]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * hugetlbpage-backed filesystem. Based on ramfs. | |
3 | * | |
6d49e352 | 4 | * Nadia Yvette Chambers, 2002 |
1da177e4 LT |
5 | * |
6 | * Copyright (C) 2002 Linus Torvalds. | |
3e89e1c5 | 7 | * License: GPL |
1da177e4 LT |
8 | */ |
9 | ||
9b857d26 AM |
10 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
11 | ||
1da177e4 LT |
12 | #include <linux/thread_info.h> |
13 | #include <asm/current.h> | |
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 | 465 | unsigned long v_offset; |
9aacdd35 | 466 | unsigned long v_end; |
1da177e4 | 467 | |
1da177e4 | 468 | /* |
856fc295 | 469 | * Can the expression below overflow on 32-bit arches? |
6b2dbba8 | 470 | * No, because the interval tree returns us only those vmas |
856fc295 HD |
471 | * which overlap the truncated area starting at pgoff, |
472 | * and no vma on a 32-bit arch can span beyond the 4GB. | |
1da177e4 | 473 | */ |
1bfad99a MK |
474 | if (vma->vm_pgoff < start) |
475 | v_offset = (start - vma->vm_pgoff) << PAGE_SHIFT; | |
856fc295 | 476 | else |
1da177e4 LT |
477 | v_offset = 0; |
478 | ||
9aacdd35 MK |
479 | if (!end) |
480 | v_end = vma->vm_end; | |
481 | else { | |
482 | v_end = ((end - vma->vm_pgoff) << PAGE_SHIFT) | |
483 | + vma->vm_start; | |
484 | if (v_end > vma->vm_end) | |
485 | v_end = vma->vm_end; | |
486 | } | |
1bfad99a | 487 | |
9aacdd35 MK |
488 | unmap_hugepage_range(vma, vma->vm_start + v_offset, v_end, |
489 | NULL); | |
1da177e4 LT |
490 | } |
491 | } | |
492 | ||
1da177e4 LT |
493 | static int hugetlb_vmtruncate(struct inode *inode, loff_t offset) |
494 | { | |
856fc295 | 495 | pgoff_t pgoff; |
1da177e4 | 496 | struct address_space *mapping = inode->i_mapping; |
a5516438 | 497 | struct hstate *h = hstate_inode(inode); |
1da177e4 | 498 | |
a5516438 | 499 | BUG_ON(offset & ~huge_page_mask(h)); |
856fc295 | 500 | pgoff = offset >> PAGE_SHIFT; |
1da177e4 | 501 | |
7aa91e10 | 502 | i_size_write(inode, offset); |
83cde9e8 | 503 | i_mmap_lock_write(mapping); |
6b2dbba8 | 504 | if (!RB_EMPTY_ROOT(&mapping->i_mmap)) |
1bfad99a | 505 | hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0); |
83cde9e8 | 506 | i_mmap_unlock_write(mapping); |
b5cec28d | 507 | remove_inode_hugepages(inode, offset, LLONG_MAX); |
1da177e4 LT |
508 | return 0; |
509 | } | |
510 | ||
70c3547e MK |
511 | static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len) |
512 | { | |
513 | struct hstate *h = hstate_inode(inode); | |
514 | loff_t hpage_size = huge_page_size(h); | |
515 | loff_t hole_start, hole_end; | |
516 | ||
517 | /* | |
518 | * For hole punch round up the beginning offset of the hole and | |
519 | * round down the end. | |
520 | */ | |
521 | hole_start = round_up(offset, hpage_size); | |
522 | hole_end = round_down(offset + len, hpage_size); | |
523 | ||
524 | if (hole_end > hole_start) { | |
525 | struct address_space *mapping = inode->i_mapping; | |
526 | ||
527 | mutex_lock(&inode->i_mutex); | |
528 | i_mmap_lock_write(mapping); | |
529 | if (!RB_EMPTY_ROOT(&mapping->i_mmap)) | |
530 | hugetlb_vmdelete_list(&mapping->i_mmap, | |
531 | hole_start >> PAGE_SHIFT, | |
532 | hole_end >> PAGE_SHIFT); | |
533 | i_mmap_unlock_write(mapping); | |
534 | remove_inode_hugepages(inode, hole_start, hole_end); | |
535 | mutex_unlock(&inode->i_mutex); | |
536 | } | |
537 | ||
538 | return 0; | |
539 | } | |
540 | ||
541 | static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset, | |
542 | loff_t len) | |
543 | { | |
544 | struct inode *inode = file_inode(file); | |
545 | struct address_space *mapping = inode->i_mapping; | |
546 | struct hstate *h = hstate_inode(inode); | |
547 | struct vm_area_struct pseudo_vma; | |
548 | struct mm_struct *mm = current->mm; | |
549 | loff_t hpage_size = huge_page_size(h); | |
550 | unsigned long hpage_shift = huge_page_shift(h); | |
551 | pgoff_t start, index, end; | |
552 | int error; | |
553 | u32 hash; | |
554 | ||
555 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
556 | return -EOPNOTSUPP; | |
557 | ||
558 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
559 | return hugetlbfs_punch_hole(inode, offset, len); | |
560 | ||
561 | /* | |
562 | * Default preallocate case. | |
563 | * For this range, start is rounded down and end is rounded up | |
564 | * as well as being converted to page offsets. | |
565 | */ | |
566 | start = offset >> hpage_shift; | |
567 | end = (offset + len + hpage_size - 1) >> hpage_shift; | |
568 | ||
569 | mutex_lock(&inode->i_mutex); | |
570 | ||
571 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
572 | error = inode_newsize_ok(inode, offset + len); | |
573 | if (error) | |
574 | goto out; | |
575 | ||
576 | /* | |
577 | * Initialize a pseudo vma as this is required by the huge page | |
578 | * allocation routines. If NUMA is configured, use page index | |
579 | * as input to create an allocation policy. | |
580 | */ | |
581 | memset(&pseudo_vma, 0, sizeof(struct vm_area_struct)); | |
582 | pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); | |
583 | pseudo_vma.vm_file = file; | |
584 | ||
585 | for (index = start; index < end; index++) { | |
586 | /* | |
587 | * This is supposed to be the vaddr where the page is being | |
588 | * faulted in, but we have no vaddr here. | |
589 | */ | |
590 | struct page *page; | |
591 | unsigned long addr; | |
592 | int avoid_reserve = 0; | |
593 | ||
594 | cond_resched(); | |
595 | ||
596 | /* | |
597 | * fallocate(2) manpage permits EINTR; we may have been | |
598 | * interrupted because we are using up too much memory. | |
599 | */ | |
600 | if (signal_pending(current)) { | |
601 | error = -EINTR; | |
602 | break; | |
603 | } | |
604 | ||
605 | /* Set numa allocation policy based on index */ | |
606 | hugetlb_set_vma_policy(&pseudo_vma, inode, index); | |
607 | ||
608 | /* addr is the offset within the file (zero based) */ | |
609 | addr = index * hpage_size; | |
610 | ||
611 | /* mutex taken here, fault path and hole punch */ | |
612 | hash = hugetlb_fault_mutex_hash(h, mm, &pseudo_vma, mapping, | |
613 | index, addr); | |
614 | mutex_lock(&hugetlb_fault_mutex_table[hash]); | |
615 | ||
616 | /* See if already present in mapping to avoid alloc/free */ | |
617 | page = find_get_page(mapping, index); | |
618 | if (page) { | |
619 | put_page(page); | |
620 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
621 | hugetlb_drop_vma_policy(&pseudo_vma); | |
622 | continue; | |
623 | } | |
624 | ||
625 | /* Allocate page and add to page cache */ | |
626 | page = alloc_huge_page(&pseudo_vma, addr, avoid_reserve); | |
627 | hugetlb_drop_vma_policy(&pseudo_vma); | |
628 | if (IS_ERR(page)) { | |
629 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
630 | error = PTR_ERR(page); | |
631 | goto out; | |
632 | } | |
633 | clear_huge_page(page, addr, pages_per_huge_page(h)); | |
634 | __SetPageUptodate(page); | |
635 | error = huge_add_to_page_cache(page, mapping, index); | |
636 | if (unlikely(error)) { | |
637 | put_page(page); | |
638 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
639 | goto out; | |
640 | } | |
641 | ||
642 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
643 | ||
644 | /* | |
645 | * page_put due to reference from alloc_huge_page() | |
646 | * unlock_page because locked by add_to_page_cache() | |
647 | */ | |
648 | put_page(page); | |
649 | unlock_page(page); | |
650 | } | |
651 | ||
652 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
653 | i_size_write(inode, offset + len); | |
654 | inode->i_ctime = CURRENT_TIME; | |
70c3547e MK |
655 | out: |
656 | mutex_unlock(&inode->i_mutex); | |
657 | return error; | |
658 | } | |
659 | ||
1da177e4 LT |
660 | static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr) |
661 | { | |
2b0143b5 | 662 | struct inode *inode = d_inode(dentry); |
a5516438 | 663 | struct hstate *h = hstate_inode(inode); |
1da177e4 LT |
664 | int error; |
665 | unsigned int ia_valid = attr->ia_valid; | |
666 | ||
667 | BUG_ON(!inode); | |
668 | ||
669 | error = inode_change_ok(inode, attr); | |
670 | if (error) | |
1025774c | 671 | return error; |
1da177e4 LT |
672 | |
673 | if (ia_valid & ATTR_SIZE) { | |
674 | error = -EINVAL; | |
1025774c CH |
675 | if (attr->ia_size & ~huge_page_mask(h)) |
676 | return -EINVAL; | |
677 | error = hugetlb_vmtruncate(inode, attr->ia_size); | |
1da177e4 | 678 | if (error) |
1025774c | 679 | return error; |
1da177e4 | 680 | } |
1025774c CH |
681 | |
682 | setattr_copy(inode, attr); | |
683 | mark_inode_dirty(inode); | |
684 | return 0; | |
1da177e4 LT |
685 | } |
686 | ||
7d54fa64 AV |
687 | static struct inode *hugetlbfs_get_root(struct super_block *sb, |
688 | struct hugetlbfs_config *config) | |
1da177e4 LT |
689 | { |
690 | struct inode *inode; | |
1da177e4 LT |
691 | |
692 | inode = new_inode(sb); | |
693 | if (inode) { | |
694 | struct hugetlbfs_inode_info *info; | |
85fe4025 | 695 | inode->i_ino = get_next_ino(); |
7d54fa64 AV |
696 | inode->i_mode = S_IFDIR | config->mode; |
697 | inode->i_uid = config->uid; | |
698 | inode->i_gid = config->gid; | |
699 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; | |
700 | info = HUGETLBFS_I(inode); | |
701 | mpol_shared_policy_init(&info->policy, NULL); | |
702 | inode->i_op = &hugetlbfs_dir_inode_operations; | |
703 | inode->i_fop = &simple_dir_operations; | |
704 | /* directory inodes start off with i_nlink == 2 (for "." entry) */ | |
705 | inc_nlink(inode); | |
65ed7601 | 706 | lockdep_annotate_inode_mutex_key(inode); |
7d54fa64 AV |
707 | } |
708 | return inode; | |
709 | } | |
710 | ||
b610ded7 | 711 | /* |
c8c06efa | 712 | * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never |
b610ded7 | 713 | * be taken from reclaim -- unlike regular filesystems. This needs an |
88f306b6 | 714 | * annotation because huge_pmd_share() does an allocation under hugetlb's |
c8c06efa | 715 | * i_mmap_rwsem. |
b610ded7 | 716 | */ |
c8c06efa | 717 | static struct lock_class_key hugetlbfs_i_mmap_rwsem_key; |
b610ded7 | 718 | |
7d54fa64 AV |
719 | static struct inode *hugetlbfs_get_inode(struct super_block *sb, |
720 | struct inode *dir, | |
18df2252 | 721 | umode_t mode, dev_t dev) |
7d54fa64 AV |
722 | { |
723 | struct inode *inode; | |
9119a41e JK |
724 | struct resv_map *resv_map; |
725 | ||
726 | resv_map = resv_map_alloc(); | |
727 | if (!resv_map) | |
728 | return NULL; | |
7d54fa64 AV |
729 | |
730 | inode = new_inode(sb); | |
731 | if (inode) { | |
732 | struct hugetlbfs_inode_info *info; | |
733 | inode->i_ino = get_next_ino(); | |
734 | inode_init_owner(inode, dir, mode); | |
c8c06efa DB |
735 | lockdep_set_class(&inode->i_mapping->i_mmap_rwsem, |
736 | &hugetlbfs_i_mmap_rwsem_key); | |
1da177e4 | 737 | inode->i_mapping->a_ops = &hugetlbfs_aops; |
1da177e4 | 738 | inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME; |
9119a41e | 739 | inode->i_mapping->private_data = resv_map; |
1da177e4 | 740 | info = HUGETLBFS_I(inode); |
6bfde05b EM |
741 | /* |
742 | * The policy is initialized here even if we are creating a | |
743 | * private inode because initialization simply creates an | |
4a8c7bb5 | 744 | * an empty rb tree and calls rwlock_init(), later when we |
6bfde05b EM |
745 | * call mpol_free_shared_policy() it will just return because |
746 | * the rb tree will still be empty. | |
747 | */ | |
71fe804b | 748 | mpol_shared_policy_init(&info->policy, NULL); |
1da177e4 LT |
749 | switch (mode & S_IFMT) { |
750 | default: | |
751 | init_special_inode(inode, mode, dev); | |
752 | break; | |
753 | case S_IFREG: | |
754 | inode->i_op = &hugetlbfs_inode_operations; | |
755 | inode->i_fop = &hugetlbfs_file_operations; | |
756 | break; | |
757 | case S_IFDIR: | |
758 | inode->i_op = &hugetlbfs_dir_inode_operations; | |
759 | inode->i_fop = &simple_dir_operations; | |
760 | ||
761 | /* directory inodes start off with i_nlink == 2 (for "." entry) */ | |
d8c76e6f | 762 | inc_nlink(inode); |
1da177e4 LT |
763 | break; |
764 | case S_IFLNK: | |
765 | inode->i_op = &page_symlink_inode_operations; | |
21fc61c7 | 766 | inode_nohighmem(inode); |
1da177e4 LT |
767 | break; |
768 | } | |
e096d0c7 | 769 | lockdep_annotate_inode_mutex_key(inode); |
9119a41e JK |
770 | } else |
771 | kref_put(&resv_map->refs, resv_map_release); | |
772 | ||
1da177e4 LT |
773 | return inode; |
774 | } | |
775 | ||
776 | /* | |
777 | * File creation. Allocate an inode, and we're done.. | |
778 | */ | |
779 | static int hugetlbfs_mknod(struct inode *dir, | |
1a67aafb | 780 | struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 LT |
781 | { |
782 | struct inode *inode; | |
783 | int error = -ENOSPC; | |
7d54fa64 AV |
784 | |
785 | inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev); | |
1da177e4 LT |
786 | if (inode) { |
787 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
788 | d_instantiate(dentry, inode); | |
789 | dget(dentry); /* Extra count - pin the dentry in core */ | |
790 | error = 0; | |
791 | } | |
792 | return error; | |
793 | } | |
794 | ||
18bb1db3 | 795 | static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
796 | { |
797 | int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0); | |
798 | if (!retval) | |
d8c76e6f | 799 | inc_nlink(dir); |
1da177e4 LT |
800 | return retval; |
801 | } | |
802 | ||
ebfc3b49 | 803 | static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl) |
1da177e4 LT |
804 | { |
805 | return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0); | |
806 | } | |
807 | ||
808 | static int hugetlbfs_symlink(struct inode *dir, | |
809 | struct dentry *dentry, const char *symname) | |
810 | { | |
811 | struct inode *inode; | |
812 | int error = -ENOSPC; | |
1da177e4 | 813 | |
7d54fa64 | 814 | inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0); |
1da177e4 LT |
815 | if (inode) { |
816 | int l = strlen(symname)+1; | |
817 | error = page_symlink(inode, symname, l); | |
818 | if (!error) { | |
819 | d_instantiate(dentry, inode); | |
820 | dget(dentry); | |
821 | } else | |
822 | iput(inode); | |
823 | } | |
824 | dir->i_ctime = dir->i_mtime = CURRENT_TIME; | |
825 | ||
826 | return error; | |
827 | } | |
828 | ||
829 | /* | |
6649a386 | 830 | * mark the head page dirty |
1da177e4 LT |
831 | */ |
832 | static int hugetlbfs_set_page_dirty(struct page *page) | |
833 | { | |
d85f3385 | 834 | struct page *head = compound_head(page); |
6649a386 KC |
835 | |
836 | SetPageDirty(head); | |
1da177e4 LT |
837 | return 0; |
838 | } | |
839 | ||
290408d4 | 840 | static int hugetlbfs_migrate_page(struct address_space *mapping, |
b969c4ab | 841 | struct page *newpage, struct page *page, |
a6bc32b8 | 842 | enum migrate_mode mode) |
290408d4 NH |
843 | { |
844 | int rc; | |
845 | ||
846 | rc = migrate_huge_page_move_mapping(mapping, newpage, page); | |
78bd5209 | 847 | if (rc != MIGRATEPAGE_SUCCESS) |
290408d4 NH |
848 | return rc; |
849 | migrate_page_copy(newpage, page); | |
850 | ||
78bd5209 | 851 | return MIGRATEPAGE_SUCCESS; |
290408d4 NH |
852 | } |
853 | ||
726c3342 | 854 | static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 855 | { |
726c3342 | 856 | struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb); |
2b0143b5 | 857 | struct hstate *h = hstate_inode(d_inode(dentry)); |
1da177e4 LT |
858 | |
859 | buf->f_type = HUGETLBFS_MAGIC; | |
a5516438 | 860 | buf->f_bsize = huge_page_size(h); |
1da177e4 LT |
861 | if (sbinfo) { |
862 | spin_lock(&sbinfo->stat_lock); | |
74a8a65c DG |
863 | /* If no limits set, just report 0 for max/free/used |
864 | * blocks, like simple_statfs() */ | |
90481622 DG |
865 | if (sbinfo->spool) { |
866 | long free_pages; | |
867 | ||
868 | spin_lock(&sbinfo->spool->lock); | |
869 | buf->f_blocks = sbinfo->spool->max_hpages; | |
870 | free_pages = sbinfo->spool->max_hpages | |
871 | - sbinfo->spool->used_hpages; | |
872 | buf->f_bavail = buf->f_bfree = free_pages; | |
873 | spin_unlock(&sbinfo->spool->lock); | |
74a8a65c DG |
874 | buf->f_files = sbinfo->max_inodes; |
875 | buf->f_ffree = sbinfo->free_inodes; | |
876 | } | |
1da177e4 LT |
877 | spin_unlock(&sbinfo->stat_lock); |
878 | } | |
879 | buf->f_namelen = NAME_MAX; | |
880 | return 0; | |
881 | } | |
882 | ||
883 | static void hugetlbfs_put_super(struct super_block *sb) | |
884 | { | |
885 | struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb); | |
886 | ||
887 | if (sbi) { | |
888 | sb->s_fs_info = NULL; | |
90481622 DG |
889 | |
890 | if (sbi->spool) | |
891 | hugepage_put_subpool(sbi->spool); | |
892 | ||
1da177e4 LT |
893 | kfree(sbi); |
894 | } | |
895 | } | |
896 | ||
96527980 CH |
897 | static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo) |
898 | { | |
899 | if (sbinfo->free_inodes >= 0) { | |
900 | spin_lock(&sbinfo->stat_lock); | |
901 | if (unlikely(!sbinfo->free_inodes)) { | |
902 | spin_unlock(&sbinfo->stat_lock); | |
903 | return 0; | |
904 | } | |
905 | sbinfo->free_inodes--; | |
906 | spin_unlock(&sbinfo->stat_lock); | |
907 | } | |
908 | ||
909 | return 1; | |
910 | } | |
911 | ||
912 | static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo) | |
913 | { | |
914 | if (sbinfo->free_inodes >= 0) { | |
915 | spin_lock(&sbinfo->stat_lock); | |
916 | sbinfo->free_inodes++; | |
917 | spin_unlock(&sbinfo->stat_lock); | |
918 | } | |
919 | } | |
920 | ||
921 | ||
e18b890b | 922 | static struct kmem_cache *hugetlbfs_inode_cachep; |
1da177e4 LT |
923 | |
924 | static struct inode *hugetlbfs_alloc_inode(struct super_block *sb) | |
925 | { | |
96527980 | 926 | struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb); |
1da177e4 LT |
927 | struct hugetlbfs_inode_info *p; |
928 | ||
96527980 CH |
929 | if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo))) |
930 | return NULL; | |
e94b1766 | 931 | p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL); |
96527980 CH |
932 | if (unlikely(!p)) { |
933 | hugetlbfs_inc_free_inodes(sbinfo); | |
1da177e4 | 934 | return NULL; |
96527980 | 935 | } |
1da177e4 LT |
936 | return &p->vfs_inode; |
937 | } | |
938 | ||
fa0d7e3d NP |
939 | static void hugetlbfs_i_callback(struct rcu_head *head) |
940 | { | |
941 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
942 | kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode)); |
943 | } | |
944 | ||
1da177e4 LT |
945 | static void hugetlbfs_destroy_inode(struct inode *inode) |
946 | { | |
96527980 | 947 | hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb)); |
1da177e4 | 948 | mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy); |
fa0d7e3d | 949 | call_rcu(&inode->i_rcu, hugetlbfs_i_callback); |
1da177e4 LT |
950 | } |
951 | ||
f5e54d6e | 952 | static const struct address_space_operations hugetlbfs_aops = { |
800d15a5 NP |
953 | .write_begin = hugetlbfs_write_begin, |
954 | .write_end = hugetlbfs_write_end, | |
1da177e4 | 955 | .set_page_dirty = hugetlbfs_set_page_dirty, |
290408d4 | 956 | .migratepage = hugetlbfs_migrate_page, |
1da177e4 LT |
957 | }; |
958 | ||
96527980 | 959 | |
51cc5068 | 960 | static void init_once(void *foo) |
96527980 CH |
961 | { |
962 | struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo; | |
963 | ||
a35afb83 | 964 | inode_init_once(&ei->vfs_inode); |
96527980 CH |
965 | } |
966 | ||
4b6f5d20 | 967 | const struct file_operations hugetlbfs_file_operations = { |
34d0640e | 968 | .read_iter = hugetlbfs_read_iter, |
1da177e4 | 969 | .mmap = hugetlbfs_file_mmap, |
1b061d92 | 970 | .fsync = noop_fsync, |
1da177e4 | 971 | .get_unmapped_area = hugetlb_get_unmapped_area, |
70c3547e MK |
972 | .llseek = default_llseek, |
973 | .fallocate = hugetlbfs_fallocate, | |
1da177e4 LT |
974 | }; |
975 | ||
92e1d5be | 976 | static const struct inode_operations hugetlbfs_dir_inode_operations = { |
1da177e4 LT |
977 | .create = hugetlbfs_create, |
978 | .lookup = simple_lookup, | |
979 | .link = simple_link, | |
980 | .unlink = simple_unlink, | |
981 | .symlink = hugetlbfs_symlink, | |
982 | .mkdir = hugetlbfs_mkdir, | |
983 | .rmdir = simple_rmdir, | |
984 | .mknod = hugetlbfs_mknod, | |
985 | .rename = simple_rename, | |
986 | .setattr = hugetlbfs_setattr, | |
987 | }; | |
988 | ||
92e1d5be | 989 | static const struct inode_operations hugetlbfs_inode_operations = { |
1da177e4 LT |
990 | .setattr = hugetlbfs_setattr, |
991 | }; | |
992 | ||
ee9b6d61 | 993 | static const struct super_operations hugetlbfs_ops = { |
1da177e4 LT |
994 | .alloc_inode = hugetlbfs_alloc_inode, |
995 | .destroy_inode = hugetlbfs_destroy_inode, | |
2bbbda30 | 996 | .evict_inode = hugetlbfs_evict_inode, |
1da177e4 | 997 | .statfs = hugetlbfs_statfs, |
1da177e4 | 998 | .put_super = hugetlbfs_put_super, |
10f19a86 | 999 | .show_options = generic_show_options, |
1da177e4 LT |
1000 | }; |
1001 | ||
7ca02d0a MK |
1002 | enum { NO_SIZE, SIZE_STD, SIZE_PERCENT }; |
1003 | ||
1004 | /* | |
1005 | * Convert size option passed from command line to number of huge pages | |
1006 | * in the pool specified by hstate. Size option could be in bytes | |
1007 | * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT). | |
1008 | */ | |
1009 | static long long | |
1010 | hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt, | |
1011 | int val_type) | |
1012 | { | |
1013 | if (val_type == NO_SIZE) | |
1014 | return -1; | |
1015 | ||
1016 | if (val_type == SIZE_PERCENT) { | |
1017 | size_opt <<= huge_page_shift(h); | |
1018 | size_opt *= h->max_huge_pages; | |
1019 | do_div(size_opt, 100); | |
1020 | } | |
1021 | ||
1022 | size_opt >>= huge_page_shift(h); | |
1023 | return size_opt; | |
1024 | } | |
1025 | ||
1da177e4 LT |
1026 | static int |
1027 | hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig) | |
1028 | { | |
e73a75fa RD |
1029 | char *p, *rest; |
1030 | substring_t args[MAX_OPT_ARGS]; | |
1031 | int option; | |
7ca02d0a MK |
1032 | unsigned long long max_size_opt = 0, min_size_opt = 0; |
1033 | int max_val_type = NO_SIZE, min_val_type = NO_SIZE; | |
1da177e4 LT |
1034 | |
1035 | if (!options) | |
1036 | return 0; | |
1da177e4 | 1037 | |
e73a75fa RD |
1038 | while ((p = strsep(&options, ",")) != NULL) { |
1039 | int token; | |
b4c07bce LS |
1040 | if (!*p) |
1041 | continue; | |
e73a75fa RD |
1042 | |
1043 | token = match_token(p, tokens, args); | |
1044 | switch (token) { | |
1045 | case Opt_uid: | |
1046 | if (match_int(&args[0], &option)) | |
1047 | goto bad_val; | |
a0eb3a05 EB |
1048 | pconfig->uid = make_kuid(current_user_ns(), option); |
1049 | if (!uid_valid(pconfig->uid)) | |
1050 | goto bad_val; | |
e73a75fa RD |
1051 | break; |
1052 | ||
1053 | case Opt_gid: | |
1054 | if (match_int(&args[0], &option)) | |
1055 | goto bad_val; | |
a0eb3a05 EB |
1056 | pconfig->gid = make_kgid(current_user_ns(), option); |
1057 | if (!gid_valid(pconfig->gid)) | |
1058 | goto bad_val; | |
e73a75fa RD |
1059 | break; |
1060 | ||
1061 | case Opt_mode: | |
1062 | if (match_octal(&args[0], &option)) | |
1063 | goto bad_val; | |
75897d60 | 1064 | pconfig->mode = option & 01777U; |
e73a75fa RD |
1065 | break; |
1066 | ||
1067 | case Opt_size: { | |
e73a75fa RD |
1068 | /* memparse() will accept a K/M/G without a digit */ |
1069 | if (!isdigit(*args[0].from)) | |
1070 | goto bad_val; | |
7ca02d0a MK |
1071 | max_size_opt = memparse(args[0].from, &rest); |
1072 | max_val_type = SIZE_STD; | |
a137e1cc | 1073 | if (*rest == '%') |
7ca02d0a | 1074 | max_val_type = SIZE_PERCENT; |
e73a75fa RD |
1075 | break; |
1076 | } | |
1da177e4 | 1077 | |
e73a75fa RD |
1078 | case Opt_nr_inodes: |
1079 | /* memparse() will accept a K/M/G without a digit */ | |
1080 | if (!isdigit(*args[0].from)) | |
1081 | goto bad_val; | |
1082 | pconfig->nr_inodes = memparse(args[0].from, &rest); | |
1083 | break; | |
1084 | ||
a137e1cc AK |
1085 | case Opt_pagesize: { |
1086 | unsigned long ps; | |
1087 | ps = memparse(args[0].from, &rest); | |
1088 | pconfig->hstate = size_to_hstate(ps); | |
1089 | if (!pconfig->hstate) { | |
9b857d26 | 1090 | pr_err("Unsupported page size %lu MB\n", |
a137e1cc AK |
1091 | ps >> 20); |
1092 | return -EINVAL; | |
1093 | } | |
1094 | break; | |
1095 | } | |
1096 | ||
7ca02d0a MK |
1097 | case Opt_min_size: { |
1098 | /* memparse() will accept a K/M/G without a digit */ | |
1099 | if (!isdigit(*args[0].from)) | |
1100 | goto bad_val; | |
1101 | min_size_opt = memparse(args[0].from, &rest); | |
1102 | min_val_type = SIZE_STD; | |
1103 | if (*rest == '%') | |
1104 | min_val_type = SIZE_PERCENT; | |
1105 | break; | |
1106 | } | |
1107 | ||
e73a75fa | 1108 | default: |
9b857d26 | 1109 | pr_err("Bad mount option: \"%s\"\n", p); |
b4c07bce | 1110 | return -EINVAL; |
e73a75fa RD |
1111 | break; |
1112 | } | |
1da177e4 | 1113 | } |
a137e1cc | 1114 | |
7ca02d0a MK |
1115 | /* |
1116 | * Use huge page pool size (in hstate) to convert the size | |
1117 | * options to number of huge pages. If NO_SIZE, -1 is returned. | |
1118 | */ | |
1119 | pconfig->max_hpages = hugetlbfs_size_to_hpages(pconfig->hstate, | |
1120 | max_size_opt, max_val_type); | |
1121 | pconfig->min_hpages = hugetlbfs_size_to_hpages(pconfig->hstate, | |
1122 | min_size_opt, min_val_type); | |
1123 | ||
1124 | /* | |
1125 | * If max_size was specified, then min_size must be smaller | |
1126 | */ | |
1127 | if (max_val_type > NO_SIZE && | |
1128 | pconfig->min_hpages > pconfig->max_hpages) { | |
1129 | pr_err("minimum size can not be greater than maximum size\n"); | |
1130 | return -EINVAL; | |
a137e1cc AK |
1131 | } |
1132 | ||
1da177e4 | 1133 | return 0; |
e73a75fa RD |
1134 | |
1135 | bad_val: | |
9b857d26 | 1136 | pr_err("Bad value '%s' for mount option '%s'\n", args[0].from, p); |
c12ddba0 | 1137 | return -EINVAL; |
1da177e4 LT |
1138 | } |
1139 | ||
1140 | static int | |
1141 | hugetlbfs_fill_super(struct super_block *sb, void *data, int silent) | |
1142 | { | |
1da177e4 LT |
1143 | int ret; |
1144 | struct hugetlbfs_config config; | |
1145 | struct hugetlbfs_sb_info *sbinfo; | |
1146 | ||
10f19a86 MS |
1147 | save_mount_options(sb, data); |
1148 | ||
7ca02d0a | 1149 | config.max_hpages = -1; /* No limit on size by default */ |
1da177e4 | 1150 | config.nr_inodes = -1; /* No limit on number of inodes by default */ |
77c70de1 DH |
1151 | config.uid = current_fsuid(); |
1152 | config.gid = current_fsgid(); | |
1da177e4 | 1153 | config.mode = 0755; |
a137e1cc | 1154 | config.hstate = &default_hstate; |
7ca02d0a | 1155 | config.min_hpages = -1; /* No default minimum size */ |
1da177e4 | 1156 | ret = hugetlbfs_parse_options(data, &config); |
1da177e4 LT |
1157 | if (ret) |
1158 | return ret; | |
1159 | ||
1160 | sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL); | |
1161 | if (!sbinfo) | |
1162 | return -ENOMEM; | |
1163 | sb->s_fs_info = sbinfo; | |
a137e1cc | 1164 | sbinfo->hstate = config.hstate; |
1da177e4 | 1165 | spin_lock_init(&sbinfo->stat_lock); |
1da177e4 LT |
1166 | sbinfo->max_inodes = config.nr_inodes; |
1167 | sbinfo->free_inodes = config.nr_inodes; | |
90481622 | 1168 | sbinfo->spool = NULL; |
7ca02d0a MK |
1169 | /* |
1170 | * Allocate and initialize subpool if maximum or minimum size is | |
1171 | * specified. Any needed reservations (for minimim size) are taken | |
1172 | * taken when the subpool is created. | |
1173 | */ | |
1174 | if (config.max_hpages != -1 || config.min_hpages != -1) { | |
1175 | sbinfo->spool = hugepage_new_subpool(config.hstate, | |
1176 | config.max_hpages, | |
1177 | config.min_hpages); | |
90481622 DG |
1178 | if (!sbinfo->spool) |
1179 | goto out_free; | |
1180 | } | |
1da177e4 | 1181 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
a137e1cc AK |
1182 | sb->s_blocksize = huge_page_size(config.hstate); |
1183 | sb->s_blocksize_bits = huge_page_shift(config.hstate); | |
1da177e4 LT |
1184 | sb->s_magic = HUGETLBFS_MAGIC; |
1185 | sb->s_op = &hugetlbfs_ops; | |
1186 | sb->s_time_gran = 1; | |
48fde701 AV |
1187 | sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config)); |
1188 | if (!sb->s_root) | |
1da177e4 | 1189 | goto out_free; |
1da177e4 LT |
1190 | return 0; |
1191 | out_free: | |
6e6870d4 | 1192 | kfree(sbinfo->spool); |
1da177e4 LT |
1193 | kfree(sbinfo); |
1194 | return -ENOMEM; | |
1195 | } | |
1196 | ||
3c26ff6e AV |
1197 | static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type, |
1198 | int flags, const char *dev_name, void *data) | |
1da177e4 | 1199 | { |
3c26ff6e | 1200 | return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super); |
1da177e4 LT |
1201 | } |
1202 | ||
1203 | static struct file_system_type hugetlbfs_fs_type = { | |
1204 | .name = "hugetlbfs", | |
3c26ff6e | 1205 | .mount = hugetlbfs_mount, |
1da177e4 LT |
1206 | .kill_sb = kill_litter_super, |
1207 | }; | |
1208 | ||
42d7395f | 1209 | static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE]; |
1da177e4 | 1210 | |
ef1ff6b8 | 1211 | static int can_do_hugetlb_shm(void) |
1da177e4 | 1212 | { |
a0eb3a05 EB |
1213 | kgid_t shm_group; |
1214 | shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group); | |
1215 | return capable(CAP_IPC_LOCK) || in_group_p(shm_group); | |
1da177e4 LT |
1216 | } |
1217 | ||
42d7395f AK |
1218 | static int get_hstate_idx(int page_size_log) |
1219 | { | |
af73e4d9 | 1220 | struct hstate *h = hstate_sizelog(page_size_log); |
42d7395f | 1221 | |
42d7395f AK |
1222 | if (!h) |
1223 | return -1; | |
1224 | return h - hstates; | |
1225 | } | |
1226 | ||
be1d2cf5 | 1227 | static const struct dentry_operations anon_ops = { |
118b2302 | 1228 | .d_dname = simple_dname |
0df4d6e5 AV |
1229 | }; |
1230 | ||
af73e4d9 NH |
1231 | /* |
1232 | * Note that size should be aligned to proper hugepage size in caller side, | |
1233 | * otherwise hugetlb_reserve_pages reserves one less hugepages than intended. | |
1234 | */ | |
1235 | struct file *hugetlb_file_setup(const char *name, size_t size, | |
1236 | vm_flags_t acctflag, struct user_struct **user, | |
42d7395f | 1237 | int creat_flags, int page_size_log) |
1da177e4 | 1238 | { |
39b65252 | 1239 | struct file *file = ERR_PTR(-ENOMEM); |
1da177e4 | 1240 | struct inode *inode; |
2c48b9c4 | 1241 | struct path path; |
0df4d6e5 | 1242 | struct super_block *sb; |
1da177e4 | 1243 | struct qstr quick_string; |
42d7395f AK |
1244 | int hstate_idx; |
1245 | ||
1246 | hstate_idx = get_hstate_idx(page_size_log); | |
1247 | if (hstate_idx < 0) | |
1248 | return ERR_PTR(-ENODEV); | |
1da177e4 | 1249 | |
353d5c30 | 1250 | *user = NULL; |
42d7395f | 1251 | if (!hugetlbfs_vfsmount[hstate_idx]) |
5bc98594 AM |
1252 | return ERR_PTR(-ENOENT); |
1253 | ||
ef1ff6b8 | 1254 | if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) { |
353d5c30 HD |
1255 | *user = current_user(); |
1256 | if (user_shm_lock(size, *user)) { | |
21a3c273 | 1257 | task_lock(current); |
9b857d26 | 1258 | pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n", |
21a3c273 DR |
1259 | current->comm, current->pid); |
1260 | task_unlock(current); | |
353d5c30 HD |
1261 | } else { |
1262 | *user = NULL; | |
2584e517 | 1263 | return ERR_PTR(-EPERM); |
353d5c30 | 1264 | } |
2584e517 | 1265 | } |
1da177e4 | 1266 | |
0df4d6e5 | 1267 | sb = hugetlbfs_vfsmount[hstate_idx]->mnt_sb; |
9d66586f | 1268 | quick_string.name = name; |
1da177e4 LT |
1269 | quick_string.len = strlen(quick_string.name); |
1270 | quick_string.hash = 0; | |
0df4d6e5 | 1271 | path.dentry = d_alloc_pseudo(sb, &quick_string); |
2c48b9c4 | 1272 | if (!path.dentry) |
1da177e4 LT |
1273 | goto out_shm_unlock; |
1274 | ||
0df4d6e5 | 1275 | d_set_d_op(path.dentry, &anon_ops); |
42d7395f | 1276 | path.mnt = mntget(hugetlbfs_vfsmount[hstate_idx]); |
39b65252 | 1277 | file = ERR_PTR(-ENOSPC); |
0df4d6e5 | 1278 | inode = hugetlbfs_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0); |
1da177e4 | 1279 | if (!inode) |
ce8d2cdf | 1280 | goto out_dentry; |
e1832f29 SS |
1281 | if (creat_flags == HUGETLB_SHMFS_INODE) |
1282 | inode->i_flags |= S_PRIVATE; | |
1da177e4 | 1283 | |
39b65252 | 1284 | file = ERR_PTR(-ENOMEM); |
af73e4d9 NH |
1285 | if (hugetlb_reserve_pages(inode, 0, |
1286 | size >> huge_page_shift(hstate_inode(inode)), NULL, | |
1287 | acctflag)) | |
b45b5bd6 DG |
1288 | goto out_inode; |
1289 | ||
2c48b9c4 | 1290 | d_instantiate(path.dentry, inode); |
1da177e4 | 1291 | inode->i_size = size; |
6d6b77f1 | 1292 | clear_nlink(inode); |
ce8d2cdf | 1293 | |
2c48b9c4 | 1294 | file = alloc_file(&path, FMODE_WRITE | FMODE_READ, |
ce8d2cdf | 1295 | &hugetlbfs_file_operations); |
39b65252 | 1296 | if (IS_ERR(file)) |
b4d232e6 | 1297 | goto out_dentry; /* inode is already attached */ |
ce8d2cdf | 1298 | |
1da177e4 LT |
1299 | return file; |
1300 | ||
b45b5bd6 DG |
1301 | out_inode: |
1302 | iput(inode); | |
1da177e4 | 1303 | out_dentry: |
2c48b9c4 | 1304 | path_put(&path); |
1da177e4 | 1305 | out_shm_unlock: |
353d5c30 HD |
1306 | if (*user) { |
1307 | user_shm_unlock(size, *user); | |
1308 | *user = NULL; | |
1309 | } | |
39b65252 | 1310 | return file; |
1da177e4 LT |
1311 | } |
1312 | ||
1313 | static int __init init_hugetlbfs_fs(void) | |
1314 | { | |
42d7395f | 1315 | struct hstate *h; |
1da177e4 | 1316 | int error; |
42d7395f | 1317 | int i; |
1da177e4 | 1318 | |
457c1b27 | 1319 | if (!hugepages_supported()) { |
9b857d26 | 1320 | pr_info("disabling because there are no supported hugepage sizes\n"); |
457c1b27 NA |
1321 | return -ENOTSUPP; |
1322 | } | |
1323 | ||
d1d5e05f | 1324 | error = -ENOMEM; |
1da177e4 LT |
1325 | hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache", |
1326 | sizeof(struct hugetlbfs_inode_info), | |
5d097056 | 1327 | 0, SLAB_ACCOUNT, init_once); |
1da177e4 | 1328 | if (hugetlbfs_inode_cachep == NULL) |
e0bf68dd | 1329 | goto out2; |
1da177e4 LT |
1330 | |
1331 | error = register_filesystem(&hugetlbfs_fs_type); | |
1332 | if (error) | |
1333 | goto out; | |
1334 | ||
42d7395f AK |
1335 | i = 0; |
1336 | for_each_hstate(h) { | |
1337 | char buf[50]; | |
1338 | unsigned ps_kb = 1U << (h->order + PAGE_SHIFT - 10); | |
1da177e4 | 1339 | |
42d7395f AK |
1340 | snprintf(buf, sizeof(buf), "pagesize=%uK", ps_kb); |
1341 | hugetlbfs_vfsmount[i] = kern_mount_data(&hugetlbfs_fs_type, | |
1342 | buf); | |
1da177e4 | 1343 | |
42d7395f | 1344 | if (IS_ERR(hugetlbfs_vfsmount[i])) { |
9b857d26 | 1345 | pr_err("Cannot mount internal hugetlbfs for " |
42d7395f AK |
1346 | "page size %uK", ps_kb); |
1347 | error = PTR_ERR(hugetlbfs_vfsmount[i]); | |
1348 | hugetlbfs_vfsmount[i] = NULL; | |
1349 | } | |
1350 | i++; | |
1351 | } | |
1352 | /* Non default hstates are optional */ | |
1353 | if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount[default_hstate_idx])) | |
1354 | return 0; | |
1da177e4 LT |
1355 | |
1356 | out: | |
d1d5e05f | 1357 | kmem_cache_destroy(hugetlbfs_inode_cachep); |
e0bf68dd | 1358 | out2: |
1da177e4 LT |
1359 | return error; |
1360 | } | |
3e89e1c5 | 1361 | fs_initcall(init_hugetlbfs_fs) |