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), | |
e0f7e2b2 | 80 | fsparam_u32oct("mode", Opt_mode), |
32021982 DH |
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); |
22247efd | 134 | struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); |
1da177e4 LT |
135 | loff_t len, vma_len; |
136 | int ret; | |
a5516438 | 137 | struct hstate *h = hstate_file(file); |
1da177e4 | 138 | |
68589bc3 | 139 | /* |
dec4ad86 DG |
140 | * vma address alignment (but not the pgoff alignment) has |
141 | * already been checked by prepare_hugepage_range. If you add | |
142 | * any error returns here, do so after setting VM_HUGETLB, so | |
143 | * is_vm_hugetlb_page tests below unmap_region go the right | |
45e55300 | 144 | * way when do_mmap unwinds (may be important on powerpc |
dec4ad86 | 145 | * and ia64). |
68589bc3 | 146 | */ |
a2fce914 | 147 | vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND; |
68589bc3 | 148 | vma->vm_ops = &hugetlb_vm_ops; |
1da177e4 | 149 | |
22247efd PX |
150 | ret = seal_check_future_write(info->seals, vma); |
151 | if (ret) | |
152 | return ret; | |
153 | ||
045c7a3f | 154 | /* |
63489f8e | 155 | * page based offset in vm_pgoff could be sufficiently large to |
5df63c2a MK |
156 | * overflow a loff_t when converted to byte offset. This can |
157 | * only happen on architectures where sizeof(loff_t) == | |
158 | * sizeof(unsigned long). So, only check in those instances. | |
045c7a3f | 159 | */ |
5df63c2a MK |
160 | if (sizeof(unsigned long) == sizeof(loff_t)) { |
161 | if (vma->vm_pgoff & PGOFF_LOFFT_MAX) | |
162 | return -EINVAL; | |
163 | } | |
045c7a3f | 164 | |
63489f8e | 165 | /* must be huge page aligned */ |
2b37c35e | 166 | if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT)) |
dec4ad86 DG |
167 | return -EINVAL; |
168 | ||
1da177e4 | 169 | vma_len = (loff_t)(vma->vm_end - vma->vm_start); |
045c7a3f MK |
170 | len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); |
171 | /* check for overflow */ | |
172 | if (len < vma_len) | |
173 | return -EINVAL; | |
1da177e4 | 174 | |
5955102c | 175 | inode_lock(inode); |
1da177e4 | 176 | file_accessed(file); |
1da177e4 LT |
177 | |
178 | ret = -ENOMEM; | |
33b8f84a | 179 | if (!hugetlb_reserve_pages(inode, |
a5516438 | 180 | vma->vm_pgoff >> huge_page_order(h), |
5a6fe125 MG |
181 | len >> huge_page_shift(h), vma, |
182 | vma->vm_flags)) | |
a43a8c39 | 183 | goto out; |
b45b5bd6 | 184 | |
4c887265 | 185 | ret = 0; |
b6174df5 | 186 | if (vma->vm_flags & VM_WRITE && inode->i_size < len) |
045c7a3f | 187 | i_size_write(inode, len); |
1da177e4 | 188 | out: |
5955102c | 189 | inode_unlock(inode); |
1da177e4 LT |
190 | |
191 | return ret; | |
192 | } | |
193 | ||
194 | /* | |
3e4e28c5 | 195 | * Called under mmap_write_lock(mm). |
1da177e4 LT |
196 | */ |
197 | ||
88590253 SH |
198 | static unsigned long |
199 | hugetlb_get_unmapped_area_bottomup(struct file *file, unsigned long addr, | |
200 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
201 | { | |
202 | struct hstate *h = hstate_file(file); | |
203 | struct vm_unmapped_area_info info; | |
204 | ||
205 | info.flags = 0; | |
206 | info.length = len; | |
207 | info.low_limit = current->mm->mmap_base; | |
2cb4de08 | 208 | info.high_limit = arch_get_mmap_end(addr, len, flags); |
88590253 SH |
209 | info.align_mask = PAGE_MASK & ~huge_page_mask(h); |
210 | info.align_offset = 0; | |
211 | return vm_unmapped_area(&info); | |
212 | } | |
213 | ||
214 | static unsigned long | |
215 | hugetlb_get_unmapped_area_topdown(struct file *file, unsigned long addr, | |
216 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
217 | { | |
218 | struct hstate *h = hstate_file(file); | |
219 | struct vm_unmapped_area_info info; | |
220 | ||
221 | info.flags = VM_UNMAPPED_AREA_TOPDOWN; | |
222 | info.length = len; | |
223 | info.low_limit = max(PAGE_SIZE, mmap_min_addr); | |
5f24d5a5 | 224 | info.high_limit = arch_get_mmap_base(addr, current->mm->mmap_base); |
88590253 SH |
225 | info.align_mask = PAGE_MASK & ~huge_page_mask(h); |
226 | info.align_offset = 0; | |
227 | addr = vm_unmapped_area(&info); | |
228 | ||
229 | /* | |
230 | * A failed mmap() very likely causes application failure, | |
231 | * so fall back to the bottom-up function here. This scenario | |
232 | * can happen with large stack limits and large mmap() | |
233 | * allocations. | |
234 | */ | |
235 | if (unlikely(offset_in_page(addr))) { | |
236 | VM_BUG_ON(addr != -ENOMEM); | |
237 | info.flags = 0; | |
238 | info.low_limit = current->mm->mmap_base; | |
2cb4de08 | 239 | info.high_limit = arch_get_mmap_end(addr, len, flags); |
88590253 SH |
240 | addr = vm_unmapped_area(&info); |
241 | } | |
242 | ||
243 | return addr; | |
244 | } | |
245 | ||
4b439e25 CL |
246 | unsigned long |
247 | generic_hugetlb_get_unmapped_area(struct file *file, unsigned long addr, | |
248 | unsigned long len, unsigned long pgoff, | |
249 | unsigned long flags) | |
1da177e4 LT |
250 | { |
251 | struct mm_struct *mm = current->mm; | |
252 | struct vm_area_struct *vma; | |
a5516438 | 253 | struct hstate *h = hstate_file(file); |
2cb4de08 | 254 | const unsigned long mmap_end = arch_get_mmap_end(addr, len, flags); |
1da177e4 | 255 | |
a5516438 | 256 | if (len & ~huge_page_mask(h)) |
1da177e4 LT |
257 | return -EINVAL; |
258 | if (len > TASK_SIZE) | |
259 | return -ENOMEM; | |
260 | ||
036e0856 | 261 | if (flags & MAP_FIXED) { |
a5516438 | 262 | if (prepare_hugepage_range(file, addr, len)) |
036e0856 BH |
263 | return -EINVAL; |
264 | return addr; | |
265 | } | |
266 | ||
1da177e4 | 267 | if (addr) { |
a5516438 | 268 | addr = ALIGN(addr, huge_page_size(h)); |
1da177e4 | 269 | vma = find_vma(mm, addr); |
5f24d5a5 | 270 | if (mmap_end - len >= addr && |
1be7107f | 271 | (!vma || addr + len <= vm_start_gap(vma))) |
1da177e4 LT |
272 | return addr; |
273 | } | |
274 | ||
88590253 SH |
275 | /* |
276 | * Use mm->get_unmapped_area value as a hint to use topdown routine. | |
277 | * If architectures have special needs, they should define their own | |
278 | * version of hugetlb_get_unmapped_area. | |
279 | */ | |
280 | if (mm->get_unmapped_area == arch_get_unmapped_area_topdown) | |
281 | return hugetlb_get_unmapped_area_topdown(file, addr, len, | |
282 | pgoff, flags); | |
283 | return hugetlb_get_unmapped_area_bottomup(file, addr, len, | |
284 | pgoff, flags); | |
1da177e4 | 285 | } |
4b439e25 CL |
286 | |
287 | #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA | |
288 | static unsigned long | |
289 | hugetlb_get_unmapped_area(struct file *file, unsigned long addr, | |
290 | unsigned long len, unsigned long pgoff, | |
291 | unsigned long flags) | |
292 | { | |
293 | return generic_hugetlb_get_unmapped_area(file, addr, len, pgoff, flags); | |
294 | } | |
1da177e4 LT |
295 | #endif |
296 | ||
34d0640e | 297 | static size_t |
e63e1e5a | 298 | hugetlbfs_read_actor(struct page *page, unsigned long offset, |
34d0640e | 299 | struct iov_iter *to, unsigned long size) |
e63e1e5a | 300 | { |
34d0640e | 301 | size_t copied = 0; |
e63e1e5a BP |
302 | int i, chunksize; |
303 | ||
e63e1e5a | 304 | /* Find which 4k chunk and offset with in that chunk */ |
09cbfeaf KS |
305 | i = offset >> PAGE_SHIFT; |
306 | offset = offset & ~PAGE_MASK; | |
e63e1e5a BP |
307 | |
308 | while (size) { | |
34d0640e | 309 | size_t n; |
09cbfeaf | 310 | chunksize = PAGE_SIZE; |
e63e1e5a BP |
311 | if (offset) |
312 | chunksize -= offset; | |
313 | if (chunksize > size) | |
314 | chunksize = size; | |
34d0640e AV |
315 | n = copy_page_to_iter(&page[i], offset, chunksize, to); |
316 | copied += n; | |
317 | if (n != chunksize) | |
318 | return copied; | |
e63e1e5a BP |
319 | offset = 0; |
320 | size -= chunksize; | |
e63e1e5a BP |
321 | i++; |
322 | } | |
34d0640e | 323 | return copied; |
e63e1e5a BP |
324 | } |
325 | ||
326 | /* | |
327 | * Support for read() - Find the page attached to f_mapping and copy out the | |
c7e285e3 | 328 | * data. Its *very* similar to generic_file_buffered_read(), we can't use that |
ea1754a0 | 329 | * since it has PAGE_SIZE assumptions. |
e63e1e5a | 330 | */ |
34d0640e | 331 | static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to) |
e63e1e5a | 332 | { |
34d0640e AV |
333 | struct file *file = iocb->ki_filp; |
334 | struct hstate *h = hstate_file(file); | |
335 | struct address_space *mapping = file->f_mapping; | |
e63e1e5a | 336 | struct inode *inode = mapping->host; |
34d0640e AV |
337 | unsigned long index = iocb->ki_pos >> huge_page_shift(h); |
338 | unsigned long offset = iocb->ki_pos & ~huge_page_mask(h); | |
e63e1e5a BP |
339 | unsigned long end_index; |
340 | loff_t isize; | |
341 | ssize_t retval = 0; | |
342 | ||
34d0640e | 343 | while (iov_iter_count(to)) { |
e63e1e5a | 344 | struct page *page; |
34d0640e | 345 | size_t nr, copied; |
e63e1e5a BP |
346 | |
347 | /* nr is the maximum number of bytes to copy from this page */ | |
a5516438 | 348 | nr = huge_page_size(h); |
a05b0855 AK |
349 | isize = i_size_read(inode); |
350 | if (!isize) | |
34d0640e | 351 | break; |
a05b0855 | 352 | end_index = (isize - 1) >> huge_page_shift(h); |
34d0640e AV |
353 | if (index > end_index) |
354 | break; | |
355 | if (index == end_index) { | |
a5516438 | 356 | nr = ((isize - 1) & ~huge_page_mask(h)) + 1; |
a05b0855 | 357 | if (nr <= offset) |
34d0640e | 358 | break; |
e63e1e5a BP |
359 | } |
360 | nr = nr - offset; | |
361 | ||
362 | /* Find the page */ | |
a05b0855 | 363 | page = find_lock_page(mapping, index); |
e63e1e5a BP |
364 | if (unlikely(page == NULL)) { |
365 | /* | |
366 | * We have a HOLE, zero out the user-buffer for the | |
367 | * length of the hole or request. | |
368 | */ | |
34d0640e | 369 | copied = iov_iter_zero(nr, to); |
e63e1e5a | 370 | } else { |
a05b0855 AK |
371 | unlock_page(page); |
372 | ||
e63e1e5a BP |
373 | /* |
374 | * We have the page, copy it to user space buffer. | |
375 | */ | |
34d0640e | 376 | copied = hugetlbfs_read_actor(page, offset, to, nr); |
09cbfeaf | 377 | put_page(page); |
e63e1e5a | 378 | } |
34d0640e AV |
379 | offset += copied; |
380 | retval += copied; | |
381 | if (copied != nr && iov_iter_count(to)) { | |
382 | if (!retval) | |
383 | retval = -EFAULT; | |
384 | break; | |
e63e1e5a | 385 | } |
a5516438 AK |
386 | index += offset >> huge_page_shift(h); |
387 | offset &= ~huge_page_mask(h); | |
e63e1e5a | 388 | } |
34d0640e | 389 | iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset; |
e63e1e5a BP |
390 | return retval; |
391 | } | |
392 | ||
800d15a5 NP |
393 | static int hugetlbfs_write_begin(struct file *file, |
394 | struct address_space *mapping, | |
9d6b0cd7 | 395 | loff_t pos, unsigned len, |
800d15a5 | 396 | struct page **pagep, void **fsdata) |
1da177e4 LT |
397 | { |
398 | return -EINVAL; | |
399 | } | |
400 | ||
800d15a5 NP |
401 | static int hugetlbfs_write_end(struct file *file, struct address_space *mapping, |
402 | loff_t pos, unsigned len, unsigned copied, | |
403 | struct page *page, void *fsdata) | |
1da177e4 | 404 | { |
800d15a5 | 405 | BUG(); |
1da177e4 LT |
406 | return -EINVAL; |
407 | } | |
408 | ||
b5cec28d | 409 | static void remove_huge_page(struct page *page) |
1da177e4 | 410 | { |
b9ea2515 | 411 | ClearPageDirty(page); |
1da177e4 | 412 | ClearPageUptodate(page); |
bd65cb86 | 413 | delete_from_page_cache(page); |
1da177e4 LT |
414 | } |
415 | ||
4aae8d1c | 416 | static void |
05e90bd0 PX |
417 | hugetlb_vmdelete_list(struct rb_root_cached *root, pgoff_t start, pgoff_t end, |
418 | zap_flags_t zap_flags) | |
4aae8d1c MK |
419 | { |
420 | struct vm_area_struct *vma; | |
421 | ||
422 | /* | |
d6aba4c8 SC |
423 | * end == 0 indicates that the entire range after start should be |
424 | * unmapped. Note, end is exclusive, whereas the interval tree takes | |
425 | * an inclusive "last". | |
4aae8d1c | 426 | */ |
d6aba4c8 | 427 | vma_interval_tree_foreach(vma, root, start, end ? end - 1 : ULONG_MAX) { |
4aae8d1c MK |
428 | unsigned long v_offset; |
429 | unsigned long v_end; | |
430 | ||
431 | /* | |
432 | * Can the expression below overflow on 32-bit arches? | |
433 | * No, because the interval tree returns us only those vmas | |
434 | * which overlap the truncated area starting at pgoff, | |
435 | * and no vma on a 32-bit arch can span beyond the 4GB. | |
436 | */ | |
437 | if (vma->vm_pgoff < start) | |
438 | v_offset = (start - vma->vm_pgoff) << PAGE_SHIFT; | |
439 | else | |
440 | v_offset = 0; | |
441 | ||
442 | if (!end) | |
443 | v_end = vma->vm_end; | |
444 | else { | |
445 | v_end = ((end - vma->vm_pgoff) << PAGE_SHIFT) | |
446 | + vma->vm_start; | |
447 | if (v_end > vma->vm_end) | |
448 | v_end = vma->vm_end; | |
449 | } | |
450 | ||
451 | unmap_hugepage_range(vma, vma->vm_start + v_offset, v_end, | |
05e90bd0 | 452 | NULL, zap_flags); |
4aae8d1c MK |
453 | } |
454 | } | |
b5cec28d MK |
455 | |
456 | /* | |
457 | * remove_inode_hugepages handles two distinct cases: truncation and hole | |
458 | * punch. There are subtle differences in operation for each case. | |
4aae8d1c | 459 | * |
b5cec28d MK |
460 | * truncation is indicated by end of range being LLONG_MAX |
461 | * In this case, we first scan the range and release found pages. | |
1935ebd3 | 462 | * After releasing pages, hugetlb_unreserve_pages cleans up region/reserve |
e7c58097 | 463 | * maps and global counts. Page faults can not race with truncation |
87bf91d3 MK |
464 | * in this routine. hugetlb_no_page() holds i_mmap_rwsem and prevents |
465 | * page faults in the truncated range by checking i_size. i_size is | |
466 | * modified while holding i_mmap_rwsem. | |
b5cec28d MK |
467 | * hole punch is indicated if end is not LLONG_MAX |
468 | * In the hole punch case we scan the range and release found pages. | |
1935ebd3 ML |
469 | * Only when releasing a page is the associated region/reserve map |
470 | * deleted. The region/reserve map for ranges without associated | |
e7c58097 MK |
471 | * pages are not modified. Page faults can race with hole punch. |
472 | * This is indicated if we find a mapped page. | |
b5cec28d MK |
473 | * Note: If the passed end of range value is beyond the end of file, but |
474 | * not LLONG_MAX this routine still performs a hole punch operation. | |
475 | */ | |
476 | static void remove_inode_hugepages(struct inode *inode, loff_t lstart, | |
477 | loff_t lend) | |
1da177e4 | 478 | { |
a5516438 | 479 | struct hstate *h = hstate_inode(inode); |
b45b5bd6 | 480 | struct address_space *mapping = &inode->i_data; |
a5516438 | 481 | const pgoff_t start = lstart >> huge_page_shift(h); |
b5cec28d | 482 | const pgoff_t end = lend >> huge_page_shift(h); |
1da177e4 | 483 | struct pagevec pvec; |
d72dc8a2 | 484 | pgoff_t next, index; |
a43a8c39 | 485 | int i, freed = 0; |
b5cec28d | 486 | bool truncate_op = (lend == LLONG_MAX); |
1da177e4 | 487 | |
86679820 | 488 | pagevec_init(&pvec); |
1da177e4 | 489 | next = start; |
b5cec28d | 490 | while (next < end) { |
b5cec28d | 491 | /* |
1817889e | 492 | * When no more pages are found, we are done. |
b5cec28d | 493 | */ |
397162ff | 494 | if (!pagevec_lookup_range(&pvec, mapping, &next, end - 1)) |
1817889e | 495 | break; |
1da177e4 LT |
496 | |
497 | for (i = 0; i < pagevec_count(&pvec); ++i) { | |
498 | struct page *page = pvec.pages[i]; | |
d4241a04 | 499 | u32 hash = 0; |
b5cec28d | 500 | |
d72dc8a2 | 501 | index = page->index; |
87bf91d3 MK |
502 | if (!truncate_op) { |
503 | /* | |
504 | * Only need to hold the fault mutex in the | |
505 | * hole punch case. This prevents races with | |
506 | * page faults. Races are not possible in the | |
507 | * case of truncation. | |
508 | */ | |
d4241a04 | 509 | hash = hugetlb_fault_mutex_hash(mapping, index); |
87bf91d3 MK |
510 | mutex_lock(&hugetlb_fault_mutex_table[hash]); |
511 | } | |
e7c58097 | 512 | |
4aae8d1c | 513 | /* |
e7c58097 MK |
514 | * If page is mapped, it was faulted in after being |
515 | * unmapped in caller. Unmap (again) now after taking | |
516 | * the fault mutex. The mutex will prevent faults | |
517 | * until we finish removing the page. | |
518 | * | |
519 | * This race can only happen in the hole punch case. | |
520 | * Getting here in a truncate operation is a bug. | |
4aae8d1c | 521 | */ |
e7c58097 MK |
522 | if (unlikely(page_mapped(page))) { |
523 | BUG_ON(truncate_op); | |
524 | ||
c0d0381a | 525 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); |
e7c58097 | 526 | i_mmap_lock_write(mapping); |
c0d0381a | 527 | mutex_lock(&hugetlb_fault_mutex_table[hash]); |
e7c58097 MK |
528 | hugetlb_vmdelete_list(&mapping->i_mmap, |
529 | index * pages_per_huge_page(h), | |
05e90bd0 PX |
530 | (index + 1) * pages_per_huge_page(h), |
531 | ZAP_FLAG_DROP_MARKER); | |
e7c58097 MK |
532 | i_mmap_unlock_write(mapping); |
533 | } | |
4aae8d1c MK |
534 | |
535 | lock_page(page); | |
536 | /* | |
537 | * We must free the huge page and remove from page | |
538 | * cache (remove_huge_page) BEFORE removing the | |
539 | * region/reserve map (hugetlb_unreserve_pages). In | |
540 | * rare out of memory conditions, removal of the | |
72e2936c | 541 | * region/reserve map could fail. Correspondingly, |
542 | * the subpool and global reserve usage count can need | |
543 | * to be adjusted. | |
4aae8d1c | 544 | */ |
e32905e5 | 545 | VM_BUG_ON(HPageRestoreReserve(page)); |
4aae8d1c MK |
546 | remove_huge_page(page); |
547 | freed++; | |
548 | if (!truncate_op) { | |
549 | if (unlikely(hugetlb_unreserve_pages(inode, | |
d72dc8a2 | 550 | index, index + 1, 1))) |
72e2936c | 551 | hugetlb_fix_reserve_counts(inode); |
b5cec28d MK |
552 | } |
553 | ||
1da177e4 | 554 | unlock_page(page); |
87bf91d3 MK |
555 | if (!truncate_op) |
556 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
1da177e4 LT |
557 | } |
558 | huge_pagevec_release(&pvec); | |
1817889e | 559 | cond_resched(); |
1da177e4 | 560 | } |
b5cec28d MK |
561 | |
562 | if (truncate_op) | |
563 | (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed); | |
1da177e4 LT |
564 | } |
565 | ||
2bbbda30 | 566 | static void hugetlbfs_evict_inode(struct inode *inode) |
1da177e4 | 567 | { |
9119a41e JK |
568 | struct resv_map *resv_map; |
569 | ||
b5cec28d | 570 | remove_inode_hugepages(inode, 0, LLONG_MAX); |
f27a5136 MK |
571 | |
572 | /* | |
573 | * Get the resv_map from the address space embedded in the inode. | |
574 | * This is the address space which points to any resv_map allocated | |
575 | * at inode creation time. If this is a device special inode, | |
576 | * i_mapping may not point to the original address space. | |
577 | */ | |
578 | resv_map = (struct resv_map *)(&inode->i_data)->private_data; | |
579 | /* Only regular and link inodes have associated reserve maps */ | |
9119a41e JK |
580 | if (resv_map) |
581 | resv_map_release(&resv_map->refs); | |
dbd5768f | 582 | clear_inode(inode); |
149f4211 CH |
583 | } |
584 | ||
e5d319de | 585 | static void hugetlb_vmtruncate(struct inode *inode, loff_t offset) |
1da177e4 | 586 | { |
856fc295 | 587 | pgoff_t pgoff; |
1da177e4 | 588 | struct address_space *mapping = inode->i_mapping; |
a5516438 | 589 | struct hstate *h = hstate_inode(inode); |
1da177e4 | 590 | |
a5516438 | 591 | BUG_ON(offset & ~huge_page_mask(h)); |
856fc295 | 592 | pgoff = offset >> PAGE_SHIFT; |
1da177e4 | 593 | |
83cde9e8 | 594 | i_mmap_lock_write(mapping); |
87bf91d3 | 595 | i_size_write(inode, offset); |
f808c13f | 596 | if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)) |
05e90bd0 PX |
597 | hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0, |
598 | ZAP_FLAG_DROP_MARKER); | |
c86aa7bb | 599 | i_mmap_unlock_write(mapping); |
e7c58097 | 600 | remove_inode_hugepages(inode, offset, LLONG_MAX); |
1da177e4 LT |
601 | } |
602 | ||
68d32527 MK |
603 | static void hugetlbfs_zero_partial_page(struct hstate *h, |
604 | struct address_space *mapping, | |
605 | loff_t start, | |
606 | loff_t end) | |
607 | { | |
608 | pgoff_t idx = start >> huge_page_shift(h); | |
609 | struct folio *folio; | |
610 | ||
611 | folio = filemap_lock_folio(mapping, idx); | |
612 | if (!folio) | |
613 | return; | |
614 | ||
615 | start = start & ~huge_page_mask(h); | |
616 | end = end & ~huge_page_mask(h); | |
617 | if (!end) | |
618 | end = huge_page_size(h); | |
619 | ||
620 | folio_zero_segment(folio, (size_t)start, (size_t)end); | |
621 | ||
622 | folio_unlock(folio); | |
623 | folio_put(folio); | |
624 | } | |
625 | ||
70c3547e MK |
626 | static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len) |
627 | { | |
68d32527 MK |
628 | struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); |
629 | struct address_space *mapping = inode->i_mapping; | |
70c3547e MK |
630 | struct hstate *h = hstate_inode(inode); |
631 | loff_t hpage_size = huge_page_size(h); | |
632 | loff_t hole_start, hole_end; | |
633 | ||
634 | /* | |
68d32527 | 635 | * hole_start and hole_end indicate the full pages within the hole. |
70c3547e MK |
636 | */ |
637 | hole_start = round_up(offset, hpage_size); | |
638 | hole_end = round_down(offset + len, hpage_size); | |
639 | ||
68d32527 MK |
640 | inode_lock(inode); |
641 | ||
642 | /* protected by i_rwsem */ | |
643 | if (info->seals & (F_SEAL_WRITE | F_SEAL_FUTURE_WRITE)) { | |
644 | inode_unlock(inode); | |
645 | return -EPERM; | |
646 | } | |
70c3547e | 647 | |
68d32527 | 648 | i_mmap_lock_write(mapping); |
ff62a342 | 649 | |
68d32527 MK |
650 | /* If range starts before first full page, zero partial page. */ |
651 | if (offset < hole_start) | |
652 | hugetlbfs_zero_partial_page(h, mapping, | |
653 | offset, min(offset + len, hole_start)); | |
ff62a342 | 654 | |
68d32527 MK |
655 | /* Unmap users of full pages in the hole. */ |
656 | if (hole_end > hole_start) { | |
f808c13f | 657 | if (!RB_EMPTY_ROOT(&mapping->i_mmap.rb_root)) |
70c3547e | 658 | hugetlb_vmdelete_list(&mapping->i_mmap, |
05e90bd0 PX |
659 | hole_start >> PAGE_SHIFT, |
660 | hole_end >> PAGE_SHIFT, 0); | |
70c3547e MK |
661 | } |
662 | ||
68d32527 MK |
663 | /* If range extends beyond last full page, zero partial page. */ |
664 | if ((offset + len) > hole_end && (offset + len) > hole_start) | |
665 | hugetlbfs_zero_partial_page(h, mapping, | |
666 | hole_end, offset + len); | |
667 | ||
668 | i_mmap_unlock_write(mapping); | |
669 | ||
670 | /* Remove full pages from the file. */ | |
671 | if (hole_end > hole_start) | |
672 | remove_inode_hugepages(inode, hole_start, hole_end); | |
673 | ||
674 | inode_unlock(inode); | |
675 | ||
70c3547e MK |
676 | return 0; |
677 | } | |
678 | ||
679 | static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset, | |
680 | loff_t len) | |
681 | { | |
682 | struct inode *inode = file_inode(file); | |
ff62a342 | 683 | struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); |
70c3547e MK |
684 | struct address_space *mapping = inode->i_mapping; |
685 | struct hstate *h = hstate_inode(inode); | |
686 | struct vm_area_struct pseudo_vma; | |
687 | struct mm_struct *mm = current->mm; | |
688 | loff_t hpage_size = huge_page_size(h); | |
689 | unsigned long hpage_shift = huge_page_shift(h); | |
690 | pgoff_t start, index, end; | |
691 | int error; | |
692 | u32 hash; | |
693 | ||
694 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
695 | return -EOPNOTSUPP; | |
696 | ||
697 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
698 | return hugetlbfs_punch_hole(inode, offset, len); | |
699 | ||
700 | /* | |
701 | * Default preallocate case. | |
702 | * For this range, start is rounded down and end is rounded up | |
703 | * as well as being converted to page offsets. | |
704 | */ | |
705 | start = offset >> hpage_shift; | |
706 | end = (offset + len + hpage_size - 1) >> hpage_shift; | |
707 | ||
5955102c | 708 | inode_lock(inode); |
70c3547e MK |
709 | |
710 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
711 | error = inode_newsize_ok(inode, offset + len); | |
712 | if (error) | |
713 | goto out; | |
714 | ||
ff62a342 MAL |
715 | if ((info->seals & F_SEAL_GROW) && offset + len > inode->i_size) { |
716 | error = -EPERM; | |
717 | goto out; | |
718 | } | |
719 | ||
70c3547e MK |
720 | /* |
721 | * Initialize a pseudo vma as this is required by the huge page | |
722 | * allocation routines. If NUMA is configured, use page index | |
723 | * as input to create an allocation policy. | |
724 | */ | |
2c4541e2 | 725 | vma_init(&pseudo_vma, mm); |
70c3547e MK |
726 | pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); |
727 | pseudo_vma.vm_file = file; | |
728 | ||
729 | for (index = start; index < end; index++) { | |
730 | /* | |
731 | * This is supposed to be the vaddr where the page is being | |
732 | * faulted in, but we have no vaddr here. | |
733 | */ | |
734 | struct page *page; | |
735 | unsigned long addr; | |
70c3547e MK |
736 | |
737 | cond_resched(); | |
738 | ||
739 | /* | |
740 | * fallocate(2) manpage permits EINTR; we may have been | |
741 | * interrupted because we are using up too much memory. | |
742 | */ | |
743 | if (signal_pending(current)) { | |
744 | error = -EINTR; | |
745 | break; | |
746 | } | |
747 | ||
748 | /* Set numa allocation policy based on index */ | |
749 | hugetlb_set_vma_policy(&pseudo_vma, inode, index); | |
750 | ||
751 | /* addr is the offset within the file (zero based) */ | |
752 | addr = index * hpage_size; | |
753 | ||
87bf91d3 MK |
754 | /* |
755 | * fault mutex taken here, protects against fault path | |
756 | * and hole punch. inode_lock previously taken protects | |
757 | * against truncation. | |
758 | */ | |
188b04a7 | 759 | hash = hugetlb_fault_mutex_hash(mapping, index); |
70c3547e MK |
760 | mutex_lock(&hugetlb_fault_mutex_table[hash]); |
761 | ||
762 | /* See if already present in mapping to avoid alloc/free */ | |
763 | page = find_get_page(mapping, index); | |
764 | if (page) { | |
765 | put_page(page); | |
766 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
767 | hugetlb_drop_vma_policy(&pseudo_vma); | |
768 | continue; | |
769 | } | |
770 | ||
88ce3fef ML |
771 | /* |
772 | * Allocate page without setting the avoid_reserve argument. | |
773 | * There certainly are no reserves associated with the | |
774 | * pseudo_vma. However, there could be shared mappings with | |
775 | * reserves for the file at the inode level. If we fallocate | |
776 | * pages in these areas, we need to consume the reserves | |
777 | * to keep reservation accounting consistent. | |
778 | */ | |
779 | page = alloc_huge_page(&pseudo_vma, addr, 0); | |
70c3547e MK |
780 | hugetlb_drop_vma_policy(&pseudo_vma); |
781 | if (IS_ERR(page)) { | |
782 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
783 | error = PTR_ERR(page); | |
784 | goto out; | |
785 | } | |
786 | clear_huge_page(page, addr, pages_per_huge_page(h)); | |
787 | __SetPageUptodate(page); | |
788 | error = huge_add_to_page_cache(page, mapping, index); | |
789 | if (unlikely(error)) { | |
846be085 | 790 | restore_reserve_on_error(h, &pseudo_vma, addr, page); |
70c3547e MK |
791 | put_page(page); |
792 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
793 | goto out; | |
794 | } | |
795 | ||
796 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
797 | ||
8f251a3d | 798 | SetHPageMigratable(page); |
70c3547e | 799 | /* |
70c3547e | 800 | * unlock_page because locked by add_to_page_cache() |
585fc0d2 | 801 | * put_page() due to reference from alloc_huge_page() |
70c3547e | 802 | */ |
70c3547e | 803 | unlock_page(page); |
72639e6d | 804 | put_page(page); |
70c3547e MK |
805 | } |
806 | ||
807 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
808 | i_size_write(inode, offset + len); | |
078cd827 | 809 | inode->i_ctime = current_time(inode); |
70c3547e | 810 | out: |
5955102c | 811 | inode_unlock(inode); |
70c3547e MK |
812 | return error; |
813 | } | |
814 | ||
549c7297 CB |
815 | static int hugetlbfs_setattr(struct user_namespace *mnt_userns, |
816 | struct dentry *dentry, struct iattr *attr) | |
1da177e4 | 817 | { |
2b0143b5 | 818 | struct inode *inode = d_inode(dentry); |
a5516438 | 819 | struct hstate *h = hstate_inode(inode); |
1da177e4 LT |
820 | int error; |
821 | unsigned int ia_valid = attr->ia_valid; | |
ff62a342 | 822 | struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); |
1da177e4 | 823 | |
2f221d6f | 824 | error = setattr_prepare(&init_user_ns, dentry, attr); |
1da177e4 | 825 | if (error) |
1025774c | 826 | return error; |
1da177e4 LT |
827 | |
828 | if (ia_valid & ATTR_SIZE) { | |
ff62a342 MAL |
829 | loff_t oldsize = inode->i_size; |
830 | loff_t newsize = attr->ia_size; | |
831 | ||
832 | if (newsize & ~huge_page_mask(h)) | |
1025774c | 833 | return -EINVAL; |
398c0da7 | 834 | /* protected by i_rwsem */ |
ff62a342 MAL |
835 | if ((newsize < oldsize && (info->seals & F_SEAL_SHRINK)) || |
836 | (newsize > oldsize && (info->seals & F_SEAL_GROW))) | |
837 | return -EPERM; | |
e5d319de | 838 | hugetlb_vmtruncate(inode, newsize); |
1da177e4 | 839 | } |
1025774c | 840 | |
2f221d6f | 841 | setattr_copy(&init_user_ns, inode, attr); |
1025774c CH |
842 | mark_inode_dirty(inode); |
843 | return 0; | |
1da177e4 LT |
844 | } |
845 | ||
7d54fa64 | 846 | static struct inode *hugetlbfs_get_root(struct super_block *sb, |
32021982 | 847 | struct hugetlbfs_fs_context *ctx) |
1da177e4 LT |
848 | { |
849 | struct inode *inode; | |
1da177e4 LT |
850 | |
851 | inode = new_inode(sb); | |
852 | if (inode) { | |
85fe4025 | 853 | inode->i_ino = get_next_ino(); |
32021982 DH |
854 | inode->i_mode = S_IFDIR | ctx->mode; |
855 | inode->i_uid = ctx->uid; | |
856 | inode->i_gid = ctx->gid; | |
078cd827 | 857 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
7d54fa64 AV |
858 | inode->i_op = &hugetlbfs_dir_inode_operations; |
859 | inode->i_fop = &simple_dir_operations; | |
860 | /* directory inodes start off with i_nlink == 2 (for "." entry) */ | |
861 | inc_nlink(inode); | |
65ed7601 | 862 | lockdep_annotate_inode_mutex_key(inode); |
7d54fa64 AV |
863 | } |
864 | return inode; | |
865 | } | |
866 | ||
b610ded7 | 867 | /* |
c8c06efa | 868 | * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never |
b610ded7 | 869 | * be taken from reclaim -- unlike regular filesystems. This needs an |
88f306b6 | 870 | * annotation because huge_pmd_share() does an allocation under hugetlb's |
c8c06efa | 871 | * i_mmap_rwsem. |
b610ded7 | 872 | */ |
c8c06efa | 873 | static struct lock_class_key hugetlbfs_i_mmap_rwsem_key; |
b610ded7 | 874 | |
7d54fa64 AV |
875 | static struct inode *hugetlbfs_get_inode(struct super_block *sb, |
876 | struct inode *dir, | |
18df2252 | 877 | umode_t mode, dev_t dev) |
7d54fa64 AV |
878 | { |
879 | struct inode *inode; | |
58b6e5e8 | 880 | struct resv_map *resv_map = NULL; |
9119a41e | 881 | |
58b6e5e8 MK |
882 | /* |
883 | * Reserve maps are only needed for inodes that can have associated | |
884 | * page allocations. | |
885 | */ | |
886 | if (S_ISREG(mode) || S_ISLNK(mode)) { | |
887 | resv_map = resv_map_alloc(); | |
888 | if (!resv_map) | |
889 | return NULL; | |
890 | } | |
7d54fa64 AV |
891 | |
892 | inode = new_inode(sb); | |
893 | if (inode) { | |
ff62a342 MAL |
894 | struct hugetlbfs_inode_info *info = HUGETLBFS_I(inode); |
895 | ||
7d54fa64 | 896 | inode->i_ino = get_next_ino(); |
21cb47be | 897 | inode_init_owner(&init_user_ns, inode, dir, mode); |
c8c06efa DB |
898 | lockdep_set_class(&inode->i_mapping->i_mmap_rwsem, |
899 | &hugetlbfs_i_mmap_rwsem_key); | |
1da177e4 | 900 | inode->i_mapping->a_ops = &hugetlbfs_aops; |
078cd827 | 901 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
9119a41e | 902 | inode->i_mapping->private_data = resv_map; |
ff62a342 | 903 | info->seals = F_SEAL_SEAL; |
1da177e4 LT |
904 | switch (mode & S_IFMT) { |
905 | default: | |
906 | init_special_inode(inode, mode, dev); | |
907 | break; | |
908 | case S_IFREG: | |
909 | inode->i_op = &hugetlbfs_inode_operations; | |
910 | inode->i_fop = &hugetlbfs_file_operations; | |
911 | break; | |
912 | case S_IFDIR: | |
913 | inode->i_op = &hugetlbfs_dir_inode_operations; | |
914 | inode->i_fop = &simple_dir_operations; | |
915 | ||
916 | /* directory inodes start off with i_nlink == 2 (for "." entry) */ | |
d8c76e6f | 917 | inc_nlink(inode); |
1da177e4 LT |
918 | break; |
919 | case S_IFLNK: | |
920 | inode->i_op = &page_symlink_inode_operations; | |
21fc61c7 | 921 | inode_nohighmem(inode); |
1da177e4 LT |
922 | break; |
923 | } | |
e096d0c7 | 924 | lockdep_annotate_inode_mutex_key(inode); |
58b6e5e8 MK |
925 | } else { |
926 | if (resv_map) | |
927 | kref_put(&resv_map->refs, resv_map_release); | |
928 | } | |
9119a41e | 929 | |
1da177e4 LT |
930 | return inode; |
931 | } | |
932 | ||
933 | /* | |
934 | * File creation. Allocate an inode, and we're done.. | |
935 | */ | |
1ab5b82f PS |
936 | static int do_hugetlbfs_mknod(struct inode *dir, |
937 | struct dentry *dentry, | |
938 | umode_t mode, | |
939 | dev_t dev, | |
940 | bool tmpfile) | |
1da177e4 LT |
941 | { |
942 | struct inode *inode; | |
943 | int error = -ENOSPC; | |
7d54fa64 AV |
944 | |
945 | inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev); | |
1da177e4 | 946 | if (inode) { |
078cd827 | 947 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1ab5b82f PS |
948 | if (tmpfile) { |
949 | d_tmpfile(dentry, inode); | |
950 | } else { | |
951 | d_instantiate(dentry, inode); | |
952 | dget(dentry);/* Extra count - pin the dentry in core */ | |
953 | } | |
1da177e4 LT |
954 | error = 0; |
955 | } | |
956 | return error; | |
957 | } | |
958 | ||
549c7297 CB |
959 | static int hugetlbfs_mknod(struct user_namespace *mnt_userns, struct inode *dir, |
960 | struct dentry *dentry, umode_t mode, dev_t dev) | |
1ab5b82f PS |
961 | { |
962 | return do_hugetlbfs_mknod(dir, dentry, mode, dev, false); | |
963 | } | |
964 | ||
549c7297 CB |
965 | static int hugetlbfs_mkdir(struct user_namespace *mnt_userns, struct inode *dir, |
966 | struct dentry *dentry, umode_t mode) | |
1da177e4 | 967 | { |
549c7297 CB |
968 | int retval = hugetlbfs_mknod(&init_user_ns, dir, dentry, |
969 | mode | S_IFDIR, 0); | |
1da177e4 | 970 | if (!retval) |
d8c76e6f | 971 | inc_nlink(dir); |
1da177e4 LT |
972 | return retval; |
973 | } | |
974 | ||
549c7297 CB |
975 | static int hugetlbfs_create(struct user_namespace *mnt_userns, |
976 | struct inode *dir, struct dentry *dentry, | |
977 | umode_t mode, bool excl) | |
1da177e4 | 978 | { |
549c7297 | 979 | return hugetlbfs_mknod(&init_user_ns, dir, dentry, mode | S_IFREG, 0); |
1da177e4 LT |
980 | } |
981 | ||
549c7297 CB |
982 | static int hugetlbfs_tmpfile(struct user_namespace *mnt_userns, |
983 | struct inode *dir, struct dentry *dentry, | |
984 | umode_t mode) | |
1ab5b82f PS |
985 | { |
986 | return do_hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0, true); | |
987 | } | |
988 | ||
549c7297 CB |
989 | static int hugetlbfs_symlink(struct user_namespace *mnt_userns, |
990 | struct inode *dir, struct dentry *dentry, | |
991 | const char *symname) | |
1da177e4 LT |
992 | { |
993 | struct inode *inode; | |
994 | int error = -ENOSPC; | |
1da177e4 | 995 | |
7d54fa64 | 996 | inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0); |
1da177e4 LT |
997 | if (inode) { |
998 | int l = strlen(symname)+1; | |
999 | error = page_symlink(inode, symname, l); | |
1000 | if (!error) { | |
1001 | d_instantiate(dentry, inode); | |
1002 | dget(dentry); | |
1003 | } else | |
1004 | iput(inode); | |
1005 | } | |
078cd827 | 1006 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1da177e4 LT |
1007 | |
1008 | return error; | |
1009 | } | |
1010 | ||
290408d4 | 1011 | static int hugetlbfs_migrate_page(struct address_space *mapping, |
b969c4ab | 1012 | struct page *newpage, struct page *page, |
a6bc32b8 | 1013 | enum migrate_mode mode) |
290408d4 NH |
1014 | { |
1015 | int rc; | |
1016 | ||
1017 | rc = migrate_huge_page_move_mapping(mapping, newpage, page); | |
78bd5209 | 1018 | if (rc != MIGRATEPAGE_SUCCESS) |
290408d4 | 1019 | return rc; |
cb6acd01 | 1020 | |
d6995da3 MK |
1021 | if (hugetlb_page_subpool(page)) { |
1022 | hugetlb_set_page_subpool(newpage, hugetlb_page_subpool(page)); | |
1023 | hugetlb_set_page_subpool(page, NULL); | |
cb6acd01 MK |
1024 | } |
1025 | ||
2916ecc0 JG |
1026 | if (mode != MIGRATE_SYNC_NO_COPY) |
1027 | migrate_page_copy(newpage, page); | |
1028 | else | |
1029 | migrate_page_states(newpage, page); | |
290408d4 | 1030 | |
78bd5209 | 1031 | return MIGRATEPAGE_SUCCESS; |
290408d4 NH |
1032 | } |
1033 | ||
78bb9203 NH |
1034 | static int hugetlbfs_error_remove_page(struct address_space *mapping, |
1035 | struct page *page) | |
1036 | { | |
1037 | struct inode *inode = mapping->host; | |
ab615a5b | 1038 | pgoff_t index = page->index; |
78bb9203 NH |
1039 | |
1040 | remove_huge_page(page); | |
ab615a5b MK |
1041 | if (unlikely(hugetlb_unreserve_pages(inode, index, index + 1, 1))) |
1042 | hugetlb_fix_reserve_counts(inode); | |
1043 | ||
78bb9203 NH |
1044 | return 0; |
1045 | } | |
1046 | ||
4a25220d DH |
1047 | /* |
1048 | * Display the mount options in /proc/mounts. | |
1049 | */ | |
1050 | static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root) | |
1051 | { | |
1052 | struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(root->d_sb); | |
1053 | struct hugepage_subpool *spool = sbinfo->spool; | |
1054 | unsigned long hpage_size = huge_page_size(sbinfo->hstate); | |
1055 | unsigned hpage_shift = huge_page_shift(sbinfo->hstate); | |
1056 | char mod; | |
1057 | ||
1058 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) | |
1059 | seq_printf(m, ",uid=%u", | |
1060 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
1061 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
1062 | seq_printf(m, ",gid=%u", | |
1063 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
1064 | if (sbinfo->mode != 0755) | |
1065 | seq_printf(m, ",mode=%o", sbinfo->mode); | |
1066 | if (sbinfo->max_inodes != -1) | |
1067 | seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes); | |
1068 | ||
1069 | hpage_size /= 1024; | |
1070 | mod = 'K'; | |
1071 | if (hpage_size >= 1024) { | |
1072 | hpage_size /= 1024; | |
1073 | mod = 'M'; | |
1074 | } | |
1075 | seq_printf(m, ",pagesize=%lu%c", hpage_size, mod); | |
1076 | if (spool) { | |
1077 | if (spool->max_hpages != -1) | |
1078 | seq_printf(m, ",size=%llu", | |
1079 | (unsigned long long)spool->max_hpages << hpage_shift); | |
1080 | if (spool->min_hpages != -1) | |
1081 | seq_printf(m, ",min_size=%llu", | |
1082 | (unsigned long long)spool->min_hpages << hpage_shift); | |
1083 | } | |
1084 | return 0; | |
1085 | } | |
1086 | ||
726c3342 | 1087 | static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 1088 | { |
726c3342 | 1089 | struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb); |
2b0143b5 | 1090 | struct hstate *h = hstate_inode(d_inode(dentry)); |
1da177e4 LT |
1091 | |
1092 | buf->f_type = HUGETLBFS_MAGIC; | |
a5516438 | 1093 | buf->f_bsize = huge_page_size(h); |
1da177e4 LT |
1094 | if (sbinfo) { |
1095 | spin_lock(&sbinfo->stat_lock); | |
74a8a65c DG |
1096 | /* If no limits set, just report 0 for max/free/used |
1097 | * blocks, like simple_statfs() */ | |
90481622 DG |
1098 | if (sbinfo->spool) { |
1099 | long free_pages; | |
1100 | ||
4b25f030 | 1101 | spin_lock_irq(&sbinfo->spool->lock); |
90481622 DG |
1102 | buf->f_blocks = sbinfo->spool->max_hpages; |
1103 | free_pages = sbinfo->spool->max_hpages | |
1104 | - sbinfo->spool->used_hpages; | |
1105 | buf->f_bavail = buf->f_bfree = free_pages; | |
4b25f030 | 1106 | spin_unlock_irq(&sbinfo->spool->lock); |
74a8a65c DG |
1107 | buf->f_files = sbinfo->max_inodes; |
1108 | buf->f_ffree = sbinfo->free_inodes; | |
1109 | } | |
1da177e4 LT |
1110 | spin_unlock(&sbinfo->stat_lock); |
1111 | } | |
1112 | buf->f_namelen = NAME_MAX; | |
1113 | return 0; | |
1114 | } | |
1115 | ||
1116 | static void hugetlbfs_put_super(struct super_block *sb) | |
1117 | { | |
1118 | struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb); | |
1119 | ||
1120 | if (sbi) { | |
1121 | sb->s_fs_info = NULL; | |
90481622 DG |
1122 | |
1123 | if (sbi->spool) | |
1124 | hugepage_put_subpool(sbi->spool); | |
1125 | ||
1da177e4 LT |
1126 | kfree(sbi); |
1127 | } | |
1128 | } | |
1129 | ||
96527980 CH |
1130 | static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo) |
1131 | { | |
1132 | if (sbinfo->free_inodes >= 0) { | |
1133 | spin_lock(&sbinfo->stat_lock); | |
1134 | if (unlikely(!sbinfo->free_inodes)) { | |
1135 | spin_unlock(&sbinfo->stat_lock); | |
1136 | return 0; | |
1137 | } | |
1138 | sbinfo->free_inodes--; | |
1139 | spin_unlock(&sbinfo->stat_lock); | |
1140 | } | |
1141 | ||
1142 | return 1; | |
1143 | } | |
1144 | ||
1145 | static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo) | |
1146 | { | |
1147 | if (sbinfo->free_inodes >= 0) { | |
1148 | spin_lock(&sbinfo->stat_lock); | |
1149 | sbinfo->free_inodes++; | |
1150 | spin_unlock(&sbinfo->stat_lock); | |
1151 | } | |
1152 | } | |
1153 | ||
1154 | ||
e18b890b | 1155 | static struct kmem_cache *hugetlbfs_inode_cachep; |
1da177e4 LT |
1156 | |
1157 | static struct inode *hugetlbfs_alloc_inode(struct super_block *sb) | |
1158 | { | |
96527980 | 1159 | struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb); |
1da177e4 LT |
1160 | struct hugetlbfs_inode_info *p; |
1161 | ||
96527980 CH |
1162 | if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo))) |
1163 | return NULL; | |
fd60b288 | 1164 | p = alloc_inode_sb(sb, hugetlbfs_inode_cachep, GFP_KERNEL); |
96527980 CH |
1165 | if (unlikely(!p)) { |
1166 | hugetlbfs_inc_free_inodes(sbinfo); | |
1da177e4 | 1167 | return NULL; |
96527980 | 1168 | } |
4742a35d MK |
1169 | |
1170 | /* | |
1171 | * Any time after allocation, hugetlbfs_destroy_inode can be called | |
1172 | * for the inode. mpol_free_shared_policy is unconditionally called | |
1173 | * as part of hugetlbfs_destroy_inode. So, initialize policy here | |
1174 | * in case of a quick call to destroy. | |
1175 | * | |
1176 | * Note that the policy is initialized even if we are creating a | |
1177 | * private inode. This simplifies hugetlbfs_destroy_inode. | |
1178 | */ | |
1179 | mpol_shared_policy_init(&p->policy, NULL); | |
1180 | ||
1da177e4 LT |
1181 | return &p->vfs_inode; |
1182 | } | |
1183 | ||
b62de322 | 1184 | static void hugetlbfs_free_inode(struct inode *inode) |
fa0d7e3d | 1185 | { |
fa0d7e3d NP |
1186 | kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode)); |
1187 | } | |
1188 | ||
1da177e4 LT |
1189 | static void hugetlbfs_destroy_inode(struct inode *inode) |
1190 | { | |
96527980 | 1191 | hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb)); |
1da177e4 | 1192 | mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy); |
1da177e4 LT |
1193 | } |
1194 | ||
f5e54d6e | 1195 | static const struct address_space_operations hugetlbfs_aops = { |
800d15a5 NP |
1196 | .write_begin = hugetlbfs_write_begin, |
1197 | .write_end = hugetlbfs_write_end, | |
46de8b97 | 1198 | .dirty_folio = noop_dirty_folio, |
290408d4 | 1199 | .migratepage = hugetlbfs_migrate_page, |
78bb9203 | 1200 | .error_remove_page = hugetlbfs_error_remove_page, |
1da177e4 LT |
1201 | }; |
1202 | ||
96527980 | 1203 | |
51cc5068 | 1204 | static void init_once(void *foo) |
96527980 CH |
1205 | { |
1206 | struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo; | |
1207 | ||
a35afb83 | 1208 | inode_init_once(&ei->vfs_inode); |
96527980 CH |
1209 | } |
1210 | ||
4b6f5d20 | 1211 | const struct file_operations hugetlbfs_file_operations = { |
34d0640e | 1212 | .read_iter = hugetlbfs_read_iter, |
1da177e4 | 1213 | .mmap = hugetlbfs_file_mmap, |
1b061d92 | 1214 | .fsync = noop_fsync, |
1da177e4 | 1215 | .get_unmapped_area = hugetlb_get_unmapped_area, |
70c3547e MK |
1216 | .llseek = default_llseek, |
1217 | .fallocate = hugetlbfs_fallocate, | |
1da177e4 LT |
1218 | }; |
1219 | ||
92e1d5be | 1220 | static const struct inode_operations hugetlbfs_dir_inode_operations = { |
1da177e4 LT |
1221 | .create = hugetlbfs_create, |
1222 | .lookup = simple_lookup, | |
1223 | .link = simple_link, | |
1224 | .unlink = simple_unlink, | |
1225 | .symlink = hugetlbfs_symlink, | |
1226 | .mkdir = hugetlbfs_mkdir, | |
1227 | .rmdir = simple_rmdir, | |
1228 | .mknod = hugetlbfs_mknod, | |
1229 | .rename = simple_rename, | |
1230 | .setattr = hugetlbfs_setattr, | |
1ab5b82f | 1231 | .tmpfile = hugetlbfs_tmpfile, |
1da177e4 LT |
1232 | }; |
1233 | ||
92e1d5be | 1234 | static const struct inode_operations hugetlbfs_inode_operations = { |
1da177e4 LT |
1235 | .setattr = hugetlbfs_setattr, |
1236 | }; | |
1237 | ||
ee9b6d61 | 1238 | static const struct super_operations hugetlbfs_ops = { |
1da177e4 | 1239 | .alloc_inode = hugetlbfs_alloc_inode, |
b62de322 | 1240 | .free_inode = hugetlbfs_free_inode, |
1da177e4 | 1241 | .destroy_inode = hugetlbfs_destroy_inode, |
2bbbda30 | 1242 | .evict_inode = hugetlbfs_evict_inode, |
1da177e4 | 1243 | .statfs = hugetlbfs_statfs, |
1da177e4 | 1244 | .put_super = hugetlbfs_put_super, |
4a25220d | 1245 | .show_options = hugetlbfs_show_options, |
1da177e4 LT |
1246 | }; |
1247 | ||
7ca02d0a MK |
1248 | /* |
1249 | * Convert size option passed from command line to number of huge pages | |
1250 | * in the pool specified by hstate. Size option could be in bytes | |
1251 | * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT). | |
1252 | */ | |
4a25220d | 1253 | static long |
7ca02d0a | 1254 | hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt, |
4a25220d | 1255 | enum hugetlbfs_size_type val_type) |
7ca02d0a MK |
1256 | { |
1257 | if (val_type == NO_SIZE) | |
1258 | return -1; | |
1259 | ||
1260 | if (val_type == SIZE_PERCENT) { | |
1261 | size_opt <<= huge_page_shift(h); | |
1262 | size_opt *= h->max_huge_pages; | |
1263 | do_div(size_opt, 100); | |
1264 | } | |
1265 | ||
1266 | size_opt >>= huge_page_shift(h); | |
1267 | return size_opt; | |
1268 | } | |
1269 | ||
32021982 DH |
1270 | /* |
1271 | * Parse one mount parameter. | |
1272 | */ | |
1273 | static int hugetlbfs_parse_param(struct fs_context *fc, struct fs_parameter *param) | |
1da177e4 | 1274 | { |
32021982 DH |
1275 | struct hugetlbfs_fs_context *ctx = fc->fs_private; |
1276 | struct fs_parse_result result; | |
1277 | char *rest; | |
1278 | unsigned long ps; | |
1279 | int opt; | |
1280 | ||
d7167b14 | 1281 | opt = fs_parse(fc, hugetlb_fs_parameters, param, &result); |
32021982 DH |
1282 | if (opt < 0) |
1283 | return opt; | |
1284 | ||
1285 | switch (opt) { | |
1286 | case Opt_uid: | |
1287 | ctx->uid = make_kuid(current_user_ns(), result.uint_32); | |
1288 | if (!uid_valid(ctx->uid)) | |
1289 | goto bad_val; | |
1da177e4 | 1290 | return 0; |
1da177e4 | 1291 | |
32021982 DH |
1292 | case Opt_gid: |
1293 | ctx->gid = make_kgid(current_user_ns(), result.uint_32); | |
1294 | if (!gid_valid(ctx->gid)) | |
1295 | goto bad_val; | |
1296 | return 0; | |
e73a75fa | 1297 | |
32021982 DH |
1298 | case Opt_mode: |
1299 | ctx->mode = result.uint_32 & 01777U; | |
1300 | return 0; | |
e73a75fa | 1301 | |
32021982 DH |
1302 | case Opt_size: |
1303 | /* memparse() will accept a K/M/G without a digit */ | |
1304 | if (!isdigit(param->string[0])) | |
1305 | goto bad_val; | |
1306 | ctx->max_size_opt = memparse(param->string, &rest); | |
1307 | ctx->max_val_type = SIZE_STD; | |
1308 | if (*rest == '%') | |
1309 | ctx->max_val_type = SIZE_PERCENT; | |
1310 | return 0; | |
e73a75fa | 1311 | |
32021982 DH |
1312 | case Opt_nr_inodes: |
1313 | /* memparse() will accept a K/M/G without a digit */ | |
1314 | if (!isdigit(param->string[0])) | |
1315 | goto bad_val; | |
1316 | ctx->nr_inodes = memparse(param->string, &rest); | |
1317 | return 0; | |
e73a75fa | 1318 | |
32021982 DH |
1319 | case Opt_pagesize: |
1320 | ps = memparse(param->string, &rest); | |
1321 | ctx->hstate = size_to_hstate(ps); | |
1322 | if (!ctx->hstate) { | |
1323 | pr_err("Unsupported page size %lu MB\n", ps >> 20); | |
1324 | return -EINVAL; | |
e73a75fa | 1325 | } |
32021982 | 1326 | return 0; |
1da177e4 | 1327 | |
32021982 DH |
1328 | case Opt_min_size: |
1329 | /* memparse() will accept a K/M/G without a digit */ | |
1330 | if (!isdigit(param->string[0])) | |
1331 | goto bad_val; | |
1332 | ctx->min_size_opt = memparse(param->string, &rest); | |
1333 | ctx->min_val_type = SIZE_STD; | |
1334 | if (*rest == '%') | |
1335 | ctx->min_val_type = SIZE_PERCENT; | |
1336 | return 0; | |
e73a75fa | 1337 | |
32021982 DH |
1338 | default: |
1339 | return -EINVAL; | |
1340 | } | |
a137e1cc | 1341 | |
32021982 | 1342 | bad_val: |
b5db30cf | 1343 | return invalfc(fc, "Bad value '%s' for mount option '%s'\n", |
32021982 DH |
1344 | param->string, param->key); |
1345 | } | |
7ca02d0a | 1346 | |
32021982 DH |
1347 | /* |
1348 | * Validate the parsed options. | |
1349 | */ | |
1350 | static int hugetlbfs_validate(struct fs_context *fc) | |
1351 | { | |
1352 | struct hugetlbfs_fs_context *ctx = fc->fs_private; | |
a137e1cc | 1353 | |
7ca02d0a MK |
1354 | /* |
1355 | * Use huge page pool size (in hstate) to convert the size | |
1356 | * options to number of huge pages. If NO_SIZE, -1 is returned. | |
1357 | */ | |
32021982 DH |
1358 | ctx->max_hpages = hugetlbfs_size_to_hpages(ctx->hstate, |
1359 | ctx->max_size_opt, | |
1360 | ctx->max_val_type); | |
1361 | ctx->min_hpages = hugetlbfs_size_to_hpages(ctx->hstate, | |
1362 | ctx->min_size_opt, | |
1363 | ctx->min_val_type); | |
7ca02d0a MK |
1364 | |
1365 | /* | |
1366 | * If max_size was specified, then min_size must be smaller | |
1367 | */ | |
32021982 DH |
1368 | if (ctx->max_val_type > NO_SIZE && |
1369 | ctx->min_hpages > ctx->max_hpages) { | |
1370 | pr_err("Minimum size can not be greater than maximum size\n"); | |
7ca02d0a | 1371 | return -EINVAL; |
a137e1cc AK |
1372 | } |
1373 | ||
1da177e4 LT |
1374 | return 0; |
1375 | } | |
1376 | ||
1377 | static int | |
32021982 | 1378 | hugetlbfs_fill_super(struct super_block *sb, struct fs_context *fc) |
1da177e4 | 1379 | { |
32021982 | 1380 | struct hugetlbfs_fs_context *ctx = fc->fs_private; |
1da177e4 LT |
1381 | struct hugetlbfs_sb_info *sbinfo; |
1382 | ||
1da177e4 LT |
1383 | sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL); |
1384 | if (!sbinfo) | |
1385 | return -ENOMEM; | |
1386 | sb->s_fs_info = sbinfo; | |
1387 | spin_lock_init(&sbinfo->stat_lock); | |
32021982 DH |
1388 | sbinfo->hstate = ctx->hstate; |
1389 | sbinfo->max_inodes = ctx->nr_inodes; | |
1390 | sbinfo->free_inodes = ctx->nr_inodes; | |
1391 | sbinfo->spool = NULL; | |
1392 | sbinfo->uid = ctx->uid; | |
1393 | sbinfo->gid = ctx->gid; | |
1394 | sbinfo->mode = ctx->mode; | |
4a25220d | 1395 | |
7ca02d0a MK |
1396 | /* |
1397 | * Allocate and initialize subpool if maximum or minimum size is | |
1935ebd3 | 1398 | * specified. Any needed reservations (for minimum size) are taken |
7ca02d0a MK |
1399 | * taken when the subpool is created. |
1400 | */ | |
32021982 DH |
1401 | if (ctx->max_hpages != -1 || ctx->min_hpages != -1) { |
1402 | sbinfo->spool = hugepage_new_subpool(ctx->hstate, | |
1403 | ctx->max_hpages, | |
1404 | ctx->min_hpages); | |
90481622 DG |
1405 | if (!sbinfo->spool) |
1406 | goto out_free; | |
1407 | } | |
1da177e4 | 1408 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
32021982 DH |
1409 | sb->s_blocksize = huge_page_size(ctx->hstate); |
1410 | sb->s_blocksize_bits = huge_page_shift(ctx->hstate); | |
1da177e4 LT |
1411 | sb->s_magic = HUGETLBFS_MAGIC; |
1412 | sb->s_op = &hugetlbfs_ops; | |
1413 | sb->s_time_gran = 1; | |
15568299 MK |
1414 | |
1415 | /* | |
1416 | * Due to the special and limited functionality of hugetlbfs, it does | |
1417 | * not work well as a stacking filesystem. | |
1418 | */ | |
1419 | sb->s_stack_depth = FILESYSTEM_MAX_STACK_DEPTH; | |
32021982 | 1420 | sb->s_root = d_make_root(hugetlbfs_get_root(sb, ctx)); |
48fde701 | 1421 | if (!sb->s_root) |
1da177e4 | 1422 | goto out_free; |
1da177e4 LT |
1423 | return 0; |
1424 | out_free: | |
6e6870d4 | 1425 | kfree(sbinfo->spool); |
1da177e4 LT |
1426 | kfree(sbinfo); |
1427 | return -ENOMEM; | |
1428 | } | |
1429 | ||
32021982 DH |
1430 | static int hugetlbfs_get_tree(struct fs_context *fc) |
1431 | { | |
1432 | int err = hugetlbfs_validate(fc); | |
1433 | if (err) | |
1434 | return err; | |
2ac295d4 | 1435 | return get_tree_nodev(fc, hugetlbfs_fill_super); |
32021982 DH |
1436 | } |
1437 | ||
1438 | static void hugetlbfs_fs_context_free(struct fs_context *fc) | |
1439 | { | |
1440 | kfree(fc->fs_private); | |
1441 | } | |
1442 | ||
1443 | static const struct fs_context_operations hugetlbfs_fs_context_ops = { | |
1444 | .free = hugetlbfs_fs_context_free, | |
1445 | .parse_param = hugetlbfs_parse_param, | |
1446 | .get_tree = hugetlbfs_get_tree, | |
1447 | }; | |
1448 | ||
1449 | static int hugetlbfs_init_fs_context(struct fs_context *fc) | |
1da177e4 | 1450 | { |
32021982 DH |
1451 | struct hugetlbfs_fs_context *ctx; |
1452 | ||
1453 | ctx = kzalloc(sizeof(struct hugetlbfs_fs_context), GFP_KERNEL); | |
1454 | if (!ctx) | |
1455 | return -ENOMEM; | |
1456 | ||
1457 | ctx->max_hpages = -1; /* No limit on size by default */ | |
1458 | ctx->nr_inodes = -1; /* No limit on number of inodes by default */ | |
1459 | ctx->uid = current_fsuid(); | |
1460 | ctx->gid = current_fsgid(); | |
1461 | ctx->mode = 0755; | |
1462 | ctx->hstate = &default_hstate; | |
1463 | ctx->min_hpages = -1; /* No default minimum size */ | |
1464 | ctx->max_val_type = NO_SIZE; | |
1465 | ctx->min_val_type = NO_SIZE; | |
1466 | fc->fs_private = ctx; | |
1467 | fc->ops = &hugetlbfs_fs_context_ops; | |
1468 | return 0; | |
1da177e4 LT |
1469 | } |
1470 | ||
1471 | static struct file_system_type hugetlbfs_fs_type = { | |
32021982 DH |
1472 | .name = "hugetlbfs", |
1473 | .init_fs_context = hugetlbfs_init_fs_context, | |
d7167b14 | 1474 | .parameters = hugetlb_fs_parameters, |
32021982 | 1475 | .kill_sb = kill_litter_super, |
1da177e4 LT |
1476 | }; |
1477 | ||
42d7395f | 1478 | static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE]; |
1da177e4 | 1479 | |
ef1ff6b8 | 1480 | static int can_do_hugetlb_shm(void) |
1da177e4 | 1481 | { |
a0eb3a05 EB |
1482 | kgid_t shm_group; |
1483 | shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group); | |
1484 | return capable(CAP_IPC_LOCK) || in_group_p(shm_group); | |
1da177e4 LT |
1485 | } |
1486 | ||
42d7395f AK |
1487 | static int get_hstate_idx(int page_size_log) |
1488 | { | |
af73e4d9 | 1489 | struct hstate *h = hstate_sizelog(page_size_log); |
42d7395f | 1490 | |
42d7395f AK |
1491 | if (!h) |
1492 | return -1; | |
04adbc3f | 1493 | return hstate_index(h); |
42d7395f AK |
1494 | } |
1495 | ||
af73e4d9 NH |
1496 | /* |
1497 | * Note that size should be aligned to proper hugepage size in caller side, | |
1498 | * otherwise hugetlb_reserve_pages reserves one less hugepages than intended. | |
1499 | */ | |
1500 | struct file *hugetlb_file_setup(const char *name, size_t size, | |
83c1fd76 | 1501 | vm_flags_t acctflag, int creat_flags, |
1502 | int page_size_log) | |
1da177e4 | 1503 | { |
1da177e4 | 1504 | struct inode *inode; |
e68375c8 | 1505 | struct vfsmount *mnt; |
42d7395f | 1506 | int hstate_idx; |
e68375c8 | 1507 | struct file *file; |
42d7395f AK |
1508 | |
1509 | hstate_idx = get_hstate_idx(page_size_log); | |
1510 | if (hstate_idx < 0) | |
1511 | return ERR_PTR(-ENODEV); | |
1da177e4 | 1512 | |
e68375c8 AV |
1513 | mnt = hugetlbfs_vfsmount[hstate_idx]; |
1514 | if (!mnt) | |
5bc98594 AM |
1515 | return ERR_PTR(-ENOENT); |
1516 | ||
ef1ff6b8 | 1517 | if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) { |
83c1fd76 | 1518 | struct ucounts *ucounts = current_ucounts(); |
1519 | ||
1520 | if (user_shm_lock(size, ucounts)) { | |
1521 | pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is obsolete\n", | |
21a3c273 | 1522 | current->comm, current->pid); |
83c1fd76 | 1523 | user_shm_unlock(size, ucounts); |
353d5c30 | 1524 | } |
83c1fd76 | 1525 | return ERR_PTR(-EPERM); |
2584e517 | 1526 | } |
1da177e4 | 1527 | |
39b65252 | 1528 | file = ERR_PTR(-ENOSPC); |
e68375c8 | 1529 | inode = hugetlbfs_get_inode(mnt->mnt_sb, NULL, S_IFREG | S_IRWXUGO, 0); |
1da177e4 | 1530 | if (!inode) |
e68375c8 | 1531 | goto out; |
e1832f29 SS |
1532 | if (creat_flags == HUGETLB_SHMFS_INODE) |
1533 | inode->i_flags |= S_PRIVATE; | |
1da177e4 | 1534 | |
1da177e4 | 1535 | inode->i_size = size; |
6d6b77f1 | 1536 | clear_nlink(inode); |
ce8d2cdf | 1537 | |
33b8f84a | 1538 | if (!hugetlb_reserve_pages(inode, 0, |
e68375c8 AV |
1539 | size >> huge_page_shift(hstate_inode(inode)), NULL, |
1540 | acctflag)) | |
1541 | file = ERR_PTR(-ENOMEM); | |
1542 | else | |
1543 | file = alloc_file_pseudo(inode, mnt, name, O_RDWR, | |
1544 | &hugetlbfs_file_operations); | |
1545 | if (!IS_ERR(file)) | |
1546 | return file; | |
1da177e4 | 1547 | |
b45b5bd6 | 1548 | iput(inode); |
e68375c8 | 1549 | out: |
39b65252 | 1550 | return file; |
1da177e4 LT |
1551 | } |
1552 | ||
32021982 DH |
1553 | static struct vfsmount *__init mount_one_hugetlbfs(struct hstate *h) |
1554 | { | |
1555 | struct fs_context *fc; | |
1556 | struct vfsmount *mnt; | |
1557 | ||
1558 | fc = fs_context_for_mount(&hugetlbfs_fs_type, SB_KERNMOUNT); | |
1559 | if (IS_ERR(fc)) { | |
1560 | mnt = ERR_CAST(fc); | |
1561 | } else { | |
1562 | struct hugetlbfs_fs_context *ctx = fc->fs_private; | |
1563 | ctx->hstate = h; | |
1564 | mnt = fc_mount(fc); | |
1565 | put_fs_context(fc); | |
1566 | } | |
1567 | if (IS_ERR(mnt)) | |
a25fddce ML |
1568 | pr_err("Cannot mount internal hugetlbfs for page size %luK", |
1569 | huge_page_size(h) >> 10); | |
32021982 DH |
1570 | return mnt; |
1571 | } | |
1572 | ||
1da177e4 LT |
1573 | static int __init init_hugetlbfs_fs(void) |
1574 | { | |
32021982 | 1575 | struct vfsmount *mnt; |
42d7395f | 1576 | struct hstate *h; |
1da177e4 | 1577 | int error; |
42d7395f | 1578 | int i; |
1da177e4 | 1579 | |
457c1b27 | 1580 | if (!hugepages_supported()) { |
9b857d26 | 1581 | pr_info("disabling because there are no supported hugepage sizes\n"); |
457c1b27 NA |
1582 | return -ENOTSUPP; |
1583 | } | |
1584 | ||
d1d5e05f | 1585 | error = -ENOMEM; |
1da177e4 LT |
1586 | hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache", |
1587 | sizeof(struct hugetlbfs_inode_info), | |
5d097056 | 1588 | 0, SLAB_ACCOUNT, init_once); |
1da177e4 | 1589 | if (hugetlbfs_inode_cachep == NULL) |
8fc312b3 | 1590 | goto out; |
1da177e4 LT |
1591 | |
1592 | error = register_filesystem(&hugetlbfs_fs_type); | |
1593 | if (error) | |
8fc312b3 | 1594 | goto out_free; |
1da177e4 | 1595 | |
8fc312b3 | 1596 | /* default hstate mount is required */ |
3b2275a8 | 1597 | mnt = mount_one_hugetlbfs(&default_hstate); |
8fc312b3 MK |
1598 | if (IS_ERR(mnt)) { |
1599 | error = PTR_ERR(mnt); | |
1600 | goto out_unreg; | |
1601 | } | |
1602 | hugetlbfs_vfsmount[default_hstate_idx] = mnt; | |
1603 | ||
1604 | /* other hstates are optional */ | |
42d7395f AK |
1605 | i = 0; |
1606 | for_each_hstate(h) { | |
15f0ec94 JS |
1607 | if (i == default_hstate_idx) { |
1608 | i++; | |
8fc312b3 | 1609 | continue; |
15f0ec94 | 1610 | } |
8fc312b3 | 1611 | |
32021982 | 1612 | mnt = mount_one_hugetlbfs(h); |
8fc312b3 MK |
1613 | if (IS_ERR(mnt)) |
1614 | hugetlbfs_vfsmount[i] = NULL; | |
1615 | else | |
1616 | hugetlbfs_vfsmount[i] = mnt; | |
42d7395f AK |
1617 | i++; |
1618 | } | |
32021982 DH |
1619 | |
1620 | return 0; | |
1da177e4 | 1621 | |
8fc312b3 MK |
1622 | out_unreg: |
1623 | (void)unregister_filesystem(&hugetlbfs_fs_type); | |
1624 | out_free: | |
d1d5e05f | 1625 | kmem_cache_destroy(hugetlbfs_inode_cachep); |
8fc312b3 | 1626 | out: |
1da177e4 LT |
1627 | return error; |
1628 | } | |
3e89e1c5 | 1629 | fs_initcall(init_hugetlbfs_fs) |