xref: /openbmc/linux/mm/shmem.c (revision 8a10bc9d)
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/ramfs.h>
29 #include <linux/pagemap.h>
30 #include <linux/file.h>
31 #include <linux/mm.h>
32 #include <linux/export.h>
33 #include <linux/swap.h>
34 #include <linux/aio.h>
35 
36 static struct vfsmount *shm_mnt;
37 
38 #ifdef CONFIG_SHMEM
39 /*
40  * This virtual memory filesystem is heavily based on the ramfs. It
41  * extends ramfs by the ability to use swap and honor resource limits
42  * which makes it a completely usable filesystem.
43  */
44 
45 #include <linux/xattr.h>
46 #include <linux/exportfs.h>
47 #include <linux/posix_acl.h>
48 #include <linux/posix_acl_xattr.h>
49 #include <linux/mman.h>
50 #include <linux/string.h>
51 #include <linux/slab.h>
52 #include <linux/backing-dev.h>
53 #include <linux/shmem_fs.h>
54 #include <linux/writeback.h>
55 #include <linux/blkdev.h>
56 #include <linux/pagevec.h>
57 #include <linux/percpu_counter.h>
58 #include <linux/falloc.h>
59 #include <linux/splice.h>
60 #include <linux/security.h>
61 #include <linux/swapops.h>
62 #include <linux/mempolicy.h>
63 #include <linux/namei.h>
64 #include <linux/ctype.h>
65 #include <linux/migrate.h>
66 #include <linux/highmem.h>
67 #include <linux/seq_file.h>
68 #include <linux/magic.h>
69 
70 #include <asm/uaccess.h>
71 #include <asm/pgtable.h>
72 
73 #define BLOCKS_PER_PAGE  (PAGE_CACHE_SIZE/512)
74 #define VM_ACCT(size)    (PAGE_CACHE_ALIGN(size) >> PAGE_SHIFT)
75 
76 /* Pretend that each entry is of this size in directory's i_size */
77 #define BOGO_DIRENT_SIZE 20
78 
79 /* Symlink up to this size is kmalloc'ed instead of using a swappable page */
80 #define SHORT_SYMLINK_LEN 128
81 
82 /*
83  * shmem_fallocate and shmem_writepage communicate via inode->i_private
84  * (with i_mutex making sure that it has only one user at a time):
85  * we would prefer not to enlarge the shmem inode just for that.
86  */
87 struct shmem_falloc {
88 	pgoff_t start;		/* start of range currently being fallocated */
89 	pgoff_t next;		/* the next page offset to be fallocated */
90 	pgoff_t nr_falloced;	/* how many new pages have been fallocated */
91 	pgoff_t nr_unswapped;	/* how often writepage refused to swap out */
92 };
93 
94 /* Flag allocation requirements to shmem_getpage */
95 enum sgp_type {
96 	SGP_READ,	/* don't exceed i_size, don't allocate page */
97 	SGP_CACHE,	/* don't exceed i_size, may allocate page */
98 	SGP_DIRTY,	/* like SGP_CACHE, but set new page dirty */
99 	SGP_WRITE,	/* may exceed i_size, may allocate !Uptodate page */
100 	SGP_FALLOC,	/* like SGP_WRITE, but make existing page Uptodate */
101 };
102 
103 #ifdef CONFIG_TMPFS
104 static unsigned long shmem_default_max_blocks(void)
105 {
106 	return totalram_pages / 2;
107 }
108 
109 static unsigned long shmem_default_max_inodes(void)
110 {
111 	return min(totalram_pages - totalhigh_pages, totalram_pages / 2);
112 }
113 #endif
114 
115 static bool shmem_should_replace_page(struct page *page, gfp_t gfp);
116 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
117 				struct shmem_inode_info *info, pgoff_t index);
118 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
119 	struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type);
120 
121 static inline int shmem_getpage(struct inode *inode, pgoff_t index,
122 	struct page **pagep, enum sgp_type sgp, int *fault_type)
123 {
124 	return shmem_getpage_gfp(inode, index, pagep, sgp,
125 			mapping_gfp_mask(inode->i_mapping), fault_type);
126 }
127 
128 static inline struct shmem_sb_info *SHMEM_SB(struct super_block *sb)
129 {
130 	return sb->s_fs_info;
131 }
132 
133 /*
134  * shmem_file_setup pre-accounts the whole fixed size of a VM object,
135  * for shared memory and for shared anonymous (/dev/zero) mappings
136  * (unless MAP_NORESERVE and sysctl_overcommit_memory <= 1),
137  * consistent with the pre-accounting of private mappings ...
138  */
139 static inline int shmem_acct_size(unsigned long flags, loff_t size)
140 {
141 	return (flags & VM_NORESERVE) ?
142 		0 : security_vm_enough_memory_mm(current->mm, VM_ACCT(size));
143 }
144 
145 static inline void shmem_unacct_size(unsigned long flags, loff_t size)
146 {
147 	if (!(flags & VM_NORESERVE))
148 		vm_unacct_memory(VM_ACCT(size));
149 }
150 
151 /*
152  * ... whereas tmpfs objects are accounted incrementally as
153  * pages are allocated, in order to allow huge sparse files.
154  * shmem_getpage reports shmem_acct_block failure as -ENOSPC not -ENOMEM,
155  * so that a failure on a sparse tmpfs mapping will give SIGBUS not OOM.
156  */
157 static inline int shmem_acct_block(unsigned long flags)
158 {
159 	return (flags & VM_NORESERVE) ?
160 		security_vm_enough_memory_mm(current->mm, VM_ACCT(PAGE_CACHE_SIZE)) : 0;
161 }
162 
163 static inline void shmem_unacct_blocks(unsigned long flags, long pages)
164 {
165 	if (flags & VM_NORESERVE)
166 		vm_unacct_memory(pages * VM_ACCT(PAGE_CACHE_SIZE));
167 }
168 
169 static const struct super_operations shmem_ops;
170 static const struct address_space_operations shmem_aops;
171 static const struct file_operations shmem_file_operations;
172 static const struct inode_operations shmem_inode_operations;
173 static const struct inode_operations shmem_dir_inode_operations;
174 static const struct inode_operations shmem_special_inode_operations;
175 static const struct vm_operations_struct shmem_vm_ops;
176 
177 static struct backing_dev_info shmem_backing_dev_info  __read_mostly = {
178 	.ra_pages	= 0,	/* No readahead */
179 	.capabilities	= BDI_CAP_NO_ACCT_AND_WRITEBACK | BDI_CAP_SWAP_BACKED,
180 };
181 
182 static LIST_HEAD(shmem_swaplist);
183 static DEFINE_MUTEX(shmem_swaplist_mutex);
184 
185 static int shmem_reserve_inode(struct super_block *sb)
186 {
187 	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
188 	if (sbinfo->max_inodes) {
189 		spin_lock(&sbinfo->stat_lock);
190 		if (!sbinfo->free_inodes) {
191 			spin_unlock(&sbinfo->stat_lock);
192 			return -ENOSPC;
193 		}
194 		sbinfo->free_inodes--;
195 		spin_unlock(&sbinfo->stat_lock);
196 	}
197 	return 0;
198 }
199 
200 static void shmem_free_inode(struct super_block *sb)
201 {
202 	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
203 	if (sbinfo->max_inodes) {
204 		spin_lock(&sbinfo->stat_lock);
205 		sbinfo->free_inodes++;
206 		spin_unlock(&sbinfo->stat_lock);
207 	}
208 }
209 
210 /**
211  * shmem_recalc_inode - recalculate the block usage of an inode
212  * @inode: inode to recalc
213  *
214  * We have to calculate the free blocks since the mm can drop
215  * undirtied hole pages behind our back.
216  *
217  * But normally   info->alloced == inode->i_mapping->nrpages + info->swapped
218  * So mm freed is info->alloced - (inode->i_mapping->nrpages + info->swapped)
219  *
220  * It has to be called with the spinlock held.
221  */
222 static void shmem_recalc_inode(struct inode *inode)
223 {
224 	struct shmem_inode_info *info = SHMEM_I(inode);
225 	long freed;
226 
227 	freed = info->alloced - info->swapped - inode->i_mapping->nrpages;
228 	if (freed > 0) {
229 		struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
230 		if (sbinfo->max_blocks)
231 			percpu_counter_add(&sbinfo->used_blocks, -freed);
232 		info->alloced -= freed;
233 		inode->i_blocks -= freed * BLOCKS_PER_PAGE;
234 		shmem_unacct_blocks(info->flags, freed);
235 	}
236 }
237 
238 /*
239  * Replace item expected in radix tree by a new item, while holding tree lock.
240  */
241 static int shmem_radix_tree_replace(struct address_space *mapping,
242 			pgoff_t index, void *expected, void *replacement)
243 {
244 	void **pslot;
245 	void *item = NULL;
246 
247 	VM_BUG_ON(!expected);
248 	pslot = radix_tree_lookup_slot(&mapping->page_tree, index);
249 	if (pslot)
250 		item = radix_tree_deref_slot_protected(pslot,
251 							&mapping->tree_lock);
252 	if (item != expected)
253 		return -ENOENT;
254 	if (replacement)
255 		radix_tree_replace_slot(pslot, replacement);
256 	else
257 		radix_tree_delete(&mapping->page_tree, index);
258 	return 0;
259 }
260 
261 /*
262  * Sometimes, before we decide whether to proceed or to fail, we must check
263  * that an entry was not already brought back from swap by a racing thread.
264  *
265  * Checking page is not enough: by the time a SwapCache page is locked, it
266  * might be reused, and again be SwapCache, using the same swap as before.
267  */
268 static bool shmem_confirm_swap(struct address_space *mapping,
269 			       pgoff_t index, swp_entry_t swap)
270 {
271 	void *item;
272 
273 	rcu_read_lock();
274 	item = radix_tree_lookup(&mapping->page_tree, index);
275 	rcu_read_unlock();
276 	return item == swp_to_radix_entry(swap);
277 }
278 
279 /*
280  * Like add_to_page_cache_locked, but error if expected item has gone.
281  */
282 static int shmem_add_to_page_cache(struct page *page,
283 				   struct address_space *mapping,
284 				   pgoff_t index, gfp_t gfp, void *expected)
285 {
286 	int error;
287 
288 	VM_BUG_ON_PAGE(!PageLocked(page), page);
289 	VM_BUG_ON_PAGE(!PageSwapBacked(page), page);
290 
291 	page_cache_get(page);
292 	page->mapping = mapping;
293 	page->index = index;
294 
295 	spin_lock_irq(&mapping->tree_lock);
296 	if (!expected)
297 		error = radix_tree_insert(&mapping->page_tree, index, page);
298 	else
299 		error = shmem_radix_tree_replace(mapping, index, expected,
300 								 page);
301 	if (!error) {
302 		mapping->nrpages++;
303 		__inc_zone_page_state(page, NR_FILE_PAGES);
304 		__inc_zone_page_state(page, NR_SHMEM);
305 		spin_unlock_irq(&mapping->tree_lock);
306 	} else {
307 		page->mapping = NULL;
308 		spin_unlock_irq(&mapping->tree_lock);
309 		page_cache_release(page);
310 	}
311 	return error;
312 }
313 
314 /*
315  * Like delete_from_page_cache, but substitutes swap for page.
316  */
317 static void shmem_delete_from_page_cache(struct page *page, void *radswap)
318 {
319 	struct address_space *mapping = page->mapping;
320 	int error;
321 
322 	spin_lock_irq(&mapping->tree_lock);
323 	error = shmem_radix_tree_replace(mapping, page->index, page, radswap);
324 	page->mapping = NULL;
325 	mapping->nrpages--;
326 	__dec_zone_page_state(page, NR_FILE_PAGES);
327 	__dec_zone_page_state(page, NR_SHMEM);
328 	spin_unlock_irq(&mapping->tree_lock);
329 	page_cache_release(page);
330 	BUG_ON(error);
331 }
332 
333 /*
334  * Like find_get_pages, but collecting swap entries as well as pages.
335  */
336 static unsigned shmem_find_get_pages_and_swap(struct address_space *mapping,
337 					pgoff_t start, unsigned int nr_pages,
338 					struct page **pages, pgoff_t *indices)
339 {
340 	void **slot;
341 	unsigned int ret = 0;
342 	struct radix_tree_iter iter;
343 
344 	if (!nr_pages)
345 		return 0;
346 
347 	rcu_read_lock();
348 restart:
349 	radix_tree_for_each_slot(slot, &mapping->page_tree, &iter, start) {
350 		struct page *page;
351 repeat:
352 		page = radix_tree_deref_slot(slot);
353 		if (unlikely(!page))
354 			continue;
355 		if (radix_tree_exception(page)) {
356 			if (radix_tree_deref_retry(page))
357 				goto restart;
358 			/*
359 			 * Otherwise, we must be storing a swap entry
360 			 * here as an exceptional entry: so return it
361 			 * without attempting to raise page count.
362 			 */
363 			goto export;
364 		}
365 		if (!page_cache_get_speculative(page))
366 			goto repeat;
367 
368 		/* Has the page moved? */
369 		if (unlikely(page != *slot)) {
370 			page_cache_release(page);
371 			goto repeat;
372 		}
373 export:
374 		indices[ret] = iter.index;
375 		pages[ret] = page;
376 		if (++ret == nr_pages)
377 			break;
378 	}
379 	rcu_read_unlock();
380 	return ret;
381 }
382 
383 /*
384  * Remove swap entry from radix tree, free the swap and its page cache.
385  */
386 static int shmem_free_swap(struct address_space *mapping,
387 			   pgoff_t index, void *radswap)
388 {
389 	int error;
390 
391 	spin_lock_irq(&mapping->tree_lock);
392 	error = shmem_radix_tree_replace(mapping, index, radswap, NULL);
393 	spin_unlock_irq(&mapping->tree_lock);
394 	if (!error)
395 		free_swap_and_cache(radix_to_swp_entry(radswap));
396 	return error;
397 }
398 
399 /*
400  * Pagevec may contain swap entries, so shuffle up pages before releasing.
401  */
402 static void shmem_deswap_pagevec(struct pagevec *pvec)
403 {
404 	int i, j;
405 
406 	for (i = 0, j = 0; i < pagevec_count(pvec); i++) {
407 		struct page *page = pvec->pages[i];
408 		if (!radix_tree_exceptional_entry(page))
409 			pvec->pages[j++] = page;
410 	}
411 	pvec->nr = j;
412 }
413 
414 /*
415  * SysV IPC SHM_UNLOCK restore Unevictable pages to their evictable lists.
416  */
417 void shmem_unlock_mapping(struct address_space *mapping)
418 {
419 	struct pagevec pvec;
420 	pgoff_t indices[PAGEVEC_SIZE];
421 	pgoff_t index = 0;
422 
423 	pagevec_init(&pvec, 0);
424 	/*
425 	 * Minor point, but we might as well stop if someone else SHM_LOCKs it.
426 	 */
427 	while (!mapping_unevictable(mapping)) {
428 		/*
429 		 * Avoid pagevec_lookup(): find_get_pages() returns 0 as if it
430 		 * has finished, if it hits a row of PAGEVEC_SIZE swap entries.
431 		 */
432 		pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
433 					PAGEVEC_SIZE, pvec.pages, indices);
434 		if (!pvec.nr)
435 			break;
436 		index = indices[pvec.nr - 1] + 1;
437 		shmem_deswap_pagevec(&pvec);
438 		check_move_unevictable_pages(pvec.pages, pvec.nr);
439 		pagevec_release(&pvec);
440 		cond_resched();
441 	}
442 }
443 
444 /*
445  * Remove range of pages and swap entries from radix tree, and free them.
446  * If !unfalloc, truncate or punch hole; if unfalloc, undo failed fallocate.
447  */
448 static void shmem_undo_range(struct inode *inode, loff_t lstart, loff_t lend,
449 								 bool unfalloc)
450 {
451 	struct address_space *mapping = inode->i_mapping;
452 	struct shmem_inode_info *info = SHMEM_I(inode);
453 	pgoff_t start = (lstart + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
454 	pgoff_t end = (lend + 1) >> PAGE_CACHE_SHIFT;
455 	unsigned int partial_start = lstart & (PAGE_CACHE_SIZE - 1);
456 	unsigned int partial_end = (lend + 1) & (PAGE_CACHE_SIZE - 1);
457 	struct pagevec pvec;
458 	pgoff_t indices[PAGEVEC_SIZE];
459 	long nr_swaps_freed = 0;
460 	pgoff_t index;
461 	int i;
462 
463 	if (lend == -1)
464 		end = -1;	/* unsigned, so actually very big */
465 
466 	pagevec_init(&pvec, 0);
467 	index = start;
468 	while (index < end) {
469 		pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
470 				min(end - index, (pgoff_t)PAGEVEC_SIZE),
471 							pvec.pages, indices);
472 		if (!pvec.nr)
473 			break;
474 		mem_cgroup_uncharge_start();
475 		for (i = 0; i < pagevec_count(&pvec); i++) {
476 			struct page *page = pvec.pages[i];
477 
478 			index = indices[i];
479 			if (index >= end)
480 				break;
481 
482 			if (radix_tree_exceptional_entry(page)) {
483 				if (unfalloc)
484 					continue;
485 				nr_swaps_freed += !shmem_free_swap(mapping,
486 								index, page);
487 				continue;
488 			}
489 
490 			if (!trylock_page(page))
491 				continue;
492 			if (!unfalloc || !PageUptodate(page)) {
493 				if (page->mapping == mapping) {
494 					VM_BUG_ON_PAGE(PageWriteback(page), page);
495 					truncate_inode_page(mapping, page);
496 				}
497 			}
498 			unlock_page(page);
499 		}
500 		shmem_deswap_pagevec(&pvec);
501 		pagevec_release(&pvec);
502 		mem_cgroup_uncharge_end();
503 		cond_resched();
504 		index++;
505 	}
506 
507 	if (partial_start) {
508 		struct page *page = NULL;
509 		shmem_getpage(inode, start - 1, &page, SGP_READ, NULL);
510 		if (page) {
511 			unsigned int top = PAGE_CACHE_SIZE;
512 			if (start > end) {
513 				top = partial_end;
514 				partial_end = 0;
515 			}
516 			zero_user_segment(page, partial_start, top);
517 			set_page_dirty(page);
518 			unlock_page(page);
519 			page_cache_release(page);
520 		}
521 	}
522 	if (partial_end) {
523 		struct page *page = NULL;
524 		shmem_getpage(inode, end, &page, SGP_READ, NULL);
525 		if (page) {
526 			zero_user_segment(page, 0, partial_end);
527 			set_page_dirty(page);
528 			unlock_page(page);
529 			page_cache_release(page);
530 		}
531 	}
532 	if (start >= end)
533 		return;
534 
535 	index = start;
536 	for ( ; ; ) {
537 		cond_resched();
538 		pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
539 				min(end - index, (pgoff_t)PAGEVEC_SIZE),
540 							pvec.pages, indices);
541 		if (!pvec.nr) {
542 			if (index == start || unfalloc)
543 				break;
544 			index = start;
545 			continue;
546 		}
547 		if ((index == start || unfalloc) && indices[0] >= end) {
548 			shmem_deswap_pagevec(&pvec);
549 			pagevec_release(&pvec);
550 			break;
551 		}
552 		mem_cgroup_uncharge_start();
553 		for (i = 0; i < pagevec_count(&pvec); i++) {
554 			struct page *page = pvec.pages[i];
555 
556 			index = indices[i];
557 			if (index >= end)
558 				break;
559 
560 			if (radix_tree_exceptional_entry(page)) {
561 				if (unfalloc)
562 					continue;
563 				nr_swaps_freed += !shmem_free_swap(mapping,
564 								index, page);
565 				continue;
566 			}
567 
568 			lock_page(page);
569 			if (!unfalloc || !PageUptodate(page)) {
570 				if (page->mapping == mapping) {
571 					VM_BUG_ON_PAGE(PageWriteback(page), page);
572 					truncate_inode_page(mapping, page);
573 				}
574 			}
575 			unlock_page(page);
576 		}
577 		shmem_deswap_pagevec(&pvec);
578 		pagevec_release(&pvec);
579 		mem_cgroup_uncharge_end();
580 		index++;
581 	}
582 
583 	spin_lock(&info->lock);
584 	info->swapped -= nr_swaps_freed;
585 	shmem_recalc_inode(inode);
586 	spin_unlock(&info->lock);
587 }
588 
589 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
590 {
591 	shmem_undo_range(inode, lstart, lend, false);
592 	inode->i_ctime = inode->i_mtime = CURRENT_TIME;
593 }
594 EXPORT_SYMBOL_GPL(shmem_truncate_range);
595 
596 static int shmem_setattr(struct dentry *dentry, struct iattr *attr)
597 {
598 	struct inode *inode = dentry->d_inode;
599 	int error;
600 
601 	error = inode_change_ok(inode, attr);
602 	if (error)
603 		return error;
604 
605 	if (S_ISREG(inode->i_mode) && (attr->ia_valid & ATTR_SIZE)) {
606 		loff_t oldsize = inode->i_size;
607 		loff_t newsize = attr->ia_size;
608 
609 		if (newsize != oldsize) {
610 			i_size_write(inode, newsize);
611 			inode->i_ctime = inode->i_mtime = CURRENT_TIME;
612 		}
613 		if (newsize < oldsize) {
614 			loff_t holebegin = round_up(newsize, PAGE_SIZE);
615 			unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
616 			shmem_truncate_range(inode, newsize, (loff_t)-1);
617 			/* unmap again to remove racily COWed private pages */
618 			unmap_mapping_range(inode->i_mapping, holebegin, 0, 1);
619 		}
620 	}
621 
622 	setattr_copy(inode, attr);
623 	if (attr->ia_valid & ATTR_MODE)
624 		error = posix_acl_chmod(inode, inode->i_mode);
625 	return error;
626 }
627 
628 static void shmem_evict_inode(struct inode *inode)
629 {
630 	struct shmem_inode_info *info = SHMEM_I(inode);
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 	simple_xattrs_free(&info->xattrs);
645 	WARN_ON(inode->i_blocks);
646 	shmem_free_inode(inode->i_sb);
647 	clear_inode(inode);
648 }
649 
650 /*
651  * If swap found in inode, free it and move page from swapcache to filecache.
652  */
653 static int shmem_unuse_inode(struct shmem_inode_info *info,
654 			     swp_entry_t swap, struct page **pagep)
655 {
656 	struct address_space *mapping = info->vfs_inode.i_mapping;
657 	void *radswap;
658 	pgoff_t index;
659 	gfp_t gfp;
660 	int error = 0;
661 
662 	radswap = swp_to_radix_entry(swap);
663 	index = radix_tree_locate_item(&mapping->page_tree, radswap);
664 	if (index == -1)
665 		return 0;
666 
667 	/*
668 	 * Move _head_ to start search for next from here.
669 	 * But be careful: shmem_evict_inode checks list_empty without taking
670 	 * mutex, and there's an instant in list_move_tail when info->swaplist
671 	 * would appear empty, if it were the only one on shmem_swaplist.
672 	 */
673 	if (shmem_swaplist.next != &info->swaplist)
674 		list_move_tail(&shmem_swaplist, &info->swaplist);
675 
676 	gfp = mapping_gfp_mask(mapping);
677 	if (shmem_should_replace_page(*pagep, gfp)) {
678 		mutex_unlock(&shmem_swaplist_mutex);
679 		error = shmem_replace_page(pagep, gfp, info, index);
680 		mutex_lock(&shmem_swaplist_mutex);
681 		/*
682 		 * We needed to drop mutex to make that restrictive page
683 		 * allocation, but the inode might have been freed while we
684 		 * dropped it: although a racing shmem_evict_inode() cannot
685 		 * complete without emptying the radix_tree, our page lock
686 		 * on this swapcache page is not enough to prevent that -
687 		 * free_swap_and_cache() of our swap entry will only
688 		 * trylock_page(), removing swap from radix_tree whatever.
689 		 *
690 		 * We must not proceed to shmem_add_to_page_cache() if the
691 		 * inode has been freed, but of course we cannot rely on
692 		 * inode or mapping or info to check that.  However, we can
693 		 * safely check if our swap entry is still in use (and here
694 		 * it can't have got reused for another page): if it's still
695 		 * in use, then the inode cannot have been freed yet, and we
696 		 * can safely proceed (if it's no longer in use, that tells
697 		 * nothing about the inode, but we don't need to unuse swap).
698 		 */
699 		if (!page_swapcount(*pagep))
700 			error = -ENOENT;
701 	}
702 
703 	/*
704 	 * We rely on shmem_swaplist_mutex, not only to protect the swaplist,
705 	 * but also to hold up shmem_evict_inode(): so inode cannot be freed
706 	 * beneath us (pagelock doesn't help until the page is in pagecache).
707 	 */
708 	if (!error)
709 		error = shmem_add_to_page_cache(*pagep, mapping, index,
710 						GFP_NOWAIT, radswap);
711 	if (error != -ENOMEM) {
712 		/*
713 		 * Truncation and eviction use free_swap_and_cache(), which
714 		 * only does trylock page: if we raced, best clean up here.
715 		 */
716 		delete_from_swap_cache(*pagep);
717 		set_page_dirty(*pagep);
718 		if (!error) {
719 			spin_lock(&info->lock);
720 			info->swapped--;
721 			spin_unlock(&info->lock);
722 			swap_free(swap);
723 		}
724 		error = 1;	/* not an error, but entry was found */
725 	}
726 	return error;
727 }
728 
729 /*
730  * Search through swapped inodes to find and replace swap by page.
731  */
732 int shmem_unuse(swp_entry_t swap, struct page *page)
733 {
734 	struct list_head *this, *next;
735 	struct shmem_inode_info *info;
736 	int found = 0;
737 	int error = 0;
738 
739 	/*
740 	 * There's a faint possibility that swap page was replaced before
741 	 * caller locked it: caller will come back later with the right page.
742 	 */
743 	if (unlikely(!PageSwapCache(page) || page_private(page) != swap.val))
744 		goto out;
745 
746 	/*
747 	 * Charge page using GFP_KERNEL while we can wait, before taking
748 	 * the shmem_swaplist_mutex which might hold up shmem_writepage().
749 	 * Charged back to the user (not to caller) when swap account is used.
750 	 */
751 	error = mem_cgroup_cache_charge(page, current->mm, GFP_KERNEL);
752 	if (error)
753 		goto out;
754 	/* No radix_tree_preload: swap entry keeps a place for page in tree */
755 
756 	mutex_lock(&shmem_swaplist_mutex);
757 	list_for_each_safe(this, next, &shmem_swaplist) {
758 		info = list_entry(this, struct shmem_inode_info, swaplist);
759 		if (info->swapped)
760 			found = shmem_unuse_inode(info, swap, &page);
761 		else
762 			list_del_init(&info->swaplist);
763 		cond_resched();
764 		if (found)
765 			break;
766 	}
767 	mutex_unlock(&shmem_swaplist_mutex);
768 
769 	if (found < 0)
770 		error = found;
771 out:
772 	unlock_page(page);
773 	page_cache_release(page);
774 	return error;
775 }
776 
777 /*
778  * Move the page from the page cache to the swap cache.
779  */
780 static int shmem_writepage(struct page *page, struct writeback_control *wbc)
781 {
782 	struct shmem_inode_info *info;
783 	struct address_space *mapping;
784 	struct inode *inode;
785 	swp_entry_t swap;
786 	pgoff_t index;
787 
788 	BUG_ON(!PageLocked(page));
789 	mapping = page->mapping;
790 	index = page->index;
791 	inode = mapping->host;
792 	info = SHMEM_I(inode);
793 	if (info->flags & VM_LOCKED)
794 		goto redirty;
795 	if (!total_swap_pages)
796 		goto redirty;
797 
798 	/*
799 	 * shmem_backing_dev_info's capabilities prevent regular writeback or
800 	 * sync from ever calling shmem_writepage; but a stacking filesystem
801 	 * might use ->writepage of its underlying filesystem, in which case
802 	 * tmpfs should write out to swap only in response to memory pressure,
803 	 * and not for the writeback threads or sync.
804 	 */
805 	if (!wbc->for_reclaim) {
806 		WARN_ON_ONCE(1);	/* Still happens? Tell us about it! */
807 		goto redirty;
808 	}
809 
810 	/*
811 	 * This is somewhat ridiculous, but without plumbing a SWAP_MAP_FALLOC
812 	 * value into swapfile.c, the only way we can correctly account for a
813 	 * fallocated page arriving here is now to initialize it and write it.
814 	 *
815 	 * That's okay for a page already fallocated earlier, but if we have
816 	 * not yet completed the fallocation, then (a) we want to keep track
817 	 * of this page in case we have to undo it, and (b) it may not be a
818 	 * good idea to continue anyway, once we're pushing into swap.  So
819 	 * reactivate the page, and let shmem_fallocate() quit when too many.
820 	 */
821 	if (!PageUptodate(page)) {
822 		if (inode->i_private) {
823 			struct shmem_falloc *shmem_falloc;
824 			spin_lock(&inode->i_lock);
825 			shmem_falloc = inode->i_private;
826 			if (shmem_falloc &&
827 			    index >= shmem_falloc->start &&
828 			    index < shmem_falloc->next)
829 				shmem_falloc->nr_unswapped++;
830 			else
831 				shmem_falloc = NULL;
832 			spin_unlock(&inode->i_lock);
833 			if (shmem_falloc)
834 				goto redirty;
835 		}
836 		clear_highpage(page);
837 		flush_dcache_page(page);
838 		SetPageUptodate(page);
839 	}
840 
841 	swap = get_swap_page();
842 	if (!swap.val)
843 		goto redirty;
844 
845 	/*
846 	 * Add inode to shmem_unuse()'s list of swapped-out inodes,
847 	 * if it's not already there.  Do it now before the page is
848 	 * moved to swap cache, when its pagelock no longer protects
849 	 * the inode from eviction.  But don't unlock the mutex until
850 	 * we've incremented swapped, because shmem_unuse_inode() will
851 	 * prune a !swapped inode from the swaplist under this mutex.
852 	 */
853 	mutex_lock(&shmem_swaplist_mutex);
854 	if (list_empty(&info->swaplist))
855 		list_add_tail(&info->swaplist, &shmem_swaplist);
856 
857 	if (add_to_swap_cache(page, swap, GFP_ATOMIC) == 0) {
858 		swap_shmem_alloc(swap);
859 		shmem_delete_from_page_cache(page, swp_to_radix_entry(swap));
860 
861 		spin_lock(&info->lock);
862 		info->swapped++;
863 		shmem_recalc_inode(inode);
864 		spin_unlock(&info->lock);
865 
866 		mutex_unlock(&shmem_swaplist_mutex);
867 		BUG_ON(page_mapped(page));
868 		swap_writepage(page, wbc);
869 		return 0;
870 	}
871 
872 	mutex_unlock(&shmem_swaplist_mutex);
873 	swapcache_free(swap, NULL);
874 redirty:
875 	set_page_dirty(page);
876 	if (wbc->for_reclaim)
877 		return AOP_WRITEPAGE_ACTIVATE;	/* Return with page locked */
878 	unlock_page(page);
879 	return 0;
880 }
881 
882 #ifdef CONFIG_NUMA
883 #ifdef CONFIG_TMPFS
884 static void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
885 {
886 	char buffer[64];
887 
888 	if (!mpol || mpol->mode == MPOL_DEFAULT)
889 		return;		/* show nothing */
890 
891 	mpol_to_str(buffer, sizeof(buffer), mpol);
892 
893 	seq_printf(seq, ",mpol=%s", buffer);
894 }
895 
896 static struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
897 {
898 	struct mempolicy *mpol = NULL;
899 	if (sbinfo->mpol) {
900 		spin_lock(&sbinfo->stat_lock);	/* prevent replace/use races */
901 		mpol = sbinfo->mpol;
902 		mpol_get(mpol);
903 		spin_unlock(&sbinfo->stat_lock);
904 	}
905 	return mpol;
906 }
907 #endif /* CONFIG_TMPFS */
908 
909 static struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
910 			struct shmem_inode_info *info, pgoff_t index)
911 {
912 	struct vm_area_struct pvma;
913 	struct page *page;
914 
915 	/* Create a pseudo vma that just contains the policy */
916 	pvma.vm_start = 0;
917 	/* Bias interleave by inode number to distribute better across nodes */
918 	pvma.vm_pgoff = index + info->vfs_inode.i_ino;
919 	pvma.vm_ops = NULL;
920 	pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
921 
922 	page = swapin_readahead(swap, gfp, &pvma, 0);
923 
924 	/* Drop reference taken by mpol_shared_policy_lookup() */
925 	mpol_cond_put(pvma.vm_policy);
926 
927 	return page;
928 }
929 
930 static struct page *shmem_alloc_page(gfp_t gfp,
931 			struct shmem_inode_info *info, pgoff_t index)
932 {
933 	struct vm_area_struct pvma;
934 	struct page *page;
935 
936 	/* Create a pseudo vma that just contains the policy */
937 	pvma.vm_start = 0;
938 	/* Bias interleave by inode number to distribute better across nodes */
939 	pvma.vm_pgoff = index + info->vfs_inode.i_ino;
940 	pvma.vm_ops = NULL;
941 	pvma.vm_policy = mpol_shared_policy_lookup(&info->policy, index);
942 
943 	page = alloc_page_vma(gfp, &pvma, 0);
944 
945 	/* Drop reference taken by mpol_shared_policy_lookup() */
946 	mpol_cond_put(pvma.vm_policy);
947 
948 	return page;
949 }
950 #else /* !CONFIG_NUMA */
951 #ifdef CONFIG_TMPFS
952 static inline void shmem_show_mpol(struct seq_file *seq, struct mempolicy *mpol)
953 {
954 }
955 #endif /* CONFIG_TMPFS */
956 
957 static inline struct page *shmem_swapin(swp_entry_t swap, gfp_t gfp,
958 			struct shmem_inode_info *info, pgoff_t index)
959 {
960 	return swapin_readahead(swap, gfp, NULL, 0);
961 }
962 
963 static inline struct page *shmem_alloc_page(gfp_t gfp,
964 			struct shmem_inode_info *info, pgoff_t index)
965 {
966 	return alloc_page(gfp);
967 }
968 #endif /* CONFIG_NUMA */
969 
970 #if !defined(CONFIG_NUMA) || !defined(CONFIG_TMPFS)
971 static inline struct mempolicy *shmem_get_sbmpol(struct shmem_sb_info *sbinfo)
972 {
973 	return NULL;
974 }
975 #endif
976 
977 /*
978  * When a page is moved from swapcache to shmem filecache (either by the
979  * usual swapin of shmem_getpage_gfp(), or by the less common swapoff of
980  * shmem_unuse_inode()), it may have been read in earlier from swap, in
981  * ignorance of the mapping it belongs to.  If that mapping has special
982  * constraints (like the gma500 GEM driver, which requires RAM below 4GB),
983  * we may need to copy to a suitable page before moving to filecache.
984  *
985  * In a future release, this may well be extended to respect cpuset and
986  * NUMA mempolicy, and applied also to anonymous pages in do_swap_page();
987  * but for now it is a simple matter of zone.
988  */
989 static bool shmem_should_replace_page(struct page *page, gfp_t gfp)
990 {
991 	return page_zonenum(page) > gfp_zone(gfp);
992 }
993 
994 static int shmem_replace_page(struct page **pagep, gfp_t gfp,
995 				struct shmem_inode_info *info, pgoff_t index)
996 {
997 	struct page *oldpage, *newpage;
998 	struct address_space *swap_mapping;
999 	pgoff_t swap_index;
1000 	int error;
1001 
1002 	oldpage = *pagep;
1003 	swap_index = page_private(oldpage);
1004 	swap_mapping = page_mapping(oldpage);
1005 
1006 	/*
1007 	 * We have arrived here because our zones are constrained, so don't
1008 	 * limit chance of success by further cpuset and node constraints.
1009 	 */
1010 	gfp &= ~GFP_CONSTRAINT_MASK;
1011 	newpage = shmem_alloc_page(gfp, info, index);
1012 	if (!newpage)
1013 		return -ENOMEM;
1014 
1015 	page_cache_get(newpage);
1016 	copy_highpage(newpage, oldpage);
1017 	flush_dcache_page(newpage);
1018 
1019 	__set_page_locked(newpage);
1020 	SetPageUptodate(newpage);
1021 	SetPageSwapBacked(newpage);
1022 	set_page_private(newpage, swap_index);
1023 	SetPageSwapCache(newpage);
1024 
1025 	/*
1026 	 * Our caller will very soon move newpage out of swapcache, but it's
1027 	 * a nice clean interface for us to replace oldpage by newpage there.
1028 	 */
1029 	spin_lock_irq(&swap_mapping->tree_lock);
1030 	error = shmem_radix_tree_replace(swap_mapping, swap_index, oldpage,
1031 								   newpage);
1032 	if (!error) {
1033 		__inc_zone_page_state(newpage, NR_FILE_PAGES);
1034 		__dec_zone_page_state(oldpage, NR_FILE_PAGES);
1035 	}
1036 	spin_unlock_irq(&swap_mapping->tree_lock);
1037 
1038 	if (unlikely(error)) {
1039 		/*
1040 		 * Is this possible?  I think not, now that our callers check
1041 		 * both PageSwapCache and page_private after getting page lock;
1042 		 * but be defensive.  Reverse old to newpage for clear and free.
1043 		 */
1044 		oldpage = newpage;
1045 	} else {
1046 		mem_cgroup_replace_page_cache(oldpage, newpage);
1047 		lru_cache_add_anon(newpage);
1048 		*pagep = newpage;
1049 	}
1050 
1051 	ClearPageSwapCache(oldpage);
1052 	set_page_private(oldpage, 0);
1053 
1054 	unlock_page(oldpage);
1055 	page_cache_release(oldpage);
1056 	page_cache_release(oldpage);
1057 	return error;
1058 }
1059 
1060 /*
1061  * shmem_getpage_gfp - find page in cache, or get from swap, or allocate
1062  *
1063  * If we allocate a new one we do not mark it dirty. That's up to the
1064  * vm. If we swap it in we mark it dirty since we also free the swap
1065  * entry since a page cannot live in both the swap and page cache
1066  */
1067 static int shmem_getpage_gfp(struct inode *inode, pgoff_t index,
1068 	struct page **pagep, enum sgp_type sgp, gfp_t gfp, int *fault_type)
1069 {
1070 	struct address_space *mapping = inode->i_mapping;
1071 	struct shmem_inode_info *info;
1072 	struct shmem_sb_info *sbinfo;
1073 	struct page *page;
1074 	swp_entry_t swap;
1075 	int error;
1076 	int once = 0;
1077 	int alloced = 0;
1078 
1079 	if (index > (MAX_LFS_FILESIZE >> PAGE_CACHE_SHIFT))
1080 		return -EFBIG;
1081 repeat:
1082 	swap.val = 0;
1083 	page = find_lock_page(mapping, index);
1084 	if (radix_tree_exceptional_entry(page)) {
1085 		swap = radix_to_swp_entry(page);
1086 		page = NULL;
1087 	}
1088 
1089 	if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1090 	    ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1091 		error = -EINVAL;
1092 		goto failed;
1093 	}
1094 
1095 	/* fallocated page? */
1096 	if (page && !PageUptodate(page)) {
1097 		if (sgp != SGP_READ)
1098 			goto clear;
1099 		unlock_page(page);
1100 		page_cache_release(page);
1101 		page = NULL;
1102 	}
1103 	if (page || (sgp == SGP_READ && !swap.val)) {
1104 		*pagep = page;
1105 		return 0;
1106 	}
1107 
1108 	/*
1109 	 * Fast cache lookup did not find it:
1110 	 * bring it back from swap or allocate.
1111 	 */
1112 	info = SHMEM_I(inode);
1113 	sbinfo = SHMEM_SB(inode->i_sb);
1114 
1115 	if (swap.val) {
1116 		/* Look it up and read it in.. */
1117 		page = lookup_swap_cache(swap);
1118 		if (!page) {
1119 			/* here we actually do the io */
1120 			if (fault_type)
1121 				*fault_type |= VM_FAULT_MAJOR;
1122 			page = shmem_swapin(swap, gfp, info, index);
1123 			if (!page) {
1124 				error = -ENOMEM;
1125 				goto failed;
1126 			}
1127 		}
1128 
1129 		/* We have to do this with page locked to prevent races */
1130 		lock_page(page);
1131 		if (!PageSwapCache(page) || page_private(page) != swap.val ||
1132 		    !shmem_confirm_swap(mapping, index, swap)) {
1133 			error = -EEXIST;	/* try again */
1134 			goto unlock;
1135 		}
1136 		if (!PageUptodate(page)) {
1137 			error = -EIO;
1138 			goto failed;
1139 		}
1140 		wait_on_page_writeback(page);
1141 
1142 		if (shmem_should_replace_page(page, gfp)) {
1143 			error = shmem_replace_page(&page, gfp, info, index);
1144 			if (error)
1145 				goto failed;
1146 		}
1147 
1148 		error = mem_cgroup_cache_charge(page, current->mm,
1149 						gfp & GFP_RECLAIM_MASK);
1150 		if (!error) {
1151 			error = shmem_add_to_page_cache(page, mapping, index,
1152 						gfp, swp_to_radix_entry(swap));
1153 			/*
1154 			 * We already confirmed swap under page lock, and make
1155 			 * no memory allocation here, so usually no possibility
1156 			 * of error; but free_swap_and_cache() only trylocks a
1157 			 * page, so it is just possible that the entry has been
1158 			 * truncated or holepunched since swap was confirmed.
1159 			 * shmem_undo_range() will have done some of the
1160 			 * unaccounting, now delete_from_swap_cache() will do
1161 			 * the rest (including mem_cgroup_uncharge_swapcache).
1162 			 * Reset swap.val? No, leave it so "failed" goes back to
1163 			 * "repeat": reading a hole and writing should succeed.
1164 			 */
1165 			if (error)
1166 				delete_from_swap_cache(page);
1167 		}
1168 		if (error)
1169 			goto failed;
1170 
1171 		spin_lock(&info->lock);
1172 		info->swapped--;
1173 		shmem_recalc_inode(inode);
1174 		spin_unlock(&info->lock);
1175 
1176 		delete_from_swap_cache(page);
1177 		set_page_dirty(page);
1178 		swap_free(swap);
1179 
1180 	} else {
1181 		if (shmem_acct_block(info->flags)) {
1182 			error = -ENOSPC;
1183 			goto failed;
1184 		}
1185 		if (sbinfo->max_blocks) {
1186 			if (percpu_counter_compare(&sbinfo->used_blocks,
1187 						sbinfo->max_blocks) >= 0) {
1188 				error = -ENOSPC;
1189 				goto unacct;
1190 			}
1191 			percpu_counter_inc(&sbinfo->used_blocks);
1192 		}
1193 
1194 		page = shmem_alloc_page(gfp, info, index);
1195 		if (!page) {
1196 			error = -ENOMEM;
1197 			goto decused;
1198 		}
1199 
1200 		SetPageSwapBacked(page);
1201 		__set_page_locked(page);
1202 		error = mem_cgroup_cache_charge(page, current->mm,
1203 						gfp & GFP_RECLAIM_MASK);
1204 		if (error)
1205 			goto decused;
1206 		error = radix_tree_maybe_preload(gfp & GFP_RECLAIM_MASK);
1207 		if (!error) {
1208 			error = shmem_add_to_page_cache(page, mapping, index,
1209 							gfp, NULL);
1210 			radix_tree_preload_end();
1211 		}
1212 		if (error) {
1213 			mem_cgroup_uncharge_cache_page(page);
1214 			goto decused;
1215 		}
1216 		lru_cache_add_anon(page);
1217 
1218 		spin_lock(&info->lock);
1219 		info->alloced++;
1220 		inode->i_blocks += BLOCKS_PER_PAGE;
1221 		shmem_recalc_inode(inode);
1222 		spin_unlock(&info->lock);
1223 		alloced = true;
1224 
1225 		/*
1226 		 * Let SGP_FALLOC use the SGP_WRITE optimization on a new page.
1227 		 */
1228 		if (sgp == SGP_FALLOC)
1229 			sgp = SGP_WRITE;
1230 clear:
1231 		/*
1232 		 * Let SGP_WRITE caller clear ends if write does not fill page;
1233 		 * but SGP_FALLOC on a page fallocated earlier must initialize
1234 		 * it now, lest undo on failure cancel our earlier guarantee.
1235 		 */
1236 		if (sgp != SGP_WRITE) {
1237 			clear_highpage(page);
1238 			flush_dcache_page(page);
1239 			SetPageUptodate(page);
1240 		}
1241 		if (sgp == SGP_DIRTY)
1242 			set_page_dirty(page);
1243 	}
1244 
1245 	/* Perhaps the file has been truncated since we checked */
1246 	if (sgp != SGP_WRITE && sgp != SGP_FALLOC &&
1247 	    ((loff_t)index << PAGE_CACHE_SHIFT) >= i_size_read(inode)) {
1248 		error = -EINVAL;
1249 		if (alloced)
1250 			goto trunc;
1251 		else
1252 			goto failed;
1253 	}
1254 	*pagep = page;
1255 	return 0;
1256 
1257 	/*
1258 	 * Error recovery.
1259 	 */
1260 trunc:
1261 	info = SHMEM_I(inode);
1262 	ClearPageDirty(page);
1263 	delete_from_page_cache(page);
1264 	spin_lock(&info->lock);
1265 	info->alloced--;
1266 	inode->i_blocks -= BLOCKS_PER_PAGE;
1267 	spin_unlock(&info->lock);
1268 decused:
1269 	sbinfo = SHMEM_SB(inode->i_sb);
1270 	if (sbinfo->max_blocks)
1271 		percpu_counter_add(&sbinfo->used_blocks, -1);
1272 unacct:
1273 	shmem_unacct_blocks(info->flags, 1);
1274 failed:
1275 	if (swap.val && error != -EINVAL &&
1276 	    !shmem_confirm_swap(mapping, index, swap))
1277 		error = -EEXIST;
1278 unlock:
1279 	if (page) {
1280 		unlock_page(page);
1281 		page_cache_release(page);
1282 	}
1283 	if (error == -ENOSPC && !once++) {
1284 		info = SHMEM_I(inode);
1285 		spin_lock(&info->lock);
1286 		shmem_recalc_inode(inode);
1287 		spin_unlock(&info->lock);
1288 		goto repeat;
1289 	}
1290 	if (error == -EEXIST)	/* from above or from radix_tree_insert */
1291 		goto repeat;
1292 	return error;
1293 }
1294 
1295 static int shmem_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1296 {
1297 	struct inode *inode = file_inode(vma->vm_file);
1298 	int error;
1299 	int ret = VM_FAULT_LOCKED;
1300 
1301 	error = shmem_getpage(inode, vmf->pgoff, &vmf->page, SGP_CACHE, &ret);
1302 	if (error)
1303 		return ((error == -ENOMEM) ? VM_FAULT_OOM : VM_FAULT_SIGBUS);
1304 
1305 	if (ret & VM_FAULT_MAJOR) {
1306 		count_vm_event(PGMAJFAULT);
1307 		mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
1308 	}
1309 	return ret;
1310 }
1311 
1312 #ifdef CONFIG_NUMA
1313 static int shmem_set_policy(struct vm_area_struct *vma, struct mempolicy *mpol)
1314 {
1315 	struct inode *inode = file_inode(vma->vm_file);
1316 	return mpol_set_shared_policy(&SHMEM_I(inode)->policy, vma, mpol);
1317 }
1318 
1319 static struct mempolicy *shmem_get_policy(struct vm_area_struct *vma,
1320 					  unsigned long addr)
1321 {
1322 	struct inode *inode = file_inode(vma->vm_file);
1323 	pgoff_t index;
1324 
1325 	index = ((addr - vma->vm_start) >> PAGE_SHIFT) + vma->vm_pgoff;
1326 	return mpol_shared_policy_lookup(&SHMEM_I(inode)->policy, index);
1327 }
1328 #endif
1329 
1330 int shmem_lock(struct file *file, int lock, struct user_struct *user)
1331 {
1332 	struct inode *inode = file_inode(file);
1333 	struct shmem_inode_info *info = SHMEM_I(inode);
1334 	int retval = -ENOMEM;
1335 
1336 	spin_lock(&info->lock);
1337 	if (lock && !(info->flags & VM_LOCKED)) {
1338 		if (!user_shm_lock(inode->i_size, user))
1339 			goto out_nomem;
1340 		info->flags |= VM_LOCKED;
1341 		mapping_set_unevictable(file->f_mapping);
1342 	}
1343 	if (!lock && (info->flags & VM_LOCKED) && user) {
1344 		user_shm_unlock(inode->i_size, user);
1345 		info->flags &= ~VM_LOCKED;
1346 		mapping_clear_unevictable(file->f_mapping);
1347 	}
1348 	retval = 0;
1349 
1350 out_nomem:
1351 	spin_unlock(&info->lock);
1352 	return retval;
1353 }
1354 
1355 static int shmem_mmap(struct file *file, struct vm_area_struct *vma)
1356 {
1357 	file_accessed(file);
1358 	vma->vm_ops = &shmem_vm_ops;
1359 	return 0;
1360 }
1361 
1362 static struct inode *shmem_get_inode(struct super_block *sb, const struct inode *dir,
1363 				     umode_t mode, dev_t dev, unsigned long flags)
1364 {
1365 	struct inode *inode;
1366 	struct shmem_inode_info *info;
1367 	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
1368 
1369 	if (shmem_reserve_inode(sb))
1370 		return NULL;
1371 
1372 	inode = new_inode(sb);
1373 	if (inode) {
1374 		inode->i_ino = get_next_ino();
1375 		inode_init_owner(inode, dir, mode);
1376 		inode->i_blocks = 0;
1377 		inode->i_mapping->backing_dev_info = &shmem_backing_dev_info;
1378 		inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
1379 		inode->i_generation = get_seconds();
1380 		info = SHMEM_I(inode);
1381 		memset(info, 0, (char *)inode - (char *)info);
1382 		spin_lock_init(&info->lock);
1383 		info->flags = flags & VM_NORESERVE;
1384 		INIT_LIST_HEAD(&info->swaplist);
1385 		simple_xattrs_init(&info->xattrs);
1386 		cache_no_acl(inode);
1387 
1388 		switch (mode & S_IFMT) {
1389 		default:
1390 			inode->i_op = &shmem_special_inode_operations;
1391 			init_special_inode(inode, mode, dev);
1392 			break;
1393 		case S_IFREG:
1394 			inode->i_mapping->a_ops = &shmem_aops;
1395 			inode->i_op = &shmem_inode_operations;
1396 			inode->i_fop = &shmem_file_operations;
1397 			mpol_shared_policy_init(&info->policy,
1398 						 shmem_get_sbmpol(sbinfo));
1399 			break;
1400 		case S_IFDIR:
1401 			inc_nlink(inode);
1402 			/* Some things misbehave if size == 0 on a directory */
1403 			inode->i_size = 2 * BOGO_DIRENT_SIZE;
1404 			inode->i_op = &shmem_dir_inode_operations;
1405 			inode->i_fop = &simple_dir_operations;
1406 			break;
1407 		case S_IFLNK:
1408 			/*
1409 			 * Must not load anything in the rbtree,
1410 			 * mpol_free_shared_policy will not be called.
1411 			 */
1412 			mpol_shared_policy_init(&info->policy, NULL);
1413 			break;
1414 		}
1415 	} else
1416 		shmem_free_inode(sb);
1417 	return inode;
1418 }
1419 
1420 #ifdef CONFIG_TMPFS
1421 static const struct inode_operations shmem_symlink_inode_operations;
1422 static const struct inode_operations shmem_short_symlink_operations;
1423 
1424 #ifdef CONFIG_TMPFS_XATTR
1425 static int shmem_initxattrs(struct inode *, const struct xattr *, void *);
1426 #else
1427 #define shmem_initxattrs NULL
1428 #endif
1429 
1430 static int
1431 shmem_write_begin(struct file *file, struct address_space *mapping,
1432 			loff_t pos, unsigned len, unsigned flags,
1433 			struct page **pagep, void **fsdata)
1434 {
1435 	struct inode *inode = mapping->host;
1436 	pgoff_t index = pos >> PAGE_CACHE_SHIFT;
1437 	return shmem_getpage(inode, index, pagep, SGP_WRITE, NULL);
1438 }
1439 
1440 static int
1441 shmem_write_end(struct file *file, struct address_space *mapping,
1442 			loff_t pos, unsigned len, unsigned copied,
1443 			struct page *page, void *fsdata)
1444 {
1445 	struct inode *inode = mapping->host;
1446 
1447 	if (pos + copied > inode->i_size)
1448 		i_size_write(inode, pos + copied);
1449 
1450 	if (!PageUptodate(page)) {
1451 		if (copied < PAGE_CACHE_SIZE) {
1452 			unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1453 			zero_user_segments(page, 0, from,
1454 					from + copied, PAGE_CACHE_SIZE);
1455 		}
1456 		SetPageUptodate(page);
1457 	}
1458 	set_page_dirty(page);
1459 	unlock_page(page);
1460 	page_cache_release(page);
1461 
1462 	return copied;
1463 }
1464 
1465 static void do_shmem_file_read(struct file *filp, loff_t *ppos, read_descriptor_t *desc, read_actor_t actor)
1466 {
1467 	struct inode *inode = file_inode(filp);
1468 	struct address_space *mapping = inode->i_mapping;
1469 	pgoff_t index;
1470 	unsigned long offset;
1471 	enum sgp_type sgp = SGP_READ;
1472 
1473 	/*
1474 	 * Might this read be for a stacking filesystem?  Then when reading
1475 	 * holes of a sparse file, we actually need to allocate those pages,
1476 	 * and even mark them dirty, so it cannot exceed the max_blocks limit.
1477 	 */
1478 	if (segment_eq(get_fs(), KERNEL_DS))
1479 		sgp = SGP_DIRTY;
1480 
1481 	index = *ppos >> PAGE_CACHE_SHIFT;
1482 	offset = *ppos & ~PAGE_CACHE_MASK;
1483 
1484 	for (;;) {
1485 		struct page *page = NULL;
1486 		pgoff_t end_index;
1487 		unsigned long nr, ret;
1488 		loff_t i_size = i_size_read(inode);
1489 
1490 		end_index = i_size >> PAGE_CACHE_SHIFT;
1491 		if (index > end_index)
1492 			break;
1493 		if (index == end_index) {
1494 			nr = i_size & ~PAGE_CACHE_MASK;
1495 			if (nr <= offset)
1496 				break;
1497 		}
1498 
1499 		desc->error = shmem_getpage(inode, index, &page, sgp, NULL);
1500 		if (desc->error) {
1501 			if (desc->error == -EINVAL)
1502 				desc->error = 0;
1503 			break;
1504 		}
1505 		if (page)
1506 			unlock_page(page);
1507 
1508 		/*
1509 		 * We must evaluate after, since reads (unlike writes)
1510 		 * are called without i_mutex protection against truncate
1511 		 */
1512 		nr = PAGE_CACHE_SIZE;
1513 		i_size = i_size_read(inode);
1514 		end_index = i_size >> PAGE_CACHE_SHIFT;
1515 		if (index == end_index) {
1516 			nr = i_size & ~PAGE_CACHE_MASK;
1517 			if (nr <= offset) {
1518 				if (page)
1519 					page_cache_release(page);
1520 				break;
1521 			}
1522 		}
1523 		nr -= offset;
1524 
1525 		if (page) {
1526 			/*
1527 			 * If users can be writing to this page using arbitrary
1528 			 * virtual addresses, take care about potential aliasing
1529 			 * before reading the page on the kernel side.
1530 			 */
1531 			if (mapping_writably_mapped(mapping))
1532 				flush_dcache_page(page);
1533 			/*
1534 			 * Mark the page accessed if we read the beginning.
1535 			 */
1536 			if (!offset)
1537 				mark_page_accessed(page);
1538 		} else {
1539 			page = ZERO_PAGE(0);
1540 			page_cache_get(page);
1541 		}
1542 
1543 		/*
1544 		 * Ok, we have the page, and it's up-to-date, so
1545 		 * now we can copy it to user space...
1546 		 *
1547 		 * The actor routine returns how many bytes were actually used..
1548 		 * NOTE! This may not be the same as how much of a user buffer
1549 		 * we filled up (we may be padding etc), so we can only update
1550 		 * "pos" here (the actor routine has to update the user buffer
1551 		 * pointers and the remaining count).
1552 		 */
1553 		ret = actor(desc, page, offset, nr);
1554 		offset += ret;
1555 		index += offset >> PAGE_CACHE_SHIFT;
1556 		offset &= ~PAGE_CACHE_MASK;
1557 
1558 		page_cache_release(page);
1559 		if (ret != nr || !desc->count)
1560 			break;
1561 
1562 		cond_resched();
1563 	}
1564 
1565 	*ppos = ((loff_t) index << PAGE_CACHE_SHIFT) + offset;
1566 	file_accessed(filp);
1567 }
1568 
1569 static ssize_t shmem_file_aio_read(struct kiocb *iocb,
1570 		const struct iovec *iov, unsigned long nr_segs, loff_t pos)
1571 {
1572 	struct file *filp = iocb->ki_filp;
1573 	ssize_t retval;
1574 	unsigned long seg;
1575 	size_t count;
1576 	loff_t *ppos = &iocb->ki_pos;
1577 
1578 	retval = generic_segment_checks(iov, &nr_segs, &count, VERIFY_WRITE);
1579 	if (retval)
1580 		return retval;
1581 
1582 	for (seg = 0; seg < nr_segs; seg++) {
1583 		read_descriptor_t desc;
1584 
1585 		desc.written = 0;
1586 		desc.arg.buf = iov[seg].iov_base;
1587 		desc.count = iov[seg].iov_len;
1588 		if (desc.count == 0)
1589 			continue;
1590 		desc.error = 0;
1591 		do_shmem_file_read(filp, ppos, &desc, file_read_actor);
1592 		retval += desc.written;
1593 		if (desc.error) {
1594 			retval = retval ?: desc.error;
1595 			break;
1596 		}
1597 		if (desc.count > 0)
1598 			break;
1599 	}
1600 	return retval;
1601 }
1602 
1603 static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
1604 				struct pipe_inode_info *pipe, size_t len,
1605 				unsigned int flags)
1606 {
1607 	struct address_space *mapping = in->f_mapping;
1608 	struct inode *inode = mapping->host;
1609 	unsigned int loff, nr_pages, req_pages;
1610 	struct page *pages[PIPE_DEF_BUFFERS];
1611 	struct partial_page partial[PIPE_DEF_BUFFERS];
1612 	struct page *page;
1613 	pgoff_t index, end_index;
1614 	loff_t isize, left;
1615 	int error, page_nr;
1616 	struct splice_pipe_desc spd = {
1617 		.pages = pages,
1618 		.partial = partial,
1619 		.nr_pages_max = PIPE_DEF_BUFFERS,
1620 		.flags = flags,
1621 		.ops = &page_cache_pipe_buf_ops,
1622 		.spd_release = spd_release_page,
1623 	};
1624 
1625 	isize = i_size_read(inode);
1626 	if (unlikely(*ppos >= isize))
1627 		return 0;
1628 
1629 	left = isize - *ppos;
1630 	if (unlikely(left < len))
1631 		len = left;
1632 
1633 	if (splice_grow_spd(pipe, &spd))
1634 		return -ENOMEM;
1635 
1636 	index = *ppos >> PAGE_CACHE_SHIFT;
1637 	loff = *ppos & ~PAGE_CACHE_MASK;
1638 	req_pages = (len + loff + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1639 	nr_pages = min(req_pages, pipe->buffers);
1640 
1641 	spd.nr_pages = find_get_pages_contig(mapping, index,
1642 						nr_pages, spd.pages);
1643 	index += spd.nr_pages;
1644 	error = 0;
1645 
1646 	while (spd.nr_pages < nr_pages) {
1647 		error = shmem_getpage(inode, index, &page, SGP_CACHE, NULL);
1648 		if (error)
1649 			break;
1650 		unlock_page(page);
1651 		spd.pages[spd.nr_pages++] = page;
1652 		index++;
1653 	}
1654 
1655 	index = *ppos >> PAGE_CACHE_SHIFT;
1656 	nr_pages = spd.nr_pages;
1657 	spd.nr_pages = 0;
1658 
1659 	for (page_nr = 0; page_nr < nr_pages; page_nr++) {
1660 		unsigned int this_len;
1661 
1662 		if (!len)
1663 			break;
1664 
1665 		this_len = min_t(unsigned long, len, PAGE_CACHE_SIZE - loff);
1666 		page = spd.pages[page_nr];
1667 
1668 		if (!PageUptodate(page) || page->mapping != mapping) {
1669 			error = shmem_getpage(inode, index, &page,
1670 							SGP_CACHE, NULL);
1671 			if (error)
1672 				break;
1673 			unlock_page(page);
1674 			page_cache_release(spd.pages[page_nr]);
1675 			spd.pages[page_nr] = page;
1676 		}
1677 
1678 		isize = i_size_read(inode);
1679 		end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
1680 		if (unlikely(!isize || index > end_index))
1681 			break;
1682 
1683 		if (end_index == index) {
1684 			unsigned int plen;
1685 
1686 			plen = ((isize - 1) & ~PAGE_CACHE_MASK) + 1;
1687 			if (plen <= loff)
1688 				break;
1689 
1690 			this_len = min(this_len, plen - loff);
1691 			len = this_len;
1692 		}
1693 
1694 		spd.partial[page_nr].offset = loff;
1695 		spd.partial[page_nr].len = this_len;
1696 		len -= this_len;
1697 		loff = 0;
1698 		spd.nr_pages++;
1699 		index++;
1700 	}
1701 
1702 	while (page_nr < nr_pages)
1703 		page_cache_release(spd.pages[page_nr++]);
1704 
1705 	if (spd.nr_pages)
1706 		error = splice_to_pipe(pipe, &spd);
1707 
1708 	splice_shrink_spd(&spd);
1709 
1710 	if (error > 0) {
1711 		*ppos += error;
1712 		file_accessed(in);
1713 	}
1714 	return error;
1715 }
1716 
1717 /*
1718  * llseek SEEK_DATA or SEEK_HOLE through the radix_tree.
1719  */
1720 static pgoff_t shmem_seek_hole_data(struct address_space *mapping,
1721 				    pgoff_t index, pgoff_t end, int whence)
1722 {
1723 	struct page *page;
1724 	struct pagevec pvec;
1725 	pgoff_t indices[PAGEVEC_SIZE];
1726 	bool done = false;
1727 	int i;
1728 
1729 	pagevec_init(&pvec, 0);
1730 	pvec.nr = 1;		/* start small: we may be there already */
1731 	while (!done) {
1732 		pvec.nr = shmem_find_get_pages_and_swap(mapping, index,
1733 					pvec.nr, pvec.pages, indices);
1734 		if (!pvec.nr) {
1735 			if (whence == SEEK_DATA)
1736 				index = end;
1737 			break;
1738 		}
1739 		for (i = 0; i < pvec.nr; i++, index++) {
1740 			if (index < indices[i]) {
1741 				if (whence == SEEK_HOLE) {
1742 					done = true;
1743 					break;
1744 				}
1745 				index = indices[i];
1746 			}
1747 			page = pvec.pages[i];
1748 			if (page && !radix_tree_exceptional_entry(page)) {
1749 				if (!PageUptodate(page))
1750 					page = NULL;
1751 			}
1752 			if (index >= end ||
1753 			    (page && whence == SEEK_DATA) ||
1754 			    (!page && whence == SEEK_HOLE)) {
1755 				done = true;
1756 				break;
1757 			}
1758 		}
1759 		shmem_deswap_pagevec(&pvec);
1760 		pagevec_release(&pvec);
1761 		pvec.nr = PAGEVEC_SIZE;
1762 		cond_resched();
1763 	}
1764 	return index;
1765 }
1766 
1767 static loff_t shmem_file_llseek(struct file *file, loff_t offset, int whence)
1768 {
1769 	struct address_space *mapping = file->f_mapping;
1770 	struct inode *inode = mapping->host;
1771 	pgoff_t start, end;
1772 	loff_t new_offset;
1773 
1774 	if (whence != SEEK_DATA && whence != SEEK_HOLE)
1775 		return generic_file_llseek_size(file, offset, whence,
1776 					MAX_LFS_FILESIZE, i_size_read(inode));
1777 	mutex_lock(&inode->i_mutex);
1778 	/* We're holding i_mutex so we can access i_size directly */
1779 
1780 	if (offset < 0)
1781 		offset = -EINVAL;
1782 	else if (offset >= inode->i_size)
1783 		offset = -ENXIO;
1784 	else {
1785 		start = offset >> PAGE_CACHE_SHIFT;
1786 		end = (inode->i_size + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1787 		new_offset = shmem_seek_hole_data(mapping, start, end, whence);
1788 		new_offset <<= PAGE_CACHE_SHIFT;
1789 		if (new_offset > offset) {
1790 			if (new_offset < inode->i_size)
1791 				offset = new_offset;
1792 			else if (whence == SEEK_DATA)
1793 				offset = -ENXIO;
1794 			else
1795 				offset = inode->i_size;
1796 		}
1797 	}
1798 
1799 	if (offset >= 0)
1800 		offset = vfs_setpos(file, offset, MAX_LFS_FILESIZE);
1801 	mutex_unlock(&inode->i_mutex);
1802 	return offset;
1803 }
1804 
1805 static long shmem_fallocate(struct file *file, int mode, loff_t offset,
1806 							 loff_t len)
1807 {
1808 	struct inode *inode = file_inode(file);
1809 	struct shmem_sb_info *sbinfo = SHMEM_SB(inode->i_sb);
1810 	struct shmem_falloc shmem_falloc;
1811 	pgoff_t start, index, end;
1812 	int error;
1813 
1814 	mutex_lock(&inode->i_mutex);
1815 
1816 	if (mode & FALLOC_FL_PUNCH_HOLE) {
1817 		struct address_space *mapping = file->f_mapping;
1818 		loff_t unmap_start = round_up(offset, PAGE_SIZE);
1819 		loff_t unmap_end = round_down(offset + len, PAGE_SIZE) - 1;
1820 
1821 		if ((u64)unmap_end > (u64)unmap_start)
1822 			unmap_mapping_range(mapping, unmap_start,
1823 					    1 + unmap_end - unmap_start, 0);
1824 		shmem_truncate_range(inode, offset, offset + len - 1);
1825 		/* No need to unmap again: hole-punching leaves COWed pages */
1826 		error = 0;
1827 		goto out;
1828 	}
1829 
1830 	/* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
1831 	error = inode_newsize_ok(inode, offset + len);
1832 	if (error)
1833 		goto out;
1834 
1835 	start = offset >> PAGE_CACHE_SHIFT;
1836 	end = (offset + len + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
1837 	/* Try to avoid a swapstorm if len is impossible to satisfy */
1838 	if (sbinfo->max_blocks && end - start > sbinfo->max_blocks) {
1839 		error = -ENOSPC;
1840 		goto out;
1841 	}
1842 
1843 	shmem_falloc.start = start;
1844 	shmem_falloc.next  = start;
1845 	shmem_falloc.nr_falloced = 0;
1846 	shmem_falloc.nr_unswapped = 0;
1847 	spin_lock(&inode->i_lock);
1848 	inode->i_private = &shmem_falloc;
1849 	spin_unlock(&inode->i_lock);
1850 
1851 	for (index = start; index < end; index++) {
1852 		struct page *page;
1853 
1854 		/*
1855 		 * Good, the fallocate(2) manpage permits EINTR: we may have
1856 		 * been interrupted because we are using up too much memory.
1857 		 */
1858 		if (signal_pending(current))
1859 			error = -EINTR;
1860 		else if (shmem_falloc.nr_unswapped > shmem_falloc.nr_falloced)
1861 			error = -ENOMEM;
1862 		else
1863 			error = shmem_getpage(inode, index, &page, SGP_FALLOC,
1864 									NULL);
1865 		if (error) {
1866 			/* Remove the !PageUptodate pages we added */
1867 			shmem_undo_range(inode,
1868 				(loff_t)start << PAGE_CACHE_SHIFT,
1869 				(loff_t)index << PAGE_CACHE_SHIFT, true);
1870 			goto undone;
1871 		}
1872 
1873 		/*
1874 		 * Inform shmem_writepage() how far we have reached.
1875 		 * No need for lock or barrier: we have the page lock.
1876 		 */
1877 		shmem_falloc.next++;
1878 		if (!PageUptodate(page))
1879 			shmem_falloc.nr_falloced++;
1880 
1881 		/*
1882 		 * If !PageUptodate, leave it that way so that freeable pages
1883 		 * can be recognized if we need to rollback on error later.
1884 		 * But set_page_dirty so that memory pressure will swap rather
1885 		 * than free the pages we are allocating (and SGP_CACHE pages
1886 		 * might still be clean: we now need to mark those dirty too).
1887 		 */
1888 		set_page_dirty(page);
1889 		unlock_page(page);
1890 		page_cache_release(page);
1891 		cond_resched();
1892 	}
1893 
1894 	if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
1895 		i_size_write(inode, offset + len);
1896 	inode->i_ctime = CURRENT_TIME;
1897 undone:
1898 	spin_lock(&inode->i_lock);
1899 	inode->i_private = NULL;
1900 	spin_unlock(&inode->i_lock);
1901 out:
1902 	mutex_unlock(&inode->i_mutex);
1903 	return error;
1904 }
1905 
1906 static int shmem_statfs(struct dentry *dentry, struct kstatfs *buf)
1907 {
1908 	struct shmem_sb_info *sbinfo = SHMEM_SB(dentry->d_sb);
1909 
1910 	buf->f_type = TMPFS_MAGIC;
1911 	buf->f_bsize = PAGE_CACHE_SIZE;
1912 	buf->f_namelen = NAME_MAX;
1913 	if (sbinfo->max_blocks) {
1914 		buf->f_blocks = sbinfo->max_blocks;
1915 		buf->f_bavail =
1916 		buf->f_bfree  = sbinfo->max_blocks -
1917 				percpu_counter_sum(&sbinfo->used_blocks);
1918 	}
1919 	if (sbinfo->max_inodes) {
1920 		buf->f_files = sbinfo->max_inodes;
1921 		buf->f_ffree = sbinfo->free_inodes;
1922 	}
1923 	/* else leave those fields 0 like simple_statfs */
1924 	return 0;
1925 }
1926 
1927 /*
1928  * File creation. Allocate an inode, and we're done..
1929  */
1930 static int
1931 shmem_mknod(struct inode *dir, struct dentry *dentry, umode_t mode, dev_t dev)
1932 {
1933 	struct inode *inode;
1934 	int error = -ENOSPC;
1935 
1936 	inode = shmem_get_inode(dir->i_sb, dir, mode, dev, VM_NORESERVE);
1937 	if (inode) {
1938 		error = simple_acl_create(dir, inode);
1939 		if (error)
1940 			goto out_iput;
1941 		error = security_inode_init_security(inode, dir,
1942 						     &dentry->d_name,
1943 						     shmem_initxattrs, NULL);
1944 		if (error && error != -EOPNOTSUPP)
1945 			goto out_iput;
1946 
1947 		error = 0;
1948 		dir->i_size += BOGO_DIRENT_SIZE;
1949 		dir->i_ctime = dir->i_mtime = CURRENT_TIME;
1950 		d_instantiate(dentry, inode);
1951 		dget(dentry); /* Extra count - pin the dentry in core */
1952 	}
1953 	return error;
1954 out_iput:
1955 	iput(inode);
1956 	return error;
1957 }
1958 
1959 static int
1960 shmem_tmpfile(struct inode *dir, struct dentry *dentry, umode_t mode)
1961 {
1962 	struct inode *inode;
1963 	int error = -ENOSPC;
1964 
1965 	inode = shmem_get_inode(dir->i_sb, dir, mode, 0, VM_NORESERVE);
1966 	if (inode) {
1967 		error = security_inode_init_security(inode, dir,
1968 						     NULL,
1969 						     shmem_initxattrs, NULL);
1970 		if (error && error != -EOPNOTSUPP)
1971 			goto out_iput;
1972 		error = simple_acl_create(dir, inode);
1973 		if (error)
1974 			goto out_iput;
1975 		d_tmpfile(dentry, inode);
1976 	}
1977 	return error;
1978 out_iput:
1979 	iput(inode);
1980 	return error;
1981 }
1982 
1983 static int shmem_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
1984 {
1985 	int error;
1986 
1987 	if ((error = shmem_mknod(dir, dentry, mode | S_IFDIR, 0)))
1988 		return error;
1989 	inc_nlink(dir);
1990 	return 0;
1991 }
1992 
1993 static int shmem_create(struct inode *dir, struct dentry *dentry, umode_t mode,
1994 		bool excl)
1995 {
1996 	return shmem_mknod(dir, dentry, mode | S_IFREG, 0);
1997 }
1998 
1999 /*
2000  * Link a file..
2001  */
2002 static int shmem_link(struct dentry *old_dentry, struct inode *dir, struct dentry *dentry)
2003 {
2004 	struct inode *inode = old_dentry->d_inode;
2005 	int ret;
2006 
2007 	/*
2008 	 * No ordinary (disk based) filesystem counts links as inodes;
2009 	 * but each new link needs a new dentry, pinning lowmem, and
2010 	 * tmpfs dentries cannot be pruned until they are unlinked.
2011 	 */
2012 	ret = shmem_reserve_inode(inode->i_sb);
2013 	if (ret)
2014 		goto out;
2015 
2016 	dir->i_size += BOGO_DIRENT_SIZE;
2017 	inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2018 	inc_nlink(inode);
2019 	ihold(inode);	/* New dentry reference */
2020 	dget(dentry);		/* Extra pinning count for the created dentry */
2021 	d_instantiate(dentry, inode);
2022 out:
2023 	return ret;
2024 }
2025 
2026 static int shmem_unlink(struct inode *dir, struct dentry *dentry)
2027 {
2028 	struct inode *inode = dentry->d_inode;
2029 
2030 	if (inode->i_nlink > 1 && !S_ISDIR(inode->i_mode))
2031 		shmem_free_inode(inode->i_sb);
2032 
2033 	dir->i_size -= BOGO_DIRENT_SIZE;
2034 	inode->i_ctime = dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2035 	drop_nlink(inode);
2036 	dput(dentry);	/* Undo the count from "create" - this does all the work */
2037 	return 0;
2038 }
2039 
2040 static int shmem_rmdir(struct inode *dir, struct dentry *dentry)
2041 {
2042 	if (!simple_empty(dentry))
2043 		return -ENOTEMPTY;
2044 
2045 	drop_nlink(dentry->d_inode);
2046 	drop_nlink(dir);
2047 	return shmem_unlink(dir, dentry);
2048 }
2049 
2050 /*
2051  * The VFS layer already does all the dentry stuff for rename,
2052  * we just have to decrement the usage count for the target if
2053  * it exists so that the VFS layer correctly free's it when it
2054  * gets overwritten.
2055  */
2056 static int shmem_rename(struct inode *old_dir, struct dentry *old_dentry, struct inode *new_dir, struct dentry *new_dentry)
2057 {
2058 	struct inode *inode = old_dentry->d_inode;
2059 	int they_are_dirs = S_ISDIR(inode->i_mode);
2060 
2061 	if (!simple_empty(new_dentry))
2062 		return -ENOTEMPTY;
2063 
2064 	if (new_dentry->d_inode) {
2065 		(void) shmem_unlink(new_dir, new_dentry);
2066 		if (they_are_dirs)
2067 			drop_nlink(old_dir);
2068 	} else if (they_are_dirs) {
2069 		drop_nlink(old_dir);
2070 		inc_nlink(new_dir);
2071 	}
2072 
2073 	old_dir->i_size -= BOGO_DIRENT_SIZE;
2074 	new_dir->i_size += BOGO_DIRENT_SIZE;
2075 	old_dir->i_ctime = old_dir->i_mtime =
2076 	new_dir->i_ctime = new_dir->i_mtime =
2077 	inode->i_ctime = CURRENT_TIME;
2078 	return 0;
2079 }
2080 
2081 static int shmem_symlink(struct inode *dir, struct dentry *dentry, const char *symname)
2082 {
2083 	int error;
2084 	int len;
2085 	struct inode *inode;
2086 	struct page *page;
2087 	char *kaddr;
2088 	struct shmem_inode_info *info;
2089 
2090 	len = strlen(symname) + 1;
2091 	if (len > PAGE_CACHE_SIZE)
2092 		return -ENAMETOOLONG;
2093 
2094 	inode = shmem_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0, VM_NORESERVE);
2095 	if (!inode)
2096 		return -ENOSPC;
2097 
2098 	error = security_inode_init_security(inode, dir, &dentry->d_name,
2099 					     shmem_initxattrs, NULL);
2100 	if (error) {
2101 		if (error != -EOPNOTSUPP) {
2102 			iput(inode);
2103 			return error;
2104 		}
2105 		error = 0;
2106 	}
2107 
2108 	info = SHMEM_I(inode);
2109 	inode->i_size = len-1;
2110 	if (len <= SHORT_SYMLINK_LEN) {
2111 		info->symlink = kmemdup(symname, len, GFP_KERNEL);
2112 		if (!info->symlink) {
2113 			iput(inode);
2114 			return -ENOMEM;
2115 		}
2116 		inode->i_op = &shmem_short_symlink_operations;
2117 	} else {
2118 		error = shmem_getpage(inode, 0, &page, SGP_WRITE, NULL);
2119 		if (error) {
2120 			iput(inode);
2121 			return error;
2122 		}
2123 		inode->i_mapping->a_ops = &shmem_aops;
2124 		inode->i_op = &shmem_symlink_inode_operations;
2125 		kaddr = kmap_atomic(page);
2126 		memcpy(kaddr, symname, len);
2127 		kunmap_atomic(kaddr);
2128 		SetPageUptodate(page);
2129 		set_page_dirty(page);
2130 		unlock_page(page);
2131 		page_cache_release(page);
2132 	}
2133 	dir->i_size += BOGO_DIRENT_SIZE;
2134 	dir->i_ctime = dir->i_mtime = CURRENT_TIME;
2135 	d_instantiate(dentry, inode);
2136 	dget(dentry);
2137 	return 0;
2138 }
2139 
2140 static void *shmem_follow_short_symlink(struct dentry *dentry, struct nameidata *nd)
2141 {
2142 	nd_set_link(nd, SHMEM_I(dentry->d_inode)->symlink);
2143 	return NULL;
2144 }
2145 
2146 static void *shmem_follow_link(struct dentry *dentry, struct nameidata *nd)
2147 {
2148 	struct page *page = NULL;
2149 	int error = shmem_getpage(dentry->d_inode, 0, &page, SGP_READ, NULL);
2150 	nd_set_link(nd, error ? ERR_PTR(error) : kmap(page));
2151 	if (page)
2152 		unlock_page(page);
2153 	return page;
2154 }
2155 
2156 static void shmem_put_link(struct dentry *dentry, struct nameidata *nd, void *cookie)
2157 {
2158 	if (!IS_ERR(nd_get_link(nd))) {
2159 		struct page *page = cookie;
2160 		kunmap(page);
2161 		mark_page_accessed(page);
2162 		page_cache_release(page);
2163 	}
2164 }
2165 
2166 #ifdef CONFIG_TMPFS_XATTR
2167 /*
2168  * Superblocks without xattr inode operations may get some security.* xattr
2169  * support from the LSM "for free". As soon as we have any other xattrs
2170  * like ACLs, we also need to implement the security.* handlers at
2171  * filesystem level, though.
2172  */
2173 
2174 /*
2175  * Callback for security_inode_init_security() for acquiring xattrs.
2176  */
2177 static int shmem_initxattrs(struct inode *inode,
2178 			    const struct xattr *xattr_array,
2179 			    void *fs_info)
2180 {
2181 	struct shmem_inode_info *info = SHMEM_I(inode);
2182 	const struct xattr *xattr;
2183 	struct simple_xattr *new_xattr;
2184 	size_t len;
2185 
2186 	for (xattr = xattr_array; xattr->name != NULL; xattr++) {
2187 		new_xattr = simple_xattr_alloc(xattr->value, xattr->value_len);
2188 		if (!new_xattr)
2189 			return -ENOMEM;
2190 
2191 		len = strlen(xattr->name) + 1;
2192 		new_xattr->name = kmalloc(XATTR_SECURITY_PREFIX_LEN + len,
2193 					  GFP_KERNEL);
2194 		if (!new_xattr->name) {
2195 			kfree(new_xattr);
2196 			return -ENOMEM;
2197 		}
2198 
2199 		memcpy(new_xattr->name, XATTR_SECURITY_PREFIX,
2200 		       XATTR_SECURITY_PREFIX_LEN);
2201 		memcpy(new_xattr->name + XATTR_SECURITY_PREFIX_LEN,
2202 		       xattr->name, len);
2203 
2204 		simple_xattr_list_add(&info->xattrs, new_xattr);
2205 	}
2206 
2207 	return 0;
2208 }
2209 
2210 static const struct xattr_handler *shmem_xattr_handlers[] = {
2211 #ifdef CONFIG_TMPFS_POSIX_ACL
2212 	&posix_acl_access_xattr_handler,
2213 	&posix_acl_default_xattr_handler,
2214 #endif
2215 	NULL
2216 };
2217 
2218 static int shmem_xattr_validate(const char *name)
2219 {
2220 	struct { const char *prefix; size_t len; } arr[] = {
2221 		{ XATTR_SECURITY_PREFIX, XATTR_SECURITY_PREFIX_LEN },
2222 		{ XATTR_TRUSTED_PREFIX, XATTR_TRUSTED_PREFIX_LEN }
2223 	};
2224 	int i;
2225 
2226 	for (i = 0; i < ARRAY_SIZE(arr); i++) {
2227 		size_t preflen = arr[i].len;
2228 		if (strncmp(name, arr[i].prefix, preflen) == 0) {
2229 			if (!name[preflen])
2230 				return -EINVAL;
2231 			return 0;
2232 		}
2233 	}
2234 	return -EOPNOTSUPP;
2235 }
2236 
2237 static ssize_t shmem_getxattr(struct dentry *dentry, const char *name,
2238 			      void *buffer, size_t size)
2239 {
2240 	struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2241 	int err;
2242 
2243 	/*
2244 	 * If this is a request for a synthetic attribute in the system.*
2245 	 * namespace use the generic infrastructure to resolve a handler
2246 	 * for it via sb->s_xattr.
2247 	 */
2248 	if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2249 		return generic_getxattr(dentry, name, buffer, size);
2250 
2251 	err = shmem_xattr_validate(name);
2252 	if (err)
2253 		return err;
2254 
2255 	return simple_xattr_get(&info->xattrs, name, buffer, size);
2256 }
2257 
2258 static int shmem_setxattr(struct dentry *dentry, const char *name,
2259 			  const void *value, size_t size, int flags)
2260 {
2261 	struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2262 	int err;
2263 
2264 	/*
2265 	 * If this is a request for a synthetic attribute in the system.*
2266 	 * namespace use the generic infrastructure to resolve a handler
2267 	 * for it via sb->s_xattr.
2268 	 */
2269 	if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2270 		return generic_setxattr(dentry, name, value, size, flags);
2271 
2272 	err = shmem_xattr_validate(name);
2273 	if (err)
2274 		return err;
2275 
2276 	return simple_xattr_set(&info->xattrs, name, value, size, flags);
2277 }
2278 
2279 static int shmem_removexattr(struct dentry *dentry, const char *name)
2280 {
2281 	struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2282 	int err;
2283 
2284 	/*
2285 	 * If this is a request for a synthetic attribute in the system.*
2286 	 * namespace use the generic infrastructure to resolve a handler
2287 	 * for it via sb->s_xattr.
2288 	 */
2289 	if (!strncmp(name, XATTR_SYSTEM_PREFIX, XATTR_SYSTEM_PREFIX_LEN))
2290 		return generic_removexattr(dentry, name);
2291 
2292 	err = shmem_xattr_validate(name);
2293 	if (err)
2294 		return err;
2295 
2296 	return simple_xattr_remove(&info->xattrs, name);
2297 }
2298 
2299 static ssize_t shmem_listxattr(struct dentry *dentry, char *buffer, size_t size)
2300 {
2301 	struct shmem_inode_info *info = SHMEM_I(dentry->d_inode);
2302 	return simple_xattr_list(&info->xattrs, buffer, size);
2303 }
2304 #endif /* CONFIG_TMPFS_XATTR */
2305 
2306 static const struct inode_operations shmem_short_symlink_operations = {
2307 	.readlink	= generic_readlink,
2308 	.follow_link	= shmem_follow_short_symlink,
2309 #ifdef CONFIG_TMPFS_XATTR
2310 	.setxattr	= shmem_setxattr,
2311 	.getxattr	= shmem_getxattr,
2312 	.listxattr	= shmem_listxattr,
2313 	.removexattr	= shmem_removexattr,
2314 #endif
2315 };
2316 
2317 static const struct inode_operations shmem_symlink_inode_operations = {
2318 	.readlink	= generic_readlink,
2319 	.follow_link	= shmem_follow_link,
2320 	.put_link	= shmem_put_link,
2321 #ifdef CONFIG_TMPFS_XATTR
2322 	.setxattr	= shmem_setxattr,
2323 	.getxattr	= shmem_getxattr,
2324 	.listxattr	= shmem_listxattr,
2325 	.removexattr	= shmem_removexattr,
2326 #endif
2327 };
2328 
2329 static struct dentry *shmem_get_parent(struct dentry *child)
2330 {
2331 	return ERR_PTR(-ESTALE);
2332 }
2333 
2334 static int shmem_match(struct inode *ino, void *vfh)
2335 {
2336 	__u32 *fh = vfh;
2337 	__u64 inum = fh[2];
2338 	inum = (inum << 32) | fh[1];
2339 	return ino->i_ino == inum && fh[0] == ino->i_generation;
2340 }
2341 
2342 static struct dentry *shmem_fh_to_dentry(struct super_block *sb,
2343 		struct fid *fid, int fh_len, int fh_type)
2344 {
2345 	struct inode *inode;
2346 	struct dentry *dentry = NULL;
2347 	u64 inum;
2348 
2349 	if (fh_len < 3)
2350 		return NULL;
2351 
2352 	inum = fid->raw[2];
2353 	inum = (inum << 32) | fid->raw[1];
2354 
2355 	inode = ilookup5(sb, (unsigned long)(inum + fid->raw[0]),
2356 			shmem_match, fid->raw);
2357 	if (inode) {
2358 		dentry = d_find_alias(inode);
2359 		iput(inode);
2360 	}
2361 
2362 	return dentry;
2363 }
2364 
2365 static int shmem_encode_fh(struct inode *inode, __u32 *fh, int *len,
2366 				struct inode *parent)
2367 {
2368 	if (*len < 3) {
2369 		*len = 3;
2370 		return FILEID_INVALID;
2371 	}
2372 
2373 	if (inode_unhashed(inode)) {
2374 		/* Unfortunately insert_inode_hash is not idempotent,
2375 		 * so as we hash inodes here rather than at creation
2376 		 * time, we need a lock to ensure we only try
2377 		 * to do it once
2378 		 */
2379 		static DEFINE_SPINLOCK(lock);
2380 		spin_lock(&lock);
2381 		if (inode_unhashed(inode))
2382 			__insert_inode_hash(inode,
2383 					    inode->i_ino + inode->i_generation);
2384 		spin_unlock(&lock);
2385 	}
2386 
2387 	fh[0] = inode->i_generation;
2388 	fh[1] = inode->i_ino;
2389 	fh[2] = ((__u64)inode->i_ino) >> 32;
2390 
2391 	*len = 3;
2392 	return 1;
2393 }
2394 
2395 static const struct export_operations shmem_export_ops = {
2396 	.get_parent     = shmem_get_parent,
2397 	.encode_fh      = shmem_encode_fh,
2398 	.fh_to_dentry	= shmem_fh_to_dentry,
2399 };
2400 
2401 static int shmem_parse_options(char *options, struct shmem_sb_info *sbinfo,
2402 			       bool remount)
2403 {
2404 	char *this_char, *value, *rest;
2405 	struct mempolicy *mpol = NULL;
2406 	uid_t uid;
2407 	gid_t gid;
2408 
2409 	while (options != NULL) {
2410 		this_char = options;
2411 		for (;;) {
2412 			/*
2413 			 * NUL-terminate this option: unfortunately,
2414 			 * mount options form a comma-separated list,
2415 			 * but mpol's nodelist may also contain commas.
2416 			 */
2417 			options = strchr(options, ',');
2418 			if (options == NULL)
2419 				break;
2420 			options++;
2421 			if (!isdigit(*options)) {
2422 				options[-1] = '\0';
2423 				break;
2424 			}
2425 		}
2426 		if (!*this_char)
2427 			continue;
2428 		if ((value = strchr(this_char,'=')) != NULL) {
2429 			*value++ = 0;
2430 		} else {
2431 			printk(KERN_ERR
2432 			    "tmpfs: No value for mount option '%s'\n",
2433 			    this_char);
2434 			goto error;
2435 		}
2436 
2437 		if (!strcmp(this_char,"size")) {
2438 			unsigned long long size;
2439 			size = memparse(value,&rest);
2440 			if (*rest == '%') {
2441 				size <<= PAGE_SHIFT;
2442 				size *= totalram_pages;
2443 				do_div(size, 100);
2444 				rest++;
2445 			}
2446 			if (*rest)
2447 				goto bad_val;
2448 			sbinfo->max_blocks =
2449 				DIV_ROUND_UP(size, PAGE_CACHE_SIZE);
2450 		} else if (!strcmp(this_char,"nr_blocks")) {
2451 			sbinfo->max_blocks = memparse(value, &rest);
2452 			if (*rest)
2453 				goto bad_val;
2454 		} else if (!strcmp(this_char,"nr_inodes")) {
2455 			sbinfo->max_inodes = memparse(value, &rest);
2456 			if (*rest)
2457 				goto bad_val;
2458 		} else if (!strcmp(this_char,"mode")) {
2459 			if (remount)
2460 				continue;
2461 			sbinfo->mode = simple_strtoul(value, &rest, 8) & 07777;
2462 			if (*rest)
2463 				goto bad_val;
2464 		} else if (!strcmp(this_char,"uid")) {
2465 			if (remount)
2466 				continue;
2467 			uid = simple_strtoul(value, &rest, 0);
2468 			if (*rest)
2469 				goto bad_val;
2470 			sbinfo->uid = make_kuid(current_user_ns(), uid);
2471 			if (!uid_valid(sbinfo->uid))
2472 				goto bad_val;
2473 		} else if (!strcmp(this_char,"gid")) {
2474 			if (remount)
2475 				continue;
2476 			gid = simple_strtoul(value, &rest, 0);
2477 			if (*rest)
2478 				goto bad_val;
2479 			sbinfo->gid = make_kgid(current_user_ns(), gid);
2480 			if (!gid_valid(sbinfo->gid))
2481 				goto bad_val;
2482 		} else if (!strcmp(this_char,"mpol")) {
2483 			mpol_put(mpol);
2484 			mpol = NULL;
2485 			if (mpol_parse_str(value, &mpol))
2486 				goto bad_val;
2487 		} else {
2488 			printk(KERN_ERR "tmpfs: Bad mount option %s\n",
2489 			       this_char);
2490 			goto error;
2491 		}
2492 	}
2493 	sbinfo->mpol = mpol;
2494 	return 0;
2495 
2496 bad_val:
2497 	printk(KERN_ERR "tmpfs: Bad value '%s' for mount option '%s'\n",
2498 	       value, this_char);
2499 error:
2500 	mpol_put(mpol);
2501 	return 1;
2502 
2503 }
2504 
2505 static int shmem_remount_fs(struct super_block *sb, int *flags, char *data)
2506 {
2507 	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2508 	struct shmem_sb_info config = *sbinfo;
2509 	unsigned long inodes;
2510 	int error = -EINVAL;
2511 
2512 	config.mpol = NULL;
2513 	if (shmem_parse_options(data, &config, true))
2514 		return error;
2515 
2516 	spin_lock(&sbinfo->stat_lock);
2517 	inodes = sbinfo->max_inodes - sbinfo->free_inodes;
2518 	if (percpu_counter_compare(&sbinfo->used_blocks, config.max_blocks) > 0)
2519 		goto out;
2520 	if (config.max_inodes < inodes)
2521 		goto out;
2522 	/*
2523 	 * Those tests disallow limited->unlimited while any are in use;
2524 	 * but we must separately disallow unlimited->limited, because
2525 	 * in that case we have no record of how much is already in use.
2526 	 */
2527 	if (config.max_blocks && !sbinfo->max_blocks)
2528 		goto out;
2529 	if (config.max_inodes && !sbinfo->max_inodes)
2530 		goto out;
2531 
2532 	error = 0;
2533 	sbinfo->max_blocks  = config.max_blocks;
2534 	sbinfo->max_inodes  = config.max_inodes;
2535 	sbinfo->free_inodes = config.max_inodes - inodes;
2536 
2537 	/*
2538 	 * Preserve previous mempolicy unless mpol remount option was specified.
2539 	 */
2540 	if (config.mpol) {
2541 		mpol_put(sbinfo->mpol);
2542 		sbinfo->mpol = config.mpol;	/* transfers initial ref */
2543 	}
2544 out:
2545 	spin_unlock(&sbinfo->stat_lock);
2546 	return error;
2547 }
2548 
2549 static int shmem_show_options(struct seq_file *seq, struct dentry *root)
2550 {
2551 	struct shmem_sb_info *sbinfo = SHMEM_SB(root->d_sb);
2552 
2553 	if (sbinfo->max_blocks != shmem_default_max_blocks())
2554 		seq_printf(seq, ",size=%luk",
2555 			sbinfo->max_blocks << (PAGE_CACHE_SHIFT - 10));
2556 	if (sbinfo->max_inodes != shmem_default_max_inodes())
2557 		seq_printf(seq, ",nr_inodes=%lu", sbinfo->max_inodes);
2558 	if (sbinfo->mode != (S_IRWXUGO | S_ISVTX))
2559 		seq_printf(seq, ",mode=%03ho", sbinfo->mode);
2560 	if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
2561 		seq_printf(seq, ",uid=%u",
2562 				from_kuid_munged(&init_user_ns, sbinfo->uid));
2563 	if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
2564 		seq_printf(seq, ",gid=%u",
2565 				from_kgid_munged(&init_user_ns, sbinfo->gid));
2566 	shmem_show_mpol(seq, sbinfo->mpol);
2567 	return 0;
2568 }
2569 #endif /* CONFIG_TMPFS */
2570 
2571 static void shmem_put_super(struct super_block *sb)
2572 {
2573 	struct shmem_sb_info *sbinfo = SHMEM_SB(sb);
2574 
2575 	percpu_counter_destroy(&sbinfo->used_blocks);
2576 	mpol_put(sbinfo->mpol);
2577 	kfree(sbinfo);
2578 	sb->s_fs_info = NULL;
2579 }
2580 
2581 int shmem_fill_super(struct super_block *sb, void *data, int silent)
2582 {
2583 	struct inode *inode;
2584 	struct shmem_sb_info *sbinfo;
2585 	int err = -ENOMEM;
2586 
2587 	/* Round up to L1_CACHE_BYTES to resist false sharing */
2588 	sbinfo = kzalloc(max((int)sizeof(struct shmem_sb_info),
2589 				L1_CACHE_BYTES), GFP_KERNEL);
2590 	if (!sbinfo)
2591 		return -ENOMEM;
2592 
2593 	sbinfo->mode = S_IRWXUGO | S_ISVTX;
2594 	sbinfo->uid = current_fsuid();
2595 	sbinfo->gid = current_fsgid();
2596 	sb->s_fs_info = sbinfo;
2597 
2598 #ifdef CONFIG_TMPFS
2599 	/*
2600 	 * Per default we only allow half of the physical ram per
2601 	 * tmpfs instance, limiting inodes to one per page of lowmem;
2602 	 * but the internal instance is left unlimited.
2603 	 */
2604 	if (!(sb->s_flags & MS_KERNMOUNT)) {
2605 		sbinfo->max_blocks = shmem_default_max_blocks();
2606 		sbinfo->max_inodes = shmem_default_max_inodes();
2607 		if (shmem_parse_options(data, sbinfo, false)) {
2608 			err = -EINVAL;
2609 			goto failed;
2610 		}
2611 	} else {
2612 		sb->s_flags |= MS_NOUSER;
2613 	}
2614 	sb->s_export_op = &shmem_export_ops;
2615 	sb->s_flags |= MS_NOSEC;
2616 #else
2617 	sb->s_flags |= MS_NOUSER;
2618 #endif
2619 
2620 	spin_lock_init(&sbinfo->stat_lock);
2621 	if (percpu_counter_init(&sbinfo->used_blocks, 0))
2622 		goto failed;
2623 	sbinfo->free_inodes = sbinfo->max_inodes;
2624 
2625 	sb->s_maxbytes = MAX_LFS_FILESIZE;
2626 	sb->s_blocksize = PAGE_CACHE_SIZE;
2627 	sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
2628 	sb->s_magic = TMPFS_MAGIC;
2629 	sb->s_op = &shmem_ops;
2630 	sb->s_time_gran = 1;
2631 #ifdef CONFIG_TMPFS_XATTR
2632 	sb->s_xattr = shmem_xattr_handlers;
2633 #endif
2634 #ifdef CONFIG_TMPFS_POSIX_ACL
2635 	sb->s_flags |= MS_POSIXACL;
2636 #endif
2637 
2638 	inode = shmem_get_inode(sb, NULL, S_IFDIR | sbinfo->mode, 0, VM_NORESERVE);
2639 	if (!inode)
2640 		goto failed;
2641 	inode->i_uid = sbinfo->uid;
2642 	inode->i_gid = sbinfo->gid;
2643 	sb->s_root = d_make_root(inode);
2644 	if (!sb->s_root)
2645 		goto failed;
2646 	return 0;
2647 
2648 failed:
2649 	shmem_put_super(sb);
2650 	return err;
2651 }
2652 
2653 static struct kmem_cache *shmem_inode_cachep;
2654 
2655 static struct inode *shmem_alloc_inode(struct super_block *sb)
2656 {
2657 	struct shmem_inode_info *info;
2658 	info = kmem_cache_alloc(shmem_inode_cachep, GFP_KERNEL);
2659 	if (!info)
2660 		return NULL;
2661 	return &info->vfs_inode;
2662 }
2663 
2664 static void shmem_destroy_callback(struct rcu_head *head)
2665 {
2666 	struct inode *inode = container_of(head, struct inode, i_rcu);
2667 	kmem_cache_free(shmem_inode_cachep, SHMEM_I(inode));
2668 }
2669 
2670 static void shmem_destroy_inode(struct inode *inode)
2671 {
2672 	if (S_ISREG(inode->i_mode))
2673 		mpol_free_shared_policy(&SHMEM_I(inode)->policy);
2674 	call_rcu(&inode->i_rcu, shmem_destroy_callback);
2675 }
2676 
2677 static void shmem_init_inode(void *foo)
2678 {
2679 	struct shmem_inode_info *info = foo;
2680 	inode_init_once(&info->vfs_inode);
2681 }
2682 
2683 static int shmem_init_inodecache(void)
2684 {
2685 	shmem_inode_cachep = kmem_cache_create("shmem_inode_cache",
2686 				sizeof(struct shmem_inode_info),
2687 				0, SLAB_PANIC, shmem_init_inode);
2688 	return 0;
2689 }
2690 
2691 static void shmem_destroy_inodecache(void)
2692 {
2693 	kmem_cache_destroy(shmem_inode_cachep);
2694 }
2695 
2696 static const struct address_space_operations shmem_aops = {
2697 	.writepage	= shmem_writepage,
2698 	.set_page_dirty	= __set_page_dirty_no_writeback,
2699 #ifdef CONFIG_TMPFS
2700 	.write_begin	= shmem_write_begin,
2701 	.write_end	= shmem_write_end,
2702 #endif
2703 	.migratepage	= migrate_page,
2704 	.error_remove_page = generic_error_remove_page,
2705 };
2706 
2707 static const struct file_operations shmem_file_operations = {
2708 	.mmap		= shmem_mmap,
2709 #ifdef CONFIG_TMPFS
2710 	.llseek		= shmem_file_llseek,
2711 	.read		= do_sync_read,
2712 	.write		= do_sync_write,
2713 	.aio_read	= shmem_file_aio_read,
2714 	.aio_write	= generic_file_aio_write,
2715 	.fsync		= noop_fsync,
2716 	.splice_read	= shmem_file_splice_read,
2717 	.splice_write	= generic_file_splice_write,
2718 	.fallocate	= shmem_fallocate,
2719 #endif
2720 };
2721 
2722 static const struct inode_operations shmem_inode_operations = {
2723 	.setattr	= shmem_setattr,
2724 #ifdef CONFIG_TMPFS_XATTR
2725 	.setxattr	= shmem_setxattr,
2726 	.getxattr	= shmem_getxattr,
2727 	.listxattr	= shmem_listxattr,
2728 	.removexattr	= shmem_removexattr,
2729 	.set_acl	= simple_set_acl,
2730 #endif
2731 };
2732 
2733 static const struct inode_operations shmem_dir_inode_operations = {
2734 #ifdef CONFIG_TMPFS
2735 	.create		= shmem_create,
2736 	.lookup		= simple_lookup,
2737 	.link		= shmem_link,
2738 	.unlink		= shmem_unlink,
2739 	.symlink	= shmem_symlink,
2740 	.mkdir		= shmem_mkdir,
2741 	.rmdir		= shmem_rmdir,
2742 	.mknod		= shmem_mknod,
2743 	.rename		= shmem_rename,
2744 	.tmpfile	= shmem_tmpfile,
2745 #endif
2746 #ifdef CONFIG_TMPFS_XATTR
2747 	.setxattr	= shmem_setxattr,
2748 	.getxattr	= shmem_getxattr,
2749 	.listxattr	= shmem_listxattr,
2750 	.removexattr	= shmem_removexattr,
2751 #endif
2752 #ifdef CONFIG_TMPFS_POSIX_ACL
2753 	.setattr	= shmem_setattr,
2754 	.set_acl	= simple_set_acl,
2755 #endif
2756 };
2757 
2758 static const struct inode_operations shmem_special_inode_operations = {
2759 #ifdef CONFIG_TMPFS_XATTR
2760 	.setxattr	= shmem_setxattr,
2761 	.getxattr	= shmem_getxattr,
2762 	.listxattr	= shmem_listxattr,
2763 	.removexattr	= shmem_removexattr,
2764 #endif
2765 #ifdef CONFIG_TMPFS_POSIX_ACL
2766 	.setattr	= shmem_setattr,
2767 	.set_acl	= simple_set_acl,
2768 #endif
2769 };
2770 
2771 static const struct super_operations shmem_ops = {
2772 	.alloc_inode	= shmem_alloc_inode,
2773 	.destroy_inode	= shmem_destroy_inode,
2774 #ifdef CONFIG_TMPFS
2775 	.statfs		= shmem_statfs,
2776 	.remount_fs	= shmem_remount_fs,
2777 	.show_options	= shmem_show_options,
2778 #endif
2779 	.evict_inode	= shmem_evict_inode,
2780 	.drop_inode	= generic_delete_inode,
2781 	.put_super	= shmem_put_super,
2782 };
2783 
2784 static const struct vm_operations_struct shmem_vm_ops = {
2785 	.fault		= shmem_fault,
2786 #ifdef CONFIG_NUMA
2787 	.set_policy     = shmem_set_policy,
2788 	.get_policy     = shmem_get_policy,
2789 #endif
2790 	.remap_pages	= generic_file_remap_pages,
2791 };
2792 
2793 static struct dentry *shmem_mount(struct file_system_type *fs_type,
2794 	int flags, const char *dev_name, void *data)
2795 {
2796 	return mount_nodev(fs_type, flags, data, shmem_fill_super);
2797 }
2798 
2799 static struct file_system_type shmem_fs_type = {
2800 	.owner		= THIS_MODULE,
2801 	.name		= "tmpfs",
2802 	.mount		= shmem_mount,
2803 	.kill_sb	= kill_litter_super,
2804 	.fs_flags	= FS_USERNS_MOUNT,
2805 };
2806 
2807 int __init shmem_init(void)
2808 {
2809 	int error;
2810 
2811 	/* If rootfs called this, don't re-init */
2812 	if (shmem_inode_cachep)
2813 		return 0;
2814 
2815 	error = bdi_init(&shmem_backing_dev_info);
2816 	if (error)
2817 		goto out4;
2818 
2819 	error = shmem_init_inodecache();
2820 	if (error)
2821 		goto out3;
2822 
2823 	error = register_filesystem(&shmem_fs_type);
2824 	if (error) {
2825 		printk(KERN_ERR "Could not register tmpfs\n");
2826 		goto out2;
2827 	}
2828 
2829 	shm_mnt = kern_mount(&shmem_fs_type);
2830 	if (IS_ERR(shm_mnt)) {
2831 		error = PTR_ERR(shm_mnt);
2832 		printk(KERN_ERR "Could not kern_mount tmpfs\n");
2833 		goto out1;
2834 	}
2835 	return 0;
2836 
2837 out1:
2838 	unregister_filesystem(&shmem_fs_type);
2839 out2:
2840 	shmem_destroy_inodecache();
2841 out3:
2842 	bdi_destroy(&shmem_backing_dev_info);
2843 out4:
2844 	shm_mnt = ERR_PTR(error);
2845 	return error;
2846 }
2847 
2848 #else /* !CONFIG_SHMEM */
2849 
2850 /*
2851  * tiny-shmem: simple shmemfs and tmpfs using ramfs code
2852  *
2853  * This is intended for small system where the benefits of the full
2854  * shmem code (swap-backed and resource-limited) are outweighed by
2855  * their complexity. On systems without swap this code should be
2856  * effectively equivalent, but much lighter weight.
2857  */
2858 
2859 static struct file_system_type shmem_fs_type = {
2860 	.name		= "tmpfs",
2861 	.mount		= ramfs_mount,
2862 	.kill_sb	= kill_litter_super,
2863 	.fs_flags	= FS_USERNS_MOUNT,
2864 };
2865 
2866 int __init shmem_init(void)
2867 {
2868 	BUG_ON(register_filesystem(&shmem_fs_type) != 0);
2869 
2870 	shm_mnt = kern_mount(&shmem_fs_type);
2871 	BUG_ON(IS_ERR(shm_mnt));
2872 
2873 	return 0;
2874 }
2875 
2876 int shmem_unuse(swp_entry_t swap, struct page *page)
2877 {
2878 	return 0;
2879 }
2880 
2881 int shmem_lock(struct file *file, int lock, struct user_struct *user)
2882 {
2883 	return 0;
2884 }
2885 
2886 void shmem_unlock_mapping(struct address_space *mapping)
2887 {
2888 }
2889 
2890 void shmem_truncate_range(struct inode *inode, loff_t lstart, loff_t lend)
2891 {
2892 	truncate_inode_pages_range(inode->i_mapping, lstart, lend);
2893 }
2894 EXPORT_SYMBOL_GPL(shmem_truncate_range);
2895 
2896 #define shmem_vm_ops				generic_file_vm_ops
2897 #define shmem_file_operations			ramfs_file_operations
2898 #define shmem_get_inode(sb, dir, mode, dev, flags)	ramfs_get_inode(sb, dir, mode, dev)
2899 #define shmem_acct_size(flags, size)		0
2900 #define shmem_unacct_size(flags, size)		do {} while (0)
2901 
2902 #endif /* CONFIG_SHMEM */
2903 
2904 /* common code */
2905 
2906 static struct dentry_operations anon_ops = {
2907 	.d_dname = simple_dname
2908 };
2909 
2910 static struct file *__shmem_file_setup(const char *name, loff_t size,
2911 				       unsigned long flags, unsigned int i_flags)
2912 {
2913 	struct file *res;
2914 	struct inode *inode;
2915 	struct path path;
2916 	struct super_block *sb;
2917 	struct qstr this;
2918 
2919 	if (IS_ERR(shm_mnt))
2920 		return ERR_CAST(shm_mnt);
2921 
2922 	if (size < 0 || size > MAX_LFS_FILESIZE)
2923 		return ERR_PTR(-EINVAL);
2924 
2925 	if (shmem_acct_size(flags, size))
2926 		return ERR_PTR(-ENOMEM);
2927 
2928 	res = ERR_PTR(-ENOMEM);
2929 	this.name = name;
2930 	this.len = strlen(name);
2931 	this.hash = 0; /* will go */
2932 	sb = shm_mnt->mnt_sb;
2933 	path.dentry = d_alloc_pseudo(sb, &this);
2934 	if (!path.dentry)
2935 		goto put_memory;
2936 	d_set_d_op(path.dentry, &anon_ops);
2937 	path.mnt = mntget(shm_mnt);
2938 
2939 	res = ERR_PTR(-ENOSPC);
2940 	inode = shmem_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0, flags);
2941 	if (!inode)
2942 		goto put_dentry;
2943 
2944 	inode->i_flags |= i_flags;
2945 	d_instantiate(path.dentry, inode);
2946 	inode->i_size = size;
2947 	clear_nlink(inode);	/* It is unlinked */
2948 	res = ERR_PTR(ramfs_nommu_expand_for_mapping(inode, size));
2949 	if (IS_ERR(res))
2950 		goto put_dentry;
2951 
2952 	res = alloc_file(&path, FMODE_WRITE | FMODE_READ,
2953 		  &shmem_file_operations);
2954 	if (IS_ERR(res))
2955 		goto put_dentry;
2956 
2957 	return res;
2958 
2959 put_dentry:
2960 	path_put(&path);
2961 put_memory:
2962 	shmem_unacct_size(flags, size);
2963 	return res;
2964 }
2965 
2966 /**
2967  * shmem_kernel_file_setup - get an unlinked file living in tmpfs which must be
2968  * 	kernel internal.  There will be NO LSM permission checks against the
2969  * 	underlying inode.  So users of this interface must do LSM checks at a
2970  * 	higher layer.  The one user is the big_key implementation.  LSM checks
2971  * 	are provided at the key level rather than the inode level.
2972  * @name: name for dentry (to be seen in /proc/<pid>/maps
2973  * @size: size to be set for the file
2974  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2975  */
2976 struct file *shmem_kernel_file_setup(const char *name, loff_t size, unsigned long flags)
2977 {
2978 	return __shmem_file_setup(name, size, flags, S_PRIVATE);
2979 }
2980 
2981 /**
2982  * shmem_file_setup - get an unlinked file living in tmpfs
2983  * @name: name for dentry (to be seen in /proc/<pid>/maps
2984  * @size: size to be set for the file
2985  * @flags: VM_NORESERVE suppresses pre-accounting of the entire object size
2986  */
2987 struct file *shmem_file_setup(const char *name, loff_t size, unsigned long flags)
2988 {
2989 	return __shmem_file_setup(name, size, flags, 0);
2990 }
2991 EXPORT_SYMBOL_GPL(shmem_file_setup);
2992 
2993 /**
2994  * shmem_zero_setup - setup a shared anonymous mapping
2995  * @vma: the vma to be mmapped is prepared by do_mmap_pgoff
2996  */
2997 int shmem_zero_setup(struct vm_area_struct *vma)
2998 {
2999 	struct file *file;
3000 	loff_t size = vma->vm_end - vma->vm_start;
3001 
3002 	file = shmem_file_setup("dev/zero", size, vma->vm_flags);
3003 	if (IS_ERR(file))
3004 		return PTR_ERR(file);
3005 
3006 	if (vma->vm_file)
3007 		fput(vma->vm_file);
3008 	vma->vm_file = file;
3009 	vma->vm_ops = &shmem_vm_ops;
3010 	return 0;
3011 }
3012 
3013 /**
3014  * shmem_read_mapping_page_gfp - read into page cache, using specified page allocation flags.
3015  * @mapping:	the page's address_space
3016  * @index:	the page index
3017  * @gfp:	the page allocator flags to use if allocating
3018  *
3019  * This behaves as a tmpfs "read_cache_page_gfp(mapping, index, gfp)",
3020  * with any new page allocations done using the specified allocation flags.
3021  * But read_cache_page_gfp() uses the ->readpage() method: which does not
3022  * suit tmpfs, since it may have pages in swapcache, and needs to find those
3023  * for itself; although drivers/gpu/drm i915 and ttm rely upon this support.
3024  *
3025  * i915_gem_object_get_pages_gtt() mixes __GFP_NORETRY | __GFP_NOWARN in
3026  * with the mapping_gfp_mask(), to avoid OOMing the machine unnecessarily.
3027  */
3028 struct page *shmem_read_mapping_page_gfp(struct address_space *mapping,
3029 					 pgoff_t index, gfp_t gfp)
3030 {
3031 #ifdef CONFIG_SHMEM
3032 	struct inode *inode = mapping->host;
3033 	struct page *page;
3034 	int error;
3035 
3036 	BUG_ON(mapping->a_ops != &shmem_aops);
3037 	error = shmem_getpage_gfp(inode, index, &page, SGP_CACHE, gfp, NULL);
3038 	if (error)
3039 		page = ERR_PTR(error);
3040 	else
3041 		unlock_page(page);
3042 	return page;
3043 #else
3044 	/*
3045 	 * The tiny !SHMEM case uses ramfs without swap
3046 	 */
3047 	return read_cache_page_gfp(mapping, index, gfp);
3048 #endif
3049 }
3050 EXPORT_SYMBOL_GPL(shmem_read_mapping_page_gfp);
3051