blob: ce501bce1c2e2369959666de3923e0085112bb73 [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
26#include <linux/config.h>
27#include <linux/module.h>
28#include <linux/init.h>
29#include <linux/devfs_fs_kernel.h>
30#include <linux/fs.h>
31#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>
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#include <asm/uaccess.h>
49#include <asm/div64.h>
50#include <asm/pgtable.h>
51
52/* This magic number is used in glibc for posix shared memory */
53#define TMPFS_MAGIC 0x01021994
54
55#define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
56#define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
57#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
58
59#define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
60#define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
61
62#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
63
64/* info->flags needs VM_flags to handle pagein/truncate races efficiently */
65#define SHMEM_PAGEIN VM_READ
66#define SHMEM_TRUNCATE VM_WRITE
67
68/* Definition to limit shmem_truncate's steps between cond_rescheds */
69#define LATENCY_LIMIT 64
70
71/* Pretend that each entry is of this size in directory's i_size */
72#define BOGO_DIRENT_SIZE 20
73
Linus Torvalds1da177e2005-04-16 15:20:36 -070074/* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
75enum sgp_type {
76 SGP_QUICK, /* don't try more than file page cache lookup */
77 SGP_READ, /* don't exceed i_size, don't allocate page */
78 SGP_CACHE, /* don't exceed i_size, may allocate page */
79 SGP_WRITE, /* may exceed i_size, may allocate page */
80};
81
82static int shmem_getpage(struct inode *inode, unsigned long idx,
83 struct page **pagep, enum sgp_type sgp, int *type);
84
Al Viro6daa0e22005-10-21 03:18:50 -040085static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -070086{
87 /*
88 * The above definition of ENTRIES_PER_PAGE, and the use of
89 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
90 * might be reconsidered if it ever diverges from PAGE_SIZE.
91 */
92 return alloc_pages(gfp_mask, PAGE_CACHE_SHIFT-PAGE_SHIFT);
93}
94
95static inline void shmem_dir_free(struct page *page)
96{
97 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
98}
99
100static struct page **shmem_dir_map(struct page *page)
101{
102 return (struct page **)kmap_atomic(page, KM_USER0);
103}
104
105static inline void shmem_dir_unmap(struct page **dir)
106{
107 kunmap_atomic(dir, KM_USER0);
108}
109
110static swp_entry_t *shmem_swp_map(struct page *page)
111{
112 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
113}
114
115static inline void shmem_swp_balance_unmap(void)
116{
117 /*
118 * When passing a pointer to an i_direct entry, to code which
119 * also handles indirect entries and so will shmem_swp_unmap,
120 * we must arrange for the preempt count to remain in balance.
121 * What kmap_atomic of a lowmem page does depends on config
122 * and architecture, so pretend to kmap_atomic some lowmem page.
123 */
124 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
125}
126
127static inline void shmem_swp_unmap(swp_entry_t *entry)
128{
129 kunmap_atomic(entry, KM_USER1);
130}
131
132static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
133{
134 return sb->s_fs_info;
135}
136
137/*
138 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
139 * for shared memory and for shared anonymous (/dev/zero) mappings
140 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
141 * consistent with the pre-accounting of private mappings ...
142 */
143static inline int shmem_acct_size(unsigned long flags, loff_t size)
144{
145 return (flags & VM_ACCOUNT)?
146 security_vm_enough_memory(VM_ACCT(size)): 0;
147}
148
149static inline void shmem_unacct_size(unsigned long flags, loff_t size)
150{
151 if (flags & VM_ACCOUNT)
152 vm_unacct_memory(VM_ACCT(size));
153}
154
155/*
156 * ... whereas tmpfs objects are accounted incrementally as
157 * pages are allocated, in order to allow huge sparse files.
158 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
159 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
160 */
161static inline int shmem_acct_block(unsigned long flags)
162{
163 return (flags & VM_ACCOUNT)?
164 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
165}
166
167static inline void shmem_unacct_blocks(unsigned long flags, long pages)
168{
169 if (!(flags & VM_ACCOUNT))
170 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
171}
172
173static struct super_operations shmem_ops;
174static struct address_space_operations shmem_aops;
175static struct file_operations shmem_file_operations;
176static struct inode_operations shmem_inode_operations;
177static struct inode_operations shmem_dir_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178static struct vm_operations_struct shmem_vm_ops;
179
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700180static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 .ra_pages = 0, /* No readahead */
182 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
183 .unplug_io_fn = default_unplug_io_fn,
184};
185
186static LIST_HEAD(shmem_swaplist);
187static DEFINE_SPINLOCK(shmem_swaplist_lock);
188
189static void shmem_free_blocks(struct inode *inode, long pages)
190{
191 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700192 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700193 spin_lock(&sbinfo->stat_lock);
194 sbinfo->free_blocks += pages;
195 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
196 spin_unlock(&sbinfo->stat_lock);
197 }
198}
199
200/*
201 * shmem_recalc_inode - recalculate the size of an inode
202 *
203 * @inode: inode to recalc
204 *
205 * We have to calculate the free blocks since the mm can drop
206 * undirtied hole pages behind our back.
207 *
208 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
209 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
210 *
211 * It has to be called with the spinlock held.
212 */
213static void shmem_recalc_inode(struct inode *inode)
214{
215 struct shmem_inode_info *info = SHMEM_I(inode);
216 long freed;
217
218 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
219 if (freed > 0) {
220 info->alloced -= freed;
221 shmem_unacct_blocks(info->flags, freed);
222 shmem_free_blocks(inode, freed);
223 }
224}
225
226/*
227 * shmem_swp_entry - find the swap vector position in the info structure
228 *
229 * @info: info structure for the inode
230 * @index: index of the page to find
231 * @page: optional page to add to the structure. Has to be preset to
232 * all zeros
233 *
234 * If there is no space allocated yet it will return NULL when
235 * page is NULL, else it will use the page for the needed block,
236 * setting it to NULL on return to indicate that it has been used.
237 *
238 * The swap vector is organized the following way:
239 *
240 * There are SHMEM_NR_DIRECT entries directly stored in the
241 * shmem_inode_info structure. So small files do not need an addional
242 * allocation.
243 *
244 * For pages with index > SHMEM_NR_DIRECT there is the pointer
245 * i_indirect which points to a page which holds in the first half
246 * doubly indirect blocks, in the second half triple indirect blocks:
247 *
248 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
249 * following layout (for SHMEM_NR_DIRECT == 16):
250 *
251 * i_indirect -> dir --> 16-19
252 * | +-> 20-23
253 * |
254 * +-->dir2 --> 24-27
255 * | +-> 28-31
256 * | +-> 32-35
257 * | +-> 36-39
258 * |
259 * +-->dir3 --> 40-43
260 * +-> 44-47
261 * +-> 48-51
262 * +-> 52-55
263 */
264static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
265{
266 unsigned long offset;
267 struct page **dir;
268 struct page *subdir;
269
270 if (index < SHMEM_NR_DIRECT) {
271 shmem_swp_balance_unmap();
272 return info->i_direct+index;
273 }
274 if (!info->i_indirect) {
275 if (page) {
276 info->i_indirect = *page;
277 *page = NULL;
278 }
279 return NULL; /* need another page */
280 }
281
282 index -= SHMEM_NR_DIRECT;
283 offset = index % ENTRIES_PER_PAGE;
284 index /= ENTRIES_PER_PAGE;
285 dir = shmem_dir_map(info->i_indirect);
286
287 if (index >= ENTRIES_PER_PAGE/2) {
288 index -= ENTRIES_PER_PAGE/2;
289 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
290 index %= ENTRIES_PER_PAGE;
291 subdir = *dir;
292 if (!subdir) {
293 if (page) {
294 *dir = *page;
295 *page = NULL;
296 }
297 shmem_dir_unmap(dir);
298 return NULL; /* need another page */
299 }
300 shmem_dir_unmap(dir);
301 dir = shmem_dir_map(subdir);
302 }
303
304 dir += index;
305 subdir = *dir;
306 if (!subdir) {
307 if (!page || !(subdir = *page)) {
308 shmem_dir_unmap(dir);
309 return NULL; /* need a page */
310 }
311 *dir = subdir;
312 *page = NULL;
313 }
314 shmem_dir_unmap(dir);
315 return shmem_swp_map(subdir) + offset;
316}
317
318static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
319{
320 long incdec = value? 1: -1;
321
322 entry->val = value;
323 info->swapped += incdec;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700324 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
325 struct page *page = kmap_atomic_to_page(entry);
326 set_page_private(page, page_private(page) + incdec);
327 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328}
329
330/*
331 * shmem_swp_alloc - get the position of the swap entry for the page.
332 * If it does not exist allocate the entry.
333 *
334 * @info: info structure for the inode
335 * @index: index of the page to find
336 * @sgp: check and recheck i_size? skip allocation?
337 */
338static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
339{
340 struct inode *inode = &info->vfs_inode;
341 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
342 struct page *page = NULL;
343 swp_entry_t *entry;
344
345 if (sgp != SGP_WRITE &&
346 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
347 return ERR_PTR(-EINVAL);
348
349 while (!(entry = shmem_swp_entry(info, index, &page))) {
350 if (sgp == SGP_READ)
351 return shmem_swp_map(ZERO_PAGE(0));
352 /*
353 * Test free_blocks against 1 not 0, since we have 1 data
354 * page (and perhaps indirect index pages) yet to allocate:
355 * a waste to allocate index if we cannot allocate data.
356 */
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700357 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 spin_lock(&sbinfo->stat_lock);
359 if (sbinfo->free_blocks <= 1) {
360 spin_unlock(&sbinfo->stat_lock);
361 return ERR_PTR(-ENOSPC);
362 }
363 sbinfo->free_blocks--;
364 inode->i_blocks += BLOCKS_PER_PAGE;
365 spin_unlock(&sbinfo->stat_lock);
366 }
367
368 spin_unlock(&info->lock);
369 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping) | __GFP_ZERO);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700370 if (page)
371 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700372 spin_lock(&info->lock);
373
374 if (!page) {
375 shmem_free_blocks(inode, 1);
376 return ERR_PTR(-ENOMEM);
377 }
378 if (sgp != SGP_WRITE &&
379 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
380 entry = ERR_PTR(-EINVAL);
381 break;
382 }
383 if (info->next_index <= index)
384 info->next_index = index + 1;
385 }
386 if (page) {
387 /* another task gave its page, or truncated the file */
388 shmem_free_blocks(inode, 1);
389 shmem_dir_free(page);
390 }
391 if (info->next_index <= index && !IS_ERR(entry))
392 info->next_index = index + 1;
393 return entry;
394}
395
396/*
397 * shmem_free_swp - free some swap entries in a directory
398 *
399 * @dir: pointer to the directory
400 * @edir: pointer after last entry of the directory
401 */
402static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir)
403{
404 swp_entry_t *ptr;
405 int freed = 0;
406
407 for (ptr = dir; ptr < edir; ptr++) {
408 if (ptr->val) {
409 free_swap_and_cache(*ptr);
410 *ptr = (swp_entry_t){0};
411 freed++;
412 }
413 }
414 return freed;
415}
416
417static int shmem_map_and_free_swp(struct page *subdir,
418 int offset, int limit, struct page ***dir)
419{
420 swp_entry_t *ptr;
421 int freed = 0;
422
423 ptr = shmem_swp_map(subdir);
424 for (; offset < limit; offset += LATENCY_LIMIT) {
425 int size = limit - offset;
426 if (size > LATENCY_LIMIT)
427 size = LATENCY_LIMIT;
428 freed += shmem_free_swp(ptr+offset, ptr+offset+size);
429 if (need_resched()) {
430 shmem_swp_unmap(ptr);
431 if (*dir) {
432 shmem_dir_unmap(*dir);
433 *dir = NULL;
434 }
435 cond_resched();
436 ptr = shmem_swp_map(subdir);
437 }
438 }
439 shmem_swp_unmap(ptr);
440 return freed;
441}
442
443static void shmem_free_pages(struct list_head *next)
444{
445 struct page *page;
446 int freed = 0;
447
448 do {
449 page = container_of(next, struct page, lru);
450 next = next->next;
451 shmem_dir_free(page);
452 freed++;
453 if (freed >= LATENCY_LIMIT) {
454 cond_resched();
455 freed = 0;
456 }
457 } while (next);
458}
459
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800460static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700461{
462 struct shmem_inode_info *info = SHMEM_I(inode);
463 unsigned long idx;
464 unsigned long size;
465 unsigned long limit;
466 unsigned long stage;
467 unsigned long diroff;
468 struct page **dir;
469 struct page *topdir;
470 struct page *middir;
471 struct page *subdir;
472 swp_entry_t *ptr;
473 LIST_HEAD(pages_to_free);
474 long nr_pages_to_free = 0;
475 long nr_swaps_freed = 0;
476 int offset;
477 int freed;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800478 int punch_hole = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479
480 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800481 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482 if (idx >= info->next_index)
483 return;
484
485 spin_lock(&info->lock);
486 info->flags |= SHMEM_TRUNCATE;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800487 if (likely(end == (loff_t) -1)) {
488 limit = info->next_index;
489 info->next_index = idx;
490 } else {
491 limit = (end + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
492 if (limit > info->next_index)
493 limit = info->next_index;
494 punch_hole = 1;
495 }
496
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497 topdir = info->i_indirect;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800498 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 info->i_indirect = NULL;
500 nr_pages_to_free++;
501 list_add(&topdir->lru, &pages_to_free);
502 }
503 spin_unlock(&info->lock);
504
505 if (info->swapped && idx < SHMEM_NR_DIRECT) {
506 ptr = info->i_direct;
507 size = limit;
508 if (size > SHMEM_NR_DIRECT)
509 size = SHMEM_NR_DIRECT;
510 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size);
511 }
512 if (!topdir)
513 goto done2;
514
515 BUG_ON(limit <= SHMEM_NR_DIRECT);
516 limit -= SHMEM_NR_DIRECT;
517 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
518 offset = idx % ENTRIES_PER_PAGE;
519 idx -= offset;
520
521 dir = shmem_dir_map(topdir);
522 stage = ENTRIES_PER_PAGEPAGE/2;
523 if (idx < ENTRIES_PER_PAGEPAGE/2) {
524 middir = topdir;
525 diroff = idx/ENTRIES_PER_PAGE;
526 } else {
527 dir += ENTRIES_PER_PAGE/2;
528 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
529 while (stage <= idx)
530 stage += ENTRIES_PER_PAGEPAGE;
531 middir = *dir;
532 if (*dir) {
533 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
534 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
535 if (!diroff && !offset) {
536 *dir = NULL;
537 nr_pages_to_free++;
538 list_add(&middir->lru, &pages_to_free);
539 }
540 shmem_dir_unmap(dir);
541 dir = shmem_dir_map(middir);
542 } else {
543 diroff = 0;
544 offset = 0;
545 idx = stage;
546 }
547 }
548
549 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
550 if (unlikely(idx == stage)) {
551 shmem_dir_unmap(dir);
552 dir = shmem_dir_map(topdir) +
553 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
554 while (!*dir) {
555 dir++;
556 idx += ENTRIES_PER_PAGEPAGE;
557 if (idx >= limit)
558 goto done1;
559 }
560 stage = idx + ENTRIES_PER_PAGEPAGE;
561 middir = *dir;
562 *dir = NULL;
563 nr_pages_to_free++;
564 list_add(&middir->lru, &pages_to_free);
565 shmem_dir_unmap(dir);
566 cond_resched();
567 dir = shmem_dir_map(middir);
568 diroff = 0;
569 }
570 subdir = dir[diroff];
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700571 if (subdir && page_private(subdir)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 size = limit - idx;
573 if (size > ENTRIES_PER_PAGE)
574 size = ENTRIES_PER_PAGE;
575 freed = shmem_map_and_free_swp(subdir,
576 offset, size, &dir);
577 if (!dir)
578 dir = shmem_dir_map(middir);
579 nr_swaps_freed += freed;
580 if (offset)
581 spin_lock(&info->lock);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700582 set_page_private(subdir, page_private(subdir) - freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 if (offset)
584 spin_unlock(&info->lock);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800585 if (!punch_hole)
586 BUG_ON(page_private(subdir) > offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700587 }
588 if (offset)
589 offset = 0;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800590 else if (subdir && !page_private(subdir)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 dir[diroff] = NULL;
592 nr_pages_to_free++;
593 list_add(&subdir->lru, &pages_to_free);
594 }
595 }
596done1:
597 shmem_dir_unmap(dir);
598done2:
599 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
600 /*
601 * Call truncate_inode_pages again: racing shmem_unuse_inode
602 * may have swizzled a page in from swap since vmtruncate or
603 * generic_delete_inode did it, before we lowered next_index.
604 * Also, though shmem_getpage checks i_size before adding to
605 * cache, no recheck after: so fix the narrow window there too.
606 */
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800607 truncate_inode_pages_range(inode->i_mapping, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 }
609
610 spin_lock(&info->lock);
611 info->flags &= ~SHMEM_TRUNCATE;
612 info->swapped -= nr_swaps_freed;
613 if (nr_pages_to_free)
614 shmem_free_blocks(inode, nr_pages_to_free);
615 shmem_recalc_inode(inode);
616 spin_unlock(&info->lock);
617
618 /*
619 * Empty swap vector directory pages to be freed?
620 */
621 if (!list_empty(&pages_to_free)) {
622 pages_to_free.prev->next = NULL;
623 shmem_free_pages(pages_to_free.next);
624 }
625}
626
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800627static void shmem_truncate(struct inode *inode)
628{
629 shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
630}
631
Linus Torvalds1da177e2005-04-16 15:20:36 -0700632static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
633{
634 struct inode *inode = dentry->d_inode;
635 struct page *page = NULL;
636 int error;
637
638 if (attr->ia_valid & ATTR_SIZE) {
639 if (attr->ia_size < inode->i_size) {
640 /*
641 * If truncating down to a partial page, then
642 * if that page is already allocated, hold it
643 * in memory until the truncation is over, so
644 * truncate_partial_page cannnot miss it were
645 * it assigned to swap.
646 */
647 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
648 (void) shmem_getpage(inode,
649 attr->ia_size>>PAGE_CACHE_SHIFT,
650 &page, SGP_READ, NULL);
651 }
652 /*
653 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
654 * detect if any pages might have been added to cache
655 * after truncate_inode_pages. But we needn't bother
656 * if it's being fully truncated to zero-length: the
657 * nrpages check is efficient enough in that case.
658 */
659 if (attr->ia_size) {
660 struct shmem_inode_info *info = SHMEM_I(inode);
661 spin_lock(&info->lock);
662 info->flags &= ~SHMEM_PAGEIN;
663 spin_unlock(&info->lock);
664 }
665 }
666 }
667
668 error = inode_change_ok(inode, attr);
669 if (!error)
670 error = inode_setattr(inode, attr);
671 if (page)
672 page_cache_release(page);
673 return error;
674}
675
676static void shmem_delete_inode(struct inode *inode)
677{
678 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
679 struct shmem_inode_info *info = SHMEM_I(inode);
680
681 if (inode->i_op->truncate == shmem_truncate) {
Mark Fashehfef26652005-09-09 13:01:31 -0700682 truncate_inode_pages(inode->i_mapping, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700683 shmem_unacct_size(info->flags, inode->i_size);
684 inode->i_size = 0;
685 shmem_truncate(inode);
686 if (!list_empty(&info->swaplist)) {
687 spin_lock(&shmem_swaplist_lock);
688 list_del_init(&info->swaplist);
689 spin_unlock(&shmem_swaplist_lock);
690 }
691 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700692 BUG_ON(inode->i_blocks);
693 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 spin_lock(&sbinfo->stat_lock);
695 sbinfo->free_inodes++;
696 spin_unlock(&sbinfo->stat_lock);
697 }
698 clear_inode(inode);
699}
700
701static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
702{
703 swp_entry_t *ptr;
704
705 for (ptr = dir; ptr < edir; ptr++) {
706 if (ptr->val == entry.val)
707 return ptr - dir;
708 }
709 return -1;
710}
711
712static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
713{
714 struct inode *inode;
715 unsigned long idx;
716 unsigned long size;
717 unsigned long limit;
718 unsigned long stage;
719 struct page **dir;
720 struct page *subdir;
721 swp_entry_t *ptr;
722 int offset;
723
724 idx = 0;
725 ptr = info->i_direct;
726 spin_lock(&info->lock);
727 limit = info->next_index;
728 size = limit;
729 if (size > SHMEM_NR_DIRECT)
730 size = SHMEM_NR_DIRECT;
731 offset = shmem_find_swp(entry, ptr, ptr+size);
732 if (offset >= 0) {
733 shmem_swp_balance_unmap();
734 goto found;
735 }
736 if (!info->i_indirect)
737 goto lost2;
738
739 dir = shmem_dir_map(info->i_indirect);
740 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
741
742 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
743 if (unlikely(idx == stage)) {
744 shmem_dir_unmap(dir-1);
745 dir = shmem_dir_map(info->i_indirect) +
746 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
747 while (!*dir) {
748 dir++;
749 idx += ENTRIES_PER_PAGEPAGE;
750 if (idx >= limit)
751 goto lost1;
752 }
753 stage = idx + ENTRIES_PER_PAGEPAGE;
754 subdir = *dir;
755 shmem_dir_unmap(dir);
756 dir = shmem_dir_map(subdir);
757 }
758 subdir = *dir;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700759 if (subdir && page_private(subdir)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 ptr = shmem_swp_map(subdir);
761 size = limit - idx;
762 if (size > ENTRIES_PER_PAGE)
763 size = ENTRIES_PER_PAGE;
764 offset = shmem_find_swp(entry, ptr, ptr+size);
765 if (offset >= 0) {
766 shmem_dir_unmap(dir);
767 goto found;
768 }
769 shmem_swp_unmap(ptr);
770 }
771 }
772lost1:
773 shmem_dir_unmap(dir-1);
774lost2:
775 spin_unlock(&info->lock);
776 return 0;
777found:
778 idx += offset;
779 inode = &info->vfs_inode;
780 if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
781 info->flags |= SHMEM_PAGEIN;
782 shmem_swp_set(info, ptr + offset, 0);
783 }
784 shmem_swp_unmap(ptr);
785 spin_unlock(&info->lock);
786 /*
787 * Decrement swap count even when the entry is left behind:
788 * try_to_unuse will skip over mms, then reincrement count.
789 */
790 swap_free(entry);
791 return 1;
792}
793
794/*
795 * shmem_unuse() search for an eventually swapped out shmem page.
796 */
797int shmem_unuse(swp_entry_t entry, struct page *page)
798{
799 struct list_head *p, *next;
800 struct shmem_inode_info *info;
801 int found = 0;
802
803 spin_lock(&shmem_swaplist_lock);
804 list_for_each_safe(p, next, &shmem_swaplist) {
805 info = list_entry(p, struct shmem_inode_info, swaplist);
806 if (!info->swapped)
807 list_del_init(&info->swaplist);
808 else if (shmem_unuse_inode(info, entry, page)) {
809 /* move head to start search for next from here */
810 list_move_tail(&shmem_swaplist, &info->swaplist);
811 found = 1;
812 break;
813 }
814 }
815 spin_unlock(&shmem_swaplist_lock);
816 return found;
817}
818
819/*
820 * Move the page from the page cache to the swap cache.
821 */
822static int shmem_writepage(struct page *page, struct writeback_control *wbc)
823{
824 struct shmem_inode_info *info;
825 swp_entry_t *entry, swap;
826 struct address_space *mapping;
827 unsigned long index;
828 struct inode *inode;
829
830 BUG_ON(!PageLocked(page));
831 BUG_ON(page_mapped(page));
832
833 mapping = page->mapping;
834 index = page->index;
835 inode = mapping->host;
836 info = SHMEM_I(inode);
837 if (info->flags & VM_LOCKED)
838 goto redirty;
839 swap = get_swap_page();
840 if (!swap.val)
841 goto redirty;
842
843 spin_lock(&info->lock);
844 shmem_recalc_inode(inode);
845 if (index >= info->next_index) {
846 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
847 goto unlock;
848 }
849 entry = shmem_swp_entry(info, index, NULL);
850 BUG_ON(!entry);
851 BUG_ON(entry->val);
852
853 if (move_to_swap_cache(page, swap) == 0) {
854 shmem_swp_set(info, entry, swap.val);
855 shmem_swp_unmap(entry);
856 spin_unlock(&info->lock);
857 if (list_empty(&info->swaplist)) {
858 spin_lock(&shmem_swaplist_lock);
859 /* move instead of add in case we're racing */
860 list_move_tail(&info->swaplist, &shmem_swaplist);
861 spin_unlock(&shmem_swaplist_lock);
862 }
863 unlock_page(page);
864 return 0;
865 }
866
867 shmem_swp_unmap(entry);
868unlock:
869 spin_unlock(&info->lock);
870 swap_free(swap);
871redirty:
872 set_page_dirty(page);
Zach Brown994fc28c2005-12-15 14:28:17 -0800873 return AOP_WRITEPAGE_ACTIVATE; /* Return with the page locked */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700874}
875
876#ifdef CONFIG_NUMA
877static struct page *shmem_swapin_async(struct shared_policy *p,
878 swp_entry_t entry, unsigned long idx)
879{
880 struct page *page;
881 struct vm_area_struct pvma;
882
883 /* Create a pseudo vma that just contains the policy */
884 memset(&pvma, 0, sizeof(struct vm_area_struct));
885 pvma.vm_end = PAGE_SIZE;
886 pvma.vm_pgoff = idx;
887 pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
888 page = read_swap_cache_async(entry, &pvma, 0);
889 mpol_free(pvma.vm_policy);
890 return page;
891}
892
893struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry,
894 unsigned long idx)
895{
896 struct shared_policy *p = &info->policy;
897 int i, num;
898 struct page *page;
899 unsigned long offset;
900
901 num = valid_swaphandles(entry, &offset);
902 for (i = 0; i < num; offset++, i++) {
903 page = shmem_swapin_async(p,
904 swp_entry(swp_type(entry), offset), idx);
905 if (!page)
906 break;
907 page_cache_release(page);
908 }
909 lru_add_drain(); /* Push any new pages onto the LRU now */
910 return shmem_swapin_async(p, entry, idx);
911}
912
913static struct page *
Al Viro6daa0e22005-10-21 03:18:50 -0400914shmem_alloc_page(gfp_t gfp, struct shmem_inode_info *info,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700915 unsigned long idx)
916{
917 struct vm_area_struct pvma;
918 struct page *page;
919
920 memset(&pvma, 0, sizeof(struct vm_area_struct));
921 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
922 pvma.vm_pgoff = idx;
923 pvma.vm_end = PAGE_SIZE;
924 page = alloc_page_vma(gfp | __GFP_ZERO, &pvma, 0);
925 mpol_free(pvma.vm_policy);
926 return page;
927}
928#else
929static inline struct page *
930shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
931{
932 swapin_readahead(entry, 0, NULL);
933 return read_swap_cache_async(entry, NULL, 0);
934}
935
936static inline struct page *
Al Virodd0fc662005-10-07 07:46:04 +0100937shmem_alloc_page(gfp_t gfp,struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700938{
939 return alloc_page(gfp | __GFP_ZERO);
940}
941#endif
942
943/*
944 * shmem_getpage - either get the page from swap or allocate a new one
945 *
946 * If we allocate a new one we do not mark it dirty. That's up to the
947 * vm. If we swap it in we mark it dirty since we also free the swap
948 * entry since a page cannot live in both the swap and page cache
949 */
950static int shmem_getpage(struct inode *inode, unsigned long idx,
951 struct page **pagep, enum sgp_type sgp, int *type)
952{
953 struct address_space *mapping = inode->i_mapping;
954 struct shmem_inode_info *info = SHMEM_I(inode);
955 struct shmem_sb_info *sbinfo;
956 struct page *filepage = *pagep;
957 struct page *swappage;
958 swp_entry_t *entry;
959 swp_entry_t swap;
960 int error;
961
962 if (idx >= SHMEM_MAX_INDEX)
963 return -EFBIG;
964 /*
965 * Normally, filepage is NULL on entry, and either found
966 * uptodate immediately, or allocated and zeroed, or read
967 * in under swappage, which is then assigned to filepage.
968 * But shmem_prepare_write passes in a locked filepage,
969 * which may be found not uptodate by other callers too,
970 * and may need to be copied from the swappage read in.
971 */
972repeat:
973 if (!filepage)
974 filepage = find_lock_page(mapping, idx);
975 if (filepage && PageUptodate(filepage))
976 goto done;
977 error = 0;
978 if (sgp == SGP_QUICK)
979 goto failed;
980
981 spin_lock(&info->lock);
982 shmem_recalc_inode(inode);
983 entry = shmem_swp_alloc(info, idx, sgp);
984 if (IS_ERR(entry)) {
985 spin_unlock(&info->lock);
986 error = PTR_ERR(entry);
987 goto failed;
988 }
989 swap = *entry;
990
991 if (swap.val) {
992 /* Look it up and read it in.. */
993 swappage = lookup_swap_cache(swap);
994 if (!swappage) {
995 shmem_swp_unmap(entry);
996 spin_unlock(&info->lock);
997 /* here we actually do the io */
998 if (type && *type == VM_FAULT_MINOR) {
999 inc_page_state(pgmajfault);
1000 *type = VM_FAULT_MAJOR;
1001 }
1002 swappage = shmem_swapin(info, swap, idx);
1003 if (!swappage) {
1004 spin_lock(&info->lock);
1005 entry = shmem_swp_alloc(info, idx, sgp);
1006 if (IS_ERR(entry))
1007 error = PTR_ERR(entry);
1008 else {
1009 if (entry->val == swap.val)
1010 error = -ENOMEM;
1011 shmem_swp_unmap(entry);
1012 }
1013 spin_unlock(&info->lock);
1014 if (error)
1015 goto failed;
1016 goto repeat;
1017 }
1018 wait_on_page_locked(swappage);
1019 page_cache_release(swappage);
1020 goto repeat;
1021 }
1022
1023 /* We have to do this with page locked to prevent races */
1024 if (TestSetPageLocked(swappage)) {
1025 shmem_swp_unmap(entry);
1026 spin_unlock(&info->lock);
1027 wait_on_page_locked(swappage);
1028 page_cache_release(swappage);
1029 goto repeat;
1030 }
1031 if (PageWriteback(swappage)) {
1032 shmem_swp_unmap(entry);
1033 spin_unlock(&info->lock);
1034 wait_on_page_writeback(swappage);
1035 unlock_page(swappage);
1036 page_cache_release(swappage);
1037 goto repeat;
1038 }
1039 if (!PageUptodate(swappage)) {
1040 shmem_swp_unmap(entry);
1041 spin_unlock(&info->lock);
1042 unlock_page(swappage);
1043 page_cache_release(swappage);
1044 error = -EIO;
1045 goto failed;
1046 }
1047
1048 if (filepage) {
1049 shmem_swp_set(info, entry, 0);
1050 shmem_swp_unmap(entry);
1051 delete_from_swap_cache(swappage);
1052 spin_unlock(&info->lock);
1053 copy_highpage(filepage, swappage);
1054 unlock_page(swappage);
1055 page_cache_release(swappage);
1056 flush_dcache_page(filepage);
1057 SetPageUptodate(filepage);
1058 set_page_dirty(filepage);
1059 swap_free(swap);
1060 } else if (!(error = move_from_swap_cache(
1061 swappage, idx, mapping))) {
1062 info->flags |= SHMEM_PAGEIN;
1063 shmem_swp_set(info, entry, 0);
1064 shmem_swp_unmap(entry);
1065 spin_unlock(&info->lock);
1066 filepage = swappage;
1067 swap_free(swap);
1068 } else {
1069 shmem_swp_unmap(entry);
1070 spin_unlock(&info->lock);
1071 unlock_page(swappage);
1072 page_cache_release(swappage);
1073 if (error == -ENOMEM) {
1074 /* let kswapd refresh zone for GFP_ATOMICs */
1075 blk_congestion_wait(WRITE, HZ/50);
1076 }
1077 goto repeat;
1078 }
1079 } else if (sgp == SGP_READ && !filepage) {
1080 shmem_swp_unmap(entry);
1081 filepage = find_get_page(mapping, idx);
1082 if (filepage &&
1083 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1084 spin_unlock(&info->lock);
1085 wait_on_page_locked(filepage);
1086 page_cache_release(filepage);
1087 filepage = NULL;
1088 goto repeat;
1089 }
1090 spin_unlock(&info->lock);
1091 } else {
1092 shmem_swp_unmap(entry);
1093 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001094 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001095 spin_lock(&sbinfo->stat_lock);
1096 if (sbinfo->free_blocks == 0 ||
1097 shmem_acct_block(info->flags)) {
1098 spin_unlock(&sbinfo->stat_lock);
1099 spin_unlock(&info->lock);
1100 error = -ENOSPC;
1101 goto failed;
1102 }
1103 sbinfo->free_blocks--;
1104 inode->i_blocks += BLOCKS_PER_PAGE;
1105 spin_unlock(&sbinfo->stat_lock);
1106 } else if (shmem_acct_block(info->flags)) {
1107 spin_unlock(&info->lock);
1108 error = -ENOSPC;
1109 goto failed;
1110 }
1111
1112 if (!filepage) {
1113 spin_unlock(&info->lock);
1114 filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
1115 info,
1116 idx);
1117 if (!filepage) {
1118 shmem_unacct_blocks(info->flags, 1);
1119 shmem_free_blocks(inode, 1);
1120 error = -ENOMEM;
1121 goto failed;
1122 }
1123
1124 spin_lock(&info->lock);
1125 entry = shmem_swp_alloc(info, idx, sgp);
1126 if (IS_ERR(entry))
1127 error = PTR_ERR(entry);
1128 else {
1129 swap = *entry;
1130 shmem_swp_unmap(entry);
1131 }
1132 if (error || swap.val || 0 != add_to_page_cache_lru(
1133 filepage, mapping, idx, GFP_ATOMIC)) {
1134 spin_unlock(&info->lock);
1135 page_cache_release(filepage);
1136 shmem_unacct_blocks(info->flags, 1);
1137 shmem_free_blocks(inode, 1);
1138 filepage = NULL;
1139 if (error)
1140 goto failed;
1141 goto repeat;
1142 }
1143 info->flags |= SHMEM_PAGEIN;
1144 }
1145
1146 info->alloced++;
1147 spin_unlock(&info->lock);
1148 flush_dcache_page(filepage);
1149 SetPageUptodate(filepage);
1150 }
1151done:
1152 if (*pagep != filepage) {
1153 unlock_page(filepage);
1154 *pagep = filepage;
1155 }
1156 return 0;
1157
1158failed:
1159 if (*pagep != filepage) {
1160 unlock_page(filepage);
1161 page_cache_release(filepage);
1162 }
1163 return error;
1164}
1165
1166struct page *shmem_nopage(struct vm_area_struct *vma, unsigned long address, int *type)
1167{
1168 struct inode *inode = vma->vm_file->f_dentry->d_inode;
1169 struct page *page = NULL;
1170 unsigned long idx;
1171 int error;
1172
1173 idx = (address - vma->vm_start) >> PAGE_SHIFT;
1174 idx += vma->vm_pgoff;
1175 idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT;
1176 if (((loff_t) idx << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1177 return NOPAGE_SIGBUS;
1178
1179 error = shmem_getpage(inode, idx, &page, SGP_CACHE, type);
1180 if (error)
1181 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS;
1182
1183 mark_page_accessed(page);
1184 return page;
1185}
1186
1187static int shmem_populate(struct vm_area_struct *vma,
1188 unsigned long addr, unsigned long len,
1189 pgprot_t prot, unsigned long pgoff, int nonblock)
1190{
1191 struct inode *inode = vma->vm_file->f_dentry->d_inode;
1192 struct mm_struct *mm = vma->vm_mm;
1193 enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE;
1194 unsigned long size;
1195
1196 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1197 if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size)
1198 return -EINVAL;
1199
1200 while ((long) len > 0) {
1201 struct page *page = NULL;
1202 int err;
1203 /*
1204 * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE
1205 */
1206 err = shmem_getpage(inode, pgoff, &page, sgp, NULL);
1207 if (err)
1208 return err;
Paolo 'Blaisorblade' Giarrussod44ed4f2005-09-03 15:54:55 -07001209 /* Page may still be null, but only if nonblock was set. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 if (page) {
1211 mark_page_accessed(page);
1212 err = install_page(mm, vma, addr, page, prot);
1213 if (err) {
1214 page_cache_release(page);
1215 return err;
1216 }
Hugh Dickins65500d22005-10-29 18:15:59 -07001217 } else if (vma->vm_flags & VM_NONLINEAR) {
Paolo 'Blaisorblade' Giarrussod44ed4f2005-09-03 15:54:55 -07001218 /* No page was found just because we can't read it in
1219 * now (being here implies nonblock != 0), but the page
1220 * may exist, so set the PTE to fault it in later. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001221 err = install_file_pte(mm, vma, addr, pgoff, prot);
1222 if (err)
1223 return err;
1224 }
1225
1226 len -= PAGE_SIZE;
1227 addr += PAGE_SIZE;
1228 pgoff++;
1229 }
1230 return 0;
1231}
1232
1233#ifdef CONFIG_NUMA
1234int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1235{
1236 struct inode *i = vma->vm_file->f_dentry->d_inode;
1237 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1238}
1239
1240struct mempolicy *
1241shmem_get_policy(struct vm_area_struct *vma, unsigned long addr)
1242{
1243 struct inode *i = vma->vm_file->f_dentry->d_inode;
1244 unsigned long idx;
1245
1246 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1247 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1248}
1249#endif
1250
1251int shmem_lock(struct file *file, int lock, struct user_struct *user)
1252{
1253 struct inode *inode = file->f_dentry->d_inode;
1254 struct shmem_inode_info *info = SHMEM_I(inode);
1255 int retval = -ENOMEM;
1256
1257 spin_lock(&info->lock);
1258 if (lock && !(info->flags & VM_LOCKED)) {
1259 if (!user_shm_lock(inode->i_size, user))
1260 goto out_nomem;
1261 info->flags |= VM_LOCKED;
1262 }
1263 if (!lock && (info->flags & VM_LOCKED) && user) {
1264 user_shm_unlock(inode->i_size, user);
1265 info->flags &= ~VM_LOCKED;
1266 }
1267 retval = 0;
1268out_nomem:
1269 spin_unlock(&info->lock);
1270 return retval;
1271}
1272
David Howellsb0e15192006-01-06 00:11:42 -08001273int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274{
1275 file_accessed(file);
1276 vma->vm_ops = &shmem_vm_ops;
1277 return 0;
1278}
1279
1280static struct inode *
1281shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1282{
1283 struct inode *inode;
1284 struct shmem_inode_info *info;
1285 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1286
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001287 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 spin_lock(&sbinfo->stat_lock);
1289 if (!sbinfo->free_inodes) {
1290 spin_unlock(&sbinfo->stat_lock);
1291 return NULL;
1292 }
1293 sbinfo->free_inodes--;
1294 spin_unlock(&sbinfo->stat_lock);
1295 }
1296
1297 inode = new_inode(sb);
1298 if (inode) {
1299 inode->i_mode = mode;
1300 inode->i_uid = current->fsuid;
1301 inode->i_gid = current->fsgid;
1302 inode->i_blksize = PAGE_CACHE_SIZE;
1303 inode->i_blocks = 0;
1304 inode->i_mapping->a_ops = &shmem_aops;
1305 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1306 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1307 info = SHMEM_I(inode);
1308 memset(info, 0, (char *)inode - (char *)info);
1309 spin_lock_init(&info->lock);
1310 INIT_LIST_HEAD(&info->swaplist);
1311
1312 switch (mode & S_IFMT) {
1313 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314 init_special_inode(inode, mode, dev);
1315 break;
1316 case S_IFREG:
1317 inode->i_op = &shmem_inode_operations;
1318 inode->i_fop = &shmem_file_operations;
Robin Holt7339ff82006-01-14 13:20:48 -08001319 mpol_shared_policy_init(&info->policy, sbinfo->policy,
1320 &sbinfo->policy_nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 break;
1322 case S_IFDIR:
1323 inode->i_nlink++;
1324 /* Some things misbehave if size == 0 on a directory */
1325 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1326 inode->i_op = &shmem_dir_inode_operations;
1327 inode->i_fop = &simple_dir_operations;
1328 break;
1329 case S_IFLNK:
1330 /*
1331 * Must not load anything in the rbtree,
1332 * mpol_free_shared_policy will not be called.
1333 */
Robin Holt7339ff82006-01-14 13:20:48 -08001334 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT,
1335 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 break;
1337 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001338 } else if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 spin_lock(&sbinfo->stat_lock);
1340 sbinfo->free_inodes++;
1341 spin_unlock(&sbinfo->stat_lock);
1342 }
1343 return inode;
1344}
1345
1346#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347static struct inode_operations shmem_symlink_inode_operations;
1348static struct inode_operations shmem_symlink_inline_operations;
1349
1350/*
1351 * Normally tmpfs makes no use of shmem_prepare_write, but it
1352 * lets a tmpfs file be used read-write below the loop driver.
1353 */
1354static int
1355shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1356{
1357 struct inode *inode = page->mapping->host;
1358 return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL);
1359}
1360
1361static ssize_t
1362shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1363{
1364 struct inode *inode = file->f_dentry->d_inode;
1365 loff_t pos;
1366 unsigned long written;
1367 ssize_t err;
1368
1369 if ((ssize_t) count < 0)
1370 return -EINVAL;
1371
1372 if (!access_ok(VERIFY_READ, buf, count))
1373 return -EFAULT;
1374
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001375 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001376
1377 pos = *ppos;
1378 written = 0;
1379
1380 err = generic_write_checks(file, &pos, &count, 0);
1381 if (err || !count)
1382 goto out;
1383
1384 err = remove_suid(file->f_dentry);
1385 if (err)
1386 goto out;
1387
1388 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1389
1390 do {
1391 struct page *page = NULL;
1392 unsigned long bytes, index, offset;
1393 char *kaddr;
1394 int left;
1395
1396 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1397 index = pos >> PAGE_CACHE_SHIFT;
1398 bytes = PAGE_CACHE_SIZE - offset;
1399 if (bytes > count)
1400 bytes = count;
1401
1402 /*
1403 * We don't hold page lock across copy from user -
1404 * what would it guard against? - so no deadlock here.
1405 * But it still may be a good idea to prefault below.
1406 */
1407
1408 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1409 if (err)
1410 break;
1411
1412 left = bytes;
1413 if (PageHighMem(page)) {
1414 volatile unsigned char dummy;
1415 __get_user(dummy, buf);
1416 __get_user(dummy, buf + bytes - 1);
1417
1418 kaddr = kmap_atomic(page, KM_USER0);
1419 left = __copy_from_user_inatomic(kaddr + offset,
1420 buf, bytes);
1421 kunmap_atomic(kaddr, KM_USER0);
1422 }
1423 if (left) {
1424 kaddr = kmap(page);
1425 left = __copy_from_user(kaddr + offset, buf, bytes);
1426 kunmap(page);
1427 }
1428
1429 written += bytes;
1430 count -= bytes;
1431 pos += bytes;
1432 buf += bytes;
1433 if (pos > inode->i_size)
1434 i_size_write(inode, pos);
1435
1436 flush_dcache_page(page);
1437 set_page_dirty(page);
1438 mark_page_accessed(page);
1439 page_cache_release(page);
1440
1441 if (left) {
1442 pos -= left;
1443 written -= left;
1444 err = -EFAULT;
1445 break;
1446 }
1447
1448 /*
1449 * Our dirty pages are not counted in nr_dirty,
1450 * and we do not attempt to balance dirty pages.
1451 */
1452
1453 cond_resched();
1454 } while (count);
1455
1456 *ppos = pos;
1457 if (written)
1458 err = written;
1459out:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001460 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461 return err;
1462}
1463
1464static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1465{
1466 struct inode *inode = filp->f_dentry->d_inode;
1467 struct address_space *mapping = inode->i_mapping;
1468 unsigned long index, offset;
1469
1470 index = *ppos >> PAGE_CACHE_SHIFT;
1471 offset = *ppos & ~PAGE_CACHE_MASK;
1472
1473 for (;;) {
1474 struct page *page = NULL;
1475 unsigned long end_index, nr, ret;
1476 loff_t i_size = i_size_read(inode);
1477
1478 end_index = i_size >> PAGE_CACHE_SHIFT;
1479 if (index > end_index)
1480 break;
1481 if (index == end_index) {
1482 nr = i_size & ~PAGE_CACHE_MASK;
1483 if (nr <= offset)
1484 break;
1485 }
1486
1487 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1488 if (desc->error) {
1489 if (desc->error == -EINVAL)
1490 desc->error = 0;
1491 break;
1492 }
1493
1494 /*
1495 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001496 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 */
1498 nr = PAGE_CACHE_SIZE;
1499 i_size = i_size_read(inode);
1500 end_index = i_size >> PAGE_CACHE_SHIFT;
1501 if (index == end_index) {
1502 nr = i_size & ~PAGE_CACHE_MASK;
1503 if (nr <= offset) {
1504 if (page)
1505 page_cache_release(page);
1506 break;
1507 }
1508 }
1509 nr -= offset;
1510
1511 if (page) {
1512 /*
1513 * If users can be writing to this page using arbitrary
1514 * virtual addresses, take care about potential aliasing
1515 * before reading the page on the kernel side.
1516 */
1517 if (mapping_writably_mapped(mapping))
1518 flush_dcache_page(page);
1519 /*
1520 * Mark the page accessed if we read the beginning.
1521 */
1522 if (!offset)
1523 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001524 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 page = ZERO_PAGE(0);
Nick Pigginb5810032005-10-29 18:16:12 -07001526 page_cache_get(page);
1527 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528
1529 /*
1530 * Ok, we have the page, and it's up-to-date, so
1531 * now we can copy it to user space...
1532 *
1533 * The actor routine returns how many bytes were actually used..
1534 * NOTE! This may not be the same as how much of a user buffer
1535 * we filled up (we may be padding etc), so we can only update
1536 * "pos" here (the actor routine has to update the user buffer
1537 * pointers and the remaining count).
1538 */
1539 ret = actor(desc, page, offset, nr);
1540 offset += ret;
1541 index += offset >> PAGE_CACHE_SHIFT;
1542 offset &= ~PAGE_CACHE_MASK;
1543
1544 page_cache_release(page);
1545 if (ret != nr || !desc->count)
1546 break;
1547
1548 cond_resched();
1549 }
1550
1551 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1552 file_accessed(filp);
1553}
1554
1555static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1556{
1557 read_descriptor_t desc;
1558
1559 if ((ssize_t) count < 0)
1560 return -EINVAL;
1561 if (!access_ok(VERIFY_WRITE, buf, count))
1562 return -EFAULT;
1563 if (!count)
1564 return 0;
1565
1566 desc.written = 0;
1567 desc.count = count;
1568 desc.arg.buf = buf;
1569 desc.error = 0;
1570
1571 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1572 if (desc.written)
1573 return desc.written;
1574 return desc.error;
1575}
1576
1577static ssize_t shmem_file_sendfile(struct file *in_file, loff_t *ppos,
1578 size_t count, read_actor_t actor, void *target)
1579{
1580 read_descriptor_t desc;
1581
1582 if (!count)
1583 return 0;
1584
1585 desc.written = 0;
1586 desc.count = count;
1587 desc.arg.data = target;
1588 desc.error = 0;
1589
1590 do_shmem_file_read(in_file, ppos, &desc, actor);
1591 if (desc.written)
1592 return desc.written;
1593 return desc.error;
1594}
1595
1596static int shmem_statfs(struct super_block *sb, struct kstatfs *buf)
1597{
1598 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1599
1600 buf->f_type = TMPFS_MAGIC;
1601 buf->f_bsize = PAGE_CACHE_SIZE;
1602 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001603 spin_lock(&sbinfo->stat_lock);
1604 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 buf->f_blocks = sbinfo->max_blocks;
1606 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001607 }
1608 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001609 buf->f_files = sbinfo->max_inodes;
1610 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 }
1612 /* else leave those fields 0 like simple_statfs */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001613 spin_unlock(&sbinfo->stat_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 return 0;
1615}
1616
1617/*
1618 * File creation. Allocate an inode, and we're done..
1619 */
1620static int
1621shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1622{
1623 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1624 int error = -ENOSPC;
1625
1626 if (inode) {
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001627 error = security_inode_init_security(inode, dir, NULL, NULL,
1628 NULL);
1629 if (error) {
1630 if (error != -EOPNOTSUPP) {
1631 iput(inode);
1632 return error;
1633 }
1634 error = 0;
1635 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636 if (dir->i_mode & S_ISGID) {
1637 inode->i_gid = dir->i_gid;
1638 if (S_ISDIR(mode))
1639 inode->i_mode |= S_ISGID;
1640 }
1641 dir->i_size += BOGO_DIRENT_SIZE;
1642 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1643 d_instantiate(dentry, inode);
1644 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645 }
1646 return error;
1647}
1648
1649static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1650{
1651 int error;
1652
1653 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1654 return error;
1655 dir->i_nlink++;
1656 return 0;
1657}
1658
1659static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1660 struct nameidata *nd)
1661{
1662 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1663}
1664
1665/*
1666 * Link a file..
1667 */
1668static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1669{
1670 struct inode *inode = old_dentry->d_inode;
1671 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1672
1673 /*
1674 * No ordinary (disk based) filesystem counts links as inodes;
1675 * but each new link needs a new dentry, pinning lowmem, and
1676 * tmpfs dentries cannot be pruned until they are unlinked.
1677 */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001678 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 spin_lock(&sbinfo->stat_lock);
1680 if (!sbinfo->free_inodes) {
1681 spin_unlock(&sbinfo->stat_lock);
1682 return -ENOSPC;
1683 }
1684 sbinfo->free_inodes--;
1685 spin_unlock(&sbinfo->stat_lock);
1686 }
1687
1688 dir->i_size += BOGO_DIRENT_SIZE;
1689 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1690 inode->i_nlink++;
1691 atomic_inc(&inode->i_count); /* New dentry reference */
1692 dget(dentry); /* Extra pinning count for the created dentry */
1693 d_instantiate(dentry, inode);
1694 return 0;
1695}
1696
1697static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1698{
1699 struct inode *inode = dentry->d_inode;
1700
1701 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) {
1702 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001703 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 spin_lock(&sbinfo->stat_lock);
1705 sbinfo->free_inodes++;
1706 spin_unlock(&sbinfo->stat_lock);
1707 }
1708 }
1709
1710 dir->i_size -= BOGO_DIRENT_SIZE;
1711 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1712 inode->i_nlink--;
1713 dput(dentry); /* Undo the count from "create" - this does all the work */
1714 return 0;
1715}
1716
1717static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1718{
1719 if (!simple_empty(dentry))
1720 return -ENOTEMPTY;
1721
1722 dir->i_nlink--;
1723 return shmem_unlink(dir, dentry);
1724}
1725
1726/*
1727 * The VFS layer already does all the dentry stuff for rename,
1728 * we just have to decrement the usage count for the target if
1729 * it exists so that the VFS layer correctly free's it when it
1730 * gets overwritten.
1731 */
1732static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1733{
1734 struct inode *inode = old_dentry->d_inode;
1735 int they_are_dirs = S_ISDIR(inode->i_mode);
1736
1737 if (!simple_empty(new_dentry))
1738 return -ENOTEMPTY;
1739
1740 if (new_dentry->d_inode) {
1741 (void) shmem_unlink(new_dir, new_dentry);
1742 if (they_are_dirs)
1743 old_dir->i_nlink--;
1744 } else if (they_are_dirs) {
1745 old_dir->i_nlink--;
1746 new_dir->i_nlink++;
1747 }
1748
1749 old_dir->i_size -= BOGO_DIRENT_SIZE;
1750 new_dir->i_size += BOGO_DIRENT_SIZE;
1751 old_dir->i_ctime = old_dir->i_mtime =
1752 new_dir->i_ctime = new_dir->i_mtime =
1753 inode->i_ctime = CURRENT_TIME;
1754 return 0;
1755}
1756
1757static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1758{
1759 int error;
1760 int len;
1761 struct inode *inode;
1762 struct page *page = NULL;
1763 char *kaddr;
1764 struct shmem_inode_info *info;
1765
1766 len = strlen(symname) + 1;
1767 if (len > PAGE_CACHE_SIZE)
1768 return -ENAMETOOLONG;
1769
1770 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1771 if (!inode)
1772 return -ENOSPC;
1773
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001774 error = security_inode_init_security(inode, dir, NULL, NULL,
1775 NULL);
1776 if (error) {
1777 if (error != -EOPNOTSUPP) {
1778 iput(inode);
1779 return error;
1780 }
1781 error = 0;
1782 }
1783
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 info = SHMEM_I(inode);
1785 inode->i_size = len-1;
1786 if (len <= (char *)inode - (char *)info) {
1787 /* do it inline */
1788 memcpy(info, symname, len);
1789 inode->i_op = &shmem_symlink_inline_operations;
1790 } else {
1791 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1792 if (error) {
1793 iput(inode);
1794 return error;
1795 }
1796 inode->i_op = &shmem_symlink_inode_operations;
1797 kaddr = kmap_atomic(page, KM_USER0);
1798 memcpy(kaddr, symname, len);
1799 kunmap_atomic(kaddr, KM_USER0);
1800 set_page_dirty(page);
1801 page_cache_release(page);
1802 }
1803 if (dir->i_mode & S_ISGID)
1804 inode->i_gid = dir->i_gid;
1805 dir->i_size += BOGO_DIRENT_SIZE;
1806 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1807 d_instantiate(dentry, inode);
1808 dget(dentry);
1809 return 0;
1810}
1811
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001812static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813{
1814 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001815 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816}
1817
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001818static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819{
1820 struct page *page = NULL;
1821 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1822 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001823 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824}
1825
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001826static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827{
1828 if (!IS_ERR(nd_get_link(nd))) {
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001829 struct page *page = cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830 kunmap(page);
1831 mark_page_accessed(page);
1832 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833 }
1834}
1835
1836static struct inode_operations shmem_symlink_inline_operations = {
1837 .readlink = generic_readlink,
1838 .follow_link = shmem_follow_link_inline,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839};
1840
1841static struct inode_operations shmem_symlink_inode_operations = {
1842 .truncate = shmem_truncate,
1843 .readlink = generic_readlink,
1844 .follow_link = shmem_follow_link,
1845 .put_link = shmem_put_link,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846};
1847
Robin Holt7339ff82006-01-14 13:20:48 -08001848static int shmem_parse_options(char *options, int *mode, uid_t *uid,
1849 gid_t *gid, unsigned long *blocks, unsigned long *inodes,
1850 int *policy, nodemask_t *policy_nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851{
1852 char *this_char, *value, *rest;
1853
1854 while ((this_char = strsep(&options, ",")) != NULL) {
1855 if (!*this_char)
1856 continue;
1857 if ((value = strchr(this_char,'=')) != NULL) {
1858 *value++ = 0;
1859 } else {
1860 printk(KERN_ERR
1861 "tmpfs: No value for mount option '%s'\n",
1862 this_char);
1863 return 1;
1864 }
1865
1866 if (!strcmp(this_char,"size")) {
1867 unsigned long long size;
1868 size = memparse(value,&rest);
1869 if (*rest == '%') {
1870 size <<= PAGE_SHIFT;
1871 size *= totalram_pages;
1872 do_div(size, 100);
1873 rest++;
1874 }
1875 if (*rest)
1876 goto bad_val;
1877 *blocks = size >> PAGE_CACHE_SHIFT;
1878 } else if (!strcmp(this_char,"nr_blocks")) {
1879 *blocks = memparse(value,&rest);
1880 if (*rest)
1881 goto bad_val;
1882 } else if (!strcmp(this_char,"nr_inodes")) {
1883 *inodes = memparse(value,&rest);
1884 if (*rest)
1885 goto bad_val;
1886 } else if (!strcmp(this_char,"mode")) {
1887 if (!mode)
1888 continue;
1889 *mode = simple_strtoul(value,&rest,8);
1890 if (*rest)
1891 goto bad_val;
1892 } else if (!strcmp(this_char,"uid")) {
1893 if (!uid)
1894 continue;
1895 *uid = simple_strtoul(value,&rest,0);
1896 if (*rest)
1897 goto bad_val;
1898 } else if (!strcmp(this_char,"gid")) {
1899 if (!gid)
1900 continue;
1901 *gid = simple_strtoul(value,&rest,0);
1902 if (*rest)
1903 goto bad_val;
Robin Holt7339ff82006-01-14 13:20:48 -08001904 } else if (!strcmp(this_char,"mpol")) {
1905 if (!strcmp(value,"default"))
1906 *policy = MPOL_DEFAULT;
1907 else if (!strcmp(value,"preferred"))
1908 *policy = MPOL_PREFERRED;
1909 else if (!strcmp(value,"bind"))
1910 *policy = MPOL_BIND;
1911 else if (!strcmp(value,"interleave"))
1912 *policy = MPOL_INTERLEAVE;
1913 else
1914 goto bad_val;
1915 } else if (!strcmp(this_char,"mpol_nodelist")) {
1916 nodelist_parse(value, *policy_nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 } else {
1918 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
1919 this_char);
1920 return 1;
1921 }
1922 }
1923 return 0;
1924
1925bad_val:
1926 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
1927 value, this_char);
1928 return 1;
1929
1930}
1931
1932static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
1933{
1934 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001935 unsigned long max_blocks = sbinfo->max_blocks;
1936 unsigned long max_inodes = sbinfo->max_inodes;
Robin Holt7339ff82006-01-14 13:20:48 -08001937 int policy = sbinfo->policy;
1938 nodemask_t policy_nodes = sbinfo->policy_nodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001939 unsigned long blocks;
1940 unsigned long inodes;
1941 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942
Robin Holt7339ff82006-01-14 13:20:48 -08001943 if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks,
1944 &max_inodes, &policy, &policy_nodes))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001945 return error;
1946
1947 spin_lock(&sbinfo->stat_lock);
1948 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
1949 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
1950 if (max_blocks < blocks)
1951 goto out;
1952 if (max_inodes < inodes)
1953 goto out;
1954 /*
1955 * Those tests also disallow limited->unlimited while any are in
1956 * use, so i_blocks will always be zero when max_blocks is zero;
1957 * but we must separately disallow unlimited->limited, because
1958 * in that case we have no record of how much is already in use.
1959 */
1960 if (max_blocks && !sbinfo->max_blocks)
1961 goto out;
1962 if (max_inodes && !sbinfo->max_inodes)
1963 goto out;
1964
1965 error = 0;
1966 sbinfo->max_blocks = max_blocks;
1967 sbinfo->free_blocks = max_blocks - blocks;
1968 sbinfo->max_inodes = max_inodes;
1969 sbinfo->free_inodes = max_inodes - inodes;
Robin Holt7339ff82006-01-14 13:20:48 -08001970 sbinfo->policy = policy;
1971 sbinfo->policy_nodes = policy_nodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001972out:
1973 spin_unlock(&sbinfo->stat_lock);
1974 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975}
1976#endif
1977
1978static void shmem_put_super(struct super_block *sb)
1979{
1980 kfree(sb->s_fs_info);
1981 sb->s_fs_info = NULL;
1982}
1983
Linus Torvalds1da177e2005-04-16 15:20:36 -07001984static int shmem_fill_super(struct super_block *sb,
1985 void *data, int silent)
1986{
1987 struct inode *inode;
1988 struct dentry *root;
1989 int mode = S_IRWXUGO | S_ISVTX;
1990 uid_t uid = current->fsuid;
1991 gid_t gid = current->fsgid;
1992 int err = -ENOMEM;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001993 struct shmem_sb_info *sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994 unsigned long blocks = 0;
1995 unsigned long inodes = 0;
Robin Holt7339ff82006-01-14 13:20:48 -08001996 int policy = MPOL_DEFAULT;
1997 nodemask_t policy_nodes = node_online_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001999#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 /*
2001 * Per default we only allow half of the physical ram per
2002 * tmpfs instance, limiting inodes to one per page of lowmem;
2003 * but the internal instance is left unlimited.
2004 */
2005 if (!(sb->s_flags & MS_NOUSER)) {
2006 blocks = totalram_pages / 2;
2007 inodes = totalram_pages - totalhigh_pages;
2008 if (inodes > blocks)
2009 inodes = blocks;
Robin Holt7339ff82006-01-14 13:20:48 -08002010 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks,
2011 &inodes, &policy, &policy_nodes))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012 return -EINVAL;
2013 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014#else
2015 sb->s_flags |= MS_NOUSER;
2016#endif
2017
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002018 /* Round up to L1_CACHE_BYTES to resist false sharing */
2019 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
2020 L1_CACHE_BYTES), GFP_KERNEL);
2021 if (!sbinfo)
2022 return -ENOMEM;
2023
2024 spin_lock_init(&sbinfo->stat_lock);
2025 sbinfo->max_blocks = blocks;
2026 sbinfo->free_blocks = blocks;
2027 sbinfo->max_inodes = inodes;
2028 sbinfo->free_inodes = inodes;
Robin Holt7339ff82006-01-14 13:20:48 -08002029 sbinfo->policy = policy;
2030 sbinfo->policy_nodes = policy_nodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002031
2032 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 sb->s_maxbytes = SHMEM_MAX_BYTES;
2034 sb->s_blocksize = PAGE_CACHE_SIZE;
2035 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2036 sb->s_magic = TMPFS_MAGIC;
2037 sb->s_op = &shmem_ops;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002038
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
2040 if (!inode)
2041 goto failed;
2042 inode->i_uid = uid;
2043 inode->i_gid = gid;
2044 root = d_alloc_root(inode);
2045 if (!root)
2046 goto failed_iput;
2047 sb->s_root = root;
2048 return 0;
2049
2050failed_iput:
2051 iput(inode);
2052failed:
2053 shmem_put_super(sb);
2054 return err;
2055}
2056
2057static kmem_cache_t *shmem_inode_cachep;
2058
2059static struct inode *shmem_alloc_inode(struct super_block *sb)
2060{
2061 struct shmem_inode_info *p;
2062 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, SLAB_KERNEL);
2063 if (!p)
2064 return NULL;
2065 return &p->vfs_inode;
2066}
2067
2068static void shmem_destroy_inode(struct inode *inode)
2069{
2070 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2071 /* only struct inode is valid if it's an inline symlink */
2072 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2073 }
2074 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2075}
2076
2077static void init_once(void *foo, kmem_cache_t *cachep, unsigned long flags)
2078{
2079 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2080
2081 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2082 SLAB_CTOR_CONSTRUCTOR) {
2083 inode_init_once(&p->vfs_inode);
2084 }
2085}
2086
2087static int init_inodecache(void)
2088{
2089 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2090 sizeof(struct shmem_inode_info),
2091 0, 0, init_once, NULL);
2092 if (shmem_inode_cachep == NULL)
2093 return -ENOMEM;
2094 return 0;
2095}
2096
2097static void destroy_inodecache(void)
2098{
2099 if (kmem_cache_destroy(shmem_inode_cachep))
2100 printk(KERN_INFO "shmem_inode_cache: not all structures were freed\n");
2101}
2102
2103static struct address_space_operations shmem_aops = {
2104 .writepage = shmem_writepage,
2105 .set_page_dirty = __set_page_dirty_nobuffers,
2106#ifdef CONFIG_TMPFS
2107 .prepare_write = shmem_prepare_write,
2108 .commit_write = simple_commit_write,
2109#endif
2110};
2111
2112static struct file_operations shmem_file_operations = {
2113 .mmap = shmem_mmap,
2114#ifdef CONFIG_TMPFS
2115 .llseek = generic_file_llseek,
2116 .read = shmem_file_read,
2117 .write = shmem_file_write,
2118 .fsync = simple_sync_file,
2119 .sendfile = shmem_file_sendfile,
2120#endif
2121};
2122
2123static struct inode_operations shmem_inode_operations = {
2124 .truncate = shmem_truncate,
2125 .setattr = shmem_notify_change,
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002126 .truncate_range = shmem_truncate_range,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127};
2128
2129static struct inode_operations shmem_dir_inode_operations = {
2130#ifdef CONFIG_TMPFS
2131 .create = shmem_create,
2132 .lookup = simple_lookup,
2133 .link = shmem_link,
2134 .unlink = shmem_unlink,
2135 .symlink = shmem_symlink,
2136 .mkdir = shmem_mkdir,
2137 .rmdir = shmem_rmdir,
2138 .mknod = shmem_mknod,
2139 .rename = shmem_rename,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140#endif
2141};
2142
2143static struct super_operations shmem_ops = {
2144 .alloc_inode = shmem_alloc_inode,
2145 .destroy_inode = shmem_destroy_inode,
2146#ifdef CONFIG_TMPFS
2147 .statfs = shmem_statfs,
2148 .remount_fs = shmem_remount_fs,
2149#endif
2150 .delete_inode = shmem_delete_inode,
2151 .drop_inode = generic_delete_inode,
2152 .put_super = shmem_put_super,
2153};
2154
2155static struct vm_operations_struct shmem_vm_ops = {
2156 .nopage = shmem_nopage,
2157 .populate = shmem_populate,
2158#ifdef CONFIG_NUMA
2159 .set_policy = shmem_set_policy,
2160 .get_policy = shmem_get_policy,
2161#endif
2162};
2163
2164
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165static struct super_block *shmem_get_sb(struct file_system_type *fs_type,
2166 int flags, const char *dev_name, void *data)
2167{
2168 return get_sb_nodev(fs_type, flags, data, shmem_fill_super);
2169}
2170
2171static struct file_system_type tmpfs_fs_type = {
2172 .owner = THIS_MODULE,
2173 .name = "tmpfs",
2174 .get_sb = shmem_get_sb,
2175 .kill_sb = kill_litter_super,
2176};
2177static struct vfsmount *shm_mnt;
2178
2179static int __init init_tmpfs(void)
2180{
2181 int error;
2182
2183 error = init_inodecache();
2184 if (error)
2185 goto out3;
2186
2187 error = register_filesystem(&tmpfs_fs_type);
2188 if (error) {
2189 printk(KERN_ERR "Could not register tmpfs\n");
2190 goto out2;
2191 }
2192#ifdef CONFIG_TMPFS
2193 devfs_mk_dir("shm");
2194#endif
2195 shm_mnt = do_kern_mount(tmpfs_fs_type.name, MS_NOUSER,
2196 tmpfs_fs_type.name, NULL);
2197 if (IS_ERR(shm_mnt)) {
2198 error = PTR_ERR(shm_mnt);
2199 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2200 goto out1;
2201 }
2202 return 0;
2203
2204out1:
2205 unregister_filesystem(&tmpfs_fs_type);
2206out2:
2207 destroy_inodecache();
2208out3:
2209 shm_mnt = ERR_PTR(error);
2210 return error;
2211}
2212module_init(init_tmpfs)
2213
2214/*
2215 * shmem_file_setup - get an unlinked file living in tmpfs
2216 *
2217 * @name: name for dentry (to be seen in /proc/<pid>/maps
2218 * @size: size to be set for the file
2219 *
2220 */
2221struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2222{
2223 int error;
2224 struct file *file;
2225 struct inode *inode;
2226 struct dentry *dentry, *root;
2227 struct qstr this;
2228
2229 if (IS_ERR(shm_mnt))
2230 return (void *)shm_mnt;
2231
2232 if (size < 0 || size > SHMEM_MAX_BYTES)
2233 return ERR_PTR(-EINVAL);
2234
2235 if (shmem_acct_size(flags, size))
2236 return ERR_PTR(-ENOMEM);
2237
2238 error = -ENOMEM;
2239 this.name = name;
2240 this.len = strlen(name);
2241 this.hash = 0; /* will go */
2242 root = shm_mnt->mnt_root;
2243 dentry = d_alloc(root, &this);
2244 if (!dentry)
2245 goto put_memory;
2246
2247 error = -ENFILE;
2248 file = get_empty_filp();
2249 if (!file)
2250 goto put_dentry;
2251
2252 error = -ENOSPC;
2253 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2254 if (!inode)
2255 goto close_file;
2256
2257 SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2258 d_instantiate(dentry, inode);
2259 inode->i_size = size;
2260 inode->i_nlink = 0; /* It is unlinked */
2261 file->f_vfsmnt = mntget(shm_mnt);
2262 file->f_dentry = dentry;
2263 file->f_mapping = inode->i_mapping;
2264 file->f_op = &shmem_file_operations;
2265 file->f_mode = FMODE_WRITE | FMODE_READ;
2266 return file;
2267
2268close_file:
2269 put_filp(file);
2270put_dentry:
2271 dput(dentry);
2272put_memory:
2273 shmem_unacct_size(flags, size);
2274 return ERR_PTR(error);
2275}
2276
2277/*
2278 * shmem_zero_setup - setup a shared anonymous mapping
2279 *
2280 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2281 */
2282int shmem_zero_setup(struct vm_area_struct *vma)
2283{
2284 struct file *file;
2285 loff_t size = vma->vm_end - vma->vm_start;
2286
2287 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2288 if (IS_ERR(file))
2289 return PTR_ERR(file);
2290
2291 if (vma->vm_file)
2292 fput(vma->vm_file);
2293 vma->vm_file = file;
2294 vma->vm_ops = &shmem_vm_ops;
2295 return 0;
2296}