blob: a06f23731d3f96db3a2b134d0d62b3c78dcef00c [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>
Ingo Molnar174cd4b2017-02-02 19:15:33 +010032#include <linux/sched/signal.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040033#include <linux/export.h>
Matt Mackall853ac432009-01-06 14:40:20 -080034#include <linux/swap.h>
Christoph Hellwige2e40f22015-02-22 08:58:50 -080035#include <linux/uio.h>
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -070036#include <linux/khugepaged.h>
Matt Mackall853ac432009-01-06 14:40:20 -080037
Andrea Arcangeli95cc09d2017-02-22 15:43:31 -080038#include <asm/tlbflush.h> /* for arch/microblaze update_mmu_cache() */
39
Matt Mackall853ac432009-01-06 14:40:20 -080040static struct vfsmount *shm_mnt;
41
42#ifdef CONFIG_SHMEM
Linus Torvalds1da177e2005-04-16 15:20:36 -070043/*
44 * This virtual memory filesystem is heavily based on the ramfs. It
45 * extends ramfs by the ability to use swap and honor resource limits
46 * which makes it a completely usable filesystem.
47 */
48
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070049#include <linux/xattr.h>
Christoph Hellwiga5694252007-07-17 04:04:28 -070050#include <linux/exportfs.h>
Christoph Hellwig1c7c4742009-11-03 16:44:44 +010051#include <linux/posix_acl.h>
Christoph Hellwigfeda8212013-12-20 05:16:54 -080052#include <linux/posix_acl_xattr.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070053#include <linux/mman.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070054#include <linux/string.h>
55#include <linux/slab.h>
56#include <linux/backing-dev.h>
57#include <linux/shmem_fs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070058#include <linux/writeback.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070059#include <linux/blkdev.h>
Hugh Dickinsbda97ea2011-08-03 16:21:21 -070060#include <linux/pagevec.h>
Hugh Dickins41ffe5d2011-08-03 16:21:21 -070061#include <linux/percpu_counter.h>
Hugh Dickins83e4fa92012-05-29 15:06:40 -070062#include <linux/falloc.h>
Hugh Dickins708e3502011-07-25 17:12:32 -070063#include <linux/splice.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070064#include <linux/security.h>
65#include <linux/swapops.h>
66#include <linux/mempolicy.h>
67#include <linux/namei.h>
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +000068#include <linux/ctype.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070069#include <linux/migrate.h>
Christoph Lameterc1f60a52006-09-25 23:31:11 -070070#include <linux/highmem.h>
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -080071#include <linux/seq_file.h>
Mimi Zohar92562922008-10-07 14:00:12 -040072#include <linux/magic.h>
David Herrmann9183df22014-08-08 14:25:29 -070073#include <linux/syscalls.h>
David Herrmann40e041a2014-08-08 14:25:27 -070074#include <linux/fcntl.h>
David Herrmann9183df22014-08-08 14:25:29 -070075#include <uapi/linux/memfd.h>
Mike Rapoportcfda0522017-02-22 15:43:37 -080076#include <linux/userfaultfd_k.h>
Mike Rapoport4c27fe42017-02-22 15:43:25 -080077#include <linux/rmap.h>
Amir Goldstein2b4db792017-05-18 15:29:33 +030078#include <linux/uuid.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070079
Linus Torvalds7c0f6ba2016-12-24 11:46:01 -080080#include <linux/uaccess.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070081#include <asm/pgtable.h>
82
Mel Gormandd56b042015-11-06 16:28:43 -080083#include "internal.h"
84
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +030085#define BLOCKS_PER_PAGE (PAGE_SIZE/512)
86#define VM_ACCT(size) (PAGE_ALIGN(size) >> PAGE_SHIFT)
Linus Torvalds1da177e2005-04-16 15:20:36 -070087
Linus Torvalds1da177e2005-04-16 15:20:36 -070088/* Pretend that each entry is of this size in directory's i_size */
89#define BOGO_DIRENT_SIZE 20
90
Hugh Dickins69f07ec2011-08-03 16:21:26 -070091/* Symlink up to this size is kmalloc'ed instead of using a swappable page */
92#define SHORT_SYMLINK_LEN 128
93
Hugh Dickins1aac1402012-05-29 15:06:42 -070094/*
Hugh Dickinsf00cdc62014-06-23 13:22:06 -070095 * shmem_fallocate communicates with shmem_fault or shmem_writepage via
96 * inode->i_private (with i_mutex making sure that it has only one user at
97 * a time): we would prefer not to enlarge the shmem inode just for that.
Hugh Dickins1aac1402012-05-29 15:06:42 -070098 */
99struct shmem_falloc {
Hugh Dickins8e205f72014-07-23 14:00:10 -0700100 wait_queue_head_t *waitq; /* faults into hole wait for punch to end */
Hugh Dickins1aac1402012-05-29 15:06:42 -0700101 pgoff_t start; /* start of range currently being fallocated */
102 pgoff_t next; /* the next page offset to be fallocated */
103 pgoff_t nr_falloced; /* how many new pages have been fallocated */
104 pgoff_t nr_unswapped; /* how often writepage refused to swap out */
105};
106
Andrew Mortonb76db732008-02-08 04:21:49 -0800107#ifdef CONFIG_TMPFS
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800108static unsigned long shmem_default_max_blocks(void)
109{
110 return totalram_pages / 2;
111}
112
113static unsigned long shmem_default_max_inodes(void)
114{
115 return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
116}
Andrew Mortonb76db732008-02-08 04:21:49 -0800117#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -0800118
Hugh Dickinsbde05d12012-05-29 15:06:38 -0700119static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
120static int shmem_replace_page(struct page **pagep, gfp_t gfp,
121 struct shmem_inode_info *info, pgoff_t index);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700122static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700123 struct page **pagep, enum sgp_type sgp,
Mike Rapoportcfda0522017-02-22 15:43:37 -0800124 gfp_t gfp, struct vm_area_struct *vma,
125 struct vm_fault *vmf, int *fault_type);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700126
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -0700127int shmem_getpage(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700128 struct page **pagep, enum sgp_type sgp)
Hugh Dickins68da9f02011-07-25 17:12:34 -0700129{
130 return shmem_getpage_gfp(inode, index, pagep, sgp,
Mike Rapoportcfda0522017-02-22 15:43:37 -0800131 mapping_gfp_mask(inode->i_mapping), NULL, NULL, NULL);
Hugh Dickins68da9f02011-07-25 17:12:34 -0700132}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
135{
136 return sb->s_fs_info;
137}
138
139/*
140 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
141 * for shared memory and for shared anonymous (/dev/zero) mappings
142 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
143 * consistent with the pre-accounting of private mappings ...
144 */
145static inline int shmem_acct_size(unsigned long flags, loff_t size)
146{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000147 return (flags & VM_NORESERVE) ?
Al Viro191c5422012-02-13 03:58:52 +0000148 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149}
150
151static inline void shmem_unacct_size(unsigned long flags, loff_t size)
152{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000153 if (!(flags & VM_NORESERVE))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154 vm_unacct_memory(VM_ACCT(size));
155}
156
Konstantin Khlebnikov77142512014-08-06 16:06:34 -0700157static inline int shmem_reacct_size(unsigned long flags,
158 loff_t oldsize, loff_t newsize)
159{
160 if (!(flags & VM_NORESERVE)) {
161 if (VM_ACCT(newsize) > VM_ACCT(oldsize))
162 return security_vm_enough_memory_mm(current->mm,
163 VM_ACCT(newsize) - VM_ACCT(oldsize));
164 else if (VM_ACCT(newsize) < VM_ACCT(oldsize))
165 vm_unacct_memory(VM_ACCT(oldsize) - VM_ACCT(newsize));
166 }
167 return 0;
168}
169
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170/*
171 * ... whereas tmpfs objects are accounted incrementally as
Hugh Dickins75edd342016-05-19 17:12:44 -0700172 * pages are allocated, in order to allow large sparse files.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
174 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
175 */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700176static inline int shmem_acct_block(unsigned long flags, long pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177{
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700178 if (!(flags & VM_NORESERVE))
179 return 0;
180
181 return security_vm_enough_memory_mm(current->mm,
182 pages * VM_ACCT(PAGE_SIZE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700183}
184
185static inline void shmem_unacct_blocks(unsigned long flags, long pages)
186{
Hugh Dickins0b0a0802009-02-24 20:51:52 +0000187 if (flags & VM_NORESERVE)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300188 vm_unacct_memory(pages * VM_ACCT(PAGE_SIZE));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189}
190
Hugh Dickins759b9772007-03-05 00:30:28 -0800191static const struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700192static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800193static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800194static const struct inode_operations shmem_inode_operations;
195static const struct inode_operations shmem_dir_inode_operations;
196static const struct inode_operations shmem_special_inode_operations;
Alexey Dobriyanf0f37e2f2009-09-27 22:29:37 +0400197static const struct vm_operations_struct shmem_vm_ops;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700198static struct file_system_type shmem_fs_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199
Mike Rapoportb0506e42017-02-22 15:43:28 -0800200bool vma_is_shmem(struct vm_area_struct *vma)
201{
202 return vma->vm_ops == &shmem_vm_ops;
203}
204
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205static LIST_HEAD(shmem_swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -0800206static DEFINE_MUTEX(shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700207
Pavel Emelyanov5b04c682008-02-04 22:28:47 -0800208static int shmem_reserve_inode(struct super_block *sb)
209{
210 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
211 if (sbinfo->max_inodes) {
212 spin_lock(&sbinfo->stat_lock);
213 if (!sbinfo->free_inodes) {
214 spin_unlock(&sbinfo->stat_lock);
215 return -ENOSPC;
216 }
217 sbinfo->free_inodes--;
218 spin_unlock(&sbinfo->stat_lock);
219 }
220 return 0;
221}
222
223static void shmem_free_inode(struct super_block *sb)
224{
225 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
226 if (sbinfo->max_inodes) {
227 spin_lock(&sbinfo->stat_lock);
228 sbinfo->free_inodes++;
229 spin_unlock(&sbinfo->stat_lock);
230 }
231}
232
Randy Dunlap46711812008-03-19 17:00:41 -0700233/**
Hugh Dickins41ffe5d2011-08-03 16:21:21 -0700234 * shmem_recalc_inode - recalculate the block usage of an inode
Linus Torvalds1da177e2005-04-16 15:20:36 -0700235 * @inode: inode to recalc
236 *
237 * We have to calculate the free blocks since the mm can drop
238 * undirtied hole pages behind our back.
239 *
240 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
241 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
242 *
243 * It has to be called with the spinlock held.
244 */
245static void shmem_recalc_inode(struct inode *inode)
246{
247 struct shmem_inode_info *info = SHMEM_I(inode);
248 long freed;
249
250 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
251 if (freed > 0) {
Hugh Dickins54af60422011-08-03 16:21:24 -0700252 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
253 if (sbinfo->max_blocks)
254 percpu_counter_add(&sbinfo->used_blocks, -freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 info->alloced -= freed;
Hugh Dickins54af60422011-08-03 16:21:24 -0700256 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 shmem_unacct_blocks(info->flags, freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258 }
259}
260
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700261bool shmem_charge(struct inode *inode, long pages)
262{
263 struct shmem_inode_info *info = SHMEM_I(inode);
264 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700265 unsigned long flags;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700266
267 if (shmem_acct_block(info->flags, pages))
268 return false;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700269 spin_lock_irqsave(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700270 info->alloced += pages;
271 inode->i_blocks += pages * BLOCKS_PER_PAGE;
272 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700273 spin_unlock_irqrestore(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700274 inode->i_mapping->nrpages += pages;
275
276 if (!sbinfo->max_blocks)
277 return true;
278 if (percpu_counter_compare(&sbinfo->used_blocks,
279 sbinfo->max_blocks - pages) > 0) {
280 inode->i_mapping->nrpages -= pages;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700281 spin_lock_irqsave(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700282 info->alloced -= pages;
283 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700284 spin_unlock_irqrestore(&info->lock, flags);
Hugh Dickins71664662016-09-23 20:24:23 -0700285 shmem_unacct_blocks(info->flags, pages);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700286 return false;
287 }
288 percpu_counter_add(&sbinfo->used_blocks, pages);
289 return true;
290}
291
292void shmem_uncharge(struct inode *inode, long pages)
293{
294 struct shmem_inode_info *info = SHMEM_I(inode);
295 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700296 unsigned long flags;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700297
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700298 spin_lock_irqsave(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700299 info->alloced -= pages;
300 inode->i_blocks -= pages * BLOCKS_PER_PAGE;
301 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700302 spin_unlock_irqrestore(&info->lock, flags);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700303
304 if (sbinfo->max_blocks)
305 percpu_counter_sub(&sbinfo->used_blocks, pages);
Hugh Dickins71664662016-09-23 20:24:23 -0700306 shmem_unacct_blocks(info->flags, pages);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700307}
308
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700309/*
310 * Replace item expected in radix tree by a new item, while holding tree lock.
311 */
312static int shmem_radix_tree_replace(struct address_space *mapping,
313 pgoff_t index, void *expected, void *replacement)
314{
Johannes Weinerf7942432016-12-12 16:43:41 -0800315 struct radix_tree_node *node;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700316 void **pslot;
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700317 void *item;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700318
319 VM_BUG_ON(!expected);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700320 VM_BUG_ON(!replacement);
Johannes Weinerf7942432016-12-12 16:43:41 -0800321 item = __radix_tree_lookup(&mapping->page_tree, index, &node, &pslot);
322 if (!item)
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700323 return -ENOENT;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700324 if (item != expected)
325 return -ENOENT;
Johannes Weiner4d693d02016-12-12 16:43:49 -0800326 __radix_tree_replace(&mapping->page_tree, node, pslot,
327 replacement, NULL, NULL);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700328 return 0;
329}
330
331/*
Hugh Dickinsd1899222012-07-11 14:02:47 -0700332 * Sometimes, before we decide whether to proceed or to fail, we must check
333 * that an entry was not already brought back from swap by a racing thread.
334 *
335 * Checking page is not enough: by the time a SwapCache page is locked, it
336 * might be reused, and again be SwapCache, using the same swap as before.
337 */
338static bool shmem_confirm_swap(struct address_space *mapping,
339 pgoff_t index, swp_entry_t swap)
340{
341 void *item;
342
343 rcu_read_lock();
344 item = radix_tree_lookup(&mapping->page_tree, index);
345 rcu_read_unlock();
346 return item == swp_to_radix_entry(swap);
347}
348
349/*
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700350 * Definitions for "huge tmpfs": tmpfs mounted with the huge= option
351 *
352 * SHMEM_HUGE_NEVER:
353 * disables huge pages for the mount;
354 * SHMEM_HUGE_ALWAYS:
355 * enables huge pages for the mount;
356 * SHMEM_HUGE_WITHIN_SIZE:
357 * only allocate huge pages if the page will be fully within i_size,
358 * also respect fadvise()/madvise() hints;
359 * SHMEM_HUGE_ADVISE:
360 * only allocate huge pages if requested with fadvise()/madvise();
361 */
362
363#define SHMEM_HUGE_NEVER 0
364#define SHMEM_HUGE_ALWAYS 1
365#define SHMEM_HUGE_WITHIN_SIZE 2
366#define SHMEM_HUGE_ADVISE 3
367
368/*
369 * Special values.
370 * Only can be set via /sys/kernel/mm/transparent_hugepage/shmem_enabled:
371 *
372 * SHMEM_HUGE_DENY:
373 * disables huge on shm_mnt and all mounts, for emergency use;
374 * SHMEM_HUGE_FORCE:
375 * enables huge on shm_mnt and all mounts, w/o needing option, for testing;
376 *
377 */
378#define SHMEM_HUGE_DENY (-1)
379#define SHMEM_HUGE_FORCE (-2)
380
Kirill A. Shutemove496cf32016-07-26 15:26:35 -0700381#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700382/* ifdef here to avoid bloating shmem.o when not necessary */
383
384int shmem_huge __read_mostly;
385
Jérémy Lefauref1f59292016-12-12 16:43:23 -0800386#if defined(CONFIG_SYSFS) || defined(CONFIG_TMPFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700387static int shmem_parse_huge(const char *str)
388{
389 if (!strcmp(str, "never"))
390 return SHMEM_HUGE_NEVER;
391 if (!strcmp(str, "always"))
392 return SHMEM_HUGE_ALWAYS;
393 if (!strcmp(str, "within_size"))
394 return SHMEM_HUGE_WITHIN_SIZE;
395 if (!strcmp(str, "advise"))
396 return SHMEM_HUGE_ADVISE;
397 if (!strcmp(str, "deny"))
398 return SHMEM_HUGE_DENY;
399 if (!strcmp(str, "force"))
400 return SHMEM_HUGE_FORCE;
401 return -EINVAL;
402}
403
404static const char *shmem_format_huge(int huge)
405{
406 switch (huge) {
407 case SHMEM_HUGE_NEVER:
408 return "never";
409 case SHMEM_HUGE_ALWAYS:
410 return "always";
411 case SHMEM_HUGE_WITHIN_SIZE:
412 return "within_size";
413 case SHMEM_HUGE_ADVISE:
414 return "advise";
415 case SHMEM_HUGE_DENY:
416 return "deny";
417 case SHMEM_HUGE_FORCE:
418 return "force";
419 default:
420 VM_BUG_ON(1);
421 return "bad_val";
422 }
423}
Jérémy Lefauref1f59292016-12-12 16:43:23 -0800424#endif
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700425
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700426static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
427 struct shrink_control *sc, unsigned long nr_to_split)
428{
429 LIST_HEAD(list), *pos, *next;
Kirill A. Shutemov253fd0f2017-02-03 13:13:15 -0800430 LIST_HEAD(to_remove);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700431 struct inode *inode;
432 struct shmem_inode_info *info;
433 struct page *page;
434 unsigned long batch = sc ? sc->nr_to_scan : 128;
435 int removed = 0, split = 0;
436
437 if (list_empty(&sbinfo->shrinklist))
438 return SHRINK_STOP;
439
440 spin_lock(&sbinfo->shrinklist_lock);
441 list_for_each_safe(pos, next, &sbinfo->shrinklist) {
442 info = list_entry(pos, struct shmem_inode_info, shrinklist);
443
444 /* pin the inode */
445 inode = igrab(&info->vfs_inode);
446
447 /* inode is about to be evicted */
448 if (!inode) {
449 list_del_init(&info->shrinklist);
450 removed++;
451 goto next;
452 }
453
454 /* Check if there's anything to gain */
455 if (round_up(inode->i_size, PAGE_SIZE) ==
456 round_up(inode->i_size, HPAGE_PMD_SIZE)) {
Kirill A. Shutemov253fd0f2017-02-03 13:13:15 -0800457 list_move(&info->shrinklist, &to_remove);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700458 removed++;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700459 goto next;
460 }
461
462 list_move(&info->shrinklist, &list);
463next:
464 if (!--batch)
465 break;
466 }
467 spin_unlock(&sbinfo->shrinklist_lock);
468
Kirill A. Shutemov253fd0f2017-02-03 13:13:15 -0800469 list_for_each_safe(pos, next, &to_remove) {
470 info = list_entry(pos, struct shmem_inode_info, shrinklist);
471 inode = &info->vfs_inode;
472 list_del_init(&info->shrinklist);
473 iput(inode);
474 }
475
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700476 list_for_each_safe(pos, next, &list) {
477 int ret;
478
479 info = list_entry(pos, struct shmem_inode_info, shrinklist);
480 inode = &info->vfs_inode;
481
482 if (nr_to_split && split >= nr_to_split) {
483 iput(inode);
484 continue;
485 }
486
487 page = find_lock_page(inode->i_mapping,
488 (inode->i_size & HPAGE_PMD_MASK) >> PAGE_SHIFT);
489 if (!page)
490 goto drop;
491
492 if (!PageTransHuge(page)) {
493 unlock_page(page);
494 put_page(page);
495 goto drop;
496 }
497
498 ret = split_huge_page(page);
499 unlock_page(page);
500 put_page(page);
501
502 if (ret) {
503 /* split failed: leave it on the list */
504 iput(inode);
505 continue;
506 }
507
508 split++;
509drop:
510 list_del_init(&info->shrinklist);
511 removed++;
512 iput(inode);
513 }
514
515 spin_lock(&sbinfo->shrinklist_lock);
516 list_splice_tail(&list, &sbinfo->shrinklist);
517 sbinfo->shrinklist_len -= removed;
518 spin_unlock(&sbinfo->shrinklist_lock);
519
520 return split;
521}
522
523static long shmem_unused_huge_scan(struct super_block *sb,
524 struct shrink_control *sc)
525{
526 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
527
528 if (!READ_ONCE(sbinfo->shrinklist_len))
529 return SHRINK_STOP;
530
531 return shmem_unused_huge_shrink(sbinfo, sc, 0);
532}
533
534static long shmem_unused_huge_count(struct super_block *sb,
535 struct shrink_control *sc)
536{
537 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
538 return READ_ONCE(sbinfo->shrinklist_len);
539}
Kirill A. Shutemove496cf32016-07-26 15:26:35 -0700540#else /* !CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700541
542#define shmem_huge SHMEM_HUGE_DENY
543
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700544static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
545 struct shrink_control *sc, unsigned long nr_to_split)
546{
547 return 0;
548}
Kirill A. Shutemove496cf32016-07-26 15:26:35 -0700549#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -0700550
551/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700552 * Like add_to_page_cache_locked, but error if expected item has gone.
553 */
554static int shmem_add_to_page_cache(struct page *page,
555 struct address_space *mapping,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -0700556 pgoff_t index, void *expected)
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700557{
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700558 int error, nr = hpage_nr_pages(page);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700559
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700560 VM_BUG_ON_PAGE(PageTail(page), page);
561 VM_BUG_ON_PAGE(index != round_down(index, nr), page);
Sasha Levin309381fea2014-01-23 15:52:54 -0800562 VM_BUG_ON_PAGE(!PageLocked(page), page);
563 VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700564 VM_BUG_ON(expected && PageTransHuge(page));
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700565
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700566 page_ref_add(page, nr);
Hugh Dickinsb065b432012-07-11 14:02:48 -0700567 page->mapping = mapping;
568 page->index = index;
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700569
Hugh Dickinsb065b432012-07-11 14:02:48 -0700570 spin_lock_irq(&mapping->tree_lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700571 if (PageTransHuge(page)) {
572 void __rcu **results;
573 pgoff_t idx;
574 int i;
575
576 error = 0;
577 if (radix_tree_gang_lookup_slot(&mapping->page_tree,
578 &results, &idx, index, 1) &&
579 idx < index + HPAGE_PMD_NR) {
580 error = -EEXIST;
581 }
582
583 if (!error) {
584 for (i = 0; i < HPAGE_PMD_NR; i++) {
585 error = radix_tree_insert(&mapping->page_tree,
586 index + i, page + i);
587 VM_BUG_ON(error);
588 }
589 count_vm_event(THP_FILE_ALLOC);
590 }
591 } else if (!expected) {
Hugh Dickinsb065b432012-07-11 14:02:48 -0700592 error = radix_tree_insert(&mapping->page_tree, index, page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700593 } else {
Hugh Dickinsb065b432012-07-11 14:02:48 -0700594 error = shmem_radix_tree_replace(mapping, index, expected,
595 page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700596 }
597
Hugh Dickinsb065b432012-07-11 14:02:48 -0700598 if (!error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700599 mapping->nrpages += nr;
600 if (PageTransHuge(page))
Mel Gorman11fb9982016-07-28 15:46:20 -0700601 __inc_node_page_state(page, NR_SHMEM_THPS);
602 __mod_node_page_state(page_pgdat(page), NR_FILE_PAGES, nr);
603 __mod_node_page_state(page_pgdat(page), NR_SHMEM, nr);
Hugh Dickinsb065b432012-07-11 14:02:48 -0700604 spin_unlock_irq(&mapping->tree_lock);
605 } else {
606 page->mapping = NULL;
607 spin_unlock_irq(&mapping->tree_lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700608 page_ref_sub(page, nr);
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700609 }
Hugh Dickins46f65ec2011-08-03 16:21:23 -0700610 return error;
611}
612
613/*
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700614 * Like delete_from_page_cache, but substitutes swap for page.
615 */
616static void shmem_delete_from_page_cache(struct page *page, void *radswap)
617{
618 struct address_space *mapping = page->mapping;
619 int error;
620
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700621 VM_BUG_ON_PAGE(PageCompound(page), page);
622
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700623 spin_lock_irq(&mapping->tree_lock);
624 error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
625 page->mapping = NULL;
626 mapping->nrpages--;
Mel Gorman11fb9982016-07-28 15:46:20 -0700627 __dec_node_page_state(page, NR_FILE_PAGES);
628 __dec_node_page_state(page, NR_SHMEM);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700629 spin_unlock_irq(&mapping->tree_lock);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300630 put_page(page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -0700631 BUG_ON(error);
632}
633
634/*
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700635 * Remove swap entry from radix tree, free the swap and its page cache.
636 */
637static int shmem_free_swap(struct address_space *mapping,
638 pgoff_t index, void *radswap)
639{
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700640 void *old;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700641
642 spin_lock_irq(&mapping->tree_lock);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700643 old = radix_tree_delete_item(&mapping->page_tree, index, radswap);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700644 spin_unlock_irq(&mapping->tree_lock);
Johannes Weiner6dbaf222014-04-03 14:47:41 -0700645 if (old != radswap)
646 return -ENOENT;
647 free_swap_and_cache(radix_to_swp_entry(radswap));
648 return 0;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700649}
650
651/*
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800652 * Determine (in bytes) how many of the shmem object's pages mapped by the
Vlastimil Babka48131e02016-01-14 15:19:23 -0800653 * given offsets are swapped out.
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800654 *
655 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
656 * as long as the inode doesn't go away and racy results are not a problem.
657 */
Vlastimil Babka48131e02016-01-14 15:19:23 -0800658unsigned long shmem_partial_swap_usage(struct address_space *mapping,
659 pgoff_t start, pgoff_t end)
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800660{
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800661 struct radix_tree_iter iter;
662 void **slot;
663 struct page *page;
Vlastimil Babka48131e02016-01-14 15:19:23 -0800664 unsigned long swapped = 0;
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800665
666 rcu_read_lock();
667
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800668 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
669 if (iter.index >= end)
670 break;
671
672 page = radix_tree_deref_slot(slot);
673
Matthew Wilcox2cf938a2016-03-17 14:22:03 -0700674 if (radix_tree_deref_retry(page)) {
675 slot = radix_tree_iter_retry(&iter);
676 continue;
677 }
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800678
679 if (radix_tree_exceptional_entry(page))
680 swapped++;
681
682 if (need_resched()) {
Matthew Wilcox148deab2016-12-14 15:08:49 -0800683 slot = radix_tree_iter_resume(slot, &iter);
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800684 cond_resched_rcu();
Vlastimil Babka6a15a372016-01-14 15:19:20 -0800685 }
686 }
687
688 rcu_read_unlock();
689
690 return swapped << PAGE_SHIFT;
691}
692
693/*
Vlastimil Babka48131e02016-01-14 15:19:23 -0800694 * Determine (in bytes) how many of the shmem object's pages mapped by the
695 * given vma is swapped out.
696 *
697 * This is safe to call without i_mutex or mapping->tree_lock thanks to RCU,
698 * as long as the inode doesn't go away and racy results are not a problem.
699 */
700unsigned long shmem_swap_usage(struct vm_area_struct *vma)
701{
702 struct inode *inode = file_inode(vma->vm_file);
703 struct shmem_inode_info *info = SHMEM_I(inode);
704 struct address_space *mapping = inode->i_mapping;
705 unsigned long swapped;
706
707 /* Be careful as we don't hold info->lock */
708 swapped = READ_ONCE(info->swapped);
709
710 /*
711 * The easier cases are when the shmem object has nothing in swap, or
712 * the vma maps it whole. Then we can simply use the stats that we
713 * already track.
714 */
715 if (!swapped)
716 return 0;
717
718 if (!vma->vm_pgoff && vma->vm_end - vma->vm_start >= inode->i_size)
719 return swapped << PAGE_SHIFT;
720
721 /* Here comes the more involved part */
722 return shmem_partial_swap_usage(mapping,
723 linear_page_index(vma, vma->vm_start),
724 linear_page_index(vma, vma->vm_end));
725}
726
727/*
Hugh Dickins24513262012-01-20 14:34:21 -0800728 * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
729 */
730void shmem_unlock_mapping(struct address_space *mapping)
731{
732 struct pagevec pvec;
733 pgoff_t indices[PAGEVEC_SIZE];
734 pgoff_t index = 0;
735
736 pagevec_init(&pvec, 0);
737 /*
738 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
739 */
740 while (!mapping_unevictable(mapping)) {
741 /*
742 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
743 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
744 */
Johannes Weiner0cd61442014-04-03 14:47:46 -0700745 pvec.nr = find_get_entries(mapping, index,
746 PAGEVEC_SIZE, pvec.pages, indices);
Hugh Dickins24513262012-01-20 14:34:21 -0800747 if (!pvec.nr)
748 break;
749 index = indices[pvec.nr - 1] + 1;
Johannes Weiner0cd61442014-04-03 14:47:46 -0700750 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800751 check_move_unevictable_pages(pvec.pages, pvec.nr);
752 pagevec_release(&pvec);
753 cond_resched();
754 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700755}
756
757/*
758 * Remove range of pages and swap entries from radix tree, and free them.
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700759 * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700760 */
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700761static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
762 bool unfalloc)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763{
Hugh Dickins285b2c42011-08-03 16:21:20 -0700764 struct address_space *mapping = inode->i_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300766 pgoff_t start = (lstart + PAGE_SIZE - 1) >> PAGE_SHIFT;
767 pgoff_t end = (lend + 1) >> PAGE_SHIFT;
768 unsigned int partial_start = lstart & (PAGE_SIZE - 1);
769 unsigned int partial_end = (lend + 1) & (PAGE_SIZE - 1);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700770 struct pagevec pvec;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700771 pgoff_t indices[PAGEVEC_SIZE];
772 long nr_swaps_freed = 0;
Hugh Dickins285b2c42011-08-03 16:21:20 -0700773 pgoff_t index;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700774 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700776 if (lend == -1)
777 end = -1; /* unsigned, so actually very big */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700778
779 pagevec_init(&pvec, 0);
780 index = start;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700781 while (index < end) {
Johannes Weiner0cd61442014-04-03 14:47:46 -0700782 pvec.nr = find_get_entries(mapping, index,
783 min(end - index, (pgoff_t)PAGEVEC_SIZE),
784 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700785 if (!pvec.nr)
786 break;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700787 for (i = 0; i < pagevec_count(&pvec); i++) {
788 struct page *page = pvec.pages[i];
789
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700790 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700791 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700792 break;
793
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700794 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700795 if (unfalloc)
796 continue;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700797 nr_swaps_freed += !shmem_free_swap(mapping,
798 index, page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700799 continue;
800 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700801
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700802 VM_BUG_ON_PAGE(page_to_pgoff(page) != index, page);
803
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700804 if (!trylock_page(page))
805 continue;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700806
807 if (PageTransTail(page)) {
808 /* Middle of THP: zero out the page */
809 clear_highpage(page);
810 unlock_page(page);
811 continue;
812 } else if (PageTransHuge(page)) {
813 if (index == round_down(end, HPAGE_PMD_NR)) {
814 /*
815 * Range ends in the middle of THP:
816 * zero out the page
817 */
818 clear_highpage(page);
819 unlock_page(page);
820 continue;
821 }
822 index += HPAGE_PMD_NR - 1;
823 i += HPAGE_PMD_NR - 1;
824 }
825
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700826 if (!unfalloc || !PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700827 VM_BUG_ON_PAGE(PageTail(page), page);
828 if (page_mapping(page) == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800829 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700830 truncate_inode_page(mapping, page);
831 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700832 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700833 unlock_page(page);
834 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700835 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800836 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700837 cond_resched();
838 index++;
839 }
840
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700841 if (partial_start) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700842 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700843 shmem_getpage(inode, start - 1, &page, SGP_READ);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700844 if (page) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300845 unsigned int top = PAGE_SIZE;
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700846 if (start > end) {
847 top = partial_end;
848 partial_end = 0;
849 }
850 zero_user_segment(page, partial_start, top);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700851 set_page_dirty(page);
852 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300853 put_page(page);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700854 }
855 }
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700856 if (partial_end) {
857 struct page *page = NULL;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -0700858 shmem_getpage(inode, end, &page, SGP_READ);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700859 if (page) {
860 zero_user_segment(page, 0, partial_end);
861 set_page_dirty(page);
862 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300863 put_page(page);
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700864 }
865 }
866 if (start >= end)
867 return;
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700868
869 index = start;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700870 while (index < end) {
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700871 cond_resched();
Johannes Weiner0cd61442014-04-03 14:47:46 -0700872
873 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700874 min(end - index, (pgoff_t)PAGEVEC_SIZE),
Johannes Weiner0cd61442014-04-03 14:47:46 -0700875 pvec.pages, indices);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700876 if (!pvec.nr) {
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700877 /* If all gone or hole-punch or unfalloc, we're done */
878 if (index == start || end != -1)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700879 break;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700880 /* But if truncating, restart to make sure all gone */
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700881 index = start;
882 continue;
883 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700884 for (i = 0; i < pagevec_count(&pvec); i++) {
885 struct page *page = pvec.pages[i];
886
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700887 index = indices[i];
Hugh Dickins83e4fa92012-05-29 15:06:40 -0700888 if (index >= end)
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700889 break;
890
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700891 if (radix_tree_exceptional_entry(page)) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700892 if (unfalloc)
893 continue;
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700894 if (shmem_free_swap(mapping, index, page)) {
895 /* Swap was replaced by page: retry */
896 index--;
897 break;
898 }
899 nr_swaps_freed++;
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700900 continue;
901 }
902
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700903 lock_page(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700904
905 if (PageTransTail(page)) {
906 /* Middle of THP: zero out the page */
907 clear_highpage(page);
908 unlock_page(page);
909 /*
910 * Partial thp truncate due 'start' in middle
911 * of THP: don't need to look on these pages
912 * again on !pvec.nr restart.
913 */
914 if (index != round_down(end, HPAGE_PMD_NR))
915 start++;
916 continue;
917 } else if (PageTransHuge(page)) {
918 if (index == round_down(end, HPAGE_PMD_NR)) {
919 /*
920 * Range ends in the middle of THP:
921 * zero out the page
922 */
923 clear_highpage(page);
924 unlock_page(page);
925 continue;
926 }
927 index += HPAGE_PMD_NR - 1;
928 i += HPAGE_PMD_NR - 1;
929 }
930
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700931 if (!unfalloc || !PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -0700932 VM_BUG_ON_PAGE(PageTail(page), page);
933 if (page_mapping(page) == mapping) {
Sasha Levin309381fea2014-01-23 15:52:54 -0800934 VM_BUG_ON_PAGE(PageWriteback(page), page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700935 truncate_inode_page(mapping, page);
Hugh Dickinsb1a36652014-07-23 14:00:13 -0700936 } else {
937 /* Page was replaced by swap: retry */
938 unlock_page(page);
939 index--;
940 break;
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700941 }
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700942 }
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700943 unlock_page(page);
944 }
Johannes Weiner0cd61442014-04-03 14:47:46 -0700945 pagevec_remove_exceptionals(&pvec);
Hugh Dickins24513262012-01-20 14:34:21 -0800946 pagevec_release(&pvec);
Hugh Dickinsbda97ea2011-08-03 16:21:21 -0700947 index++;
948 }
Hugh Dickins94c1e622011-06-27 16:18:03 -0700949
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700950 spin_lock_irq(&info->lock);
Hugh Dickins7a5d0fb2011-08-03 16:21:22 -0700951 info->swapped -= nr_swaps_freed;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700953 spin_unlock_irq(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700954}
Linus Torvalds1da177e2005-04-16 15:20:36 -0700955
Hugh Dickins1635f6a2012-05-29 15:06:42 -0700956void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
957{
958 shmem_undo_range(inode, lstart, lend, false);
Deepa Dinamani078cd822016-09-14 07:48:04 -0700959 inode->i_ctime = inode->i_mtime = current_time(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960}
Hugh Dickins94c1e622011-06-27 16:18:03 -0700961EXPORT_SYMBOL_GPL(shmem_truncate_range);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962
David Howellsa528d352017-01-31 16:46:22 +0000963static int shmem_getattr(const struct path *path, struct kstat *stat,
964 u32 request_mask, unsigned int query_flags)
Yu Zhao44a30222015-09-08 15:03:33 -0700965{
David Howellsa528d352017-01-31 16:46:22 +0000966 struct inode *inode = path->dentry->d_inode;
Yu Zhao44a30222015-09-08 15:03:33 -0700967 struct shmem_inode_info *info = SHMEM_I(inode);
968
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800969 if (info->alloced - info->swapped != inode->i_mapping->nrpages) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700970 spin_lock_irq(&info->lock);
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800971 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -0700972 spin_unlock_irq(&info->lock);
Hugh Dickinsd0424c42015-11-05 18:50:34 -0800973 }
Yu Zhao44a30222015-09-08 15:03:33 -0700974 generic_fillattr(inode, stat);
Yu Zhao44a30222015-09-08 15:03:33 -0700975 return 0;
976}
977
Hugh Dickins94c1e622011-06-27 16:18:03 -0700978static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979{
David Howells75c3cfa2015-03-17 22:26:12 +0000980 struct inode *inode = d_inode(dentry);
David Herrmann40e041a2014-08-08 14:25:27 -0700981 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -0700982 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 int error;
984
Jan Kara31051c82016-05-26 16:55:18 +0200985 error = setattr_prepare(dentry, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +0200986 if (error)
987 return error;
988
Hugh Dickins94c1e622011-06-27 16:18:03 -0700989 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
990 loff_t oldsize = inode->i_size;
991 loff_t newsize = attr->ia_size;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +1000992
David Herrmann40e041a2014-08-08 14:25:27 -0700993 /* protected by i_mutex */
994 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
995 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
996 return -EPERM;
997
Hugh Dickins94c1e622011-06-27 16:18:03 -0700998 if (newsize != oldsize) {
Konstantin Khlebnikov77142512014-08-06 16:06:34 -0700999 error = shmem_reacct_size(SHMEM_I(inode)->flags,
1000 oldsize, newsize);
1001 if (error)
1002 return error;
Hugh Dickins94c1e622011-06-27 16:18:03 -07001003 i_size_write(inode, newsize);
Deepa Dinamani078cd822016-09-14 07:48:04 -07001004 inode->i_ctime = inode->i_mtime = current_time(inode);
Hugh Dickins94c1e622011-06-27 16:18:03 -07001005 }
Josef Bacikafa2db22015-06-24 16:58:45 -07001006 if (newsize <= oldsize) {
Hugh Dickins94c1e622011-06-27 16:18:03 -07001007 loff_t holebegin = round_up(newsize, PAGE_SIZE);
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001008 if (oldsize > holebegin)
1009 unmap_mapping_range(inode->i_mapping,
1010 holebegin, 0, 1);
1011 if (info->alloced)
1012 shmem_truncate_range(inode,
1013 newsize, (loff_t)-1);
Hugh Dickins94c1e622011-06-27 16:18:03 -07001014 /* unmap again to remove racily COWed private pages */
Hugh Dickinsd0424c42015-11-05 18:50:34 -08001015 if (oldsize > holebegin)
1016 unmap_mapping_range(inode->i_mapping,
1017 holebegin, 0, 1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001018
1019 /*
1020 * Part of the huge page can be beyond i_size: subject
1021 * to shrink under memory pressure.
1022 */
1023 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE)) {
1024 spin_lock(&sbinfo->shrinklist_lock);
1025 if (list_empty(&info->shrinklist)) {
1026 list_add_tail(&info->shrinklist,
1027 &sbinfo->shrinklist);
1028 sbinfo->shrinklist_len++;
1029 }
1030 spin_unlock(&sbinfo->shrinklist_lock);
1031 }
Hugh Dickins94c1e622011-06-27 16:18:03 -07001032 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 }
1034
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001035 setattr_copy(inode, attr);
Christoph Hellwigdb78b872010-06-04 11:30:03 +02001036 if (attr->ia_valid & ATTR_MODE)
Christoph Hellwigfeda8212013-12-20 05:16:54 -08001037 error = posix_acl_chmod(inode, inode->i_mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 return error;
1039}
1040
Al Viro1f895f72010-06-05 19:10:41 -04001041static void shmem_evict_inode(struct inode *inode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001044 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001046 if (inode->i_mapping->a_ops == &shmem_aops) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 shmem_unacct_size(info->flags, inode->i_size);
1048 inode->i_size = 0;
npiggin@suse.de3889e6e2010-05-27 01:05:36 +10001049 shmem_truncate_range(inode, 0, (loff_t)-1);
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001050 if (!list_empty(&info->shrinklist)) {
1051 spin_lock(&sbinfo->shrinklist_lock);
1052 if (!list_empty(&info->shrinklist)) {
1053 list_del_init(&info->shrinklist);
1054 sbinfo->shrinklist_len--;
1055 }
1056 spin_unlock(&sbinfo->shrinklist_lock);
1057 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 if (!list_empty(&info->swaplist)) {
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001059 mutex_lock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001060 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001061 mutex_unlock(&shmem_swaplist_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 }
Al Viro3ed47db2016-01-22 18:08:52 -05001063 }
Eric Parisb09e0fa2011-05-24 17:12:39 -07001064
Aristeu Rozanski38f38652012-08-23 16:53:28 -04001065 simple_xattrs_free(&info->xattrs);
Hugh Dickins0f3c42f2012-11-16 14:15:04 -08001066 WARN_ON(inode->i_blocks);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08001067 shmem_free_inode(inode->i_sb);
Jan Karadbd57682012-05-03 14:48:02 +02001068 clear_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001069}
1070
Matthew Wilcox478922e2016-12-14 15:08:52 -08001071static unsigned long find_swap_entry(struct radix_tree_root *root, void *item)
1072{
1073 struct radix_tree_iter iter;
1074 void **slot;
1075 unsigned long found = -1;
1076 unsigned int checked = 0;
1077
1078 rcu_read_lock();
1079 radix_tree_for_each_slot(slot, root, &iter, 0) {
1080 if (*slot == item) {
1081 found = iter.index;
1082 break;
1083 }
1084 checked++;
1085 if ((checked % 4096) != 0)
1086 continue;
1087 slot = radix_tree_iter_resume(slot, &iter);
1088 cond_resched_rcu();
1089 }
1090
1091 rcu_read_unlock();
1092 return found;
1093}
1094
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001095/*
1096 * If swap found in inode, free it and move page from swapcache to filecache.
1097 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001098static int shmem_unuse_inode(struct shmem_inode_info *info,
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001099 swp_entry_t swap, struct page **pagep)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100{
Hugh Dickins285b2c42011-08-03 16:21:20 -07001101 struct address_space *mapping = info->vfs_inode.i_mapping;
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001102 void *radswap;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001103 pgoff_t index;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001104 gfp_t gfp;
1105 int error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001107 radswap = swp_to_radix_entry(swap);
Matthew Wilcox478922e2016-12-14 15:08:52 -08001108 index = find_swap_entry(&mapping->page_tree, radswap);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001109 if (index == -1)
Johannes Weiner00501b52014-08-08 14:19:20 -07001110 return -EAGAIN; /* tell shmem_unuse we found nothing */
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001111
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001112 /*
1113 * Move _head_ to start search for next from here.
Al Viro1f895f72010-06-05 19:10:41 -04001114 * But be careful: shmem_evict_inode checks list_empty without taking
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001115 * mutex, and there's an instant in list_move_tail when info->swaplist
Hugh Dickins285b2c42011-08-03 16:21:20 -07001116 * would appear empty, if it were the only one on shmem_swaplist.
Hugh Dickins1b1b32f2008-02-04 22:28:55 -08001117 */
1118 if (shmem_swaplist.next != &info->swaplist)
1119 list_move_tail(&shmem_swaplist, &info->swaplist);
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001120
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001121 gfp = mapping_gfp_mask(mapping);
1122 if (shmem_should_replace_page(*pagep, gfp)) {
1123 mutex_unlock(&shmem_swaplist_mutex);
1124 error = shmem_replace_page(pagep, gfp, info, index);
1125 mutex_lock(&shmem_swaplist_mutex);
1126 /*
1127 * We needed to drop mutex to make that restrictive page
Hugh Dickins0142ef62012-06-07 14:21:09 -07001128 * allocation, but the inode might have been freed while we
1129 * dropped it: although a racing shmem_evict_inode() cannot
1130 * complete without emptying the radix_tree, our page lock
1131 * on this swapcache page is not enough to prevent that -
1132 * free_swap_and_cache() of our swap entry will only
1133 * trylock_page(), removing swap from radix_tree whatever.
1134 *
1135 * We must not proceed to shmem_add_to_page_cache() if the
1136 * inode has been freed, but of course we cannot rely on
1137 * inode or mapping or info to check that. However, we can
1138 * safely check if our swap entry is still in use (and here
1139 * it can't have got reused for another page): if it's still
1140 * in use, then the inode cannot have been freed yet, and we
1141 * can safely proceed (if it's no longer in use, that tells
1142 * nothing about the inode, but we don't need to unuse swap).
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001143 */
1144 if (!page_swapcount(*pagep))
1145 error = -ENOENT;
1146 }
1147
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001148 /*
Hugh Dickins778dd892011-05-11 15:13:37 -07001149 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
1150 * but also to hold up shmem_evict_inode(): so inode cannot be freed
1151 * beneath us (pagelock doesn't help until the page is in pagecache).
KAMEZAWA Hiroyukid13d1442009-01-07 18:07:56 -08001152 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001153 if (!error)
1154 error = shmem_add_to_page_cache(*pagep, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001155 radswap);
Hugh Dickins48f170f2011-07-25 17:12:37 -07001156 if (error != -ENOMEM) {
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001157 /*
1158 * Truncation and eviction use free_swap_and_cache(), which
1159 * only does trylock page: if we raced, best clean up here.
1160 */
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001161 delete_from_swap_cache(*pagep);
1162 set_page_dirty(*pagep);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001163 if (!error) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001164 spin_lock_irq(&info->lock);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001165 info->swapped--;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001166 spin_unlock_irq(&info->lock);
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001167 swap_free(swap);
1168 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 }
Hugh Dickins2e0e26c2008-02-04 22:28:53 -08001170 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171}
1172
1173/*
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001174 * Search through swapped inodes to find and replace swap by page.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001175 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001176int shmem_unuse(swp_entry_t swap, struct page *page)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001178 struct list_head *this, *next;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 struct shmem_inode_info *info;
Johannes Weiner00501b52014-08-08 14:19:20 -07001180 struct mem_cgroup *memcg;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001181 int error = 0;
1182
1183 /*
1184 * There's a faint possibility that swap page was replaced before
Hugh Dickins0142ef62012-06-07 14:21:09 -07001185 * caller locked it: caller will come back later with the right page.
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001186 */
Hugh Dickins0142ef62012-06-07 14:21:09 -07001187 if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001188 goto out;
Hugh Dickins778dd892011-05-11 15:13:37 -07001189
1190 /*
1191 * Charge page using GFP_KERNEL while we can wait, before taking
1192 * the shmem_swaplist_mutex which might hold up shmem_writepage().
1193 * Charged back to the user (not to caller) when swap account is used.
Hugh Dickins778dd892011-05-11 15:13:37 -07001194 */
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001195 error = mem_cgroup_try_charge(page, current->mm, GFP_KERNEL, &memcg,
1196 false);
Hugh Dickins778dd892011-05-11 15:13:37 -07001197 if (error)
1198 goto out;
Hugh Dickins46f65ec2011-08-03 16:21:23 -07001199 /* No radix_tree_preload: swap entry keeps a place for page in tree */
Johannes Weiner00501b52014-08-08 14:19:20 -07001200 error = -EAGAIN;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001202 mutex_lock(&shmem_swaplist_mutex);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001203 list_for_each_safe(this, next, &shmem_swaplist) {
1204 info = list_entry(this, struct shmem_inode_info, swaplist);
Hugh Dickins285b2c42011-08-03 16:21:20 -07001205 if (info->swapped)
Johannes Weiner00501b52014-08-08 14:19:20 -07001206 error = shmem_unuse_inode(info, swap, &page);
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001207 else
1208 list_del_init(&info->swaplist);
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001209 cond_resched();
Johannes Weiner00501b52014-08-08 14:19:20 -07001210 if (error != -EAGAIN)
Hugh Dickins778dd892011-05-11 15:13:37 -07001211 break;
Johannes Weiner00501b52014-08-08 14:19:20 -07001212 /* found nothing in this: move on to search the next */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 }
Hugh Dickinscb5f7b92008-02-04 22:28:52 -08001214 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickins778dd892011-05-11 15:13:37 -07001215
Johannes Weiner00501b52014-08-08 14:19:20 -07001216 if (error) {
1217 if (error != -ENOMEM)
1218 error = 0;
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001219 mem_cgroup_cancel_charge(page, memcg, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001220 } else
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001221 mem_cgroup_commit_charge(page, memcg, true, false);
Hugh Dickins778dd892011-05-11 15:13:37 -07001222out:
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001223 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001224 put_page(page);
Hugh Dickins778dd892011-05-11 15:13:37 -07001225 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001226}
1227
1228/*
1229 * Move the page from the page cache to the swap cache.
1230 */
1231static int shmem_writepage(struct page *page, struct writeback_control *wbc)
1232{
1233 struct shmem_inode_info *info;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 struct inode *inode;
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001236 swp_entry_t swap;
1237 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001238
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001239 VM_BUG_ON_PAGE(PageCompound(page), page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 mapping = page->mapping;
1242 index = page->index;
1243 inode = mapping->host;
1244 info = SHMEM_I(inode);
1245 if (info->flags & VM_LOCKED)
1246 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001247 if (!total_swap_pages)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 goto redirty;
1249
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001250 /*
Christoph Hellwig97b713b2015-01-14 10:42:31 +01001251 * Our capabilities prevent regular writeback or sync from ever calling
1252 * shmem_writepage; but a stacking filesystem might use ->writepage of
1253 * its underlying filesystem, in which case tmpfs should write out to
1254 * swap only in response to memory pressure, and not for the writeback
1255 * threads or sync.
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001256 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001257 if (!wbc->for_reclaim) {
1258 WARN_ON_ONCE(1); /* Still happens? Tell us about it! */
1259 goto redirty;
1260 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001261
1262 /*
1263 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
1264 * value into swapfile.c, the only way we can correctly account for a
1265 * fallocated page arriving here is now to initialize it and write it.
Hugh Dickins1aac1402012-05-29 15:06:42 -07001266 *
1267 * That's okay for a page already fallocated earlier, but if we have
1268 * not yet completed the fallocation, then (a) we want to keep track
1269 * of this page in case we have to undo it, and (b) it may not be a
1270 * good idea to continue anyway, once we're pushing into swap. So
1271 * reactivate the page, and let shmem_fallocate() quit when too many.
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001272 */
1273 if (!PageUptodate(page)) {
Hugh Dickins1aac1402012-05-29 15:06:42 -07001274 if (inode->i_private) {
1275 struct shmem_falloc *shmem_falloc;
1276 spin_lock(&inode->i_lock);
1277 shmem_falloc = inode->i_private;
1278 if (shmem_falloc &&
Hugh Dickins8e205f72014-07-23 14:00:10 -07001279 !shmem_falloc->waitq &&
Hugh Dickins1aac1402012-05-29 15:06:42 -07001280 index >= shmem_falloc->start &&
1281 index < shmem_falloc->next)
1282 shmem_falloc->nr_unswapped++;
1283 else
1284 shmem_falloc = NULL;
1285 spin_unlock(&inode->i_lock);
1286 if (shmem_falloc)
1287 goto redirty;
1288 }
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001289 clear_highpage(page);
1290 flush_dcache_page(page);
1291 SetPageUptodate(page);
1292 }
1293
Huang Ying38d8b4e2017-07-06 15:37:18 -07001294 swap = get_swap_page(page);
Hugh Dickins48f170f2011-07-25 17:12:37 -07001295 if (!swap.val)
1296 goto redirty;
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001297
Vladimir Davydov37e84352016-01-20 15:02:56 -08001298 if (mem_cgroup_try_charge_swap(page, swap))
1299 goto free_swap;
1300
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001301 /*
1302 * Add inode to shmem_unuse()'s list of swapped-out inodes,
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001303 * if it's not already there. Do it now before the page is
1304 * moved to swap cache, when its pagelock no longer protects
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001305 * the inode from eviction. But don't unlock the mutex until
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001306 * we've incremented swapped, because shmem_unuse_inode() will
1307 * prune a !swapped inode from the swaplist under this mutex.
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001308 */
Hugh Dickins48f170f2011-07-25 17:12:37 -07001309 mutex_lock(&shmem_swaplist_mutex);
1310 if (list_empty(&info->swaplist))
1311 list_add_tail(&info->swaplist, &shmem_swaplist);
Hugh Dickinsb1dea802011-05-11 15:13:36 -07001312
Hugh Dickins48f170f2011-07-25 17:12:37 -07001313 if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001314 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001315 shmem_recalc_inode(inode);
1316 info->swapped++;
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001317 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001318
Hugh Dickinsaaa46862009-12-14 17:58:47 -08001319 swap_shmem_alloc(swap);
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001320 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
1321
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001322 mutex_unlock(&shmem_swaplist_mutex);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001323 BUG_ON(page_mapped(page));
Hugh Dickins9fab5612009-03-31 15:23:33 -07001324 swap_writepage(page, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 return 0;
1326 }
1327
Hugh Dickins6922c0c2011-08-03 16:21:25 -07001328 mutex_unlock(&shmem_swaplist_mutex);
Vladimir Davydov37e84352016-01-20 15:02:56 -08001329free_swap:
Minchan Kim75f6d6d2017-07-06 15:37:21 -07001330 put_swap_page(page, swap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331redirty:
1332 set_page_dirty(page);
Hugh Dickinsd9fe5262008-02-04 22:28:51 -08001333 if (wbc->for_reclaim)
1334 return AOP_WRITEPAGE_ACTIVATE; /* Return with page locked */
1335 unlock_page(page);
1336 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337}
1338
Hugh Dickins75edd342016-05-19 17:12:44 -07001339#if defined(CONFIG_NUMA) && defined(CONFIG_TMPFS)
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001340static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001341{
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001342 char buffer[64];
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001343
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001344 if (!mpol || mpol->mode == MPOL_DEFAULT)
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001345 return; /* show nothing */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001346
Hugh Dickinsa7a88b22013-01-02 02:04:23 -08001347 mpol_to_str(buffer, sizeof(buffer), mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001348
Lee Schermerhorn095f1fc2008-04-28 02:13:23 -07001349 seq_printf(seq, ",mpol=%s", buffer);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001350}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001351
1352static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1353{
1354 struct mempolicy *mpol = NULL;
1355 if (sbinfo->mpol) {
1356 spin_lock(&sbinfo->stat_lock); /* prevent replace/use races */
1357 mpol = sbinfo->mpol;
1358 mpol_get(mpol);
1359 spin_unlock(&sbinfo->stat_lock);
1360 }
1361 return mpol;
1362}
Hugh Dickins75edd342016-05-19 17:12:44 -07001363#else /* !CONFIG_NUMA || !CONFIG_TMPFS */
1364static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
1365{
1366}
1367static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
1368{
1369 return NULL;
1370}
1371#endif /* CONFIG_NUMA && CONFIG_TMPFS */
1372#ifndef CONFIG_NUMA
1373#define vm_policy vm_private_data
1374#endif
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08001375
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001376static void shmem_pseudo_vma_init(struct vm_area_struct *vma,
1377 struct shmem_inode_info *info, pgoff_t index)
1378{
1379 /* Create a pseudo vma that just contains the policy */
1380 vma->vm_start = 0;
1381 /* Bias interleave by inode number to distribute better across nodes */
1382 vma->vm_pgoff = index + info->vfs_inode.i_ino;
1383 vma->vm_ops = NULL;
1384 vma->vm_policy = mpol_shared_policy_lookup(&info->policy, index);
1385}
1386
1387static void shmem_pseudo_vma_destroy(struct vm_area_struct *vma)
1388{
1389 /* Drop reference taken by mpol_shared_policy_lookup() */
1390 mpol_cond_put(vma->vm_policy);
1391}
1392
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001393static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
1394 struct shmem_inode_info *info, pgoff_t index)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 struct vm_area_struct pvma;
Mel Gorman18a2f372012-12-05 14:01:41 -08001397 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001399 shmem_pseudo_vma_init(&pvma, info, index);
Mel Gorman18a2f372012-12-05 14:01:41 -08001400 page = swapin_readahead(swap, gfp, &pvma, 0);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001401 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001402
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001403 return page;
1404}
Mel Gorman18a2f372012-12-05 14:01:41 -08001405
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001406static struct page *shmem_alloc_hugepage(gfp_t gfp,
1407 struct shmem_inode_info *info, pgoff_t index)
1408{
1409 struct vm_area_struct pvma;
1410 struct inode *inode = &info->vfs_inode;
1411 struct address_space *mapping = inode->i_mapping;
Geert Uytterhoeven4620a062016-08-03 19:58:19 +02001412 pgoff_t idx, hindex;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001413 void __rcu **results;
1414 struct page *page;
1415
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001416 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001417 return NULL;
1418
Geert Uytterhoeven4620a062016-08-03 19:58:19 +02001419 hindex = round_down(index, HPAGE_PMD_NR);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001420 rcu_read_lock();
1421 if (radix_tree_gang_lookup_slot(&mapping->page_tree, &results, &idx,
1422 hindex, 1) && idx < hindex + HPAGE_PMD_NR) {
1423 rcu_read_unlock();
1424 return NULL;
1425 }
1426 rcu_read_unlock();
1427
1428 shmem_pseudo_vma_init(&pvma, info, hindex);
1429 page = alloc_pages_vma(gfp | __GFP_COMP | __GFP_NORETRY | __GFP_NOWARN,
1430 HPAGE_PMD_ORDER, &pvma, 0, numa_node_id(), true);
1431 shmem_pseudo_vma_destroy(&pvma);
1432 if (page)
1433 prep_transhuge_page(page);
Mel Gorman18a2f372012-12-05 14:01:41 -08001434 return page;
1435}
1436
1437static struct page *shmem_alloc_page(gfp_t gfp,
1438 struct shmem_inode_info *info, pgoff_t index)
1439{
1440 struct vm_area_struct pvma;
1441 struct page *page;
1442
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001443 shmem_pseudo_vma_init(&pvma, info, index);
1444 page = alloc_page_vma(gfp, &pvma, 0);
1445 shmem_pseudo_vma_destroy(&pvma);
Mel Gorman18a2f372012-12-05 14:01:41 -08001446
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001447 return page;
1448}
1449
1450static struct page *shmem_alloc_and_acct_page(gfp_t gfp,
1451 struct shmem_inode_info *info, struct shmem_sb_info *sbinfo,
1452 pgoff_t index, bool huge)
1453{
1454 struct page *page;
1455 int nr;
1456 int err = -ENOSPC;
1457
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07001458 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001459 huge = false;
1460 nr = huge ? HPAGE_PMD_NR : 1;
1461
1462 if (shmem_acct_block(info->flags, nr))
1463 goto failed;
1464 if (sbinfo->max_blocks) {
1465 if (percpu_counter_compare(&sbinfo->used_blocks,
1466 sbinfo->max_blocks - nr) > 0)
1467 goto unacct;
1468 percpu_counter_add(&sbinfo->used_blocks, nr);
1469 }
1470
1471 if (huge)
1472 page = shmem_alloc_hugepage(gfp, info, index);
1473 else
1474 page = shmem_alloc_page(gfp, info, index);
Hugh Dickins75edd342016-05-19 17:12:44 -07001475 if (page) {
1476 __SetPageLocked(page);
1477 __SetPageSwapBacked(page);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001478 return page;
Hugh Dickins75edd342016-05-19 17:12:44 -07001479 }
Mel Gorman18a2f372012-12-05 14:01:41 -08001480
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001481 err = -ENOMEM;
1482 if (sbinfo->max_blocks)
1483 percpu_counter_add(&sbinfo->used_blocks, -nr);
1484unacct:
1485 shmem_unacct_blocks(info->flags, nr);
1486failed:
1487 return ERR_PTR(err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488}
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07001489
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490/*
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001491 * When a page is moved from swapcache to shmem filecache (either by the
1492 * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
1493 * shmem_unuse_inode()), it may have been read in earlier from swap, in
1494 * ignorance of the mapping it belongs to. If that mapping has special
1495 * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
1496 * we may need to copy to a suitable page before moving to filecache.
1497 *
1498 * In a future release, this may well be extended to respect cpuset and
1499 * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
1500 * but for now it is a simple matter of zone.
1501 */
1502static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
1503{
1504 return page_zonenum(page) > gfp_zone(gfp);
1505}
1506
1507static int shmem_replace_page(struct page **pagep, gfp_t gfp,
1508 struct shmem_inode_info *info, pgoff_t index)
1509{
1510 struct page *oldpage, *newpage;
1511 struct address_space *swap_mapping;
1512 pgoff_t swap_index;
1513 int error;
1514
1515 oldpage = *pagep;
1516 swap_index = page_private(oldpage);
1517 swap_mapping = page_mapping(oldpage);
1518
1519 /*
1520 * We have arrived here because our zones are constrained, so don't
1521 * limit chance of success by further cpuset and node constraints.
1522 */
1523 gfp &= ~GFP_CONSTRAINT_MASK;
1524 newpage = shmem_alloc_page(gfp, info, index);
1525 if (!newpage)
1526 return -ENOMEM;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001527
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001528 get_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001529 copy_highpage(newpage, oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001530 flush_dcache_page(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001531
Hugh Dickins9956edf2016-11-10 10:46:11 -08001532 __SetPageLocked(newpage);
1533 __SetPageSwapBacked(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001534 SetPageUptodate(newpage);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001535 set_page_private(newpage, swap_index);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001536 SetPageSwapCache(newpage);
1537
1538 /*
1539 * Our caller will very soon move newpage out of swapcache, but it's
1540 * a nice clean interface for us to replace oldpage by newpage there.
1541 */
1542 spin_lock_irq(&swap_mapping->tree_lock);
1543 error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1544 newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001545 if (!error) {
Mel Gorman11fb9982016-07-28 15:46:20 -07001546 __inc_node_page_state(newpage, NR_FILE_PAGES);
1547 __dec_node_page_state(oldpage, NR_FILE_PAGES);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001548 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001549 spin_unlock_irq(&swap_mapping->tree_lock);
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001550
Hugh Dickins0142ef62012-06-07 14:21:09 -07001551 if (unlikely(error)) {
1552 /*
1553 * Is this possible? I think not, now that our callers check
1554 * both PageSwapCache and page_private after getting page lock;
1555 * but be defensive. Reverse old to newpage for clear and free.
1556 */
1557 oldpage = newpage;
1558 } else {
Johannes Weiner6a93ca82016-03-15 14:57:19 -07001559 mem_cgroup_migrate(oldpage, newpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001560 lru_cache_add_anon(newpage);
1561 *pagep = newpage;
1562 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001563
1564 ClearPageSwapCache(oldpage);
1565 set_page_private(oldpage, 0);
1566
1567 unlock_page(oldpage);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001568 put_page(oldpage);
1569 put_page(oldpage);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001570 return error;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001571}
1572
1573/*
Hugh Dickins68da9f02011-07-25 17:12:34 -07001574 * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001575 *
1576 * If we allocate a new one we do not mark it dirty. That's up to the
1577 * vm. If we swap it in we mark it dirty since we also free the swap
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001578 * entry since a page cannot live in both the swap and page cache.
1579 *
1580 * fault_mm and fault_type are only supplied by shmem_fault:
1581 * otherwise they are NULL.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001583static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001584 struct page **pagep, enum sgp_type sgp, gfp_t gfp,
Mike Rapoportcfda0522017-02-22 15:43:37 -08001585 struct vm_area_struct *vma, struct vm_fault *vmf, int *fault_type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586{
1587 struct address_space *mapping = inode->i_mapping;
Arnd Bergmann23f919d2016-12-12 16:42:28 -08001588 struct shmem_inode_info *info = SHMEM_I(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001589 struct shmem_sb_info *sbinfo;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001590 struct mm_struct *charge_mm;
Johannes Weiner00501b52014-08-08 14:19:20 -07001591 struct mem_cgroup *memcg;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001592 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001593 swp_entry_t swap;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001594 enum sgp_type sgp_huge = sgp;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001595 pgoff_t hindex = index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 int error;
Hugh Dickins54af60422011-08-03 16:21:24 -07001597 int once = 0;
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001598 int alloced = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001600 if (index > (MAX_LFS_FILESIZE >> PAGE_SHIFT))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 return -EFBIG;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001602 if (sgp == SGP_NOHUGE || sgp == SGP_HUGE)
1603 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604repeat:
Hugh Dickins54af60422011-08-03 16:21:24 -07001605 swap.val = 0;
Johannes Weiner0cd61442014-04-03 14:47:46 -07001606 page = find_lock_entry(mapping, index);
Hugh Dickins54af60422011-08-03 16:21:24 -07001607 if (radix_tree_exceptional_entry(page)) {
1608 swap = radix_to_swp_entry(page);
1609 page = NULL;
1610 }
1611
Hugh Dickins75edd342016-05-19 17:12:44 -07001612 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001613 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001614 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001615 goto unlock;
Hugh Dickins54af60422011-08-03 16:21:24 -07001616 }
1617
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001618 if (page && sgp == SGP_WRITE)
1619 mark_page_accessed(page);
1620
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001621 /* fallocated page? */
1622 if (page && !PageUptodate(page)) {
1623 if (sgp != SGP_READ)
1624 goto clear;
1625 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001626 put_page(page);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001627 page = NULL;
1628 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001629 if (page || (sgp == SGP_READ && !swap.val)) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001630 *pagep = page;
1631 return 0;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001632 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001633
1634 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001635 * Fast cache lookup did not find it:
1636 * bring it back from swap or allocate.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001638 sbinfo = SHMEM_SB(inode->i_sb);
Mike Rapoportcfda0522017-02-22 15:43:37 -08001639 charge_mm = vma ? vma->vm_mm : current->mm;
Hugh Dickins27ab7002011-07-25 17:12:36 -07001640
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 if (swap.val) {
1642 /* Look it up and read it in.. */
Hugh Dickins27ab7002011-07-25 17:12:36 -07001643 page = lookup_swap_cache(swap);
1644 if (!page) {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001645 /* Or update major stats only when swapin succeeds?? */
1646 if (fault_type) {
Hugh Dickins68da9f02011-07-25 17:12:34 -07001647 *fault_type |= VM_FAULT_MAJOR;
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001648 count_vm_event(PGMAJFAULT);
Mike Rapoportcfda0522017-02-22 15:43:37 -08001649 mem_cgroup_count_vm_event(charge_mm,
1650 PGMAJFAULT);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001651 }
1652 /* Here we actually start the io */
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07001653 page = shmem_swapin(swap, gfp, info, index);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001654 if (!page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001655 error = -ENOMEM;
1656 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001658 }
1659
1660 /* We have to do this with page locked to prevent races */
Hugh Dickins54af60422011-08-03 16:21:24 -07001661 lock_page(page);
Hugh Dickins0142ef62012-06-07 14:21:09 -07001662 if (!PageSwapCache(page) || page_private(page) != swap.val ||
Hugh Dickinsd1899222012-07-11 14:02:47 -07001663 !shmem_confirm_swap(mapping, index, swap)) {
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001664 error = -EEXIST; /* try again */
Hugh Dickinsd1899222012-07-11 14:02:47 -07001665 goto unlock;
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001666 }
Hugh Dickins27ab7002011-07-25 17:12:36 -07001667 if (!PageUptodate(page)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668 error = -EIO;
Hugh Dickins54af60422011-08-03 16:21:24 -07001669 goto failed;
1670 }
1671 wait_on_page_writeback(page);
1672
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001673 if (shmem_should_replace_page(page, gfp)) {
1674 error = shmem_replace_page(&page, gfp, info, index);
1675 if (error)
1676 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001677 }
1678
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001679 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001680 false);
Hugh Dickinsd1899222012-07-11 14:02:47 -07001681 if (!error) {
Hugh Dickinsaa3b1892011-08-03 16:21:24 -07001682 error = shmem_add_to_page_cache(page, mapping, index,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001683 swp_to_radix_entry(swap));
Hugh Dickins215c02b2012-11-16 14:15:03 -08001684 /*
1685 * We already confirmed swap under page lock, and make
1686 * no memory allocation here, so usually no possibility
1687 * of error; but free_swap_and_cache() only trylocks a
1688 * page, so it is just possible that the entry has been
1689 * truncated or holepunched since swap was confirmed.
1690 * shmem_undo_range() will have done some of the
1691 * unaccounting, now delete_from_swap_cache() will do
Vladimir Davydov93aa7d92015-02-11 15:24:59 -08001692 * the rest.
Hugh Dickins215c02b2012-11-16 14:15:03 -08001693 * Reset swap.val? No, leave it so "failed" goes back to
1694 * "repeat": reading a hole and writing should succeed.
1695 */
Johannes Weiner00501b52014-08-08 14:19:20 -07001696 if (error) {
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001697 mem_cgroup_cancel_charge(page, memcg, false);
Hugh Dickins215c02b2012-11-16 14:15:03 -08001698 delete_from_swap_cache(page);
Johannes Weiner00501b52014-08-08 14:19:20 -07001699 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001700 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001701 if (error)
1702 goto failed;
1703
Kirill A. Shutemovf627c2f2016-01-15 16:52:20 -08001704 mem_cgroup_commit_charge(page, memcg, true, false);
Johannes Weiner00501b52014-08-08 14:19:20 -07001705
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001706 spin_lock_irq(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001707 info->swapped--;
1708 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001709 spin_unlock_irq(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001710
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001711 if (sgp == SGP_WRITE)
1712 mark_page_accessed(page);
1713
Hugh Dickins27ab7002011-07-25 17:12:36 -07001714 delete_from_swap_cache(page);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001715 set_page_dirty(page);
1716 swap_free(swap);
1717
Hugh Dickins54af60422011-08-03 16:21:24 -07001718 } else {
Mike Rapoportcfda0522017-02-22 15:43:37 -08001719 if (vma && userfaultfd_missing(vma)) {
1720 *fault_type = handle_userfault(vmf, VM_UFFD_MISSING);
1721 return 0;
1722 }
1723
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001724 /* shmem_symlink() */
1725 if (mapping->a_ops != &shmem_aops)
1726 goto alloc_nohuge;
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001727 if (shmem_huge == SHMEM_HUGE_DENY || sgp_huge == SGP_NOHUGE)
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001728 goto alloc_nohuge;
1729 if (shmem_huge == SHMEM_HUGE_FORCE)
1730 goto alloc_huge;
1731 switch (sbinfo->huge) {
1732 loff_t i_size;
1733 pgoff_t off;
1734 case SHMEM_HUGE_NEVER:
1735 goto alloc_nohuge;
1736 case SHMEM_HUGE_WITHIN_SIZE:
1737 off = round_up(index, HPAGE_PMD_NR);
1738 i_size = round_up(i_size_read(inode), PAGE_SIZE);
1739 if (i_size >= HPAGE_PMD_SIZE &&
1740 i_size >> PAGE_SHIFT >= off)
1741 goto alloc_huge;
1742 /* fallthrough */
1743 case SHMEM_HUGE_ADVISE:
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001744 if (sgp_huge == SGP_HUGE)
1745 goto alloc_huge;
1746 /* TODO: implement fadvise() hints */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001747 goto alloc_nohuge;
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001748 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001750alloc_huge:
1751 page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1752 index, true);
1753 if (IS_ERR(page)) {
1754alloc_nohuge: page = shmem_alloc_and_acct_page(gfp, info, sbinfo,
1755 index, false);
Hugh Dickins54af60422011-08-03 16:21:24 -07001756 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001757 if (IS_ERR(page)) {
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001758 int retry = 5;
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001759 error = PTR_ERR(page);
1760 page = NULL;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001761 if (error != -ENOSPC)
1762 goto failed;
1763 /*
1764 * Try to reclaim some spece by splitting a huge page
1765 * beyond i_size on the filesystem.
1766 */
1767 while (retry--) {
1768 int ret;
1769 ret = shmem_unused_huge_shrink(sbinfo, NULL, 1);
1770 if (ret == SHRINK_STOP)
1771 break;
1772 if (ret)
1773 goto alloc_nohuge;
1774 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001775 goto failed;
1776 }
1777
1778 if (PageTransHuge(page))
1779 hindex = round_down(index, HPAGE_PMD_NR);
1780 else
1781 hindex = index;
1782
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001783 if (sgp == SGP_WRITE)
Hugh Dickinseb39d612014-08-06 16:06:43 -07001784 __SetPageReferenced(page);
Hugh Dickins66d2f4d2014-07-02 15:22:38 -07001785
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001786 error = mem_cgroup_try_charge(page, charge_mm, gfp, &memcg,
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001787 PageTransHuge(page));
Hugh Dickins54af60422011-08-03 16:21:24 -07001788 if (error)
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001789 goto unacct;
1790 error = radix_tree_maybe_preload_order(gfp & GFP_RECLAIM_MASK,
1791 compound_order(page));
Hugh Dickinsb065b432012-07-11 14:02:48 -07001792 if (!error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001793 error = shmem_add_to_page_cache(page, mapping, hindex,
Wang Sheng-Huifed400a2014-08-06 16:07:26 -07001794 NULL);
Hugh Dickinsb065b432012-07-11 14:02:48 -07001795 radix_tree_preload_end();
1796 }
1797 if (error) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001798 mem_cgroup_cancel_charge(page, memcg,
1799 PageTransHuge(page));
1800 goto unacct;
Hugh Dickinsb065b432012-07-11 14:02:48 -07001801 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001802 mem_cgroup_commit_charge(page, memcg, false,
1803 PageTransHuge(page));
Hugh Dickins54af60422011-08-03 16:21:24 -07001804 lru_cache_add_anon(page);
1805
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001806 spin_lock_irq(&info->lock);
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001807 info->alloced += 1 << compound_order(page);
1808 inode->i_blocks += BLOCKS_PER_PAGE << compound_order(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001809 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001810 spin_unlock_irq(&info->lock);
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001811 alloced = true;
Hugh Dickins54af60422011-08-03 16:21:24 -07001812
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07001813 if (PageTransHuge(page) &&
1814 DIV_ROUND_UP(i_size_read(inode), PAGE_SIZE) <
1815 hindex + HPAGE_PMD_NR - 1) {
1816 /*
1817 * Part of the huge page is beyond i_size: subject
1818 * to shrink under memory pressure.
1819 */
1820 spin_lock(&sbinfo->shrinklist_lock);
1821 if (list_empty(&info->shrinklist)) {
1822 list_add_tail(&info->shrinklist,
1823 &sbinfo->shrinklist);
1824 sbinfo->shrinklist_len++;
1825 }
1826 spin_unlock(&sbinfo->shrinklist_lock);
1827 }
1828
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001829 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07001830 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1831 */
1832 if (sgp == SGP_FALLOC)
1833 sgp = SGP_WRITE;
1834clear:
1835 /*
1836 * Let SGP_WRITE caller clear ends if write does not fill page;
1837 * but SGP_FALLOC on a page fallocated earlier must initialize
1838 * it now, lest undo on failure cancel our earlier guarantee.
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001839 */
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001840 if (sgp != SGP_WRITE && !PageUptodate(page)) {
1841 struct page *head = compound_head(page);
1842 int i;
1843
1844 for (i = 0; i < (1 << compound_order(head)); i++) {
1845 clear_highpage(head + i);
1846 flush_dcache_page(head + i);
1847 }
1848 SetPageUptodate(head);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07001849 }
Hugh Dickins59a16ea2011-05-11 15:13:38 -07001850 }
Hugh Dickinsbde05d12012-05-29 15:06:38 -07001851
Hugh Dickins54af60422011-08-03 16:21:24 -07001852 /* Perhaps the file has been truncated since we checked */
Hugh Dickins75edd342016-05-19 17:12:44 -07001853 if (sgp <= SGP_CACHE &&
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001854 ((loff_t)index << PAGE_SHIFT) >= i_size_read(inode)) {
Hugh Dickins267a4c72015-12-11 13:40:55 -08001855 if (alloced) {
1856 ClearPageDirty(page);
1857 delete_from_page_cache(page);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001858 spin_lock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001859 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001860 spin_unlock_irq(&info->lock);
Hugh Dickins267a4c72015-12-11 13:40:55 -08001861 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001862 error = -EINVAL;
Hugh Dickins267a4c72015-12-11 13:40:55 -08001863 goto unlock;
Shaohua Liff36b8012010-08-09 17:19:06 -07001864 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001865 *pagep = page + index - hindex;
Hugh Dickins54af60422011-08-03 16:21:24 -07001866 return 0;
Nick Piggind00806b2007-07-19 01:46:57 -07001867
Nick Piggind0217ac2007-07-19 01:47:03 -07001868 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07001869 * Error recovery.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 */
Hugh Dickins54af60422011-08-03 16:21:24 -07001871unacct:
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07001872 if (sbinfo->max_blocks)
1873 percpu_counter_sub(&sbinfo->used_blocks,
1874 1 << compound_order(page));
1875 shmem_unacct_blocks(info->flags, 1 << compound_order(page));
1876
1877 if (PageTransHuge(page)) {
1878 unlock_page(page);
1879 put_page(page);
1880 goto alloc_nohuge;
1881 }
Hugh Dickins54af60422011-08-03 16:21:24 -07001882failed:
Hugh Dickins267a4c72015-12-11 13:40:55 -08001883 if (swap.val && !shmem_confirm_swap(mapping, index, swap))
Hugh Dickinsd1899222012-07-11 14:02:47 -07001884 error = -EEXIST;
1885unlock:
Hugh Dickins27ab7002011-07-25 17:12:36 -07001886 if (page) {
Hugh Dickins54af60422011-08-03 16:21:24 -07001887 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001888 put_page(page);
Hugh Dickins54af60422011-08-03 16:21:24 -07001889 }
1890 if (error == -ENOSPC && !once++) {
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001891 spin_lock_irq(&info->lock);
Hugh Dickins54af60422011-08-03 16:21:24 -07001892 shmem_recalc_inode(inode);
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07001893 spin_unlock_irq(&info->lock);
Hugh Dickins27ab7002011-07-25 17:12:36 -07001894 goto repeat;
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001895 }
Hugh Dickinsd1899222012-07-11 14:02:47 -07001896 if (error == -EEXIST) /* from above or from radix_tree_insert */
Hugh Dickins54af60422011-08-03 16:21:24 -07001897 goto repeat;
1898 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899}
1900
Linus Torvalds10d20bd2016-12-05 12:10:29 -08001901/*
1902 * This is like autoremove_wake_function, but it removes the wait queue
1903 * entry unconditionally - even if something else had already woken the
1904 * target.
1905 */
Ingo Molnarac6424b2017-06-20 12:06:13 +02001906static int synchronous_wake_function(wait_queue_entry_t *wait, unsigned mode, int sync, void *key)
Linus Torvalds10d20bd2016-12-05 12:10:29 -08001907{
1908 int ret = default_wake_function(wait, mode, sync, key);
Ingo Molnar2055da92017-06-20 12:06:46 +02001909 list_del_init(&wait->entry);
Linus Torvalds10d20bd2016-12-05 12:10:29 -08001910 return ret;
1911}
1912
Dave Jiang11bac802017-02-24 14:56:41 -08001913static int shmem_fault(struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914{
Dave Jiang11bac802017-02-24 14:56:41 -08001915 struct vm_area_struct *vma = vmf->vma;
Al Viro496ad9a2013-01-23 17:07:38 -05001916 struct inode *inode = file_inode(vma->vm_file);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07001917 gfp_t gfp = mapping_gfp_mask(inode->i_mapping);
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001918 enum sgp_type sgp;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919 int error;
Hugh Dickins68da9f02011-07-25 17:12:34 -07001920 int ret = VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001921
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001922 /*
1923 * Trinity finds that probing a hole which tmpfs is punching can
1924 * prevent the hole-punch from ever completing: which in turn
1925 * locks writers out with its hold on i_mutex. So refrain from
Hugh Dickins8e205f72014-07-23 14:00:10 -07001926 * faulting pages into the hole while it's being punched. Although
1927 * shmem_undo_range() does remove the additions, it may be unable to
1928 * keep up, as each new page needs its own unmap_mapping_range() call,
1929 * and the i_mmap tree grows ever slower to scan if new vmas are added.
1930 *
1931 * It does not matter if we sometimes reach this check just before the
1932 * hole-punch begins, so that one fault then races with the punch:
1933 * we just need to make racing faults a rare case.
1934 *
1935 * The implementation below would be much simpler if we just used a
1936 * standard mutex or completion: but we cannot take i_mutex in fault,
1937 * and bloating every shmem inode for this unlikely case would be sad.
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001938 */
1939 if (unlikely(inode->i_private)) {
1940 struct shmem_falloc *shmem_falloc;
1941
1942 spin_lock(&inode->i_lock);
1943 shmem_falloc = inode->i_private;
Hugh Dickins8e205f72014-07-23 14:00:10 -07001944 if (shmem_falloc &&
1945 shmem_falloc->waitq &&
1946 vmf->pgoff >= shmem_falloc->start &&
1947 vmf->pgoff < shmem_falloc->next) {
1948 wait_queue_head_t *shmem_falloc_waitq;
Linus Torvalds10d20bd2016-12-05 12:10:29 -08001949 DEFINE_WAIT_FUNC(shmem_fault_wait, synchronous_wake_function);
Hugh Dickins8e205f72014-07-23 14:00:10 -07001950
1951 ret = VM_FAULT_NOPAGE;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001952 if ((vmf->flags & FAULT_FLAG_ALLOW_RETRY) &&
1953 !(vmf->flags & FAULT_FLAG_RETRY_NOWAIT)) {
Hugh Dickins8e205f72014-07-23 14:00:10 -07001954 /* It's polite to up mmap_sem if we can */
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001955 up_read(&vma->vm_mm->mmap_sem);
Hugh Dickins8e205f72014-07-23 14:00:10 -07001956 ret = VM_FAULT_RETRY;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001957 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001958
1959 shmem_falloc_waitq = shmem_falloc->waitq;
1960 prepare_to_wait(shmem_falloc_waitq, &shmem_fault_wait,
1961 TASK_UNINTERRUPTIBLE);
1962 spin_unlock(&inode->i_lock);
1963 schedule();
1964
1965 /*
1966 * shmem_falloc_waitq points into the shmem_fallocate()
1967 * stack of the hole-punching task: shmem_falloc_waitq
1968 * is usually invalid by the time we reach here, but
1969 * finish_wait() does not dereference it in that case;
1970 * though i_lock needed lest racing with wake_up_all().
1971 */
1972 spin_lock(&inode->i_lock);
1973 finish_wait(shmem_falloc_waitq, &shmem_fault_wait);
1974 spin_unlock(&inode->i_lock);
1975 return ret;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001976 }
Hugh Dickins8e205f72014-07-23 14:00:10 -07001977 spin_unlock(&inode->i_lock);
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07001978 }
1979
Kirill A. Shutemov657e3032016-07-26 15:26:21 -07001980 sgp = SGP_CACHE;
1981 if (vma->vm_flags & VM_HUGEPAGE)
1982 sgp = SGP_HUGE;
1983 else if (vma->vm_flags & VM_NOHUGEPAGE)
1984 sgp = SGP_NOHUGE;
1985
1986 error = shmem_getpage_gfp(inode, vmf->pgoff, &vmf->page, sgp,
Mike Rapoportcfda0522017-02-22 15:43:37 -08001987 gfp, vma, vmf, &ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988 if (error)
1989 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Hugh Dickins68da9f02011-07-25 17:12:34 -07001990 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001991}
1992
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07001993unsigned long shmem_get_unmapped_area(struct file *file,
1994 unsigned long uaddr, unsigned long len,
1995 unsigned long pgoff, unsigned long flags)
1996{
1997 unsigned long (*get_area)(struct file *,
1998 unsigned long, unsigned long, unsigned long, unsigned long);
1999 unsigned long addr;
2000 unsigned long offset;
2001 unsigned long inflated_len;
2002 unsigned long inflated_addr;
2003 unsigned long inflated_offset;
2004
2005 if (len > TASK_SIZE)
2006 return -ENOMEM;
2007
2008 get_area = current->mm->get_unmapped_area;
2009 addr = get_area(file, uaddr, len, pgoff, flags);
2010
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07002011 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE))
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002012 return addr;
2013 if (IS_ERR_VALUE(addr))
2014 return addr;
2015 if (addr & ~PAGE_MASK)
2016 return addr;
2017 if (addr > TASK_SIZE - len)
2018 return addr;
2019
2020 if (shmem_huge == SHMEM_HUGE_DENY)
2021 return addr;
2022 if (len < HPAGE_PMD_SIZE)
2023 return addr;
2024 if (flags & MAP_FIXED)
2025 return addr;
2026 /*
2027 * Our priority is to support MAP_SHARED mapped hugely;
2028 * and support MAP_PRIVATE mapped hugely too, until it is COWed.
2029 * But if caller specified an address hint, respect that as before.
2030 */
2031 if (uaddr)
2032 return addr;
2033
2034 if (shmem_huge != SHMEM_HUGE_FORCE) {
2035 struct super_block *sb;
2036
2037 if (file) {
2038 VM_BUG_ON(file->f_op != &shmem_file_operations);
2039 sb = file_inode(file)->i_sb;
2040 } else {
2041 /*
2042 * Called directly from mm/mmap.c, or drivers/char/mem.c
2043 * for "/dev/zero", to create a shared anonymous object.
2044 */
2045 if (IS_ERR(shm_mnt))
2046 return addr;
2047 sb = shm_mnt->mnt_sb;
2048 }
Toshi Kani3089bf62016-09-23 20:21:56 -07002049 if (SHMEM_SB(sb)->huge == SHMEM_HUGE_NEVER)
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07002050 return addr;
2051 }
2052
2053 offset = (pgoff << PAGE_SHIFT) & (HPAGE_PMD_SIZE-1);
2054 if (offset && offset + len < 2 * HPAGE_PMD_SIZE)
2055 return addr;
2056 if ((addr & (HPAGE_PMD_SIZE-1)) == offset)
2057 return addr;
2058
2059 inflated_len = len + HPAGE_PMD_SIZE - PAGE_SIZE;
2060 if (inflated_len > TASK_SIZE)
2061 return addr;
2062 if (inflated_len < len)
2063 return addr;
2064
2065 inflated_addr = get_area(NULL, 0, inflated_len, 0, flags);
2066 if (IS_ERR_VALUE(inflated_addr))
2067 return addr;
2068 if (inflated_addr & ~PAGE_MASK)
2069 return addr;
2070
2071 inflated_offset = inflated_addr & (HPAGE_PMD_SIZE-1);
2072 inflated_addr += offset - inflated_offset;
2073 if (inflated_offset > offset)
2074 inflated_addr += HPAGE_PMD_SIZE;
2075
2076 if (inflated_addr > TASK_SIZE - len)
2077 return addr;
2078 return inflated_addr;
2079}
2080
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081#ifdef CONFIG_NUMA
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002082static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083{
Al Viro496ad9a2013-01-23 17:07:38 -05002084 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002085 return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086}
2087
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07002088static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
2089 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002090{
Al Viro496ad9a2013-01-23 17:07:38 -05002091 struct inode *inode = file_inode(vma->vm_file);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002092 pgoff_t index;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002094 index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
2095 return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096}
2097#endif
2098
2099int shmem_lock(struct file *file, int lock, struct user_struct *user)
2100{
Al Viro496ad9a2013-01-23 17:07:38 -05002101 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102 struct shmem_inode_info *info = SHMEM_I(inode);
2103 int retval = -ENOMEM;
2104
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002105 spin_lock_irq(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002106 if (lock && !(info->flags & VM_LOCKED)) {
2107 if (!user_shm_lock(inode->i_size, user))
2108 goto out_nomem;
2109 info->flags |= VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002110 mapping_set_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 }
2112 if (!lock && (info->flags & VM_LOCKED) && user) {
2113 user_shm_unlock(inode->i_size, user);
2114 info->flags &= ~VM_LOCKED;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002115 mapping_clear_unevictable(file->f_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 }
2117 retval = 0;
Lee Schermerhorn89e004ea2008-10-18 20:26:43 -07002118
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119out_nomem:
Kirill A. Shutemov4595ef82016-07-26 15:26:29 -07002120 spin_unlock_irq(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 return retval;
2122}
2123
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08002124static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002125{
2126 file_accessed(file);
2127 vma->vm_ops = &shmem_vm_ops;
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07002128 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07002129 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
2130 (vma->vm_end & HPAGE_PMD_MASK)) {
2131 khugepaged_enter(vma, vma->vm_flags);
2132 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 return 0;
2134}
2135
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002136static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
Al Viro09208d12011-07-26 03:15:03 -04002137 umode_t mode, dev_t dev, unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138{
2139 struct inode *inode;
2140 struct shmem_inode_info *info;
2141 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2142
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002143 if (shmem_reserve_inode(sb))
2144 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145
2146 inode = new_inode(sb);
2147 if (inode) {
Christoph Hellwig85fe4022010-10-23 11:19:54 -04002148 inode->i_ino = get_next_ino();
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002149 inode_init_owner(inode, dir, mode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002150 inode->i_blocks = 0;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002151 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002152 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153 info = SHMEM_I(inode);
2154 memset(info, 0, (char *)inode - (char *)info);
2155 spin_lock_init(&info->lock);
David Herrmann40e041a2014-08-08 14:25:27 -07002156 info->seals = F_SEAL_SEAL;
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002157 info->flags = flags & VM_NORESERVE;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07002158 INIT_LIST_HEAD(&info->shrinklist);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002159 INIT_LIST_HEAD(&info->swaplist);
Aristeu Rozanski38f38652012-08-23 16:53:28 -04002160 simple_xattrs_init(&info->xattrs);
Al Viro72c04902009-06-24 16:58:48 -04002161 cache_no_acl(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002162
2163 switch (mode & S_IFMT) {
2164 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002165 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166 init_special_inode(inode, mode, dev);
2167 break;
2168 case S_IFREG:
Hugh Dickins14fcc232008-07-28 15:46:19 -07002169 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 inode->i_op = &shmem_inode_operations;
2171 inode->i_fop = &shmem_file_operations;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002172 mpol_shared_policy_init(&info->policy,
2173 shmem_get_sbmpol(sbinfo));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002174 break;
2175 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07002176 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 /* Some things misbehave if size == 0 on a directory */
2178 inode->i_size = 2 * BOGO_DIRENT_SIZE;
2179 inode->i_op = &shmem_dir_inode_operations;
2180 inode->i_fop = &simple_dir_operations;
2181 break;
2182 case S_IFLNK:
2183 /*
2184 * Must not load anything in the rbtree,
2185 * mpol_free_shared_policy will not be called.
2186 */
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07002187 mpol_shared_policy_init(&info->policy, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 break;
2189 }
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08002190 } else
2191 shmem_free_inode(sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 return inode;
2193}
2194
Johannes Weiner0cd61442014-04-03 14:47:46 -07002195bool shmem_mapping(struct address_space *mapping)
2196{
Hugh Dickinsf8005452017-02-22 15:41:42 -08002197 return mapping->a_ops == &shmem_aops;
Johannes Weiner0cd61442014-04-03 14:47:46 -07002198}
2199
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002200int shmem_mcopy_atomic_pte(struct mm_struct *dst_mm,
2201 pmd_t *dst_pmd,
2202 struct vm_area_struct *dst_vma,
2203 unsigned long dst_addr,
2204 unsigned long src_addr,
2205 struct page **pagep)
2206{
2207 struct inode *inode = file_inode(dst_vma->vm_file);
2208 struct shmem_inode_info *info = SHMEM_I(inode);
2209 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
2210 struct address_space *mapping = inode->i_mapping;
2211 gfp_t gfp = mapping_gfp_mask(mapping);
2212 pgoff_t pgoff = linear_page_index(dst_vma, dst_addr);
2213 struct mem_cgroup *memcg;
2214 spinlock_t *ptl;
2215 void *page_kaddr;
2216 struct page *page;
2217 pte_t _dst_pte, *dst_pte;
2218 int ret;
2219
Andrea Arcangelicb658a42017-02-22 15:43:55 -08002220 ret = -ENOMEM;
2221 if (shmem_acct_block(info->flags, 1))
2222 goto out;
2223 if (sbinfo->max_blocks) {
2224 if (percpu_counter_compare(&sbinfo->used_blocks,
2225 sbinfo->max_blocks) >= 0)
2226 goto out_unacct_blocks;
2227 percpu_counter_inc(&sbinfo->used_blocks);
2228 }
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002229
Andrea Arcangelicb658a42017-02-22 15:43:55 -08002230 if (!*pagep) {
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002231 page = shmem_alloc_page(gfp, info, pgoff);
2232 if (!page)
2233 goto out_dec_used_blocks;
2234
2235 page_kaddr = kmap_atomic(page);
2236 ret = copy_from_user(page_kaddr, (const void __user *)src_addr,
2237 PAGE_SIZE);
2238 kunmap_atomic(page_kaddr);
2239
2240 /* fallback to copy_from_user outside mmap_sem */
2241 if (unlikely(ret)) {
2242 *pagep = page;
Andrea Arcangelicb658a42017-02-22 15:43:55 -08002243 if (sbinfo->max_blocks)
2244 percpu_counter_add(&sbinfo->used_blocks, -1);
2245 shmem_unacct_blocks(info->flags, 1);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002246 /* don't free the page */
2247 return -EFAULT;
2248 }
2249 } else {
2250 page = *pagep;
2251 *pagep = NULL;
2252 }
2253
Andrea Arcangeli9cc90c62017-02-22 15:43:49 -08002254 VM_BUG_ON(PageLocked(page) || PageSwapBacked(page));
2255 __SetPageLocked(page);
2256 __SetPageSwapBacked(page);
Andrea Arcangelia425d352017-02-22 15:43:52 -08002257 __SetPageUptodate(page);
Andrea Arcangeli9cc90c62017-02-22 15:43:49 -08002258
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002259 ret = mem_cgroup_try_charge(page, dst_mm, gfp, &memcg, false);
2260 if (ret)
2261 goto out_release;
2262
2263 ret = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK);
2264 if (!ret) {
2265 ret = shmem_add_to_page_cache(page, mapping, pgoff, NULL);
2266 radix_tree_preload_end();
2267 }
2268 if (ret)
2269 goto out_release_uncharge;
2270
2271 mem_cgroup_commit_charge(page, memcg, false, false);
2272
2273 _dst_pte = mk_pte(page, dst_vma->vm_page_prot);
2274 if (dst_vma->vm_flags & VM_WRITE)
2275 _dst_pte = pte_mkwrite(pte_mkdirty(_dst_pte));
2276
2277 ret = -EEXIST;
2278 dst_pte = pte_offset_map_lock(dst_mm, dst_pmd, dst_addr, &ptl);
2279 if (!pte_none(*dst_pte))
2280 goto out_release_uncharge_unlock;
2281
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002282 lru_cache_add_anon(page);
2283
2284 spin_lock(&info->lock);
2285 info->alloced++;
2286 inode->i_blocks += BLOCKS_PER_PAGE;
2287 shmem_recalc_inode(inode);
2288 spin_unlock(&info->lock);
2289
2290 inc_mm_counter(dst_mm, mm_counter_file(page));
2291 page_add_file_rmap(page, false);
2292 set_pte_at(dst_mm, dst_addr, dst_pte, _dst_pte);
2293
2294 /* No need to invalidate - it was non-present before */
2295 update_mmu_cache(dst_vma, dst_addr, dst_pte);
2296 unlock_page(page);
2297 pte_unmap_unlock(dst_pte, ptl);
2298 ret = 0;
2299out:
2300 return ret;
2301out_release_uncharge_unlock:
2302 pte_unmap_unlock(dst_pte, ptl);
2303out_release_uncharge:
2304 mem_cgroup_cancel_charge(page, memcg, false);
2305out_release:
Andrea Arcangeli9cc90c62017-02-22 15:43:49 -08002306 unlock_page(page);
Mike Rapoport4c27fe42017-02-22 15:43:25 -08002307 put_page(page);
2308out_dec_used_blocks:
2309 if (sbinfo->max_blocks)
2310 percpu_counter_add(&sbinfo->used_blocks, -1);
2311out_unacct_blocks:
2312 shmem_unacct_blocks(info->flags, 1);
2313 goto out;
2314}
2315
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002317static const struct inode_operations shmem_symlink_inode_operations;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07002318static const struct inode_operations shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002320#ifdef CONFIG_TMPFS_XATTR
2321static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
2322#else
2323#define shmem_initxattrs NULL
2324#endif
2325
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326static int
Nick Piggin800d15a2007-10-16 01:25:03 -07002327shmem_write_begin(struct file *file, struct address_space *mapping,
2328 loff_t pos, unsigned len, unsigned flags,
2329 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002330{
Nick Piggin800d15a2007-10-16 01:25:03 -07002331 struct inode *inode = mapping->host;
David Herrmann40e041a2014-08-08 14:25:27 -07002332 struct shmem_inode_info *info = SHMEM_I(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002333 pgoff_t index = pos >> PAGE_SHIFT;
David Herrmann40e041a2014-08-08 14:25:27 -07002334
2335 /* i_mutex is held by caller */
Steven Rostedt (VMware)3f472cc2017-02-24 14:59:10 -08002336 if (unlikely(info->seals & (F_SEAL_WRITE | F_SEAL_GROW))) {
David Herrmann40e041a2014-08-08 14:25:27 -07002337 if (info->seals & F_SEAL_WRITE)
2338 return -EPERM;
2339 if ((info->seals & F_SEAL_GROW) && pos + len > inode->i_size)
2340 return -EPERM;
2341 }
2342
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002343 return shmem_getpage(inode, index, pagep, SGP_WRITE);
Nick Piggin800d15a2007-10-16 01:25:03 -07002344}
2345
2346static int
2347shmem_write_end(struct file *file, struct address_space *mapping,
2348 loff_t pos, unsigned len, unsigned copied,
2349 struct page *page, void *fsdata)
2350{
2351 struct inode *inode = mapping->host;
2352
Hugh Dickinsd3602442008-02-04 22:28:44 -08002353 if (pos + copied > inode->i_size)
2354 i_size_write(inode, pos + copied);
2355
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002356 if (!PageUptodate(page)) {
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002357 struct page *head = compound_head(page);
2358 if (PageTransCompound(page)) {
2359 int i;
2360
2361 for (i = 0; i < HPAGE_PMD_NR; i++) {
2362 if (head + i == page)
2363 continue;
2364 clear_highpage(head + i);
2365 flush_dcache_page(head + i);
2366 }
2367 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002368 if (copied < PAGE_SIZE) {
2369 unsigned from = pos & (PAGE_SIZE - 1);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002370 zero_user_segments(page, 0, from,
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002371 from + copied, PAGE_SIZE);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002372 }
Kirill A. Shutemov800d8c62016-07-26 15:26:18 -07002373 SetPageUptodate(head);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07002374 }
Nick Piggin800d15a2007-10-16 01:25:03 -07002375 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02002376 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002377 put_page(page);
Nick Piggin800d15a2007-10-16 01:25:03 -07002378
Nick Piggin800d15a2007-10-16 01:25:03 -07002379 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380}
2381
Al Viro2ba5bbe2014-04-02 20:00:02 -04002382static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383{
Al Viro6e58e792014-02-03 17:07:03 -05002384 struct file *file = iocb->ki_filp;
2385 struct inode *inode = file_inode(file);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386 struct address_space *mapping = inode->i_mapping;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002387 pgoff_t index;
2388 unsigned long offset;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002389 enum sgp_type sgp = SGP_READ;
Geert Uytterhoevenf7c1d072014-04-13 20:46:22 +02002390 int error = 0;
Al Virocb66a7a2014-03-04 15:24:06 -05002391 ssize_t retval = 0;
Al Viro6e58e792014-02-03 17:07:03 -05002392 loff_t *ppos = &iocb->ki_pos;
Hugh Dickinsa0ee5ec2008-02-04 22:28:51 -08002393
2394 /*
2395 * Might this read be for a stacking filesystem? Then when reading
2396 * holes of a sparse file, we actually need to allocate those pages,
2397 * and even mark them dirty, so it cannot exceed the max_blocks limit.
2398 */
Al Viro777eda22014-12-17 04:46:46 -05002399 if (!iter_is_iovec(to))
Hugh Dickins75edd342016-05-19 17:12:44 -07002400 sgp = SGP_CACHE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002402 index = *ppos >> PAGE_SHIFT;
2403 offset = *ppos & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404
2405 for (;;) {
2406 struct page *page = NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002407 pgoff_t end_index;
2408 unsigned long nr, ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409 loff_t i_size = i_size_read(inode);
2410
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002411 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002412 if (index > end_index)
2413 break;
2414 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002415 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416 if (nr <= offset)
2417 break;
2418 }
2419
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002420 error = shmem_getpage(inode, index, &page, sgp);
Al Viro6e58e792014-02-03 17:07:03 -05002421 if (error) {
2422 if (error == -EINVAL)
2423 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002424 break;
2425 }
Hugh Dickins75edd342016-05-19 17:12:44 -07002426 if (page) {
2427 if (sgp == SGP_CACHE)
2428 set_page_dirty(page);
Hugh Dickinsd3602442008-02-04 22:28:44 -08002429 unlock_page(page);
Hugh Dickins75edd342016-05-19 17:12:44 -07002430 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431
2432 /*
2433 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08002434 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002436 nr = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002438 end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439 if (index == end_index) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002440 nr = i_size & ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441 if (nr <= offset) {
2442 if (page)
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002443 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444 break;
2445 }
2446 }
2447 nr -= offset;
2448
2449 if (page) {
2450 /*
2451 * If users can be writing to this page using arbitrary
2452 * virtual addresses, take care about potential aliasing
2453 * before reading the page on the kernel side.
2454 */
2455 if (mapping_writably_mapped(mapping))
2456 flush_dcache_page(page);
2457 /*
2458 * Mark the page accessed if we read the beginning.
2459 */
2460 if (!offset)
2461 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002462 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 page = ZERO_PAGE(0);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002464 get_page(page);
Nick Pigginb5810032005-10-29 18:16:12 -07002465 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466
2467 /*
2468 * Ok, we have the page, and it's up-to-date, so
2469 * now we can copy it to user space...
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470 */
Al Viro2ba5bbe2014-04-02 20:00:02 -04002471 ret = copy_page_to_iter(page, offset, nr, to);
Al Viro6e58e792014-02-03 17:07:03 -05002472 retval += ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002473 offset += ret;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002474 index += offset >> PAGE_SHIFT;
2475 offset &= ~PAGE_MASK;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002477 put_page(page);
Al Viro2ba5bbe2014-04-02 20:00:02 -04002478 if (!iov_iter_count(to))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479 break;
Al Viro6e58e792014-02-03 17:07:03 -05002480 if (ret < nr) {
2481 error = -EFAULT;
2482 break;
2483 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484 cond_resched();
2485 }
2486
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002487 *ppos = ((loff_t) index << PAGE_SHIFT) + offset;
Al Viro6e58e792014-02-03 17:07:03 -05002488 file_accessed(file);
2489 return retval ? retval : error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490}
2491
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002492/*
2493 * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
2494 */
2495static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
Andrew Morton965c8e52012-12-17 15:59:39 -08002496 pgoff_t index, pgoff_t end, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002497{
2498 struct page *page;
2499 struct pagevec pvec;
2500 pgoff_t indices[PAGEVEC_SIZE];
2501 bool done = false;
2502 int i;
2503
2504 pagevec_init(&pvec, 0);
2505 pvec.nr = 1; /* start small: we may be there already */
2506 while (!done) {
Johannes Weiner0cd61442014-04-03 14:47:46 -07002507 pvec.nr = find_get_entries(mapping, index,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002508 pvec.nr, pvec.pages, indices);
2509 if (!pvec.nr) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002510 if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002511 index = end;
2512 break;
2513 }
2514 for (i = 0; i < pvec.nr; i++, index++) {
2515 if (index < indices[i]) {
Andrew Morton965c8e52012-12-17 15:59:39 -08002516 if (whence == SEEK_HOLE) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002517 done = true;
2518 break;
2519 }
2520 index = indices[i];
2521 }
2522 page = pvec.pages[i];
2523 if (page && !radix_tree_exceptional_entry(page)) {
2524 if (!PageUptodate(page))
2525 page = NULL;
2526 }
2527 if (index >= end ||
Andrew Morton965c8e52012-12-17 15:59:39 -08002528 (page && whence == SEEK_DATA) ||
2529 (!page && whence == SEEK_HOLE)) {
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002530 done = true;
2531 break;
2532 }
2533 }
Johannes Weiner0cd61442014-04-03 14:47:46 -07002534 pagevec_remove_exceptionals(&pvec);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002535 pagevec_release(&pvec);
2536 pvec.nr = PAGEVEC_SIZE;
2537 cond_resched();
2538 }
2539 return index;
2540}
2541
Andrew Morton965c8e52012-12-17 15:59:39 -08002542static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002543{
2544 struct address_space *mapping = file->f_mapping;
2545 struct inode *inode = mapping->host;
2546 pgoff_t start, end;
2547 loff_t new_offset;
2548
Andrew Morton965c8e52012-12-17 15:59:39 -08002549 if (whence != SEEK_DATA && whence != SEEK_HOLE)
2550 return generic_file_llseek_size(file, offset, whence,
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002551 MAX_LFS_FILESIZE, i_size_read(inode));
Al Viro59551022016-01-22 15:40:57 -05002552 inode_lock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002553 /* We're holding i_mutex so we can access i_size directly */
2554
2555 if (offset < 0)
2556 offset = -EINVAL;
2557 else if (offset >= inode->i_size)
2558 offset = -ENXIO;
2559 else {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002560 start = offset >> PAGE_SHIFT;
2561 end = (inode->i_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
Andrew Morton965c8e52012-12-17 15:59:39 -08002562 new_offset = shmem_seek_hole_data(mapping, start, end, whence);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002563 new_offset <<= PAGE_SHIFT;
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002564 if (new_offset > offset) {
2565 if (new_offset < inode->i_size)
2566 offset = new_offset;
Andrew Morton965c8e52012-12-17 15:59:39 -08002567 else if (whence == SEEK_DATA)
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002568 offset = -ENXIO;
2569 else
2570 offset = inode->i_size;
2571 }
2572 }
2573
Hugh Dickins387aae62013-08-04 11:30:25 -07002574 if (offset >= 0)
2575 offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
Al Viro59551022016-01-22 15:40:57 -05002576 inode_unlock(inode);
Hugh Dickins220f2ac2012-12-12 13:52:21 -08002577 return offset;
2578}
2579
David Herrmann05f65b52014-08-08 14:25:36 -07002580/*
2581 * We need a tag: a new tag would expand every radix_tree_node by 8 bytes,
2582 * so reuse a tag which we firmly believe is never set or cleared on shmem.
2583 */
2584#define SHMEM_TAG_PINNED PAGECACHE_TAG_TOWRITE
2585#define LAST_SCAN 4 /* about 150ms max */
2586
2587static void shmem_tag_pins(struct address_space *mapping)
2588{
2589 struct radix_tree_iter iter;
2590 void **slot;
2591 pgoff_t start;
2592 struct page *page;
2593
2594 lru_add_drain();
2595 start = 0;
2596 rcu_read_lock();
2597
David Herrmann05f65b52014-08-08 14:25:36 -07002598 radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
2599 page = radix_tree_deref_slot(slot);
2600 if (!page || radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002601 if (radix_tree_deref_retry(page)) {
2602 slot = radix_tree_iter_retry(&iter);
2603 continue;
2604 }
David Herrmann05f65b52014-08-08 14:25:36 -07002605 } else if (page_count(page) - page_mapcount(page) > 1) {
2606 spin_lock_irq(&mapping->tree_lock);
2607 radix_tree_tag_set(&mapping->page_tree, iter.index,
2608 SHMEM_TAG_PINNED);
2609 spin_unlock_irq(&mapping->tree_lock);
2610 }
2611
2612 if (need_resched()) {
Matthew Wilcox148deab2016-12-14 15:08:49 -08002613 slot = radix_tree_iter_resume(slot, &iter);
David Herrmann05f65b52014-08-08 14:25:36 -07002614 cond_resched_rcu();
David Herrmann05f65b52014-08-08 14:25:36 -07002615 }
2616 }
2617 rcu_read_unlock();
2618}
2619
2620/*
2621 * Setting SEAL_WRITE requires us to verify there's no pending writer. However,
2622 * via get_user_pages(), drivers might have some pending I/O without any active
2623 * user-space mappings (eg., direct-IO, AIO). Therefore, we look at all pages
2624 * and see whether it has an elevated ref-count. If so, we tag them and wait for
2625 * them to be dropped.
2626 * The caller must guarantee that no new user will acquire writable references
2627 * to those pages to avoid races.
2628 */
David Herrmann40e041a2014-08-08 14:25:27 -07002629static int shmem_wait_for_pins(struct address_space *mapping)
2630{
David Herrmann05f65b52014-08-08 14:25:36 -07002631 struct radix_tree_iter iter;
2632 void **slot;
2633 pgoff_t start;
2634 struct page *page;
2635 int error, scan;
2636
2637 shmem_tag_pins(mapping);
2638
2639 error = 0;
2640 for (scan = 0; scan <= LAST_SCAN; scan++) {
2641 if (!radix_tree_tagged(&mapping->page_tree, SHMEM_TAG_PINNED))
2642 break;
2643
2644 if (!scan)
2645 lru_add_drain_all();
2646 else if (schedule_timeout_killable((HZ << scan) / 200))
2647 scan = LAST_SCAN;
2648
2649 start = 0;
2650 rcu_read_lock();
David Herrmann05f65b52014-08-08 14:25:36 -07002651 radix_tree_for_each_tagged(slot, &mapping->page_tree, &iter,
2652 start, SHMEM_TAG_PINNED) {
2653
2654 page = radix_tree_deref_slot(slot);
2655 if (radix_tree_exception(page)) {
Matthew Wilcox2cf938a2016-03-17 14:22:03 -07002656 if (radix_tree_deref_retry(page)) {
2657 slot = radix_tree_iter_retry(&iter);
2658 continue;
2659 }
David Herrmann05f65b52014-08-08 14:25:36 -07002660
2661 page = NULL;
2662 }
2663
2664 if (page &&
2665 page_count(page) - page_mapcount(page) != 1) {
2666 if (scan < LAST_SCAN)
2667 goto continue_resched;
2668
2669 /*
2670 * On the last scan, we clean up all those tags
2671 * we inserted; but make a note that we still
2672 * found pages pinned.
2673 */
2674 error = -EBUSY;
2675 }
2676
2677 spin_lock_irq(&mapping->tree_lock);
2678 radix_tree_tag_clear(&mapping->page_tree,
2679 iter.index, SHMEM_TAG_PINNED);
2680 spin_unlock_irq(&mapping->tree_lock);
2681continue_resched:
2682 if (need_resched()) {
Matthew Wilcox148deab2016-12-14 15:08:49 -08002683 slot = radix_tree_iter_resume(slot, &iter);
David Herrmann05f65b52014-08-08 14:25:36 -07002684 cond_resched_rcu();
David Herrmann05f65b52014-08-08 14:25:36 -07002685 }
2686 }
2687 rcu_read_unlock();
2688 }
2689
2690 return error;
David Herrmann40e041a2014-08-08 14:25:27 -07002691}
2692
2693#define F_ALL_SEALS (F_SEAL_SEAL | \
2694 F_SEAL_SHRINK | \
2695 F_SEAL_GROW | \
2696 F_SEAL_WRITE)
2697
2698int shmem_add_seals(struct file *file, unsigned int seals)
2699{
2700 struct inode *inode = file_inode(file);
2701 struct shmem_inode_info *info = SHMEM_I(inode);
2702 int error;
2703
2704 /*
2705 * SEALING
2706 * Sealing allows multiple parties to share a shmem-file but restrict
2707 * access to a specific subset of file operations. Seals can only be
2708 * added, but never removed. This way, mutually untrusted parties can
2709 * share common memory regions with a well-defined policy. A malicious
2710 * peer can thus never perform unwanted operations on a shared object.
2711 *
2712 * Seals are only supported on special shmem-files and always affect
2713 * the whole underlying inode. Once a seal is set, it may prevent some
2714 * kinds of access to the file. Currently, the following seals are
2715 * defined:
2716 * SEAL_SEAL: Prevent further seals from being set on this file
2717 * SEAL_SHRINK: Prevent the file from shrinking
2718 * SEAL_GROW: Prevent the file from growing
2719 * SEAL_WRITE: Prevent write access to the file
2720 *
2721 * As we don't require any trust relationship between two parties, we
2722 * must prevent seals from being removed. Therefore, sealing a file
2723 * only adds a given set of seals to the file, it never touches
2724 * existing seals. Furthermore, the "setting seals"-operation can be
2725 * sealed itself, which basically prevents any further seal from being
2726 * added.
2727 *
2728 * Semantics of sealing are only defined on volatile files. Only
2729 * anonymous shmem files support sealing. More importantly, seals are
2730 * never written to disk. Therefore, there's no plan to support it on
2731 * other file types.
2732 */
2733
2734 if (file->f_op != &shmem_file_operations)
2735 return -EINVAL;
2736 if (!(file->f_mode & FMODE_WRITE))
2737 return -EPERM;
2738 if (seals & ~(unsigned int)F_ALL_SEALS)
2739 return -EINVAL;
2740
Al Viro59551022016-01-22 15:40:57 -05002741 inode_lock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002742
2743 if (info->seals & F_SEAL_SEAL) {
2744 error = -EPERM;
2745 goto unlock;
2746 }
2747
2748 if ((seals & F_SEAL_WRITE) && !(info->seals & F_SEAL_WRITE)) {
2749 error = mapping_deny_writable(file->f_mapping);
2750 if (error)
2751 goto unlock;
2752
2753 error = shmem_wait_for_pins(file->f_mapping);
2754 if (error) {
2755 mapping_allow_writable(file->f_mapping);
2756 goto unlock;
2757 }
2758 }
2759
2760 info->seals |= seals;
2761 error = 0;
2762
2763unlock:
Al Viro59551022016-01-22 15:40:57 -05002764 inode_unlock(inode);
David Herrmann40e041a2014-08-08 14:25:27 -07002765 return error;
2766}
2767EXPORT_SYMBOL_GPL(shmem_add_seals);
2768
2769int shmem_get_seals(struct file *file)
2770{
2771 if (file->f_op != &shmem_file_operations)
2772 return -EINVAL;
2773
2774 return SHMEM_I(file_inode(file))->seals;
2775}
2776EXPORT_SYMBOL_GPL(shmem_get_seals);
2777
2778long shmem_fcntl(struct file *file, unsigned int cmd, unsigned long arg)
2779{
2780 long error;
2781
2782 switch (cmd) {
2783 case F_ADD_SEALS:
2784 /* disallow upper 32bit */
2785 if (arg > UINT_MAX)
2786 return -EINVAL;
2787
2788 error = shmem_add_seals(file, arg);
2789 break;
2790 case F_GET_SEALS:
2791 error = shmem_get_seals(file);
2792 break;
2793 default:
2794 error = -EINVAL;
2795 break;
2796 }
2797
2798 return error;
2799}
2800
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002801static long shmem_fallocate(struct file *file, int mode, loff_t offset,
2802 loff_t len)
2803{
Al Viro496ad9a2013-01-23 17:07:38 -05002804 struct inode *inode = file_inode(file);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002805 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
David Herrmann40e041a2014-08-08 14:25:27 -07002806 struct shmem_inode_info *info = SHMEM_I(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002807 struct shmem_falloc shmem_falloc;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002808 pgoff_t start, index, end;
2809 int error;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002810
Hugh Dickins13ace4d2014-06-23 13:22:03 -07002811 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
2812 return -EOPNOTSUPP;
2813
Al Viro59551022016-01-22 15:40:57 -05002814 inode_lock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002815
2816 if (mode & FALLOC_FL_PUNCH_HOLE) {
2817 struct address_space *mapping = file->f_mapping;
2818 loff_t unmap_start = round_up(offset, PAGE_SIZE);
2819 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002820 DECLARE_WAIT_QUEUE_HEAD_ONSTACK(shmem_falloc_waitq);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002821
David Herrmann40e041a2014-08-08 14:25:27 -07002822 /* protected by i_mutex */
2823 if (info->seals & F_SEAL_WRITE) {
2824 error = -EPERM;
2825 goto out;
2826 }
2827
Hugh Dickins8e205f72014-07-23 14:00:10 -07002828 shmem_falloc.waitq = &shmem_falloc_waitq;
Hugh Dickinsf00cdc62014-06-23 13:22:06 -07002829 shmem_falloc.start = unmap_start >> PAGE_SHIFT;
2830 shmem_falloc.next = (unmap_end + 1) >> PAGE_SHIFT;
2831 spin_lock(&inode->i_lock);
2832 inode->i_private = &shmem_falloc;
2833 spin_unlock(&inode->i_lock);
2834
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002835 if ((u64)unmap_end > (u64)unmap_start)
2836 unmap_mapping_range(mapping, unmap_start,
2837 1 + unmap_end - unmap_start, 0);
2838 shmem_truncate_range(inode, offset, offset + len - 1);
2839 /* No need to unmap again: hole-punching leaves COWed pages */
Hugh Dickins8e205f72014-07-23 14:00:10 -07002840
2841 spin_lock(&inode->i_lock);
2842 inode->i_private = NULL;
2843 wake_up_all(&shmem_falloc_waitq);
Ingo Molnar2055da92017-06-20 12:06:46 +02002844 WARN_ON_ONCE(!list_empty(&shmem_falloc_waitq.head));
Hugh Dickins8e205f72014-07-23 14:00:10 -07002845 spin_unlock(&inode->i_lock);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002846 error = 0;
Hugh Dickins8e205f72014-07-23 14:00:10 -07002847 goto out;
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002848 }
2849
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002850 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
2851 error = inode_newsize_ok(inode, offset + len);
2852 if (error)
2853 goto out;
2854
David Herrmann40e041a2014-08-08 14:25:27 -07002855 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
2856 error = -EPERM;
2857 goto out;
2858 }
2859
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002860 start = offset >> PAGE_SHIFT;
2861 end = (offset + len + PAGE_SIZE - 1) >> PAGE_SHIFT;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002862 /* Try to avoid a swapstorm if len is impossible to satisfy */
2863 if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
2864 error = -ENOSPC;
2865 goto out;
2866 }
2867
Hugh Dickins8e205f72014-07-23 14:00:10 -07002868 shmem_falloc.waitq = NULL;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002869 shmem_falloc.start = start;
2870 shmem_falloc.next = start;
2871 shmem_falloc.nr_falloced = 0;
2872 shmem_falloc.nr_unswapped = 0;
2873 spin_lock(&inode->i_lock);
2874 inode->i_private = &shmem_falloc;
2875 spin_unlock(&inode->i_lock);
2876
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002877 for (index = start; index < end; index++) {
2878 struct page *page;
2879
2880 /*
2881 * Good, the fallocate(2) manpage permits EINTR: we may have
2882 * been interrupted because we are using up too much memory.
2883 */
2884 if (signal_pending(current))
2885 error = -EINTR;
Hugh Dickins1aac1402012-05-29 15:06:42 -07002886 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
2887 error = -ENOMEM;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002888 else
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07002889 error = shmem_getpage(inode, index, &page, SGP_FALLOC);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002890 if (error) {
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002891 /* Remove the !PageUptodate pages we added */
Hugh Dickins7f556562016-07-10 16:46:32 -07002892 if (index > start) {
2893 shmem_undo_range(inode,
2894 (loff_t)start << PAGE_SHIFT,
2895 ((loff_t)index << PAGE_SHIFT) - 1, true);
2896 }
Hugh Dickins1aac1402012-05-29 15:06:42 -07002897 goto undone;
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002898 }
2899
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002900 /*
Hugh Dickins1aac1402012-05-29 15:06:42 -07002901 * Inform shmem_writepage() how far we have reached.
2902 * No need for lock or barrier: we have the page lock.
2903 */
2904 shmem_falloc.next++;
2905 if (!PageUptodate(page))
2906 shmem_falloc.nr_falloced++;
2907
2908 /*
Hugh Dickins1635f6a2012-05-29 15:06:42 -07002909 * If !PageUptodate, leave it that way so that freeable pages
2910 * can be recognized if we need to rollback on error later.
2911 * But set_page_dirty so that memory pressure will swap rather
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002912 * than free the pages we are allocating (and SGP_CACHE pages
2913 * might still be clean: we now need to mark those dirty too).
2914 */
2915 set_page_dirty(page);
2916 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002917 put_page(page);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002918 cond_resched();
2919 }
2920
2921 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
2922 i_size_write(inode, offset + len);
Deepa Dinamani078cd822016-09-14 07:48:04 -07002923 inode->i_ctime = current_time(inode);
Hugh Dickins1aac1402012-05-29 15:06:42 -07002924undone:
2925 spin_lock(&inode->i_lock);
2926 inode->i_private = NULL;
2927 spin_unlock(&inode->i_lock);
Hugh Dickinse2d12e22012-05-29 15:06:41 -07002928out:
Al Viro59551022016-01-22 15:40:57 -05002929 inode_unlock(inode);
Hugh Dickins83e4fa92012-05-29 15:06:40 -07002930 return error;
2931}
2932
David Howells726c3342006-06-23 02:02:58 -07002933static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934{
David Howells726c3342006-06-23 02:02:58 -07002935 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002936
2937 buf->f_type = TMPFS_MAGIC;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002938 buf->f_bsize = PAGE_SIZE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002939 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002940 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002941 buf->f_blocks = sbinfo->max_blocks;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07002942 buf->f_bavail =
2943 buf->f_bfree = sbinfo->max_blocks -
2944 percpu_counter_sum(&sbinfo->used_blocks);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002945 }
2946 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002947 buf->f_files = sbinfo->max_inodes;
2948 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 }
2950 /* else leave those fields 0 like simple_statfs */
2951 return 0;
2952}
2953
2954/*
2955 * File creation. Allocate an inode, and we're done..
2956 */
2957static int
Al Viro1a67aaf2011-07-26 01:52:52 -04002958shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002959{
Hugh Dickins0b0a0802009-02-24 20:51:52 +00002960 struct inode *inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961 int error = -ENOSPC;
2962
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03002963 inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002964 if (inode) {
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002965 error = simple_acl_create(dir, inode);
2966 if (error)
2967 goto out_iput;
Eric Paris2a7dba32011-02-01 11:05:39 -05002968 error = security_inode_init_security(inode, dir,
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04002969 &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07002970 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002971 if (error && error != -EOPNOTSUPP)
2972 goto out_iput;
Mimi Zohar37ec43c2013-04-14 09:21:47 -04002973
Al Viro718deb62009-12-16 19:35:36 -05002974 error = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07002976 dir->i_ctime = dir->i_mtime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002977 d_instantiate(dentry, inode);
2978 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979 }
2980 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002981out_iput:
2982 iput(inode);
2983 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002984}
2985
Al Viro60545d02013-06-07 01:20:27 -04002986static int
2987shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
2988{
2989 struct inode *inode;
2990 int error = -ENOSPC;
2991
2992 inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
2993 if (inode) {
2994 error = security_inode_init_security(inode, dir,
2995 NULL,
2996 shmem_initxattrs, NULL);
Christoph Hellwigfeda8212013-12-20 05:16:54 -08002997 if (error && error != -EOPNOTSUPP)
2998 goto out_iput;
2999 error = simple_acl_create(dir, inode);
3000 if (error)
3001 goto out_iput;
Al Viro60545d02013-06-07 01:20:27 -04003002 d_tmpfile(dentry, inode);
3003 }
3004 return error;
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003005out_iput:
3006 iput(inode);
3007 return error;
Al Viro60545d02013-06-07 01:20:27 -04003008}
3009
Al Viro18bb1db2011-07-26 01:41:39 -04003010static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011{
3012 int error;
3013
3014 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
3015 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07003016 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017 return 0;
3018}
3019
Al Viro4acdaf22011-07-26 01:42:34 -04003020static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
Al Viroebfc3b42012-06-10 18:05:36 -04003021 bool excl)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003022{
3023 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
3024}
3025
3026/*
3027 * Link a file..
3028 */
3029static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
3030{
David Howells75c3cfa2015-03-17 22:26:12 +00003031 struct inode *inode = d_inode(old_dentry);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08003032 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003033
3034 /*
3035 * No ordinary (disk based) filesystem counts links as inodes;
3036 * but each new link needs a new dentry, pinning lowmem, and
3037 * tmpfs dentries cannot be pruned until they are unlinked.
3038 */
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08003039 ret = shmem_reserve_inode(inode->i_sb);
3040 if (ret)
3041 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003042
3043 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07003044 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
Dave Hansend8c76e62006-09-30 23:29:04 -07003045 inc_nlink(inode);
Al Viro7de9c6ee2010-10-23 11:11:40 -04003046 ihold(inode); /* New dentry reference */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003047 dget(dentry); /* Extra pinning count for the created dentry */
3048 d_instantiate(dentry, inode);
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08003049out:
3050 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051}
3052
3053static int shmem_unlink(struct inode *dir, struct dentry *dentry)
3054{
David Howells75c3cfa2015-03-17 22:26:12 +00003055 struct inode *inode = d_inode(dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003056
Pavel Emelyanov5b04c682008-02-04 22:28:47 -08003057 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
3058 shmem_free_inode(inode->i_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003059
3060 dir->i_size -= BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07003061 inode->i_ctime = dir->i_ctime = dir->i_mtime = current_time(inode);
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003062 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003063 dput(dentry); /* Undo the count from "create" - this does all the work */
3064 return 0;
3065}
3066
3067static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
3068{
3069 if (!simple_empty(dentry))
3070 return -ENOTEMPTY;
3071
David Howells75c3cfa2015-03-17 22:26:12 +00003072 drop_nlink(d_inode(dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003073 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003074 return shmem_unlink(dir, dentry);
3075}
3076
Miklos Szeredi37456772014-07-23 15:15:34 +02003077static int shmem_exchange(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
3078{
David Howellse36cb0b2015-01-29 12:02:35 +00003079 bool old_is_dir = d_is_dir(old_dentry);
3080 bool new_is_dir = d_is_dir(new_dentry);
Miklos Szeredi37456772014-07-23 15:15:34 +02003081
3082 if (old_dir != new_dir && old_is_dir != new_is_dir) {
3083 if (old_is_dir) {
3084 drop_nlink(old_dir);
3085 inc_nlink(new_dir);
3086 } else {
3087 drop_nlink(new_dir);
3088 inc_nlink(old_dir);
3089 }
3090 }
3091 old_dir->i_ctime = old_dir->i_mtime =
3092 new_dir->i_ctime = new_dir->i_mtime =
David Howells75c3cfa2015-03-17 22:26:12 +00003093 d_inode(old_dentry)->i_ctime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07003094 d_inode(new_dentry)->i_ctime = current_time(old_dir);
Miklos Szeredi37456772014-07-23 15:15:34 +02003095
3096 return 0;
3097}
3098
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003099static int shmem_whiteout(struct inode *old_dir, struct dentry *old_dentry)
3100{
3101 struct dentry *whiteout;
3102 int error;
3103
3104 whiteout = d_alloc(old_dentry->d_parent, &old_dentry->d_name);
3105 if (!whiteout)
3106 return -ENOMEM;
3107
3108 error = shmem_mknod(old_dir, whiteout,
3109 S_IFCHR | WHITEOUT_MODE, WHITEOUT_DEV);
3110 dput(whiteout);
3111 if (error)
3112 return error;
3113
3114 /*
3115 * Cheat and hash the whiteout while the old dentry is still in
3116 * place, instead of playing games with FS_RENAME_DOES_D_MOVE.
3117 *
3118 * d_lookup() will consistently find one of them at this point,
3119 * not sure which one, but that isn't even important.
3120 */
3121 d_rehash(whiteout);
3122 return 0;
3123}
3124
Linus Torvalds1da177e2005-04-16 15:20:36 -07003125/*
3126 * The VFS layer already does all the dentry stuff for rename,
3127 * we just have to decrement the usage count for the target if
3128 * it exists so that the VFS layer correctly free's it when it
3129 * gets overwritten.
3130 */
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003131static int shmem_rename2(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry, unsigned int flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003132{
David Howells75c3cfa2015-03-17 22:26:12 +00003133 struct inode *inode = d_inode(old_dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003134 int they_are_dirs = S_ISDIR(inode->i_mode);
3135
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003136 if (flags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT))
Miklos Szeredi3b69ff52014-07-23 15:15:33 +02003137 return -EINVAL;
3138
Miklos Szeredi37456772014-07-23 15:15:34 +02003139 if (flags & RENAME_EXCHANGE)
3140 return shmem_exchange(old_dir, old_dentry, new_dir, new_dentry);
3141
Linus Torvalds1da177e2005-04-16 15:20:36 -07003142 if (!simple_empty(new_dentry))
3143 return -ENOTEMPTY;
3144
Miklos Szeredi46fdb792014-10-24 00:14:37 +02003145 if (flags & RENAME_WHITEOUT) {
3146 int error;
3147
3148 error = shmem_whiteout(old_dir, old_dentry);
3149 if (error)
3150 return error;
3151 }
3152
David Howells75c3cfa2015-03-17 22:26:12 +00003153 if (d_really_is_positive(new_dentry)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154 (void) shmem_unlink(new_dir, new_dentry);
Miklos Szeredib9280952014-09-24 17:56:17 +02003155 if (they_are_dirs) {
David Howells75c3cfa2015-03-17 22:26:12 +00003156 drop_nlink(d_inode(new_dentry));
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003157 drop_nlink(old_dir);
Miklos Szeredib9280952014-09-24 17:56:17 +02003158 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003159 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07003160 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07003161 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003162 }
3163
3164 old_dir->i_size -= BOGO_DIRENT_SIZE;
3165 new_dir->i_size += BOGO_DIRENT_SIZE;
3166 old_dir->i_ctime = old_dir->i_mtime =
3167 new_dir->i_ctime = new_dir->i_mtime =
Deepa Dinamani078cd822016-09-14 07:48:04 -07003168 inode->i_ctime = current_time(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003169 return 0;
3170}
3171
3172static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
3173{
3174 int error;
3175 int len;
3176 struct inode *inode;
Hugh Dickins9276aad2011-07-25 17:12:34 -07003177 struct page *page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003178 struct shmem_inode_info *info;
3179
3180 len = strlen(symname) + 1;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003181 if (len > PAGE_SIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003182 return -ENAMETOOLONG;
3183
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003184 inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185 if (!inode)
3186 return -ENOSPC;
3187
Mimi Zohar9d8f13b2011-06-06 15:29:25 -04003188 error = security_inode_init_security(inode, dir, &dentry->d_name,
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003189 shmem_initxattrs, NULL);
Stephen Smalley570bc1c2005-09-09 13:01:43 -07003190 if (error) {
3191 if (error != -EOPNOTSUPP) {
3192 iput(inode);
3193 return error;
3194 }
3195 error = 0;
3196 }
3197
Linus Torvalds1da177e2005-04-16 15:20:36 -07003198 info = SHMEM_I(inode);
3199 inode->i_size = len-1;
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003200 if (len <= SHORT_SYMLINK_LEN) {
Al Viro3ed47db2016-01-22 18:08:52 -05003201 inode->i_link = kmemdup(symname, len, GFP_KERNEL);
3202 if (!inode->i_link) {
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003203 iput(inode);
3204 return -ENOMEM;
3205 }
3206 inode->i_op = &shmem_short_symlink_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003207 } else {
Al Viroe8ecde22016-01-14 17:52:59 -05003208 inode_nohighmem(inode);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003209 error = shmem_getpage(inode, 0, &page, SGP_WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003210 if (error) {
3211 iput(inode);
3212 return error;
3213 }
Hugh Dickins14fcc232008-07-28 15:46:19 -07003214 inode->i_mapping->a_ops = &shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003215 inode->i_op = &shmem_symlink_inode_operations;
Al Viro21fc61c2015-11-17 01:07:57 -05003216 memcpy(page_address(page), symname, len);
Hugh Dickinsec9516f2012-05-29 15:06:39 -07003217 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003218 set_page_dirty(page);
Wu Fengguang6746aff2009-09-16 11:50:14 +02003219 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003220 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003221 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003222 dir->i_size += BOGO_DIRENT_SIZE;
Deepa Dinamani078cd822016-09-14 07:48:04 -07003223 dir->i_ctime = dir->i_mtime = current_time(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003224 d_instantiate(dentry, inode);
3225 dget(dentry);
3226 return 0;
3227}
3228
Al Virofceef392015-12-29 15:58:39 -05003229static void shmem_put_link(void *arg)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003230{
Al Virofceef392015-12-29 15:58:39 -05003231 mark_page_accessed(arg);
3232 put_page(arg);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003233}
3234
Al Viro6b255392015-11-17 10:20:54 -05003235static const char *shmem_get_link(struct dentry *dentry,
Al Virofceef392015-12-29 15:58:39 -05003236 struct inode *inode,
3237 struct delayed_call *done)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003238{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003239 struct page *page = NULL;
Al Viro6b255392015-11-17 10:20:54 -05003240 int error;
Al Viro6a6c9902015-11-17 10:54:32 -05003241 if (!dentry) {
3242 page = find_get_page(inode->i_mapping, 0);
3243 if (!page)
3244 return ERR_PTR(-ECHILD);
3245 if (!PageUptodate(page)) {
3246 put_page(page);
3247 return ERR_PTR(-ECHILD);
3248 }
3249 } else {
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07003250 error = shmem_getpage(inode, 0, &page, SGP_READ);
Al Viro6a6c9902015-11-17 10:54:32 -05003251 if (error)
3252 return ERR_PTR(error);
3253 unlock_page(page);
3254 }
Al Virofceef392015-12-29 15:58:39 -05003255 set_delayed_call(done, shmem_put_link, page);
Al Viro21fc61c2015-11-17 01:07:57 -05003256 return page_address(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003257}
3258
Eric Parisb09e0fa2011-05-24 17:12:39 -07003259#ifdef CONFIG_TMPFS_XATTR
3260/*
3261 * Superblocks without xattr inode operations may get some security.* xattr
3262 * support from the LSM "for free". As soon as we have any other xattrs
3263 * like ACLs, we also need to implement the security.* handlers at
3264 * filesystem level, though.
3265 */
3266
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003267/*
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003268 * Callback for security_inode_init_security() for acquiring xattrs.
3269 */
3270static int shmem_initxattrs(struct inode *inode,
3271 const struct xattr *xattr_array,
3272 void *fs_info)
3273{
3274 struct shmem_inode_info *info = SHMEM_I(inode);
3275 const struct xattr *xattr;
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003276 struct simple_xattr *new_xattr;
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003277 size_t len;
3278
3279 for (xattr = xattr_array; xattr->name != NULL; xattr++) {
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003280 new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003281 if (!new_xattr)
3282 return -ENOMEM;
3283
3284 len = strlen(xattr->name) + 1;
3285 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
3286 GFP_KERNEL);
3287 if (!new_xattr->name) {
3288 kfree(new_xattr);
3289 return -ENOMEM;
3290 }
3291
3292 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
3293 XATTR_SECURITY_PREFIX_LEN);
3294 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
3295 xattr->name, len);
3296
Aristeu Rozanski38f38652012-08-23 16:53:28 -04003297 simple_xattr_list_add(&info->xattrs, new_xattr);
Jarkko Sakkinen6d9d88d2012-03-21 16:34:05 -07003298 }
3299
3300 return 0;
3301}
3302
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003303static int shmem_xattr_handler_get(const struct xattr_handler *handler,
Al Virob2968212016-04-10 20:48:24 -04003304 struct dentry *unused, struct inode *inode,
3305 const char *name, void *buffer, size_t size)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003306{
Al Virob2968212016-04-10 20:48:24 -04003307 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003308
3309 name = xattr_full_name(handler, name);
3310 return simple_xattr_get(&info->xattrs, name, buffer, size);
3311}
3312
3313static int shmem_xattr_handler_set(const struct xattr_handler *handler,
Al Viro59301222016-05-27 10:19:30 -04003314 struct dentry *unused, struct inode *inode,
3315 const char *name, const void *value,
3316 size_t size, int flags)
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003317{
Al Viro59301222016-05-27 10:19:30 -04003318 struct shmem_inode_info *info = SHMEM_I(inode);
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003319
3320 name = xattr_full_name(handler, name);
3321 return simple_xattr_set(&info->xattrs, name, value, size, flags);
3322}
3323
3324static const struct xattr_handler shmem_security_xattr_handler = {
3325 .prefix = XATTR_SECURITY_PREFIX,
3326 .get = shmem_xattr_handler_get,
3327 .set = shmem_xattr_handler_set,
3328};
3329
3330static const struct xattr_handler shmem_trusted_xattr_handler = {
3331 .prefix = XATTR_TRUSTED_PREFIX,
3332 .get = shmem_xattr_handler_get,
3333 .set = shmem_xattr_handler_set,
3334};
3335
Eric Parisb09e0fa2011-05-24 17:12:39 -07003336static const struct xattr_handler *shmem_xattr_handlers[] = {
3337#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003338 &posix_acl_access_xattr_handler,
3339 &posix_acl_default_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003340#endif
Andreas Gruenbacheraa7c5242015-12-02 14:44:38 +01003341 &shmem_security_xattr_handler,
3342 &shmem_trusted_xattr_handler,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003343 NULL
3344};
3345
Eric Parisb09e0fa2011-05-24 17:12:39 -07003346static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
3347{
David Howells75c3cfa2015-03-17 22:26:12 +00003348 struct shmem_inode_info *info = SHMEM_I(d_inode(dentry));
Andreas Gruenbacher786534b2015-12-02 14:44:39 +01003349 return simple_xattr_list(d_inode(dentry), &info->xattrs, buffer, size);
Eric Parisb09e0fa2011-05-24 17:12:39 -07003350}
3351#endif /* CONFIG_TMPFS_XATTR */
3352
Hugh Dickins69f07ec2011-08-03 16:21:26 -07003353static const struct inode_operations shmem_short_symlink_operations = {
Al Viro6b255392015-11-17 10:20:54 -05003354 .get_link = simple_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003355#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003356 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003357#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003358};
3359
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003360static const struct inode_operations shmem_symlink_inode_operations = {
Al Viro6b255392015-11-17 10:20:54 -05003361 .get_link = shmem_get_link,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003362#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003363 .listxattr = shmem_listxattr,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003364#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003365};
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003366
David M. Grimes91828a42006-10-17 00:09:45 -07003367static struct dentry *shmem_get_parent(struct dentry *child)
3368{
3369 return ERR_PTR(-ESTALE);
3370}
3371
3372static int shmem_match(struct inode *ino, void *vfh)
3373{
3374 __u32 *fh = vfh;
3375 __u64 inum = fh[2];
3376 inum = (inum << 32) | fh[1];
3377 return ino->i_ino == inum && fh[0] == ino->i_generation;
3378}
3379
Christoph Hellwig480b1162007-10-21 16:42:13 -07003380static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
3381 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07003382{
David M. Grimes91828a42006-10-17 00:09:45 -07003383 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07003384 struct dentry *dentry = NULL;
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003385 u64 inum;
David M. Grimes91828a42006-10-17 00:09:45 -07003386
Christoph Hellwig480b1162007-10-21 16:42:13 -07003387 if (fh_len < 3)
3388 return NULL;
3389
Hugh Dickins35c2a7f2012-10-07 20:32:51 -07003390 inum = fid->raw[2];
3391 inum = (inum << 32) | fid->raw[1];
3392
Christoph Hellwig480b1162007-10-21 16:42:13 -07003393 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
3394 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07003395 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07003396 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07003397 iput(inode);
3398 }
3399
Christoph Hellwig480b1162007-10-21 16:42:13 -07003400 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07003401}
3402
Al Virob0b03822012-04-02 14:34:06 -04003403static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
3404 struct inode *parent)
David M. Grimes91828a42006-10-17 00:09:45 -07003405{
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303406 if (*len < 3) {
3407 *len = 3;
Namjae Jeon94e07a752013-02-17 15:48:11 +09003408 return FILEID_INVALID;
Aneesh Kumar K.V5fe0c232011-01-29 18:43:25 +05303409 }
David M. Grimes91828a42006-10-17 00:09:45 -07003410
Al Viro1d3382cb2010-10-23 15:19:20 -04003411 if (inode_unhashed(inode)) {
David M. Grimes91828a42006-10-17 00:09:45 -07003412 /* Unfortunately insert_inode_hash is not idempotent,
3413 * so as we hash inodes here rather than at creation
3414 * time, we need a lock to ensure we only try
3415 * to do it once
3416 */
3417 static DEFINE_SPINLOCK(lock);
3418 spin_lock(&lock);
Al Viro1d3382cb2010-10-23 15:19:20 -04003419 if (inode_unhashed(inode))
David M. Grimes91828a42006-10-17 00:09:45 -07003420 __insert_inode_hash(inode,
3421 inode->i_ino + inode->i_generation);
3422 spin_unlock(&lock);
3423 }
3424
3425 fh[0] = inode->i_generation;
3426 fh[1] = inode->i_ino;
3427 fh[2] = ((__u64)inode->i_ino) >> 32;
3428
3429 *len = 3;
3430 return 1;
3431}
3432
Christoph Hellwig39655162007-10-21 16:42:17 -07003433static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07003434 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07003435 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07003436 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07003437};
3438
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003439static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
3440 bool remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003441{
3442 char *this_char, *value, *rest;
Greg Thelen49cd0a52013-02-22 16:36:02 -08003443 struct mempolicy *mpol = NULL;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003444 uid_t uid;
3445 gid_t gid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003446
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00003447 while (options != NULL) {
3448 this_char = options;
3449 for (;;) {
3450 /*
3451 * NUL-terminate this option: unfortunately,
3452 * mount options form a comma-separated list,
3453 * but mpol's nodelist may also contain commas.
3454 */
3455 options = strchr(options, ',');
3456 if (options == NULL)
3457 break;
3458 options++;
3459 if (!isdigit(*options)) {
3460 options[-1] = '\0';
3461 break;
3462 }
3463 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07003464 if (!*this_char)
3465 continue;
3466 if ((value = strchr(this_char,'=')) != NULL) {
3467 *value++ = 0;
3468 } else {
Joe Perches11705322016-03-17 14:19:50 -07003469 pr_err("tmpfs: No value for mount option '%s'\n",
3470 this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003471 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003472 }
3473
3474 if (!strcmp(this_char,"size")) {
3475 unsigned long long size;
3476 size = memparse(value,&rest);
3477 if (*rest == '%') {
3478 size <<= PAGE_SHIFT;
3479 size *= totalram_pages;
3480 do_div(size, 100);
3481 rest++;
3482 }
3483 if (*rest)
3484 goto bad_val;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003485 sbinfo->max_blocks =
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003486 DIV_ROUND_UP(size, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003487 } else if (!strcmp(this_char,"nr_blocks")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003488 sbinfo->max_blocks = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003489 if (*rest)
3490 goto bad_val;
3491 } else if (!strcmp(this_char,"nr_inodes")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003492 sbinfo->max_inodes = memparse(value, &rest);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003493 if (*rest)
3494 goto bad_val;
3495 } else if (!strcmp(this_char,"mode")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003496 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003497 continue;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003498 sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003499 if (*rest)
3500 goto bad_val;
3501 } else if (!strcmp(this_char,"uid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003502 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003503 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003504 uid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003505 if (*rest)
3506 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003507 sbinfo->uid = make_kuid(current_user_ns(), uid);
3508 if (!uid_valid(sbinfo->uid))
3509 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003510 } else if (!strcmp(this_char,"gid")) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003511 if (remount)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003512 continue;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003513 gid = simple_strtoul(value, &rest, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003514 if (*rest)
3515 goto bad_val;
Eric W. Biederman8751e032012-02-07 16:46:12 -08003516 sbinfo->gid = make_kgid(current_user_ns(), gid);
3517 if (!gid_valid(sbinfo->gid))
3518 goto bad_val;
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003519#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003520 } else if (!strcmp(this_char, "huge")) {
3521 int huge;
3522 huge = shmem_parse_huge(value);
3523 if (huge < 0)
3524 goto bad_val;
3525 if (!has_transparent_hugepage() &&
3526 huge != SHMEM_HUGE_NEVER)
3527 goto bad_val;
3528 sbinfo->huge = huge;
3529#endif
3530#ifdef CONFIG_NUMA
Robin Holt7339ff82006-01-14 13:20:48 -08003531 } else if (!strcmp(this_char,"mpol")) {
Greg Thelen49cd0a52013-02-22 16:36:02 -08003532 mpol_put(mpol);
3533 mpol = NULL;
3534 if (mpol_parse_str(value, &mpol))
Robin Holt7339ff82006-01-14 13:20:48 -08003535 goto bad_val;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003536#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003537 } else {
Joe Perches11705322016-03-17 14:19:50 -07003538 pr_err("tmpfs: Bad mount option %s\n", this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003539 goto error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003540 }
3541 }
Greg Thelen49cd0a52013-02-22 16:36:02 -08003542 sbinfo->mpol = mpol;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003543 return 0;
3544
3545bad_val:
Joe Perches11705322016-03-17 14:19:50 -07003546 pr_err("tmpfs: Bad value '%s' for mount option '%s'\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -07003547 value, this_char);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003548error:
3549 mpol_put(mpol);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003550 return 1;
3551
3552}
3553
3554static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
3555{
3556 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003557 struct shmem_sb_info config = *sbinfo;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003558 unsigned long inodes;
3559 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003560
Greg Thelen5f001102013-02-22 16:36:01 -08003561 config.mpol = NULL;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003562 if (shmem_parse_options(data, &config, true))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003563 return error;
3564
3565 spin_lock(&sbinfo->stat_lock);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003566 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
Tim Chen7e496292010-08-09 17:19:05 -07003567 if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003568 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003569 if (config.max_inodes < inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003570 goto out;
3571 /*
Hugh Dickins54af60422011-08-03 16:21:24 -07003572 * Those tests disallow limited->unlimited while any are in use;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003573 * but we must separately disallow unlimited->limited, because
3574 * in that case we have no record of how much is already in use.
3575 */
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003576 if (config.max_blocks && !sbinfo->max_blocks)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003577 goto out;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003578 if (config.max_inodes && !sbinfo->max_inodes)
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003579 goto out;
3580
3581 error = 0;
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003582 sbinfo->huge = config.huge;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003583 sbinfo->max_blocks = config.max_blocks;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003584 sbinfo->max_inodes = config.max_inodes;
3585 sbinfo->free_inodes = config.max_inodes - inodes;
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003586
Greg Thelen5f001102013-02-22 16:36:01 -08003587 /*
3588 * Preserve previous mempolicy unless mpol remount option was specified.
3589 */
3590 if (config.mpol) {
3591 mpol_put(sbinfo->mpol);
3592 sbinfo->mpol = config.mpol; /* transfers initial ref */
3593 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003594out:
3595 spin_unlock(&sbinfo->stat_lock);
3596 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003597}
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003598
Al Viro34c80b12011-12-08 21:32:45 -05003599static int shmem_show_options(struct seq_file *seq, struct dentry *root)
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003600{
Al Viro34c80b12011-12-08 21:32:45 -05003601 struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003602
3603 if (sbinfo->max_blocks != shmem_default_max_blocks())
3604 seq_printf(seq, ",size=%luk",
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003605 sbinfo->max_blocks << (PAGE_SHIFT - 10));
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003606 if (sbinfo->max_inodes != shmem_default_max_inodes())
3607 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
3608 if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
Al Viro09208d12011-07-26 03:15:03 -04003609 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
Eric W. Biederman8751e032012-02-07 16:46:12 -08003610 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
3611 seq_printf(seq, ",uid=%u",
3612 from_kuid_munged(&init_user_ns, sbinfo->uid));
3613 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
3614 seq_printf(seq, ",gid=%u",
3615 from_kgid_munged(&init_user_ns, sbinfo->gid));
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003616#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003617 /* Rightly or wrongly, show huge mount option unmasked by shmem_huge */
3618 if (sbinfo->huge)
3619 seq_printf(seq, ",huge=%s", shmem_format_huge(sbinfo->huge));
3620#endif
Lee Schermerhorn71fe8042008-04-28 02:13:26 -07003621 shmem_show_mpol(seq, sbinfo->mpol);
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003622 return 0;
3623}
David Herrmann9183df22014-08-08 14:25:29 -07003624
3625#define MFD_NAME_PREFIX "memfd:"
3626#define MFD_NAME_PREFIX_LEN (sizeof(MFD_NAME_PREFIX) - 1)
3627#define MFD_NAME_MAX_LEN (NAME_MAX - MFD_NAME_PREFIX_LEN)
3628
3629#define MFD_ALL_FLAGS (MFD_CLOEXEC | MFD_ALLOW_SEALING)
3630
3631SYSCALL_DEFINE2(memfd_create,
3632 const char __user *, uname,
3633 unsigned int, flags)
3634{
3635 struct shmem_inode_info *info;
3636 struct file *file;
3637 int fd, error;
3638 char *name;
3639 long len;
3640
3641 if (flags & ~(unsigned int)MFD_ALL_FLAGS)
3642 return -EINVAL;
3643
3644 /* length includes terminating zero */
3645 len = strnlen_user(uname, MFD_NAME_MAX_LEN + 1);
3646 if (len <= 0)
3647 return -EFAULT;
3648 if (len > MFD_NAME_MAX_LEN + 1)
3649 return -EINVAL;
3650
3651 name = kmalloc(len + MFD_NAME_PREFIX_LEN, GFP_TEMPORARY);
3652 if (!name)
3653 return -ENOMEM;
3654
3655 strcpy(name, MFD_NAME_PREFIX);
3656 if (copy_from_user(&name[MFD_NAME_PREFIX_LEN], uname, len)) {
3657 error = -EFAULT;
3658 goto err_name;
3659 }
3660
3661 /* terminating-zero may have changed after strnlen_user() returned */
3662 if (name[len + MFD_NAME_PREFIX_LEN - 1]) {
3663 error = -EFAULT;
3664 goto err_name;
3665 }
3666
3667 fd = get_unused_fd_flags((flags & MFD_CLOEXEC) ? O_CLOEXEC : 0);
3668 if (fd < 0) {
3669 error = fd;
3670 goto err_name;
3671 }
3672
3673 file = shmem_file_setup(name, 0, VM_NORESERVE);
3674 if (IS_ERR(file)) {
3675 error = PTR_ERR(file);
3676 goto err_fd;
3677 }
3678 info = SHMEM_I(file_inode(file));
3679 file->f_mode |= FMODE_LSEEK | FMODE_PREAD | FMODE_PWRITE;
3680 file->f_flags |= O_RDWR | O_LARGEFILE;
3681 if (flags & MFD_ALLOW_SEALING)
3682 info->seals &= ~F_SEAL_SEAL;
3683
3684 fd_install(fd, file);
3685 kfree(name);
3686 return fd;
3687
3688err_fd:
3689 put_unused_fd(fd);
3690err_name:
3691 kfree(name);
3692 return error;
3693}
3694
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003695#endif /* CONFIG_TMPFS */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003696
3697static void shmem_put_super(struct super_block *sb)
3698{
Hugh Dickins602586a2010-08-17 15:23:56 -07003699 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
3700
3701 percpu_counter_destroy(&sbinfo->used_blocks);
Greg Thelen49cd0a52013-02-22 16:36:02 -08003702 mpol_put(sbinfo->mpol);
Hugh Dickins602586a2010-08-17 15:23:56 -07003703 kfree(sbinfo);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003704 sb->s_fs_info = NULL;
3705}
3706
Kay Sievers2b2af542009-04-30 15:23:42 +02003707int shmem_fill_super(struct super_block *sb, void *data, int silent)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003708{
3709 struct inode *inode;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003710 struct shmem_sb_info *sbinfo;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003711 int err = -ENOMEM;
3712
3713 /* Round up to L1_CACHE_BYTES to resist false sharing */
Pekka Enberg425fbf02009-09-21 17:03:50 -07003714 sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003715 L1_CACHE_BYTES), GFP_KERNEL);
3716 if (!sbinfo)
3717 return -ENOMEM;
3718
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003719 sbinfo->mode = S_IRWXUGO | S_ISVTX;
David Howells76aac0e2008-11-14 10:39:12 +11003720 sbinfo->uid = current_fsuid();
3721 sbinfo->gid = current_fsgid();
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003722 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003723
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003724#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07003725 /*
3726 * Per default we only allow half of the physical ram per
3727 * tmpfs instance, limiting inodes to one per page of lowmem;
3728 * but the internal instance is left unlimited.
3729 */
Al Viroca4e0512013-08-31 12:57:10 -04003730 if (!(sb->s_flags & MS_KERNMOUNT)) {
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003731 sbinfo->max_blocks = shmem_default_max_blocks();
3732 sbinfo->max_inodes = shmem_default_max_inodes();
3733 if (shmem_parse_options(data, sbinfo, false)) {
3734 err = -EINVAL;
3735 goto failed;
3736 }
Al Viroca4e0512013-08-31 12:57:10 -04003737 } else {
3738 sb->s_flags |= MS_NOUSER;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003739 }
David M. Grimes91828a42006-10-17 00:09:45 -07003740 sb->s_export_op = &shmem_export_ops;
Hugh Dickins2f6e38f2012-05-29 15:06:38 -07003741 sb->s_flags |= MS_NOSEC;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003742#else
3743 sb->s_flags |= MS_NOUSER;
3744#endif
3745
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003746 spin_lock_init(&sbinfo->stat_lock);
Tejun Heo908c7f12014-09-08 09:51:29 +09003747 if (percpu_counter_init(&sbinfo->used_blocks, 0, GFP_KERNEL))
Hugh Dickins602586a2010-08-17 15:23:56 -07003748 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003749 sbinfo->free_inodes = sbinfo->max_inodes;
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003750 spin_lock_init(&sbinfo->shrinklist_lock);
3751 INIT_LIST_HEAD(&sbinfo->shrinklist);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003752
Hugh Dickins285b2c42011-08-03 16:21:20 -07003753 sb->s_maxbytes = MAX_LFS_FILESIZE;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003754 sb->s_blocksize = PAGE_SIZE;
3755 sb->s_blocksize_bits = PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003756 sb->s_magic = TMPFS_MAGIC;
3757 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01003758 sb->s_time_gran = 1;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003759#ifdef CONFIG_TMPFS_XATTR
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003760 sb->s_xattr = shmem_xattr_handlers;
Eric Parisb09e0fa2011-05-24 17:12:39 -07003761#endif
3762#ifdef CONFIG_TMPFS_POSIX_ACL
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003763 sb->s_flags |= MS_POSIXACL;
3764#endif
Amir Goldstein2b4db792017-05-18 15:29:33 +03003765 uuid_gen(&sb->s_uuid);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07003766
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03003767 inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003768 if (!inode)
3769 goto failed;
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003770 inode->i_uid = sbinfo->uid;
3771 inode->i_gid = sbinfo->gid;
Al Viro318ceed2012-02-12 22:08:01 -05003772 sb->s_root = d_make_root(inode);
3773 if (!sb->s_root)
Al Viro48fde702012-01-08 22:15:13 -05003774 goto failed;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003775 return 0;
3776
Linus Torvalds1da177e2005-04-16 15:20:36 -07003777failed:
3778 shmem_put_super(sb);
3779 return err;
3780}
3781
Pekka Enbergfcc234f2006-03-22 00:08:13 -08003782static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003783
3784static struct inode *shmem_alloc_inode(struct super_block *sb)
3785{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003786 struct shmem_inode_info *info;
3787 info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
3788 if (!info)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003789 return NULL;
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003790 return &info->vfs_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003791}
3792
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003793static void shmem_destroy_callback(struct rcu_head *head)
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003794{
3795 struct inode *inode = container_of(head, struct inode, i_rcu);
Al Viro84e710d2016-04-15 00:58:55 -04003796 if (S_ISLNK(inode->i_mode))
3797 kfree(inode->i_link);
Nick Pigginfa0d7e3d2011-01-07 17:49:49 +11003798 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
3799}
3800
Linus Torvalds1da177e2005-04-16 15:20:36 -07003801static void shmem_destroy_inode(struct inode *inode)
3802{
Al Viro09208d12011-07-26 03:15:03 -04003803 if (S_ISREG(inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003804 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003805 call_rcu(&inode->i_rcu, shmem_destroy_callback);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003806}
3807
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003808static void shmem_init_inode(void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003809{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003810 struct shmem_inode_info *info = foo;
3811 inode_init_once(&info->vfs_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003812}
3813
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003814static int shmem_init_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003815{
3816 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
3817 sizeof(struct shmem_inode_info),
Vladimir Davydov5d097052016-01-14 15:18:21 -08003818 0, SLAB_PANIC|SLAB_ACCOUNT, shmem_init_inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003819 return 0;
3820}
3821
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003822static void shmem_destroy_inodecache(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003823{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07003824 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003825}
3826
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07003827static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003828 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08003829 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003830#ifdef CONFIG_TMPFS
Nick Piggin800d15a2007-10-16 01:25:03 -07003831 .write_begin = shmem_write_begin,
3832 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003833#endif
Andrew Morton1c939232014-10-09 15:27:59 -07003834#ifdef CONFIG_MIGRATION
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07003835 .migratepage = migrate_page,
Andrew Morton1c939232014-10-09 15:27:59 -07003836#endif
Andi Kleenaa261f52009-09-16 11:50:16 +02003837 .error_remove_page = generic_error_remove_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003838};
3839
Helge Deller15ad7cd2006-12-06 20:40:36 -08003840static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003841 .mmap = shmem_mmap,
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07003842 .get_unmapped_area = shmem_get_unmapped_area,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003843#ifdef CONFIG_TMPFS
Hugh Dickins220f2ac2012-12-12 13:52:21 -08003844 .llseek = shmem_file_llseek,
Al Viro2ba5bbe2014-04-02 20:00:02 -04003845 .read_iter = shmem_file_read_iter,
Al Viro81742022014-04-03 03:17:43 -04003846 .write_iter = generic_file_write_iter,
Christoph Hellwig1b061d92010-05-26 17:53:41 +02003847 .fsync = noop_fsync,
Al Viro82c156f2016-09-22 23:35:42 -04003848 .splice_read = generic_file_splice_read,
Al Virof6cb85d2014-04-05 04:38:56 -04003849 .splice_write = iter_file_splice_write,
Hugh Dickins83e4fa92012-05-29 15:06:40 -07003850 .fallocate = shmem_fallocate,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003851#endif
3852};
3853
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003854static const struct inode_operations shmem_inode_operations = {
Yu Zhao44a30222015-09-08 15:03:33 -07003855 .getattr = shmem_getattr,
Hugh Dickins94c1e622011-06-27 16:18:03 -07003856 .setattr = shmem_setattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003857#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003858 .listxattr = shmem_listxattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003859 .set_acl = simple_set_acl,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003860#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003861};
3862
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003863static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003864#ifdef CONFIG_TMPFS
3865 .create = shmem_create,
3866 .lookup = simple_lookup,
3867 .link = shmem_link,
3868 .unlink = shmem_unlink,
3869 .symlink = shmem_symlink,
3870 .mkdir = shmem_mkdir,
3871 .rmdir = shmem_rmdir,
3872 .mknod = shmem_mknod,
Miklos Szeredi2773bf02016-09-27 11:03:58 +02003873 .rename = shmem_rename2,
Al Viro60545d02013-06-07 01:20:27 -04003874 .tmpfile = shmem_tmpfile,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003875#endif
Eric Parisb09e0fa2011-05-24 17:12:39 -07003876#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003877 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003878#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003879#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003880 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003881 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003882#endif
3883};
3884
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08003885static const struct inode_operations shmem_special_inode_operations = {
Eric Parisb09e0fa2011-05-24 17:12:39 -07003886#ifdef CONFIG_TMPFS_XATTR
Eric Parisb09e0fa2011-05-24 17:12:39 -07003887 .listxattr = shmem_listxattr,
Eric Parisb09e0fa2011-05-24 17:12:39 -07003888#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003889#ifdef CONFIG_TMPFS_POSIX_ACL
Hugh Dickins94c1e622011-06-27 16:18:03 -07003890 .setattr = shmem_setattr,
Christoph Hellwigfeda8212013-12-20 05:16:54 -08003891 .set_acl = simple_set_acl,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07003892#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003893};
3894
Hugh Dickins759b9772007-03-05 00:30:28 -08003895static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003896 .alloc_inode = shmem_alloc_inode,
3897 .destroy_inode = shmem_destroy_inode,
3898#ifdef CONFIG_TMPFS
3899 .statfs = shmem_statfs,
3900 .remount_fs = shmem_remount_fs,
akpm@linux-foundation.org680d7942008-02-08 04:21:48 -08003901 .show_options = shmem_show_options,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003902#endif
Al Viro1f895f72010-06-05 19:10:41 -04003903 .evict_inode = shmem_evict_inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003904 .drop_inode = generic_delete_inode,
3905 .put_super = shmem_put_super,
Kirill A. Shutemov779750d2016-07-26 15:26:38 -07003906#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
3907 .nr_cached_objects = shmem_unused_huge_count,
3908 .free_cached_objects = shmem_unused_huge_scan,
3909#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003910};
3911
Alexey Dobriyanf0f37e2f2009-09-27 22:29:37 +04003912static const struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07003913 .fault = shmem_fault,
Ning Qud7c17552014-04-07 15:37:24 -07003914 .map_pages = filemap_map_pages,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003915#ifdef CONFIG_NUMA
3916 .set_policy = shmem_set_policy,
3917 .get_policy = shmem_get_policy,
3918#endif
3919};
3920
Al Viro3c26ff62010-07-25 11:46:36 +04003921static struct dentry *shmem_mount(struct file_system_type *fs_type,
3922 int flags, const char *dev_name, void *data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003923{
Al Viro3c26ff62010-07-25 11:46:36 +04003924 return mount_nodev(fs_type, flags, data, shmem_fill_super);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003925}
3926
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003927static struct file_system_type shmem_fs_type = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003928 .owner = THIS_MODULE,
3929 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04003930 .mount = shmem_mount,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003931 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08003932 .fs_flags = FS_USERNS_MOUNT,
Linus Torvalds1da177e2005-04-16 15:20:36 -07003933};
Linus Torvalds1da177e2005-04-16 15:20:36 -07003934
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003935int __init shmem_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003936{
3937 int error;
3938
Rob Landley16203a72013-09-11 14:26:12 -07003939 /* If rootfs called this, don't re-init */
3940 if (shmem_inode_cachep)
3941 return 0;
3942
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003943 error = shmem_init_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003944 if (error)
3945 goto out3;
3946
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003947 error = register_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003948 if (error) {
Joe Perches11705322016-03-17 14:19:50 -07003949 pr_err("Could not register tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003950 goto out2;
3951 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07003952
Al Viroca4e0512013-08-31 12:57:10 -04003953 shm_mnt = kern_mount(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003954 if (IS_ERR(shm_mnt)) {
3955 error = PTR_ERR(shm_mnt);
Joe Perches11705322016-03-17 14:19:50 -07003956 pr_err("Could not kern_mount tmpfs\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07003957 goto out1;
3958 }
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003959
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003960#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003961 if (has_transparent_hugepage() && shmem_huge < SHMEM_HUGE_DENY)
3962 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
3963 else
3964 shmem_huge = 0; /* just in case it was patched */
3965#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003966 return 0;
3967
3968out1:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003969 unregister_filesystem(&shmem_fs_type);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003970out2:
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07003971 shmem_destroy_inodecache();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003972out3:
3973 shm_mnt = ERR_PTR(error);
3974 return error;
3975}
Matt Mackall853ac432009-01-06 14:40:20 -08003976
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07003977#if defined(CONFIG_TRANSPARENT_HUGE_PAGECACHE) && defined(CONFIG_SYSFS)
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07003978static ssize_t shmem_enabled_show(struct kobject *kobj,
3979 struct kobj_attribute *attr, char *buf)
3980{
3981 int values[] = {
3982 SHMEM_HUGE_ALWAYS,
3983 SHMEM_HUGE_WITHIN_SIZE,
3984 SHMEM_HUGE_ADVISE,
3985 SHMEM_HUGE_NEVER,
3986 SHMEM_HUGE_DENY,
3987 SHMEM_HUGE_FORCE,
3988 };
3989 int i, count;
3990
3991 for (i = 0, count = 0; i < ARRAY_SIZE(values); i++) {
3992 const char *fmt = shmem_huge == values[i] ? "[%s] " : "%s ";
3993
3994 count += sprintf(buf + count, fmt,
3995 shmem_format_huge(values[i]));
3996 }
3997 buf[count - 1] = '\n';
3998 return count;
3999}
4000
4001static ssize_t shmem_enabled_store(struct kobject *kobj,
4002 struct kobj_attribute *attr, const char *buf, size_t count)
4003{
4004 char tmp[16];
4005 int huge;
4006
4007 if (count + 1 > sizeof(tmp))
4008 return -EINVAL;
4009 memcpy(tmp, buf, count);
4010 tmp[count] = '\0';
4011 if (count && tmp[count - 1] == '\n')
4012 tmp[count - 1] = '\0';
4013
4014 huge = shmem_parse_huge(tmp);
4015 if (huge == -EINVAL)
4016 return -EINVAL;
4017 if (!has_transparent_hugepage() &&
4018 huge != SHMEM_HUGE_NEVER && huge != SHMEM_HUGE_DENY)
4019 return -EINVAL;
4020
4021 shmem_huge = huge;
4022 if (shmem_huge < SHMEM_HUGE_DENY)
4023 SHMEM_SB(shm_mnt->mnt_sb)->huge = shmem_huge;
4024 return count;
4025}
4026
4027struct kobj_attribute shmem_enabled_attr =
4028 __ATTR(shmem_enabled, 0644, shmem_enabled_show, shmem_enabled_store);
Arnd Bergmann3b337192016-08-10 16:27:44 -07004029#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE && CONFIG_SYSFS */
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004030
Arnd Bergmann3b337192016-08-10 16:27:44 -07004031#ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004032bool shmem_huge_enabled(struct vm_area_struct *vma)
4033{
4034 struct inode *inode = file_inode(vma->vm_file);
4035 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
4036 loff_t i_size;
4037 pgoff_t off;
4038
4039 if (shmem_huge == SHMEM_HUGE_FORCE)
4040 return true;
4041 if (shmem_huge == SHMEM_HUGE_DENY)
4042 return false;
4043 switch (sbinfo->huge) {
4044 case SHMEM_HUGE_NEVER:
4045 return false;
4046 case SHMEM_HUGE_ALWAYS:
4047 return true;
4048 case SHMEM_HUGE_WITHIN_SIZE:
4049 off = round_up(vma->vm_pgoff, HPAGE_PMD_NR);
4050 i_size = round_up(i_size_read(inode), PAGE_SIZE);
4051 if (i_size >= HPAGE_PMD_SIZE &&
4052 i_size >> PAGE_SHIFT >= off)
4053 return true;
4054 case SHMEM_HUGE_ADVISE:
4055 /* TODO: implement fadvise() hints */
4056 return (vma->vm_flags & VM_HUGEPAGE);
4057 default:
4058 VM_BUG_ON(1);
4059 return false;
4060 }
4061}
Arnd Bergmann3b337192016-08-10 16:27:44 -07004062#endif /* CONFIG_TRANSPARENT_HUGE_PAGECACHE */
Kirill A. Shutemov5a6e75f2016-07-26 15:26:13 -07004063
Matt Mackall853ac432009-01-06 14:40:20 -08004064#else /* !CONFIG_SHMEM */
4065
4066/*
4067 * tiny-shmem: simple shmemfs and tmpfs using ramfs code
4068 *
4069 * This is intended for small system where the benefits of the full
4070 * shmem code (swap-backed and resource-limited) are outweighed by
4071 * their complexity. On systems without swap this code should be
4072 * effectively equivalent, but much lighter weight.
4073 */
4074
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004075static struct file_system_type shmem_fs_type = {
Matt Mackall853ac432009-01-06 14:40:20 -08004076 .name = "tmpfs",
Al Viro3c26ff62010-07-25 11:46:36 +04004077 .mount = ramfs_mount,
Matt Mackall853ac432009-01-06 14:40:20 -08004078 .kill_sb = kill_litter_super,
Eric W. Biederman2b8576c2013-01-25 16:32:10 -08004079 .fs_flags = FS_USERNS_MOUNT,
Matt Mackall853ac432009-01-06 14:40:20 -08004080};
4081
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004082int __init shmem_init(void)
Matt Mackall853ac432009-01-06 14:40:20 -08004083{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004084 BUG_ON(register_filesystem(&shmem_fs_type) != 0);
Matt Mackall853ac432009-01-06 14:40:20 -08004085
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004086 shm_mnt = kern_mount(&shmem_fs_type);
Matt Mackall853ac432009-01-06 14:40:20 -08004087 BUG_ON(IS_ERR(shm_mnt));
4088
4089 return 0;
4090}
4091
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004092int shmem_unuse(swp_entry_t swap, struct page *page)
Matt Mackall853ac432009-01-06 14:40:20 -08004093{
4094 return 0;
4095}
4096
Hugh Dickins3f96b792009-09-21 17:03:37 -07004097int shmem_lock(struct file *file, int lock, struct user_struct *user)
4098{
4099 return 0;
4100}
4101
Hugh Dickins24513262012-01-20 14:34:21 -08004102void shmem_unlock_mapping(struct address_space *mapping)
4103{
4104}
4105
Hugh Dickinsc01d5b32016-07-26 15:26:15 -07004106#ifdef CONFIG_MMU
4107unsigned long shmem_get_unmapped_area(struct file *file,
4108 unsigned long addr, unsigned long len,
4109 unsigned long pgoff, unsigned long flags)
4110{
4111 return current->mm->get_unmapped_area(file, addr, len, pgoff, flags);
4112}
4113#endif
4114
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004115void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
Hugh Dickins94c1e622011-06-27 16:18:03 -07004116{
Hugh Dickins41ffe5d2011-08-03 16:21:21 -07004117 truncate_inode_pages_range(inode->i_mapping, lstart, lend);
Hugh Dickins94c1e622011-06-27 16:18:03 -07004118}
4119EXPORT_SYMBOL_GPL(shmem_truncate_range);
4120
Hugh Dickins0b0a0802009-02-24 20:51:52 +00004121#define shmem_vm_ops generic_file_vm_ops
4122#define shmem_file_operations ramfs_file_operations
Dmitry Monakhov454abaf2010-03-04 17:32:18 +03004123#define shmem_get_inode(sb, dir, mode, dev, flags) ramfs_get_inode(sb, dir, mode, dev)
Hugh Dickins0b0a0802009-02-24 20:51:52 +00004124#define shmem_acct_size(flags, size) 0
4125#define shmem_unacct_size(flags, size) do {} while (0)
Matt Mackall853ac432009-01-06 14:40:20 -08004126
4127#endif /* CONFIG_SHMEM */
4128
4129/* common code */
Linus Torvalds1da177e2005-04-16 15:20:36 -07004130
Rasmus Villemoes19938e32016-10-07 17:01:01 -07004131static const struct dentry_operations anon_ops = {
Al Viro118b2302013-08-24 12:08:17 -04004132 .d_dname = simple_dname
Al Viro34515382013-02-14 22:38:02 -05004133};
4134
Eric Parisc7277092013-12-02 11:24:19 +00004135static struct file *__shmem_file_setup(const char *name, loff_t size,
4136 unsigned long flags, unsigned int i_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004137{
Al Viro6b4d0b22013-02-14 21:37:26 -05004138 struct file *res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004139 struct inode *inode;
Al Viro2c48b9c2009-08-09 00:52:35 +04004140 struct path path;
Al Viro34515382013-02-14 22:38:02 -05004141 struct super_block *sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004142 struct qstr this;
4143
4144 if (IS_ERR(shm_mnt))
Al Viro6b4d0b22013-02-14 21:37:26 -05004145 return ERR_CAST(shm_mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004146
Hugh Dickins285b2c42011-08-03 16:21:20 -07004147 if (size < 0 || size > MAX_LFS_FILESIZE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004148 return ERR_PTR(-EINVAL);
4149
4150 if (shmem_acct_size(flags, size))
4151 return ERR_PTR(-ENOMEM);
4152
Al Viro6b4d0b22013-02-14 21:37:26 -05004153 res = ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004154 this.name = name;
4155 this.len = strlen(name);
4156 this.hash = 0; /* will go */
Al Viro34515382013-02-14 22:38:02 -05004157 sb = shm_mnt->mnt_sb;
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004158 path.mnt = mntget(shm_mnt);
Al Viro34515382013-02-14 22:38:02 -05004159 path.dentry = d_alloc_pseudo(sb, &this);
Al Viro2c48b9c2009-08-09 00:52:35 +04004160 if (!path.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07004161 goto put_memory;
Al Viro34515382013-02-14 22:38:02 -05004162 d_set_d_op(path.dentry, &anon_ops);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004163
Al Viro6b4d0b22013-02-14 21:37:26 -05004164 res = ERR_PTR(-ENOSPC);
Al Viro34515382013-02-14 22:38:02 -05004165 inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004166 if (!inode)
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004167 goto put_memory;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004168
Eric Parisc7277092013-12-02 11:24:19 +00004169 inode->i_flags |= i_flags;
Al Viro2c48b9c2009-08-09 00:52:35 +04004170 d_instantiate(path.dentry, inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004171 inode->i_size = size;
Miklos Szeredi6d6b77f2011-10-28 14:13:28 +02004172 clear_nlink(inode); /* It is unlinked */
Al Viro26567cd2013-03-01 20:22:53 -05004173 res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
4174 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004175 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04004176
Al Viro6b4d0b22013-02-14 21:37:26 -05004177 res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
Al Viro4b42af82009-08-05 18:25:56 +04004178 &shmem_file_operations);
Al Viro6b4d0b22013-02-14 21:37:26 -05004179 if (IS_ERR(res))
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004180 goto put_path;
Al Viro4b42af82009-08-05 18:25:56 +04004181
Al Viro6b4d0b22013-02-14 21:37:26 -05004182 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004183
Linus Torvalds1da177e2005-04-16 15:20:36 -07004184put_memory:
4185 shmem_unacct_size(flags, size);
Konstantin Khlebnikov66ee4b82014-08-06 16:06:32 -07004186put_path:
4187 path_put(&path);
Al Viro6b4d0b22013-02-14 21:37:26 -05004188 return res;
Linus Torvalds1da177e2005-04-16 15:20:36 -07004189}
Eric Parisc7277092013-12-02 11:24:19 +00004190
4191/**
4192 * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
4193 * kernel internal. There will be NO LSM permission checks against the
4194 * underlying inode. So users of this interface must do LSM checks at a
Stephen Smalleye1832f22015-08-06 15:46:55 -07004195 * higher layer. The users are the big_key and shm implementations. LSM
4196 * checks are provided at the key or shm level rather than the inode.
Eric Parisc7277092013-12-02 11:24:19 +00004197 * @name: name for dentry (to be seen in /proc/<pid>/maps
4198 * @size: size to be set for the file
4199 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4200 */
4201struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
4202{
4203 return __shmem_file_setup(name, size, flags, S_PRIVATE);
4204}
4205
4206/**
4207 * shmem_file_setup - get an unlinked file living in tmpfs
4208 * @name: name for dentry (to be seen in /proc/<pid>/maps
4209 * @size: size to be set for the file
4210 * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
4211 */
4212struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
4213{
4214 return __shmem_file_setup(name, size, flags, 0);
4215}
Keith Packard395e0dd2008-06-20 00:08:06 -07004216EXPORT_SYMBOL_GPL(shmem_file_setup);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004217
Randy Dunlap46711812008-03-19 17:00:41 -07004218/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07004219 * shmem_zero_setup - setup a shared anonymous mapping
Linus Torvalds1da177e2005-04-16 15:20:36 -07004220 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
4221 */
4222int shmem_zero_setup(struct vm_area_struct *vma)
4223{
4224 struct file *file;
4225 loff_t size = vma->vm_end - vma->vm_start;
4226
Hugh Dickins66fc1302015-06-14 09:48:09 -07004227 /*
4228 * Cloning a new file under mmap_sem leads to a lock ordering conflict
4229 * between XFS directory reading and selinux: since this file is only
4230 * accessible to the user through its mapping, use S_PRIVATE flag to
4231 * bypass file security, in the same way as shmem_kernel_file_setup().
4232 */
4233 file = __shmem_file_setup("dev/zero", size, vma->vm_flags, S_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07004234 if (IS_ERR(file))
4235 return PTR_ERR(file);
4236
4237 if (vma->vm_file)
4238 fput(vma->vm_file);
4239 vma->vm_file = file;
4240 vma->vm_ops = &shmem_vm_ops;
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004241
Kirill A. Shutemove496cf32016-07-26 15:26:35 -07004242 if (IS_ENABLED(CONFIG_TRANSPARENT_HUGE_PAGECACHE) &&
Kirill A. Shutemovf3f0e1d2016-07-26 15:26:32 -07004243 ((vma->vm_start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK) <
4244 (vma->vm_end & HPAGE_PMD_MASK)) {
4245 khugepaged_enter(vma, vma->vm_flags);
4246 }
4247
Linus Torvalds1da177e2005-04-16 15:20:36 -07004248 return 0;
4249}
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004250
4251/**
4252 * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
4253 * @mapping: the page's address_space
4254 * @index: the page index
4255 * @gfp: the page allocator flags to use if allocating
4256 *
4257 * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
4258 * with any new page allocations done using the specified allocation flags.
4259 * But read_cache_page_gfp() uses the ->readpage() method: which does not
4260 * suit tmpfs, since it may have pages in swapcache, and needs to find those
4261 * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
4262 *
Hugh Dickins68da9f02011-07-25 17:12:34 -07004263 * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
4264 * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004265 */
4266struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
4267 pgoff_t index, gfp_t gfp)
4268{
Hugh Dickins68da9f02011-07-25 17:12:34 -07004269#ifdef CONFIG_SHMEM
4270 struct inode *inode = mapping->host;
Hugh Dickins9276aad2011-07-25 17:12:34 -07004271 struct page *page;
Hugh Dickins68da9f02011-07-25 17:12:34 -07004272 int error;
4273
4274 BUG_ON(mapping->a_ops != &shmem_aops);
Andres Lagar-Cavilla9e18eb22016-05-19 17:12:47 -07004275 error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE,
Mike Rapoportcfda0522017-02-22 15:43:37 -08004276 gfp, NULL, NULL, NULL);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004277 if (error)
4278 page = ERR_PTR(error);
4279 else
4280 unlock_page(page);
4281 return page;
4282#else
4283 /*
4284 * The tiny !SHMEM case uses ramfs without swap
4285 */
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004286 return read_cache_page_gfp(mapping, index, gfp);
Hugh Dickins68da9f02011-07-25 17:12:34 -07004287#endif
Hugh Dickinsd9d90e52011-06-27 16:18:04 -07004288}
4289EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);