?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
x86/oprofile: Fix bogus GCC-8 warning in nmi_setup()
[cris-mirror.git] / fs / hugetlbfs / inode.c
blob8fe1b0aa2896bc547955b8760bffea0ec326d7e8
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 int sysctl_hugetlb_shm_group;
59
60 enum {
61 Opt_size, Opt_nr_inodes,
62 Opt_mode, Opt_uid, Opt_gid,
63 Opt_pagesize, Opt_min_size,
64 Opt_err,
65 };
66
67 static const match_table_t tokens = {
68 {Opt_size, "size=%s"},
69 {Opt_nr_inodes, "nr_inodes=%s"},
70 {Opt_mode, "mode=%o"},
71 {Opt_uid, "uid=%u"},
72 {Opt_gid, "gid=%u"},
73 {Opt_pagesize, "pagesize=%s"},
74 {Opt_min_size, "min_size=%s"},
75 {Opt_err, NULL},
76 };
77
78 #ifdef CONFIG_NUMA
79 static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma,
80 struct inode *inode, pgoff_t index)
81 {
82 vma->vm_policy = mpol_shared_policy_lookup(&HUGETLBFS_I(inode)->policy,
83 index);
84 }
85
86 static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma)
87 {
88 mpol_cond_put(vma->vm_policy);
89 }
90 #else
91 static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma,
92 struct inode *inode, pgoff_t index)
93 {
94 }
95
96 static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma)
97 {
98 }
99 #endif
100
101 static void huge_pagevec_release(struct pagevec *pvec)
102 {
103 int i;
104
105 for (i = 0; i < pagevec_count(pvec); ++i)
106 put_page(pvec->pages[i]);
107
108 pagevec_reinit(pvec);
109 }
110
111 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
112 {
113 struct inode *inode = file_inode(file);
114 loff_t len, vma_len;
115 int ret;
116 struct hstate *h = hstate_file(file);
117
118 /*
119 * vma address alignment (but not the pgoff alignment) has
120 * already been checked by prepare_hugepage_range. If you add
121 * any error returns here, do so after setting VM_HUGETLB, so
122 * is_vm_hugetlb_page tests below unmap_region go the right
123 * way when do_mmap_pgoff unwinds (may be important on powerpc
124 * and ia64).
125 */
126 vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND;
127 vma->vm_ops = &hugetlb_vm_ops;
128
129 /*
130 * Offset passed to mmap (before page shift) could have been
131 * negative when represented as a (l)off_t.
132 */
133 if (((loff_t)vma->vm_pgoff << PAGE_SHIFT) < 0)
134 return -EINVAL;
135
136 if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT))
137 return -EINVAL;
138
139 vma_len = (loff_t)(vma->vm_end - vma->vm_start);
140 len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
141 /* check for overflow */
142 if (len < vma_len)
143 return -EINVAL;
144
145 inode_lock(inode);
146 file_accessed(file);
147
148 ret = -ENOMEM;
149 if (hugetlb_reserve_pages(inode,
150 vma->vm_pgoff >> huge_page_order(h),
151 len >> huge_page_shift(h), vma,
152 vma->vm_flags))
153 goto out;
154
155 ret = 0;
156 if (vma->vm_flags & VM_WRITE && inode->i_size < len)
157 i_size_write(inode, len);
158 out:
159 inode_unlock(inode);
160
161 return ret;
162 }
163
164 /*
165 * Called under down_write(mmap_sem).
166 */
167
168 #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA
169 static unsigned long
170 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
171 unsigned long len, unsigned long pgoff, unsigned long flags)
172 {
173 struct mm_struct *mm = current->mm;
174 struct vm_area_struct *vma;
175 struct hstate *h = hstate_file(file);
176 struct vm_unmapped_area_info info;
177
178 if (len & ~huge_page_mask(h))
179 return -EINVAL;
180 if (len > TASK_SIZE)
181 return -ENOMEM;
182
183 if (flags & MAP_FIXED) {
184 if (prepare_hugepage_range(file, addr, len))
185 return -EINVAL;
186 return addr;
187 }
188
189 if (addr) {
190 addr = ALIGN(addr, huge_page_size(h));
191 vma = find_vma(mm, addr);
192 if (TASK_SIZE - len >= addr &&
193 (!vma || addr + len <= vm_start_gap(vma)))
194 return addr;
195 }
196
197 info.flags = 0;
198 info.length = len;
199 info.low_limit = TASK_UNMAPPED_BASE;
200 info.high_limit = TASK_SIZE;
201 info.align_mask = PAGE_MASK & ~huge_page_mask(h);
202 info.align_offset = 0;
203 return vm_unmapped_area(&info);
204 }
205 #endif
206
207 static size_t
208 hugetlbfs_read_actor(struct page *page, unsigned long offset,
209 struct iov_iter *to, unsigned long size)
210 {
211 size_t copied = 0;
212 int i, chunksize;
213
214 /* Find which 4k chunk and offset with in that chunk */
215 i = offset >> PAGE_SHIFT;
216 offset = offset & ~PAGE_MASK;
217
218 while (size) {
219 size_t n;
220 chunksize = PAGE_SIZE;
221 if (offset)
222 chunksize -= offset;
223 if (chunksize > size)
224 chunksize = size;
225 n = copy_page_to_iter(&page[i], offset, chunksize, to);
226 copied += n;
227 if (n != chunksize)
228 return copied;
229 offset = 0;
230 size -= chunksize;
231 i++;
232 }
233 return copied;
234 }
235
236 /*
237 * Support for read() - Find the page attached to f_mapping and copy out the
238 * data. Its *very* similar to do_generic_mapping_read(), we can't use that
239 * since it has PAGE_SIZE assumptions.
240 */
241 static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to)
242 {
243 struct file *file = iocb->ki_filp;
244 struct hstate *h = hstate_file(file);
245 struct address_space *mapping = file->f_mapping;
246 struct inode *inode = mapping->host;
247 unsigned long index = iocb->ki_pos >> huge_page_shift(h);
248 unsigned long offset = iocb->ki_pos & ~huge_page_mask(h);
249 unsigned long end_index;
250 loff_t isize;
251 ssize_t retval = 0;
252
253 while (iov_iter_count(to)) {
254 struct page *page;
255 size_t nr, copied;
256
257 /* nr is the maximum number of bytes to copy from this page */
258 nr = huge_page_size(h);
259 isize = i_size_read(inode);
260 if (!isize)
261 break;
262 end_index = (isize - 1) >> huge_page_shift(h);
263 if (index > end_index)
264 break;
265 if (index == end_index) {
266 nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
267 if (nr <= offset)
268 break;
269 }
270 nr = nr - offset;
271
272 /* Find the page */
273 page = find_lock_page(mapping, index);
274 if (unlikely(page == NULL)) {
275 /*
276 * We have a HOLE, zero out the user-buffer for the
277 * length of the hole or request.
278 */
279 copied = iov_iter_zero(nr, to);
280 } else {
281 unlock_page(page);
282
283 /*
284 * We have the page, copy it to user space buffer.
285 */
286 copied = hugetlbfs_read_actor(page, offset, to, nr);
287 put_page(page);
288 }
289 offset += copied;
290 retval += copied;
291 if (copied != nr && iov_iter_count(to)) {
292 if (!retval)
293 retval = -EFAULT;
294 break;
295 }
296 index += offset >> huge_page_shift(h);
297 offset &= ~huge_page_mask(h);
298 }
299 iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset;
300 return retval;
301 }
302
303 static int hugetlbfs_write_begin(struct file *file,
304 struct address_space *mapping,
305 loff_t pos, unsigned len, unsigned flags,
306 struct page **pagep, void **fsdata)
307 {
308 return -EINVAL;
309 }
310
311 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
312 loff_t pos, unsigned len, unsigned copied,
313 struct page *page, void *fsdata)
314 {
315 BUG();
316 return -EINVAL;
317 }
318
319 static void remove_huge_page(struct page *page)
320 {
321 ClearPageDirty(page);
322 ClearPageUptodate(page);
323 delete_from_page_cache(page);
324 }
325
326 static void
327 hugetlb_vmdelete_list(struct rb_root_cached *root, pgoff_t start, pgoff_t end)
328 {
329 struct vm_area_struct *vma;
330
331 /*
332 * end == 0 indicates that the entire range after
333 * start should be unmapped.
334 */
335 vma_interval_tree_foreach(vma, root, start, end ? end : ULONG_MAX) {
336 unsigned long v_offset;
337 unsigned long v_end;
338
339 /*
340 * Can the expression below overflow on 32-bit arches?
341 * No, because the interval tree returns us only those vmas
342 * which overlap the truncated area starting at pgoff,
343 * and no vma on a 32-bit arch can span beyond the 4GB.
344 */
345 if (vma->vm_pgoff < start)
346 v_offset = (start - vma->vm_pgoff) << PAGE_SHIFT;
347 else
348 v_offset = 0;
349
350 if (!end)
351 v_end = vma->vm_end;
352 else {
353 v_end = ((end - vma->vm_pgoff) << PAGE_SHIFT)
354 + vma->vm_start;
355 if (v_end > vma->vm_end)
356 v_end = vma->vm_end;
357 }
358
359 unmap_hugepage_range(vma, vma->vm_start + v_offset, v_end,
360 NULL);
361 }
362 }
363
364 /*
365 * remove_inode_hugepages handles two distinct cases: truncation and hole
366 * punch. There are subtle differences in operation for each case.
367 *
368 * truncation is indicated by end of range being LLONG_MAX
369 * In this case, we first scan the range and release found pages.
370 * After releasing pages, hugetlb_unreserve_pages cleans up region/reserv
371 * maps and global counts. Page faults can not race with truncation
372 * in this routine. hugetlb_no_page() prevents page faults in the
373 * truncated range. It checks i_size before allocation, and again after
374 * with the page table lock for the page held. The same lock must be
375 * acquired to unmap a page.
376 * hole punch is indicated if end is not LLONG_MAX
377 * In the hole punch case we scan the range and release found pages.
378 * Only when releasing a page is the associated region/reserv map
379 * deleted. The region/reserv map for ranges without associated
380 * pages are not modified. Page faults can race with hole punch.
381 * This is indicated if we find a mapped page.
382 * Note: If the passed end of range value is beyond the end of file, but
383 * not LLONG_MAX this routine still performs a hole punch operation.
384 */
385 static void remove_inode_hugepages(struct inode *inode, loff_t lstart,
386 loff_t lend)
387 {
388 struct hstate *h = hstate_inode(inode);
389 struct address_space *mapping = &inode->i_data;
390 const pgoff_t start = lstart >> huge_page_shift(h);
391 const pgoff_t end = lend >> huge_page_shift(h);
392 struct vm_area_struct pseudo_vma;
393 struct pagevec pvec;
394 pgoff_t next, index;
395 int i, freed = 0;
396 bool truncate_op = (lend == LLONG_MAX);
397
398 memset(&pseudo_vma, 0, sizeof(struct vm_area_struct));
399 pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED);
400 pagevec_init(&pvec);
401 next = start;
402 while (next < end) {
403 /*
404 * When no more pages are found, we are done.
405 */
406 if (!pagevec_lookup_range(&pvec, mapping, &next, end - 1))
407 break;
408
409 for (i = 0; i < pagevec_count(&pvec); ++i) {
410 struct page *page = pvec.pages[i];
411 u32 hash;
412
413 index = page->index;
414 hash = hugetlb_fault_mutex_hash(h, current->mm,
415 &pseudo_vma,
416 mapping, index, 0);
417 mutex_lock(&hugetlb_fault_mutex_table[hash]);
418
419 /*
420 * If page is mapped, it was faulted in after being
421 * unmapped in caller. Unmap (again) now after taking
422 * the fault mutex. The mutex will prevent faults
423 * until we finish removing the page.
424 *
425 * This race can only happen in the hole punch case.
426 * Getting here in a truncate operation is a bug.
427 */
428 if (unlikely(page_mapped(page))) {
429 BUG_ON(truncate_op);
430
431 i_mmap_lock_write(mapping);
432 hugetlb_vmdelete_list(&mapping->i_mmap,
433 index * pages_per_huge_page(h),
434 (index + 1) * pages_per_huge_page(h));
435 i_mmap_unlock_write(mapping);
436 }
437
438 lock_page(page);
439 /*
440 * We must free the huge page and remove from page
441 * cache (remove_huge_page) BEFORE removing the
442 * region/reserve map (hugetlb_unreserve_pages). In
443 * rare out of memory conditions, removal of the
444 * region/reserve map could fail. Correspondingly,
445 * the subpool and global reserve usage count can need
446 * to be adjusted.
447 */
448 VM_BUG_ON(PagePrivate(page));
449 remove_huge_page(page);
450 freed++;
451 if (!truncate_op) {
452 if (unlikely(hugetlb_unreserve_pages(inode,
453 index, index + 1, 1)))
454 hugetlb_fix_reserve_counts(inode);
455 }
456
457 unlock_page(page);
458 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
459 }
460 huge_pagevec_release(&pvec);
461 cond_resched();
462 }
463
464 if (truncate_op)
465 (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed);
466 }
467
468 static void hugetlbfs_evict_inode(struct inode *inode)
469 {
470 struct resv_map *resv_map;
471
472 remove_inode_hugepages(inode, 0, LLONG_MAX);
473 resv_map = (struct resv_map *)inode->i_mapping->private_data;
474 /* root inode doesn't have the resv_map, so we should check it */
475 if (resv_map)
476 resv_map_release(&resv_map->refs);
477 clear_inode(inode);
478 }
479
480 static int hugetlb_vmtruncate(struct inode *inode, loff_t offset)
481 {
482 pgoff_t pgoff;
483 struct address_space *mapping = inode->i_mapping;
484 struct hstate *h = hstate_inode(inode);
485
486 BUG_ON(offset & ~huge_page_mask(h));
487 pgoff = offset >> PAGE_SHIFT;
488
489 i_size_write(inode, offset);
490 i_mmap_lock_write(mapping);
491 if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))
492 hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0);
493 i_mmap_unlock_write(mapping);
494 remove_inode_hugepages(inode, offset, LLONG_MAX);
495 return 0;
496 }
497
498 static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
499 {
500 struct hstate *h = hstate_inode(inode);
501 loff_t hpage_size = huge_page_size(h);
502 loff_t hole_start, hole_end;
503
504 /*
505 * For hole punch round up the beginning offset of the hole and
506 * round down the end.
507 */
508 hole_start = round_up(offset, hpage_size);
509 hole_end = round_down(offset + len, hpage_size);
510
511 if (hole_end > hole_start) {
512 struct address_space *mapping = inode->i_mapping;
513 struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
514
515 inode_lock(inode);
516
517 /* protected by i_mutex */
518 if (info->seals & F_SEAL_WRITE) {
519 inode_unlock(inode);
520 return -EPERM;
521 }
522
523 i_mmap_lock_write(mapping);
524 if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))
525 hugetlb_vmdelete_list(&mapping->i_mmap,
526 hole_start >> PAGE_SHIFT,
527 hole_end >> PAGE_SHIFT);
528 i_mmap_unlock_write(mapping);
529 remove_inode_hugepages(inode, hole_start, hole_end);
530 inode_unlock(inode);
531 }
532
533 return 0;
534 }
535
536 static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset,
537 loff_t len)
538 {
539 struct inode *inode = file_inode(file);
540 struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
541 struct address_space *mapping = inode->i_mapping;
542 struct hstate *h = hstate_inode(inode);
543 struct vm_area_struct pseudo_vma;
544 struct mm_struct *mm = current->mm;
545 loff_t hpage_size = huge_page_size(h);
546 unsigned long hpage_shift = huge_page_shift(h);
547 pgoff_t start, index, end;
548 int error;
549 u32 hash;
550
551 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
552 return -EOPNOTSUPP;
553
554 if (mode & FALLOC_FL_PUNCH_HOLE)
555 return hugetlbfs_punch_hole(inode, offset, len);
556
557 /*
558 * Default preallocate case.
559 * For this range, start is rounded down and end is rounded up
560 * as well as being converted to page offsets.
561 */
562 start = offset >> hpage_shift;
563 end = (offset + len + hpage_size - 1) >> hpage_shift;
564
565 inode_lock(inode);
566
567 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
568 error = inode_newsize_ok(inode, offset + len);
569 if (error)
570 goto out;
571
572 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
573 error = -EPERM;
574 goto out;
575 }
576
577 /*
578 * Initialize a pseudo vma as this is required by the huge page
579 * allocation routines. If NUMA is configured, use page index
580 * as input to create an allocation policy.
581 */
582 memset(&pseudo_vma, 0, sizeof(struct vm_area_struct));
583 pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED);
584 pseudo_vma.vm_file = file;
585
586 for (index = start; index < end; index++) {
587 /*
588 * This is supposed to be the vaddr where the page is being
589 * faulted in, but we have no vaddr here.
590 */
591 struct page *page;
592 unsigned long addr;
593 int avoid_reserve = 0;
594
595 cond_resched();
596
597 /*
598 * fallocate(2) manpage permits EINTR; we may have been
599 * interrupted because we are using up too much memory.
600 */
601 if (signal_pending(current)) {
602 error = -EINTR;
603 break;
604 }
605
606 /* Set numa allocation policy based on index */
607 hugetlb_set_vma_policy(&pseudo_vma, inode, index);
608
609 /* addr is the offset within the file (zero based) */
610 addr = index * hpage_size;
611
612 /* mutex taken here, fault path and hole punch */
613 hash = hugetlb_fault_mutex_hash(h, mm, &pseudo_vma, mapping,
614 index, addr);
615 mutex_lock(&hugetlb_fault_mutex_table[hash]);
616
617 /* See if already present in mapping to avoid alloc/free */
618 page = find_get_page(mapping, index);
619 if (page) {
620 put_page(page);
621 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
622 hugetlb_drop_vma_policy(&pseudo_vma);
623 continue;
624 }
625
626 /* Allocate page and add to page cache */
627 page = alloc_huge_page(&pseudo_vma, addr, avoid_reserve);
628 hugetlb_drop_vma_policy(&pseudo_vma);
629 if (IS_ERR(page)) {
630 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
631 error = PTR_ERR(page);
632 goto out;
633 }
634 clear_huge_page(page, addr, pages_per_huge_page(h));
635 __SetPageUptodate(page);
636 error = huge_add_to_page_cache(page, mapping, index);
637 if (unlikely(error)) {
638 put_page(page);
639 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
640 goto out;
641 }
642
643 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
644
645 /*
646 * unlock_page because locked by add_to_page_cache()
647 * page_put due to reference from alloc_huge_page()
648 */
649 unlock_page(page);
650 put_page(page);
651 }
652
653 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
654 i_size_write(inode, offset + len);
655 inode->i_ctime = current_time(inode);
656 out:
657 inode_unlock(inode);
658 return error;
659 }
660
661 static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr)
662 {
663 struct inode *inode = d_inode(dentry);
664 struct hstate *h = hstate_inode(inode);
665 int error;
666 unsigned int ia_valid = attr->ia_valid;
667 struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
668
669 BUG_ON(!inode);
670
671 error = setattr_prepare(dentry, attr);
672 if (error)
673 return error;
674
675 if (ia_valid & ATTR_SIZE) {
676 loff_t oldsize = inode->i_size;
677 loff_t newsize = attr->ia_size;
678
679 if (newsize & ~huge_page_mask(h))
680 return -EINVAL;
681 /* protected by i_mutex */
682 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
683 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
684 return -EPERM;
685 error = hugetlb_vmtruncate(inode, newsize);
686 if (error)
687 return error;
688 }
689
690 setattr_copy(inode, attr);
691 mark_inode_dirty(inode);
692 return 0;
693 }
694
695 static struct inode *hugetlbfs_get_root(struct super_block *sb,
696 struct hugetlbfs_config *config)
697 {
698 struct inode *inode;
699
700 inode = new_inode(sb);
701 if (inode) {
702 inode->i_ino = get_next_ino();
703 inode->i_mode = S_IFDIR | config->mode;
704 inode->i_uid = config->uid;
705 inode->i_gid = config->gid;
706 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
707 inode->i_op = &hugetlbfs_dir_inode_operations;
708 inode->i_fop = &simple_dir_operations;
709 /* directory inodes start off with i_nlink == 2 (for "." entry) */
710 inc_nlink(inode);
711 lockdep_annotate_inode_mutex_key(inode);
712 }
713 return inode;
714 }
715
716 /*
717 * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never
718 * be taken from reclaim -- unlike regular filesystems. This needs an
719 * annotation because huge_pmd_share() does an allocation under hugetlb's
720 * i_mmap_rwsem.
721 */
722 static struct lock_class_key hugetlbfs_i_mmap_rwsem_key;
723
724 static struct inode *hugetlbfs_get_inode(struct super_block *sb,
725 struct inode *dir,
726 umode_t mode, dev_t dev)
727 {
728 struct inode *inode;
729 struct resv_map *resv_map;
730
731 resv_map = resv_map_alloc();
732 if (!resv_map)
733 return NULL;
734
735 inode = new_inode(sb);
736 if (inode) {
737 struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
738
739 inode->i_ino = get_next_ino();
740 inode_init_owner(inode, dir, mode);
741 lockdep_set_class(&inode->i_mapping->i_mmap_rwsem,
742 &hugetlbfs_i_mmap_rwsem_key);
743 inode->i_mapping->a_ops = &hugetlbfs_aops;
744 inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode);
745 inode->i_mapping->private_data = resv_map;
746 info->seals = F_SEAL_SEAL;
747 switch (mode & S_IFMT) {
748 default:
749 init_special_inode(inode, mode, dev);
750 break;
751 case S_IFREG:
752 inode->i_op = &hugetlbfs_inode_operations;
753 inode->i_fop = &hugetlbfs_file_operations;
754 break;
755 case S_IFDIR:
756 inode->i_op = &hugetlbfs_dir_inode_operations;
757 inode->i_fop = &simple_dir_operations;
758
759 /* directory inodes start off with i_nlink == 2 (for "." entry) */
760 inc_nlink(inode);
761 break;
762 case S_IFLNK:
763 inode->i_op = &page_symlink_inode_operations;
764 inode_nohighmem(inode);
765 break;
766 }
767 lockdep_annotate_inode_mutex_key(inode);
768 } else
769 kref_put(&resv_map->refs, resv_map_release);
770
771 return inode;
772 }
773
774 /*
775 * File creation. Allocate an inode, and we're done..
776 */
777 static int hugetlbfs_mknod(struct inode *dir,
778 struct dentry *dentry, umode_t mode, dev_t dev)
779 {
780 struct inode *inode;
781 int error = -ENOSPC;
782
783 inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev);
784 if (inode) {
785 dir->i_ctime = dir->i_mtime = current_time(dir);
786 d_instantiate(dentry, inode);
787 dget(dentry); /* Extra count - pin the dentry in core */
788 error = 0;
789 }
790 return error;
791 }
792
793 static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode)
794 {
795 int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0);
796 if (!retval)
797 inc_nlink(dir);
798 return retval;
799 }
800
801 static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl)
802 {
803 return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0);
804 }
805
806 static int hugetlbfs_symlink(struct inode *dir,
807 struct dentry *dentry, const char *symname)
808 {
809 struct inode *inode;
810 int error = -ENOSPC;
811
812 inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0);
813 if (inode) {
814 int l = strlen(symname)+1;
815 error = page_symlink(inode, symname, l);
816 if (!error) {
817 d_instantiate(dentry, inode);
818 dget(dentry);
819 } else
820 iput(inode);
821 }
822 dir->i_ctime = dir->i_mtime = current_time(dir);
823
824 return error;
825 }
826
827 /*
828 * mark the head page dirty
829 */
830 static int hugetlbfs_set_page_dirty(struct page *page)
831 {
832 struct page *head = compound_head(page);
833
834 SetPageDirty(head);
835 return 0;
836 }
837
838 static int hugetlbfs_migrate_page(struct address_space *mapping,
839 struct page *newpage, struct page *page,
840 enum migrate_mode mode)
841 {
842 int rc;
843
844 rc = migrate_huge_page_move_mapping(mapping, newpage, page);
845 if (rc != MIGRATEPAGE_SUCCESS)
846 return rc;
847 if (mode != MIGRATE_SYNC_NO_COPY)
848 migrate_page_copy(newpage, page);
849 else
850 migrate_page_states(newpage, page);
851
852 return MIGRATEPAGE_SUCCESS;
853 }
854
855 static int hugetlbfs_error_remove_page(struct address_space *mapping,
856 struct page *page)
857 {
858 struct inode *inode = mapping->host;
859 pgoff_t index = page->index;
860
861 remove_huge_page(page);
862 if (unlikely(hugetlb_unreserve_pages(inode, index, index + 1, 1)))
863 hugetlb_fix_reserve_counts(inode);
864
865 return 0;
866 }
867
868 /*
869 * Display the mount options in /proc/mounts.
870 */
871 static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root)
872 {
873 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(root->d_sb);
874 struct hugepage_subpool *spool = sbinfo->spool;
875 unsigned long hpage_size = huge_page_size(sbinfo->hstate);
876 unsigned hpage_shift = huge_page_shift(sbinfo->hstate);
877 char mod;
878
879 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
880 seq_printf(m, ",uid=%u",
881 from_kuid_munged(&init_user_ns, sbinfo->uid));
882 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
883 seq_printf(m, ",gid=%u",
884 from_kgid_munged(&init_user_ns, sbinfo->gid));
885 if (sbinfo->mode != 0755)
886 seq_printf(m, ",mode=%o", sbinfo->mode);
887 if (sbinfo->max_inodes != -1)
888 seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes);
889
890 hpage_size /= 1024;
891 mod = 'K';
892 if (hpage_size >= 1024) {
893 hpage_size /= 1024;
894 mod = 'M';
895 }
896 seq_printf(m, ",pagesize=%lu%c", hpage_size, mod);
897 if (spool) {
898 if (spool->max_hpages != -1)
899 seq_printf(m, ",size=%llu",
900 (unsigned long long)spool->max_hpages << hpage_shift);
901 if (spool->min_hpages != -1)
902 seq_printf(m, ",min_size=%llu",
903 (unsigned long long)spool->min_hpages << hpage_shift);
904 }
905 return 0;
906 }
907
908 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
909 {
910 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
911 struct hstate *h = hstate_inode(d_inode(dentry));
912
913 buf->f_type = HUGETLBFS_MAGIC;
914 buf->f_bsize = huge_page_size(h);
915 if (sbinfo) {
916 spin_lock(&sbinfo->stat_lock);
917 /* If no limits set, just report 0 for max/free/used
918 * blocks, like simple_statfs() */
919 if (sbinfo->spool) {
920 long free_pages;
921
922 spin_lock(&sbinfo->spool->lock);
923 buf->f_blocks = sbinfo->spool->max_hpages;
924 free_pages = sbinfo->spool->max_hpages
925 - sbinfo->spool->used_hpages;
926 buf->f_bavail = buf->f_bfree = free_pages;
927 spin_unlock(&sbinfo->spool->lock);
928 buf->f_files = sbinfo->max_inodes;
929 buf->f_ffree = sbinfo->free_inodes;
930 }
931 spin_unlock(&sbinfo->stat_lock);
932 }
933 buf->f_namelen = NAME_MAX;
934 return 0;
935 }
936
937 static void hugetlbfs_put_super(struct super_block *sb)
938 {
939 struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
940
941 if (sbi) {
942 sb->s_fs_info = NULL;
943
944 if (sbi->spool)
945 hugepage_put_subpool(sbi->spool);
946
947 kfree(sbi);
948 }
949 }
950
951 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
952 {
953 if (sbinfo->free_inodes >= 0) {
954 spin_lock(&sbinfo->stat_lock);
955 if (unlikely(!sbinfo->free_inodes)) {
956 spin_unlock(&sbinfo->stat_lock);
957 return 0;
958 }
959 sbinfo->free_inodes--;
960 spin_unlock(&sbinfo->stat_lock);
961 }
962
963 return 1;
964 }
965
966 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
967 {
968 if (sbinfo->free_inodes >= 0) {
969 spin_lock(&sbinfo->stat_lock);
970 sbinfo->free_inodes++;
971 spin_unlock(&sbinfo->stat_lock);
972 }
973 }
974
975
976 static struct kmem_cache *hugetlbfs_inode_cachep;
977
978 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
979 {
980 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
981 struct hugetlbfs_inode_info *p;
982
983 if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
984 return NULL;
985 p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL);
986 if (unlikely(!p)) {
987 hugetlbfs_inc_free_inodes(sbinfo);
988 return NULL;
989 }
990
991 /*
992 * Any time after allocation, hugetlbfs_destroy_inode can be called
993 * for the inode. mpol_free_shared_policy is unconditionally called
994 * as part of hugetlbfs_destroy_inode. So, initialize policy here
995 * in case of a quick call to destroy.
996 *
997 * Note that the policy is initialized even if we are creating a
998 * private inode. This simplifies hugetlbfs_destroy_inode.
999 */
1000 mpol_shared_policy_init(&p->policy, NULL);
1001
1002 return &p->vfs_inode;
1003 }
1004
1005 static void hugetlbfs_i_callback(struct rcu_head *head)
1006 {
1007 struct inode *inode = container_of(head, struct inode, i_rcu);
1008 kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
1009 }
1010
1011 static void hugetlbfs_destroy_inode(struct inode *inode)
1012 {
1013 hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
1014 mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy);
1015 call_rcu(&inode->i_rcu, hugetlbfs_i_callback);
1016 }
1017
1018 static const struct address_space_operations hugetlbfs_aops = {
1019 .write_begin = hugetlbfs_write_begin,
1020 .write_end = hugetlbfs_write_end,
1021 .set_page_dirty = hugetlbfs_set_page_dirty,
1022 .migratepage = hugetlbfs_migrate_page,
1023 .error_remove_page = hugetlbfs_error_remove_page,
1024 };
1025
1026
1027 static void init_once(void *foo)
1028 {
1029 struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo;
1030
1031 inode_init_once(&ei->vfs_inode);
1032 }
1033
1034 const struct file_operations hugetlbfs_file_operations = {
1035 .read_iter = hugetlbfs_read_iter,
1036 .mmap = hugetlbfs_file_mmap,
1037 .fsync = noop_fsync,
1038 .get_unmapped_area = hugetlb_get_unmapped_area,
1039 .llseek = default_llseek,
1040 .fallocate = hugetlbfs_fallocate,
1041 };
1042
1043 static const struct inode_operations hugetlbfs_dir_inode_operations = {
1044 .create = hugetlbfs_create,
1045 .lookup = simple_lookup,
1046 .link = simple_link,
1047 .unlink = simple_unlink,
1048 .symlink = hugetlbfs_symlink,
1049 .mkdir = hugetlbfs_mkdir,
1050 .rmdir = simple_rmdir,
1051 .mknod = hugetlbfs_mknod,
1052 .rename = simple_rename,
1053 .setattr = hugetlbfs_setattr,
1054 };
1055
1056 static const struct inode_operations hugetlbfs_inode_operations = {
1057 .setattr = hugetlbfs_setattr,
1058 };
1059
1060 static const struct super_operations hugetlbfs_ops = {
1061 .alloc_inode = hugetlbfs_alloc_inode,
1062 .destroy_inode = hugetlbfs_destroy_inode,
1063 .evict_inode = hugetlbfs_evict_inode,
1064 .statfs = hugetlbfs_statfs,
1065 .put_super = hugetlbfs_put_super,
1066 .show_options = hugetlbfs_show_options,
1067 };
1068
1069 enum hugetlbfs_size_type { NO_SIZE, SIZE_STD, SIZE_PERCENT };
1070
1071 /*
1072 * Convert size option passed from command line to number of huge pages
1073 * in the pool specified by hstate. Size option could be in bytes
1074 * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT).
1075 */
1076 static long
1077 hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt,
1078 enum hugetlbfs_size_type val_type)
1079 {
1080 if (val_type == NO_SIZE)
1081 return -1;
1082
1083 if (val_type == SIZE_PERCENT) {
1084 size_opt <<= huge_page_shift(h);
1085 size_opt *= h->max_huge_pages;
1086 do_div(size_opt, 100);
1087 }
1088
1089 size_opt >>= huge_page_shift(h);
1090 return size_opt;
1091 }
1092
1093 static int
1094 hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig)
1095 {
1096 char *p, *rest;
1097 substring_t args[MAX_OPT_ARGS];
1098 int option;
1099 unsigned long long max_size_opt = 0, min_size_opt = 0;
1100 enum hugetlbfs_size_type max_val_type = NO_SIZE, min_val_type = NO_SIZE;
1101
1102 if (!options)
1103 return 0;
1104
1105 while ((p = strsep(&options, ",")) != NULL) {
1106 int token;
1107 if (!*p)
1108 continue;
1109
1110 token = match_token(p, tokens, args);
1111 switch (token) {
1112 case Opt_uid:
1113 if (match_int(&args[0], &option))
1114 goto bad_val;
1115 pconfig->uid = make_kuid(current_user_ns(), option);
1116 if (!uid_valid(pconfig->uid))
1117 goto bad_val;
1118 break;
1119
1120 case Opt_gid:
1121 if (match_int(&args[0], &option))
1122 goto bad_val;
1123 pconfig->gid = make_kgid(current_user_ns(), option);
1124 if (!gid_valid(pconfig->gid))
1125 goto bad_val;
1126 break;
1127
1128 case Opt_mode:
1129 if (match_octal(&args[0], &option))
1130 goto bad_val;
1131 pconfig->mode = option & 01777U;
1132 break;
1133
1134 case Opt_size: {
1135 /* memparse() will accept a K/M/G without a digit */
1136 if (!isdigit(*args[0].from))
1137 goto bad_val;
1138 max_size_opt = memparse(args[0].from, &rest);
1139 max_val_type = SIZE_STD;
1140 if (*rest == '%')
1141 max_val_type = SIZE_PERCENT;
1142 break;
1143 }
1144
1145 case Opt_nr_inodes:
1146 /* memparse() will accept a K/M/G without a digit */
1147 if (!isdigit(*args[0].from))
1148 goto bad_val;
1149 pconfig->nr_inodes = memparse(args[0].from, &rest);
1150 break;
1151
1152 case Opt_pagesize: {
1153 unsigned long ps;
1154 ps = memparse(args[0].from, &rest);
1155 pconfig->hstate = size_to_hstate(ps);
1156 if (!pconfig->hstate) {
1157 pr_err("Unsupported page size %lu MB\n",
1158 ps >> 20);
1159 return -EINVAL;
1160 }
1161 break;
1162 }
1163
1164 case Opt_min_size: {
1165 /* memparse() will accept a K/M/G without a digit */
1166 if (!isdigit(*args[0].from))
1167 goto bad_val;
1168 min_size_opt = memparse(args[0].from, &rest);
1169 min_val_type = SIZE_STD;
1170 if (*rest == '%')
1171 min_val_type = SIZE_PERCENT;
1172 break;
1173 }
1174
1175 default:
1176 pr_err("Bad mount option: \"%s\"\n", p);
1177 return -EINVAL;
1178 break;
1179 }
1180 }
1181
1182 /*
1183 * Use huge page pool size (in hstate) to convert the size
1184 * options to number of huge pages. If NO_SIZE, -1 is returned.
1185 */
1186 pconfig->max_hpages = hugetlbfs_size_to_hpages(pconfig->hstate,
1187 max_size_opt, max_val_type);
1188 pconfig->min_hpages = hugetlbfs_size_to_hpages(pconfig->hstate,
1189 min_size_opt, min_val_type);
1190
1191 /*
1192 * If max_size was specified, then min_size must be smaller
1193 */
1194 if (max_val_type > NO_SIZE &&
1195 pconfig->min_hpages > pconfig->max_hpages) {
1196 pr_err("minimum size can not be greater than maximum size\n");
1197 return -EINVAL;
1198 }
1199
1200 return 0;
1201
1202 bad_val:
1203 pr_err("Bad value '%s' for mount option '%s'\n", args[0].from, p);
1204 return -EINVAL;
1205 }
1206
1207 static int
1208 hugetlbfs_fill_super(struct super_block *sb, void *data, int silent)
1209 {
1210 int ret;
1211 struct hugetlbfs_config config;
1212 struct hugetlbfs_sb_info *sbinfo;
1213
1214 config.max_hpages = -1; /* No limit on size by default */
1215 config.nr_inodes = -1; /* No limit on number of inodes by default */
1216 config.uid = current_fsuid();
1217 config.gid = current_fsgid();
1218 config.mode = 0755;
1219 config.hstate = &default_hstate;
1220 config.min_hpages = -1; /* No default minimum size */
1221 ret = hugetlbfs_parse_options(data, &config);
1222 if (ret)
1223 return ret;
1224
1225 sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
1226 if (!sbinfo)
1227 return -ENOMEM;
1228 sb->s_fs_info = sbinfo;
1229 sbinfo->hstate = config.hstate;
1230 spin_lock_init(&sbinfo->stat_lock);
1231 sbinfo->max_inodes = config.nr_inodes;
1232 sbinfo->free_inodes = config.nr_inodes;
1233 sbinfo->spool = NULL;
1234 sbinfo->uid = config.uid;
1235 sbinfo->gid = config.gid;
1236 sbinfo->mode = config.mode;
1237
1238 /*
1239 * Allocate and initialize subpool if maximum or minimum size is
1240 * specified. Any needed reservations (for minimim size) are taken
1241 * taken when the subpool is created.
1242 */
1243 if (config.max_hpages != -1 || config.min_hpages != -1) {
1244 sbinfo->spool = hugepage_new_subpool(config.hstate,
1245 config.max_hpages,
1246 config.min_hpages);
1247 if (!sbinfo->spool)
1248 goto out_free;
1249 }
1250 sb->s_maxbytes = MAX_LFS_FILESIZE;
1251 sb->s_blocksize = huge_page_size(config.hstate);
1252 sb->s_blocksize_bits = huge_page_shift(config.hstate);
1253 sb->s_magic = HUGETLBFS_MAGIC;
1254 sb->s_op = &hugetlbfs_ops;
1255 sb->s_time_gran = 1;
1256 sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config));
1257 if (!sb->s_root)
1258 goto out_free;
1259 return 0;
1260 out_free:
1261 kfree(sbinfo->spool);
1262 kfree(sbinfo);
1263 return -ENOMEM;
1264 }
1265
1266 static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type,
1267 int flags, const char *dev_name, void *data)
1268 {
1269 return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super);
1270 }
1271
1272 static struct file_system_type hugetlbfs_fs_type = {
1273 .name = "hugetlbfs",
1274 .mount = hugetlbfs_mount,
1275 .kill_sb = kill_litter_super,
1276 };
1277
1278 static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE];
1279
1280 static int can_do_hugetlb_shm(void)
1281 {
1282 kgid_t shm_group;
1283 shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group);
1284 return capable(CAP_IPC_LOCK) || in_group_p(shm_group);
1285 }
1286
1287 static int get_hstate_idx(int page_size_log)
1288 {
1289 struct hstate *h = hstate_sizelog(page_size_log);
1290
1291 if (!h)
1292 return -1;
1293 return h - hstates;
1294 }
1295
1296 static const struct dentry_operations anon_ops = {
1297 .d_dname = simple_dname
1298 };
1299
1300 /*
1301 * Note that size should be aligned to proper hugepage size in caller side,
1302 * otherwise hugetlb_reserve_pages reserves one less hugepages than intended.
1303 */
1304 struct file *hugetlb_file_setup(const char *name, size_t size,
1305 vm_flags_t acctflag, struct user_struct **user,
1306 int creat_flags, int page_size_log)
1307 {
1308 struct file *file = ERR_PTR(-ENOMEM);
1309 struct inode *inode;
1310 struct path path;
1311 struct super_block *sb;
1312 struct qstr quick_string;
1313 int hstate_idx;
1314
1315 hstate_idx = get_hstate_idx(page_size_log);
1316 if (hstate_idx < 0)
1317 return ERR_PTR(-ENODEV);
1318
1319 *user = NULL;
1320 if (!hugetlbfs_vfsmount[hstate_idx])
1321 return ERR_PTR(-ENOENT);
1322
1323 if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
1324 *user = current_user();
1325 if (user_shm_lock(size, *user)) {
1326 task_lock(current);
1327 pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n",
1328 current->comm, current->pid);
1329 task_unlock(current);
1330 } else {
1331 *user = NULL;
1332 return ERR_PTR(-EPERM);
1333 }
1334 }
1335
1336 sb = hugetlbfs_vfsmount[hstate_idx]->mnt_sb;
1337 quick_string.name = name;
1338 quick_string.len = strlen(quick_string.name);
1339 quick_string.hash = 0;
1340 path.dentry = d_alloc_pseudo(sb, &quick_string);
1341 if (!path.dentry)
1342 goto out_shm_unlock;
1343
1344 d_set_d_op(path.dentry, &anon_ops);
1345 path.mnt = mntget(hugetlbfs_vfsmount[hstate_idx]);
1346 file = ERR_PTR(-ENOSPC);
1347 inode = hugetlbfs_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0);
1348 if (!inode)
1349 goto out_dentry;
1350 if (creat_flags == HUGETLB_SHMFS_INODE)
1351 inode->i_flags |= S_PRIVATE;
1352
1353 file = ERR_PTR(-ENOMEM);
1354 if (hugetlb_reserve_pages(inode, 0,
1355 size >> huge_page_shift(hstate_inode(inode)), NULL,
1356 acctflag))
1357 goto out_inode;
1358
1359 d_instantiate(path.dentry, inode);
1360 inode->i_size = size;
1361 clear_nlink(inode);
1362
1363 file = alloc_file(&path, FMODE_WRITE | FMODE_READ,
1364 &hugetlbfs_file_operations);
1365 if (IS_ERR(file))
1366 goto out_dentry; /* inode is already attached */
1367
1368 return file;
1369
1370 out_inode:
1371 iput(inode);
1372 out_dentry:
1373 path_put(&path);
1374 out_shm_unlock:
1375 if (*user) {
1376 user_shm_unlock(size, *user);
1377 *user = NULL;
1378 }
1379 return file;
1380 }
1381
1382 static int __init init_hugetlbfs_fs(void)
1383 {
1384 struct hstate *h;
1385 int error;
1386 int i;
1387
1388 if (!hugepages_supported()) {
1389 pr_info("disabling because there are no supported hugepage sizes\n");
1390 return -ENOTSUPP;
1391 }
1392
1393 error = -ENOMEM;
1394 hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1395 sizeof(struct hugetlbfs_inode_info),
1396 0, SLAB_ACCOUNT, init_once);
1397 if (hugetlbfs_inode_cachep == NULL)
1398 goto out2;
1399
1400 error = register_filesystem(&hugetlbfs_fs_type);
1401 if (error)
1402 goto out;
1403
1404 i = 0;
1405 for_each_hstate(h) {
1406 char buf[50];
1407 unsigned ps_kb = 1U << (h->order + PAGE_SHIFT - 10);
1408
1409 snprintf(buf, sizeof(buf), "pagesize=%uK", ps_kb);
1410 hugetlbfs_vfsmount[i] = kern_mount_data(&hugetlbfs_fs_type,
1411 buf);
1412
1413 if (IS_ERR(hugetlbfs_vfsmount[i])) {
1414 pr_err("Cannot mount internal hugetlbfs for "
1415 "page size %uK", ps_kb);
1416 error = PTR_ERR(hugetlbfs_vfsmount[i]);
1417 hugetlbfs_vfsmount[i] = NULL;
1418 }
1419 i++;
1420 }
1421 /* Non default hstates are optional */
1422 if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount[default_hstate_idx]))
1423 return 0;
1424
1425 out:
1426 kmem_cache_destroy(hugetlbfs_inode_cachep);
1427 out2:
1428 return error;
1429 }
1430 fs_initcall(init_hugetlbfs_fs)
CRIS linux kernel tree