 swap = radix_to_swp_entry(page);
1088 page = NULL;
1089 }
1090
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001091 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
Hugh Dickins54af60422011-08-03 16:21:24 -07001092 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1093 error = -EINVAL;
1094 goto failed;
1095 }
1096
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001097 /* fallocated page? */
1098 if (page && !PageUptodate(page)) {
1099 if (sgp != SGP_READ)
1100 goto clear;
1101 unlock_page(page);
1102 page_cache_release(page);
1103 page = NULL;
1104 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001105 if (page || (sgp == SGP_READ && !swap.val)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001106 *pagep = page;
1107 return 0;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001108 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109
1110 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001111 * Fast cache lookup did not find it:
1112 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001114 info = SHMEM_I(inode);
1115 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001116
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 if (swap.val) {
1118 /* Look it up and read it in.. */
Hugh Dickins27ab7002011-07-25 17:12:36 -07001119 page = lookup_swap_cache(swap);
1120 if (!page) {
Ying Han456f9982011-05-26 16:25:38 -07001121 /* here we actually do the io */
Hugh Dickins68da9f02011-07-25 17:12:34 -07001122 if (fault_type)
1123 *fault_type |= VM_FAULT_MAJOR;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001124 page = shmem_swapin(swap, gfp, info, index);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001125 if (!page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001126 error = -ENOMEM;
1127 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001128 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001129 }
1130
1131 /* We have to do this with page locked to prevent races */
Hugh Dickins54af60422011-08-03 16:21:24 -07001132 lock_page(page);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001133 if (!PageSwapCache(page) || page_private(page) != swap.val ||
Hugh Dickinsd1899222012-07-11 14:02:47 -07001134 !shmem_confirm_swap(mapping, index, swap)) {
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001135 error = -EEXIST; /* try again */
Hugh Dickinsd1899222012-07-11 14:02:47 -07001136 goto unlock;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001137 }
Hugh Dickins27ab7002011-07-25 17:12:36 -07001138 if (!PageUptodate(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 error = -EIO;
Hugh Dickins54af60422011-08-03 16:21:24 -07001140 goto failed;
1141 }
1142 wait_on_page_writeback(page);
1143
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001144 if (shmem_should_replace_page(page, gfp)) {
1145 error = shmem_replace_page(&page, gfp, info, index);
1146 if (error)
1147 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 }
1149
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001150 error = mem_cgroup_cache_charge(page, current->mm,
1151 gfp & GFP_RECLAIM_MASK);
Hugh Dickinsd1899222012-07-11 14:02:47 -07001152 if (!error) {
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001153 error = shmem_add_to_page_cache(page, mapping, index,
1154 gfp, swp_to_radix_entry(swap));
Hugh Dickins215c02b2012-11-16 14:15:03 -08001155 /*
1156 * We already confirmed swap under page lock, and make
1157 * no memory allocation here, so usually no possibility
1158 * of error; but free_swap_and_cache() only trylocks a
1159 * page, so it is just possible that the entry has been
1160 * truncated or holepunched since swap was confirmed.
1161 * shmem_undo_range() will have done some of the
1162 * unaccounting, now delete_from_swap_cache() will do
1163 * the rest (including mem_cgroup_uncharge_swapcache).
1164 * Reset swap.val? No, leave it so "failed" goes back to
1165 * "repeat": reading a hole and writing should succeed.
1166 */
1167 if (error)
1168 delete_from_swap_cache(page);
Hugh Dickinsd1899222012-07-11 14:02:47 -07001169 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001170 if (error)
1171 goto failed;
1172
1173 spin_lock(&info->lock);
1174 info->swapped--;
1175 shmem_recalc_inode(inode);
1176 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001177
Hugh Dickins27ab7002011-07-25 17:12:36 -07001178 delete_from_swap_cache(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001179 set_page_dirty(page);
1180 swap_free(swap);
1181
Hugh Dickins54af60422011-08-03 16:21:24 -07001182 } else {
1183 if (shmem_acct_block(info->flags)) {
1184 error = -ENOSPC;
1185 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001187 if (sbinfo->max_blocks) {
Hugh Dickinsfc5da222011-04-14 15:22:07 -07001188 if (percpu_counter_compare(&sbinfo->used_blocks,
Hugh Dickins54af60422011-08-03 16:21:24 -07001189 sbinfo->max_blocks) >= 0) {
1190 error = -ENOSPC;
1191 goto unacct;
1192 }
Tim Chen7e496292010-08-09 17:19:05 -07001193 percpu_counter_inc(&sbinfo->used_blocks);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001194 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195
Hugh Dickins54af60422011-08-03 16:21:24 -07001196 page = shmem_alloc_page(gfp, info, index);
1197 if (!page) {
1198 error = -ENOMEM;
1199 goto decused;
1200 }
1201
1202 SetPageSwapBacked(page);
1203 __set_page_locked(page);
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001204 error = mem_cgroup_cache_charge(page, current->mm,
1205 gfp & GFP_RECLAIM_MASK);
Hugh Dickins54af60422011-08-03 16:21:24 -07001206 if (error)
1207 goto decused;
Jan Kara5e4c0d972013-09-11 14:26:05 -07001208 error = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001209 if (!error) {
1210 error = shmem_add_to_page_cache(page, mapping, index,
1211 gfp, NULL);
1212 radix_tree_preload_end();
1213 }
1214 if (error) {
1215 mem_cgroup_uncharge_cache_page(page);
1216 goto decused;
1217 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001218 lru_cache_add_anon(page);
1219
1220 spin_lock(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 info->alloced++;
Hugh Dickins54af60422011-08-03 16:21:24 -07001222 inode->i_blocks += BLOCKS_PER_PAGE;
1223 shmem_recalc_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 spin_unlock(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001225 alloced = true;
Hugh Dickins54af60422011-08-03 16:21:24 -07001226
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001227 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001228 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1229 */
1230 if (sgp == SGP_FALLOC)
1231 sgp = SGP_WRITE;
1232clear:
1233 /*
1234 * Let SGP_WRITE caller clear ends if write does not fill page;
1235 * but SGP_FALLOC on a page fallocated earlier must initialize
1236 * it now, lest undo on failure cancel our earlier guarantee.
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001237 */
1238 if (sgp != SGP_WRITE) {
1239 clear_highpage(page);
1240 flush_dcache_page(page);
1241 SetPageUptodate(page);
1242 }
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001243 if (sgp == SGP_DIRTY)
Hugh Dickins27ab7002011-07-25 17:12:36 -07001244 set_page_dirty(page);
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001245 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001246
Hugh Dickins54af60422011-08-03 16:21:24 -07001247 /* Perhaps the file has been truncated since we checked */
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001248 if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
Hugh Dickins54af60422011-08-03 16:21:24 -07001249 ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1250 error = -EINVAL;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001251 if (alloced)
1252 goto trunc;
1253 else
1254 goto failed;
Shaohua Liff36b8012010-08-09 17:19:06 -07001255 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001256 *pagep = page;
1257 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07001258
Nick Piggind0217ac2007-07-19 01:47:03 -07001259 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001260 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001262trunc:
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001263 info = SHMEM_I(inode);
Hugh Dickins54af60422011-08-03 16:21:24 -07001264 ClearPageDirty(page);
1265 delete_from_page_cache(page);
1266 spin_lock(&info->lock);
1267 info->alloced--;
1268 inode->i_blocks -= BLOCKS_PER_PAGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001269 spin_unlock(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001270decused:
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001271 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins54af60422011-08-03 16:21:24 -07001272 if (sbinfo->max_blocks)
1273 percpu_counter_add(&sbinfo->used_blocks, -1);
1274unacct:
1275 shmem_unacct_blocks(info->flags, 1);
1276failed:
Hugh Dickinsd1899222012-07-11 14:02:47 -07001277 if (swap.val && error != -EINVAL &&
1278 !shmem_confirm_swap(mapping, index, swap))
1279 error = -EEXIST;
1280unlock:
Hugh Dickins27ab7002011-07-25 17:12:36 -07001281 if (page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001282 unlock_page(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001283 page_cache_release(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001284 }
1285 if (error == -ENOSPC && !once++) {
1286 info = SHMEM_I(inode);
1287 spin_lock(&info->lock);
1288 shmem_recalc_inode(inode);
1289 spin_unlock(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001290 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001291 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001292 if (error == -EEXIST) /* from above or from radix_tree_insert */
Hugh Dickins54af60422011-08-03 16:21:24 -07001293 goto repeat;
1294 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295}
1296
1297static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1298{
Al Viro496ad9a2013-01-23 17:07:38 -05001299 struct inode *inode = file_inode(vma->vm_file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001300 int error;
Hugh Dickins68da9f02011-07-25 17:12:34 -07001301 int ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1304 if (error)
1305 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Hugh Dickins68da9f02011-07-25 17:12:34 -07001306
Ying Han456f9982011-05-26 16:25:38 -07001307 if (ret & VM_FAULT_MAJOR) {
1308 count_vm_event(PGMAJFAULT);
1309 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1310 }
Hugh Dickins68da9f02011-07-25 17:12:34 -07001311 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312}
1313
1314#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001315static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316{
Al Viro496ad9a2013-01-23 17:07:38 -05001317 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001318 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319}
1320
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001321static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1322 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323{
Al Viro496ad9a2013-01-23 17:07:38 -05001324 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001325 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001327 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1328 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329}
1330#endif
1331
1332int shmem_lock(struct file *file, int lock, struct user_struct *user)
1333{
Al Viro496ad9a2013-01-23 17:07:38 -05001334 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 struct shmem_inode_info *info = SHMEM_I(inode);
1336 int retval = -ENOMEM;
1337
1338 spin_lock(&info->lock);
1339 if (lock && !(info->flags & VM_LOCKED)) {
1340 if (!user_shm_lock(inode->i_size, user))
1341 goto out_nomem;
1342 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001343 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 }
1345 if (!lock && (info->flags & VM_LOCKED) && user) {
1346 user_shm_unlock(inode->i_size, user);
1347 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001348 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 }
1350 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07001351
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352out_nomem:
1353 spin_unlock(&info->lock);
1354 return retval;
1355}
1356
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08001357static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001358{
1359 file_accessed(file);
1360 vma->vm_ops = &shmem_vm_ops;
1361 return 0;
1362}
1363
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001364static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04001365 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366{
1367 struct inode *inode;
1368 struct shmem_inode_info *info;
1369 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1370
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001371 if (shmem_reserve_inode(sb))
1372 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373
1374 inode = new_inode(sb);
1375 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -04001376 inode->i_ino = get_next_ino();
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001377 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 inode->i_blocks = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1380 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
David M. Grimes91828a42006-10-17 00:09:45 -07001381 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382 info = SHMEM_I(inode);
1383 memset(info, 0, (char *)inode - (char *)info);
1384 spin_lock_init(&info->lock);
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001385 info->flags = flags & VM_NORESERVE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 INIT_LIST_HEAD(&info->swaplist);
Aristeu Rozanski38f38652012-08-23 16:53:28 -04001387 simple_xattrs_init(&info->xattrs);
Al Viro72c04902009-06-24 16:58:48 -04001388 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389
1390 switch (mode & S_IFMT) {
1391 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001392 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001393 init_special_inode(inode, mode, dev);
1394 break;
1395 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07001396 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 inode->i_op = &shmem_inode_operations;
1398 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001399 mpol_shared_policy_init(&info->policy,
1400 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 break;
1402 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07001403 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404 /* Some things misbehave if size == 0 on a directory */
1405 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1406 inode->i_op = &shmem_dir_inode_operations;
1407 inode->i_fop = &simple_dir_operations;
1408 break;
1409 case S_IFLNK:
1410 /*
1411 * Must not load anything in the rbtree,
1412 * mpol_free_shared_policy will not be called.
1413 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001414 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 break;
1416 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001417 } else
1418 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 return inode;
1420}
1421
1422#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001423static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07001424static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001426#ifdef CONFIG_TMPFS_XATTR
1427static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1428#else
1429#define shmem_initxattrs NULL
1430#endif
1431
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432static int
Nick Piggin800d15a2007-10-16 01:25:03 -07001433shmem_write_begin(struct file *file, struct address_space *mapping,
1434 loff_t pos, unsigned len, unsigned flags,
1435 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436{
Nick Piggin800d15a2007-10-16 01:25:03 -07001437 struct inode *inode = mapping->host;
1438 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
Nick Piggin800d15a2007-10-16 01:25:03 -07001439 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1440}
1441
1442static int
1443shmem_write_end(struct file *file, struct address_space *mapping,
1444 loff_t pos, unsigned len, unsigned copied,
1445 struct page *page, void *fsdata)
1446{
1447 struct inode *inode = mapping->host;
1448
Hugh Dickinsd3602442008-02-04 22:28:44 -08001449 if (pos + copied > inode->i_size)
1450 i_size_write(inode, pos + copied);
1451
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001452 if (!PageUptodate(page)) {
1453 if (copied < PAGE_CACHE_SIZE) {
1454 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1455 zero_user_segments(page, 0, from,
1456 from + copied, PAGE_CACHE_SIZE);
1457 }
1458 SetPageUptodate(page);
1459 }
Nick Piggin800d15a2007-10-16 01:25:03 -07001460 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02001461 unlock_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07001462 page_cache_release(page);
1463
Nick Piggin800d15a2007-10-16 01:25:03 -07001464 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465}
1466
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1468{
Al Viro496ad9a2013-01-23 17:07:38 -05001469 struct inode *inode = file_inode(filp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001471 pgoff_t index;
1472 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001473 enum sgp_type sgp = SGP_READ;
1474
1475 /*
1476 * Might this read be for a stacking filesystem? Then when reading
1477 * holes of a sparse file, we actually need to allocate those pages,
1478 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1479 */
1480 if (segment_eq(get_fs(), KERNEL_DS))
1481 sgp = SGP_DIRTY;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482
1483 index = *ppos >> PAGE_CACHE_SHIFT;
1484 offset = *ppos & ~PAGE_CACHE_MASK;
1485
1486 for (;;) {
1487 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001488 pgoff_t end_index;
1489 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490 loff_t i_size = i_size_read(inode);
1491
1492 end_index = i_size >> PAGE_CACHE_SHIFT;
1493 if (index > end_index)
1494 break;
1495 if (index == end_index) {
1496 nr = i_size & ~PAGE_CACHE_MASK;
1497 if (nr <= offset)
1498 break;
1499 }
1500
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08001501 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 if (desc->error) {
1503 if (desc->error == -EINVAL)
1504 desc->error = 0;
1505 break;
1506 }
Hugh Dickinsd3602442008-02-04 22:28:44 -08001507 if (page)
1508 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509
1510 /*
1511 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001512 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 */
1514 nr = PAGE_CACHE_SIZE;
1515 i_size = i_size_read(inode);
1516 end_index = i_size >> PAGE_CACHE_SHIFT;
1517 if (index == end_index) {
1518 nr = i_size & ~PAGE_CACHE_MASK;
1519 if (nr <= offset) {
1520 if (page)
1521 page_cache_release(page);
1522 break;
1523 }
1524 }
1525 nr -= offset;
1526
1527 if (page) {
1528 /*
1529 * If users can be writing to this page using arbitrary
1530 * virtual addresses, take care about potential aliasing
1531 * before reading the page on the kernel side.
1532 */
1533 if (mapping_writably_mapped(mapping))
1534 flush_dcache_page(page);
1535 /*
1536 * Mark the page accessed if we read the beginning.
1537 */
1538 if (!offset)
1539 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001540 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541 page = ZERO_PAGE(0);
Nick Pigginb5810032005-10-29 18:16:12 -07001542 page_cache_get(page);
1543 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544
1545 /*
1546 * Ok, we have the page, and it's up-to-date, so
1547 * now we can copy it to user space...
1548 *
1549 * The actor routine returns how many bytes were actually used..
1550 * NOTE! This may not be the same as how much of a user buffer
1551 * we filled up (we may be padding etc), so we can only update
1552 * "pos" here (the actor routine has to update the user buffer
1553 * pointers and the remaining count).
1554 */
1555 ret = actor(desc, page, offset, nr);
1556 offset += ret;
1557 index += offset >> PAGE_CACHE_SHIFT;
1558 offset &= ~PAGE_CACHE_MASK;
1559
1560 page_cache_release(page);
1561 if (ret != nr || !desc->count)
1562 break;
1563
1564 cond_resched();
1565 }
1566
1567 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1568 file_accessed(filp);
1569}
1570
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001571static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1572 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573{
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001574 struct file *filp = iocb->ki_filp;
1575 ssize_t retval;
1576 unsigned long seg;
1577 size_t count;
1578 loff_t *ppos = &iocb->ki_pos;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001580 retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1581 if (retval)
1582 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001584 for (seg = 0; seg < nr_segs; seg++) {
1585 read_descriptor_t desc;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07001587 desc.written = 0;
1588 desc.arg.buf = iov[seg].iov_base;
1589 desc.count = iov[seg].iov_len;
1590 if (desc.count == 0)
1591 continue;
1592 desc.error = 0;
1593 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1594 retval += desc.written;
1595 if (desc.error) {
1596 retval = retval ?: desc.error;
1597 break;
1598 }
1599 if (desc.count > 0)
1600 break;
1601 }
1602 return retval;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603}
1604
Hugh Dickins708e3502011-07-25 17:12:32 -07001605static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1606 struct pipe_inode_info *pipe, size_t len,
1607 unsigned int flags)
1608{
1609 struct address_space *mapping = in->f_mapping;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001610 struct inode *inode = mapping->host;
Hugh Dickins708e3502011-07-25 17:12:32 -07001611 unsigned int loff, nr_pages, req_pages;
1612 struct page *pages[PIPE_DEF_BUFFERS];
1613 struct partial_page partial[PIPE_DEF_BUFFERS];
1614 struct page *page;
1615 pgoff_t index, end_index;
1616 loff_t isize, left;
1617 int error, page_nr;
1618 struct splice_pipe_desc spd = {
1619 .pages = pages,
1620 .partial = partial,
Eric Dumazet047fe362012-06-12 15:24:40 +02001621 .nr_pages_max = PIPE_DEF_BUFFERS,
Hugh Dickins708e3502011-07-25 17:12:32 -07001622 .flags = flags,
1623 .ops = &page_cache_pipe_buf_ops,
1624 .spd_release = spd_release_page,
1625 };
1626
Hugh Dickins71f0e072011-07-25 17:12:33 -07001627 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07001628 if (unlikely(*ppos >= isize))
1629 return 0;
1630
1631 left = isize - *ppos;
1632 if (unlikely(left < len))
1633 len = left;
1634
1635 if (splice_grow_spd(pipe, &spd))
1636 return -ENOMEM;
1637
1638 index = *ppos >> PAGE_CACHE_SHIFT;
1639 loff = *ppos & ~PAGE_CACHE_MASK;
1640 req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1641 nr_pages = min(req_pages, pipe->buffers);
1642
Hugh Dickins708e3502011-07-25 17:12:32 -07001643 spd.nr_pages = find_get_pages_contig(mapping, index,
1644 nr_pages, spd.pages);
1645 index += spd.nr_pages;
Hugh Dickins708e3502011-07-25 17:12:32 -07001646 error = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001647
Hugh Dickins708e3502011-07-25 17:12:32 -07001648 while (spd.nr_pages < nr_pages) {
Hugh Dickins71f0e072011-07-25 17:12:33 -07001649 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1650 if (error)
1651 break;
1652 unlock_page(page);
Hugh Dickins708e3502011-07-25 17:12:32 -07001653 spd.pages[spd.nr_pages++] = page;
1654 index++;
1655 }
1656
Hugh Dickins708e3502011-07-25 17:12:32 -07001657 index = *ppos >> PAGE_CACHE_SHIFT;
1658 nr_pages = spd.nr_pages;
1659 spd.nr_pages = 0;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001660
Hugh Dickins708e3502011-07-25 17:12:32 -07001661 for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1662 unsigned int this_len;
1663
1664 if (!len)
1665 break;
1666
Hugh Dickins708e3502011-07-25 17:12:32 -07001667 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1668 page = spd.pages[page_nr];
1669
Hugh Dickins71f0e072011-07-25 17:12:33 -07001670 if (!PageUptodate(page) || page->mapping != mapping) {
Hugh Dickins71f0e072011-07-25 17:12:33 -07001671 error = shmem_getpage(inode, index, &page,
1672 SGP_CACHE, NULL);
1673 if (error)
Hugh Dickins708e3502011-07-25 17:12:32 -07001674 break;
Hugh Dickins71f0e072011-07-25 17:12:33 -07001675 unlock_page(page);
1676 page_cache_release(spd.pages[page_nr]);
1677 spd.pages[page_nr] = page;
Hugh Dickins708e3502011-07-25 17:12:32 -07001678 }
Hugh Dickins71f0e072011-07-25 17:12:33 -07001679
1680 isize = i_size_read(inode);
Hugh Dickins708e3502011-07-25 17:12:32 -07001681 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1682 if (unlikely(!isize || index > end_index))
1683 break;
1684
Hugh Dickins708e3502011-07-25 17:12:32 -07001685 if (end_index == index) {
1686 unsigned int plen;
1687
Hugh Dickins708e3502011-07-25 17:12:32 -07001688 plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1689 if (plen <= loff)
1690 break;
1691
Hugh Dickins708e3502011-07-25 17:12:32 -07001692 this_len = min(this_len, plen - loff);
1693 len = this_len;
1694 }
1695
1696 spd.partial[page_nr].offset = loff;
1697 spd.partial[page_nr].len = this_len;
1698 len -= this_len;
1699 loff = 0;
1700 spd.nr_pages++;
1701 index++;
1702 }
1703
Hugh Dickins708e3502011-07-25 17:12:32 -07001704 while (page_nr < nr_pages)
1705 page_cache_release(spd.pages[page_nr++]);
Hugh Dickins708e3502011-07-25 17:12:32 -07001706
1707 if (spd.nr_pages)
1708 error = splice_to_pipe(pipe, &spd);
1709
Eric Dumazet047fe362012-06-12 15:24:40 +02001710 splice_shrink_spd(&spd);
Hugh Dickins708e3502011-07-25 17:12:32 -07001711
1712 if (error > 0) {
1713 *ppos += error;
1714 file_accessed(in);
1715 }
1716 return error;
1717}
1718
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001719/*
1720 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1721 */
1722static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
Andrew Morton965c8e52012-12-17 15:59:39 -08001723 pgoff_t index, pgoff_t end, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001724{
1725 struct page *page;
1726 struct pagevec pvec;
1727 pgoff_t indices[PAGEVEC_SIZE];
1728 bool done = false;
1729 int i;
1730
1731 pagevec_init(&pvec, 0);
1732 pvec.nr = 1; /* start small: we may be there already */
1733 while (!done) {
1734 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
1735 pvec.nr, pvec.pages, indices);
1736 if (!pvec.nr) {
Andrew Morton965c8e52012-12-17 15:59:39 -08001737 if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001738 index = end;
1739 break;
1740 }
1741 for (i = 0; i < pvec.nr; i++, index++) {
1742 if (index < indices[i]) {
Andrew Morton965c8e52012-12-17 15:59:39 -08001743 if (whence == SEEK_HOLE) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001744 done = true;
1745 break;
1746 }
1747 index = indices[i];
1748 }
1749 page = pvec.pages[i];
1750 if (page && !radix_tree_exceptional_entry(page)) {
1751 if (!PageUptodate(page))
1752 page = NULL;
1753 }
1754 if (index >= end ||
Andrew Morton965c8e52012-12-17 15:59:39 -08001755 (page && whence == SEEK_DATA) ||
1756 (!page && whence == SEEK_HOLE)) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001757 done = true;
1758 break;
1759 }
1760 }
1761 shmem_deswap_pagevec(&pvec);
1762 pagevec_release(&pvec);
1763 pvec.nr = PAGEVEC_SIZE;
1764 cond_resched();
1765 }
1766 return index;
1767}
1768
Andrew Morton965c8e52012-12-17 15:59:39 -08001769static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001770{
1771 struct address_space *mapping = file->f_mapping;
1772 struct inode *inode = mapping->host;
1773 pgoff_t start, end;
1774 loff_t new_offset;
1775
Andrew Morton965c8e52012-12-17 15:59:39 -08001776 if (whence != SEEK_DATA && whence != SEEK_HOLE)
1777 return generic_file_llseek_size(file, offset, whence,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001778 MAX_LFS_FILESIZE, i_size_read(inode));
1779 mutex_lock(&inode->i_mutex);
1780 /* We're holding i_mutex so we can access i_size directly */
1781
1782 if (offset < 0)
1783 offset = -EINVAL;
1784 else if (offset >= inode->i_size)
1785 offset = -ENXIO;
1786 else {
1787 start = offset >> PAGE_CACHE_SHIFT;
1788 end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Andrew Morton965c8e52012-12-17 15:59:39 -08001789 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001790 new_offset <<= PAGE_CACHE_SHIFT;
1791 if (new_offset > offset) {
1792 if (new_offset < inode->i_size)
1793 offset = new_offset;
Andrew Morton965c8e52012-12-17 15:59:39 -08001794 else if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001795 offset = -ENXIO;
1796 else
1797 offset = inode->i_size;
1798 }
1799 }
1800
Hugh Dickins387aae62013-08-04 11:30:25 -07001801 if (offset >= 0)
1802 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08001803 mutex_unlock(&inode->i_mutex);
1804 return offset;
1805}
1806
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001807static long shmem_fallocate(struct file *file, int mode, loff_t offset,
1808 loff_t len)
1809{
Al Viro496ad9a2013-01-23 17:07:38 -05001810 struct inode *inode = file_inode(file);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001811 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins1aac1402012-05-29 15:06:42 -07001812 struct shmem_falloc shmem_falloc;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001813 pgoff_t start, index, end;
1814 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001815
1816 mutex_lock(&inode->i_mutex);
1817
1818 if (mode & FALLOC_FL_PUNCH_HOLE) {
1819 struct address_space *mapping = file->f_mapping;
1820 loff_t unmap_start = round_up(offset, PAGE_SIZE);
1821 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
1822
1823 if ((u64)unmap_end > (u64)unmap_start)
1824 unmap_mapping_range(mapping, unmap_start,
1825 1 + unmap_end - unmap_start, 0);
1826 shmem_truncate_range(inode, offset, offset + len - 1);
1827 /* No need to unmap again: hole-punching leaves COWed pages */
1828 error = 0;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001829 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001830 }
1831
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001832 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1833 error = inode_newsize_ok(inode, offset + len);
1834 if (error)
1835 goto out;
1836
1837 start = offset >> PAGE_CACHE_SHIFT;
1838 end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1839 /* Try to avoid a swapstorm if len is impossible to satisfy */
1840 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
1841 error = -ENOSPC;
1842 goto out;
1843 }
1844
Hugh Dickins1aac1402012-05-29 15:06:42 -07001845 shmem_falloc.start = start;
1846 shmem_falloc.next = start;
1847 shmem_falloc.nr_falloced = 0;
1848 shmem_falloc.nr_unswapped = 0;
1849 spin_lock(&inode->i_lock);
1850 inode->i_private = &shmem_falloc;
1851 spin_unlock(&inode->i_lock);
1852
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001853 for (index = start; index < end; index++) {
1854 struct page *page;
1855
1856 /*
1857 * Good, the fallocate(2) manpage permits EINTR: we may have
1858 * been interrupted because we are using up too much memory.
1859 */
1860 if (signal_pending(current))
1861 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07001862 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
1863 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001864 else
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001865 error = shmem_getpage(inode, index, &page, SGP_FALLOC,
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001866 NULL);
1867 if (error) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001868 /* Remove the !PageUptodate pages we added */
1869 shmem_undo_range(inode,
1870 (loff_t)start << PAGE_CACHE_SHIFT,
1871 (loff_t)index << PAGE_CACHE_SHIFT, true);
Hugh Dickins1aac1402012-05-29 15:06:42 -07001872 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001873 }
1874
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001875 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07001876 * Inform shmem_writepage() how far we have reached.
1877 * No need for lock or barrier: we have the page lock.
1878 */
1879 shmem_falloc.next++;
1880 if (!PageUptodate(page))
1881 shmem_falloc.nr_falloced++;
1882
1883 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001884 * If !PageUptodate, leave it that way so that freeable pages
1885 * can be recognized if we need to rollback on error later.
1886 * But set_page_dirty so that memory pressure will swap rather
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001887 * than free the pages we are allocating (and SGP_CACHE pages
1888 * might still be clean: we now need to mark those dirty too).
1889 */
1890 set_page_dirty(page);
1891 unlock_page(page);
1892 page_cache_release(page);
1893 cond_resched();
1894 }
1895
1896 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
1897 i_size_write(inode, offset + len);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001898 inode->i_ctime = CURRENT_TIME;
Hugh Dickins1aac1402012-05-29 15:06:42 -07001899undone:
1900 spin_lock(&inode->i_lock);
1901 inode->i_private = NULL;
1902 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07001903out:
Hugh Dickins83e4fa92012-05-29 15:06:40 -07001904 mutex_unlock(&inode->i_mutex);
1905 return error;
1906}
1907
David Howells726c3342006-06-23 02:02:58 -07001908static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909{
David Howells726c3342006-06-23 02:02:58 -07001910 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911
1912 buf->f_type = TMPFS_MAGIC;
1913 buf->f_bsize = PAGE_CACHE_SIZE;
1914 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001915 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001917 buf->f_bavail =
1918 buf->f_bfree = sbinfo->max_blocks -
1919 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001920 }
1921 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922 buf->f_files = sbinfo->max_inodes;
1923 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 }
1925 /* else leave those fields 0 like simple_statfs */
1926 return 0;
1927}
1928
1929/*
1930 * File creation. Allocate an inode, and we're done..
1931 */
1932static int
Al Viro1a67aaf2011-07-26 01:52:52 -04001933shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00001935 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001936 int error = -ENOSPC;
1937
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03001938 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 if (inode) {
Mimi Zohar37ec43c2013-04-14 09:21:47 -04001940#ifdef CONFIG_TMPFS_POSIX_ACL
1941 error = generic_acl_init(inode, dir);
1942 if (error) {
1943 iput(inode);
1944 return error;
1945 }
1946#endif
Eric Paris2a7dba32011-02-01 11:05:39 -05001947 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04001948 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07001949 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001950 if (error) {
1951 if (error != -EOPNOTSUPP) {
1952 iput(inode);
1953 return error;
1954 }
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001955 }
Mimi Zohar37ec43c2013-04-14 09:21:47 -04001956
Al Viro718deb62009-12-16 19:35:36 -05001957 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 dir->i_size += BOGO_DIRENT_SIZE;
1959 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1960 d_instantiate(dentry, inode);
1961 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962 }
1963 return error;
1964}
1965
Al Viro60545d02013-06-07 01:20:27 -04001966static int
1967shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
1968{
1969 struct inode *inode;
1970 int error = -ENOSPC;
1971
1972 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
1973 if (inode) {
1974 error = security_inode_init_security(inode, dir,
1975 NULL,
1976 shmem_initxattrs, NULL);
1977 if (error) {
1978 if (error != -EOPNOTSUPP) {
1979 iput(inode);
1980 return error;
1981 }
1982 }
1983#ifdef CONFIG_TMPFS_POSIX_ACL
1984 error = generic_acl_init(inode, dir);
1985 if (error) {
1986 iput(inode);
1987 return error;
1988 }
1989#else
1990 error = 0;
1991#endif
1992 d_tmpfile(dentry, inode);
1993 }
1994 return error;
1995}
1996
Al Viro18bb1db2011-07-26 01:41:39 -04001997static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998{
1999 int error;
2000
2001 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
2002 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07002003 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 return 0;
2005}
2006
Al Viro4acdaf22011-07-26 01:42:34 -04002007static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -04002008 bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009{
2010 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
2011}
2012
2013/*
2014 * Link a file..
2015 */
2016static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2017{
2018 struct inode *inode = old_dentry->d_inode;
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002019 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020
2021 /*
2022 * No ordinary (disk based) filesystem counts links as inodes;
2023 * but each new link needs a new dentry, pinning lowmem, and
2024 * tmpfs dentries cannot be pruned until they are unlinked.
2025 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002026 ret = shmem_reserve_inode(inode->i_sb);
2027 if (ret)
2028 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029
2030 dir->i_size += BOGO_DIRENT_SIZE;
2031 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansend8c76e62006-09-30 23:29:04 -07002032 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04002033 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034 dget(dentry); /* Extra pinning count for the created dentry */
2035 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002036out:
2037 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038}
2039
2040static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2041{
2042 struct inode *inode = dentry->d_inode;
2043
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002044 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2045 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046
2047 dir->i_size -= BOGO_DIRENT_SIZE;
2048 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002049 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 dput(dentry); /* Undo the count from "create" - this does all the work */
2051 return 0;
2052}
2053
2054static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2055{
2056 if (!simple_empty(dentry))
2057 return -ENOTEMPTY;
2058
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002059 drop_nlink(dentry->d_inode);
2060 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 return shmem_unlink(dir, dentry);
2062}
2063
2064/*
2065 * The VFS layer already does all the dentry stuff for rename,
2066 * we just have to decrement the usage count for the target if
2067 * it exists so that the VFS layer correctly free's it when it
2068 * gets overwritten.
2069 */
2070static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2071{
2072 struct inode *inode = old_dentry->d_inode;
2073 int they_are_dirs = S_ISDIR(inode->i_mode);
2074
2075 if (!simple_empty(new_dentry))
2076 return -ENOTEMPTY;
2077
2078 if (new_dentry->d_inode) {
2079 (void) shmem_unlink(new_dir, new_dentry);
2080 if (they_are_dirs)
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002081 drop_nlink(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002082 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07002083 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07002084 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 }
2086
2087 old_dir->i_size -= BOGO_DIRENT_SIZE;
2088 new_dir->i_size += BOGO_DIRENT_SIZE;
2089 old_dir->i_ctime = old_dir->i_mtime =
2090 new_dir->i_ctime = new_dir->i_mtime =
2091 inode->i_ctime = CURRENT_TIME;
2092 return 0;
2093}
2094
2095static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2096{
2097 int error;
2098 int len;
2099 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07002100 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002101 char *kaddr;
2102 struct shmem_inode_info *info;
2103
2104 len = strlen(symname) + 1;
2105 if (len > PAGE_CACHE_SIZE)
2106 return -ENAMETOOLONG;
2107
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002108 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 if (!inode)
2110 return -ENOSPC;
2111
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002112 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002113 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07002114 if (error) {
2115 if (error != -EOPNOTSUPP) {
2116 iput(inode);
2117 return error;
2118 }
2119 error = 0;
2120 }
2121
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 info = SHMEM_I(inode);
2123 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002124 if (len <= SHORT_SYMLINK_LEN) {
2125 info->symlink = kmemdup(symname, len, GFP_KERNEL);
2126 if (!info->symlink) {
2127 iput(inode);
2128 return -ENOMEM;
2129 }
2130 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002131 } else {
2132 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2133 if (error) {
2134 iput(inode);
2135 return error;
2136 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07002137 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 inode->i_op = &shmem_symlink_inode_operations;
Cong Wang9b04c5f2011-11-25 23:14:39 +08002139 kaddr = kmap_atomic(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 memcpy(kaddr, symname, len);
Cong Wang9b04c5f2011-11-25 23:14:39 +08002141 kunmap_atomic(kaddr);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002142 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002143 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002144 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 page_cache_release(page);
2146 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147 dir->i_size += BOGO_DIRENT_SIZE;
2148 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2149 d_instantiate(dentry, inode);
2150 dget(dentry);
2151 return 0;
2152}
2153
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002154static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155{
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002156 nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002157 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158}
2159
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002160static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161{
2162 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002163 int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2164 nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
Hugh Dickinsd3602442008-02-04 22:28:44 -08002165 if (page)
2166 unlock_page(page);
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002167 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168}
2169
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002170static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171{
2172 if (!IS_ERR(nd_get_link(nd))) {
Linus Torvaldscc314ee2005-08-19 18:02:56 -07002173 struct page *page = cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 kunmap(page);
2175 mark_page_accessed(page);
2176 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 }
2178}
2179
Eric Parisb09e0fa2011-05-24 17:12:39 -07002180#ifdef CONFIG_TMPFS_XATTR
2181/*
2182 * Superblocks without xattr inode operations may get some security.* xattr
2183 * support from the LSM "for free". As soon as we have any other xattrs
2184 * like ACLs, we also need to implement the security.* handlers at
2185 * filesystem level, though.
2186 */
2187
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002188/*
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002189 * Callback for security_inode_init_security() for acquiring xattrs.
2190 */
2191static int shmem_initxattrs(struct inode *inode,
2192 const struct xattr *xattr_array,
2193 void *fs_info)
2194{
2195 struct shmem_inode_info *info = SHMEM_I(inode);
2196 const struct xattr *xattr;
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002197 struct simple_xattr *new_xattr;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002198 size_t len;
2199
2200 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002201 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002202 if (!new_xattr)
2203 return -ENOMEM;
2204
2205 len = strlen(xattr->name) + 1;
2206 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2207 GFP_KERNEL);
2208 if (!new_xattr->name) {
2209 kfree(new_xattr);
2210 return -ENOMEM;
2211 }
2212
2213 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2214 XATTR_SECURITY_PREFIX_LEN);
2215 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2216 xattr->name, len);
2217
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002218 simple_xattr_list_add(&info->xattrs, new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002219 }
2220
2221 return 0;
2222}
2223
Eric Parisb09e0fa2011-05-24 17:12:39 -07002224static const struct xattr_handler *shmem_xattr_handlers[] = {
2225#ifdef CONFIG_TMPFS_POSIX_ACL
2226 &generic_acl_access_handler,
2227 &generic_acl_default_handler,
2228#endif
2229 NULL
2230};
2231
2232static int shmem_xattr_validate(const char *name)
2233{
2234 struct { const char *prefix; size_t len; } arr[] = {
2235 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2236 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2237 };
2238 int i;
2239
2240 for (i = 0; i < ARRAY_SIZE(arr); i++) {
2241 size_t preflen = arr[i].len;
2242 if (strncmp(name, arr[i].prefix, preflen) == 0) {
2243 if (!name[preflen])
2244 return -EINVAL;
2245 return 0;
2246 }
2247 }
2248 return -EOPNOTSUPP;
2249}
2250
2251static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2252 void *buffer, size_t size)
2253{
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002254 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002255 int err;
2256
2257 /*
2258 * If this is a request for a synthetic attribute in the system.*
2259 * namespace use the generic infrastructure to resolve a handler
2260 * for it via sb->s_xattr.
2261 */
2262 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2263 return generic_getxattr(dentry, name, buffer, size);
2264
2265 err = shmem_xattr_validate(name);
2266 if (err)
2267 return err;
2268
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002269 return simple_xattr_get(&info->xattrs, name, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002270}
2271
2272static int shmem_setxattr(struct dentry *dentry, const char *name,
2273 const void *value, size_t size, int flags)
2274{
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002275 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002276 int err;
2277
2278 /*
2279 * If this is a request for a synthetic attribute in the system.*
2280 * namespace use the generic infrastructure to resolve a handler
2281 * for it via sb->s_xattr.
2282 */
2283 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2284 return generic_setxattr(dentry, name, value, size, flags);
2285
2286 err = shmem_xattr_validate(name);
2287 if (err)
2288 return err;
2289
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002290 return simple_xattr_set(&info->xattrs, name, value, size, flags);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002291}
2292
2293static int shmem_removexattr(struct dentry *dentry, const char *name)
2294{
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002295 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002296 int err;
2297
2298 /*
2299 * If this is a request for a synthetic attribute in the system.*
2300 * namespace use the generic infrastructure to resolve a handler
2301 * for it via sb->s_xattr.
2302 */
2303 if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2304 return generic_removexattr(dentry, name);
2305
2306 err = shmem_xattr_validate(name);
2307 if (err)
2308 return err;
2309
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002310 return simple_xattr_remove(&info->xattrs, name);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002311}
2312
2313static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2314{
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002315 struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2316 return simple_xattr_list(&info->xattrs, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07002317}
2318#endif /* CONFIG_TMPFS_XATTR */
2319
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002320static const struct inode_operations shmem_short_symlink_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 .readlink = generic_readlink,
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002322 .follow_link = shmem_follow_short_symlink,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002323#ifdef CONFIG_TMPFS_XATTR
2324 .setxattr = shmem_setxattr,
2325 .getxattr = shmem_getxattr,
2326 .listxattr = shmem_listxattr,
2327 .removexattr = shmem_removexattr,
2328#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329};
2330
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002331static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332 .readlink = generic_readlink,
2333 .follow_link = shmem_follow_link,
2334 .put_link = shmem_put_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002335#ifdef CONFIG_TMPFS_XATTR
2336 .setxattr = shmem_setxattr,
2337 .getxattr = shmem_getxattr,
2338 .listxattr = shmem_listxattr,
2339 .removexattr = shmem_removexattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002340#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07002341};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002342
David M. Grimes91828a42006-10-17 00:09:45 -07002343static struct dentry *shmem_get_parent(struct dentry *child)
2344{
2345 return ERR_PTR(-ESTALE);
2346}
2347
2348static int shmem_match(struct inode *ino, void *vfh)
2349{
2350 __u32 *fh = vfh;
2351 __u64 inum = fh[2];
2352 inum = (inum << 32) | fh[1];
2353 return ino->i_ino == inum && fh[0] == ino->i_generation;
2354}
2355
Christoph Hellwig480b1162007-10-21 16:42:13 -07002356static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2357 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07002358{
David M. Grimes91828a42006-10-17 00:09:45 -07002359 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07002360 struct dentry *dentry = NULL;
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07002361 u64 inum;
David M. Grimes91828a42006-10-17 00:09:45 -07002362
Christoph Hellwig480b1162007-10-21 16:42:13 -07002363 if (fh_len < 3)
2364 return NULL;
2365
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07002366 inum = fid->raw[2];
2367 inum = (inum << 32) | fid->raw[1];
2368
Christoph Hellwig480b1162007-10-21 16:42:13 -07002369 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2370 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07002371 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07002372 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002373 iput(inode);
2374 }
2375
Christoph Hellwig480b1162007-10-21 16:42:13 -07002376 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07002377}
2378
Al Virob0b03822012-04-02 14:34:06 -04002379static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2380 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07002381{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302382 if (*len < 3) {
2383 *len = 3;
Namjae Jeon94e07a752013-02-17 15:48:11 +09002384 return FILEID_INVALID;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05302385 }
David M. Grimes91828a42006-10-17 00:09:45 -07002386
Al Viro1d3382cb2010-10-23 15:19:20 -04002387 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07002388 /* Unfortunately insert_inode_hash is not idempotent,
2389 * so as we hash inodes here rather than at creation
2390 * time, we need a lock to ensure we only try
2391 * to do it once
2392 */
2393 static DEFINE_SPINLOCK(lock);
2394 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04002395 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07002396 __insert_inode_hash(inode,
2397 inode->i_ino + inode->i_generation);
2398 spin_unlock(&lock);
2399 }
2400
2401 fh[0] = inode->i_generation;
2402 fh[1] = inode->i_ino;
2403 fh[2] = ((__u64)inode->i_ino) >> 32;
2404
2405 *len = 3;
2406 return 1;
2407}
2408
Christoph Hellwig39655162007-10-21 16:42:17 -07002409static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07002410 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07002411 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07002412 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07002413};
2414
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002415static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2416 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002417{
2418 char *this_char, *value, *rest;
Greg Thelen49cd0a52013-02-22 16:36:02 -08002419 struct mempolicy *mpol = NULL;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002420 uid_t uid;
2421 gid_t gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002422
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002423 while (options != NULL) {
2424 this_char = options;
2425 for (;;) {
2426 /*
2427 * NUL-terminate this option: unfortunately,
2428 * mount options form a comma-separated list,
2429 * but mpol's nodelist may also contain commas.
2430 */
2431 options = strchr(options, ',');
2432 if (options == NULL)
2433 break;
2434 options++;
2435 if (!isdigit(*options)) {
2436 options[-1] = '\0';
2437 break;
2438 }
2439 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440 if (!*this_char)
2441 continue;
2442 if ((value = strchr(this_char,'=')) != NULL) {
2443 *value++ = 0;
2444 } else {
2445 printk(KERN_ERR
2446 "tmpfs: No value for mount option '%s'\n",
2447 this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08002448 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449 }
2450
2451 if (!strcmp(this_char,"size")) {
2452 unsigned long long size;
2453 size = memparse(value,&rest);
2454 if (*rest == '%') {
2455 size <<= PAGE_SHIFT;
2456 size *= totalram_pages;
2457 do_div(size, 100);
2458 rest++;
2459 }
2460 if (*rest)
2461 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002462 sbinfo->max_blocks =
2463 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002465 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466 if (*rest)
2467 goto bad_val;
2468 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002469 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470 if (*rest)
2471 goto bad_val;
2472 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002473 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002475 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476 if (*rest)
2477 goto bad_val;
2478 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002479 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002480 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002481 uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002482 if (*rest)
2483 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002484 sbinfo->uid = make_kuid(current_user_ns(), uid);
2485 if (!uid_valid(sbinfo->uid))
2486 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002487 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002488 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002489 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002490 gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491 if (*rest)
2492 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08002493 sbinfo->gid = make_kgid(current_user_ns(), gid);
2494 if (!gid_valid(sbinfo->gid))
2495 goto bad_val;
Robin Holt7339ff82006-01-14 13:20:48 -08002496 } else if (!strcmp(this_char,"mpol")) {
Greg Thelen49cd0a52013-02-22 16:36:02 -08002497 mpol_put(mpol);
2498 mpol = NULL;
2499 if (mpol_parse_str(value, &mpol))
Robin Holt7339ff82006-01-14 13:20:48 -08002500 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501 } else {
2502 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2503 this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08002504 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505 }
2506 }
Greg Thelen49cd0a52013-02-22 16:36:02 -08002507 sbinfo->mpol = mpol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 return 0;
2509
2510bad_val:
2511 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2512 value, this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08002513error:
2514 mpol_put(mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515 return 1;
2516
2517}
2518
2519static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2520{
2521 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002522 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002523 unsigned long inodes;
2524 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002525
Greg Thelen5f001102013-02-22 16:36:01 -08002526 config.mpol = NULL;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002527 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002528 return error;
2529
2530 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002531 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
Tim Chen7e496292010-08-09 17:19:05 -07002532 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002533 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002534 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002535 goto out;
2536 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07002537 * Those tests disallow limited->unlimited while any are in use;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002538 * but we must separately disallow unlimited->limited, because
2539 * in that case we have no record of how much is already in use.
2540 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002541 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002542 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002543 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002544 goto out;
2545
2546 error = 0;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002547 sbinfo->max_blocks = config.max_blocks;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002548 sbinfo->max_inodes = config.max_inodes;
2549 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002550
Greg Thelen5f001102013-02-22 16:36:01 -08002551 /*
2552 * Preserve previous mempolicy unless mpol remount option was specified.
2553 */
2554 if (config.mpol) {
2555 mpol_put(sbinfo->mpol);
2556 sbinfo->mpol = config.mpol; /* transfers initial ref */
2557 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002558out:
2559 spin_unlock(&sbinfo->stat_lock);
2560 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002562
Al Viro34c80b12011-12-08 21:32:45 -05002563static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002564{
Al Viro34c80b12011-12-08 21:32:45 -05002565 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002566
2567 if (sbinfo->max_blocks != shmem_default_max_blocks())
2568 seq_printf(seq, ",size=%luk",
2569 sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2570 if (sbinfo->max_inodes != shmem_default_max_inodes())
2571 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2572 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04002573 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08002574 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2575 seq_printf(seq, ",uid=%u",
2576 from_kuid_munged(&init_user_ns, sbinfo->uid));
2577 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2578 seq_printf(seq, ",gid=%u",
2579 from_kgid_munged(&init_user_ns, sbinfo->gid));
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002580 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002581 return 0;
2582}
2583#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584
2585static void shmem_put_super(struct super_block *sb)
2586{
Hugh Dickins602586a2010-08-17 15:23:56 -07002587 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2588
2589 percpu_counter_destroy(&sbinfo->used_blocks);
Greg Thelen49cd0a52013-02-22 16:36:02 -08002590 mpol_put(sbinfo->mpol);
Hugh Dickins602586a2010-08-17 15:23:56 -07002591 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592 sb->s_fs_info = NULL;
2593}
2594
Kay Sievers2b2af542009-04-30 15:23:42 +02002595int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596{
2597 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002598 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002599 int err = -ENOMEM;
2600
2601 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07002602 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002603 L1_CACHE_BYTES), GFP_KERNEL);
2604 if (!sbinfo)
2605 return -ENOMEM;
2606
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002607 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11002608 sbinfo->uid = current_fsuid();
2609 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002610 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002611
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002612#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07002613 /*
2614 * Per default we only allow half of the physical ram per
2615 * tmpfs instance, limiting inodes to one per page of lowmem;
2616 * but the internal instance is left unlimited.
2617 */
Al Viroca4e0512013-08-31 12:57:10 -04002618 if (!(sb->s_flags & MS_KERNMOUNT)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002619 sbinfo->max_blocks = shmem_default_max_blocks();
2620 sbinfo->max_inodes = shmem_default_max_inodes();
2621 if (shmem_parse_options(data, sbinfo, false)) {
2622 err = -EINVAL;
2623 goto failed;
2624 }
Al Viroca4e0512013-08-31 12:57:10 -04002625 } else {
2626 sb->s_flags |= MS_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 }
David M. Grimes91828a42006-10-17 00:09:45 -07002628 sb->s_export_op = &shmem_export_ops;
Hugh Dickins2f6e38f2012-05-29 15:06:38 -07002629 sb->s_flags |= MS_NOSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630#else
2631 sb->s_flags |= MS_NOUSER;
2632#endif
2633
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002634 spin_lock_init(&sbinfo->stat_lock);
Hugh Dickins602586a2010-08-17 15:23:56 -07002635 if (percpu_counter_init(&sbinfo->used_blocks, 0))
2636 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002637 sbinfo->free_inodes = sbinfo->max_inodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002638
Hugh Dickins285b2c42011-08-03 16:21:20 -07002639 sb->s_maxbytes = MAX_LFS_FILESIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640 sb->s_blocksize = PAGE_CACHE_SIZE;
2641 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2642 sb->s_magic = TMPFS_MAGIC;
2643 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01002644 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07002645#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002646 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07002647#endif
2648#ifdef CONFIG_TMPFS_POSIX_ACL
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002649 sb->s_flags |= MS_POSIXACL;
2650#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002651
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002652 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653 if (!inode)
2654 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002655 inode->i_uid = sbinfo->uid;
2656 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05002657 sb->s_root = d_make_root(inode);
2658 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05002659 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660 return 0;
2661
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662failed:
2663 shmem_put_super(sb);
2664 return err;
2665}
2666
Pekka Enbergfcc234f2006-03-22 00:08:13 -08002667static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668
2669static struct inode *shmem_alloc_inode(struct super_block *sb)
2670{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002671 struct shmem_inode_info *info;
2672 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2673 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002675 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676}
2677
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002678static void shmem_destroy_callback(struct rcu_head *head)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11002679{
2680 struct inode *inode = container_of(head, struct inode, i_rcu);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11002681 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2682}
2683
Linus Torvalds1da177e2005-04-16 15:20:36 -07002684static void shmem_destroy_inode(struct inode *inode)
2685{
Al Viro09208d12011-07-26 03:15:03 -04002686 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002688 call_rcu(&inode->i_rcu, shmem_destroy_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689}
2690
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002691static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002693 struct shmem_inode_info *info = foo;
2694 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695}
2696
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002697static int shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698{
2699 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2700 sizeof(struct shmem_inode_info),
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002701 0, SLAB_PANIC, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002702 return 0;
2703}
2704
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002705static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07002707 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708}
2709
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002710static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08002712 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07002714 .write_begin = shmem_write_begin,
2715 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716#endif
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07002717 .migratepage = migrate_page,
Andi Kleenaa261f52009-09-16 11:50:16 +02002718 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719};
2720
Helge Deller15ad7cd2006-12-06 20:40:36 -08002721static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722 .mmap = shmem_mmap,
2723#ifdef CONFIG_TMPFS
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002724 .llseek = shmem_file_llseek,
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07002725 .read = do_sync_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002726 .write = do_sync_write,
Hugh Dickinsbcd78e42008-07-23 21:27:35 -07002727 .aio_read = shmem_file_aio_read,
Hugh Dickins5402b972008-02-04 22:28:44 -08002728 .aio_write = generic_file_aio_write,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02002729 .fsync = noop_fsync,
Hugh Dickins708e3502011-07-25 17:12:32 -07002730 .splice_read = shmem_file_splice_read,
Hugh Dickinsae9764162007-06-04 10:00:39 +02002731 .splice_write = generic_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002732 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733#endif
2734};
2735
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002736static const struct inode_operations shmem_inode_operations = {
Hugh Dickins94c1e622011-06-27 16:18:03 -07002737 .setattr = shmem_setattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07002738#ifdef CONFIG_TMPFS_XATTR
2739 .setxattr = shmem_setxattr,
2740 .getxattr = shmem_getxattr,
2741 .listxattr = shmem_listxattr,
2742 .removexattr = shmem_removexattr,
2743#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744};
2745
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002746static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747#ifdef CONFIG_TMPFS
2748 .create = shmem_create,
2749 .lookup = simple_lookup,
2750 .link = shmem_link,
2751 .unlink = shmem_unlink,
2752 .symlink = shmem_symlink,
2753 .mkdir = shmem_mkdir,
2754 .rmdir = shmem_rmdir,
2755 .mknod = shmem_mknod,
2756 .rename = shmem_rename,
Al Viro60545d02013-06-07 01:20:27 -04002757 .tmpfile = shmem_tmpfile,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07002759#ifdef CONFIG_TMPFS_XATTR
2760 .setxattr = shmem_setxattr,
2761 .getxattr = shmem_getxattr,
2762 .listxattr = shmem_listxattr,
2763 .removexattr = shmem_removexattr,
2764#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002765#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07002766 .setattr = shmem_setattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002767#endif
2768};
2769
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002770static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07002771#ifdef CONFIG_TMPFS_XATTR
2772 .setxattr = shmem_setxattr,
2773 .getxattr = shmem_getxattr,
2774 .listxattr = shmem_listxattr,
2775 .removexattr = shmem_removexattr,
2776#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002777#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07002778 .setattr = shmem_setattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002779#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780};
2781
Hugh Dickins759b9772007-03-05 00:30:28 -08002782static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783 .alloc_inode = shmem_alloc_inode,
2784 .destroy_inode = shmem_destroy_inode,
2785#ifdef CONFIG_TMPFS
2786 .statfs = shmem_statfs,
2787 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08002788 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789#endif
Al Viro1f895f72010-06-05 19:10:41 -04002790 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791 .drop_inode = generic_delete_inode,
2792 .put_super = shmem_put_super,
2793};
2794
Alexey Dobriyanf0f37e22009-09-27 22:29:37 +04002795static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002796 .fault = shmem_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797#ifdef CONFIG_NUMA
2798 .set_policy = shmem_set_policy,
2799 .get_policy = shmem_get_policy,
2800#endif
Konstantin Khlebnikov0b173bc2012-10-08 16:28:46 -07002801 .remap_pages = generic_file_remap_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802};
2803
Al Viro3c26ff62010-07-25 11:46:36 +04002804static struct dentry *shmem_mount(struct file_system_type *fs_type,
2805 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002806{
Al Viro3c26ff62010-07-25 11:46:36 +04002807 return mount_nodev(fs_type, flags, data, shmem_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002808}
2809
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002810static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811 .owner = THIS_MODULE,
2812 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04002813 .mount = shmem_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08002815 .fs_flags = FS_USERNS_MOUNT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002816};
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002818int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819{
2820 int error;
2821
Rob Landley16203a72013-09-11 14:26:12 -07002822 /* If rootfs called this, don't re-init */
2823 if (shmem_inode_cachep)
2824 return 0;
2825
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002826 error = bdi_init(&shmem_backing_dev_info);
2827 if (error)
2828 goto out4;
2829
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002830 error = shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002831 if (error)
2832 goto out3;
2833
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002834 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835 if (error) {
2836 printk(KERN_ERR "Could not register tmpfs\n");
2837 goto out2;
2838 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07002839
Al Viroca4e0512013-08-31 12:57:10 -04002840 shm_mnt = kern_mount(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841 if (IS_ERR(shm_mnt)) {
2842 error = PTR_ERR(shm_mnt);
2843 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2844 goto out1;
2845 }
2846 return 0;
2847
2848out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002849 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002850out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002851 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852out3:
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002853 bdi_destroy(&shmem_backing_dev_info);
2854out4:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002855 shm_mnt = ERR_PTR(error);
2856 return error;
2857}
Matt Mackall853ac432009-01-06 14:40:20 -08002858
2859#else /* !CONFIG_SHMEM */
2860
2861/*
2862 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2863 *
2864 * This is intended for small system where the benefits of the full
2865 * shmem code (swap-backed and resource-limited) are outweighed by
2866 * their complexity. On systems without swap this code should be
2867 * effectively equivalent, but much lighter weight.
2868 */
2869
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002870static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08002871 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04002872 .mount = ramfs_mount,
Matt Mackall853ac432009-01-06 14:40:20 -08002873 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08002874 .fs_flags = FS_USERNS_MOUNT,
Matt Mackall853ac432009-01-06 14:40:20 -08002875};
2876
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002877int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08002878{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002879 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08002880
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002881 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08002882 BUG_ON(IS_ERR(shm_mnt));
2883
2884 return 0;
2885}
2886
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002887int shmem_unuse(swp_entry_t swap, struct page *page)
Matt Mackall853ac432009-01-06 14:40:20 -08002888{
2889 return 0;
2890}
2891
Hugh Dickins3f96b792009-09-21 17:03:37 -07002892int shmem_lock(struct file *file, int lock, struct user_struct *user)
2893{
2894 return 0;
2895}
2896
Hugh Dickins24513262012-01-20 14:34:21 -08002897void shmem_unlock_mapping(struct address_space *mapping)
2898{
2899}
2900
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002901void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07002902{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002903 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07002904}
2905EXPORT_SYMBOL_GPL(shmem_truncate_range);
2906
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002907#define shmem_vm_ops generic_file_vm_ops
2908#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002909#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002910#define shmem_acct_size(flags, size) 0
2911#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08002912
2913#endif /* CONFIG_SHMEM */
2914
2915/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916
Al Viro34515382013-02-14 22:38:02 -05002917static struct dentry_operations anon_ops = {
Al Viro118b2302013-08-24 12:08:17 -04002918 .d_dname = simple_dname
Al Viro34515382013-02-14 22:38:02 -05002919};
2920
Eric Parisc7277092013-12-02 11:24:19 +00002921static struct file *__shmem_file_setup(const char *name, loff_t size,
2922 unsigned long flags, unsigned int i_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923{
Al Viro6b4d0b22013-02-14 21:37:26 -05002924 struct file *res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002925 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04002926 struct path path;
Al Viro34515382013-02-14 22:38:02 -05002927 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928 struct qstr this;
2929
2930 if (IS_ERR(shm_mnt))
Al Viro6b4d0b22013-02-14 21:37:26 -05002931 return ERR_CAST(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932
Hugh Dickins285b2c42011-08-03 16:21:20 -07002933 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934 return ERR_PTR(-EINVAL);
2935
2936 if (shmem_acct_size(flags, size))
2937 return ERR_PTR(-ENOMEM);
2938
Al Viro6b4d0b22013-02-14 21:37:26 -05002939 res = ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002940 this.name = name;
2941 this.len = strlen(name);
2942 this.hash = 0; /* will go */
Al Viro34515382013-02-14 22:38:02 -05002943 sb = shm_mnt->mnt_sb;
2944 path.dentry = d_alloc_pseudo(sb, &this);
Al Viro2c48b9c2009-08-09 00:52:35 +04002945 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946 goto put_memory;
Al Viro34515382013-02-14 22:38:02 -05002947 d_set_d_op(path.dentry, &anon_ops);
Al Viro2c48b9c2009-08-09 00:52:35 +04002948 path.mnt = mntget(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949
Al Viro6b4d0b22013-02-14 21:37:26 -05002950 res = ERR_PTR(-ENOSPC);
Al Viro34515382013-02-14 22:38:02 -05002951 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002952 if (!inode)
Al Viro4b42af82009-08-05 18:25:56 +04002953 goto put_dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002954
Eric Parisc7277092013-12-02 11:24:19 +00002955 inode->i_flags |= i_flags;
Al Viro2c48b9c2009-08-09 00:52:35 +04002956 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002957 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02002958 clear_nlink(inode); /* It is unlinked */
Al Viro26567cd2013-03-01 20:22:53 -05002959 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
2960 if (IS_ERR(res))
Al Viro4b42af82009-08-05 18:25:56 +04002961 goto put_dentry;
Al Viro4b42af82009-08-05 18:25:56 +04002962
Al Viro6b4d0b22013-02-14 21:37:26 -05002963 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Al Viro4b42af82009-08-05 18:25:56 +04002964 &shmem_file_operations);
Al Viro6b4d0b22013-02-14 21:37:26 -05002965 if (IS_ERR(res))
Al Viro4b42af82009-08-05 18:25:56 +04002966 goto put_dentry;
2967
Al Viro6b4d0b22013-02-14 21:37:26 -05002968 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969
Linus Torvalds1da177e2005-04-16 15:20:36 -07002970put_dentry:
Al Viro2c48b9c2009-08-09 00:52:35 +04002971 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972put_memory:
2973 shmem_unacct_size(flags, size);
Al Viro6b4d0b22013-02-14 21:37:26 -05002974 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975}
Eric Parisc7277092013-12-02 11:24:19 +00002976
2977/**
2978 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
2979 * kernel internal. There will be NO LSM permission checks against the
2980 * underlying inode. So users of this interface must do LSM checks at a
2981 * higher layer. The one user is the big_key implementation. LSM checks
2982 * are provided at the key level rather than the inode level.
2983 * @name: name for dentry (to be seen in /proc/<pid>/maps
2984 * @size: size to be set for the file
2985 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2986 */
2987struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
2988{
2989 return __shmem_file_setup(name, size, flags, S_PRIVATE);
2990}
2991
2992/**
2993 * shmem_file_setup - get an unlinked file living in tmpfs
2994 * @name: name for dentry (to be seen in /proc/<pid>/maps
2995 * @size: size to be set for the file
2996 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2997 */
2998struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2999{
3000 return __shmem_file_setup(name, size, flags, 0);
3001}
Keith Packard395e0dd2008-06-20 00:08:06 -07003002EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003003
Randy Dunlap46711812008-03-19 17:00:41 -07003004/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3007 */
3008int shmem_zero_setup(struct vm_area_struct *vma)
3009{
3010 struct file *file;
3011 loff_t size = vma->vm_end - vma->vm_start;
3012
3013 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
3014 if (IS_ERR(file))
3015 return PTR_ERR(file);
3016
3017 if (vma->vm_file)
3018 fput(vma->vm_file);
3019 vma->vm_file = file;
3020 vma->vm_ops = &shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021 return 0;
3022}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003023
3024/**
3025 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3026 * @mapping: the page's address_space
3027 * @index: the page index
3028 * @gfp: the page allocator flags to use if allocating
3029 *
3030 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3031 * with any new page allocations done using the specified allocation flags.
3032 * But read_cache_page_gfp() uses the ->readpage() method: which does not
3033 * suit tmpfs, since it may have pages in swapcache, and needs to find those
3034 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3035 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07003036 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3037 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003038 */
3039struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
3040 pgoff_t index, gfp_t gfp)
3041{
Hugh Dickins68da9f02011-07-25 17:12:34 -07003042#ifdef CONFIG_SHMEM
3043 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07003044 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07003045 int error;
3046
3047 BUG_ON(mapping->a_ops != &shmem_aops);
3048 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
3049 if (error)
3050 page = ERR_PTR(error);
3051 else
3052 unlock_page(page);
3053 return page;
3054#else
3055 /*
3056 * The tiny !SHMEM case uses ramfs without swap
3057 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003058 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07003059#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07003060}
3061EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);