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. | |
7 | */ | |
8 | ||
9b857d26 AM |
9 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
10 | ||
1da177e4 LT |
11 | #include <linux/module.h> |
12 | #include <linux/thread_info.h> | |
13 | #include <asm/current.h> | |
14 | #include <linux/sched.h> /* remove ASAP */ | |
15 | #include <linux/fs.h> | |
16 | #include <linux/mount.h> | |
17 | #include <linux/file.h> | |
e73a75fa | 18 | #include <linux/kernel.h> |
1da177e4 LT |
19 | #include <linux/writeback.h> |
20 | #include <linux/pagemap.h> | |
21 | #include <linux/highmem.h> | |
22 | #include <linux/init.h> | |
23 | #include <linux/string.h> | |
16f7e0fe | 24 | #include <linux/capability.h> |
e73a75fa | 25 | #include <linux/ctype.h> |
1da177e4 LT |
26 | #include <linux/backing-dev.h> |
27 | #include <linux/hugetlb.h> | |
28 | #include <linux/pagevec.h> | |
e73a75fa | 29 | #include <linux/parser.h> |
036e0856 | 30 | #include <linux/mman.h> |
1da177e4 LT |
31 | #include <linux/slab.h> |
32 | #include <linux/dnotify.h> | |
33 | #include <linux/statfs.h> | |
34 | #include <linux/security.h> | |
1fd7317d | 35 | #include <linux/magic.h> |
290408d4 | 36 | #include <linux/migrate.h> |
34d0640e | 37 | #include <linux/uio.h> |
1da177e4 LT |
38 | |
39 | #include <asm/uaccess.h> | |
40 | ||
ee9b6d61 | 41 | static const struct super_operations hugetlbfs_ops; |
f5e54d6e | 42 | static const struct address_space_operations hugetlbfs_aops; |
4b6f5d20 | 43 | const struct file_operations hugetlbfs_file_operations; |
92e1d5be AV |
44 | static const struct inode_operations hugetlbfs_dir_inode_operations; |
45 | static const struct inode_operations hugetlbfs_inode_operations; | |
1da177e4 | 46 | |
a1d776ee | 47 | struct hugetlbfs_config { |
a0eb3a05 EB |
48 | kuid_t uid; |
49 | kgid_t gid; | |
a1d776ee DG |
50 | umode_t mode; |
51 | long nr_blocks; | |
52 | long nr_inodes; | |
53 | struct hstate *hstate; | |
54 | }; | |
55 | ||
56 | struct hugetlbfs_inode_info { | |
57 | struct shared_policy policy; | |
58 | struct inode vfs_inode; | |
59 | }; | |
60 | ||
61 | static inline struct hugetlbfs_inode_info *HUGETLBFS_I(struct inode *inode) | |
62 | { | |
63 | return container_of(inode, struct hugetlbfs_inode_info, vfs_inode); | |
64 | } | |
65 | ||
1da177e4 LT |
66 | int sysctl_hugetlb_shm_group; |
67 | ||
e73a75fa RD |
68 | enum { |
69 | Opt_size, Opt_nr_inodes, | |
70 | Opt_mode, Opt_uid, Opt_gid, | |
a137e1cc | 71 | Opt_pagesize, |
e73a75fa RD |
72 | Opt_err, |
73 | }; | |
74 | ||
a447c093 | 75 | static const match_table_t tokens = { |
e73a75fa RD |
76 | {Opt_size, "size=%s"}, |
77 | {Opt_nr_inodes, "nr_inodes=%s"}, | |
78 | {Opt_mode, "mode=%o"}, | |
79 | {Opt_uid, "uid=%u"}, | |
80 | {Opt_gid, "gid=%u"}, | |
a137e1cc | 81 | {Opt_pagesize, "pagesize=%s"}, |
e73a75fa RD |
82 | {Opt_err, NULL}, |
83 | }; | |
84 | ||
2e9b367c AL |
85 | static void huge_pagevec_release(struct pagevec *pvec) |
86 | { | |
87 | int i; | |
88 | ||
89 | for (i = 0; i < pagevec_count(pvec); ++i) | |
90 | put_page(pvec->pages[i]); | |
91 | ||
92 | pagevec_reinit(pvec); | |
93 | } | |
94 | ||
1da177e4 LT |
95 | static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma) |
96 | { | |
496ad9aa | 97 | struct inode *inode = file_inode(file); |
1da177e4 LT |
98 | loff_t len, vma_len; |
99 | int ret; | |
a5516438 | 100 | struct hstate *h = hstate_file(file); |
1da177e4 | 101 | |
68589bc3 | 102 | /* |
dec4ad86 DG |
103 | * vma address alignment (but not the pgoff alignment) has |
104 | * already been checked by prepare_hugepage_range. If you add | |
105 | * any error returns here, do so after setting VM_HUGETLB, so | |
106 | * is_vm_hugetlb_page tests below unmap_region go the right | |
107 | * way when do_mmap_pgoff unwinds (may be important on powerpc | |
108 | * and ia64). | |
68589bc3 | 109 | */ |
a2fce914 | 110 | vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND; |
68589bc3 | 111 | vma->vm_ops = &hugetlb_vm_ops; |
1da177e4 | 112 | |
2b37c35e | 113 | if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT)) |
dec4ad86 DG |
114 | return -EINVAL; |
115 | ||
1da177e4 LT |
116 | vma_len = (loff_t)(vma->vm_end - vma->vm_start); |
117 | ||
1b1dcc1b | 118 | mutex_lock(&inode->i_mutex); |
1da177e4 | 119 | file_accessed(file); |
1da177e4 LT |
120 | |
121 | ret = -ENOMEM; | |
122 | len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); | |
1da177e4 | 123 | |
a1e78772 | 124 | if (hugetlb_reserve_pages(inode, |
a5516438 | 125 | vma->vm_pgoff >> huge_page_order(h), |
5a6fe125 MG |
126 | len >> huge_page_shift(h), vma, |
127 | vma->vm_flags)) | |
a43a8c39 | 128 | goto out; |
b45b5bd6 | 129 | |
4c887265 AL |
130 | ret = 0; |
131 | hugetlb_prefault_arch_hook(vma->vm_mm); | |
b6174df5 | 132 | if (vma->vm_flags & VM_WRITE && inode->i_size < len) |
1da177e4 LT |
133 | inode->i_size = len; |
134 | out: | |
1b1dcc1b | 135 | mutex_unlock(&inode->i_mutex); |
1da177e4 LT |
136 | |
137 | return ret; | |
138 | } | |
139 | ||
140 | /* | |
508034a3 | 141 | * Called under down_write(mmap_sem). |
1da177e4 LT |
142 | */ |
143 | ||
d2ba27e8 | 144 | #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA |
1da177e4 LT |
145 | static unsigned long |
146 | hugetlb_get_unmapped_area(struct file *file, unsigned long addr, | |
147 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
148 | { | |
149 | struct mm_struct *mm = current->mm; | |
150 | struct vm_area_struct *vma; | |
a5516438 | 151 | struct hstate *h = hstate_file(file); |
08659355 | 152 | struct vm_unmapped_area_info info; |
1da177e4 | 153 | |
a5516438 | 154 | if (len & ~huge_page_mask(h)) |
1da177e4 LT |
155 | return -EINVAL; |
156 | if (len > TASK_SIZE) | |
157 | return -ENOMEM; | |
158 | ||
036e0856 | 159 | if (flags & MAP_FIXED) { |
a5516438 | 160 | if (prepare_hugepage_range(file, addr, len)) |
036e0856 BH |
161 | return -EINVAL; |
162 | return addr; | |
163 | } | |
164 | ||
1da177e4 | 165 | if (addr) { |
a5516438 | 166 | addr = ALIGN(addr, huge_page_size(h)); |
1da177e4 LT |
167 | vma = find_vma(mm, addr); |
168 | if (TASK_SIZE - len >= addr && | |
169 | (!vma || addr + len <= vma->vm_start)) | |
170 | return addr; | |
171 | } | |
172 | ||
08659355 ML |
173 | info.flags = 0; |
174 | info.length = len; | |
175 | info.low_limit = TASK_UNMAPPED_BASE; | |
176 | info.high_limit = TASK_SIZE; | |
177 | info.align_mask = PAGE_MASK & ~huge_page_mask(h); | |
178 | info.align_offset = 0; | |
179 | return vm_unmapped_area(&info); | |
1da177e4 LT |
180 | } |
181 | #endif | |
182 | ||
34d0640e | 183 | static size_t |
e63e1e5a | 184 | hugetlbfs_read_actor(struct page *page, unsigned long offset, |
34d0640e | 185 | struct iov_iter *to, unsigned long size) |
e63e1e5a | 186 | { |
34d0640e | 187 | size_t copied = 0; |
e63e1e5a BP |
188 | int i, chunksize; |
189 | ||
e63e1e5a BP |
190 | /* Find which 4k chunk and offset with in that chunk */ |
191 | i = offset >> PAGE_CACHE_SHIFT; | |
192 | offset = offset & ~PAGE_CACHE_MASK; | |
193 | ||
194 | while (size) { | |
34d0640e | 195 | size_t n; |
e63e1e5a BP |
196 | chunksize = PAGE_CACHE_SIZE; |
197 | if (offset) | |
198 | chunksize -= offset; | |
199 | if (chunksize > size) | |
200 | chunksize = size; | |
34d0640e AV |
201 | n = copy_page_to_iter(&page[i], offset, chunksize, to); |
202 | copied += n; | |
203 | if (n != chunksize) | |
204 | return copied; | |
e63e1e5a BP |
205 | offset = 0; |
206 | size -= chunksize; | |
e63e1e5a BP |
207 | i++; |
208 | } | |
34d0640e | 209 | return copied; |
e63e1e5a BP |
210 | } |
211 | ||
212 | /* | |
213 | * Support for read() - Find the page attached to f_mapping and copy out the | |
214 | * data. Its *very* similar to do_generic_mapping_read(), we can't use that | |
215 | * since it has PAGE_CACHE_SIZE assumptions. | |
216 | */ | |
34d0640e | 217 | static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to) |
e63e1e5a | 218 | { |
34d0640e AV |
219 | struct file *file = iocb->ki_filp; |
220 | struct hstate *h = hstate_file(file); | |
221 | struct address_space *mapping = file->f_mapping; | |
e63e1e5a | 222 | struct inode *inode = mapping->host; |
34d0640e AV |
223 | unsigned long index = iocb->ki_pos >> huge_page_shift(h); |
224 | unsigned long offset = iocb->ki_pos & ~huge_page_mask(h); | |
e63e1e5a BP |
225 | unsigned long end_index; |
226 | loff_t isize; | |
227 | ssize_t retval = 0; | |
228 | ||
34d0640e | 229 | while (iov_iter_count(to)) { |
e63e1e5a | 230 | struct page *page; |
34d0640e | 231 | size_t nr, copied; |
e63e1e5a BP |
232 | |
233 | /* nr is the maximum number of bytes to copy from this page */ | |
a5516438 | 234 | nr = huge_page_size(h); |
a05b0855 AK |
235 | isize = i_size_read(inode); |
236 | if (!isize) | |
34d0640e | 237 | break; |
a05b0855 | 238 | end_index = (isize - 1) >> huge_page_shift(h); |
34d0640e AV |
239 | if (index > end_index) |
240 | break; | |
241 | if (index == end_index) { | |
a5516438 | 242 | nr = ((isize - 1) & ~huge_page_mask(h)) + 1; |
a05b0855 | 243 | if (nr <= offset) |
34d0640e | 244 | break; |
e63e1e5a BP |
245 | } |
246 | nr = nr - offset; | |
247 | ||
248 | /* Find the page */ | |
a05b0855 | 249 | page = find_lock_page(mapping, index); |
e63e1e5a BP |
250 | if (unlikely(page == NULL)) { |
251 | /* | |
252 | * We have a HOLE, zero out the user-buffer for the | |
253 | * length of the hole or request. | |
254 | */ | |
34d0640e | 255 | copied = iov_iter_zero(nr, to); |
e63e1e5a | 256 | } else { |
a05b0855 AK |
257 | unlock_page(page); |
258 | ||
e63e1e5a BP |
259 | /* |
260 | * We have the page, copy it to user space buffer. | |
261 | */ | |
34d0640e | 262 | copied = hugetlbfs_read_actor(page, offset, to, nr); |
a05b0855 | 263 | page_cache_release(page); |
e63e1e5a | 264 | } |
34d0640e AV |
265 | offset += copied; |
266 | retval += copied; | |
267 | if (copied != nr && iov_iter_count(to)) { | |
268 | if (!retval) | |
269 | retval = -EFAULT; | |
270 | break; | |
e63e1e5a | 271 | } |
a5516438 AK |
272 | index += offset >> huge_page_shift(h); |
273 | offset &= ~huge_page_mask(h); | |
e63e1e5a | 274 | } |
34d0640e | 275 | iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset; |
e63e1e5a BP |
276 | return retval; |
277 | } | |
278 | ||
800d15a5 NP |
279 | static int hugetlbfs_write_begin(struct file *file, |
280 | struct address_space *mapping, | |
281 | loff_t pos, unsigned len, unsigned flags, | |
282 | struct page **pagep, void **fsdata) | |
1da177e4 LT |
283 | { |
284 | return -EINVAL; | |
285 | } | |
286 | ||
800d15a5 NP |
287 | static int hugetlbfs_write_end(struct file *file, struct address_space *mapping, |
288 | loff_t pos, unsigned len, unsigned copied, | |
289 | struct page *page, void *fsdata) | |
1da177e4 | 290 | { |
800d15a5 | 291 | BUG(); |
1da177e4 LT |
292 | return -EINVAL; |
293 | } | |
294 | ||
1da177e4 LT |
295 | static void truncate_huge_page(struct page *page) |
296 | { | |
fba2591b | 297 | cancel_dirty_page(page, /* No IO accounting for huge pages? */0); |
1da177e4 | 298 | ClearPageUptodate(page); |
bd65cb86 | 299 | delete_from_page_cache(page); |
1da177e4 LT |
300 | } |
301 | ||
b45b5bd6 | 302 | static void truncate_hugepages(struct inode *inode, loff_t lstart) |
1da177e4 | 303 | { |
a5516438 | 304 | struct hstate *h = hstate_inode(inode); |
b45b5bd6 | 305 | struct address_space *mapping = &inode->i_data; |
a5516438 | 306 | const pgoff_t start = lstart >> huge_page_shift(h); |
1da177e4 LT |
307 | struct pagevec pvec; |
308 | pgoff_t next; | |
a43a8c39 | 309 | int i, freed = 0; |
1da177e4 LT |
310 | |
311 | pagevec_init(&pvec, 0); | |
312 | next = start; | |
313 | while (1) { | |
314 | if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { | |
315 | if (next == start) | |
316 | break; | |
317 | next = start; | |
318 | continue; | |
319 | } | |
320 | ||
321 | for (i = 0; i < pagevec_count(&pvec); ++i) { | |
322 | struct page *page = pvec.pages[i]; | |
323 | ||
324 | lock_page(page); | |
325 | if (page->index > next) | |
326 | next = page->index; | |
327 | ++next; | |
328 | truncate_huge_page(page); | |
329 | unlock_page(page); | |
a43a8c39 | 330 | freed++; |
1da177e4 LT |
331 | } |
332 | huge_pagevec_release(&pvec); | |
333 | } | |
334 | BUG_ON(!lstart && mapping->nrpages); | |
a43a8c39 | 335 | hugetlb_unreserve_pages(inode, start, freed); |
1da177e4 LT |
336 | } |
337 | ||
2bbbda30 | 338 | static void hugetlbfs_evict_inode(struct inode *inode) |
1da177e4 | 339 | { |
9119a41e JK |
340 | struct resv_map *resv_map; |
341 | ||
b45b5bd6 | 342 | truncate_hugepages(inode, 0); |
9119a41e JK |
343 | resv_map = (struct resv_map *)inode->i_mapping->private_data; |
344 | /* root inode doesn't have the resv_map, so we should check it */ | |
345 | if (resv_map) | |
346 | resv_map_release(&resv_map->refs); | |
dbd5768f | 347 | clear_inode(inode); |
149f4211 CH |
348 | } |
349 | ||
1da177e4 | 350 | static inline void |
6b2dbba8 | 351 | hugetlb_vmtruncate_list(struct rb_root *root, pgoff_t pgoff) |
1da177e4 LT |
352 | { |
353 | struct vm_area_struct *vma; | |
1da177e4 | 354 | |
6b2dbba8 | 355 | vma_interval_tree_foreach(vma, root, pgoff, ULONG_MAX) { |
1da177e4 LT |
356 | unsigned long v_offset; |
357 | ||
1da177e4 | 358 | /* |
856fc295 | 359 | * Can the expression below overflow on 32-bit arches? |
6b2dbba8 | 360 | * No, because the interval tree returns us only those vmas |
856fc295 HD |
361 | * which overlap the truncated area starting at pgoff, |
362 | * and no vma on a 32-bit arch can span beyond the 4GB. | |
1da177e4 | 363 | */ |
856fc295 HD |
364 | if (vma->vm_pgoff < pgoff) |
365 | v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT; | |
366 | else | |
1da177e4 LT |
367 | v_offset = 0; |
368 | ||
24669e58 AK |
369 | unmap_hugepage_range(vma, vma->vm_start + v_offset, |
370 | vma->vm_end, NULL); | |
1da177e4 LT |
371 | } |
372 | } | |
373 | ||
1da177e4 LT |
374 | static int hugetlb_vmtruncate(struct inode *inode, loff_t offset) |
375 | { | |
856fc295 | 376 | pgoff_t pgoff; |
1da177e4 | 377 | struct address_space *mapping = inode->i_mapping; |
a5516438 | 378 | struct hstate *h = hstate_inode(inode); |
1da177e4 | 379 | |
a5516438 | 380 | BUG_ON(offset & ~huge_page_mask(h)); |
856fc295 | 381 | pgoff = offset >> PAGE_SHIFT; |
1da177e4 | 382 | |
7aa91e10 | 383 | i_size_write(inode, offset); |
83cde9e8 | 384 | i_mmap_lock_write(mapping); |
6b2dbba8 | 385 | if (!RB_EMPTY_ROOT(&mapping->i_mmap)) |
1da177e4 | 386 | hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff); |
83cde9e8 | 387 | i_mmap_unlock_write(mapping); |
b45b5bd6 | 388 | truncate_hugepages(inode, offset); |
1da177e4 LT |
389 | return 0; |
390 | } | |
391 | ||
392 | static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr) | |
393 | { | |
2b0143b5 | 394 | struct inode *inode = d_inode(dentry); |
a5516438 | 395 | struct hstate *h = hstate_inode(inode); |
1da177e4 LT |
396 | int error; |
397 | unsigned int ia_valid = attr->ia_valid; | |
398 | ||
399 | BUG_ON(!inode); | |
400 | ||
401 | error = inode_change_ok(inode, attr); | |
402 | if (error) | |
1025774c | 403 | return error; |
1da177e4 LT |
404 | |
405 | if (ia_valid & ATTR_SIZE) { | |
406 | error = -EINVAL; | |
1025774c CH |
407 | if (attr->ia_size & ~huge_page_mask(h)) |
408 | return -EINVAL; | |
409 | error = hugetlb_vmtruncate(inode, attr->ia_size); | |
1da177e4 | 410 | if (error) |
1025774c | 411 | return error; |
1da177e4 | 412 | } |
1025774c CH |
413 | |
414 | setattr_copy(inode, attr); | |
415 | mark_inode_dirty(inode); | |
416 | return 0; | |
1da177e4 LT |
417 | } |
418 | ||
7d54fa64 AV |
419 | static struct inode *hugetlbfs_get_root(struct super_block *sb, |
420 | struct hugetlbfs_config *config) | |
1da177e4 LT |
421 | { |
422 | struct inode *inode; | |
1da177e4 LT |
423 | |
424 | inode = new_inode(sb); | |
425 | if (inode) { | |
426 | struct hugetlbfs_inode_info *info; | |
85fe4025 | 427 | inode->i_ino = get_next_ino(); |
7d54fa64 AV |
428 | inode->i_mode = S_IFDIR | config->mode; |
429 | inode->i_uid = config->uid; | |
430 | inode->i_gid = config->gid; | |
431 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
432 | info = HUGETLBFS_I(inode); | |
433 | mpol_shared_policy_init(&info->policy, NULL); | |
434 | inode->i_op = &hugetlbfs_dir_inode_operations; | |
435 | inode->i_fop = &simple_dir_operations; | |
436 | /* directory inodes start off with i_nlink == 2 (for "." entry) */ | |
437 | inc_nlink(inode); | |
65ed7601 | 438 | lockdep_annotate_inode_mutex_key(inode); |
7d54fa64 AV |
439 | } |
440 | return inode; | |
441 | } | |
442 | ||
b610ded7 | 443 | /* |
c8c06efa | 444 | * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never |
b610ded7 MH |
445 | * be taken from reclaim -- unlike regular filesystems. This needs an |
446 | * annotation because huge_pmd_share() does an allocation under | |
c8c06efa | 447 | * i_mmap_rwsem. |
b610ded7 | 448 | */ |
c8c06efa | 449 | static struct lock_class_key hugetlbfs_i_mmap_rwsem_key; |
b610ded7 | 450 | |
7d54fa64 AV |
451 | static struct inode *hugetlbfs_get_inode(struct super_block *sb, |
452 | struct inode *dir, | |
18df2252 | 453 | umode_t mode, dev_t dev) |
7d54fa64 AV |
454 | { |
455 | struct inode *inode; | |
9119a41e JK |
456 | struct resv_map *resv_map; |
457 | ||
458 | resv_map = resv_map_alloc(); | |
459 | if (!resv_map) | |
460 | return NULL; | |
7d54fa64 AV |
461 | |
462 | inode = new_inode(sb); | |
463 | if (inode) { | |
464 | struct hugetlbfs_inode_info *info; | |
465 | inode->i_ino = get_next_ino(); | |
466 | inode_init_owner(inode, dir, mode); | |
c8c06efa DB |
467 | lockdep_set_class(&inode->i_mapping->i_mmap_rwsem, |
468 | &hugetlbfs_i_mmap_rwsem_key); | |
1da177e4 | 469 | inode->i_mapping->a_ops = &hugetlbfs_aops; |
1da177e4 | 470 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
9119a41e | 471 | inode->i_mapping->private_data = resv_map; |
1da177e4 | 472 | info = HUGETLBFS_I(inode); |
6bfde05b EM |
473 | /* |
474 | * The policy is initialized here even if we are creating a | |
475 | * private inode because initialization simply creates an | |
476 | * an empty rb tree and calls spin_lock_init(), later when we | |
477 | * call mpol_free_shared_policy() it will just return because | |
478 | * the rb tree will still be empty. | |
479 | */ | |
71fe804b | 480 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
481 | switch (mode & S_IFMT) { |
482 | default: | |
483 | init_special_inode(inode, mode, dev); | |
484 | break; | |
485 | case S_IFREG: | |
486 | inode->i_op = &hugetlbfs_inode_operations; | |
487 | inode->i_fop = &hugetlbfs_file_operations; | |
488 | break; | |
489 | case S_IFDIR: | |
490 | inode->i_op = &hugetlbfs_dir_inode_operations; | |
491 | inode->i_fop = &simple_dir_operations; | |
492 | ||
493 | /* directory inodes start off with i_nlink == 2 (for "." entry) */ | |
d8c76e6f | 494 | inc_nlink(inode); |
1da177e4 LT |
495 | break; |
496 | case S_IFLNK: | |
497 | inode->i_op = &page_symlink_inode_operations; | |
498 | break; | |
499 | } | |
e096d0c7 | 500 | lockdep_annotate_inode_mutex_key(inode); |
9119a41e JK |
501 | } else |
502 | kref_put(&resv_map->refs, resv_map_release); | |
503 | ||
1da177e4 LT |
504 | return inode; |
505 | } | |
506 | ||
507 | /* | |
508 | * File creation. Allocate an inode, and we're done.. | |
509 | */ | |
510 | static int hugetlbfs_mknod(struct inode *dir, | |
1a67aafb | 511 | struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 LT |
512 | { |
513 | struct inode *inode; | |
514 | int error = -ENOSPC; | |
7d54fa64 AV |
515 | |
516 | inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev); | |
1da177e4 LT |
517 | if (inode) { |
518 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
519 | d_instantiate(dentry, inode); | |
520 | dget(dentry); /* Extra count - pin the dentry in core */ | |
521 | error = 0; | |
522 | } | |
523 | return error; | |
524 | } | |
525 | ||
18bb1db3 | 526 | static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
527 | { |
528 | int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0); | |
529 | if (!retval) | |
d8c76e6f | 530 | inc_nlink(dir); |
1da177e4 LT |
531 | return retval; |
532 | } | |
533 | ||
ebfc3b49 | 534 | static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl) |
1da177e4 LT |
535 | { |
536 | return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0); | |
537 | } | |
538 | ||
539 | static int hugetlbfs_symlink(struct inode *dir, | |
540 | struct dentry *dentry, const char *symname) | |
541 | { | |
542 | struct inode *inode; | |
543 | int error = -ENOSPC; | |
1da177e4 | 544 | |
7d54fa64 | 545 | inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0); |
1da177e4 LT |
546 | if (inode) { |
547 | int l = strlen(symname)+1; | |
548 | error = page_symlink(inode, symname, l); | |
549 | if (!error) { | |
550 | d_instantiate(dentry, inode); | |
551 | dget(dentry); | |
552 | } else | |
553 | iput(inode); | |
554 | } | |
555 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
556 | ||
557 | return error; | |
558 | } | |
559 | ||
560 | /* | |
6649a386 | 561 | * mark the head page dirty |
1da177e4 LT |
562 | */ |
563 | static int hugetlbfs_set_page_dirty(struct page *page) | |
564 | { | |
d85f3385 | 565 | struct page *head = compound_head(page); |
6649a386 KC |
566 | |
567 | SetPageDirty(head); | |
1da177e4 LT |
568 | return 0; |
569 | } | |
570 | ||
290408d4 | 571 | static int hugetlbfs_migrate_page(struct address_space *mapping, |
b969c4ab | 572 | struct page *newpage, struct page *page, |
a6bc32b8 | 573 | enum migrate_mode mode) |
290408d4 NH |
574 | { |
575 | int rc; | |
576 | ||
577 | rc = migrate_huge_page_move_mapping(mapping, newpage, page); | |
78bd5209 | 578 | if (rc != MIGRATEPAGE_SUCCESS) |
290408d4 NH |
579 | return rc; |
580 | migrate_page_copy(newpage, page); | |
581 | ||
78bd5209 | 582 | return MIGRATEPAGE_SUCCESS; |
290408d4 NH |
583 | } |
584 | ||
726c3342 | 585 | static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 586 | { |
726c3342 | 587 | struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb); |
2b0143b5 | 588 | struct hstate *h = hstate_inode(d_inode(dentry)); |
1da177e4 LT |
589 | |
590 | buf->f_type = HUGETLBFS_MAGIC; | |
a5516438 | 591 | buf->f_bsize = huge_page_size(h); |
1da177e4 LT |
592 | if (sbinfo) { |
593 | spin_lock(&sbinfo->stat_lock); | |
74a8a65c DG |
594 | /* If no limits set, just report 0 for max/free/used |
595 | * blocks, like simple_statfs() */ | |
90481622 DG |
596 | if (sbinfo->spool) { |
597 | long free_pages; | |
598 | ||
599 | spin_lock(&sbinfo->spool->lock); | |
600 | buf->f_blocks = sbinfo->spool->max_hpages; | |
601 | free_pages = sbinfo->spool->max_hpages | |
602 | - sbinfo->spool->used_hpages; | |
603 | buf->f_bavail = buf->f_bfree = free_pages; | |
604 | spin_unlock(&sbinfo->spool->lock); | |
74a8a65c DG |
605 | buf->f_files = sbinfo->max_inodes; |
606 | buf->f_ffree = sbinfo->free_inodes; | |
607 | } | |
1da177e4 LT |
608 | spin_unlock(&sbinfo->stat_lock); |
609 | } | |
610 | buf->f_namelen = NAME_MAX; | |
611 | return 0; | |
612 | } | |
613 | ||
614 | static void hugetlbfs_put_super(struct super_block *sb) | |
615 | { | |
616 | struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb); | |
617 | ||
618 | if (sbi) { | |
619 | sb->s_fs_info = NULL; | |
90481622 DG |
620 | |
621 | if (sbi->spool) | |
622 | hugepage_put_subpool(sbi->spool); | |
623 | ||
1da177e4 LT |
624 | kfree(sbi); |
625 | } | |
626 | } | |
627 | ||
96527980 CH |
628 | static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo) |
629 | { | |
630 | if (sbinfo->free_inodes >= 0) { | |
631 | spin_lock(&sbinfo->stat_lock); | |
632 | if (unlikely(!sbinfo->free_inodes)) { | |
633 | spin_unlock(&sbinfo->stat_lock); | |
634 | return 0; | |
635 | } | |
636 | sbinfo->free_inodes--; | |
637 | spin_unlock(&sbinfo->stat_lock); | |
638 | } | |
639 | ||
640 | return 1; | |
641 | } | |
642 | ||
643 | static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo) | |
644 | { | |
645 | if (sbinfo->free_inodes >= 0) { | |
646 | spin_lock(&sbinfo->stat_lock); | |
647 | sbinfo->free_inodes++; | |
648 | spin_unlock(&sbinfo->stat_lock); | |
649 | } | |
650 | } | |
651 | ||
652 | ||
e18b890b | 653 | static struct kmem_cache *hugetlbfs_inode_cachep; |
1da177e4 LT |
654 | |
655 | static struct inode *hugetlbfs_alloc_inode(struct super_block *sb) | |
656 | { | |
96527980 | 657 | struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb); |
1da177e4 LT |
658 | struct hugetlbfs_inode_info *p; |
659 | ||
96527980 CH |
660 | if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo))) |
661 | return NULL; | |
e94b1766 | 662 | p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL); |
96527980 CH |
663 | if (unlikely(!p)) { |
664 | hugetlbfs_inc_free_inodes(sbinfo); | |
1da177e4 | 665 | return NULL; |
96527980 | 666 | } |
1da177e4 LT |
667 | return &p->vfs_inode; |
668 | } | |
669 | ||
fa0d7e3d NP |
670 | static void hugetlbfs_i_callback(struct rcu_head *head) |
671 | { | |
672 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
673 | kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode)); |
674 | } | |
675 | ||
1da177e4 LT |
676 | static void hugetlbfs_destroy_inode(struct inode *inode) |
677 | { | |
96527980 | 678 | hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb)); |
1da177e4 | 679 | mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy); |
fa0d7e3d | 680 | call_rcu(&inode->i_rcu, hugetlbfs_i_callback); |
1da177e4 LT |
681 | } |
682 | ||
f5e54d6e | 683 | static const struct address_space_operations hugetlbfs_aops = { |
800d15a5 NP |
684 | .write_begin = hugetlbfs_write_begin, |
685 | .write_end = hugetlbfs_write_end, | |
1da177e4 | 686 | .set_page_dirty = hugetlbfs_set_page_dirty, |
290408d4 | 687 | .migratepage = hugetlbfs_migrate_page, |
1da177e4 LT |
688 | }; |
689 | ||
96527980 | 690 | |
51cc5068 | 691 | static void init_once(void *foo) |
96527980 CH |
692 | { |
693 | struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo; | |
694 | ||
a35afb83 | 695 | inode_init_once(&ei->vfs_inode); |
96527980 CH |
696 | } |
697 | ||
4b6f5d20 | 698 | const struct file_operations hugetlbfs_file_operations = { |
34d0640e | 699 | .read_iter = hugetlbfs_read_iter, |
1da177e4 | 700 | .mmap = hugetlbfs_file_mmap, |
1b061d92 | 701 | .fsync = noop_fsync, |
1da177e4 | 702 | .get_unmapped_area = hugetlb_get_unmapped_area, |
6038f373 | 703 | .llseek = default_llseek, |
1da177e4 LT |
704 | }; |
705 | ||
92e1d5be | 706 | static const struct inode_operations hugetlbfs_dir_inode_operations = { |
1da177e4 LT |
707 | .create = hugetlbfs_create, |
708 | .lookup = simple_lookup, | |
709 | .link = simple_link, | |
710 | .unlink = simple_unlink, | |
711 | .symlink = hugetlbfs_symlink, | |
712 | .mkdir = hugetlbfs_mkdir, | |
713 | .rmdir = simple_rmdir, | |
714 | .mknod = hugetlbfs_mknod, | |
715 | .rename = simple_rename, | |
716 | .setattr = hugetlbfs_setattr, | |
717 | }; | |
718 | ||
92e1d5be | 719 | static const struct inode_operations hugetlbfs_inode_operations = { |
1da177e4 LT |
720 | .setattr = hugetlbfs_setattr, |
721 | }; | |
722 | ||
ee9b6d61 | 723 | static const struct super_operations hugetlbfs_ops = { |
1da177e4 LT |
724 | .alloc_inode = hugetlbfs_alloc_inode, |
725 | .destroy_inode = hugetlbfs_destroy_inode, | |
2bbbda30 | 726 | .evict_inode = hugetlbfs_evict_inode, |
1da177e4 | 727 | .statfs = hugetlbfs_statfs, |
1da177e4 | 728 | .put_super = hugetlbfs_put_super, |
10f19a86 | 729 | .show_options = generic_show_options, |
1da177e4 LT |
730 | }; |
731 | ||
732 | static int | |
733 | hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig) | |
734 | { | |
e73a75fa RD |
735 | char *p, *rest; |
736 | substring_t args[MAX_OPT_ARGS]; | |
737 | int option; | |
a137e1cc AK |
738 | unsigned long long size = 0; |
739 | enum { NO_SIZE, SIZE_STD, SIZE_PERCENT } setsize = NO_SIZE; | |
1da177e4 LT |
740 | |
741 | if (!options) | |
742 | return 0; | |
1da177e4 | 743 | |
e73a75fa RD |
744 | while ((p = strsep(&options, ",")) != NULL) { |
745 | int token; | |
b4c07bce LS |
746 | if (!*p) |
747 | continue; | |
e73a75fa RD |
748 | |
749 | token = match_token(p, tokens, args); | |
750 | switch (token) { | |
751 | case Opt_uid: | |
752 | if (match_int(&args[0], &option)) | |
753 | goto bad_val; | |
a0eb3a05 EB |
754 | pconfig->uid = make_kuid(current_user_ns(), option); |
755 | if (!uid_valid(pconfig->uid)) | |
756 | goto bad_val; | |
e73a75fa RD |
757 | break; |
758 | ||
759 | case Opt_gid: | |
760 | if (match_int(&args[0], &option)) | |
761 | goto bad_val; | |
a0eb3a05 EB |
762 | pconfig->gid = make_kgid(current_user_ns(), option); |
763 | if (!gid_valid(pconfig->gid)) | |
764 | goto bad_val; | |
e73a75fa RD |
765 | break; |
766 | ||
767 | case Opt_mode: | |
768 | if (match_octal(&args[0], &option)) | |
769 | goto bad_val; | |
75897d60 | 770 | pconfig->mode = option & 01777U; |
e73a75fa RD |
771 | break; |
772 | ||
773 | case Opt_size: { | |
e73a75fa RD |
774 | /* memparse() will accept a K/M/G without a digit */ |
775 | if (!isdigit(*args[0].from)) | |
776 | goto bad_val; | |
777 | size = memparse(args[0].from, &rest); | |
a137e1cc AK |
778 | setsize = SIZE_STD; |
779 | if (*rest == '%') | |
780 | setsize = SIZE_PERCENT; | |
e73a75fa RD |
781 | break; |
782 | } | |
1da177e4 | 783 | |
e73a75fa RD |
784 | case Opt_nr_inodes: |
785 | /* memparse() will accept a K/M/G without a digit */ | |
786 | if (!isdigit(*args[0].from)) | |
787 | goto bad_val; | |
788 | pconfig->nr_inodes = memparse(args[0].from, &rest); | |
789 | break; | |
790 | ||
a137e1cc AK |
791 | case Opt_pagesize: { |
792 | unsigned long ps; | |
793 | ps = memparse(args[0].from, &rest); | |
794 | pconfig->hstate = size_to_hstate(ps); | |
795 | if (!pconfig->hstate) { | |
9b857d26 | 796 | pr_err("Unsupported page size %lu MB\n", |
a137e1cc AK |
797 | ps >> 20); |
798 | return -EINVAL; | |
799 | } | |
800 | break; | |
801 | } | |
802 | ||
e73a75fa | 803 | default: |
9b857d26 | 804 | pr_err("Bad mount option: \"%s\"\n", p); |
b4c07bce | 805 | return -EINVAL; |
e73a75fa RD |
806 | break; |
807 | } | |
1da177e4 | 808 | } |
a137e1cc AK |
809 | |
810 | /* Do size after hstate is set up */ | |
811 | if (setsize > NO_SIZE) { | |
812 | struct hstate *h = pconfig->hstate; | |
813 | if (setsize == SIZE_PERCENT) { | |
814 | size <<= huge_page_shift(h); | |
815 | size *= h->max_huge_pages; | |
816 | do_div(size, 100); | |
817 | } | |
818 | pconfig->nr_blocks = (size >> huge_page_shift(h)); | |
819 | } | |
820 | ||
1da177e4 | 821 | return 0; |
e73a75fa RD |
822 | |
823 | bad_val: | |
9b857d26 | 824 | pr_err("Bad value '%s' for mount option '%s'\n", args[0].from, p); |
c12ddba0 | 825 | return -EINVAL; |
1da177e4 LT |
826 | } |
827 | ||
828 | static int | |
829 | hugetlbfs_fill_super(struct super_block *sb, void *data, int silent) | |
830 | { | |
1da177e4 LT |
831 | int ret; |
832 | struct hugetlbfs_config config; | |
833 | struct hugetlbfs_sb_info *sbinfo; | |
834 | ||
10f19a86 MS |
835 | save_mount_options(sb, data); |
836 | ||
1da177e4 LT |
837 | config.nr_blocks = -1; /* No limit on size by default */ |
838 | config.nr_inodes = -1; /* No limit on number of inodes by default */ | |
77c70de1 DH |
839 | config.uid = current_fsuid(); |
840 | config.gid = current_fsgid(); | |
1da177e4 | 841 | config.mode = 0755; |
a137e1cc | 842 | config.hstate = &default_hstate; |
1da177e4 | 843 | ret = hugetlbfs_parse_options(data, &config); |
1da177e4 LT |
844 | if (ret) |
845 | return ret; | |
846 | ||
847 | sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL); | |
848 | if (!sbinfo) | |
849 | return -ENOMEM; | |
850 | sb->s_fs_info = sbinfo; | |
a137e1cc | 851 | sbinfo->hstate = config.hstate; |
1da177e4 | 852 | spin_lock_init(&sbinfo->stat_lock); |
1da177e4 LT |
853 | sbinfo->max_inodes = config.nr_inodes; |
854 | sbinfo->free_inodes = config.nr_inodes; | |
90481622 DG |
855 | sbinfo->spool = NULL; |
856 | if (config.nr_blocks != -1) { | |
857 | sbinfo->spool = hugepage_new_subpool(config.nr_blocks); | |
858 | if (!sbinfo->spool) | |
859 | goto out_free; | |
860 | } | |
1da177e4 | 861 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
a137e1cc AK |
862 | sb->s_blocksize = huge_page_size(config.hstate); |
863 | sb->s_blocksize_bits = huge_page_shift(config.hstate); | |
1da177e4 LT |
864 | sb->s_magic = HUGETLBFS_MAGIC; |
865 | sb->s_op = &hugetlbfs_ops; | |
866 | sb->s_time_gran = 1; | |
48fde701 AV |
867 | sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config)); |
868 | if (!sb->s_root) | |
1da177e4 | 869 | goto out_free; |
1da177e4 LT |
870 | return 0; |
871 | out_free: | |
6e6870d4 | 872 | kfree(sbinfo->spool); |
1da177e4 LT |
873 | kfree(sbinfo); |
874 | return -ENOMEM; | |
875 | } | |
876 | ||
3c26ff6e AV |
877 | static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type, |
878 | int flags, const char *dev_name, void *data) | |
1da177e4 | 879 | { |
3c26ff6e | 880 | return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super); |
1da177e4 LT |
881 | } |
882 | ||
883 | static struct file_system_type hugetlbfs_fs_type = { | |
884 | .name = "hugetlbfs", | |
3c26ff6e | 885 | .mount = hugetlbfs_mount, |
1da177e4 LT |
886 | .kill_sb = kill_litter_super, |
887 | }; | |
7f78e035 | 888 | MODULE_ALIAS_FS("hugetlbfs"); |
1da177e4 | 889 | |
42d7395f | 890 | static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE]; |
1da177e4 | 891 | |
ef1ff6b8 | 892 | static int can_do_hugetlb_shm(void) |
1da177e4 | 893 | { |
a0eb3a05 EB |
894 | kgid_t shm_group; |
895 | shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group); | |
896 | return capable(CAP_IPC_LOCK) || in_group_p(shm_group); | |
1da177e4 LT |
897 | } |
898 | ||
42d7395f AK |
899 | static int get_hstate_idx(int page_size_log) |
900 | { | |
af73e4d9 | 901 | struct hstate *h = hstate_sizelog(page_size_log); |
42d7395f | 902 | |
42d7395f AK |
903 | if (!h) |
904 | return -1; | |
905 | return h - hstates; | |
906 | } | |
907 | ||
be1d2cf5 | 908 | static const struct dentry_operations anon_ops = { |
118b2302 | 909 | .d_dname = simple_dname |
0df4d6e5 AV |
910 | }; |
911 | ||
af73e4d9 NH |
912 | /* |
913 | * Note that size should be aligned to proper hugepage size in caller side, | |
914 | * otherwise hugetlb_reserve_pages reserves one less hugepages than intended. | |
915 | */ | |
916 | struct file *hugetlb_file_setup(const char *name, size_t size, | |
917 | vm_flags_t acctflag, struct user_struct **user, | |
42d7395f | 918 | int creat_flags, int page_size_log) |
1da177e4 | 919 | { |
39b65252 | 920 | struct file *file = ERR_PTR(-ENOMEM); |
1da177e4 | 921 | struct inode *inode; |
2c48b9c4 | 922 | struct path path; |
0df4d6e5 | 923 | struct super_block *sb; |
1da177e4 | 924 | struct qstr quick_string; |
42d7395f AK |
925 | int hstate_idx; |
926 | ||
927 | hstate_idx = get_hstate_idx(page_size_log); | |
928 | if (hstate_idx < 0) | |
929 | return ERR_PTR(-ENODEV); | |
1da177e4 | 930 | |
353d5c30 | 931 | *user = NULL; |
42d7395f | 932 | if (!hugetlbfs_vfsmount[hstate_idx]) |
5bc98594 AM |
933 | return ERR_PTR(-ENOENT); |
934 | ||
ef1ff6b8 | 935 | if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) { |
353d5c30 HD |
936 | *user = current_user(); |
937 | if (user_shm_lock(size, *user)) { | |
21a3c273 | 938 | task_lock(current); |
9b857d26 | 939 | pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n", |
21a3c273 DR |
940 | current->comm, current->pid); |
941 | task_unlock(current); | |
353d5c30 HD |
942 | } else { |
943 | *user = NULL; | |
2584e517 | 944 | return ERR_PTR(-EPERM); |
353d5c30 | 945 | } |
2584e517 | 946 | } |
1da177e4 | 947 | |
0df4d6e5 | 948 | sb = hugetlbfs_vfsmount[hstate_idx]->mnt_sb; |
9d66586f | 949 | quick_string.name = name; |
1da177e4 LT |
950 | quick_string.len = strlen(quick_string.name); |
951 | quick_string.hash = 0; | |
0df4d6e5 | 952 | path.dentry = d_alloc_pseudo(sb, &quick_string); |
2c48b9c4 | 953 | if (!path.dentry) |
1da177e4 LT |
954 | goto out_shm_unlock; |
955 | ||
0df4d6e5 | 956 | d_set_d_op(path.dentry, &anon_ops); |
42d7395f | 957 | path.mnt = mntget(hugetlbfs_vfsmount[hstate_idx]); |
39b65252 | 958 | file = ERR_PTR(-ENOSPC); |
0df4d6e5 | 959 | inode = hugetlbfs_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0); |
1da177e4 | 960 | if (!inode) |
ce8d2cdf | 961 | goto out_dentry; |
1da177e4 | 962 | |
39b65252 | 963 | file = ERR_PTR(-ENOMEM); |
af73e4d9 NH |
964 | if (hugetlb_reserve_pages(inode, 0, |
965 | size >> huge_page_shift(hstate_inode(inode)), NULL, | |
966 | acctflag)) | |
b45b5bd6 DG |
967 | goto out_inode; |
968 | ||
2c48b9c4 | 969 | d_instantiate(path.dentry, inode); |
1da177e4 | 970 | inode->i_size = size; |
6d6b77f1 | 971 | clear_nlink(inode); |
ce8d2cdf | 972 | |
2c48b9c4 | 973 | file = alloc_file(&path, FMODE_WRITE | FMODE_READ, |
ce8d2cdf | 974 | &hugetlbfs_file_operations); |
39b65252 | 975 | if (IS_ERR(file)) |
b4d232e6 | 976 | goto out_dentry; /* inode is already attached */ |
ce8d2cdf | 977 | |
1da177e4 LT |
978 | return file; |
979 | ||
b45b5bd6 DG |
980 | out_inode: |
981 | iput(inode); | |
1da177e4 | 982 | out_dentry: |
2c48b9c4 | 983 | path_put(&path); |
1da177e4 | 984 | out_shm_unlock: |
353d5c30 HD |
985 | if (*user) { |
986 | user_shm_unlock(size, *user); | |
987 | *user = NULL; | |
988 | } | |
39b65252 | 989 | return file; |
1da177e4 LT |
990 | } |
991 | ||
992 | static int __init init_hugetlbfs_fs(void) | |
993 | { | |
42d7395f | 994 | struct hstate *h; |
1da177e4 | 995 | int error; |
42d7395f | 996 | int i; |
1da177e4 | 997 | |
457c1b27 | 998 | if (!hugepages_supported()) { |
9b857d26 | 999 | pr_info("disabling because there are no supported hugepage sizes\n"); |
457c1b27 NA |
1000 | return -ENOTSUPP; |
1001 | } | |
1002 | ||
d1d5e05f | 1003 | error = -ENOMEM; |
1da177e4 LT |
1004 | hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache", |
1005 | sizeof(struct hugetlbfs_inode_info), | |
20c2df83 | 1006 | 0, 0, init_once); |
1da177e4 | 1007 | if (hugetlbfs_inode_cachep == NULL) |
e0bf68dd | 1008 | goto out2; |
1da177e4 LT |
1009 | |
1010 | error = register_filesystem(&hugetlbfs_fs_type); | |
1011 | if (error) | |
1012 | goto out; | |
1013 | ||
42d7395f AK |
1014 | i = 0; |
1015 | for_each_hstate(h) { | |
1016 | char buf[50]; | |
1017 | unsigned ps_kb = 1U << (h->order + PAGE_SHIFT - 10); | |
1da177e4 | 1018 | |
42d7395f AK |
1019 | snprintf(buf, sizeof(buf), "pagesize=%uK", ps_kb); |
1020 | hugetlbfs_vfsmount[i] = kern_mount_data(&hugetlbfs_fs_type, | |
1021 | buf); | |
1da177e4 | 1022 | |
42d7395f | 1023 | if (IS_ERR(hugetlbfs_vfsmount[i])) { |
9b857d26 | 1024 | pr_err("Cannot mount internal hugetlbfs for " |
42d7395f AK |
1025 | "page size %uK", ps_kb); |
1026 | error = PTR_ERR(hugetlbfs_vfsmount[i]); | |
1027 | hugetlbfs_vfsmount[i] = NULL; | |
1028 | } | |
1029 | i++; | |
1030 | } | |
1031 | /* Non default hstates are optional */ | |
1032 | if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount[default_hstate_idx])) | |
1033 | return 0; | |
1da177e4 LT |
1034 | |
1035 | out: | |
d1d5e05f | 1036 | kmem_cache_destroy(hugetlbfs_inode_cachep); |
e0bf68dd | 1037 | out2: |
1da177e4 LT |
1038 | return error; |
1039 | } | |
1040 | ||
1041 | static void __exit exit_hugetlbfs_fs(void) | |
1042 | { | |
42d7395f AK |
1043 | struct hstate *h; |
1044 | int i; | |
1045 | ||
1046 | ||
8c0a8537 KS |
1047 | /* |
1048 | * Make sure all delayed rcu free inodes are flushed before we | |
1049 | * destroy cache. | |
1050 | */ | |
1051 | rcu_barrier(); | |
1da177e4 | 1052 | kmem_cache_destroy(hugetlbfs_inode_cachep); |
42d7395f AK |
1053 | i = 0; |
1054 | for_each_hstate(h) | |
1055 | kern_unmount(hugetlbfs_vfsmount[i++]); | |
1da177e4 LT |
1056 | unregister_filesystem(&hugetlbfs_fs_type); |
1057 | } | |
1058 | ||
1059 | module_init(init_hugetlbfs_fs) | |
1060 | module_exit(exit_hugetlbfs_fs) | |
1061 | ||
1062 | MODULE_LICENSE("GPL"); |