blob: 204865750fe4610362454c27f27753022b0d7425 [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));
918 BUG_ON(page_mapped(page));
919
920 mapping = page->mapping;
921 index = page->index;
922 inode = mapping->host;
923 info = SHMEM_I(inode);
924 if (info->flags & VM_LOCKED)
925 goto redirty;
926 swap = get_swap_page();
927 if (!swap.val)
928 goto redirty;
929
930 spin_lock(&info->lock);
931 shmem_recalc_inode(inode);
932 if (index >= info->next_index) {
933 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
934 goto unlock;
935 }
936 entry = shmem_swp_entry(info, index, NULL);
937 BUG_ON(!entry);
938 BUG_ON(entry->val);
939
940 if (move_to_swap_cache(page, swap) == 0) {
941 shmem_swp_set(info, entry, swap.val);
942 shmem_swp_unmap(entry);
943 spin_unlock(&info->lock);
944 if (list_empty(&info->swaplist)) {
945 spin_lock(&shmem_swaplist_lock);
946 /* move instead of add in case we're racing */
947 list_move_tail(&info->swaplist, &shmem_swaplist);
948 spin_unlock(&shmem_swaplist_lock);
949 }
950 unlock_page(page);
951 return 0;
952 }
953
954 shmem_swp_unmap(entry);
955unlock:
956 spin_unlock(&info->lock);
957 swap_free(swap);
958redirty:
959 set_page_dirty(page);
Zach Brown994fc28c2005-12-15 14:28:17 -0800960 return AOP_WRITEPAGE_ACTIVATE; /* Return with the page locked */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961}
962
963#ifdef CONFIG_NUMA
Hugh Dickinsd15c0232006-03-22 00:08:46 -0800964static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +0000965{
966 char *nodelist = strchr(value, ':');
967 int err = 1;
968
969 if (nodelist) {
970 /* NUL-terminate policy string */
971 *nodelist++ = '\0';
972 if (nodelist_parse(nodelist, *policy_nodes))
973 goto out;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -0700974 if (!nodes_subset(*policy_nodes, node_states[N_HIGH_MEMORY]))
Hugh Dickinsa2109062007-06-08 13:46:46 -0700975 goto out;
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +0000976 }
977 if (!strcmp(value, "default")) {
978 *policy = MPOL_DEFAULT;
979 /* Don't allow a nodelist */
980 if (!nodelist)
981 err = 0;
982 } else if (!strcmp(value, "prefer")) {
983 *policy = MPOL_PREFERRED;
984 /* Insist on a nodelist of one node only */
985 if (nodelist) {
986 char *rest = nodelist;
987 while (isdigit(*rest))
988 rest++;
989 if (!*rest)
990 err = 0;
991 }
992 } else if (!strcmp(value, "bind")) {
993 *policy = MPOL_BIND;
994 /* Insist on a nodelist */
995 if (nodelist)
996 err = 0;
997 } else if (!strcmp(value, "interleave")) {
998 *policy = MPOL_INTERLEAVE;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -0700999 /*
1000 * Default to online nodes with memory if no nodelist
1001 */
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00001002 if (!nodelist)
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001003 *policy_nodes = node_states[N_HIGH_MEMORY];
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00001004 err = 0;
1005 }
1006out:
1007 /* Restore string for error message */
1008 if (nodelist)
1009 *--nodelist = ':';
1010 return err;
1011}
1012
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013static struct page *shmem_swapin_async(struct shared_policy *p,
1014 swp_entry_t entry, unsigned long idx)
1015{
1016 struct page *page;
1017 struct vm_area_struct pvma;
1018
1019 /* Create a pseudo vma that just contains the policy */
1020 memset(&pvma, 0, sizeof(struct vm_area_struct));
1021 pvma.vm_end = PAGE_SIZE;
1022 pvma.vm_pgoff = idx;
1023 pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
1024 page = read_swap_cache_async(entry, &pvma, 0);
1025 mpol_free(pvma.vm_policy);
1026 return page;
1027}
1028
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001029static struct page *shmem_swapin(struct shmem_inode_info *info,
1030 swp_entry_t entry, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001031{
1032 struct shared_policy *p = &info->policy;
1033 int i, num;
1034 struct page *page;
1035 unsigned long offset;
1036
1037 num = valid_swaphandles(entry, &offset);
1038 for (i = 0; i < num; offset++, i++) {
1039 page = shmem_swapin_async(p,
1040 swp_entry(swp_type(entry), offset), idx);
1041 if (!page)
1042 break;
1043 page_cache_release(page);
1044 }
1045 lru_add_drain(); /* Push any new pages onto the LRU now */
1046 return shmem_swapin_async(p, entry, idx);
1047}
1048
1049static struct page *
Al Viro6daa0e22005-10-21 03:18:50 -04001050shmem_alloc_page(gfp_t gfp, struct shmem_inode_info *info,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 unsigned long idx)
1052{
1053 struct vm_area_struct pvma;
1054 struct page *page;
1055
1056 memset(&pvma, 0, sizeof(struct vm_area_struct));
1057 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
1058 pvma.vm_pgoff = idx;
1059 pvma.vm_end = PAGE_SIZE;
1060 page = alloc_page_vma(gfp | __GFP_ZERO, &pvma, 0);
1061 mpol_free(pvma.vm_policy);
1062 return page;
1063}
1064#else
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07001065static inline int shmem_parse_mpol(char *value, int *policy,
1066 nodemask_t *policy_nodes)
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00001067{
1068 return 1;
1069}
1070
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071static inline struct page *
1072shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
1073{
1074 swapin_readahead(entry, 0, NULL);
1075 return read_swap_cache_async(entry, NULL, 0);
1076}
1077
1078static inline struct page *
Al Virodd0fc662005-10-07 07:46:04 +01001079shmem_alloc_page(gfp_t gfp,struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080{
1081 return alloc_page(gfp | __GFP_ZERO);
1082}
1083#endif
1084
1085/*
1086 * shmem_getpage - either get the page from swap or allocate a new one
1087 *
1088 * If we allocate a new one we do not mark it dirty. That's up to the
1089 * vm. If we swap it in we mark it dirty since we also free the swap
1090 * entry since a page cannot live in both the swap and page cache
1091 */
1092static int shmem_getpage(struct inode *inode, unsigned long idx,
1093 struct page **pagep, enum sgp_type sgp, int *type)
1094{
1095 struct address_space *mapping = inode->i_mapping;
1096 struct shmem_inode_info *info = SHMEM_I(inode);
1097 struct shmem_sb_info *sbinfo;
1098 struct page *filepage = *pagep;
1099 struct page *swappage;
1100 swp_entry_t *entry;
1101 swp_entry_t swap;
1102 int error;
1103
1104 if (idx >= SHMEM_MAX_INDEX)
1105 return -EFBIG;
Nick Piggin54cb8822007-07-19 01:46:59 -07001106
1107 if (type)
Nick Piggin83c54072007-07-19 01:47:05 -07001108 *type = 0;
Nick Piggin54cb8822007-07-19 01:46:59 -07001109
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 /*
1111 * Normally, filepage is NULL on entry, and either found
1112 * uptodate immediately, or allocated and zeroed, or read
1113 * in under swappage, which is then assigned to filepage.
Nick Piggin800d15a2007-10-16 01:25:03 -07001114 * But shmem_readpage and shmem_write_begin pass in a locked
Hugh Dickinsae976412007-06-04 10:00:39 +02001115 * filepage, which may be found not uptodate by other callers
1116 * too, and may need to be copied from the swappage read in.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117 */
1118repeat:
1119 if (!filepage)
1120 filepage = find_lock_page(mapping, idx);
1121 if (filepage && PageUptodate(filepage))
1122 goto done;
1123 error = 0;
1124 if (sgp == SGP_QUICK)
1125 goto failed;
1126
1127 spin_lock(&info->lock);
1128 shmem_recalc_inode(inode);
1129 entry = shmem_swp_alloc(info, idx, sgp);
1130 if (IS_ERR(entry)) {
1131 spin_unlock(&info->lock);
1132 error = PTR_ERR(entry);
1133 goto failed;
1134 }
1135 swap = *entry;
1136
1137 if (swap.val) {
1138 /* Look it up and read it in.. */
1139 swappage = lookup_swap_cache(swap);
1140 if (!swappage) {
1141 shmem_swp_unmap(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001142 /* here we actually do the io */
Nick Piggin83c54072007-07-19 01:47:05 -07001143 if (type && !(*type & VM_FAULT_MAJOR)) {
Christoph Lameterf8891e52006-06-30 01:55:45 -07001144 __count_vm_event(PGMAJFAULT);
Nick Piggin83c54072007-07-19 01:47:05 -07001145 *type |= VM_FAULT_MAJOR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 }
Christoph Lameterf8891e52006-06-30 01:55:45 -07001147 spin_unlock(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 swappage = shmem_swapin(info, swap, idx);
1149 if (!swappage) {
1150 spin_lock(&info->lock);
1151 entry = shmem_swp_alloc(info, idx, sgp);
1152 if (IS_ERR(entry))
1153 error = PTR_ERR(entry);
1154 else {
1155 if (entry->val == swap.val)
1156 error = -ENOMEM;
1157 shmem_swp_unmap(entry);
1158 }
1159 spin_unlock(&info->lock);
1160 if (error)
1161 goto failed;
1162 goto repeat;
1163 }
1164 wait_on_page_locked(swappage);
1165 page_cache_release(swappage);
1166 goto repeat;
1167 }
1168
1169 /* We have to do this with page locked to prevent races */
1170 if (TestSetPageLocked(swappage)) {
1171 shmem_swp_unmap(entry);
1172 spin_unlock(&info->lock);
1173 wait_on_page_locked(swappage);
1174 page_cache_release(swappage);
1175 goto repeat;
1176 }
1177 if (PageWriteback(swappage)) {
1178 shmem_swp_unmap(entry);
1179 spin_unlock(&info->lock);
1180 wait_on_page_writeback(swappage);
1181 unlock_page(swappage);
1182 page_cache_release(swappage);
1183 goto repeat;
1184 }
1185 if (!PageUptodate(swappage)) {
1186 shmem_swp_unmap(entry);
1187 spin_unlock(&info->lock);
1188 unlock_page(swappage);
1189 page_cache_release(swappage);
1190 error = -EIO;
1191 goto failed;
1192 }
1193
1194 if (filepage) {
1195 shmem_swp_set(info, entry, 0);
1196 shmem_swp_unmap(entry);
1197 delete_from_swap_cache(swappage);
1198 spin_unlock(&info->lock);
1199 copy_highpage(filepage, swappage);
1200 unlock_page(swappage);
1201 page_cache_release(swappage);
1202 flush_dcache_page(filepage);
1203 SetPageUptodate(filepage);
1204 set_page_dirty(filepage);
1205 swap_free(swap);
1206 } else if (!(error = move_from_swap_cache(
1207 swappage, idx, mapping))) {
1208 info->flags |= SHMEM_PAGEIN;
1209 shmem_swp_set(info, entry, 0);
1210 shmem_swp_unmap(entry);
1211 spin_unlock(&info->lock);
1212 filepage = swappage;
1213 swap_free(swap);
1214 } else {
1215 shmem_swp_unmap(entry);
1216 spin_unlock(&info->lock);
1217 unlock_page(swappage);
1218 page_cache_release(swappage);
1219 if (error == -ENOMEM) {
1220 /* let kswapd refresh zone for GFP_ATOMICs */
Andrew Morton3fcfab12006-10-19 23:28:16 -07001221 congestion_wait(WRITE, HZ/50);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222 }
1223 goto repeat;
1224 }
1225 } else if (sgp == SGP_READ && !filepage) {
1226 shmem_swp_unmap(entry);
1227 filepage = find_get_page(mapping, idx);
1228 if (filepage &&
1229 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1230 spin_unlock(&info->lock);
1231 wait_on_page_locked(filepage);
1232 page_cache_release(filepage);
1233 filepage = NULL;
1234 goto repeat;
1235 }
1236 spin_unlock(&info->lock);
1237 } else {
1238 shmem_swp_unmap(entry);
1239 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001240 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241 spin_lock(&sbinfo->stat_lock);
1242 if (sbinfo->free_blocks == 0 ||
1243 shmem_acct_block(info->flags)) {
1244 spin_unlock(&sbinfo->stat_lock);
1245 spin_unlock(&info->lock);
1246 error = -ENOSPC;
1247 goto failed;
1248 }
1249 sbinfo->free_blocks--;
1250 inode->i_blocks += BLOCKS_PER_PAGE;
1251 spin_unlock(&sbinfo->stat_lock);
1252 } else if (shmem_acct_block(info->flags)) {
1253 spin_unlock(&info->lock);
1254 error = -ENOSPC;
1255 goto failed;
1256 }
1257
1258 if (!filepage) {
1259 spin_unlock(&info->lock);
1260 filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
1261 info,
1262 idx);
1263 if (!filepage) {
1264 shmem_unacct_blocks(info->flags, 1);
1265 shmem_free_blocks(inode, 1);
1266 error = -ENOMEM;
1267 goto failed;
1268 }
1269
1270 spin_lock(&info->lock);
1271 entry = shmem_swp_alloc(info, idx, sgp);
1272 if (IS_ERR(entry))
1273 error = PTR_ERR(entry);
1274 else {
1275 swap = *entry;
1276 shmem_swp_unmap(entry);
1277 }
1278 if (error || swap.val || 0 != add_to_page_cache_lru(
1279 filepage, mapping, idx, GFP_ATOMIC)) {
1280 spin_unlock(&info->lock);
1281 page_cache_release(filepage);
1282 shmem_unacct_blocks(info->flags, 1);
1283 shmem_free_blocks(inode, 1);
1284 filepage = NULL;
1285 if (error)
1286 goto failed;
1287 goto repeat;
1288 }
1289 info->flags |= SHMEM_PAGEIN;
1290 }
1291
1292 info->alloced++;
1293 spin_unlock(&info->lock);
1294 flush_dcache_page(filepage);
1295 SetPageUptodate(filepage);
1296 }
1297done:
1298 if (*pagep != filepage) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299 *pagep = filepage;
Nick Piggin54cb8822007-07-19 01:46:59 -07001300 if (sgp != SGP_FAULT)
Nick Piggind00806b2007-07-19 01:46:57 -07001301 unlock_page(filepage);
1302
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303 }
1304 return 0;
1305
1306failed:
1307 if (*pagep != filepage) {
1308 unlock_page(filepage);
1309 page_cache_release(filepage);
1310 }
1311 return error;
1312}
1313
Nick Piggind0217ac2007-07-19 01:47:03 -07001314static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001315{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001316 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317 int error;
Nick Piggind0217ac2007-07-19 01:47:03 -07001318 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319
Nick Piggind0217ac2007-07-19 01:47:03 -07001320 if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1321 return VM_FAULT_SIGBUS;
Nick Piggind00806b2007-07-19 01:46:57 -07001322
Nick Piggind0217ac2007-07-19 01:47:03 -07001323 error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_FAULT, &ret);
1324 if (error)
1325 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326
Nick Piggind0217ac2007-07-19 01:47:03 -07001327 mark_page_accessed(vmf->page);
Nick Piggin83c54072007-07-19 01:47:05 -07001328 return ret | VM_FAULT_LOCKED;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329}
1330
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331#ifdef CONFIG_NUMA
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001332static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001334 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1336}
1337
Adrian Bunkd8dc74f2007-10-16 01:26:26 -07001338static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1339 unsigned long addr)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001341 struct inode *i = vma->vm_file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 unsigned long idx;
1343
1344 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1345 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1346}
1347#endif
1348
1349int shmem_lock(struct file *file, int lock, struct user_struct *user)
1350{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001351 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 struct shmem_inode_info *info = SHMEM_I(inode);
1353 int retval = -ENOMEM;
1354
1355 spin_lock(&info->lock);
1356 if (lock && !(info->flags & VM_LOCKED)) {
1357 if (!user_shm_lock(inode->i_size, user))
1358 goto out_nomem;
1359 info->flags |= VM_LOCKED;
1360 }
1361 if (!lock && (info->flags & VM_LOCKED) && user) {
1362 user_shm_unlock(inode->i_size, user);
1363 info->flags &= ~VM_LOCKED;
1364 }
1365 retval = 0;
1366out_nomem:
1367 spin_unlock(&info->lock);
1368 return retval;
1369}
1370
Adrian Bunk9b83a6a2007-02-28 20:11:03 -08001371static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372{
1373 file_accessed(file);
1374 vma->vm_ops = &shmem_vm_ops;
Nick Piggind0217ac2007-07-19 01:47:03 -07001375 vma->vm_flags |= VM_CAN_NONLINEAR;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376 return 0;
1377}
1378
1379static struct inode *
1380shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1381{
1382 struct inode *inode;
1383 struct shmem_inode_info *info;
1384 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1385
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001386 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 spin_lock(&sbinfo->stat_lock);
1388 if (!sbinfo->free_inodes) {
1389 spin_unlock(&sbinfo->stat_lock);
1390 return NULL;
1391 }
1392 sbinfo->free_inodes--;
1393 spin_unlock(&sbinfo->stat_lock);
1394 }
1395
1396 inode = new_inode(sb);
1397 if (inode) {
1398 inode->i_mode = mode;
1399 inode->i_uid = current->fsuid;
1400 inode->i_gid = current->fsgid;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 inode->i_blocks = 0;
1402 inode->i_mapping->a_ops = &shmem_aops;
1403 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1404 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
David M. Grimes91828a42006-10-17 00:09:45 -07001405 inode->i_generation = get_seconds();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406 info = SHMEM_I(inode);
1407 memset(info, 0, (char *)inode - (char *)info);
1408 spin_lock_init(&info->lock);
1409 INIT_LIST_HEAD(&info->swaplist);
1410
1411 switch (mode & S_IFMT) {
1412 default:
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001413 inode->i_op = &shmem_special_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 init_special_inode(inode, mode, dev);
1415 break;
1416 case S_IFREG:
1417 inode->i_op = &shmem_inode_operations;
1418 inode->i_fop = &shmem_file_operations;
Robin Holt7339ff82006-01-14 13:20:48 -08001419 mpol_shared_policy_init(&info->policy, sbinfo->policy,
1420 &sbinfo->policy_nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421 break;
1422 case S_IFDIR:
Dave Hansend8c76e62006-09-30 23:29:04 -07001423 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 /* Some things misbehave if size == 0 on a directory */
1425 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1426 inode->i_op = &shmem_dir_inode_operations;
1427 inode->i_fop = &simple_dir_operations;
1428 break;
1429 case S_IFLNK:
1430 /*
1431 * Must not load anything in the rbtree,
1432 * mpol_free_shared_policy will not be called.
1433 */
Robin Holt7339ff82006-01-14 13:20:48 -08001434 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT,
1435 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001436 break;
1437 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001438 } else if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 spin_lock(&sbinfo->stat_lock);
1440 sbinfo->free_inodes++;
1441 spin_unlock(&sbinfo->stat_lock);
1442 }
1443 return inode;
1444}
1445
1446#ifdef CONFIG_TMPFS
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001447static const struct inode_operations shmem_symlink_inode_operations;
1448static const struct inode_operations shmem_symlink_inline_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449
1450/*
Nick Piggin800d15a2007-10-16 01:25:03 -07001451 * Normally tmpfs avoids the use of shmem_readpage and shmem_write_begin;
Hugh Dickinsae976412007-06-04 10:00:39 +02001452 * but providing them allows a tmpfs file to be used for splice, sendfile, and
1453 * below the loop driver, in the generic fashion that many filesystems support.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001454 */
Hugh Dickinsae976412007-06-04 10:00:39 +02001455static int shmem_readpage(struct file *file, struct page *page)
1456{
1457 struct inode *inode = page->mapping->host;
1458 int error = shmem_getpage(inode, page->index, &page, SGP_CACHE, NULL);
1459 unlock_page(page);
1460 return error;
1461}
1462
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463static int
Nick Piggin800d15a2007-10-16 01:25:03 -07001464shmem_write_begin(struct file *file, struct address_space *mapping,
1465 loff_t pos, unsigned len, unsigned flags,
1466 struct page **pagep, void **fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467{
Nick Piggin800d15a2007-10-16 01:25:03 -07001468 struct inode *inode = mapping->host;
1469 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1470 *pagep = NULL;
1471 return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1472}
1473
1474static int
1475shmem_write_end(struct file *file, struct address_space *mapping,
1476 loff_t pos, unsigned len, unsigned copied,
1477 struct page *page, void *fsdata)
1478{
1479 struct inode *inode = mapping->host;
1480
1481 set_page_dirty(page);
1482 page_cache_release(page);
1483
1484 if (pos+copied > inode->i_size)
1485 i_size_write(inode, pos+copied);
1486
1487 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488}
1489
1490static ssize_t
1491shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1492{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001493 struct inode *inode = file->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494 loff_t pos;
1495 unsigned long written;
1496 ssize_t err;
1497
1498 if ((ssize_t) count < 0)
1499 return -EINVAL;
1500
1501 if (!access_ok(VERIFY_READ, buf, count))
1502 return -EFAULT;
1503
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001504 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505
1506 pos = *ppos;
1507 written = 0;
1508
1509 err = generic_write_checks(file, &pos, &count, 0);
1510 if (err || !count)
1511 goto out;
1512
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001513 err = remove_suid(file->f_path.dentry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 if (err)
1515 goto out;
1516
1517 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1518
1519 do {
1520 struct page *page = NULL;
1521 unsigned long bytes, index, offset;
1522 char *kaddr;
1523 int left;
1524
1525 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1526 index = pos >> PAGE_CACHE_SHIFT;
1527 bytes = PAGE_CACHE_SIZE - offset;
1528 if (bytes > count)
1529 bytes = count;
1530
1531 /*
1532 * We don't hold page lock across copy from user -
1533 * what would it guard against? - so no deadlock here.
1534 * But it still may be a good idea to prefault below.
1535 */
1536
1537 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1538 if (err)
1539 break;
1540
1541 left = bytes;
1542 if (PageHighMem(page)) {
1543 volatile unsigned char dummy;
1544 __get_user(dummy, buf);
1545 __get_user(dummy, buf + bytes - 1);
1546
1547 kaddr = kmap_atomic(page, KM_USER0);
1548 left = __copy_from_user_inatomic(kaddr + offset,
1549 buf, bytes);
1550 kunmap_atomic(kaddr, KM_USER0);
1551 }
1552 if (left) {
1553 kaddr = kmap(page);
1554 left = __copy_from_user(kaddr + offset, buf, bytes);
1555 kunmap(page);
1556 }
1557
1558 written += bytes;
1559 count -= bytes;
1560 pos += bytes;
1561 buf += bytes;
1562 if (pos > inode->i_size)
1563 i_size_write(inode, pos);
1564
1565 flush_dcache_page(page);
1566 set_page_dirty(page);
1567 mark_page_accessed(page);
1568 page_cache_release(page);
1569
1570 if (left) {
1571 pos -= left;
1572 written -= left;
1573 err = -EFAULT;
1574 break;
1575 }
1576
1577 /*
1578 * Our dirty pages are not counted in nr_dirty,
1579 * and we do not attempt to balance dirty pages.
1580 */
1581
1582 cond_resched();
1583 } while (count);
1584
1585 *ppos = pos;
1586 if (written)
1587 err = written;
1588out:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001589 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001590 return err;
1591}
1592
1593static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1594{
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08001595 struct inode *inode = filp->f_path.dentry->d_inode;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596 struct address_space *mapping = inode->i_mapping;
1597 unsigned long index, offset;
1598
1599 index = *ppos >> PAGE_CACHE_SHIFT;
1600 offset = *ppos & ~PAGE_CACHE_MASK;
1601
1602 for (;;) {
1603 struct page *page = NULL;
1604 unsigned long end_index, nr, ret;
1605 loff_t i_size = i_size_read(inode);
1606
1607 end_index = i_size >> PAGE_CACHE_SHIFT;
1608 if (index > end_index)
1609 break;
1610 if (index == end_index) {
1611 nr = i_size & ~PAGE_CACHE_MASK;
1612 if (nr <= offset)
1613 break;
1614 }
1615
1616 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1617 if (desc->error) {
1618 if (desc->error == -EINVAL)
1619 desc->error = 0;
1620 break;
1621 }
1622
1623 /*
1624 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001625 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626 */
1627 nr = PAGE_CACHE_SIZE;
1628 i_size = i_size_read(inode);
1629 end_index = i_size >> PAGE_CACHE_SHIFT;
1630 if (index == end_index) {
1631 nr = i_size & ~PAGE_CACHE_MASK;
1632 if (nr <= offset) {
1633 if (page)
1634 page_cache_release(page);
1635 break;
1636 }
1637 }
1638 nr -= offset;
1639
1640 if (page) {
1641 /*
1642 * If users can be writing to this page using arbitrary
1643 * virtual addresses, take care about potential aliasing
1644 * before reading the page on the kernel side.
1645 */
1646 if (mapping_writably_mapped(mapping))
1647 flush_dcache_page(page);
1648 /*
1649 * Mark the page accessed if we read the beginning.
1650 */
1651 if (!offset)
1652 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001653 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001654 page = ZERO_PAGE(0);
Nick Pigginb5810032005-10-29 18:16:12 -07001655 page_cache_get(page);
1656 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657
1658 /*
1659 * Ok, we have the page, and it's up-to-date, so
1660 * now we can copy it to user space...
1661 *
1662 * The actor routine returns how many bytes were actually used..
1663 * NOTE! This may not be the same as how much of a user buffer
1664 * we filled up (we may be padding etc), so we can only update
1665 * "pos" here (the actor routine has to update the user buffer
1666 * pointers and the remaining count).
1667 */
1668 ret = actor(desc, page, offset, nr);
1669 offset += ret;
1670 index += offset >> PAGE_CACHE_SHIFT;
1671 offset &= ~PAGE_CACHE_MASK;
1672
1673 page_cache_release(page);
1674 if (ret != nr || !desc->count)
1675 break;
1676
1677 cond_resched();
1678 }
1679
1680 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1681 file_accessed(filp);
1682}
1683
1684static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1685{
1686 read_descriptor_t desc;
1687
1688 if ((ssize_t) count < 0)
1689 return -EINVAL;
1690 if (!access_ok(VERIFY_WRITE, buf, count))
1691 return -EFAULT;
1692 if (!count)
1693 return 0;
1694
1695 desc.written = 0;
1696 desc.count = count;
1697 desc.arg.buf = buf;
1698 desc.error = 0;
1699
1700 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1701 if (desc.written)
1702 return desc.written;
1703 return desc.error;
1704}
1705
David Howells726c3342006-06-23 02:02:58 -07001706static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707{
David Howells726c3342006-06-23 02:02:58 -07001708 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709
1710 buf->f_type = TMPFS_MAGIC;
1711 buf->f_bsize = PAGE_CACHE_SIZE;
1712 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001713 spin_lock(&sbinfo->stat_lock);
1714 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 buf->f_blocks = sbinfo->max_blocks;
1716 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001717 }
1718 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 buf->f_files = sbinfo->max_inodes;
1720 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 }
1722 /* else leave those fields 0 like simple_statfs */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001723 spin_unlock(&sbinfo->stat_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724 return 0;
1725}
1726
1727/*
1728 * File creation. Allocate an inode, and we're done..
1729 */
1730static int
1731shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1732{
1733 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1734 int error = -ENOSPC;
1735
1736 if (inode) {
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001737 error = security_inode_init_security(inode, dir, NULL, NULL,
1738 NULL);
1739 if (error) {
1740 if (error != -EOPNOTSUPP) {
1741 iput(inode);
1742 return error;
1743 }
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001744 }
1745 error = shmem_acl_init(inode, dir);
1746 if (error) {
1747 iput(inode);
1748 return error;
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001749 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 if (dir->i_mode & S_ISGID) {
1751 inode->i_gid = dir->i_gid;
1752 if (S_ISDIR(mode))
1753 inode->i_mode |= S_ISGID;
1754 }
1755 dir->i_size += BOGO_DIRENT_SIZE;
1756 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1757 d_instantiate(dentry, inode);
1758 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 }
1760 return error;
1761}
1762
1763static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1764{
1765 int error;
1766
1767 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1768 return error;
Dave Hansend8c76e62006-09-30 23:29:04 -07001769 inc_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 return 0;
1771}
1772
1773static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1774 struct nameidata *nd)
1775{
1776 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1777}
1778
1779/*
1780 * Link a file..
1781 */
1782static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1783{
1784 struct inode *inode = old_dentry->d_inode;
1785 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1786
1787 /*
1788 * No ordinary (disk based) filesystem counts links as inodes;
1789 * but each new link needs a new dentry, pinning lowmem, and
1790 * tmpfs dentries cannot be pruned until they are unlinked.
1791 */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001792 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 spin_lock(&sbinfo->stat_lock);
1794 if (!sbinfo->free_inodes) {
1795 spin_unlock(&sbinfo->stat_lock);
1796 return -ENOSPC;
1797 }
1798 sbinfo->free_inodes--;
1799 spin_unlock(&sbinfo->stat_lock);
1800 }
1801
1802 dir->i_size += BOGO_DIRENT_SIZE;
1803 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansend8c76e62006-09-30 23:29:04 -07001804 inc_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 atomic_inc(&inode->i_count); /* New dentry reference */
1806 dget(dentry); /* Extra pinning count for the created dentry */
1807 d_instantiate(dentry, inode);
1808 return 0;
1809}
1810
1811static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1812{
1813 struct inode *inode = dentry->d_inode;
1814
1815 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) {
1816 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001817 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818 spin_lock(&sbinfo->stat_lock);
1819 sbinfo->free_inodes++;
1820 spin_unlock(&sbinfo->stat_lock);
1821 }
1822 }
1823
1824 dir->i_size -= BOGO_DIRENT_SIZE;
1825 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001826 drop_nlink(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 dput(dentry); /* Undo the count from "create" - this does all the work */
1828 return 0;
1829}
1830
1831static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1832{
1833 if (!simple_empty(dentry))
1834 return -ENOTEMPTY;
1835
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001836 drop_nlink(dentry->d_inode);
1837 drop_nlink(dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 return shmem_unlink(dir, dentry);
1839}
1840
1841/*
1842 * The VFS layer already does all the dentry stuff for rename,
1843 * we just have to decrement the usage count for the target if
1844 * it exists so that the VFS layer correctly free's it when it
1845 * gets overwritten.
1846 */
1847static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1848{
1849 struct inode *inode = old_dentry->d_inode;
1850 int they_are_dirs = S_ISDIR(inode->i_mode);
1851
1852 if (!simple_empty(new_dentry))
1853 return -ENOTEMPTY;
1854
1855 if (new_dentry->d_inode) {
1856 (void) shmem_unlink(new_dir, new_dentry);
1857 if (they_are_dirs)
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001858 drop_nlink(old_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859 } else if (they_are_dirs) {
Dave Hansen9a53c3a2006-09-30 23:29:03 -07001860 drop_nlink(old_dir);
Dave Hansend8c76e62006-09-30 23:29:04 -07001861 inc_nlink(new_dir);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 }
1863
1864 old_dir->i_size -= BOGO_DIRENT_SIZE;
1865 new_dir->i_size += BOGO_DIRENT_SIZE;
1866 old_dir->i_ctime = old_dir->i_mtime =
1867 new_dir->i_ctime = new_dir->i_mtime =
1868 inode->i_ctime = CURRENT_TIME;
1869 return 0;
1870}
1871
1872static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1873{
1874 int error;
1875 int len;
1876 struct inode *inode;
1877 struct page *page = NULL;
1878 char *kaddr;
1879 struct shmem_inode_info *info;
1880
1881 len = strlen(symname) + 1;
1882 if (len > PAGE_CACHE_SIZE)
1883 return -ENAMETOOLONG;
1884
1885 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1886 if (!inode)
1887 return -ENOSPC;
1888
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001889 error = security_inode_init_security(inode, dir, NULL, NULL,
1890 NULL);
1891 if (error) {
1892 if (error != -EOPNOTSUPP) {
1893 iput(inode);
1894 return error;
1895 }
1896 error = 0;
1897 }
1898
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899 info = SHMEM_I(inode);
1900 inode->i_size = len-1;
1901 if (len <= (char *)inode - (char *)info) {
1902 /* do it inline */
1903 memcpy(info, symname, len);
1904 inode->i_op = &shmem_symlink_inline_operations;
1905 } else {
1906 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1907 if (error) {
1908 iput(inode);
1909 return error;
1910 }
1911 inode->i_op = &shmem_symlink_inode_operations;
1912 kaddr = kmap_atomic(page, KM_USER0);
1913 memcpy(kaddr, symname, len);
1914 kunmap_atomic(kaddr, KM_USER0);
1915 set_page_dirty(page);
1916 page_cache_release(page);
1917 }
1918 if (dir->i_mode & S_ISGID)
1919 inode->i_gid = dir->i_gid;
1920 dir->i_size += BOGO_DIRENT_SIZE;
1921 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1922 d_instantiate(dentry, inode);
1923 dget(dentry);
1924 return 0;
1925}
1926
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001927static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928{
1929 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001930 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931}
1932
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001933static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934{
1935 struct page *page = NULL;
1936 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1937 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001938 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939}
1940
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001941static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942{
1943 if (!IS_ERR(nd_get_link(nd))) {
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001944 struct page *page = cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 kunmap(page);
1946 mark_page_accessed(page);
1947 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001948 }
1949}
1950
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001951static const struct inode_operations shmem_symlink_inline_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 .readlink = generic_readlink,
1953 .follow_link = shmem_follow_link_inline,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954};
1955
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08001956static const struct inode_operations shmem_symlink_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 .truncate = shmem_truncate,
1958 .readlink = generic_readlink,
1959 .follow_link = shmem_follow_link,
1960 .put_link = shmem_put_link,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001961};
1962
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001963#ifdef CONFIG_TMPFS_POSIX_ACL
1964/**
1965 * Superblocks without xattr inode operations will get security.* xattr
1966 * support from the VFS "for free". As soon as we have any other xattrs
1967 * like ACLs, we also need to implement the security.* handlers at
1968 * filesystem level, though.
1969 */
1970
1971static size_t shmem_xattr_security_list(struct inode *inode, char *list,
1972 size_t list_len, const char *name,
1973 size_t name_len)
1974{
1975 return security_inode_listsecurity(inode, list, list_len);
1976}
1977
1978static int shmem_xattr_security_get(struct inode *inode, const char *name,
1979 void *buffer, size_t size)
1980{
1981 if (strcmp(name, "") == 0)
1982 return -EINVAL;
1983 return security_inode_getsecurity(inode, name, buffer, size,
1984 -EOPNOTSUPP);
1985}
1986
1987static int shmem_xattr_security_set(struct inode *inode, const char *name,
1988 const void *value, size_t size, int flags)
1989{
1990 if (strcmp(name, "") == 0)
1991 return -EINVAL;
1992 return security_inode_setsecurity(inode, name, value, size, flags);
1993}
1994
Adrian Bunk1f370a22006-12-06 20:38:21 -08001995static struct xattr_handler shmem_xattr_security_handler = {
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07001996 .prefix = XATTR_SECURITY_PREFIX,
1997 .list = shmem_xattr_security_list,
1998 .get = shmem_xattr_security_get,
1999 .set = shmem_xattr_security_set,
2000};
2001
2002static struct xattr_handler *shmem_xattr_handlers[] = {
2003 &shmem_xattr_acl_access_handler,
2004 &shmem_xattr_acl_default_handler,
2005 &shmem_xattr_security_handler,
2006 NULL
2007};
2008#endif
2009
David M. Grimes91828a42006-10-17 00:09:45 -07002010static struct dentry *shmem_get_parent(struct dentry *child)
2011{
2012 return ERR_PTR(-ESTALE);
2013}
2014
2015static int shmem_match(struct inode *ino, void *vfh)
2016{
2017 __u32 *fh = vfh;
2018 __u64 inum = fh[2];
2019 inum = (inum << 32) | fh[1];
2020 return ino->i_ino == inum && fh[0] == ino->i_generation;
2021}
2022
2023static struct dentry *shmem_get_dentry(struct super_block *sb, void *vfh)
2024{
2025 struct dentry *de = NULL;
2026 struct inode *inode;
2027 __u32 *fh = vfh;
2028 __u64 inum = fh[2];
2029 inum = (inum << 32) | fh[1];
2030
2031 inode = ilookup5(sb, (unsigned long)(inum+fh[0]), shmem_match, vfh);
2032 if (inode) {
2033 de = d_find_alias(inode);
2034 iput(inode);
2035 }
2036
2037 return de? de: ERR_PTR(-ESTALE);
2038}
2039
2040static struct dentry *shmem_decode_fh(struct super_block *sb, __u32 *fh,
2041 int len, int type,
2042 int (*acceptable)(void *context, struct dentry *de),
2043 void *context)
2044{
2045 if (len < 3)
2046 return ERR_PTR(-ESTALE);
2047
2048 return sb->s_export_op->find_exported_dentry(sb, fh, NULL, acceptable,
2049 context);
2050}
2051
2052static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2053 int connectable)
2054{
2055 struct inode *inode = dentry->d_inode;
2056
2057 if (*len < 3)
2058 return 255;
2059
2060 if (hlist_unhashed(&inode->i_hash)) {
2061 /* Unfortunately insert_inode_hash is not idempotent,
2062 * so as we hash inodes here rather than at creation
2063 * time, we need a lock to ensure we only try
2064 * to do it once
2065 */
2066 static DEFINE_SPINLOCK(lock);
2067 spin_lock(&lock);
2068 if (hlist_unhashed(&inode->i_hash))
2069 __insert_inode_hash(inode,
2070 inode->i_ino + inode->i_generation);
2071 spin_unlock(&lock);
2072 }
2073
2074 fh[0] = inode->i_generation;
2075 fh[1] = inode->i_ino;
2076 fh[2] = ((__u64)inode->i_ino) >> 32;
2077
2078 *len = 3;
2079 return 1;
2080}
2081
2082static struct export_operations shmem_export_ops = {
2083 .get_parent = shmem_get_parent,
2084 .get_dentry = shmem_get_dentry,
2085 .encode_fh = shmem_encode_fh,
2086 .decode_fh = shmem_decode_fh,
2087};
2088
Robin Holt7339ff82006-01-14 13:20:48 -08002089static int shmem_parse_options(char *options, int *mode, uid_t *uid,
2090 gid_t *gid, unsigned long *blocks, unsigned long *inodes,
2091 int *policy, nodemask_t *policy_nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092{
2093 char *this_char, *value, *rest;
2094
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002095 while (options != NULL) {
2096 this_char = options;
2097 for (;;) {
2098 /*
2099 * NUL-terminate this option: unfortunately,
2100 * mount options form a comma-separated list,
2101 * but mpol's nodelist may also contain commas.
2102 */
2103 options = strchr(options, ',');
2104 if (options == NULL)
2105 break;
2106 options++;
2107 if (!isdigit(*options)) {
2108 options[-1] = '\0';
2109 break;
2110 }
2111 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 if (!*this_char)
2113 continue;
2114 if ((value = strchr(this_char,'=')) != NULL) {
2115 *value++ = 0;
2116 } else {
2117 printk(KERN_ERR
2118 "tmpfs: No value for mount option '%s'\n",
2119 this_char);
2120 return 1;
2121 }
2122
2123 if (!strcmp(this_char,"size")) {
2124 unsigned long long size;
2125 size = memparse(value,&rest);
2126 if (*rest == '%') {
2127 size <<= PAGE_SHIFT;
2128 size *= totalram_pages;
2129 do_div(size, 100);
2130 rest++;
2131 }
2132 if (*rest)
2133 goto bad_val;
2134 *blocks = size >> PAGE_CACHE_SHIFT;
2135 } else if (!strcmp(this_char,"nr_blocks")) {
2136 *blocks = memparse(value,&rest);
2137 if (*rest)
2138 goto bad_val;
2139 } else if (!strcmp(this_char,"nr_inodes")) {
2140 *inodes = memparse(value,&rest);
2141 if (*rest)
2142 goto bad_val;
2143 } else if (!strcmp(this_char,"mode")) {
2144 if (!mode)
2145 continue;
2146 *mode = simple_strtoul(value,&rest,8);
2147 if (*rest)
2148 goto bad_val;
2149 } else if (!strcmp(this_char,"uid")) {
2150 if (!uid)
2151 continue;
2152 *uid = simple_strtoul(value,&rest,0);
2153 if (*rest)
2154 goto bad_val;
2155 } else if (!strcmp(this_char,"gid")) {
2156 if (!gid)
2157 continue;
2158 *gid = simple_strtoul(value,&rest,0);
2159 if (*rest)
2160 goto bad_val;
Robin Holt7339ff82006-01-14 13:20:48 -08002161 } else if (!strcmp(this_char,"mpol")) {
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00002162 if (shmem_parse_mpol(value,policy,policy_nodes))
Robin Holt7339ff82006-01-14 13:20:48 -08002163 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 } else {
2165 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2166 this_char);
2167 return 1;
2168 }
2169 }
2170 return 0;
2171
2172bad_val:
2173 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2174 value, this_char);
2175 return 1;
2176
2177}
2178
2179static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2180{
2181 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002182 unsigned long max_blocks = sbinfo->max_blocks;
2183 unsigned long max_inodes = sbinfo->max_inodes;
Robin Holt7339ff82006-01-14 13:20:48 -08002184 int policy = sbinfo->policy;
2185 nodemask_t policy_nodes = sbinfo->policy_nodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002186 unsigned long blocks;
2187 unsigned long inodes;
2188 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002189
Robin Holt7339ff82006-01-14 13:20:48 -08002190 if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks,
2191 &max_inodes, &policy, &policy_nodes))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002192 return error;
2193
2194 spin_lock(&sbinfo->stat_lock);
2195 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2196 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2197 if (max_blocks < blocks)
2198 goto out;
2199 if (max_inodes < inodes)
2200 goto out;
2201 /*
2202 * Those tests also disallow limited->unlimited while any are in
2203 * use, so i_blocks will always be zero when max_blocks is zero;
2204 * but we must separately disallow unlimited->limited, because
2205 * in that case we have no record of how much is already in use.
2206 */
2207 if (max_blocks && !sbinfo->max_blocks)
2208 goto out;
2209 if (max_inodes && !sbinfo->max_inodes)
2210 goto out;
2211
2212 error = 0;
2213 sbinfo->max_blocks = max_blocks;
2214 sbinfo->free_blocks = max_blocks - blocks;
2215 sbinfo->max_inodes = max_inodes;
2216 sbinfo->free_inodes = max_inodes - inodes;
Robin Holt7339ff82006-01-14 13:20:48 -08002217 sbinfo->policy = policy;
2218 sbinfo->policy_nodes = policy_nodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002219out:
2220 spin_unlock(&sbinfo->stat_lock);
2221 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222}
2223#endif
2224
2225static void shmem_put_super(struct super_block *sb)
2226{
2227 kfree(sb->s_fs_info);
2228 sb->s_fs_info = NULL;
2229}
2230
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231static int shmem_fill_super(struct super_block *sb,
2232 void *data, int silent)
2233{
2234 struct inode *inode;
2235 struct dentry *root;
2236 int mode = S_IRWXUGO | S_ISVTX;
2237 uid_t uid = current->fsuid;
2238 gid_t gid = current->fsgid;
2239 int err = -ENOMEM;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002240 struct shmem_sb_info *sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 unsigned long blocks = 0;
2242 unsigned long inodes = 0;
Robin Holt7339ff82006-01-14 13:20:48 -08002243 int policy = MPOL_DEFAULT;
Lee Schermerhorn37b07e42007-10-16 01:25:39 -07002244 nodemask_t policy_nodes = node_states[N_HIGH_MEMORY];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002245
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002246#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 /*
2248 * Per default we only allow half of the physical ram per
2249 * tmpfs instance, limiting inodes to one per page of lowmem;
2250 * but the internal instance is left unlimited.
2251 */
2252 if (!(sb->s_flags & MS_NOUSER)) {
2253 blocks = totalram_pages / 2;
2254 inodes = totalram_pages - totalhigh_pages;
2255 if (inodes > blocks)
2256 inodes = blocks;
Robin Holt7339ff82006-01-14 13:20:48 -08002257 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks,
2258 &inodes, &policy, &policy_nodes))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259 return -EINVAL;
2260 }
David M. Grimes91828a42006-10-17 00:09:45 -07002261 sb->s_export_op = &shmem_export_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262#else
2263 sb->s_flags |= MS_NOUSER;
2264#endif
2265
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002266 /* Round up to L1_CACHE_BYTES to resist false sharing */
2267 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
2268 L1_CACHE_BYTES), GFP_KERNEL);
2269 if (!sbinfo)
2270 return -ENOMEM;
2271
2272 spin_lock_init(&sbinfo->stat_lock);
2273 sbinfo->max_blocks = blocks;
2274 sbinfo->free_blocks = blocks;
2275 sbinfo->max_inodes = inodes;
2276 sbinfo->free_inodes = inodes;
Robin Holt7339ff82006-01-14 13:20:48 -08002277 sbinfo->policy = policy;
2278 sbinfo->policy_nodes = policy_nodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002279
2280 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 sb->s_maxbytes = SHMEM_MAX_BYTES;
2282 sb->s_blocksize = PAGE_CACHE_SIZE;
2283 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2284 sb->s_magic = TMPFS_MAGIC;
2285 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01002286 sb->s_time_gran = 1;
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002287#ifdef CONFIG_TMPFS_POSIX_ACL
2288 sb->s_xattr = shmem_xattr_handlers;
2289 sb->s_flags |= MS_POSIXACL;
2290#endif
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002291
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
2293 if (!inode)
2294 goto failed;
2295 inode->i_uid = uid;
2296 inode->i_gid = gid;
2297 root = d_alloc_root(inode);
2298 if (!root)
2299 goto failed_iput;
2300 sb->s_root = root;
2301 return 0;
2302
2303failed_iput:
2304 iput(inode);
2305failed:
2306 shmem_put_super(sb);
2307 return err;
2308}
2309
Pekka Enbergfcc234f2006-03-22 00:08:13 -08002310static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311
2312static struct inode *shmem_alloc_inode(struct super_block *sb)
2313{
2314 struct shmem_inode_info *p;
Christoph Lametere94b1762006-12-06 20:33:17 -08002315 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 if (!p)
2317 return NULL;
2318 return &p->vfs_inode;
2319}
2320
2321static void shmem_destroy_inode(struct inode *inode)
2322{
2323 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2324 /* only struct inode is valid if it's an inline symlink */
2325 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2326 }
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002327 shmem_acl_destroy_inode(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2329}
2330
Christoph Lameter4ba9b9d2007-10-16 23:25:51 -07002331static void init_once(struct kmem_cache *cachep, void *foo)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332{
2333 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2334
Christoph Lametera35afb82007-05-16 22:10:57 -07002335 inode_init_once(&p->vfs_inode);
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002336#ifdef CONFIG_TMPFS_POSIX_ACL
Christoph Lametera35afb82007-05-16 22:10:57 -07002337 p->i_acl = NULL;
2338 p->i_default_acl = NULL;
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002339#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340}
2341
2342static int init_inodecache(void)
2343{
2344 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2345 sizeof(struct shmem_inode_info),
Paul Mundt20c2df82007-07-20 10:11:58 +09002346 0, 0, init_once);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347 if (shmem_inode_cachep == NULL)
2348 return -ENOMEM;
2349 return 0;
2350}
2351
2352static void destroy_inodecache(void)
2353{
Alexey Dobriyan1a1d92c2006-09-27 01:49:40 -07002354 kmem_cache_destroy(shmem_inode_cachep);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002355}
2356
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002357static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358 .writepage = shmem_writepage,
Ken Chen76719322007-02-10 01:43:15 -08002359 .set_page_dirty = __set_page_dirty_no_writeback,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360#ifdef CONFIG_TMPFS
Hugh Dickinsae976412007-06-04 10:00:39 +02002361 .readpage = shmem_readpage,
Nick Piggin800d15a2007-10-16 01:25:03 -07002362 .write_begin = shmem_write_begin,
2363 .write_end = shmem_write_end,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364#endif
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07002365 .migratepage = migrate_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366};
2367
Helge Deller15ad7cd2006-12-06 20:40:36 -08002368static const struct file_operations shmem_file_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 .mmap = shmem_mmap,
2370#ifdef CONFIG_TMPFS
2371 .llseek = generic_file_llseek,
2372 .read = shmem_file_read,
2373 .write = shmem_file_write,
2374 .fsync = simple_sync_file,
Hugh Dickinsae976412007-06-04 10:00:39 +02002375 .splice_read = generic_file_splice_read,
2376 .splice_write = generic_file_splice_write,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377#endif
2378};
2379
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002380static const struct inode_operations shmem_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 .truncate = shmem_truncate,
2382 .setattr = shmem_notify_change,
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002383 .truncate_range = shmem_truncate_range,
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002384#ifdef CONFIG_TMPFS_POSIX_ACL
2385 .setxattr = generic_setxattr,
2386 .getxattr = generic_getxattr,
2387 .listxattr = generic_listxattr,
2388 .removexattr = generic_removexattr,
2389 .permission = shmem_permission,
2390#endif
2391
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392};
2393
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002394static const struct inode_operations shmem_dir_inode_operations = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395#ifdef CONFIG_TMPFS
2396 .create = shmem_create,
2397 .lookup = simple_lookup,
2398 .link = shmem_link,
2399 .unlink = shmem_unlink,
2400 .symlink = shmem_symlink,
2401 .mkdir = shmem_mkdir,
2402 .rmdir = shmem_rmdir,
2403 .mknod = shmem_mknod,
2404 .rename = shmem_rename,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405#endif
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002406#ifdef CONFIG_TMPFS_POSIX_ACL
2407 .setattr = shmem_notify_change,
2408 .setxattr = generic_setxattr,
2409 .getxattr = generic_getxattr,
2410 .listxattr = generic_listxattr,
2411 .removexattr = generic_removexattr,
2412 .permission = shmem_permission,
2413#endif
2414};
2415
Arjan van de Ven92e1d5b2007-02-12 00:55:39 -08002416static const struct inode_operations shmem_special_inode_operations = {
Andreas Gruenbacher39f02472006-09-29 02:01:35 -07002417#ifdef CONFIG_TMPFS_POSIX_ACL
2418 .setattr = shmem_notify_change,
2419 .setxattr = generic_setxattr,
2420 .getxattr = generic_getxattr,
2421 .listxattr = generic_listxattr,
2422 .removexattr = generic_removexattr,
2423 .permission = shmem_permission,
2424#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002425};
2426
Hugh Dickins759b9772007-03-05 00:30:28 -08002427static const struct super_operations shmem_ops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428 .alloc_inode = shmem_alloc_inode,
2429 .destroy_inode = shmem_destroy_inode,
2430#ifdef CONFIG_TMPFS
2431 .statfs = shmem_statfs,
2432 .remount_fs = shmem_remount_fs,
2433#endif
2434 .delete_inode = shmem_delete_inode,
2435 .drop_inode = generic_delete_inode,
2436 .put_super = shmem_put_super,
2437};
2438
2439static struct vm_operations_struct shmem_vm_ops = {
Nick Piggin54cb8822007-07-19 01:46:59 -07002440 .fault = shmem_fault,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441#ifdef CONFIG_NUMA
2442 .set_policy = shmem_set_policy,
2443 .get_policy = shmem_get_policy,
2444#endif
2445};
2446
2447
David Howells454e2392006-06-23 02:02:57 -07002448static int shmem_get_sb(struct file_system_type *fs_type,
2449 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002450{
David Howells454e2392006-06-23 02:02:57 -07002451 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452}
2453
2454static struct file_system_type tmpfs_fs_type = {
2455 .owner = THIS_MODULE,
2456 .name = "tmpfs",
2457 .get_sb = shmem_get_sb,
2458 .kill_sb = kill_litter_super,
2459};
2460static struct vfsmount *shm_mnt;
2461
2462static int __init init_tmpfs(void)
2463{
2464 int error;
2465
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002466 error = bdi_init(&shmem_backing_dev_info);
2467 if (error)
2468 goto out4;
2469
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470 error = init_inodecache();
2471 if (error)
2472 goto out3;
2473
2474 error = register_filesystem(&tmpfs_fs_type);
2475 if (error) {
2476 printk(KERN_ERR "Could not register tmpfs\n");
2477 goto out2;
2478 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07002479
Trond Myklebust1f5ce9e2006-06-09 09:34:16 -04002480 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 tmpfs_fs_type.name, NULL);
2482 if (IS_ERR(shm_mnt)) {
2483 error = PTR_ERR(shm_mnt);
2484 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2485 goto out1;
2486 }
2487 return 0;
2488
2489out1:
2490 unregister_filesystem(&tmpfs_fs_type);
2491out2:
2492 destroy_inodecache();
2493out3:
Peter Zijlstrae0bf68d2007-10-16 23:25:46 -07002494 bdi_destroy(&shmem_backing_dev_info);
2495out4:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496 shm_mnt = ERR_PTR(error);
2497 return error;
2498}
2499module_init(init_tmpfs)
2500
2501/*
2502 * shmem_file_setup - get an unlinked file living in tmpfs
2503 *
2504 * @name: name for dentry (to be seen in /proc/<pid>/maps
2505 * @size: size to be set for the file
2506 *
2507 */
2508struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2509{
2510 int error;
2511 struct file *file;
2512 struct inode *inode;
2513 struct dentry *dentry, *root;
2514 struct qstr this;
2515
2516 if (IS_ERR(shm_mnt))
2517 return (void *)shm_mnt;
2518
2519 if (size < 0 || size > SHMEM_MAX_BYTES)
2520 return ERR_PTR(-EINVAL);
2521
2522 if (shmem_acct_size(flags, size))
2523 return ERR_PTR(-ENOMEM);
2524
2525 error = -ENOMEM;
2526 this.name = name;
2527 this.len = strlen(name);
2528 this.hash = 0; /* will go */
2529 root = shm_mnt->mnt_root;
2530 dentry = d_alloc(root, &this);
2531 if (!dentry)
2532 goto put_memory;
2533
2534 error = -ENFILE;
2535 file = get_empty_filp();
2536 if (!file)
2537 goto put_dentry;
2538
2539 error = -ENOSPC;
2540 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2541 if (!inode)
2542 goto close_file;
2543
2544 SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2545 d_instantiate(dentry, inode);
2546 inode->i_size = size;
2547 inode->i_nlink = 0; /* It is unlinked */
Josef "Jeff" Sipekd3ac7f82006-12-08 02:36:44 -08002548 file->f_path.mnt = mntget(shm_mnt);
2549 file->f_path.dentry = dentry;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002550 file->f_mapping = inode->i_mapping;
2551 file->f_op = &shmem_file_operations;
2552 file->f_mode = FMODE_WRITE | FMODE_READ;
2553 return file;
2554
2555close_file:
2556 put_filp(file);
2557put_dentry:
2558 dput(dentry);
2559put_memory:
2560 shmem_unacct_size(flags, size);
2561 return ERR_PTR(error);
2562}
2563
2564/*
2565 * shmem_zero_setup - setup a shared anonymous mapping
2566 *
2567 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2568 */
2569int shmem_zero_setup(struct vm_area_struct *vma)
2570{
2571 struct file *file;
2572 loff_t size = vma->vm_end - vma->vm_start;
2573
2574 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2575 if (IS_ERR(file))
2576 return PTR_ERR(file);
2577
2578 if (vma->vm_file)
2579 fput(vma->vm_file);
2580 vma->vm_file = file;
2581 vma->vm_ops = &shmem_vm_ops;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002582 return 0;
2583}