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