blob: db21c51531ca894f647d1aea6b79f72b0d6fc56d [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>
29#include <linux/mm.h>
30#include <linux/mman.h>
31#include <linux/file.h>
32#include <linux/swap.h>
33#include <linux/pagemap.h>
34#include <linux/string.h>
35#include <linux/slab.h>
36#include <linux/backing-dev.h>
37#include <linux/shmem_fs.h>
38#include <linux/mount.h>
39#include <linux/writeback.h>
40#include <linux/vfs.h>
41#include <linux/blkdev.h>
42#include <linux/security.h>
43#include <linux/swapops.h>
44#include <linux/mempolicy.h>
45#include <linux/namei.h>
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +000046#include <linux/ctype.h>
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -070047#include <linux/migrate.h>
48
Linus Torvalds1da177e2005-04-16 15:20:36 -070049#include <asm/uaccess.h>
50#include <asm/div64.h>
51#include <asm/pgtable.h>
52
53/* This magic number is used in glibc for posix shared memory */
54#define TMPFS_MAGIC 0x01021994
55
56#define ENTRIES_PER_PAGE (PAGE_CACHE_SIZE/sizeof(unsigned long))
57#define ENTRIES_PER_PAGEPAGE (ENTRIES_PER_PAGE*ENTRIES_PER_PAGE)
58#define BLOCKS_PER_PAGE (PAGE_CACHE_SIZE/512)
59
60#define SHMEM_MAX_INDEX (SHMEM_NR_DIRECT + (ENTRIES_PER_PAGEPAGE/2) * (ENTRIES_PER_PAGE+1))
61#define SHMEM_MAX_BYTES ((unsigned long long)SHMEM_MAX_INDEX << PAGE_CACHE_SHIFT)
62
63#define VM_ACCT(size) (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
64
65/* info->flags needs VM_flags to handle pagein/truncate races efficiently */
66#define SHMEM_PAGEIN VM_READ
67#define SHMEM_TRUNCATE VM_WRITE
68
69/* Definition to limit shmem_truncate's steps between cond_rescheds */
70#define LATENCY_LIMIT 64
71
72/* Pretend that each entry is of this size in directory's i_size */
73#define BOGO_DIRENT_SIZE 20
74
Linus Torvalds1da177e2005-04-16 15:20:36 -070075/* Flag allocation requirements to shmem_getpage and shmem_swp_alloc */
76enum sgp_type {
77 SGP_QUICK, /* don't try more than file page cache lookup */
78 SGP_READ, /* don't exceed i_size, don't allocate page */
79 SGP_CACHE, /* don't exceed i_size, may allocate page */
80 SGP_WRITE, /* may exceed i_size, may allocate page */
81};
82
83static int shmem_getpage(struct inode *inode, unsigned long idx,
84 struct page **pagep, enum sgp_type sgp, int *type);
85
Al Viro6daa0e22005-10-21 03:18:50 -040086static inline struct page *shmem_dir_alloc(gfp_t gfp_mask)
Linus Torvalds1da177e2005-04-16 15:20:36 -070087{
88 /*
89 * The above definition of ENTRIES_PER_PAGE, and the use of
90 * BLOCKS_PER_PAGE on indirect pages, assume PAGE_CACHE_SIZE:
91 * might be reconsidered if it ever diverges from PAGE_SIZE.
92 */
93 return alloc_pages(gfp_mask, PAGE_CACHE_SHIFT-PAGE_SHIFT);
94}
95
96static inline void shmem_dir_free(struct page *page)
97{
98 __free_pages(page, PAGE_CACHE_SHIFT-PAGE_SHIFT);
99}
100
101static struct page **shmem_dir_map(struct page *page)
102{
103 return (struct page **)kmap_atomic(page, KM_USER0);
104}
105
106static inline void shmem_dir_unmap(struct page **dir)
107{
108 kunmap_atomic(dir, KM_USER0);
109}
110
111static swp_entry_t *shmem_swp_map(struct page *page)
112{
113 return (swp_entry_t *)kmap_atomic(page, KM_USER1);
114}
115
116static inline void shmem_swp_balance_unmap(void)
117{
118 /*
119 * When passing a pointer to an i_direct entry, to code which
120 * also handles indirect entries and so will shmem_swp_unmap,
121 * we must arrange for the preempt count to remain in balance.
122 * What kmap_atomic of a lowmem page does depends on config
123 * and architecture, so pretend to kmap_atomic some lowmem page.
124 */
125 (void) kmap_atomic(ZERO_PAGE(0), KM_USER1);
126}
127
128static inline void shmem_swp_unmap(swp_entry_t *entry)
129{
130 kunmap_atomic(entry, KM_USER1);
131}
132
133static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
134{
135 return sb->s_fs_info;
136}
137
138/*
139 * shmem_file_setup pre-accounts the whole fixed size of a VM object,
140 * for shared memory and for shared anonymous (/dev/zero) mappings
141 * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
142 * consistent with the pre-accounting of private mappings ...
143 */
144static inline int shmem_acct_size(unsigned long flags, loff_t size)
145{
146 return (flags & VM_ACCOUNT)?
147 security_vm_enough_memory(VM_ACCT(size)): 0;
148}
149
150static inline void shmem_unacct_size(unsigned long flags, loff_t size)
151{
152 if (flags & VM_ACCOUNT)
153 vm_unacct_memory(VM_ACCT(size));
154}
155
156/*
157 * ... whereas tmpfs objects are accounted incrementally as
158 * pages are allocated, in order to allow huge sparse files.
159 * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
160 * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
161 */
162static inline int shmem_acct_block(unsigned long flags)
163{
164 return (flags & VM_ACCOUNT)?
165 0: security_vm_enough_memory(VM_ACCT(PAGE_CACHE_SIZE));
166}
167
168static inline void shmem_unacct_blocks(unsigned long flags, long pages)
169{
170 if (!(flags & VM_ACCOUNT))
171 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
172}
173
174static struct super_operations shmem_ops;
Christoph Hellwigf5e54d62006-06-28 04:26:44 -0700175static const struct address_space_operations shmem_aops;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176static struct file_operations shmem_file_operations;
177static struct inode_operations shmem_inode_operations;
178static struct inode_operations shmem_dir_inode_operations;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179static struct vm_operations_struct shmem_vm_ops;
180
Ravikiran G Thirumalai6c231b72005-09-06 15:17:45 -0700181static struct backing_dev_info shmem_backing_dev_info __read_mostly = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 .ra_pages = 0, /* No readahead */
183 .capabilities = BDI_CAP_NO_ACCT_DIRTY | BDI_CAP_NO_WRITEBACK,
184 .unplug_io_fn = default_unplug_io_fn,
185};
186
187static LIST_HEAD(shmem_swaplist);
188static DEFINE_SPINLOCK(shmem_swaplist_lock);
189
190static void shmem_free_blocks(struct inode *inode, long pages)
191{
192 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700193 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194 spin_lock(&sbinfo->stat_lock);
195 sbinfo->free_blocks += pages;
196 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
197 spin_unlock(&sbinfo->stat_lock);
198 }
199}
200
201/*
202 * shmem_recalc_inode - recalculate the size of an inode
203 *
204 * @inode: inode to recalc
205 *
206 * We have to calculate the free blocks since the mm can drop
207 * undirtied hole pages behind our back.
208 *
209 * But normally info->alloced == inode->i_mapping->nrpages + info->swapped
210 * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
211 *
212 * It has to be called with the spinlock held.
213 */
214static void shmem_recalc_inode(struct inode *inode)
215{
216 struct shmem_inode_info *info = SHMEM_I(inode);
217 long freed;
218
219 freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
220 if (freed > 0) {
221 info->alloced -= freed;
222 shmem_unacct_blocks(info->flags, freed);
223 shmem_free_blocks(inode, freed);
224 }
225}
226
227/*
228 * shmem_swp_entry - find the swap vector position in the info structure
229 *
230 * @info: info structure for the inode
231 * @index: index of the page to find
232 * @page: optional page to add to the structure. Has to be preset to
233 * all zeros
234 *
235 * If there is no space allocated yet it will return NULL when
236 * page is NULL, else it will use the page for the needed block,
237 * setting it to NULL on return to indicate that it has been used.
238 *
239 * The swap vector is organized the following way:
240 *
241 * There are SHMEM_NR_DIRECT entries directly stored in the
242 * shmem_inode_info structure. So small files do not need an addional
243 * allocation.
244 *
245 * For pages with index > SHMEM_NR_DIRECT there is the pointer
246 * i_indirect which points to a page which holds in the first half
247 * doubly indirect blocks, in the second half triple indirect blocks:
248 *
249 * For an artificial ENTRIES_PER_PAGE = 4 this would lead to the
250 * following layout (for SHMEM_NR_DIRECT == 16):
251 *
252 * i_indirect -> dir --> 16-19
253 * | +-> 20-23
254 * |
255 * +-->dir2 --> 24-27
256 * | +-> 28-31
257 * | +-> 32-35
258 * | +-> 36-39
259 * |
260 * +-->dir3 --> 40-43
261 * +-> 44-47
262 * +-> 48-51
263 * +-> 52-55
264 */
265static swp_entry_t *shmem_swp_entry(struct shmem_inode_info *info, unsigned long index, struct page **page)
266{
267 unsigned long offset;
268 struct page **dir;
269 struct page *subdir;
270
271 if (index < SHMEM_NR_DIRECT) {
272 shmem_swp_balance_unmap();
273 return info->i_direct+index;
274 }
275 if (!info->i_indirect) {
276 if (page) {
277 info->i_indirect = *page;
278 *page = NULL;
279 }
280 return NULL; /* need another page */
281 }
282
283 index -= SHMEM_NR_DIRECT;
284 offset = index % ENTRIES_PER_PAGE;
285 index /= ENTRIES_PER_PAGE;
286 dir = shmem_dir_map(info->i_indirect);
287
288 if (index >= ENTRIES_PER_PAGE/2) {
289 index -= ENTRIES_PER_PAGE/2;
290 dir += ENTRIES_PER_PAGE/2 + index/ENTRIES_PER_PAGE;
291 index %= ENTRIES_PER_PAGE;
292 subdir = *dir;
293 if (!subdir) {
294 if (page) {
295 *dir = *page;
296 *page = NULL;
297 }
298 shmem_dir_unmap(dir);
299 return NULL; /* need another page */
300 }
301 shmem_dir_unmap(dir);
302 dir = shmem_dir_map(subdir);
303 }
304
305 dir += index;
306 subdir = *dir;
307 if (!subdir) {
308 if (!page || !(subdir = *page)) {
309 shmem_dir_unmap(dir);
310 return NULL; /* need a page */
311 }
312 *dir = subdir;
313 *page = NULL;
314 }
315 shmem_dir_unmap(dir);
316 return shmem_swp_map(subdir) + offset;
317}
318
319static void shmem_swp_set(struct shmem_inode_info *info, swp_entry_t *entry, unsigned long value)
320{
321 long incdec = value? 1: -1;
322
323 entry->val = value;
324 info->swapped += incdec;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700325 if ((unsigned long)(entry - info->i_direct) >= SHMEM_NR_DIRECT) {
326 struct page *page = kmap_atomic_to_page(entry);
327 set_page_private(page, page_private(page) + incdec);
328 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700329}
330
331/*
332 * shmem_swp_alloc - get the position of the swap entry for the page.
333 * If it does not exist allocate the entry.
334 *
335 * @info: info structure for the inode
336 * @index: index of the page to find
337 * @sgp: check and recheck i_size? skip allocation?
338 */
339static swp_entry_t *shmem_swp_alloc(struct shmem_inode_info *info, unsigned long index, enum sgp_type sgp)
340{
341 struct inode *inode = &info->vfs_inode;
342 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
343 struct page *page = NULL;
344 swp_entry_t *entry;
345
346 if (sgp != SGP_WRITE &&
347 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode))
348 return ERR_PTR(-EINVAL);
349
350 while (!(entry = shmem_swp_entry(info, index, &page))) {
351 if (sgp == SGP_READ)
352 return shmem_swp_map(ZERO_PAGE(0));
353 /*
354 * Test free_blocks against 1 not 0, since we have 1 data
355 * page (and perhaps indirect index pages) yet to allocate:
356 * a waste to allocate index if we cannot allocate data.
357 */
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700358 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700359 spin_lock(&sbinfo->stat_lock);
360 if (sbinfo->free_blocks <= 1) {
361 spin_unlock(&sbinfo->stat_lock);
362 return ERR_PTR(-ENOSPC);
363 }
364 sbinfo->free_blocks--;
365 inode->i_blocks += BLOCKS_PER_PAGE;
366 spin_unlock(&sbinfo->stat_lock);
367 }
368
369 spin_unlock(&info->lock);
370 page = shmem_dir_alloc(mapping_gfp_mask(inode->i_mapping) | __GFP_ZERO);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700371 if (page)
372 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700373 spin_lock(&info->lock);
374
375 if (!page) {
376 shmem_free_blocks(inode, 1);
377 return ERR_PTR(-ENOMEM);
378 }
379 if (sgp != SGP_WRITE &&
380 ((loff_t) index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
381 entry = ERR_PTR(-EINVAL);
382 break;
383 }
384 if (info->next_index <= index)
385 info->next_index = index + 1;
386 }
387 if (page) {
388 /* another task gave its page, or truncated the file */
389 shmem_free_blocks(inode, 1);
390 shmem_dir_free(page);
391 }
392 if (info->next_index <= index && !IS_ERR(entry))
393 info->next_index = index + 1;
394 return entry;
395}
396
397/*
398 * shmem_free_swp - free some swap entries in a directory
399 *
400 * @dir: pointer to the directory
401 * @edir: pointer after last entry of the directory
402 */
403static int shmem_free_swp(swp_entry_t *dir, swp_entry_t *edir)
404{
405 swp_entry_t *ptr;
406 int freed = 0;
407
408 for (ptr = dir; ptr < edir; ptr++) {
409 if (ptr->val) {
410 free_swap_and_cache(*ptr);
411 *ptr = (swp_entry_t){0};
412 freed++;
413 }
414 }
415 return freed;
416}
417
418static int shmem_map_and_free_swp(struct page *subdir,
419 int offset, int limit, struct page ***dir)
420{
421 swp_entry_t *ptr;
422 int freed = 0;
423
424 ptr = shmem_swp_map(subdir);
425 for (; offset < limit; offset += LATENCY_LIMIT) {
426 int size = limit - offset;
427 if (size > LATENCY_LIMIT)
428 size = LATENCY_LIMIT;
429 freed += shmem_free_swp(ptr+offset, ptr+offset+size);
430 if (need_resched()) {
431 shmem_swp_unmap(ptr);
432 if (*dir) {
433 shmem_dir_unmap(*dir);
434 *dir = NULL;
435 }
436 cond_resched();
437 ptr = shmem_swp_map(subdir);
438 }
439 }
440 shmem_swp_unmap(ptr);
441 return freed;
442}
443
444static void shmem_free_pages(struct list_head *next)
445{
446 struct page *page;
447 int freed = 0;
448
449 do {
450 page = container_of(next, struct page, lru);
451 next = next->next;
452 shmem_dir_free(page);
453 freed++;
454 if (freed >= LATENCY_LIMIT) {
455 cond_resched();
456 freed = 0;
457 }
458 } while (next);
459}
460
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800461static void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462{
463 struct shmem_inode_info *info = SHMEM_I(inode);
464 unsigned long idx;
465 unsigned long size;
466 unsigned long limit;
467 unsigned long stage;
468 unsigned long diroff;
469 struct page **dir;
470 struct page *topdir;
471 struct page *middir;
472 struct page *subdir;
473 swp_entry_t *ptr;
474 LIST_HEAD(pages_to_free);
475 long nr_pages_to_free = 0;
476 long nr_swaps_freed = 0;
477 int offset;
478 int freed;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800479 int punch_hole = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480
481 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800482 idx = (start + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 if (idx >= info->next_index)
484 return;
485
486 spin_lock(&info->lock);
487 info->flags |= SHMEM_TRUNCATE;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800488 if (likely(end == (loff_t) -1)) {
489 limit = info->next_index;
490 info->next_index = idx;
491 } else {
492 limit = (end + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
493 if (limit > info->next_index)
494 limit = info->next_index;
495 punch_hole = 1;
496 }
497
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 topdir = info->i_indirect;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800499 if (topdir && idx <= SHMEM_NR_DIRECT && !punch_hole) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 info->i_indirect = NULL;
501 nr_pages_to_free++;
502 list_add(&topdir->lru, &pages_to_free);
503 }
504 spin_unlock(&info->lock);
505
506 if (info->swapped && idx < SHMEM_NR_DIRECT) {
507 ptr = info->i_direct;
508 size = limit;
509 if (size > SHMEM_NR_DIRECT)
510 size = SHMEM_NR_DIRECT;
511 nr_swaps_freed = shmem_free_swp(ptr+idx, ptr+size);
512 }
513 if (!topdir)
514 goto done2;
515
516 BUG_ON(limit <= SHMEM_NR_DIRECT);
517 limit -= SHMEM_NR_DIRECT;
518 idx = (idx > SHMEM_NR_DIRECT)? (idx - SHMEM_NR_DIRECT): 0;
519 offset = idx % ENTRIES_PER_PAGE;
520 idx -= offset;
521
522 dir = shmem_dir_map(topdir);
523 stage = ENTRIES_PER_PAGEPAGE/2;
524 if (idx < ENTRIES_PER_PAGEPAGE/2) {
525 middir = topdir;
526 diroff = idx/ENTRIES_PER_PAGE;
527 } else {
528 dir += ENTRIES_PER_PAGE/2;
529 dir += (idx - ENTRIES_PER_PAGEPAGE/2)/ENTRIES_PER_PAGEPAGE;
530 while (stage <= idx)
531 stage += ENTRIES_PER_PAGEPAGE;
532 middir = *dir;
533 if (*dir) {
534 diroff = ((idx - ENTRIES_PER_PAGEPAGE/2) %
535 ENTRIES_PER_PAGEPAGE) / ENTRIES_PER_PAGE;
536 if (!diroff && !offset) {
537 *dir = NULL;
538 nr_pages_to_free++;
539 list_add(&middir->lru, &pages_to_free);
540 }
541 shmem_dir_unmap(dir);
542 dir = shmem_dir_map(middir);
543 } else {
544 diroff = 0;
545 offset = 0;
546 idx = stage;
547 }
548 }
549
550 for (; idx < limit; idx += ENTRIES_PER_PAGE, diroff++) {
551 if (unlikely(idx == stage)) {
552 shmem_dir_unmap(dir);
553 dir = shmem_dir_map(topdir) +
554 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
555 while (!*dir) {
556 dir++;
557 idx += ENTRIES_PER_PAGEPAGE;
558 if (idx >= limit)
559 goto done1;
560 }
561 stage = idx + ENTRIES_PER_PAGEPAGE;
562 middir = *dir;
563 *dir = NULL;
564 nr_pages_to_free++;
565 list_add(&middir->lru, &pages_to_free);
566 shmem_dir_unmap(dir);
567 cond_resched();
568 dir = shmem_dir_map(middir);
569 diroff = 0;
570 }
571 subdir = dir[diroff];
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700572 if (subdir && page_private(subdir)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 size = limit - idx;
574 if (size > ENTRIES_PER_PAGE)
575 size = ENTRIES_PER_PAGE;
576 freed = shmem_map_and_free_swp(subdir,
577 offset, size, &dir);
578 if (!dir)
579 dir = shmem_dir_map(middir);
580 nr_swaps_freed += freed;
581 if (offset)
582 spin_lock(&info->lock);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700583 set_page_private(subdir, page_private(subdir) - freed);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 if (offset)
585 spin_unlock(&info->lock);
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800586 if (!punch_hole)
587 BUG_ON(page_private(subdir) > offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588 }
589 if (offset)
590 offset = 0;
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800591 else if (subdir && !page_private(subdir)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 dir[diroff] = NULL;
593 nr_pages_to_free++;
594 list_add(&subdir->lru, &pages_to_free);
595 }
596 }
597done1:
598 shmem_dir_unmap(dir);
599done2:
600 if (inode->i_mapping->nrpages && (info->flags & SHMEM_PAGEIN)) {
601 /*
602 * Call truncate_inode_pages again: racing shmem_unuse_inode
603 * may have swizzled a page in from swap since vmtruncate or
604 * generic_delete_inode did it, before we lowered next_index.
605 * Also, though shmem_getpage checks i_size before adding to
606 * cache, no recheck after: so fix the narrow window there too.
607 */
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800608 truncate_inode_pages_range(inode->i_mapping, start, end);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 }
610
611 spin_lock(&info->lock);
612 info->flags &= ~SHMEM_TRUNCATE;
613 info->swapped -= nr_swaps_freed;
614 if (nr_pages_to_free)
615 shmem_free_blocks(inode, nr_pages_to_free);
616 shmem_recalc_inode(inode);
617 spin_unlock(&info->lock);
618
619 /*
620 * Empty swap vector directory pages to be freed?
621 */
622 if (!list_empty(&pages_to_free)) {
623 pages_to_free.prev->next = NULL;
624 shmem_free_pages(pages_to_free.next);
625 }
626}
627
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -0800628static void shmem_truncate(struct inode *inode)
629{
630 shmem_truncate_range(inode, inode->i_size, (loff_t)-1);
631}
632
Linus Torvalds1da177e2005-04-16 15:20:36 -0700633static int shmem_notify_change(struct dentry *dentry, struct iattr *attr)
634{
635 struct inode *inode = dentry->d_inode;
636 struct page *page = NULL;
637 int error;
638
639 if (attr->ia_valid & ATTR_SIZE) {
640 if (attr->ia_size < inode->i_size) {
641 /*
642 * If truncating down to a partial page, then
643 * if that page is already allocated, hold it
644 * in memory until the truncation is over, so
645 * truncate_partial_page cannnot miss it were
646 * it assigned to swap.
647 */
648 if (attr->ia_size & (PAGE_CACHE_SIZE-1)) {
649 (void) shmem_getpage(inode,
650 attr->ia_size>>PAGE_CACHE_SHIFT,
651 &page, SGP_READ, NULL);
652 }
653 /*
654 * Reset SHMEM_PAGEIN flag so that shmem_truncate can
655 * detect if any pages might have been added to cache
656 * after truncate_inode_pages. But we needn't bother
657 * if it's being fully truncated to zero-length: the
658 * nrpages check is efficient enough in that case.
659 */
660 if (attr->ia_size) {
661 struct shmem_inode_info *info = SHMEM_I(inode);
662 spin_lock(&info->lock);
663 info->flags &= ~SHMEM_PAGEIN;
664 spin_unlock(&info->lock);
665 }
666 }
667 }
668
669 error = inode_change_ok(inode, attr);
670 if (!error)
671 error = inode_setattr(inode, attr);
672 if (page)
673 page_cache_release(page);
674 return error;
675}
676
677static void shmem_delete_inode(struct inode *inode)
678{
679 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
680 struct shmem_inode_info *info = SHMEM_I(inode);
681
682 if (inode->i_op->truncate == shmem_truncate) {
Mark Fashehfef26652005-09-09 13:01:31 -0700683 truncate_inode_pages(inode->i_mapping, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684 shmem_unacct_size(info->flags, inode->i_size);
685 inode->i_size = 0;
686 shmem_truncate(inode);
687 if (!list_empty(&info->swaplist)) {
688 spin_lock(&shmem_swaplist_lock);
689 list_del_init(&info->swaplist);
690 spin_unlock(&shmem_swaplist_lock);
691 }
692 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -0700693 BUG_ON(inode->i_blocks);
694 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700695 spin_lock(&sbinfo->stat_lock);
696 sbinfo->free_inodes++;
697 spin_unlock(&sbinfo->stat_lock);
698 }
699 clear_inode(inode);
700}
701
702static inline int shmem_find_swp(swp_entry_t entry, swp_entry_t *dir, swp_entry_t *edir)
703{
704 swp_entry_t *ptr;
705
706 for (ptr = dir; ptr < edir; ptr++) {
707 if (ptr->val == entry.val)
708 return ptr - dir;
709 }
710 return -1;
711}
712
713static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
714{
715 struct inode *inode;
716 unsigned long idx;
717 unsigned long size;
718 unsigned long limit;
719 unsigned long stage;
720 struct page **dir;
721 struct page *subdir;
722 swp_entry_t *ptr;
723 int offset;
724
725 idx = 0;
726 ptr = info->i_direct;
727 spin_lock(&info->lock);
728 limit = info->next_index;
729 size = limit;
730 if (size > SHMEM_NR_DIRECT)
731 size = SHMEM_NR_DIRECT;
732 offset = shmem_find_swp(entry, ptr, ptr+size);
733 if (offset >= 0) {
734 shmem_swp_balance_unmap();
735 goto found;
736 }
737 if (!info->i_indirect)
738 goto lost2;
739
740 dir = shmem_dir_map(info->i_indirect);
741 stage = SHMEM_NR_DIRECT + ENTRIES_PER_PAGEPAGE/2;
742
743 for (idx = SHMEM_NR_DIRECT; idx < limit; idx += ENTRIES_PER_PAGE, dir++) {
744 if (unlikely(idx == stage)) {
745 shmem_dir_unmap(dir-1);
746 dir = shmem_dir_map(info->i_indirect) +
747 ENTRIES_PER_PAGE/2 + idx/ENTRIES_PER_PAGEPAGE;
748 while (!*dir) {
749 dir++;
750 idx += ENTRIES_PER_PAGEPAGE;
751 if (idx >= limit)
752 goto lost1;
753 }
754 stage = idx + ENTRIES_PER_PAGEPAGE;
755 subdir = *dir;
756 shmem_dir_unmap(dir);
757 dir = shmem_dir_map(subdir);
758 }
759 subdir = *dir;
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700760 if (subdir && page_private(subdir)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761 ptr = shmem_swp_map(subdir);
762 size = limit - idx;
763 if (size > ENTRIES_PER_PAGE)
764 size = ENTRIES_PER_PAGE;
765 offset = shmem_find_swp(entry, ptr, ptr+size);
766 if (offset >= 0) {
767 shmem_dir_unmap(dir);
768 goto found;
769 }
770 shmem_swp_unmap(ptr);
771 }
772 }
773lost1:
774 shmem_dir_unmap(dir-1);
775lost2:
776 spin_unlock(&info->lock);
777 return 0;
778found:
779 idx += offset;
780 inode = &info->vfs_inode;
781 if (move_from_swap_cache(page, idx, inode->i_mapping) == 0) {
782 info->flags |= SHMEM_PAGEIN;
783 shmem_swp_set(info, ptr + offset, 0);
784 }
785 shmem_swp_unmap(ptr);
786 spin_unlock(&info->lock);
787 /*
788 * Decrement swap count even when the entry is left behind:
789 * try_to_unuse will skip over mms, then reincrement count.
790 */
791 swap_free(entry);
792 return 1;
793}
794
795/*
796 * shmem_unuse() search for an eventually swapped out shmem page.
797 */
798int shmem_unuse(swp_entry_t entry, struct page *page)
799{
800 struct list_head *p, *next;
801 struct shmem_inode_info *info;
802 int found = 0;
803
804 spin_lock(&shmem_swaplist_lock);
805 list_for_each_safe(p, next, &shmem_swaplist) {
806 info = list_entry(p, struct shmem_inode_info, swaplist);
807 if (!info->swapped)
808 list_del_init(&info->swaplist);
809 else if (shmem_unuse_inode(info, entry, page)) {
810 /* move head to start search for next from here */
811 list_move_tail(&shmem_swaplist, &info->swaplist);
812 found = 1;
813 break;
814 }
815 }
816 spin_unlock(&shmem_swaplist_lock);
817 return found;
818}
819
820/*
821 * Move the page from the page cache to the swap cache.
822 */
823static int shmem_writepage(struct page *page, struct writeback_control *wbc)
824{
825 struct shmem_inode_info *info;
826 swp_entry_t *entry, swap;
827 struct address_space *mapping;
828 unsigned long index;
829 struct inode *inode;
830
831 BUG_ON(!PageLocked(page));
832 BUG_ON(page_mapped(page));
833
834 mapping = page->mapping;
835 index = page->index;
836 inode = mapping->host;
837 info = SHMEM_I(inode);
838 if (info->flags & VM_LOCKED)
839 goto redirty;
840 swap = get_swap_page();
841 if (!swap.val)
842 goto redirty;
843
844 spin_lock(&info->lock);
845 shmem_recalc_inode(inode);
846 if (index >= info->next_index) {
847 BUG_ON(!(info->flags & SHMEM_TRUNCATE));
848 goto unlock;
849 }
850 entry = shmem_swp_entry(info, index, NULL);
851 BUG_ON(!entry);
852 BUG_ON(entry->val);
853
854 if (move_to_swap_cache(page, swap) == 0) {
855 shmem_swp_set(info, entry, swap.val);
856 shmem_swp_unmap(entry);
857 spin_unlock(&info->lock);
858 if (list_empty(&info->swaplist)) {
859 spin_lock(&shmem_swaplist_lock);
860 /* move instead of add in case we're racing */
861 list_move_tail(&info->swaplist, &shmem_swaplist);
862 spin_unlock(&shmem_swaplist_lock);
863 }
864 unlock_page(page);
865 return 0;
866 }
867
868 shmem_swp_unmap(entry);
869unlock:
870 spin_unlock(&info->lock);
871 swap_free(swap);
872redirty:
873 set_page_dirty(page);
Zach Brown994fc28c2005-12-15 14:28:17 -0800874 return AOP_WRITEPAGE_ACTIVATE; /* Return with the page locked */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875}
876
877#ifdef CONFIG_NUMA
Hugh Dickinsd15c0232006-03-22 00:08:46 -0800878static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +0000879{
880 char *nodelist = strchr(value, ':');
881 int err = 1;
882
883 if (nodelist) {
884 /* NUL-terminate policy string */
885 *nodelist++ = '\0';
886 if (nodelist_parse(nodelist, *policy_nodes))
887 goto out;
888 }
889 if (!strcmp(value, "default")) {
890 *policy = MPOL_DEFAULT;
891 /* Don't allow a nodelist */
892 if (!nodelist)
893 err = 0;
894 } else if (!strcmp(value, "prefer")) {
895 *policy = MPOL_PREFERRED;
896 /* Insist on a nodelist of one node only */
897 if (nodelist) {
898 char *rest = nodelist;
899 while (isdigit(*rest))
900 rest++;
901 if (!*rest)
902 err = 0;
903 }
904 } else if (!strcmp(value, "bind")) {
905 *policy = MPOL_BIND;
906 /* Insist on a nodelist */
907 if (nodelist)
908 err = 0;
909 } else if (!strcmp(value, "interleave")) {
910 *policy = MPOL_INTERLEAVE;
911 /* Default to nodes online if no nodelist */
912 if (!nodelist)
913 *policy_nodes = node_online_map;
914 err = 0;
915 }
916out:
917 /* Restore string for error message */
918 if (nodelist)
919 *--nodelist = ':';
920 return err;
921}
922
Linus Torvalds1da177e2005-04-16 15:20:36 -0700923static struct page *shmem_swapin_async(struct shared_policy *p,
924 swp_entry_t entry, unsigned long idx)
925{
926 struct page *page;
927 struct vm_area_struct pvma;
928
929 /* Create a pseudo vma that just contains the policy */
930 memset(&pvma, 0, sizeof(struct vm_area_struct));
931 pvma.vm_end = PAGE_SIZE;
932 pvma.vm_pgoff = idx;
933 pvma.vm_policy = mpol_shared_policy_lookup(p, idx);
934 page = read_swap_cache_async(entry, &pvma, 0);
935 mpol_free(pvma.vm_policy);
936 return page;
937}
938
939struct page *shmem_swapin(struct shmem_inode_info *info, swp_entry_t entry,
940 unsigned long idx)
941{
942 struct shared_policy *p = &info->policy;
943 int i, num;
944 struct page *page;
945 unsigned long offset;
946
947 num = valid_swaphandles(entry, &offset);
948 for (i = 0; i < num; offset++, i++) {
949 page = shmem_swapin_async(p,
950 swp_entry(swp_type(entry), offset), idx);
951 if (!page)
952 break;
953 page_cache_release(page);
954 }
955 lru_add_drain(); /* Push any new pages onto the LRU now */
956 return shmem_swapin_async(p, entry, idx);
957}
958
959static struct page *
Al Viro6daa0e22005-10-21 03:18:50 -0400960shmem_alloc_page(gfp_t gfp, struct shmem_inode_info *info,
Linus Torvalds1da177e2005-04-16 15:20:36 -0700961 unsigned long idx)
962{
963 struct vm_area_struct pvma;
964 struct page *page;
965
966 memset(&pvma, 0, sizeof(struct vm_area_struct));
967 pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
968 pvma.vm_pgoff = idx;
969 pvma.vm_end = PAGE_SIZE;
970 page = alloc_page_vma(gfp | __GFP_ZERO, &pvma, 0);
971 mpol_free(pvma.vm_policy);
972 return page;
973}
974#else
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +0000975static inline int shmem_parse_mpol(char *value, int *policy, nodemask_t *policy_nodes)
976{
977 return 1;
978}
979
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980static inline struct page *
981shmem_swapin(struct shmem_inode_info *info,swp_entry_t entry,unsigned long idx)
982{
983 swapin_readahead(entry, 0, NULL);
984 return read_swap_cache_async(entry, NULL, 0);
985}
986
987static inline struct page *
Al Virodd0fc662005-10-07 07:46:04 +0100988shmem_alloc_page(gfp_t gfp,struct shmem_inode_info *info, unsigned long idx)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989{
990 return alloc_page(gfp | __GFP_ZERO);
991}
992#endif
993
994/*
995 * shmem_getpage - either get the page from swap or allocate a new one
996 *
997 * If we allocate a new one we do not mark it dirty. That's up to the
998 * vm. If we swap it in we mark it dirty since we also free the swap
999 * entry since a page cannot live in both the swap and page cache
1000 */
1001static int shmem_getpage(struct inode *inode, unsigned long idx,
1002 struct page **pagep, enum sgp_type sgp, int *type)
1003{
1004 struct address_space *mapping = inode->i_mapping;
1005 struct shmem_inode_info *info = SHMEM_I(inode);
1006 struct shmem_sb_info *sbinfo;
1007 struct page *filepage = *pagep;
1008 struct page *swappage;
1009 swp_entry_t *entry;
1010 swp_entry_t swap;
1011 int error;
1012
1013 if (idx >= SHMEM_MAX_INDEX)
1014 return -EFBIG;
1015 /*
1016 * Normally, filepage is NULL on entry, and either found
1017 * uptodate immediately, or allocated and zeroed, or read
1018 * in under swappage, which is then assigned to filepage.
1019 * But shmem_prepare_write passes in a locked filepage,
1020 * which may be found not uptodate by other callers too,
1021 * and may need to be copied from the swappage read in.
1022 */
1023repeat:
1024 if (!filepage)
1025 filepage = find_lock_page(mapping, idx);
1026 if (filepage && PageUptodate(filepage))
1027 goto done;
1028 error = 0;
1029 if (sgp == SGP_QUICK)
1030 goto failed;
1031
1032 spin_lock(&info->lock);
1033 shmem_recalc_inode(inode);
1034 entry = shmem_swp_alloc(info, idx, sgp);
1035 if (IS_ERR(entry)) {
1036 spin_unlock(&info->lock);
1037 error = PTR_ERR(entry);
1038 goto failed;
1039 }
1040 swap = *entry;
1041
1042 if (swap.val) {
1043 /* Look it up and read it in.. */
1044 swappage = lookup_swap_cache(swap);
1045 if (!swappage) {
1046 shmem_swp_unmap(entry);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047 /* here we actually do the io */
1048 if (type && *type == VM_FAULT_MINOR) {
Christoph Lameterf8891e52006-06-30 01:55:45 -07001049 __count_vm_event(PGMAJFAULT);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 *type = VM_FAULT_MAJOR;
1051 }
Christoph Lameterf8891e52006-06-30 01:55:45 -07001052 spin_unlock(&info->lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053 swappage = shmem_swapin(info, swap, idx);
1054 if (!swappage) {
1055 spin_lock(&info->lock);
1056 entry = shmem_swp_alloc(info, idx, sgp);
1057 if (IS_ERR(entry))
1058 error = PTR_ERR(entry);
1059 else {
1060 if (entry->val == swap.val)
1061 error = -ENOMEM;
1062 shmem_swp_unmap(entry);
1063 }
1064 spin_unlock(&info->lock);
1065 if (error)
1066 goto failed;
1067 goto repeat;
1068 }
1069 wait_on_page_locked(swappage);
1070 page_cache_release(swappage);
1071 goto repeat;
1072 }
1073
1074 /* We have to do this with page locked to prevent races */
1075 if (TestSetPageLocked(swappage)) {
1076 shmem_swp_unmap(entry);
1077 spin_unlock(&info->lock);
1078 wait_on_page_locked(swappage);
1079 page_cache_release(swappage);
1080 goto repeat;
1081 }
1082 if (PageWriteback(swappage)) {
1083 shmem_swp_unmap(entry);
1084 spin_unlock(&info->lock);
1085 wait_on_page_writeback(swappage);
1086 unlock_page(swappage);
1087 page_cache_release(swappage);
1088 goto repeat;
1089 }
1090 if (!PageUptodate(swappage)) {
1091 shmem_swp_unmap(entry);
1092 spin_unlock(&info->lock);
1093 unlock_page(swappage);
1094 page_cache_release(swappage);
1095 error = -EIO;
1096 goto failed;
1097 }
1098
1099 if (filepage) {
1100 shmem_swp_set(info, entry, 0);
1101 shmem_swp_unmap(entry);
1102 delete_from_swap_cache(swappage);
1103 spin_unlock(&info->lock);
1104 copy_highpage(filepage, swappage);
1105 unlock_page(swappage);
1106 page_cache_release(swappage);
1107 flush_dcache_page(filepage);
1108 SetPageUptodate(filepage);
1109 set_page_dirty(filepage);
1110 swap_free(swap);
1111 } else if (!(error = move_from_swap_cache(
1112 swappage, idx, mapping))) {
1113 info->flags |= SHMEM_PAGEIN;
1114 shmem_swp_set(info, entry, 0);
1115 shmem_swp_unmap(entry);
1116 spin_unlock(&info->lock);
1117 filepage = swappage;
1118 swap_free(swap);
1119 } else {
1120 shmem_swp_unmap(entry);
1121 spin_unlock(&info->lock);
1122 unlock_page(swappage);
1123 page_cache_release(swappage);
1124 if (error == -ENOMEM) {
1125 /* let kswapd refresh zone for GFP_ATOMICs */
1126 blk_congestion_wait(WRITE, HZ/50);
1127 }
1128 goto repeat;
1129 }
1130 } else if (sgp == SGP_READ && !filepage) {
1131 shmem_swp_unmap(entry);
1132 filepage = find_get_page(mapping, idx);
1133 if (filepage &&
1134 (!PageUptodate(filepage) || TestSetPageLocked(filepage))) {
1135 spin_unlock(&info->lock);
1136 wait_on_page_locked(filepage);
1137 page_cache_release(filepage);
1138 filepage = NULL;
1139 goto repeat;
1140 }
1141 spin_unlock(&info->lock);
1142 } else {
1143 shmem_swp_unmap(entry);
1144 sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001145 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 spin_lock(&sbinfo->stat_lock);
1147 if (sbinfo->free_blocks == 0 ||
1148 shmem_acct_block(info->flags)) {
1149 spin_unlock(&sbinfo->stat_lock);
1150 spin_unlock(&info->lock);
1151 error = -ENOSPC;
1152 goto failed;
1153 }
1154 sbinfo->free_blocks--;
1155 inode->i_blocks += BLOCKS_PER_PAGE;
1156 spin_unlock(&sbinfo->stat_lock);
1157 } else if (shmem_acct_block(info->flags)) {
1158 spin_unlock(&info->lock);
1159 error = -ENOSPC;
1160 goto failed;
1161 }
1162
1163 if (!filepage) {
1164 spin_unlock(&info->lock);
1165 filepage = shmem_alloc_page(mapping_gfp_mask(mapping),
1166 info,
1167 idx);
1168 if (!filepage) {
1169 shmem_unacct_blocks(info->flags, 1);
1170 shmem_free_blocks(inode, 1);
1171 error = -ENOMEM;
1172 goto failed;
1173 }
1174
1175 spin_lock(&info->lock);
1176 entry = shmem_swp_alloc(info, idx, sgp);
1177 if (IS_ERR(entry))
1178 error = PTR_ERR(entry);
1179 else {
1180 swap = *entry;
1181 shmem_swp_unmap(entry);
1182 }
1183 if (error || swap.val || 0 != add_to_page_cache_lru(
1184 filepage, mapping, idx, GFP_ATOMIC)) {
1185 spin_unlock(&info->lock);
1186 page_cache_release(filepage);
1187 shmem_unacct_blocks(info->flags, 1);
1188 shmem_free_blocks(inode, 1);
1189 filepage = NULL;
1190 if (error)
1191 goto failed;
1192 goto repeat;
1193 }
1194 info->flags |= SHMEM_PAGEIN;
1195 }
1196
1197 info->alloced++;
1198 spin_unlock(&info->lock);
1199 flush_dcache_page(filepage);
1200 SetPageUptodate(filepage);
1201 }
1202done:
1203 if (*pagep != filepage) {
1204 unlock_page(filepage);
1205 *pagep = filepage;
1206 }
1207 return 0;
1208
1209failed:
1210 if (*pagep != filepage) {
1211 unlock_page(filepage);
1212 page_cache_release(filepage);
1213 }
1214 return error;
1215}
1216
1217struct page *shmem_nopage(struct vm_area_struct *vma, unsigned long address, int *type)
1218{
1219 struct inode *inode = vma->vm_file->f_dentry->d_inode;
1220 struct page *page = NULL;
1221 unsigned long idx;
1222 int error;
1223
1224 idx = (address - vma->vm_start) >> PAGE_SHIFT;
1225 idx += vma->vm_pgoff;
1226 idx >>= PAGE_CACHE_SHIFT - PAGE_SHIFT;
1227 if (((loff_t) idx << PAGE_CACHE_SHIFT) >= i_size_read(inode))
1228 return NOPAGE_SIGBUS;
1229
1230 error = shmem_getpage(inode, idx, &page, SGP_CACHE, type);
1231 if (error)
1232 return (error == -ENOMEM)? NOPAGE_OOM: NOPAGE_SIGBUS;
1233
1234 mark_page_accessed(page);
1235 return page;
1236}
1237
1238static int shmem_populate(struct vm_area_struct *vma,
1239 unsigned long addr, unsigned long len,
1240 pgprot_t prot, unsigned long pgoff, int nonblock)
1241{
1242 struct inode *inode = vma->vm_file->f_dentry->d_inode;
1243 struct mm_struct *mm = vma->vm_mm;
1244 enum sgp_type sgp = nonblock? SGP_QUICK: SGP_CACHE;
1245 unsigned long size;
1246
1247 size = (i_size_read(inode) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1248 if (pgoff >= size || pgoff + (len >> PAGE_SHIFT) > size)
1249 return -EINVAL;
1250
1251 while ((long) len > 0) {
1252 struct page *page = NULL;
1253 int err;
1254 /*
1255 * Will need changing if PAGE_CACHE_SIZE != PAGE_SIZE
1256 */
1257 err = shmem_getpage(inode, pgoff, &page, sgp, NULL);
1258 if (err)
1259 return err;
Paolo 'Blaisorblade' Giarrussod44ed4f2005-09-03 15:54:55 -07001260 /* Page may still be null, but only if nonblock was set. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001261 if (page) {
1262 mark_page_accessed(page);
1263 err = install_page(mm, vma, addr, page, prot);
1264 if (err) {
1265 page_cache_release(page);
1266 return err;
1267 }
Hugh Dickins65500d22005-10-29 18:15:59 -07001268 } else if (vma->vm_flags & VM_NONLINEAR) {
Paolo 'Blaisorblade' Giarrussod44ed4f2005-09-03 15:54:55 -07001269 /* No page was found just because we can't read it in
1270 * now (being here implies nonblock != 0), but the page
1271 * may exist, so set the PTE to fault it in later. */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 err = install_file_pte(mm, vma, addr, pgoff, prot);
1273 if (err)
1274 return err;
1275 }
1276
1277 len -= PAGE_SIZE;
1278 addr += PAGE_SIZE;
1279 pgoff++;
1280 }
1281 return 0;
1282}
1283
1284#ifdef CONFIG_NUMA
1285int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
1286{
1287 struct inode *i = vma->vm_file->f_dentry->d_inode;
1288 return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
1289}
1290
1291struct mempolicy *
1292shmem_get_policy(struct vm_area_struct *vma, unsigned long addr)
1293{
1294 struct inode *i = vma->vm_file->f_dentry->d_inode;
1295 unsigned long idx;
1296
1297 idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1298 return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
1299}
1300#endif
1301
1302int shmem_lock(struct file *file, int lock, struct user_struct *user)
1303{
1304 struct inode *inode = file->f_dentry->d_inode;
1305 struct shmem_inode_info *info = SHMEM_I(inode);
1306 int retval = -ENOMEM;
1307
1308 spin_lock(&info->lock);
1309 if (lock && !(info->flags & VM_LOCKED)) {
1310 if (!user_shm_lock(inode->i_size, user))
1311 goto out_nomem;
1312 info->flags |= VM_LOCKED;
1313 }
1314 if (!lock && (info->flags & VM_LOCKED) && user) {
1315 user_shm_unlock(inode->i_size, user);
1316 info->flags &= ~VM_LOCKED;
1317 }
1318 retval = 0;
1319out_nomem:
1320 spin_unlock(&info->lock);
1321 return retval;
1322}
1323
David Howellsb0e15192006-01-06 00:11:42 -08001324int shmem_mmap(struct file *file, struct vm_area_struct *vma)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325{
1326 file_accessed(file);
1327 vma->vm_ops = &shmem_vm_ops;
1328 return 0;
1329}
1330
1331static struct inode *
1332shmem_get_inode(struct super_block *sb, int mode, dev_t dev)
1333{
1334 struct inode *inode;
1335 struct shmem_inode_info *info;
1336 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1337
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001338 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 spin_lock(&sbinfo->stat_lock);
1340 if (!sbinfo->free_inodes) {
1341 spin_unlock(&sbinfo->stat_lock);
1342 return NULL;
1343 }
1344 sbinfo->free_inodes--;
1345 spin_unlock(&sbinfo->stat_lock);
1346 }
1347
1348 inode = new_inode(sb);
1349 if (inode) {
1350 inode->i_mode = mode;
1351 inode->i_uid = current->fsuid;
1352 inode->i_gid = current->fsgid;
1353 inode->i_blksize = PAGE_CACHE_SIZE;
1354 inode->i_blocks = 0;
1355 inode->i_mapping->a_ops = &shmem_aops;
1356 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1357 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1358 info = SHMEM_I(inode);
1359 memset(info, 0, (char *)inode - (char *)info);
1360 spin_lock_init(&info->lock);
1361 INIT_LIST_HEAD(&info->swaplist);
1362
1363 switch (mode & S_IFMT) {
1364 default:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 init_special_inode(inode, mode, dev);
1366 break;
1367 case S_IFREG:
1368 inode->i_op = &shmem_inode_operations;
1369 inode->i_fop = &shmem_file_operations;
Robin Holt7339ff82006-01-14 13:20:48 -08001370 mpol_shared_policy_init(&info->policy, sbinfo->policy,
1371 &sbinfo->policy_nodes);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 break;
1373 case S_IFDIR:
1374 inode->i_nlink++;
1375 /* Some things misbehave if size == 0 on a directory */
1376 inode->i_size = 2 * BOGO_DIRENT_SIZE;
1377 inode->i_op = &shmem_dir_inode_operations;
1378 inode->i_fop = &simple_dir_operations;
1379 break;
1380 case S_IFLNK:
1381 /*
1382 * Must not load anything in the rbtree,
1383 * mpol_free_shared_policy will not be called.
1384 */
Robin Holt7339ff82006-01-14 13:20:48 -08001385 mpol_shared_policy_init(&info->policy, MPOL_DEFAULT,
1386 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 break;
1388 }
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001389 } else if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 spin_lock(&sbinfo->stat_lock);
1391 sbinfo->free_inodes++;
1392 spin_unlock(&sbinfo->stat_lock);
1393 }
1394 return inode;
1395}
1396
1397#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398static struct inode_operations shmem_symlink_inode_operations;
1399static struct inode_operations shmem_symlink_inline_operations;
1400
1401/*
1402 * Normally tmpfs makes no use of shmem_prepare_write, but it
1403 * lets a tmpfs file be used read-write below the loop driver.
1404 */
1405static int
1406shmem_prepare_write(struct file *file, struct page *page, unsigned offset, unsigned to)
1407{
1408 struct inode *inode = page->mapping->host;
1409 return shmem_getpage(inode, page->index, &page, SGP_WRITE, NULL);
1410}
1411
1412static ssize_t
1413shmem_file_write(struct file *file, const char __user *buf, size_t count, loff_t *ppos)
1414{
1415 struct inode *inode = file->f_dentry->d_inode;
1416 loff_t pos;
1417 unsigned long written;
1418 ssize_t err;
1419
1420 if ((ssize_t) count < 0)
1421 return -EINVAL;
1422
1423 if (!access_ok(VERIFY_READ, buf, count))
1424 return -EFAULT;
1425
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001426 mutex_lock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427
1428 pos = *ppos;
1429 written = 0;
1430
1431 err = generic_write_checks(file, &pos, &count, 0);
1432 if (err || !count)
1433 goto out;
1434
1435 err = remove_suid(file->f_dentry);
1436 if (err)
1437 goto out;
1438
1439 inode->i_ctime = inode->i_mtime = CURRENT_TIME;
1440
1441 do {
1442 struct page *page = NULL;
1443 unsigned long bytes, index, offset;
1444 char *kaddr;
1445 int left;
1446
1447 offset = (pos & (PAGE_CACHE_SIZE -1)); /* Within page */
1448 index = pos >> PAGE_CACHE_SHIFT;
1449 bytes = PAGE_CACHE_SIZE - offset;
1450 if (bytes > count)
1451 bytes = count;
1452
1453 /*
1454 * We don't hold page lock across copy from user -
1455 * what would it guard against? - so no deadlock here.
1456 * But it still may be a good idea to prefault below.
1457 */
1458
1459 err = shmem_getpage(inode, index, &page, SGP_WRITE, NULL);
1460 if (err)
1461 break;
1462
1463 left = bytes;
1464 if (PageHighMem(page)) {
1465 volatile unsigned char dummy;
1466 __get_user(dummy, buf);
1467 __get_user(dummy, buf + bytes - 1);
1468
1469 kaddr = kmap_atomic(page, KM_USER0);
1470 left = __copy_from_user_inatomic(kaddr + offset,
1471 buf, bytes);
1472 kunmap_atomic(kaddr, KM_USER0);
1473 }
1474 if (left) {
1475 kaddr = kmap(page);
1476 left = __copy_from_user(kaddr + offset, buf, bytes);
1477 kunmap(page);
1478 }
1479
1480 written += bytes;
1481 count -= bytes;
1482 pos += bytes;
1483 buf += bytes;
1484 if (pos > inode->i_size)
1485 i_size_write(inode, pos);
1486
1487 flush_dcache_page(page);
1488 set_page_dirty(page);
1489 mark_page_accessed(page);
1490 page_cache_release(page);
1491
1492 if (left) {
1493 pos -= left;
1494 written -= left;
1495 err = -EFAULT;
1496 break;
1497 }
1498
1499 /*
1500 * Our dirty pages are not counted in nr_dirty,
1501 * and we do not attempt to balance dirty pages.
1502 */
1503
1504 cond_resched();
1505 } while (count);
1506
1507 *ppos = pos;
1508 if (written)
1509 err = written;
1510out:
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001511 mutex_unlock(&inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 return err;
1513}
1514
1515static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1516{
1517 struct inode *inode = filp->f_dentry->d_inode;
1518 struct address_space *mapping = inode->i_mapping;
1519 unsigned long index, offset;
1520
1521 index = *ppos >> PAGE_CACHE_SHIFT;
1522 offset = *ppos & ~PAGE_CACHE_MASK;
1523
1524 for (;;) {
1525 struct page *page = NULL;
1526 unsigned long end_index, nr, ret;
1527 loff_t i_size = i_size_read(inode);
1528
1529 end_index = i_size >> PAGE_CACHE_SHIFT;
1530 if (index > end_index)
1531 break;
1532 if (index == end_index) {
1533 nr = i_size & ~PAGE_CACHE_MASK;
1534 if (nr <= offset)
1535 break;
1536 }
1537
1538 desc->error = shmem_getpage(inode, index, &page, SGP_READ, NULL);
1539 if (desc->error) {
1540 if (desc->error == -EINVAL)
1541 desc->error = 0;
1542 break;
1543 }
1544
1545 /*
1546 * We must evaluate after, since reads (unlike writes)
Jes Sorensen1b1dcc12006-01-09 15:59:24 -08001547 * are called without i_mutex protection against truncate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548 */
1549 nr = PAGE_CACHE_SIZE;
1550 i_size = i_size_read(inode);
1551 end_index = i_size >> PAGE_CACHE_SHIFT;
1552 if (index == end_index) {
1553 nr = i_size & ~PAGE_CACHE_MASK;
1554 if (nr <= offset) {
1555 if (page)
1556 page_cache_release(page);
1557 break;
1558 }
1559 }
1560 nr -= offset;
1561
1562 if (page) {
1563 /*
1564 * If users can be writing to this page using arbitrary
1565 * virtual addresses, take care about potential aliasing
1566 * before reading the page on the kernel side.
1567 */
1568 if (mapping_writably_mapped(mapping))
1569 flush_dcache_page(page);
1570 /*
1571 * Mark the page accessed if we read the beginning.
1572 */
1573 if (!offset)
1574 mark_page_accessed(page);
Nick Pigginb5810032005-10-29 18:16:12 -07001575 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 page = ZERO_PAGE(0);
Nick Pigginb5810032005-10-29 18:16:12 -07001577 page_cache_get(page);
1578 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579
1580 /*
1581 * Ok, we have the page, and it's up-to-date, so
1582 * now we can copy it to user space...
1583 *
1584 * The actor routine returns how many bytes were actually used..
1585 * NOTE! This may not be the same as how much of a user buffer
1586 * we filled up (we may be padding etc), so we can only update
1587 * "pos" here (the actor routine has to update the user buffer
1588 * pointers and the remaining count).
1589 */
1590 ret = actor(desc, page, offset, nr);
1591 offset += ret;
1592 index += offset >> PAGE_CACHE_SHIFT;
1593 offset &= ~PAGE_CACHE_MASK;
1594
1595 page_cache_release(page);
1596 if (ret != nr || !desc->count)
1597 break;
1598
1599 cond_resched();
1600 }
1601
1602 *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1603 file_accessed(filp);
1604}
1605
1606static ssize_t shmem_file_read(struct file *filp, char __user *buf, size_t count, loff_t *ppos)
1607{
1608 read_descriptor_t desc;
1609
1610 if ((ssize_t) count < 0)
1611 return -EINVAL;
1612 if (!access_ok(VERIFY_WRITE, buf, count))
1613 return -EFAULT;
1614 if (!count)
1615 return 0;
1616
1617 desc.written = 0;
1618 desc.count = count;
1619 desc.arg.buf = buf;
1620 desc.error = 0;
1621
1622 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1623 if (desc.written)
1624 return desc.written;
1625 return desc.error;
1626}
1627
1628static ssize_t shmem_file_sendfile(struct file *in_file, loff_t *ppos,
1629 size_t count, read_actor_t actor, void *target)
1630{
1631 read_descriptor_t desc;
1632
1633 if (!count)
1634 return 0;
1635
1636 desc.written = 0;
1637 desc.count = count;
1638 desc.arg.data = target;
1639 desc.error = 0;
1640
1641 do_shmem_file_read(in_file, ppos, &desc, actor);
1642 if (desc.written)
1643 return desc.written;
1644 return desc.error;
1645}
1646
David Howells726c3342006-06-23 02:02:58 -07001647static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648{
David Howells726c3342006-06-23 02:02:58 -07001649 struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650
1651 buf->f_type = TMPFS_MAGIC;
1652 buf->f_bsize = PAGE_CACHE_SIZE;
1653 buf->f_namelen = NAME_MAX;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001654 spin_lock(&sbinfo->stat_lock);
1655 if (sbinfo->max_blocks) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 buf->f_blocks = sbinfo->max_blocks;
1657 buf->f_bavail = buf->f_bfree = sbinfo->free_blocks;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001658 }
1659 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 buf->f_files = sbinfo->max_inodes;
1661 buf->f_ffree = sbinfo->free_inodes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 }
1663 /* else leave those fields 0 like simple_statfs */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001664 spin_unlock(&sbinfo->stat_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 return 0;
1666}
1667
1668/*
1669 * File creation. Allocate an inode, and we're done..
1670 */
1671static int
1672shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1673{
1674 struct inode *inode = shmem_get_inode(dir->i_sb, mode, dev);
1675 int error = -ENOSPC;
1676
1677 if (inode) {
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001678 error = security_inode_init_security(inode, dir, NULL, NULL,
1679 NULL);
1680 if (error) {
1681 if (error != -EOPNOTSUPP) {
1682 iput(inode);
1683 return error;
1684 }
1685 error = 0;
1686 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687 if (dir->i_mode & S_ISGID) {
1688 inode->i_gid = dir->i_gid;
1689 if (S_ISDIR(mode))
1690 inode->i_mode |= S_ISGID;
1691 }
1692 dir->i_size += BOGO_DIRENT_SIZE;
1693 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1694 d_instantiate(dentry, inode);
1695 dget(dentry); /* Extra count - pin the dentry in core */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696 }
1697 return error;
1698}
1699
1700static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1701{
1702 int error;
1703
1704 if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1705 return error;
1706 dir->i_nlink++;
1707 return 0;
1708}
1709
1710static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1711 struct nameidata *nd)
1712{
1713 return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1714}
1715
1716/*
1717 * Link a file..
1718 */
1719static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1720{
1721 struct inode *inode = old_dentry->d_inode;
1722 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1723
1724 /*
1725 * No ordinary (disk based) filesystem counts links as inodes;
1726 * but each new link needs a new dentry, pinning lowmem, and
1727 * tmpfs dentries cannot be pruned until they are unlinked.
1728 */
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001729 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001730 spin_lock(&sbinfo->stat_lock);
1731 if (!sbinfo->free_inodes) {
1732 spin_unlock(&sbinfo->stat_lock);
1733 return -ENOSPC;
1734 }
1735 sbinfo->free_inodes--;
1736 spin_unlock(&sbinfo->stat_lock);
1737 }
1738
1739 dir->i_size += BOGO_DIRENT_SIZE;
1740 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1741 inode->i_nlink++;
1742 atomic_inc(&inode->i_count); /* New dentry reference */
1743 dget(dentry); /* Extra pinning count for the created dentry */
1744 d_instantiate(dentry, inode);
1745 return 0;
1746}
1747
1748static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1749{
1750 struct inode *inode = dentry->d_inode;
1751
1752 if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode)) {
1753 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001754 if (sbinfo->max_inodes) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755 spin_lock(&sbinfo->stat_lock);
1756 sbinfo->free_inodes++;
1757 spin_unlock(&sbinfo->stat_lock);
1758 }
1759 }
1760
1761 dir->i_size -= BOGO_DIRENT_SIZE;
1762 inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1763 inode->i_nlink--;
1764 dput(dentry); /* Undo the count from "create" - this does all the work */
1765 return 0;
1766}
1767
1768static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1769{
1770 if (!simple_empty(dentry))
1771 return -ENOTEMPTY;
1772
Sergey Vlasov86bc8432006-06-12 21:53:23 +01001773 dentry->d_inode->i_nlink--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 dir->i_nlink--;
1775 return shmem_unlink(dir, dentry);
1776}
1777
1778/*
1779 * The VFS layer already does all the dentry stuff for rename,
1780 * we just have to decrement the usage count for the target if
1781 * it exists so that the VFS layer correctly free's it when it
1782 * gets overwritten.
1783 */
1784static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1785{
1786 struct inode *inode = old_dentry->d_inode;
1787 int they_are_dirs = S_ISDIR(inode->i_mode);
1788
1789 if (!simple_empty(new_dentry))
1790 return -ENOTEMPTY;
1791
1792 if (new_dentry->d_inode) {
1793 (void) shmem_unlink(new_dir, new_dentry);
1794 if (they_are_dirs)
1795 old_dir->i_nlink--;
1796 } else if (they_are_dirs) {
1797 old_dir->i_nlink--;
1798 new_dir->i_nlink++;
1799 }
1800
1801 old_dir->i_size -= BOGO_DIRENT_SIZE;
1802 new_dir->i_size += BOGO_DIRENT_SIZE;
1803 old_dir->i_ctime = old_dir->i_mtime =
1804 new_dir->i_ctime = new_dir->i_mtime =
1805 inode->i_ctime = CURRENT_TIME;
1806 return 0;
1807}
1808
1809static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1810{
1811 int error;
1812 int len;
1813 struct inode *inode;
1814 struct page *page = NULL;
1815 char *kaddr;
1816 struct shmem_inode_info *info;
1817
1818 len = strlen(symname) + 1;
1819 if (len > PAGE_CACHE_SIZE)
1820 return -ENAMETOOLONG;
1821
1822 inode = shmem_get_inode(dir->i_sb, S_IFLNK|S_IRWXUGO, 0);
1823 if (!inode)
1824 return -ENOSPC;
1825
Stephen Smalley570bc1c2005-09-09 13:01:43 -07001826 error = security_inode_init_security(inode, dir, NULL, NULL,
1827 NULL);
1828 if (error) {
1829 if (error != -EOPNOTSUPP) {
1830 iput(inode);
1831 return error;
1832 }
1833 error = 0;
1834 }
1835
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 info = SHMEM_I(inode);
1837 inode->i_size = len-1;
1838 if (len <= (char *)inode - (char *)info) {
1839 /* do it inline */
1840 memcpy(info, symname, len);
1841 inode->i_op = &shmem_symlink_inline_operations;
1842 } else {
1843 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1844 if (error) {
1845 iput(inode);
1846 return error;
1847 }
1848 inode->i_op = &shmem_symlink_inode_operations;
1849 kaddr = kmap_atomic(page, KM_USER0);
1850 memcpy(kaddr, symname, len);
1851 kunmap_atomic(kaddr, KM_USER0);
1852 set_page_dirty(page);
1853 page_cache_release(page);
1854 }
1855 if (dir->i_mode & S_ISGID)
1856 inode->i_gid = dir->i_gid;
1857 dir->i_size += BOGO_DIRENT_SIZE;
1858 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1859 d_instantiate(dentry, inode);
1860 dget(dentry);
1861 return 0;
1862}
1863
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001864static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001865{
1866 nd_set_link(nd, (char *)SHMEM_I(dentry->d_inode));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001867 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868}
1869
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001870static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871{
1872 struct page *page = NULL;
1873 int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1874 nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001875 return page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876}
1877
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001878static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879{
1880 if (!IS_ERR(nd_get_link(nd))) {
Linus Torvaldscc314ee2005-08-19 18:02:56 -07001881 struct page *page = cookie;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 kunmap(page);
1883 mark_page_accessed(page);
1884 page_cache_release(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 }
1886}
1887
1888static struct inode_operations shmem_symlink_inline_operations = {
1889 .readlink = generic_readlink,
1890 .follow_link = shmem_follow_link_inline,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891};
1892
1893static struct inode_operations shmem_symlink_inode_operations = {
1894 .truncate = shmem_truncate,
1895 .readlink = generic_readlink,
1896 .follow_link = shmem_follow_link,
1897 .put_link = shmem_put_link,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898};
1899
Robin Holt7339ff82006-01-14 13:20:48 -08001900static int shmem_parse_options(char *options, int *mode, uid_t *uid,
1901 gid_t *gid, unsigned long *blocks, unsigned long *inodes,
1902 int *policy, nodemask_t *policy_nodes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903{
1904 char *this_char, *value, *rest;
1905
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00001906 while (options != NULL) {
1907 this_char = options;
1908 for (;;) {
1909 /*
1910 * NUL-terminate this option: unfortunately,
1911 * mount options form a comma-separated list,
1912 * but mpol's nodelist may also contain commas.
1913 */
1914 options = strchr(options, ',');
1915 if (options == NULL)
1916 break;
1917 options++;
1918 if (!isdigit(*options)) {
1919 options[-1] = '\0';
1920 break;
1921 }
1922 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001923 if (!*this_char)
1924 continue;
1925 if ((value = strchr(this_char,'=')) != NULL) {
1926 *value++ = 0;
1927 } else {
1928 printk(KERN_ERR
1929 "tmpfs: No value for mount option '%s'\n",
1930 this_char);
1931 return 1;
1932 }
1933
1934 if (!strcmp(this_char,"size")) {
1935 unsigned long long size;
1936 size = memparse(value,&rest);
1937 if (*rest == '%') {
1938 size <<= PAGE_SHIFT;
1939 size *= totalram_pages;
1940 do_div(size, 100);
1941 rest++;
1942 }
1943 if (*rest)
1944 goto bad_val;
1945 *blocks = size >> PAGE_CACHE_SHIFT;
1946 } else if (!strcmp(this_char,"nr_blocks")) {
1947 *blocks = memparse(value,&rest);
1948 if (*rest)
1949 goto bad_val;
1950 } else if (!strcmp(this_char,"nr_inodes")) {
1951 *inodes = memparse(value,&rest);
1952 if (*rest)
1953 goto bad_val;
1954 } else if (!strcmp(this_char,"mode")) {
1955 if (!mode)
1956 continue;
1957 *mode = simple_strtoul(value,&rest,8);
1958 if (*rest)
1959 goto bad_val;
1960 } else if (!strcmp(this_char,"uid")) {
1961 if (!uid)
1962 continue;
1963 *uid = simple_strtoul(value,&rest,0);
1964 if (*rest)
1965 goto bad_val;
1966 } else if (!strcmp(this_char,"gid")) {
1967 if (!gid)
1968 continue;
1969 *gid = simple_strtoul(value,&rest,0);
1970 if (*rest)
1971 goto bad_val;
Robin Holt7339ff82006-01-14 13:20:48 -08001972 } else if (!strcmp(this_char,"mpol")) {
Hugh Dickinsb00dc3a2006-02-21 23:49:47 +00001973 if (shmem_parse_mpol(value,policy,policy_nodes))
Robin Holt7339ff82006-01-14 13:20:48 -08001974 goto bad_val;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 } else {
1976 printk(KERN_ERR "tmpfs: Bad mount option %s\n",
1977 this_char);
1978 return 1;
1979 }
1980 }
1981 return 0;
1982
1983bad_val:
1984 printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
1985 value, this_char);
1986 return 1;
1987
1988}
1989
1990static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
1991{
1992 struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001993 unsigned long max_blocks = sbinfo->max_blocks;
1994 unsigned long max_inodes = sbinfo->max_inodes;
Robin Holt7339ff82006-01-14 13:20:48 -08001995 int policy = sbinfo->policy;
1996 nodemask_t policy_nodes = sbinfo->policy_nodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07001997 unsigned long blocks;
1998 unsigned long inodes;
1999 int error = -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000
Robin Holt7339ff82006-01-14 13:20:48 -08002001 if (shmem_parse_options(data, NULL, NULL, NULL, &max_blocks,
2002 &max_inodes, &policy, &policy_nodes))
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002003 return error;
2004
2005 spin_lock(&sbinfo->stat_lock);
2006 blocks = sbinfo->max_blocks - sbinfo->free_blocks;
2007 inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2008 if (max_blocks < blocks)
2009 goto out;
2010 if (max_inodes < inodes)
2011 goto out;
2012 /*
2013 * Those tests also disallow limited->unlimited while any are in
2014 * use, so i_blocks will always be zero when max_blocks is zero;
2015 * but we must separately disallow unlimited->limited, because
2016 * in that case we have no record of how much is already in use.
2017 */
2018 if (max_blocks && !sbinfo->max_blocks)
2019 goto out;
2020 if (max_inodes && !sbinfo->max_inodes)
2021 goto out;
2022
2023 error = 0;
2024 sbinfo->max_blocks = max_blocks;
2025 sbinfo->free_blocks = max_blocks - blocks;
2026 sbinfo->max_inodes = max_inodes;
2027 sbinfo->free_inodes = max_inodes - inodes;
Robin Holt7339ff82006-01-14 13:20:48 -08002028 sbinfo->policy = policy;
2029 sbinfo->policy_nodes = policy_nodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002030out:
2031 spin_unlock(&sbinfo->stat_lock);
2032 return error;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033}
2034#endif
2035
2036static void shmem_put_super(struct super_block *sb)
2037{
2038 kfree(sb->s_fs_info);
2039 sb->s_fs_info = NULL;
2040}
2041
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042static int shmem_fill_super(struct super_block *sb,
2043 void *data, int silent)
2044{
2045 struct inode *inode;
2046 struct dentry *root;
2047 int mode = S_IRWXUGO | S_ISVTX;
2048 uid_t uid = current->fsuid;
2049 gid_t gid = current->fsgid;
2050 int err = -ENOMEM;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002051 struct shmem_sb_info *sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 unsigned long blocks = 0;
2053 unsigned long inodes = 0;
Robin Holt7339ff82006-01-14 13:20:48 -08002054 int policy = MPOL_DEFAULT;
2055 nodemask_t policy_nodes = node_online_map;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002057#ifdef CONFIG_TMPFS
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058 /*
2059 * Per default we only allow half of the physical ram per
2060 * tmpfs instance, limiting inodes to one per page of lowmem;
2061 * but the internal instance is left unlimited.
2062 */
2063 if (!(sb->s_flags & MS_NOUSER)) {
2064 blocks = totalram_pages / 2;
2065 inodes = totalram_pages - totalhigh_pages;
2066 if (inodes > blocks)
2067 inodes = blocks;
Robin Holt7339ff82006-01-14 13:20:48 -08002068 if (shmem_parse_options(data, &mode, &uid, &gid, &blocks,
2069 &inodes, &policy, &policy_nodes))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 return -EINVAL;
2071 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072#else
2073 sb->s_flags |= MS_NOUSER;
2074#endif
2075
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002076 /* Round up to L1_CACHE_BYTES to resist false sharing */
2077 sbinfo = kmalloc(max((int)sizeof(struct shmem_sb_info),
2078 L1_CACHE_BYTES), GFP_KERNEL);
2079 if (!sbinfo)
2080 return -ENOMEM;
2081
2082 spin_lock_init(&sbinfo->stat_lock);
2083 sbinfo->max_blocks = blocks;
2084 sbinfo->free_blocks = blocks;
2085 sbinfo->max_inodes = inodes;
2086 sbinfo->free_inodes = inodes;
Robin Holt7339ff82006-01-14 13:20:48 -08002087 sbinfo->policy = policy;
2088 sbinfo->policy_nodes = policy_nodes;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002089
2090 sb->s_fs_info = sbinfo;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 sb->s_maxbytes = SHMEM_MAX_BYTES;
2092 sb->s_blocksize = PAGE_CACHE_SIZE;
2093 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2094 sb->s_magic = TMPFS_MAGIC;
2095 sb->s_op = &shmem_ops;
Robin H. Johnsoncfd95a92006-06-12 21:50:25 +01002096 sb->s_time_gran = 1;
Hugh Dickins0edd73b2005-06-21 17:15:04 -07002097
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098 inode = shmem_get_inode(sb, S_IFDIR | mode, 0);
2099 if (!inode)
2100 goto failed;
2101 inode->i_uid = uid;
2102 inode->i_gid = gid;
2103 root = d_alloc_root(inode);
2104 if (!root)
2105 goto failed_iput;
2106 sb->s_root = root;
2107 return 0;
2108
2109failed_iput:
2110 iput(inode);
2111failed:
2112 shmem_put_super(sb);
2113 return err;
2114}
2115
Pekka Enbergfcc234f2006-03-22 00:08:13 -08002116static struct kmem_cache *shmem_inode_cachep;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117
2118static struct inode *shmem_alloc_inode(struct super_block *sb)
2119{
2120 struct shmem_inode_info *p;
2121 p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, SLAB_KERNEL);
2122 if (!p)
2123 return NULL;
2124 return &p->vfs_inode;
2125}
2126
2127static void shmem_destroy_inode(struct inode *inode)
2128{
2129 if ((inode->i_mode & S_IFMT) == S_IFREG) {
2130 /* only struct inode is valid if it's an inline symlink */
2131 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2132 }
2133 kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2134}
2135
Pekka Enbergfcc234f2006-03-22 00:08:13 -08002136static void init_once(void *foo, struct kmem_cache *cachep,
2137 unsigned long flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138{
2139 struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2140
2141 if ((flags & (SLAB_CTOR_VERIFY|SLAB_CTOR_CONSTRUCTOR)) ==
2142 SLAB_CTOR_CONSTRUCTOR) {
2143 inode_init_once(&p->vfs_inode);
2144 }
2145}
2146
2147static int init_inodecache(void)
2148{
2149 shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2150 sizeof(struct shmem_inode_info),
2151 0, 0, init_once, NULL);
2152 if (shmem_inode_cachep == NULL)
2153 return -ENOMEM;
2154 return 0;
2155}
2156
2157static void destroy_inodecache(void)
2158{
2159 if (kmem_cache_destroy(shmem_inode_cachep))
2160 printk(KERN_INFO "shmem_inode_cache: not all structures were freed\n");
2161}
2162
Christoph Hellwigf5e54d62006-06-28 04:26:44 -07002163static const struct address_space_operations shmem_aops = {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 .writepage = shmem_writepage,
2165 .set_page_dirty = __set_page_dirty_nobuffers,
2166#ifdef CONFIG_TMPFS
2167 .prepare_write = shmem_prepare_write,
2168 .commit_write = simple_commit_write,
2169#endif
Lee Schermerhorn304dbdb2006-04-22 02:35:48 -07002170 .migratepage = migrate_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002171};
2172
2173static struct file_operations shmem_file_operations = {
2174 .mmap = shmem_mmap,
2175#ifdef CONFIG_TMPFS
2176 .llseek = generic_file_llseek,
2177 .read = shmem_file_read,
2178 .write = shmem_file_write,
2179 .fsync = simple_sync_file,
2180 .sendfile = shmem_file_sendfile,
2181#endif
2182};
2183
2184static struct inode_operations shmem_inode_operations = {
2185 .truncate = shmem_truncate,
2186 .setattr = shmem_notify_change,
Badari Pulavartyf6b3ec22006-01-06 00:10:38 -08002187 .truncate_range = shmem_truncate_range,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188};
2189
2190static struct inode_operations shmem_dir_inode_operations = {
2191#ifdef CONFIG_TMPFS
2192 .create = shmem_create,
2193 .lookup = simple_lookup,
2194 .link = shmem_link,
2195 .unlink = shmem_unlink,
2196 .symlink = shmem_symlink,
2197 .mkdir = shmem_mkdir,
2198 .rmdir = shmem_rmdir,
2199 .mknod = shmem_mknod,
2200 .rename = shmem_rename,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201#endif
2202};
2203
2204static struct super_operations shmem_ops = {
2205 .alloc_inode = shmem_alloc_inode,
2206 .destroy_inode = shmem_destroy_inode,
2207#ifdef CONFIG_TMPFS
2208 .statfs = shmem_statfs,
2209 .remount_fs = shmem_remount_fs,
2210#endif
2211 .delete_inode = shmem_delete_inode,
2212 .drop_inode = generic_delete_inode,
2213 .put_super = shmem_put_super,
2214};
2215
2216static struct vm_operations_struct shmem_vm_ops = {
2217 .nopage = shmem_nopage,
2218 .populate = shmem_populate,
2219#ifdef CONFIG_NUMA
2220 .set_policy = shmem_set_policy,
2221 .get_policy = shmem_get_policy,
2222#endif
2223};
2224
2225
David Howells454e2392006-06-23 02:02:57 -07002226static int shmem_get_sb(struct file_system_type *fs_type,
2227 int flags, const char *dev_name, void *data, struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002228{
David Howells454e2392006-06-23 02:02:57 -07002229 return get_sb_nodev(fs_type, flags, data, shmem_fill_super, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230}
2231
2232static struct file_system_type tmpfs_fs_type = {
2233 .owner = THIS_MODULE,
2234 .name = "tmpfs",
2235 .get_sb = shmem_get_sb,
2236 .kill_sb = kill_litter_super,
2237};
2238static struct vfsmount *shm_mnt;
2239
2240static int __init init_tmpfs(void)
2241{
2242 int error;
2243
2244 error = init_inodecache();
2245 if (error)
2246 goto out3;
2247
2248 error = register_filesystem(&tmpfs_fs_type);
2249 if (error) {
2250 printk(KERN_ERR "Could not register tmpfs\n");
2251 goto out2;
2252 }
Greg Kroah-Hartman95dc1122005-06-20 21:15:16 -07002253
Trond Myklebust1f5ce9e2006-06-09 09:34:16 -04002254 shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 tmpfs_fs_type.name, NULL);
2256 if (IS_ERR(shm_mnt)) {
2257 error = PTR_ERR(shm_mnt);
2258 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2259 goto out1;
2260 }
2261 return 0;
2262
2263out1:
2264 unregister_filesystem(&tmpfs_fs_type);
2265out2:
2266 destroy_inodecache();
2267out3:
2268 shm_mnt = ERR_PTR(error);
2269 return error;
2270}
2271module_init(init_tmpfs)
2272
2273/*
2274 * shmem_file_setup - get an unlinked file living in tmpfs
2275 *
2276 * @name: name for dentry (to be seen in /proc/<pid>/maps
2277 * @size: size to be set for the file
2278 *
2279 */
2280struct file *shmem_file_setup(char *name, loff_t size, unsigned long flags)
2281{
2282 int error;
2283 struct file *file;
2284 struct inode *inode;
2285 struct dentry *dentry, *root;
2286 struct qstr this;
2287
2288 if (IS_ERR(shm_mnt))
2289 return (void *)shm_mnt;
2290
2291 if (size < 0 || size > SHMEM_MAX_BYTES)
2292 return ERR_PTR(-EINVAL);
2293
2294 if (shmem_acct_size(flags, size))
2295 return ERR_PTR(-ENOMEM);
2296
2297 error = -ENOMEM;
2298 this.name = name;
2299 this.len = strlen(name);
2300 this.hash = 0; /* will go */
2301 root = shm_mnt->mnt_root;
2302 dentry = d_alloc(root, &this);
2303 if (!dentry)
2304 goto put_memory;
2305
2306 error = -ENFILE;
2307 file = get_empty_filp();
2308 if (!file)
2309 goto put_dentry;
2310
2311 error = -ENOSPC;
2312 inode = shmem_get_inode(root->d_sb, S_IFREG | S_IRWXUGO, 0);
2313 if (!inode)
2314 goto close_file;
2315
2316 SHMEM_I(inode)->flags = flags & VM_ACCOUNT;
2317 d_instantiate(dentry, inode);
2318 inode->i_size = size;
2319 inode->i_nlink = 0; /* It is unlinked */
2320 file->f_vfsmnt = mntget(shm_mnt);
2321 file->f_dentry = dentry;
2322 file->f_mapping = inode->i_mapping;
2323 file->f_op = &shmem_file_operations;
2324 file->f_mode = FMODE_WRITE | FMODE_READ;
2325 return file;
2326
2327close_file:
2328 put_filp(file);
2329put_dentry:
2330 dput(dentry);
2331put_memory:
2332 shmem_unacct_size(flags, size);
2333 return ERR_PTR(error);
2334}
2335
2336/*
2337 * shmem_zero_setup - setup a shared anonymous mapping
2338 *
2339 * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2340 */
2341int shmem_zero_setup(struct vm_area_struct *vma)
2342{
2343 struct file *file;
2344 loff_t size = vma->vm_end - vma->vm_start;
2345
2346 file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2347 if (IS_ERR(file))
2348 return PTR_ERR(file);
2349
2350 if (vma->vm_file)
2351 fput(vma->vm_file);
2352 vma->vm_file = file;
2353 vma->vm_ops = &shmem_vm_ops;
2354 return 0;
2355}