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