blob: b6aae2b33393d96d694145326c6591acf6f83b4b [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>
30#include <linux/generic_acl.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <linux/mm.h>
32#include <linux/mman.h>
33#include <linux/file.h>
34#include <linux/swap.h>
35#include <linux/pagemap.h>
36#include <linux/string.h>
37#include <linux/slab.h>
38#include <linux/backing-dev.h>
39#include <linux/shmem_fs.h>
40#include <linux/mount.h>
41#include <linux/writeback.h>
42#include <linux/vfs.h>
43#include <linux/blkdev.h>
44#include <linux/security.h>
45#include <linux/swapops.h>
46#include <linux/mempolicy.h>
47#include <linux/namei.h>
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +000048#include <linux/ctype.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070049#include <linux/migrate.h>
Christoph Lameterc1f60a52006-09-25 23:31:11 -070050#include <linux/highmem.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -070051#include <linux/backing-dev.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 */
85};
86
87static int shmem_getpage(struct inode *inode, unsigned long idx,
88 struct page **pagep, enum sgp_type sgp, int *type);
89
Al Viro6daa0e22005-10-21 03:18:50 -040090static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -070091{
92 /*
93 * The above definition of ENTRIES_PER_PAGE, and the use of
94 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
95 * might be reconsidered if it ever diverges from PAGE_SIZE.
96 */
97 return alloc_pages(gfp_mask, PAGE_CACHE_SHIFT-PAGE_SHIFT);
98}
99
100static inline void shmem_dir_free(struct page *page)
101{
102 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
103}
104
105static struct page **shmem_dir_map(struct page *page)
106{
107 return (struct page **)kmap_atomic(page, KM_USER0);
108}
109
110static inline void shmem_dir_unmap(struct page **dir)
111{
112 kunmap_atomic(dir, KM_USER0);
113}
114
115static swp_entry_t *shmem_swp_map(struct page *page)
116{
117 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
118}
119
120static inline void shmem_swp_balance_unmap(void)
121{
122 /*
123 * When passing a pointer to an i_direct entry, to code which
124 * also handles indirect entries and so will shmem_swp_unmap,
125 * we must arrange for the preempt count to remain in balance.
126 * What kmap_atomic of a lowmem page does depends on config
127 * and architecture, so pretend to kmap_atomic some lowmem page.
128 */
129 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
130}
131
132static inline void shmem_swp_unmap(swp_entry_t *entry)
133{
134 kunmap_atomic(entry, KM_USER1);
135}
136
137static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
138{
139 return sb->s_fs_info;
140}
141
142/*
143 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
144 * for shared memory and for shared anonymous (/dev/zero) mappings
145 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
146 * consistent with the pre-accounting of private mappings ...
147 */
148static inline int shmem_acct_size(unsigned long flags, loff_t size)
149{
150 return (flags & VM_ACCOUNT)?
151 security_vm_enough_memory(VM_ACCT(size)): 0;
152}
153
154static inline void shmem_unacct_size(unsigned long flags, loff_t size)
155{
156 if (flags & VM_ACCOUNT)
157 vm_unacct_memory(VM_ACCT(size));
158}
159
160/*
161 * ... whereas tmpfs objects are accounted incrementally as
162 * pages are allocated, in order to allow huge sparse files.
163 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
164 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
165 */
166static inline int shmem_acct_block(unsigned long flags)
167{
168 return (flags & VM_ACCOUNT)?
169 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
170}
171
172static inline void shmem_unacct_blocks(unsigned long flags, long pages)
173{
174 if (!(flags & VM_ACCOUNT))
175 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
176}
177
Hugh Dickins759b9772007-03-05 00:30:28 -0800178static const struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700179static const struct address_space_operations shmem_aops;
Helge Deller15ad7cd2006-12-06 20:40:36 -0800180static const struct file_operations shmem_file_operations;
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -0800181static const struct inode_operations shmem_inode_operations;
182static const struct inode_operations shmem_dir_inode_operations;
183static const struct inode_operations shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700184static struct vm_operations_struct shmem_vm_ops;
185
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700186static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 .ra_pages = 0, /* No readahead */
188 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
189 .unplug_io_fn = default_unplug_io_fn,
190};
191
192static LIST_HEAD(shmem_swaplist);
193static DEFINE_SPINLOCK(shmem_swaplist_lock);
194
195static void shmem_free_blocks(struct inode *inode, long pages)
196{
197 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700198 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199 spin_lock(&sbinfo->stat_lock);
200 sbinfo->free_blocks += pages;
201 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
202 spin_unlock(&sbinfo->stat_lock);
203 }
204}
205
206/*
207 * shmem_recalc_inode - recalculate the size of an inode
208 *
209 * @inode: inode to recalc
210 *
211 * We have to calculate the free blocks since the mm can drop
212 * undirtied hole pages behind our back.
213 *
214 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
215 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
216 *
217 * It has to be called with the spinlock held.
218 */
219static void shmem_recalc_inode(struct inode *inode)
220{
221 struct shmem_inode_info *info = SHMEM_I(inode);
222 long freed;
223
224 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
225 if (freed > 0) {
226 info->alloced -= freed;
227 shmem_unacct_blocks(info->flags, freed);
228 shmem_free_blocks(inode, freed);
229 }
230}
231
232/*
233 * shmem_swp_entry - find the swap vector position in the info structure
234 *
235 * @info: info structure for the inode
236 * @index: index of the page to find
237 * @page: optional page to add to the structure. Has to be preset to
238 * all zeros
239 *
240 * If there is no space allocated yet it will return NULL when
241 * page is NULL, else it will use the page for the needed block,
242 * setting it to NULL on return to indicate that it has been used.
243 *
244 * The swap vector is organized the following way:
245 *
246 * There are SHMEM_NR_DIRECT entries directly stored in the
247 * shmem_inode_info structure. So small files do not need an addional
248 * allocation.
249 *
250 * For pages with index > SHMEM_NR_DIRECT there is the pointer
251 * i_indirect which points to a page which holds in the first half
252 * doubly indirect blocks, in the second half triple indirect blocks:
253 *
254 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
255 * following layout (for SHMEM_NR_DIRECT == 16):
256 *
257 * i_indirect -> dir --> 16-19
258 * | +-> 20-23
259 * |
260 * +-->dir2 --> 24-27
261 * | +-> 28-31
262 * | +-> 32-35
263 * | +-> 36-39
264 * |
265 * +-->dir3 --> 40-43
266 * +-> 44-47
267 * +-> 48-51
268 * +-> 52-55
269 */
270static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
271{
272 unsigned long offset;
273 struct page **dir;
274 struct page *subdir;
275
276 if (index < SHMEM_NR_DIRECT) {
277 shmem_swp_balance_unmap();
278 return info->i_direct+index;
279 }
280 if (!info->i_indirect) {
281 if (page) {
282 info->i_indirect = *page;
283 *page = NULL;
284 }
285 return NULL; /* need another page */
286 }
287
288 index -= SHMEM_NR_DIRECT;
289 offset = index % ENTRIES_PER_PAGE;
290 index /= ENTRIES_PER_PAGE;
291 dir = shmem_dir_map(info->i_indirect);
292
293 if (index >= ENTRIES_PER_PAGE/2) {
294 index -= ENTRIES_PER_PAGE/2;
295 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
296 index %= ENTRIES_PER_PAGE;
297 subdir = *dir;
298 if (!subdir) {
299 if (page) {
300 *dir = *page;
301 *page = NULL;
302 }
303 shmem_dir_unmap(dir);
304 return NULL; /* need another page */
305 }
306 shmem_dir_unmap(dir);
307 dir = shmem_dir_map(subdir);
308 }
309
310 dir += index;
311 subdir = *dir;
312 if (!subdir) {
313 if (!page || !(subdir = *page)) {
314 shmem_dir_unmap(dir);
315 return NULL; /* need a page */
316 }
317 *dir = subdir;
318 *page = NULL;
319 }
320 shmem_dir_unmap(dir);
321 return shmem_swp_map(subdir) + offset;
322}
323
324static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
325{
326 long incdec = value? 1: -1;
327
328 entry->val = value;
329 info->swapped += incdec;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700330 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
331 struct page *page = kmap_atomic_to_page(entry);
332 set_page_private(page, page_private(page) + incdec);
333 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334}
335
336/*
337 * shmem_swp_alloc - get the position of the swap entry for the page.
338 * If it does not exist allocate the entry.
339 *
340 * @info: info structure for the inode
341 * @index: index of the page to find
342 * @sgp: check and recheck i_size? skip allocation?
343 */
344static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
345{
346 struct inode *inode = &info->vfs_inode;
347 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
348 struct page *page = NULL;
349 swp_entry_t *entry;
350
351 if (sgp != SGP_WRITE &&
352 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
353 return ERR_PTR(-EINVAL);
354
355 while (!(entry = shmem_swp_entry(info, index, &page))) {
356 if (sgp == SGP_READ)
357 return shmem_swp_map(ZERO_PAGE(0));
358 /*
359 * Test free_blocks against 1 not 0, since we have 1 data
360 * page (and perhaps indirect index pages) yet to allocate:
361 * a waste to allocate index if we cannot allocate data.
362 */
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700363 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 spin_lock(&sbinfo->stat_lock);
365 if (sbinfo->free_blocks <= 1) {
366 spin_unlock(&sbinfo->stat_lock);
367 return ERR_PTR(-ENOSPC);
368 }
369 sbinfo->free_blocks--;
370 inode->i_blocks += BLOCKS_PER_PAGE;
371 spin_unlock(&sbinfo->stat_lock);
372 }
373
374 spin_unlock(&info->lock);
375 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping) | __GFP_ZERO);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700376 if (page)
377 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700378 spin_lock(&info->lock);
379
380 if (!page) {
381 shmem_free_blocks(inode, 1);
382 return ERR_PTR(-ENOMEM);
383 }
384 if (sgp != SGP_WRITE &&
385 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
386 entry = ERR_PTR(-EINVAL);
387 break;
388 }
389 if (info->next_index <= index)
390 info->next_index = index + 1;
391 }
392 if (page) {
393 /* another task gave its page, or truncated the file */
394 shmem_free_blocks(inode, 1);
395 shmem_dir_free(page);
396 }
397 if (info->next_index <= index && !IS_ERR(entry))
398 info->next_index = index + 1;
399 return entry;
400}
401
402/*
403 * shmem_free_swp - free some swap entries in a directory
404 *
Hugh Dickins1ae70002007-03-29 01:20:36 -0700405 * @dir: pointer to the directory
406 * @edir: pointer after last entry of the directory
407 * @punch_lock: pointer to spinlock when needed for the holepunch case
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 */
Hugh Dickins1ae70002007-03-29 01:20:36 -0700409static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir,
410 spinlock_t *punch_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411{
Hugh Dickins1ae70002007-03-29 01:20:36 -0700412 spinlock_t *punch_unlock = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 swp_entry_t *ptr;
414 int freed = 0;
415
416 for (ptr = dir; ptr < edir; ptr++) {
417 if (ptr->val) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700418 if (unlikely(punch_lock)) {
419 punch_unlock = punch_lock;
420 punch_lock = NULL;
421 spin_lock(punch_unlock);
422 if (!ptr->val)
423 continue;
424 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 free_swap_and_cache(*ptr);
426 *ptr = (swp_entry_t){0};
427 freed++;
428 }
429 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700430 if (punch_unlock)
431 spin_unlock(punch_unlock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700432 return freed;
433}
434
Hugh Dickins1ae70002007-03-29 01:20:36 -0700435static int shmem_map_and_free_swp(struct page *subdir, int offset,
436 int limit, struct page ***dir, spinlock_t *punch_lock)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437{
438 swp_entry_t *ptr;
439 int freed = 0;
440
441 ptr = shmem_swp_map(subdir);
442 for (; offset < limit; offset += LATENCY_LIMIT) {
443 int size = limit - offset;
444 if (size > LATENCY_LIMIT)
445 size = LATENCY_LIMIT;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700446 freed += shmem_free_swp(ptr+offset, ptr+offset+size,
447 punch_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700448 if (need_resched()) {
449 shmem_swp_unmap(ptr);
450 if (*dir) {
451 shmem_dir_unmap(*dir);
452 *dir = NULL;
453 }
454 cond_resched();
455 ptr = shmem_swp_map(subdir);
456 }
457 }
458 shmem_swp_unmap(ptr);
459 return freed;
460}
461
462static void shmem_free_pages(struct list_head *next)
463{
464 struct page *page;
465 int freed = 0;
466
467 do {
468 page = container_of(next, struct page, lru);
469 next = next->next;
470 shmem_dir_free(page);
471 freed++;
472 if (freed >= LATENCY_LIMIT) {
473 cond_resched();
474 freed = 0;
475 }
476 } while (next);
477}
478
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800479static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480{
481 struct shmem_inode_info *info = SHMEM_I(inode);
482 unsigned long idx;
483 unsigned long size;
484 unsigned long limit;
485 unsigned long stage;
486 unsigned long diroff;
487 struct page **dir;
488 struct page *topdir;
489 struct page *middir;
490 struct page *subdir;
491 swp_entry_t *ptr;
492 LIST_HEAD(pages_to_free);
493 long nr_pages_to_free = 0;
494 long nr_swaps_freed = 0;
495 int offset;
496 int freed;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700497 int punch_hole;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700498 spinlock_t *needs_lock;
499 spinlock_t *punch_lock;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700500 unsigned long upper_limit;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700501
502 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800503 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700504 if (idx >= info->next_index)
505 return;
506
507 spin_lock(&info->lock);
508 info->flags |= SHMEM_TRUNCATE;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800509 if (likely(end == (loff_t) -1)) {
510 limit = info->next_index;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700511 upper_limit = SHMEM_MAX_INDEX;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800512 info->next_index = idx;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700513 needs_lock = NULL;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700514 punch_hole = 0;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800515 } else {
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700516 if (end + 1 >= inode->i_size) { /* we may free a little more */
517 limit = (inode->i_size + PAGE_CACHE_SIZE - 1) >>
518 PAGE_CACHE_SHIFT;
519 upper_limit = SHMEM_MAX_INDEX;
520 } else {
521 limit = (end + 1) >> PAGE_CACHE_SHIFT;
522 upper_limit = limit;
523 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700524 needs_lock = &info->lock;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800525 punch_hole = 1;
526 }
527
Linus Torvalds1da177e2005-04-16 15:20:36 -0700528 topdir = info->i_indirect;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800529 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530 info->i_indirect = NULL;
531 nr_pages_to_free++;
532 list_add(&topdir->lru, &pages_to_free);
533 }
534 spin_unlock(&info->lock);
535
536 if (info->swapped && idx < SHMEM_NR_DIRECT) {
537 ptr = info->i_direct;
538 size = limit;
539 if (size > SHMEM_NR_DIRECT)
540 size = SHMEM_NR_DIRECT;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700541 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size, needs_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700542 }
Badari Pulavarty92a3d032006-12-22 01:06:23 -0800543
544 /*
545 * If there are no indirect blocks or we are punching a hole
546 * below indirect blocks, nothing to be done.
547 */
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700548 if (!topdir || limit <= SHMEM_NR_DIRECT)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700549 goto done2;
550
Hugh Dickins1ae70002007-03-29 01:20:36 -0700551 /*
552 * The truncation case has already dropped info->lock, and we're safe
553 * because i_size and next_index have already been lowered, preventing
554 * access beyond. But in the punch_hole case, we still need to take
555 * the lock when updating the swap directory, because there might be
556 * racing accesses by shmem_getpage(SGP_CACHE), shmem_unuse_inode or
557 * shmem_writepage. However, whenever we find we can remove a whole
558 * directory page (not at the misaligned start or end of the range),
559 * we first NULLify its pointer in the level above, and then have no
560 * need to take the lock when updating its contents: needs_lock and
561 * punch_lock (either pointing to info->lock or NULL) manage this.
562 */
563
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700564 upper_limit -= SHMEM_NR_DIRECT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700565 limit -= SHMEM_NR_DIRECT;
566 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
567 offset = idx % ENTRIES_PER_PAGE;
568 idx -= offset;
569
570 dir = shmem_dir_map(topdir);
571 stage = ENTRIES_PER_PAGEPAGE/2;
572 if (idx < ENTRIES_PER_PAGEPAGE/2) {
573 middir = topdir;
574 diroff = idx/ENTRIES_PER_PAGE;
575 } else {
576 dir += ENTRIES_PER_PAGE/2;
577 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
578 while (stage <= idx)
579 stage += ENTRIES_PER_PAGEPAGE;
580 middir = *dir;
581 if (*dir) {
582 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
583 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700584 if (!diroff && !offset && upper_limit >= stage) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700585 if (needs_lock) {
586 spin_lock(needs_lock);
587 *dir = NULL;
588 spin_unlock(needs_lock);
589 needs_lock = NULL;
590 } else
591 *dir = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 nr_pages_to_free++;
593 list_add(&middir->lru, &pages_to_free);
594 }
595 shmem_dir_unmap(dir);
596 dir = shmem_dir_map(middir);
597 } else {
598 diroff = 0;
599 offset = 0;
600 idx = stage;
601 }
602 }
603
604 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
605 if (unlikely(idx == stage)) {
606 shmem_dir_unmap(dir);
607 dir = shmem_dir_map(topdir) +
608 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
609 while (!*dir) {
610 dir++;
611 idx += ENTRIES_PER_PAGEPAGE;
612 if (idx >= limit)
613 goto done1;
614 }
615 stage = idx + ENTRIES_PER_PAGEPAGE;
616 middir = *dir;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700617 if (punch_hole)
618 needs_lock = &info->lock;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700619 if (upper_limit >= stage) {
Hugh Dickins1ae70002007-03-29 01:20:36 -0700620 if (needs_lock) {
621 spin_lock(needs_lock);
622 *dir = NULL;
623 spin_unlock(needs_lock);
624 needs_lock = NULL;
625 } else
626 *dir = NULL;
Hugh Dickinsa2646d12007-03-29 01:20:35 -0700627 nr_pages_to_free++;
628 list_add(&middir->lru, &pages_to_free);
629 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700630 shmem_dir_unmap(dir);
631 cond_resched();
632 dir = shmem_dir_map(middir);
633 diroff = 0;
634 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700635 punch_lock = needs_lock;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700636 subdir = dir[diroff];
Hugh Dickins1ae70002007-03-29 01:20:36 -0700637 if (subdir && !offset && upper_limit-idx >= ENTRIES_PER_PAGE) {
638 if (needs_lock) {
639 spin_lock(needs_lock);
640 dir[diroff] = NULL;
641 spin_unlock(needs_lock);
642 punch_lock = NULL;
643 } else
644 dir[diroff] = NULL;
645 nr_pages_to_free++;
646 list_add(&subdir->lru, &pages_to_free);
647 }
648 if (subdir && page_private(subdir) /* has swap entries */) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 size = limit - idx;
650 if (size > ENTRIES_PER_PAGE)
651 size = ENTRIES_PER_PAGE;
652 freed = shmem_map_and_free_swp(subdir,
Hugh Dickins1ae70002007-03-29 01:20:36 -0700653 offset, size, &dir, punch_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654 if (!dir)
655 dir = shmem_dir_map(middir);
656 nr_swaps_freed += freed;
Hugh Dickins1ae70002007-03-29 01:20:36 -0700657 if (offset || punch_lock) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658 spin_lock(&info->lock);
Hugh Dickins1ae70002007-03-29 01:20:36 -0700659 set_page_private(subdir,
660 page_private(subdir) - freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 spin_unlock(&info->lock);
Hugh Dickins1ae70002007-03-29 01:20:36 -0700662 } else
663 BUG_ON(page_private(subdir) != freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 }
Hugh Dickins1ae70002007-03-29 01:20:36 -0700665 offset = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 }
667done1:
668 shmem_dir_unmap(dir);
669done2:
670 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
671 /*
672 * Call truncate_inode_pages again: racing shmem_unuse_inode
673 * may have swizzled a page in from swap since vmtruncate or
674 * generic_delete_inode did it, before we lowered next_index.
675 * Also, though shmem_getpage checks i_size before adding to
676 * cache, no recheck after: so fix the narrow window there too.
Hugh Dickins16a10012007-03-29 01:20:37 -0700677 *
678 * Recalling truncate_inode_pages_range and unmap_mapping_range
679 * every time for punch_hole (which never got a chance to clear
680 * SHMEM_PAGEIN at the start of vmtruncate_range) is expensive,
681 * yet hardly ever necessary: try to optimize them out later.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 */
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800683 truncate_inode_pages_range(inode->i_mapping, start, end);
Hugh Dickins16a10012007-03-29 01:20:37 -0700684 if (punch_hole)
685 unmap_mapping_range(inode->i_mapping, start,
686 end - start, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700687 }
688
689 spin_lock(&info->lock);
690 info->flags &= ~SHMEM_TRUNCATE;
691 info->swapped -= nr_swaps_freed;
692 if (nr_pages_to_free)
693 shmem_free_blocks(inode, nr_pages_to_free);
694 shmem_recalc_inode(inode);
695 spin_unlock(&info->lock);
696
697 /*
698 * Empty swap vector directory pages to be freed?
699 */
700 if (!list_empty(&pages_to_free)) {
701 pages_to_free.prev->next = NULL;
702 shmem_free_pages(pages_to_free.next);
703 }
704}
705
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800706static void shmem_truncate(struct inode *inode)
707{
708 shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
709}
710
Linus Torvalds1da177e2005-04-16 15:20:36 -0700711static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
712{
713 struct inode *inode = dentry->d_inode;
714 struct page *page = NULL;
715 int error;
716
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700717 if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718 if (attr->ia_size < inode->i_size) {
719 /*
720 * If truncating down to a partial page, then
721 * if that page is already allocated, hold it
722 * in memory until the truncation is over, so
723 * truncate_partial_page cannnot miss it were
724 * it assigned to swap.
725 */
726 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
727 (void) shmem_getpage(inode,
728 attr->ia_size>>PAGE_CACHE_SHIFT,
729 &page, SGP_READ, NULL);
730 }
731 /*
732 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
733 * detect if any pages might have been added to cache
734 * after truncate_inode_pages. But we needn't bother
735 * if it's being fully truncated to zero-length: the
736 * nrpages check is efficient enough in that case.
737 */
738 if (attr->ia_size) {
739 struct shmem_inode_info *info = SHMEM_I(inode);
740 spin_lock(&info->lock);
741 info->flags &= ~SHMEM_PAGEIN;
742 spin_unlock(&info->lock);
743 }
744 }
745 }
746
747 error = inode_change_ok(inode, attr);
748 if (!error)
749 error = inode_setattr(inode, attr);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -0700750#ifdef CONFIG_TMPFS_POSIX_ACL
751 if (!error && (attr->ia_valid & ATTR_MODE))
752 error = generic_acl_chmod(inode, &shmem_acl_ops);
753#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 if (page)
755 page_cache_release(page);
756 return error;
757}
758
759static void shmem_delete_inode(struct inode *inode)
760{
761 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
762 struct shmem_inode_info *info = SHMEM_I(inode);
763
764 if (inode->i_op->truncate == shmem_truncate) {
Mark Fashehfef26652005-09-09 13:01:31 -0700765 truncate_inode_pages(inode->i_mapping, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700766 shmem_unacct_size(info->flags, inode->i_size);
767 inode->i_size = 0;
768 shmem_truncate(inode);
769 if (!list_empty(&info->swaplist)) {
770 spin_lock(&shmem_swaplist_lock);
771 list_del_init(&info->swaplist);
772 spin_unlock(&shmem_swaplist_lock);
773 }
774 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700775 BUG_ON(inode->i_blocks);
776 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 spin_lock(&sbinfo->stat_lock);
778 sbinfo->free_inodes++;
779 spin_unlock(&sbinfo->stat_lock);
780 }
781 clear_inode(inode);
782}
783
784static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
785{
786 swp_entry_t *ptr;
787
788 for (ptr = dir; ptr < edir; ptr++) {
789 if (ptr->val == entry.val)
790 return ptr - dir;
791 }
792 return -1;
793}
794
795static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
796{
797 struct inode *inode;
798 unsigned long idx;
799 unsigned long size;
800 unsigned long limit;
801 unsigned long stage;
802 struct page **dir;
803 struct page *subdir;
804 swp_entry_t *ptr;
805 int offset;
806
807 idx = 0;
808 ptr = info->i_direct;
809 spin_lock(&info->lock);
810 limit = info->next_index;
811 size = limit;
812 if (size > SHMEM_NR_DIRECT)
813 size = SHMEM_NR_DIRECT;
814 offset = shmem_find_swp(entry, ptr, ptr+size);
815 if (offset >= 0) {
816 shmem_swp_balance_unmap();
817 goto found;
818 }
819 if (!info->i_indirect)
820 goto lost2;
821
822 dir = shmem_dir_map(info->i_indirect);
823 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
824
825 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
826 if (unlikely(idx == stage)) {
827 shmem_dir_unmap(dir-1);
828 dir = shmem_dir_map(info->i_indirect) +
829 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
830 while (!*dir) {
831 dir++;
832 idx += ENTRIES_PER_PAGEPAGE;
833 if (idx >= limit)
834 goto lost1;
835 }
836 stage = idx + ENTRIES_PER_PAGEPAGE;
837 subdir = *dir;
838 shmem_dir_unmap(dir);
839 dir = shmem_dir_map(subdir);
840 }
841 subdir = *dir;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700842 if (subdir && page_private(subdir)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 ptr = shmem_swp_map(subdir);
844 size = limit - idx;
845 if (size > ENTRIES_PER_PAGE)
846 size = ENTRIES_PER_PAGE;
847 offset = shmem_find_swp(entry, ptr, ptr+size);
848 if (offset >= 0) {
849 shmem_dir_unmap(dir);
850 goto found;
851 }
852 shmem_swp_unmap(ptr);
853 }
854 }
855lost1:
856 shmem_dir_unmap(dir-1);
857lost2:
858 spin_unlock(&info->lock);
859 return 0;
860found:
861 idx += offset;
862 inode = &info->vfs_inode;
863 if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
864 info->flags |= SHMEM_PAGEIN;
865 shmem_swp_set(info, ptr + offset, 0);
866 }
867 shmem_swp_unmap(ptr);
868 spin_unlock(&info->lock);
869 /*
870 * Decrement swap count even when the entry is left behind:
871 * try_to_unuse will skip over mms, then reincrement count.
872 */
873 swap_free(entry);
874 return 1;
875}
876
877/*
878 * shmem_unuse() search for an eventually swapped out shmem page.
879 */
880int shmem_unuse(swp_entry_t entry, struct page *page)
881{
882 struct list_head *p, *next;
883 struct shmem_inode_info *info;
884 int found = 0;
885
886 spin_lock(&shmem_swaplist_lock);
887 list_for_each_safe(p, next, &shmem_swaplist) {
888 info = list_entry(p, struct shmem_inode_info, swaplist);
889 if (!info->swapped)
890 list_del_init(&info->swaplist);
891 else if (shmem_unuse_inode(info, entry, page)) {
892 /* move head to start search for next from here */
893 list_move_tail(&shmem_swaplist, &info->swaplist);
894 found = 1;
895 break;
896 }
897 }
898 spin_unlock(&shmem_swaplist_lock);
899 return found;
900}
901
902/*
903 * Move the page from the page cache to the swap cache.
904 */
905static int shmem_writepage(struct page *page, struct writeback_control *wbc)
906{
907 struct shmem_inode_info *info;
908 swp_entry_t *entry, swap;
909 struct address_space *mapping;
910 unsigned long index;
911 struct inode *inode;
912
913 BUG_ON(!PageLocked(page));
914 BUG_ON(page_mapped(page));
915
916 mapping = page->mapping;
917 index = page->index;
918 inode = mapping->host;
919 info = SHMEM_I(inode);
920 if (info->flags & VM_LOCKED)
921 goto redirty;
922 swap = get_swap_page();
923 if (!swap.val)
924 goto redirty;
925
926 spin_lock(&info->lock);
927 shmem_recalc_inode(inode);
928 if (index >= info->next_index) {
929 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
930 goto unlock;
931 }
932 entry = shmem_swp_entry(info, index, NULL);
933 BUG_ON(!entry);
934 BUG_ON(entry->val);
935
936 if (move_to_swap_cache(page, swap) == 0) {
937 shmem_swp_set(info, entry, swap.val);
938 shmem_swp_unmap(entry);
939 spin_unlock(&info->lock);
940 if (list_empty(&info->swaplist)) {
941 spin_lock(&shmem_swaplist_lock);
942 /* move instead of add in case we're racing */
943 list_move_tail(&info->swaplist, &shmem_swaplist);
944 spin_unlock(&shmem_swaplist_lock);
945 }
946 unlock_page(page);
947 return 0;
948 }
949
950 shmem_swp_unmap(entry);
951unlock:
952 spin_unlock(&info->lock);
953 swap_free(swap);
954redirty:
955 set_page_dirty(page);
Zach Brown994fc28c2005-12-15 14:28:17 -0800956 return AOP_WRITEPAGE_ACTIVATE; /* Return with the page locked */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957}
958
959#ifdef CONFIG_NUMA
Hugh Dickinsd15c0232006-03-22 00:08:46 -0800960static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +0000961{
962 char *nodelist = strchr(value, ':');
963 int err = 1;
964
965 if (nodelist) {
966 /* NUL-terminate policy string */
967 *nodelist++ = '\0';
968 if (nodelist_parse(nodelist, *policy_nodes))
969 goto out;
Hugh Dickinsa2109062007-06-08 13:46:46 -0700970 if (!nodes_subset(*policy_nodes, node_online_map))
971 goto out;
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +0000972 }
973 if (!strcmp(value, "default")) {
974 *policy = MPOL_DEFAULT;
975 /* Don't allow a nodelist */
976 if (!nodelist)
977 err = 0;
978 } else if (!strcmp(value, "prefer")) {
979 *policy = MPOL_PREFERRED;
980 /* Insist on a nodelist of one node only */
981 if (nodelist) {
982 char *rest = nodelist;
983 while (isdigit(*rest))
984 rest++;
985 if (!*rest)
986 err = 0;
987 }
988 } else if (!strcmp(value, "bind")) {
989 *policy = MPOL_BIND;
990 /* Insist on a nodelist */
991 if (nodelist)
992 err = 0;
993 } else if (!strcmp(value, "interleave")) {
994 *policy = MPOL_INTERLEAVE;
995 /* Default to nodes online if no nodelist */
996 if (!nodelist)
997 *policy_nodes = node_online_map;
998 err = 0;
999 }
1000out:
1001 /* Restore string for error message */
1002 if (nodelist)
1003 *--nodelist = ':';
1004 return err;
1005}
1006
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007static struct page *shmem_swapin_async(struct shared_policy *p,
1008 swp_entry_t entry, unsigned long idx)
1009{
1010 struct page *page;
1011 struct vm_area_struct pvma;
1012
1013 /* Create a pseudo vma that just contains the policy */
1014 memset(&pvma, 0, sizeof(struct vm_area_struct));
1015 pvma.vm_end = PAGE_SIZE;
1016 pvma.vm_pgoff = idx;
1017 pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
1018 page = read_swap_cache_async(entry, &pvma, 0);
1019 mpol_free(pvma.vm_policy);
1020 return page;
1021}
1022
1023struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry,
1024 unsigned long idx)
1025{
1026 struct shared_policy *p = &info->policy;
1027 int i, num;
1028 struct page *page;
1029 unsigned long offset;
1030
1031 num = valid_swaphandles(entry, &offset);
1032 for (i = 0; i < num; offset++, i++) {
1033 page = shmem_swapin_async(p,
1034 swp_entry(swp_type(entry), offset), idx);
1035 if (!page)
1036 break;
1037 page_cache_release(page);
1038 }
1039 lru_add_drain(); /* Push any new pages onto the LRU now */
1040 return shmem_swapin_async(p, entry, idx);
1041}
1042
1043static struct page *
Al Viro6daa0e22005-10-21 03:18:50 -04001044shmem_alloc_page(gfp_t gfp, struct shmem_inode_info *info,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045 unsigned long idx)
1046{
1047 struct vm_area_struct pvma;
1048 struct page *page;
1049
1050 memset(&pvma, 0, sizeof(struct vm_area_struct));
1051 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1052 pvma.vm_pgoff = idx;
1053 pvma.vm_end = PAGE_SIZE;
1054 page = alloc_page_vma(gfp | __GFP_ZERO, &pvma, 0);
1055 mpol_free(pvma.vm_policy);
1056 return page;
1057}
1058#else
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00001059static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
1060{
1061 return 1;
1062}
1063
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064static inline struct page *
1065shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
1066{
1067 swapin_readahead(entry, 0, NULL);
1068 return read_swap_cache_async(entry, NULL, 0);
1069}
1070
1071static inline struct page *
Al Virodd0fc662005-10-07 07:46:04 +01001072shmem_alloc_page(gfp_t gfp,struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001073{
1074 return alloc_page(gfp | __GFP_ZERO);
1075}
1076#endif
1077
1078/*
1079 * shmem_getpage - either get the page from swap or allocate a new one
1080 *
1081 * If we allocate a new one we do not mark it dirty. That's up to the
1082 * vm. If we swap it in we mark it dirty since we also free the swap
1083 * entry since a page cannot live in both the swap and page cache
1084 */
1085static int shmem_getpage(struct inode *inode, unsigned long idx,
1086 struct page **pagep, enum sgp_type sgp, int *type)
1087{
1088 struct address_space *mapping = inode->i_mapping;
1089 struct shmem_inode_info *info = SHMEM_I(inode);
1090 struct shmem_sb_info *sbinfo;
1091 struct page *filepage = *pagep;
1092 struct page *swappage;
1093 swp_entry_t *entry;
1094 swp_entry_t swap;
1095 int error;
1096
1097 if (idx >= SHMEM_MAX_INDEX)
1098 return -EFBIG;
1099 /*
1100 * Normally, filepage is NULL on entry, and either found
1101 * uptodate immediately, or allocated and zeroed, or read
1102 * in under swappage, which is then assigned to filepage.
1103 * But shmem_prepare_write passes in a locked filepage,
1104 * which may be found not uptodate by other callers too,
1105 * and may need to be copied from the swappage read in.
1106 */
1107repeat:
1108 if (!filepage)
1109 filepage = find_lock_page(mapping, idx);
1110 if (filepage && PageUptodate(filepage))
1111 goto done;
1112 error = 0;
1113 if (sgp == SGP_QUICK)
1114 goto failed;
1115
1116 spin_lock(&info->lock);
1117 shmem_recalc_inode(inode);
1118 entry = shmem_swp_alloc(info, idx, sgp);
1119 if (IS_ERR(entry)) {
1120 spin_unlock(&info->lock);
1121 error = PTR_ERR(entry);
1122 goto failed;
1123 }
1124 swap = *entry;
1125
1126 if (swap.val) {
1127 /* Look it up and read it in.. */
1128 swappage = lookup_swap_cache(swap);
1129 if (!swappage) {
1130 shmem_swp_unmap(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001131 /* here we actually do the io */
1132 if (type && *type == VM_FAULT_MINOR) {
Christoph Lameterf8891e52006-06-30 01:55:45 -07001133 __count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 *type = VM_FAULT_MAJOR;
1135 }
Christoph Lameterf8891e52006-06-30 01:55:45 -07001136 spin_unlock(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 swappage = shmem_swapin(info, swap, idx);
1138 if (!swappage) {
1139 spin_lock(&info->lock);
1140 entry = shmem_swp_alloc(info, idx, sgp);
1141 if (IS_ERR(entry))
1142 error = PTR_ERR(entry);
1143 else {
1144 if (entry->val == swap.val)
1145 error = -ENOMEM;
1146 shmem_swp_unmap(entry);
1147 }
1148 spin_unlock(&info->lock);
1149 if (error)
1150 goto failed;
1151 goto repeat;
1152 }
1153 wait_on_page_locked(swappage);
1154 page_cache_release(swappage);
1155 goto repeat;
1156 }
1157
1158 /* We have to do this with page locked to prevent races */
1159 if (TestSetPageLocked(swappage)) {
1160 shmem_swp_unmap(entry);
1161 spin_unlock(&info->lock);
1162 wait_on_page_locked(swappage);
1163 page_cache_release(swappage);
1164 goto repeat;
1165 }
1166 if (PageWriteback(swappage)) {
1167 shmem_swp_unmap(entry);
1168 spin_unlock(&info->lock);
1169 wait_on_page_writeback(swappage);
1170 unlock_page(swappage);
1171 page_cache_release(swappage);
1172 goto repeat;
1173 }
1174 if (!PageUptodate(swappage)) {
1175 shmem_swp_unmap(entry);
1176 spin_unlock(&info->lock);
1177 unlock_page(swappage);
1178 page_cache_release(swappage);
1179 error = -EIO;
1180 goto failed;
1181 }
1182
1183 if (filepage) {
1184 shmem_swp_set(info, entry, 0);
1185 shmem_swp_unmap(entry);
1186 delete_from_swap_cache(swappage);
1187 spin_unlock(&info->lock);
1188 copy_highpage(filepage, swappage);
1189 unlock_page(swappage);
1190 page_cache_release(swappage);
1191 flush_dcache_page(filepage);
1192 SetPageUptodate(filepage);
1193 set_page_dirty(filepage);
1194 swap_free(swap);
1195 } else if (!(error = move_from_swap_cache(
1196 swappage, idx, mapping))) {
1197 info->flags |= SHMEM_PAGEIN;
1198 shmem_swp_set(info, entry, 0);
1199 shmem_swp_unmap(entry);
1200 spin_unlock(&info->lock);
1201 filepage = swappage;
1202 swap_free(swap);
1203 } else {
1204 shmem_swp_unmap(entry);
1205 spin_unlock(&info->lock);
1206 unlock_page(swappage);
1207 page_cache_release(swappage);
1208 if (error == -ENOMEM) {
1209 /* let kswapd refresh zone for GFP_ATOMICs */
Andrew Morton3fcfab12006-10-19 23:28:16 -07001210 congestion_wait(WRITE, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001211 }
1212 goto repeat;
1213 }
1214 } else if (sgp == SGP_READ && !filepage) {
1215 shmem_swp_unmap(entry);
1216 filepage = find_get_page(mapping, idx);
1217 if (filepage &&
1218 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1219 spin_unlock(&info->lock);
1220 wait_on_page_locked(filepage);
1221 page_cache_release(filepage);
1222 filepage = NULL;
1223 goto repeat;
1224 }
1225 spin_unlock(&info->lock);
1226 } else {
1227 shmem_swp_unmap(entry);
1228 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001229 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 spin_lock(&sbinfo->stat_lock);
1231 if (sbinfo->free_blocks == 0 ||
1232 shmem_acct_block(info->flags)) {
1233 spin_unlock(&sbinfo->stat_lock);
1234 spin_unlock(&info->lock);
1235 error = -ENOSPC;
1236 goto failed;
1237 }
1238 sbinfo->free_blocks--;
1239 inode->i_blocks += BLOCKS_PER_PAGE;
1240 spin_unlock(&sbinfo->stat_lock);
1241 } else if (shmem_acct_block(info->flags)) {
1242 spin_unlock(&info->lock);
1243 error = -ENOSPC;
1244 goto failed;
1245 }
1246
1247 if (!filepage) {
1248 spin_unlock(&info->lock);
1249 filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
1250 info,
1251 idx);
1252 if (!filepage) {
1253 shmem_unacct_blocks(info->flags, 1);
1254 shmem_free_blocks(inode, 1);
1255 error = -ENOMEM;
1256 goto failed;
1257 }
1258
1259 spin_lock(&info->lock);
1260 entry = shmem_swp_alloc(info, idx, sgp);
1261 if (IS_ERR(entry))
1262 error = PTR_ERR(entry);
1263 else {
1264 swap = *entry;
1265 shmem_swp_unmap(entry);
1266 }
1267 if (error || swap.val || 0 != add_to_page_cache_lru(
1268 filepage, mapping, idx, GFP_ATOMIC)) {
1269 spin_unlock(&info->lock);
1270 page_cache_release(filepage);
1271 shmem_unacct_blocks(info->flags, 1);
1272 shmem_free_blocks(inode, 1);
1273 filepage = NULL;
1274 if (error)
1275 goto failed;
1276 goto repeat;
1277 }
1278 info->flags |= SHMEM_PAGEIN;
1279 }
1280
1281 info->alloced++;
1282 spin_unlock(&info->lock);
1283 flush_dcache_page(filepage);
1284 SetPageUptodate(filepage);
1285 }
1286done:
1287 if (*pagep != filepage) {
1288 unlock_page(filepage);
1289 *pagep = filepage;
1290 }
1291 return 0;
1292
1293failed:
1294 if (*pagep != filepage) {
1295 unlock_page(filepage);
1296 page_cache_release(filepage);
1297 }
1298 return error;
1299}
1300
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08001301static struct page *shmem_nopage(struct vm_area_struct *vma,
1302 unsigned long address, int *type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001304 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001305 struct page *page = NULL;
1306 unsigned long idx;
1307 int error;
1308
1309 idx = (address - vma->vm_start) >> PAGE_SHIFT;
1310 idx += vma->vm_pgoff;
1311 idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT;
1312 if (((loff_t) idx << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1313 return NOPAGE_SIGBUS;
1314
1315 error = shmem_getpage(inode, idx, &page, SGP_CACHE, type);
1316 if (error)
1317 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS;
1318
1319 mark_page_accessed(page);
1320 return page;
1321}
1322
1323static int shmem_populate(struct vm_area_struct *vma,
1324 unsigned long addr, unsigned long len,
1325 pgprot_t prot, unsigned long pgoff, int nonblock)
1326{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001327 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 struct mm_struct *mm = vma->vm_mm;
1329 enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE;
1330 unsigned long size;
1331
1332 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1333 if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size)
1334 return -EINVAL;
1335
1336 while ((long) len > 0) {
1337 struct page *page = NULL;
1338 int err;
1339 /*
1340 * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE
1341 */
1342 err = shmem_getpage(inode, pgoff, &page, sgp, NULL);
1343 if (err)
1344 return err;
Paolo 'Blaisorblade' Giarrussod44ed4f2005-09-03 15:54:55 -07001345 /* Page may still be null, but only if nonblock was set. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 if (page) {
1347 mark_page_accessed(page);
1348 err = install_page(mm, vma, addr, page, prot);
1349 if (err) {
1350 page_cache_release(page);
1351 return err;
1352 }
Hugh Dickins65500d22005-10-29 18:15:59 -07001353 } else if (vma->vm_flags & VM_NONLINEAR) {
Paolo 'Blaisorblade' Giarrussod44ed4f2005-09-03 15:54:55 -07001354 /* No page was found just because we can't read it in
1355 * now (being here implies nonblock != 0), but the page
1356 * may exist, so set the PTE to fault it in later. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001357 err = install_file_pte(mm, vma, addr, pgoff, prot);
1358 if (err)
1359 return err;
1360 }
1361
1362 len -= PAGE_SIZE;
1363 addr += PAGE_SIZE;
1364 pgoff++;
1365 }
1366 return 0;
1367}
1368
1369#ifdef CONFIG_NUMA
1370int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1371{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001372 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1374}
1375
1376struct mempolicy *
1377shmem_get_policy(struct vm_area_struct *vma, unsigned long addr)
1378{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001379 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380 unsigned long idx;
1381
1382 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1383 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1384}
1385#endif
1386
1387int shmem_lock(struct file *file, int lock, struct user_struct *user)
1388{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001389 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 struct shmem_inode_info *info = SHMEM_I(inode);
1391 int retval = -ENOMEM;
1392
1393 spin_lock(&info->lock);
1394 if (lock && !(info->flags & VM_LOCKED)) {
1395 if (!user_shm_lock(inode->i_size, user))
1396 goto out_nomem;
1397 info->flags |= VM_LOCKED;
1398 }
1399 if (!lock && (info->flags & VM_LOCKED) && user) {
1400 user_shm_unlock(inode->i_size, user);
1401 info->flags &= ~VM_LOCKED;
1402 }
1403 retval = 0;
1404out_nomem:
1405 spin_unlock(&info->lock);
1406 return retval;
1407}
1408
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08001409static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001410{
1411 file_accessed(file);
1412 vma->vm_ops = &shmem_vm_ops;
1413 return 0;
1414}
1415
1416static struct inode *
1417shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1418{
1419 struct inode *inode;
1420 struct shmem_inode_info *info;
1421 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1422
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001423 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 spin_lock(&sbinfo->stat_lock);
1425 if (!sbinfo->free_inodes) {
1426 spin_unlock(&sbinfo->stat_lock);
1427 return NULL;
1428 }
1429 sbinfo->free_inodes--;
1430 spin_unlock(&sbinfo->stat_lock);
1431 }
1432
1433 inode = new_inode(sb);
1434 if (inode) {
1435 inode->i_mode = mode;
1436 inode->i_uid = current->fsuid;
1437 inode->i_gid = current->fsgid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438 inode->i_blocks = 0;
1439 inode->i_mapping->a_ops = &shmem_aops;
1440 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1441 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
David M. Grimes91828a42006-10-17 00:09:45 -07001442 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 info = SHMEM_I(inode);
1444 memset(info, 0, (char *)inode - (char *)info);
1445 spin_lock_init(&info->lock);
1446 INIT_LIST_HEAD(&info->swaplist);
1447
1448 switch (mode & S_IFMT) {
1449 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001450 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451 init_special_inode(inode, mode, dev);
1452 break;
1453 case S_IFREG:
1454 inode->i_op = &shmem_inode_operations;
1455 inode->i_fop = &shmem_file_operations;
Robin Holt7339ff82006-01-14 13:20:48 -08001456 mpol_shared_policy_init(&info->policy, sbinfo->policy,
1457 &sbinfo->policy_nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458 break;
1459 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07001460 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 /* Some things misbehave if size == 0 on a directory */
1462 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1463 inode->i_op = &shmem_dir_inode_operations;
1464 inode->i_fop = &simple_dir_operations;
1465 break;
1466 case S_IFLNK:
1467 /*
1468 * Must not load anything in the rbtree,
1469 * mpol_free_shared_policy will not be called.
1470 */
Robin Holt7339ff82006-01-14 13:20:48 -08001471 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT,
1472 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 break;
1474 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001475 } else if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 spin_lock(&sbinfo->stat_lock);
1477 sbinfo->free_inodes++;
1478 spin_unlock(&sbinfo->stat_lock);
1479 }
1480 return inode;
1481}
1482
1483#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001484static const struct inode_operations shmem_symlink_inode_operations;
1485static const struct inode_operations shmem_symlink_inline_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486
1487/*
1488 * Normally tmpfs makes no use of shmem_prepare_write, but it
1489 * lets a tmpfs file be used read-write below the loop driver.
1490 */
1491static int
1492shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1493{
1494 struct inode *inode = page->mapping->host;
1495 return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL);
1496}
1497
1498static ssize_t
1499shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1500{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001501 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502 loff_t pos;
1503 unsigned long written;
1504 ssize_t err;
1505
1506 if ((ssize_t) count < 0)
1507 return -EINVAL;
1508
1509 if (!access_ok(VERIFY_READ, buf, count))
1510 return -EFAULT;
1511
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001512 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513
1514 pos = *ppos;
1515 written = 0;
1516
1517 err = generic_write_checks(file, &pos, &count, 0);
1518 if (err || !count)
1519 goto out;
1520
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001521 err = remove_suid(file->f_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001522 if (err)
1523 goto out;
1524
1525 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1526
1527 do {
1528 struct page *page = NULL;
1529 unsigned long bytes, index, offset;
1530 char *kaddr;
1531 int left;
1532
1533 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1534 index = pos >> PAGE_CACHE_SHIFT;
1535 bytes = PAGE_CACHE_SIZE - offset;
1536 if (bytes > count)
1537 bytes = count;
1538
1539 /*
1540 * We don't hold page lock across copy from user -
1541 * what would it guard against? - so no deadlock here.
1542 * But it still may be a good idea to prefault below.
1543 */
1544
1545 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1546 if (err)
1547 break;
1548
1549 left = bytes;
1550 if (PageHighMem(page)) {
1551 volatile unsigned char dummy;
1552 __get_user(dummy, buf);
1553 __get_user(dummy, buf + bytes - 1);
1554
1555 kaddr = kmap_atomic(page, KM_USER0);
1556 left = __copy_from_user_inatomic(kaddr + offset,
1557 buf, bytes);
1558 kunmap_atomic(kaddr, KM_USER0);
1559 }
1560 if (left) {
1561 kaddr = kmap(page);
1562 left = __copy_from_user(kaddr + offset, buf, bytes);
1563 kunmap(page);
1564 }
1565
1566 written += bytes;
1567 count -= bytes;
1568 pos += bytes;
1569 buf += bytes;
1570 if (pos > inode->i_size)
1571 i_size_write(inode, pos);
1572
1573 flush_dcache_page(page);
1574 set_page_dirty(page);
1575 mark_page_accessed(page);
1576 page_cache_release(page);
1577
1578 if (left) {
1579 pos -= left;
1580 written -= left;
1581 err = -EFAULT;
1582 break;
1583 }
1584
1585 /*
1586 * Our dirty pages are not counted in nr_dirty,
1587 * and we do not attempt to balance dirty pages.
1588 */
1589
1590 cond_resched();
1591 } while (count);
1592
1593 *ppos = pos;
1594 if (written)
1595 err = written;
1596out:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001597 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 return err;
1599}
1600
1601static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1602{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001603 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 struct address_space *mapping = inode->i_mapping;
1605 unsigned long index, offset;
1606
1607 index = *ppos >> PAGE_CACHE_SHIFT;
1608 offset = *ppos & ~PAGE_CACHE_MASK;
1609
1610 for (;;) {
1611 struct page *page = NULL;
1612 unsigned long end_index, nr, ret;
1613 loff_t i_size = i_size_read(inode);
1614
1615 end_index = i_size >> PAGE_CACHE_SHIFT;
1616 if (index > end_index)
1617 break;
1618 if (index == end_index) {
1619 nr = i_size & ~PAGE_CACHE_MASK;
1620 if (nr <= offset)
1621 break;
1622 }
1623
1624 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1625 if (desc->error) {
1626 if (desc->error == -EINVAL)
1627 desc->error = 0;
1628 break;
1629 }
1630
1631 /*
1632 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001633 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001634 */
1635 nr = PAGE_CACHE_SIZE;
1636 i_size = i_size_read(inode);
1637 end_index = i_size >> PAGE_CACHE_SHIFT;
1638 if (index == end_index) {
1639 nr = i_size & ~PAGE_CACHE_MASK;
1640 if (nr <= offset) {
1641 if (page)
1642 page_cache_release(page);
1643 break;
1644 }
1645 }
1646 nr -= offset;
1647
1648 if (page) {
1649 /*
1650 * If users can be writing to this page using arbitrary
1651 * virtual addresses, take care about potential aliasing
1652 * before reading the page on the kernel side.
1653 */
1654 if (mapping_writably_mapped(mapping))
1655 flush_dcache_page(page);
1656 /*
1657 * Mark the page accessed if we read the beginning.
1658 */
1659 if (!offset)
1660 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001661 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 page = ZERO_PAGE(0);
Nick Pigginb5810032005-10-29 18:16:12 -07001663 page_cache_get(page);
1664 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665
1666 /*
1667 * Ok, we have the page, and it's up-to-date, so
1668 * now we can copy it to user space...
1669 *
1670 * The actor routine returns how many bytes were actually used..
1671 * NOTE! This may not be the same as how much of a user buffer
1672 * we filled up (we may be padding etc), so we can only update
1673 * "pos" here (the actor routine has to update the user buffer
1674 * pointers and the remaining count).
1675 */
1676 ret = actor(desc, page, offset, nr);
1677 offset += ret;
1678 index += offset >> PAGE_CACHE_SHIFT;
1679 offset &= ~PAGE_CACHE_MASK;
1680
1681 page_cache_release(page);
1682 if (ret != nr || !desc->count)
1683 break;
1684
1685 cond_resched();
1686 }
1687
1688 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1689 file_accessed(filp);
1690}
1691
1692static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1693{
1694 read_descriptor_t desc;
1695
1696 if ((ssize_t) count < 0)
1697 return -EINVAL;
1698 if (!access_ok(VERIFY_WRITE, buf, count))
1699 return -EFAULT;
1700 if (!count)
1701 return 0;
1702
1703 desc.written = 0;
1704 desc.count = count;
1705 desc.arg.buf = buf;
1706 desc.error = 0;
1707
1708 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1709 if (desc.written)
1710 return desc.written;
1711 return desc.error;
1712}
1713
1714static ssize_t shmem_file_sendfile(struct file *in_file, loff_t *ppos,
1715 size_t count, read_actor_t actor, void *target)
1716{
1717 read_descriptor_t desc;
1718
1719 if (!count)
1720 return 0;
1721
1722 desc.written = 0;
1723 desc.count = count;
1724 desc.arg.data = target;
1725 desc.error = 0;
1726
1727 do_shmem_file_read(in_file, ppos, &desc, actor);
1728 if (desc.written)
1729 return desc.written;
1730 return desc.error;
1731}
1732
David Howells726c3342006-06-23 02:02:58 -07001733static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734{
David Howells726c3342006-06-23 02:02:58 -07001735 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736
1737 buf->f_type = TMPFS_MAGIC;
1738 buf->f_bsize = PAGE_CACHE_SIZE;
1739 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001740 spin_lock(&sbinfo->stat_lock);
1741 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 buf->f_blocks = sbinfo->max_blocks;
1743 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001744 }
1745 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 buf->f_files = sbinfo->max_inodes;
1747 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 }
1749 /* else leave those fields 0 like simple_statfs */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001750 spin_unlock(&sbinfo->stat_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751 return 0;
1752}
1753
1754/*
1755 * File creation. Allocate an inode, and we're done..
1756 */
1757static int
1758shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1759{
1760 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1761 int error = -ENOSPC;
1762
1763 if (inode) {
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001764 error = security_inode_init_security(inode, dir, NULL, NULL,
1765 NULL);
1766 if (error) {
1767 if (error != -EOPNOTSUPP) {
1768 iput(inode);
1769 return error;
1770 }
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001771 }
1772 error = shmem_acl_init(inode, dir);
1773 if (error) {
1774 iput(inode);
1775 return error;
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001776 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 if (dir->i_mode & S_ISGID) {
1778 inode->i_gid = dir->i_gid;
1779 if (S_ISDIR(mode))
1780 inode->i_mode |= S_ISGID;
1781 }
1782 dir->i_size += BOGO_DIRENT_SIZE;
1783 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1784 d_instantiate(dentry, inode);
1785 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786 }
1787 return error;
1788}
1789
1790static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1791{
1792 int error;
1793
1794 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1795 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07001796 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001797 return 0;
1798}
1799
1800static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1801 struct nameidata *nd)
1802{
1803 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1804}
1805
1806/*
1807 * Link a file..
1808 */
1809static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1810{
1811 struct inode *inode = old_dentry->d_inode;
1812 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1813
1814 /*
1815 * No ordinary (disk based) filesystem counts links as inodes;
1816 * but each new link needs a new dentry, pinning lowmem, and
1817 * tmpfs dentries cannot be pruned until they are unlinked.
1818 */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001819 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001820 spin_lock(&sbinfo->stat_lock);
1821 if (!sbinfo->free_inodes) {
1822 spin_unlock(&sbinfo->stat_lock);
1823 return -ENOSPC;
1824 }
1825 sbinfo->free_inodes--;
1826 spin_unlock(&sbinfo->stat_lock);
1827 }
1828
1829 dir->i_size += BOGO_DIRENT_SIZE;
1830 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansend8c76e62006-09-30 23:29:04 -07001831 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832 atomic_inc(&inode->i_count); /* New dentry reference */
1833 dget(dentry); /* Extra pinning count for the created dentry */
1834 d_instantiate(dentry, inode);
1835 return 0;
1836}
1837
1838static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1839{
1840 struct inode *inode = dentry->d_inode;
1841
1842 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) {
1843 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001844 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 spin_lock(&sbinfo->stat_lock);
1846 sbinfo->free_inodes++;
1847 spin_unlock(&sbinfo->stat_lock);
1848 }
1849 }
1850
1851 dir->i_size -= BOGO_DIRENT_SIZE;
1852 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001853 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854 dput(dentry); /* Undo the count from "create" - this does all the work */
1855 return 0;
1856}
1857
1858static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1859{
1860 if (!simple_empty(dentry))
1861 return -ENOTEMPTY;
1862
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001863 drop_nlink(dentry->d_inode);
1864 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865 return shmem_unlink(dir, dentry);
1866}
1867
1868/*
1869 * The VFS layer already does all the dentry stuff for rename,
1870 * we just have to decrement the usage count for the target if
1871 * it exists so that the VFS layer correctly free's it when it
1872 * gets overwritten.
1873 */
1874static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1875{
1876 struct inode *inode = old_dentry->d_inode;
1877 int they_are_dirs = S_ISDIR(inode->i_mode);
1878
1879 if (!simple_empty(new_dentry))
1880 return -ENOTEMPTY;
1881
1882 if (new_dentry->d_inode) {
1883 (void) shmem_unlink(new_dir, new_dentry);
1884 if (they_are_dirs)
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001885 drop_nlink(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001886 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001887 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07001888 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 }
1890
1891 old_dir->i_size -= BOGO_DIRENT_SIZE;
1892 new_dir->i_size += BOGO_DIRENT_SIZE;
1893 old_dir->i_ctime = old_dir->i_mtime =
1894 new_dir->i_ctime = new_dir->i_mtime =
1895 inode->i_ctime = CURRENT_TIME;
1896 return 0;
1897}
1898
1899static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1900{
1901 int error;
1902 int len;
1903 struct inode *inode;
1904 struct page *page = NULL;
1905 char *kaddr;
1906 struct shmem_inode_info *info;
1907
1908 len = strlen(symname) + 1;
1909 if (len > PAGE_CACHE_SIZE)
1910 return -ENAMETOOLONG;
1911
1912 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1913 if (!inode)
1914 return -ENOSPC;
1915
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001916 error = security_inode_init_security(inode, dir, NULL, NULL,
1917 NULL);
1918 if (error) {
1919 if (error != -EOPNOTSUPP) {
1920 iput(inode);
1921 return error;
1922 }
1923 error = 0;
1924 }
1925
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 info = SHMEM_I(inode);
1927 inode->i_size = len-1;
1928 if (len <= (char *)inode - (char *)info) {
1929 /* do it inline */
1930 memcpy(info, symname, len);
1931 inode->i_op = &shmem_symlink_inline_operations;
1932 } else {
1933 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1934 if (error) {
1935 iput(inode);
1936 return error;
1937 }
1938 inode->i_op = &shmem_symlink_inode_operations;
1939 kaddr = kmap_atomic(page, KM_USER0);
1940 memcpy(kaddr, symname, len);
1941 kunmap_atomic(kaddr, KM_USER0);
1942 set_page_dirty(page);
1943 page_cache_release(page);
1944 }
1945 if (dir->i_mode & S_ISGID)
1946 inode->i_gid = dir->i_gid;
1947 dir->i_size += BOGO_DIRENT_SIZE;
1948 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1949 d_instantiate(dentry, inode);
1950 dget(dentry);
1951 return 0;
1952}
1953
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001954static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955{
1956 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001957 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958}
1959
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001960static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961{
1962 struct page *page = NULL;
1963 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1964 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001965 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966}
1967
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001968static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969{
1970 if (!IS_ERR(nd_get_link(nd))) {
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001971 struct page *page = cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 kunmap(page);
1973 mark_page_accessed(page);
1974 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 }
1976}
1977
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001978static const struct inode_operations shmem_symlink_inline_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001979 .readlink = generic_readlink,
1980 .follow_link = shmem_follow_link_inline,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981};
1982
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001983static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984 .truncate = shmem_truncate,
1985 .readlink = generic_readlink,
1986 .follow_link = shmem_follow_link,
1987 .put_link = shmem_put_link,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988};
1989
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001990#ifdef CONFIG_TMPFS_POSIX_ACL
1991/**
1992 * Superblocks without xattr inode operations will get security.* xattr
1993 * support from the VFS "for free". As soon as we have any other xattrs
1994 * like ACLs, we also need to implement the security.* handlers at
1995 * filesystem level, though.
1996 */
1997
1998static size_t shmem_xattr_security_list(struct inode *inode, char *list,
1999 size_t list_len, const char *name,
2000 size_t name_len)
2001{
2002 return security_inode_listsecurity(inode, list, list_len);
2003}
2004
2005static int shmem_xattr_security_get(struct inode *inode, const char *name,
2006 void *buffer, size_t size)
2007{
2008 if (strcmp(name, "") == 0)
2009 return -EINVAL;
2010 return security_inode_getsecurity(inode, name, buffer, size,
2011 -EOPNOTSUPP);
2012}
2013
2014static int shmem_xattr_security_set(struct inode *inode, const char *name,
2015 const void *value, size_t size, int flags)
2016{
2017 if (strcmp(name, "") == 0)
2018 return -EINVAL;
2019 return security_inode_setsecurity(inode, name, value, size, flags);
2020}
2021
Adrian Bunk1f370a22006-12-06 20:38:21 -08002022static struct xattr_handler shmem_xattr_security_handler = {
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002023 .prefix = XATTR_SECURITY_PREFIX,
2024 .list = shmem_xattr_security_list,
2025 .get = shmem_xattr_security_get,
2026 .set = shmem_xattr_security_set,
2027};
2028
2029static struct xattr_handler *shmem_xattr_handlers[] = {
2030 &shmem_xattr_acl_access_handler,
2031 &shmem_xattr_acl_default_handler,
2032 &shmem_xattr_security_handler,
2033 NULL
2034};
2035#endif
2036
David M. Grimes91828a42006-10-17 00:09:45 -07002037static struct dentry *shmem_get_parent(struct dentry *child)
2038{
2039 return ERR_PTR(-ESTALE);
2040}
2041
2042static int shmem_match(struct inode *ino, void *vfh)
2043{
2044 __u32 *fh = vfh;
2045 __u64 inum = fh[2];
2046 inum = (inum << 32) | fh[1];
2047 return ino->i_ino == inum && fh[0] == ino->i_generation;
2048}
2049
2050static struct dentry *shmem_get_dentry(struct super_block *sb, void *vfh)
2051{
2052 struct dentry *de = NULL;
2053 struct inode *inode;
2054 __u32 *fh = vfh;
2055 __u64 inum = fh[2];
2056 inum = (inum << 32) | fh[1];
2057
2058 inode = ilookup5(sb, (unsigned long)(inum+fh[0]), shmem_match, vfh);
2059 if (inode) {
2060 de = d_find_alias(inode);
2061 iput(inode);
2062 }
2063
2064 return de? de: ERR_PTR(-ESTALE);
2065}
2066
2067static struct dentry *shmem_decode_fh(struct super_block *sb, __u32 *fh,
2068 int len, int type,
2069 int (*acceptable)(void *context, struct dentry *de),
2070 void *context)
2071{
2072 if (len < 3)
2073 return ERR_PTR(-ESTALE);
2074
2075 return sb->s_export_op->find_exported_dentry(sb, fh, NULL, acceptable,
2076 context);
2077}
2078
2079static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2080 int connectable)
2081{
2082 struct inode *inode = dentry->d_inode;
2083
2084 if (*len < 3)
2085 return 255;
2086
2087 if (hlist_unhashed(&inode->i_hash)) {
2088 /* Unfortunately insert_inode_hash is not idempotent,
2089 * so as we hash inodes here rather than at creation
2090 * time, we need a lock to ensure we only try
2091 * to do it once
2092 */
2093 static DEFINE_SPINLOCK(lock);
2094 spin_lock(&lock);
2095 if (hlist_unhashed(&inode->i_hash))
2096 __insert_inode_hash(inode,
2097 inode->i_ino + inode->i_generation);
2098 spin_unlock(&lock);
2099 }
2100
2101 fh[0] = inode->i_generation;
2102 fh[1] = inode->i_ino;
2103 fh[2] = ((__u64)inode->i_ino) >> 32;
2104
2105 *len = 3;
2106 return 1;
2107}
2108
2109static struct export_operations shmem_export_ops = {
2110 .get_parent = shmem_get_parent,
2111 .get_dentry = shmem_get_dentry,
2112 .encode_fh = shmem_encode_fh,
2113 .decode_fh = shmem_decode_fh,
2114};
2115
Robin Holt7339ff82006-01-14 13:20:48 -08002116static int shmem_parse_options(char *options, int *mode, uid_t *uid,
2117 gid_t *gid, unsigned long *blocks, unsigned long *inodes,
2118 int *policy, nodemask_t *policy_nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119{
2120 char *this_char, *value, *rest;
2121
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002122 while (options != NULL) {
2123 this_char = options;
2124 for (;;) {
2125 /*
2126 * NUL-terminate this option: unfortunately,
2127 * mount options form a comma-separated list,
2128 * but mpol's nodelist may also contain commas.
2129 */
2130 options = strchr(options, ',');
2131 if (options == NULL)
2132 break;
2133 options++;
2134 if (!isdigit(*options)) {
2135 options[-1] = '\0';
2136 break;
2137 }
2138 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002139 if (!*this_char)
2140 continue;
2141 if ((value = strchr(this_char,'=')) != NULL) {
2142 *value++ = 0;
2143 } else {
2144 printk(KERN_ERR
2145 "tmpfs: No value for mount option '%s'\n",
2146 this_char);
2147 return 1;
2148 }
2149
2150 if (!strcmp(this_char,"size")) {
2151 unsigned long long size;
2152 size = memparse(value,&rest);
2153 if (*rest == '%') {
2154 size <<= PAGE_SHIFT;
2155 size *= totalram_pages;
2156 do_div(size, 100);
2157 rest++;
2158 }
2159 if (*rest)
2160 goto bad_val;
2161 *blocks = size >> PAGE_CACHE_SHIFT;
2162 } else if (!strcmp(this_char,"nr_blocks")) {
2163 *blocks = memparse(value,&rest);
2164 if (*rest)
2165 goto bad_val;
2166 } else if (!strcmp(this_char,"nr_inodes")) {
2167 *inodes = memparse(value,&rest);
2168 if (*rest)
2169 goto bad_val;
2170 } else if (!strcmp(this_char,"mode")) {
2171 if (!mode)
2172 continue;
2173 *mode = simple_strtoul(value,&rest,8);
2174 if (*rest)
2175 goto bad_val;
2176 } else if (!strcmp(this_char,"uid")) {
2177 if (!uid)
2178 continue;
2179 *uid = simple_strtoul(value,&rest,0);
2180 if (*rest)
2181 goto bad_val;
2182 } else if (!strcmp(this_char,"gid")) {
2183 if (!gid)
2184 continue;
2185 *gid = simple_strtoul(value,&rest,0);
2186 if (*rest)
2187 goto bad_val;
Robin Holt7339ff82006-01-14 13:20:48 -08002188 } else if (!strcmp(this_char,"mpol")) {
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002189 if (shmem_parse_mpol(value,policy,policy_nodes))
Robin Holt7339ff82006-01-14 13:20:48 -08002190 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 } else {
2192 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2193 this_char);
2194 return 1;
2195 }
2196 }
2197 return 0;
2198
2199bad_val:
2200 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2201 value, this_char);
2202 return 1;
2203
2204}
2205
2206static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2207{
2208 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002209 unsigned long max_blocks = sbinfo->max_blocks;
2210 unsigned long max_inodes = sbinfo->max_inodes;
Robin Holt7339ff82006-01-14 13:20:48 -08002211 int policy = sbinfo->policy;
2212 nodemask_t policy_nodes = sbinfo->policy_nodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002213 unsigned long blocks;
2214 unsigned long inodes;
2215 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216
Robin Holt7339ff82006-01-14 13:20:48 -08002217 if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks,
2218 &max_inodes, &policy, &policy_nodes))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002219 return error;
2220
2221 spin_lock(&sbinfo->stat_lock);
2222 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2223 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2224 if (max_blocks < blocks)
2225 goto out;
2226 if (max_inodes < inodes)
2227 goto out;
2228 /*
2229 * Those tests also disallow limited->unlimited while any are in
2230 * use, so i_blocks will always be zero when max_blocks is zero;
2231 * but we must separately disallow unlimited->limited, because
2232 * in that case we have no record of how much is already in use.
2233 */
2234 if (max_blocks && !sbinfo->max_blocks)
2235 goto out;
2236 if (max_inodes && !sbinfo->max_inodes)
2237 goto out;
2238
2239 error = 0;
2240 sbinfo->max_blocks = max_blocks;
2241 sbinfo->free_blocks = max_blocks - blocks;
2242 sbinfo->max_inodes = max_inodes;
2243 sbinfo->free_inodes = max_inodes - inodes;
Robin Holt7339ff82006-01-14 13:20:48 -08002244 sbinfo->policy = policy;
2245 sbinfo->policy_nodes = policy_nodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002246out:
2247 spin_unlock(&sbinfo->stat_lock);
2248 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249}
2250#endif
2251
2252static void shmem_put_super(struct super_block *sb)
2253{
2254 kfree(sb->s_fs_info);
2255 sb->s_fs_info = NULL;
2256}
2257
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258static int shmem_fill_super(struct super_block *sb,
2259 void *data, int silent)
2260{
2261 struct inode *inode;
2262 struct dentry *root;
2263 int mode = S_IRWXUGO | S_ISVTX;
2264 uid_t uid = current->fsuid;
2265 gid_t gid = current->fsgid;
2266 int err = -ENOMEM;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002267 struct shmem_sb_info *sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002268 unsigned long blocks = 0;
2269 unsigned long inodes = 0;
Robin Holt7339ff82006-01-14 13:20:48 -08002270 int policy = MPOL_DEFAULT;
2271 nodemask_t policy_nodes = node_online_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002273#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 /*
2275 * Per default we only allow half of the physical ram per
2276 * tmpfs instance, limiting inodes to one per page of lowmem;
2277 * but the internal instance is left unlimited.
2278 */
2279 if (!(sb->s_flags & MS_NOUSER)) {
2280 blocks = totalram_pages / 2;
2281 inodes = totalram_pages - totalhigh_pages;
2282 if (inodes > blocks)
2283 inodes = blocks;
Robin Holt7339ff82006-01-14 13:20:48 -08002284 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks,
2285 &inodes, &policy, &policy_nodes))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 return -EINVAL;
2287 }
David M. Grimes91828a42006-10-17 00:09:45 -07002288 sb->s_export_op = &shmem_export_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002289#else
2290 sb->s_flags |= MS_NOUSER;
2291#endif
2292
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002293 /* Round up to L1_CACHE_BYTES to resist false sharing */
2294 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
2295 L1_CACHE_BYTES), GFP_KERNEL);
2296 if (!sbinfo)
2297 return -ENOMEM;
2298
2299 spin_lock_init(&sbinfo->stat_lock);
2300 sbinfo->max_blocks = blocks;
2301 sbinfo->free_blocks = blocks;
2302 sbinfo->max_inodes = inodes;
2303 sbinfo->free_inodes = inodes;
Robin Holt7339ff82006-01-14 13:20:48 -08002304 sbinfo->policy = policy;
2305 sbinfo->policy_nodes = policy_nodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002306
2307 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002308 sb->s_maxbytes = SHMEM_MAX_BYTES;
2309 sb->s_blocksize = PAGE_CACHE_SIZE;
2310 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2311 sb->s_magic = TMPFS_MAGIC;
2312 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01002313 sb->s_time_gran = 1;
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002314#ifdef CONFIG_TMPFS_POSIX_ACL
2315 sb->s_xattr = shmem_xattr_handlers;
2316 sb->s_flags |= MS_POSIXACL;
2317#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002318
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
2320 if (!inode)
2321 goto failed;
2322 inode->i_uid = uid;
2323 inode->i_gid = gid;
2324 root = d_alloc_root(inode);
2325 if (!root)
2326 goto failed_iput;
2327 sb->s_root = root;
2328 return 0;
2329
2330failed_iput:
2331 iput(inode);
2332failed:
2333 shmem_put_super(sb);
2334 return err;
2335}
2336
Pekka Enbergfcc234f2006-03-22 00:08:13 -08002337static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002338
2339static struct inode *shmem_alloc_inode(struct super_block *sb)
2340{
2341 struct shmem_inode_info *p;
Christoph Lametere94b1762006-12-06 20:33:17 -08002342 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 if (!p)
2344 return NULL;
2345 return &p->vfs_inode;
2346}
2347
2348static void shmem_destroy_inode(struct inode *inode)
2349{
2350 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2351 /* only struct inode is valid if it's an inline symlink */
2352 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2353 }
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002354 shmem_acl_destroy_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2356}
2357
Pekka Enbergfcc234f2006-03-22 00:08:13 -08002358static void init_once(void *foo, struct kmem_cache *cachep,
2359 unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360{
2361 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2362
Christoph Lametera35afb82007-05-16 22:10:57 -07002363 inode_init_once(&p->vfs_inode);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002364#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Lametera35afb82007-05-16 22:10:57 -07002365 p->i_acl = NULL;
2366 p->i_default_acl = NULL;
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002367#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368}
2369
2370static int init_inodecache(void)
2371{
2372 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2373 sizeof(struct shmem_inode_info),
2374 0, 0, init_once, NULL);
2375 if (shmem_inode_cachep == NULL)
2376 return -ENOMEM;
2377 return 0;
2378}
2379
2380static void destroy_inodecache(void)
2381{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07002382 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383}
2384
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002385static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08002387 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388#ifdef CONFIG_TMPFS
2389 .prepare_write = shmem_prepare_write,
2390 .commit_write = simple_commit_write,
2391#endif
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07002392 .migratepage = migrate_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393};
2394
Helge Deller15ad7cd2006-12-06 20:40:36 -08002395static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396 .mmap = shmem_mmap,
2397#ifdef CONFIG_TMPFS
2398 .llseek = generic_file_llseek,
2399 .read = shmem_file_read,
2400 .write = shmem_file_write,
2401 .fsync = simple_sync_file,
2402 .sendfile = shmem_file_sendfile,
2403#endif
2404};
2405
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002406static const struct inode_operations shmem_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002407 .truncate = shmem_truncate,
2408 .setattr = shmem_notify_change,
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002409 .truncate_range = shmem_truncate_range,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002410#ifdef CONFIG_TMPFS_POSIX_ACL
2411 .setxattr = generic_setxattr,
2412 .getxattr = generic_getxattr,
2413 .listxattr = generic_listxattr,
2414 .removexattr = generic_removexattr,
2415 .permission = shmem_permission,
2416#endif
2417
Linus Torvalds1da177e2005-04-16 15:20:36 -07002418};
2419
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002420static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421#ifdef CONFIG_TMPFS
2422 .create = shmem_create,
2423 .lookup = simple_lookup,
2424 .link = shmem_link,
2425 .unlink = shmem_unlink,
2426 .symlink = shmem_symlink,
2427 .mkdir = shmem_mkdir,
2428 .rmdir = shmem_rmdir,
2429 .mknod = shmem_mknod,
2430 .rename = shmem_rename,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002431#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002432#ifdef CONFIG_TMPFS_POSIX_ACL
2433 .setattr = shmem_notify_change,
2434 .setxattr = generic_setxattr,
2435 .getxattr = generic_getxattr,
2436 .listxattr = generic_listxattr,
2437 .removexattr = generic_removexattr,
2438 .permission = shmem_permission,
2439#endif
2440};
2441
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002442static const struct inode_operations shmem_special_inode_operations = {
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002443#ifdef CONFIG_TMPFS_POSIX_ACL
2444 .setattr = shmem_notify_change,
2445 .setxattr = generic_setxattr,
2446 .getxattr = generic_getxattr,
2447 .listxattr = generic_listxattr,
2448 .removexattr = generic_removexattr,
2449 .permission = shmem_permission,
2450#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002451};
2452
Hugh Dickins759b9772007-03-05 00:30:28 -08002453static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454 .alloc_inode = shmem_alloc_inode,
2455 .destroy_inode = shmem_destroy_inode,
2456#ifdef CONFIG_TMPFS
2457 .statfs = shmem_statfs,
2458 .remount_fs = shmem_remount_fs,
2459#endif
2460 .delete_inode = shmem_delete_inode,
2461 .drop_inode = generic_delete_inode,
2462 .put_super = shmem_put_super,
2463};
2464
2465static struct vm_operations_struct shmem_vm_ops = {
2466 .nopage = shmem_nopage,
2467 .populate = shmem_populate,
2468#ifdef CONFIG_NUMA
2469 .set_policy = shmem_set_policy,
2470 .get_policy = shmem_get_policy,
2471#endif
2472};
2473
2474
David Howells454e2392006-06-23 02:02:57 -07002475static int shmem_get_sb(struct file_system_type *fs_type,
2476 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477{
David Howells454e2392006-06-23 02:02:57 -07002478 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002479}
2480
2481static struct file_system_type tmpfs_fs_type = {
2482 .owner = THIS_MODULE,
2483 .name = "tmpfs",
2484 .get_sb = shmem_get_sb,
2485 .kill_sb = kill_litter_super,
2486};
2487static struct vfsmount *shm_mnt;
2488
2489static int __init init_tmpfs(void)
2490{
2491 int error;
2492
2493 error = init_inodecache();
2494 if (error)
2495 goto out3;
2496
2497 error = register_filesystem(&tmpfs_fs_type);
2498 if (error) {
2499 printk(KERN_ERR "Could not register tmpfs\n");
2500 goto out2;
2501 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07002502
Trond Myklebust1f5ce9e2006-06-09 09:34:16 -04002503 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 tmpfs_fs_type.name, NULL);
2505 if (IS_ERR(shm_mnt)) {
2506 error = PTR_ERR(shm_mnt);
2507 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2508 goto out1;
2509 }
2510 return 0;
2511
2512out1:
2513 unregister_filesystem(&tmpfs_fs_type);
2514out2:
2515 destroy_inodecache();
2516out3:
2517 shm_mnt = ERR_PTR(error);
2518 return error;
2519}
2520module_init(init_tmpfs)
2521
2522/*
2523 * shmem_file_setup - get an unlinked file living in tmpfs
2524 *
2525 * @name: name for dentry (to be seen in /proc/<pid>/maps
2526 * @size: size to be set for the file
2527 *
2528 */
2529struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2530{
2531 int error;
2532 struct file *file;
2533 struct inode *inode;
2534 struct dentry *dentry, *root;
2535 struct qstr this;
2536
2537 if (IS_ERR(shm_mnt))
2538 return (void *)shm_mnt;
2539
2540 if (size < 0 || size > SHMEM_MAX_BYTES)
2541 return ERR_PTR(-EINVAL);
2542
2543 if (shmem_acct_size(flags, size))
2544 return ERR_PTR(-ENOMEM);
2545
2546 error = -ENOMEM;
2547 this.name = name;
2548 this.len = strlen(name);
2549 this.hash = 0; /* will go */
2550 root = shm_mnt->mnt_root;
2551 dentry = d_alloc(root, &this);
2552 if (!dentry)
2553 goto put_memory;
2554
2555 error = -ENFILE;
2556 file = get_empty_filp();
2557 if (!file)
2558 goto put_dentry;
2559
2560 error = -ENOSPC;
2561 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2562 if (!inode)
2563 goto close_file;
2564
2565 SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2566 d_instantiate(dentry, inode);
2567 inode->i_size = size;
2568 inode->i_nlink = 0; /* It is unlinked */
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08002569 file->f_path.mnt = mntget(shm_mnt);
2570 file->f_path.dentry = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 file->f_mapping = inode->i_mapping;
2572 file->f_op = &shmem_file_operations;
2573 file->f_mode = FMODE_WRITE | FMODE_READ;
2574 return file;
2575
2576close_file:
2577 put_filp(file);
2578put_dentry:
2579 dput(dentry);
2580put_memory:
2581 shmem_unacct_size(flags, size);
2582 return ERR_PTR(error);
2583}
2584
2585/*
2586 * shmem_zero_setup - setup a shared anonymous mapping
2587 *
2588 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2589 */
2590int shmem_zero_setup(struct vm_area_struct *vma)
2591{
2592 struct file *file;
2593 loff_t size = vma->vm_end - vma->vm_start;
2594
2595 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2596 if (IS_ERR(file))
2597 return PTR_ERR(file);
2598
2599 if (vma->vm_file)
2600 fput(vma->vm_file);
2601 vma->vm_file = file;
2602 vma->vm_ops = &shmem_vm_ops;
2603 return 0;
2604}