]>
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 */ | |
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 LT |
39 | |
40 | #include <asm/uaccess.h> | |
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 { |
a0eb3a05 EB |
49 | kuid_t uid; |
50 | kgid_t gid; | |
a1d776ee | 51 | umode_t mode; |
7ca02d0a | 52 | long max_hpages; |
a1d776ee DG |
53 | long nr_inodes; |
54 | struct hstate *hstate; | |
7ca02d0a | 55 | long min_hpages; |
a1d776ee DG |
56 | }; |
57 | ||
58 | struct hugetlbfs_inode_info { | |
59 | struct shared_policy policy; | |
60 | struct inode vfs_inode; | |
61 | }; | |
62 | ||
63 | static inline struct hugetlbfs_inode_info *HUGETLBFS_I(struct inode *inode) | |
64 | { | |
65 | return container_of(inode, struct hugetlbfs_inode_info, vfs_inode); | |
66 | } | |
67 | ||
1da177e4 LT |
68 | int sysctl_hugetlb_shm_group; |
69 | ||
e73a75fa RD |
70 | enum { |
71 | Opt_size, Opt_nr_inodes, | |
72 | Opt_mode, Opt_uid, Opt_gid, | |
7ca02d0a | 73 | Opt_pagesize, Opt_min_size, |
e73a75fa RD |
74 | Opt_err, |
75 | }; | |
76 | ||
a447c093 | 77 | static const match_table_t tokens = { |
e73a75fa RD |
78 | {Opt_size, "size=%s"}, |
79 | {Opt_nr_inodes, "nr_inodes=%s"}, | |
80 | {Opt_mode, "mode=%o"}, | |
81 | {Opt_uid, "uid=%u"}, | |
82 | {Opt_gid, "gid=%u"}, | |
a137e1cc | 83 | {Opt_pagesize, "pagesize=%s"}, |
7ca02d0a | 84 | {Opt_min_size, "min_size=%s"}, |
e73a75fa RD |
85 | {Opt_err, NULL}, |
86 | }; | |
87 | ||
70c3547e MK |
88 | #ifdef CONFIG_NUMA |
89 | static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma, | |
90 | struct inode *inode, pgoff_t index) | |
91 | { | |
92 | vma->vm_policy = mpol_shared_policy_lookup(&HUGETLBFS_I(inode)->policy, | |
93 | index); | |
94 | } | |
95 | ||
96 | static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma) | |
97 | { | |
98 | mpol_cond_put(vma->vm_policy); | |
99 | } | |
100 | #else | |
101 | static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma, | |
102 | struct inode *inode, pgoff_t index) | |
103 | { | |
104 | } | |
105 | ||
106 | static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma) | |
107 | { | |
108 | } | |
109 | #endif | |
110 | ||
2e9b367c AL |
111 | static void huge_pagevec_release(struct pagevec *pvec) |
112 | { | |
113 | int i; | |
114 | ||
115 | for (i = 0; i < pagevec_count(pvec); ++i) | |
116 | put_page(pvec->pages[i]); | |
117 | ||
118 | pagevec_reinit(pvec); | |
119 | } | |
120 | ||
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 | |
2b37c35e | 139 | if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT)) |
dec4ad86 DG |
140 | return -EINVAL; |
141 | ||
1da177e4 LT |
142 | vma_len = (loff_t)(vma->vm_end - vma->vm_start); |
143 | ||
1b1dcc1b | 144 | mutex_lock(&inode->i_mutex); |
1da177e4 | 145 | file_accessed(file); |
1da177e4 LT |
146 | |
147 | ret = -ENOMEM; | |
148 | len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); | |
1da177e4 | 149 | |
a1e78772 | 150 | if (hugetlb_reserve_pages(inode, |
a5516438 | 151 | vma->vm_pgoff >> huge_page_order(h), |
5a6fe125 MG |
152 | len >> huge_page_shift(h), vma, |
153 | vma->vm_flags)) | |
a43a8c39 | 154 | goto out; |
b45b5bd6 | 155 | |
4c887265 | 156 | ret = 0; |
b6174df5 | 157 | if (vma->vm_flags & VM_WRITE && inode->i_size < len) |
1da177e4 LT |
158 | inode->i_size = len; |
159 | out: | |
1b1dcc1b | 160 | mutex_unlock(&inode->i_mutex); |
1da177e4 LT |
161 | |
162 | return ret; | |
163 | } | |
164 | ||
165 | /* | |
508034a3 | 166 | * Called under down_write(mmap_sem). |
1da177e4 LT |
167 | */ |
168 | ||
d2ba27e8 | 169 | #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA |
1da177e4 LT |
170 | static unsigned long |
171 | hugetlb_get_unmapped_area(struct file *file, unsigned long addr, | |
172 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
173 | { | |
174 | struct mm_struct *mm = current->mm; | |
175 | struct vm_area_struct *vma; | |
a5516438 | 176 | struct hstate *h = hstate_file(file); |
08659355 | 177 | struct vm_unmapped_area_info info; |
1da177e4 | 178 | |
a5516438 | 179 | if (len & ~huge_page_mask(h)) |
1da177e4 LT |
180 | return -EINVAL; |
181 | if (len > TASK_SIZE) | |
182 | return -ENOMEM; | |
183 | ||
036e0856 | 184 | if (flags & MAP_FIXED) { |
a5516438 | 185 | if (prepare_hugepage_range(file, addr, len)) |
036e0856 BH |
186 | return -EINVAL; |
187 | return addr; | |
188 | } | |
189 | ||
1da177e4 | 190 | if (addr) { |
a5516438 | 191 | addr = ALIGN(addr, huge_page_size(h)); |
1da177e4 LT |
192 | vma = find_vma(mm, addr); |
193 | if (TASK_SIZE - len >= addr && | |
194 | (!vma || addr + len <= vma->vm_start)) | |
195 | return addr; | |
196 | } | |
197 | ||
08659355 ML |
198 | info.flags = 0; |
199 | info.length = len; | |
200 | info.low_limit = TASK_UNMAPPED_BASE; | |
201 | info.high_limit = TASK_SIZE; | |
202 | info.align_mask = PAGE_MASK & ~huge_page_mask(h); | |
203 | info.align_offset = 0; | |
204 | return vm_unmapped_area(&info); | |
1da177e4 LT |
205 | } |
206 | #endif | |
207 | ||
34d0640e | 208 | static size_t |
e63e1e5a | 209 | hugetlbfs_read_actor(struct page *page, unsigned long offset, |
34d0640e | 210 | struct iov_iter *to, unsigned long size) |
e63e1e5a | 211 | { |
34d0640e | 212 | size_t copied = 0; |
e63e1e5a BP |
213 | int i, chunksize; |
214 | ||
e63e1e5a BP |
215 | /* Find which 4k chunk and offset with in that chunk */ |
216 | i = offset >> PAGE_CACHE_SHIFT; | |
217 | offset = offset & ~PAGE_CACHE_MASK; | |
218 | ||
219 | while (size) { | |
34d0640e | 220 | size_t n; |
e63e1e5a BP |
221 | chunksize = PAGE_CACHE_SIZE; |
222 | if (offset) | |
223 | chunksize -= offset; | |
224 | if (chunksize > size) | |
225 | chunksize = size; | |
34d0640e AV |
226 | n = copy_page_to_iter(&page[i], offset, chunksize, to); |
227 | copied += n; | |
228 | if (n != chunksize) | |
229 | return copied; | |
e63e1e5a BP |
230 | offset = 0; |
231 | size -= chunksize; | |
e63e1e5a BP |
232 | i++; |
233 | } | |
34d0640e | 234 | return copied; |
e63e1e5a BP |
235 | } |
236 | ||
237 | /* | |
238 | * Support for read() - Find the page attached to f_mapping and copy out the | |
239 | * data. Its *very* similar to do_generic_mapping_read(), we can't use that | |
240 | * since it has PAGE_CACHE_SIZE assumptions. | |
241 | */ | |
34d0640e | 242 | static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to) |
e63e1e5a | 243 | { |
34d0640e AV |
244 | struct file *file = iocb->ki_filp; |
245 | struct hstate *h = hstate_file(file); | |
246 | struct address_space *mapping = file->f_mapping; | |
e63e1e5a | 247 | struct inode *inode = mapping->host; |
34d0640e AV |
248 | unsigned long index = iocb->ki_pos >> huge_page_shift(h); |
249 | unsigned long offset = iocb->ki_pos & ~huge_page_mask(h); | |
e63e1e5a BP |
250 | unsigned long end_index; |
251 | loff_t isize; | |
252 | ssize_t retval = 0; | |
253 | ||
34d0640e | 254 | while (iov_iter_count(to)) { |
e63e1e5a | 255 | struct page *page; |
34d0640e | 256 | size_t nr, copied; |
e63e1e5a BP |
257 | |
258 | /* nr is the maximum number of bytes to copy from this page */ | |
a5516438 | 259 | nr = huge_page_size(h); |
a05b0855 AK |
260 | isize = i_size_read(inode); |
261 | if (!isize) | |
34d0640e | 262 | break; |
a05b0855 | 263 | end_index = (isize - 1) >> huge_page_shift(h); |
34d0640e AV |
264 | if (index > end_index) |
265 | break; | |
266 | if (index == end_index) { | |
a5516438 | 267 | nr = ((isize - 1) & ~huge_page_mask(h)) + 1; |
a05b0855 | 268 | if (nr <= offset) |
34d0640e | 269 | break; |
e63e1e5a BP |
270 | } |
271 | nr = nr - offset; | |
272 | ||
273 | /* Find the page */ | |
a05b0855 | 274 | page = find_lock_page(mapping, index); |
e63e1e5a BP |
275 | if (unlikely(page == NULL)) { |
276 | /* | |
277 | * We have a HOLE, zero out the user-buffer for the | |
278 | * length of the hole or request. | |
279 | */ | |
34d0640e | 280 | copied = iov_iter_zero(nr, to); |
e63e1e5a | 281 | } else { |
a05b0855 AK |
282 | unlock_page(page); |
283 | ||
e63e1e5a BP |
284 | /* |
285 | * We have the page, copy it to user space buffer. | |
286 | */ | |
34d0640e | 287 | copied = hugetlbfs_read_actor(page, offset, to, nr); |
a05b0855 | 288 | page_cache_release(page); |
e63e1e5a | 289 | } |
34d0640e AV |
290 | offset += copied; |
291 | retval += copied; | |
292 | if (copied != nr && iov_iter_count(to)) { | |
293 | if (!retval) | |
294 | retval = -EFAULT; | |
295 | break; | |
e63e1e5a | 296 | } |
a5516438 AK |
297 | index += offset >> huge_page_shift(h); |
298 | offset &= ~huge_page_mask(h); | |
e63e1e5a | 299 | } |
34d0640e | 300 | iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset; |
e63e1e5a BP |
301 | return retval; |
302 | } | |
303 | ||
800d15a5 NP |
304 | static int hugetlbfs_write_begin(struct file *file, |
305 | struct address_space *mapping, | |
306 | loff_t pos, unsigned len, unsigned flags, | |
307 | struct page **pagep, void **fsdata) | |
1da177e4 LT |
308 | { |
309 | return -EINVAL; | |
310 | } | |
311 | ||
800d15a5 NP |
312 | static int hugetlbfs_write_end(struct file *file, struct address_space *mapping, |
313 | loff_t pos, unsigned len, unsigned copied, | |
314 | struct page *page, void *fsdata) | |
1da177e4 | 315 | { |
800d15a5 | 316 | BUG(); |
1da177e4 LT |
317 | return -EINVAL; |
318 | } | |
319 | ||
b5cec28d | 320 | static void remove_huge_page(struct page *page) |
1da177e4 | 321 | { |
b9ea2515 | 322 | ClearPageDirty(page); |
1da177e4 | 323 | ClearPageUptodate(page); |
bd65cb86 | 324 | delete_from_page_cache(page); |
1da177e4 LT |
325 | } |
326 | ||
b5cec28d MK |
327 | |
328 | /* | |
329 | * remove_inode_hugepages handles two distinct cases: truncation and hole | |
330 | * punch. There are subtle differences in operation for each case. | |
331 | ||
332 | * truncation is indicated by end of range being LLONG_MAX | |
333 | * In this case, we first scan the range and release found pages. | |
334 | * After releasing pages, hugetlb_unreserve_pages cleans up region/reserv | |
1817889e MK |
335 | * maps and global counts. Page faults can not race with truncation |
336 | * in this routine. hugetlb_no_page() prevents page faults in the | |
337 | * truncated range. It checks i_size before allocation, and again after | |
338 | * with the page table lock for the page held. The same lock must be | |
339 | * acquired to unmap a page. | |
b5cec28d MK |
340 | * hole punch is indicated if end is not LLONG_MAX |
341 | * In the hole punch case we scan the range and release found pages. | |
342 | * Only when releasing a page is the associated region/reserv map | |
343 | * deleted. The region/reserv map for ranges without associated | |
1817889e MK |
344 | * pages are not modified. Page faults can race with hole punch. |
345 | * This is indicated if we find a mapped page. | |
b5cec28d MK |
346 | * Note: If the passed end of range value is beyond the end of file, but |
347 | * not LLONG_MAX this routine still performs a hole punch operation. | |
348 | */ | |
349 | static void remove_inode_hugepages(struct inode *inode, loff_t lstart, | |
350 | loff_t lend) | |
1da177e4 | 351 | { |
a5516438 | 352 | struct hstate *h = hstate_inode(inode); |
b45b5bd6 | 353 | struct address_space *mapping = &inode->i_data; |
a5516438 | 354 | const pgoff_t start = lstart >> huge_page_shift(h); |
b5cec28d MK |
355 | const pgoff_t end = lend >> huge_page_shift(h); |
356 | struct vm_area_struct pseudo_vma; | |
1da177e4 LT |
357 | struct pagevec pvec; |
358 | pgoff_t next; | |
a43a8c39 | 359 | int i, freed = 0; |
b5cec28d MK |
360 | long lookup_nr = PAGEVEC_SIZE; |
361 | bool truncate_op = (lend == LLONG_MAX); | |
1da177e4 | 362 | |
b5cec28d MK |
363 | memset(&pseudo_vma, 0, sizeof(struct vm_area_struct)); |
364 | pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); | |
1da177e4 LT |
365 | pagevec_init(&pvec, 0); |
366 | next = start; | |
b5cec28d MK |
367 | while (next < end) { |
368 | /* | |
1817889e | 369 | * Don't grab more pages than the number left in the range. |
b5cec28d MK |
370 | */ |
371 | if (end - next < lookup_nr) | |
372 | lookup_nr = end - next; | |
373 | ||
374 | /* | |
1817889e | 375 | * When no more pages are found, we are done. |
b5cec28d | 376 | */ |
1817889e MK |
377 | if (!pagevec_lookup(&pvec, mapping, next, lookup_nr)) |
378 | break; | |
1da177e4 LT |
379 | |
380 | for (i = 0; i < pagevec_count(&pvec); ++i) { | |
381 | struct page *page = pvec.pages[i]; | |
b5cec28d MK |
382 | u32 hash; |
383 | ||
1817889e MK |
384 | /* |
385 | * The page (index) could be beyond end. This is | |
386 | * only possible in the punch hole case as end is | |
387 | * max page offset in the truncate case. | |
388 | */ | |
389 | next = page->index; | |
390 | if (next >= end) | |
391 | break; | |
392 | ||
b5cec28d MK |
393 | hash = hugetlb_fault_mutex_hash(h, current->mm, |
394 | &pseudo_vma, | |
395 | mapping, next, 0); | |
396 | mutex_lock(&hugetlb_fault_mutex_table[hash]); | |
1da177e4 LT |
397 | |
398 | lock_page(page); | |
1817889e | 399 | if (likely(!page_mapped(page))) { |
b5cec28d MK |
400 | bool rsv_on_error = !PagePrivate(page); |
401 | /* | |
402 | * We must free the huge page and remove | |
403 | * from page cache (remove_huge_page) BEFORE | |
404 | * removing the region/reserve map | |
405 | * (hugetlb_unreserve_pages). In rare out | |
406 | * of memory conditions, removal of the | |
407 | * region/reserve map could fail. Before | |
408 | * free'ing the page, note PagePrivate which | |
409 | * is used in case of error. | |
410 | */ | |
411 | remove_huge_page(page); | |
412 | freed++; | |
413 | if (!truncate_op) { | |
414 | if (unlikely(hugetlb_unreserve_pages( | |
415 | inode, next, | |
416 | next + 1, 1))) | |
417 | hugetlb_fix_reserve_counts( | |
418 | inode, rsv_on_error); | |
419 | } | |
1817889e MK |
420 | } else { |
421 | /* | |
422 | * If page is mapped, it was faulted in after | |
423 | * being unmapped. It indicates a race between | |
424 | * hole punch and page fault. Do nothing in | |
425 | * this case. Getting here in a truncate | |
426 | * operation is a bug. | |
427 | */ | |
428 | BUG_ON(truncate_op); | |
b5cec28d MK |
429 | } |
430 | ||
1da177e4 | 431 | unlock_page(page); |
b5cec28d | 432 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); |
1da177e4 | 433 | } |
1817889e | 434 | ++next; |
1da177e4 | 435 | huge_pagevec_release(&pvec); |
1817889e | 436 | cond_resched(); |
1da177e4 | 437 | } |
b5cec28d MK |
438 | |
439 | if (truncate_op) | |
440 | (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed); | |
1da177e4 LT |
441 | } |
442 | ||
2bbbda30 | 443 | static void hugetlbfs_evict_inode(struct inode *inode) |
1da177e4 | 444 | { |
9119a41e JK |
445 | struct resv_map *resv_map; |
446 | ||
b5cec28d | 447 | remove_inode_hugepages(inode, 0, LLONG_MAX); |
9119a41e JK |
448 | resv_map = (struct resv_map *)inode->i_mapping->private_data; |
449 | /* root inode doesn't have the resv_map, so we should check it */ | |
450 | if (resv_map) | |
451 | resv_map_release(&resv_map->refs); | |
dbd5768f | 452 | clear_inode(inode); |
149f4211 CH |
453 | } |
454 | ||
1da177e4 | 455 | static inline void |
1bfad99a | 456 | hugetlb_vmdelete_list(struct rb_root *root, pgoff_t start, pgoff_t end) |
1da177e4 LT |
457 | { |
458 | struct vm_area_struct *vma; | |
1da177e4 | 459 | |
1bfad99a MK |
460 | /* |
461 | * end == 0 indicates that the entire range after | |
462 | * start should be unmapped. | |
463 | */ | |
464 | vma_interval_tree_foreach(vma, root, start, end ? end : ULONG_MAX) { | |
1da177e4 LT |
465 | unsigned long v_offset; |
466 | ||
1da177e4 | 467 | /* |
856fc295 | 468 | * Can the expression below overflow on 32-bit arches? |
6b2dbba8 | 469 | * No, because the interval tree returns us only those vmas |
856fc295 HD |
470 | * which overlap the truncated area starting at pgoff, |
471 | * and no vma on a 32-bit arch can span beyond the 4GB. | |
1da177e4 | 472 | */ |
1bfad99a MK |
473 | if (vma->vm_pgoff < start) |
474 | v_offset = (start - vma->vm_pgoff) << PAGE_SHIFT; | |
856fc295 | 475 | else |
1da177e4 LT |
476 | v_offset = 0; |
477 | ||
1bfad99a MK |
478 | if (end) { |
479 | end = ((end - start) << PAGE_SHIFT) + | |
480 | vma->vm_start + v_offset; | |
481 | if (end > vma->vm_end) | |
482 | end = vma->vm_end; | |
483 | } else | |
484 | end = vma->vm_end; | |
485 | ||
486 | unmap_hugepage_range(vma, vma->vm_start + v_offset, end, NULL); | |
1da177e4 LT |
487 | } |
488 | } | |
489 | ||
1da177e4 LT |
490 | static int hugetlb_vmtruncate(struct inode *inode, loff_t offset) |
491 | { | |
856fc295 | 492 | pgoff_t pgoff; |
1da177e4 | 493 | struct address_space *mapping = inode->i_mapping; |
a5516438 | 494 | struct hstate *h = hstate_inode(inode); |
1da177e4 | 495 | |
a5516438 | 496 | BUG_ON(offset & ~huge_page_mask(h)); |
856fc295 | 497 | pgoff = offset >> PAGE_SHIFT; |
1da177e4 | 498 | |
7aa91e10 | 499 | i_size_write(inode, offset); |
83cde9e8 | 500 | i_mmap_lock_write(mapping); |
6b2dbba8 | 501 | if (!RB_EMPTY_ROOT(&mapping->i_mmap)) |
1bfad99a | 502 | hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0); |
83cde9e8 | 503 | i_mmap_unlock_write(mapping); |
b5cec28d | 504 | remove_inode_hugepages(inode, offset, LLONG_MAX); |
1da177e4 LT |
505 | return 0; |
506 | } | |
507 | ||
70c3547e MK |
508 | static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len) |
509 | { | |
510 | struct hstate *h = hstate_inode(inode); | |
511 | loff_t hpage_size = huge_page_size(h); | |
512 | loff_t hole_start, hole_end; | |
513 | ||
514 | /* | |
515 | * For hole punch round up the beginning offset of the hole and | |
516 | * round down the end. | |
517 | */ | |
518 | hole_start = round_up(offset, hpage_size); | |
519 | hole_end = round_down(offset + len, hpage_size); | |
520 | ||
521 | if (hole_end > hole_start) { | |
522 | struct address_space *mapping = inode->i_mapping; | |
523 | ||
524 | mutex_lock(&inode->i_mutex); | |
525 | i_mmap_lock_write(mapping); | |
526 | if (!RB_EMPTY_ROOT(&mapping->i_mmap)) | |
527 | hugetlb_vmdelete_list(&mapping->i_mmap, | |
528 | hole_start >> PAGE_SHIFT, | |
529 | hole_end >> PAGE_SHIFT); | |
530 | i_mmap_unlock_write(mapping); | |
531 | remove_inode_hugepages(inode, hole_start, hole_end); | |
532 | mutex_unlock(&inode->i_mutex); | |
533 | } | |
534 | ||
535 | return 0; | |
536 | } | |
537 | ||
538 | static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset, | |
539 | loff_t len) | |
540 | { | |
541 | struct inode *inode = file_inode(file); | |
542 | struct address_space *mapping = inode->i_mapping; | |
543 | struct hstate *h = hstate_inode(inode); | |
544 | struct vm_area_struct pseudo_vma; | |
545 | struct mm_struct *mm = current->mm; | |
546 | loff_t hpage_size = huge_page_size(h); | |
547 | unsigned long hpage_shift = huge_page_shift(h); | |
548 | pgoff_t start, index, end; | |
549 | int error; | |
550 | u32 hash; | |
551 | ||
552 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
553 | return -EOPNOTSUPP; | |
554 | ||
555 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
556 | return hugetlbfs_punch_hole(inode, offset, len); | |
557 | ||
558 | /* | |
559 | * Default preallocate case. | |
560 | * For this range, start is rounded down and end is rounded up | |
561 | * as well as being converted to page offsets. | |
562 | */ | |
563 | start = offset >> hpage_shift; | |
564 | end = (offset + len + hpage_size - 1) >> hpage_shift; | |
565 | ||
566 | mutex_lock(&inode->i_mutex); | |
567 | ||
568 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
569 | error = inode_newsize_ok(inode, offset + len); | |
570 | if (error) | |
571 | goto out; | |
572 | ||
573 | /* | |
574 | * Initialize a pseudo vma as this is required by the huge page | |
575 | * allocation routines. If NUMA is configured, use page index | |
576 | * as input to create an allocation policy. | |
577 | */ | |
578 | memset(&pseudo_vma, 0, sizeof(struct vm_area_struct)); | |
579 | pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); | |
580 | pseudo_vma.vm_file = file; | |
581 | ||
582 | for (index = start; index < end; index++) { | |
583 | /* | |
584 | * This is supposed to be the vaddr where the page is being | |
585 | * faulted in, but we have no vaddr here. | |
586 | */ | |
587 | struct page *page; | |
588 | unsigned long addr; | |
589 | int avoid_reserve = 0; | |
590 | ||
591 | cond_resched(); | |
592 | ||
593 | /* | |
594 | * fallocate(2) manpage permits EINTR; we may have been | |
595 | * interrupted because we are using up too much memory. | |
596 | */ | |
597 | if (signal_pending(current)) { | |
598 | error = -EINTR; | |
599 | break; | |
600 | } | |
601 | ||
602 | /* Set numa allocation policy based on index */ | |
603 | hugetlb_set_vma_policy(&pseudo_vma, inode, index); | |
604 | ||
605 | /* addr is the offset within the file (zero based) */ | |
606 | addr = index * hpage_size; | |
607 | ||
608 | /* mutex taken here, fault path and hole punch */ | |
609 | hash = hugetlb_fault_mutex_hash(h, mm, &pseudo_vma, mapping, | |
610 | index, addr); | |
611 | mutex_lock(&hugetlb_fault_mutex_table[hash]); | |
612 | ||
613 | /* See if already present in mapping to avoid alloc/free */ | |
614 | page = find_get_page(mapping, index); | |
615 | if (page) { | |
616 | put_page(page); | |
617 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
618 | hugetlb_drop_vma_policy(&pseudo_vma); | |
619 | continue; | |
620 | } | |
621 | ||
622 | /* Allocate page and add to page cache */ | |
623 | page = alloc_huge_page(&pseudo_vma, addr, avoid_reserve); | |
624 | hugetlb_drop_vma_policy(&pseudo_vma); | |
625 | if (IS_ERR(page)) { | |
626 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
627 | error = PTR_ERR(page); | |
628 | goto out; | |
629 | } | |
630 | clear_huge_page(page, addr, pages_per_huge_page(h)); | |
631 | __SetPageUptodate(page); | |
632 | error = huge_add_to_page_cache(page, mapping, index); | |
633 | if (unlikely(error)) { | |
634 | put_page(page); | |
635 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
636 | goto out; | |
637 | } | |
638 | ||
639 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
640 | ||
641 | /* | |
642 | * page_put due to reference from alloc_huge_page() | |
643 | * unlock_page because locked by add_to_page_cache() | |
644 | */ | |
645 | put_page(page); | |
646 | unlock_page(page); | |
647 | } | |
648 | ||
649 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
650 | i_size_write(inode, offset + len); | |
651 | inode->i_ctime = CURRENT_TIME; | |
70c3547e MK |
652 | out: |
653 | mutex_unlock(&inode->i_mutex); | |
654 | return error; | |
655 | } | |
656 | ||
1da177e4 LT |
657 | static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr) |
658 | { | |
2b0143b5 | 659 | struct inode *inode = d_inode(dentry); |
a5516438 | 660 | struct hstate *h = hstate_inode(inode); |
1da177e4 LT |
661 | int error; |
662 | unsigned int ia_valid = attr->ia_valid; | |
663 | ||
664 | BUG_ON(!inode); | |
665 | ||
666 | error = inode_change_ok(inode, attr); | |
667 | if (error) | |
1025774c | 668 | return error; |
1da177e4 LT |
669 | |
670 | if (ia_valid & ATTR_SIZE) { | |
671 | error = -EINVAL; | |
1025774c CH |
672 | if (attr->ia_size & ~huge_page_mask(h)) |
673 | return -EINVAL; | |
674 | error = hugetlb_vmtruncate(inode, attr->ia_size); | |
1da177e4 | 675 | if (error) |
1025774c | 676 | return error; |
1da177e4 | 677 | } |
1025774c CH |
678 | |
679 | setattr_copy(inode, attr); | |
680 | mark_inode_dirty(inode); | |
681 | return 0; | |
1da177e4 LT |
682 | } |
683 | ||
7d54fa64 AV |
684 | static struct inode *hugetlbfs_get_root(struct super_block *sb, |
685 | struct hugetlbfs_config *config) | |
1da177e4 LT |
686 | { |
687 | struct inode *inode; | |
1da177e4 LT |
688 | |
689 | inode = new_inode(sb); | |
690 | if (inode) { | |
691 | struct hugetlbfs_inode_info *info; | |
85fe4025 | 692 | inode->i_ino = get_next_ino(); |
7d54fa64 AV |
693 | inode->i_mode = S_IFDIR | config->mode; |
694 | inode->i_uid = config->uid; | |
695 | inode->i_gid = config->gid; | |
696 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
697 | info = HUGETLBFS_I(inode); | |
698 | mpol_shared_policy_init(&info->policy, NULL); | |
699 | inode->i_op = &hugetlbfs_dir_inode_operations; | |
700 | inode->i_fop = &simple_dir_operations; | |
701 | /* directory inodes start off with i_nlink == 2 (for "." entry) */ | |
702 | inc_nlink(inode); | |
65ed7601 | 703 | lockdep_annotate_inode_mutex_key(inode); |
7d54fa64 AV |
704 | } |
705 | return inode; | |
706 | } | |
707 | ||
b610ded7 | 708 | /* |
c8c06efa | 709 | * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never |
b610ded7 MH |
710 | * be taken from reclaim -- unlike regular filesystems. This needs an |
711 | * annotation because huge_pmd_share() does an allocation under | |
c8c06efa | 712 | * i_mmap_rwsem. |
b610ded7 | 713 | */ |
c8c06efa | 714 | static struct lock_class_key hugetlbfs_i_mmap_rwsem_key; |
b610ded7 | 715 | |
7d54fa64 AV |
716 | static struct inode *hugetlbfs_get_inode(struct super_block *sb, |
717 | struct inode *dir, | |
18df2252 | 718 | umode_t mode, dev_t dev) |
7d54fa64 AV |
719 | { |
720 | struct inode *inode; | |
9119a41e JK |
721 | struct resv_map *resv_map; |
722 | ||
723 | resv_map = resv_map_alloc(); | |
724 | if (!resv_map) | |
725 | return NULL; | |
7d54fa64 AV |
726 | |
727 | inode = new_inode(sb); | |
728 | if (inode) { | |
729 | struct hugetlbfs_inode_info *info; | |
730 | inode->i_ino = get_next_ino(); | |
731 | inode_init_owner(inode, dir, mode); | |
c8c06efa DB |
732 | lockdep_set_class(&inode->i_mapping->i_mmap_rwsem, |
733 | &hugetlbfs_i_mmap_rwsem_key); | |
1da177e4 | 734 | inode->i_mapping->a_ops = &hugetlbfs_aops; |
1da177e4 | 735 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
9119a41e | 736 | inode->i_mapping->private_data = resv_map; |
1da177e4 | 737 | info = HUGETLBFS_I(inode); |
6bfde05b EM |
738 | /* |
739 | * The policy is initialized here even if we are creating a | |
740 | * private inode because initialization simply creates an | |
741 | * an empty rb tree and calls spin_lock_init(), later when we | |
742 | * call mpol_free_shared_policy() it will just return because | |
743 | * the rb tree will still be empty. | |
744 | */ | |
71fe804b | 745 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
746 | switch (mode & S_IFMT) { |
747 | default: | |
748 | init_special_inode(inode, mode, dev); | |
749 | break; | |
750 | case S_IFREG: | |
751 | inode->i_op = &hugetlbfs_inode_operations; | |
752 | inode->i_fop = &hugetlbfs_file_operations; | |
753 | break; | |
754 | case S_IFDIR: | |
755 | inode->i_op = &hugetlbfs_dir_inode_operations; | |
756 | inode->i_fop = &simple_dir_operations; | |
757 | ||
758 | /* directory inodes start off with i_nlink == 2 (for "." entry) */ | |
d8c76e6f | 759 | inc_nlink(inode); |
1da177e4 LT |
760 | break; |
761 | case S_IFLNK: | |
762 | inode->i_op = &page_symlink_inode_operations; | |
21fc61c7 | 763 | inode_nohighmem(inode); |
1da177e4 LT |
764 | break; |
765 | } | |
e096d0c7 | 766 | lockdep_annotate_inode_mutex_key(inode); |
9119a41e JK |
767 | } else |
768 | kref_put(&resv_map->refs, resv_map_release); | |
769 | ||
1da177e4 LT |
770 | return inode; |
771 | } | |
772 | ||
773 | /* | |
774 | * File creation. Allocate an inode, and we're done.. | |
775 | */ | |
776 | static int hugetlbfs_mknod(struct inode *dir, | |
1a67aafb | 777 | struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 LT |
778 | { |
779 | struct inode *inode; | |
780 | int error = -ENOSPC; | |
7d54fa64 AV |
781 | |
782 | inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev); | |
1da177e4 LT |
783 | if (inode) { |
784 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
785 | d_instantiate(dentry, inode); | |
786 | dget(dentry); /* Extra count - pin the dentry in core */ | |
787 | error = 0; | |
788 | } | |
789 | return error; | |
790 | } | |
791 | ||
18bb1db3 | 792 | static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
793 | { |
794 | int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0); | |
795 | if (!retval) | |
d8c76e6f | 796 | inc_nlink(dir); |
1da177e4 LT |
797 | return retval; |
798 | } | |
799 | ||
ebfc3b49 | 800 | static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl) |
1da177e4 LT |
801 | { |
802 | return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0); | |
803 | } | |
804 | ||
805 | static int hugetlbfs_symlink(struct inode *dir, | |
806 | struct dentry *dentry, const char *symname) | |
807 | { | |
808 | struct inode *inode; | |
809 | int error = -ENOSPC; | |
1da177e4 | 810 | |
7d54fa64 | 811 | inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0); |
1da177e4 LT |
812 | if (inode) { |
813 | int l = strlen(symname)+1; | |
814 | error = page_symlink(inode, symname, l); | |
815 | if (!error) { | |
816 | d_instantiate(dentry, inode); | |
817 | dget(dentry); | |
818 | } else | |
819 | iput(inode); | |
820 | } | |
821 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
822 | ||
823 | return error; | |
824 | } | |
825 | ||
826 | /* | |
6649a386 | 827 | * mark the head page dirty |
1da177e4 LT |
828 | */ |
829 | static int hugetlbfs_set_page_dirty(struct page *page) | |
830 | { | |
d85f3385 | 831 | struct page *head = compound_head(page); |
6649a386 KC |
832 | |
833 | SetPageDirty(head); | |
1da177e4 LT |
834 | return 0; |
835 | } | |
836 | ||
290408d4 | 837 | static int hugetlbfs_migrate_page(struct address_space *mapping, |
b969c4ab | 838 | struct page *newpage, struct page *page, |
a6bc32b8 | 839 | enum migrate_mode mode) |
290408d4 NH |
840 | { |
841 | int rc; | |
842 | ||
843 | rc = migrate_huge_page_move_mapping(mapping, newpage, page); | |
78bd5209 | 844 | if (rc != MIGRATEPAGE_SUCCESS) |
290408d4 NH |
845 | return rc; |
846 | migrate_page_copy(newpage, page); | |
847 | ||
78bd5209 | 848 | return MIGRATEPAGE_SUCCESS; |
290408d4 NH |
849 | } |
850 | ||
726c3342 | 851 | static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 852 | { |
726c3342 | 853 | struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb); |
2b0143b5 | 854 | struct hstate *h = hstate_inode(d_inode(dentry)); |
1da177e4 LT |
855 | |
856 | buf->f_type = HUGETLBFS_MAGIC; | |
a5516438 | 857 | buf->f_bsize = huge_page_size(h); |
1da177e4 LT |
858 | if (sbinfo) { |
859 | spin_lock(&sbinfo->stat_lock); | |
74a8a65c DG |
860 | /* If no limits set, just report 0 for max/free/used |
861 | * blocks, like simple_statfs() */ | |
90481622 DG |
862 | if (sbinfo->spool) { |
863 | long free_pages; | |
864 | ||
865 | spin_lock(&sbinfo->spool->lock); | |
866 | buf->f_blocks = sbinfo->spool->max_hpages; | |
867 | free_pages = sbinfo->spool->max_hpages | |
868 | - sbinfo->spool->used_hpages; | |
869 | buf->f_bavail = buf->f_bfree = free_pages; | |
870 | spin_unlock(&sbinfo->spool->lock); | |
74a8a65c DG |
871 | buf->f_files = sbinfo->max_inodes; |
872 | buf->f_ffree = sbinfo->free_inodes; | |
873 | } | |
1da177e4 LT |
874 | spin_unlock(&sbinfo->stat_lock); |
875 | } | |
876 | buf->f_namelen = NAME_MAX; | |
877 | return 0; | |
878 | } | |
879 | ||
880 | static void hugetlbfs_put_super(struct super_block *sb) | |
881 | { | |
882 | struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb); | |
883 | ||
884 | if (sbi) { | |
885 | sb->s_fs_info = NULL; | |
90481622 DG |
886 | |
887 | if (sbi->spool) | |
888 | hugepage_put_subpool(sbi->spool); | |
889 | ||
1da177e4 LT |
890 | kfree(sbi); |
891 | } | |
892 | } | |
893 | ||
96527980 CH |
894 | static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo) |
895 | { | |
896 | if (sbinfo->free_inodes >= 0) { | |
897 | spin_lock(&sbinfo->stat_lock); | |
898 | if (unlikely(!sbinfo->free_inodes)) { | |
899 | spin_unlock(&sbinfo->stat_lock); | |
900 | return 0; | |
901 | } | |
902 | sbinfo->free_inodes--; | |
903 | spin_unlock(&sbinfo->stat_lock); | |
904 | } | |
905 | ||
906 | return 1; | |
907 | } | |
908 | ||
909 | static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo) | |
910 | { | |
911 | if (sbinfo->free_inodes >= 0) { | |
912 | spin_lock(&sbinfo->stat_lock); | |
913 | sbinfo->free_inodes++; | |
914 | spin_unlock(&sbinfo->stat_lock); | |
915 | } | |
916 | } | |
917 | ||
918 | ||
e18b890b | 919 | static struct kmem_cache *hugetlbfs_inode_cachep; |
1da177e4 LT |
920 | |
921 | static struct inode *hugetlbfs_alloc_inode(struct super_block *sb) | |
922 | { | |
96527980 | 923 | struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb); |
1da177e4 LT |
924 | struct hugetlbfs_inode_info *p; |
925 | ||
96527980 CH |
926 | if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo))) |
927 | return NULL; | |
e94b1766 | 928 | p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL); |
96527980 CH |
929 | if (unlikely(!p)) { |
930 | hugetlbfs_inc_free_inodes(sbinfo); | |
1da177e4 | 931 | return NULL; |
96527980 | 932 | } |
1da177e4 LT |
933 | return &p->vfs_inode; |
934 | } | |
935 | ||
fa0d7e3d NP |
936 | static void hugetlbfs_i_callback(struct rcu_head *head) |
937 | { | |
938 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
939 | kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode)); |
940 | } | |
941 | ||
1da177e4 LT |
942 | static void hugetlbfs_destroy_inode(struct inode *inode) |
943 | { | |
96527980 | 944 | hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb)); |
1da177e4 | 945 | mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy); |
fa0d7e3d | 946 | call_rcu(&inode->i_rcu, hugetlbfs_i_callback); |
1da177e4 LT |
947 | } |
948 | ||
f5e54d6e | 949 | static const struct address_space_operations hugetlbfs_aops = { |
800d15a5 NP |
950 | .write_begin = hugetlbfs_write_begin, |
951 | .write_end = hugetlbfs_write_end, | |
1da177e4 | 952 | .set_page_dirty = hugetlbfs_set_page_dirty, |
290408d4 | 953 | .migratepage = hugetlbfs_migrate_page, |
1da177e4 LT |
954 | }; |
955 | ||
96527980 | 956 | |
51cc5068 | 957 | static void init_once(void *foo) |
96527980 CH |
958 | { |
959 | struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo; | |
960 | ||
a35afb83 | 961 | inode_init_once(&ei->vfs_inode); |
96527980 CH |
962 | } |
963 | ||
4b6f5d20 | 964 | const struct file_operations hugetlbfs_file_operations = { |
34d0640e | 965 | .read_iter = hugetlbfs_read_iter, |
1da177e4 | 966 | .mmap = hugetlbfs_file_mmap, |
1b061d92 | 967 | .fsync = noop_fsync, |
1da177e4 | 968 | .get_unmapped_area = hugetlb_get_unmapped_area, |
70c3547e MK |
969 | .llseek = default_llseek, |
970 | .fallocate = hugetlbfs_fallocate, | |
1da177e4 LT |
971 | }; |
972 | ||
92e1d5be | 973 | static const struct inode_operations hugetlbfs_dir_inode_operations = { |
1da177e4 LT |
974 | .create = hugetlbfs_create, |
975 | .lookup = simple_lookup, | |
976 | .link = simple_link, | |
977 | .unlink = simple_unlink, | |
978 | .symlink = hugetlbfs_symlink, | |
979 | .mkdir = hugetlbfs_mkdir, | |
980 | .rmdir = simple_rmdir, | |
981 | .mknod = hugetlbfs_mknod, | |
982 | .rename = simple_rename, | |
983 | .setattr = hugetlbfs_setattr, | |
984 | }; | |
985 | ||
92e1d5be | 986 | static const struct inode_operations hugetlbfs_inode_operations = { |
1da177e4 LT |
987 | .setattr = hugetlbfs_setattr, |
988 | }; | |
989 | ||
ee9b6d61 | 990 | static const struct super_operations hugetlbfs_ops = { |
1da177e4 LT |
991 | .alloc_inode = hugetlbfs_alloc_inode, |
992 | .destroy_inode = hugetlbfs_destroy_inode, | |
2bbbda30 | 993 | .evict_inode = hugetlbfs_evict_inode, |
1da177e4 | 994 | .statfs = hugetlbfs_statfs, |
1da177e4 | 995 | .put_super = hugetlbfs_put_super, |
10f19a86 | 996 | .show_options = generic_show_options, |
1da177e4 LT |
997 | }; |
998 | ||
7ca02d0a MK |
999 | enum { NO_SIZE, SIZE_STD, SIZE_PERCENT }; |
1000 | ||
1001 | /* | |
1002 | * Convert size option passed from command line to number of huge pages | |
1003 | * in the pool specified by hstate. Size option could be in bytes | |
1004 | * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT). | |
1005 | */ | |
1006 | static long long | |
1007 | hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt, | |
1008 | int val_type) | |
1009 | { | |
1010 | if (val_type == NO_SIZE) | |
1011 | return -1; | |
1012 | ||
1013 | if (val_type == SIZE_PERCENT) { | |
1014 | size_opt <<= huge_page_shift(h); | |
1015 | size_opt *= h->max_huge_pages; | |
1016 | do_div(size_opt, 100); | |
1017 | } | |
1018 | ||
1019 | size_opt >>= huge_page_shift(h); | |
1020 | return size_opt; | |
1021 | } | |
1022 | ||
1da177e4 LT |
1023 | static int |
1024 | hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig) | |
1025 | { | |
e73a75fa RD |
1026 | char *p, *rest; |
1027 | substring_t args[MAX_OPT_ARGS]; | |
1028 | int option; | |
7ca02d0a MK |
1029 | unsigned long long max_size_opt = 0, min_size_opt = 0; |
1030 | int max_val_type = NO_SIZE, min_val_type = NO_SIZE; | |
1da177e4 LT |
1031 | |
1032 | if (!options) | |
1033 | return 0; | |
1da177e4 | 1034 | |
e73a75fa RD |
1035 | while ((p = strsep(&options, ",")) != NULL) { |
1036 | int token; | |
b4c07bce LS |
1037 | if (!*p) |
1038 | continue; | |
e73a75fa RD |
1039 | |
1040 | token = match_token(p, tokens, args); | |
1041 | switch (token) { | |
1042 | case Opt_uid: | |
1043 | if (match_int(&args[0], &option)) | |
1044 | goto bad_val; | |
a0eb3a05 EB |
1045 | pconfig->uid = make_kuid(current_user_ns(), option); |
1046 | if (!uid_valid(pconfig->uid)) | |
1047 | goto bad_val; | |
e73a75fa RD |
1048 | break; |
1049 | ||
1050 | case Opt_gid: | |
1051 | if (match_int(&args[0], &option)) | |
1052 | goto bad_val; | |
a0eb3a05 EB |
1053 | pconfig->gid = make_kgid(current_user_ns(), option); |
1054 | if (!gid_valid(pconfig->gid)) | |
1055 | goto bad_val; | |
e73a75fa RD |
1056 | break; |
1057 | ||
1058 | case Opt_mode: | |
1059 | if (match_octal(&args[0], &option)) | |
1060 | goto bad_val; | |
75897d60 | 1061 | pconfig->mode = option & 01777U; |
e73a75fa RD |
1062 | break; |
1063 | ||
1064 | case Opt_size: { | |
e73a75fa RD |
1065 | /* memparse() will accept a K/M/G without a digit */ |
1066 | if (!isdigit(*args[0].from)) | |
1067 | goto bad_val; | |
7ca02d0a MK |
1068 | max_size_opt = memparse(args[0].from, &rest); |
1069 | max_val_type = SIZE_STD; | |
a137e1cc | 1070 | if (*rest == '%') |
7ca02d0a | 1071 | max_val_type = SIZE_PERCENT; |
e73a75fa RD |
1072 | break; |
1073 | } | |
1da177e4 | 1074 | |
e73a75fa RD |
1075 | case Opt_nr_inodes: |
1076 | /* memparse() will accept a K/M/G without a digit */ | |
1077 | if (!isdigit(*args[0].from)) | |
1078 | goto bad_val; | |
1079 | pconfig->nr_inodes = memparse(args[0].from, &rest); | |
1080 | break; | |
1081 | ||
a137e1cc AK |
1082 | case Opt_pagesize: { |
1083 | unsigned long ps; | |
1084 | ps = memparse(args[0].from, &rest); | |
1085 | pconfig->hstate = size_to_hstate(ps); | |
1086 | if (!pconfig->hstate) { | |
9b857d26 | 1087 | pr_err("Unsupported page size %lu MB\n", |
a137e1cc AK |
1088 | ps >> 20); |
1089 | return -EINVAL; | |
1090 | } | |
1091 | break; | |
1092 | } | |
1093 | ||
7ca02d0a MK |
1094 | case Opt_min_size: { |
1095 | /* memparse() will accept a K/M/G without a digit */ | |
1096 | if (!isdigit(*args[0].from)) | |
1097 | goto bad_val; | |
1098 | min_size_opt = memparse(args[0].from, &rest); | |
1099 | min_val_type = SIZE_STD; | |
1100 | if (*rest == '%') | |
1101 | min_val_type = SIZE_PERCENT; | |
1102 | break; | |
1103 | } | |
1104 | ||
e73a75fa | 1105 | default: |
9b857d26 | 1106 | pr_err("Bad mount option: \"%s\"\n", p); |
b4c07bce | 1107 | return -EINVAL; |
e73a75fa RD |
1108 | break; |
1109 | } | |
1da177e4 | 1110 | } |
a137e1cc | 1111 | |
7ca02d0a MK |
1112 | /* |
1113 | * Use huge page pool size (in hstate) to convert the size | |
1114 | * options to number of huge pages. If NO_SIZE, -1 is returned. | |
1115 | */ | |
1116 | pconfig->max_hpages = hugetlbfs_size_to_hpages(pconfig->hstate, | |
1117 | max_size_opt, max_val_type); | |
1118 | pconfig->min_hpages = hugetlbfs_size_to_hpages(pconfig->hstate, | |
1119 | min_size_opt, min_val_type); | |
1120 | ||
1121 | /* | |
1122 | * If max_size was specified, then min_size must be smaller | |
1123 | */ | |
1124 | if (max_val_type > NO_SIZE && | |
1125 | pconfig->min_hpages > pconfig->max_hpages) { | |
1126 | pr_err("minimum size can not be greater than maximum size\n"); | |
1127 | return -EINVAL; | |
a137e1cc AK |
1128 | } |
1129 | ||
1da177e4 | 1130 | return 0; |
e73a75fa RD |
1131 | |
1132 | bad_val: | |
9b857d26 | 1133 | pr_err("Bad value '%s' for mount option '%s'\n", args[0].from, p); |
c12ddba0 | 1134 | return -EINVAL; |
1da177e4 LT |
1135 | } |
1136 | ||
1137 | static int | |
1138 | hugetlbfs_fill_super(struct super_block *sb, void *data, int silent) | |
1139 | { | |
1da177e4 LT |
1140 | int ret; |
1141 | struct hugetlbfs_config config; | |
1142 | struct hugetlbfs_sb_info *sbinfo; | |
1143 | ||
10f19a86 MS |
1144 | save_mount_options(sb, data); |
1145 | ||
7ca02d0a | 1146 | config.max_hpages = -1; /* No limit on size by default */ |
1da177e4 | 1147 | config.nr_inodes = -1; /* No limit on number of inodes by default */ |
77c70de1 DH |
1148 | config.uid = current_fsuid(); |
1149 | config.gid = current_fsgid(); | |
1da177e4 | 1150 | config.mode = 0755; |
a137e1cc | 1151 | config.hstate = &default_hstate; |
7ca02d0a | 1152 | config.min_hpages = -1; /* No default minimum size */ |
1da177e4 | 1153 | ret = hugetlbfs_parse_options(data, &config); |
1da177e4 LT |
1154 | if (ret) |
1155 | return ret; | |
1156 | ||
1157 | sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL); | |
1158 | if (!sbinfo) | |
1159 | return -ENOMEM; | |
1160 | sb->s_fs_info = sbinfo; | |
a137e1cc | 1161 | sbinfo->hstate = config.hstate; |
1da177e4 | 1162 | spin_lock_init(&sbinfo->stat_lock); |
1da177e4 LT |
1163 | sbinfo->max_inodes = config.nr_inodes; |
1164 | sbinfo->free_inodes = config.nr_inodes; | |
90481622 | 1165 | sbinfo->spool = NULL; |
7ca02d0a MK |
1166 | /* |
1167 | * Allocate and initialize subpool if maximum or minimum size is | |
1168 | * specified. Any needed reservations (for minimim size) are taken | |
1169 | * taken when the subpool is created. | |
1170 | */ | |
1171 | if (config.max_hpages != -1 || config.min_hpages != -1) { | |
1172 | sbinfo->spool = hugepage_new_subpool(config.hstate, | |
1173 | config.max_hpages, | |
1174 | config.min_hpages); | |
90481622 DG |
1175 | if (!sbinfo->spool) |
1176 | goto out_free; | |
1177 | } | |
1da177e4 | 1178 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
a137e1cc AK |
1179 | sb->s_blocksize = huge_page_size(config.hstate); |
1180 | sb->s_blocksize_bits = huge_page_shift(config.hstate); | |
1da177e4 LT |
1181 | sb->s_magic = HUGETLBFS_MAGIC; |
1182 | sb->s_op = &hugetlbfs_ops; | |
1183 | sb->s_time_gran = 1; | |
48fde701 AV |
1184 | sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config)); |
1185 | if (!sb->s_root) | |
1da177e4 | 1186 | goto out_free; |
1da177e4 LT |
1187 | return 0; |
1188 | out_free: | |
6e6870d4 | 1189 | kfree(sbinfo->spool); |
1da177e4 LT |
1190 | kfree(sbinfo); |
1191 | return -ENOMEM; | |
1192 | } | |
1193 | ||
3c26ff6e AV |
1194 | static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type, |
1195 | int flags, const char *dev_name, void *data) | |
1da177e4 | 1196 | { |
3c26ff6e | 1197 | return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super); |
1da177e4 LT |
1198 | } |
1199 | ||
1200 | static struct file_system_type hugetlbfs_fs_type = { | |
1201 | .name = "hugetlbfs", | |
3c26ff6e | 1202 | .mount = hugetlbfs_mount, |
1da177e4 LT |
1203 | .kill_sb = kill_litter_super, |
1204 | }; | |
7f78e035 | 1205 | MODULE_ALIAS_FS("hugetlbfs"); |
1da177e4 | 1206 | |
42d7395f | 1207 | static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE]; |
1da177e4 | 1208 | |
ef1ff6b8 | 1209 | static int can_do_hugetlb_shm(void) |
1da177e4 | 1210 | { |
a0eb3a05 EB |
1211 | kgid_t shm_group; |
1212 | shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group); | |
1213 | return capable(CAP_IPC_LOCK) || in_group_p(shm_group); | |
1da177e4 LT |
1214 | } |
1215 | ||
42d7395f AK |
1216 | static int get_hstate_idx(int page_size_log) |
1217 | { | |
af73e4d9 | 1218 | struct hstate *h = hstate_sizelog(page_size_log); |
42d7395f | 1219 | |
42d7395f AK |
1220 | if (!h) |
1221 | return -1; | |
1222 | return h - hstates; | |
1223 | } | |
1224 | ||
be1d2cf5 | 1225 | static const struct dentry_operations anon_ops = { |
118b2302 | 1226 | .d_dname = simple_dname |
0df4d6e5 AV |
1227 | }; |
1228 | ||
af73e4d9 NH |
1229 | /* |
1230 | * Note that size should be aligned to proper hugepage size in caller side, | |
1231 | * otherwise hugetlb_reserve_pages reserves one less hugepages than intended. | |
1232 | */ | |
1233 | struct file *hugetlb_file_setup(const char *name, size_t size, | |
1234 | vm_flags_t acctflag, struct user_struct **user, | |
42d7395f | 1235 | int creat_flags, int page_size_log) |
1da177e4 | 1236 | { |
39b65252 | 1237 | struct file *file = ERR_PTR(-ENOMEM); |
1da177e4 | 1238 | struct inode *inode; |
2c48b9c4 | 1239 | struct path path; |
0df4d6e5 | 1240 | struct super_block *sb; |
1da177e4 | 1241 | struct qstr quick_string; |
42d7395f AK |
1242 | int hstate_idx; |
1243 | ||
1244 | hstate_idx = get_hstate_idx(page_size_log); | |
1245 | if (hstate_idx < 0) | |
1246 | return ERR_PTR(-ENODEV); | |
1da177e4 | 1247 | |
353d5c30 | 1248 | *user = NULL; |
42d7395f | 1249 | if (!hugetlbfs_vfsmount[hstate_idx]) |
5bc98594 AM |
1250 | return ERR_PTR(-ENOENT); |
1251 | ||
ef1ff6b8 | 1252 | if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) { |
353d5c30 HD |
1253 | *user = current_user(); |
1254 | if (user_shm_lock(size, *user)) { | |
21a3c273 | 1255 | task_lock(current); |
9b857d26 | 1256 | pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n", |
21a3c273 DR |
1257 | current->comm, current->pid); |
1258 | task_unlock(current); | |
353d5c30 HD |
1259 | } else { |
1260 | *user = NULL; | |
2584e517 | 1261 | return ERR_PTR(-EPERM); |
353d5c30 | 1262 | } |
2584e517 | 1263 | } |
1da177e4 | 1264 | |
0df4d6e5 | 1265 | sb = hugetlbfs_vfsmount[hstate_idx]->mnt_sb; |
9d66586f | 1266 | quick_string.name = name; |
1da177e4 LT |
1267 | quick_string.len = strlen(quick_string.name); |
1268 | quick_string.hash = 0; | |
0df4d6e5 | 1269 | path.dentry = d_alloc_pseudo(sb, &quick_string); |
2c48b9c4 | 1270 | if (!path.dentry) |
1da177e4 LT |
1271 | goto out_shm_unlock; |
1272 | ||
0df4d6e5 | 1273 | d_set_d_op(path.dentry, &anon_ops); |
42d7395f | 1274 | path.mnt = mntget(hugetlbfs_vfsmount[hstate_idx]); |
39b65252 | 1275 | file = ERR_PTR(-ENOSPC); |
0df4d6e5 | 1276 | inode = hugetlbfs_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0); |
1da177e4 | 1277 | if (!inode) |
ce8d2cdf | 1278 | goto out_dentry; |
e1832f29 SS |
1279 | if (creat_flags == HUGETLB_SHMFS_INODE) |
1280 | inode->i_flags |= S_PRIVATE; | |
1da177e4 | 1281 | |
39b65252 | 1282 | file = ERR_PTR(-ENOMEM); |
af73e4d9 NH |
1283 | if (hugetlb_reserve_pages(inode, 0, |
1284 | size >> huge_page_shift(hstate_inode(inode)), NULL, | |
1285 | acctflag)) | |
b45b5bd6 DG |
1286 | goto out_inode; |
1287 | ||
2c48b9c4 | 1288 | d_instantiate(path.dentry, inode); |
1da177e4 | 1289 | inode->i_size = size; |
6d6b77f1 | 1290 | clear_nlink(inode); |
ce8d2cdf | 1291 | |
2c48b9c4 | 1292 | file = alloc_file(&path, FMODE_WRITE | FMODE_READ, |
ce8d2cdf | 1293 | &hugetlbfs_file_operations); |
39b65252 | 1294 | if (IS_ERR(file)) |
b4d232e6 | 1295 | goto out_dentry; /* inode is already attached */ |
ce8d2cdf | 1296 | |
1da177e4 LT |
1297 | return file; |
1298 | ||
b45b5bd6 DG |
1299 | out_inode: |
1300 | iput(inode); | |
1da177e4 | 1301 | out_dentry: |
2c48b9c4 | 1302 | path_put(&path); |
1da177e4 | 1303 | out_shm_unlock: |
353d5c30 HD |
1304 | if (*user) { |
1305 | user_shm_unlock(size, *user); | |
1306 | *user = NULL; | |
1307 | } | |
39b65252 | 1308 | return file; |
1da177e4 LT |
1309 | } |
1310 | ||
1311 | static int __init init_hugetlbfs_fs(void) | |
1312 | { | |
42d7395f | 1313 | struct hstate *h; |
1da177e4 | 1314 | int error; |
42d7395f | 1315 | int i; |
1da177e4 | 1316 | |
457c1b27 | 1317 | if (!hugepages_supported()) { |
9b857d26 | 1318 | pr_info("disabling because there are no supported hugepage sizes\n"); |
457c1b27 NA |
1319 | return -ENOTSUPP; |
1320 | } | |
1321 | ||
d1d5e05f | 1322 | error = -ENOMEM; |
1da177e4 LT |
1323 | hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache", |
1324 | sizeof(struct hugetlbfs_inode_info), | |
5d097056 | 1325 | 0, SLAB_ACCOUNT, init_once); |
1da177e4 | 1326 | if (hugetlbfs_inode_cachep == NULL) |
e0bf68dd | 1327 | goto out2; |
1da177e4 LT |
1328 | |
1329 | error = register_filesystem(&hugetlbfs_fs_type); | |
1330 | if (error) | |
1331 | goto out; | |
1332 | ||
42d7395f AK |
1333 | i = 0; |
1334 | for_each_hstate(h) { | |
1335 | char buf[50]; | |
1336 | unsigned ps_kb = 1U << (h->order + PAGE_SHIFT - 10); | |
1da177e4 | 1337 | |
42d7395f AK |
1338 | snprintf(buf, sizeof(buf), "pagesize=%uK", ps_kb); |
1339 | hugetlbfs_vfsmount[i] = kern_mount_data(&hugetlbfs_fs_type, | |
1340 | buf); | |
1da177e4 | 1341 | |
42d7395f | 1342 | if (IS_ERR(hugetlbfs_vfsmount[i])) { |
9b857d26 | 1343 | pr_err("Cannot mount internal hugetlbfs for " |
42d7395f AK |
1344 | "page size %uK", ps_kb); |
1345 | error = PTR_ERR(hugetlbfs_vfsmount[i]); | |
1346 | hugetlbfs_vfsmount[i] = NULL; | |
1347 | } | |
1348 | i++; | |
1349 | } | |
1350 | /* Non default hstates are optional */ | |
1351 | if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount[default_hstate_idx])) | |
1352 | return 0; | |
1da177e4 LT |
1353 | |
1354 | out: | |
d1d5e05f | 1355 | kmem_cache_destroy(hugetlbfs_inode_cachep); |
e0bf68dd | 1356 | out2: |
1da177e4 LT |
1357 | return error; |
1358 | } | |
1359 | ||
1360 | static void __exit exit_hugetlbfs_fs(void) | |
1361 | { | |
42d7395f AK |
1362 | struct hstate *h; |
1363 | int i; | |
1364 | ||
1365 | ||
8c0a8537 KS |
1366 | /* |
1367 | * Make sure all delayed rcu free inodes are flushed before we | |
1368 | * destroy cache. | |
1369 | */ | |
1370 | rcu_barrier(); | |
1da177e4 | 1371 | kmem_cache_destroy(hugetlbfs_inode_cachep); |
42d7395f AK |
1372 | i = 0; |
1373 | for_each_hstate(h) | |
1374 | kern_unmount(hugetlbfs_vfsmount[i++]); | |
1da177e4 LT |
1375 | unregister_filesystem(&hugetlbfs_fs_type); |
1376 | } | |
1377 | ||
1378 | module_init(init_hugetlbfs_fs) | |
1379 | module_exit(exit_hugetlbfs_fs) | |
1380 | ||
1381 | MODULE_LICENSE("GPL"); |