?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge tag 'for-linus' of git://git.kernel.org/pub/scm/virt/kvm/kvm
[linux-stable.git] / fs / hugetlbfs / inode.c
blobfc1ae513212716a4b86cc62c27d41576f246e27f
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/falloc.h>
15 #include <linux/fs.h>
16 #include <linux/mount.h>
17 #include <linux/file.h>
18 #include <linux/kernel.h>
19 #include <linux/writeback.h>
20 #include <linux/pagemap.h>
21 #include <linux/highmem.h>
22 #include <linux/init.h>
23 #include <linux/string.h>
24 #include <linux/capability.h>
25 #include <linux/ctype.h>
26 #include <linux/backing-dev.h>
27 #include <linux/hugetlb.h>
28 #include <linux/pagevec.h>
29 #include <linux/fs_parser.h>
30 #include <linux/mman.h>
31 #include <linux/slab.h>
32 #include <linux/dnotify.h>
33 #include <linux/statfs.h>
34 #include <linux/security.h>
35 #include <linux/magic.h>
36 #include <linux/migrate.h>
37 #include <linux/uio.h>
38
39 #include <linux/uaccess.h>
40 #include <linux/sched/mm.h>
41
42 #define CREATE_TRACE_POINTS
43 #include <trace/events/hugetlbfs.h>
44
45 static const struct address_space_operations hugetlbfs_aops;
46 static const struct file_operations hugetlbfs_file_operations;
47 static const struct inode_operations hugetlbfs_dir_inode_operations;
48 static const struct inode_operations hugetlbfs_inode_operations;
49
50 enum hugetlbfs_size_type { NO_SIZE, SIZE_STD, SIZE_PERCENT };
51
52 struct hugetlbfs_fs_context {
53 struct hstate *hstate;
54 unsigned long long max_size_opt;
55 unsigned long long min_size_opt;
56 long max_hpages;
57 long nr_inodes;
58 long min_hpages;
59 enum hugetlbfs_size_type max_val_type;
60 enum hugetlbfs_size_type min_val_type;
61 kuid_t uid;
62 kgid_t gid;
63 umode_t mode;
64 };
65
66 int sysctl_hugetlb_shm_group;
67
68 enum hugetlb_param {
69 Opt_gid,
70 Opt_min_size,
71 Opt_mode,
72 Opt_nr_inodes,
73 Opt_pagesize,
74 Opt_size,
75 Opt_uid,
76 };
77
78 static const struct fs_parameter_spec hugetlb_fs_parameters[] = {
79 fsparam_gid ("gid", Opt_gid),
80 fsparam_string("min_size", Opt_min_size),
81 fsparam_u32oct("mode", Opt_mode),
82 fsparam_string("nr_inodes", Opt_nr_inodes),
83 fsparam_string("pagesize", Opt_pagesize),
84 fsparam_string("size", Opt_size),
85 fsparam_uid ("uid", Opt_uid),
86 {}
87 };
88
89 /*
90 * Mask used when checking the page offset value passed in via system
91 * calls. This value will be converted to a loff_t which is signed.
92 * Therefore, we want to check the upper PAGE_SHIFT + 1 bits of the
93 * value. The extra bit (- 1 in the shift value) is to take the sign
94 * bit into account.
95 */
96 #define PGOFF_LOFFT_MAX \
97 (((1UL << (PAGE_SHIFT + 1)) - 1) << (BITS_PER_LONG - (PAGE_SHIFT + 1)))
98
99 static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma)
100 {
101 struct inode *inode = file_inode(file);
102 struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
103 loff_t len, vma_len;
104 int ret;
105 struct hstate *h = hstate_file(file);
106 vm_flags_t vm_flags;
107
108 /*
109 * vma address alignment (but not the pgoff alignment) has
110 * already been checked by prepare_hugepage_range. If you add
111 * any error returns here, do so after setting VM_HUGETLB, so
112 * is_vm_hugetlb_page tests below unmap_region go the right
113 * way when do_mmap unwinds (may be important on powerpc
114 * and ia64).
115 */
116 vm_flags_set(vma, VM_HUGETLB | VM_DONTEXPAND);
117 vma->vm_ops = &hugetlb_vm_ops;
118
119 ret = seal_check_write(info->seals, vma);
120 if (ret)
121 return ret;
122
123 /*
124 * page based offset in vm_pgoff could be sufficiently large to
125 * overflow a loff_t when converted to byte offset. This can
126 * only happen on architectures where sizeof(loff_t) ==
127 * sizeof(unsigned long). So, only check in those instances.
128 */
129 if (sizeof(unsigned long) == sizeof(loff_t)) {
130 if (vma->vm_pgoff & PGOFF_LOFFT_MAX)
131 return -EINVAL;
132 }
133
134 /* must be huge page aligned */
135 if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT))
136 return -EINVAL;
137
138 vma_len = (loff_t)(vma->vm_end - vma->vm_start);
139 len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT);
140 /* check for overflow */
141 if (len < vma_len)
142 return -EINVAL;
143
144 inode_lock(inode);
145 file_accessed(file);
146
147 ret = -ENOMEM;
148
149 vm_flags = vma->vm_flags;
150 /*
151 * for SHM_HUGETLB, the pages are reserved in the shmget() call so skip
152 * reserving here. Note: only for SHM hugetlbfs file, the inode
153 * flag S_PRIVATE is set.
154 */
155 if (inode->i_flags & S_PRIVATE)
156 vm_flags |= VM_NORESERVE;
157
158 if (!hugetlb_reserve_pages(inode,
159 vma->vm_pgoff >> huge_page_order(h),
160 len >> huge_page_shift(h), vma,
161 vm_flags))
162 goto out;
163
164 ret = 0;
165 if (vma->vm_flags & VM_WRITE && inode->i_size < len)
166 i_size_write(inode, len);
167 out:
168 inode_unlock(inode);
169
170 return ret;
171 }
172
173 /*
174 * Called under mmap_write_lock(mm).
175 */
176
177 unsigned long
178 hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
179 unsigned long len, unsigned long pgoff,
180 unsigned long flags)
181 {
182 unsigned long addr0 = 0;
183 struct hstate *h = hstate_file(file);
184
185 if (len & ~huge_page_mask(h))
186 return -EINVAL;
187 if (flags & MAP_FIXED) {
188 if (addr & ~huge_page_mask(h))
189 return -EINVAL;
190 if (prepare_hugepage_range(file, addr, len))
191 return -EINVAL;
192 }
193 if (addr)
194 addr0 = ALIGN(addr, huge_page_size(h));
195
196 return mm_get_unmapped_area_vmflags(current->mm, file, addr0, len, pgoff,
197 flags, 0);
198 }
199
200 /*
201 * Someone wants to read @bytes from a HWPOISON hugetlb @page from @offset.
202 * Returns the maximum number of bytes one can read without touching the 1st raw
203 * HWPOISON subpage.
204 *
205 * The implementation borrows the iteration logic from copy_page_to_iter*.
206 */
207 static size_t adjust_range_hwpoison(struct page *page, size_t offset, size_t bytes)
208 {
209 size_t n = 0;
210 size_t res = 0;
211
212 /* First subpage to start the loop. */
213 page = nth_page(page, offset / PAGE_SIZE);
214 offset %= PAGE_SIZE;
215 while (1) {
216 if (is_raw_hwpoison_page_in_hugepage(page))
217 break;
218
219 /* Safe to read n bytes without touching HWPOISON subpage. */
220 n = min(bytes, (size_t)PAGE_SIZE - offset);
221 res += n;
222 bytes -= n;
223 if (!bytes || !n)
224 break;
225 offset += n;
226 if (offset == PAGE_SIZE) {
227 page = nth_page(page, 1);
228 offset = 0;
229 }
230 }
231
232 return res;
233 }
234
235 /*
236 * Support for read() - Find the page attached to f_mapping and copy out the
237 * data. This provides functionality similar to filemap_read().
238 */
239 static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to)
240 {
241 struct file *file = iocb->ki_filp;
242 struct hstate *h = hstate_file(file);
243 struct address_space *mapping = file->f_mapping;
244 struct inode *inode = mapping->host;
245 unsigned long index = iocb->ki_pos >> huge_page_shift(h);
246 unsigned long offset = iocb->ki_pos & ~huge_page_mask(h);
247 unsigned long end_index;
248 loff_t isize;
249 ssize_t retval = 0;
250
251 while (iov_iter_count(to)) {
252 struct folio *folio;
253 size_t nr, copied, want;
254
255 /* nr is the maximum number of bytes to copy from this page */
256 nr = huge_page_size(h);
257 isize = i_size_read(inode);
258 if (!isize)
259 break;
260 end_index = (isize - 1) >> huge_page_shift(h);
261 if (index > end_index)
262 break;
263 if (index == end_index) {
264 nr = ((isize - 1) & ~huge_page_mask(h)) + 1;
265 if (nr <= offset)
266 break;
267 }
268 nr = nr - offset;
269
270 /* Find the folio */
271 folio = filemap_lock_hugetlb_folio(h, mapping, index);
272 if (IS_ERR(folio)) {
273 /*
274 * We have a HOLE, zero out the user-buffer for the
275 * length of the hole or request.
276 */
277 copied = iov_iter_zero(nr, to);
278 } else {
279 folio_unlock(folio);
280
281 if (!folio_test_hwpoison(folio))
282 want = nr;
283 else {
284 /*
285 * Adjust how many bytes safe to read without
286 * touching the 1st raw HWPOISON subpage after
287 * offset.
288 */
289 want = adjust_range_hwpoison(&folio->page, offset, nr);
290 if (want == 0) {
291 folio_put(folio);
292 retval = -EIO;
293 break;
294 }
295 }
296
297 /*
298 * We have the folio, copy it to user space buffer.
299 */
300 copied = copy_folio_to_iter(folio, offset, want, to);
301 folio_put(folio);
302 }
303 offset += copied;
304 retval += copied;
305 if (copied != nr && iov_iter_count(to)) {
306 if (!retval)
307 retval = -EFAULT;
308 break;
309 }
310 index += offset >> huge_page_shift(h);
311 offset &= ~huge_page_mask(h);
312 }
313 iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset;
314 return retval;
315 }
316
317 static int hugetlbfs_write_begin(struct file *file,
318 struct address_space *mapping,
319 loff_t pos, unsigned len,
320 struct folio **foliop, void **fsdata)
321 {
322 return -EINVAL;
323 }
324
325 static int hugetlbfs_write_end(struct file *file, struct address_space *mapping,
326 loff_t pos, unsigned len, unsigned copied,
327 struct folio *folio, void *fsdata)
328 {
329 BUG();
330 return -EINVAL;
331 }
332
333 static void hugetlb_delete_from_page_cache(struct folio *folio)
334 {
335 folio_clear_dirty(folio);
336 folio_clear_uptodate(folio);
337 filemap_remove_folio(folio);
338 }
339
340 /*
341 * Called with i_mmap_rwsem held for inode based vma maps. This makes
342 * sure vma (and vm_mm) will not go away. We also hold the hugetlb fault
343 * mutex for the page in the mapping. So, we can not race with page being
344 * faulted into the vma.
345 */
346 static bool hugetlb_vma_maps_page(struct vm_area_struct *vma,
347 unsigned long addr, struct page *page)
348 {
349 pte_t *ptep, pte;
350
351 ptep = hugetlb_walk(vma, addr, huge_page_size(hstate_vma(vma)));
352 if (!ptep)
353 return false;
354
355 pte = huge_ptep_get(vma->vm_mm, addr, ptep);
356 if (huge_pte_none(pte) || !pte_present(pte))
357 return false;
358
359 if (pte_page(pte) == page)
360 return true;
361
362 return false;
363 }
364
365 /*
366 * Can vma_offset_start/vma_offset_end overflow on 32-bit arches?
367 * No, because the interval tree returns us only those vmas
368 * which overlap the truncated area starting at pgoff,
369 * and no vma on a 32-bit arch can span beyond the 4GB.
370 */
371 static unsigned long vma_offset_start(struct vm_area_struct *vma, pgoff_t start)
372 {
373 unsigned long offset = 0;
374
375 if (vma->vm_pgoff < start)
376 offset = (start - vma->vm_pgoff) << PAGE_SHIFT;
377
378 return vma->vm_start + offset;
379 }
380
381 static unsigned long vma_offset_end(struct vm_area_struct *vma, pgoff_t end)
382 {
383 unsigned long t_end;
384
385 if (!end)
386 return vma->vm_end;
387
388 t_end = ((end - vma->vm_pgoff) << PAGE_SHIFT) + vma->vm_start;
389 if (t_end > vma->vm_end)
390 t_end = vma->vm_end;
391 return t_end;
392 }
393
394 /*
395 * Called with hugetlb fault mutex held. Therefore, no more mappings to
396 * this folio can be created while executing the routine.
397 */
398 static void hugetlb_unmap_file_folio(struct hstate *h,
399 struct address_space *mapping,
400 struct folio *folio, pgoff_t index)
401 {
402 struct rb_root_cached *root = &mapping->i_mmap;
403 struct hugetlb_vma_lock *vma_lock;
404 struct page *page = &folio->page;
405 struct vm_area_struct *vma;
406 unsigned long v_start;
407 unsigned long v_end;
408 pgoff_t start, end;
409
410 start = index * pages_per_huge_page(h);
411 end = (index + 1) * pages_per_huge_page(h);
412
413 i_mmap_lock_write(mapping);
414 retry:
415 vma_lock = NULL;
416 vma_interval_tree_foreach(vma, root, start, end - 1) {
417 v_start = vma_offset_start(vma, start);
418 v_end = vma_offset_end(vma, end);
419
420 if (!hugetlb_vma_maps_page(vma, v_start, page))
421 continue;
422
423 if (!hugetlb_vma_trylock_write(vma)) {
424 vma_lock = vma->vm_private_data;
425 /*
426 * If we can not get vma lock, we need to drop
427 * immap_sema and take locks in order. First,
428 * take a ref on the vma_lock structure so that
429 * we can be guaranteed it will not go away when
430 * dropping immap_sema.
431 */
432 kref_get(&vma_lock->refs);
433 break;
434 }
435
436 unmap_hugepage_range(vma, v_start, v_end, NULL,
437 ZAP_FLAG_DROP_MARKER);
438 hugetlb_vma_unlock_write(vma);
439 }
440
441 i_mmap_unlock_write(mapping);
442
443 if (vma_lock) {
444 /*
445 * Wait on vma_lock. We know it is still valid as we have
446 * a reference. We must 'open code' vma locking as we do
447 * not know if vma_lock is still attached to vma.
448 */
449 down_write(&vma_lock->rw_sema);
450 i_mmap_lock_write(mapping);
451
452 vma = vma_lock->vma;
453 if (!vma) {
454 /*
455 * If lock is no longer attached to vma, then just
456 * unlock, drop our reference and retry looking for
457 * other vmas.
458 */
459 up_write(&vma_lock->rw_sema);
460 kref_put(&vma_lock->refs, hugetlb_vma_lock_release);
461 goto retry;
462 }
463
464 /*
465 * vma_lock is still attached to vma. Check to see if vma
466 * still maps page and if so, unmap.
467 */
468 v_start = vma_offset_start(vma, start);
469 v_end = vma_offset_end(vma, end);
470 if (hugetlb_vma_maps_page(vma, v_start, page))
471 unmap_hugepage_range(vma, v_start, v_end, NULL,
472 ZAP_FLAG_DROP_MARKER);
473
474 kref_put(&vma_lock->refs, hugetlb_vma_lock_release);
475 hugetlb_vma_unlock_write(vma);
476
477 goto retry;
478 }
479 }
480
481 static void
482 hugetlb_vmdelete_list(struct rb_root_cached *root, pgoff_t start, pgoff_t end,
483 zap_flags_t zap_flags)
484 {
485 struct vm_area_struct *vma;
486
487 /*
488 * end == 0 indicates that the entire range after start should be
489 * unmapped. Note, end is exclusive, whereas the interval tree takes
490 * an inclusive "last".
491 */
492 vma_interval_tree_foreach(vma, root, start, end ? end - 1 : ULONG_MAX) {
493 unsigned long v_start;
494 unsigned long v_end;
495
496 if (!hugetlb_vma_trylock_write(vma))
497 continue;
498
499 v_start = vma_offset_start(vma, start);
500 v_end = vma_offset_end(vma, end);
501
502 unmap_hugepage_range(vma, v_start, v_end, NULL, zap_flags);
503
504 /*
505 * Note that vma lock only exists for shared/non-private
506 * vmas. Therefore, lock is not held when calling
507 * unmap_hugepage_range for private vmas.
508 */
509 hugetlb_vma_unlock_write(vma);
510 }
511 }
512
513 /*
514 * Called with hugetlb fault mutex held.
515 * Returns true if page was actually removed, false otherwise.
516 */
517 static bool remove_inode_single_folio(struct hstate *h, struct inode *inode,
518 struct address_space *mapping,
519 struct folio *folio, pgoff_t index,
520 bool truncate_op)
521 {
522 bool ret = false;
523
524 /*
525 * If folio is mapped, it was faulted in after being
526 * unmapped in caller. Unmap (again) while holding
527 * the fault mutex. The mutex will prevent faults
528 * until we finish removing the folio.
529 */
530 if (unlikely(folio_mapped(folio)))
531 hugetlb_unmap_file_folio(h, mapping, folio, index);
532
533 folio_lock(folio);
534 /*
535 * We must remove the folio from page cache before removing
536 * the region/ reserve map (hugetlb_unreserve_pages). In
537 * rare out of memory conditions, removal of the region/reserve
538 * map could fail. Correspondingly, the subpool and global
539 * reserve usage count can need to be adjusted.
540 */
541 VM_BUG_ON_FOLIO(folio_test_hugetlb_restore_reserve(folio), folio);
542 hugetlb_delete_from_page_cache(folio);
543 ret = true;
544 if (!truncate_op) {
545 if (unlikely(hugetlb_unreserve_pages(inode, index,
546 index + 1, 1)))
547 hugetlb_fix_reserve_counts(inode);
548 }
549
550 folio_unlock(folio);
551 return ret;
552 }
553
554 /*
555 * remove_inode_hugepages handles two distinct cases: truncation and hole
556 * punch. There are subtle differences in operation for each case.
557 *
558 * truncation is indicated by end of range being LLONG_MAX
559 * In this case, we first scan the range and release found pages.
560 * After releasing pages, hugetlb_unreserve_pages cleans up region/reserve
561 * maps and global counts. Page faults can race with truncation.
562 * During faults, hugetlb_no_page() checks i_size before page allocation,
563 * and again after obtaining page table lock. It will 'back out'
564 * allocations in the truncated range.
565 * hole punch is indicated if end is not LLONG_MAX
566 * In the hole punch case we scan the range and release found pages.
567 * Only when releasing a page is the associated region/reserve map
568 * deleted. The region/reserve map for ranges without associated
569 * pages are not modified. Page faults can race with hole punch.
570 * This is indicated if we find a mapped page.
571 * Note: If the passed end of range value is beyond the end of file, but
572 * not LLONG_MAX this routine still performs a hole punch operation.
573 */
574 static void remove_inode_hugepages(struct inode *inode, loff_t lstart,
575 loff_t lend)
576 {
577 struct hstate *h = hstate_inode(inode);
578 struct address_space *mapping = &inode->i_data;
579 const pgoff_t end = lend >> PAGE_SHIFT;
580 struct folio_batch fbatch;
581 pgoff_t next, index;
582 int i, freed = 0;
583 bool truncate_op = (lend == LLONG_MAX);
584
585 folio_batch_init(&fbatch);
586 next = lstart >> PAGE_SHIFT;
587 while (filemap_get_folios(mapping, &next, end - 1, &fbatch)) {
588 for (i = 0; i < folio_batch_count(&fbatch); ++i) {
589 struct folio *folio = fbatch.folios[i];
590 u32 hash = 0;
591
592 index = folio->index >> huge_page_order(h);
593 hash = hugetlb_fault_mutex_hash(mapping, index);
594 mutex_lock(&hugetlb_fault_mutex_table[hash]);
595
596 /*
597 * Remove folio that was part of folio_batch.
598 */
599 if (remove_inode_single_folio(h, inode, mapping, folio,
600 index, truncate_op))
601 freed++;
602
603 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
604 }
605 folio_batch_release(&fbatch);
606 cond_resched();
607 }
608
609 if (truncate_op)
610 (void)hugetlb_unreserve_pages(inode,
611 lstart >> huge_page_shift(h),
612 LONG_MAX, freed);
613 }
614
615 static void hugetlbfs_evict_inode(struct inode *inode)
616 {
617 struct resv_map *resv_map;
618
619 trace_hugetlbfs_evict_inode(inode);
620 remove_inode_hugepages(inode, 0, LLONG_MAX);
621
622 /*
623 * Get the resv_map from the address space embedded in the inode.
624 * This is the address space which points to any resv_map allocated
625 * at inode creation time. If this is a device special inode,
626 * i_mapping may not point to the original address space.
627 */
628 resv_map = (struct resv_map *)(&inode->i_data)->i_private_data;
629 /* Only regular and link inodes have associated reserve maps */
630 if (resv_map)
631 resv_map_release(&resv_map->refs);
632 clear_inode(inode);
633 }
634
635 static void hugetlb_vmtruncate(struct inode *inode, loff_t offset)
636 {
637 pgoff_t pgoff;
638 struct address_space *mapping = inode->i_mapping;
639 struct hstate *h = hstate_inode(inode);
640
641 BUG_ON(offset & ~huge_page_mask(h));
642 pgoff = offset >> PAGE_SHIFT;
643
644 i_size_write(inode, offset);
645 i_mmap_lock_write(mapping);
646 if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))
647 hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0,
648 ZAP_FLAG_DROP_MARKER);
649 i_mmap_unlock_write(mapping);
650 remove_inode_hugepages(inode, offset, LLONG_MAX);
651 }
652
653 static void hugetlbfs_zero_partial_page(struct hstate *h,
654 struct address_space *mapping,
655 loff_t start,
656 loff_t end)
657 {
658 pgoff_t idx = start >> huge_page_shift(h);
659 struct folio *folio;
660
661 folio = filemap_lock_hugetlb_folio(h, mapping, idx);
662 if (IS_ERR(folio))
663 return;
664
665 start = start & ~huge_page_mask(h);
666 end = end & ~huge_page_mask(h);
667 if (!end)
668 end = huge_page_size(h);
669
670 folio_zero_segment(folio, (size_t)start, (size_t)end);
671
672 folio_unlock(folio);
673 folio_put(folio);
674 }
675
676 static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len)
677 {
678 struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
679 struct address_space *mapping = inode->i_mapping;
680 struct hstate *h = hstate_inode(inode);
681 loff_t hpage_size = huge_page_size(h);
682 loff_t hole_start, hole_end;
683
684 /*
685 * hole_start and hole_end indicate the full pages within the hole.
686 */
687 hole_start = round_up(offset, hpage_size);
688 hole_end = round_down(offset + len, hpage_size);
689
690 inode_lock(inode);
691
692 /* protected by i_rwsem */
693 if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) {
694 inode_unlock(inode);
695 return -EPERM;
696 }
697
698 i_mmap_lock_write(mapping);
699
700 /* If range starts before first full page, zero partial page. */
701 if (offset < hole_start)
702 hugetlbfs_zero_partial_page(h, mapping,
703 offset, min(offset + len, hole_start));
704
705 /* Unmap users of full pages in the hole. */
706 if (hole_end > hole_start) {
707 if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root))
708 hugetlb_vmdelete_list(&mapping->i_mmap,
709 hole_start >> PAGE_SHIFT,
710 hole_end >> PAGE_SHIFT, 0);
711 }
712
713 /* If range extends beyond last full page, zero partial page. */
714 if ((offset + len) > hole_end && (offset + len) > hole_start)
715 hugetlbfs_zero_partial_page(h, mapping,
716 hole_end, offset + len);
717
718 i_mmap_unlock_write(mapping);
719
720 /* Remove full pages from the file. */
721 if (hole_end > hole_start)
722 remove_inode_hugepages(inode, hole_start, hole_end);
723
724 inode_unlock(inode);
725
726 return 0;
727 }
728
729 static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset,
730 loff_t len)
731 {
732 struct inode *inode = file_inode(file);
733 struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
734 struct address_space *mapping = inode->i_mapping;
735 struct hstate *h = hstate_inode(inode);
736 struct vm_area_struct pseudo_vma;
737 struct mm_struct *mm = current->mm;
738 loff_t hpage_size = huge_page_size(h);
739 unsigned long hpage_shift = huge_page_shift(h);
740 pgoff_t start, index, end;
741 int error;
742 u32 hash;
743
744 if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE))
745 return -EOPNOTSUPP;
746
747 if (mode & FALLOC_FL_PUNCH_HOLE) {
748 error = hugetlbfs_punch_hole(inode, offset, len);
749 goto out_nolock;
750 }
751
752 /*
753 * Default preallocate case.
754 * For this range, start is rounded down and end is rounded up
755 * as well as being converted to page offsets.
756 */
757 start = offset >> hpage_shift;
758 end = (offset + len + hpage_size - 1) >> hpage_shift;
759
760 inode_lock(inode);
761
762 /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */
763 error = inode_newsize_ok(inode, offset + len);
764 if (error)
765 goto out;
766
767 if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) {
768 error = -EPERM;
769 goto out;
770 }
771
772 /*
773 * Initialize a pseudo vma as this is required by the huge page
774 * allocation routines.
775 */
776 vma_init(&pseudo_vma, mm);
777 vm_flags_init(&pseudo_vma, VM_HUGETLB | VM_MAYSHARE | VM_SHARED);
778 pseudo_vma.vm_file = file;
779
780 for (index = start; index < end; index++) {
781 /*
782 * This is supposed to be the vaddr where the page is being
783 * faulted in, but we have no vaddr here.
784 */
785 struct folio *folio;
786 unsigned long addr;
787
788 cond_resched();
789
790 /*
791 * fallocate(2) manpage permits EINTR; we may have been
792 * interrupted because we are using up too much memory.
793 */
794 if (signal_pending(current)) {
795 error = -EINTR;
796 break;
797 }
798
799 /* addr is the offset within the file (zero based) */
800 addr = index * hpage_size;
801
802 /* mutex taken here, fault path and hole punch */
803 hash = hugetlb_fault_mutex_hash(mapping, index);
804 mutex_lock(&hugetlb_fault_mutex_table[hash]);
805
806 /* See if already present in mapping to avoid alloc/free */
807 folio = filemap_get_folio(mapping, index << huge_page_order(h));
808 if (!IS_ERR(folio)) {
809 folio_put(folio);
810 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
811 continue;
812 }
813
814 /*
815 * Allocate folio without setting the avoid_reserve argument.
816 * There certainly are no reserves associated with the
817 * pseudo_vma. However, there could be shared mappings with
818 * reserves for the file at the inode level. If we fallocate
819 * folios in these areas, we need to consume the reserves
820 * to keep reservation accounting consistent.
821 */
822 folio = alloc_hugetlb_folio(&pseudo_vma, addr, 0);
823 if (IS_ERR(folio)) {
824 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
825 error = PTR_ERR(folio);
826 goto out;
827 }
828 folio_zero_user(folio, addr);
829 __folio_mark_uptodate(folio);
830 error = hugetlb_add_to_page_cache(folio, mapping, index);
831 if (unlikely(error)) {
832 restore_reserve_on_error(h, &pseudo_vma, addr, folio);
833 folio_put(folio);
834 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
835 goto out;
836 }
837
838 mutex_unlock(&hugetlb_fault_mutex_table[hash]);
839
840 folio_set_hugetlb_migratable(folio);
841 /*
842 * folio_unlock because locked by hugetlb_add_to_page_cache()
843 * folio_put() due to reference from alloc_hugetlb_folio()
844 */
845 folio_unlock(folio);
846 folio_put(folio);
847 }
848
849 if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size)
850 i_size_write(inode, offset + len);
851 inode_set_ctime_current(inode);
852 out:
853 inode_unlock(inode);
854
855 out_nolock:
856 trace_hugetlbfs_fallocate(inode, mode, offset, len, error);
857 return error;
858 }
859
860 static int hugetlbfs_setattr(struct mnt_idmap *idmap,
861 struct dentry *dentry, struct iattr *attr)
862 {
863 struct inode *inode = d_inode(dentry);
864 struct hstate *h = hstate_inode(inode);
865 int error;
866 unsigned int ia_valid = attr->ia_valid;
867 struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
868
869 error = setattr_prepare(idmap, dentry, attr);
870 if (error)
871 return error;
872
873 trace_hugetlbfs_setattr(inode, dentry, attr);
874
875 if (ia_valid & ATTR_SIZE) {
876 loff_t oldsize = inode->i_size;
877 loff_t newsize = attr->ia_size;
878
879 if (newsize & ~huge_page_mask(h))
880 return -EINVAL;
881 /* protected by i_rwsem */
882 if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) ||
883 (newsize > oldsize && (info->seals & F_SEAL_GROW)))
884 return -EPERM;
885 hugetlb_vmtruncate(inode, newsize);
886 }
887
888 setattr_copy(idmap, inode, attr);
889 mark_inode_dirty(inode);
890 return 0;
891 }
892
893 static struct inode *hugetlbfs_get_root(struct super_block *sb,
894 struct hugetlbfs_fs_context *ctx)
895 {
896 struct inode *inode;
897
898 inode = new_inode(sb);
899 if (inode) {
900 inode->i_ino = get_next_ino();
901 inode->i_mode = S_IFDIR | ctx->mode;
902 inode->i_uid = ctx->uid;
903 inode->i_gid = ctx->gid;
904 simple_inode_init_ts(inode);
905 inode->i_op = &hugetlbfs_dir_inode_operations;
906 inode->i_fop = &simple_dir_operations;
907 /* directory inodes start off with i_nlink == 2 (for "." entry) */
908 inc_nlink(inode);
909 lockdep_annotate_inode_mutex_key(inode);
910 }
911 return inode;
912 }
913
914 /*
915 * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never
916 * be taken from reclaim -- unlike regular filesystems. This needs an
917 * annotation because huge_pmd_share() does an allocation under hugetlb's
918 * i_mmap_rwsem.
919 */
920 static struct lock_class_key hugetlbfs_i_mmap_rwsem_key;
921
922 static struct inode *hugetlbfs_get_inode(struct super_block *sb,
923 struct mnt_idmap *idmap,
924 struct inode *dir,
925 umode_t mode, dev_t dev)
926 {
927 struct inode *inode;
928 struct resv_map *resv_map = NULL;
929
930 /*
931 * Reserve maps are only needed for inodes that can have associated
932 * page allocations.
933 */
934 if (S_ISREG(mode) || S_ISLNK(mode)) {
935 resv_map = resv_map_alloc();
936 if (!resv_map)
937 return NULL;
938 }
939
940 inode = new_inode(sb);
941 if (inode) {
942 struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode);
943
944 inode->i_ino = get_next_ino();
945 inode_init_owner(idmap, inode, dir, mode);
946 lockdep_set_class(&inode->i_mapping->i_mmap_rwsem,
947 &hugetlbfs_i_mmap_rwsem_key);
948 inode->i_mapping->a_ops = &hugetlbfs_aops;
949 simple_inode_init_ts(inode);
950 inode->i_mapping->i_private_data = resv_map;
951 info->seals = F_SEAL_SEAL;
952 switch (mode & S_IFMT) {
953 default:
954 init_special_inode(inode, mode, dev);
955 break;
956 case S_IFREG:
957 inode->i_op = &hugetlbfs_inode_operations;
958 inode->i_fop = &hugetlbfs_file_operations;
959 break;
960 case S_IFDIR:
961 inode->i_op = &hugetlbfs_dir_inode_operations;
962 inode->i_fop = &simple_dir_operations;
963
964 /* directory inodes start off with i_nlink == 2 (for "." entry) */
965 inc_nlink(inode);
966 break;
967 case S_IFLNK:
968 inode->i_op = &page_symlink_inode_operations;
969 inode_nohighmem(inode);
970 break;
971 }
972 lockdep_annotate_inode_mutex_key(inode);
973 trace_hugetlbfs_alloc_inode(inode, dir, mode);
974 } else {
975 if (resv_map)
976 kref_put(&resv_map->refs, resv_map_release);
977 }
978
979 return inode;
980 }
981
982 /*
983 * File creation. Allocate an inode, and we're done..
984 */
985 static int hugetlbfs_mknod(struct mnt_idmap *idmap, struct inode *dir,
986 struct dentry *dentry, umode_t mode, dev_t dev)
987 {
988 struct inode *inode;
989
990 inode = hugetlbfs_get_inode(dir->i_sb, idmap, dir, mode, dev);
991 if (!inode)
992 return -ENOSPC;
993 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
994 d_instantiate(dentry, inode);
995 dget(dentry);/* Extra count - pin the dentry in core */
996 return 0;
997 }
998
999 static int hugetlbfs_mkdir(struct mnt_idmap *idmap, struct inode *dir,
1000 struct dentry *dentry, umode_t mode)
1001 {
1002 int retval = hugetlbfs_mknod(idmap, dir, dentry,
1003 mode | S_IFDIR, 0);
1004 if (!retval)
1005 inc_nlink(dir);
1006 return retval;
1007 }
1008
1009 static int hugetlbfs_create(struct mnt_idmap *idmap,
1010 struct inode *dir, struct dentry *dentry,
1011 umode_t mode, bool excl)
1012 {
1013 return hugetlbfs_mknod(idmap, dir, dentry, mode | S_IFREG, 0);
1014 }
1015
1016 static int hugetlbfs_tmpfile(struct mnt_idmap *idmap,
1017 struct inode *dir, struct file *file,
1018 umode_t mode)
1019 {
1020 struct inode *inode;
1021
1022 inode = hugetlbfs_get_inode(dir->i_sb, idmap, dir, mode | S_IFREG, 0);
1023 if (!inode)
1024 return -ENOSPC;
1025 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
1026 d_tmpfile(file, inode);
1027 return finish_open_simple(file, 0);
1028 }
1029
1030 static int hugetlbfs_symlink(struct mnt_idmap *idmap,
1031 struct inode *dir, struct dentry *dentry,
1032 const char *symname)
1033 {
1034 const umode_t mode = S_IFLNK|S_IRWXUGO;
1035 struct inode *inode;
1036 int error = -ENOSPC;
1037
1038 inode = hugetlbfs_get_inode(dir->i_sb, idmap, dir, mode, 0);
1039 if (inode) {
1040 int l = strlen(symname)+1;
1041 error = page_symlink(inode, symname, l);
1042 if (!error) {
1043 d_instantiate(dentry, inode);
1044 dget(dentry);
1045 } else
1046 iput(inode);
1047 }
1048 inode_set_mtime_to_ts(dir, inode_set_ctime_current(dir));
1049
1050 return error;
1051 }
1052
1053 #ifdef CONFIG_MIGRATION
1054 static int hugetlbfs_migrate_folio(struct address_space *mapping,
1055 struct folio *dst, struct folio *src,
1056 enum migrate_mode mode)
1057 {
1058 int rc;
1059
1060 rc = migrate_huge_page_move_mapping(mapping, dst, src);
1061 if (rc != MIGRATEPAGE_SUCCESS)
1062 return rc;
1063
1064 if (hugetlb_folio_subpool(src)) {
1065 hugetlb_set_folio_subpool(dst,
1066 hugetlb_folio_subpool(src));
1067 hugetlb_set_folio_subpool(src, NULL);
1068 }
1069
1070 folio_migrate_flags(dst, src);
1071
1072 return MIGRATEPAGE_SUCCESS;
1073 }
1074 #else
1075 #define hugetlbfs_migrate_folio NULL
1076 #endif
1077
1078 static int hugetlbfs_error_remove_folio(struct address_space *mapping,
1079 struct folio *folio)
1080 {
1081 return 0;
1082 }
1083
1084 /*
1085 * Display the mount options in /proc/mounts.
1086 */
1087 static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root)
1088 {
1089 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(root->d_sb);
1090 struct hugepage_subpool *spool = sbinfo->spool;
1091 unsigned long hpage_size = huge_page_size(sbinfo->hstate);
1092 unsigned hpage_shift = huge_page_shift(sbinfo->hstate);
1093 char mod;
1094
1095 if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID))
1096 seq_printf(m, ",uid=%u",
1097 from_kuid_munged(&init_user_ns, sbinfo->uid));
1098 if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID))
1099 seq_printf(m, ",gid=%u",
1100 from_kgid_munged(&init_user_ns, sbinfo->gid));
1101 if (sbinfo->mode != 0755)
1102 seq_printf(m, ",mode=%o", sbinfo->mode);
1103 if (sbinfo->max_inodes != -1)
1104 seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes);
1105
1106 hpage_size /= 1024;
1107 mod = 'K';
1108 if (hpage_size >= 1024) {
1109 hpage_size /= 1024;
1110 mod = 'M';
1111 }
1112 seq_printf(m, ",pagesize=%lu%c", hpage_size, mod);
1113 if (spool) {
1114 if (spool->max_hpages != -1)
1115 seq_printf(m, ",size=%llu",
1116 (unsigned long long)spool->max_hpages << hpage_shift);
1117 if (spool->min_hpages != -1)
1118 seq_printf(m, ",min_size=%llu",
1119 (unsigned long long)spool->min_hpages << hpage_shift);
1120 }
1121 return 0;
1122 }
1123
1124 static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf)
1125 {
1126 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb);
1127 struct hstate *h = hstate_inode(d_inode(dentry));
1128 u64 id = huge_encode_dev(dentry->d_sb->s_dev);
1129
1130 buf->f_fsid = u64_to_fsid(id);
1131 buf->f_type = HUGETLBFS_MAGIC;
1132 buf->f_bsize = huge_page_size(h);
1133 if (sbinfo) {
1134 spin_lock(&sbinfo->stat_lock);
1135 /* If no limits set, just report 0 or -1 for max/free/used
1136 * blocks, like simple_statfs() */
1137 if (sbinfo->spool) {
1138 long free_pages;
1139
1140 spin_lock_irq(&sbinfo->spool->lock);
1141 buf->f_blocks = sbinfo->spool->max_hpages;
1142 free_pages = sbinfo->spool->max_hpages
1143 - sbinfo->spool->used_hpages;
1144 buf->f_bavail = buf->f_bfree = free_pages;
1145 spin_unlock_irq(&sbinfo->spool->lock);
1146 buf->f_files = sbinfo->max_inodes;
1147 buf->f_ffree = sbinfo->free_inodes;
1148 }
1149 spin_unlock(&sbinfo->stat_lock);
1150 }
1151 buf->f_namelen = NAME_MAX;
1152 return 0;
1153 }
1154
1155 static void hugetlbfs_put_super(struct super_block *sb)
1156 {
1157 struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb);
1158
1159 if (sbi) {
1160 sb->s_fs_info = NULL;
1161
1162 if (sbi->spool)
1163 hugepage_put_subpool(sbi->spool);
1164
1165 kfree(sbi);
1166 }
1167 }
1168
1169 static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo)
1170 {
1171 if (sbinfo->free_inodes >= 0) {
1172 spin_lock(&sbinfo->stat_lock);
1173 if (unlikely(!sbinfo->free_inodes)) {
1174 spin_unlock(&sbinfo->stat_lock);
1175 return 0;
1176 }
1177 sbinfo->free_inodes--;
1178 spin_unlock(&sbinfo->stat_lock);
1179 }
1180
1181 return 1;
1182 }
1183
1184 static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo)
1185 {
1186 if (sbinfo->free_inodes >= 0) {
1187 spin_lock(&sbinfo->stat_lock);
1188 sbinfo->free_inodes++;
1189 spin_unlock(&sbinfo->stat_lock);
1190 }
1191 }
1192
1193
1194 static struct kmem_cache *hugetlbfs_inode_cachep;
1195
1196 static struct inode *hugetlbfs_alloc_inode(struct super_block *sb)
1197 {
1198 struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb);
1199 struct hugetlbfs_inode_info *p;
1200
1201 if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo)))
1202 return NULL;
1203 p = alloc_inode_sb(sb, hugetlbfs_inode_cachep, GFP_KERNEL);
1204 if (unlikely(!p)) {
1205 hugetlbfs_inc_free_inodes(sbinfo);
1206 return NULL;
1207 }
1208 return &p->vfs_inode;
1209 }
1210
1211 static void hugetlbfs_free_inode(struct inode *inode)
1212 {
1213 trace_hugetlbfs_free_inode(inode);
1214 kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode));
1215 }
1216
1217 static void hugetlbfs_destroy_inode(struct inode *inode)
1218 {
1219 hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb));
1220 }
1221
1222 static const struct address_space_operations hugetlbfs_aops = {
1223 .write_begin = hugetlbfs_write_begin,
1224 .write_end = hugetlbfs_write_end,
1225 .dirty_folio = noop_dirty_folio,
1226 .migrate_folio = hugetlbfs_migrate_folio,
1227 .error_remove_folio = hugetlbfs_error_remove_folio,
1228 };
1229
1230
1231 static void init_once(void *foo)
1232 {
1233 struct hugetlbfs_inode_info *ei = foo;
1234
1235 inode_init_once(&ei->vfs_inode);
1236 }
1237
1238 static const struct file_operations hugetlbfs_file_operations = {
1239 .read_iter = hugetlbfs_read_iter,
1240 .mmap = hugetlbfs_file_mmap,
1241 .fsync = noop_fsync,
1242 .get_unmapped_area = hugetlb_get_unmapped_area,
1243 .llseek = default_llseek,
1244 .fallocate = hugetlbfs_fallocate,
1245 .fop_flags = FOP_HUGE_PAGES,
1246 };
1247
1248 static const struct inode_operations hugetlbfs_dir_inode_operations = {
1249 .create = hugetlbfs_create,
1250 .lookup = simple_lookup,
1251 .link = simple_link,
1252 .unlink = simple_unlink,
1253 .symlink = hugetlbfs_symlink,
1254 .mkdir = hugetlbfs_mkdir,
1255 .rmdir = simple_rmdir,
1256 .mknod = hugetlbfs_mknod,
1257 .rename = simple_rename,
1258 .setattr = hugetlbfs_setattr,
1259 .tmpfile = hugetlbfs_tmpfile,
1260 };
1261
1262 static const struct inode_operations hugetlbfs_inode_operations = {
1263 .setattr = hugetlbfs_setattr,
1264 };
1265
1266 static const struct super_operations hugetlbfs_ops = {
1267 .alloc_inode = hugetlbfs_alloc_inode,
1268 .free_inode = hugetlbfs_free_inode,
1269 .destroy_inode = hugetlbfs_destroy_inode,
1270 .evict_inode = hugetlbfs_evict_inode,
1271 .statfs = hugetlbfs_statfs,
1272 .put_super = hugetlbfs_put_super,
1273 .show_options = hugetlbfs_show_options,
1274 };
1275
1276 /*
1277 * Convert size option passed from command line to number of huge pages
1278 * in the pool specified by hstate. Size option could be in bytes
1279 * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT).
1280 */
1281 static long
1282 hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt,
1283 enum hugetlbfs_size_type val_type)
1284 {
1285 if (val_type == NO_SIZE)
1286 return -1;
1287
1288 if (val_type == SIZE_PERCENT) {
1289 size_opt <<= huge_page_shift(h);
1290 size_opt *= h->max_huge_pages;
1291 do_div(size_opt, 100);
1292 }
1293
1294 size_opt >>= huge_page_shift(h);
1295 return size_opt;
1296 }
1297
1298 /*
1299 * Parse one mount parameter.
1300 */
1301 static int hugetlbfs_parse_param(struct fs_context *fc, struct fs_parameter *param)
1302 {
1303 struct hugetlbfs_fs_context *ctx = fc->fs_private;
1304 struct fs_parse_result result;
1305 struct hstate *h;
1306 char *rest;
1307 unsigned long ps;
1308 int opt;
1309
1310 opt = fs_parse(fc, hugetlb_fs_parameters, param, &result);
1311 if (opt < 0)
1312 return opt;
1313
1314 switch (opt) {
1315 case Opt_uid:
1316 ctx->uid = result.uid;
1317 return 0;
1318
1319 case Opt_gid:
1320 ctx->gid = result.gid;
1321 return 0;
1322
1323 case Opt_mode:
1324 ctx->mode = result.uint_32 & 01777U;
1325 return 0;
1326
1327 case Opt_size:
1328 /* memparse() will accept a K/M/G without a digit */
1329 if (!param->string || !isdigit(param->string[0]))
1330 goto bad_val;
1331 ctx->max_size_opt = memparse(param->string, &rest);
1332 ctx->max_val_type = SIZE_STD;
1333 if (*rest == '%')
1334 ctx->max_val_type = SIZE_PERCENT;
1335 return 0;
1336
1337 case Opt_nr_inodes:
1338 /* memparse() will accept a K/M/G without a digit */
1339 if (!param->string || !isdigit(param->string[0]))
1340 goto bad_val;
1341 ctx->nr_inodes = memparse(param->string, &rest);
1342 return 0;
1343
1344 case Opt_pagesize:
1345 ps = memparse(param->string, &rest);
1346 h = size_to_hstate(ps);
1347 if (!h) {
1348 pr_err("Unsupported page size %lu MB\n", ps / SZ_1M);
1349 return -EINVAL;
1350 }
1351 ctx->hstate = h;
1352 return 0;
1353
1354 case Opt_min_size:
1355 /* memparse() will accept a K/M/G without a digit */
1356 if (!param->string || !isdigit(param->string[0]))
1357 goto bad_val;
1358 ctx->min_size_opt = memparse(param->string, &rest);
1359 ctx->min_val_type = SIZE_STD;
1360 if (*rest == '%')
1361 ctx->min_val_type = SIZE_PERCENT;
1362 return 0;
1363
1364 default:
1365 return -EINVAL;
1366 }
1367
1368 bad_val:
1369 return invalfc(fc, "Bad value '%s' for mount option '%s'\n",
1370 param->string, param->key);
1371 }
1372
1373 /*
1374 * Validate the parsed options.
1375 */
1376 static int hugetlbfs_validate(struct fs_context *fc)
1377 {
1378 struct hugetlbfs_fs_context *ctx = fc->fs_private;
1379
1380 /*
1381 * Use huge page pool size (in hstate) to convert the size
1382 * options to number of huge pages. If NO_SIZE, -1 is returned.
1383 */
1384 ctx->max_hpages = hugetlbfs_size_to_hpages(ctx->hstate,
1385 ctx->max_size_opt,
1386 ctx->max_val_type);
1387 ctx->min_hpages = hugetlbfs_size_to_hpages(ctx->hstate,
1388 ctx->min_size_opt,
1389 ctx->min_val_type);
1390
1391 /*
1392 * If max_size was specified, then min_size must be smaller
1393 */
1394 if (ctx->max_val_type > NO_SIZE &&
1395 ctx->min_hpages > ctx->max_hpages) {
1396 pr_err("Minimum size can not be greater than maximum size\n");
1397 return -EINVAL;
1398 }
1399
1400 return 0;
1401 }
1402
1403 static int
1404 hugetlbfs_fill_super(struct super_block *sb, struct fs_context *fc)
1405 {
1406 struct hugetlbfs_fs_context *ctx = fc->fs_private;
1407 struct hugetlbfs_sb_info *sbinfo;
1408
1409 sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL);
1410 if (!sbinfo)
1411 return -ENOMEM;
1412 sb->s_fs_info = sbinfo;
1413 spin_lock_init(&sbinfo->stat_lock);
1414 sbinfo->hstate = ctx->hstate;
1415 sbinfo->max_inodes = ctx->nr_inodes;
1416 sbinfo->free_inodes = ctx->nr_inodes;
1417 sbinfo->spool = NULL;
1418 sbinfo->uid = ctx->uid;
1419 sbinfo->gid = ctx->gid;
1420 sbinfo->mode = ctx->mode;
1421
1422 /*
1423 * Allocate and initialize subpool if maximum or minimum size is
1424 * specified. Any needed reservations (for minimum size) are taken
1425 * when the subpool is created.
1426 */
1427 if (ctx->max_hpages != -1 || ctx->min_hpages != -1) {
1428 sbinfo->spool = hugepage_new_subpool(ctx->hstate,
1429 ctx->max_hpages,
1430 ctx->min_hpages);
1431 if (!sbinfo->spool)
1432 goto out_free;
1433 }
1434 sb->s_maxbytes = MAX_LFS_FILESIZE;
1435 sb->s_blocksize = huge_page_size(ctx->hstate);
1436 sb->s_blocksize_bits = huge_page_shift(ctx->hstate);
1437 sb->s_magic = HUGETLBFS_MAGIC;
1438 sb->s_op = &hugetlbfs_ops;
1439 sb->s_time_gran = 1;
1440
1441 /*
1442 * Due to the special and limited functionality of hugetlbfs, it does
1443 * not work well as a stacking filesystem.
1444 */
1445 sb->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH;
1446 sb->s_root = d_make_root(hugetlbfs_get_root(sb, ctx));
1447 if (!sb->s_root)
1448 goto out_free;
1449 return 0;
1450 out_free:
1451 kfree(sbinfo->spool);
1452 kfree(sbinfo);
1453 return -ENOMEM;
1454 }
1455
1456 static int hugetlbfs_get_tree(struct fs_context *fc)
1457 {
1458 int err = hugetlbfs_validate(fc);
1459 if (err)
1460 return err;
1461 return get_tree_nodev(fc, hugetlbfs_fill_super);
1462 }
1463
1464 static void hugetlbfs_fs_context_free(struct fs_context *fc)
1465 {
1466 kfree(fc->fs_private);
1467 }
1468
1469 static const struct fs_context_operations hugetlbfs_fs_context_ops = {
1470 .free = hugetlbfs_fs_context_free,
1471 .parse_param = hugetlbfs_parse_param,
1472 .get_tree = hugetlbfs_get_tree,
1473 };
1474
1475 static int hugetlbfs_init_fs_context(struct fs_context *fc)
1476 {
1477 struct hugetlbfs_fs_context *ctx;
1478
1479 ctx = kzalloc(sizeof(struct hugetlbfs_fs_context), GFP_KERNEL);
1480 if (!ctx)
1481 return -ENOMEM;
1482
1483 ctx->max_hpages = -1; /* No limit on size by default */
1484 ctx->nr_inodes = -1; /* No limit on number of inodes by default */
1485 ctx->uid = current_fsuid();
1486 ctx->gid = current_fsgid();
1487 ctx->mode = 0755;
1488 ctx->hstate = &default_hstate;
1489 ctx->min_hpages = -1; /* No default minimum size */
1490 ctx->max_val_type = NO_SIZE;
1491 ctx->min_val_type = NO_SIZE;
1492 fc->fs_private = ctx;
1493 fc->ops = &hugetlbfs_fs_context_ops;
1494 return 0;
1495 }
1496
1497 static struct file_system_type hugetlbfs_fs_type = {
1498 .name = "hugetlbfs",
1499 .init_fs_context = hugetlbfs_init_fs_context,
1500 .parameters = hugetlb_fs_parameters,
1501 .kill_sb = kill_litter_super,
1502 .fs_flags = FS_ALLOW_IDMAP,
1503 };
1504
1505 static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE];
1506
1507 static int can_do_hugetlb_shm(void)
1508 {
1509 kgid_t shm_group;
1510 shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group);
1511 return capable(CAP_IPC_LOCK) || in_group_p(shm_group);
1512 }
1513
1514 static int get_hstate_idx(int page_size_log)
1515 {
1516 struct hstate *h = hstate_sizelog(page_size_log);
1517
1518 if (!h)
1519 return -1;
1520 return hstate_index(h);
1521 }
1522
1523 /*
1524 * Note that size should be aligned to proper hugepage size in caller side,
1525 * otherwise hugetlb_reserve_pages reserves one less hugepages than intended.
1526 */
1527 struct file *hugetlb_file_setup(const char *name, size_t size,
1528 vm_flags_t acctflag, int creat_flags,
1529 int page_size_log)
1530 {
1531 struct inode *inode;
1532 struct vfsmount *mnt;
1533 int hstate_idx;
1534 struct file *file;
1535
1536 hstate_idx = get_hstate_idx(page_size_log);
1537 if (hstate_idx < 0)
1538 return ERR_PTR(-ENODEV);
1539
1540 mnt = hugetlbfs_vfsmount[hstate_idx];
1541 if (!mnt)
1542 return ERR_PTR(-ENOENT);
1543
1544 if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) {
1545 struct ucounts *ucounts = current_ucounts();
1546
1547 if (user_shm_lock(size, ucounts)) {
1548 pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is obsolete\n",
1549 current->comm, current->pid);
1550 user_shm_unlock(size, ucounts);
1551 }
1552 return ERR_PTR(-EPERM);
1553 }
1554
1555 file = ERR_PTR(-ENOSPC);
1556 /* hugetlbfs_vfsmount[] mounts do not use idmapped mounts. */
1557 inode = hugetlbfs_get_inode(mnt->mnt_sb, &nop_mnt_idmap, NULL,
1558 S_IFREG | S_IRWXUGO, 0);
1559 if (!inode)
1560 goto out;
1561 if (creat_flags == HUGETLB_SHMFS_INODE)
1562 inode->i_flags |= S_PRIVATE;
1563
1564 inode->i_size = size;
1565 clear_nlink(inode);
1566
1567 if (!hugetlb_reserve_pages(inode, 0,
1568 size >> huge_page_shift(hstate_inode(inode)), NULL,
1569 acctflag))
1570 file = ERR_PTR(-ENOMEM);
1571 else
1572 file = alloc_file_pseudo(inode, mnt, name, O_RDWR,
1573 &hugetlbfs_file_operations);
1574 if (!IS_ERR(file))
1575 return file;
1576
1577 iput(inode);
1578 out:
1579 return file;
1580 }
1581
1582 static struct vfsmount *__init mount_one_hugetlbfs(struct hstate *h)
1583 {
1584 struct fs_context *fc;
1585 struct vfsmount *mnt;
1586
1587 fc = fs_context_for_mount(&hugetlbfs_fs_type, SB_KERNMOUNT);
1588 if (IS_ERR(fc)) {
1589 mnt = ERR_CAST(fc);
1590 } else {
1591 struct hugetlbfs_fs_context *ctx = fc->fs_private;
1592 ctx->hstate = h;
1593 mnt = fc_mount(fc);
1594 put_fs_context(fc);
1595 }
1596 if (IS_ERR(mnt))
1597 pr_err("Cannot mount internal hugetlbfs for page size %luK",
1598 huge_page_size(h) / SZ_1K);
1599 return mnt;
1600 }
1601
1602 static int __init init_hugetlbfs_fs(void)
1603 {
1604 struct vfsmount *mnt;
1605 struct hstate *h;
1606 int error;
1607 int i;
1608
1609 if (!hugepages_supported()) {
1610 pr_info("disabling because there are no supported hugepage sizes\n");
1611 return -ENOTSUPP;
1612 }
1613
1614 error = -ENOMEM;
1615 hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache",
1616 sizeof(struct hugetlbfs_inode_info),
1617 0, SLAB_ACCOUNT, init_once);
1618 if (hugetlbfs_inode_cachep == NULL)
1619 goto out;
1620
1621 error = register_filesystem(&hugetlbfs_fs_type);
1622 if (error)
1623 goto out_free;
1624
1625 /* default hstate mount is required */
1626 mnt = mount_one_hugetlbfs(&default_hstate);
1627 if (IS_ERR(mnt)) {
1628 error = PTR_ERR(mnt);
1629 goto out_unreg;
1630 }
1631 hugetlbfs_vfsmount[default_hstate_idx] = mnt;
1632
1633 /* other hstates are optional */
1634 i = 0;
1635 for_each_hstate(h) {
1636 if (i == default_hstate_idx) {
1637 i++;
1638 continue;
1639 }
1640
1641 mnt = mount_one_hugetlbfs(h);
1642 if (IS_ERR(mnt))
1643 hugetlbfs_vfsmount[i] = NULL;
1644 else
1645 hugetlbfs_vfsmount[i] = mnt;
1646 i++;
1647 }
1648
1649 return 0;
1650
1651 out_unreg:
1652 (void)unregister_filesystem(&hugetlbfs_fs_type);
1653 out_free:
1654 kmem_cache_destroy(hugetlbfs_inode_cachep);
1655 out:
1656 return error;
1657 }
1658 fs_initcall(init_hugetlbfs_fs)
Linux kernel stable tree