]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * hugetlbpage-backed filesystem. Based on ramfs. | |
3 | * | |
6d49e352 | 4 | * Nadia Yvette Chambers, 2002 |
1da177e4 LT |
5 | * |
6 | * Copyright (C) 2002 Linus Torvalds. | |
3e89e1c5 | 7 | * License: GPL |
1da177e4 LT |
8 | */ |
9 | ||
9b857d26 AM |
10 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
11 | ||
1da177e4 LT |
12 | #include <linux/thread_info.h> |
13 | #include <asm/current.h> | |
174cd4b1 | 14 | #include <linux/sched/signal.h> /* remove ASAP */ |
70c3547e | 15 | #include <linux/falloc.h> |
1da177e4 LT |
16 | #include <linux/fs.h> |
17 | #include <linux/mount.h> | |
18 | #include <linux/file.h> | |
e73a75fa | 19 | #include <linux/kernel.h> |
1da177e4 LT |
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> | |
16f7e0fe | 25 | #include <linux/capability.h> |
e73a75fa | 26 | #include <linux/ctype.h> |
1da177e4 LT |
27 | #include <linux/backing-dev.h> |
28 | #include <linux/hugetlb.h> | |
29 | #include <linux/pagevec.h> | |
e73a75fa | 30 | #include <linux/parser.h> |
036e0856 | 31 | #include <linux/mman.h> |
1da177e4 LT |
32 | #include <linux/slab.h> |
33 | #include <linux/dnotify.h> | |
34 | #include <linux/statfs.h> | |
35 | #include <linux/security.h> | |
1fd7317d | 36 | #include <linux/magic.h> |
290408d4 | 37 | #include <linux/migrate.h> |
34d0640e | 38 | #include <linux/uio.h> |
1da177e4 | 39 | |
7c0f6ba6 | 40 | #include <linux/uaccess.h> |
1da177e4 | 41 | |
ee9b6d61 | 42 | static const struct super_operations hugetlbfs_ops; |
f5e54d6e | 43 | static const struct address_space_operations hugetlbfs_aops; |
4b6f5d20 | 44 | const struct file_operations hugetlbfs_file_operations; |
92e1d5be AV |
45 | static const struct inode_operations hugetlbfs_dir_inode_operations; |
46 | static const struct inode_operations hugetlbfs_inode_operations; | |
1da177e4 | 47 | |
a1d776ee | 48 | struct hugetlbfs_config { |
4a25220d DH |
49 | struct hstate *hstate; |
50 | long max_hpages; | |
51 | long nr_inodes; | |
52 | long min_hpages; | |
53 | kuid_t uid; | |
54 | kgid_t gid; | |
55 | umode_t mode; | |
a1d776ee DG |
56 | }; |
57 | ||
1da177e4 LT |
58 | int sysctl_hugetlb_shm_group; |
59 | ||
e73a75fa RD |
60 | enum { |
61 | Opt_size, Opt_nr_inodes, | |
62 | Opt_mode, Opt_uid, Opt_gid, | |
7ca02d0a | 63 | Opt_pagesize, Opt_min_size, |
e73a75fa RD |
64 | Opt_err, |
65 | }; | |
66 | ||
a447c093 | 67 | static const match_table_t tokens = { |
e73a75fa RD |
68 | {Opt_size, "size=%s"}, |
69 | {Opt_nr_inodes, "nr_inodes=%s"}, | |
70 | {Opt_mode, "mode=%o"}, | |
71 | {Opt_uid, "uid=%u"}, | |
72 | {Opt_gid, "gid=%u"}, | |
a137e1cc | 73 | {Opt_pagesize, "pagesize=%s"}, |
7ca02d0a | 74 | {Opt_min_size, "min_size=%s"}, |
e73a75fa RD |
75 | {Opt_err, NULL}, |
76 | }; | |
77 | ||
70c3547e MK |
78 | #ifdef CONFIG_NUMA |
79 | static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma, | |
80 | struct inode *inode, pgoff_t index) | |
81 | { | |
82 | vma->vm_policy = mpol_shared_policy_lookup(&HUGETLBFS_I(inode)->policy, | |
83 | index); | |
84 | } | |
85 | ||
86 | static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma) | |
87 | { | |
88 | mpol_cond_put(vma->vm_policy); | |
89 | } | |
90 | #else | |
91 | static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma, | |
92 | struct inode *inode, pgoff_t index) | |
93 | { | |
94 | } | |
95 | ||
96 | static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma) | |
97 | { | |
98 | } | |
99 | #endif | |
100 | ||
2e9b367c AL |
101 | static void huge_pagevec_release(struct pagevec *pvec) |
102 | { | |
103 | int i; | |
104 | ||
105 | for (i = 0; i < pagevec_count(pvec); ++i) | |
106 | put_page(pvec->pages[i]); | |
107 | ||
108 | pagevec_reinit(pvec); | |
109 | } | |
110 | ||
63489f8e MK |
111 | /* |
112 | * Mask used when checking the page offset value passed in via system | |
113 | * calls. This value will be converted to a loff_t which is signed. | |
114 | * Therefore, we want to check the upper PAGE_SHIFT + 1 bits of the | |
115 | * value. The extra bit (- 1 in the shift value) is to take the sign | |
116 | * bit into account. | |
117 | */ | |
118 | #define PGOFF_LOFFT_MAX \ | |
119 | (((1UL << (PAGE_SHIFT + 1)) - 1) << (BITS_PER_LONG - (PAGE_SHIFT + 1))) | |
120 | ||
1da177e4 LT |
121 | static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma) |
122 | { | |
496ad9aa | 123 | struct inode *inode = file_inode(file); |
1da177e4 LT |
124 | loff_t len, vma_len; |
125 | int ret; | |
a5516438 | 126 | struct hstate *h = hstate_file(file); |
1da177e4 | 127 | |
68589bc3 | 128 | /* |
dec4ad86 DG |
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). | |
68589bc3 | 135 | */ |
a2fce914 | 136 | vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND; |
68589bc3 | 137 | vma->vm_ops = &hugetlb_vm_ops; |
1da177e4 | 138 | |
045c7a3f | 139 | /* |
63489f8e MK |
140 | * page based offset in vm_pgoff could be sufficiently large to |
141 | * overflow a (l)off_t when converted to byte offset. | |
045c7a3f | 142 | */ |
63489f8e | 143 | if (vma->vm_pgoff & PGOFF_LOFFT_MAX) |
045c7a3f MK |
144 | return -EINVAL; |
145 | ||
63489f8e | 146 | /* must be huge page aligned */ |
2b37c35e | 147 | if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT)) |
dec4ad86 DG |
148 | return -EINVAL; |
149 | ||
1da177e4 | 150 | vma_len = (loff_t)(vma->vm_end - vma->vm_start); |
045c7a3f MK |
151 | len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); |
152 | /* check for overflow */ | |
153 | if (len < vma_len) | |
154 | return -EINVAL; | |
1da177e4 | 155 | |
5955102c | 156 | inode_lock(inode); |
1da177e4 | 157 | file_accessed(file); |
1da177e4 LT |
158 | |
159 | ret = -ENOMEM; | |
a1e78772 | 160 | if (hugetlb_reserve_pages(inode, |
a5516438 | 161 | vma->vm_pgoff >> huge_page_order(h), |
5a6fe125 MG |
162 | len >> huge_page_shift(h), vma, |
163 | vma->vm_flags)) | |
a43a8c39 | 164 | goto out; |
b45b5bd6 | 165 | |
4c887265 | 166 | ret = 0; |
b6174df5 | 167 | if (vma->vm_flags & VM_WRITE && inode->i_size < len) |
045c7a3f | 168 | i_size_write(inode, len); |
1da177e4 | 169 | out: |
5955102c | 170 | inode_unlock(inode); |
1da177e4 LT |
171 | |
172 | return ret; | |
173 | } | |
174 | ||
175 | /* | |
508034a3 | 176 | * Called under down_write(mmap_sem). |
1da177e4 LT |
177 | */ |
178 | ||
d2ba27e8 | 179 | #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA |
1da177e4 LT |
180 | static unsigned long |
181 | hugetlb_get_unmapped_area(struct file *file, unsigned long addr, | |
182 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
183 | { | |
184 | struct mm_struct *mm = current->mm; | |
185 | struct vm_area_struct *vma; | |
a5516438 | 186 | struct hstate *h = hstate_file(file); |
08659355 | 187 | struct vm_unmapped_area_info info; |
1da177e4 | 188 | |
a5516438 | 189 | if (len & ~huge_page_mask(h)) |
1da177e4 LT |
190 | return -EINVAL; |
191 | if (len > TASK_SIZE) | |
192 | return -ENOMEM; | |
193 | ||
036e0856 | 194 | if (flags & MAP_FIXED) { |
a5516438 | 195 | if (prepare_hugepage_range(file, addr, len)) |
036e0856 BH |
196 | return -EINVAL; |
197 | return addr; | |
198 | } | |
199 | ||
1da177e4 | 200 | if (addr) { |
a5516438 | 201 | addr = ALIGN(addr, huge_page_size(h)); |
1da177e4 LT |
202 | vma = find_vma(mm, addr); |
203 | if (TASK_SIZE - len >= addr && | |
1be7107f | 204 | (!vma || addr + len <= vm_start_gap(vma))) |
1da177e4 LT |
205 | return addr; |
206 | } | |
207 | ||
08659355 ML |
208 | info.flags = 0; |
209 | info.length = len; | |
210 | info.low_limit = TASK_UNMAPPED_BASE; | |
211 | info.high_limit = TASK_SIZE; | |
212 | info.align_mask = PAGE_MASK & ~huge_page_mask(h); | |
213 | info.align_offset = 0; | |
214 | return vm_unmapped_area(&info); | |
1da177e4 LT |
215 | } |
216 | #endif | |
217 | ||
34d0640e | 218 | static size_t |
e63e1e5a | 219 | hugetlbfs_read_actor(struct page *page, unsigned long offset, |
34d0640e | 220 | struct iov_iter *to, unsigned long size) |
e63e1e5a | 221 | { |
34d0640e | 222 | size_t copied = 0; |
e63e1e5a BP |
223 | int i, chunksize; |
224 | ||
e63e1e5a | 225 | /* Find which 4k chunk and offset with in that chunk */ |
09cbfeaf KS |
226 | i = offset >> PAGE_SHIFT; |
227 | offset = offset & ~PAGE_MASK; | |
e63e1e5a BP |
228 | |
229 | while (size) { | |
34d0640e | 230 | size_t n; |
09cbfeaf | 231 | chunksize = PAGE_SIZE; |
e63e1e5a BP |
232 | if (offset) |
233 | chunksize -= offset; | |
234 | if (chunksize > size) | |
235 | chunksize = size; | |
34d0640e AV |
236 | n = copy_page_to_iter(&page[i], offset, chunksize, to); |
237 | copied += n; | |
238 | if (n != chunksize) | |
239 | return copied; | |
e63e1e5a BP |
240 | offset = 0; |
241 | size -= chunksize; | |
e63e1e5a BP |
242 | i++; |
243 | } | |
34d0640e | 244 | return copied; |
e63e1e5a BP |
245 | } |
246 | ||
247 | /* | |
248 | * Support for read() - Find the page attached to f_mapping and copy out the | |
249 | * data. Its *very* similar to do_generic_mapping_read(), we can't use that | |
ea1754a0 | 250 | * since it has PAGE_SIZE assumptions. |
e63e1e5a | 251 | */ |
34d0640e | 252 | static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to) |
e63e1e5a | 253 | { |
34d0640e AV |
254 | struct file *file = iocb->ki_filp; |
255 | struct hstate *h = hstate_file(file); | |
256 | struct address_space *mapping = file->f_mapping; | |
e63e1e5a | 257 | struct inode *inode = mapping->host; |
34d0640e AV |
258 | unsigned long index = iocb->ki_pos >> huge_page_shift(h); |
259 | unsigned long offset = iocb->ki_pos & ~huge_page_mask(h); | |
e63e1e5a BP |
260 | unsigned long end_index; |
261 | loff_t isize; | |
262 | ssize_t retval = 0; | |
263 | ||
34d0640e | 264 | while (iov_iter_count(to)) { |
e63e1e5a | 265 | struct page *page; |
34d0640e | 266 | size_t nr, copied; |
e63e1e5a BP |
267 | |
268 | /* nr is the maximum number of bytes to copy from this page */ | |
a5516438 | 269 | nr = huge_page_size(h); |
a05b0855 AK |
270 | isize = i_size_read(inode); |
271 | if (!isize) | |
34d0640e | 272 | break; |
a05b0855 | 273 | end_index = (isize - 1) >> huge_page_shift(h); |
34d0640e AV |
274 | if (index > end_index) |
275 | break; | |
276 | if (index == end_index) { | |
a5516438 | 277 | nr = ((isize - 1) & ~huge_page_mask(h)) + 1; |
a05b0855 | 278 | if (nr <= offset) |
34d0640e | 279 | break; |
e63e1e5a BP |
280 | } |
281 | nr = nr - offset; | |
282 | ||
283 | /* Find the page */ | |
a05b0855 | 284 | page = find_lock_page(mapping, index); |
e63e1e5a BP |
285 | if (unlikely(page == NULL)) { |
286 | /* | |
287 | * We have a HOLE, zero out the user-buffer for the | |
288 | * length of the hole or request. | |
289 | */ | |
34d0640e | 290 | copied = iov_iter_zero(nr, to); |
e63e1e5a | 291 | } else { |
a05b0855 AK |
292 | unlock_page(page); |
293 | ||
e63e1e5a BP |
294 | /* |
295 | * We have the page, copy it to user space buffer. | |
296 | */ | |
34d0640e | 297 | copied = hugetlbfs_read_actor(page, offset, to, nr); |
09cbfeaf | 298 | put_page(page); |
e63e1e5a | 299 | } |
34d0640e AV |
300 | offset += copied; |
301 | retval += copied; | |
302 | if (copied != nr && iov_iter_count(to)) { | |
303 | if (!retval) | |
304 | retval = -EFAULT; | |
305 | break; | |
e63e1e5a | 306 | } |
a5516438 AK |
307 | index += offset >> huge_page_shift(h); |
308 | offset &= ~huge_page_mask(h); | |
e63e1e5a | 309 | } |
34d0640e | 310 | iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset; |
e63e1e5a BP |
311 | return retval; |
312 | } | |
313 | ||
800d15a5 NP |
314 | static int hugetlbfs_write_begin(struct file *file, |
315 | struct address_space *mapping, | |
316 | loff_t pos, unsigned len, unsigned flags, | |
317 | struct page **pagep, void **fsdata) | |
1da177e4 LT |
318 | { |
319 | return -EINVAL; | |
320 | } | |
321 | ||
800d15a5 NP |
322 | static int hugetlbfs_write_end(struct file *file, struct address_space *mapping, |
323 | loff_t pos, unsigned len, unsigned copied, | |
324 | struct page *page, void *fsdata) | |
1da177e4 | 325 | { |
800d15a5 | 326 | BUG(); |
1da177e4 LT |
327 | return -EINVAL; |
328 | } | |
329 | ||
b5cec28d | 330 | static void remove_huge_page(struct page *page) |
1da177e4 | 331 | { |
b9ea2515 | 332 | ClearPageDirty(page); |
1da177e4 | 333 | ClearPageUptodate(page); |
bd65cb86 | 334 | delete_from_page_cache(page); |
1da177e4 LT |
335 | } |
336 | ||
4aae8d1c | 337 | static void |
f808c13f | 338 | hugetlb_vmdelete_list(struct rb_root_cached *root, pgoff_t start, pgoff_t end) |
4aae8d1c MK |
339 | { |
340 | struct vm_area_struct *vma; | |
341 | ||
342 | /* | |
343 | * end == 0 indicates that the entire range after | |
344 | * start should be unmapped. | |
345 | */ | |
346 | vma_interval_tree_foreach(vma, root, start, end ? end : ULONG_MAX) { | |
347 | unsigned long v_offset; | |
348 | unsigned long v_end; | |
349 | ||
350 | /* | |
351 | * Can the expression below overflow on 32-bit arches? | |
352 | * No, because the interval tree returns us only those vmas | |
353 | * which overlap the truncated area starting at pgoff, | |
354 | * and no vma on a 32-bit arch can span beyond the 4GB. | |
355 | */ | |
356 | if (vma->vm_pgoff < start) | |
357 | v_offset = (start - vma->vm_pgoff) << PAGE_SHIFT; | |
358 | else | |
359 | v_offset = 0; | |
360 | ||
361 | if (!end) | |
362 | v_end = vma->vm_end; | |
363 | else { | |
364 | v_end = ((end - vma->vm_pgoff) << PAGE_SHIFT) | |
365 | + vma->vm_start; | |
366 | if (v_end > vma->vm_end) | |
367 | v_end = vma->vm_end; | |
368 | } | |
369 | ||
370 | unmap_hugepage_range(vma, vma->vm_start + v_offset, v_end, | |
371 | NULL); | |
372 | } | |
373 | } | |
b5cec28d MK |
374 | |
375 | /* | |
376 | * remove_inode_hugepages handles two distinct cases: truncation and hole | |
377 | * punch. There are subtle differences in operation for each case. | |
4aae8d1c | 378 | * |
b5cec28d MK |
379 | * truncation is indicated by end of range being LLONG_MAX |
380 | * In this case, we first scan the range and release found pages. | |
381 | * After releasing pages, hugetlb_unreserve_pages cleans up region/reserv | |
1817889e MK |
382 | * maps and global counts. Page faults can not race with truncation |
383 | * in this routine. hugetlb_no_page() prevents page faults in the | |
384 | * truncated range. It checks i_size before allocation, and again after | |
385 | * with the page table lock for the page held. The same lock must be | |
386 | * acquired to unmap a page. | |
b5cec28d MK |
387 | * hole punch is indicated if end is not LLONG_MAX |
388 | * In the hole punch case we scan the range and release found pages. | |
389 | * Only when releasing a page is the associated region/reserv map | |
390 | * deleted. The region/reserv map for ranges without associated | |
1817889e MK |
391 | * pages are not modified. Page faults can race with hole punch. |
392 | * This is indicated if we find a mapped page. | |
b5cec28d MK |
393 | * Note: If the passed end of range value is beyond the end of file, but |
394 | * not LLONG_MAX this routine still performs a hole punch operation. | |
395 | */ | |
396 | static void remove_inode_hugepages(struct inode *inode, loff_t lstart, | |
397 | loff_t lend) | |
1da177e4 | 398 | { |
a5516438 | 399 | struct hstate *h = hstate_inode(inode); |
b45b5bd6 | 400 | struct address_space *mapping = &inode->i_data; |
a5516438 | 401 | const pgoff_t start = lstart >> huge_page_shift(h); |
b5cec28d MK |
402 | const pgoff_t end = lend >> huge_page_shift(h); |
403 | struct vm_area_struct pseudo_vma; | |
1da177e4 | 404 | struct pagevec pvec; |
d72dc8a2 | 405 | pgoff_t next, index; |
a43a8c39 | 406 | int i, freed = 0; |
b5cec28d | 407 | bool truncate_op = (lend == LLONG_MAX); |
1da177e4 | 408 | |
b5cec28d MK |
409 | memset(&pseudo_vma, 0, sizeof(struct vm_area_struct)); |
410 | pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); | |
86679820 | 411 | pagevec_init(&pvec); |
1da177e4 | 412 | next = start; |
b5cec28d | 413 | while (next < end) { |
b5cec28d | 414 | /* |
1817889e | 415 | * When no more pages are found, we are done. |
b5cec28d | 416 | */ |
397162ff | 417 | if (!pagevec_lookup_range(&pvec, mapping, &next, end - 1)) |
1817889e | 418 | break; |
1da177e4 LT |
419 | |
420 | for (i = 0; i < pagevec_count(&pvec); ++i) { | |
421 | struct page *page = pvec.pages[i]; | |
b5cec28d MK |
422 | u32 hash; |
423 | ||
d72dc8a2 | 424 | index = page->index; |
b5cec28d MK |
425 | hash = hugetlb_fault_mutex_hash(h, current->mm, |
426 | &pseudo_vma, | |
d72dc8a2 | 427 | mapping, index, 0); |
b5cec28d | 428 | mutex_lock(&hugetlb_fault_mutex_table[hash]); |
1da177e4 | 429 | |
4aae8d1c MK |
430 | /* |
431 | * If page is mapped, it was faulted in after being | |
432 | * unmapped in caller. Unmap (again) now after taking | |
433 | * the fault mutex. The mutex will prevent faults | |
434 | * until we finish removing the page. | |
435 | * | |
436 | * This race can only happen in the hole punch case. | |
437 | * Getting here in a truncate operation is a bug. | |
438 | */ | |
439 | if (unlikely(page_mapped(page))) { | |
1817889e | 440 | BUG_ON(truncate_op); |
4aae8d1c MK |
441 | |
442 | i_mmap_lock_write(mapping); | |
443 | hugetlb_vmdelete_list(&mapping->i_mmap, | |
d72dc8a2 JK |
444 | index * pages_per_huge_page(h), |
445 | (index + 1) * pages_per_huge_page(h)); | |
4aae8d1c MK |
446 | i_mmap_unlock_write(mapping); |
447 | } | |
448 | ||
449 | lock_page(page); | |
450 | /* | |
451 | * We must free the huge page and remove from page | |
452 | * cache (remove_huge_page) BEFORE removing the | |
453 | * region/reserve map (hugetlb_unreserve_pages). In | |
454 | * rare out of memory conditions, removal of the | |
72e2936c | 455 | * region/reserve map could fail. Correspondingly, |
456 | * the subpool and global reserve usage count can need | |
457 | * to be adjusted. | |
4aae8d1c | 458 | */ |
72e2936c | 459 | VM_BUG_ON(PagePrivate(page)); |
4aae8d1c MK |
460 | remove_huge_page(page); |
461 | freed++; | |
462 | if (!truncate_op) { | |
463 | if (unlikely(hugetlb_unreserve_pages(inode, | |
d72dc8a2 | 464 | index, index + 1, 1))) |
72e2936c | 465 | hugetlb_fix_reserve_counts(inode); |
b5cec28d MK |
466 | } |
467 | ||
1da177e4 | 468 | unlock_page(page); |
b5cec28d | 469 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); |
1da177e4 LT |
470 | } |
471 | huge_pagevec_release(&pvec); | |
1817889e | 472 | cond_resched(); |
1da177e4 | 473 | } |
b5cec28d MK |
474 | |
475 | if (truncate_op) | |
476 | (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed); | |
1da177e4 LT |
477 | } |
478 | ||
2bbbda30 | 479 | static void hugetlbfs_evict_inode(struct inode *inode) |
1da177e4 | 480 | { |
9119a41e JK |
481 | struct resv_map *resv_map; |
482 | ||
b5cec28d | 483 | remove_inode_hugepages(inode, 0, LLONG_MAX); |
9119a41e JK |
484 | resv_map = (struct resv_map *)inode->i_mapping->private_data; |
485 | /* root inode doesn't have the resv_map, so we should check it */ | |
486 | if (resv_map) | |
487 | resv_map_release(&resv_map->refs); | |
dbd5768f | 488 | clear_inode(inode); |
149f4211 CH |
489 | } |
490 | ||
1da177e4 LT |
491 | static int hugetlb_vmtruncate(struct inode *inode, loff_t offset) |
492 | { | |
856fc295 | 493 | pgoff_t pgoff; |
1da177e4 | 494 | struct address_space *mapping = inode->i_mapping; |
a5516438 | 495 | struct hstate *h = hstate_inode(inode); |
1da177e4 | 496 | |
a5516438 | 497 | BUG_ON(offset & ~huge_page_mask(h)); |
856fc295 | 498 | pgoff = offset >> PAGE_SHIFT; |
1da177e4 | 499 | |
7aa91e10 | 500 | i_size_write(inode, offset); |
83cde9e8 | 501 | i_mmap_lock_write(mapping); |
f808c13f | 502 | if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)) |
1bfad99a | 503 | hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0); |
83cde9e8 | 504 | i_mmap_unlock_write(mapping); |
b5cec28d | 505 | remove_inode_hugepages(inode, offset, LLONG_MAX); |
1da177e4 LT |
506 | return 0; |
507 | } | |
508 | ||
70c3547e MK |
509 | static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len) |
510 | { | |
511 | struct hstate *h = hstate_inode(inode); | |
512 | loff_t hpage_size = huge_page_size(h); | |
513 | loff_t hole_start, hole_end; | |
514 | ||
515 | /* | |
516 | * For hole punch round up the beginning offset of the hole and | |
517 | * round down the end. | |
518 | */ | |
519 | hole_start = round_up(offset, hpage_size); | |
520 | hole_end = round_down(offset + len, hpage_size); | |
521 | ||
522 | if (hole_end > hole_start) { | |
523 | struct address_space *mapping = inode->i_mapping; | |
ff62a342 | 524 | struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); |
70c3547e | 525 | |
5955102c | 526 | inode_lock(inode); |
ff62a342 MAL |
527 | |
528 | /* protected by i_mutex */ | |
529 | if (info->seals & F_SEAL_WRITE) { | |
530 | inode_unlock(inode); | |
531 | return -EPERM; | |
532 | } | |
533 | ||
70c3547e | 534 | i_mmap_lock_write(mapping); |
f808c13f | 535 | if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)) |
70c3547e MK |
536 | hugetlb_vmdelete_list(&mapping->i_mmap, |
537 | hole_start >> PAGE_SHIFT, | |
538 | hole_end >> PAGE_SHIFT); | |
539 | i_mmap_unlock_write(mapping); | |
540 | remove_inode_hugepages(inode, hole_start, hole_end); | |
5955102c | 541 | inode_unlock(inode); |
70c3547e MK |
542 | } |
543 | ||
544 | return 0; | |
545 | } | |
546 | ||
547 | static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset, | |
548 | loff_t len) | |
549 | { | |
550 | struct inode *inode = file_inode(file); | |
ff62a342 | 551 | struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); |
70c3547e MK |
552 | struct address_space *mapping = inode->i_mapping; |
553 | struct hstate *h = hstate_inode(inode); | |
554 | struct vm_area_struct pseudo_vma; | |
555 | struct mm_struct *mm = current->mm; | |
556 | loff_t hpage_size = huge_page_size(h); | |
557 | unsigned long hpage_shift = huge_page_shift(h); | |
558 | pgoff_t start, index, end; | |
559 | int error; | |
560 | u32 hash; | |
561 | ||
562 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
563 | return -EOPNOTSUPP; | |
564 | ||
565 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
566 | return hugetlbfs_punch_hole(inode, offset, len); | |
567 | ||
568 | /* | |
569 | * Default preallocate case. | |
570 | * For this range, start is rounded down and end is rounded up | |
571 | * as well as being converted to page offsets. | |
572 | */ | |
573 | start = offset >> hpage_shift; | |
574 | end = (offset + len + hpage_size - 1) >> hpage_shift; | |
575 | ||
5955102c | 576 | inode_lock(inode); |
70c3547e MK |
577 | |
578 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
579 | error = inode_newsize_ok(inode, offset + len); | |
580 | if (error) | |
581 | goto out; | |
582 | ||
ff62a342 MAL |
583 | if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { |
584 | error = -EPERM; | |
585 | goto out; | |
586 | } | |
587 | ||
70c3547e MK |
588 | /* |
589 | * Initialize a pseudo vma as this is required by the huge page | |
590 | * allocation routines. If NUMA is configured, use page index | |
591 | * as input to create an allocation policy. | |
592 | */ | |
593 | memset(&pseudo_vma, 0, sizeof(struct vm_area_struct)); | |
594 | pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); | |
595 | pseudo_vma.vm_file = file; | |
596 | ||
597 | for (index = start; index < end; index++) { | |
598 | /* | |
599 | * This is supposed to be the vaddr where the page is being | |
600 | * faulted in, but we have no vaddr here. | |
601 | */ | |
602 | struct page *page; | |
603 | unsigned long addr; | |
604 | int avoid_reserve = 0; | |
605 | ||
606 | cond_resched(); | |
607 | ||
608 | /* | |
609 | * fallocate(2) manpage permits EINTR; we may have been | |
610 | * interrupted because we are using up too much memory. | |
611 | */ | |
612 | if (signal_pending(current)) { | |
613 | error = -EINTR; | |
614 | break; | |
615 | } | |
616 | ||
617 | /* Set numa allocation policy based on index */ | |
618 | hugetlb_set_vma_policy(&pseudo_vma, inode, index); | |
619 | ||
620 | /* addr is the offset within the file (zero based) */ | |
621 | addr = index * hpage_size; | |
622 | ||
623 | /* mutex taken here, fault path and hole punch */ | |
624 | hash = hugetlb_fault_mutex_hash(h, mm, &pseudo_vma, mapping, | |
625 | index, addr); | |
626 | mutex_lock(&hugetlb_fault_mutex_table[hash]); | |
627 | ||
628 | /* See if already present in mapping to avoid alloc/free */ | |
629 | page = find_get_page(mapping, index); | |
630 | if (page) { | |
631 | put_page(page); | |
632 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
633 | hugetlb_drop_vma_policy(&pseudo_vma); | |
634 | continue; | |
635 | } | |
636 | ||
637 | /* Allocate page and add to page cache */ | |
638 | page = alloc_huge_page(&pseudo_vma, addr, avoid_reserve); | |
639 | hugetlb_drop_vma_policy(&pseudo_vma); | |
640 | if (IS_ERR(page)) { | |
641 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
642 | error = PTR_ERR(page); | |
643 | goto out; | |
644 | } | |
645 | clear_huge_page(page, addr, pages_per_huge_page(h)); | |
646 | __SetPageUptodate(page); | |
647 | error = huge_add_to_page_cache(page, mapping, index); | |
648 | if (unlikely(error)) { | |
649 | put_page(page); | |
650 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
651 | goto out; | |
652 | } | |
653 | ||
654 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
655 | ||
656 | /* | |
70c3547e | 657 | * unlock_page because locked by add_to_page_cache() |
72639e6d | 658 | * page_put due to reference from alloc_huge_page() |
70c3547e | 659 | */ |
70c3547e | 660 | unlock_page(page); |
72639e6d | 661 | put_page(page); |
70c3547e MK |
662 | } |
663 | ||
664 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
665 | i_size_write(inode, offset + len); | |
078cd827 | 666 | inode->i_ctime = current_time(inode); |
70c3547e | 667 | out: |
5955102c | 668 | inode_unlock(inode); |
70c3547e MK |
669 | return error; |
670 | } | |
671 | ||
1da177e4 LT |
672 | static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr) |
673 | { | |
2b0143b5 | 674 | struct inode *inode = d_inode(dentry); |
a5516438 | 675 | struct hstate *h = hstate_inode(inode); |
1da177e4 LT |
676 | int error; |
677 | unsigned int ia_valid = attr->ia_valid; | |
ff62a342 | 678 | struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); |
1da177e4 LT |
679 | |
680 | BUG_ON(!inode); | |
681 | ||
31051c85 | 682 | error = setattr_prepare(dentry, attr); |
1da177e4 | 683 | if (error) |
1025774c | 684 | return error; |
1da177e4 LT |
685 | |
686 | if (ia_valid & ATTR_SIZE) { | |
ff62a342 MAL |
687 | loff_t oldsize = inode->i_size; |
688 | loff_t newsize = attr->ia_size; | |
689 | ||
690 | if (newsize & ~huge_page_mask(h)) | |
1025774c | 691 | return -EINVAL; |
ff62a342 MAL |
692 | /* protected by i_mutex */ |
693 | if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || | |
694 | (newsize > oldsize && (info->seals & F_SEAL_GROW))) | |
695 | return -EPERM; | |
696 | error = hugetlb_vmtruncate(inode, newsize); | |
1da177e4 | 697 | if (error) |
1025774c | 698 | return error; |
1da177e4 | 699 | } |
1025774c CH |
700 | |
701 | setattr_copy(inode, attr); | |
702 | mark_inode_dirty(inode); | |
703 | return 0; | |
1da177e4 LT |
704 | } |
705 | ||
7d54fa64 AV |
706 | static struct inode *hugetlbfs_get_root(struct super_block *sb, |
707 | struct hugetlbfs_config *config) | |
1da177e4 LT |
708 | { |
709 | struct inode *inode; | |
1da177e4 LT |
710 | |
711 | inode = new_inode(sb); | |
712 | if (inode) { | |
85fe4025 | 713 | inode->i_ino = get_next_ino(); |
7d54fa64 AV |
714 | inode->i_mode = S_IFDIR | config->mode; |
715 | inode->i_uid = config->uid; | |
716 | inode->i_gid = config->gid; | |
078cd827 | 717 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
7d54fa64 AV |
718 | inode->i_op = &hugetlbfs_dir_inode_operations; |
719 | inode->i_fop = &simple_dir_operations; | |
720 | /* directory inodes start off with i_nlink == 2 (for "." entry) */ | |
721 | inc_nlink(inode); | |
65ed7601 | 722 | lockdep_annotate_inode_mutex_key(inode); |
7d54fa64 AV |
723 | } |
724 | return inode; | |
725 | } | |
726 | ||
b610ded7 | 727 | /* |
c8c06efa | 728 | * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never |
b610ded7 | 729 | * be taken from reclaim -- unlike regular filesystems. This needs an |
88f306b6 | 730 | * annotation because huge_pmd_share() does an allocation under hugetlb's |
c8c06efa | 731 | * i_mmap_rwsem. |
b610ded7 | 732 | */ |
c8c06efa | 733 | static struct lock_class_key hugetlbfs_i_mmap_rwsem_key; |
b610ded7 | 734 | |
7d54fa64 AV |
735 | static struct inode *hugetlbfs_get_inode(struct super_block *sb, |
736 | struct inode *dir, | |
18df2252 | 737 | umode_t mode, dev_t dev) |
7d54fa64 AV |
738 | { |
739 | struct inode *inode; | |
9119a41e JK |
740 | struct resv_map *resv_map; |
741 | ||
742 | resv_map = resv_map_alloc(); | |
743 | if (!resv_map) | |
744 | return NULL; | |
7d54fa64 AV |
745 | |
746 | inode = new_inode(sb); | |
747 | if (inode) { | |
ff62a342 MAL |
748 | struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); |
749 | ||
7d54fa64 AV |
750 | inode->i_ino = get_next_ino(); |
751 | inode_init_owner(inode, dir, mode); | |
c8c06efa DB |
752 | lockdep_set_class(&inode->i_mapping->i_mmap_rwsem, |
753 | &hugetlbfs_i_mmap_rwsem_key); | |
1da177e4 | 754 | inode->i_mapping->a_ops = &hugetlbfs_aops; |
078cd827 | 755 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
9119a41e | 756 | inode->i_mapping->private_data = resv_map; |
ff62a342 | 757 | info->seals = F_SEAL_SEAL; |
1da177e4 LT |
758 | switch (mode & S_IFMT) { |
759 | default: | |
760 | init_special_inode(inode, mode, dev); | |
761 | break; | |
762 | case S_IFREG: | |
763 | inode->i_op = &hugetlbfs_inode_operations; | |
764 | inode->i_fop = &hugetlbfs_file_operations; | |
765 | break; | |
766 | case S_IFDIR: | |
767 | inode->i_op = &hugetlbfs_dir_inode_operations; | |
768 | inode->i_fop = &simple_dir_operations; | |
769 | ||
770 | /* directory inodes start off with i_nlink == 2 (for "." entry) */ | |
d8c76e6f | 771 | inc_nlink(inode); |
1da177e4 LT |
772 | break; |
773 | case S_IFLNK: | |
774 | inode->i_op = &page_symlink_inode_operations; | |
21fc61c7 | 775 | inode_nohighmem(inode); |
1da177e4 LT |
776 | break; |
777 | } | |
e096d0c7 | 778 | lockdep_annotate_inode_mutex_key(inode); |
9119a41e JK |
779 | } else |
780 | kref_put(&resv_map->refs, resv_map_release); | |
781 | ||
1da177e4 LT |
782 | return inode; |
783 | } | |
784 | ||
785 | /* | |
786 | * File creation. Allocate an inode, and we're done.. | |
787 | */ | |
788 | static int hugetlbfs_mknod(struct inode *dir, | |
1a67aafb | 789 | struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 LT |
790 | { |
791 | struct inode *inode; | |
792 | int error = -ENOSPC; | |
7d54fa64 AV |
793 | |
794 | inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev); | |
1da177e4 | 795 | if (inode) { |
078cd827 | 796 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1da177e4 LT |
797 | d_instantiate(dentry, inode); |
798 | dget(dentry); /* Extra count - pin the dentry in core */ | |
799 | error = 0; | |
800 | } | |
801 | return error; | |
802 | } | |
803 | ||
18bb1db3 | 804 | static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
805 | { |
806 | int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0); | |
807 | if (!retval) | |
d8c76e6f | 808 | inc_nlink(dir); |
1da177e4 LT |
809 | return retval; |
810 | } | |
811 | ||
ebfc3b49 | 812 | static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl) |
1da177e4 LT |
813 | { |
814 | return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0); | |
815 | } | |
816 | ||
817 | static int hugetlbfs_symlink(struct inode *dir, | |
818 | struct dentry *dentry, const char *symname) | |
819 | { | |
820 | struct inode *inode; | |
821 | int error = -ENOSPC; | |
1da177e4 | 822 | |
7d54fa64 | 823 | inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0); |
1da177e4 LT |
824 | if (inode) { |
825 | int l = strlen(symname)+1; | |
826 | error = page_symlink(inode, symname, l); | |
827 | if (!error) { | |
828 | d_instantiate(dentry, inode); | |
829 | dget(dentry); | |
830 | } else | |
831 | iput(inode); | |
832 | } | |
078cd827 | 833 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1da177e4 LT |
834 | |
835 | return error; | |
836 | } | |
837 | ||
838 | /* | |
6649a386 | 839 | * mark the head page dirty |
1da177e4 LT |
840 | */ |
841 | static int hugetlbfs_set_page_dirty(struct page *page) | |
842 | { | |
d85f3385 | 843 | struct page *head = compound_head(page); |
6649a386 KC |
844 | |
845 | SetPageDirty(head); | |
1da177e4 LT |
846 | return 0; |
847 | } | |
848 | ||
290408d4 | 849 | static int hugetlbfs_migrate_page(struct address_space *mapping, |
b969c4ab | 850 | struct page *newpage, struct page *page, |
a6bc32b8 | 851 | enum migrate_mode mode) |
290408d4 NH |
852 | { |
853 | int rc; | |
854 | ||
855 | rc = migrate_huge_page_move_mapping(mapping, newpage, page); | |
78bd5209 | 856 | if (rc != MIGRATEPAGE_SUCCESS) |
290408d4 | 857 | return rc; |
2916ecc0 JG |
858 | if (mode != MIGRATE_SYNC_NO_COPY) |
859 | migrate_page_copy(newpage, page); | |
860 | else | |
861 | migrate_page_states(newpage, page); | |
290408d4 | 862 | |
78bd5209 | 863 | return MIGRATEPAGE_SUCCESS; |
290408d4 NH |
864 | } |
865 | ||
78bb9203 NH |
866 | static int hugetlbfs_error_remove_page(struct address_space *mapping, |
867 | struct page *page) | |
868 | { | |
869 | struct inode *inode = mapping->host; | |
ab615a5b | 870 | pgoff_t index = page->index; |
78bb9203 NH |
871 | |
872 | remove_huge_page(page); | |
ab615a5b MK |
873 | if (unlikely(hugetlb_unreserve_pages(inode, index, index + 1, 1))) |
874 | hugetlb_fix_reserve_counts(inode); | |
875 | ||
78bb9203 NH |
876 | return 0; |
877 | } | |
878 | ||
4a25220d DH |
879 | /* |
880 | * Display the mount options in /proc/mounts. | |
881 | */ | |
882 | static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root) | |
883 | { | |
884 | struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(root->d_sb); | |
885 | struct hugepage_subpool *spool = sbinfo->spool; | |
886 | unsigned long hpage_size = huge_page_size(sbinfo->hstate); | |
887 | unsigned hpage_shift = huge_page_shift(sbinfo->hstate); | |
888 | char mod; | |
889 | ||
890 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) | |
891 | seq_printf(m, ",uid=%u", | |
892 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
893 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
894 | seq_printf(m, ",gid=%u", | |
895 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
896 | if (sbinfo->mode != 0755) | |
897 | seq_printf(m, ",mode=%o", sbinfo->mode); | |
898 | if (sbinfo->max_inodes != -1) | |
899 | seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes); | |
900 | ||
901 | hpage_size /= 1024; | |
902 | mod = 'K'; | |
903 | if (hpage_size >= 1024) { | |
904 | hpage_size /= 1024; | |
905 | mod = 'M'; | |
906 | } | |
907 | seq_printf(m, ",pagesize=%lu%c", hpage_size, mod); | |
908 | if (spool) { | |
909 | if (spool->max_hpages != -1) | |
910 | seq_printf(m, ",size=%llu", | |
911 | (unsigned long long)spool->max_hpages << hpage_shift); | |
912 | if (spool->min_hpages != -1) | |
913 | seq_printf(m, ",min_size=%llu", | |
914 | (unsigned long long)spool->min_hpages << hpage_shift); | |
915 | } | |
916 | return 0; | |
917 | } | |
918 | ||
726c3342 | 919 | static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 920 | { |
726c3342 | 921 | struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb); |
2b0143b5 | 922 | struct hstate *h = hstate_inode(d_inode(dentry)); |
1da177e4 LT |
923 | |
924 | buf->f_type = HUGETLBFS_MAGIC; | |
a5516438 | 925 | buf->f_bsize = huge_page_size(h); |
1da177e4 LT |
926 | if (sbinfo) { |
927 | spin_lock(&sbinfo->stat_lock); | |
74a8a65c DG |
928 | /* If no limits set, just report 0 for max/free/used |
929 | * blocks, like simple_statfs() */ | |
90481622 DG |
930 | if (sbinfo->spool) { |
931 | long free_pages; | |
932 | ||
933 | spin_lock(&sbinfo->spool->lock); | |
934 | buf->f_blocks = sbinfo->spool->max_hpages; | |
935 | free_pages = sbinfo->spool->max_hpages | |
936 | - sbinfo->spool->used_hpages; | |
937 | buf->f_bavail = buf->f_bfree = free_pages; | |
938 | spin_unlock(&sbinfo->spool->lock); | |
74a8a65c DG |
939 | buf->f_files = sbinfo->max_inodes; |
940 | buf->f_ffree = sbinfo->free_inodes; | |
941 | } | |
1da177e4 LT |
942 | spin_unlock(&sbinfo->stat_lock); |
943 | } | |
944 | buf->f_namelen = NAME_MAX; | |
945 | return 0; | |
946 | } | |
947 | ||
948 | static void hugetlbfs_put_super(struct super_block *sb) | |
949 | { | |
950 | struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb); | |
951 | ||
952 | if (sbi) { | |
953 | sb->s_fs_info = NULL; | |
90481622 DG |
954 | |
955 | if (sbi->spool) | |
956 | hugepage_put_subpool(sbi->spool); | |
957 | ||
1da177e4 LT |
958 | kfree(sbi); |
959 | } | |
960 | } | |
961 | ||
96527980 CH |
962 | static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo) |
963 | { | |
964 | if (sbinfo->free_inodes >= 0) { | |
965 | spin_lock(&sbinfo->stat_lock); | |
966 | if (unlikely(!sbinfo->free_inodes)) { | |
967 | spin_unlock(&sbinfo->stat_lock); | |
968 | return 0; | |
969 | } | |
970 | sbinfo->free_inodes--; | |
971 | spin_unlock(&sbinfo->stat_lock); | |
972 | } | |
973 | ||
974 | return 1; | |
975 | } | |
976 | ||
977 | static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo) | |
978 | { | |
979 | if (sbinfo->free_inodes >= 0) { | |
980 | spin_lock(&sbinfo->stat_lock); | |
981 | sbinfo->free_inodes++; | |
982 | spin_unlock(&sbinfo->stat_lock); | |
983 | } | |
984 | } | |
985 | ||
986 | ||
e18b890b | 987 | static struct kmem_cache *hugetlbfs_inode_cachep; |
1da177e4 LT |
988 | |
989 | static struct inode *hugetlbfs_alloc_inode(struct super_block *sb) | |
990 | { | |
96527980 | 991 | struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb); |
1da177e4 LT |
992 | struct hugetlbfs_inode_info *p; |
993 | ||
96527980 CH |
994 | if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo))) |
995 | return NULL; | |
e94b1766 | 996 | p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL); |
96527980 CH |
997 | if (unlikely(!p)) { |
998 | hugetlbfs_inc_free_inodes(sbinfo); | |
1da177e4 | 999 | return NULL; |
96527980 | 1000 | } |
4742a35d MK |
1001 | |
1002 | /* | |
1003 | * Any time after allocation, hugetlbfs_destroy_inode can be called | |
1004 | * for the inode. mpol_free_shared_policy is unconditionally called | |
1005 | * as part of hugetlbfs_destroy_inode. So, initialize policy here | |
1006 | * in case of a quick call to destroy. | |
1007 | * | |
1008 | * Note that the policy is initialized even if we are creating a | |
1009 | * private inode. This simplifies hugetlbfs_destroy_inode. | |
1010 | */ | |
1011 | mpol_shared_policy_init(&p->policy, NULL); | |
1012 | ||
1da177e4 LT |
1013 | return &p->vfs_inode; |
1014 | } | |
1015 | ||
fa0d7e3d NP |
1016 | static void hugetlbfs_i_callback(struct rcu_head *head) |
1017 | { | |
1018 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
1019 | kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode)); |
1020 | } | |
1021 | ||
1da177e4 LT |
1022 | static void hugetlbfs_destroy_inode(struct inode *inode) |
1023 | { | |
96527980 | 1024 | hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb)); |
1da177e4 | 1025 | mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy); |
fa0d7e3d | 1026 | call_rcu(&inode->i_rcu, hugetlbfs_i_callback); |
1da177e4 LT |
1027 | } |
1028 | ||
f5e54d6e | 1029 | static const struct address_space_operations hugetlbfs_aops = { |
800d15a5 NP |
1030 | .write_begin = hugetlbfs_write_begin, |
1031 | .write_end = hugetlbfs_write_end, | |
1da177e4 | 1032 | .set_page_dirty = hugetlbfs_set_page_dirty, |
290408d4 | 1033 | .migratepage = hugetlbfs_migrate_page, |
78bb9203 | 1034 | .error_remove_page = hugetlbfs_error_remove_page, |
1da177e4 LT |
1035 | }; |
1036 | ||
96527980 | 1037 | |
51cc5068 | 1038 | static void init_once(void *foo) |
96527980 CH |
1039 | { |
1040 | struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo; | |
1041 | ||
a35afb83 | 1042 | inode_init_once(&ei->vfs_inode); |
96527980 CH |
1043 | } |
1044 | ||
4b6f5d20 | 1045 | const struct file_operations hugetlbfs_file_operations = { |
34d0640e | 1046 | .read_iter = hugetlbfs_read_iter, |
1da177e4 | 1047 | .mmap = hugetlbfs_file_mmap, |
1b061d92 | 1048 | .fsync = noop_fsync, |
1da177e4 | 1049 | .get_unmapped_area = hugetlb_get_unmapped_area, |
70c3547e MK |
1050 | .llseek = default_llseek, |
1051 | .fallocate = hugetlbfs_fallocate, | |
1da177e4 LT |
1052 | }; |
1053 | ||
92e1d5be | 1054 | static const struct inode_operations hugetlbfs_dir_inode_operations = { |
1da177e4 LT |
1055 | .create = hugetlbfs_create, |
1056 | .lookup = simple_lookup, | |
1057 | .link = simple_link, | |
1058 | .unlink = simple_unlink, | |
1059 | .symlink = hugetlbfs_symlink, | |
1060 | .mkdir = hugetlbfs_mkdir, | |
1061 | .rmdir = simple_rmdir, | |
1062 | .mknod = hugetlbfs_mknod, | |
1063 | .rename = simple_rename, | |
1064 | .setattr = hugetlbfs_setattr, | |
1065 | }; | |
1066 | ||
92e1d5be | 1067 | static const struct inode_operations hugetlbfs_inode_operations = { |
1da177e4 LT |
1068 | .setattr = hugetlbfs_setattr, |
1069 | }; | |
1070 | ||
ee9b6d61 | 1071 | static const struct super_operations hugetlbfs_ops = { |
1da177e4 LT |
1072 | .alloc_inode = hugetlbfs_alloc_inode, |
1073 | .destroy_inode = hugetlbfs_destroy_inode, | |
2bbbda30 | 1074 | .evict_inode = hugetlbfs_evict_inode, |
1da177e4 | 1075 | .statfs = hugetlbfs_statfs, |
1da177e4 | 1076 | .put_super = hugetlbfs_put_super, |
4a25220d | 1077 | .show_options = hugetlbfs_show_options, |
1da177e4 LT |
1078 | }; |
1079 | ||
4a25220d | 1080 | enum hugetlbfs_size_type { NO_SIZE, SIZE_STD, SIZE_PERCENT }; |
7ca02d0a MK |
1081 | |
1082 | /* | |
1083 | * Convert size option passed from command line to number of huge pages | |
1084 | * in the pool specified by hstate. Size option could be in bytes | |
1085 | * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT). | |
1086 | */ | |
4a25220d | 1087 | static long |
7ca02d0a | 1088 | hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt, |
4a25220d | 1089 | enum hugetlbfs_size_type val_type) |
7ca02d0a MK |
1090 | { |
1091 | if (val_type == NO_SIZE) | |
1092 | return -1; | |
1093 | ||
1094 | if (val_type == SIZE_PERCENT) { | |
1095 | size_opt <<= huge_page_shift(h); | |
1096 | size_opt *= h->max_huge_pages; | |
1097 | do_div(size_opt, 100); | |
1098 | } | |
1099 | ||
1100 | size_opt >>= huge_page_shift(h); | |
1101 | return size_opt; | |
1102 | } | |
1103 | ||
1da177e4 LT |
1104 | static int |
1105 | hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig) | |
1106 | { | |
e73a75fa RD |
1107 | char *p, *rest; |
1108 | substring_t args[MAX_OPT_ARGS]; | |
1109 | int option; | |
7ca02d0a | 1110 | unsigned long long max_size_opt = 0, min_size_opt = 0; |
4a25220d | 1111 | enum hugetlbfs_size_type max_val_type = NO_SIZE, min_val_type = NO_SIZE; |
1da177e4 LT |
1112 | |
1113 | if (!options) | |
1114 | return 0; | |
1da177e4 | 1115 | |
e73a75fa RD |
1116 | while ((p = strsep(&options, ",")) != NULL) { |
1117 | int token; | |
b4c07bce LS |
1118 | if (!*p) |
1119 | continue; | |
e73a75fa RD |
1120 | |
1121 | token = match_token(p, tokens, args); | |
1122 | switch (token) { | |
1123 | case Opt_uid: | |
1124 | if (match_int(&args[0], &option)) | |
1125 | goto bad_val; | |
a0eb3a05 EB |
1126 | pconfig->uid = make_kuid(current_user_ns(), option); |
1127 | if (!uid_valid(pconfig->uid)) | |
1128 | goto bad_val; | |
e73a75fa RD |
1129 | break; |
1130 | ||
1131 | case Opt_gid: | |
1132 | if (match_int(&args[0], &option)) | |
1133 | goto bad_val; | |
a0eb3a05 EB |
1134 | pconfig->gid = make_kgid(current_user_ns(), option); |
1135 | if (!gid_valid(pconfig->gid)) | |
1136 | goto bad_val; | |
e73a75fa RD |
1137 | break; |
1138 | ||
1139 | case Opt_mode: | |
1140 | if (match_octal(&args[0], &option)) | |
1141 | goto bad_val; | |
75897d60 | 1142 | pconfig->mode = option & 01777U; |
e73a75fa RD |
1143 | break; |
1144 | ||
1145 | case Opt_size: { | |
e73a75fa RD |
1146 | /* memparse() will accept a K/M/G without a digit */ |
1147 | if (!isdigit(*args[0].from)) | |
1148 | goto bad_val; | |
7ca02d0a MK |
1149 | max_size_opt = memparse(args[0].from, &rest); |
1150 | max_val_type = SIZE_STD; | |
a137e1cc | 1151 | if (*rest == '%') |
7ca02d0a | 1152 | max_val_type = SIZE_PERCENT; |
e73a75fa RD |
1153 | break; |
1154 | } | |
1da177e4 | 1155 | |
e73a75fa RD |
1156 | case Opt_nr_inodes: |
1157 | /* memparse() will accept a K/M/G without a digit */ | |
1158 | if (!isdigit(*args[0].from)) | |
1159 | goto bad_val; | |
1160 | pconfig->nr_inodes = memparse(args[0].from, &rest); | |
1161 | break; | |
1162 | ||
a137e1cc AK |
1163 | case Opt_pagesize: { |
1164 | unsigned long ps; | |
1165 | ps = memparse(args[0].from, &rest); | |
1166 | pconfig->hstate = size_to_hstate(ps); | |
1167 | if (!pconfig->hstate) { | |
9b857d26 | 1168 | pr_err("Unsupported page size %lu MB\n", |
a137e1cc AK |
1169 | ps >> 20); |
1170 | return -EINVAL; | |
1171 | } | |
1172 | break; | |
1173 | } | |
1174 | ||
7ca02d0a MK |
1175 | case Opt_min_size: { |
1176 | /* memparse() will accept a K/M/G without a digit */ | |
1177 | if (!isdigit(*args[0].from)) | |
1178 | goto bad_val; | |
1179 | min_size_opt = memparse(args[0].from, &rest); | |
1180 | min_val_type = SIZE_STD; | |
1181 | if (*rest == '%') | |
1182 | min_val_type = SIZE_PERCENT; | |
1183 | break; | |
1184 | } | |
1185 | ||
e73a75fa | 1186 | default: |
9b857d26 | 1187 | pr_err("Bad mount option: \"%s\"\n", p); |
b4c07bce | 1188 | return -EINVAL; |
e73a75fa RD |
1189 | break; |
1190 | } | |
1da177e4 | 1191 | } |
a137e1cc | 1192 | |
7ca02d0a MK |
1193 | /* |
1194 | * Use huge page pool size (in hstate) to convert the size | |
1195 | * options to number of huge pages. If NO_SIZE, -1 is returned. | |
1196 | */ | |
1197 | pconfig->max_hpages = hugetlbfs_size_to_hpages(pconfig->hstate, | |
1198 | max_size_opt, max_val_type); | |
1199 | pconfig->min_hpages = hugetlbfs_size_to_hpages(pconfig->hstate, | |
1200 | min_size_opt, min_val_type); | |
1201 | ||
1202 | /* | |
1203 | * If max_size was specified, then min_size must be smaller | |
1204 | */ | |
1205 | if (max_val_type > NO_SIZE && | |
1206 | pconfig->min_hpages > pconfig->max_hpages) { | |
1207 | pr_err("minimum size can not be greater than maximum size\n"); | |
1208 | return -EINVAL; | |
a137e1cc AK |
1209 | } |
1210 | ||
1da177e4 | 1211 | return 0; |
e73a75fa RD |
1212 | |
1213 | bad_val: | |
9b857d26 | 1214 | pr_err("Bad value '%s' for mount option '%s'\n", args[0].from, p); |
c12ddba0 | 1215 | return -EINVAL; |
1da177e4 LT |
1216 | } |
1217 | ||
1218 | static int | |
1219 | hugetlbfs_fill_super(struct super_block *sb, void *data, int silent) | |
1220 | { | |
1da177e4 LT |
1221 | int ret; |
1222 | struct hugetlbfs_config config; | |
1223 | struct hugetlbfs_sb_info *sbinfo; | |
1224 | ||
7ca02d0a | 1225 | config.max_hpages = -1; /* No limit on size by default */ |
1da177e4 | 1226 | config.nr_inodes = -1; /* No limit on number of inodes by default */ |
77c70de1 DH |
1227 | config.uid = current_fsuid(); |
1228 | config.gid = current_fsgid(); | |
1da177e4 | 1229 | config.mode = 0755; |
a137e1cc | 1230 | config.hstate = &default_hstate; |
7ca02d0a | 1231 | config.min_hpages = -1; /* No default minimum size */ |
1da177e4 | 1232 | ret = hugetlbfs_parse_options(data, &config); |
1da177e4 LT |
1233 | if (ret) |
1234 | return ret; | |
1235 | ||
1236 | sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL); | |
1237 | if (!sbinfo) | |
1238 | return -ENOMEM; | |
1239 | sb->s_fs_info = sbinfo; | |
a137e1cc | 1240 | sbinfo->hstate = config.hstate; |
1da177e4 | 1241 | spin_lock_init(&sbinfo->stat_lock); |
1da177e4 LT |
1242 | sbinfo->max_inodes = config.nr_inodes; |
1243 | sbinfo->free_inodes = config.nr_inodes; | |
90481622 | 1244 | sbinfo->spool = NULL; |
4a25220d DH |
1245 | sbinfo->uid = config.uid; |
1246 | sbinfo->gid = config.gid; | |
1247 | sbinfo->mode = config.mode; | |
1248 | ||
7ca02d0a MK |
1249 | /* |
1250 | * Allocate and initialize subpool if maximum or minimum size is | |
1251 | * specified. Any needed reservations (for minimim size) are taken | |
1252 | * taken when the subpool is created. | |
1253 | */ | |
1254 | if (config.max_hpages != -1 || config.min_hpages != -1) { | |
1255 | sbinfo->spool = hugepage_new_subpool(config.hstate, | |
1256 | config.max_hpages, | |
1257 | config.min_hpages); | |
90481622 DG |
1258 | if (!sbinfo->spool) |
1259 | goto out_free; | |
1260 | } | |
1da177e4 | 1261 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
a137e1cc AK |
1262 | sb->s_blocksize = huge_page_size(config.hstate); |
1263 | sb->s_blocksize_bits = huge_page_shift(config.hstate); | |
1da177e4 LT |
1264 | sb->s_magic = HUGETLBFS_MAGIC; |
1265 | sb->s_op = &hugetlbfs_ops; | |
1266 | sb->s_time_gran = 1; | |
48fde701 AV |
1267 | sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config)); |
1268 | if (!sb->s_root) | |
1da177e4 | 1269 | goto out_free; |
1da177e4 LT |
1270 | return 0; |
1271 | out_free: | |
6e6870d4 | 1272 | kfree(sbinfo->spool); |
1da177e4 LT |
1273 | kfree(sbinfo); |
1274 | return -ENOMEM; | |
1275 | } | |
1276 | ||
3c26ff6e AV |
1277 | static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type, |
1278 | int flags, const char *dev_name, void *data) | |
1da177e4 | 1279 | { |
3c26ff6e | 1280 | return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super); |
1da177e4 LT |
1281 | } |
1282 | ||
1283 | static struct file_system_type hugetlbfs_fs_type = { | |
1284 | .name = "hugetlbfs", | |
3c26ff6e | 1285 | .mount = hugetlbfs_mount, |
1da177e4 LT |
1286 | .kill_sb = kill_litter_super, |
1287 | }; | |
1288 | ||
42d7395f | 1289 | static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE]; |
1da177e4 | 1290 | |
ef1ff6b8 | 1291 | static int can_do_hugetlb_shm(void) |
1da177e4 | 1292 | { |
a0eb3a05 EB |
1293 | kgid_t shm_group; |
1294 | shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group); | |
1295 | return capable(CAP_IPC_LOCK) || in_group_p(shm_group); | |
1da177e4 LT |
1296 | } |
1297 | ||
42d7395f AK |
1298 | static int get_hstate_idx(int page_size_log) |
1299 | { | |
af73e4d9 | 1300 | struct hstate *h = hstate_sizelog(page_size_log); |
42d7395f | 1301 | |
42d7395f AK |
1302 | if (!h) |
1303 | return -1; | |
1304 | return h - hstates; | |
1305 | } | |
1306 | ||
be1d2cf5 | 1307 | static const struct dentry_operations anon_ops = { |
118b2302 | 1308 | .d_dname = simple_dname |
0df4d6e5 AV |
1309 | }; |
1310 | ||
af73e4d9 NH |
1311 | /* |
1312 | * Note that size should be aligned to proper hugepage size in caller side, | |
1313 | * otherwise hugetlb_reserve_pages reserves one less hugepages than intended. | |
1314 | */ | |
1315 | struct file *hugetlb_file_setup(const char *name, size_t size, | |
1316 | vm_flags_t acctflag, struct user_struct **user, | |
42d7395f | 1317 | int creat_flags, int page_size_log) |
1da177e4 | 1318 | { |
39b65252 | 1319 | struct file *file = ERR_PTR(-ENOMEM); |
1da177e4 | 1320 | struct inode *inode; |
2c48b9c4 | 1321 | struct path path; |
0df4d6e5 | 1322 | struct super_block *sb; |
1da177e4 | 1323 | struct qstr quick_string; |
42d7395f AK |
1324 | int hstate_idx; |
1325 | ||
1326 | hstate_idx = get_hstate_idx(page_size_log); | |
1327 | if (hstate_idx < 0) | |
1328 | return ERR_PTR(-ENODEV); | |
1da177e4 | 1329 | |
353d5c30 | 1330 | *user = NULL; |
42d7395f | 1331 | if (!hugetlbfs_vfsmount[hstate_idx]) |
5bc98594 AM |
1332 | return ERR_PTR(-ENOENT); |
1333 | ||
ef1ff6b8 | 1334 | if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) { |
353d5c30 HD |
1335 | *user = current_user(); |
1336 | if (user_shm_lock(size, *user)) { | |
21a3c273 | 1337 | task_lock(current); |
9b857d26 | 1338 | pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n", |
21a3c273 DR |
1339 | current->comm, current->pid); |
1340 | task_unlock(current); | |
353d5c30 HD |
1341 | } else { |
1342 | *user = NULL; | |
2584e517 | 1343 | return ERR_PTR(-EPERM); |
353d5c30 | 1344 | } |
2584e517 | 1345 | } |
1da177e4 | 1346 | |
0df4d6e5 | 1347 | sb = hugetlbfs_vfsmount[hstate_idx]->mnt_sb; |
9d66586f | 1348 | quick_string.name = name; |
1da177e4 LT |
1349 | quick_string.len = strlen(quick_string.name); |
1350 | quick_string.hash = 0; | |
0df4d6e5 | 1351 | path.dentry = d_alloc_pseudo(sb, &quick_string); |
2c48b9c4 | 1352 | if (!path.dentry) |
1da177e4 LT |
1353 | goto out_shm_unlock; |
1354 | ||
0df4d6e5 | 1355 | d_set_d_op(path.dentry, &anon_ops); |
42d7395f | 1356 | path.mnt = mntget(hugetlbfs_vfsmount[hstate_idx]); |
39b65252 | 1357 | file = ERR_PTR(-ENOSPC); |
0df4d6e5 | 1358 | inode = hugetlbfs_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0); |
1da177e4 | 1359 | if (!inode) |
ce8d2cdf | 1360 | goto out_dentry; |
e1832f29 SS |
1361 | if (creat_flags == HUGETLB_SHMFS_INODE) |
1362 | inode->i_flags |= S_PRIVATE; | |
1da177e4 | 1363 | |
39b65252 | 1364 | file = ERR_PTR(-ENOMEM); |
af73e4d9 NH |
1365 | if (hugetlb_reserve_pages(inode, 0, |
1366 | size >> huge_page_shift(hstate_inode(inode)), NULL, | |
1367 | acctflag)) | |
b45b5bd6 DG |
1368 | goto out_inode; |
1369 | ||
2c48b9c4 | 1370 | d_instantiate(path.dentry, inode); |
1da177e4 | 1371 | inode->i_size = size; |
6d6b77f1 | 1372 | clear_nlink(inode); |
ce8d2cdf | 1373 | |
2c48b9c4 | 1374 | file = alloc_file(&path, FMODE_WRITE | FMODE_READ, |
ce8d2cdf | 1375 | &hugetlbfs_file_operations); |
39b65252 | 1376 | if (IS_ERR(file)) |
b4d232e6 | 1377 | goto out_dentry; /* inode is already attached */ |
ce8d2cdf | 1378 | |
1da177e4 LT |
1379 | return file; |
1380 | ||
b45b5bd6 DG |
1381 | out_inode: |
1382 | iput(inode); | |
1da177e4 | 1383 | out_dentry: |
2c48b9c4 | 1384 | path_put(&path); |
1da177e4 | 1385 | out_shm_unlock: |
353d5c30 HD |
1386 | if (*user) { |
1387 | user_shm_unlock(size, *user); | |
1388 | *user = NULL; | |
1389 | } | |
39b65252 | 1390 | return file; |
1da177e4 LT |
1391 | } |
1392 | ||
1393 | static int __init init_hugetlbfs_fs(void) | |
1394 | { | |
42d7395f | 1395 | struct hstate *h; |
1da177e4 | 1396 | int error; |
42d7395f | 1397 | int i; |
1da177e4 | 1398 | |
457c1b27 | 1399 | if (!hugepages_supported()) { |
9b857d26 | 1400 | pr_info("disabling because there are no supported hugepage sizes\n"); |
457c1b27 NA |
1401 | return -ENOTSUPP; |
1402 | } | |
1403 | ||
d1d5e05f | 1404 | error = -ENOMEM; |
1da177e4 LT |
1405 | hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache", |
1406 | sizeof(struct hugetlbfs_inode_info), | |
5d097056 | 1407 | 0, SLAB_ACCOUNT, init_once); |
1da177e4 | 1408 | if (hugetlbfs_inode_cachep == NULL) |
e0bf68dd | 1409 | goto out2; |
1da177e4 LT |
1410 | |
1411 | error = register_filesystem(&hugetlbfs_fs_type); | |
1412 | if (error) | |
1413 | goto out; | |
1414 | ||
42d7395f AK |
1415 | i = 0; |
1416 | for_each_hstate(h) { | |
1417 | char buf[50]; | |
1418 | unsigned ps_kb = 1U << (h->order + PAGE_SHIFT - 10); | |
1da177e4 | 1419 | |
42d7395f AK |
1420 | snprintf(buf, sizeof(buf), "pagesize=%uK", ps_kb); |
1421 | hugetlbfs_vfsmount[i] = kern_mount_data(&hugetlbfs_fs_type, | |
1422 | buf); | |
1da177e4 | 1423 | |
42d7395f | 1424 | if (IS_ERR(hugetlbfs_vfsmount[i])) { |
9b857d26 | 1425 | pr_err("Cannot mount internal hugetlbfs for " |
42d7395f AK |
1426 | "page size %uK", ps_kb); |
1427 | error = PTR_ERR(hugetlbfs_vfsmount[i]); | |
1428 | hugetlbfs_vfsmount[i] = NULL; | |
1429 | } | |
1430 | i++; | |
1431 | } | |
1432 | /* Non default hstates are optional */ | |
1433 | if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount[default_hstate_idx])) | |
1434 | return 0; | |
1da177e4 LT |
1435 | |
1436 | out: | |
d1d5e05f | 1437 | kmem_cache_destroy(hugetlbfs_inode_cachep); |
e0bf68dd | 1438 | out2: |
1da177e4 LT |
1439 | return error; |
1440 | } | |
3e89e1c5 | 1441 | fs_initcall(init_hugetlbfs_fs) |