]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * hugetlbpage-backed filesystem. Based on ramfs. | |
3 | * | |
6d49e352 | 4 | * Nadia Yvette Chambers, 2002 |
1da177e4 LT |
5 | * |
6 | * Copyright (C) 2002 Linus Torvalds. | |
3e89e1c5 | 7 | * License: GPL |
1da177e4 LT |
8 | */ |
9 | ||
9b857d26 AM |
10 | #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt |
11 | ||
1da177e4 LT |
12 | #include <linux/thread_info.h> |
13 | #include <asm/current.h> | |
174cd4b1 | 14 | #include <linux/sched/signal.h> /* remove ASAP */ |
70c3547e | 15 | #include <linux/falloc.h> |
1da177e4 LT |
16 | #include <linux/fs.h> |
17 | #include <linux/mount.h> | |
18 | #include <linux/file.h> | |
e73a75fa | 19 | #include <linux/kernel.h> |
1da177e4 LT |
20 | #include <linux/writeback.h> |
21 | #include <linux/pagemap.h> | |
22 | #include <linux/highmem.h> | |
23 | #include <linux/init.h> | |
24 | #include <linux/string.h> | |
16f7e0fe | 25 | #include <linux/capability.h> |
e73a75fa | 26 | #include <linux/ctype.h> |
1da177e4 LT |
27 | #include <linux/backing-dev.h> |
28 | #include <linux/hugetlb.h> | |
29 | #include <linux/pagevec.h> | |
32021982 | 30 | #include <linux/fs_parser.h> |
036e0856 | 31 | #include <linux/mman.h> |
1da177e4 LT |
32 | #include <linux/slab.h> |
33 | #include <linux/dnotify.h> | |
34 | #include <linux/statfs.h> | |
35 | #include <linux/security.h> | |
1fd7317d | 36 | #include <linux/magic.h> |
290408d4 | 37 | #include <linux/migrate.h> |
34d0640e | 38 | #include <linux/uio.h> |
1da177e4 | 39 | |
7c0f6ba6 | 40 | #include <linux/uaccess.h> |
88590253 | 41 | #include <linux/sched/mm.h> |
1da177e4 | 42 | |
ee9b6d61 | 43 | static const struct super_operations hugetlbfs_ops; |
f5e54d6e | 44 | static const struct address_space_operations hugetlbfs_aops; |
4b6f5d20 | 45 | const struct file_operations hugetlbfs_file_operations; |
92e1d5be AV |
46 | static const struct inode_operations hugetlbfs_dir_inode_operations; |
47 | static const struct inode_operations hugetlbfs_inode_operations; | |
1da177e4 | 48 | |
32021982 DH |
49 | enum hugetlbfs_size_type { NO_SIZE, SIZE_STD, SIZE_PERCENT }; |
50 | ||
51 | struct hugetlbfs_fs_context { | |
4a25220d | 52 | struct hstate *hstate; |
32021982 DH |
53 | unsigned long long max_size_opt; |
54 | unsigned long long min_size_opt; | |
4a25220d DH |
55 | long max_hpages; |
56 | long nr_inodes; | |
57 | long min_hpages; | |
32021982 DH |
58 | enum hugetlbfs_size_type max_val_type; |
59 | enum hugetlbfs_size_type min_val_type; | |
4a25220d DH |
60 | kuid_t uid; |
61 | kgid_t gid; | |
62 | umode_t mode; | |
a1d776ee DG |
63 | }; |
64 | ||
1da177e4 LT |
65 | int sysctl_hugetlb_shm_group; |
66 | ||
32021982 DH |
67 | enum hugetlb_param { |
68 | Opt_gid, | |
69 | Opt_min_size, | |
70 | Opt_mode, | |
71 | Opt_nr_inodes, | |
72 | Opt_pagesize, | |
73 | Opt_size, | |
74 | Opt_uid, | |
e73a75fa RD |
75 | }; |
76 | ||
d7167b14 | 77 | static const struct fs_parameter_spec hugetlb_fs_parameters[] = { |
32021982 DH |
78 | fsparam_u32 ("gid", Opt_gid), |
79 | fsparam_string("min_size", Opt_min_size), | |
80 | fsparam_u32 ("mode", Opt_mode), | |
81 | fsparam_string("nr_inodes", Opt_nr_inodes), | |
82 | fsparam_string("pagesize", Opt_pagesize), | |
83 | fsparam_string("size", Opt_size), | |
84 | fsparam_u32 ("uid", Opt_uid), | |
85 | {} | |
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 | ||
63489f8e MK |
121 | /* |
122 | * Mask used when checking the page offset value passed in via system | |
123 | * calls. This value will be converted to a loff_t which is signed. | |
124 | * Therefore, we want to check the upper PAGE_SHIFT + 1 bits of the | |
125 | * value. The extra bit (- 1 in the shift value) is to take the sign | |
126 | * bit into account. | |
127 | */ | |
128 | #define PGOFF_LOFFT_MAX \ | |
129 | (((1UL << (PAGE_SHIFT + 1)) - 1) << (BITS_PER_LONG - (PAGE_SHIFT + 1))) | |
130 | ||
1da177e4 LT |
131 | static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma) |
132 | { | |
496ad9aa | 133 | struct inode *inode = file_inode(file); |
1da177e4 LT |
134 | loff_t len, vma_len; |
135 | int ret; | |
a5516438 | 136 | struct hstate *h = hstate_file(file); |
1da177e4 | 137 | |
68589bc3 | 138 | /* |
dec4ad86 DG |
139 | * vma address alignment (but not the pgoff alignment) has |
140 | * already been checked by prepare_hugepage_range. If you add | |
141 | * any error returns here, do so after setting VM_HUGETLB, so | |
142 | * is_vm_hugetlb_page tests below unmap_region go the right | |
143 | * way when do_mmap_pgoff unwinds (may be important on powerpc | |
144 | * and ia64). | |
68589bc3 | 145 | */ |
a2fce914 | 146 | vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND; |
68589bc3 | 147 | vma->vm_ops = &hugetlb_vm_ops; |
1da177e4 | 148 | |
045c7a3f | 149 | /* |
63489f8e | 150 | * page based offset in vm_pgoff could be sufficiently large to |
5df63c2a MK |
151 | * overflow a loff_t when converted to byte offset. This can |
152 | * only happen on architectures where sizeof(loff_t) == | |
153 | * sizeof(unsigned long). So, only check in those instances. | |
045c7a3f | 154 | */ |
5df63c2a MK |
155 | if (sizeof(unsigned long) == sizeof(loff_t)) { |
156 | if (vma->vm_pgoff & PGOFF_LOFFT_MAX) | |
157 | return -EINVAL; | |
158 | } | |
045c7a3f | 159 | |
63489f8e | 160 | /* must be huge page aligned */ |
2b37c35e | 161 | if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT)) |
dec4ad86 DG |
162 | return -EINVAL; |
163 | ||
1da177e4 | 164 | vma_len = (loff_t)(vma->vm_end - vma->vm_start); |
045c7a3f MK |
165 | len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); |
166 | /* check for overflow */ | |
167 | if (len < vma_len) | |
168 | return -EINVAL; | |
1da177e4 | 169 | |
5955102c | 170 | inode_lock(inode); |
1da177e4 | 171 | file_accessed(file); |
1da177e4 LT |
172 | |
173 | ret = -ENOMEM; | |
a1e78772 | 174 | if (hugetlb_reserve_pages(inode, |
a5516438 | 175 | vma->vm_pgoff >> huge_page_order(h), |
5a6fe125 MG |
176 | len >> huge_page_shift(h), vma, |
177 | vma->vm_flags)) | |
a43a8c39 | 178 | goto out; |
b45b5bd6 | 179 | |
4c887265 | 180 | ret = 0; |
b6174df5 | 181 | if (vma->vm_flags & VM_WRITE && inode->i_size < len) |
045c7a3f | 182 | i_size_write(inode, len); |
1da177e4 | 183 | out: |
5955102c | 184 | inode_unlock(inode); |
1da177e4 LT |
185 | |
186 | return ret; | |
187 | } | |
188 | ||
189 | /* | |
3e4e28c5 | 190 | * Called under mmap_write_lock(mm). |
1da177e4 LT |
191 | */ |
192 | ||
d2ba27e8 | 193 | #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA |
88590253 SH |
194 | static unsigned long |
195 | hugetlb_get_unmapped_area_bottomup(struct file *file, unsigned long addr, | |
196 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
197 | { | |
198 | struct hstate *h = hstate_file(file); | |
199 | struct vm_unmapped_area_info info; | |
200 | ||
201 | info.flags = 0; | |
202 | info.length = len; | |
203 | info.low_limit = current->mm->mmap_base; | |
204 | info.high_limit = TASK_SIZE; | |
205 | info.align_mask = PAGE_MASK & ~huge_page_mask(h); | |
206 | info.align_offset = 0; | |
207 | return vm_unmapped_area(&info); | |
208 | } | |
209 | ||
210 | static unsigned long | |
211 | hugetlb_get_unmapped_area_topdown(struct file *file, unsigned long addr, | |
212 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
213 | { | |
214 | struct hstate *h = hstate_file(file); | |
215 | struct vm_unmapped_area_info info; | |
216 | ||
217 | info.flags = VM_UNMAPPED_AREA_TOPDOWN; | |
218 | info.length = len; | |
219 | info.low_limit = max(PAGE_SIZE, mmap_min_addr); | |
220 | info.high_limit = current->mm->mmap_base; | |
221 | info.align_mask = PAGE_MASK & ~huge_page_mask(h); | |
222 | info.align_offset = 0; | |
223 | addr = vm_unmapped_area(&info); | |
224 | ||
225 | /* | |
226 | * A failed mmap() very likely causes application failure, | |
227 | * so fall back to the bottom-up function here. This scenario | |
228 | * can happen with large stack limits and large mmap() | |
229 | * allocations. | |
230 | */ | |
231 | if (unlikely(offset_in_page(addr))) { | |
232 | VM_BUG_ON(addr != -ENOMEM); | |
233 | info.flags = 0; | |
234 | info.low_limit = current->mm->mmap_base; | |
235 | info.high_limit = TASK_SIZE; | |
236 | addr = vm_unmapped_area(&info); | |
237 | } | |
238 | ||
239 | return addr; | |
240 | } | |
241 | ||
1da177e4 LT |
242 | static unsigned long |
243 | hugetlb_get_unmapped_area(struct file *file, unsigned long addr, | |
244 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
245 | { | |
246 | struct mm_struct *mm = current->mm; | |
247 | struct vm_area_struct *vma; | |
a5516438 | 248 | struct hstate *h = hstate_file(file); |
1da177e4 | 249 | |
a5516438 | 250 | if (len & ~huge_page_mask(h)) |
1da177e4 LT |
251 | return -EINVAL; |
252 | if (len > TASK_SIZE) | |
253 | return -ENOMEM; | |
254 | ||
036e0856 | 255 | if (flags & MAP_FIXED) { |
a5516438 | 256 | if (prepare_hugepage_range(file, addr, len)) |
036e0856 BH |
257 | return -EINVAL; |
258 | return addr; | |
259 | } | |
260 | ||
1da177e4 | 261 | if (addr) { |
a5516438 | 262 | addr = ALIGN(addr, huge_page_size(h)); |
1da177e4 LT |
263 | vma = find_vma(mm, addr); |
264 | if (TASK_SIZE - len >= addr && | |
1be7107f | 265 | (!vma || addr + len <= vm_start_gap(vma))) |
1da177e4 LT |
266 | return addr; |
267 | } | |
268 | ||
88590253 SH |
269 | /* |
270 | * Use mm->get_unmapped_area value as a hint to use topdown routine. | |
271 | * If architectures have special needs, they should define their own | |
272 | * version of hugetlb_get_unmapped_area. | |
273 | */ | |
274 | if (mm->get_unmapped_area == arch_get_unmapped_area_topdown) | |
275 | return hugetlb_get_unmapped_area_topdown(file, addr, len, | |
276 | pgoff, flags); | |
277 | return hugetlb_get_unmapped_area_bottomup(file, addr, len, | |
278 | pgoff, flags); | |
1da177e4 LT |
279 | } |
280 | #endif | |
281 | ||
34d0640e | 282 | static size_t |
e63e1e5a | 283 | hugetlbfs_read_actor(struct page *page, unsigned long offset, |
34d0640e | 284 | struct iov_iter *to, unsigned long size) |
e63e1e5a | 285 | { |
34d0640e | 286 | size_t copied = 0; |
e63e1e5a BP |
287 | int i, chunksize; |
288 | ||
e63e1e5a | 289 | /* Find which 4k chunk and offset with in that chunk */ |
09cbfeaf KS |
290 | i = offset >> PAGE_SHIFT; |
291 | offset = offset & ~PAGE_MASK; | |
e63e1e5a BP |
292 | |
293 | while (size) { | |
34d0640e | 294 | size_t n; |
09cbfeaf | 295 | chunksize = PAGE_SIZE; |
e63e1e5a BP |
296 | if (offset) |
297 | chunksize -= offset; | |
298 | if (chunksize > size) | |
299 | chunksize = size; | |
34d0640e AV |
300 | n = copy_page_to_iter(&page[i], offset, chunksize, to); |
301 | copied += n; | |
302 | if (n != chunksize) | |
303 | return copied; | |
e63e1e5a BP |
304 | offset = 0; |
305 | size -= chunksize; | |
e63e1e5a BP |
306 | i++; |
307 | } | |
34d0640e | 308 | return copied; |
e63e1e5a BP |
309 | } |
310 | ||
311 | /* | |
312 | * Support for read() - Find the page attached to f_mapping and copy out the | |
313 | * data. Its *very* similar to do_generic_mapping_read(), we can't use that | |
ea1754a0 | 314 | * since it has PAGE_SIZE assumptions. |
e63e1e5a | 315 | */ |
34d0640e | 316 | static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to) |
e63e1e5a | 317 | { |
34d0640e AV |
318 | struct file *file = iocb->ki_filp; |
319 | struct hstate *h = hstate_file(file); | |
320 | struct address_space *mapping = file->f_mapping; | |
e63e1e5a | 321 | struct inode *inode = mapping->host; |
34d0640e AV |
322 | unsigned long index = iocb->ki_pos >> huge_page_shift(h); |
323 | unsigned long offset = iocb->ki_pos & ~huge_page_mask(h); | |
e63e1e5a BP |
324 | unsigned long end_index; |
325 | loff_t isize; | |
326 | ssize_t retval = 0; | |
327 | ||
34d0640e | 328 | while (iov_iter_count(to)) { |
e63e1e5a | 329 | struct page *page; |
34d0640e | 330 | size_t nr, copied; |
e63e1e5a BP |
331 | |
332 | /* nr is the maximum number of bytes to copy from this page */ | |
a5516438 | 333 | nr = huge_page_size(h); |
a05b0855 AK |
334 | isize = i_size_read(inode); |
335 | if (!isize) | |
34d0640e | 336 | break; |
a05b0855 | 337 | end_index = (isize - 1) >> huge_page_shift(h); |
34d0640e AV |
338 | if (index > end_index) |
339 | break; | |
340 | if (index == end_index) { | |
a5516438 | 341 | nr = ((isize - 1) & ~huge_page_mask(h)) + 1; |
a05b0855 | 342 | if (nr <= offset) |
34d0640e | 343 | break; |
e63e1e5a BP |
344 | } |
345 | nr = nr - offset; | |
346 | ||
347 | /* Find the page */ | |
a05b0855 | 348 | page = find_lock_page(mapping, index); |
e63e1e5a BP |
349 | if (unlikely(page == NULL)) { |
350 | /* | |
351 | * We have a HOLE, zero out the user-buffer for the | |
352 | * length of the hole or request. | |
353 | */ | |
34d0640e | 354 | copied = iov_iter_zero(nr, to); |
e63e1e5a | 355 | } else { |
a05b0855 AK |
356 | unlock_page(page); |
357 | ||
e63e1e5a BP |
358 | /* |
359 | * We have the page, copy it to user space buffer. | |
360 | */ | |
34d0640e | 361 | copied = hugetlbfs_read_actor(page, offset, to, nr); |
09cbfeaf | 362 | put_page(page); |
e63e1e5a | 363 | } |
34d0640e AV |
364 | offset += copied; |
365 | retval += copied; | |
366 | if (copied != nr && iov_iter_count(to)) { | |
367 | if (!retval) | |
368 | retval = -EFAULT; | |
369 | break; | |
e63e1e5a | 370 | } |
a5516438 AK |
371 | index += offset >> huge_page_shift(h); |
372 | offset &= ~huge_page_mask(h); | |
e63e1e5a | 373 | } |
34d0640e | 374 | iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset; |
e63e1e5a BP |
375 | return retval; |
376 | } | |
377 | ||
800d15a5 NP |
378 | static int hugetlbfs_write_begin(struct file *file, |
379 | struct address_space *mapping, | |
380 | loff_t pos, unsigned len, unsigned flags, | |
381 | struct page **pagep, void **fsdata) | |
1da177e4 LT |
382 | { |
383 | return -EINVAL; | |
384 | } | |
385 | ||
800d15a5 NP |
386 | static int hugetlbfs_write_end(struct file *file, struct address_space *mapping, |
387 | loff_t pos, unsigned len, unsigned copied, | |
388 | struct page *page, void *fsdata) | |
1da177e4 | 389 | { |
800d15a5 | 390 | BUG(); |
1da177e4 LT |
391 | return -EINVAL; |
392 | } | |
393 | ||
b5cec28d | 394 | static void remove_huge_page(struct page *page) |
1da177e4 | 395 | { |
b9ea2515 | 396 | ClearPageDirty(page); |
1da177e4 | 397 | ClearPageUptodate(page); |
bd65cb86 | 398 | delete_from_page_cache(page); |
1da177e4 LT |
399 | } |
400 | ||
4aae8d1c | 401 | static void |
f808c13f | 402 | hugetlb_vmdelete_list(struct rb_root_cached *root, pgoff_t start, pgoff_t end) |
4aae8d1c MK |
403 | { |
404 | struct vm_area_struct *vma; | |
405 | ||
406 | /* | |
407 | * end == 0 indicates that the entire range after | |
408 | * start should be unmapped. | |
409 | */ | |
410 | vma_interval_tree_foreach(vma, root, start, end ? end : ULONG_MAX) { | |
411 | unsigned long v_offset; | |
412 | unsigned long v_end; | |
413 | ||
414 | /* | |
415 | * Can the expression below overflow on 32-bit arches? | |
416 | * No, because the interval tree returns us only those vmas | |
417 | * which overlap the truncated area starting at pgoff, | |
418 | * and no vma on a 32-bit arch can span beyond the 4GB. | |
419 | */ | |
420 | if (vma->vm_pgoff < start) | |
421 | v_offset = (start - vma->vm_pgoff) << PAGE_SHIFT; | |
422 | else | |
423 | v_offset = 0; | |
424 | ||
425 | if (!end) | |
426 | v_end = vma->vm_end; | |
427 | else { | |
428 | v_end = ((end - vma->vm_pgoff) << PAGE_SHIFT) | |
429 | + vma->vm_start; | |
430 | if (v_end > vma->vm_end) | |
431 | v_end = vma->vm_end; | |
432 | } | |
433 | ||
434 | unmap_hugepage_range(vma, vma->vm_start + v_offset, v_end, | |
435 | NULL); | |
436 | } | |
437 | } | |
b5cec28d MK |
438 | |
439 | /* | |
440 | * remove_inode_hugepages handles two distinct cases: truncation and hole | |
441 | * punch. There are subtle differences in operation for each case. | |
4aae8d1c | 442 | * |
b5cec28d MK |
443 | * truncation is indicated by end of range being LLONG_MAX |
444 | * In this case, we first scan the range and release found pages. | |
445 | * After releasing pages, hugetlb_unreserve_pages cleans up region/reserv | |
e7c58097 | 446 | * maps and global counts. Page faults can not race with truncation |
87bf91d3 MK |
447 | * in this routine. hugetlb_no_page() holds i_mmap_rwsem and prevents |
448 | * page faults in the truncated range by checking i_size. i_size is | |
449 | * modified while holding i_mmap_rwsem. | |
b5cec28d MK |
450 | * hole punch is indicated if end is not LLONG_MAX |
451 | * In the hole punch case we scan the range and release found pages. | |
452 | * Only when releasing a page is the associated region/reserv map | |
453 | * deleted. The region/reserv map for ranges without associated | |
e7c58097 MK |
454 | * pages are not modified. Page faults can race with hole punch. |
455 | * This is indicated if we find a mapped page. | |
b5cec28d MK |
456 | * Note: If the passed end of range value is beyond the end of file, but |
457 | * not LLONG_MAX this routine still performs a hole punch operation. | |
458 | */ | |
459 | static void remove_inode_hugepages(struct inode *inode, loff_t lstart, | |
460 | loff_t lend) | |
1da177e4 | 461 | { |
a5516438 | 462 | struct hstate *h = hstate_inode(inode); |
b45b5bd6 | 463 | struct address_space *mapping = &inode->i_data; |
a5516438 | 464 | const pgoff_t start = lstart >> huge_page_shift(h); |
b5cec28d MK |
465 | const pgoff_t end = lend >> huge_page_shift(h); |
466 | struct vm_area_struct pseudo_vma; | |
1da177e4 | 467 | struct pagevec pvec; |
d72dc8a2 | 468 | pgoff_t next, index; |
a43a8c39 | 469 | int i, freed = 0; |
b5cec28d | 470 | bool truncate_op = (lend == LLONG_MAX); |
1da177e4 | 471 | |
2c4541e2 | 472 | vma_init(&pseudo_vma, current->mm); |
b5cec28d | 473 | pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); |
86679820 | 474 | pagevec_init(&pvec); |
1da177e4 | 475 | next = start; |
b5cec28d | 476 | while (next < end) { |
b5cec28d | 477 | /* |
1817889e | 478 | * When no more pages are found, we are done. |
b5cec28d | 479 | */ |
397162ff | 480 | if (!pagevec_lookup_range(&pvec, mapping, &next, end - 1)) |
1817889e | 481 | break; |
1da177e4 LT |
482 | |
483 | for (i = 0; i < pagevec_count(&pvec); ++i) { | |
484 | struct page *page = pvec.pages[i]; | |
e7c58097 | 485 | u32 hash; |
b5cec28d | 486 | |
d72dc8a2 | 487 | index = page->index; |
188b04a7 | 488 | hash = hugetlb_fault_mutex_hash(mapping, index); |
87bf91d3 MK |
489 | if (!truncate_op) { |
490 | /* | |
491 | * Only need to hold the fault mutex in the | |
492 | * hole punch case. This prevents races with | |
493 | * page faults. Races are not possible in the | |
494 | * case of truncation. | |
495 | */ | |
496 | mutex_lock(&hugetlb_fault_mutex_table[hash]); | |
497 | } | |
e7c58097 | 498 | |
4aae8d1c | 499 | /* |
e7c58097 MK |
500 | * If page is mapped, it was faulted in after being |
501 | * unmapped in caller. Unmap (again) now after taking | |
502 | * the fault mutex. The mutex will prevent faults | |
503 | * until we finish removing the page. | |
504 | * | |
505 | * This race can only happen in the hole punch case. | |
506 | * Getting here in a truncate operation is a bug. | |
4aae8d1c | 507 | */ |
e7c58097 MK |
508 | if (unlikely(page_mapped(page))) { |
509 | BUG_ON(truncate_op); | |
510 | ||
c0d0381a | 511 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); |
e7c58097 | 512 | i_mmap_lock_write(mapping); |
c0d0381a | 513 | mutex_lock(&hugetlb_fault_mutex_table[hash]); |
e7c58097 MK |
514 | hugetlb_vmdelete_list(&mapping->i_mmap, |
515 | index * pages_per_huge_page(h), | |
516 | (index + 1) * pages_per_huge_page(h)); | |
517 | i_mmap_unlock_write(mapping); | |
518 | } | |
4aae8d1c MK |
519 | |
520 | lock_page(page); | |
521 | /* | |
522 | * We must free the huge page and remove from page | |
523 | * cache (remove_huge_page) BEFORE removing the | |
524 | * region/reserve map (hugetlb_unreserve_pages). In | |
525 | * rare out of memory conditions, removal of the | |
72e2936c | 526 | * region/reserve map could fail. Correspondingly, |
527 | * the subpool and global reserve usage count can need | |
528 | * to be adjusted. | |
4aae8d1c | 529 | */ |
72e2936c | 530 | VM_BUG_ON(PagePrivate(page)); |
4aae8d1c MK |
531 | remove_huge_page(page); |
532 | freed++; | |
533 | if (!truncate_op) { | |
534 | if (unlikely(hugetlb_unreserve_pages(inode, | |
d72dc8a2 | 535 | index, index + 1, 1))) |
72e2936c | 536 | hugetlb_fix_reserve_counts(inode); |
b5cec28d MK |
537 | } |
538 | ||
1da177e4 | 539 | unlock_page(page); |
87bf91d3 MK |
540 | if (!truncate_op) |
541 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
1da177e4 LT |
542 | } |
543 | huge_pagevec_release(&pvec); | |
1817889e | 544 | cond_resched(); |
1da177e4 | 545 | } |
b5cec28d MK |
546 | |
547 | if (truncate_op) | |
548 | (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed); | |
1da177e4 LT |
549 | } |
550 | ||
2bbbda30 | 551 | static void hugetlbfs_evict_inode(struct inode *inode) |
1da177e4 | 552 | { |
9119a41e JK |
553 | struct resv_map *resv_map; |
554 | ||
b5cec28d | 555 | remove_inode_hugepages(inode, 0, LLONG_MAX); |
f27a5136 MK |
556 | |
557 | /* | |
558 | * Get the resv_map from the address space embedded in the inode. | |
559 | * This is the address space which points to any resv_map allocated | |
560 | * at inode creation time. If this is a device special inode, | |
561 | * i_mapping may not point to the original address space. | |
562 | */ | |
563 | resv_map = (struct resv_map *)(&inode->i_data)->private_data; | |
564 | /* Only regular and link inodes have associated reserve maps */ | |
9119a41e JK |
565 | if (resv_map) |
566 | resv_map_release(&resv_map->refs); | |
dbd5768f | 567 | clear_inode(inode); |
149f4211 CH |
568 | } |
569 | ||
1da177e4 LT |
570 | static int hugetlb_vmtruncate(struct inode *inode, loff_t offset) |
571 | { | |
856fc295 | 572 | pgoff_t pgoff; |
1da177e4 | 573 | struct address_space *mapping = inode->i_mapping; |
a5516438 | 574 | struct hstate *h = hstate_inode(inode); |
1da177e4 | 575 | |
a5516438 | 576 | BUG_ON(offset & ~huge_page_mask(h)); |
856fc295 | 577 | pgoff = offset >> PAGE_SHIFT; |
1da177e4 | 578 | |
83cde9e8 | 579 | i_mmap_lock_write(mapping); |
87bf91d3 | 580 | i_size_write(inode, offset); |
f808c13f | 581 | if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)) |
1bfad99a | 582 | hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0); |
c86aa7bb | 583 | i_mmap_unlock_write(mapping); |
e7c58097 | 584 | remove_inode_hugepages(inode, offset, LLONG_MAX); |
1da177e4 LT |
585 | return 0; |
586 | } | |
587 | ||
70c3547e MK |
588 | static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len) |
589 | { | |
590 | struct hstate *h = hstate_inode(inode); | |
591 | loff_t hpage_size = huge_page_size(h); | |
592 | loff_t hole_start, hole_end; | |
593 | ||
594 | /* | |
595 | * For hole punch round up the beginning offset of the hole and | |
596 | * round down the end. | |
597 | */ | |
598 | hole_start = round_up(offset, hpage_size); | |
599 | hole_end = round_down(offset + len, hpage_size); | |
600 | ||
601 | if (hole_end > hole_start) { | |
602 | struct address_space *mapping = inode->i_mapping; | |
ff62a342 | 603 | struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); |
70c3547e | 604 | |
5955102c | 605 | inode_lock(inode); |
ff62a342 MAL |
606 | |
607 | /* protected by i_mutex */ | |
ab3948f5 | 608 | if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) { |
ff62a342 MAL |
609 | inode_unlock(inode); |
610 | return -EPERM; | |
611 | } | |
612 | ||
70c3547e | 613 | i_mmap_lock_write(mapping); |
f808c13f | 614 | if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)) |
70c3547e MK |
615 | hugetlb_vmdelete_list(&mapping->i_mmap, |
616 | hole_start >> PAGE_SHIFT, | |
617 | hole_end >> PAGE_SHIFT); | |
c86aa7bb | 618 | i_mmap_unlock_write(mapping); |
e7c58097 | 619 | remove_inode_hugepages(inode, hole_start, hole_end); |
5955102c | 620 | inode_unlock(inode); |
70c3547e MK |
621 | } |
622 | ||
623 | return 0; | |
624 | } | |
625 | ||
626 | static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset, | |
627 | loff_t len) | |
628 | { | |
629 | struct inode *inode = file_inode(file); | |
ff62a342 | 630 | struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); |
70c3547e MK |
631 | struct address_space *mapping = inode->i_mapping; |
632 | struct hstate *h = hstate_inode(inode); | |
633 | struct vm_area_struct pseudo_vma; | |
634 | struct mm_struct *mm = current->mm; | |
635 | loff_t hpage_size = huge_page_size(h); | |
636 | unsigned long hpage_shift = huge_page_shift(h); | |
637 | pgoff_t start, index, end; | |
638 | int error; | |
639 | u32 hash; | |
640 | ||
641 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
642 | return -EOPNOTSUPP; | |
643 | ||
644 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
645 | return hugetlbfs_punch_hole(inode, offset, len); | |
646 | ||
647 | /* | |
648 | * Default preallocate case. | |
649 | * For this range, start is rounded down and end is rounded up | |
650 | * as well as being converted to page offsets. | |
651 | */ | |
652 | start = offset >> hpage_shift; | |
653 | end = (offset + len + hpage_size - 1) >> hpage_shift; | |
654 | ||
5955102c | 655 | inode_lock(inode); |
70c3547e MK |
656 | |
657 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
658 | error = inode_newsize_ok(inode, offset + len); | |
659 | if (error) | |
660 | goto out; | |
661 | ||
ff62a342 MAL |
662 | if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { |
663 | error = -EPERM; | |
664 | goto out; | |
665 | } | |
666 | ||
70c3547e MK |
667 | /* |
668 | * Initialize a pseudo vma as this is required by the huge page | |
669 | * allocation routines. If NUMA is configured, use page index | |
670 | * as input to create an allocation policy. | |
671 | */ | |
2c4541e2 | 672 | vma_init(&pseudo_vma, mm); |
70c3547e MK |
673 | pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); |
674 | pseudo_vma.vm_file = file; | |
675 | ||
676 | for (index = start; index < end; index++) { | |
677 | /* | |
678 | * This is supposed to be the vaddr where the page is being | |
679 | * faulted in, but we have no vaddr here. | |
680 | */ | |
681 | struct page *page; | |
682 | unsigned long addr; | |
683 | int avoid_reserve = 0; | |
684 | ||
685 | cond_resched(); | |
686 | ||
687 | /* | |
688 | * fallocate(2) manpage permits EINTR; we may have been | |
689 | * interrupted because we are using up too much memory. | |
690 | */ | |
691 | if (signal_pending(current)) { | |
692 | error = -EINTR; | |
693 | break; | |
694 | } | |
695 | ||
696 | /* Set numa allocation policy based on index */ | |
697 | hugetlb_set_vma_policy(&pseudo_vma, inode, index); | |
698 | ||
699 | /* addr is the offset within the file (zero based) */ | |
700 | addr = index * hpage_size; | |
701 | ||
87bf91d3 MK |
702 | /* |
703 | * fault mutex taken here, protects against fault path | |
704 | * and hole punch. inode_lock previously taken protects | |
705 | * against truncation. | |
706 | */ | |
188b04a7 | 707 | hash = hugetlb_fault_mutex_hash(mapping, index); |
70c3547e MK |
708 | mutex_lock(&hugetlb_fault_mutex_table[hash]); |
709 | ||
710 | /* See if already present in mapping to avoid alloc/free */ | |
711 | page = find_get_page(mapping, index); | |
712 | if (page) { | |
713 | put_page(page); | |
714 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
715 | hugetlb_drop_vma_policy(&pseudo_vma); | |
716 | continue; | |
717 | } | |
718 | ||
719 | /* Allocate page and add to page cache */ | |
720 | page = alloc_huge_page(&pseudo_vma, addr, avoid_reserve); | |
721 | hugetlb_drop_vma_policy(&pseudo_vma); | |
722 | if (IS_ERR(page)) { | |
723 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
724 | error = PTR_ERR(page); | |
725 | goto out; | |
726 | } | |
727 | clear_huge_page(page, addr, pages_per_huge_page(h)); | |
728 | __SetPageUptodate(page); | |
729 | error = huge_add_to_page_cache(page, mapping, index); | |
730 | if (unlikely(error)) { | |
731 | put_page(page); | |
732 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
733 | goto out; | |
734 | } | |
735 | ||
736 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
737 | ||
738 | /* | |
70c3547e | 739 | * unlock_page because locked by add_to_page_cache() |
72639e6d | 740 | * page_put due to reference from alloc_huge_page() |
70c3547e | 741 | */ |
70c3547e | 742 | unlock_page(page); |
72639e6d | 743 | put_page(page); |
70c3547e MK |
744 | } |
745 | ||
746 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
747 | i_size_write(inode, offset + len); | |
078cd827 | 748 | inode->i_ctime = current_time(inode); |
70c3547e | 749 | out: |
5955102c | 750 | inode_unlock(inode); |
70c3547e MK |
751 | return error; |
752 | } | |
753 | ||
1da177e4 LT |
754 | static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr) |
755 | { | |
2b0143b5 | 756 | struct inode *inode = d_inode(dentry); |
a5516438 | 757 | struct hstate *h = hstate_inode(inode); |
1da177e4 LT |
758 | int error; |
759 | unsigned int ia_valid = attr->ia_valid; | |
ff62a342 | 760 | struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); |
1da177e4 LT |
761 | |
762 | BUG_ON(!inode); | |
763 | ||
31051c85 | 764 | error = setattr_prepare(dentry, attr); |
1da177e4 | 765 | if (error) |
1025774c | 766 | return error; |
1da177e4 LT |
767 | |
768 | if (ia_valid & ATTR_SIZE) { | |
ff62a342 MAL |
769 | loff_t oldsize = inode->i_size; |
770 | loff_t newsize = attr->ia_size; | |
771 | ||
772 | if (newsize & ~huge_page_mask(h)) | |
1025774c | 773 | return -EINVAL; |
ff62a342 MAL |
774 | /* protected by i_mutex */ |
775 | if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || | |
776 | (newsize > oldsize && (info->seals & F_SEAL_GROW))) | |
777 | return -EPERM; | |
778 | error = hugetlb_vmtruncate(inode, newsize); | |
1da177e4 | 779 | if (error) |
1025774c | 780 | return error; |
1da177e4 | 781 | } |
1025774c CH |
782 | |
783 | setattr_copy(inode, attr); | |
784 | mark_inode_dirty(inode); | |
785 | return 0; | |
1da177e4 LT |
786 | } |
787 | ||
7d54fa64 | 788 | static struct inode *hugetlbfs_get_root(struct super_block *sb, |
32021982 | 789 | struct hugetlbfs_fs_context *ctx) |
1da177e4 LT |
790 | { |
791 | struct inode *inode; | |
1da177e4 LT |
792 | |
793 | inode = new_inode(sb); | |
794 | if (inode) { | |
85fe4025 | 795 | inode->i_ino = get_next_ino(); |
32021982 DH |
796 | inode->i_mode = S_IFDIR | ctx->mode; |
797 | inode->i_uid = ctx->uid; | |
798 | inode->i_gid = ctx->gid; | |
078cd827 | 799 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
7d54fa64 AV |
800 | inode->i_op = &hugetlbfs_dir_inode_operations; |
801 | inode->i_fop = &simple_dir_operations; | |
802 | /* directory inodes start off with i_nlink == 2 (for "." entry) */ | |
803 | inc_nlink(inode); | |
65ed7601 | 804 | lockdep_annotate_inode_mutex_key(inode); |
7d54fa64 AV |
805 | } |
806 | return inode; | |
807 | } | |
808 | ||
b610ded7 | 809 | /* |
c8c06efa | 810 | * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never |
b610ded7 | 811 | * be taken from reclaim -- unlike regular filesystems. This needs an |
88f306b6 | 812 | * annotation because huge_pmd_share() does an allocation under hugetlb's |
c8c06efa | 813 | * i_mmap_rwsem. |
b610ded7 | 814 | */ |
c8c06efa | 815 | static struct lock_class_key hugetlbfs_i_mmap_rwsem_key; |
b610ded7 | 816 | |
7d54fa64 AV |
817 | static struct inode *hugetlbfs_get_inode(struct super_block *sb, |
818 | struct inode *dir, | |
18df2252 | 819 | umode_t mode, dev_t dev) |
7d54fa64 AV |
820 | { |
821 | struct inode *inode; | |
58b6e5e8 | 822 | struct resv_map *resv_map = NULL; |
9119a41e | 823 | |
58b6e5e8 MK |
824 | /* |
825 | * Reserve maps are only needed for inodes that can have associated | |
826 | * page allocations. | |
827 | */ | |
828 | if (S_ISREG(mode) || S_ISLNK(mode)) { | |
829 | resv_map = resv_map_alloc(); | |
830 | if (!resv_map) | |
831 | return NULL; | |
832 | } | |
7d54fa64 AV |
833 | |
834 | inode = new_inode(sb); | |
835 | if (inode) { | |
ff62a342 MAL |
836 | struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); |
837 | ||
7d54fa64 AV |
838 | inode->i_ino = get_next_ino(); |
839 | inode_init_owner(inode, dir, mode); | |
c8c06efa DB |
840 | lockdep_set_class(&inode->i_mapping->i_mmap_rwsem, |
841 | &hugetlbfs_i_mmap_rwsem_key); | |
1da177e4 | 842 | inode->i_mapping->a_ops = &hugetlbfs_aops; |
078cd827 | 843 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
9119a41e | 844 | inode->i_mapping->private_data = resv_map; |
ff62a342 | 845 | info->seals = F_SEAL_SEAL; |
1da177e4 LT |
846 | switch (mode & S_IFMT) { |
847 | default: | |
848 | init_special_inode(inode, mode, dev); | |
849 | break; | |
850 | case S_IFREG: | |
851 | inode->i_op = &hugetlbfs_inode_operations; | |
852 | inode->i_fop = &hugetlbfs_file_operations; | |
853 | break; | |
854 | case S_IFDIR: | |
855 | inode->i_op = &hugetlbfs_dir_inode_operations; | |
856 | inode->i_fop = &simple_dir_operations; | |
857 | ||
858 | /* directory inodes start off with i_nlink == 2 (for "." entry) */ | |
d8c76e6f | 859 | inc_nlink(inode); |
1da177e4 LT |
860 | break; |
861 | case S_IFLNK: | |
862 | inode->i_op = &page_symlink_inode_operations; | |
21fc61c7 | 863 | inode_nohighmem(inode); |
1da177e4 LT |
864 | break; |
865 | } | |
e096d0c7 | 866 | lockdep_annotate_inode_mutex_key(inode); |
58b6e5e8 MK |
867 | } else { |
868 | if (resv_map) | |
869 | kref_put(&resv_map->refs, resv_map_release); | |
870 | } | |
9119a41e | 871 | |
1da177e4 LT |
872 | return inode; |
873 | } | |
874 | ||
875 | /* | |
876 | * File creation. Allocate an inode, and we're done.. | |
877 | */ | |
1ab5b82f PS |
878 | static int do_hugetlbfs_mknod(struct inode *dir, |
879 | struct dentry *dentry, | |
880 | umode_t mode, | |
881 | dev_t dev, | |
882 | bool tmpfile) | |
1da177e4 LT |
883 | { |
884 | struct inode *inode; | |
885 | int error = -ENOSPC; | |
7d54fa64 AV |
886 | |
887 | inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev); | |
1da177e4 | 888 | if (inode) { |
078cd827 | 889 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1ab5b82f PS |
890 | if (tmpfile) { |
891 | d_tmpfile(dentry, inode); | |
892 | } else { | |
893 | d_instantiate(dentry, inode); | |
894 | dget(dentry);/* Extra count - pin the dentry in core */ | |
895 | } | |
1da177e4 LT |
896 | error = 0; |
897 | } | |
898 | return error; | |
899 | } | |
900 | ||
1ab5b82f PS |
901 | static int hugetlbfs_mknod(struct inode *dir, |
902 | struct dentry *dentry, umode_t mode, dev_t dev) | |
903 | { | |
904 | return do_hugetlbfs_mknod(dir, dentry, mode, dev, false); | |
905 | } | |
906 | ||
18bb1db3 | 907 | static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
908 | { |
909 | int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0); | |
910 | if (!retval) | |
d8c76e6f | 911 | inc_nlink(dir); |
1da177e4 LT |
912 | return retval; |
913 | } | |
914 | ||
ebfc3b49 | 915 | static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl) |
1da177e4 LT |
916 | { |
917 | return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0); | |
918 | } | |
919 | ||
1ab5b82f PS |
920 | static int hugetlbfs_tmpfile(struct inode *dir, |
921 | struct dentry *dentry, umode_t mode) | |
922 | { | |
923 | return do_hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0, true); | |
924 | } | |
925 | ||
1da177e4 LT |
926 | static int hugetlbfs_symlink(struct inode *dir, |
927 | struct dentry *dentry, const char *symname) | |
928 | { | |
929 | struct inode *inode; | |
930 | int error = -ENOSPC; | |
1da177e4 | 931 | |
7d54fa64 | 932 | inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0); |
1da177e4 LT |
933 | if (inode) { |
934 | int l = strlen(symname)+1; | |
935 | error = page_symlink(inode, symname, l); | |
936 | if (!error) { | |
937 | d_instantiate(dentry, inode); | |
938 | dget(dentry); | |
939 | } else | |
940 | iput(inode); | |
941 | } | |
078cd827 | 942 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1da177e4 LT |
943 | |
944 | return error; | |
945 | } | |
946 | ||
947 | /* | |
6649a386 | 948 | * mark the head page dirty |
1da177e4 LT |
949 | */ |
950 | static int hugetlbfs_set_page_dirty(struct page *page) | |
951 | { | |
d85f3385 | 952 | struct page *head = compound_head(page); |
6649a386 KC |
953 | |
954 | SetPageDirty(head); | |
1da177e4 LT |
955 | return 0; |
956 | } | |
957 | ||
290408d4 | 958 | static int hugetlbfs_migrate_page(struct address_space *mapping, |
b969c4ab | 959 | struct page *newpage, struct page *page, |
a6bc32b8 | 960 | enum migrate_mode mode) |
290408d4 NH |
961 | { |
962 | int rc; | |
963 | ||
964 | rc = migrate_huge_page_move_mapping(mapping, newpage, page); | |
78bd5209 | 965 | if (rc != MIGRATEPAGE_SUCCESS) |
290408d4 | 966 | return rc; |
cb6acd01 MK |
967 | |
968 | /* | |
969 | * page_private is subpool pointer in hugetlb pages. Transfer to | |
970 | * new page. PagePrivate is not associated with page_private for | |
971 | * hugetlb pages and can not be set here as only page_huge_active | |
972 | * pages can be migrated. | |
973 | */ | |
974 | if (page_private(page)) { | |
975 | set_page_private(newpage, page_private(page)); | |
976 | set_page_private(page, 0); | |
977 | } | |
978 | ||
2916ecc0 JG |
979 | if (mode != MIGRATE_SYNC_NO_COPY) |
980 | migrate_page_copy(newpage, page); | |
981 | else | |
982 | migrate_page_states(newpage, page); | |
290408d4 | 983 | |
78bd5209 | 984 | return MIGRATEPAGE_SUCCESS; |
290408d4 NH |
985 | } |
986 | ||
78bb9203 NH |
987 | static int hugetlbfs_error_remove_page(struct address_space *mapping, |
988 | struct page *page) | |
989 | { | |
990 | struct inode *inode = mapping->host; | |
ab615a5b | 991 | pgoff_t index = page->index; |
78bb9203 NH |
992 | |
993 | remove_huge_page(page); | |
ab615a5b MK |
994 | if (unlikely(hugetlb_unreserve_pages(inode, index, index + 1, 1))) |
995 | hugetlb_fix_reserve_counts(inode); | |
996 | ||
78bb9203 NH |
997 | return 0; |
998 | } | |
999 | ||
4a25220d DH |
1000 | /* |
1001 | * Display the mount options in /proc/mounts. | |
1002 | */ | |
1003 | static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root) | |
1004 | { | |
1005 | struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(root->d_sb); | |
1006 | struct hugepage_subpool *spool = sbinfo->spool; | |
1007 | unsigned long hpage_size = huge_page_size(sbinfo->hstate); | |
1008 | unsigned hpage_shift = huge_page_shift(sbinfo->hstate); | |
1009 | char mod; | |
1010 | ||
1011 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) | |
1012 | seq_printf(m, ",uid=%u", | |
1013 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
1014 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
1015 | seq_printf(m, ",gid=%u", | |
1016 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
1017 | if (sbinfo->mode != 0755) | |
1018 | seq_printf(m, ",mode=%o", sbinfo->mode); | |
1019 | if (sbinfo->max_inodes != -1) | |
1020 | seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes); | |
1021 | ||
1022 | hpage_size /= 1024; | |
1023 | mod = 'K'; | |
1024 | if (hpage_size >= 1024) { | |
1025 | hpage_size /= 1024; | |
1026 | mod = 'M'; | |
1027 | } | |
1028 | seq_printf(m, ",pagesize=%lu%c", hpage_size, mod); | |
1029 | if (spool) { | |
1030 | if (spool->max_hpages != -1) | |
1031 | seq_printf(m, ",size=%llu", | |
1032 | (unsigned long long)spool->max_hpages << hpage_shift); | |
1033 | if (spool->min_hpages != -1) | |
1034 | seq_printf(m, ",min_size=%llu", | |
1035 | (unsigned long long)spool->min_hpages << hpage_shift); | |
1036 | } | |
1037 | return 0; | |
1038 | } | |
1039 | ||
726c3342 | 1040 | static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 1041 | { |
726c3342 | 1042 | struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb); |
2b0143b5 | 1043 | struct hstate *h = hstate_inode(d_inode(dentry)); |
1da177e4 LT |
1044 | |
1045 | buf->f_type = HUGETLBFS_MAGIC; | |
a5516438 | 1046 | buf->f_bsize = huge_page_size(h); |
1da177e4 LT |
1047 | if (sbinfo) { |
1048 | spin_lock(&sbinfo->stat_lock); | |
74a8a65c DG |
1049 | /* If no limits set, just report 0 for max/free/used |
1050 | * blocks, like simple_statfs() */ | |
90481622 DG |
1051 | if (sbinfo->spool) { |
1052 | long free_pages; | |
1053 | ||
1054 | spin_lock(&sbinfo->spool->lock); | |
1055 | buf->f_blocks = sbinfo->spool->max_hpages; | |
1056 | free_pages = sbinfo->spool->max_hpages | |
1057 | - sbinfo->spool->used_hpages; | |
1058 | buf->f_bavail = buf->f_bfree = free_pages; | |
1059 | spin_unlock(&sbinfo->spool->lock); | |
74a8a65c DG |
1060 | buf->f_files = sbinfo->max_inodes; |
1061 | buf->f_ffree = sbinfo->free_inodes; | |
1062 | } | |
1da177e4 LT |
1063 | spin_unlock(&sbinfo->stat_lock); |
1064 | } | |
1065 | buf->f_namelen = NAME_MAX; | |
1066 | return 0; | |
1067 | } | |
1068 | ||
1069 | static void hugetlbfs_put_super(struct super_block *sb) | |
1070 | { | |
1071 | struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb); | |
1072 | ||
1073 | if (sbi) { | |
1074 | sb->s_fs_info = NULL; | |
90481622 DG |
1075 | |
1076 | if (sbi->spool) | |
1077 | hugepage_put_subpool(sbi->spool); | |
1078 | ||
1da177e4 LT |
1079 | kfree(sbi); |
1080 | } | |
1081 | } | |
1082 | ||
96527980 CH |
1083 | static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo) |
1084 | { | |
1085 | if (sbinfo->free_inodes >= 0) { | |
1086 | spin_lock(&sbinfo->stat_lock); | |
1087 | if (unlikely(!sbinfo->free_inodes)) { | |
1088 | spin_unlock(&sbinfo->stat_lock); | |
1089 | return 0; | |
1090 | } | |
1091 | sbinfo->free_inodes--; | |
1092 | spin_unlock(&sbinfo->stat_lock); | |
1093 | } | |
1094 | ||
1095 | return 1; | |
1096 | } | |
1097 | ||
1098 | static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo) | |
1099 | { | |
1100 | if (sbinfo->free_inodes >= 0) { | |
1101 | spin_lock(&sbinfo->stat_lock); | |
1102 | sbinfo->free_inodes++; | |
1103 | spin_unlock(&sbinfo->stat_lock); | |
1104 | } | |
1105 | } | |
1106 | ||
1107 | ||
e18b890b | 1108 | static struct kmem_cache *hugetlbfs_inode_cachep; |
1da177e4 LT |
1109 | |
1110 | static struct inode *hugetlbfs_alloc_inode(struct super_block *sb) | |
1111 | { | |
96527980 | 1112 | struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb); |
1da177e4 LT |
1113 | struct hugetlbfs_inode_info *p; |
1114 | ||
96527980 CH |
1115 | if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo))) |
1116 | return NULL; | |
e94b1766 | 1117 | p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL); |
96527980 CH |
1118 | if (unlikely(!p)) { |
1119 | hugetlbfs_inc_free_inodes(sbinfo); | |
1da177e4 | 1120 | return NULL; |
96527980 | 1121 | } |
4742a35d MK |
1122 | |
1123 | /* | |
1124 | * Any time after allocation, hugetlbfs_destroy_inode can be called | |
1125 | * for the inode. mpol_free_shared_policy is unconditionally called | |
1126 | * as part of hugetlbfs_destroy_inode. So, initialize policy here | |
1127 | * in case of a quick call to destroy. | |
1128 | * | |
1129 | * Note that the policy is initialized even if we are creating a | |
1130 | * private inode. This simplifies hugetlbfs_destroy_inode. | |
1131 | */ | |
1132 | mpol_shared_policy_init(&p->policy, NULL); | |
1133 | ||
1da177e4 LT |
1134 | return &p->vfs_inode; |
1135 | } | |
1136 | ||
b62de322 | 1137 | static void hugetlbfs_free_inode(struct inode *inode) |
fa0d7e3d | 1138 | { |
fa0d7e3d NP |
1139 | kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode)); |
1140 | } | |
1141 | ||
1da177e4 LT |
1142 | static void hugetlbfs_destroy_inode(struct inode *inode) |
1143 | { | |
96527980 | 1144 | hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb)); |
1da177e4 | 1145 | mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy); |
1da177e4 LT |
1146 | } |
1147 | ||
f5e54d6e | 1148 | static const struct address_space_operations hugetlbfs_aops = { |
800d15a5 NP |
1149 | .write_begin = hugetlbfs_write_begin, |
1150 | .write_end = hugetlbfs_write_end, | |
1da177e4 | 1151 | .set_page_dirty = hugetlbfs_set_page_dirty, |
290408d4 | 1152 | .migratepage = hugetlbfs_migrate_page, |
78bb9203 | 1153 | .error_remove_page = hugetlbfs_error_remove_page, |
1da177e4 LT |
1154 | }; |
1155 | ||
96527980 | 1156 | |
51cc5068 | 1157 | static void init_once(void *foo) |
96527980 CH |
1158 | { |
1159 | struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo; | |
1160 | ||
a35afb83 | 1161 | inode_init_once(&ei->vfs_inode); |
96527980 CH |
1162 | } |
1163 | ||
4b6f5d20 | 1164 | const struct file_operations hugetlbfs_file_operations = { |
34d0640e | 1165 | .read_iter = hugetlbfs_read_iter, |
1da177e4 | 1166 | .mmap = hugetlbfs_file_mmap, |
1b061d92 | 1167 | .fsync = noop_fsync, |
1da177e4 | 1168 | .get_unmapped_area = hugetlb_get_unmapped_area, |
70c3547e MK |
1169 | .llseek = default_llseek, |
1170 | .fallocate = hugetlbfs_fallocate, | |
1da177e4 LT |
1171 | }; |
1172 | ||
92e1d5be | 1173 | static const struct inode_operations hugetlbfs_dir_inode_operations = { |
1da177e4 LT |
1174 | .create = hugetlbfs_create, |
1175 | .lookup = simple_lookup, | |
1176 | .link = simple_link, | |
1177 | .unlink = simple_unlink, | |
1178 | .symlink = hugetlbfs_symlink, | |
1179 | .mkdir = hugetlbfs_mkdir, | |
1180 | .rmdir = simple_rmdir, | |
1181 | .mknod = hugetlbfs_mknod, | |
1182 | .rename = simple_rename, | |
1183 | .setattr = hugetlbfs_setattr, | |
1ab5b82f | 1184 | .tmpfile = hugetlbfs_tmpfile, |
1da177e4 LT |
1185 | }; |
1186 | ||
92e1d5be | 1187 | static const struct inode_operations hugetlbfs_inode_operations = { |
1da177e4 LT |
1188 | .setattr = hugetlbfs_setattr, |
1189 | }; | |
1190 | ||
ee9b6d61 | 1191 | static const struct super_operations hugetlbfs_ops = { |
1da177e4 | 1192 | .alloc_inode = hugetlbfs_alloc_inode, |
b62de322 | 1193 | .free_inode = hugetlbfs_free_inode, |
1da177e4 | 1194 | .destroy_inode = hugetlbfs_destroy_inode, |
2bbbda30 | 1195 | .evict_inode = hugetlbfs_evict_inode, |
1da177e4 | 1196 | .statfs = hugetlbfs_statfs, |
1da177e4 | 1197 | .put_super = hugetlbfs_put_super, |
4a25220d | 1198 | .show_options = hugetlbfs_show_options, |
1da177e4 LT |
1199 | }; |
1200 | ||
7ca02d0a MK |
1201 | /* |
1202 | * Convert size option passed from command line to number of huge pages | |
1203 | * in the pool specified by hstate. Size option could be in bytes | |
1204 | * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT). | |
1205 | */ | |
4a25220d | 1206 | static long |
7ca02d0a | 1207 | hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt, |
4a25220d | 1208 | enum hugetlbfs_size_type val_type) |
7ca02d0a MK |
1209 | { |
1210 | if (val_type == NO_SIZE) | |
1211 | return -1; | |
1212 | ||
1213 | if (val_type == SIZE_PERCENT) { | |
1214 | size_opt <<= huge_page_shift(h); | |
1215 | size_opt *= h->max_huge_pages; | |
1216 | do_div(size_opt, 100); | |
1217 | } | |
1218 | ||
1219 | size_opt >>= huge_page_shift(h); | |
1220 | return size_opt; | |
1221 | } | |
1222 | ||
32021982 DH |
1223 | /* |
1224 | * Parse one mount parameter. | |
1225 | */ | |
1226 | static int hugetlbfs_parse_param(struct fs_context *fc, struct fs_parameter *param) | |
1da177e4 | 1227 | { |
32021982 DH |
1228 | struct hugetlbfs_fs_context *ctx = fc->fs_private; |
1229 | struct fs_parse_result result; | |
1230 | char *rest; | |
1231 | unsigned long ps; | |
1232 | int opt; | |
1233 | ||
d7167b14 | 1234 | opt = fs_parse(fc, hugetlb_fs_parameters, param, &result); |
32021982 DH |
1235 | if (opt < 0) |
1236 | return opt; | |
1237 | ||
1238 | switch (opt) { | |
1239 | case Opt_uid: | |
1240 | ctx->uid = make_kuid(current_user_ns(), result.uint_32); | |
1241 | if (!uid_valid(ctx->uid)) | |
1242 | goto bad_val; | |
1da177e4 | 1243 | return 0; |
1da177e4 | 1244 | |
32021982 DH |
1245 | case Opt_gid: |
1246 | ctx->gid = make_kgid(current_user_ns(), result.uint_32); | |
1247 | if (!gid_valid(ctx->gid)) | |
1248 | goto bad_val; | |
1249 | return 0; | |
e73a75fa | 1250 | |
32021982 DH |
1251 | case Opt_mode: |
1252 | ctx->mode = result.uint_32 & 01777U; | |
1253 | return 0; | |
e73a75fa | 1254 | |
32021982 DH |
1255 | case Opt_size: |
1256 | /* memparse() will accept a K/M/G without a digit */ | |
1257 | if (!isdigit(param->string[0])) | |
1258 | goto bad_val; | |
1259 | ctx->max_size_opt = memparse(param->string, &rest); | |
1260 | ctx->max_val_type = SIZE_STD; | |
1261 | if (*rest == '%') | |
1262 | ctx->max_val_type = SIZE_PERCENT; | |
1263 | return 0; | |
e73a75fa | 1264 | |
32021982 DH |
1265 | case Opt_nr_inodes: |
1266 | /* memparse() will accept a K/M/G without a digit */ | |
1267 | if (!isdigit(param->string[0])) | |
1268 | goto bad_val; | |
1269 | ctx->nr_inodes = memparse(param->string, &rest); | |
1270 | return 0; | |
e73a75fa | 1271 | |
32021982 DH |
1272 | case Opt_pagesize: |
1273 | ps = memparse(param->string, &rest); | |
1274 | ctx->hstate = size_to_hstate(ps); | |
1275 | if (!ctx->hstate) { | |
1276 | pr_err("Unsupported page size %lu MB\n", ps >> 20); | |
1277 | return -EINVAL; | |
e73a75fa | 1278 | } |
32021982 | 1279 | return 0; |
1da177e4 | 1280 | |
32021982 DH |
1281 | case Opt_min_size: |
1282 | /* memparse() will accept a K/M/G without a digit */ | |
1283 | if (!isdigit(param->string[0])) | |
1284 | goto bad_val; | |
1285 | ctx->min_size_opt = memparse(param->string, &rest); | |
1286 | ctx->min_val_type = SIZE_STD; | |
1287 | if (*rest == '%') | |
1288 | ctx->min_val_type = SIZE_PERCENT; | |
1289 | return 0; | |
e73a75fa | 1290 | |
32021982 DH |
1291 | default: |
1292 | return -EINVAL; | |
1293 | } | |
a137e1cc | 1294 | |
32021982 | 1295 | bad_val: |
b5db30cf | 1296 | return invalfc(fc, "Bad value '%s' for mount option '%s'\n", |
32021982 DH |
1297 | param->string, param->key); |
1298 | } | |
7ca02d0a | 1299 | |
32021982 DH |
1300 | /* |
1301 | * Validate the parsed options. | |
1302 | */ | |
1303 | static int hugetlbfs_validate(struct fs_context *fc) | |
1304 | { | |
1305 | struct hugetlbfs_fs_context *ctx = fc->fs_private; | |
a137e1cc | 1306 | |
7ca02d0a MK |
1307 | /* |
1308 | * Use huge page pool size (in hstate) to convert the size | |
1309 | * options to number of huge pages. If NO_SIZE, -1 is returned. | |
1310 | */ | |
32021982 DH |
1311 | ctx->max_hpages = hugetlbfs_size_to_hpages(ctx->hstate, |
1312 | ctx->max_size_opt, | |
1313 | ctx->max_val_type); | |
1314 | ctx->min_hpages = hugetlbfs_size_to_hpages(ctx->hstate, | |
1315 | ctx->min_size_opt, | |
1316 | ctx->min_val_type); | |
7ca02d0a MK |
1317 | |
1318 | /* | |
1319 | * If max_size was specified, then min_size must be smaller | |
1320 | */ | |
32021982 DH |
1321 | if (ctx->max_val_type > NO_SIZE && |
1322 | ctx->min_hpages > ctx->max_hpages) { | |
1323 | pr_err("Minimum size can not be greater than maximum size\n"); | |
7ca02d0a | 1324 | return -EINVAL; |
a137e1cc AK |
1325 | } |
1326 | ||
1da177e4 LT |
1327 | return 0; |
1328 | } | |
1329 | ||
1330 | static int | |
32021982 | 1331 | hugetlbfs_fill_super(struct super_block *sb, struct fs_context *fc) |
1da177e4 | 1332 | { |
32021982 | 1333 | struct hugetlbfs_fs_context *ctx = fc->fs_private; |
1da177e4 LT |
1334 | struct hugetlbfs_sb_info *sbinfo; |
1335 | ||
1da177e4 LT |
1336 | sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL); |
1337 | if (!sbinfo) | |
1338 | return -ENOMEM; | |
1339 | sb->s_fs_info = sbinfo; | |
1340 | spin_lock_init(&sbinfo->stat_lock); | |
32021982 DH |
1341 | sbinfo->hstate = ctx->hstate; |
1342 | sbinfo->max_inodes = ctx->nr_inodes; | |
1343 | sbinfo->free_inodes = ctx->nr_inodes; | |
1344 | sbinfo->spool = NULL; | |
1345 | sbinfo->uid = ctx->uid; | |
1346 | sbinfo->gid = ctx->gid; | |
1347 | sbinfo->mode = ctx->mode; | |
4a25220d | 1348 | |
7ca02d0a MK |
1349 | /* |
1350 | * Allocate and initialize subpool if maximum or minimum size is | |
1351 | * specified. Any needed reservations (for minimim size) are taken | |
1352 | * taken when the subpool is created. | |
1353 | */ | |
32021982 DH |
1354 | if (ctx->max_hpages != -1 || ctx->min_hpages != -1) { |
1355 | sbinfo->spool = hugepage_new_subpool(ctx->hstate, | |
1356 | ctx->max_hpages, | |
1357 | ctx->min_hpages); | |
90481622 DG |
1358 | if (!sbinfo->spool) |
1359 | goto out_free; | |
1360 | } | |
1da177e4 | 1361 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
32021982 DH |
1362 | sb->s_blocksize = huge_page_size(ctx->hstate); |
1363 | sb->s_blocksize_bits = huge_page_shift(ctx->hstate); | |
1da177e4 LT |
1364 | sb->s_magic = HUGETLBFS_MAGIC; |
1365 | sb->s_op = &hugetlbfs_ops; | |
1366 | sb->s_time_gran = 1; | |
32021982 | 1367 | sb->s_root = d_make_root(hugetlbfs_get_root(sb, ctx)); |
48fde701 | 1368 | if (!sb->s_root) |
1da177e4 | 1369 | goto out_free; |
1da177e4 LT |
1370 | return 0; |
1371 | out_free: | |
6e6870d4 | 1372 | kfree(sbinfo->spool); |
1da177e4 LT |
1373 | kfree(sbinfo); |
1374 | return -ENOMEM; | |
1375 | } | |
1376 | ||
32021982 DH |
1377 | static int hugetlbfs_get_tree(struct fs_context *fc) |
1378 | { | |
1379 | int err = hugetlbfs_validate(fc); | |
1380 | if (err) | |
1381 | return err; | |
2ac295d4 | 1382 | return get_tree_nodev(fc, hugetlbfs_fill_super); |
32021982 DH |
1383 | } |
1384 | ||
1385 | static void hugetlbfs_fs_context_free(struct fs_context *fc) | |
1386 | { | |
1387 | kfree(fc->fs_private); | |
1388 | } | |
1389 | ||
1390 | static const struct fs_context_operations hugetlbfs_fs_context_ops = { | |
1391 | .free = hugetlbfs_fs_context_free, | |
1392 | .parse_param = hugetlbfs_parse_param, | |
1393 | .get_tree = hugetlbfs_get_tree, | |
1394 | }; | |
1395 | ||
1396 | static int hugetlbfs_init_fs_context(struct fs_context *fc) | |
1da177e4 | 1397 | { |
32021982 DH |
1398 | struct hugetlbfs_fs_context *ctx; |
1399 | ||
1400 | ctx = kzalloc(sizeof(struct hugetlbfs_fs_context), GFP_KERNEL); | |
1401 | if (!ctx) | |
1402 | return -ENOMEM; | |
1403 | ||
1404 | ctx->max_hpages = -1; /* No limit on size by default */ | |
1405 | ctx->nr_inodes = -1; /* No limit on number of inodes by default */ | |
1406 | ctx->uid = current_fsuid(); | |
1407 | ctx->gid = current_fsgid(); | |
1408 | ctx->mode = 0755; | |
1409 | ctx->hstate = &default_hstate; | |
1410 | ctx->min_hpages = -1; /* No default minimum size */ | |
1411 | ctx->max_val_type = NO_SIZE; | |
1412 | ctx->min_val_type = NO_SIZE; | |
1413 | fc->fs_private = ctx; | |
1414 | fc->ops = &hugetlbfs_fs_context_ops; | |
1415 | return 0; | |
1da177e4 LT |
1416 | } |
1417 | ||
1418 | static struct file_system_type hugetlbfs_fs_type = { | |
32021982 DH |
1419 | .name = "hugetlbfs", |
1420 | .init_fs_context = hugetlbfs_init_fs_context, | |
d7167b14 | 1421 | .parameters = hugetlb_fs_parameters, |
32021982 | 1422 | .kill_sb = kill_litter_super, |
1da177e4 LT |
1423 | }; |
1424 | ||
42d7395f | 1425 | static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE]; |
1da177e4 | 1426 | |
ef1ff6b8 | 1427 | static int can_do_hugetlb_shm(void) |
1da177e4 | 1428 | { |
a0eb3a05 EB |
1429 | kgid_t shm_group; |
1430 | shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group); | |
1431 | return capable(CAP_IPC_LOCK) || in_group_p(shm_group); | |
1da177e4 LT |
1432 | } |
1433 | ||
42d7395f AK |
1434 | static int get_hstate_idx(int page_size_log) |
1435 | { | |
af73e4d9 | 1436 | struct hstate *h = hstate_sizelog(page_size_log); |
42d7395f | 1437 | |
42d7395f AK |
1438 | if (!h) |
1439 | return -1; | |
1440 | return h - hstates; | |
1441 | } | |
1442 | ||
af73e4d9 NH |
1443 | /* |
1444 | * Note that size should be aligned to proper hugepage size in caller side, | |
1445 | * otherwise hugetlb_reserve_pages reserves one less hugepages than intended. | |
1446 | */ | |
1447 | struct file *hugetlb_file_setup(const char *name, size_t size, | |
1448 | vm_flags_t acctflag, struct user_struct **user, | |
42d7395f | 1449 | int creat_flags, int page_size_log) |
1da177e4 | 1450 | { |
1da177e4 | 1451 | struct inode *inode; |
e68375c8 | 1452 | struct vfsmount *mnt; |
42d7395f | 1453 | int hstate_idx; |
e68375c8 | 1454 | struct file *file; |
42d7395f AK |
1455 | |
1456 | hstate_idx = get_hstate_idx(page_size_log); | |
1457 | if (hstate_idx < 0) | |
1458 | return ERR_PTR(-ENODEV); | |
1da177e4 | 1459 | |
353d5c30 | 1460 | *user = NULL; |
e68375c8 AV |
1461 | mnt = hugetlbfs_vfsmount[hstate_idx]; |
1462 | if (!mnt) | |
5bc98594 AM |
1463 | return ERR_PTR(-ENOENT); |
1464 | ||
ef1ff6b8 | 1465 | if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) { |
353d5c30 HD |
1466 | *user = current_user(); |
1467 | if (user_shm_lock(size, *user)) { | |
21a3c273 | 1468 | task_lock(current); |
9b857d26 | 1469 | pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n", |
21a3c273 DR |
1470 | current->comm, current->pid); |
1471 | task_unlock(current); | |
353d5c30 HD |
1472 | } else { |
1473 | *user = NULL; | |
2584e517 | 1474 | return ERR_PTR(-EPERM); |
353d5c30 | 1475 | } |
2584e517 | 1476 | } |
1da177e4 | 1477 | |
39b65252 | 1478 | file = ERR_PTR(-ENOSPC); |
e68375c8 | 1479 | inode = hugetlbfs_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0); |
1da177e4 | 1480 | if (!inode) |
e68375c8 | 1481 | goto out; |
e1832f29 SS |
1482 | if (creat_flags == HUGETLB_SHMFS_INODE) |
1483 | inode->i_flags |= S_PRIVATE; | |
1da177e4 | 1484 | |
1da177e4 | 1485 | inode->i_size = size; |
6d6b77f1 | 1486 | clear_nlink(inode); |
ce8d2cdf | 1487 | |
e68375c8 AV |
1488 | if (hugetlb_reserve_pages(inode, 0, |
1489 | size >> huge_page_shift(hstate_inode(inode)), NULL, | |
1490 | acctflag)) | |
1491 | file = ERR_PTR(-ENOMEM); | |
1492 | else | |
1493 | file = alloc_file_pseudo(inode, mnt, name, O_RDWR, | |
1494 | &hugetlbfs_file_operations); | |
1495 | if (!IS_ERR(file)) | |
1496 | return file; | |
1da177e4 | 1497 | |
b45b5bd6 | 1498 | iput(inode); |
e68375c8 | 1499 | out: |
353d5c30 HD |
1500 | if (*user) { |
1501 | user_shm_unlock(size, *user); | |
1502 | *user = NULL; | |
1503 | } | |
39b65252 | 1504 | return file; |
1da177e4 LT |
1505 | } |
1506 | ||
32021982 DH |
1507 | static struct vfsmount *__init mount_one_hugetlbfs(struct hstate *h) |
1508 | { | |
1509 | struct fs_context *fc; | |
1510 | struct vfsmount *mnt; | |
1511 | ||
1512 | fc = fs_context_for_mount(&hugetlbfs_fs_type, SB_KERNMOUNT); | |
1513 | if (IS_ERR(fc)) { | |
1514 | mnt = ERR_CAST(fc); | |
1515 | } else { | |
1516 | struct hugetlbfs_fs_context *ctx = fc->fs_private; | |
1517 | ctx->hstate = h; | |
1518 | mnt = fc_mount(fc); | |
1519 | put_fs_context(fc); | |
1520 | } | |
1521 | if (IS_ERR(mnt)) | |
1522 | pr_err("Cannot mount internal hugetlbfs for page size %uK", | |
1523 | 1U << (h->order + PAGE_SHIFT - 10)); | |
1524 | return mnt; | |
1525 | } | |
1526 | ||
1da177e4 LT |
1527 | static int __init init_hugetlbfs_fs(void) |
1528 | { | |
32021982 | 1529 | struct vfsmount *mnt; |
42d7395f | 1530 | struct hstate *h; |
1da177e4 | 1531 | int error; |
42d7395f | 1532 | int i; |
1da177e4 | 1533 | |
457c1b27 | 1534 | if (!hugepages_supported()) { |
9b857d26 | 1535 | pr_info("disabling because there are no supported hugepage sizes\n"); |
457c1b27 NA |
1536 | return -ENOTSUPP; |
1537 | } | |
1538 | ||
d1d5e05f | 1539 | error = -ENOMEM; |
1da177e4 LT |
1540 | hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache", |
1541 | sizeof(struct hugetlbfs_inode_info), | |
5d097056 | 1542 | 0, SLAB_ACCOUNT, init_once); |
1da177e4 | 1543 | if (hugetlbfs_inode_cachep == NULL) |
8fc312b3 | 1544 | goto out; |
1da177e4 LT |
1545 | |
1546 | error = register_filesystem(&hugetlbfs_fs_type); | |
1547 | if (error) | |
8fc312b3 | 1548 | goto out_free; |
1da177e4 | 1549 | |
8fc312b3 MK |
1550 | /* default hstate mount is required */ |
1551 | mnt = mount_one_hugetlbfs(&hstates[default_hstate_idx]); | |
1552 | if (IS_ERR(mnt)) { | |
1553 | error = PTR_ERR(mnt); | |
1554 | goto out_unreg; | |
1555 | } | |
1556 | hugetlbfs_vfsmount[default_hstate_idx] = mnt; | |
1557 | ||
1558 | /* other hstates are optional */ | |
42d7395f AK |
1559 | i = 0; |
1560 | for_each_hstate(h) { | |
15f0ec94 JS |
1561 | if (i == default_hstate_idx) { |
1562 | i++; | |
8fc312b3 | 1563 | continue; |
15f0ec94 | 1564 | } |
8fc312b3 | 1565 | |
32021982 | 1566 | mnt = mount_one_hugetlbfs(h); |
8fc312b3 MK |
1567 | if (IS_ERR(mnt)) |
1568 | hugetlbfs_vfsmount[i] = NULL; | |
1569 | else | |
1570 | hugetlbfs_vfsmount[i] = mnt; | |
42d7395f AK |
1571 | i++; |
1572 | } | |
32021982 DH |
1573 | |
1574 | return 0; | |
1da177e4 | 1575 | |
8fc312b3 MK |
1576 | out_unreg: |
1577 | (void)unregister_filesystem(&hugetlbfs_fs_type); | |
1578 | out_free: | |
d1d5e05f | 1579 | kmem_cache_destroy(hugetlbfs_inode_cachep); |
8fc312b3 | 1580 | out: |
1da177e4 LT |
1581 | return error; |
1582 | } | |
3e89e1c5 | 1583 | fs_initcall(init_hugetlbfs_fs) |