]> git.openfabrics.org - ~shefty/rdma-dev.git/blob - mm/shmem.c
tmpfs: support SEEK_DATA and SEEK_HOLE
[~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) 2002-2011 Hugh Dickins.
10  * Copyright (C) 2011 Google Inc.
11  * Copyright (C) 2002-2005 VERITAS Software Corporation.
12  * Copyright (C) 2004 Andi Kleen, SuSE Labs
13  *
14  * Extended attribute support for tmpfs:
15  * Copyright (c) 2004, Luke Kenneth Casson Leighton <lkcl@lkcl.net>
16  * Copyright (c) 2004 Red Hat, Inc., James Morris <jmorris@redhat.com>
17  *
18  * tiny-shmem:
19  * Copyright (c) 2004, 2008 Matt Mackall <mpm@selenic.com>
20  *
21  * This file is released under the GPL.
22  */
23
24 #include <linux/fs.h>
25 #include <linux/init.h>
26 #include <linux/vfs.h>
27 #include <linux/mount.h>
28 #include <linux/pagemap.h>
29 #include <linux/file.h>
30 #include <linux/mm.h>
31 #include <linux/export.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/pagevec.h>
55 #include <linux/percpu_counter.h>
56 #include <linux/falloc.h>
57 #include <linux/splice.h>
58 #include <linux/security.h>
59 #include <linux/swapops.h>
60 #include <linux/mempolicy.h>
61 #include <linux/namei.h>
62 #include <linux/ctype.h>
63 #include <linux/migrate.h>
64 #include <linux/highmem.h>
65 #include <linux/seq_file.h>
66 #include <linux/magic.h>
67
68 #include <asm/uaccess.h>
69 #include <asm/pgtable.h>
70
71 #define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
72 #define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
73
74 /* Pretend that each entry is of this size in directory's i_size */
75 #define BOGO_DIRENT_SIZE 20
76
77 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
78 #define SHORT_SYMLINK_LEN 128
79
80 struct shmem_xattr {
81         struct list_head list;  /* anchored by shmem_inode_info->xattr_list */
82         char *name;             /* xattr name */
83         size_t size;
84         char value[0];
85 };
86
87 /*
88  * shmem_fallocate and shmem_writepage communicate via inode->i_private
89  * (with i_mutex making sure that it has only one user at a time):
90  * we would prefer not to enlarge the shmem inode just for that.
91  */
92 struct shmem_falloc {
93         pgoff_t start;          /* start of range currently being fallocated */
94         pgoff_t next;           /* the next page offset to be fallocated */
95         pgoff_t nr_falloced;    /* how many new pages have been fallocated */
96         pgoff_t nr_unswapped;   /* how often writepage refused to swap out */
97 };
98
99 /* Flag allocation requirements to shmem_getpage */
100 enum sgp_type {
101         SGP_READ,       /* don't exceed i_size, don't allocate page */
102         SGP_CACHE,      /* don't exceed i_size, may allocate page */
103         SGP_DIRTY,      /* like SGP_CACHE, but set new page dirty */
104         SGP_WRITE,      /* may exceed i_size, may allocate !Uptodate page */
105         SGP_FALLOC,     /* like SGP_WRITE, but make existing page Uptodate */
106 };
107
108 #ifdef CONFIG_TMPFS
109 static unsigned long shmem_default_max_blocks(void)
110 {
111         return totalram_pages / 2;
112 }
113
114 static unsigned long shmem_default_max_inodes(void)
115 {
116         return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
117 }
118 #endif
119
120 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
121 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
122                                 struct shmem_inode_info *info, pgoff_t index);
123 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
124         struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
125
126 static inline int shmem_getpage(struct inode *inode, pgoff_t index,
127         struct page **pagep, enum sgp_type sgp, int *fault_type)
128 {
129         return shmem_getpage_gfp(inode, index, pagep, sgp,
130                         mapping_gfp_mask(inode->i_mapping), fault_type);
131 }
132
133 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
134 {
135         return sb->s_fs_info;
136 }
137
138 /*
139  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
140  * for shared memory and for shared anonymous (/dev/zero) mappings
141  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
142  * consistent with the pre-accounting of private mappings ...
143  */
144 static inline int shmem_acct_size(unsigned long flags, loff_t size)
145 {
146         return (flags & VM_NORESERVE) ?
147                 0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
148 }
149
150 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
151 {
152         if (!(flags & VM_NORESERVE))
153                 vm_unacct_memory(VM_ACCT(size));
154 }
155
156 /*
157  * ... whereas tmpfs objects are accounted incrementally as
158  * pages are allocated, in order to allow huge sparse files.
159  * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
160  * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
161  */
162 static inline int shmem_acct_block(unsigned long flags)
163 {
164         return (flags & VM_NORESERVE) ?
165                 security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0;
166 }
167
168 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
169 {
170         if (flags & VM_NORESERVE)
171                 vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
172 }
173
174 static const struct super_operations shmem_ops;
175 static const struct address_space_operations shmem_aops;
176 static const struct file_operations shmem_file_operations;
177 static const struct inode_operations shmem_inode_operations;
178 static const struct inode_operations shmem_dir_inode_operations;
179 static const struct inode_operations shmem_special_inode_operations;
180 static const struct vm_operations_struct shmem_vm_ops;
181
182 static struct backing_dev_info shmem_backing_dev_info  __read_mostly = {
183         .ra_pages       = 0,    /* No readahead */
184         .capabilities   = BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
185 };
186
187 static LIST_HEAD(shmem_swaplist);
188 static DEFINE_MUTEX(shmem_swaplist_mutex);
189
190 static int shmem_reserve_inode(struct super_block *sb)
191 {
192         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
193         if (sbinfo->max_inodes) {
194                 spin_lock(&sbinfo->stat_lock);
195                 if (!sbinfo->free_inodes) {
196                         spin_unlock(&sbinfo->stat_lock);
197                         return -ENOSPC;
198                 }
199                 sbinfo->free_inodes--;
200                 spin_unlock(&sbinfo->stat_lock);
201         }
202         return 0;
203 }
204
205 static void shmem_free_inode(struct super_block *sb)
206 {
207         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
208         if (sbinfo->max_inodes) {
209                 spin_lock(&sbinfo->stat_lock);
210                 sbinfo->free_inodes++;
211                 spin_unlock(&sbinfo->stat_lock);
212         }
213 }
214
215 /**
216  * shmem_recalc_inode - recalculate the block usage of an inode
217  * @inode: inode to recalc
218  *
219  * We have to calculate the free blocks since the mm can drop
220  * undirtied hole pages behind our back.
221  *
222  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
223  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
224  *
225  * It has to be called with the spinlock held.
226  */
227 static void shmem_recalc_inode(struct inode *inode)
228 {
229         struct shmem_inode_info *info = SHMEM_I(inode);
230         long freed;
231
232         freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
233         if (freed > 0) {
234                 struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
235                 if (sbinfo->max_blocks)
236                         percpu_counter_add(&sbinfo->used_blocks, -freed);
237                 info->alloced -= freed;
238                 inode->i_blocks -= freed * BLOCKS_PER_PAGE;
239                 shmem_unacct_blocks(info->flags, freed);
240         }
241 }
242
243 /*
244  * Replace item expected in radix tree by a new item, while holding tree lock.
245  */
246 static int shmem_radix_tree_replace(struct address_space *mapping,
247                         pgoff_t index, void *expected, void *replacement)
248 {
249         void **pslot;
250         void *item = NULL;
251
252         VM_BUG_ON(!expected);
253         pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
254         if (pslot)
255                 item = radix_tree_deref_slot_protected(pslot,
256                                                         &mapping->tree_lock);
257         if (item != expected)
258                 return -ENOENT;
259         if (replacement)
260                 radix_tree_replace_slot(pslot, replacement);
261         else
262                 radix_tree_delete(&mapping->page_tree, index);
263         return 0;
264 }
265
266 /*
267  * Like add_to_page_cache_locked, but error if expected item has gone.
268  */
269 static int shmem_add_to_page_cache(struct page *page,
270                                    struct address_space *mapping,
271                                    pgoff_t index, gfp_t gfp, void *expected)
272 {
273         int error = 0;
274
275         VM_BUG_ON(!PageLocked(page));
276         VM_BUG_ON(!PageSwapBacked(page));
277
278         if (!expected)
279                 error = radix_tree_preload(gfp & GFP_RECLAIM_MASK);
280         if (!error) {
281                 page_cache_get(page);
282                 page->mapping = mapping;
283                 page->index = index;
284
285                 spin_lock_irq(&mapping->tree_lock);
286                 if (!expected)
287                         error = radix_tree_insert(&mapping->page_tree,
288                                                         index, page);
289                 else
290                         error = shmem_radix_tree_replace(mapping, index,
291                                                         expected, page);
292                 if (!error) {
293                         mapping->nrpages++;
294                         __inc_zone_page_state(page, NR_FILE_PAGES);
295                         __inc_zone_page_state(page, NR_SHMEM);
296                         spin_unlock_irq(&mapping->tree_lock);
297                 } else {
298                         page->mapping = NULL;
299                         spin_unlock_irq(&mapping->tree_lock);
300                         page_cache_release(page);
301                 }
302                 if (!expected)
303                         radix_tree_preload_end();
304         }
305         if (error)
306                 mem_cgroup_uncharge_cache_page(page);
307         return error;
308 }
309
310 /*
311  * Like delete_from_page_cache, but substitutes swap for page.
312  */
313 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
314 {
315         struct address_space *mapping = page->mapping;
316         int error;
317
318         spin_lock_irq(&mapping->tree_lock);
319         error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
320         page->mapping = NULL;
321         mapping->nrpages--;
322         __dec_zone_page_state(page, NR_FILE_PAGES);
323         __dec_zone_page_state(page, NR_SHMEM);
324         spin_unlock_irq(&mapping->tree_lock);
325         page_cache_release(page);
326         BUG_ON(error);
327 }
328
329 /*
330  * Like find_get_pages, but collecting swap entries as well as pages.
331  */
332 static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping,
333                                         pgoff_t start, unsigned int nr_pages,
334                                         struct page **pages, pgoff_t *indices)
335 {
336         unsigned int i;
337         unsigned int ret;
338         unsigned int nr_found;
339
340         rcu_read_lock();
341 restart:
342         nr_found = radix_tree_gang_lookup_slot(&mapping->page_tree,
343                                 (void ***)pages, indices, start, nr_pages);
344         ret = 0;
345         for (i = 0; i < nr_found; i++) {
346                 struct page *page;
347 repeat:
348                 page = radix_tree_deref_slot((void **)pages[i]);
349                 if (unlikely(!page))
350                         continue;
351                 if (radix_tree_exception(page)) {
352                         if (radix_tree_deref_retry(page))
353                                 goto restart;
354                         /*
355                          * Otherwise, we must be storing a swap entry
356                          * here as an exceptional entry: so return it
357                          * without attempting to raise page count.
358                          */
359                         goto export;
360                 }
361                 if (!page_cache_get_speculative(page))
362                         goto repeat;
363
364                 /* Has the page moved? */
365                 if (unlikely(page != *((void **)pages[i]))) {
366                         page_cache_release(page);
367                         goto repeat;
368                 }
369 export:
370                 indices[ret] = indices[i];
371                 pages[ret] = page;
372                 ret++;
373         }
374         if (unlikely(!ret && nr_found))
375                 goto restart;
376         rcu_read_unlock();
377         return ret;
378 }
379
380 /*
381  * Remove swap entry from radix tree, free the swap and its page cache.
382  */
383 static int shmem_free_swap(struct address_space *mapping,
384                            pgoff_t index, void *radswap)
385 {
386         int error;
387
388         spin_lock_irq(&mapping->tree_lock);
389         error = shmem_radix_tree_replace(mapping, index, radswap, NULL);
390         spin_unlock_irq(&mapping->tree_lock);
391         if (!error)
392                 free_swap_and_cache(radix_to_swp_entry(radswap));
393         return error;
394 }
395
396 /*
397  * Pagevec may contain swap entries, so shuffle up pages before releasing.
398  */
399 static void shmem_deswap_pagevec(struct pagevec *pvec)
400 {
401         int i, j;
402
403         for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
404                 struct page *page = pvec->pages[i];
405                 if (!radix_tree_exceptional_entry(page))
406                         pvec->pages[j++] = page;
407         }
408         pvec->nr = j;
409 }
410
411 /*
412  * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
413  */
414 void shmem_unlock_mapping(struct address_space *mapping)
415 {
416         struct pagevec pvec;
417         pgoff_t indices[PAGEVEC_SIZE];
418         pgoff_t index = 0;
419
420         pagevec_init(&pvec, 0);
421         /*
422          * Minor point, but we might as well stop if someone else SHM_LOCKs it.
423          */
424         while (!mapping_unevictable(mapping)) {
425                 /*
426                  * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
427                  * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
428                  */
429                 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
430                                         PAGEVEC_SIZE, pvec.pages, indices);
431                 if (!pvec.nr)
432                         break;
433                 index = indices[pvec.nr - 1] + 1;
434                 shmem_deswap_pagevec(&pvec);
435                 check_move_unevictable_pages(pvec.pages, pvec.nr);
436                 pagevec_release(&pvec);
437                 cond_resched();
438         }
439 }
440
441 /*
442  * Remove range of pages and swap entries from radix tree, and free them.
443  * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
444  */
445 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
446                                                                  bool unfalloc)
447 {
448         struct address_space *mapping = inode->i_mapping;
449         struct shmem_inode_info *info = SHMEM_I(inode);
450         pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
451         pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
452         unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
453         unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
454         struct pagevec pvec;
455         pgoff_t indices[PAGEVEC_SIZE];
456         long nr_swaps_freed = 0;
457         pgoff_t index;
458         int i;
459
460         if (lend == -1)
461                 end = -1;       /* unsigned, so actually very big */
462
463         pagevec_init(&pvec, 0);
464         index = start;
465         while (index < end) {
466                 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
467                                 min(end - index, (pgoff_t)PAGEVEC_SIZE),
468                                                         pvec.pages, indices);
469                 if (!pvec.nr)
470                         break;
471                 mem_cgroup_uncharge_start();
472                 for (i = 0; i < pagevec_count(&pvec); i++) {
473                         struct page *page = pvec.pages[i];
474
475                         index = indices[i];
476                         if (index >= end)
477                                 break;
478
479                         if (radix_tree_exceptional_entry(page)) {
480                                 if (unfalloc)
481                                         continue;
482                                 nr_swaps_freed += !shmem_free_swap(mapping,
483                                                                 index, page);
484                                 continue;
485                         }
486
487                         if (!trylock_page(page))
488                                 continue;
489                         if (!unfalloc || !PageUptodate(page)) {
490                                 if (page->mapping == mapping) {
491                                         VM_BUG_ON(PageWriteback(page));
492                                         truncate_inode_page(mapping, page);
493                                 }
494                         }
495                         unlock_page(page);
496                 }
497                 shmem_deswap_pagevec(&pvec);
498                 pagevec_release(&pvec);
499                 mem_cgroup_uncharge_end();
500                 cond_resched();
501                 index++;
502         }
503
504         if (partial_start) {
505                 struct page *page = NULL;
506                 shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
507                 if (page) {
508                         unsigned int top = PAGE_CACHE_SIZE;
509                         if (start > end) {
510                                 top = partial_end;
511                                 partial_end = 0;
512                         }
513                         zero_user_segment(page, partial_start, top);
514                         set_page_dirty(page);
515                         unlock_page(page);
516                         page_cache_release(page);
517                 }
518         }
519         if (partial_end) {
520                 struct page *page = NULL;
521                 shmem_getpage(inode, end, &page, SGP_READ, NULL);
522                 if (page) {
523                         zero_user_segment(page, 0, partial_end);
524                         set_page_dirty(page);
525                         unlock_page(page);
526                         page_cache_release(page);
527                 }
528         }
529         if (start >= end)
530                 return;
531
532         index = start;
533         for ( ; ; ) {
534                 cond_resched();
535                 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
536                                 min(end - index, (pgoff_t)PAGEVEC_SIZE),
537                                                         pvec.pages, indices);
538                 if (!pvec.nr) {
539                         if (index == start || unfalloc)
540                                 break;
541                         index = start;
542                         continue;
543                 }
544                 if ((index == start || unfalloc) && indices[0] >= end) {
545                         shmem_deswap_pagevec(&pvec);
546                         pagevec_release(&pvec);
547                         break;
548                 }
549                 mem_cgroup_uncharge_start();
550                 for (i = 0; i < pagevec_count(&pvec); i++) {
551                         struct page *page = pvec.pages[i];
552
553                         index = indices[i];
554                         if (index >= end)
555                                 break;
556
557                         if (radix_tree_exceptional_entry(page)) {
558                                 if (unfalloc)
559                                         continue;
560                                 nr_swaps_freed += !shmem_free_swap(mapping,
561                                                                 index, page);
562                                 continue;
563                         }
564
565                         lock_page(page);
566                         if (!unfalloc || !PageUptodate(page)) {
567                                 if (page->mapping == mapping) {
568                                         VM_BUG_ON(PageWriteback(page));
569                                         truncate_inode_page(mapping, page);
570                                 }
571                         }
572                         unlock_page(page);
573                 }
574                 shmem_deswap_pagevec(&pvec);
575                 pagevec_release(&pvec);
576                 mem_cgroup_uncharge_end();
577                 index++;
578         }
579
580         spin_lock(&info->lock);
581         info->swapped -= nr_swaps_freed;
582         shmem_recalc_inode(inode);
583         spin_unlock(&info->lock);
584 }
585
586 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
587 {
588         shmem_undo_range(inode, lstart, lend, false);
589         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
590 }
591 EXPORT_SYMBOL_GPL(shmem_truncate_range);
592
593 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
594 {
595         struct inode *inode = dentry->d_inode;
596         int error;
597
598         error = inode_change_ok(inode, attr);
599         if (error)
600                 return error;
601
602         if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
603                 loff_t oldsize = inode->i_size;
604                 loff_t newsize = attr->ia_size;
605
606                 if (newsize != oldsize) {
607                         i_size_write(inode, newsize);
608                         inode->i_ctime = inode->i_mtime = CURRENT_TIME;
609                 }
610                 if (newsize < oldsize) {
611                         loff_t holebegin = round_up(newsize, PAGE_SIZE);
612                         unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
613                         shmem_truncate_range(inode, newsize, (loff_t)-1);
614                         /* unmap again to remove racily COWed private pages */
615                         unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
616                 }
617         }
618
619         setattr_copy(inode, attr);
620 #ifdef CONFIG_TMPFS_POSIX_ACL
621         if (attr->ia_valid & ATTR_MODE)
622                 error = generic_acl_chmod(inode);
623 #endif
624         return error;
625 }
626
627 static void shmem_evict_inode(struct inode *inode)
628 {
629         struct shmem_inode_info *info = SHMEM_I(inode);
630         struct shmem_xattr *xattr, *nxattr;
631
632         if (inode->i_mapping->a_ops == &shmem_aops) {
633                 shmem_unacct_size(info->flags, inode->i_size);
634                 inode->i_size = 0;
635                 shmem_truncate_range(inode, 0, (loff_t)-1);
636                 if (!list_empty(&info->swaplist)) {
637                         mutex_lock(&shmem_swaplist_mutex);
638                         list_del_init(&info->swaplist);
639                         mutex_unlock(&shmem_swaplist_mutex);
640                 }
641         } else
642                 kfree(info->symlink);
643
644         list_for_each_entry_safe(xattr, nxattr, &info->xattr_list, list) {
645                 kfree(xattr->name);
646                 kfree(xattr);
647         }
648         BUG_ON(inode->i_blocks);
649         shmem_free_inode(inode->i_sb);
650         clear_inode(inode);
651 }
652
653 /*
654  * If swap found in inode, free it and move page from swapcache to filecache.
655  */
656 static int shmem_unuse_inode(struct shmem_inode_info *info,
657                              swp_entry_t swap, struct page **pagep)
658 {
659         struct address_space *mapping = info->vfs_inode.i_mapping;
660         void *radswap;
661         pgoff_t index;
662         gfp_t gfp;
663         int error = 0;
664
665         radswap = swp_to_radix_entry(swap);
666         index = radix_tree_locate_item(&mapping->page_tree, radswap);
667         if (index == -1)
668                 return 0;
669
670         /*
671          * Move _head_ to start search for next from here.
672          * But be careful: shmem_evict_inode checks list_empty without taking
673          * mutex, and there's an instant in list_move_tail when info->swaplist
674          * would appear empty, if it were the only one on shmem_swaplist.
675          */
676         if (shmem_swaplist.next != &info->swaplist)
677                 list_move_tail(&shmem_swaplist, &info->swaplist);
678
679         gfp = mapping_gfp_mask(mapping);
680         if (shmem_should_replace_page(*pagep, gfp)) {
681                 mutex_unlock(&shmem_swaplist_mutex);
682                 error = shmem_replace_page(pagep, gfp, info, index);
683                 mutex_lock(&shmem_swaplist_mutex);
684                 /*
685                  * We needed to drop mutex to make that restrictive page
686                  * allocation; but the inode might already be freed by now,
687                  * and we cannot refer to inode or mapping or info to check.
688                  * However, we do hold page lock on the PageSwapCache page,
689                  * so can check if that still has our reference remaining.
690                  */
691                 if (!page_swapcount(*pagep))
692                         error = -ENOENT;
693         }
694
695         /*
696          * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
697          * but also to hold up shmem_evict_inode(): so inode cannot be freed
698          * beneath us (pagelock doesn't help until the page is in pagecache).
699          */
700         if (!error)
701                 error = shmem_add_to_page_cache(*pagep, mapping, index,
702                                                 GFP_NOWAIT, radswap);
703         if (error != -ENOMEM) {
704                 /*
705                  * Truncation and eviction use free_swap_and_cache(), which
706                  * only does trylock page: if we raced, best clean up here.
707                  */
708                 delete_from_swap_cache(*pagep);
709                 set_page_dirty(*pagep);
710                 if (!error) {
711                         spin_lock(&info->lock);
712                         info->swapped--;
713                         spin_unlock(&info->lock);
714                         swap_free(swap);
715                 }
716                 error = 1;      /* not an error, but entry was found */
717         }
718         return error;
719 }
720
721 /*
722  * Search through swapped inodes to find and replace swap by page.
723  */
724 int shmem_unuse(swp_entry_t swap, struct page *page)
725 {
726         struct list_head *this, *next;
727         struct shmem_inode_info *info;
728         int found = 0;
729         int error = 0;
730
731         /*
732          * There's a faint possibility that swap page was replaced before
733          * caller locked it: it will come back later with the right page.
734          */
735         if (unlikely(!PageSwapCache(page)))
736                 goto out;
737
738         /*
739          * Charge page using GFP_KERNEL while we can wait, before taking
740          * the shmem_swaplist_mutex which might hold up shmem_writepage().
741          * Charged back to the user (not to caller) when swap account is used.
742          */
743         error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
744         if (error)
745                 goto out;
746         /* No radix_tree_preload: swap entry keeps a place for page in tree */
747
748         mutex_lock(&shmem_swaplist_mutex);
749         list_for_each_safe(this, next, &shmem_swaplist) {
750                 info = list_entry(this, struct shmem_inode_info, swaplist);
751                 if (info->swapped)
752                         found = shmem_unuse_inode(info, swap, &page);
753                 else
754                         list_del_init(&info->swaplist);
755                 cond_resched();
756                 if (found)
757                         break;
758         }
759         mutex_unlock(&shmem_swaplist_mutex);
760
761         if (found < 0)
762                 error = found;
763 out:
764         unlock_page(page);
765         page_cache_release(page);
766         return error;
767 }
768
769 /*
770  * Move the page from the page cache to the swap cache.
771  */
772 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
773 {
774         struct shmem_inode_info *info;
775         struct address_space *mapping;
776         struct inode *inode;
777         swp_entry_t swap;
778         pgoff_t index;
779
780         BUG_ON(!PageLocked(page));
781         mapping = page->mapping;
782         index = page->index;
783         inode = mapping->host;
784         info = SHMEM_I(inode);
785         if (info->flags & VM_LOCKED)
786                 goto redirty;
787         if (!total_swap_pages)
788                 goto redirty;
789
790         /*
791          * shmem_backing_dev_info's capabilities prevent regular writeback or
792          * sync from ever calling shmem_writepage; but a stacking filesystem
793          * might use ->writepage of its underlying filesystem, in which case
794          * tmpfs should write out to swap only in response to memory pressure,
795          * and not for the writeback threads or sync.
796          */
797         if (!wbc->for_reclaim) {
798                 WARN_ON_ONCE(1);        /* Still happens? Tell us about it! */
799                 goto redirty;
800         }
801
802         /*
803          * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
804          * value into swapfile.c, the only way we can correctly account for a
805          * fallocated page arriving here is now to initialize it and write it.
806          *
807          * That's okay for a page already fallocated earlier, but if we have
808          * not yet completed the fallocation, then (a) we want to keep track
809          * of this page in case we have to undo it, and (b) it may not be a
810          * good idea to continue anyway, once we're pushing into swap.  So
811          * reactivate the page, and let shmem_fallocate() quit when too many.
812          */
813         if (!PageUptodate(page)) {
814                 if (inode->i_private) {
815                         struct shmem_falloc *shmem_falloc;
816                         spin_lock(&inode->i_lock);
817                         shmem_falloc = inode->i_private;
818                         if (shmem_falloc &&
819                             index >= shmem_falloc->start &&
820                             index < shmem_falloc->next)
821                                 shmem_falloc->nr_unswapped++;
822                         else
823                                 shmem_falloc = NULL;
824                         spin_unlock(&inode->i_lock);
825                         if (shmem_falloc)
826                                 goto redirty;
827                 }
828                 clear_highpage(page);
829                 flush_dcache_page(page);
830                 SetPageUptodate(page);
831         }
832
833         swap = get_swap_page();
834         if (!swap.val)
835                 goto redirty;
836
837         /*
838          * Add inode to shmem_unuse()'s list of swapped-out inodes,
839          * if it's not already there.  Do it now before the page is
840          * moved to swap cache, when its pagelock no longer protects
841          * the inode from eviction.  But don't unlock the mutex until
842          * we've incremented swapped, because shmem_unuse_inode() will
843          * prune a !swapped inode from the swaplist under this mutex.
844          */
845         mutex_lock(&shmem_swaplist_mutex);
846         if (list_empty(&info->swaplist))
847                 list_add_tail(&info->swaplist, &shmem_swaplist);
848
849         if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
850                 swap_shmem_alloc(swap);
851                 shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
852
853                 spin_lock(&info->lock);
854                 info->swapped++;
855                 shmem_recalc_inode(inode);
856                 spin_unlock(&info->lock);
857
858                 mutex_unlock(&shmem_swaplist_mutex);
859                 BUG_ON(page_mapped(page));
860                 swap_writepage(page, wbc);
861                 return 0;
862         }
863
864         mutex_unlock(&shmem_swaplist_mutex);
865         swapcache_free(swap, NULL);
866 redirty:
867         set_page_dirty(page);
868         if (wbc->for_reclaim)
869                 return AOP_WRITEPAGE_ACTIVATE;  /* Return with page locked */
870         unlock_page(page);
871         return 0;
872 }
873
874 #ifdef CONFIG_NUMA
875 #ifdef CONFIG_TMPFS
876 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
877 {
878         char buffer[64];
879
880         if (!mpol || mpol->mode == MPOL_DEFAULT)
881                 return;         /* show nothing */
882
883         mpol_to_str(buffer, sizeof(buffer), mpol, 1);
884
885         seq_printf(seq, ",mpol=%s", buffer);
886 }
887
888 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
889 {
890         struct mempolicy *mpol = NULL;
891         if (sbinfo->mpol) {
892                 spin_lock(&sbinfo->stat_lock);  /* prevent replace/use races */
893                 mpol = sbinfo->mpol;
894                 mpol_get(mpol);
895                 spin_unlock(&sbinfo->stat_lock);
896         }
897         return mpol;
898 }
899 #endif /* CONFIG_TMPFS */
900
901 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
902                         struct shmem_inode_info *info, pgoff_t index)
903 {
904         struct mempolicy mpol, *spol;
905         struct vm_area_struct pvma;
906
907         spol = mpol_cond_copy(&mpol,
908                         mpol_shared_policy_lookup(&info->policy, index));
909
910         /* Create a pseudo vma that just contains the policy */
911         pvma.vm_start = 0;
912         pvma.vm_pgoff = index;
913         pvma.vm_ops = NULL;
914         pvma.vm_policy = spol;
915         return swapin_readahead(swap, gfp, &pvma, 0);
916 }
917
918 static struct page *shmem_alloc_page(gfp_t gfp,
919                         struct shmem_inode_info *info, pgoff_t index)
920 {
921         struct vm_area_struct pvma;
922
923         /* Create a pseudo vma that just contains the policy */
924         pvma.vm_start = 0;
925         pvma.vm_pgoff = index;
926         pvma.vm_ops = NULL;
927         pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
928
929         /*
930          * alloc_page_vma() will drop the shared policy reference
931          */
932         return alloc_page_vma(gfp, &pvma, 0);
933 }
934 #else /* !CONFIG_NUMA */
935 #ifdef CONFIG_TMPFS
936 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
937 {
938 }
939 #endif /* CONFIG_TMPFS */
940
941 static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
942                         struct shmem_inode_info *info, pgoff_t index)
943 {
944         return swapin_readahead(swap, gfp, NULL, 0);
945 }
946
947 static inline struct page *shmem_alloc_page(gfp_t gfp,
948                         struct shmem_inode_info *info, pgoff_t index)
949 {
950         return alloc_page(gfp);
951 }
952 #endif /* CONFIG_NUMA */
953
954 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
955 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
956 {
957         return NULL;
958 }
959 #endif
960
961 /*
962  * When a page is moved from swapcache to shmem filecache (either by the
963  * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
964  * shmem_unuse_inode()), it may have been read in earlier from swap, in
965  * ignorance of the mapping it belongs to.  If that mapping has special
966  * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
967  * we may need to copy to a suitable page before moving to filecache.
968  *
969  * In a future release, this may well be extended to respect cpuset and
970  * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
971  * but for now it is a simple matter of zone.
972  */
973 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
974 {
975         return page_zonenum(page) > gfp_zone(gfp);
976 }
977
978 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
979                                 struct shmem_inode_info *info, pgoff_t index)
980 {
981         struct page *oldpage, *newpage;
982         struct address_space *swap_mapping;
983         pgoff_t swap_index;
984         int error;
985
986         oldpage = *pagep;
987         swap_index = page_private(oldpage);
988         swap_mapping = page_mapping(oldpage);
989
990         /*
991          * We have arrived here because our zones are constrained, so don't
992          * limit chance of success by further cpuset and node constraints.
993          */
994         gfp &= ~GFP_CONSTRAINT_MASK;
995         newpage = shmem_alloc_page(gfp, info, index);
996         if (!newpage)
997                 return -ENOMEM;
998         VM_BUG_ON(shmem_should_replace_page(newpage, gfp));
999
1000         *pagep = newpage;
1001         page_cache_get(newpage);
1002         copy_highpage(newpage, oldpage);
1003
1004         VM_BUG_ON(!PageLocked(oldpage));
1005         __set_page_locked(newpage);
1006         VM_BUG_ON(!PageUptodate(oldpage));
1007         SetPageUptodate(newpage);
1008         VM_BUG_ON(!PageSwapBacked(oldpage));
1009         SetPageSwapBacked(newpage);
1010         VM_BUG_ON(!swap_index);
1011         set_page_private(newpage, swap_index);
1012         VM_BUG_ON(!PageSwapCache(oldpage));
1013         SetPageSwapCache(newpage);
1014
1015         /*
1016          * Our caller will very soon move newpage out of swapcache, but it's
1017          * a nice clean interface for us to replace oldpage by newpage there.
1018          */
1019         spin_lock_irq(&swap_mapping->tree_lock);
1020         error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1021                                                                    newpage);
1022         __inc_zone_page_state(newpage, NR_FILE_PAGES);
1023         __dec_zone_page_state(oldpage, NR_FILE_PAGES);
1024         spin_unlock_irq(&swap_mapping->tree_lock);
1025         BUG_ON(error);
1026
1027         mem_cgroup_replace_page_cache(oldpage, newpage);
1028         lru_cache_add_anon(newpage);
1029
1030         ClearPageSwapCache(oldpage);
1031         set_page_private(oldpage, 0);
1032
1033         unlock_page(oldpage);
1034         page_cache_release(oldpage);
1035         page_cache_release(oldpage);
1036         return 0;
1037 }
1038
1039 /*
1040  * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1041  *
1042  * If we allocate a new one we do not mark it dirty. That's up to the
1043  * vm. If we swap it in we mark it dirty since we also free the swap
1044  * entry since a page cannot live in both the swap and page cache
1045  */
1046 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1047         struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
1048 {
1049         struct address_space *mapping = inode->i_mapping;
1050         struct shmem_inode_info *info;
1051         struct shmem_sb_info *sbinfo;
1052         struct page *page;
1053         swp_entry_t swap;
1054         int error;
1055         int once = 0;
1056         int alloced = 0;
1057
1058         if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
1059                 return -EFBIG;
1060 repeat:
1061         swap.val = 0;
1062         page = find_lock_page(mapping, index);
1063         if (radix_tree_exceptional_entry(page)) {
1064                 swap = radix_to_swp_entry(page);
1065                 page = NULL;
1066         }
1067
1068         if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1069             ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1070                 error = -EINVAL;
1071                 goto failed;
1072         }
1073
1074         /* fallocated page? */
1075         if (page && !PageUptodate(page)) {
1076                 if (sgp != SGP_READ)
1077                         goto clear;
1078                 unlock_page(page);
1079                 page_cache_release(page);
1080                 page = NULL;
1081         }
1082         if (page || (sgp == SGP_READ && !swap.val)) {
1083                 *pagep = page;
1084                 return 0;
1085         }
1086
1087         /*
1088          * Fast cache lookup did not find it:
1089          * bring it back from swap or allocate.
1090          */
1091         info = SHMEM_I(inode);
1092         sbinfo = SHMEM_SB(inode->i_sb);
1093
1094         if (swap.val) {
1095                 /* Look it up and read it in.. */
1096                 page = lookup_swap_cache(swap);
1097                 if (!page) {
1098                         /* here we actually do the io */
1099                         if (fault_type)
1100                                 *fault_type |= VM_FAULT_MAJOR;
1101                         page = shmem_swapin(swap, gfp, info, index);
1102                         if (!page) {
1103                                 error = -ENOMEM;
1104                                 goto failed;
1105                         }
1106                 }
1107
1108                 /* We have to do this with page locked to prevent races */
1109                 lock_page(page);
1110                 if (!PageSwapCache(page) || page->mapping) {
1111                         error = -EEXIST;        /* try again */
1112                         goto failed;
1113                 }
1114                 if (!PageUptodate(page)) {
1115                         error = -EIO;
1116                         goto failed;
1117                 }
1118                 wait_on_page_writeback(page);
1119
1120                 if (shmem_should_replace_page(page, gfp)) {
1121                         error = shmem_replace_page(&page, gfp, info, index);
1122                         if (error)
1123                                 goto failed;
1124                 }
1125
1126                 error = mem_cgroup_cache_charge(page, current->mm,
1127                                                 gfp & GFP_RECLAIM_MASK);
1128                 if (!error)
1129                         error = shmem_add_to_page_cache(page, mapping, index,
1130                                                 gfp, swp_to_radix_entry(swap));
1131                 if (error)
1132                         goto failed;
1133
1134                 spin_lock(&info->lock);
1135                 info->swapped--;
1136                 shmem_recalc_inode(inode);
1137                 spin_unlock(&info->lock);
1138
1139                 delete_from_swap_cache(page);
1140                 set_page_dirty(page);
1141                 swap_free(swap);
1142
1143         } else {
1144                 if (shmem_acct_block(info->flags)) {
1145                         error = -ENOSPC;
1146                         goto failed;
1147                 }
1148                 if (sbinfo->max_blocks) {
1149                         if (percpu_counter_compare(&sbinfo->used_blocks,
1150                                                 sbinfo->max_blocks) >= 0) {
1151                                 error = -ENOSPC;
1152                                 goto unacct;
1153                         }
1154                         percpu_counter_inc(&sbinfo->used_blocks);
1155                 }
1156
1157                 page = shmem_alloc_page(gfp, info, index);
1158                 if (!page) {
1159                         error = -ENOMEM;
1160                         goto decused;
1161                 }
1162
1163                 SetPageSwapBacked(page);
1164                 __set_page_locked(page);
1165                 error = mem_cgroup_cache_charge(page, current->mm,
1166                                                 gfp & GFP_RECLAIM_MASK);
1167                 if (!error)
1168                         error = shmem_add_to_page_cache(page, mapping, index,
1169                                                 gfp, NULL);
1170                 if (error)
1171                         goto decused;
1172                 lru_cache_add_anon(page);
1173
1174                 spin_lock(&info->lock);
1175                 info->alloced++;
1176                 inode->i_blocks += BLOCKS_PER_PAGE;
1177                 shmem_recalc_inode(inode);
1178                 spin_unlock(&info->lock);
1179                 alloced = true;
1180
1181                 /*
1182                  * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1183                  */
1184                 if (sgp == SGP_FALLOC)
1185                         sgp = SGP_WRITE;
1186 clear:
1187                 /*
1188                  * Let SGP_WRITE caller clear ends if write does not fill page;
1189                  * but SGP_FALLOC on a page fallocated earlier must initialize
1190                  * it now, lest undo on failure cancel our earlier guarantee.
1191                  */
1192                 if (sgp != SGP_WRITE) {
1193                         clear_highpage(page);
1194                         flush_dcache_page(page);
1195                         SetPageUptodate(page);
1196                 }
1197                 if (sgp == SGP_DIRTY)
1198                         set_page_dirty(page);
1199         }
1200
1201         /* Perhaps the file has been truncated since we checked */
1202         if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1203             ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1204                 error = -EINVAL;
1205                 if (alloced)
1206                         goto trunc;
1207                 else
1208                         goto failed;
1209         }
1210         *pagep = page;
1211         return 0;
1212
1213         /*
1214          * Error recovery.
1215          */
1216 trunc:
1217         info = SHMEM_I(inode);
1218         ClearPageDirty(page);
1219         delete_from_page_cache(page);
1220         spin_lock(&info->lock);
1221         info->alloced--;
1222         inode->i_blocks -= BLOCKS_PER_PAGE;
1223         spin_unlock(&info->lock);
1224 decused:
1225         sbinfo = SHMEM_SB(inode->i_sb);
1226         if (sbinfo->max_blocks)
1227                 percpu_counter_add(&sbinfo->used_blocks, -1);
1228 unacct:
1229         shmem_unacct_blocks(info->flags, 1);
1230 failed:
1231         if (swap.val && error != -EINVAL) {
1232                 struct page *test = find_get_page(mapping, index);
1233                 if (test && !radix_tree_exceptional_entry(test))
1234                         page_cache_release(test);
1235                 /* Have another try if the entry has changed */
1236                 if (test != swp_to_radix_entry(swap))
1237                         error = -EEXIST;
1238         }
1239         if (page) {
1240                 unlock_page(page);
1241                 page_cache_release(page);
1242         }
1243         if (error == -ENOSPC && !once++) {
1244                 info = SHMEM_I(inode);
1245                 spin_lock(&info->lock);
1246                 shmem_recalc_inode(inode);
1247                 spin_unlock(&info->lock);
1248                 goto repeat;
1249         }
1250         if (error == -EEXIST)
1251                 goto repeat;
1252         return error;
1253 }
1254
1255 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1256 {
1257         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1258         int error;
1259         int ret = VM_FAULT_LOCKED;
1260
1261         error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1262         if (error)
1263                 return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1264
1265         if (ret & VM_FAULT_MAJOR) {
1266                 count_vm_event(PGMAJFAULT);
1267                 mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1268         }
1269         return ret;
1270 }
1271
1272 #ifdef CONFIG_NUMA
1273 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
1274 {
1275         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1276         return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
1277 }
1278
1279 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1280                                           unsigned long addr)
1281 {
1282         struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
1283         pgoff_t index;
1284
1285         index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1286         return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
1287 }
1288 #endif
1289
1290 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1291 {
1292         struct inode *inode = file->f_path.dentry->d_inode;
1293         struct shmem_inode_info *info = SHMEM_I(inode);
1294         int retval = -ENOMEM;
1295
1296         spin_lock(&info->lock);
1297         if (lock && !(info->flags & VM_LOCKED)) {
1298                 if (!user_shm_lock(inode->i_size, user))
1299                         goto out_nomem;
1300                 info->flags |= VM_LOCKED;
1301                 mapping_set_unevictable(file->f_mapping);
1302         }
1303         if (!lock && (info->flags & VM_LOCKED) && user) {
1304                 user_shm_unlock(inode->i_size, user);
1305                 info->flags &= ~VM_LOCKED;
1306                 mapping_clear_unevictable(file->f_mapping);
1307         }
1308         retval = 0;
1309
1310 out_nomem:
1311         spin_unlock(&info->lock);
1312         return retval;
1313 }
1314
1315 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1316 {
1317         file_accessed(file);
1318         vma->vm_ops = &shmem_vm_ops;
1319         vma->vm_flags |= VM_CAN_NONLINEAR;
1320         return 0;
1321 }
1322
1323 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1324                                      umode_t mode, dev_t dev, unsigned long flags)
1325 {
1326         struct inode *inode;
1327         struct shmem_inode_info *info;
1328         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1329
1330         if (shmem_reserve_inode(sb))
1331                 return NULL;
1332
1333         inode = new_inode(sb);
1334         if (inode) {
1335                 inode->i_ino = get_next_ino();
1336                 inode_init_owner(inode, dir, mode);
1337                 inode->i_blocks = 0;
1338                 inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1339                 inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1340                 inode->i_generation = get_seconds();
1341                 info = SHMEM_I(inode);
1342                 memset(info, 0, (char *)inode - (char *)info);
1343                 spin_lock_init(&info->lock);
1344                 info->flags = flags & VM_NORESERVE;
1345                 INIT_LIST_HEAD(&info->swaplist);
1346                 INIT_LIST_HEAD(&info->xattr_list);
1347                 cache_no_acl(inode);
1348
1349                 switch (mode & S_IFMT) {
1350                 default:
1351                         inode->i_op = &shmem_special_inode_operations;
1352                         init_special_inode(inode, mode, dev);
1353                         break;
1354                 case S_IFREG:
1355                         inode->i_mapping->a_ops = &shmem_aops;
1356                         inode->i_op = &shmem_inode_operations;
1357                         inode->i_fop = &shmem_file_operations;
1358                         mpol_shared_policy_init(&info->policy,
1359                                                  shmem_get_sbmpol(sbinfo));
1360                         break;
1361                 case S_IFDIR:
1362                         inc_nlink(inode);
1363                         /* Some things misbehave if size == 0 on a directory */
1364                         inode->i_size = 2 * BOGO_DIRENT_SIZE;
1365                         inode->i_op = &shmem_dir_inode_operations;
1366                         inode->i_fop = &simple_dir_operations;
1367                         break;
1368                 case S_IFLNK:
1369                         /*
1370                          * Must not load anything in the rbtree,
1371                          * mpol_free_shared_policy will not be called.
1372                          */
1373                         mpol_shared_policy_init(&info->policy, NULL);
1374                         break;
1375                 }
1376         } else
1377                 shmem_free_inode(sb);
1378         return inode;
1379 }
1380
1381 #ifdef CONFIG_TMPFS
1382 static const struct inode_operations shmem_symlink_inode_operations;
1383 static const struct inode_operations shmem_short_symlink_operations;
1384
1385 #ifdef CONFIG_TMPFS_XATTR
1386 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1387 #else
1388 #define shmem_initxattrs NULL
1389 #endif
1390
1391 static int
1392 shmem_write_begin(struct file *file, struct address_space *mapping,
1393                         loff_t pos, unsigned len, unsigned flags,
1394                         struct page **pagep, void **fsdata)
1395 {
1396         struct inode *inode = mapping->host;
1397         pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1398         return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1399 }
1400
1401 static int
1402 shmem_write_end(struct file *file, struct address_space *mapping,
1403                         loff_t pos, unsigned len, unsigned copied,
1404                         struct page *page, void *fsdata)
1405 {
1406         struct inode *inode = mapping->host;
1407
1408         if (pos + copied > inode->i_size)
1409                 i_size_write(inode, pos + copied);
1410
1411         if (!PageUptodate(page)) {
1412                 if (copied < PAGE_CACHE_SIZE) {
1413                         unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1414                         zero_user_segments(page, 0, from,
1415                                         from + copied, PAGE_CACHE_SIZE);
1416                 }
1417                 SetPageUptodate(page);
1418         }
1419         set_page_dirty(page);
1420         unlock_page(page);
1421         page_cache_release(page);
1422
1423         return copied;
1424 }
1425
1426 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1427 {
1428         struct inode *inode = filp->f_path.dentry->d_inode;
1429         struct address_space *mapping = inode->i_mapping;
1430         pgoff_t index;
1431         unsigned long offset;
1432         enum sgp_type sgp = SGP_READ;
1433
1434         /*
1435          * Might this read be for a stacking filesystem?  Then when reading
1436          * holes of a sparse file, we actually need to allocate those pages,
1437          * and even mark them dirty, so it cannot exceed the max_blocks limit.
1438          */
1439         if (segment_eq(get_fs(), KERNEL_DS))
1440                 sgp = SGP_DIRTY;
1441
1442         index = *ppos >> PAGE_CACHE_SHIFT;
1443         offset = *ppos & ~PAGE_CACHE_MASK;
1444
1445         for (;;) {
1446                 struct page *page = NULL;
1447                 pgoff_t end_index;
1448                 unsigned long nr, ret;
1449                 loff_t i_size = i_size_read(inode);
1450
1451                 end_index = i_size >> PAGE_CACHE_SHIFT;
1452                 if (index > end_index)
1453                         break;
1454                 if (index == end_index) {
1455                         nr = i_size & ~PAGE_CACHE_MASK;
1456                         if (nr <= offset)
1457                                 break;
1458                 }
1459
1460                 desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1461                 if (desc->error) {
1462                         if (desc->error == -EINVAL)
1463                                 desc->error = 0;
1464                         break;
1465                 }
1466                 if (page)
1467                         unlock_page(page);
1468
1469                 /*
1470                  * We must evaluate after, since reads (unlike writes)
1471                  * are called without i_mutex protection against truncate
1472                  */
1473                 nr = PAGE_CACHE_SIZE;
1474                 i_size = i_size_read(inode);
1475                 end_index = i_size >> PAGE_CACHE_SHIFT;
1476                 if (index == end_index) {
1477                         nr = i_size & ~PAGE_CACHE_MASK;
1478                         if (nr <= offset) {
1479                                 if (page)
1480                                         page_cache_release(page);
1481                                 break;
1482                         }
1483                 }
1484                 nr -= offset;
1485
1486                 if (page) {
1487                         /*
1488                          * If users can be writing to this page using arbitrary
1489                          * virtual addresses, take care about potential aliasing
1490                          * before reading the page on the kernel side.
1491                          */
1492                         if (mapping_writably_mapped(mapping))
1493                                 flush_dcache_page(page);
1494                         /*
1495                          * Mark the page accessed if we read the beginning.
1496                          */
1497                         if (!offset)
1498                                 mark_page_accessed(page);
1499                 } else {
1500                         page = ZERO_PAGE(0);
1501                         page_cache_get(page);
1502                 }
1503
1504                 /*
1505                  * Ok, we have the page, and it's up-to-date, so
1506                  * now we can copy it to user space...
1507                  *
1508                  * The actor routine returns how many bytes were actually used..
1509                  * NOTE! This may not be the same as how much of a user buffer
1510                  * we filled up (we may be padding etc), so we can only update
1511                  * "pos" here (the actor routine has to update the user buffer
1512                  * pointers and the remaining count).
1513                  */
1514                 ret = actor(desc, page, offset, nr);
1515                 offset += ret;
1516                 index += offset >> PAGE_CACHE_SHIFT;
1517                 offset &= ~PAGE_CACHE_MASK;
1518
1519                 page_cache_release(page);
1520                 if (ret != nr || !desc->count)
1521                         break;
1522
1523                 cond_resched();
1524         }
1525
1526         *ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1527         file_accessed(filp);
1528 }
1529
1530 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1531                 const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1532 {
1533         struct file *filp = iocb->ki_filp;
1534         ssize_t retval;
1535         unsigned long seg;
1536         size_t count;
1537         loff_t *ppos = &iocb->ki_pos;
1538
1539         retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1540         if (retval)
1541                 return retval;
1542
1543         for (seg = 0; seg < nr_segs; seg++) {
1544                 read_descriptor_t desc;
1545
1546                 desc.written = 0;
1547                 desc.arg.buf = iov[seg].iov_base;
1548                 desc.count = iov[seg].iov_len;
1549                 if (desc.count == 0)
1550                         continue;
1551                 desc.error = 0;
1552                 do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1553                 retval += desc.written;
1554                 if (desc.error) {
1555                         retval = retval ?: desc.error;
1556                         break;
1557                 }
1558                 if (desc.count > 0)
1559                         break;
1560         }
1561         return retval;
1562 }
1563
1564 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1565                                 struct pipe_inode_info *pipe, size_t len,
1566                                 unsigned int flags)
1567 {
1568         struct address_space *mapping = in->f_mapping;
1569         struct inode *inode = mapping->host;
1570         unsigned int loff, nr_pages, req_pages;
1571         struct page *pages[PIPE_DEF_BUFFERS];
1572         struct partial_page partial[PIPE_DEF_BUFFERS];
1573         struct page *page;
1574         pgoff_t index, end_index;
1575         loff_t isize, left;
1576         int error, page_nr;
1577         struct splice_pipe_desc spd = {
1578                 .pages = pages,
1579                 .partial = partial,
1580                 .flags = flags,
1581                 .ops = &page_cache_pipe_buf_ops,
1582                 .spd_release = spd_release_page,
1583         };
1584
1585         isize = i_size_read(inode);
1586         if (unlikely(*ppos >= isize))
1587                 return 0;
1588
1589         left = isize - *ppos;
1590         if (unlikely(left < len))
1591                 len = left;
1592
1593         if (splice_grow_spd(pipe, &spd))
1594                 return -ENOMEM;
1595
1596         index = *ppos >> PAGE_CACHE_SHIFT;
1597         loff = *ppos & ~PAGE_CACHE_MASK;
1598         req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1599         nr_pages = min(req_pages, pipe->buffers);
1600
1601         spd.nr_pages = find_get_pages_contig(mapping, index,
1602                                                 nr_pages, spd.pages);
1603         index += spd.nr_pages;
1604         error = 0;
1605
1606         while (spd.nr_pages < nr_pages) {
1607                 error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1608                 if (error)
1609                         break;
1610                 unlock_page(page);
1611                 spd.pages[spd.nr_pages++] = page;
1612                 index++;
1613         }
1614
1615         index = *ppos >> PAGE_CACHE_SHIFT;
1616         nr_pages = spd.nr_pages;
1617         spd.nr_pages = 0;
1618
1619         for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1620                 unsigned int this_len;
1621
1622                 if (!len)
1623                         break;
1624
1625                 this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1626                 page = spd.pages[page_nr];
1627
1628                 if (!PageUptodate(page) || page->mapping != mapping) {
1629                         error = shmem_getpage(inode, index, &page,
1630                                                         SGP_CACHE, NULL);
1631                         if (error)
1632                                 break;
1633                         unlock_page(page);
1634                         page_cache_release(spd.pages[page_nr]);
1635                         spd.pages[page_nr] = page;
1636                 }
1637
1638                 isize = i_size_read(inode);
1639                 end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1640                 if (unlikely(!isize || index > end_index))
1641                         break;
1642
1643                 if (end_index == index) {
1644                         unsigned int plen;
1645
1646                         plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1647                         if (plen <= loff)
1648                                 break;
1649
1650                         this_len = min(this_len, plen - loff);
1651                         len = this_len;
1652                 }
1653
1654                 spd.partial[page_nr].offset = loff;
1655                 spd.partial[page_nr].len = this_len;
1656                 len -= this_len;
1657                 loff = 0;
1658                 spd.nr_pages++;
1659                 index++;
1660         }
1661
1662         while (page_nr < nr_pages)
1663                 page_cache_release(spd.pages[page_nr++]);
1664
1665         if (spd.nr_pages)
1666                 error = splice_to_pipe(pipe, &spd);
1667
1668         splice_shrink_spd(pipe, &spd);
1669
1670         if (error > 0) {
1671                 *ppos += error;
1672                 file_accessed(in);
1673         }
1674         return error;
1675 }
1676
1677 /*
1678  * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1679  */
1680 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
1681                                     pgoff_t index, pgoff_t end, int origin)
1682 {
1683         struct page *page;
1684         struct pagevec pvec;
1685         pgoff_t indices[PAGEVEC_SIZE];
1686         bool done = false;
1687         int i;
1688
1689         pagevec_init(&pvec, 0);
1690         pvec.nr = 1;            /* start small: we may be there already */
1691         while (!done) {
1692                 pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
1693                                         pvec.nr, pvec.pages, indices);
1694                 if (!pvec.nr) {
1695                         if (origin == SEEK_DATA)
1696                                 index = end;
1697                         break;
1698                 }
1699                 for (i = 0; i < pvec.nr; i++, index++) {
1700                         if (index < indices[i]) {
1701                                 if (origin == SEEK_HOLE) {
1702                                         done = true;
1703                                         break;
1704                                 }
1705                                 index = indices[i];
1706                         }
1707                         page = pvec.pages[i];
1708                         if (page && !radix_tree_exceptional_entry(page)) {
1709                                 if (!PageUptodate(page))
1710                                         page = NULL;
1711                         }
1712                         if (index >= end ||
1713                             (page && origin == SEEK_DATA) ||
1714                             (!page && origin == SEEK_HOLE)) {
1715                                 done = true;
1716                                 break;
1717                         }
1718                 }
1719                 shmem_deswap_pagevec(&pvec);
1720                 pagevec_release(&pvec);
1721                 pvec.nr = PAGEVEC_SIZE;
1722                 cond_resched();
1723         }
1724         return index;
1725 }
1726
1727 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int origin)
1728 {
1729         struct address_space *mapping;
1730         struct inode *inode;
1731         pgoff_t start, end;
1732         loff_t new_offset;
1733
1734         if (origin != SEEK_DATA && origin != SEEK_HOLE)
1735                 return generic_file_llseek_size(file, offset, origin,
1736                                                         MAX_LFS_FILESIZE);
1737         mapping = file->f_mapping;
1738         inode = mapping->host;
1739         mutex_lock(&inode->i_mutex);
1740         /* We're holding i_mutex so we can access i_size directly */
1741
1742         if (offset < 0)
1743                 offset = -EINVAL;
1744         else if (offset >= inode->i_size)
1745                 offset = -ENXIO;
1746         else {
1747                 start = offset >> PAGE_CACHE_SHIFT;
1748                 end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1749                 new_offset = shmem_seek_hole_data(mapping, start, end, origin);
1750                 new_offset <<= PAGE_CACHE_SHIFT;
1751                 if (new_offset > offset) {
1752                         if (new_offset < inode->i_size)
1753                                 offset = new_offset;
1754                         else if (origin == SEEK_DATA)
1755                                 offset = -ENXIO;
1756                         else
1757                                 offset = inode->i_size;
1758                 }
1759         }
1760
1761         if (offset >= 0 && offset != file->f_pos) {
1762                 file->f_pos = offset;
1763                 file->f_version = 0;
1764         }
1765         mutex_unlock(&inode->i_mutex);
1766         return offset;
1767 }
1768
1769 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
1770                                                          loff_t len)
1771 {
1772         struct inode *inode = file->f_path.dentry->d_inode;
1773         struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1774         struct shmem_falloc shmem_falloc;
1775         pgoff_t start, index, end;
1776         int error;
1777
1778         mutex_lock(&inode->i_mutex);
1779
1780         if (mode & FALLOC_FL_PUNCH_HOLE) {
1781                 struct address_space *mapping = file->f_mapping;
1782                 loff_t unmap_start = round_up(offset, PAGE_SIZE);
1783                 loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
1784
1785                 if ((u64)unmap_end > (u64)unmap_start)
1786                         unmap_mapping_range(mapping, unmap_start,
1787                                             1 + unmap_end - unmap_start, 0);
1788                 shmem_truncate_range(inode, offset, offset + len - 1);
1789                 /* No need to unmap again: hole-punching leaves COWed pages */
1790                 error = 0;
1791                 goto out;
1792         }
1793
1794         /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1795         error = inode_newsize_ok(inode, offset + len);
1796         if (error)
1797                 goto out;
1798
1799         start = offset >> PAGE_CACHE_SHIFT;
1800         end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1801         /* Try to avoid a swapstorm if len is impossible to satisfy */
1802         if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
1803                 error = -ENOSPC;
1804                 goto out;
1805         }
1806
1807         shmem_falloc.start = start;
1808         shmem_falloc.next  = start;
1809         shmem_falloc.nr_falloced = 0;
1810         shmem_falloc.nr_unswapped = 0;
1811         spin_lock(&inode->i_lock);
1812         inode->i_private = &shmem_falloc;
1813         spin_unlock(&inode->i_lock);
1814
1815         for (index = start; index < end; index++) {
1816                 struct page *page;
1817
1818                 /*
1819                  * Good, the fallocate(2) manpage permits EINTR: we may have
1820                  * been interrupted because we are using up too much memory.
1821                  */
1822                 if (signal_pending(current))
1823                         error = -EINTR;
1824                 else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
1825                         error = -ENOMEM;
1826                 else
1827                         error = shmem_getpage(inode, index, &page, SGP_FALLOC,
1828                                                                         NULL);
1829                 if (error) {
1830                         /* Remove the !PageUptodate pages we added */
1831                         shmem_undo_range(inode,
1832                                 (loff_t)start << PAGE_CACHE_SHIFT,
1833                                 (loff_t)index << PAGE_CACHE_SHIFT, true);
1834                         goto undone;
1835                 }
1836
1837                 /*
1838                  * Inform shmem_writepage() how far we have reached.
1839                  * No need for lock or barrier: we have the page lock.
1840                  */
1841                 shmem_falloc.next++;
1842                 if (!PageUptodate(page))
1843                         shmem_falloc.nr_falloced++;
1844
1845                 /*
1846                  * If !PageUptodate, leave it that way so that freeable pages
1847                  * can be recognized if we need to rollback on error later.
1848                  * But set_page_dirty so that memory pressure will swap rather
1849                  * than free the pages we are allocating (and SGP_CACHE pages
1850                  * might still be clean: we now need to mark those dirty too).
1851                  */
1852                 set_page_dirty(page);
1853                 unlock_page(page);
1854                 page_cache_release(page);
1855                 cond_resched();
1856         }
1857
1858         if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
1859                 i_size_write(inode, offset + len);
1860         inode->i_ctime = CURRENT_TIME;
1861 undone:
1862         spin_lock(&inode->i_lock);
1863         inode->i_private = NULL;
1864         spin_unlock(&inode->i_lock);
1865 out:
1866         mutex_unlock(&inode->i_mutex);
1867         return error;
1868 }
1869
1870 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1871 {
1872         struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1873
1874         buf->f_type = TMPFS_MAGIC;
1875         buf->f_bsize = PAGE_CACHE_SIZE;
1876         buf->f_namelen = NAME_MAX;
1877         if (sbinfo->max_blocks) {
1878                 buf->f_blocks = sbinfo->max_blocks;
1879                 buf->f_bavail =
1880                 buf->f_bfree  = sbinfo->max_blocks -
1881                                 percpu_counter_sum(&sbinfo->used_blocks);
1882         }
1883         if (sbinfo->max_inodes) {
1884                 buf->f_files = sbinfo->max_inodes;
1885                 buf->f_ffree = sbinfo->free_inodes;
1886         }
1887         /* else leave those fields 0 like simple_statfs */
1888         return 0;
1889 }
1890
1891 /*
1892  * File creation. Allocate an inode, and we're done..
1893  */
1894 static int
1895 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
1896 {
1897         struct inode *inode;
1898         int error = -ENOSPC;
1899
1900         inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1901         if (inode) {
1902                 error = security_inode_init_security(inode, dir,
1903                                                      &dentry->d_name,
1904                                                      shmem_initxattrs, NULL);
1905                 if (error) {
1906                         if (error != -EOPNOTSUPP) {
1907                                 iput(inode);
1908                                 return error;
1909                         }
1910                 }
1911 #ifdef CONFIG_TMPFS_POSIX_ACL
1912                 error = generic_acl_init(inode, dir);
1913                 if (error) {
1914                         iput(inode);
1915                         return error;
1916                 }
1917 #else
1918                 error = 0;
1919 #endif
1920                 dir->i_size += BOGO_DIRENT_SIZE;
1921                 dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1922                 d_instantiate(dentry, inode);
1923                 dget(dentry); /* Extra count - pin the dentry in core */
1924         }
1925         return error;
1926 }
1927
1928 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1929 {
1930         int error;
1931
1932         if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1933                 return error;
1934         inc_nlink(dir);
1935         return 0;
1936 }
1937
1938 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1939                 struct nameidata *nd)
1940 {
1941         return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1942 }
1943
1944 /*
1945  * Link a file..
1946  */
1947 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
1948 {
1949         struct inode *inode = old_dentry->d_inode;
1950         int ret;
1951
1952         /*
1953          * No ordinary (disk based) filesystem counts links as inodes;
1954          * but each new link needs a new dentry, pinning lowmem, and
1955          * tmpfs dentries cannot be pruned until they are unlinked.
1956          */
1957         ret = shmem_reserve_inode(inode->i_sb);
1958         if (ret)
1959                 goto out;
1960
1961         dir->i_size += BOGO_DIRENT_SIZE;
1962         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1963         inc_nlink(inode);
1964         ihold(inode);   /* New dentry reference */
1965         dget(dentry);           /* Extra pinning count for the created dentry */
1966         d_instantiate(dentry, inode);
1967 out:
1968         return ret;
1969 }
1970
1971 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
1972 {
1973         struct inode *inode = dentry->d_inode;
1974
1975         if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
1976                 shmem_free_inode(inode->i_sb);
1977
1978         dir->i_size -= BOGO_DIRENT_SIZE;
1979         inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1980         drop_nlink(inode);
1981         dput(dentry);   /* Undo the count from "create" - this does all the work */
1982         return 0;
1983 }
1984
1985 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
1986 {
1987         if (!simple_empty(dentry))
1988                 return -ENOTEMPTY;
1989
1990         drop_nlink(dentry->d_inode);
1991         drop_nlink(dir);
1992         return shmem_unlink(dir, dentry);
1993 }
1994
1995 /*
1996  * The VFS layer already does all the dentry stuff for rename,
1997  * we just have to decrement the usage count for the target if
1998  * it exists so that the VFS layer correctly free's it when it
1999  * gets overwritten.
2000  */
2001 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2002 {
2003         struct inode *inode = old_dentry->d_inode;
2004         int they_are_dirs = S_ISDIR(inode->i_mode);
2005
2006         if (!simple_empty(new_dentry))
2007                 return -ENOTEMPTY;
2008
2009         if (new_dentry->d_inode) {
2010                 (void) shmem_unlink(new_dir, new_dentry);
2011                 if (they_are_dirs)
2012                         drop_nlink(old_dir);
2013         } else if (they_are_dirs) {
2014                 drop_nlink(old_dir);
2015                 inc_nlink(new_dir);
2016         }
2017
2018         old_dir->i_size -= BOGO_DIRENT_SIZE;
2019         new_dir->i_size += BOGO_DIRENT_SIZE;
2020         old_dir->i_ctime = old_dir->i_mtime =
2021         new_dir->i_ctime = new_dir->i_mtime =
2022         inode->i_ctime = CURRENT_TIME;
2023         return 0;
2024 }
2025
2026 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2027 {
2028         int error;
2029         int len;
2030         struct inode *inode;
2031         struct page *page;
2032         char *kaddr;
2033         struct shmem_inode_info *info;
2034
2035         len = strlen(symname) + 1;
2036         if (len > PAGE_CACHE_SIZE)
2037                 return -ENAMETOOLONG;
2038
2039         inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
2040         if (!inode)
2041                 return -ENOSPC;
2042
2043         error = security_inode_init_security(inode, dir, &dentry->d_name,
2044                                              shmem_initxattrs, NULL);
2045         if (error) {
2046                 if (error != -EOPNOTSUPP) {
2047                         iput(inode);
2048                         return error;
2049                 }
2050                 error = 0;
2051         }
2052
2053         info = SHMEM_I(inode);
2054         inode->i_size = len-1;
2055         if (len <= SHORT_SYMLINK_LEN) {
2056                 info->symlink = kmemdup(symname, len, GFP_KERNEL);
2057                 if (!info->symlink) {
2058                         iput(inode);
2059                         return -ENOMEM;
2060                 }
2061                 inode->i_op = &shmem_short_symlink_operations;
2062         } else {
2063                 error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2064                 if (error) {
2065                         iput(inode);
2066                         return error;
2067                 }
2068                 inode->i_mapping->a_ops = &shmem_aops;
2069                 inode->i_op = &shmem_symlink_inode_operations;
2070                 kaddr = kmap_atomic(page);
2071                 memcpy(kaddr, symname, len);
2072                 kunmap_atomic(kaddr);
2073                 SetPageUptodate(page);
2074                 set_page_dirty(page);
2075                 unlock_page(page);
2076                 page_cache_release(page);
2077         }
2078         dir->i_size += BOGO_DIRENT_SIZE;
2079         dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2080         d_instantiate(dentry, inode);
2081         dget(dentry);
2082         return 0;
2083 }
2084
2085 static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
2086 {
2087         nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
2088         return NULL;
2089 }
2090
2091 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
2092 {
2093         struct page *page = NULL;
2094         int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2095         nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
2096         if (page)
2097                 unlock_page(page);
2098         return page;
2099 }
2100
2101 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2102 {
2103         if (!IS_ERR(nd_get_link(nd))) {
2104                 struct page *page = cookie;
2105                 kunmap(page);
2106                 mark_page_accessed(page);
2107                 page_cache_release(page);
2108         }
2109 }
2110
2111 #ifdef CONFIG_TMPFS_XATTR
2112 /*
2113  * Superblocks without xattr inode operations may get some security.* xattr
2114  * support from the LSM "for free". As soon as we have any other xattrs
2115  * like ACLs, we also need to implement the security.* handlers at
2116  * filesystem level, though.
2117  */
2118
2119 /*
2120  * Allocate new xattr and copy in the value; but leave the name to callers.
2121  */
2122 static struct shmem_xattr *shmem_xattr_alloc(const void *value, size_t size)
2123 {
2124         struct shmem_xattr *new_xattr;
2125         size_t len;
2126
2127         /* wrap around? */
2128         len = sizeof(*new_xattr) + size;
2129         if (len <= sizeof(*new_xattr))
2130                 return NULL;
2131
2132         new_xattr = kmalloc(len, GFP_KERNEL);
2133         if (!new_xattr)
2134                 return NULL;
2135
2136         new_xattr->size = size;
2137         memcpy(new_xattr->value, value, size);
2138         return new_xattr;
2139 }
2140
2141 /*
2142  * Callback for security_inode_init_security() for acquiring xattrs.
2143  */
2144 static int shmem_initxattrs(struct inode *inode,
2145                             const struct xattr *xattr_array,
2146                             void *fs_info)
2147 {
2148         struct shmem_inode_info *info = SHMEM_I(inode);
2149         const struct xattr *xattr;
2150         struct shmem_xattr *new_xattr;
2151         size_t len;
2152
2153         for (xattr = xattr_array; xattr->name != NULL; xattr++) {
2154                 new_xattr = shmem_xattr_alloc(xattr->value, xattr->value_len);
2155                 if (!new_xattr)
2156                         return -ENOMEM;
2157
2158                 len = strlen(xattr->name) + 1;
2159                 new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2160                                           GFP_KERNEL);
2161                 if (!new_xattr->name) {
2162                         kfree(new_xattr);
2163                         return -ENOMEM;
2164                 }
2165
2166                 memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2167                        XATTR_SECURITY_PREFIX_LEN);
2168                 memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2169                        xattr->name, len);
2170
2171                 spin_lock(&info->lock);
2172                 list_add(&new_xattr->list, &info->xattr_list);
2173                 spin_unlock(&info->lock);
2174         }
2175
2176         return 0;
2177 }
2178
2179 static int shmem_xattr_get(struct dentry *dentry, const char *name,
2180                            void *buffer, size_t size)
2181 {
2182         struct shmem_inode_info *info;
2183         struct shmem_xattr *xattr;
2184         int ret = -ENODATA;
2185
2186         info = SHMEM_I(dentry->d_inode);
2187
2188         spin_lock(&info->lock);
2189         list_for_each_entry(xattr, &info->xattr_list, list) {
2190                 if (strcmp(name, xattr->name))
2191                         continue;
2192
2193                 ret = xattr->size;
2194                 if (buffer) {
2195                         if (size < xattr->size)
2196                                 ret = -ERANGE;
2197                         else
2198                                 memcpy(buffer, xattr->value, xattr->size);
2199                 }
2200                 break;
2201         }
2202         spin_unlock(&info->lock);
2203         return ret;
2204 }
2205
2206 static int shmem_xattr_set(struct inode *inode, const char *name,
2207                            const void *value, size_t size, int flags)
2208 {
2209         struct shmem_inode_info *info = SHMEM_I(inode);
2210         struct shmem_xattr *xattr;
2211         struct shmem_xattr *new_xattr = NULL;
2212         int err = 0;
2213
2214         /* value == NULL means remove */
2215         if (value) {
2216                 new_xattr = shmem_xattr_alloc(value, size);
2217                 if (!new_xattr)
2218                         return -ENOMEM;
2219
2220                 new_xattr->name = kstrdup(name, GFP_KERNEL);
2221                 if (!new_xattr->name) {
2222                         kfree(new_xattr);
2223                         return -ENOMEM;
2224                 }
2225         }
2226
2227         spin_lock(&info->lock);
2228         list_for_each_entry(xattr, &info->xattr_list, list) {
2229                 if (!strcmp(name, xattr->name)) {
2230                         if (flags & XATTR_CREATE) {
2231                                 xattr = new_xattr;
2232                                 err = -EEXIST;
2233                         } else if (new_xattr) {
2234                                 list_replace(&xattr->list, &new_xattr->list);
2235                         } else {
2236                                 list_del(&xattr->list);
2237                         }
2238                         goto out;
2239                 }
2240         }
2241         if (flags & XATTR_REPLACE) {
2242                 xattr = new_xattr;
2243                 err = -ENODATA;
2244         } else {
2245                 list_add(&new_xattr->list, &info->xattr_list);
2246                 xattr = NULL;
2247         }
2248 out:
2249         spin_unlock(&info->lock);
2250         if (xattr)
2251                 kfree(xattr->name);
2252         kfree(xattr);
2253         return err;
2254 }
2255
2256 static const struct xattr_handler *shmem_xattr_handlers[] = {
2257 #ifdef CONFIG_TMPFS_POSIX_ACL
2258         &generic_acl_access_handler,
2259         &generic_acl_default_handler,
2260 #endif
2261         NULL
2262 };
2263
2264 static int shmem_xattr_validate(const char *name)
2265 {
2266         struct { const char *prefix; size_t len; } arr[] = {
2267                 { XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2268                 { XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2269         };
2270         int i;
2271
2272         for (i = 0; i < ARRAY_SIZE(arr); i++) {
2273                 size_t preflen = arr[i].len;
2274                 if (strncmp(name, arr[i].prefix, preflen) == 0) {
2275                         if (!name[preflen])
2276                                 return -EINVAL;
2277                         return 0;
2278                 }
2279         }
2280         return -EOPNOTSUPP;
2281 }
2282
2283 static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2284                               void *buffer, size_t size)
2285 {
2286         int err;
2287
2288         /*
2289          * If this is a request for a synthetic attribute in the system.*
2290          * namespace use the generic infrastructure to resolve a handler
2291          * for it via sb->s_xattr.
2292          */
2293         if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2294                 return generic_getxattr(dentry, name, buffer, size);
2295
2296         err = shmem_xattr_validate(name);
2297         if (err)
2298                 return err;
2299
2300         return shmem_xattr_get(dentry, name, buffer, size);
2301 }
2302
2303 static int shmem_setxattr(struct dentry *dentry, const char *name,
2304                           const void *value, size_t size, int flags)
2305 {
2306         int err;
2307
2308         /*
2309          * If this is a request for a synthetic attribute in the system.*
2310          * namespace use the generic infrastructure to resolve a handler
2311          * for it via sb->s_xattr.
2312          */
2313         if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2314                 return generic_setxattr(dentry, name, value, size, flags);
2315
2316         err = shmem_xattr_validate(name);
2317         if (err)
2318                 return err;
2319
2320         if (size == 0)
2321                 value = "";  /* empty EA, do not remove */
2322
2323         return shmem_xattr_set(dentry->d_inode, name, value, size, flags);
2324
2325 }
2326
2327 static int shmem_removexattr(struct dentry *dentry, const char *name)
2328 {
2329         int err;
2330
2331         /*
2332          * If this is a request for a synthetic attribute in the system.*
2333          * namespace use the generic infrastructure to resolve a handler
2334          * for it via sb->s_xattr.
2335          */
2336         if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2337                 return generic_removexattr(dentry, name);
2338
2339         err = shmem_xattr_validate(name);
2340         if (err)
2341                 return err;
2342
2343         return shmem_xattr_set(dentry->d_inode, name, NULL, 0, XATTR_REPLACE);
2344 }
2345
2346 static bool xattr_is_trusted(const char *name)
2347 {
2348         return !strncmp(name, XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN);
2349 }
2350
2351 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2352 {
2353         bool trusted = capable(CAP_SYS_ADMIN);
2354         struct shmem_xattr *xattr;
2355         struct shmem_inode_info *info;
2356         size_t used = 0;
2357
2358         info = SHMEM_I(dentry->d_inode);
2359
2360         spin_lock(&info->lock);
2361         list_for_each_entry(xattr, &info->xattr_list, list) {
2362                 size_t len;
2363
2364                 /* skip "trusted." attributes for unprivileged callers */
2365                 if (!trusted && xattr_is_trusted(xattr->name))
2366                         continue;
2367
2368                 len = strlen(xattr->name) + 1;
2369                 used += len;
2370                 if (buffer) {
2371                         if (size < used) {
2372                                 used = -ERANGE;
2373                                 break;
2374                         }
2375                         memcpy(buffer, xattr->name, len);
2376                         buffer += len;
2377                 }
2378         }
2379         spin_unlock(&info->lock);
2380
2381         return used;
2382 }
2383 #endif /* CONFIG_TMPFS_XATTR */
2384
2385 static const struct inode_operations shmem_short_symlink_operations = {
2386         .readlink       = generic_readlink,
2387         .follow_link    = shmem_follow_short_symlink,
2388 #ifdef CONFIG_TMPFS_XATTR
2389         .setxattr       = shmem_setxattr,
2390         .getxattr       = shmem_getxattr,
2391         .listxattr      = shmem_listxattr,
2392         .removexattr    = shmem_removexattr,
2393 #endif
2394 };
2395
2396 static const struct inode_operations shmem_symlink_inode_operations = {
2397         .readlink       = generic_readlink,
2398         .follow_link    = shmem_follow_link,
2399         .put_link       = shmem_put_link,
2400 #ifdef CONFIG_TMPFS_XATTR
2401         .setxattr       = shmem_setxattr,
2402         .getxattr       = shmem_getxattr,
2403         .listxattr      = shmem_listxattr,
2404         .removexattr    = shmem_removexattr,
2405 #endif
2406 };
2407
2408 static struct dentry *shmem_get_parent(struct dentry *child)
2409 {
2410         return ERR_PTR(-ESTALE);
2411 }
2412
2413 static int shmem_match(struct inode *ino, void *vfh)
2414 {
2415         __u32 *fh = vfh;
2416         __u64 inum = fh[2];
2417         inum = (inum << 32) | fh[1];
2418         return ino->i_ino == inum && fh[0] == ino->i_generation;
2419 }
2420
2421 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2422                 struct fid *fid, int fh_len, int fh_type)
2423 {
2424         struct inode *inode;
2425         struct dentry *dentry = NULL;
2426         u64 inum = fid->raw[2];
2427         inum = (inum << 32) | fid->raw[1];
2428
2429         if (fh_len < 3)
2430                 return NULL;
2431
2432         inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2433                         shmem_match, fid->raw);
2434         if (inode) {
2435                 dentry = d_find_alias(inode);
2436                 iput(inode);
2437         }
2438
2439         return dentry;
2440 }
2441
2442 static int shmem_encode_fh(struct dentry *dentry, __u32 *fh, int *len,
2443                                 int connectable)
2444 {
2445         struct inode *inode = dentry->d_inode;
2446
2447         if (*len < 3) {
2448                 *len = 3;
2449                 return 255;
2450         }
2451
2452         if (inode_unhashed(inode)) {
2453                 /* Unfortunately insert_inode_hash is not idempotent,
2454                  * so as we hash inodes here rather than at creation
2455                  * time, we need a lock to ensure we only try
2456                  * to do it once
2457                  */
2458                 static DEFINE_SPINLOCK(lock);
2459                 spin_lock(&lock);
2460                 if (inode_unhashed(inode))
2461                         __insert_inode_hash(inode,
2462                                             inode->i_ino + inode->i_generation);
2463                 spin_unlock(&lock);
2464         }
2465
2466         fh[0] = inode->i_generation;
2467         fh[1] = inode->i_ino;
2468         fh[2] = ((__u64)inode->i_ino) >> 32;
2469
2470         *len = 3;
2471         return 1;
2472 }
2473
2474 static const struct export_operations shmem_export_ops = {
2475         .get_parent     = shmem_get_parent,
2476         .encode_fh      = shmem_encode_fh,
2477         .fh_to_dentry   = shmem_fh_to_dentry,
2478 };
2479
2480 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2481                                bool remount)
2482 {
2483         char *this_char, *value, *rest;
2484         uid_t uid;
2485         gid_t gid;
2486
2487         while (options != NULL) {
2488                 this_char = options;
2489                 for (;;) {
2490                         /*
2491                          * NUL-terminate this option: unfortunately,
2492                          * mount options form a comma-separated list,
2493                          * but mpol's nodelist may also contain commas.
2494                          */
2495                         options = strchr(options, ',');
2496                         if (options == NULL)
2497                                 break;
2498                         options++;
2499                         if (!isdigit(*options)) {
2500                                 options[-1] = '\0';
2501                                 break;
2502                         }
2503                 }
2504                 if (!*this_char)
2505                         continue;
2506                 if ((value = strchr(this_char,'=')) != NULL) {
2507                         *value++ = 0;
2508                 } else {
2509                         printk(KERN_ERR
2510                             "tmpfs: No value for mount option '%s'\n",
2511                             this_char);
2512                         return 1;
2513                 }
2514
2515                 if (!strcmp(this_char,"size")) {
2516                         unsigned long long size;
2517                         size = memparse(value,&rest);
2518                         if (*rest == '%') {
2519                                 size <<= PAGE_SHIFT;
2520                                 size *= totalram_pages;
2521                                 do_div(size, 100);
2522                                 rest++;
2523                         }
2524                         if (*rest)
2525                                 goto bad_val;
2526                         sbinfo->max_blocks =
2527                                 DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2528                 } else if (!strcmp(this_char,"nr_blocks")) {
2529                         sbinfo->max_blocks = memparse(value, &rest);
2530                         if (*rest)
2531                                 goto bad_val;
2532                 } else if (!strcmp(this_char,"nr_inodes")) {
2533                         sbinfo->max_inodes = memparse(value, &rest);
2534                         if (*rest)
2535                                 goto bad_val;
2536                 } else if (!strcmp(this_char,"mode")) {
2537                         if (remount)
2538                                 continue;
2539                         sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2540                         if (*rest)
2541                                 goto bad_val;
2542                 } else if (!strcmp(this_char,"uid")) {
2543                         if (remount)
2544                                 continue;
2545                         uid = simple_strtoul(value, &rest, 0);
2546                         if (*rest)
2547                                 goto bad_val;
2548                         sbinfo->uid = make_kuid(current_user_ns(), uid);
2549                         if (!uid_valid(sbinfo->uid))
2550                                 goto bad_val;
2551                 } else if (!strcmp(this_char,"gid")) {
2552                         if (remount)
2553                                 continue;
2554                         gid = simple_strtoul(value, &rest, 0);
2555                         if (*rest)
2556                                 goto bad_val;
2557                         sbinfo->gid = make_kgid(current_user_ns(), gid);
2558                         if (!gid_valid(sbinfo->gid))
2559                                 goto bad_val;
2560                 } else if (!strcmp(this_char,"mpol")) {
2561                         if (mpol_parse_str(value, &sbinfo->mpol, 1))
2562                                 goto bad_val;
2563                 } else {
2564                         printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2565                                this_char);
2566                         return 1;
2567                 }
2568         }
2569         return 0;
2570
2571 bad_val:
2572         printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2573                value, this_char);
2574         return 1;
2575
2576 }
2577
2578 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2579 {
2580         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2581         struct shmem_sb_info config = *sbinfo;
2582         unsigned long inodes;
2583         int error = -EINVAL;
2584
2585         if (shmem_parse_options(data, &config, true))
2586                 return error;
2587
2588         spin_lock(&sbinfo->stat_lock);
2589         inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2590         if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2591                 goto out;
2592         if (config.max_inodes < inodes)
2593                 goto out;
2594         /*
2595          * Those tests disallow limited->unlimited while any are in use;
2596          * but we must separately disallow unlimited->limited, because
2597          * in that case we have no record of how much is already in use.
2598          */
2599         if (config.max_blocks && !sbinfo->max_blocks)
2600                 goto out;
2601         if (config.max_inodes && !sbinfo->max_inodes)
2602                 goto out;
2603
2604         error = 0;
2605         sbinfo->max_blocks  = config.max_blocks;
2606         sbinfo->max_inodes  = config.max_inodes;
2607         sbinfo->free_inodes = config.max_inodes - inodes;
2608
2609         mpol_put(sbinfo->mpol);
2610         sbinfo->mpol        = config.mpol;      /* transfers initial ref */
2611 out:
2612         spin_unlock(&sbinfo->stat_lock);
2613         return error;
2614 }
2615
2616 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
2617 {
2618         struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
2619
2620         if (sbinfo->max_blocks != shmem_default_max_blocks())
2621                 seq_printf(seq, ",size=%luk",
2622                         sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2623         if (sbinfo->max_inodes != shmem_default_max_inodes())
2624                 seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2625         if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2626                 seq_printf(seq, ",mode=%03ho", sbinfo->mode);
2627         if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2628                 seq_printf(seq, ",uid=%u",
2629                                 from_kuid_munged(&init_user_ns, sbinfo->uid));
2630         if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2631                 seq_printf(seq, ",gid=%u",
2632                                 from_kgid_munged(&init_user_ns, sbinfo->gid));
2633         shmem_show_mpol(seq, sbinfo->mpol);
2634         return 0;
2635 }
2636 #endif /* CONFIG_TMPFS */
2637
2638 static void shmem_put_super(struct super_block *sb)
2639 {
2640         struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2641
2642         percpu_counter_destroy(&sbinfo->used_blocks);
2643         kfree(sbinfo);
2644         sb->s_fs_info = NULL;
2645 }
2646
2647 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2648 {
2649         struct inode *inode;
2650         struct shmem_sb_info *sbinfo;
2651         int err = -ENOMEM;
2652
2653         /* Round up to L1_CACHE_BYTES to resist false sharing */
2654         sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2655                                 L1_CACHE_BYTES), GFP_KERNEL);
2656         if (!sbinfo)
2657                 return -ENOMEM;
2658
2659         sbinfo->mode = S_IRWXUGO | S_ISVTX;
2660         sbinfo->uid = current_fsuid();
2661         sbinfo->gid = current_fsgid();
2662         sb->s_fs_info = sbinfo;
2663
2664 #ifdef CONFIG_TMPFS
2665         /*
2666          * Per default we only allow half of the physical ram per
2667          * tmpfs instance, limiting inodes to one per page of lowmem;
2668          * but the internal instance is left unlimited.
2669          */
2670         if (!(sb->s_flags & MS_NOUSER)) {
2671                 sbinfo->max_blocks = shmem_default_max_blocks();
2672                 sbinfo->max_inodes = shmem_default_max_inodes();
2673                 if (shmem_parse_options(data, sbinfo, false)) {
2674                         err = -EINVAL;
2675                         goto failed;
2676                 }
2677         }
2678         sb->s_export_op = &shmem_export_ops;
2679         sb->s_flags |= MS_NOSEC;
2680 #else
2681         sb->s_flags |= MS_NOUSER;
2682 #endif
2683
2684         spin_lock_init(&sbinfo->stat_lock);
2685         if (percpu_counter_init(&sbinfo->used_blocks, 0))
2686                 goto failed;
2687         sbinfo->free_inodes = sbinfo->max_inodes;
2688
2689         sb->s_maxbytes = MAX_LFS_FILESIZE;
2690         sb->s_blocksize = PAGE_CACHE_SIZE;
2691         sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2692         sb->s_magic = TMPFS_MAGIC;
2693         sb->s_op = &shmem_ops;
2694         sb->s_time_gran = 1;
2695 #ifdef CONFIG_TMPFS_XATTR
2696         sb->s_xattr = shmem_xattr_handlers;
2697 #endif
2698 #ifdef CONFIG_TMPFS_POSIX_ACL
2699         sb->s_flags |= MS_POSIXACL;
2700 #endif
2701
2702         inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2703         if (!inode)
2704                 goto failed;
2705         inode->i_uid = sbinfo->uid;
2706         inode->i_gid = sbinfo->gid;
2707         sb->s_root = d_make_root(inode);
2708         if (!sb->s_root)
2709                 goto failed;
2710         return 0;
2711
2712 failed:
2713         shmem_put_super(sb);
2714         return err;
2715 }
2716
2717 static struct kmem_cache *shmem_inode_cachep;
2718
2719 static struct inode *shmem_alloc_inode(struct super_block *sb)
2720 {
2721         struct shmem_inode_info *info;
2722         info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2723         if (!info)
2724                 return NULL;
2725         return &info->vfs_inode;
2726 }
2727
2728 static void shmem_destroy_callback(struct rcu_head *head)
2729 {
2730         struct inode *inode = container_of(head, struct inode, i_rcu);
2731         kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2732 }
2733
2734 static void shmem_destroy_inode(struct inode *inode)
2735 {
2736         if (S_ISREG(inode->i_mode))
2737                 mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2738         call_rcu(&inode->i_rcu, shmem_destroy_callback);
2739 }
2740
2741 static void shmem_init_inode(void *foo)
2742 {
2743         struct shmem_inode_info *info = foo;
2744         inode_init_once(&info->vfs_inode);
2745 }
2746
2747 static int shmem_init_inodecache(void)
2748 {
2749         shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2750                                 sizeof(struct shmem_inode_info),
2751                                 0, SLAB_PANIC, shmem_init_inode);
2752         return 0;
2753 }
2754
2755 static void shmem_destroy_inodecache(void)
2756 {
2757         kmem_cache_destroy(shmem_inode_cachep);
2758 }
2759
2760 static const struct address_space_operations shmem_aops = {
2761         .writepage      = shmem_writepage,
2762         .set_page_dirty = __set_page_dirty_no_writeback,
2763 #ifdef CONFIG_TMPFS
2764         .write_begin    = shmem_write_begin,
2765         .write_end      = shmem_write_end,
2766 #endif
2767         .migratepage    = migrate_page,
2768         .error_remove_page = generic_error_remove_page,
2769 };
2770
2771 static const struct file_operations shmem_file_operations = {
2772         .mmap           = shmem_mmap,
2773 #ifdef CONFIG_TMPFS
2774         .llseek         = shmem_file_llseek,
2775         .read           = do_sync_read,
2776         .write          = do_sync_write,
2777         .aio_read       = shmem_file_aio_read,
2778         .aio_write      = generic_file_aio_write,
2779         .fsync          = noop_fsync,
2780         .splice_read    = shmem_file_splice_read,
2781         .splice_write   = generic_file_splice_write,
2782         .fallocate      = shmem_fallocate,
2783 #endif
2784 };
2785
2786 static const struct inode_operations shmem_inode_operations = {
2787         .setattr        = shmem_setattr,
2788 #ifdef CONFIG_TMPFS_XATTR
2789         .setxattr       = shmem_setxattr,
2790         .getxattr       = shmem_getxattr,
2791         .listxattr      = shmem_listxattr,
2792         .removexattr    = shmem_removexattr,
2793 #endif
2794 };
2795
2796 static const struct inode_operations shmem_dir_inode_operations = {
2797 #ifdef CONFIG_TMPFS
2798         .create         = shmem_create,
2799         .lookup         = simple_lookup,
2800         .link           = shmem_link,
2801         .unlink         = shmem_unlink,
2802         .symlink        = shmem_symlink,
2803         .mkdir          = shmem_mkdir,
2804         .rmdir          = shmem_rmdir,
2805         .mknod          = shmem_mknod,
2806         .rename         = shmem_rename,
2807 #endif
2808 #ifdef CONFIG_TMPFS_XATTR
2809         .setxattr       = shmem_setxattr,
2810         .getxattr       = shmem_getxattr,
2811         .listxattr      = shmem_listxattr,
2812         .removexattr    = shmem_removexattr,
2813 #endif
2814 #ifdef CONFIG_TMPFS_POSIX_ACL
2815         .setattr        = shmem_setattr,
2816 #endif
2817 };
2818
2819 static const struct inode_operations shmem_special_inode_operations = {
2820 #ifdef CONFIG_TMPFS_XATTR
2821         .setxattr       = shmem_setxattr,
2822         .getxattr       = shmem_getxattr,
2823         .listxattr      = shmem_listxattr,
2824         .removexattr    = shmem_removexattr,
2825 #endif
2826 #ifdef CONFIG_TMPFS_POSIX_ACL
2827         .setattr        = shmem_setattr,
2828 #endif
2829 };
2830
2831 static const struct super_operations shmem_ops = {
2832         .alloc_inode    = shmem_alloc_inode,
2833         .destroy_inode  = shmem_destroy_inode,
2834 #ifdef CONFIG_TMPFS
2835         .statfs         = shmem_statfs,
2836         .remount_fs     = shmem_remount_fs,
2837         .show_options   = shmem_show_options,
2838 #endif
2839         .evict_inode    = shmem_evict_inode,
2840         .drop_inode     = generic_delete_inode,
2841         .put_super      = shmem_put_super,
2842 };
2843
2844 static const struct vm_operations_struct shmem_vm_ops = {
2845         .fault          = shmem_fault,
2846 #ifdef CONFIG_NUMA
2847         .set_policy     = shmem_set_policy,
2848         .get_policy     = shmem_get_policy,
2849 #endif
2850 };
2851
2852 static struct dentry *shmem_mount(struct file_system_type *fs_type,
2853         int flags, const char *dev_name, void *data)
2854 {
2855         return mount_nodev(fs_type, flags, data, shmem_fill_super);
2856 }
2857
2858 static struct file_system_type shmem_fs_type = {
2859         .owner          = THIS_MODULE,
2860         .name           = "tmpfs",
2861         .mount          = shmem_mount,
2862         .kill_sb        = kill_litter_super,
2863 };
2864
2865 int __init shmem_init(void)
2866 {
2867         int error;
2868
2869         error = bdi_init(&shmem_backing_dev_info);
2870         if (error)
2871                 goto out4;
2872
2873         error = shmem_init_inodecache();
2874         if (error)
2875                 goto out3;
2876
2877         error = register_filesystem(&shmem_fs_type);
2878         if (error) {
2879                 printk(KERN_ERR "Could not register tmpfs\n");
2880                 goto out2;
2881         }
2882
2883         shm_mnt = vfs_kern_mount(&shmem_fs_type, MS_NOUSER,
2884                                  shmem_fs_type.name, NULL);
2885         if (IS_ERR(shm_mnt)) {
2886                 error = PTR_ERR(shm_mnt);
2887                 printk(KERN_ERR "Could not kern_mount tmpfs\n");
2888                 goto out1;
2889         }
2890         return 0;
2891
2892 out1:
2893         unregister_filesystem(&shmem_fs_type);
2894 out2:
2895         shmem_destroy_inodecache();
2896 out3:
2897         bdi_destroy(&shmem_backing_dev_info);
2898 out4:
2899         shm_mnt = ERR_PTR(error);
2900         return error;
2901 }
2902
2903 #else /* !CONFIG_SHMEM */
2904
2905 /*
2906  * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2907  *
2908  * This is intended for small system where the benefits of the full
2909  * shmem code (swap-backed and resource-limited) are outweighed by
2910  * their complexity. On systems without swap this code should be
2911  * effectively equivalent, but much lighter weight.
2912  */
2913
2914 #include <linux/ramfs.h>
2915
2916 static struct file_system_type shmem_fs_type = {
2917         .name           = "tmpfs",
2918         .mount          = ramfs_mount,
2919         .kill_sb        = kill_litter_super,
2920 };
2921
2922 int __init shmem_init(void)
2923 {
2924         BUG_ON(register_filesystem(&shmem_fs_type) != 0);
2925
2926         shm_mnt = kern_mount(&shmem_fs_type);
2927         BUG_ON(IS_ERR(shm_mnt));
2928
2929         return 0;
2930 }
2931
2932 int shmem_unuse(swp_entry_t swap, struct page *page)
2933 {
2934         return 0;
2935 }
2936
2937 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2938 {
2939         return 0;
2940 }
2941
2942 void shmem_unlock_mapping(struct address_space *mapping)
2943 {
2944 }
2945
2946 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
2947 {
2948         truncate_inode_pages_range(inode->i_mapping, lstart, lend);
2949 }
2950 EXPORT_SYMBOL_GPL(shmem_truncate_range);
2951
2952 #define shmem_vm_ops                            generic_file_vm_ops
2953 #define shmem_file_operations                   ramfs_file_operations
2954 #define shmem_get_inode(sb, dir, mode, dev, flags)      ramfs_get_inode(sb, dir, mode, dev)
2955 #define shmem_acct_size(flags, size)            0
2956 #define shmem_unacct_size(flags, size)          do {} while (0)
2957
2958 #endif /* CONFIG_SHMEM */
2959
2960 /* common code */
2961
2962 /**
2963  * shmem_file_setup - get an unlinked file living in tmpfs
2964  * @name: name for dentry (to be seen in /proc/<pid>/maps
2965  * @size: size to be set for the file
2966  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2967  */
2968 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2969 {
2970         int error;
2971         struct file *file;
2972         struct inode *inode;
2973         struct path path;
2974         struct dentry *root;
2975         struct qstr this;
2976
2977         if (IS_ERR(shm_mnt))
2978                 return (void *)shm_mnt;
2979
2980         if (size < 0 || size > MAX_LFS_FILESIZE)
2981                 return ERR_PTR(-EINVAL);
2982
2983         if (shmem_acct_size(flags, size))
2984                 return ERR_PTR(-ENOMEM);
2985
2986         error = -ENOMEM;
2987         this.name = name;
2988         this.len = strlen(name);
2989         this.hash = 0; /* will go */
2990         root = shm_mnt->mnt_root;
2991         path.dentry = d_alloc(root, &this);
2992         if (!path.dentry)
2993                 goto put_memory;
2994         path.mnt = mntget(shm_mnt);
2995
2996         error = -ENOSPC;
2997         inode = shmem_get_inode(root->d_sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2998         if (!inode)
2999                 goto put_dentry;
3000
3001         d_instantiate(path.dentry, inode);
3002         inode->i_size = size;
3003         clear_nlink(inode);     /* It is unlinked */
3004 #ifndef CONFIG_MMU
3005         error = ramfs_nommu_expand_for_mapping(inode, size);
3006         if (error)
3007                 goto put_dentry;
3008 #endif
3009
3010         error = -ENFILE;
3011         file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
3012                   &shmem_file_operations);
3013         if (!file)
3014                 goto put_dentry;
3015
3016         return file;
3017
3018 put_dentry:
3019         path_put(&path);
3020 put_memory:
3021         shmem_unacct_size(flags, size);
3022         return ERR_PTR(error);
3023 }
3024 EXPORT_SYMBOL_GPL(shmem_file_setup);
3025
3026 /**
3027  * shmem_zero_setup - setup a shared anonymous mapping
3028  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
3029  */
3030 int shmem_zero_setup(struct vm_area_struct *vma)
3031 {
3032         struct file *file;
3033         loff_t size = vma->vm_end - vma->vm_start;
3034
3035         file = shmem_file_setup("dev/zero", size, vma->vm_flags);
3036         if (IS_ERR(file))
3037                 return PTR_ERR(file);
3038
3039         if (vma->vm_file)
3040                 fput(vma->vm_file);
3041         vma->vm_file = file;
3042         vma->vm_ops = &shmem_vm_ops;
3043         vma->vm_flags |= VM_CAN_NONLINEAR;
3044         return 0;
3045 }
3046
3047 /**
3048  * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3049  * @mapping:    the page's address_space
3050  * @index:      the page index
3051  * @gfp:        the page allocator flags to use if allocating
3052  *
3053  * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3054  * with any new page allocations done using the specified allocation flags.
3055  * But read_cache_page_gfp() uses the ->readpage() method: which does not
3056  * suit tmpfs, since it may have pages in swapcache, and needs to find those
3057  * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3058  *
3059  * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3060  * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
3061  */
3062 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
3063                                          pgoff_t index, gfp_t gfp)
3064 {
3065 #ifdef CONFIG_SHMEM
3066         struct inode *inode = mapping->host;
3067         struct page *page;
3068         int error;
3069
3070         BUG_ON(mapping->a_ops != &shmem_aops);
3071         error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
3072         if (error)
3073                 page = ERR_PTR(error);
3074         else
3075                 unlock_page(page);
3076         return page;
3077 #else
3078         /*
3079          * The tiny !SHMEM case uses ramfs without swap
3080          */
3081         return read_cache_page_gfp(mapping, index, gfp);
3082 #endif
3083 }
3084 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);