blob: 253d205914bad382e48abfd49393f6fe87179646 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Resizable virtual memory filesystem for Linux.
3 *
4 * Copyright (C) 2000 Linus Torvalds.
5 * 2000 Transmeta Corp.
6 * 2000-2001 Christoph Rohland
7 * 2000-2001 SAP AG
8 * 2002 Red Hat Inc.
Hugh Dickins0edd73b2005-06-21 17:15:04 -07009 * Copyright (C) 2002-2005 Hugh Dickins.
10 * Copyright (C) 2002-2005 VERITAS Software Corporation.
Linus Torvalds1da177e2005-04-16 15:20:36 -070011 * Copyright (C) 2004 Andi Kleen, SuSE Labs
12 *
13 * Extended attribute support for tmpfs:
14 * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
15 * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
16 *
17 * This file is released under the GPL.
18 */
19
20/*
21 * This virtual memory filesystem is heavily based on the ramfs. It
22 * extends ramfs by the ability to use swap and honor resource limits
23 * which makes it a completely usable filesystem.
24 */
25
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/module.h>
27#include <linux/init.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/fs.h>
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070029#include <linux/xattr.h>
Christoph Hellwiga5694252007-07-17 04:04:28 -070030#include <linux/exportfs.h>
Andreas Gruenbacher39f02472006-09-29 02:01:35 -070031#include <linux/generic_acl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <linux/mm.h>
33#include <linux/mman.h>
34#include <linux/file.h>
35#include <linux/swap.h>
36#include <linux/pagemap.h>
37#include <linux/string.h>
38#include <linux/slab.h>
39#include <linux/backing-dev.h>
40#include <linux/shmem_fs.h>
41#include <linux/mount.h>
42#include <linux/writeback.h>
43#include <linux/vfs.h>
44#include <linux/blkdev.h>
45#include <linux/security.h>
46#include <linux/swapops.h>
47#include <linux/mempolicy.h>
48#include <linux/namei.h>
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +000049#include <linux/ctype.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070050#include <linux/migrate.h>
Christoph Lameterc1f60a52006-09-25 23:31:11 -070051#include <linux/highmem.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070052
Linus Torvalds1da177e2005-04-16 15:20:36 -070053#include <asm/uaccess.h>
54#include <asm/div64.h>
55#include <asm/pgtable.h>
56
57/* This magic number is used in glibc for posix shared memory */
58#define TMPFS_MAGIC 0x01021994
59
60#define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
61#define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
62#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
63
64#define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
65#define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
66
67#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
68
69/* info->flags needs VM_flags to handle pagein/truncate races efficiently */
70#define SHMEM_PAGEIN VM_READ
71#define SHMEM_TRUNCATE VM_WRITE
72
73/* Definition to limit shmem_truncate's steps between cond_rescheds */
74#define LATENCY_LIMIT 64
75
76/* Pretend that each entry is of this size in directory's i_size */
77#define BOGO_DIRENT_SIZE 20
78
Linus Torvalds1da177e2005-04-16 15:20:36 -070079/* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
80enum sgp_type {
81 SGP_QUICK, /* don't try more than file page cache lookup */
82 SGP_READ, /* don't exceed i_size, don't allocate page */
83 SGP_CACHE, /* don't exceed i_size, may allocate page */
84 SGP_WRITE, /* may exceed i_size, may allocate page */
Nick Piggin54cb8822007-07-19 01:46:59 -070085 SGP_FAULT, /* same as SGP_CACHE, return with page locked */
Linus Torvalds1da177e2005-04-16 15:20:36 -070086};
87
88static int shmem_getpage(struct inode *inode, unsigned long idx,
89 struct page **pagep, enum sgp_type sgp, int *type);
90
Al Viro6daa0e22005-10-21 03:18:50 -040091static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -070092{
93 /*
94 * The above definition of ENTRIES_PER_PAGE, and the use of
95 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
96 * might be reconsidered if it ever diverges from PAGE_SIZE.
Mel Gorman769848c2007-07-17 04:03:05 -070097 *
Mel Gormane12ba742007-10-16 01:25:52 -070098 * Mobility flags are masked out as swap vectors cannot move
Linus Torvalds1da177e2005-04-16 15:20:36 -070099 */
Mel Gormane12ba742007-10-16 01:25:52 -0700100 return alloc_pages((gfp_mask & ~GFP_MOVABLE_MASK) | __GFP_ZERO,
Mel Gorman769848c2007-07-17 04:03:05 -0700101 PAGE_CACHE_SHIFT-PAGE_SHIFT);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700102}
103
104static inline void shmem_dir_free(struct page *page)
105{
106 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
107}
108
109static struct page **shmem_dir_map(struct page *page)
110{
111 return (struct page **)kmap_atomic(page, KM_USER0);
112}
113
114static inline void shmem_dir_unmap(struct page **dir)
115{
116 kunmap_atomic(dir, KM_USER0);
117}
118
119static swp_entry_t *shmem_swp_map(struct page *page)
120{
121 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
122}
123
124static inline void shmem_swp_balance_unmap(void)
125{
126 /*
127 * When passing a pointer to an i_direct entry, to code which
128 * also handles indirect entries and so will shmem_swp_unmap,
129 * we must arrange for the preempt count to remain in balance.
130 * What kmap_atomic of a lowmem page does depends on config
131 * and architecture, so pretend to kmap_atomic some lowmem page.
132 */
133 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
134}
135
136static inline void shmem_swp_unmap(swp_entry_t *entry)
137{
138 kunmap_atomic(entry, KM_USER1);
139}
140
141static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
142{
143 return sb->s_fs_info;
144}
145
146/*
147 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
148 * for shared memory and for shared anonymous (/dev/zero) mappings
149 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
150 * consistent with the pre-accounting of private mappings ...
151 */
152static inline int shmem_acct_size(unsigned long flags, loff_t size)
153{
154 return (flags & VM_ACCOUNT)?
155 security_vm_enough_memory(VM_ACCT(size)): 0;
156}
157
158static inline void shmem_unacct_size(unsigned long flags, loff_t size)
159{
160 if (flags & VM_ACCOUNT)
161 vm_unacct_memory(VM_ACCT(size));
162}
163
164/*
165 * ... whereas tmpfs objects are accounted incrementally as
166 * pages are allocated, in order to allow huge sparse files.
167 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
168 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
169 */
170static inline int shmem_acct_block(unsigned long flags)
171{
172 return (flags & VM_ACCOUNT)?
173 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
174}
175
176static inline void shmem_unacct_blocks(unsigned long flags, long pages)
177{
178 if (!(flags & VM_ACCOUNT))
179 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
180}
181
Hugh Dickins759b9772007-03-05 00:30:28 -0800182static const struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700183static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800184static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800185static const struct inode_operations shmem_inode_operations;
186static const struct inode_operations shmem_dir_inode_operations;
187static const struct inode_operations shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188static struct vm_operations_struct shmem_vm_ops;
189
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700190static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 .ra_pages = 0, /* No readahead */
192 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
193 .unplug_io_fn = default_unplug_io_fn,
194};
195
196static LIST_HEAD(shmem_swaplist);
197static DEFINE_SPINLOCK(shmem_swaplist_lock);
198
199static void shmem_free_blocks(struct inode *inode, long pages)
200{
201 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700202 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 spin_lock(&sbinfo->stat_lock);
204 sbinfo->free_blocks += pages;
205 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
206 spin_unlock(&sbinfo->stat_lock);
207 }
208}
209
210/*
211 * shmem_recalc_inode - recalculate the size of an inode
212 *
213 * @inode: inode to recalc
214 *
215 * We have to calculate the free blocks since the mm can drop
216 * undirtied hole pages behind our back.
217 *
218 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
219 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
220 *
221 * It has to be called with the spinlock held.
222 */
223static void shmem_recalc_inode(struct inode *inode)
224{
225 struct shmem_inode_info *info = SHMEM_I(inode);
226 long freed;
227
228 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
229 if (freed > 0) {
230 info->alloced -= freed;
231 shmem_unacct_blocks(info->flags, freed);
232 shmem_free_blocks(inode, freed);
233 }
234}
235
236/*
237 * shmem_swp_entry - find the swap vector position in the info structure
238 *
239 * @info: info structure for the inode
240 * @index: index of the page to find
241 * @page: optional page to add to the structure. Has to be preset to
242 * all zeros
243 *
244 * If there is no space allocated yet it will return NULL when
245 * page is NULL, else it will use the page for the needed block,
246 * setting it to NULL on return to indicate that it has been used.
247 *
248 * The swap vector is organized the following way:
249 *
250 * There are SHMEM_NR_DIRECT entries directly stored in the
251 * shmem_inode_info structure. So small files do not need an addional
252 * allocation.
253 *
254 * For pages with index > SHMEM_NR_DIRECT there is the pointer
255 * i_indirect which points to a page which holds in the first half
256 * doubly indirect blocks, in the second half triple indirect blocks:
257 *
258 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
259 * following layout (for SHMEM_NR_DIRECT == 16):
260 *
261 * i_indirect -> dir --> 16-19
262 * | +-> 20-23
263 * |
264 * +-->dir2 --> 24-27
265 * | +-> 28-31
266 * | +-> 32-35
267 * | +-> 36-39
268 * |
269 * +-->dir3 --> 40-43
270 * +-> 44-47
271 * +-> 48-51
272 * +-> 52-55
273 */
274static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
275{
276 unsigned long offset;
277 struct page **dir;
278 struct page *subdir;
279
280 if (index < SHMEM_NR_DIRECT) {
281 shmem_swp_balance_unmap();
282 return info->i_direct+index;
283 }
284 if (!info->i_indirect) {
285 if (page) {
286 info->i_indirect = *page;
287 *page = NULL;
288 }
289 return NULL; /* need another page */
290 }
291
292 index -= SHMEM_NR_DIRECT;
293 offset = index % ENTRIES_PER_PAGE;
294 index /= ENTRIES_PER_PAGE;
295 dir = shmem_dir_map(info->i_indirect);
296
297 if (index >= ENTRIES_PER_PAGE/2) {
298 index -= ENTRIES_PER_PAGE/2;
299 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
300 index %= ENTRIES_PER_PAGE;
301 subdir = *dir;
302 if (!subdir) {
303 if (page) {
304 *dir = *page;
305 *page = NULL;
306 }
307 shmem_dir_unmap(dir);
308 return NULL; /* need another page */
309 }
310 shmem_dir_unmap(dir);
311 dir = shmem_dir_map(subdir);
312 }
313
314 dir += index;
315 subdir = *dir;
316 if (!subdir) {
317 if (!page || !(subdir = *page)) {
318 shmem_dir_unmap(dir);
319 return NULL; /* need a page */
320 }
321 *dir = subdir;
322 *page = NULL;
323 }
324 shmem_dir_unmap(dir);
325 return shmem_swp_map(subdir) + offset;
326}
327
328static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
329{
330 long incdec = value? 1: -1;
331
332 entry->val = value;
333 info->swapped += incdec;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700334 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
335 struct page *page = kmap_atomic_to_page(entry);
336 set_page_private(page, page_private(page) + incdec);
337 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338}
339
340/*
341 * shmem_swp_alloc - get the position of the swap entry for the page.
342 * If it does not exist allocate the entry.
343 *
344 * @info: info structure for the inode
345 * @index: index of the page to find
346 * @sgp: check and recheck i_size? skip allocation?
347 */
348static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
349{
350 struct inode *inode = &info->vfs_inode;
351 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
352 struct page *page = NULL;
353 swp_entry_t *entry;
354
355 if (sgp != SGP_WRITE &&
356 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
357 return ERR_PTR(-EINVAL);
358
359 while (!(entry = shmem_swp_entry(info, index, &page))) {
360 if (sgp == SGP_READ)
361 return shmem_swp_map(ZERO_PAGE(0));
362 /*
363 * Test free_blocks against 1 not 0, since we have 1 data
364 * page (and perhaps indirect index pages) yet to allocate:
365 * a waste to allocate index if we cannot allocate data.
366 */
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700367 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 spin_lock(&sbinfo->stat_lock);
369 if (sbinfo->free_blocks <= 1) {
370 spin_unlock(&sbinfo->stat_lock);
371 return ERR_PTR(-ENOSPC);
372 }
373 sbinfo->free_blocks--;
374 inode->i_blocks += BLOCKS_PER_PAGE;
375 spin_unlock(&sbinfo->stat_lock);
376 }
377
378 spin_unlock(&info->lock);
Mel Gorman769848c2007-07-17 04:03:05 -0700379 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping));
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700380 if (page)
381 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 spin_lock(&info->lock);
383
384 if (!page) {
385 shmem_free_blocks(inode, 1);
386 return ERR_PTR(-ENOMEM);
387 }
388 if (sgp != SGP_WRITE &&
389 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
390 entry = ERR_PTR(-EINVAL);
391 break;
392 }
393 if (info->next_index <= index)
394 info->next_index = index + 1;
395 }
396 if (page) {
397 /* another task gave its page, or truncated the file */
398 shmem_free_blocks(inode, 1);
399 shmem_dir_free(page);
400 }
401 if (info->next_index <= index && !IS_ERR(entry))
402 info->next_index = index + 1;
403 return entry;
404}
405
406/*
407 * shmem_free_swp - free some swap entries in a directory
408 *
Hugh Dickins1ae70002007-03-29 01:20:36 -0700409 * @dir: pointer to the directory
410 * @edir: pointer after last entry of the directory
411 * @punch_lock: pointer to spinlock when needed for the holepunch case
Linus Torvalds1da177e2005-04-16 15:20:36 -0700412 */
Hugh Dickins1ae70002007-03-29 01:20:36 -0700413static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
414 spinlock_t *punch_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415{
Hugh Dickins1ae70002007-03-29 01:20:36 -0700416 spinlock_t *punch_unlock = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 swp_entry_t *ptr;
418 int freed = 0;
419
420 for (ptr = dir; ptr < edir; ptr++) {
421 if (ptr->val) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700422 if (unlikely(punch_lock)) {
423 punch_unlock = punch_lock;
424 punch_lock = NULL;
425 spin_lock(punch_unlock);
426 if (!ptr->val)
427 continue;
428 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 free_swap_and_cache(*ptr);
430 *ptr = (swp_entry_t){0};
431 freed++;
432 }
433 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700434 if (punch_unlock)
435 spin_unlock(punch_unlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 return freed;
437}
438
Hugh Dickins1ae70002007-03-29 01:20:36 -0700439static int shmem_map_and_free_swp(struct page *subdir, int offset,
440 int limit, struct page ***dir, spinlock_t *punch_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441{
442 swp_entry_t *ptr;
443 int freed = 0;
444
445 ptr = shmem_swp_map(subdir);
446 for (; offset < limit; offset += LATENCY_LIMIT) {
447 int size = limit - offset;
448 if (size > LATENCY_LIMIT)
449 size = LATENCY_LIMIT;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700450 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
451 punch_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700452 if (need_resched()) {
453 shmem_swp_unmap(ptr);
454 if (*dir) {
455 shmem_dir_unmap(*dir);
456 *dir = NULL;
457 }
458 cond_resched();
459 ptr = shmem_swp_map(subdir);
460 }
461 }
462 shmem_swp_unmap(ptr);
463 return freed;
464}
465
466static void shmem_free_pages(struct list_head *next)
467{
468 struct page *page;
469 int freed = 0;
470
471 do {
472 page = container_of(next, struct page, lru);
473 next = next->next;
474 shmem_dir_free(page);
475 freed++;
476 if (freed >= LATENCY_LIMIT) {
477 cond_resched();
478 freed = 0;
479 }
480 } while (next);
481}
482
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800483static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484{
485 struct shmem_inode_info *info = SHMEM_I(inode);
486 unsigned long idx;
487 unsigned long size;
488 unsigned long limit;
489 unsigned long stage;
490 unsigned long diroff;
491 struct page **dir;
492 struct page *topdir;
493 struct page *middir;
494 struct page *subdir;
495 swp_entry_t *ptr;
496 LIST_HEAD(pages_to_free);
497 long nr_pages_to_free = 0;
498 long nr_swaps_freed = 0;
499 int offset;
500 int freed;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700501 int punch_hole;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700502 spinlock_t *needs_lock;
503 spinlock_t *punch_lock;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700504 unsigned long upper_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700505
506 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800507 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700508 if (idx >= info->next_index)
509 return;
510
511 spin_lock(&info->lock);
512 info->flags |= SHMEM_TRUNCATE;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800513 if (likely(end == (loff_t) -1)) {
514 limit = info->next_index;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700515 upper_limit = SHMEM_MAX_INDEX;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800516 info->next_index = idx;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700517 needs_lock = NULL;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700518 punch_hole = 0;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800519 } else {
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700520 if (end + 1 >= inode->i_size) { /* we may free a little more */
521 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
522 PAGE_CACHE_SHIFT;
523 upper_limit = SHMEM_MAX_INDEX;
524 } else {
525 limit = (end + 1) >> PAGE_CACHE_SHIFT;
526 upper_limit = limit;
527 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700528 needs_lock = &info->lock;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800529 punch_hole = 1;
530 }
531
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532 topdir = info->i_indirect;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800533 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 info->i_indirect = NULL;
535 nr_pages_to_free++;
536 list_add(&topdir->lru, &pages_to_free);
537 }
538 spin_unlock(&info->lock);
539
540 if (info->swapped && idx < SHMEM_NR_DIRECT) {
541 ptr = info->i_direct;
542 size = limit;
543 if (size > SHMEM_NR_DIRECT)
544 size = SHMEM_NR_DIRECT;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700545 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700546 }
Badari Pulavarty92a3d032006-12-22 01:06:23 -0800547
548 /*
549 * If there are no indirect blocks or we are punching a hole
550 * below indirect blocks, nothing to be done.
551 */
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700552 if (!topdir || limit <= SHMEM_NR_DIRECT)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 goto done2;
554
Hugh Dickins1ae70002007-03-29 01:20:36 -0700555 /*
556 * The truncation case has already dropped info->lock, and we're safe
557 * because i_size and next_index have already been lowered, preventing
558 * access beyond. But in the punch_hole case, we still need to take
559 * the lock when updating the swap directory, because there might be
560 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
561 * shmem_writepage. However, whenever we find we can remove a whole
562 * directory page (not at the misaligned start or end of the range),
563 * we first NULLify its pointer in the level above, and then have no
564 * need to take the lock when updating its contents: needs_lock and
565 * punch_lock (either pointing to info->lock or NULL) manage this.
566 */
567
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700568 upper_limit -= SHMEM_NR_DIRECT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 limit -= SHMEM_NR_DIRECT;
570 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
571 offset = idx % ENTRIES_PER_PAGE;
572 idx -= offset;
573
574 dir = shmem_dir_map(topdir);
575 stage = ENTRIES_PER_PAGEPAGE/2;
576 if (idx < ENTRIES_PER_PAGEPAGE/2) {
577 middir = topdir;
578 diroff = idx/ENTRIES_PER_PAGE;
579 } else {
580 dir += ENTRIES_PER_PAGE/2;
581 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
582 while (stage <= idx)
583 stage += ENTRIES_PER_PAGEPAGE;
584 middir = *dir;
585 if (*dir) {
586 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
587 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700588 if (!diroff && !offset && upper_limit >= stage) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700589 if (needs_lock) {
590 spin_lock(needs_lock);
591 *dir = NULL;
592 spin_unlock(needs_lock);
593 needs_lock = NULL;
594 } else
595 *dir = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 nr_pages_to_free++;
597 list_add(&middir->lru, &pages_to_free);
598 }
599 shmem_dir_unmap(dir);
600 dir = shmem_dir_map(middir);
601 } else {
602 diroff = 0;
603 offset = 0;
604 idx = stage;
605 }
606 }
607
608 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
609 if (unlikely(idx == stage)) {
610 shmem_dir_unmap(dir);
611 dir = shmem_dir_map(topdir) +
612 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
613 while (!*dir) {
614 dir++;
615 idx += ENTRIES_PER_PAGEPAGE;
616 if (idx >= limit)
617 goto done1;
618 }
619 stage = idx + ENTRIES_PER_PAGEPAGE;
620 middir = *dir;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700621 if (punch_hole)
622 needs_lock = &info->lock;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700623 if (upper_limit >= stage) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700624 if (needs_lock) {
625 spin_lock(needs_lock);
626 *dir = NULL;
627 spin_unlock(needs_lock);
628 needs_lock = NULL;
629 } else
630 *dir = NULL;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700631 nr_pages_to_free++;
632 list_add(&middir->lru, &pages_to_free);
633 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634 shmem_dir_unmap(dir);
635 cond_resched();
636 dir = shmem_dir_map(middir);
637 diroff = 0;
638 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700639 punch_lock = needs_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 subdir = dir[diroff];
Hugh Dickins1ae70002007-03-29 01:20:36 -0700641 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
642 if (needs_lock) {
643 spin_lock(needs_lock);
644 dir[diroff] = NULL;
645 spin_unlock(needs_lock);
646 punch_lock = NULL;
647 } else
648 dir[diroff] = NULL;
649 nr_pages_to_free++;
650 list_add(&subdir->lru, &pages_to_free);
651 }
652 if (subdir && page_private(subdir) /* has swap entries */) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700653 size = limit - idx;
654 if (size > ENTRIES_PER_PAGE)
655 size = ENTRIES_PER_PAGE;
656 freed = shmem_map_and_free_swp(subdir,
Hugh Dickins1ae70002007-03-29 01:20:36 -0700657 offset, size, &dir, punch_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 if (!dir)
659 dir = shmem_dir_map(middir);
660 nr_swaps_freed += freed;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700661 if (offset || punch_lock) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 spin_lock(&info->lock);
Hugh Dickins1ae70002007-03-29 01:20:36 -0700663 set_page_private(subdir,
664 page_private(subdir) - freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665 spin_unlock(&info->lock);
Hugh Dickins1ae70002007-03-29 01:20:36 -0700666 } else
667 BUG_ON(page_private(subdir) != freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700668 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700669 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 }
671done1:
672 shmem_dir_unmap(dir);
673done2:
674 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
675 /*
676 * Call truncate_inode_pages again: racing shmem_unuse_inode
677 * may have swizzled a page in from swap since vmtruncate or
678 * generic_delete_inode did it, before we lowered next_index.
679 * Also, though shmem_getpage checks i_size before adding to
680 * cache, no recheck after: so fix the narrow window there too.
Hugh Dickins16a10012007-03-29 01:20:37 -0700681 *
682 * Recalling truncate_inode_pages_range and unmap_mapping_range
683 * every time for punch_hole (which never got a chance to clear
684 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
685 * yet hardly ever necessary: try to optimize them out later.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700686 */
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800687 truncate_inode_pages_range(inode->i_mapping, start, end);
Hugh Dickins16a10012007-03-29 01:20:37 -0700688 if (punch_hole)
689 unmap_mapping_range(inode->i_mapping, start,
690 end - start, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 }
692
693 spin_lock(&info->lock);
694 info->flags &= ~SHMEM_TRUNCATE;
695 info->swapped -= nr_swaps_freed;
696 if (nr_pages_to_free)
697 shmem_free_blocks(inode, nr_pages_to_free);
698 shmem_recalc_inode(inode);
699 spin_unlock(&info->lock);
700
701 /*
702 * Empty swap vector directory pages to be freed?
703 */
704 if (!list_empty(&pages_to_free)) {
705 pages_to_free.prev->next = NULL;
706 shmem_free_pages(pages_to_free.next);
707 }
708}
709
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800710static void shmem_truncate(struct inode *inode)
711{
712 shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
713}
714
Linus Torvalds1da177e2005-04-16 15:20:36 -0700715static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
716{
717 struct inode *inode = dentry->d_inode;
718 struct page *page = NULL;
719 int error;
720
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700721 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 if (attr->ia_size < inode->i_size) {
723 /*
724 * If truncating down to a partial page, then
725 * if that page is already allocated, hold it
726 * in memory until the truncation is over, so
727 * truncate_partial_page cannnot miss it were
728 * it assigned to swap.
729 */
730 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
731 (void) shmem_getpage(inode,
732 attr->ia_size>>PAGE_CACHE_SHIFT,
733 &page, SGP_READ, NULL);
734 }
735 /*
736 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
737 * detect if any pages might have been added to cache
738 * after truncate_inode_pages. But we needn't bother
739 * if it's being fully truncated to zero-length: the
740 * nrpages check is efficient enough in that case.
741 */
742 if (attr->ia_size) {
743 struct shmem_inode_info *info = SHMEM_I(inode);
744 spin_lock(&info->lock);
745 info->flags &= ~SHMEM_PAGEIN;
746 spin_unlock(&info->lock);
747 }
748 }
749 }
750
751 error = inode_change_ok(inode, attr);
752 if (!error)
753 error = inode_setattr(inode, attr);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700754#ifdef CONFIG_TMPFS_POSIX_ACL
755 if (!error && (attr->ia_valid & ATTR_MODE))
756 error = generic_acl_chmod(inode, &shmem_acl_ops);
757#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700758 if (page)
759 page_cache_release(page);
760 return error;
761}
762
763static void shmem_delete_inode(struct inode *inode)
764{
765 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
766 struct shmem_inode_info *info = SHMEM_I(inode);
767
768 if (inode->i_op->truncate == shmem_truncate) {
Mark Fashehfef26652005-09-09 13:01:31 -0700769 truncate_inode_pages(inode->i_mapping, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 shmem_unacct_size(info->flags, inode->i_size);
771 inode->i_size = 0;
772 shmem_truncate(inode);
773 if (!list_empty(&info->swaplist)) {
774 spin_lock(&shmem_swaplist_lock);
775 list_del_init(&info->swaplist);
776 spin_unlock(&shmem_swaplist_lock);
777 }
778 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700779 BUG_ON(inode->i_blocks);
780 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 spin_lock(&sbinfo->stat_lock);
782 sbinfo->free_inodes++;
783 spin_unlock(&sbinfo->stat_lock);
784 }
785 clear_inode(inode);
786}
787
788static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
789{
790 swp_entry_t *ptr;
791
792 for (ptr = dir; ptr < edir; ptr++) {
793 if (ptr->val == entry.val)
794 return ptr - dir;
795 }
796 return -1;
797}
798
799static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
800{
801 struct inode *inode;
802 unsigned long idx;
803 unsigned long size;
804 unsigned long limit;
805 unsigned long stage;
806 struct page **dir;
807 struct page *subdir;
808 swp_entry_t *ptr;
809 int offset;
810
811 idx = 0;
812 ptr = info->i_direct;
813 spin_lock(&info->lock);
814 limit = info->next_index;
815 size = limit;
816 if (size > SHMEM_NR_DIRECT)
817 size = SHMEM_NR_DIRECT;
818 offset = shmem_find_swp(entry, ptr, ptr+size);
819 if (offset >= 0) {
820 shmem_swp_balance_unmap();
821 goto found;
822 }
823 if (!info->i_indirect)
824 goto lost2;
825
826 dir = shmem_dir_map(info->i_indirect);
827 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
828
829 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
830 if (unlikely(idx == stage)) {
831 shmem_dir_unmap(dir-1);
832 dir = shmem_dir_map(info->i_indirect) +
833 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
834 while (!*dir) {
835 dir++;
836 idx += ENTRIES_PER_PAGEPAGE;
837 if (idx >= limit)
838 goto lost1;
839 }
840 stage = idx + ENTRIES_PER_PAGEPAGE;
841 subdir = *dir;
842 shmem_dir_unmap(dir);
843 dir = shmem_dir_map(subdir);
844 }
845 subdir = *dir;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700846 if (subdir && page_private(subdir)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847 ptr = shmem_swp_map(subdir);
848 size = limit - idx;
849 if (size > ENTRIES_PER_PAGE)
850 size = ENTRIES_PER_PAGE;
851 offset = shmem_find_swp(entry, ptr, ptr+size);
852 if (offset >= 0) {
853 shmem_dir_unmap(dir);
854 goto found;
855 }
856 shmem_swp_unmap(ptr);
857 }
858 }
859lost1:
860 shmem_dir_unmap(dir-1);
861lost2:
862 spin_unlock(&info->lock);
863 return 0;
864found:
865 idx += offset;
866 inode = &info->vfs_inode;
867 if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
868 info->flags |= SHMEM_PAGEIN;
869 shmem_swp_set(info, ptr + offset, 0);
870 }
871 shmem_swp_unmap(ptr);
872 spin_unlock(&info->lock);
873 /*
874 * Decrement swap count even when the entry is left behind:
875 * try_to_unuse will skip over mms, then reincrement count.
876 */
877 swap_free(entry);
878 return 1;
879}
880
881/*
882 * shmem_unuse() search for an eventually swapped out shmem page.
883 */
884int shmem_unuse(swp_entry_t entry, struct page *page)
885{
886 struct list_head *p, *next;
887 struct shmem_inode_info *info;
888 int found = 0;
889
890 spin_lock(&shmem_swaplist_lock);
891 list_for_each_safe(p, next, &shmem_swaplist) {
892 info = list_entry(p, struct shmem_inode_info, swaplist);
893 if (!info->swapped)
894 list_del_init(&info->swaplist);
895 else if (shmem_unuse_inode(info, entry, page)) {
896 /* move head to start search for next from here */
897 list_move_tail(&shmem_swaplist, &info->swaplist);
898 found = 1;
899 break;
900 }
901 }
902 spin_unlock(&shmem_swaplist_lock);
903 return found;
904}
905
906/*
907 * Move the page from the page cache to the swap cache.
908 */
909static int shmem_writepage(struct page *page, struct writeback_control *wbc)
910{
911 struct shmem_inode_info *info;
912 swp_entry_t *entry, swap;
913 struct address_space *mapping;
914 unsigned long index;
915 struct inode *inode;
916
917 BUG_ON(!PageLocked(page));
Hugh Dickins487e9bf2007-10-29 14:37:20 -0700918 /*
919 * shmem_backing_dev_info's capabilities prevent regular writeback or
920 * sync from ever calling shmem_writepage; but a stacking filesystem
921 * may use the ->writepage of its underlying filesystem, in which case
922 * we want to do nothing when that underlying filesystem is tmpfs
923 * (writing out to swap is useful as a response to memory pressure, but
924 * of no use to stabilize the data) - just redirty the page, unlock it
925 * and claim success in this case. AOP_WRITEPAGE_ACTIVATE, and the
926 * page_mapped check below, must be avoided unless we're in reclaim.
927 */
928 if (!wbc->for_reclaim) {
929 set_page_dirty(page);
930 unlock_page(page);
931 return 0;
932 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933 BUG_ON(page_mapped(page));
934
935 mapping = page->mapping;
936 index = page->index;
937 inode = mapping->host;
938 info = SHMEM_I(inode);
939 if (info->flags & VM_LOCKED)
940 goto redirty;
941 swap = get_swap_page();
942 if (!swap.val)
943 goto redirty;
944
945 spin_lock(&info->lock);
946 shmem_recalc_inode(inode);
947 if (index >= info->next_index) {
948 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
949 goto unlock;
950 }
951 entry = shmem_swp_entry(info, index, NULL);
952 BUG_ON(!entry);
953 BUG_ON(entry->val);
954
955 if (move_to_swap_cache(page, swap) == 0) {
956 shmem_swp_set(info, entry, swap.val);
957 shmem_swp_unmap(entry);
958 spin_unlock(&info->lock);
959 if (list_empty(&info->swaplist)) {
960 spin_lock(&shmem_swaplist_lock);
961 /* move instead of add in case we're racing */
962 list_move_tail(&info->swaplist, &shmem_swaplist);
963 spin_unlock(&shmem_swaplist_lock);
964 }
965 unlock_page(page);
966 return 0;
967 }
968
969 shmem_swp_unmap(entry);
970unlock:
971 spin_unlock(&info->lock);
972 swap_free(swap);
973redirty:
974 set_page_dirty(page);
Zach Brown994fc28c2005-12-15 14:28:17 -0800975 return AOP_WRITEPAGE_ACTIVATE; /* Return with the page locked */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976}
977
978#ifdef CONFIG_NUMA
Hugh Dickinsd15c0232006-03-22 00:08:46 -0800979static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +0000980{
981 char *nodelist = strchr(value, ':');
982 int err = 1;
983
984 if (nodelist) {
985 /* NUL-terminate policy string */
986 *nodelist++ = '\0';
987 if (nodelist_parse(nodelist, *policy_nodes))
988 goto out;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -0700989 if (!nodes_subset(*policy_nodes, node_states[N_HIGH_MEMORY]))
Hugh Dickinsa2109062007-06-08 13:46:46 -0700990 goto out;
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +0000991 }
992 if (!strcmp(value, "default")) {
993 *policy = MPOL_DEFAULT;
994 /* Don't allow a nodelist */
995 if (!nodelist)
996 err = 0;
997 } else if (!strcmp(value, "prefer")) {
998 *policy = MPOL_PREFERRED;
999 /* Insist on a nodelist of one node only */
1000 if (nodelist) {
1001 char *rest = nodelist;
1002 while (isdigit(*rest))
1003 rest++;
1004 if (!*rest)
1005 err = 0;
1006 }
1007 } else if (!strcmp(value, "bind")) {
1008 *policy = MPOL_BIND;
1009 /* Insist on a nodelist */
1010 if (nodelist)
1011 err = 0;
1012 } else if (!strcmp(value, "interleave")) {
1013 *policy = MPOL_INTERLEAVE;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001014 /*
1015 * Default to online nodes with memory if no nodelist
1016 */
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00001017 if (!nodelist)
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001018 *policy_nodes = node_states[N_HIGH_MEMORY];
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00001019 err = 0;
1020 }
1021out:
1022 /* Restore string for error message */
1023 if (nodelist)
1024 *--nodelist = ':';
1025 return err;
1026}
1027
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028static struct page *shmem_swapin_async(struct shared_policy *p,
1029 swp_entry_t entry, unsigned long idx)
1030{
1031 struct page *page;
1032 struct vm_area_struct pvma;
1033
1034 /* Create a pseudo vma that just contains the policy */
1035 memset(&pvma, 0, sizeof(struct vm_area_struct));
1036 pvma.vm_end = PAGE_SIZE;
1037 pvma.vm_pgoff = idx;
1038 pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
1039 page = read_swap_cache_async(entry, &pvma, 0);
1040 mpol_free(pvma.vm_policy);
1041 return page;
1042}
1043
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001044static struct page *shmem_swapin(struct shmem_inode_info *info,
1045 swp_entry_t entry, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046{
1047 struct shared_policy *p = &info->policy;
1048 int i, num;
1049 struct page *page;
1050 unsigned long offset;
1051
1052 num = valid_swaphandles(entry, &offset);
1053 for (i = 0; i < num; offset++, i++) {
1054 page = shmem_swapin_async(p,
1055 swp_entry(swp_type(entry), offset), idx);
1056 if (!page)
1057 break;
1058 page_cache_release(page);
1059 }
1060 lru_add_drain(); /* Push any new pages onto the LRU now */
1061 return shmem_swapin_async(p, entry, idx);
1062}
1063
1064static struct page *
Al Viro6daa0e22005-10-21 03:18:50 -04001065shmem_alloc_page(gfp_t gfp, struct shmem_inode_info *info,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 unsigned long idx)
1067{
1068 struct vm_area_struct pvma;
1069 struct page *page;
1070
1071 memset(&pvma, 0, sizeof(struct vm_area_struct));
1072 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1073 pvma.vm_pgoff = idx;
1074 pvma.vm_end = PAGE_SIZE;
1075 page = alloc_page_vma(gfp | __GFP_ZERO, &pvma, 0);
1076 mpol_free(pvma.vm_policy);
1077 return page;
1078}
1079#else
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001080static inline int shmem_parse_mpol(char *value, int *policy,
1081 nodemask_t *policy_nodes)
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00001082{
1083 return 1;
1084}
1085
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086static inline struct page *
1087shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
1088{
1089 swapin_readahead(entry, 0, NULL);
1090 return read_swap_cache_async(entry, NULL, 0);
1091}
1092
1093static inline struct page *
Al Virodd0fc662005-10-07 07:46:04 +01001094shmem_alloc_page(gfp_t gfp,struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095{
1096 return alloc_page(gfp | __GFP_ZERO);
1097}
1098#endif
1099
1100/*
1101 * shmem_getpage - either get the page from swap or allocate a new one
1102 *
1103 * If we allocate a new one we do not mark it dirty. That's up to the
1104 * vm. If we swap it in we mark it dirty since we also free the swap
1105 * entry since a page cannot live in both the swap and page cache
1106 */
1107static int shmem_getpage(struct inode *inode, unsigned long idx,
1108 struct page **pagep, enum sgp_type sgp, int *type)
1109{
1110 struct address_space *mapping = inode->i_mapping;
1111 struct shmem_inode_info *info = SHMEM_I(inode);
1112 struct shmem_sb_info *sbinfo;
1113 struct page *filepage = *pagep;
1114 struct page *swappage;
1115 swp_entry_t *entry;
1116 swp_entry_t swap;
1117 int error;
1118
1119 if (idx >= SHMEM_MAX_INDEX)
1120 return -EFBIG;
Nick Piggin54cb8822007-07-19 01:46:59 -07001121
1122 if (type)
Nick Piggin83c54072007-07-19 01:47:05 -07001123 *type = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07001124
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 /*
1126 * Normally, filepage is NULL on entry, and either found
1127 * uptodate immediately, or allocated and zeroed, or read
1128 * in under swappage, which is then assigned to filepage.
Nick Piggin800d15a2007-10-16 01:25:03 -07001129 * But shmem_readpage and shmem_write_begin pass in a locked
Hugh Dickinsae976412007-06-04 10:00:39 +02001130 * filepage, which may be found not uptodate by other callers
1131 * too, and may need to be copied from the swappage read in.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132 */
1133repeat:
1134 if (!filepage)
1135 filepage = find_lock_page(mapping, idx);
1136 if (filepage && PageUptodate(filepage))
1137 goto done;
1138 error = 0;
1139 if (sgp == SGP_QUICK)
1140 goto failed;
1141
1142 spin_lock(&info->lock);
1143 shmem_recalc_inode(inode);
1144 entry = shmem_swp_alloc(info, idx, sgp);
1145 if (IS_ERR(entry)) {
1146 spin_unlock(&info->lock);
1147 error = PTR_ERR(entry);
1148 goto failed;
1149 }
1150 swap = *entry;
1151
1152 if (swap.val) {
1153 /* Look it up and read it in.. */
1154 swappage = lookup_swap_cache(swap);
1155 if (!swappage) {
1156 shmem_swp_unmap(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 /* here we actually do the io */
Nick Piggin83c54072007-07-19 01:47:05 -07001158 if (type && !(*type & VM_FAULT_MAJOR)) {
Christoph Lameterf8891e52006-06-30 01:55:45 -07001159 __count_vm_event(PGMAJFAULT);
Nick Piggin83c54072007-07-19 01:47:05 -07001160 *type |= VM_FAULT_MAJOR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 }
Christoph Lameterf8891e52006-06-30 01:55:45 -07001162 spin_unlock(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001163 swappage = shmem_swapin(info, swap, idx);
1164 if (!swappage) {
1165 spin_lock(&info->lock);
1166 entry = shmem_swp_alloc(info, idx, sgp);
1167 if (IS_ERR(entry))
1168 error = PTR_ERR(entry);
1169 else {
1170 if (entry->val == swap.val)
1171 error = -ENOMEM;
1172 shmem_swp_unmap(entry);
1173 }
1174 spin_unlock(&info->lock);
1175 if (error)
1176 goto failed;
1177 goto repeat;
1178 }
1179 wait_on_page_locked(swappage);
1180 page_cache_release(swappage);
1181 goto repeat;
1182 }
1183
1184 /* We have to do this with page locked to prevent races */
1185 if (TestSetPageLocked(swappage)) {
1186 shmem_swp_unmap(entry);
1187 spin_unlock(&info->lock);
1188 wait_on_page_locked(swappage);
1189 page_cache_release(swappage);
1190 goto repeat;
1191 }
1192 if (PageWriteback(swappage)) {
1193 shmem_swp_unmap(entry);
1194 spin_unlock(&info->lock);
1195 wait_on_page_writeback(swappage);
1196 unlock_page(swappage);
1197 page_cache_release(swappage);
1198 goto repeat;
1199 }
1200 if (!PageUptodate(swappage)) {
1201 shmem_swp_unmap(entry);
1202 spin_unlock(&info->lock);
1203 unlock_page(swappage);
1204 page_cache_release(swappage);
1205 error = -EIO;
1206 goto failed;
1207 }
1208
1209 if (filepage) {
1210 shmem_swp_set(info, entry, 0);
1211 shmem_swp_unmap(entry);
1212 delete_from_swap_cache(swappage);
1213 spin_unlock(&info->lock);
1214 copy_highpage(filepage, swappage);
1215 unlock_page(swappage);
1216 page_cache_release(swappage);
1217 flush_dcache_page(filepage);
1218 SetPageUptodate(filepage);
1219 set_page_dirty(filepage);
1220 swap_free(swap);
1221 } else if (!(error = move_from_swap_cache(
1222 swappage, idx, mapping))) {
1223 info->flags |= SHMEM_PAGEIN;
1224 shmem_swp_set(info, entry, 0);
1225 shmem_swp_unmap(entry);
1226 spin_unlock(&info->lock);
1227 filepage = swappage;
1228 swap_free(swap);
1229 } else {
1230 shmem_swp_unmap(entry);
1231 spin_unlock(&info->lock);
1232 unlock_page(swappage);
1233 page_cache_release(swappage);
1234 if (error == -ENOMEM) {
1235 /* let kswapd refresh zone for GFP_ATOMICs */
Andrew Morton3fcfab12006-10-19 23:28:16 -07001236 congestion_wait(WRITE, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 }
1238 goto repeat;
1239 }
1240 } else if (sgp == SGP_READ && !filepage) {
1241 shmem_swp_unmap(entry);
1242 filepage = find_get_page(mapping, idx);
1243 if (filepage &&
1244 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1245 spin_unlock(&info->lock);
1246 wait_on_page_locked(filepage);
1247 page_cache_release(filepage);
1248 filepage = NULL;
1249 goto repeat;
1250 }
1251 spin_unlock(&info->lock);
1252 } else {
1253 shmem_swp_unmap(entry);
1254 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001255 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 spin_lock(&sbinfo->stat_lock);
1257 if (sbinfo->free_blocks == 0 ||
1258 shmem_acct_block(info->flags)) {
1259 spin_unlock(&sbinfo->stat_lock);
1260 spin_unlock(&info->lock);
1261 error = -ENOSPC;
1262 goto failed;
1263 }
1264 sbinfo->free_blocks--;
1265 inode->i_blocks += BLOCKS_PER_PAGE;
1266 spin_unlock(&sbinfo->stat_lock);
1267 } else if (shmem_acct_block(info->flags)) {
1268 spin_unlock(&info->lock);
1269 error = -ENOSPC;
1270 goto failed;
1271 }
1272
1273 if (!filepage) {
1274 spin_unlock(&info->lock);
1275 filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
1276 info,
1277 idx);
1278 if (!filepage) {
1279 shmem_unacct_blocks(info->flags, 1);
1280 shmem_free_blocks(inode, 1);
1281 error = -ENOMEM;
1282 goto failed;
1283 }
1284
1285 spin_lock(&info->lock);
1286 entry = shmem_swp_alloc(info, idx, sgp);
1287 if (IS_ERR(entry))
1288 error = PTR_ERR(entry);
1289 else {
1290 swap = *entry;
1291 shmem_swp_unmap(entry);
1292 }
1293 if (error || swap.val || 0 != add_to_page_cache_lru(
1294 filepage, mapping, idx, GFP_ATOMIC)) {
1295 spin_unlock(&info->lock);
1296 page_cache_release(filepage);
1297 shmem_unacct_blocks(info->flags, 1);
1298 shmem_free_blocks(inode, 1);
1299 filepage = NULL;
1300 if (error)
1301 goto failed;
1302 goto repeat;
1303 }
1304 info->flags |= SHMEM_PAGEIN;
1305 }
1306
1307 info->alloced++;
1308 spin_unlock(&info->lock);
1309 flush_dcache_page(filepage);
1310 SetPageUptodate(filepage);
1311 }
1312done:
1313 if (*pagep != filepage) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 *pagep = filepage;
Nick Piggin54cb8822007-07-19 01:46:59 -07001315 if (sgp != SGP_FAULT)
Nick Piggind00806b2007-07-19 01:46:57 -07001316 unlock_page(filepage);
1317
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318 }
1319 return 0;
1320
1321failed:
1322 if (*pagep != filepage) {
1323 unlock_page(filepage);
1324 page_cache_release(filepage);
1325 }
1326 return error;
1327}
1328
Nick Piggind0217ac2007-07-19 01:47:03 -07001329static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001331 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 int error;
Nick Piggind0217ac2007-07-19 01:47:03 -07001333 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334
Nick Piggind0217ac2007-07-19 01:47:03 -07001335 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1336 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001337
Nick Piggind0217ac2007-07-19 01:47:03 -07001338 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_FAULT, &ret);
1339 if (error)
1340 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001341
Nick Piggind0217ac2007-07-19 01:47:03 -07001342 mark_page_accessed(vmf->page);
Nick Piggin83c54072007-07-19 01:47:05 -07001343 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344}
1345
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346#ifdef CONFIG_NUMA
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001347static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001349 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001350 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1351}
1352
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001353static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1354 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001356 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 unsigned long idx;
1358
1359 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1360 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1361}
1362#endif
1363
1364int shmem_lock(struct file *file, int lock, struct user_struct *user)
1365{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001366 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 struct shmem_inode_info *info = SHMEM_I(inode);
1368 int retval = -ENOMEM;
1369
1370 spin_lock(&info->lock);
1371 if (lock && !(info->flags & VM_LOCKED)) {
1372 if (!user_shm_lock(inode->i_size, user))
1373 goto out_nomem;
1374 info->flags |= VM_LOCKED;
1375 }
1376 if (!lock && (info->flags & VM_LOCKED) && user) {
1377 user_shm_unlock(inode->i_size, user);
1378 info->flags &= ~VM_LOCKED;
1379 }
1380 retval = 0;
1381out_nomem:
1382 spin_unlock(&info->lock);
1383 return retval;
1384}
1385
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08001386static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387{
1388 file_accessed(file);
1389 vma->vm_ops = &shmem_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001390 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 return 0;
1392}
1393
1394static struct inode *
1395shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1396{
1397 struct inode *inode;
1398 struct shmem_inode_info *info;
1399 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1400
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001401 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402 spin_lock(&sbinfo->stat_lock);
1403 if (!sbinfo->free_inodes) {
1404 spin_unlock(&sbinfo->stat_lock);
1405 return NULL;
1406 }
1407 sbinfo->free_inodes--;
1408 spin_unlock(&sbinfo->stat_lock);
1409 }
1410
1411 inode = new_inode(sb);
1412 if (inode) {
1413 inode->i_mode = mode;
1414 inode->i_uid = current->fsuid;
1415 inode->i_gid = current->fsgid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 inode->i_blocks = 0;
1417 inode->i_mapping->a_ops = &shmem_aops;
1418 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1419 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
David M. Grimes91828a42006-10-17 00:09:45 -07001420 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 info = SHMEM_I(inode);
1422 memset(info, 0, (char *)inode - (char *)info);
1423 spin_lock_init(&info->lock);
1424 INIT_LIST_HEAD(&info->swaplist);
1425
1426 switch (mode & S_IFMT) {
1427 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001428 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001429 init_special_inode(inode, mode, dev);
1430 break;
1431 case S_IFREG:
1432 inode->i_op = &shmem_inode_operations;
1433 inode->i_fop = &shmem_file_operations;
Robin Holt7339ff82006-01-14 13:20:48 -08001434 mpol_shared_policy_init(&info->policy, sbinfo->policy,
1435 &sbinfo->policy_nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 break;
1437 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07001438 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 /* Some things misbehave if size == 0 on a directory */
1440 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1441 inode->i_op = &shmem_dir_inode_operations;
1442 inode->i_fop = &simple_dir_operations;
1443 break;
1444 case S_IFLNK:
1445 /*
1446 * Must not load anything in the rbtree,
1447 * mpol_free_shared_policy will not be called.
1448 */
Robin Holt7339ff82006-01-14 13:20:48 -08001449 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT,
1450 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 break;
1452 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001453 } else if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 spin_lock(&sbinfo->stat_lock);
1455 sbinfo->free_inodes++;
1456 spin_unlock(&sbinfo->stat_lock);
1457 }
1458 return inode;
1459}
1460
1461#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001462static const struct inode_operations shmem_symlink_inode_operations;
1463static const struct inode_operations shmem_symlink_inline_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464
1465/*
Nick Piggin800d15a2007-10-16 01:25:03 -07001466 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
Hugh Dickinsae976412007-06-04 10:00:39 +02001467 * but providing them allows a tmpfs file to be used for splice, sendfile, and
1468 * below the loop driver, in the generic fashion that many filesystems support.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 */
Hugh Dickinsae976412007-06-04 10:00:39 +02001470static int shmem_readpage(struct file *file, struct page *page)
1471{
1472 struct inode *inode = page->mapping->host;
1473 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1474 unlock_page(page);
1475 return error;
1476}
1477
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478static int
Nick Piggin800d15a2007-10-16 01:25:03 -07001479shmem_write_begin(struct file *file, struct address_space *mapping,
1480 loff_t pos, unsigned len, unsigned flags,
1481 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482{
Nick Piggin800d15a2007-10-16 01:25:03 -07001483 struct inode *inode = mapping->host;
1484 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1485 *pagep = NULL;
1486 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1487}
1488
1489static int
1490shmem_write_end(struct file *file, struct address_space *mapping,
1491 loff_t pos, unsigned len, unsigned copied,
1492 struct page *page, void *fsdata)
1493{
1494 struct inode *inode = mapping->host;
1495
1496 set_page_dirty(page);
1497 page_cache_release(page);
1498
1499 if (pos+copied > inode->i_size)
1500 i_size_write(inode, pos+copied);
1501
1502 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503}
1504
1505static ssize_t
1506shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1507{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001508 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 loff_t pos;
1510 unsigned long written;
1511 ssize_t err;
1512
1513 if ((ssize_t) count < 0)
1514 return -EINVAL;
1515
1516 if (!access_ok(VERIFY_READ, buf, count))
1517 return -EFAULT;
1518
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001519 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520
1521 pos = *ppos;
1522 written = 0;
1523
1524 err = generic_write_checks(file, &pos, &count, 0);
1525 if (err || !count)
1526 goto out;
1527
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001528 err = remove_suid(file->f_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529 if (err)
1530 goto out;
1531
1532 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1533
1534 do {
1535 struct page *page = NULL;
1536 unsigned long bytes, index, offset;
1537 char *kaddr;
1538 int left;
1539
1540 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1541 index = pos >> PAGE_CACHE_SHIFT;
1542 bytes = PAGE_CACHE_SIZE - offset;
1543 if (bytes > count)
1544 bytes = count;
1545
1546 /*
1547 * We don't hold page lock across copy from user -
1548 * what would it guard against? - so no deadlock here.
1549 * But it still may be a good idea to prefault below.
1550 */
1551
1552 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1553 if (err)
1554 break;
1555
1556 left = bytes;
1557 if (PageHighMem(page)) {
1558 volatile unsigned char dummy;
1559 __get_user(dummy, buf);
1560 __get_user(dummy, buf + bytes - 1);
1561
1562 kaddr = kmap_atomic(page, KM_USER0);
1563 left = __copy_from_user_inatomic(kaddr + offset,
1564 buf, bytes);
1565 kunmap_atomic(kaddr, KM_USER0);
1566 }
1567 if (left) {
1568 kaddr = kmap(page);
1569 left = __copy_from_user(kaddr + offset, buf, bytes);
1570 kunmap(page);
1571 }
1572
1573 written += bytes;
1574 count -= bytes;
1575 pos += bytes;
1576 buf += bytes;
1577 if (pos > inode->i_size)
1578 i_size_write(inode, pos);
1579
1580 flush_dcache_page(page);
1581 set_page_dirty(page);
1582 mark_page_accessed(page);
1583 page_cache_release(page);
1584
1585 if (left) {
1586 pos -= left;
1587 written -= left;
1588 err = -EFAULT;
1589 break;
1590 }
1591
1592 /*
1593 * Our dirty pages are not counted in nr_dirty,
1594 * and we do not attempt to balance dirty pages.
1595 */
1596
1597 cond_resched();
1598 } while (count);
1599
1600 *ppos = pos;
1601 if (written)
1602 err = written;
1603out:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001604 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 return err;
1606}
1607
1608static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1609{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001610 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 struct address_space *mapping = inode->i_mapping;
1612 unsigned long index, offset;
1613
1614 index = *ppos >> PAGE_CACHE_SHIFT;
1615 offset = *ppos & ~PAGE_CACHE_MASK;
1616
1617 for (;;) {
1618 struct page *page = NULL;
1619 unsigned long end_index, nr, ret;
1620 loff_t i_size = i_size_read(inode);
1621
1622 end_index = i_size >> PAGE_CACHE_SHIFT;
1623 if (index > end_index)
1624 break;
1625 if (index == end_index) {
1626 nr = i_size & ~PAGE_CACHE_MASK;
1627 if (nr <= offset)
1628 break;
1629 }
1630
1631 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1632 if (desc->error) {
1633 if (desc->error == -EINVAL)
1634 desc->error = 0;
1635 break;
1636 }
1637
1638 /*
1639 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001640 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 */
1642 nr = PAGE_CACHE_SIZE;
1643 i_size = i_size_read(inode);
1644 end_index = i_size >> PAGE_CACHE_SHIFT;
1645 if (index == end_index) {
1646 nr = i_size & ~PAGE_CACHE_MASK;
1647 if (nr <= offset) {
1648 if (page)
1649 page_cache_release(page);
1650 break;
1651 }
1652 }
1653 nr -= offset;
1654
1655 if (page) {
1656 /*
1657 * If users can be writing to this page using arbitrary
1658 * virtual addresses, take care about potential aliasing
1659 * before reading the page on the kernel side.
1660 */
1661 if (mapping_writably_mapped(mapping))
1662 flush_dcache_page(page);
1663 /*
1664 * Mark the page accessed if we read the beginning.
1665 */
1666 if (!offset)
1667 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001668 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001669 page = ZERO_PAGE(0);
Nick Pigginb5810032005-10-29 18:16:12 -07001670 page_cache_get(page);
1671 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001672
1673 /*
1674 * Ok, we have the page, and it's up-to-date, so
1675 * now we can copy it to user space...
1676 *
1677 * The actor routine returns how many bytes were actually used..
1678 * NOTE! This may not be the same as how much of a user buffer
1679 * we filled up (we may be padding etc), so we can only update
1680 * "pos" here (the actor routine has to update the user buffer
1681 * pointers and the remaining count).
1682 */
1683 ret = actor(desc, page, offset, nr);
1684 offset += ret;
1685 index += offset >> PAGE_CACHE_SHIFT;
1686 offset &= ~PAGE_CACHE_MASK;
1687
1688 page_cache_release(page);
1689 if (ret != nr || !desc->count)
1690 break;
1691
1692 cond_resched();
1693 }
1694
1695 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1696 file_accessed(filp);
1697}
1698
1699static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1700{
1701 read_descriptor_t desc;
1702
1703 if ((ssize_t) count < 0)
1704 return -EINVAL;
1705 if (!access_ok(VERIFY_WRITE, buf, count))
1706 return -EFAULT;
1707 if (!count)
1708 return 0;
1709
1710 desc.written = 0;
1711 desc.count = count;
1712 desc.arg.buf = buf;
1713 desc.error = 0;
1714
1715 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1716 if (desc.written)
1717 return desc.written;
1718 return desc.error;
1719}
1720
David Howells726c3342006-06-23 02:02:58 -07001721static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722{
David Howells726c3342006-06-23 02:02:58 -07001723 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724
1725 buf->f_type = TMPFS_MAGIC;
1726 buf->f_bsize = PAGE_CACHE_SIZE;
1727 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001728 spin_lock(&sbinfo->stat_lock);
1729 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 buf->f_blocks = sbinfo->max_blocks;
1731 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001732 }
1733 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 buf->f_files = sbinfo->max_inodes;
1735 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 }
1737 /* else leave those fields 0 like simple_statfs */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001738 spin_unlock(&sbinfo->stat_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 return 0;
1740}
1741
1742/*
1743 * File creation. Allocate an inode, and we're done..
1744 */
1745static int
1746shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1747{
1748 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1749 int error = -ENOSPC;
1750
1751 if (inode) {
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001752 error = security_inode_init_security(inode, dir, NULL, NULL,
1753 NULL);
1754 if (error) {
1755 if (error != -EOPNOTSUPP) {
1756 iput(inode);
1757 return error;
1758 }
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001759 }
1760 error = shmem_acl_init(inode, dir);
1761 if (error) {
1762 iput(inode);
1763 return error;
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001764 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 if (dir->i_mode & S_ISGID) {
1766 inode->i_gid = dir->i_gid;
1767 if (S_ISDIR(mode))
1768 inode->i_mode |= S_ISGID;
1769 }
1770 dir->i_size += BOGO_DIRENT_SIZE;
1771 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1772 d_instantiate(dentry, inode);
1773 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 }
1775 return error;
1776}
1777
1778static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1779{
1780 int error;
1781
1782 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1783 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07001784 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 return 0;
1786}
1787
1788static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1789 struct nameidata *nd)
1790{
1791 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1792}
1793
1794/*
1795 * Link a file..
1796 */
1797static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1798{
1799 struct inode *inode = old_dentry->d_inode;
1800 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1801
1802 /*
1803 * No ordinary (disk based) filesystem counts links as inodes;
1804 * but each new link needs a new dentry, pinning lowmem, and
1805 * tmpfs dentries cannot be pruned until they are unlinked.
1806 */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001807 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 spin_lock(&sbinfo->stat_lock);
1809 if (!sbinfo->free_inodes) {
1810 spin_unlock(&sbinfo->stat_lock);
1811 return -ENOSPC;
1812 }
1813 sbinfo->free_inodes--;
1814 spin_unlock(&sbinfo->stat_lock);
1815 }
1816
1817 dir->i_size += BOGO_DIRENT_SIZE;
1818 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansend8c76e62006-09-30 23:29:04 -07001819 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 atomic_inc(&inode->i_count); /* New dentry reference */
1821 dget(dentry); /* Extra pinning count for the created dentry */
1822 d_instantiate(dentry, inode);
1823 return 0;
1824}
1825
1826static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1827{
1828 struct inode *inode = dentry->d_inode;
1829
1830 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) {
1831 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001832 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833 spin_lock(&sbinfo->stat_lock);
1834 sbinfo->free_inodes++;
1835 spin_unlock(&sbinfo->stat_lock);
1836 }
1837 }
1838
1839 dir->i_size -= BOGO_DIRENT_SIZE;
1840 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001841 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001842 dput(dentry); /* Undo the count from "create" - this does all the work */
1843 return 0;
1844}
1845
1846static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1847{
1848 if (!simple_empty(dentry))
1849 return -ENOTEMPTY;
1850
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001851 drop_nlink(dentry->d_inode);
1852 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 return shmem_unlink(dir, dentry);
1854}
1855
1856/*
1857 * The VFS layer already does all the dentry stuff for rename,
1858 * we just have to decrement the usage count for the target if
1859 * it exists so that the VFS layer correctly free's it when it
1860 * gets overwritten.
1861 */
1862static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1863{
1864 struct inode *inode = old_dentry->d_inode;
1865 int they_are_dirs = S_ISDIR(inode->i_mode);
1866
1867 if (!simple_empty(new_dentry))
1868 return -ENOTEMPTY;
1869
1870 if (new_dentry->d_inode) {
1871 (void) shmem_unlink(new_dir, new_dentry);
1872 if (they_are_dirs)
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001873 drop_nlink(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001874 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001875 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07001876 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877 }
1878
1879 old_dir->i_size -= BOGO_DIRENT_SIZE;
1880 new_dir->i_size += BOGO_DIRENT_SIZE;
1881 old_dir->i_ctime = old_dir->i_mtime =
1882 new_dir->i_ctime = new_dir->i_mtime =
1883 inode->i_ctime = CURRENT_TIME;
1884 return 0;
1885}
1886
1887static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1888{
1889 int error;
1890 int len;
1891 struct inode *inode;
1892 struct page *page = NULL;
1893 char *kaddr;
1894 struct shmem_inode_info *info;
1895
1896 len = strlen(symname) + 1;
1897 if (len > PAGE_CACHE_SIZE)
1898 return -ENAMETOOLONG;
1899
1900 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1901 if (!inode)
1902 return -ENOSPC;
1903
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001904 error = security_inode_init_security(inode, dir, NULL, NULL,
1905 NULL);
1906 if (error) {
1907 if (error != -EOPNOTSUPP) {
1908 iput(inode);
1909 return error;
1910 }
1911 error = 0;
1912 }
1913
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 info = SHMEM_I(inode);
1915 inode->i_size = len-1;
1916 if (len <= (char *)inode - (char *)info) {
1917 /* do it inline */
1918 memcpy(info, symname, len);
1919 inode->i_op = &shmem_symlink_inline_operations;
1920 } else {
1921 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1922 if (error) {
1923 iput(inode);
1924 return error;
1925 }
1926 inode->i_op = &shmem_symlink_inode_operations;
1927 kaddr = kmap_atomic(page, KM_USER0);
1928 memcpy(kaddr, symname, len);
1929 kunmap_atomic(kaddr, KM_USER0);
1930 set_page_dirty(page);
1931 page_cache_release(page);
1932 }
1933 if (dir->i_mode & S_ISGID)
1934 inode->i_gid = dir->i_gid;
1935 dir->i_size += BOGO_DIRENT_SIZE;
1936 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1937 d_instantiate(dentry, inode);
1938 dget(dentry);
1939 return 0;
1940}
1941
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001942static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943{
1944 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001945 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946}
1947
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001948static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949{
1950 struct page *page = NULL;
1951 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1952 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001953 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954}
1955
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001956static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957{
1958 if (!IS_ERR(nd_get_link(nd))) {
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001959 struct page *page = cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960 kunmap(page);
1961 mark_page_accessed(page);
1962 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963 }
1964}
1965
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001966static const struct inode_operations shmem_symlink_inline_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 .readlink = generic_readlink,
1968 .follow_link = shmem_follow_link_inline,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969};
1970
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001971static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 .truncate = shmem_truncate,
1973 .readlink = generic_readlink,
1974 .follow_link = shmem_follow_link,
1975 .put_link = shmem_put_link,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001976};
1977
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001978#ifdef CONFIG_TMPFS_POSIX_ACL
1979/**
1980 * Superblocks without xattr inode operations will get security.* xattr
1981 * support from the VFS "for free". As soon as we have any other xattrs
1982 * like ACLs, we also need to implement the security.* handlers at
1983 * filesystem level, though.
1984 */
1985
1986static size_t shmem_xattr_security_list(struct inode *inode, char *list,
1987 size_t list_len, const char *name,
1988 size_t name_len)
1989{
1990 return security_inode_listsecurity(inode, list, list_len);
1991}
1992
1993static int shmem_xattr_security_get(struct inode *inode, const char *name,
1994 void *buffer, size_t size)
1995{
1996 if (strcmp(name, "") == 0)
1997 return -EINVAL;
1998 return security_inode_getsecurity(inode, name, buffer, size,
1999 -EOPNOTSUPP);
2000}
2001
2002static int shmem_xattr_security_set(struct inode *inode, const char *name,
2003 const void *value, size_t size, int flags)
2004{
2005 if (strcmp(name, "") == 0)
2006 return -EINVAL;
2007 return security_inode_setsecurity(inode, name, value, size, flags);
2008}
2009
Adrian Bunk1f370a22006-12-06 20:38:21 -08002010static struct xattr_handler shmem_xattr_security_handler = {
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002011 .prefix = XATTR_SECURITY_PREFIX,
2012 .list = shmem_xattr_security_list,
2013 .get = shmem_xattr_security_get,
2014 .set = shmem_xattr_security_set,
2015};
2016
2017static struct xattr_handler *shmem_xattr_handlers[] = {
2018 &shmem_xattr_acl_access_handler,
2019 &shmem_xattr_acl_default_handler,
2020 &shmem_xattr_security_handler,
2021 NULL
2022};
2023#endif
2024
David M. Grimes91828a42006-10-17 00:09:45 -07002025static struct dentry *shmem_get_parent(struct dentry *child)
2026{
2027 return ERR_PTR(-ESTALE);
2028}
2029
2030static int shmem_match(struct inode *ino, void *vfh)
2031{
2032 __u32 *fh = vfh;
2033 __u64 inum = fh[2];
2034 inum = (inum << 32) | fh[1];
2035 return ino->i_ino == inum && fh[0] == ino->i_generation;
2036}
2037
Christoph Hellwig480b1162007-10-21 16:42:13 -07002038static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2039 struct fid *fid, int fh_len, int fh_type)
David M. Grimes91828a42006-10-17 00:09:45 -07002040{
David M. Grimes91828a42006-10-17 00:09:45 -07002041 struct inode *inode;
Christoph Hellwig480b1162007-10-21 16:42:13 -07002042 struct dentry *dentry = NULL;
2043 u64 inum = fid->raw[2];
2044 inum = (inum << 32) | fid->raw[1];
David M. Grimes91828a42006-10-17 00:09:45 -07002045
Christoph Hellwig480b1162007-10-21 16:42:13 -07002046 if (fh_len < 3)
2047 return NULL;
2048
2049 inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2050 shmem_match, fid->raw);
David M. Grimes91828a42006-10-17 00:09:45 -07002051 if (inode) {
Christoph Hellwig480b1162007-10-21 16:42:13 -07002052 dentry = d_find_alias(inode);
David M. Grimes91828a42006-10-17 00:09:45 -07002053 iput(inode);
2054 }
2055
Christoph Hellwig480b1162007-10-21 16:42:13 -07002056 return dentry;
David M. Grimes91828a42006-10-17 00:09:45 -07002057}
2058
2059static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2060 int connectable)
2061{
2062 struct inode *inode = dentry->d_inode;
2063
2064 if (*len < 3)
2065 return 255;
2066
2067 if (hlist_unhashed(&inode->i_hash)) {
2068 /* Unfortunately insert_inode_hash is not idempotent,
2069 * so as we hash inodes here rather than at creation
2070 * time, we need a lock to ensure we only try
2071 * to do it once
2072 */
2073 static DEFINE_SPINLOCK(lock);
2074 spin_lock(&lock);
2075 if (hlist_unhashed(&inode->i_hash))
2076 __insert_inode_hash(inode,
2077 inode->i_ino + inode->i_generation);
2078 spin_unlock(&lock);
2079 }
2080
2081 fh[0] = inode->i_generation;
2082 fh[1] = inode->i_ino;
2083 fh[2] = ((__u64)inode->i_ino) >> 32;
2084
2085 *len = 3;
2086 return 1;
2087}
2088
Christoph Hellwig39655162007-10-21 16:42:17 -07002089static const struct export_operations shmem_export_ops = {
David M. Grimes91828a42006-10-17 00:09:45 -07002090 .get_parent = shmem_get_parent,
David M. Grimes91828a42006-10-17 00:09:45 -07002091 .encode_fh = shmem_encode_fh,
Christoph Hellwig480b1162007-10-21 16:42:13 -07002092 .fh_to_dentry = shmem_fh_to_dentry,
David M. Grimes91828a42006-10-17 00:09:45 -07002093};
2094
Robin Holt7339ff82006-01-14 13:20:48 -08002095static int shmem_parse_options(char *options, int *mode, uid_t *uid,
2096 gid_t *gid, unsigned long *blocks, unsigned long *inodes,
2097 int *policy, nodemask_t *policy_nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098{
2099 char *this_char, *value, *rest;
2100
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002101 while (options != NULL) {
2102 this_char = options;
2103 for (;;) {
2104 /*
2105 * NUL-terminate this option: unfortunately,
2106 * mount options form a comma-separated list,
2107 * but mpol's nodelist may also contain commas.
2108 */
2109 options = strchr(options, ',');
2110 if (options == NULL)
2111 break;
2112 options++;
2113 if (!isdigit(*options)) {
2114 options[-1] = '\0';
2115 break;
2116 }
2117 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118 if (!*this_char)
2119 continue;
2120 if ((value = strchr(this_char,'=')) != NULL) {
2121 *value++ = 0;
2122 } else {
2123 printk(KERN_ERR
2124 "tmpfs: No value for mount option '%s'\n",
2125 this_char);
2126 return 1;
2127 }
2128
2129 if (!strcmp(this_char,"size")) {
2130 unsigned long long size;
2131 size = memparse(value,&rest);
2132 if (*rest == '%') {
2133 size <<= PAGE_SHIFT;
2134 size *= totalram_pages;
2135 do_div(size, 100);
2136 rest++;
2137 }
2138 if (*rest)
2139 goto bad_val;
2140 *blocks = size >> PAGE_CACHE_SHIFT;
2141 } else if (!strcmp(this_char,"nr_blocks")) {
2142 *blocks = memparse(value,&rest);
2143 if (*rest)
2144 goto bad_val;
2145 } else if (!strcmp(this_char,"nr_inodes")) {
2146 *inodes = memparse(value,&rest);
2147 if (*rest)
2148 goto bad_val;
2149 } else if (!strcmp(this_char,"mode")) {
2150 if (!mode)
2151 continue;
2152 *mode = simple_strtoul(value,&rest,8);
2153 if (*rest)
2154 goto bad_val;
2155 } else if (!strcmp(this_char,"uid")) {
2156 if (!uid)
2157 continue;
2158 *uid = simple_strtoul(value,&rest,0);
2159 if (*rest)
2160 goto bad_val;
2161 } else if (!strcmp(this_char,"gid")) {
2162 if (!gid)
2163 continue;
2164 *gid = simple_strtoul(value,&rest,0);
2165 if (*rest)
2166 goto bad_val;
Robin Holt7339ff82006-01-14 13:20:48 -08002167 } else if (!strcmp(this_char,"mpol")) {
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002168 if (shmem_parse_mpol(value,policy,policy_nodes))
Robin Holt7339ff82006-01-14 13:20:48 -08002169 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 } else {
2171 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2172 this_char);
2173 return 1;
2174 }
2175 }
2176 return 0;
2177
2178bad_val:
2179 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2180 value, this_char);
2181 return 1;
2182
2183}
2184
2185static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2186{
2187 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002188 unsigned long max_blocks = sbinfo->max_blocks;
2189 unsigned long max_inodes = sbinfo->max_inodes;
Robin Holt7339ff82006-01-14 13:20:48 -08002190 int policy = sbinfo->policy;
2191 nodemask_t policy_nodes = sbinfo->policy_nodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002192 unsigned long blocks;
2193 unsigned long inodes;
2194 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195
Robin Holt7339ff82006-01-14 13:20:48 -08002196 if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks,
2197 &max_inodes, &policy, &policy_nodes))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002198 return error;
2199
2200 spin_lock(&sbinfo->stat_lock);
2201 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2202 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2203 if (max_blocks < blocks)
2204 goto out;
2205 if (max_inodes < inodes)
2206 goto out;
2207 /*
2208 * Those tests also disallow limited->unlimited while any are in
2209 * use, so i_blocks will always be zero when max_blocks is zero;
2210 * but we must separately disallow unlimited->limited, because
2211 * in that case we have no record of how much is already in use.
2212 */
2213 if (max_blocks && !sbinfo->max_blocks)
2214 goto out;
2215 if (max_inodes && !sbinfo->max_inodes)
2216 goto out;
2217
2218 error = 0;
2219 sbinfo->max_blocks = max_blocks;
2220 sbinfo->free_blocks = max_blocks - blocks;
2221 sbinfo->max_inodes = max_inodes;
2222 sbinfo->free_inodes = max_inodes - inodes;
Robin Holt7339ff82006-01-14 13:20:48 -08002223 sbinfo->policy = policy;
2224 sbinfo->policy_nodes = policy_nodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002225out:
2226 spin_unlock(&sbinfo->stat_lock);
2227 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228}
2229#endif
2230
2231static void shmem_put_super(struct super_block *sb)
2232{
2233 kfree(sb->s_fs_info);
2234 sb->s_fs_info = NULL;
2235}
2236
Linus Torvalds1da177e2005-04-16 15:20:36 -07002237static int shmem_fill_super(struct super_block *sb,
2238 void *data, int silent)
2239{
2240 struct inode *inode;
2241 struct dentry *root;
2242 int mode = S_IRWXUGO | S_ISVTX;
2243 uid_t uid = current->fsuid;
2244 gid_t gid = current->fsgid;
2245 int err = -ENOMEM;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002246 struct shmem_sb_info *sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 unsigned long blocks = 0;
2248 unsigned long inodes = 0;
Robin Holt7339ff82006-01-14 13:20:48 -08002249 int policy = MPOL_DEFAULT;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002250 nodemask_t policy_nodes = node_states[N_HIGH_MEMORY];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002252#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 /*
2254 * Per default we only allow half of the physical ram per
2255 * tmpfs instance, limiting inodes to one per page of lowmem;
2256 * but the internal instance is left unlimited.
2257 */
2258 if (!(sb->s_flags & MS_NOUSER)) {
2259 blocks = totalram_pages / 2;
2260 inodes = totalram_pages - totalhigh_pages;
2261 if (inodes > blocks)
2262 inodes = blocks;
Robin Holt7339ff82006-01-14 13:20:48 -08002263 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks,
2264 &inodes, &policy, &policy_nodes))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265 return -EINVAL;
2266 }
David M. Grimes91828a42006-10-17 00:09:45 -07002267 sb->s_export_op = &shmem_export_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268#else
2269 sb->s_flags |= MS_NOUSER;
2270#endif
2271
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002272 /* Round up to L1_CACHE_BYTES to resist false sharing */
2273 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
2274 L1_CACHE_BYTES), GFP_KERNEL);
2275 if (!sbinfo)
2276 return -ENOMEM;
2277
2278 spin_lock_init(&sbinfo->stat_lock);
2279 sbinfo->max_blocks = blocks;
2280 sbinfo->free_blocks = blocks;
2281 sbinfo->max_inodes = inodes;
2282 sbinfo->free_inodes = inodes;
Robin Holt7339ff82006-01-14 13:20:48 -08002283 sbinfo->policy = policy;
2284 sbinfo->policy_nodes = policy_nodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002285
2286 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 sb->s_maxbytes = SHMEM_MAX_BYTES;
2288 sb->s_blocksize = PAGE_CACHE_SIZE;
2289 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2290 sb->s_magic = TMPFS_MAGIC;
2291 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01002292 sb->s_time_gran = 1;
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002293#ifdef CONFIG_TMPFS_POSIX_ACL
2294 sb->s_xattr = shmem_xattr_handlers;
2295 sb->s_flags |= MS_POSIXACL;
2296#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002297
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298 inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
2299 if (!inode)
2300 goto failed;
2301 inode->i_uid = uid;
2302 inode->i_gid = gid;
2303 root = d_alloc_root(inode);
2304 if (!root)
2305 goto failed_iput;
2306 sb->s_root = root;
2307 return 0;
2308
2309failed_iput:
2310 iput(inode);
2311failed:
2312 shmem_put_super(sb);
2313 return err;
2314}
2315
Pekka Enbergfcc234f2006-03-22 00:08:13 -08002316static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002317
2318static struct inode *shmem_alloc_inode(struct super_block *sb)
2319{
2320 struct shmem_inode_info *p;
Christoph Lametere94b1762006-12-06 20:33:17 -08002321 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322 if (!p)
2323 return NULL;
2324 return &p->vfs_inode;
2325}
2326
2327static void shmem_destroy_inode(struct inode *inode)
2328{
2329 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2330 /* only struct inode is valid if it's an inline symlink */
2331 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2332 }
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002333 shmem_acl_destroy_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2335}
2336
Christoph Lameter4ba9b9d2007-10-16 23:25:51 -07002337static void init_once(struct kmem_cache *cachep, void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338{
2339 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2340
Christoph Lametera35afb82007-05-16 22:10:57 -07002341 inode_init_once(&p->vfs_inode);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002342#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Lametera35afb82007-05-16 22:10:57 -07002343 p->i_acl = NULL;
2344 p->i_default_acl = NULL;
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002345#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002346}
2347
2348static int init_inodecache(void)
2349{
2350 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2351 sizeof(struct shmem_inode_info),
Alexey Dobriyan040b5c62007-10-16 23:26:10 -07002352 0, SLAB_PANIC, init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 return 0;
2354}
2355
2356static void destroy_inodecache(void)
2357{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07002358 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359}
2360
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002361static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08002363 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364#ifdef CONFIG_TMPFS
Hugh Dickinsae976412007-06-04 10:00:39 +02002365 .readpage = shmem_readpage,
Nick Piggin800d15a2007-10-16 01:25:03 -07002366 .write_begin = shmem_write_begin,
2367 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368#endif
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07002369 .migratepage = migrate_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370};
2371
Helge Deller15ad7cd2006-12-06 20:40:36 -08002372static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373 .mmap = shmem_mmap,
2374#ifdef CONFIG_TMPFS
2375 .llseek = generic_file_llseek,
2376 .read = shmem_file_read,
2377 .write = shmem_file_write,
2378 .fsync = simple_sync_file,
Hugh Dickinsae976412007-06-04 10:00:39 +02002379 .splice_read = generic_file_splice_read,
2380 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381#endif
2382};
2383
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002384static const struct inode_operations shmem_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385 .truncate = shmem_truncate,
2386 .setattr = shmem_notify_change,
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002387 .truncate_range = shmem_truncate_range,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002388#ifdef CONFIG_TMPFS_POSIX_ACL
2389 .setxattr = generic_setxattr,
2390 .getxattr = generic_getxattr,
2391 .listxattr = generic_listxattr,
2392 .removexattr = generic_removexattr,
2393 .permission = shmem_permission,
2394#endif
2395
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396};
2397
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002398static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399#ifdef CONFIG_TMPFS
2400 .create = shmem_create,
2401 .lookup = simple_lookup,
2402 .link = shmem_link,
2403 .unlink = shmem_unlink,
2404 .symlink = shmem_symlink,
2405 .mkdir = shmem_mkdir,
2406 .rmdir = shmem_rmdir,
2407 .mknod = shmem_mknod,
2408 .rename = shmem_rename,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002410#ifdef CONFIG_TMPFS_POSIX_ACL
2411 .setattr = shmem_notify_change,
2412 .setxattr = generic_setxattr,
2413 .getxattr = generic_getxattr,
2414 .listxattr = generic_listxattr,
2415 .removexattr = generic_removexattr,
2416 .permission = shmem_permission,
2417#endif
2418};
2419
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002420static const struct inode_operations shmem_special_inode_operations = {
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002421#ifdef CONFIG_TMPFS_POSIX_ACL
2422 .setattr = shmem_notify_change,
2423 .setxattr = generic_setxattr,
2424 .getxattr = generic_getxattr,
2425 .listxattr = generic_listxattr,
2426 .removexattr = generic_removexattr,
2427 .permission = shmem_permission,
2428#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002429};
2430
Hugh Dickins759b9772007-03-05 00:30:28 -08002431static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432 .alloc_inode = shmem_alloc_inode,
2433 .destroy_inode = shmem_destroy_inode,
2434#ifdef CONFIG_TMPFS
2435 .statfs = shmem_statfs,
2436 .remount_fs = shmem_remount_fs,
2437#endif
2438 .delete_inode = shmem_delete_inode,
2439 .drop_inode = generic_delete_inode,
2440 .put_super = shmem_put_super,
2441};
2442
2443static struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002444 .fault = shmem_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445#ifdef CONFIG_NUMA
2446 .set_policy = shmem_set_policy,
2447 .get_policy = shmem_get_policy,
2448#endif
2449};
2450
2451
David Howells454e2392006-06-23 02:02:57 -07002452static int shmem_get_sb(struct file_system_type *fs_type,
2453 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454{
David Howells454e2392006-06-23 02:02:57 -07002455 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002456}
2457
2458static struct file_system_type tmpfs_fs_type = {
2459 .owner = THIS_MODULE,
2460 .name = "tmpfs",
2461 .get_sb = shmem_get_sb,
2462 .kill_sb = kill_litter_super,
2463};
2464static struct vfsmount *shm_mnt;
2465
2466static int __init init_tmpfs(void)
2467{
2468 int error;
2469
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002470 error = bdi_init(&shmem_backing_dev_info);
2471 if (error)
2472 goto out4;
2473
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474 error = init_inodecache();
2475 if (error)
2476 goto out3;
2477
2478 error = register_filesystem(&tmpfs_fs_type);
2479 if (error) {
2480 printk(KERN_ERR "Could not register tmpfs\n");
2481 goto out2;
2482 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07002483
Trond Myklebust1f5ce9e2006-06-09 09:34:16 -04002484 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002485 tmpfs_fs_type.name, NULL);
2486 if (IS_ERR(shm_mnt)) {
2487 error = PTR_ERR(shm_mnt);
2488 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2489 goto out1;
2490 }
2491 return 0;
2492
2493out1:
2494 unregister_filesystem(&tmpfs_fs_type);
2495out2:
2496 destroy_inodecache();
2497out3:
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002498 bdi_destroy(&shmem_backing_dev_info);
2499out4:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500 shm_mnt = ERR_PTR(error);
2501 return error;
2502}
2503module_init(init_tmpfs)
2504
2505/*
2506 * shmem_file_setup - get an unlinked file living in tmpfs
2507 *
2508 * @name: name for dentry (to be seen in /proc/<pid>/maps
2509 * @size: size to be set for the file
2510 *
2511 */
2512struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2513{
2514 int error;
2515 struct file *file;
2516 struct inode *inode;
2517 struct dentry *dentry, *root;
2518 struct qstr this;
2519
2520 if (IS_ERR(shm_mnt))
2521 return (void *)shm_mnt;
2522
2523 if (size < 0 || size > SHMEM_MAX_BYTES)
2524 return ERR_PTR(-EINVAL);
2525
2526 if (shmem_acct_size(flags, size))
2527 return ERR_PTR(-ENOMEM);
2528
2529 error = -ENOMEM;
2530 this.name = name;
2531 this.len = strlen(name);
2532 this.hash = 0; /* will go */
2533 root = shm_mnt->mnt_root;
2534 dentry = d_alloc(root, &this);
2535 if (!dentry)
2536 goto put_memory;
2537
2538 error = -ENFILE;
2539 file = get_empty_filp();
2540 if (!file)
2541 goto put_dentry;
2542
2543 error = -ENOSPC;
2544 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2545 if (!inode)
2546 goto close_file;
2547
2548 SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2549 d_instantiate(dentry, inode);
2550 inode->i_size = size;
2551 inode->i_nlink = 0; /* It is unlinked */
Dave Hansence8d2cd2007-10-16 23:31:13 -07002552 init_file(file, shm_mnt, dentry, FMODE_WRITE | FMODE_READ,
2553 &shmem_file_operations);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554 return file;
2555
2556close_file:
2557 put_filp(file);
2558put_dentry:
2559 dput(dentry);
2560put_memory:
2561 shmem_unacct_size(flags, size);
2562 return ERR_PTR(error);
2563}
2564
2565/*
2566 * shmem_zero_setup - setup a shared anonymous mapping
2567 *
2568 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2569 */
2570int shmem_zero_setup(struct vm_area_struct *vma)
2571{
2572 struct file *file;
2573 loff_t size = vma->vm_end - vma->vm_start;
2574
2575 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2576 if (IS_ERR(file))
2577 return PTR_ERR(file);
2578
2579 if (vma->vm_file)
2580 fput(vma->vm_file);
2581 vma->vm_file = file;
2582 vma->vm_ops = &shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 return 0;
2584}