]> git.openfabrics.org - ~shefty/rdma-dev.git/blob - mm/shmem.c
tmpfs: demolish old swap vector support
[~shefty/rdma-dev.git] / mm / shmem.c
1 /*
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.
9  * Copyright (C) 2002-2005 Hugh Dickins.
10  * Copyright (C) 2002-2005 VERITAS Software Corporation.
11  * 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  * tiny-shmem:
18  * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
19  *
20  * This file is released under the GPL.
21  */
22
23 #include <linux/fs.h>
24 #include <linux/init.h>
25 #include <linux/vfs.h>
26 #include <linux/mount.h>
27 #include <linux/pagemap.h>
28 #include <linux/file.h>
29 #include <linux/mm.h>
30 #include <linux/module.h>
31 #include <linux/percpu_counter.h>
32 #include <linux/swap.h>
33
34 static struct vfsmount *shm_mnt;
35
36 #ifdef CONFIG_SHMEM
37 /*
38  * This virtual memory filesystem is heavily based on the ramfs. It
39  * extends ramfs by the ability to use swap and honor resource limits
40  * which makes it a completely usable filesystem.
41  */
42
43 #include <linux/xattr.h>
44 #include <linux/exportfs.h>
45 #include <linux/posix_acl.h>
46 #include <linux/generic_acl.h>
47 #include <linux/mman.h>
48 #include <linux/string.h>
49 #include <linux/slab.h>
50 #include <linux/backing-dev.h>
51 #include <linux/shmem_fs.h>
52 #include <linux/writeback.h>
53 #include <linux/blkdev.h>
54 #include <linux/splice.h>
55 #include <linux/security.h>
56 #include <linux/swapops.h>
57 #include <linux/mempolicy.h>
58 #include <linux/namei.h>
59 #include <linux/ctype.h>
60 #include <linux/migrate.h>
61 #include <linux/highmem.h>
62 #include <linux/seq_file.h>
63 #include <linux/magic.h>
64
65 #include <asm/uaccess.h>
66 #include <asm/div64.h>
67 #include <asm/pgtable.h>
68
69 #define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
70 #define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
71
72 /* Pretend that each entry is of this size in directory's i_size */
73 #define BOGO_DIRENT_SIZE 20
74
75 struct shmem_xattr {
76         struct list_head list;  /* anchored by shmem_inode_info->xattr_list */
77         char *name;             /* xattr name */
78         size_t size;
79         char value[0];
80 };
81
82 /* Flag allocation requirements to shmem_getpage */
83 enum sgp_type {
84         SGP_READ,       /* don't exceed i_size, don't allocate page */
85         SGP_CACHE,      /* don't exceed i_size, may allocate page */
86         SGP_DIRTY,      /* like SGP_CACHE, but set new page dirty */
87         SGP_WRITE,      /* may exceed i_size, may allocate page */
88 };
89
90 #ifdef CONFIG_TMPFS
91 static unsigned long shmem_default_max_blocks(void)
92 {
93         return totalram_pages / 2;
94 }
95
96 static unsigned long shmem_default_max_inodes(void)
97 {
98         return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
99 }
100 #endif
101
102 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
103         struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
104
105 static inline int shmem_getpage(struct inode *inode, pgoff_t index,
106         struct page **pagep, enum sgp_type sgp, int *fault_type)
107 {
108         return shmem_getpage_gfp(inode, index, pagep, sgp,
109                         mapping_gfp_mask(inode->i_mapping), fault_type);
110 }
111
112 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
113 {
114         return sb->s_fs_info;
115 }
116
117 /*
118  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
119  * for shared memory and for shared anonymous (/dev/zero) mappings
120  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
121  * consistent with the pre-accounting of private mappings ...
122  */
123 static inline int shmem_acct_size(unsigned long flags, loff_t size)
124 {
125         return (flags & VM_NORESERVE) ?
126                 0 : security_vm_enough_memory_kern(VM_ACCT(size));
127 }
128
129 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
130 {
131         if (!(flags & VM_NORESERVE))
132                 vm_unacct_memory(VM_ACCT(size));
133 }
134
135 /*
136  * ... whereas tmpfs objects are accounted incrementally as
137  * pages are allocated, in order to allow huge sparse files.
138  * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
139  * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
140  */
141 static inline int shmem_acct_block(unsigned long flags)
142 {
143         return (flags & VM_NORESERVE) ?
144                 security_vm_enough_memory_kern(VM_ACCT(PAGE_CACHE_SIZE)) : 0;
145 }
146
147 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
148 {
149         if (flags & VM_NORESERVE)
150                 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
151 }
152
153 static const struct super_operations shmem_ops;
154 static const struct address_space_operations shmem_aops;
155 static const struct file_operations shmem_file_operations;
156 static const struct inode_operations shmem_inode_operations;
157 static const struct inode_operations shmem_dir_inode_operations;
158 static const struct inode_operations shmem_special_inode_operations;
159 static const struct vm_operations_struct shmem_vm_ops;
160
161 static struct backing_dev_info shmem_backing_dev_info  __read_mostly = {
162         .ra_pages       = 0,    /* No readahead */
163         .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
164 };
165
166 static LIST_HEAD(shmem_swaplist);
167 static DEFINE_MUTEX(shmem_swaplist_mutex);
168
169 static void shmem_free_blocks(struct inode *inode, long pages)
170 {
171         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
172         if (sbinfo->max_blocks) {
173                 percpu_counter_add(&sbinfo->used_blocks, -pages);
174                 inode->i_blocks -= pages*BLOCKS_PER_PAGE;
175         }
176 }
177
178 static int shmem_reserve_inode(struct super_block *sb)
179 {
180         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
181         if (sbinfo->max_inodes) {
182                 spin_lock(&sbinfo->stat_lock);
183                 if (!sbinfo->free_inodes) {
184                         spin_unlock(&sbinfo->stat_lock);
185                         return -ENOSPC;
186                 }
187                 sbinfo->free_inodes--;
188                 spin_unlock(&sbinfo->stat_lock);
189         }
190         return 0;
191 }
192
193 static void shmem_free_inode(struct super_block *sb)
194 {
195         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
196         if (sbinfo->max_inodes) {
197                 spin_lock(&sbinfo->stat_lock);
198                 sbinfo->free_inodes++;
199                 spin_unlock(&sbinfo->stat_lock);
200         }
201 }
202
203 /**
204  * shmem_recalc_inode - recalculate the size of an inode
205  * @inode: inode to recalc
206  *
207  * We have to calculate the free blocks since the mm can drop
208  * undirtied hole pages behind our back.
209  *
210  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
211  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
212  *
213  * It has to be called with the spinlock held.
214  */
215 static void shmem_recalc_inode(struct inode *inode)
216 {
217         struct shmem_inode_info *info = SHMEM_I(inode);
218         long freed;
219
220         freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
221         if (freed > 0) {
222                 info->alloced -= freed;
223                 shmem_unacct_blocks(info->flags, freed);
224                 shmem_free_blocks(inode, freed);
225         }
226 }
227
228 static void shmem_put_swap(struct shmem_inode_info *info, pgoff_t index,
229                            swp_entry_t swap)
230 {
231         if (index < SHMEM_NR_DIRECT)
232                 info->i_direct[index] = swap;
233 }
234
235 static swp_entry_t shmem_get_swap(struct shmem_inode_info *info, pgoff_t index)
236 {
237         return (index < SHMEM_NR_DIRECT) ?
238                 info->i_direct[index] : (swp_entry_t){0};
239 }
240
241 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
242 {
243         struct address_space *mapping = inode->i_mapping;
244         struct shmem_inode_info *info = SHMEM_I(inode);
245         pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
246         pgoff_t end = (lend >> PAGE_CACHE_SHIFT);
247         pgoff_t index;
248         swp_entry_t swap;
249
250         truncate_inode_pages_range(mapping, lstart, lend);
251
252         if (end > SHMEM_NR_DIRECT)
253                 end = SHMEM_NR_DIRECT;
254
255         spin_lock(&info->lock);
256         for (index = start; index < end; index++) {
257                 swap = shmem_get_swap(info, index);
258                 if (swap.val) {
259                         free_swap_and_cache(swap);
260                         shmem_put_swap(info, index, (swp_entry_t){0});
261                         info->swapped--;
262                 }
263         }
264
265         if (mapping->nrpages) {
266                 spin_unlock(&info->lock);
267                 /*
268                  * A page may have meanwhile sneaked in from swap.
269                  */
270                 truncate_inode_pages_range(mapping, lstart, lend);
271                 spin_lock(&info->lock);
272         }
273
274         shmem_recalc_inode(inode);
275         spin_unlock(&info->lock);
276
277         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
278 }
279 EXPORT_SYMBOL_GPL(shmem_truncate_range);
280
281 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
282 {
283         struct inode *inode = dentry->d_inode;
284         int error;
285
286         error = inode_change_ok(inode, attr);
287         if (error)
288                 return error;
289
290         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
291                 loff_t oldsize = inode->i_size;
292                 loff_t newsize = attr->ia_size;
293                 struct page *page = NULL;
294
295                 if (newsize < oldsize) {
296                         /*
297                          * If truncating down to a partial page, then
298                          * if that page is already allocated, hold it
299                          * in memory until the truncation is over, so
300                          * truncate_partial_page cannot miss it were
301                          * it assigned to swap.
302                          */
303                         if (newsize & (PAGE_CACHE_SIZE-1)) {
304                                 (void) shmem_getpage(inode,
305                                         newsize >> PAGE_CACHE_SHIFT,
306                                                 &page, SGP_READ, NULL);
307                                 if (page)
308                                         unlock_page(page);
309                         }
310                 }
311                 if (newsize != oldsize) {
312                         i_size_write(inode, newsize);
313                         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
314                 }
315                 if (newsize < oldsize) {
316                         loff_t holebegin = round_up(newsize, PAGE_SIZE);
317                         unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
318                         shmem_truncate_range(inode, newsize, (loff_t)-1);
319                         /* unmap again to remove racily COWed private pages */
320                         unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
321                 }
322                 if (page)
323                         page_cache_release(page);
324         }
325
326         setattr_copy(inode, attr);
327 #ifdef CONFIG_TMPFS_POSIX_ACL
328         if (attr->ia_valid & ATTR_MODE)
329                 error = generic_acl_chmod(inode);
330 #endif
331         return error;
332 }
333
334 static void shmem_evict_inode(struct inode *inode)
335 {
336         struct shmem_inode_info *info = SHMEM_I(inode);
337         struct shmem_xattr *xattr, *nxattr;
338
339         if (inode->i_mapping->a_ops == &shmem_aops) {
340                 shmem_unacct_size(info->flags, inode->i_size);
341                 inode->i_size = 0;
342                 shmem_truncate_range(inode, 0, (loff_t)-1);
343                 if (!list_empty(&info->swaplist)) {
344                         mutex_lock(&shmem_swaplist_mutex);
345                         list_del_init(&info->swaplist);
346                         mutex_unlock(&shmem_swaplist_mutex);
347                 }
348         }
349
350         list_for_each_entry_safe(xattr, nxattr, &info->xattr_list, list) {
351                 kfree(xattr->name);
352                 kfree(xattr);
353         }
354         BUG_ON(inode->i_blocks);
355         shmem_free_inode(inode->i_sb);
356         end_writeback(inode);
357 }
358
359 static int shmem_unuse_inode(struct shmem_inode_info *info, swp_entry_t entry, struct page *page)
360 {
361         struct address_space *mapping = info->vfs_inode.i_mapping;
362         unsigned long idx;
363         int error;
364
365         for (idx = 0; idx < SHMEM_NR_DIRECT; idx++)
366                 if (shmem_get_swap(info, idx).val == entry.val)
367                         goto found;
368         return 0;
369 found:
370         spin_lock(&info->lock);
371         if (shmem_get_swap(info, idx).val != entry.val) {
372                 spin_unlock(&info->lock);
373                 return 0;
374         }
375
376         /*
377          * Move _head_ to start search for next from here.
378          * But be careful: shmem_evict_inode checks list_empty without taking
379          * mutex, and there's an instant in list_move_tail when info->swaplist
380          * would appear empty, if it were the only one on shmem_swaplist.
381          */
382         if (shmem_swaplist.next != &info->swaplist)
383                 list_move_tail(&shmem_swaplist, &info->swaplist);
384
385         /*
386          * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
387          * but also to hold up shmem_evict_inode(): so inode cannot be freed
388          * beneath us (pagelock doesn't help until the page is in pagecache).
389          */
390         error = add_to_page_cache_locked(page, mapping, idx, GFP_NOWAIT);
391         /* which does mem_cgroup_uncharge_cache_page on error */
392
393         if (error != -ENOMEM) {
394                 delete_from_swap_cache(page);
395                 set_page_dirty(page);
396                 shmem_put_swap(info, idx, (swp_entry_t){0});
397                 info->swapped--;
398                 swap_free(entry);
399                 error = 1;      /* not an error, but entry was found */
400         }
401         spin_unlock(&info->lock);
402         return error;
403 }
404
405 /*
406  * shmem_unuse() search for an eventually swapped out shmem page.
407  */
408 int shmem_unuse(swp_entry_t entry, struct page *page)
409 {
410         struct list_head *p, *next;
411         struct shmem_inode_info *info;
412         int found = 0;
413         int error;
414
415         /*
416          * Charge page using GFP_KERNEL while we can wait, before taking
417          * the shmem_swaplist_mutex which might hold up shmem_writepage().
418          * Charged back to the user (not to caller) when swap account is used.
419          * add_to_page_cache() will be called with GFP_NOWAIT.
420          */
421         error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
422         if (error)
423                 goto out;
424         /*
425          * Try to preload while we can wait, to not make a habit of
426          * draining atomic reserves; but don't latch on to this cpu,
427          * it's okay if sometimes we get rescheduled after this.
428          */
429         error = radix_tree_preload(GFP_KERNEL);
430         if (error)
431                 goto uncharge;
432         radix_tree_preload_end();
433
434         mutex_lock(&shmem_swaplist_mutex);
435         list_for_each_safe(p, next, &shmem_swaplist) {
436                 info = list_entry(p, struct shmem_inode_info, swaplist);
437                 if (!info->swapped) {
438                         spin_lock(&info->lock);
439                         if (!info->swapped)
440                                 list_del_init(&info->swaplist);
441                         spin_unlock(&info->lock);
442                 }
443                 if (info->swapped)
444                         found = shmem_unuse_inode(info, entry, page);
445                 cond_resched();
446                 if (found)
447                         break;
448         }
449         mutex_unlock(&shmem_swaplist_mutex);
450
451 uncharge:
452         if (!found)
453                 mem_cgroup_uncharge_cache_page(page);
454         if (found < 0)
455                 error = found;
456 out:
457         unlock_page(page);
458         page_cache_release(page);
459         return error;
460 }
461
462 /*
463  * Move the page from the page cache to the swap cache.
464  */
465 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
466 {
467         struct shmem_inode_info *info;
468         swp_entry_t swap, oswap;
469         struct address_space *mapping;
470         unsigned long index;
471         struct inode *inode;
472
473         BUG_ON(!PageLocked(page));
474         mapping = page->mapping;
475         index = page->index;
476         inode = mapping->host;
477         info = SHMEM_I(inode);
478         if (info->flags & VM_LOCKED)
479                 goto redirty;
480         if (!total_swap_pages)
481                 goto redirty;
482
483         /*
484          * shmem_backing_dev_info's capabilities prevent regular writeback or
485          * sync from ever calling shmem_writepage; but a stacking filesystem
486          * might use ->writepage of its underlying filesystem, in which case
487          * tmpfs should write out to swap only in response to memory pressure,
488          * and not for the writeback threads or sync.
489          */
490         if (!wbc->for_reclaim) {
491                 WARN_ON_ONCE(1);        /* Still happens? Tell us about it! */
492                 goto redirty;
493         }
494
495         /*
496          * Just for this patch, we have a toy implementation,
497          * which can swap out only the first SHMEM_NR_DIRECT pages:
498          * for simple demonstration of where we need to think about swap.
499          */
500         if (index >= SHMEM_NR_DIRECT)
501                 goto redirty;
502
503         swap = get_swap_page();
504         if (!swap.val)
505                 goto redirty;
506
507         /*
508          * Add inode to shmem_unuse()'s list of swapped-out inodes,
509          * if it's not already there.  Do it now because we cannot take
510          * mutex while holding spinlock, and must do so before the page
511          * is moved to swap cache, when its pagelock no longer protects
512          * the inode from eviction.  But don't unlock the mutex until
513          * we've taken the spinlock, because shmem_unuse_inode() will
514          * prune a !swapped inode from the swaplist under both locks.
515          */
516         mutex_lock(&shmem_swaplist_mutex);
517         if (list_empty(&info->swaplist))
518                 list_add_tail(&info->swaplist, &shmem_swaplist);
519
520         spin_lock(&info->lock);
521         mutex_unlock(&shmem_swaplist_mutex);
522
523         oswap = shmem_get_swap(info, index);
524         if (oswap.val) {
525                 WARN_ON_ONCE(1);        /* Still happens? Tell us about it! */
526                 free_swap_and_cache(oswap);
527                 shmem_put_swap(info, index, (swp_entry_t){0});
528                 info->swapped--;
529         }
530         shmem_recalc_inode(inode);
531
532         if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
533                 delete_from_page_cache(page);
534                 shmem_put_swap(info, index, swap);
535                 info->swapped++;
536                 swap_shmem_alloc(swap);
537                 spin_unlock(&info->lock);
538                 BUG_ON(page_mapped(page));
539                 swap_writepage(page, wbc);
540                 return 0;
541         }
542
543         spin_unlock(&info->lock);
544         swapcache_free(swap, NULL);
545 redirty:
546         set_page_dirty(page);
547         if (wbc->for_reclaim)
548                 return AOP_WRITEPAGE_ACTIVATE;  /* Return with page locked */
549         unlock_page(page);
550         return 0;
551 }
552
553 #ifdef CONFIG_NUMA
554 #ifdef CONFIG_TMPFS
555 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
556 {
557         char buffer[64];
558
559         if (!mpol || mpol->mode == MPOL_DEFAULT)
560                 return;         /* show nothing */
561
562         mpol_to_str(buffer, sizeof(buffer), mpol, 1);
563
564         seq_printf(seq, ",mpol=%s", buffer);
565 }
566
567 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
568 {
569         struct mempolicy *mpol = NULL;
570         if (sbinfo->mpol) {
571                 spin_lock(&sbinfo->stat_lock);  /* prevent replace/use races */
572                 mpol = sbinfo->mpol;
573                 mpol_get(mpol);
574                 spin_unlock(&sbinfo->stat_lock);
575         }
576         return mpol;
577 }
578 #endif /* CONFIG_TMPFS */
579
580 static struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
581                         struct shmem_inode_info *info, unsigned long idx)
582 {
583         struct mempolicy mpol, *spol;
584         struct vm_area_struct pvma;
585         struct page *page;
586
587         spol = mpol_cond_copy(&mpol,
588                                 mpol_shared_policy_lookup(&info->policy, idx));
589
590         /* Create a pseudo vma that just contains the policy */
591         pvma.vm_start = 0;
592         pvma.vm_pgoff = idx;
593         pvma.vm_ops = NULL;
594         pvma.vm_policy = spol;
595         page = swapin_readahead(entry, gfp, &pvma, 0);
596         return page;
597 }
598
599 static struct page *shmem_alloc_page(gfp_t gfp,
600                         struct shmem_inode_info *info, unsigned long idx)
601 {
602         struct vm_area_struct pvma;
603
604         /* Create a pseudo vma that just contains the policy */
605         pvma.vm_start = 0;
606         pvma.vm_pgoff = idx;
607         pvma.vm_ops = NULL;
608         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, idx);
609
610         /*
611          * alloc_page_vma() will drop the shared policy reference
612          */
613         return alloc_page_vma(gfp, &pvma, 0);
614 }
615 #else /* !CONFIG_NUMA */
616 #ifdef CONFIG_TMPFS
617 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *p)
618 {
619 }
620 #endif /* CONFIG_TMPFS */
621
622 static inline struct page *shmem_swapin(swp_entry_t entry, gfp_t gfp,
623                         struct shmem_inode_info *info, unsigned long idx)
624 {
625         return swapin_readahead(entry, gfp, NULL, 0);
626 }
627
628 static inline struct page *shmem_alloc_page(gfp_t gfp,
629                         struct shmem_inode_info *info, unsigned long idx)
630 {
631         return alloc_page(gfp);
632 }
633 #endif /* CONFIG_NUMA */
634
635 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
636 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
637 {
638         return NULL;
639 }
640 #endif
641
642 /*
643  * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
644  *
645  * If we allocate a new one we do not mark it dirty. That's up to the
646  * vm. If we swap it in we mark it dirty since we also free the swap
647  * entry since a page cannot live in both the swap and page cache
648  */
649 static int shmem_getpage_gfp(struct inode *inode, pgoff_t idx,
650         struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
651 {
652         struct address_space *mapping = inode->i_mapping;
653         struct shmem_inode_info *info = SHMEM_I(inode);
654         struct shmem_sb_info *sbinfo;
655         struct page *page;
656         struct page *prealloc_page = NULL;
657         swp_entry_t swap;
658         int error;
659
660         if (idx > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
661                 return -EFBIG;
662 repeat:
663         page = find_lock_page(mapping, idx);
664         if (page) {
665                 /*
666                  * Once we can get the page lock, it must be uptodate:
667                  * if there were an error in reading back from swap,
668                  * the page would not be inserted into the filecache.
669                  */
670                 BUG_ON(!PageUptodate(page));
671                 goto done;
672         }
673
674         /*
675          * Try to preload while we can wait, to not make a habit of
676          * draining atomic reserves; but don't latch on to this cpu.
677          */
678         error = radix_tree_preload(gfp & GFP_RECLAIM_MASK);
679         if (error)
680                 goto out;
681         radix_tree_preload_end();
682
683         if (sgp != SGP_READ && !prealloc_page) {
684                 prealloc_page = shmem_alloc_page(gfp, info, idx);
685                 if (prealloc_page) {
686                         SetPageSwapBacked(prealloc_page);
687                         if (mem_cgroup_cache_charge(prealloc_page,
688                                         current->mm, GFP_KERNEL)) {
689                                 page_cache_release(prealloc_page);
690                                 prealloc_page = NULL;
691                         }
692                 }
693         }
694
695         spin_lock(&info->lock);
696         shmem_recalc_inode(inode);
697         swap = shmem_get_swap(info, idx);
698         if (swap.val) {
699                 /* Look it up and read it in.. */
700                 page = lookup_swap_cache(swap);
701                 if (!page) {
702                         spin_unlock(&info->lock);
703                         /* here we actually do the io */
704                         if (fault_type)
705                                 *fault_type |= VM_FAULT_MAJOR;
706                         page = shmem_swapin(swap, gfp, info, idx);
707                         if (!page) {
708                                 swp_entry_t nswap = shmem_get_swap(info, idx);
709                                 if (nswap.val == swap.val) {
710                                         error = -ENOMEM;
711                                         goto out;
712                                 }
713                                 goto repeat;
714                         }
715                         wait_on_page_locked(page);
716                         page_cache_release(page);
717                         goto repeat;
718                 }
719
720                 /* We have to do this with page locked to prevent races */
721                 if (!trylock_page(page)) {
722                         spin_unlock(&info->lock);
723                         wait_on_page_locked(page);
724                         page_cache_release(page);
725                         goto repeat;
726                 }
727                 if (PageWriteback(page)) {
728                         spin_unlock(&info->lock);
729                         wait_on_page_writeback(page);
730                         unlock_page(page);
731                         page_cache_release(page);
732                         goto repeat;
733                 }
734                 if (!PageUptodate(page)) {
735                         spin_unlock(&info->lock);
736                         unlock_page(page);
737                         page_cache_release(page);
738                         error = -EIO;
739                         goto out;
740                 }
741
742                 error = add_to_page_cache_locked(page, mapping,
743                                                  idx, GFP_NOWAIT);
744                 if (error) {
745                         spin_unlock(&info->lock);
746                         if (error == -ENOMEM) {
747                                 /*
748                                  * reclaim from proper memory cgroup and
749                                  * call memcg's OOM if needed.
750                                  */
751                                 error = mem_cgroup_shmem_charge_fallback(
752                                                 page, current->mm, gfp);
753                                 if (error) {
754                                         unlock_page(page);
755                                         page_cache_release(page);
756                                         goto out;
757                                 }
758                         }
759                         unlock_page(page);
760                         page_cache_release(page);
761                         goto repeat;
762                 }
763
764                 delete_from_swap_cache(page);
765                 shmem_put_swap(info, idx, (swp_entry_t){0});
766                 info->swapped--;
767                 spin_unlock(&info->lock);
768                 set_page_dirty(page);
769                 swap_free(swap);
770
771         } else if (sgp == SGP_READ) {
772                 page = find_get_page(mapping, idx);
773                 if (page && !trylock_page(page)) {
774                         spin_unlock(&info->lock);
775                         wait_on_page_locked(page);
776                         page_cache_release(page);
777                         goto repeat;
778                 }
779                 spin_unlock(&info->lock);
780
781         } else if (prealloc_page) {
782                 sbinfo = SHMEM_SB(inode->i_sb);
783                 if (sbinfo->max_blocks) {
784                         if (percpu_counter_compare(&sbinfo->used_blocks,
785                                                 sbinfo->max_blocks) >= 0 ||
786                             shmem_acct_block(info->flags))
787                                 goto nospace;
788                         percpu_counter_inc(&sbinfo->used_blocks);
789                         inode->i_blocks += BLOCKS_PER_PAGE;
790                 } else if (shmem_acct_block(info->flags))
791                         goto nospace;
792
793                 page = prealloc_page;
794                 prealloc_page = NULL;
795
796                 swap = shmem_get_swap(info, idx);
797                 if (swap.val)
798                         mem_cgroup_uncharge_cache_page(page);
799                 else
800                         error = add_to_page_cache_lru(page, mapping,
801                                                 idx, GFP_NOWAIT);
802                 /*
803                  * At add_to_page_cache_lru() failure,
804                  * uncharge will be done automatically.
805                  */
806                 if (swap.val || error) {
807                         shmem_unacct_blocks(info->flags, 1);
808                         shmem_free_blocks(inode, 1);
809                         spin_unlock(&info->lock);
810                         page_cache_release(page);
811                         goto repeat;
812                 }
813
814                 info->alloced++;
815                 spin_unlock(&info->lock);
816                 clear_highpage(page);
817                 flush_dcache_page(page);
818                 SetPageUptodate(page);
819                 if (sgp == SGP_DIRTY)
820                         set_page_dirty(page);
821
822         } else {
823                 spin_unlock(&info->lock);
824                 error = -ENOMEM;
825                 goto out;
826         }
827 done:
828         *pagep = page;
829         error = 0;
830 out:
831         if (prealloc_page) {
832                 mem_cgroup_uncharge_cache_page(prealloc_page);
833                 page_cache_release(prealloc_page);
834         }
835         return error;
836
837 nospace:
838         /*
839          * Perhaps the page was brought in from swap between find_lock_page
840          * and taking info->lock?  We allow for that at add_to_page_cache_lru,
841          * but must also avoid reporting a spurious ENOSPC while working on a
842          * full tmpfs.
843          */
844         page = find_get_page(mapping, idx);
845         spin_unlock(&info->lock);
846         if (page) {
847                 page_cache_release(page);
848                 goto repeat;
849         }
850         error = -ENOSPC;
851         goto out;
852 }
853
854 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
855 {
856         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
857         int error;
858         int ret = VM_FAULT_LOCKED;
859
860         if (((loff_t)vmf->pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
861                 return VM_FAULT_SIGBUS;
862
863         error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
864         if (error)
865                 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
866
867         if (ret & VM_FAULT_MAJOR) {
868                 count_vm_event(PGMAJFAULT);
869                 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
870         }
871         return ret;
872 }
873
874 #ifdef CONFIG_NUMA
875 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *new)
876 {
877         struct inode *i = vma->vm_file->f_path.dentry->d_inode;
878         return mpol_set_shared_policy(&SHMEM_I(i)->policy, vma, new);
879 }
880
881 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
882                                           unsigned long addr)
883 {
884         struct inode *i = vma->vm_file->f_path.dentry->d_inode;
885         unsigned long idx;
886
887         idx = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
888         return mpol_shared_policy_lookup(&SHMEM_I(i)->policy, idx);
889 }
890 #endif
891
892 int shmem_lock(struct file *file, int lock, struct user_struct *user)
893 {
894         struct inode *inode = file->f_path.dentry->d_inode;
895         struct shmem_inode_info *info = SHMEM_I(inode);
896         int retval = -ENOMEM;
897
898         spin_lock(&info->lock);
899         if (lock && !(info->flags & VM_LOCKED)) {
900                 if (!user_shm_lock(inode->i_size, user))
901                         goto out_nomem;
902                 info->flags |= VM_LOCKED;
903                 mapping_set_unevictable(file->f_mapping);
904         }
905         if (!lock && (info->flags & VM_LOCKED) && user) {
906                 user_shm_unlock(inode->i_size, user);
907                 info->flags &= ~VM_LOCKED;
908                 mapping_clear_unevictable(file->f_mapping);
909                 scan_mapping_unevictable_pages(file->f_mapping);
910         }
911         retval = 0;
912
913 out_nomem:
914         spin_unlock(&info->lock);
915         return retval;
916 }
917
918 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
919 {
920         file_accessed(file);
921         vma->vm_ops = &shmem_vm_ops;
922         vma->vm_flags |= VM_CAN_NONLINEAR;
923         return 0;
924 }
925
926 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
927                                      int mode, dev_t dev, unsigned long flags)
928 {
929         struct inode *inode;
930         struct shmem_inode_info *info;
931         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
932
933         if (shmem_reserve_inode(sb))
934                 return NULL;
935
936         inode = new_inode(sb);
937         if (inode) {
938                 inode->i_ino = get_next_ino();
939                 inode_init_owner(inode, dir, mode);
940                 inode->i_blocks = 0;
941                 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
942                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
943                 inode->i_generation = get_seconds();
944                 info = SHMEM_I(inode);
945                 memset(info, 0, (char *)inode - (char *)info);
946                 spin_lock_init(&info->lock);
947                 info->flags = flags & VM_NORESERVE;
948                 INIT_LIST_HEAD(&info->swaplist);
949                 INIT_LIST_HEAD(&info->xattr_list);
950                 cache_no_acl(inode);
951
952                 switch (mode & S_IFMT) {
953                 default:
954                         inode->i_op = &shmem_special_inode_operations;
955                         init_special_inode(inode, mode, dev);
956                         break;
957                 case S_IFREG:
958                         inode->i_mapping->a_ops = &shmem_aops;
959                         inode->i_op = &shmem_inode_operations;
960                         inode->i_fop = &shmem_file_operations;
961                         mpol_shared_policy_init(&info->policy,
962                                                  shmem_get_sbmpol(sbinfo));
963                         break;
964                 case S_IFDIR:
965                         inc_nlink(inode);
966                         /* Some things misbehave if size == 0 on a directory */
967                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
968                         inode->i_op = &shmem_dir_inode_operations;
969                         inode->i_fop = &simple_dir_operations;
970                         break;
971                 case S_IFLNK:
972                         /*
973                          * Must not load anything in the rbtree,
974                          * mpol_free_shared_policy will not be called.
975                          */
976                         mpol_shared_policy_init(&info->policy, NULL);
977                         break;
978                 }
979         } else
980                 shmem_free_inode(sb);
981         return inode;
982 }
983
984 #ifdef CONFIG_TMPFS
985 static const struct inode_operations shmem_symlink_inode_operations;
986 static const struct inode_operations shmem_symlink_inline_operations;
987
988 static int
989 shmem_write_begin(struct file *file, struct address_space *mapping,
990                         loff_t pos, unsigned len, unsigned flags,
991                         struct page **pagep, void **fsdata)
992 {
993         struct inode *inode = mapping->host;
994         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
995         return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
996 }
997
998 static int
999 shmem_write_end(struct file *file, struct address_space *mapping,
1000                         loff_t pos, unsigned len, unsigned copied,
1001                         struct page *page, void *fsdata)
1002 {
1003         struct inode *inode = mapping->host;
1004
1005         if (pos + copied > inode->i_size)
1006                 i_size_write(inode, pos + copied);
1007
1008         set_page_dirty(page);
1009         unlock_page(page);
1010         page_cache_release(page);
1011
1012         return copied;
1013 }
1014
1015 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1016 {
1017         struct inode *inode = filp->f_path.dentry->d_inode;
1018         struct address_space *mapping = inode->i_mapping;
1019         unsigned long index, offset;
1020         enum sgp_type sgp = SGP_READ;
1021
1022         /*
1023          * Might this read be for a stacking filesystem?  Then when reading
1024          * holes of a sparse file, we actually need to allocate those pages,
1025          * and even mark them dirty, so it cannot exceed the max_blocks limit.
1026          */
1027         if (segment_eq(get_fs(), KERNEL_DS))
1028                 sgp = SGP_DIRTY;
1029
1030         index = *ppos >> PAGE_CACHE_SHIFT;
1031         offset = *ppos & ~PAGE_CACHE_MASK;
1032
1033         for (;;) {
1034                 struct page *page = NULL;
1035                 unsigned long end_index, nr, ret;
1036                 loff_t i_size = i_size_read(inode);
1037
1038                 end_index = i_size >> PAGE_CACHE_SHIFT;
1039                 if (index > end_index)
1040                         break;
1041                 if (index == end_index) {
1042                         nr = i_size & ~PAGE_CACHE_MASK;
1043                         if (nr <= offset)
1044                                 break;
1045                 }
1046
1047                 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1048                 if (desc->error) {
1049                         if (desc->error == -EINVAL)
1050                                 desc->error = 0;
1051                         break;
1052                 }
1053                 if (page)
1054                         unlock_page(page);
1055
1056                 /*
1057                  * We must evaluate after, since reads (unlike writes)
1058                  * are called without i_mutex protection against truncate
1059                  */
1060                 nr = PAGE_CACHE_SIZE;
1061                 i_size = i_size_read(inode);
1062                 end_index = i_size >> PAGE_CACHE_SHIFT;
1063                 if (index == end_index) {
1064                         nr = i_size & ~PAGE_CACHE_MASK;
1065                         if (nr <= offset) {
1066                                 if (page)
1067                                         page_cache_release(page);
1068                                 break;
1069                         }
1070                 }
1071                 nr -= offset;
1072
1073                 if (page) {
1074                         /*
1075                          * If users can be writing to this page using arbitrary
1076                          * virtual addresses, take care about potential aliasing
1077                          * before reading the page on the kernel side.
1078                          */
1079                         if (mapping_writably_mapped(mapping))
1080                                 flush_dcache_page(page);
1081                         /*
1082                          * Mark the page accessed if we read the beginning.
1083                          */
1084                         if (!offset)
1085                                 mark_page_accessed(page);
1086                 } else {
1087                         page = ZERO_PAGE(0);
1088                         page_cache_get(page);
1089                 }
1090
1091                 /*
1092                  * Ok, we have the page, and it's up-to-date, so
1093                  * now we can copy it to user space...
1094                  *
1095                  * The actor routine returns how many bytes were actually used..
1096                  * NOTE! This may not be the same as how much of a user buffer
1097                  * we filled up (we may be padding etc), so we can only update
1098                  * "pos" here (the actor routine has to update the user buffer
1099                  * pointers and the remaining count).
1100                  */
1101                 ret = actor(desc, page, offset, nr);
1102                 offset += ret;
1103                 index += offset >> PAGE_CACHE_SHIFT;
1104                 offset &= ~PAGE_CACHE_MASK;
1105
1106                 page_cache_release(page);
1107                 if (ret != nr || !desc->count)
1108                         break;
1109
1110                 cond_resched();
1111         }
1112
1113         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1114         file_accessed(filp);
1115 }
1116
1117 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1118                 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1119 {
1120         struct file *filp = iocb->ki_filp;
1121         ssize_t retval;
1122         unsigned long seg;
1123         size_t count;
1124         loff_t *ppos = &iocb->ki_pos;
1125
1126         retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1127         if (retval)
1128                 return retval;
1129
1130         for (seg = 0; seg < nr_segs; seg++) {
1131                 read_descriptor_t desc;
1132
1133                 desc.written = 0;
1134                 desc.arg.buf = iov[seg].iov_base;
1135                 desc.count = iov[seg].iov_len;
1136                 if (desc.count == 0)
1137                         continue;
1138                 desc.error = 0;
1139                 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1140                 retval += desc.written;
1141                 if (desc.error) {
1142                         retval = retval ?: desc.error;
1143                         break;
1144                 }
1145                 if (desc.count > 0)
1146                         break;
1147         }
1148         return retval;
1149 }
1150
1151 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1152                                 struct pipe_inode_info *pipe, size_t len,
1153                                 unsigned int flags)
1154 {
1155         struct address_space *mapping = in->f_mapping;
1156         struct inode *inode = mapping->host;
1157         unsigned int loff, nr_pages, req_pages;
1158         struct page *pages[PIPE_DEF_BUFFERS];
1159         struct partial_page partial[PIPE_DEF_BUFFERS];
1160         struct page *page;
1161         pgoff_t index, end_index;
1162         loff_t isize, left;
1163         int error, page_nr;
1164         struct splice_pipe_desc spd = {
1165                 .pages = pages,
1166                 .partial = partial,
1167                 .flags = flags,
1168                 .ops = &page_cache_pipe_buf_ops,
1169                 .spd_release = spd_release_page,
1170         };
1171
1172         isize = i_size_read(inode);
1173         if (unlikely(*ppos >= isize))
1174                 return 0;
1175
1176         left = isize - *ppos;
1177         if (unlikely(left < len))
1178                 len = left;
1179
1180         if (splice_grow_spd(pipe, &spd))
1181                 return -ENOMEM;
1182
1183         index = *ppos >> PAGE_CACHE_SHIFT;
1184         loff = *ppos & ~PAGE_CACHE_MASK;
1185         req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1186         nr_pages = min(req_pages, pipe->buffers);
1187
1188         spd.nr_pages = find_get_pages_contig(mapping, index,
1189                                                 nr_pages, spd.pages);
1190         index += spd.nr_pages;
1191         error = 0;
1192
1193         while (spd.nr_pages < nr_pages) {
1194                 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1195                 if (error)
1196                         break;
1197                 unlock_page(page);
1198                 spd.pages[spd.nr_pages++] = page;
1199                 index++;
1200         }
1201
1202         index = *ppos >> PAGE_CACHE_SHIFT;
1203         nr_pages = spd.nr_pages;
1204         spd.nr_pages = 0;
1205
1206         for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1207                 unsigned int this_len;
1208
1209                 if (!len)
1210                         break;
1211
1212                 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1213                 page = spd.pages[page_nr];
1214
1215                 if (!PageUptodate(page) || page->mapping != mapping) {
1216                         error = shmem_getpage(inode, index, &page,
1217                                                         SGP_CACHE, NULL);
1218                         if (error)
1219                                 break;
1220                         unlock_page(page);
1221                         page_cache_release(spd.pages[page_nr]);
1222                         spd.pages[page_nr] = page;
1223                 }
1224
1225                 isize = i_size_read(inode);
1226                 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1227                 if (unlikely(!isize || index > end_index))
1228                         break;
1229
1230                 if (end_index == index) {
1231                         unsigned int plen;
1232
1233                         plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1234                         if (plen <= loff)
1235                                 break;
1236
1237                         this_len = min(this_len, plen - loff);
1238                         len = this_len;
1239                 }
1240
1241                 spd.partial[page_nr].offset = loff;
1242                 spd.partial[page_nr].len = this_len;
1243                 len -= this_len;
1244                 loff = 0;
1245                 spd.nr_pages++;
1246                 index++;
1247         }
1248
1249         while (page_nr < nr_pages)
1250                 page_cache_release(spd.pages[page_nr++]);
1251
1252         if (spd.nr_pages)
1253                 error = splice_to_pipe(pipe, &spd);
1254
1255         splice_shrink_spd(pipe, &spd);
1256
1257         if (error > 0) {
1258                 *ppos += error;
1259                 file_accessed(in);
1260         }
1261         return error;
1262 }
1263
1264 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1265 {
1266         struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1267
1268         buf->f_type = TMPFS_MAGIC;
1269         buf->f_bsize = PAGE_CACHE_SIZE;
1270         buf->f_namelen = NAME_MAX;
1271         if (sbinfo->max_blocks) {
1272                 buf->f_blocks = sbinfo->max_blocks;
1273                 buf->f_bavail = buf->f_bfree =
1274                                 sbinfo->max_blocks - percpu_counter_sum(&sbinfo->used_blocks);
1275         }
1276         if (sbinfo->max_inodes) {
1277                 buf->f_files = sbinfo->max_inodes;
1278                 buf->f_ffree = sbinfo->free_inodes;
1279         }
1280         /* else leave those fields 0 like simple_statfs */
1281         return 0;
1282 }
1283
1284 /*
1285  * File creation. Allocate an inode, and we're done..
1286  */
1287 static int
1288 shmem_mknod(struct inode *dir, struct dentry *dentry, int mode, dev_t dev)
1289 {
1290         struct inode *inode;
1291         int error = -ENOSPC;
1292
1293         inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1294         if (inode) {
1295                 error = security_inode_init_security(inode, dir,
1296                                                      &dentry->d_name, NULL,
1297                                                      NULL, NULL);
1298                 if (error) {
1299                         if (error != -EOPNOTSUPP) {
1300                                 iput(inode);
1301                                 return error;
1302                         }
1303                 }
1304 #ifdef CONFIG_TMPFS_POSIX_ACL
1305                 error = generic_acl_init(inode, dir);
1306                 if (error) {
1307                         iput(inode);
1308                         return error;
1309                 }
1310 #else
1311                 error = 0;
1312 #endif
1313                 dir->i_size += BOGO_DIRENT_SIZE;
1314                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1315                 d_instantiate(dentry, inode);
1316                 dget(dentry); /* Extra count - pin the dentry in core */
1317         }
1318         return error;
1319 }
1320
1321 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, int mode)
1322 {
1323         int error;
1324
1325         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1326                 return error;
1327         inc_nlink(dir);
1328         return 0;
1329 }
1330
1331 static int shmem_create(struct inode *dir, struct dentry *dentry, int mode,
1332                 struct nameidata *nd)
1333 {
1334         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1335 }
1336
1337 /*
1338  * Link a file..
1339  */
1340 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1341 {
1342         struct inode *inode = old_dentry->d_inode;
1343         int ret;
1344
1345         /*
1346          * No ordinary (disk based) filesystem counts links as inodes;
1347          * but each new link needs a new dentry, pinning lowmem, and
1348          * tmpfs dentries cannot be pruned until they are unlinked.
1349          */
1350         ret = shmem_reserve_inode(inode->i_sb);
1351         if (ret)
1352                 goto out;
1353
1354         dir->i_size += BOGO_DIRENT_SIZE;
1355         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1356         inc_nlink(inode);
1357         ihold(inode);   /* New dentry reference */
1358         dget(dentry);           /* Extra pinning count for the created dentry */
1359         d_instantiate(dentry, inode);
1360 out:
1361         return ret;
1362 }
1363
1364 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1365 {
1366         struct inode *inode = dentry->d_inode;
1367
1368         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1369                 shmem_free_inode(inode->i_sb);
1370
1371         dir->i_size -= BOGO_DIRENT_SIZE;
1372         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1373         drop_nlink(inode);
1374         dput(dentry);   /* Undo the count from "create" - this does all the work */
1375         return 0;
1376 }
1377
1378 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1379 {
1380         if (!simple_empty(dentry))
1381                 return -ENOTEMPTY;
1382
1383         drop_nlink(dentry->d_inode);
1384         drop_nlink(dir);
1385         return shmem_unlink(dir, dentry);
1386 }
1387
1388 /*
1389  * The VFS layer already does all the dentry stuff for rename,
1390  * we just have to decrement the usage count for the target if
1391  * it exists so that the VFS layer correctly free's it when it
1392  * gets overwritten.
1393  */
1394 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
1395 {
1396         struct inode *inode = old_dentry->d_inode;
1397         int they_are_dirs = S_ISDIR(inode->i_mode);
1398
1399         if (!simple_empty(new_dentry))
1400                 return -ENOTEMPTY;
1401
1402         if (new_dentry->d_inode) {
1403                 (void) shmem_unlink(new_dir, new_dentry);
1404                 if (they_are_dirs)
1405                         drop_nlink(old_dir);
1406         } else if (they_are_dirs) {
1407                 drop_nlink(old_dir);
1408                 inc_nlink(new_dir);
1409         }
1410
1411         old_dir->i_size -= BOGO_DIRENT_SIZE;
1412         new_dir->i_size += BOGO_DIRENT_SIZE;
1413         old_dir->i_ctime = old_dir->i_mtime =
1414         new_dir->i_ctime = new_dir->i_mtime =
1415         inode->i_ctime = CURRENT_TIME;
1416         return 0;
1417 }
1418
1419 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
1420 {
1421         int error;
1422         int len;
1423         struct inode *inode;
1424         struct page *page;
1425         char *kaddr;
1426         struct shmem_inode_info *info;
1427
1428         len = strlen(symname) + 1;
1429         if (len > PAGE_CACHE_SIZE)
1430                 return -ENAMETOOLONG;
1431
1432         inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
1433         if (!inode)
1434                 return -ENOSPC;
1435
1436         error = security_inode_init_security(inode, dir, &dentry->d_name, NULL,
1437                                              NULL, NULL);
1438         if (error) {
1439                 if (error != -EOPNOTSUPP) {
1440                         iput(inode);
1441                         return error;
1442                 }
1443                 error = 0;
1444         }
1445
1446         info = SHMEM_I(inode);
1447         inode->i_size = len-1;
1448         if (len <= SHMEM_SYMLINK_INLINE_LEN) {
1449                 /* do it inline */
1450                 memcpy(info->inline_symlink, symname, len);
1451                 inode->i_op = &shmem_symlink_inline_operations;
1452         } else {
1453                 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
1454                 if (error) {
1455                         iput(inode);
1456                         return error;
1457                 }
1458                 inode->i_mapping->a_ops = &shmem_aops;
1459                 inode->i_op = &shmem_symlink_inode_operations;
1460                 kaddr = kmap_atomic(page, KM_USER0);
1461                 memcpy(kaddr, symname, len);
1462                 kunmap_atomic(kaddr, KM_USER0);
1463                 set_page_dirty(page);
1464                 unlock_page(page);
1465                 page_cache_release(page);
1466         }
1467         dir->i_size += BOGO_DIRENT_SIZE;
1468         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1469         d_instantiate(dentry, inode);
1470         dget(dentry);
1471         return 0;
1472 }
1473
1474 static void *shmem_follow_link_inline(struct dentry *dentry, struct nameidata *nd)
1475 {
1476         nd_set_link(nd, SHMEM_I(dentry->d_inode)->inline_symlink);
1477         return NULL;
1478 }
1479
1480 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
1481 {
1482         struct page *page = NULL;
1483         int res = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
1484         nd_set_link(nd, res ? ERR_PTR(res) : kmap(page));
1485         if (page)
1486                 unlock_page(page);
1487         return page;
1488 }
1489
1490 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
1491 {
1492         if (!IS_ERR(nd_get_link(nd))) {
1493                 struct page *page = cookie;
1494                 kunmap(page);
1495                 mark_page_accessed(page);
1496                 page_cache_release(page);
1497         }
1498 }
1499
1500 #ifdef CONFIG_TMPFS_XATTR
1501 /*
1502  * Superblocks without xattr inode operations may get some security.* xattr
1503  * support from the LSM "for free". As soon as we have any other xattrs
1504  * like ACLs, we also need to implement the security.* handlers at
1505  * filesystem level, though.
1506  */
1507
1508 static int shmem_xattr_get(struct dentry *dentry, const char *name,
1509                            void *buffer, size_t size)
1510 {
1511         struct shmem_inode_info *info;
1512         struct shmem_xattr *xattr;
1513         int ret = -ENODATA;
1514
1515         info = SHMEM_I(dentry->d_inode);
1516
1517         spin_lock(&info->lock);
1518         list_for_each_entry(xattr, &info->xattr_list, list) {
1519                 if (strcmp(name, xattr->name))
1520                         continue;
1521
1522                 ret = xattr->size;
1523                 if (buffer) {
1524                         if (size < xattr->size)
1525                                 ret = -ERANGE;
1526                         else
1527                                 memcpy(buffer, xattr->value, xattr->size);
1528                 }
1529                 break;
1530         }
1531         spin_unlock(&info->lock);
1532         return ret;
1533 }
1534
1535 static int shmem_xattr_set(struct dentry *dentry, const char *name,
1536                            const void *value, size_t size, int flags)
1537 {
1538         struct inode *inode = dentry->d_inode;
1539         struct shmem_inode_info *info = SHMEM_I(inode);
1540         struct shmem_xattr *xattr;
1541         struct shmem_xattr *new_xattr = NULL;
1542         size_t len;
1543         int err = 0;
1544
1545         /* value == NULL means remove */
1546         if (value) {
1547                 /* wrap around? */
1548                 len = sizeof(*new_xattr) + size;
1549                 if (len <= sizeof(*new_xattr))
1550                         return -ENOMEM;
1551
1552                 new_xattr = kmalloc(len, GFP_KERNEL);
1553                 if (!new_xattr)
1554                         return -ENOMEM;
1555
1556                 new_xattr->name = kstrdup(name, GFP_KERNEL);
1557                 if (!new_xattr->name) {
1558                         kfree(new_xattr);
1559                         return -ENOMEM;
1560                 }
1561
1562                 new_xattr->size = size;
1563                 memcpy(new_xattr->value, value, size);
1564         }
1565
1566         spin_lock(&info->lock);
1567         list_for_each_entry(xattr, &info->xattr_list, list) {
1568                 if (!strcmp(name, xattr->name)) {
1569                         if (flags & XATTR_CREATE) {
1570                                 xattr = new_xattr;
1571                                 err = -EEXIST;
1572                         } else if (new_xattr) {
1573                                 list_replace(&xattr->list, &new_xattr->list);
1574                         } else {
1575                                 list_del(&xattr->list);
1576                         }
1577                         goto out;
1578                 }
1579         }
1580         if (flags & XATTR_REPLACE) {
1581                 xattr = new_xattr;
1582                 err = -ENODATA;
1583         } else {
1584                 list_add(&new_xattr->list, &info->xattr_list);
1585                 xattr = NULL;
1586         }
1587 out:
1588         spin_unlock(&info->lock);
1589         if (xattr)
1590                 kfree(xattr->name);
1591         kfree(xattr);
1592         return err;
1593 }
1594
1595
1596 static const struct xattr_handler *shmem_xattr_handlers[] = {
1597 #ifdef CONFIG_TMPFS_POSIX_ACL
1598         &generic_acl_access_handler,
1599         &generic_acl_default_handler,
1600 #endif
1601         NULL
1602 };
1603
1604 static int shmem_xattr_validate(const char *name)
1605 {
1606         struct { const char *prefix; size_t len; } arr[] = {
1607                 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
1608                 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
1609         };
1610         int i;
1611
1612         for (i = 0; i < ARRAY_SIZE(arr); i++) {
1613                 size_t preflen = arr[i].len;
1614                 if (strncmp(name, arr[i].prefix, preflen) == 0) {
1615                         if (!name[preflen])
1616                                 return -EINVAL;
1617                         return 0;
1618                 }
1619         }
1620         return -EOPNOTSUPP;
1621 }
1622
1623 static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
1624                               void *buffer, size_t size)
1625 {
1626         int err;
1627
1628         /*
1629          * If this is a request for a synthetic attribute in the system.*
1630          * namespace use the generic infrastructure to resolve a handler
1631          * for it via sb->s_xattr.
1632          */
1633         if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1634                 return generic_getxattr(dentry, name, buffer, size);
1635
1636         err = shmem_xattr_validate(name);
1637         if (err)
1638                 return err;
1639
1640         return shmem_xattr_get(dentry, name, buffer, size);
1641 }
1642
1643 static int shmem_setxattr(struct dentry *dentry, const char *name,
1644                           const void *value, size_t size, int flags)
1645 {
1646         int err;
1647
1648         /*
1649          * If this is a request for a synthetic attribute in the system.*
1650          * namespace use the generic infrastructure to resolve a handler
1651          * for it via sb->s_xattr.
1652          */
1653         if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1654                 return generic_setxattr(dentry, name, value, size, flags);
1655
1656         err = shmem_xattr_validate(name);
1657         if (err)
1658                 return err;
1659
1660         if (size == 0)
1661                 value = "";  /* empty EA, do not remove */
1662
1663         return shmem_xattr_set(dentry, name, value, size, flags);
1664
1665 }
1666
1667 static int shmem_removexattr(struct dentry *dentry, const char *name)
1668 {
1669         int err;
1670
1671         /*
1672          * If this is a request for a synthetic attribute in the system.*
1673          * namespace use the generic infrastructure to resolve a handler
1674          * for it via sb->s_xattr.
1675          */
1676         if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
1677                 return generic_removexattr(dentry, name);
1678
1679         err = shmem_xattr_validate(name);
1680         if (err)
1681                 return err;
1682
1683         return shmem_xattr_set(dentry, name, NULL, 0, XATTR_REPLACE);
1684 }
1685
1686 static bool xattr_is_trusted(const char *name)
1687 {
1688         return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN);
1689 }
1690
1691 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
1692 {
1693         bool trusted = capable(CAP_SYS_ADMIN);
1694         struct shmem_xattr *xattr;
1695         struct shmem_inode_info *info;
1696         size_t used = 0;
1697
1698         info = SHMEM_I(dentry->d_inode);
1699
1700         spin_lock(&info->lock);
1701         list_for_each_entry(xattr, &info->xattr_list, list) {
1702                 size_t len;
1703
1704                 /* skip "trusted." attributes for unprivileged callers */
1705                 if (!trusted && xattr_is_trusted(xattr->name))
1706                         continue;
1707
1708                 len = strlen(xattr->name) + 1;
1709                 used += len;
1710                 if (buffer) {
1711                         if (size < used) {
1712                                 used = -ERANGE;
1713                                 break;
1714                         }
1715                         memcpy(buffer, xattr->name, len);
1716                         buffer += len;
1717                 }
1718         }
1719         spin_unlock(&info->lock);
1720
1721         return used;
1722 }
1723 #endif /* CONFIG_TMPFS_XATTR */
1724
1725 static const struct inode_operations shmem_symlink_inline_operations = {
1726         .readlink       = generic_readlink,
1727         .follow_link    = shmem_follow_link_inline,
1728 #ifdef CONFIG_TMPFS_XATTR
1729         .setxattr       = shmem_setxattr,
1730         .getxattr       = shmem_getxattr,
1731         .listxattr      = shmem_listxattr,
1732         .removexattr    = shmem_removexattr,
1733 #endif
1734 };
1735
1736 static const struct inode_operations shmem_symlink_inode_operations = {
1737         .readlink       = generic_readlink,
1738         .follow_link    = shmem_follow_link,
1739         .put_link       = shmem_put_link,
1740 #ifdef CONFIG_TMPFS_XATTR
1741         .setxattr       = shmem_setxattr,
1742         .getxattr       = shmem_getxattr,
1743         .listxattr      = shmem_listxattr,
1744         .removexattr    = shmem_removexattr,
1745 #endif
1746 };
1747
1748 static struct dentry *shmem_get_parent(struct dentry *child)
1749 {
1750         return ERR_PTR(-ESTALE);
1751 }
1752
1753 static int shmem_match(struct inode *ino, void *vfh)
1754 {
1755         __u32 *fh = vfh;
1756         __u64 inum = fh[2];
1757         inum = (inum << 32) | fh[1];
1758         return ino->i_ino == inum && fh[0] == ino->i_generation;
1759 }
1760
1761 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
1762                 struct fid *fid, int fh_len, int fh_type)
1763 {
1764         struct inode *inode;
1765         struct dentry *dentry = NULL;
1766         u64 inum = fid->raw[2];
1767         inum = (inum << 32) | fid->raw[1];
1768
1769         if (fh_len < 3)
1770                 return NULL;
1771
1772         inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
1773                         shmem_match, fid->raw);
1774         if (inode) {
1775                 dentry = d_find_alias(inode);
1776                 iput(inode);
1777         }
1778
1779         return dentry;
1780 }
1781
1782 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
1783                                 int connectable)
1784 {
1785         struct inode *inode = dentry->d_inode;
1786
1787         if (*len < 3) {
1788                 *len = 3;
1789                 return 255;
1790         }
1791
1792         if (inode_unhashed(inode)) {
1793                 /* Unfortunately insert_inode_hash is not idempotent,
1794                  * so as we hash inodes here rather than at creation
1795                  * time, we need a lock to ensure we only try
1796                  * to do it once
1797                  */
1798                 static DEFINE_SPINLOCK(lock);
1799                 spin_lock(&lock);
1800                 if (inode_unhashed(inode))
1801                         __insert_inode_hash(inode,
1802                                             inode->i_ino + inode->i_generation);
1803                 spin_unlock(&lock);
1804         }
1805
1806         fh[0] = inode->i_generation;
1807         fh[1] = inode->i_ino;
1808         fh[2] = ((__u64)inode->i_ino) >> 32;
1809
1810         *len = 3;
1811         return 1;
1812 }
1813
1814 static const struct export_operations shmem_export_ops = {
1815         .get_parent     = shmem_get_parent,
1816         .encode_fh      = shmem_encode_fh,
1817         .fh_to_dentry   = shmem_fh_to_dentry,
1818 };
1819
1820 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
1821                                bool remount)
1822 {
1823         char *this_char, *value, *rest;
1824
1825         while (options != NULL) {
1826                 this_char = options;
1827                 for (;;) {
1828                         /*
1829                          * NUL-terminate this option: unfortunately,
1830                          * mount options form a comma-separated list,
1831                          * but mpol's nodelist may also contain commas.
1832                          */
1833                         options = strchr(options, ',');
1834                         if (options == NULL)
1835                                 break;
1836                         options++;
1837                         if (!isdigit(*options)) {
1838                                 options[-1] = '\0';
1839                                 break;
1840                         }
1841                 }
1842                 if (!*this_char)
1843                         continue;
1844                 if ((value = strchr(this_char,'=')) != NULL) {
1845                         *value++ = 0;
1846                 } else {
1847                         printk(KERN_ERR
1848                             "tmpfs: No value for mount option '%s'\n",
1849                             this_char);
1850                         return 1;
1851                 }
1852
1853                 if (!strcmp(this_char,"size")) {
1854                         unsigned long long size;
1855                         size = memparse(value,&rest);
1856                         if (*rest == '%') {
1857                                 size <<= PAGE_SHIFT;
1858                                 size *= totalram_pages;
1859                                 do_div(size, 100);
1860                                 rest++;
1861                         }
1862                         if (*rest)
1863                                 goto bad_val;
1864                         sbinfo->max_blocks =
1865                                 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
1866                 } else if (!strcmp(this_char,"nr_blocks")) {
1867                         sbinfo->max_blocks = memparse(value, &rest);
1868                         if (*rest)
1869                                 goto bad_val;
1870                 } else if (!strcmp(this_char,"nr_inodes")) {
1871                         sbinfo->max_inodes = memparse(value, &rest);
1872                         if (*rest)
1873                                 goto bad_val;
1874                 } else if (!strcmp(this_char,"mode")) {
1875                         if (remount)
1876                                 continue;
1877                         sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
1878                         if (*rest)
1879                                 goto bad_val;
1880                 } else if (!strcmp(this_char,"uid")) {
1881                         if (remount)
1882                                 continue;
1883                         sbinfo->uid = simple_strtoul(value, &rest, 0);
1884                         if (*rest)
1885                                 goto bad_val;
1886                 } else if (!strcmp(this_char,"gid")) {
1887                         if (remount)
1888                                 continue;
1889                         sbinfo->gid = simple_strtoul(value, &rest, 0);
1890                         if (*rest)
1891                                 goto bad_val;
1892                 } else if (!strcmp(this_char,"mpol")) {
1893                         if (mpol_parse_str(value, &sbinfo->mpol, 1))
1894                                 goto bad_val;
1895                 } else {
1896                         printk(KERN_ERR "tmpfs: Bad mount option %s\n",
1897                                this_char);
1898                         return 1;
1899                 }
1900         }
1901         return 0;
1902
1903 bad_val:
1904         printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
1905                value, this_char);
1906         return 1;
1907
1908 }
1909
1910 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
1911 {
1912         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1913         struct shmem_sb_info config = *sbinfo;
1914         unsigned long inodes;
1915         int error = -EINVAL;
1916
1917         if (shmem_parse_options(data, &config, true))
1918                 return error;
1919
1920         spin_lock(&sbinfo->stat_lock);
1921         inodes = sbinfo->max_inodes - sbinfo->free_inodes;
1922         if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
1923                 goto out;
1924         if (config.max_inodes < inodes)
1925                 goto out;
1926         /*
1927          * Those tests also disallow limited->unlimited while any are in
1928          * use, so i_blocks will always be zero when max_blocks is zero;
1929          * but we must separately disallow unlimited->limited, because
1930          * in that case we have no record of how much is already in use.
1931          */
1932         if (config.max_blocks && !sbinfo->max_blocks)
1933                 goto out;
1934         if (config.max_inodes && !sbinfo->max_inodes)
1935                 goto out;
1936
1937         error = 0;
1938         sbinfo->max_blocks  = config.max_blocks;
1939         sbinfo->max_inodes  = config.max_inodes;
1940         sbinfo->free_inodes = config.max_inodes - inodes;
1941
1942         mpol_put(sbinfo->mpol);
1943         sbinfo->mpol        = config.mpol;      /* transfers initial ref */
1944 out:
1945         spin_unlock(&sbinfo->stat_lock);
1946         return error;
1947 }
1948
1949 static int shmem_show_options(struct seq_file *seq, struct vfsmount *vfs)
1950 {
1951         struct shmem_sb_info *sbinfo = SHMEM_SB(vfs->mnt_sb);
1952
1953         if (sbinfo->max_blocks != shmem_default_max_blocks())
1954                 seq_printf(seq, ",size=%luk",
1955                         sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
1956         if (sbinfo->max_inodes != shmem_default_max_inodes())
1957                 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
1958         if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
1959                 seq_printf(seq, ",mode=%03o", sbinfo->mode);
1960         if (sbinfo->uid != 0)
1961                 seq_printf(seq, ",uid=%u", sbinfo->uid);
1962         if (sbinfo->gid != 0)
1963                 seq_printf(seq, ",gid=%u", sbinfo->gid);
1964         shmem_show_mpol(seq, sbinfo->mpol);
1965         return 0;
1966 }
1967 #endif /* CONFIG_TMPFS */
1968
1969 static void shmem_put_super(struct super_block *sb)
1970 {
1971         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1972
1973         percpu_counter_destroy(&sbinfo->used_blocks);
1974         kfree(sbinfo);
1975         sb->s_fs_info = NULL;
1976 }
1977
1978 int shmem_fill_super(struct super_block *sb, void *data, int silent)
1979 {
1980         struct inode *inode;
1981         struct dentry *root;
1982         struct shmem_sb_info *sbinfo;
1983         int err = -ENOMEM;
1984
1985         /* Round up to L1_CACHE_BYTES to resist false sharing */
1986         sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
1987                                 L1_CACHE_BYTES), GFP_KERNEL);
1988         if (!sbinfo)
1989                 return -ENOMEM;
1990
1991         sbinfo->mode = S_IRWXUGO | S_ISVTX;
1992         sbinfo->uid = current_fsuid();
1993         sbinfo->gid = current_fsgid();
1994         sb->s_fs_info = sbinfo;
1995
1996 #ifdef CONFIG_TMPFS
1997         /*
1998          * Per default we only allow half of the physical ram per
1999          * tmpfs instance, limiting inodes to one per page of lowmem;
2000          * but the internal instance is left unlimited.
2001          */
2002         if (!(sb->s_flags & MS_NOUSER)) {
2003                 sbinfo->max_blocks = shmem_default_max_blocks();
2004                 sbinfo->max_inodes = shmem_default_max_inodes();
2005                 if (shmem_parse_options(data, sbinfo, false)) {
2006                         err = -EINVAL;
2007                         goto failed;
2008                 }
2009         }
2010         sb->s_export_op = &shmem_export_ops;
2011 #else
2012         sb->s_flags |= MS_NOUSER;
2013 #endif
2014
2015         spin_lock_init(&sbinfo->stat_lock);
2016         if (percpu_counter_init(&sbinfo->used_blocks, 0))
2017                 goto failed;
2018         sbinfo->free_inodes = sbinfo->max_inodes;
2019
2020         sb->s_maxbytes = MAX_LFS_FILESIZE;
2021         sb->s_blocksize = PAGE_CACHE_SIZE;
2022         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2023         sb->s_magic = TMPFS_MAGIC;
2024         sb->s_op = &shmem_ops;
2025         sb->s_time_gran = 1;
2026 #ifdef CONFIG_TMPFS_XATTR
2027         sb->s_xattr = shmem_xattr_handlers;
2028 #endif
2029 #ifdef CONFIG_TMPFS_POSIX_ACL
2030         sb->s_flags |= MS_POSIXACL;
2031 #endif
2032
2033         inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2034         if (!inode)
2035                 goto failed;
2036         inode->i_uid = sbinfo->uid;
2037         inode->i_gid = sbinfo->gid;
2038         root = d_alloc_root(inode);
2039         if (!root)
2040                 goto failed_iput;
2041         sb->s_root = root;
2042         return 0;
2043
2044 failed_iput:
2045         iput(inode);
2046 failed:
2047         shmem_put_super(sb);
2048         return err;
2049 }
2050
2051 static struct kmem_cache *shmem_inode_cachep;
2052
2053 static struct inode *shmem_alloc_inode(struct super_block *sb)
2054 {
2055         struct shmem_inode_info *p;
2056         p = (struct shmem_inode_info *)kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2057         if (!p)
2058                 return NULL;
2059         return &p->vfs_inode;
2060 }
2061
2062 static void shmem_i_callback(struct rcu_head *head)
2063 {
2064         struct inode *inode = container_of(head, struct inode, i_rcu);
2065         INIT_LIST_HEAD(&inode->i_dentry);
2066         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2067 }
2068
2069 static void shmem_destroy_inode(struct inode *inode)
2070 {
2071         if ((inode->i_mode & S_IFMT) == S_IFREG) {
2072                 /* only struct inode is valid if it's an inline symlink */
2073                 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2074         }
2075         call_rcu(&inode->i_rcu, shmem_i_callback);
2076 }
2077
2078 static void init_once(void *foo)
2079 {
2080         struct shmem_inode_info *p = (struct shmem_inode_info *) foo;
2081
2082         inode_init_once(&p->vfs_inode);
2083 }
2084
2085 static int init_inodecache(void)
2086 {
2087         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2088                                 sizeof(struct shmem_inode_info),
2089                                 0, SLAB_PANIC, init_once);
2090         return 0;
2091 }
2092
2093 static void destroy_inodecache(void)
2094 {
2095         kmem_cache_destroy(shmem_inode_cachep);
2096 }
2097
2098 static const struct address_space_operations shmem_aops = {
2099         .writepage      = shmem_writepage,
2100         .set_page_dirty = __set_page_dirty_no_writeback,
2101 #ifdef CONFIG_TMPFS
2102         .write_begin    = shmem_write_begin,
2103         .write_end      = shmem_write_end,
2104 #endif
2105         .migratepage    = migrate_page,
2106         .error_remove_page = generic_error_remove_page,
2107 };
2108
2109 static const struct file_operations shmem_file_operations = {
2110         .mmap           = shmem_mmap,
2111 #ifdef CONFIG_TMPFS
2112         .llseek         = generic_file_llseek,
2113         .read           = do_sync_read,
2114         .write          = do_sync_write,
2115         .aio_read       = shmem_file_aio_read,
2116         .aio_write      = generic_file_aio_write,
2117         .fsync          = noop_fsync,
2118         .splice_read    = shmem_file_splice_read,
2119         .splice_write   = generic_file_splice_write,
2120 #endif
2121 };
2122
2123 static const struct inode_operations shmem_inode_operations = {
2124         .setattr        = shmem_setattr,
2125         .truncate_range = shmem_truncate_range,
2126 #ifdef CONFIG_TMPFS_XATTR
2127         .setxattr       = shmem_setxattr,
2128         .getxattr       = shmem_getxattr,
2129         .listxattr      = shmem_listxattr,
2130         .removexattr    = shmem_removexattr,
2131 #endif
2132 };
2133
2134 static const struct inode_operations shmem_dir_inode_operations = {
2135 #ifdef CONFIG_TMPFS
2136         .create         = shmem_create,
2137         .lookup         = simple_lookup,
2138         .link           = shmem_link,
2139         .unlink         = shmem_unlink,
2140         .symlink        = shmem_symlink,
2141         .mkdir          = shmem_mkdir,
2142         .rmdir          = shmem_rmdir,
2143         .mknod          = shmem_mknod,
2144         .rename         = shmem_rename,
2145 #endif
2146 #ifdef CONFIG_TMPFS_XATTR
2147         .setxattr       = shmem_setxattr,
2148         .getxattr       = shmem_getxattr,
2149         .listxattr      = shmem_listxattr,
2150         .removexattr    = shmem_removexattr,
2151 #endif
2152 #ifdef CONFIG_TMPFS_POSIX_ACL
2153         .setattr        = shmem_setattr,
2154 #endif
2155 };
2156
2157 static const struct inode_operations shmem_special_inode_operations = {
2158 #ifdef CONFIG_TMPFS_XATTR
2159         .setxattr       = shmem_setxattr,
2160         .getxattr       = shmem_getxattr,
2161         .listxattr      = shmem_listxattr,
2162         .removexattr    = shmem_removexattr,
2163 #endif
2164 #ifdef CONFIG_TMPFS_POSIX_ACL
2165         .setattr        = shmem_setattr,
2166 #endif
2167 };
2168
2169 static const struct super_operations shmem_ops = {
2170         .alloc_inode    = shmem_alloc_inode,
2171         .destroy_inode  = shmem_destroy_inode,
2172 #ifdef CONFIG_TMPFS
2173         .statfs         = shmem_statfs,
2174         .remount_fs     = shmem_remount_fs,
2175         .show_options   = shmem_show_options,
2176 #endif
2177         .evict_inode    = shmem_evict_inode,
2178         .drop_inode     = generic_delete_inode,
2179         .put_super      = shmem_put_super,
2180 };
2181
2182 static const struct vm_operations_struct shmem_vm_ops = {
2183         .fault          = shmem_fault,
2184 #ifdef CONFIG_NUMA
2185         .set_policy     = shmem_set_policy,
2186         .get_policy     = shmem_get_policy,
2187 #endif
2188 };
2189
2190
2191 static struct dentry *shmem_mount(struct file_system_type *fs_type,
2192         int flags, const char *dev_name, void *data)
2193 {
2194         return mount_nodev(fs_type, flags, data, shmem_fill_super);
2195 }
2196
2197 static struct file_system_type tmpfs_fs_type = {
2198         .owner          = THIS_MODULE,
2199         .name           = "tmpfs",
2200         .mount          = shmem_mount,
2201         .kill_sb        = kill_litter_super,
2202 };
2203
2204 int __init init_tmpfs(void)
2205 {
2206         int error;
2207
2208         error = bdi_init(&shmem_backing_dev_info);
2209         if (error)
2210                 goto out4;
2211
2212         error = init_inodecache();
2213         if (error)
2214                 goto out3;
2215
2216         error = register_filesystem(&tmpfs_fs_type);
2217         if (error) {
2218                 printk(KERN_ERR "Could not register tmpfs\n");
2219                 goto out2;
2220         }
2221
2222         shm_mnt = vfs_kern_mount(&tmpfs_fs_type, MS_NOUSER,
2223                                 tmpfs_fs_type.name, NULL);
2224         if (IS_ERR(shm_mnt)) {
2225                 error = PTR_ERR(shm_mnt);
2226                 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2227                 goto out1;
2228         }
2229         return 0;
2230
2231 out1:
2232         unregister_filesystem(&tmpfs_fs_type);
2233 out2:
2234         destroy_inodecache();
2235 out3:
2236         bdi_destroy(&shmem_backing_dev_info);
2237 out4:
2238         shm_mnt = ERR_PTR(error);
2239         return error;
2240 }
2241
2242 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
2243 /**
2244  * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file
2245  * @inode: the inode to be searched
2246  * @pgoff: the offset to be searched
2247  * @pagep: the pointer for the found page to be stored
2248  * @ent: the pointer for the found swap entry to be stored
2249  *
2250  * If a page is found, refcount of it is incremented. Callers should handle
2251  * these refcount.
2252  */
2253 void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t pgoff,
2254                                         struct page **pagep, swp_entry_t *ent)
2255 {
2256         swp_entry_t entry = { .val = 0 };
2257         struct page *page = NULL;
2258         struct shmem_inode_info *info = SHMEM_I(inode);
2259
2260         if ((pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
2261                 goto out;
2262
2263         spin_lock(&info->lock);
2264 #ifdef CONFIG_SWAP
2265         entry = shmem_get_swap(info, pgoff);
2266         if (entry.val)
2267                 page = find_get_page(&swapper_space, entry.val);
2268         else
2269 #endif
2270                 page = find_get_page(inode->i_mapping, pgoff);
2271         spin_unlock(&info->lock);
2272 out:
2273         *pagep = page;
2274         *ent = entry;
2275 }
2276 #endif
2277
2278 #else /* !CONFIG_SHMEM */
2279
2280 /*
2281  * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2282  *
2283  * This is intended for small system where the benefits of the full
2284  * shmem code (swap-backed and resource-limited) are outweighed by
2285  * their complexity. On systems without swap this code should be
2286  * effectively equivalent, but much lighter weight.
2287  */
2288
2289 #include <linux/ramfs.h>
2290
2291 static struct file_system_type tmpfs_fs_type = {
2292         .name           = "tmpfs",
2293         .mount          = ramfs_mount,
2294         .kill_sb        = kill_litter_super,
2295 };
2296
2297 int __init init_tmpfs(void)
2298 {
2299         BUG_ON(register_filesystem(&tmpfs_fs_type) != 0);
2300
2301         shm_mnt = kern_mount(&tmpfs_fs_type);
2302         BUG_ON(IS_ERR(shm_mnt));
2303
2304         return 0;
2305 }
2306
2307 int shmem_unuse(swp_entry_t entry, struct page *page)
2308 {
2309         return 0;
2310 }
2311
2312 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2313 {
2314         return 0;
2315 }
2316
2317 void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end)
2318 {
2319         truncate_inode_pages_range(inode->i_mapping, start, end);
2320 }
2321 EXPORT_SYMBOL_GPL(shmem_truncate_range);
2322
2323 #ifdef CONFIG_CGROUP_MEM_RES_CTLR
2324 /**
2325  * mem_cgroup_get_shmem_target - find a page or entry assigned to the shmem file
2326  * @inode: the inode to be searched
2327  * @pgoff: the offset to be searched
2328  * @pagep: the pointer for the found page to be stored
2329  * @ent: the pointer for the found swap entry to be stored
2330  *
2331  * If a page is found, refcount of it is incremented. Callers should handle
2332  * these refcount.
2333  */
2334 void mem_cgroup_get_shmem_target(struct inode *inode, pgoff_t pgoff,
2335                                         struct page **pagep, swp_entry_t *ent)
2336 {
2337         struct page *page = NULL;
2338
2339         if ((pgoff << PAGE_CACHE_SHIFT) >= i_size_read(inode))
2340                 goto out;
2341         page = find_get_page(inode->i_mapping, pgoff);
2342 out:
2343         *pagep = page;
2344         *ent = (swp_entry_t){ .val = 0 };
2345 }
2346 #endif
2347
2348 #define shmem_vm_ops                            generic_file_vm_ops
2349 #define shmem_file_operations                   ramfs_file_operations
2350 #define shmem_get_inode(sb, dir, mode, dev, flags)      ramfs_get_inode(sb, dir, mode, dev)
2351 #define shmem_acct_size(flags, size)            0
2352 #define shmem_unacct_size(flags, size)          do {} while (0)
2353
2354 #endif /* CONFIG_SHMEM */
2355
2356 /* common code */
2357
2358 /**
2359  * shmem_file_setup - get an unlinked file living in tmpfs
2360  * @name: name for dentry (to be seen in /proc/<pid>/maps
2361  * @size: size to be set for the file
2362  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2363  */
2364 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2365 {
2366         int error;
2367         struct file *file;
2368         struct inode *inode;
2369         struct path path;
2370         struct dentry *root;
2371         struct qstr this;
2372
2373         if (IS_ERR(shm_mnt))
2374                 return (void *)shm_mnt;
2375
2376         if (size < 0 || size > MAX_LFS_FILESIZE)
2377                 return ERR_PTR(-EINVAL);
2378
2379         if (shmem_acct_size(flags, size))
2380                 return ERR_PTR(-ENOMEM);
2381
2382         error = -ENOMEM;
2383         this.name = name;
2384         this.len = strlen(name);
2385         this.hash = 0; /* will go */
2386         root = shm_mnt->mnt_root;
2387         path.dentry = d_alloc(root, &this);
2388         if (!path.dentry)
2389                 goto put_memory;
2390         path.mnt = mntget(shm_mnt);
2391
2392         error = -ENOSPC;
2393         inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2394         if (!inode)
2395                 goto put_dentry;
2396
2397         d_instantiate(path.dentry, inode);
2398         inode->i_size = size;
2399         inode->i_nlink = 0;     /* It is unlinked */
2400 #ifndef CONFIG_MMU
2401         error = ramfs_nommu_expand_for_mapping(inode, size);
2402         if (error)
2403                 goto put_dentry;
2404 #endif
2405
2406         error = -ENFILE;
2407         file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2408                   &shmem_file_operations);
2409         if (!file)
2410                 goto put_dentry;
2411
2412         return file;
2413
2414 put_dentry:
2415         path_put(&path);
2416 put_memory:
2417         shmem_unacct_size(flags, size);
2418         return ERR_PTR(error);
2419 }
2420 EXPORT_SYMBOL_GPL(shmem_file_setup);
2421
2422 /**
2423  * shmem_zero_setup - setup a shared anonymous mapping
2424  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2425  */
2426 int shmem_zero_setup(struct vm_area_struct *vma)
2427 {
2428         struct file *file;
2429         loff_t size = vma->vm_end - vma->vm_start;
2430
2431         file = shmem_file_setup("dev/zero", size, vma->vm_flags);
2432         if (IS_ERR(file))
2433                 return PTR_ERR(file);
2434
2435         if (vma->vm_file)
2436                 fput(vma->vm_file);
2437         vma->vm_file = file;
2438         vma->vm_ops = &shmem_vm_ops;
2439         vma->vm_flags |= VM_CAN_NONLINEAR;
2440         return 0;
2441 }
2442
2443 /**
2444  * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
2445  * @mapping:    the page's address_space
2446  * @index:      the page index
2447  * @gfp:        the page allocator flags to use if allocating
2448  *
2449  * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
2450  * with any new page allocations done using the specified allocation flags.
2451  * But read_cache_page_gfp() uses the ->readpage() method: which does not
2452  * suit tmpfs, since it may have pages in swapcache, and needs to find those
2453  * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
2454  *
2455  * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
2456  * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
2457  */
2458 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
2459                                          pgoff_t index, gfp_t gfp)
2460 {
2461 #ifdef CONFIG_SHMEM
2462         struct inode *inode = mapping->host;
2463         struct page *page;
2464         int error;
2465
2466         BUG_ON(mapping->a_ops != &shmem_aops);
2467         error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
2468         if (error)
2469                 page = ERR_PTR(error);
2470         else
2471                 unlock_page(page);
2472         return page;
2473 #else
2474         /*
2475          * The tiny !SHMEM case uses ramfs without swap
2476          */
2477         return read_cache_page_gfp(mapping, index, gfp);
2478 #endif
2479 }
2480 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);