]>
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 | 50 | umode_t mode; |
7ca02d0a | 51 | long max_hpages; |
a1d776ee DG |
52 | long nr_inodes; |
53 | struct hstate *hstate; | |
7ca02d0a | 54 | long min_hpages; |
a1d776ee DG |
55 | }; |
56 | ||
57 | struct hugetlbfs_inode_info { | |
58 | struct shared_policy policy; | |
59 | struct inode vfs_inode; | |
60 | }; | |
61 | ||
62 | static inline struct hugetlbfs_inode_info *HUGETLBFS_I(struct inode *inode) | |
63 | { | |
64 | return container_of(inode, struct hugetlbfs_inode_info, vfs_inode); | |
65 | } | |
66 | ||
1da177e4 LT |
67 | int sysctl_hugetlb_shm_group; |
68 | ||
e73a75fa RD |
69 | enum { |
70 | Opt_size, Opt_nr_inodes, | |
71 | Opt_mode, Opt_uid, Opt_gid, | |
7ca02d0a | 72 | Opt_pagesize, Opt_min_size, |
e73a75fa RD |
73 | Opt_err, |
74 | }; | |
75 | ||
a447c093 | 76 | static const match_table_t tokens = { |
e73a75fa RD |
77 | {Opt_size, "size=%s"}, |
78 | {Opt_nr_inodes, "nr_inodes=%s"}, | |
79 | {Opt_mode, "mode=%o"}, | |
80 | {Opt_uid, "uid=%u"}, | |
81 | {Opt_gid, "gid=%u"}, | |
a137e1cc | 82 | {Opt_pagesize, "pagesize=%s"}, |
7ca02d0a | 83 | {Opt_min_size, "min_size=%s"}, |
e73a75fa RD |
84 | {Opt_err, NULL}, |
85 | }; | |
86 | ||
2e9b367c AL |
87 | static void huge_pagevec_release(struct pagevec *pvec) |
88 | { | |
89 | int i; | |
90 | ||
91 | for (i = 0; i < pagevec_count(pvec); ++i) | |
92 | put_page(pvec->pages[i]); | |
93 | ||
94 | pagevec_reinit(pvec); | |
95 | } | |
96 | ||
1da177e4 LT |
97 | static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma) |
98 | { | |
496ad9aa | 99 | struct inode *inode = file_inode(file); |
1da177e4 LT |
100 | loff_t len, vma_len; |
101 | int ret; | |
a5516438 | 102 | struct hstate *h = hstate_file(file); |
1da177e4 | 103 | |
68589bc3 | 104 | /* |
dec4ad86 DG |
105 | * vma address alignment (but not the pgoff alignment) has |
106 | * already been checked by prepare_hugepage_range. If you add | |
107 | * any error returns here, do so after setting VM_HUGETLB, so | |
108 | * is_vm_hugetlb_page tests below unmap_region go the right | |
109 | * way when do_mmap_pgoff unwinds (may be important on powerpc | |
110 | * and ia64). | |
68589bc3 | 111 | */ |
a2fce914 | 112 | vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND; |
68589bc3 | 113 | vma->vm_ops = &hugetlb_vm_ops; |
1da177e4 | 114 | |
2b37c35e | 115 | if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT)) |
dec4ad86 DG |
116 | return -EINVAL; |
117 | ||
1da177e4 LT |
118 | vma_len = (loff_t)(vma->vm_end - vma->vm_start); |
119 | ||
1b1dcc1b | 120 | mutex_lock(&inode->i_mutex); |
1da177e4 | 121 | file_accessed(file); |
1da177e4 LT |
122 | |
123 | ret = -ENOMEM; | |
124 | len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); | |
1da177e4 | 125 | |
a1e78772 | 126 | if (hugetlb_reserve_pages(inode, |
a5516438 | 127 | vma->vm_pgoff >> huge_page_order(h), |
5a6fe125 MG |
128 | len >> huge_page_shift(h), vma, |
129 | vma->vm_flags)) | |
a43a8c39 | 130 | goto out; |
b45b5bd6 | 131 | |
4c887265 | 132 | ret = 0; |
b6174df5 | 133 | if (vma->vm_flags & VM_WRITE && inode->i_size < len) |
1da177e4 LT |
134 | inode->i_size = len; |
135 | out: | |
1b1dcc1b | 136 | mutex_unlock(&inode->i_mutex); |
1da177e4 LT |
137 | |
138 | return ret; | |
139 | } | |
140 | ||
141 | /* | |
508034a3 | 142 | * Called under down_write(mmap_sem). |
1da177e4 LT |
143 | */ |
144 | ||
d2ba27e8 | 145 | #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA |
1da177e4 LT |
146 | static unsigned long |
147 | hugetlb_get_unmapped_area(struct file *file, unsigned long addr, | |
148 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
149 | { | |
150 | struct mm_struct *mm = current->mm; | |
151 | struct vm_area_struct *vma; | |
a5516438 | 152 | struct hstate *h = hstate_file(file); |
08659355 | 153 | struct vm_unmapped_area_info info; |
1da177e4 | 154 | |
a5516438 | 155 | if (len & ~huge_page_mask(h)) |
1da177e4 LT |
156 | return -EINVAL; |
157 | if (len > TASK_SIZE) | |
158 | return -ENOMEM; | |
159 | ||
036e0856 | 160 | if (flags & MAP_FIXED) { |
a5516438 | 161 | if (prepare_hugepage_range(file, addr, len)) |
036e0856 BH |
162 | return -EINVAL; |
163 | return addr; | |
164 | } | |
165 | ||
1da177e4 | 166 | if (addr) { |
a5516438 | 167 | addr = ALIGN(addr, huge_page_size(h)); |
1da177e4 LT |
168 | vma = find_vma(mm, addr); |
169 | if (TASK_SIZE - len >= addr && | |
170 | (!vma || addr + len <= vma->vm_start)) | |
171 | return addr; | |
172 | } | |
173 | ||
08659355 ML |
174 | info.flags = 0; |
175 | info.length = len; | |
176 | info.low_limit = TASK_UNMAPPED_BASE; | |
177 | info.high_limit = TASK_SIZE; | |
178 | info.align_mask = PAGE_MASK & ~huge_page_mask(h); | |
179 | info.align_offset = 0; | |
180 | return vm_unmapped_area(&info); | |
1da177e4 LT |
181 | } |
182 | #endif | |
183 | ||
34d0640e | 184 | static size_t |
e63e1e5a | 185 | hugetlbfs_read_actor(struct page *page, unsigned long offset, |
34d0640e | 186 | struct iov_iter *to, unsigned long size) |
e63e1e5a | 187 | { |
34d0640e | 188 | size_t copied = 0; |
e63e1e5a BP |
189 | int i, chunksize; |
190 | ||
e63e1e5a BP |
191 | /* Find which 4k chunk and offset with in that chunk */ |
192 | i = offset >> PAGE_CACHE_SHIFT; | |
193 | offset = offset & ~PAGE_CACHE_MASK; | |
194 | ||
195 | while (size) { | |
34d0640e | 196 | size_t n; |
e63e1e5a BP |
197 | chunksize = PAGE_CACHE_SIZE; |
198 | if (offset) | |
199 | chunksize -= offset; | |
200 | if (chunksize > size) | |
201 | chunksize = size; | |
34d0640e AV |
202 | n = copy_page_to_iter(&page[i], offset, chunksize, to); |
203 | copied += n; | |
204 | if (n != chunksize) | |
205 | return copied; | |
e63e1e5a BP |
206 | offset = 0; |
207 | size -= chunksize; | |
e63e1e5a BP |
208 | i++; |
209 | } | |
34d0640e | 210 | return copied; |
e63e1e5a BP |
211 | } |
212 | ||
213 | /* | |
214 | * Support for read() - Find the page attached to f_mapping and copy out the | |
215 | * data. Its *very* similar to do_generic_mapping_read(), we can't use that | |
216 | * since it has PAGE_CACHE_SIZE assumptions. | |
217 | */ | |
34d0640e | 218 | static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to) |
e63e1e5a | 219 | { |
34d0640e AV |
220 | struct file *file = iocb->ki_filp; |
221 | struct hstate *h = hstate_file(file); | |
222 | struct address_space *mapping = file->f_mapping; | |
e63e1e5a | 223 | struct inode *inode = mapping->host; |
34d0640e AV |
224 | unsigned long index = iocb->ki_pos >> huge_page_shift(h); |
225 | unsigned long offset = iocb->ki_pos & ~huge_page_mask(h); | |
e63e1e5a BP |
226 | unsigned long end_index; |
227 | loff_t isize; | |
228 | ssize_t retval = 0; | |
229 | ||
34d0640e | 230 | while (iov_iter_count(to)) { |
e63e1e5a | 231 | struct page *page; |
34d0640e | 232 | size_t nr, copied; |
e63e1e5a BP |
233 | |
234 | /* nr is the maximum number of bytes to copy from this page */ | |
a5516438 | 235 | nr = huge_page_size(h); |
a05b0855 AK |
236 | isize = i_size_read(inode); |
237 | if (!isize) | |
34d0640e | 238 | break; |
a05b0855 | 239 | end_index = (isize - 1) >> huge_page_shift(h); |
34d0640e AV |
240 | if (index > end_index) |
241 | break; | |
242 | if (index == end_index) { | |
a5516438 | 243 | nr = ((isize - 1) & ~huge_page_mask(h)) + 1; |
a05b0855 | 244 | if (nr <= offset) |
34d0640e | 245 | break; |
e63e1e5a BP |
246 | } |
247 | nr = nr - offset; | |
248 | ||
249 | /* Find the page */ | |
a05b0855 | 250 | page = find_lock_page(mapping, index); |
e63e1e5a BP |
251 | if (unlikely(page == NULL)) { |
252 | /* | |
253 | * We have a HOLE, zero out the user-buffer for the | |
254 | * length of the hole or request. | |
255 | */ | |
34d0640e | 256 | copied = iov_iter_zero(nr, to); |
e63e1e5a | 257 | } else { |
a05b0855 AK |
258 | unlock_page(page); |
259 | ||
e63e1e5a BP |
260 | /* |
261 | * We have the page, copy it to user space buffer. | |
262 | */ | |
34d0640e | 263 | copied = hugetlbfs_read_actor(page, offset, to, nr); |
a05b0855 | 264 | page_cache_release(page); |
e63e1e5a | 265 | } |
34d0640e AV |
266 | offset += copied; |
267 | retval += copied; | |
268 | if (copied != nr && iov_iter_count(to)) { | |
269 | if (!retval) | |
270 | retval = -EFAULT; | |
271 | break; | |
e63e1e5a | 272 | } |
a5516438 AK |
273 | index += offset >> huge_page_shift(h); |
274 | offset &= ~huge_page_mask(h); | |
e63e1e5a | 275 | } |
34d0640e | 276 | iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset; |
e63e1e5a BP |
277 | return retval; |
278 | } | |
279 | ||
800d15a5 NP |
280 | static int hugetlbfs_write_begin(struct file *file, |
281 | struct address_space *mapping, | |
282 | loff_t pos, unsigned len, unsigned flags, | |
283 | struct page **pagep, void **fsdata) | |
1da177e4 LT |
284 | { |
285 | return -EINVAL; | |
286 | } | |
287 | ||
800d15a5 NP |
288 | static int hugetlbfs_write_end(struct file *file, struct address_space *mapping, |
289 | loff_t pos, unsigned len, unsigned copied, | |
290 | struct page *page, void *fsdata) | |
1da177e4 | 291 | { |
800d15a5 | 292 | BUG(); |
1da177e4 LT |
293 | return -EINVAL; |
294 | } | |
295 | ||
1da177e4 LT |
296 | static void truncate_huge_page(struct page *page) |
297 | { | |
b9ea2515 | 298 | ClearPageDirty(page); |
1da177e4 | 299 | ClearPageUptodate(page); |
bd65cb86 | 300 | delete_from_page_cache(page); |
1da177e4 LT |
301 | } |
302 | ||
b45b5bd6 | 303 | static void truncate_hugepages(struct inode *inode, loff_t lstart) |
1da177e4 | 304 | { |
a5516438 | 305 | struct hstate *h = hstate_inode(inode); |
b45b5bd6 | 306 | struct address_space *mapping = &inode->i_data; |
a5516438 | 307 | const pgoff_t start = lstart >> huge_page_shift(h); |
1da177e4 LT |
308 | struct pagevec pvec; |
309 | pgoff_t next; | |
a43a8c39 | 310 | int i, freed = 0; |
1da177e4 LT |
311 | |
312 | pagevec_init(&pvec, 0); | |
313 | next = start; | |
314 | while (1) { | |
315 | if (!pagevec_lookup(&pvec, mapping, next, PAGEVEC_SIZE)) { | |
316 | if (next == start) | |
317 | break; | |
318 | next = start; | |
319 | continue; | |
320 | } | |
321 | ||
322 | for (i = 0; i < pagevec_count(&pvec); ++i) { | |
323 | struct page *page = pvec.pages[i]; | |
324 | ||
325 | lock_page(page); | |
326 | if (page->index > next) | |
327 | next = page->index; | |
328 | ++next; | |
329 | truncate_huge_page(page); | |
330 | unlock_page(page); | |
a43a8c39 | 331 | freed++; |
1da177e4 LT |
332 | } |
333 | huge_pagevec_release(&pvec); | |
334 | } | |
335 | BUG_ON(!lstart && mapping->nrpages); | |
a43a8c39 | 336 | hugetlb_unreserve_pages(inode, start, freed); |
1da177e4 LT |
337 | } |
338 | ||
2bbbda30 | 339 | static void hugetlbfs_evict_inode(struct inode *inode) |
1da177e4 | 340 | { |
9119a41e JK |
341 | struct resv_map *resv_map; |
342 | ||
b45b5bd6 | 343 | truncate_hugepages(inode, 0); |
9119a41e JK |
344 | resv_map = (struct resv_map *)inode->i_mapping->private_data; |
345 | /* root inode doesn't have the resv_map, so we should check it */ | |
346 | if (resv_map) | |
347 | resv_map_release(&resv_map->refs); | |
dbd5768f | 348 | clear_inode(inode); |
149f4211 CH |
349 | } |
350 | ||
1da177e4 | 351 | static inline void |
6b2dbba8 | 352 | hugetlb_vmtruncate_list(struct rb_root *root, pgoff_t pgoff) |
1da177e4 LT |
353 | { |
354 | struct vm_area_struct *vma; | |
1da177e4 | 355 | |
6b2dbba8 | 356 | vma_interval_tree_foreach(vma, root, pgoff, ULONG_MAX) { |
1da177e4 LT |
357 | unsigned long v_offset; |
358 | ||
1da177e4 | 359 | /* |
856fc295 | 360 | * Can the expression below overflow on 32-bit arches? |
6b2dbba8 | 361 | * No, because the interval tree returns us only those vmas |
856fc295 HD |
362 | * which overlap the truncated area starting at pgoff, |
363 | * and no vma on a 32-bit arch can span beyond the 4GB. | |
1da177e4 | 364 | */ |
856fc295 HD |
365 | if (vma->vm_pgoff < pgoff) |
366 | v_offset = (pgoff - vma->vm_pgoff) << PAGE_SHIFT; | |
367 | else | |
1da177e4 LT |
368 | v_offset = 0; |
369 | ||
24669e58 AK |
370 | unmap_hugepage_range(vma, vma->vm_start + v_offset, |
371 | vma->vm_end, NULL); | |
1da177e4 LT |
372 | } |
373 | } | |
374 | ||
1da177e4 LT |
375 | static int hugetlb_vmtruncate(struct inode *inode, loff_t offset) |
376 | { | |
856fc295 | 377 | pgoff_t pgoff; |
1da177e4 | 378 | struct address_space *mapping = inode->i_mapping; |
a5516438 | 379 | struct hstate *h = hstate_inode(inode); |
1da177e4 | 380 | |
a5516438 | 381 | BUG_ON(offset & ~huge_page_mask(h)); |
856fc295 | 382 | pgoff = offset >> PAGE_SHIFT; |
1da177e4 | 383 | |
7aa91e10 | 384 | i_size_write(inode, offset); |
83cde9e8 | 385 | i_mmap_lock_write(mapping); |
6b2dbba8 | 386 | if (!RB_EMPTY_ROOT(&mapping->i_mmap)) |
1da177e4 | 387 | hugetlb_vmtruncate_list(&mapping->i_mmap, pgoff); |
83cde9e8 | 388 | i_mmap_unlock_write(mapping); |
b45b5bd6 | 389 | truncate_hugepages(inode, offset); |
1da177e4 LT |
390 | return 0; |
391 | } | |
392 | ||
393 | static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr) | |
394 | { | |
2b0143b5 | 395 | struct inode *inode = d_inode(dentry); |
a5516438 | 396 | struct hstate *h = hstate_inode(inode); |
1da177e4 LT |
397 | int error; |
398 | unsigned int ia_valid = attr->ia_valid; | |
399 | ||
400 | BUG_ON(!inode); | |
401 | ||
402 | error = inode_change_ok(inode, attr); | |
403 | if (error) | |
1025774c | 404 | return error; |
1da177e4 LT |
405 | |
406 | if (ia_valid & ATTR_SIZE) { | |
407 | error = -EINVAL; | |
1025774c CH |
408 | if (attr->ia_size & ~huge_page_mask(h)) |
409 | return -EINVAL; | |
410 | error = hugetlb_vmtruncate(inode, attr->ia_size); | |
1da177e4 | 411 | if (error) |
1025774c | 412 | return error; |
1da177e4 | 413 | } |
1025774c CH |
414 | |
415 | setattr_copy(inode, attr); | |
416 | mark_inode_dirty(inode); | |
417 | return 0; | |
1da177e4 LT |
418 | } |
419 | ||
7d54fa64 AV |
420 | static struct inode *hugetlbfs_get_root(struct super_block *sb, |
421 | struct hugetlbfs_config *config) | |
1da177e4 LT |
422 | { |
423 | struct inode *inode; | |
1da177e4 LT |
424 | |
425 | inode = new_inode(sb); | |
426 | if (inode) { | |
427 | struct hugetlbfs_inode_info *info; | |
85fe4025 | 428 | inode->i_ino = get_next_ino(); |
7d54fa64 AV |
429 | inode->i_mode = S_IFDIR | config->mode; |
430 | inode->i_uid = config->uid; | |
431 | inode->i_gid = config->gid; | |
432 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
433 | info = HUGETLBFS_I(inode); | |
434 | mpol_shared_policy_init(&info->policy, NULL); | |
435 | inode->i_op = &hugetlbfs_dir_inode_operations; | |
436 | inode->i_fop = &simple_dir_operations; | |
437 | /* directory inodes start off with i_nlink == 2 (for "." entry) */ | |
438 | inc_nlink(inode); | |
65ed7601 | 439 | lockdep_annotate_inode_mutex_key(inode); |
7d54fa64 AV |
440 | } |
441 | return inode; | |
442 | } | |
443 | ||
b610ded7 | 444 | /* |
c8c06efa | 445 | * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never |
b610ded7 MH |
446 | * be taken from reclaim -- unlike regular filesystems. This needs an |
447 | * annotation because huge_pmd_share() does an allocation under | |
c8c06efa | 448 | * i_mmap_rwsem. |
b610ded7 | 449 | */ |
c8c06efa | 450 | static struct lock_class_key hugetlbfs_i_mmap_rwsem_key; |
b610ded7 | 451 | |
7d54fa64 AV |
452 | static struct inode *hugetlbfs_get_inode(struct super_block *sb, |
453 | struct inode *dir, | |
18df2252 | 454 | umode_t mode, dev_t dev) |
7d54fa64 AV |
455 | { |
456 | struct inode *inode; | |
9119a41e JK |
457 | struct resv_map *resv_map; |
458 | ||
459 | resv_map = resv_map_alloc(); | |
460 | if (!resv_map) | |
461 | return NULL; | |
7d54fa64 AV |
462 | |
463 | inode = new_inode(sb); | |
464 | if (inode) { | |
465 | struct hugetlbfs_inode_info *info; | |
466 | inode->i_ino = get_next_ino(); | |
467 | inode_init_owner(inode, dir, mode); | |
c8c06efa DB |
468 | lockdep_set_class(&inode->i_mapping->i_mmap_rwsem, |
469 | &hugetlbfs_i_mmap_rwsem_key); | |
1da177e4 | 470 | inode->i_mapping->a_ops = &hugetlbfs_aops; |
1da177e4 | 471 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
9119a41e | 472 | inode->i_mapping->private_data = resv_map; |
1da177e4 | 473 | info = HUGETLBFS_I(inode); |
6bfde05b EM |
474 | /* |
475 | * The policy is initialized here even if we are creating a | |
476 | * private inode because initialization simply creates an | |
477 | * an empty rb tree and calls spin_lock_init(), later when we | |
478 | * call mpol_free_shared_policy() it will just return because | |
479 | * the rb tree will still be empty. | |
480 | */ | |
71fe804b | 481 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
482 | switch (mode & S_IFMT) { |
483 | default: | |
484 | init_special_inode(inode, mode, dev); | |
485 | break; | |
486 | case S_IFREG: | |
487 | inode->i_op = &hugetlbfs_inode_operations; | |
488 | inode->i_fop = &hugetlbfs_file_operations; | |
489 | break; | |
490 | case S_IFDIR: | |
491 | inode->i_op = &hugetlbfs_dir_inode_operations; | |
492 | inode->i_fop = &simple_dir_operations; | |
493 | ||
494 | /* directory inodes start off with i_nlink == 2 (for "." entry) */ | |
d8c76e6f | 495 | inc_nlink(inode); |
1da177e4 LT |
496 | break; |
497 | case S_IFLNK: | |
498 | inode->i_op = &page_symlink_inode_operations; | |
499 | break; | |
500 | } | |
e096d0c7 | 501 | lockdep_annotate_inode_mutex_key(inode); |
9119a41e JK |
502 | } else |
503 | kref_put(&resv_map->refs, resv_map_release); | |
504 | ||
1da177e4 LT |
505 | return inode; |
506 | } | |
507 | ||
508 | /* | |
509 | * File creation. Allocate an inode, and we're done.. | |
510 | */ | |
511 | static int hugetlbfs_mknod(struct inode *dir, | |
1a67aafb | 512 | struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 LT |
513 | { |
514 | struct inode *inode; | |
515 | int error = -ENOSPC; | |
7d54fa64 AV |
516 | |
517 | inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev); | |
1da177e4 LT |
518 | if (inode) { |
519 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
520 | d_instantiate(dentry, inode); | |
521 | dget(dentry); /* Extra count - pin the dentry in core */ | |
522 | error = 0; | |
523 | } | |
524 | return error; | |
525 | } | |
526 | ||
18bb1db3 | 527 | static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
528 | { |
529 | int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0); | |
530 | if (!retval) | |
d8c76e6f | 531 | inc_nlink(dir); |
1da177e4 LT |
532 | return retval; |
533 | } | |
534 | ||
ebfc3b49 | 535 | static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl) |
1da177e4 LT |
536 | { |
537 | return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0); | |
538 | } | |
539 | ||
540 | static int hugetlbfs_symlink(struct inode *dir, | |
541 | struct dentry *dentry, const char *symname) | |
542 | { | |
543 | struct inode *inode; | |
544 | int error = -ENOSPC; | |
1da177e4 | 545 | |
7d54fa64 | 546 | inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0); |
1da177e4 LT |
547 | if (inode) { |
548 | int l = strlen(symname)+1; | |
549 | error = page_symlink(inode, symname, l); | |
550 | if (!error) { | |
551 | d_instantiate(dentry, inode); | |
552 | dget(dentry); | |
553 | } else | |
554 | iput(inode); | |
555 | } | |
556 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
557 | ||
558 | return error; | |
559 | } | |
560 | ||
561 | /* | |
6649a386 | 562 | * mark the head page dirty |
1da177e4 LT |
563 | */ |
564 | static int hugetlbfs_set_page_dirty(struct page *page) | |
565 | { | |
d85f3385 | 566 | struct page *head = compound_head(page); |
6649a386 KC |
567 | |
568 | SetPageDirty(head); | |
1da177e4 LT |
569 | return 0; |
570 | } | |
571 | ||
290408d4 | 572 | static int hugetlbfs_migrate_page(struct address_space *mapping, |
b969c4ab | 573 | struct page *newpage, struct page *page, |
a6bc32b8 | 574 | enum migrate_mode mode) |
290408d4 NH |
575 | { |
576 | int rc; | |
577 | ||
578 | rc = migrate_huge_page_move_mapping(mapping, newpage, page); | |
78bd5209 | 579 | if (rc != MIGRATEPAGE_SUCCESS) |
290408d4 NH |
580 | return rc; |
581 | migrate_page_copy(newpage, page); | |
582 | ||
78bd5209 | 583 | return MIGRATEPAGE_SUCCESS; |
290408d4 NH |
584 | } |
585 | ||
726c3342 | 586 | static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 587 | { |
726c3342 | 588 | struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb); |
2b0143b5 | 589 | struct hstate *h = hstate_inode(d_inode(dentry)); |
1da177e4 LT |
590 | |
591 | buf->f_type = HUGETLBFS_MAGIC; | |
a5516438 | 592 | buf->f_bsize = huge_page_size(h); |
1da177e4 LT |
593 | if (sbinfo) { |
594 | spin_lock(&sbinfo->stat_lock); | |
74a8a65c DG |
595 | /* If no limits set, just report 0 for max/free/used |
596 | * blocks, like simple_statfs() */ | |
90481622 DG |
597 | if (sbinfo->spool) { |
598 | long free_pages; | |
599 | ||
600 | spin_lock(&sbinfo->spool->lock); | |
601 | buf->f_blocks = sbinfo->spool->max_hpages; | |
602 | free_pages = sbinfo->spool->max_hpages | |
603 | - sbinfo->spool->used_hpages; | |
604 | buf->f_bavail = buf->f_bfree = free_pages; | |
605 | spin_unlock(&sbinfo->spool->lock); | |
74a8a65c DG |
606 | buf->f_files = sbinfo->max_inodes; |
607 | buf->f_ffree = sbinfo->free_inodes; | |
608 | } | |
1da177e4 LT |
609 | spin_unlock(&sbinfo->stat_lock); |
610 | } | |
611 | buf->f_namelen = NAME_MAX; | |
612 | return 0; | |
613 | } | |
614 | ||
615 | static void hugetlbfs_put_super(struct super_block *sb) | |
616 | { | |
617 | struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb); | |
618 | ||
619 | if (sbi) { | |
620 | sb->s_fs_info = NULL; | |
90481622 DG |
621 | |
622 | if (sbi->spool) | |
623 | hugepage_put_subpool(sbi->spool); | |
624 | ||
1da177e4 LT |
625 | kfree(sbi); |
626 | } | |
627 | } | |
628 | ||
96527980 CH |
629 | static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo) |
630 | { | |
631 | if (sbinfo->free_inodes >= 0) { | |
632 | spin_lock(&sbinfo->stat_lock); | |
633 | if (unlikely(!sbinfo->free_inodes)) { | |
634 | spin_unlock(&sbinfo->stat_lock); | |
635 | return 0; | |
636 | } | |
637 | sbinfo->free_inodes--; | |
638 | spin_unlock(&sbinfo->stat_lock); | |
639 | } | |
640 | ||
641 | return 1; | |
642 | } | |
643 | ||
644 | static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo) | |
645 | { | |
646 | if (sbinfo->free_inodes >= 0) { | |
647 | spin_lock(&sbinfo->stat_lock); | |
648 | sbinfo->free_inodes++; | |
649 | spin_unlock(&sbinfo->stat_lock); | |
650 | } | |
651 | } | |
652 | ||
653 | ||
e18b890b | 654 | static struct kmem_cache *hugetlbfs_inode_cachep; |
1da177e4 LT |
655 | |
656 | static struct inode *hugetlbfs_alloc_inode(struct super_block *sb) | |
657 | { | |
96527980 | 658 | struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb); |
1da177e4 LT |
659 | struct hugetlbfs_inode_info *p; |
660 | ||
96527980 CH |
661 | if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo))) |
662 | return NULL; | |
e94b1766 | 663 | p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL); |
96527980 CH |
664 | if (unlikely(!p)) { |
665 | hugetlbfs_inc_free_inodes(sbinfo); | |
1da177e4 | 666 | return NULL; |
96527980 | 667 | } |
1da177e4 LT |
668 | return &p->vfs_inode; |
669 | } | |
670 | ||
fa0d7e3d NP |
671 | static void hugetlbfs_i_callback(struct rcu_head *head) |
672 | { | |
673 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
674 | kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode)); |
675 | } | |
676 | ||
1da177e4 LT |
677 | static void hugetlbfs_destroy_inode(struct inode *inode) |
678 | { | |
96527980 | 679 | hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb)); |
1da177e4 | 680 | mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy); |
fa0d7e3d | 681 | call_rcu(&inode->i_rcu, hugetlbfs_i_callback); |
1da177e4 LT |
682 | } |
683 | ||
f5e54d6e | 684 | static const struct address_space_operations hugetlbfs_aops = { |
800d15a5 NP |
685 | .write_begin = hugetlbfs_write_begin, |
686 | .write_end = hugetlbfs_write_end, | |
1da177e4 | 687 | .set_page_dirty = hugetlbfs_set_page_dirty, |
290408d4 | 688 | .migratepage = hugetlbfs_migrate_page, |
1da177e4 LT |
689 | }; |
690 | ||
96527980 | 691 | |
51cc5068 | 692 | static void init_once(void *foo) |
96527980 CH |
693 | { |
694 | struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo; | |
695 | ||
a35afb83 | 696 | inode_init_once(&ei->vfs_inode); |
96527980 CH |
697 | } |
698 | ||
4b6f5d20 | 699 | const struct file_operations hugetlbfs_file_operations = { |
34d0640e | 700 | .read_iter = hugetlbfs_read_iter, |
1da177e4 | 701 | .mmap = hugetlbfs_file_mmap, |
1b061d92 | 702 | .fsync = noop_fsync, |
1da177e4 | 703 | .get_unmapped_area = hugetlb_get_unmapped_area, |
6038f373 | 704 | .llseek = default_llseek, |
1da177e4 LT |
705 | }; |
706 | ||
92e1d5be | 707 | static const struct inode_operations hugetlbfs_dir_inode_operations = { |
1da177e4 LT |
708 | .create = hugetlbfs_create, |
709 | .lookup = simple_lookup, | |
710 | .link = simple_link, | |
711 | .unlink = simple_unlink, | |
712 | .symlink = hugetlbfs_symlink, | |
713 | .mkdir = hugetlbfs_mkdir, | |
714 | .rmdir = simple_rmdir, | |
715 | .mknod = hugetlbfs_mknod, | |
716 | .rename = simple_rename, | |
717 | .setattr = hugetlbfs_setattr, | |
718 | }; | |
719 | ||
92e1d5be | 720 | static const struct inode_operations hugetlbfs_inode_operations = { |
1da177e4 LT |
721 | .setattr = hugetlbfs_setattr, |
722 | }; | |
723 | ||
ee9b6d61 | 724 | static const struct super_operations hugetlbfs_ops = { |
1da177e4 LT |
725 | .alloc_inode = hugetlbfs_alloc_inode, |
726 | .destroy_inode = hugetlbfs_destroy_inode, | |
2bbbda30 | 727 | .evict_inode = hugetlbfs_evict_inode, |
1da177e4 | 728 | .statfs = hugetlbfs_statfs, |
1da177e4 | 729 | .put_super = hugetlbfs_put_super, |
10f19a86 | 730 | .show_options = generic_show_options, |
1da177e4 LT |
731 | }; |
732 | ||
7ca02d0a MK |
733 | enum { NO_SIZE, SIZE_STD, SIZE_PERCENT }; |
734 | ||
735 | /* | |
736 | * Convert size option passed from command line to number of huge pages | |
737 | * in the pool specified by hstate. Size option could be in bytes | |
738 | * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT). | |
739 | */ | |
740 | static long long | |
741 | hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt, | |
742 | int val_type) | |
743 | { | |
744 | if (val_type == NO_SIZE) | |
745 | return -1; | |
746 | ||
747 | if (val_type == SIZE_PERCENT) { | |
748 | size_opt <<= huge_page_shift(h); | |
749 | size_opt *= h->max_huge_pages; | |
750 | do_div(size_opt, 100); | |
751 | } | |
752 | ||
753 | size_opt >>= huge_page_shift(h); | |
754 | return size_opt; | |
755 | } | |
756 | ||
1da177e4 LT |
757 | static int |
758 | hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig) | |
759 | { | |
e73a75fa RD |
760 | char *p, *rest; |
761 | substring_t args[MAX_OPT_ARGS]; | |
762 | int option; | |
7ca02d0a MK |
763 | unsigned long long max_size_opt = 0, min_size_opt = 0; |
764 | int max_val_type = NO_SIZE, min_val_type = NO_SIZE; | |
1da177e4 LT |
765 | |
766 | if (!options) | |
767 | return 0; | |
1da177e4 | 768 | |
e73a75fa RD |
769 | while ((p = strsep(&options, ",")) != NULL) { |
770 | int token; | |
b4c07bce LS |
771 | if (!*p) |
772 | continue; | |
e73a75fa RD |
773 | |
774 | token = match_token(p, tokens, args); | |
775 | switch (token) { | |
776 | case Opt_uid: | |
777 | if (match_int(&args[0], &option)) | |
778 | goto bad_val; | |
a0eb3a05 EB |
779 | pconfig->uid = make_kuid(current_user_ns(), option); |
780 | if (!uid_valid(pconfig->uid)) | |
781 | goto bad_val; | |
e73a75fa RD |
782 | break; |
783 | ||
784 | case Opt_gid: | |
785 | if (match_int(&args[0], &option)) | |
786 | goto bad_val; | |
a0eb3a05 EB |
787 | pconfig->gid = make_kgid(current_user_ns(), option); |
788 | if (!gid_valid(pconfig->gid)) | |
789 | goto bad_val; | |
e73a75fa RD |
790 | break; |
791 | ||
792 | case Opt_mode: | |
793 | if (match_octal(&args[0], &option)) | |
794 | goto bad_val; | |
75897d60 | 795 | pconfig->mode = option & 01777U; |
e73a75fa RD |
796 | break; |
797 | ||
798 | case Opt_size: { | |
e73a75fa RD |
799 | /* memparse() will accept a K/M/G without a digit */ |
800 | if (!isdigit(*args[0].from)) | |
801 | goto bad_val; | |
7ca02d0a MK |
802 | max_size_opt = memparse(args[0].from, &rest); |
803 | max_val_type = SIZE_STD; | |
a137e1cc | 804 | if (*rest == '%') |
7ca02d0a | 805 | max_val_type = SIZE_PERCENT; |
e73a75fa RD |
806 | break; |
807 | } | |
1da177e4 | 808 | |
e73a75fa RD |
809 | case Opt_nr_inodes: |
810 | /* memparse() will accept a K/M/G without a digit */ | |
811 | if (!isdigit(*args[0].from)) | |
812 | goto bad_val; | |
813 | pconfig->nr_inodes = memparse(args[0].from, &rest); | |
814 | break; | |
815 | ||
a137e1cc AK |
816 | case Opt_pagesize: { |
817 | unsigned long ps; | |
818 | ps = memparse(args[0].from, &rest); | |
819 | pconfig->hstate = size_to_hstate(ps); | |
820 | if (!pconfig->hstate) { | |
9b857d26 | 821 | pr_err("Unsupported page size %lu MB\n", |
a137e1cc AK |
822 | ps >> 20); |
823 | return -EINVAL; | |
824 | } | |
825 | break; | |
826 | } | |
827 | ||
7ca02d0a MK |
828 | case Opt_min_size: { |
829 | /* memparse() will accept a K/M/G without a digit */ | |
830 | if (!isdigit(*args[0].from)) | |
831 | goto bad_val; | |
832 | min_size_opt = memparse(args[0].from, &rest); | |
833 | min_val_type = SIZE_STD; | |
834 | if (*rest == '%') | |
835 | min_val_type = SIZE_PERCENT; | |
836 | break; | |
837 | } | |
838 | ||
e73a75fa | 839 | default: |
9b857d26 | 840 | pr_err("Bad mount option: \"%s\"\n", p); |
b4c07bce | 841 | return -EINVAL; |
e73a75fa RD |
842 | break; |
843 | } | |
1da177e4 | 844 | } |
a137e1cc | 845 | |
7ca02d0a MK |
846 | /* |
847 | * Use huge page pool size (in hstate) to convert the size | |
848 | * options to number of huge pages. If NO_SIZE, -1 is returned. | |
849 | */ | |
850 | pconfig->max_hpages = hugetlbfs_size_to_hpages(pconfig->hstate, | |
851 | max_size_opt, max_val_type); | |
852 | pconfig->min_hpages = hugetlbfs_size_to_hpages(pconfig->hstate, | |
853 | min_size_opt, min_val_type); | |
854 | ||
855 | /* | |
856 | * If max_size was specified, then min_size must be smaller | |
857 | */ | |
858 | if (max_val_type > NO_SIZE && | |
859 | pconfig->min_hpages > pconfig->max_hpages) { | |
860 | pr_err("minimum size can not be greater than maximum size\n"); | |
861 | return -EINVAL; | |
a137e1cc AK |
862 | } |
863 | ||
1da177e4 | 864 | return 0; |
e73a75fa RD |
865 | |
866 | bad_val: | |
9b857d26 | 867 | pr_err("Bad value '%s' for mount option '%s'\n", args[0].from, p); |
c12ddba0 | 868 | return -EINVAL; |
1da177e4 LT |
869 | } |
870 | ||
871 | static int | |
872 | hugetlbfs_fill_super(struct super_block *sb, void *data, int silent) | |
873 | { | |
1da177e4 LT |
874 | int ret; |
875 | struct hugetlbfs_config config; | |
876 | struct hugetlbfs_sb_info *sbinfo; | |
877 | ||
10f19a86 MS |
878 | save_mount_options(sb, data); |
879 | ||
7ca02d0a | 880 | config.max_hpages = -1; /* No limit on size by default */ |
1da177e4 | 881 | config.nr_inodes = -1; /* No limit on number of inodes by default */ |
77c70de1 DH |
882 | config.uid = current_fsuid(); |
883 | config.gid = current_fsgid(); | |
1da177e4 | 884 | config.mode = 0755; |
a137e1cc | 885 | config.hstate = &default_hstate; |
7ca02d0a | 886 | config.min_hpages = -1; /* No default minimum size */ |
1da177e4 | 887 | ret = hugetlbfs_parse_options(data, &config); |
1da177e4 LT |
888 | if (ret) |
889 | return ret; | |
890 | ||
891 | sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL); | |
892 | if (!sbinfo) | |
893 | return -ENOMEM; | |
894 | sb->s_fs_info = sbinfo; | |
a137e1cc | 895 | sbinfo->hstate = config.hstate; |
1da177e4 | 896 | spin_lock_init(&sbinfo->stat_lock); |
1da177e4 LT |
897 | sbinfo->max_inodes = config.nr_inodes; |
898 | sbinfo->free_inodes = config.nr_inodes; | |
90481622 | 899 | sbinfo->spool = NULL; |
7ca02d0a MK |
900 | /* |
901 | * Allocate and initialize subpool if maximum or minimum size is | |
902 | * specified. Any needed reservations (for minimim size) are taken | |
903 | * taken when the subpool is created. | |
904 | */ | |
905 | if (config.max_hpages != -1 || config.min_hpages != -1) { | |
906 | sbinfo->spool = hugepage_new_subpool(config.hstate, | |
907 | config.max_hpages, | |
908 | config.min_hpages); | |
90481622 DG |
909 | if (!sbinfo->spool) |
910 | goto out_free; | |
911 | } | |
1da177e4 | 912 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
a137e1cc AK |
913 | sb->s_blocksize = huge_page_size(config.hstate); |
914 | sb->s_blocksize_bits = huge_page_shift(config.hstate); | |
1da177e4 LT |
915 | sb->s_magic = HUGETLBFS_MAGIC; |
916 | sb->s_op = &hugetlbfs_ops; | |
917 | sb->s_time_gran = 1; | |
48fde701 AV |
918 | sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config)); |
919 | if (!sb->s_root) | |
1da177e4 | 920 | goto out_free; |
1da177e4 LT |
921 | return 0; |
922 | out_free: | |
6e6870d4 | 923 | kfree(sbinfo->spool); |
1da177e4 LT |
924 | kfree(sbinfo); |
925 | return -ENOMEM; | |
926 | } | |
927 | ||
3c26ff6e AV |
928 | static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type, |
929 | int flags, const char *dev_name, void *data) | |
1da177e4 | 930 | { |
3c26ff6e | 931 | return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super); |
1da177e4 LT |
932 | } |
933 | ||
934 | static struct file_system_type hugetlbfs_fs_type = { | |
935 | .name = "hugetlbfs", | |
3c26ff6e | 936 | .mount = hugetlbfs_mount, |
1da177e4 LT |
937 | .kill_sb = kill_litter_super, |
938 | }; | |
7f78e035 | 939 | MODULE_ALIAS_FS("hugetlbfs"); |
1da177e4 | 940 | |
42d7395f | 941 | static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE]; |
1da177e4 | 942 | |
ef1ff6b8 | 943 | static int can_do_hugetlb_shm(void) |
1da177e4 | 944 | { |
a0eb3a05 EB |
945 | kgid_t shm_group; |
946 | shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group); | |
947 | return capable(CAP_IPC_LOCK) || in_group_p(shm_group); | |
1da177e4 LT |
948 | } |
949 | ||
42d7395f AK |
950 | static int get_hstate_idx(int page_size_log) |
951 | { | |
af73e4d9 | 952 | struct hstate *h = hstate_sizelog(page_size_log); |
42d7395f | 953 | |
42d7395f AK |
954 | if (!h) |
955 | return -1; | |
956 | return h - hstates; | |
957 | } | |
958 | ||
be1d2cf5 | 959 | static const struct dentry_operations anon_ops = { |
118b2302 | 960 | .d_dname = simple_dname |
0df4d6e5 AV |
961 | }; |
962 | ||
af73e4d9 NH |
963 | /* |
964 | * Note that size should be aligned to proper hugepage size in caller side, | |
965 | * otherwise hugetlb_reserve_pages reserves one less hugepages than intended. | |
966 | */ | |
967 | struct file *hugetlb_file_setup(const char *name, size_t size, | |
968 | vm_flags_t acctflag, struct user_struct **user, | |
42d7395f | 969 | int creat_flags, int page_size_log) |
1da177e4 | 970 | { |
39b65252 | 971 | struct file *file = ERR_PTR(-ENOMEM); |
1da177e4 | 972 | struct inode *inode; |
2c48b9c4 | 973 | struct path path; |
0df4d6e5 | 974 | struct super_block *sb; |
1da177e4 | 975 | struct qstr quick_string; |
42d7395f AK |
976 | int hstate_idx; |
977 | ||
978 | hstate_idx = get_hstate_idx(page_size_log); | |
979 | if (hstate_idx < 0) | |
980 | return ERR_PTR(-ENODEV); | |
1da177e4 | 981 | |
353d5c30 | 982 | *user = NULL; |
42d7395f | 983 | if (!hugetlbfs_vfsmount[hstate_idx]) |
5bc98594 AM |
984 | return ERR_PTR(-ENOENT); |
985 | ||
ef1ff6b8 | 986 | if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) { |
353d5c30 HD |
987 | *user = current_user(); |
988 | if (user_shm_lock(size, *user)) { | |
21a3c273 | 989 | task_lock(current); |
9b857d26 | 990 | pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n", |
21a3c273 DR |
991 | current->comm, current->pid); |
992 | task_unlock(current); | |
353d5c30 HD |
993 | } else { |
994 | *user = NULL; | |
2584e517 | 995 | return ERR_PTR(-EPERM); |
353d5c30 | 996 | } |
2584e517 | 997 | } |
1da177e4 | 998 | |
0df4d6e5 | 999 | sb = hugetlbfs_vfsmount[hstate_idx]->mnt_sb; |
9d66586f | 1000 | quick_string.name = name; |
1da177e4 LT |
1001 | quick_string.len = strlen(quick_string.name); |
1002 | quick_string.hash = 0; | |
0df4d6e5 | 1003 | path.dentry = d_alloc_pseudo(sb, &quick_string); |
2c48b9c4 | 1004 | if (!path.dentry) |
1da177e4 LT |
1005 | goto out_shm_unlock; |
1006 | ||
0df4d6e5 | 1007 | d_set_d_op(path.dentry, &anon_ops); |
42d7395f | 1008 | path.mnt = mntget(hugetlbfs_vfsmount[hstate_idx]); |
39b65252 | 1009 | file = ERR_PTR(-ENOSPC); |
0df4d6e5 | 1010 | inode = hugetlbfs_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0); |
1da177e4 | 1011 | if (!inode) |
ce8d2cdf | 1012 | goto out_dentry; |
e1832f29 SS |
1013 | if (creat_flags == HUGETLB_SHMFS_INODE) |
1014 | inode->i_flags |= S_PRIVATE; | |
1da177e4 | 1015 | |
39b65252 | 1016 | file = ERR_PTR(-ENOMEM); |
af73e4d9 NH |
1017 | if (hugetlb_reserve_pages(inode, 0, |
1018 | size >> huge_page_shift(hstate_inode(inode)), NULL, | |
1019 | acctflag)) | |
b45b5bd6 DG |
1020 | goto out_inode; |
1021 | ||
2c48b9c4 | 1022 | d_instantiate(path.dentry, inode); |
1da177e4 | 1023 | inode->i_size = size; |
6d6b77f1 | 1024 | clear_nlink(inode); |
ce8d2cdf | 1025 | |
2c48b9c4 | 1026 | file = alloc_file(&path, FMODE_WRITE | FMODE_READ, |
ce8d2cdf | 1027 | &hugetlbfs_file_operations); |
39b65252 | 1028 | if (IS_ERR(file)) |
b4d232e6 | 1029 | goto out_dentry; /* inode is already attached */ |
ce8d2cdf | 1030 | |
1da177e4 LT |
1031 | return file; |
1032 | ||
b45b5bd6 DG |
1033 | out_inode: |
1034 | iput(inode); | |
1da177e4 | 1035 | out_dentry: |
2c48b9c4 | 1036 | path_put(&path); |
1da177e4 | 1037 | out_shm_unlock: |
353d5c30 HD |
1038 | if (*user) { |
1039 | user_shm_unlock(size, *user); | |
1040 | *user = NULL; | |
1041 | } | |
39b65252 | 1042 | return file; |
1da177e4 LT |
1043 | } |
1044 | ||
1045 | static int __init init_hugetlbfs_fs(void) | |
1046 | { | |
42d7395f | 1047 | struct hstate *h; |
1da177e4 | 1048 | int error; |
42d7395f | 1049 | int i; |
1da177e4 | 1050 | |
457c1b27 | 1051 | if (!hugepages_supported()) { |
9b857d26 | 1052 | pr_info("disabling because there are no supported hugepage sizes\n"); |
457c1b27 NA |
1053 | return -ENOTSUPP; |
1054 | } | |
1055 | ||
d1d5e05f | 1056 | error = -ENOMEM; |
1da177e4 LT |
1057 | hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache", |
1058 | sizeof(struct hugetlbfs_inode_info), | |
20c2df83 | 1059 | 0, 0, init_once); |
1da177e4 | 1060 | if (hugetlbfs_inode_cachep == NULL) |
e0bf68dd | 1061 | goto out2; |
1da177e4 LT |
1062 | |
1063 | error = register_filesystem(&hugetlbfs_fs_type); | |
1064 | if (error) | |
1065 | goto out; | |
1066 | ||
42d7395f AK |
1067 | i = 0; |
1068 | for_each_hstate(h) { | |
1069 | char buf[50]; | |
1070 | unsigned ps_kb = 1U << (h->order + PAGE_SHIFT - 10); | |
1da177e4 | 1071 | |
42d7395f AK |
1072 | snprintf(buf, sizeof(buf), "pagesize=%uK", ps_kb); |
1073 | hugetlbfs_vfsmount[i] = kern_mount_data(&hugetlbfs_fs_type, | |
1074 | buf); | |
1da177e4 | 1075 | |
42d7395f | 1076 | if (IS_ERR(hugetlbfs_vfsmount[i])) { |
9b857d26 | 1077 | pr_err("Cannot mount internal hugetlbfs for " |
42d7395f AK |
1078 | "page size %uK", ps_kb); |
1079 | error = PTR_ERR(hugetlbfs_vfsmount[i]); | |
1080 | hugetlbfs_vfsmount[i] = NULL; | |
1081 | } | |
1082 | i++; | |
1083 | } | |
1084 | /* Non default hstates are optional */ | |
1085 | if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount[default_hstate_idx])) | |
1086 | return 0; | |
1da177e4 LT |
1087 | |
1088 | out: | |
d1d5e05f | 1089 | kmem_cache_destroy(hugetlbfs_inode_cachep); |
e0bf68dd | 1090 | out2: |
1da177e4 LT |
1091 | return error; |
1092 | } | |
1093 | ||
1094 | static void __exit exit_hugetlbfs_fs(void) | |
1095 | { | |
42d7395f AK |
1096 | struct hstate *h; |
1097 | int i; | |
1098 | ||
1099 | ||
8c0a8537 KS |
1100 | /* |
1101 | * Make sure all delayed rcu free inodes are flushed before we | |
1102 | * destroy cache. | |
1103 | */ | |
1104 | rcu_barrier(); | |
1da177e4 | 1105 | kmem_cache_destroy(hugetlbfs_inode_cachep); |
42d7395f AK |
1106 | i = 0; |
1107 | for_each_hstate(h) | |
1108 | kern_unmount(hugetlbfs_vfsmount[i++]); | |
1da177e4 LT |
1109 | unregister_filesystem(&hugetlbfs_fs_type); |
1110 | } | |
1111 | ||
1112 | module_init(init_hugetlbfs_fs) | |
1113 | module_exit(exit_hugetlbfs_fs) | |
1114 | ||
1115 | MODULE_LICENSE("GPL"); |