mm: make pagevec_lookup() update index
[sfrench/cifs-2.6.git] / fs / hugetlbfs / inode.c
1 /*
2  * hugetlbpage-backed filesystem.  Based on ramfs.
3  *
4  * Nadia Yvette Chambers, 2002
5  *
6  * Copyright (C) 2002 Linus Torvalds.
7  * License: GPL
8  */
9
10 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
11
12 #include <linux/thread_info.h>
13 #include <asm/current.h>
14 #include <linux/sched/signal.h>         /* remove ASAP */
15 #include <linux/falloc.h>
16 #include <linux/fs.h>
17 #include <linux/mount.h>
18 #include <linux/file.h>
19 #include <linux/kernel.h>
20 #include <linux/writeback.h>
21 #include <linux/pagemap.h>
22 #include <linux/highmem.h>
23 #include <linux/init.h>
24 #include <linux/string.h>
25 #include <linux/capability.h>
26 #include <linux/ctype.h>
27 #include <linux/backing-dev.h>
28 #include <linux/hugetlb.h>
29 #include <linux/pagevec.h>
30 #include <linux/parser.h>
31 #include <linux/mman.h>
32 #include <linux/slab.h>
33 #include <linux/dnotify.h>
34 #include <linux/statfs.h>
35 #include <linux/security.h>
36 #include <linux/magic.h>
37 #include <linux/migrate.h>
38 #include <linux/uio.h>
39
40 #include <linux/uaccess.h>
41
42 static const struct super_operations hugetlbfs_ops;
43 static const struct address_space_operations hugetlbfs_aops;
44 const struct file_operations hugetlbfs_file_operations;
45 static const struct inode_operations hugetlbfs_dir_inode_operations;
46 static const struct inode_operations hugetlbfs_inode_operations;
47
48 struct hugetlbfs_config {
49         struct hstate           *hstate;
50         long                    max_hpages;
51         long                    nr_inodes;
52         long                    min_hpages;
53         kuid_t                  uid;
54         kgid_t                  gid;
55         umode_t                 mode;
56 };
57
58 struct hugetlbfs_inode_info {
59         struct shared_policy policy;
60         struct inode vfs_inode;
61 };
62
63 static inline struct hugetlbfs_inode_info *HUGETLBFS_I(struct inode *inode)
64 {
65         return container_of(inode, struct hugetlbfs_inode_info, vfs_inode);
66 }
67
68 int sysctl_hugetlb_shm_group;
69
70 enum {
71         Opt_size, Opt_nr_inodes,
72         Opt_mode, Opt_uid, Opt_gid,
73         Opt_pagesize, Opt_min_size,
74         Opt_err,
75 };
76
77 static const match_table_t tokens = {
78         {Opt_size,      "size=%s"},
79         {Opt_nr_inodes, "nr_inodes=%s"},
80         {Opt_mode,      "mode=%o"},
81         {Opt_uid,       "uid=%u"},
82         {Opt_gid,       "gid=%u"},
83         {Opt_pagesize,  "pagesize=%s"},
84         {Opt_min_size,  "min_size=%s"},
85         {Opt_err,       NULL},
86 };
87
88 #ifdef CONFIG_NUMA
89 static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma,
90                                         struct inode *inode, pgoff_t index)
91 {
92         vma->vm_policy = mpol_shared_policy_lookup(&HUGETLBFS_I(inode)->policy,
93                                                         index);
94 }
95
96 static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma)
97 {
98         mpol_cond_put(vma->vm_policy);
99 }
100 #else
101 static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma,
102                                         struct inode *inode, pgoff_t index)
103 {
104 }
105
106 static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma)
107 {
108 }
109 #endif
110
111 static void huge_pagevec_release(struct pagevec *pvec)
112 {
113         int i;
114
115         for (i = 0; i < pagevec_count(pvec); ++i)
116                 put_page(pvec->pages[i]);
117
118         pagevec_reinit(pvec);
119 }
120
121 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
122 {
123         struct inode *inode = file_inode(file);
124         loff_t len, vma_len;
125         int ret;
126         struct hstate *h = hstate_file(file);
127
128         /*
129          * vma address alignment (but not the pgoff alignment) has
130          * already been checked by prepare_hugepage_range.  If you add
131          * any error returns here, do so after setting VM_HUGETLB, so
132          * is_vm_hugetlb_page tests below unmap_region go the right
133          * way when do_mmap_pgoff unwinds (may be important on powerpc
134          * and ia64).
135          */
136         vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND;
137         vma->vm_ops = &hugetlb_vm_ops;
138
139         /*
140          * Offset passed to mmap (before page shift) could have been
141          * negative when represented as a (l)off_t.
142          */
143         if (((loff_t)vma->vm_pgoff << PAGE_SHIFT) < 0)
144                 return -EINVAL;
145
146         if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT))
147                 return -EINVAL;
148
149         vma_len = (loff_t)(vma->vm_end - vma->vm_start);
150         len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
151         /* check for overflow */
152         if (len < vma_len)
153                 return -EINVAL;
154
155         inode_lock(inode);
156         file_accessed(file);
157
158         ret = -ENOMEM;
159         if (hugetlb_reserve_pages(inode,
160                                 vma->vm_pgoff >> huge_page_order(h),
161                                 len >> huge_page_shift(h), vma,
162                                 vma->vm_flags))
163                 goto out;
164
165         ret = 0;
166         if (vma->vm_flags & VM_WRITE && inode->i_size < len)
167                 i_size_write(inode, len);
168 out:
169         inode_unlock(inode);
170
171         return ret;
172 }
173
174 /*
175  * Called under down_write(mmap_sem).
176  */
177
178 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
179 static unsigned long
180 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
181                 unsigned long len, unsigned long pgoff, unsigned long flags)
182 {
183         struct mm_struct *mm = current->mm;
184         struct vm_area_struct *vma;
185         struct hstate *h = hstate_file(file);
186         struct vm_unmapped_area_info info;
187
188         if (len & ~huge_page_mask(h))
189                 return -EINVAL;
190         if (len > TASK_SIZE)
191                 return -ENOMEM;
192
193         if (flags & MAP_FIXED) {
194                 if (prepare_hugepage_range(file, addr, len))
195                         return -EINVAL;
196                 return addr;
197         }
198
199         if (addr) {
200                 addr = ALIGN(addr, huge_page_size(h));
201                 vma = find_vma(mm, addr);
202                 if (TASK_SIZE - len >= addr &&
203                     (!vma || addr + len <= vm_start_gap(vma)))
204                         return addr;
205         }
206
207         info.flags = 0;
208         info.length = len;
209         info.low_limit = TASK_UNMAPPED_BASE;
210         info.high_limit = TASK_SIZE;
211         info.align_mask = PAGE_MASK & ~huge_page_mask(h);
212         info.align_offset = 0;
213         return vm_unmapped_area(&info);
214 }
215 #endif
216
217 static size_t
218 hugetlbfs_read_actor(struct page *page, unsigned long offset,
219                         struct iov_iter *to, unsigned long size)
220 {
221         size_t copied = 0;
222         int i, chunksize;
223
224         /* Find which 4k chunk and offset with in that chunk */
225         i = offset >> PAGE_SHIFT;
226         offset = offset & ~PAGE_MASK;
227
228         while (size) {
229                 size_t n;
230                 chunksize = PAGE_SIZE;
231                 if (offset)
232                         chunksize -= offset;
233                 if (chunksize > size)
234                         chunksize = size;
235                 n = copy_page_to_iter(&page[i], offset, chunksize, to);
236                 copied += n;
237                 if (n != chunksize)
238                         return copied;
239                 offset = 0;
240                 size -= chunksize;
241                 i++;
242         }
243         return copied;
244 }
245
246 /*
247  * Support for read() - Find the page attached to f_mapping and copy out the
248  * data. Its *very* similar to do_generic_mapping_read(), we can't use that
249  * since it has PAGE_SIZE assumptions.
250  */
251 static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to)
252 {
253         struct file *file = iocb->ki_filp;
254         struct hstate *h = hstate_file(file);
255         struct address_space *mapping = file->f_mapping;
256         struct inode *inode = mapping->host;
257         unsigned long index = iocb->ki_pos >> huge_page_shift(h);
258         unsigned long offset = iocb->ki_pos & ~huge_page_mask(h);
259         unsigned long end_index;
260         loff_t isize;
261         ssize_t retval = 0;
262
263         while (iov_iter_count(to)) {
264                 struct page *page;
265                 size_t nr, copied;
266
267                 /* nr is the maximum number of bytes to copy from this page */
268                 nr = huge_page_size(h);
269                 isize = i_size_read(inode);
270                 if (!isize)
271                         break;
272                 end_index = (isize - 1) >> huge_page_shift(h);
273                 if (index > end_index)
274                         break;
275                 if (index == end_index) {
276                         nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
277                         if (nr <= offset)
278                                 break;
279                 }
280                 nr = nr - offset;
281
282                 /* Find the page */
283                 page = find_lock_page(mapping, index);
284                 if (unlikely(page == NULL)) {
285                         /*
286                          * We have a HOLE, zero out the user-buffer for the
287                          * length of the hole or request.
288                          */
289                         copied = iov_iter_zero(nr, to);
290                 } else {
291                         unlock_page(page);
292
293                         /*
294                          * We have the page, copy it to user space buffer.
295                          */
296                         copied = hugetlbfs_read_actor(page, offset, to, nr);
297                         put_page(page);
298                 }
299                 offset += copied;
300                 retval += copied;
301                 if (copied != nr && iov_iter_count(to)) {
302                         if (!retval)
303                                 retval = -EFAULT;
304                         break;
305                 }
306                 index += offset >> huge_page_shift(h);
307                 offset &= ~huge_page_mask(h);
308         }
309         iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset;
310         return retval;
311 }
312
313 static int hugetlbfs_write_begin(struct file *file,
314                         struct address_space *mapping,
315                         loff_t pos, unsigned len, unsigned flags,
316                         struct page **pagep, void **fsdata)
317 {
318         return -EINVAL;
319 }
320
321 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
322                         loff_t pos, unsigned len, unsigned copied,
323                         struct page *page, void *fsdata)
324 {
325         BUG();
326         return -EINVAL;
327 }
328
329 static void remove_huge_page(struct page *page)
330 {
331         ClearPageDirty(page);
332         ClearPageUptodate(page);
333         delete_from_page_cache(page);
334 }
335
336 static void
337 hugetlb_vmdelete_list(struct rb_root *root, pgoff_t start, pgoff_t end)
338 {
339         struct vm_area_struct *vma;
340
341         /*
342          * end == 0 indicates that the entire range after
343          * start should be unmapped.
344          */
345         vma_interval_tree_foreach(vma, root, start, end ? end : ULONG_MAX) {
346                 unsigned long v_offset;
347                 unsigned long v_end;
348
349                 /*
350                  * Can the expression below overflow on 32-bit arches?
351                  * No, because the interval tree returns us only those vmas
352                  * which overlap the truncated area starting at pgoff,
353                  * and no vma on a 32-bit arch can span beyond the 4GB.
354                  */
355                 if (vma->vm_pgoff < start)
356                         v_offset = (start - vma->vm_pgoff) << PAGE_SHIFT;
357                 else
358                         v_offset = 0;
359
360                 if (!end)
361                         v_end = vma->vm_end;
362                 else {
363                         v_end = ((end - vma->vm_pgoff) << PAGE_SHIFT)
364                                                         + vma->vm_start;
365                         if (v_end > vma->vm_end)
366                                 v_end = vma->vm_end;
367                 }
368
369                 unmap_hugepage_range(vma, vma->vm_start + v_offset, v_end,
370                                                                         NULL);
371         }
372 }
373
374 /*
375  * remove_inode_hugepages handles two distinct cases: truncation and hole
376  * punch.  There are subtle differences in operation for each case.
377  *
378  * truncation is indicated by end of range being LLONG_MAX
379  *      In this case, we first scan the range and release found pages.
380  *      After releasing pages, hugetlb_unreserve_pages cleans up region/reserv
381  *      maps and global counts.  Page faults can not race with truncation
382  *      in this routine.  hugetlb_no_page() prevents page faults in the
383  *      truncated range.  It checks i_size before allocation, and again after
384  *      with the page table lock for the page held.  The same lock must be
385  *      acquired to unmap a page.
386  * hole punch is indicated if end is not LLONG_MAX
387  *      In the hole punch case we scan the range and release found pages.
388  *      Only when releasing a page is the associated region/reserv map
389  *      deleted.  The region/reserv map for ranges without associated
390  *      pages are not modified.  Page faults can race with hole punch.
391  *      This is indicated if we find a mapped page.
392  * Note: If the passed end of range value is beyond the end of file, but
393  * not LLONG_MAX this routine still performs a hole punch operation.
394  */
395 static void remove_inode_hugepages(struct inode *inode, loff_t lstart,
396                                    loff_t lend)
397 {
398         struct hstate *h = hstate_inode(inode);
399         struct address_space *mapping = &inode->i_data;
400         const pgoff_t start = lstart >> huge_page_shift(h);
401         const pgoff_t end = lend >> huge_page_shift(h);
402         struct vm_area_struct pseudo_vma;
403         struct pagevec pvec;
404         pgoff_t next, index;
405         int i, freed = 0;
406         long lookup_nr = PAGEVEC_SIZE;
407         bool truncate_op = (lend == LLONG_MAX);
408
409         memset(&pseudo_vma, 0, sizeof(struct vm_area_struct));
410         pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED);
411         pagevec_init(&pvec, 0);
412         next = start;
413         while (next < end) {
414                 /*
415                  * Don't grab more pages than the number left in the range.
416                  */
417                 if (end - next < lookup_nr)
418                         lookup_nr = end - next;
419
420                 /*
421                  * When no more pages are found, we are done.
422                  */
423                 if (!pagevec_lookup(&pvec, mapping, &next, lookup_nr))
424                         break;
425
426                 for (i = 0; i < pagevec_count(&pvec); ++i) {
427                         struct page *page = pvec.pages[i];
428                         u32 hash;
429
430                         /*
431                          * The page (index) could be beyond end.  This is
432                          * only possible in the punch hole case as end is
433                          * max page offset in the truncate case.
434                          */
435                         index = page->index;
436                         if (index >= end)
437                                 break;
438
439                         hash = hugetlb_fault_mutex_hash(h, current->mm,
440                                                         &pseudo_vma,
441                                                         mapping, index, 0);
442                         mutex_lock(&hugetlb_fault_mutex_table[hash]);
443
444                         /*
445                          * If page is mapped, it was faulted in after being
446                          * unmapped in caller.  Unmap (again) now after taking
447                          * the fault mutex.  The mutex will prevent faults
448                          * until we finish removing the page.
449                          *
450                          * This race can only happen in the hole punch case.
451                          * Getting here in a truncate operation is a bug.
452                          */
453                         if (unlikely(page_mapped(page))) {
454                                 BUG_ON(truncate_op);
455
456                                 i_mmap_lock_write(mapping);
457                                 hugetlb_vmdelete_list(&mapping->i_mmap,
458                                         index * pages_per_huge_page(h),
459                                         (index + 1) * pages_per_huge_page(h));
460                                 i_mmap_unlock_write(mapping);
461                         }
462
463                         lock_page(page);
464                         /*
465                          * We must free the huge page and remove from page
466                          * cache (remove_huge_page) BEFORE removing the
467                          * region/reserve map (hugetlb_unreserve_pages).  In
468                          * rare out of memory conditions, removal of the
469                          * region/reserve map could fail. Correspondingly,
470                          * the subpool and global reserve usage count can need
471                          * to be adjusted.
472                          */
473                         VM_BUG_ON(PagePrivate(page));
474                         remove_huge_page(page);
475                         freed++;
476                         if (!truncate_op) {
477                                 if (unlikely(hugetlb_unreserve_pages(inode,
478                                                         index, index + 1, 1)))
479                                         hugetlb_fix_reserve_counts(inode);
480                         }
481
482                         unlock_page(page);
483                         mutex_unlock(&hugetlb_fault_mutex_table[hash]);
484                 }
485                 huge_pagevec_release(&pvec);
486                 cond_resched();
487         }
488
489         if (truncate_op)
490                 (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed);
491 }
492
493 static void hugetlbfs_evict_inode(struct inode *inode)
494 {
495         struct resv_map *resv_map;
496
497         remove_inode_hugepages(inode, 0, LLONG_MAX);
498         resv_map = (struct resv_map *)inode->i_mapping->private_data;
499         /* root inode doesn't have the resv_map, so we should check it */
500         if (resv_map)
501                 resv_map_release(&resv_map->refs);
502         clear_inode(inode);
503 }
504
505 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
506 {
507         pgoff_t pgoff;
508         struct address_space *mapping = inode->i_mapping;
509         struct hstate *h = hstate_inode(inode);
510
511         BUG_ON(offset & ~huge_page_mask(h));
512         pgoff = offset >> PAGE_SHIFT;
513
514         i_size_write(inode, offset);
515         i_mmap_lock_write(mapping);
516         if (!RB_EMPTY_ROOT(&mapping->i_mmap))
517                 hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0);
518         i_mmap_unlock_write(mapping);
519         remove_inode_hugepages(inode, offset, LLONG_MAX);
520         return 0;
521 }
522
523 static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
524 {
525         struct hstate *h = hstate_inode(inode);
526         loff_t hpage_size = huge_page_size(h);
527         loff_t hole_start, hole_end;
528
529         /*
530          * For hole punch round up the beginning offset of the hole and
531          * round down the end.
532          */
533         hole_start = round_up(offset, hpage_size);
534         hole_end = round_down(offset + len, hpage_size);
535
536         if (hole_end > hole_start) {
537                 struct address_space *mapping = inode->i_mapping;
538
539                 inode_lock(inode);
540                 i_mmap_lock_write(mapping);
541                 if (!RB_EMPTY_ROOT(&mapping->i_mmap))
542                         hugetlb_vmdelete_list(&mapping->i_mmap,
543                                                 hole_start >> PAGE_SHIFT,
544                                                 hole_end  >> PAGE_SHIFT);
545                 i_mmap_unlock_write(mapping);
546                 remove_inode_hugepages(inode, hole_start, hole_end);
547                 inode_unlock(inode);
548         }
549
550         return 0;
551 }
552
553 static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset,
554                                 loff_t len)
555 {
556         struct inode *inode = file_inode(file);
557         struct address_space *mapping = inode->i_mapping;
558         struct hstate *h = hstate_inode(inode);
559         struct vm_area_struct pseudo_vma;
560         struct mm_struct *mm = current->mm;
561         loff_t hpage_size = huge_page_size(h);
562         unsigned long hpage_shift = huge_page_shift(h);
563         pgoff_t start, index, end;
564         int error;
565         u32 hash;
566
567         if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
568                 return -EOPNOTSUPP;
569
570         if (mode & FALLOC_FL_PUNCH_HOLE)
571                 return hugetlbfs_punch_hole(inode, offset, len);
572
573         /*
574          * Default preallocate case.
575          * For this range, start is rounded down and end is rounded up
576          * as well as being converted to page offsets.
577          */
578         start = offset >> hpage_shift;
579         end = (offset + len + hpage_size - 1) >> hpage_shift;
580
581         inode_lock(inode);
582
583         /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
584         error = inode_newsize_ok(inode, offset + len);
585         if (error)
586                 goto out;
587
588         /*
589          * Initialize a pseudo vma as this is required by the huge page
590          * allocation routines.  If NUMA is configured, use page index
591          * as input to create an allocation policy.
592          */
593         memset(&pseudo_vma, 0, sizeof(struct vm_area_struct));
594         pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED);
595         pseudo_vma.vm_file = file;
596
597         for (index = start; index < end; index++) {
598                 /*
599                  * This is supposed to be the vaddr where the page is being
600                  * faulted in, but we have no vaddr here.
601                  */
602                 struct page *page;
603                 unsigned long addr;
604                 int avoid_reserve = 0;
605
606                 cond_resched();
607
608                 /*
609                  * fallocate(2) manpage permits EINTR; we may have been
610                  * interrupted because we are using up too much memory.
611                  */
612                 if (signal_pending(current)) {
613                         error = -EINTR;
614                         break;
615                 }
616
617                 /* Set numa allocation policy based on index */
618                 hugetlb_set_vma_policy(&pseudo_vma, inode, index);
619
620                 /* addr is the offset within the file (zero based) */
621                 addr = index * hpage_size;
622
623                 /* mutex taken here, fault path and hole punch */
624                 hash = hugetlb_fault_mutex_hash(h, mm, &pseudo_vma, mapping,
625                                                 index, addr);
626                 mutex_lock(&hugetlb_fault_mutex_table[hash]);
627
628                 /* See if already present in mapping to avoid alloc/free */
629                 page = find_get_page(mapping, index);
630                 if (page) {
631                         put_page(page);
632                         mutex_unlock(&hugetlb_fault_mutex_table[hash]);
633                         hugetlb_drop_vma_policy(&pseudo_vma);
634                         continue;
635                 }
636
637                 /* Allocate page and add to page cache */
638                 page = alloc_huge_page(&pseudo_vma, addr, avoid_reserve);
639                 hugetlb_drop_vma_policy(&pseudo_vma);
640                 if (IS_ERR(page)) {
641                         mutex_unlock(&hugetlb_fault_mutex_table[hash]);
642                         error = PTR_ERR(page);
643                         goto out;
644                 }
645                 clear_huge_page(page, addr, pages_per_huge_page(h));
646                 __SetPageUptodate(page);
647                 error = huge_add_to_page_cache(page, mapping, index);
648                 if (unlikely(error)) {
649                         put_page(page);
650                         mutex_unlock(&hugetlb_fault_mutex_table[hash]);
651                         goto out;
652                 }
653
654                 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
655
656                 /*
657                  * page_put due to reference from alloc_huge_page()
658                  * unlock_page because locked by add_to_page_cache()
659                  */
660                 put_page(page);
661                 unlock_page(page);
662         }
663
664         if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
665                 i_size_write(inode, offset + len);
666         inode->i_ctime = current_time(inode);
667 out:
668         inode_unlock(inode);
669         return error;
670 }
671
672 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
673 {
674         struct inode *inode = d_inode(dentry);
675         struct hstate *h = hstate_inode(inode);
676         int error;
677         unsigned int ia_valid = attr->ia_valid;
678
679         BUG_ON(!inode);
680
681         error = setattr_prepare(dentry, attr);
682         if (error)
683                 return error;
684
685         if (ia_valid & ATTR_SIZE) {
686                 error = -EINVAL;
687                 if (attr->ia_size & ~huge_page_mask(h))
688                         return -EINVAL;
689                 error = hugetlb_vmtruncate(inode, attr->ia_size);
690                 if (error)
691                         return error;
692         }
693
694         setattr_copy(inode, attr);
695         mark_inode_dirty(inode);
696         return 0;
697 }
698
699 static struct inode *hugetlbfs_get_root(struct super_block *sb,
700                                         struct hugetlbfs_config *config)
701 {
702         struct inode *inode;
703
704         inode = new_inode(sb);
705         if (inode) {
706                 inode->i_ino = get_next_ino();
707                 inode->i_mode = S_IFDIR | config->mode;
708                 inode->i_uid = config->uid;
709                 inode->i_gid = config->gid;
710                 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
711                 inode->i_op = &hugetlbfs_dir_inode_operations;
712                 inode->i_fop = &simple_dir_operations;
713                 /* directory inodes start off with i_nlink == 2 (for "." entry) */
714                 inc_nlink(inode);
715                 lockdep_annotate_inode_mutex_key(inode);
716         }
717         return inode;
718 }
719
720 /*
721  * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never
722  * be taken from reclaim -- unlike regular filesystems. This needs an
723  * annotation because huge_pmd_share() does an allocation under hugetlb's
724  * i_mmap_rwsem.
725  */
726 static struct lock_class_key hugetlbfs_i_mmap_rwsem_key;
727
728 static struct inode *hugetlbfs_get_inode(struct super_block *sb,
729                                         struct inode *dir,
730                                         umode_t mode, dev_t dev)
731 {
732         struct inode *inode;
733         struct resv_map *resv_map;
734
735         resv_map = resv_map_alloc();
736         if (!resv_map)
737                 return NULL;
738
739         inode = new_inode(sb);
740         if (inode) {
741                 inode->i_ino = get_next_ino();
742                 inode_init_owner(inode, dir, mode);
743                 lockdep_set_class(&inode->i_mapping->i_mmap_rwsem,
744                                 &hugetlbfs_i_mmap_rwsem_key);
745                 inode->i_mapping->a_ops = &hugetlbfs_aops;
746                 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
747                 inode->i_mapping->private_data = resv_map;
748                 switch (mode & S_IFMT) {
749                 default:
750                         init_special_inode(inode, mode, dev);
751                         break;
752                 case S_IFREG:
753                         inode->i_op = &hugetlbfs_inode_operations;
754                         inode->i_fop = &hugetlbfs_file_operations;
755                         break;
756                 case S_IFDIR:
757                         inode->i_op = &hugetlbfs_dir_inode_operations;
758                         inode->i_fop = &simple_dir_operations;
759
760                         /* directory inodes start off with i_nlink == 2 (for "." entry) */
761                         inc_nlink(inode);
762                         break;
763                 case S_IFLNK:
764                         inode->i_op = &page_symlink_inode_operations;
765                         inode_nohighmem(inode);
766                         break;
767                 }
768                 lockdep_annotate_inode_mutex_key(inode);
769         } else
770                 kref_put(&resv_map->refs, resv_map_release);
771
772         return inode;
773 }
774
775 /*
776  * File creation. Allocate an inode, and we're done..
777  */
778 static int hugetlbfs_mknod(struct inode *dir,
779                         struct dentry *dentry, umode_t mode, dev_t dev)
780 {
781         struct inode *inode;
782         int error = -ENOSPC;
783
784         inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev);
785         if (inode) {
786                 dir->i_ctime = dir->i_mtime = current_time(dir);
787                 d_instantiate(dentry, inode);
788                 dget(dentry);   /* Extra count - pin the dentry in core */
789                 error = 0;
790         }
791         return error;
792 }
793
794 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
795 {
796         int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
797         if (!retval)
798                 inc_nlink(dir);
799         return retval;
800 }
801
802 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl)
803 {
804         return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
805 }
806
807 static int hugetlbfs_symlink(struct inode *dir,
808                         struct dentry *dentry, const char *symname)
809 {
810         struct inode *inode;
811         int error = -ENOSPC;
812
813         inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0);
814         if (inode) {
815                 int l = strlen(symname)+1;
816                 error = page_symlink(inode, symname, l);
817                 if (!error) {
818                         d_instantiate(dentry, inode);
819                         dget(dentry);
820                 } else
821                         iput(inode);
822         }
823         dir->i_ctime = dir->i_mtime = current_time(dir);
824
825         return error;
826 }
827
828 /*
829  * mark the head page dirty
830  */
831 static int hugetlbfs_set_page_dirty(struct page *page)
832 {
833         struct page *head = compound_head(page);
834
835         SetPageDirty(head);
836         return 0;
837 }
838
839 static int hugetlbfs_migrate_page(struct address_space *mapping,
840                                 struct page *newpage, struct page *page,
841                                 enum migrate_mode mode)
842 {
843         int rc;
844
845         rc = migrate_huge_page_move_mapping(mapping, newpage, page);
846         if (rc != MIGRATEPAGE_SUCCESS)
847                 return rc;
848         migrate_page_copy(newpage, page);
849
850         return MIGRATEPAGE_SUCCESS;
851 }
852
853 static int hugetlbfs_error_remove_page(struct address_space *mapping,
854                                 struct page *page)
855 {
856         struct inode *inode = mapping->host;
857
858         remove_huge_page(page);
859         hugetlb_fix_reserve_counts(inode);
860         return 0;
861 }
862
863 /*
864  * Display the mount options in /proc/mounts.
865  */
866 static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root)
867 {
868         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(root->d_sb);
869         struct hugepage_subpool *spool = sbinfo->spool;
870         unsigned long hpage_size = huge_page_size(sbinfo->hstate);
871         unsigned hpage_shift = huge_page_shift(sbinfo->hstate);
872         char mod;
873
874         if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
875                 seq_printf(m, ",uid=%u",
876                            from_kuid_munged(&init_user_ns, sbinfo->uid));
877         if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
878                 seq_printf(m, ",gid=%u",
879                            from_kgid_munged(&init_user_ns, sbinfo->gid));
880         if (sbinfo->mode != 0755)
881                 seq_printf(m, ",mode=%o", sbinfo->mode);
882         if (sbinfo->max_inodes != -1)
883                 seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes);
884
885         hpage_size /= 1024;
886         mod = 'K';
887         if (hpage_size >= 1024) {
888                 hpage_size /= 1024;
889                 mod = 'M';
890         }
891         seq_printf(m, ",pagesize=%lu%c", hpage_size, mod);
892         if (spool) {
893                 if (spool->max_hpages != -1)
894                         seq_printf(m, ",size=%llu",
895                                    (unsigned long long)spool->max_hpages << hpage_shift);
896                 if (spool->min_hpages != -1)
897                         seq_printf(m, ",min_size=%llu",
898                                    (unsigned long long)spool->min_hpages << hpage_shift);
899         }
900         return 0;
901 }
902
903 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
904 {
905         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
906         struct hstate *h = hstate_inode(d_inode(dentry));
907
908         buf->f_type = HUGETLBFS_MAGIC;
909         buf->f_bsize = huge_page_size(h);
910         if (sbinfo) {
911                 spin_lock(&sbinfo->stat_lock);
912                 /* If no limits set, just report 0 for max/free/used
913                  * blocks, like simple_statfs() */
914                 if (sbinfo->spool) {
915                         long free_pages;
916
917                         spin_lock(&sbinfo->spool->lock);
918                         buf->f_blocks = sbinfo->spool->max_hpages;
919                         free_pages = sbinfo->spool->max_hpages
920                                 - sbinfo->spool->used_hpages;
921                         buf->f_bavail = buf->f_bfree = free_pages;
922                         spin_unlock(&sbinfo->spool->lock);
923                         buf->f_files = sbinfo->max_inodes;
924                         buf->f_ffree = sbinfo->free_inodes;
925                 }
926                 spin_unlock(&sbinfo->stat_lock);
927         }
928         buf->f_namelen = NAME_MAX;
929         return 0;
930 }
931
932 static void hugetlbfs_put_super(struct super_block *sb)
933 {
934         struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
935
936         if (sbi) {
937                 sb->s_fs_info = NULL;
938
939                 if (sbi->spool)
940                         hugepage_put_subpool(sbi->spool);
941
942                 kfree(sbi);
943         }
944 }
945
946 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
947 {
948         if (sbinfo->free_inodes >= 0) {
949                 spin_lock(&sbinfo->stat_lock);
950                 if (unlikely(!sbinfo->free_inodes)) {
951                         spin_unlock(&sbinfo->stat_lock);
952                         return 0;
953                 }
954                 sbinfo->free_inodes--;
955                 spin_unlock(&sbinfo->stat_lock);
956         }
957
958         return 1;
959 }
960
961 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
962 {
963         if (sbinfo->free_inodes >= 0) {
964                 spin_lock(&sbinfo->stat_lock);
965                 sbinfo->free_inodes++;
966                 spin_unlock(&sbinfo->stat_lock);
967         }
968 }
969
970
971 static struct kmem_cache *hugetlbfs_inode_cachep;
972
973 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
974 {
975         struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
976         struct hugetlbfs_inode_info *p;
977
978         if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
979                 return NULL;
980         p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
981         if (unlikely(!p)) {
982                 hugetlbfs_inc_free_inodes(sbinfo);
983                 return NULL;
984         }
985
986         /*
987          * Any time after allocation, hugetlbfs_destroy_inode can be called
988          * for the inode.  mpol_free_shared_policy is unconditionally called
989          * as part of hugetlbfs_destroy_inode.  So, initialize policy here
990          * in case of a quick call to destroy.
991          *
992          * Note that the policy is initialized even if we are creating a
993          * private inode.  This simplifies hugetlbfs_destroy_inode.
994          */
995         mpol_shared_policy_init(&p->policy, NULL);
996
997         return &p->vfs_inode;
998 }
999
1000 static void hugetlbfs_i_callback(struct rcu_head *head)
1001 {
1002         struct inode *inode = container_of(head, struct inode, i_rcu);
1003         kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
1004 }
1005
1006 static void hugetlbfs_destroy_inode(struct inode *inode)
1007 {
1008         hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
1009         mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
1010         call_rcu(&inode->i_rcu, hugetlbfs_i_callback);
1011 }
1012
1013 static const struct address_space_operations hugetlbfs_aops = {
1014         .write_begin    = hugetlbfs_write_begin,
1015         .write_end      = hugetlbfs_write_end,
1016         .set_page_dirty = hugetlbfs_set_page_dirty,
1017         .migratepage    = hugetlbfs_migrate_page,
1018         .error_remove_page      = hugetlbfs_error_remove_page,
1019 };
1020
1021
1022 static void init_once(void *foo)
1023 {
1024         struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
1025
1026         inode_init_once(&ei->vfs_inode);
1027 }
1028
1029 const struct file_operations hugetlbfs_file_operations = {
1030         .read_iter              = hugetlbfs_read_iter,
1031         .mmap                   = hugetlbfs_file_mmap,
1032         .fsync                  = noop_fsync,
1033         .get_unmapped_area      = hugetlb_get_unmapped_area,
1034         .llseek                 = default_llseek,
1035         .fallocate              = hugetlbfs_fallocate,
1036 };
1037
1038 static const struct inode_operations hugetlbfs_dir_inode_operations = {
1039         .create         = hugetlbfs_create,
1040         .lookup         = simple_lookup,
1041         .link           = simple_link,
1042         .unlink         = simple_unlink,
1043         .symlink        = hugetlbfs_symlink,
1044         .mkdir          = hugetlbfs_mkdir,
1045         .rmdir          = simple_rmdir,
1046         .mknod          = hugetlbfs_mknod,
1047         .rename         = simple_rename,
1048         .setattr        = hugetlbfs_setattr,
1049 };
1050
1051 static const struct inode_operations hugetlbfs_inode_operations = {
1052         .setattr        = hugetlbfs_setattr,
1053 };
1054
1055 static const struct super_operations hugetlbfs_ops = {
1056         .alloc_inode    = hugetlbfs_alloc_inode,
1057         .destroy_inode  = hugetlbfs_destroy_inode,
1058         .evict_inode    = hugetlbfs_evict_inode,
1059         .statfs         = hugetlbfs_statfs,
1060         .put_super      = hugetlbfs_put_super,
1061         .show_options   = hugetlbfs_show_options,
1062 };
1063
1064 enum hugetlbfs_size_type { NO_SIZE, SIZE_STD, SIZE_PERCENT };
1065
1066 /*
1067  * Convert size option passed from command line to number of huge pages
1068  * in the pool specified by hstate.  Size option could be in bytes
1069  * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT).
1070  */
1071 static long
1072 hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt,
1073                          enum hugetlbfs_size_type val_type)
1074 {
1075         if (val_type == NO_SIZE)
1076                 return -1;
1077
1078         if (val_type == SIZE_PERCENT) {
1079                 size_opt <<= huge_page_shift(h);
1080                 size_opt *= h->max_huge_pages;
1081                 do_div(size_opt, 100);
1082         }
1083
1084         size_opt >>= huge_page_shift(h);
1085         return size_opt;
1086 }
1087
1088 static int
1089 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
1090 {
1091         char *p, *rest;
1092         substring_t args[MAX_OPT_ARGS];
1093         int option;
1094         unsigned long long max_size_opt = 0, min_size_opt = 0;
1095         enum hugetlbfs_size_type max_val_type = NO_SIZE, min_val_type = NO_SIZE;
1096
1097         if (!options)
1098                 return 0;
1099
1100         while ((p = strsep(&options, ",")) != NULL) {
1101                 int token;
1102                 if (!*p)
1103                         continue;
1104
1105                 token = match_token(p, tokens, args);
1106                 switch (token) {
1107                 case Opt_uid:
1108                         if (match_int(&args[0], &option))
1109                                 goto bad_val;
1110                         pconfig->uid = make_kuid(current_user_ns(), option);
1111                         if (!uid_valid(pconfig->uid))
1112                                 goto bad_val;
1113                         break;
1114
1115                 case Opt_gid:
1116                         if (match_int(&args[0], &option))
1117                                 goto bad_val;
1118                         pconfig->gid = make_kgid(current_user_ns(), option);
1119                         if (!gid_valid(pconfig->gid))
1120                                 goto bad_val;
1121                         break;
1122
1123                 case Opt_mode:
1124                         if (match_octal(&args[0], &option))
1125                                 goto bad_val;
1126                         pconfig->mode = option & 01777U;
1127                         break;
1128
1129                 case Opt_size: {
1130                         /* memparse() will accept a K/M/G without a digit */
1131                         if (!isdigit(*args[0].from))
1132                                 goto bad_val;
1133                         max_size_opt = memparse(args[0].from, &rest);
1134                         max_val_type = SIZE_STD;
1135                         if (*rest == '%')
1136                                 max_val_type = SIZE_PERCENT;
1137                         break;
1138                 }
1139
1140                 case Opt_nr_inodes:
1141                         /* memparse() will accept a K/M/G without a digit */
1142                         if (!isdigit(*args[0].from))
1143                                 goto bad_val;
1144                         pconfig->nr_inodes = memparse(args[0].from, &rest);
1145                         break;
1146
1147                 case Opt_pagesize: {
1148                         unsigned long ps;
1149                         ps = memparse(args[0].from, &rest);
1150                         pconfig->hstate = size_to_hstate(ps);
1151                         if (!pconfig->hstate) {
1152                                 pr_err("Unsupported page size %lu MB\n",
1153                                         ps >> 20);
1154                                 return -EINVAL;
1155                         }
1156                         break;
1157                 }
1158
1159                 case Opt_min_size: {
1160                         /* memparse() will accept a K/M/G without a digit */
1161                         if (!isdigit(*args[0].from))
1162                                 goto bad_val;
1163                         min_size_opt = memparse(args[0].from, &rest);
1164                         min_val_type = SIZE_STD;
1165                         if (*rest == '%')
1166                                 min_val_type = SIZE_PERCENT;
1167                         break;
1168                 }
1169
1170                 default:
1171                         pr_err("Bad mount option: \"%s\"\n", p);
1172                         return -EINVAL;
1173                         break;
1174                 }
1175         }
1176
1177         /*
1178          * Use huge page pool size (in hstate) to convert the size
1179          * options to number of huge pages.  If NO_SIZE, -1 is returned.
1180          */
1181         pconfig->max_hpages = hugetlbfs_size_to_hpages(pconfig->hstate,
1182                                                 max_size_opt, max_val_type);
1183         pconfig->min_hpages = hugetlbfs_size_to_hpages(pconfig->hstate,
1184                                                 min_size_opt, min_val_type);
1185
1186         /*
1187          * If max_size was specified, then min_size must be smaller
1188          */
1189         if (max_val_type > NO_SIZE &&
1190             pconfig->min_hpages > pconfig->max_hpages) {
1191                 pr_err("minimum size can not be greater than maximum size\n");
1192                 return -EINVAL;
1193         }
1194
1195         return 0;
1196
1197 bad_val:
1198         pr_err("Bad value '%s' for mount option '%s'\n", args[0].from, p);
1199         return -EINVAL;
1200 }
1201
1202 static int
1203 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
1204 {
1205         int ret;
1206         struct hugetlbfs_config config;
1207         struct hugetlbfs_sb_info *sbinfo;
1208
1209         config.max_hpages = -1; /* No limit on size by default */
1210         config.nr_inodes = -1; /* No limit on number of inodes by default */
1211         config.uid = current_fsuid();
1212         config.gid = current_fsgid();
1213         config.mode = 0755;
1214         config.hstate = &default_hstate;
1215         config.min_hpages = -1; /* No default minimum size */
1216         ret = hugetlbfs_parse_options(data, &config);
1217         if (ret)
1218                 return ret;
1219
1220         sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
1221         if (!sbinfo)
1222                 return -ENOMEM;
1223         sb->s_fs_info = sbinfo;
1224         sbinfo->hstate = config.hstate;
1225         spin_lock_init(&sbinfo->stat_lock);
1226         sbinfo->max_inodes = config.nr_inodes;
1227         sbinfo->free_inodes = config.nr_inodes;
1228         sbinfo->spool = NULL;
1229         sbinfo->uid = config.uid;
1230         sbinfo->gid = config.gid;
1231         sbinfo->mode = config.mode;
1232
1233         /*
1234          * Allocate and initialize subpool if maximum or minimum size is
1235          * specified.  Any needed reservations (for minimim size) are taken
1236          * taken when the subpool is created.
1237          */
1238         if (config.max_hpages != -1 || config.min_hpages != -1) {
1239                 sbinfo->spool = hugepage_new_subpool(config.hstate,
1240                                                         config.max_hpages,
1241                                                         config.min_hpages);
1242                 if (!sbinfo->spool)
1243                         goto out_free;
1244         }
1245         sb->s_maxbytes = MAX_LFS_FILESIZE;
1246         sb->s_blocksize = huge_page_size(config.hstate);
1247         sb->s_blocksize_bits = huge_page_shift(config.hstate);
1248         sb->s_magic = HUGETLBFS_MAGIC;
1249         sb->s_op = &hugetlbfs_ops;
1250         sb->s_time_gran = 1;
1251         sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config));
1252         if (!sb->s_root)
1253                 goto out_free;
1254         return 0;
1255 out_free:
1256         kfree(sbinfo->spool);
1257         kfree(sbinfo);
1258         return -ENOMEM;
1259 }
1260
1261 static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type,
1262         int flags, const char *dev_name, void *data)
1263 {
1264         return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super);
1265 }
1266
1267 static struct file_system_type hugetlbfs_fs_type = {
1268         .name           = "hugetlbfs",
1269         .mount          = hugetlbfs_mount,
1270         .kill_sb        = kill_litter_super,
1271 };
1272
1273 static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE];
1274
1275 static int can_do_hugetlb_shm(void)
1276 {
1277         kgid_t shm_group;
1278         shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group);
1279         return capable(CAP_IPC_LOCK) || in_group_p(shm_group);
1280 }
1281
1282 static int get_hstate_idx(int page_size_log)
1283 {
1284         struct hstate *h = hstate_sizelog(page_size_log);
1285
1286         if (!h)
1287                 return -1;
1288         return h - hstates;
1289 }
1290
1291 static const struct dentry_operations anon_ops = {
1292         .d_dname = simple_dname
1293 };
1294
1295 /*
1296  * Note that size should be aligned to proper hugepage size in caller side,
1297  * otherwise hugetlb_reserve_pages reserves one less hugepages than intended.
1298  */
1299 struct file *hugetlb_file_setup(const char *name, size_t size,
1300                                 vm_flags_t acctflag, struct user_struct **user,
1301                                 int creat_flags, int page_size_log)
1302 {
1303         struct file *file = ERR_PTR(-ENOMEM);
1304         struct inode *inode;
1305         struct path path;
1306         struct super_block *sb;
1307         struct qstr quick_string;
1308         int hstate_idx;
1309
1310         hstate_idx = get_hstate_idx(page_size_log);
1311         if (hstate_idx < 0)
1312                 return ERR_PTR(-ENODEV);
1313
1314         *user = NULL;
1315         if (!hugetlbfs_vfsmount[hstate_idx])
1316                 return ERR_PTR(-ENOENT);
1317
1318         if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
1319                 *user = current_user();
1320                 if (user_shm_lock(size, *user)) {
1321                         task_lock(current);
1322                         pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n",
1323                                 current->comm, current->pid);
1324                         task_unlock(current);
1325                 } else {
1326                         *user = NULL;
1327                         return ERR_PTR(-EPERM);
1328                 }
1329         }
1330
1331         sb = hugetlbfs_vfsmount[hstate_idx]->mnt_sb;
1332         quick_string.name = name;
1333         quick_string.len = strlen(quick_string.name);
1334         quick_string.hash = 0;
1335         path.dentry = d_alloc_pseudo(sb, &quick_string);
1336         if (!path.dentry)
1337                 goto out_shm_unlock;
1338
1339         d_set_d_op(path.dentry, &anon_ops);
1340         path.mnt = mntget(hugetlbfs_vfsmount[hstate_idx]);
1341         file = ERR_PTR(-ENOSPC);
1342         inode = hugetlbfs_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0);
1343         if (!inode)
1344                 goto out_dentry;
1345         if (creat_flags == HUGETLB_SHMFS_INODE)
1346                 inode->i_flags |= S_PRIVATE;
1347
1348         file = ERR_PTR(-ENOMEM);
1349         if (hugetlb_reserve_pages(inode, 0,
1350                         size >> huge_page_shift(hstate_inode(inode)), NULL,
1351                         acctflag))
1352                 goto out_inode;
1353
1354         d_instantiate(path.dentry, inode);
1355         inode->i_size = size;
1356         clear_nlink(inode);
1357
1358         file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
1359                         &hugetlbfs_file_operations);
1360         if (IS_ERR(file))
1361                 goto out_dentry; /* inode is already attached */
1362
1363         return file;
1364
1365 out_inode:
1366         iput(inode);
1367 out_dentry:
1368         path_put(&path);
1369 out_shm_unlock:
1370         if (*user) {
1371                 user_shm_unlock(size, *user);
1372                 *user = NULL;
1373         }
1374         return file;
1375 }
1376
1377 static int __init init_hugetlbfs_fs(void)
1378 {
1379         struct hstate *h;
1380         int error;
1381         int i;
1382
1383         if (!hugepages_supported()) {
1384                 pr_info("disabling because there are no supported hugepage sizes\n");
1385                 return -ENOTSUPP;
1386         }
1387
1388         error = -ENOMEM;
1389         hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1390                                         sizeof(struct hugetlbfs_inode_info),
1391                                         0, SLAB_ACCOUNT, init_once);
1392         if (hugetlbfs_inode_cachep == NULL)
1393                 goto out2;
1394
1395         error = register_filesystem(&hugetlbfs_fs_type);
1396         if (error)
1397                 goto out;
1398
1399         i = 0;
1400         for_each_hstate(h) {
1401                 char buf[50];
1402                 unsigned ps_kb = 1U << (h->order + PAGE_SHIFT - 10);
1403
1404                 snprintf(buf, sizeof(buf), "pagesize=%uK", ps_kb);
1405                 hugetlbfs_vfsmount[i] = kern_mount_data(&hugetlbfs_fs_type,
1406                                                         buf);
1407
1408                 if (IS_ERR(hugetlbfs_vfsmount[i])) {
1409                         pr_err("Cannot mount internal hugetlbfs for "
1410                                 "page size %uK", ps_kb);
1411                         error = PTR_ERR(hugetlbfs_vfsmount[i]);
1412                         hugetlbfs_vfsmount[i] = NULL;
1413                 }
1414                 i++;
1415         }
1416         /* Non default hstates are optional */
1417         if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount[default_hstate_idx]))
1418                 return 0;
1419
1420  out:
1421         kmem_cache_destroy(hugetlbfs_inode_cachep);
1422  out2:
1423         return error;
1424 }
1425 fs_initcall(init_hugetlbfs_fs)