]>
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> | |
e73a75fa | 30 | #include <linux/parser.h> |
036e0856 | 31 | #include <linux/mman.h> |
1da177e4 LT |
32 | #include <linux/slab.h> |
33 | #include <linux/dnotify.h> | |
34 | #include <linux/statfs.h> | |
35 | #include <linux/security.h> | |
1fd7317d | 36 | #include <linux/magic.h> |
290408d4 | 37 | #include <linux/migrate.h> |
34d0640e | 38 | #include <linux/uio.h> |
1da177e4 | 39 | |
7c0f6ba6 | 40 | #include <linux/uaccess.h> |
1da177e4 | 41 | |
ee9b6d61 | 42 | static const struct super_operations hugetlbfs_ops; |
f5e54d6e | 43 | static const struct address_space_operations hugetlbfs_aops; |
4b6f5d20 | 44 | const struct file_operations hugetlbfs_file_operations; |
92e1d5be AV |
45 | static const struct inode_operations hugetlbfs_dir_inode_operations; |
46 | static const struct inode_operations hugetlbfs_inode_operations; | |
1da177e4 | 47 | |
a1d776ee | 48 | struct hugetlbfs_config { |
4a25220d DH |
49 | struct hstate *hstate; |
50 | long max_hpages; | |
51 | long nr_inodes; | |
52 | long min_hpages; | |
53 | kuid_t uid; | |
54 | kgid_t gid; | |
55 | umode_t mode; | |
a1d776ee DG |
56 | }; |
57 | ||
58 | struct hugetlbfs_inode_info { | |
59 | struct shared_policy policy; | |
60 | struct inode vfs_inode; | |
61 | }; | |
62 | ||
63 | static inline struct hugetlbfs_inode_info *HUGETLBFS_I(struct inode *inode) | |
64 | { | |
65 | return container_of(inode, struct hugetlbfs_inode_info, vfs_inode); | |
66 | } | |
67 | ||
1da177e4 LT |
68 | int sysctl_hugetlb_shm_group; |
69 | ||
e73a75fa RD |
70 | enum { |
71 | Opt_size, Opt_nr_inodes, | |
72 | Opt_mode, Opt_uid, Opt_gid, | |
7ca02d0a | 73 | Opt_pagesize, Opt_min_size, |
e73a75fa RD |
74 | Opt_err, |
75 | }; | |
76 | ||
a447c093 | 77 | static const match_table_t tokens = { |
e73a75fa RD |
78 | {Opt_size, "size=%s"}, |
79 | {Opt_nr_inodes, "nr_inodes=%s"}, | |
80 | {Opt_mode, "mode=%o"}, | |
81 | {Opt_uid, "uid=%u"}, | |
82 | {Opt_gid, "gid=%u"}, | |
a137e1cc | 83 | {Opt_pagesize, "pagesize=%s"}, |
7ca02d0a | 84 | {Opt_min_size, "min_size=%s"}, |
e73a75fa RD |
85 | {Opt_err, NULL}, |
86 | }; | |
87 | ||
70c3547e MK |
88 | #ifdef CONFIG_NUMA |
89 | static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma, | |
90 | struct inode *inode, pgoff_t index) | |
91 | { | |
92 | vma->vm_policy = mpol_shared_policy_lookup(&HUGETLBFS_I(inode)->policy, | |
93 | index); | |
94 | } | |
95 | ||
96 | static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma) | |
97 | { | |
98 | mpol_cond_put(vma->vm_policy); | |
99 | } | |
100 | #else | |
101 | static inline void hugetlb_set_vma_policy(struct vm_area_struct *vma, | |
102 | struct inode *inode, pgoff_t index) | |
103 | { | |
104 | } | |
105 | ||
106 | static inline void hugetlb_drop_vma_policy(struct vm_area_struct *vma) | |
107 | { | |
108 | } | |
109 | #endif | |
110 | ||
2e9b367c AL |
111 | static void huge_pagevec_release(struct pagevec *pvec) |
112 | { | |
113 | int i; | |
114 | ||
115 | for (i = 0; i < pagevec_count(pvec); ++i) | |
116 | put_page(pvec->pages[i]); | |
117 | ||
118 | pagevec_reinit(pvec); | |
119 | } | |
120 | ||
1da177e4 LT |
121 | static int hugetlbfs_file_mmap(struct file *file, struct vm_area_struct *vma) |
122 | { | |
496ad9aa | 123 | struct inode *inode = file_inode(file); |
1da177e4 LT |
124 | loff_t len, vma_len; |
125 | int ret; | |
a5516438 | 126 | struct hstate *h = hstate_file(file); |
1da177e4 | 127 | |
68589bc3 | 128 | /* |
dec4ad86 DG |
129 | * vma address alignment (but not the pgoff alignment) has |
130 | * already been checked by prepare_hugepage_range. If you add | |
131 | * any error returns here, do so after setting VM_HUGETLB, so | |
132 | * is_vm_hugetlb_page tests below unmap_region go the right | |
133 | * way when do_mmap_pgoff unwinds (may be important on powerpc | |
134 | * and ia64). | |
68589bc3 | 135 | */ |
a2fce914 | 136 | vma->vm_flags |= VM_HUGETLB | VM_DONTEXPAND; |
68589bc3 | 137 | vma->vm_ops = &hugetlb_vm_ops; |
1da177e4 | 138 | |
045c7a3f MK |
139 | /* |
140 | * Offset passed to mmap (before page shift) could have been | |
141 | * negative when represented as a (l)off_t. | |
142 | */ | |
143 | if (((loff_t)vma->vm_pgoff << PAGE_SHIFT) < 0) | |
144 | return -EINVAL; | |
145 | ||
2b37c35e | 146 | if (vma->vm_pgoff & (~huge_page_mask(h) >> PAGE_SHIFT)) |
dec4ad86 DG |
147 | return -EINVAL; |
148 | ||
1da177e4 | 149 | vma_len = (loff_t)(vma->vm_end - vma->vm_start); |
045c7a3f MK |
150 | len = vma_len + ((loff_t)vma->vm_pgoff << PAGE_SHIFT); |
151 | /* check for overflow */ | |
152 | if (len < vma_len) | |
153 | return -EINVAL; | |
1da177e4 | 154 | |
5955102c | 155 | inode_lock(inode); |
1da177e4 | 156 | file_accessed(file); |
1da177e4 LT |
157 | |
158 | ret = -ENOMEM; | |
a1e78772 | 159 | if (hugetlb_reserve_pages(inode, |
a5516438 | 160 | vma->vm_pgoff >> huge_page_order(h), |
5a6fe125 MG |
161 | len >> huge_page_shift(h), vma, |
162 | vma->vm_flags)) | |
a43a8c39 | 163 | goto out; |
b45b5bd6 | 164 | |
4c887265 | 165 | ret = 0; |
b6174df5 | 166 | if (vma->vm_flags & VM_WRITE && inode->i_size < len) |
045c7a3f | 167 | i_size_write(inode, len); |
1da177e4 | 168 | out: |
5955102c | 169 | inode_unlock(inode); |
1da177e4 LT |
170 | |
171 | return ret; | |
172 | } | |
173 | ||
174 | /* | |
508034a3 | 175 | * Called under down_write(mmap_sem). |
1da177e4 LT |
176 | */ |
177 | ||
d2ba27e8 | 178 | #ifndef HAVE_ARCH_HUGETLB_UNMAPPED_AREA |
1da177e4 LT |
179 | static unsigned long |
180 | hugetlb_get_unmapped_area(struct file *file, unsigned long addr, | |
181 | unsigned long len, unsigned long pgoff, unsigned long flags) | |
182 | { | |
183 | struct mm_struct *mm = current->mm; | |
184 | struct vm_area_struct *vma; | |
a5516438 | 185 | struct hstate *h = hstate_file(file); |
08659355 | 186 | struct vm_unmapped_area_info info; |
1da177e4 | 187 | |
a5516438 | 188 | if (len & ~huge_page_mask(h)) |
1da177e4 LT |
189 | return -EINVAL; |
190 | if (len > TASK_SIZE) | |
191 | return -ENOMEM; | |
192 | ||
036e0856 | 193 | if (flags & MAP_FIXED) { |
a5516438 | 194 | if (prepare_hugepage_range(file, addr, len)) |
036e0856 BH |
195 | return -EINVAL; |
196 | return addr; | |
197 | } | |
198 | ||
1da177e4 | 199 | if (addr) { |
a5516438 | 200 | addr = ALIGN(addr, huge_page_size(h)); |
1da177e4 LT |
201 | vma = find_vma(mm, addr); |
202 | if (TASK_SIZE - len >= addr && | |
1be7107f | 203 | (!vma || addr + len <= vm_start_gap(vma))) |
1da177e4 LT |
204 | return addr; |
205 | } | |
206 | ||
08659355 ML |
207 | info.flags = 0; |
208 | info.length = len; | |
209 | info.low_limit = TASK_UNMAPPED_BASE; | |
210 | info.high_limit = TASK_SIZE; | |
211 | info.align_mask = PAGE_MASK & ~huge_page_mask(h); | |
212 | info.align_offset = 0; | |
213 | return vm_unmapped_area(&info); | |
1da177e4 LT |
214 | } |
215 | #endif | |
216 | ||
34d0640e | 217 | static size_t |
e63e1e5a | 218 | hugetlbfs_read_actor(struct page *page, unsigned long offset, |
34d0640e | 219 | struct iov_iter *to, unsigned long size) |
e63e1e5a | 220 | { |
34d0640e | 221 | size_t copied = 0; |
e63e1e5a BP |
222 | int i, chunksize; |
223 | ||
e63e1e5a | 224 | /* Find which 4k chunk and offset with in that chunk */ |
09cbfeaf KS |
225 | i = offset >> PAGE_SHIFT; |
226 | offset = offset & ~PAGE_MASK; | |
e63e1e5a BP |
227 | |
228 | while (size) { | |
34d0640e | 229 | size_t n; |
09cbfeaf | 230 | chunksize = PAGE_SIZE; |
e63e1e5a BP |
231 | if (offset) |
232 | chunksize -= offset; | |
233 | if (chunksize > size) | |
234 | chunksize = size; | |
34d0640e AV |
235 | n = copy_page_to_iter(&page[i], offset, chunksize, to); |
236 | copied += n; | |
237 | if (n != chunksize) | |
238 | return copied; | |
e63e1e5a BP |
239 | offset = 0; |
240 | size -= chunksize; | |
e63e1e5a BP |
241 | i++; |
242 | } | |
34d0640e | 243 | return copied; |
e63e1e5a BP |
244 | } |
245 | ||
246 | /* | |
247 | * Support for read() - Find the page attached to f_mapping and copy out the | |
248 | * data. Its *very* similar to do_generic_mapping_read(), we can't use that | |
ea1754a0 | 249 | * since it has PAGE_SIZE assumptions. |
e63e1e5a | 250 | */ |
34d0640e | 251 | static ssize_t hugetlbfs_read_iter(struct kiocb *iocb, struct iov_iter *to) |
e63e1e5a | 252 | { |
34d0640e AV |
253 | struct file *file = iocb->ki_filp; |
254 | struct hstate *h = hstate_file(file); | |
255 | struct address_space *mapping = file->f_mapping; | |
e63e1e5a | 256 | struct inode *inode = mapping->host; |
34d0640e AV |
257 | unsigned long index = iocb->ki_pos >> huge_page_shift(h); |
258 | unsigned long offset = iocb->ki_pos & ~huge_page_mask(h); | |
e63e1e5a BP |
259 | unsigned long end_index; |
260 | loff_t isize; | |
261 | ssize_t retval = 0; | |
262 | ||
34d0640e | 263 | while (iov_iter_count(to)) { |
e63e1e5a | 264 | struct page *page; |
34d0640e | 265 | size_t nr, copied; |
e63e1e5a BP |
266 | |
267 | /* nr is the maximum number of bytes to copy from this page */ | |
a5516438 | 268 | nr = huge_page_size(h); |
a05b0855 AK |
269 | isize = i_size_read(inode); |
270 | if (!isize) | |
34d0640e | 271 | break; |
a05b0855 | 272 | end_index = (isize - 1) >> huge_page_shift(h); |
34d0640e AV |
273 | if (index > end_index) |
274 | break; | |
275 | if (index == end_index) { | |
a5516438 | 276 | nr = ((isize - 1) & ~huge_page_mask(h)) + 1; |
a05b0855 | 277 | if (nr <= offset) |
34d0640e | 278 | break; |
e63e1e5a BP |
279 | } |
280 | nr = nr - offset; | |
281 | ||
282 | /* Find the page */ | |
a05b0855 | 283 | page = find_lock_page(mapping, index); |
e63e1e5a BP |
284 | if (unlikely(page == NULL)) { |
285 | /* | |
286 | * We have a HOLE, zero out the user-buffer for the | |
287 | * length of the hole or request. | |
288 | */ | |
34d0640e | 289 | copied = iov_iter_zero(nr, to); |
e63e1e5a | 290 | } else { |
a05b0855 AK |
291 | unlock_page(page); |
292 | ||
e63e1e5a BP |
293 | /* |
294 | * We have the page, copy it to user space buffer. | |
295 | */ | |
34d0640e | 296 | copied = hugetlbfs_read_actor(page, offset, to, nr); |
09cbfeaf | 297 | put_page(page); |
e63e1e5a | 298 | } |
34d0640e AV |
299 | offset += copied; |
300 | retval += copied; | |
301 | if (copied != nr && iov_iter_count(to)) { | |
302 | if (!retval) | |
303 | retval = -EFAULT; | |
304 | break; | |
e63e1e5a | 305 | } |
a5516438 AK |
306 | index += offset >> huge_page_shift(h); |
307 | offset &= ~huge_page_mask(h); | |
e63e1e5a | 308 | } |
34d0640e | 309 | iocb->ki_pos = ((loff_t)index << huge_page_shift(h)) + offset; |
e63e1e5a BP |
310 | return retval; |
311 | } | |
312 | ||
800d15a5 NP |
313 | static int hugetlbfs_write_begin(struct file *file, |
314 | struct address_space *mapping, | |
315 | loff_t pos, unsigned len, unsigned flags, | |
316 | struct page **pagep, void **fsdata) | |
1da177e4 LT |
317 | { |
318 | return -EINVAL; | |
319 | } | |
320 | ||
800d15a5 NP |
321 | static int hugetlbfs_write_end(struct file *file, struct address_space *mapping, |
322 | loff_t pos, unsigned len, unsigned copied, | |
323 | struct page *page, void *fsdata) | |
1da177e4 | 324 | { |
800d15a5 | 325 | BUG(); |
1da177e4 LT |
326 | return -EINVAL; |
327 | } | |
328 | ||
b5cec28d | 329 | static void remove_huge_page(struct page *page) |
1da177e4 | 330 | { |
b9ea2515 | 331 | ClearPageDirty(page); |
1da177e4 | 332 | ClearPageUptodate(page); |
bd65cb86 | 333 | delete_from_page_cache(page); |
1da177e4 LT |
334 | } |
335 | ||
4aae8d1c MK |
336 | static void |
337 | hugetlb_vmdelete_list(struct rb_root *root, pgoff_t start, pgoff_t end) | |
338 | { | |
339 | struct vm_area_struct *vma; | |
340 | ||
341 | /* | |
342 | * end == 0 indicates that the entire range after | |
343 | * start should be unmapped. | |
344 | */ | |
345 | vma_interval_tree_foreach(vma, root, start, end ? end : ULONG_MAX) { | |
346 | unsigned long v_offset; | |
347 | unsigned long v_end; | |
348 | ||
349 | /* | |
350 | * Can the expression below overflow on 32-bit arches? | |
351 | * No, because the interval tree returns us only those vmas | |
352 | * which overlap the truncated area starting at pgoff, | |
353 | * and no vma on a 32-bit arch can span beyond the 4GB. | |
354 | */ | |
355 | if (vma->vm_pgoff < start) | |
356 | v_offset = (start - vma->vm_pgoff) << PAGE_SHIFT; | |
357 | else | |
358 | v_offset = 0; | |
359 | ||
360 | if (!end) | |
361 | v_end = vma->vm_end; | |
362 | else { | |
363 | v_end = ((end - vma->vm_pgoff) << PAGE_SHIFT) | |
364 | + vma->vm_start; | |
365 | if (v_end > vma->vm_end) | |
366 | v_end = vma->vm_end; | |
367 | } | |
368 | ||
369 | unmap_hugepage_range(vma, vma->vm_start + v_offset, v_end, | |
370 | NULL); | |
371 | } | |
372 | } | |
b5cec28d MK |
373 | |
374 | /* | |
375 | * remove_inode_hugepages handles two distinct cases: truncation and hole | |
376 | * punch. There are subtle differences in operation for each case. | |
4aae8d1c | 377 | * |
b5cec28d MK |
378 | * truncation is indicated by end of range being LLONG_MAX |
379 | * In this case, we first scan the range and release found pages. | |
380 | * After releasing pages, hugetlb_unreserve_pages cleans up region/reserv | |
1817889e MK |
381 | * maps and global counts. Page faults can not race with truncation |
382 | * in this routine. hugetlb_no_page() prevents page faults in the | |
383 | * truncated range. It checks i_size before allocation, and again after | |
384 | * with the page table lock for the page held. The same lock must be | |
385 | * acquired to unmap a page. | |
b5cec28d MK |
386 | * hole punch is indicated if end is not LLONG_MAX |
387 | * In the hole punch case we scan the range and release found pages. | |
388 | * Only when releasing a page is the associated region/reserv map | |
389 | * deleted. The region/reserv map for ranges without associated | |
1817889e MK |
390 | * pages are not modified. Page faults can race with hole punch. |
391 | * This is indicated if we find a mapped page. | |
b5cec28d MK |
392 | * Note: If the passed end of range value is beyond the end of file, but |
393 | * not LLONG_MAX this routine still performs a hole punch operation. | |
394 | */ | |
395 | static void remove_inode_hugepages(struct inode *inode, loff_t lstart, | |
396 | loff_t lend) | |
1da177e4 | 397 | { |
a5516438 | 398 | struct hstate *h = hstate_inode(inode); |
b45b5bd6 | 399 | struct address_space *mapping = &inode->i_data; |
a5516438 | 400 | const pgoff_t start = lstart >> huge_page_shift(h); |
b5cec28d MK |
401 | const pgoff_t end = lend >> huge_page_shift(h); |
402 | struct vm_area_struct pseudo_vma; | |
1da177e4 | 403 | struct pagevec pvec; |
d72dc8a2 | 404 | pgoff_t next, index; |
a43a8c39 | 405 | int i, freed = 0; |
b5cec28d | 406 | bool truncate_op = (lend == LLONG_MAX); |
1da177e4 | 407 | |
b5cec28d MK |
408 | memset(&pseudo_vma, 0, sizeof(struct vm_area_struct)); |
409 | pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); | |
1da177e4 LT |
410 | pagevec_init(&pvec, 0); |
411 | next = start; | |
b5cec28d | 412 | while (next < end) { |
b5cec28d | 413 | /* |
1817889e | 414 | * When no more pages are found, we are done. |
b5cec28d | 415 | */ |
48f2301c JK |
416 | if (!pagevec_lookup_range(&pvec, mapping, &next, end - 1, |
417 | PAGEVEC_SIZE)) | |
1817889e | 418 | break; |
1da177e4 LT |
419 | |
420 | for (i = 0; i < pagevec_count(&pvec); ++i) { | |
421 | struct page *page = pvec.pages[i]; | |
b5cec28d MK |
422 | u32 hash; |
423 | ||
d72dc8a2 | 424 | index = page->index; |
b5cec28d MK |
425 | hash = hugetlb_fault_mutex_hash(h, current->mm, |
426 | &pseudo_vma, | |
d72dc8a2 | 427 | mapping, index, 0); |
b5cec28d | 428 | mutex_lock(&hugetlb_fault_mutex_table[hash]); |
1da177e4 | 429 | |
4aae8d1c MK |
430 | /* |
431 | * If page is mapped, it was faulted in after being | |
432 | * unmapped in caller. Unmap (again) now after taking | |
433 | * the fault mutex. The mutex will prevent faults | |
434 | * until we finish removing the page. | |
435 | * | |
436 | * This race can only happen in the hole punch case. | |
437 | * Getting here in a truncate operation is a bug. | |
438 | */ | |
439 | if (unlikely(page_mapped(page))) { | |
1817889e | 440 | BUG_ON(truncate_op); |
4aae8d1c MK |
441 | |
442 | i_mmap_lock_write(mapping); | |
443 | hugetlb_vmdelete_list(&mapping->i_mmap, | |
d72dc8a2 JK |
444 | index * pages_per_huge_page(h), |
445 | (index + 1) * pages_per_huge_page(h)); | |
4aae8d1c MK |
446 | i_mmap_unlock_write(mapping); |
447 | } | |
448 | ||
449 | lock_page(page); | |
450 | /* | |
451 | * We must free the huge page and remove from page | |
452 | * cache (remove_huge_page) BEFORE removing the | |
453 | * region/reserve map (hugetlb_unreserve_pages). In | |
454 | * rare out of memory conditions, removal of the | |
72e2936c | 455 | * region/reserve map could fail. Correspondingly, |
456 | * the subpool and global reserve usage count can need | |
457 | * to be adjusted. | |
4aae8d1c | 458 | */ |
72e2936c | 459 | VM_BUG_ON(PagePrivate(page)); |
4aae8d1c MK |
460 | remove_huge_page(page); |
461 | freed++; | |
462 | if (!truncate_op) { | |
463 | if (unlikely(hugetlb_unreserve_pages(inode, | |
d72dc8a2 | 464 | index, index + 1, 1))) |
72e2936c | 465 | hugetlb_fix_reserve_counts(inode); |
b5cec28d MK |
466 | } |
467 | ||
1da177e4 | 468 | unlock_page(page); |
b5cec28d | 469 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); |
1da177e4 LT |
470 | } |
471 | huge_pagevec_release(&pvec); | |
1817889e | 472 | cond_resched(); |
1da177e4 | 473 | } |
b5cec28d MK |
474 | |
475 | if (truncate_op) | |
476 | (void)hugetlb_unreserve_pages(inode, start, LONG_MAX, freed); | |
1da177e4 LT |
477 | } |
478 | ||
2bbbda30 | 479 | static void hugetlbfs_evict_inode(struct inode *inode) |
1da177e4 | 480 | { |
9119a41e JK |
481 | struct resv_map *resv_map; |
482 | ||
b5cec28d | 483 | remove_inode_hugepages(inode, 0, LLONG_MAX); |
9119a41e JK |
484 | resv_map = (struct resv_map *)inode->i_mapping->private_data; |
485 | /* root inode doesn't have the resv_map, so we should check it */ | |
486 | if (resv_map) | |
487 | resv_map_release(&resv_map->refs); | |
dbd5768f | 488 | clear_inode(inode); |
149f4211 CH |
489 | } |
490 | ||
1da177e4 LT |
491 | static int hugetlb_vmtruncate(struct inode *inode, loff_t offset) |
492 | { | |
856fc295 | 493 | pgoff_t pgoff; |
1da177e4 | 494 | struct address_space *mapping = inode->i_mapping; |
a5516438 | 495 | struct hstate *h = hstate_inode(inode); |
1da177e4 | 496 | |
a5516438 | 497 | BUG_ON(offset & ~huge_page_mask(h)); |
856fc295 | 498 | pgoff = offset >> PAGE_SHIFT; |
1da177e4 | 499 | |
7aa91e10 | 500 | i_size_write(inode, offset); |
83cde9e8 | 501 | i_mmap_lock_write(mapping); |
6b2dbba8 | 502 | if (!RB_EMPTY_ROOT(&mapping->i_mmap)) |
1bfad99a | 503 | hugetlb_vmdelete_list(&mapping->i_mmap, pgoff, 0); |
83cde9e8 | 504 | i_mmap_unlock_write(mapping); |
b5cec28d | 505 | remove_inode_hugepages(inode, offset, LLONG_MAX); |
1da177e4 LT |
506 | return 0; |
507 | } | |
508 | ||
70c3547e MK |
509 | static long hugetlbfs_punch_hole(struct inode *inode, loff_t offset, loff_t len) |
510 | { | |
511 | struct hstate *h = hstate_inode(inode); | |
512 | loff_t hpage_size = huge_page_size(h); | |
513 | loff_t hole_start, hole_end; | |
514 | ||
515 | /* | |
516 | * For hole punch round up the beginning offset of the hole and | |
517 | * round down the end. | |
518 | */ | |
519 | hole_start = round_up(offset, hpage_size); | |
520 | hole_end = round_down(offset + len, hpage_size); | |
521 | ||
522 | if (hole_end > hole_start) { | |
523 | struct address_space *mapping = inode->i_mapping; | |
524 | ||
5955102c | 525 | inode_lock(inode); |
70c3547e MK |
526 | i_mmap_lock_write(mapping); |
527 | if (!RB_EMPTY_ROOT(&mapping->i_mmap)) | |
528 | hugetlb_vmdelete_list(&mapping->i_mmap, | |
529 | hole_start >> PAGE_SHIFT, | |
530 | hole_end >> PAGE_SHIFT); | |
531 | i_mmap_unlock_write(mapping); | |
532 | remove_inode_hugepages(inode, hole_start, hole_end); | |
5955102c | 533 | inode_unlock(inode); |
70c3547e MK |
534 | } |
535 | ||
536 | return 0; | |
537 | } | |
538 | ||
539 | static long hugetlbfs_fallocate(struct file *file, int mode, loff_t offset, | |
540 | loff_t len) | |
541 | { | |
542 | struct inode *inode = file_inode(file); | |
543 | struct address_space *mapping = inode->i_mapping; | |
544 | struct hstate *h = hstate_inode(inode); | |
545 | struct vm_area_struct pseudo_vma; | |
546 | struct mm_struct *mm = current->mm; | |
547 | loff_t hpage_size = huge_page_size(h); | |
548 | unsigned long hpage_shift = huge_page_shift(h); | |
549 | pgoff_t start, index, end; | |
550 | int error; | |
551 | u32 hash; | |
552 | ||
553 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
554 | return -EOPNOTSUPP; | |
555 | ||
556 | if (mode & FALLOC_FL_PUNCH_HOLE) | |
557 | return hugetlbfs_punch_hole(inode, offset, len); | |
558 | ||
559 | /* | |
560 | * Default preallocate case. | |
561 | * For this range, start is rounded down and end is rounded up | |
562 | * as well as being converted to page offsets. | |
563 | */ | |
564 | start = offset >> hpage_shift; | |
565 | end = (offset + len + hpage_size - 1) >> hpage_shift; | |
566 | ||
5955102c | 567 | inode_lock(inode); |
70c3547e MK |
568 | |
569 | /* We need to check rlimit even when FALLOC_FL_KEEP_SIZE */ | |
570 | error = inode_newsize_ok(inode, offset + len); | |
571 | if (error) | |
572 | goto out; | |
573 | ||
574 | /* | |
575 | * Initialize a pseudo vma as this is required by the huge page | |
576 | * allocation routines. If NUMA is configured, use page index | |
577 | * as input to create an allocation policy. | |
578 | */ | |
579 | memset(&pseudo_vma, 0, sizeof(struct vm_area_struct)); | |
580 | pseudo_vma.vm_flags = (VM_HUGETLB | VM_MAYSHARE | VM_SHARED); | |
581 | pseudo_vma.vm_file = file; | |
582 | ||
583 | for (index = start; index < end; index++) { | |
584 | /* | |
585 | * This is supposed to be the vaddr where the page is being | |
586 | * faulted in, but we have no vaddr here. | |
587 | */ | |
588 | struct page *page; | |
589 | unsigned long addr; | |
590 | int avoid_reserve = 0; | |
591 | ||
592 | cond_resched(); | |
593 | ||
594 | /* | |
595 | * fallocate(2) manpage permits EINTR; we may have been | |
596 | * interrupted because we are using up too much memory. | |
597 | */ | |
598 | if (signal_pending(current)) { | |
599 | error = -EINTR; | |
600 | break; | |
601 | } | |
602 | ||
603 | /* Set numa allocation policy based on index */ | |
604 | hugetlb_set_vma_policy(&pseudo_vma, inode, index); | |
605 | ||
606 | /* addr is the offset within the file (zero based) */ | |
607 | addr = index * hpage_size; | |
608 | ||
609 | /* mutex taken here, fault path and hole punch */ | |
610 | hash = hugetlb_fault_mutex_hash(h, mm, &pseudo_vma, mapping, | |
611 | index, addr); | |
612 | mutex_lock(&hugetlb_fault_mutex_table[hash]); | |
613 | ||
614 | /* See if already present in mapping to avoid alloc/free */ | |
615 | page = find_get_page(mapping, index); | |
616 | if (page) { | |
617 | put_page(page); | |
618 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
619 | hugetlb_drop_vma_policy(&pseudo_vma); | |
620 | continue; | |
621 | } | |
622 | ||
623 | /* Allocate page and add to page cache */ | |
624 | page = alloc_huge_page(&pseudo_vma, addr, avoid_reserve); | |
625 | hugetlb_drop_vma_policy(&pseudo_vma); | |
626 | if (IS_ERR(page)) { | |
627 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
628 | error = PTR_ERR(page); | |
629 | goto out; | |
630 | } | |
631 | clear_huge_page(page, addr, pages_per_huge_page(h)); | |
632 | __SetPageUptodate(page); | |
633 | error = huge_add_to_page_cache(page, mapping, index); | |
634 | if (unlikely(error)) { | |
635 | put_page(page); | |
636 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
637 | goto out; | |
638 | } | |
639 | ||
640 | mutex_unlock(&hugetlb_fault_mutex_table[hash]); | |
641 | ||
642 | /* | |
643 | * page_put due to reference from alloc_huge_page() | |
644 | * unlock_page because locked by add_to_page_cache() | |
645 | */ | |
646 | put_page(page); | |
647 | unlock_page(page); | |
648 | } | |
649 | ||
650 | if (!(mode & FALLOC_FL_KEEP_SIZE) && offset + len > inode->i_size) | |
651 | i_size_write(inode, offset + len); | |
078cd827 | 652 | inode->i_ctime = current_time(inode); |
70c3547e | 653 | out: |
5955102c | 654 | inode_unlock(inode); |
70c3547e MK |
655 | return error; |
656 | } | |
657 | ||
1da177e4 LT |
658 | static int hugetlbfs_setattr(struct dentry *dentry, struct iattr *attr) |
659 | { | |
2b0143b5 | 660 | struct inode *inode = d_inode(dentry); |
a5516438 | 661 | struct hstate *h = hstate_inode(inode); |
1da177e4 LT |
662 | int error; |
663 | unsigned int ia_valid = attr->ia_valid; | |
664 | ||
665 | BUG_ON(!inode); | |
666 | ||
31051c85 | 667 | error = setattr_prepare(dentry, attr); |
1da177e4 | 668 | if (error) |
1025774c | 669 | return error; |
1da177e4 LT |
670 | |
671 | if (ia_valid & ATTR_SIZE) { | |
672 | error = -EINVAL; | |
1025774c CH |
673 | if (attr->ia_size & ~huge_page_mask(h)) |
674 | return -EINVAL; | |
675 | error = hugetlb_vmtruncate(inode, attr->ia_size); | |
1da177e4 | 676 | if (error) |
1025774c | 677 | return error; |
1da177e4 | 678 | } |
1025774c CH |
679 | |
680 | setattr_copy(inode, attr); | |
681 | mark_inode_dirty(inode); | |
682 | return 0; | |
1da177e4 LT |
683 | } |
684 | ||
7d54fa64 AV |
685 | static struct inode *hugetlbfs_get_root(struct super_block *sb, |
686 | struct hugetlbfs_config *config) | |
1da177e4 LT |
687 | { |
688 | struct inode *inode; | |
1da177e4 LT |
689 | |
690 | inode = new_inode(sb); | |
691 | if (inode) { | |
85fe4025 | 692 | inode->i_ino = get_next_ino(); |
7d54fa64 AV |
693 | inode->i_mode = S_IFDIR | config->mode; |
694 | inode->i_uid = config->uid; | |
695 | inode->i_gid = config->gid; | |
078cd827 | 696 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
7d54fa64 AV |
697 | inode->i_op = &hugetlbfs_dir_inode_operations; |
698 | inode->i_fop = &simple_dir_operations; | |
699 | /* directory inodes start off with i_nlink == 2 (for "." entry) */ | |
700 | inc_nlink(inode); | |
65ed7601 | 701 | lockdep_annotate_inode_mutex_key(inode); |
7d54fa64 AV |
702 | } |
703 | return inode; | |
704 | } | |
705 | ||
b610ded7 | 706 | /* |
c8c06efa | 707 | * Hugetlbfs is not reclaimable; therefore its i_mmap_rwsem will never |
b610ded7 | 708 | * be taken from reclaim -- unlike regular filesystems. This needs an |
88f306b6 | 709 | * annotation because huge_pmd_share() does an allocation under hugetlb's |
c8c06efa | 710 | * i_mmap_rwsem. |
b610ded7 | 711 | */ |
c8c06efa | 712 | static struct lock_class_key hugetlbfs_i_mmap_rwsem_key; |
b610ded7 | 713 | |
7d54fa64 AV |
714 | static struct inode *hugetlbfs_get_inode(struct super_block *sb, |
715 | struct inode *dir, | |
18df2252 | 716 | umode_t mode, dev_t dev) |
7d54fa64 AV |
717 | { |
718 | struct inode *inode; | |
9119a41e JK |
719 | struct resv_map *resv_map; |
720 | ||
721 | resv_map = resv_map_alloc(); | |
722 | if (!resv_map) | |
723 | return NULL; | |
7d54fa64 AV |
724 | |
725 | inode = new_inode(sb); | |
726 | if (inode) { | |
7d54fa64 AV |
727 | inode->i_ino = get_next_ino(); |
728 | inode_init_owner(inode, dir, mode); | |
c8c06efa DB |
729 | lockdep_set_class(&inode->i_mapping->i_mmap_rwsem, |
730 | &hugetlbfs_i_mmap_rwsem_key); | |
1da177e4 | 731 | inode->i_mapping->a_ops = &hugetlbfs_aops; |
078cd827 | 732 | inode->i_atime = inode->i_mtime = inode->i_ctime = current_time(inode); |
9119a41e | 733 | inode->i_mapping->private_data = resv_map; |
1da177e4 LT |
734 | switch (mode & S_IFMT) { |
735 | default: | |
736 | init_special_inode(inode, mode, dev); | |
737 | break; | |
738 | case S_IFREG: | |
739 | inode->i_op = &hugetlbfs_inode_operations; | |
740 | inode->i_fop = &hugetlbfs_file_operations; | |
741 | break; | |
742 | case S_IFDIR: | |
743 | inode->i_op = &hugetlbfs_dir_inode_operations; | |
744 | inode->i_fop = &simple_dir_operations; | |
745 | ||
746 | /* directory inodes start off with i_nlink == 2 (for "." entry) */ | |
d8c76e6f | 747 | inc_nlink(inode); |
1da177e4 LT |
748 | break; |
749 | case S_IFLNK: | |
750 | inode->i_op = &page_symlink_inode_operations; | |
21fc61c7 | 751 | inode_nohighmem(inode); |
1da177e4 LT |
752 | break; |
753 | } | |
e096d0c7 | 754 | lockdep_annotate_inode_mutex_key(inode); |
9119a41e JK |
755 | } else |
756 | kref_put(&resv_map->refs, resv_map_release); | |
757 | ||
1da177e4 LT |
758 | return inode; |
759 | } | |
760 | ||
761 | /* | |
762 | * File creation. Allocate an inode, and we're done.. | |
763 | */ | |
764 | static int hugetlbfs_mknod(struct inode *dir, | |
1a67aafb | 765 | struct dentry *dentry, umode_t mode, dev_t dev) |
1da177e4 LT |
766 | { |
767 | struct inode *inode; | |
768 | int error = -ENOSPC; | |
7d54fa64 AV |
769 | |
770 | inode = hugetlbfs_get_inode(dir->i_sb, dir, mode, dev); | |
1da177e4 | 771 | if (inode) { |
078cd827 | 772 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1da177e4 LT |
773 | d_instantiate(dentry, inode); |
774 | dget(dentry); /* Extra count - pin the dentry in core */ | |
775 | error = 0; | |
776 | } | |
777 | return error; | |
778 | } | |
779 | ||
18bb1db3 | 780 | static int hugetlbfs_mkdir(struct inode *dir, struct dentry *dentry, umode_t mode) |
1da177e4 LT |
781 | { |
782 | int retval = hugetlbfs_mknod(dir, dentry, mode | S_IFDIR, 0); | |
783 | if (!retval) | |
d8c76e6f | 784 | inc_nlink(dir); |
1da177e4 LT |
785 | return retval; |
786 | } | |
787 | ||
ebfc3b49 | 788 | static int hugetlbfs_create(struct inode *dir, struct dentry *dentry, umode_t mode, bool excl) |
1da177e4 LT |
789 | { |
790 | return hugetlbfs_mknod(dir, dentry, mode | S_IFREG, 0); | |
791 | } | |
792 | ||
793 | static int hugetlbfs_symlink(struct inode *dir, | |
794 | struct dentry *dentry, const char *symname) | |
795 | { | |
796 | struct inode *inode; | |
797 | int error = -ENOSPC; | |
1da177e4 | 798 | |
7d54fa64 | 799 | inode = hugetlbfs_get_inode(dir->i_sb, dir, S_IFLNK|S_IRWXUGO, 0); |
1da177e4 LT |
800 | if (inode) { |
801 | int l = strlen(symname)+1; | |
802 | error = page_symlink(inode, symname, l); | |
803 | if (!error) { | |
804 | d_instantiate(dentry, inode); | |
805 | dget(dentry); | |
806 | } else | |
807 | iput(inode); | |
808 | } | |
078cd827 | 809 | dir->i_ctime = dir->i_mtime = current_time(dir); |
1da177e4 LT |
810 | |
811 | return error; | |
812 | } | |
813 | ||
814 | /* | |
6649a386 | 815 | * mark the head page dirty |
1da177e4 LT |
816 | */ |
817 | static int hugetlbfs_set_page_dirty(struct page *page) | |
818 | { | |
d85f3385 | 819 | struct page *head = compound_head(page); |
6649a386 KC |
820 | |
821 | SetPageDirty(head); | |
1da177e4 LT |
822 | return 0; |
823 | } | |
824 | ||
290408d4 | 825 | static int hugetlbfs_migrate_page(struct address_space *mapping, |
b969c4ab | 826 | struct page *newpage, struct page *page, |
a6bc32b8 | 827 | enum migrate_mode mode) |
290408d4 NH |
828 | { |
829 | int rc; | |
830 | ||
831 | rc = migrate_huge_page_move_mapping(mapping, newpage, page); | |
78bd5209 | 832 | if (rc != MIGRATEPAGE_SUCCESS) |
290408d4 NH |
833 | return rc; |
834 | migrate_page_copy(newpage, page); | |
835 | ||
78bd5209 | 836 | return MIGRATEPAGE_SUCCESS; |
290408d4 NH |
837 | } |
838 | ||
78bb9203 NH |
839 | static int hugetlbfs_error_remove_page(struct address_space *mapping, |
840 | struct page *page) | |
841 | { | |
842 | struct inode *inode = mapping->host; | |
843 | ||
844 | remove_huge_page(page); | |
845 | hugetlb_fix_reserve_counts(inode); | |
846 | return 0; | |
847 | } | |
848 | ||
4a25220d DH |
849 | /* |
850 | * Display the mount options in /proc/mounts. | |
851 | */ | |
852 | static int hugetlbfs_show_options(struct seq_file *m, struct dentry *root) | |
853 | { | |
854 | struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(root->d_sb); | |
855 | struct hugepage_subpool *spool = sbinfo->spool; | |
856 | unsigned long hpage_size = huge_page_size(sbinfo->hstate); | |
857 | unsigned hpage_shift = huge_page_shift(sbinfo->hstate); | |
858 | char mod; | |
859 | ||
860 | if (!uid_eq(sbinfo->uid, GLOBAL_ROOT_UID)) | |
861 | seq_printf(m, ",uid=%u", | |
862 | from_kuid_munged(&init_user_ns, sbinfo->uid)); | |
863 | if (!gid_eq(sbinfo->gid, GLOBAL_ROOT_GID)) | |
864 | seq_printf(m, ",gid=%u", | |
865 | from_kgid_munged(&init_user_ns, sbinfo->gid)); | |
866 | if (sbinfo->mode != 0755) | |
867 | seq_printf(m, ",mode=%o", sbinfo->mode); | |
868 | if (sbinfo->max_inodes != -1) | |
869 | seq_printf(m, ",nr_inodes=%lu", sbinfo->max_inodes); | |
870 | ||
871 | hpage_size /= 1024; | |
872 | mod = 'K'; | |
873 | if (hpage_size >= 1024) { | |
874 | hpage_size /= 1024; | |
875 | mod = 'M'; | |
876 | } | |
877 | seq_printf(m, ",pagesize=%lu%c", hpage_size, mod); | |
878 | if (spool) { | |
879 | if (spool->max_hpages != -1) | |
880 | seq_printf(m, ",size=%llu", | |
881 | (unsigned long long)spool->max_hpages << hpage_shift); | |
882 | if (spool->min_hpages != -1) | |
883 | seq_printf(m, ",min_size=%llu", | |
884 | (unsigned long long)spool->min_hpages << hpage_shift); | |
885 | } | |
886 | return 0; | |
887 | } | |
888 | ||
726c3342 | 889 | static int hugetlbfs_statfs(struct dentry *dentry, struct kstatfs *buf) |
1da177e4 | 890 | { |
726c3342 | 891 | struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(dentry->d_sb); |
2b0143b5 | 892 | struct hstate *h = hstate_inode(d_inode(dentry)); |
1da177e4 LT |
893 | |
894 | buf->f_type = HUGETLBFS_MAGIC; | |
a5516438 | 895 | buf->f_bsize = huge_page_size(h); |
1da177e4 LT |
896 | if (sbinfo) { |
897 | spin_lock(&sbinfo->stat_lock); | |
74a8a65c DG |
898 | /* If no limits set, just report 0 for max/free/used |
899 | * blocks, like simple_statfs() */ | |
90481622 DG |
900 | if (sbinfo->spool) { |
901 | long free_pages; | |
902 | ||
903 | spin_lock(&sbinfo->spool->lock); | |
904 | buf->f_blocks = sbinfo->spool->max_hpages; | |
905 | free_pages = sbinfo->spool->max_hpages | |
906 | - sbinfo->spool->used_hpages; | |
907 | buf->f_bavail = buf->f_bfree = free_pages; | |
908 | spin_unlock(&sbinfo->spool->lock); | |
74a8a65c DG |
909 | buf->f_files = sbinfo->max_inodes; |
910 | buf->f_ffree = sbinfo->free_inodes; | |
911 | } | |
1da177e4 LT |
912 | spin_unlock(&sbinfo->stat_lock); |
913 | } | |
914 | buf->f_namelen = NAME_MAX; | |
915 | return 0; | |
916 | } | |
917 | ||
918 | static void hugetlbfs_put_super(struct super_block *sb) | |
919 | { | |
920 | struct hugetlbfs_sb_info *sbi = HUGETLBFS_SB(sb); | |
921 | ||
922 | if (sbi) { | |
923 | sb->s_fs_info = NULL; | |
90481622 DG |
924 | |
925 | if (sbi->spool) | |
926 | hugepage_put_subpool(sbi->spool); | |
927 | ||
1da177e4 LT |
928 | kfree(sbi); |
929 | } | |
930 | } | |
931 | ||
96527980 CH |
932 | static inline int hugetlbfs_dec_free_inodes(struct hugetlbfs_sb_info *sbinfo) |
933 | { | |
934 | if (sbinfo->free_inodes >= 0) { | |
935 | spin_lock(&sbinfo->stat_lock); | |
936 | if (unlikely(!sbinfo->free_inodes)) { | |
937 | spin_unlock(&sbinfo->stat_lock); | |
938 | return 0; | |
939 | } | |
940 | sbinfo->free_inodes--; | |
941 | spin_unlock(&sbinfo->stat_lock); | |
942 | } | |
943 | ||
944 | return 1; | |
945 | } | |
946 | ||
947 | static void hugetlbfs_inc_free_inodes(struct hugetlbfs_sb_info *sbinfo) | |
948 | { | |
949 | if (sbinfo->free_inodes >= 0) { | |
950 | spin_lock(&sbinfo->stat_lock); | |
951 | sbinfo->free_inodes++; | |
952 | spin_unlock(&sbinfo->stat_lock); | |
953 | } | |
954 | } | |
955 | ||
956 | ||
e18b890b | 957 | static struct kmem_cache *hugetlbfs_inode_cachep; |
1da177e4 LT |
958 | |
959 | static struct inode *hugetlbfs_alloc_inode(struct super_block *sb) | |
960 | { | |
96527980 | 961 | struct hugetlbfs_sb_info *sbinfo = HUGETLBFS_SB(sb); |
1da177e4 LT |
962 | struct hugetlbfs_inode_info *p; |
963 | ||
96527980 CH |
964 | if (unlikely(!hugetlbfs_dec_free_inodes(sbinfo))) |
965 | return NULL; | |
e94b1766 | 966 | p = kmem_cache_alloc(hugetlbfs_inode_cachep, GFP_KERNEL); |
96527980 CH |
967 | if (unlikely(!p)) { |
968 | hugetlbfs_inc_free_inodes(sbinfo); | |
1da177e4 | 969 | return NULL; |
96527980 | 970 | } |
4742a35d MK |
971 | |
972 | /* | |
973 | * Any time after allocation, hugetlbfs_destroy_inode can be called | |
974 | * for the inode. mpol_free_shared_policy is unconditionally called | |
975 | * as part of hugetlbfs_destroy_inode. So, initialize policy here | |
976 | * in case of a quick call to destroy. | |
977 | * | |
978 | * Note that the policy is initialized even if we are creating a | |
979 | * private inode. This simplifies hugetlbfs_destroy_inode. | |
980 | */ | |
981 | mpol_shared_policy_init(&p->policy, NULL); | |
982 | ||
1da177e4 LT |
983 | return &p->vfs_inode; |
984 | } | |
985 | ||
fa0d7e3d NP |
986 | static void hugetlbfs_i_callback(struct rcu_head *head) |
987 | { | |
988 | struct inode *inode = container_of(head, struct inode, i_rcu); | |
fa0d7e3d NP |
989 | kmem_cache_free(hugetlbfs_inode_cachep, HUGETLBFS_I(inode)); |
990 | } | |
991 | ||
1da177e4 LT |
992 | static void hugetlbfs_destroy_inode(struct inode *inode) |
993 | { | |
96527980 | 994 | hugetlbfs_inc_free_inodes(HUGETLBFS_SB(inode->i_sb)); |
1da177e4 | 995 | mpol_free_shared_policy(&HUGETLBFS_I(inode)->policy); |
fa0d7e3d | 996 | call_rcu(&inode->i_rcu, hugetlbfs_i_callback); |
1da177e4 LT |
997 | } |
998 | ||
f5e54d6e | 999 | static const struct address_space_operations hugetlbfs_aops = { |
800d15a5 NP |
1000 | .write_begin = hugetlbfs_write_begin, |
1001 | .write_end = hugetlbfs_write_end, | |
1da177e4 | 1002 | .set_page_dirty = hugetlbfs_set_page_dirty, |
290408d4 | 1003 | .migratepage = hugetlbfs_migrate_page, |
78bb9203 | 1004 | .error_remove_page = hugetlbfs_error_remove_page, |
1da177e4 LT |
1005 | }; |
1006 | ||
96527980 | 1007 | |
51cc5068 | 1008 | static void init_once(void *foo) |
96527980 CH |
1009 | { |
1010 | struct hugetlbfs_inode_info *ei = (struct hugetlbfs_inode_info *)foo; | |
1011 | ||
a35afb83 | 1012 | inode_init_once(&ei->vfs_inode); |
96527980 CH |
1013 | } |
1014 | ||
4b6f5d20 | 1015 | const struct file_operations hugetlbfs_file_operations = { |
34d0640e | 1016 | .read_iter = hugetlbfs_read_iter, |
1da177e4 | 1017 | .mmap = hugetlbfs_file_mmap, |
1b061d92 | 1018 | .fsync = noop_fsync, |
1da177e4 | 1019 | .get_unmapped_area = hugetlb_get_unmapped_area, |
70c3547e MK |
1020 | .llseek = default_llseek, |
1021 | .fallocate = hugetlbfs_fallocate, | |
1da177e4 LT |
1022 | }; |
1023 | ||
92e1d5be | 1024 | static const struct inode_operations hugetlbfs_dir_inode_operations = { |
1da177e4 LT |
1025 | .create = hugetlbfs_create, |
1026 | .lookup = simple_lookup, | |
1027 | .link = simple_link, | |
1028 | .unlink = simple_unlink, | |
1029 | .symlink = hugetlbfs_symlink, | |
1030 | .mkdir = hugetlbfs_mkdir, | |
1031 | .rmdir = simple_rmdir, | |
1032 | .mknod = hugetlbfs_mknod, | |
1033 | .rename = simple_rename, | |
1034 | .setattr = hugetlbfs_setattr, | |
1035 | }; | |
1036 | ||
92e1d5be | 1037 | static const struct inode_operations hugetlbfs_inode_operations = { |
1da177e4 LT |
1038 | .setattr = hugetlbfs_setattr, |
1039 | }; | |
1040 | ||
ee9b6d61 | 1041 | static const struct super_operations hugetlbfs_ops = { |
1da177e4 LT |
1042 | .alloc_inode = hugetlbfs_alloc_inode, |
1043 | .destroy_inode = hugetlbfs_destroy_inode, | |
2bbbda30 | 1044 | .evict_inode = hugetlbfs_evict_inode, |
1da177e4 | 1045 | .statfs = hugetlbfs_statfs, |
1da177e4 | 1046 | .put_super = hugetlbfs_put_super, |
4a25220d | 1047 | .show_options = hugetlbfs_show_options, |
1da177e4 LT |
1048 | }; |
1049 | ||
4a25220d | 1050 | enum hugetlbfs_size_type { NO_SIZE, SIZE_STD, SIZE_PERCENT }; |
7ca02d0a MK |
1051 | |
1052 | /* | |
1053 | * Convert size option passed from command line to number of huge pages | |
1054 | * in the pool specified by hstate. Size option could be in bytes | |
1055 | * (val_type == SIZE_STD) or percentage of the pool (val_type == SIZE_PERCENT). | |
1056 | */ | |
4a25220d | 1057 | static long |
7ca02d0a | 1058 | hugetlbfs_size_to_hpages(struct hstate *h, unsigned long long size_opt, |
4a25220d | 1059 | enum hugetlbfs_size_type val_type) |
7ca02d0a MK |
1060 | { |
1061 | if (val_type == NO_SIZE) | |
1062 | return -1; | |
1063 | ||
1064 | if (val_type == SIZE_PERCENT) { | |
1065 | size_opt <<= huge_page_shift(h); | |
1066 | size_opt *= h->max_huge_pages; | |
1067 | do_div(size_opt, 100); | |
1068 | } | |
1069 | ||
1070 | size_opt >>= huge_page_shift(h); | |
1071 | return size_opt; | |
1072 | } | |
1073 | ||
1da177e4 LT |
1074 | static int |
1075 | hugetlbfs_parse_options(char *options, struct hugetlbfs_config *pconfig) | |
1076 | { | |
e73a75fa RD |
1077 | char *p, *rest; |
1078 | substring_t args[MAX_OPT_ARGS]; | |
1079 | int option; | |
7ca02d0a | 1080 | unsigned long long max_size_opt = 0, min_size_opt = 0; |
4a25220d | 1081 | enum hugetlbfs_size_type max_val_type = NO_SIZE, min_val_type = NO_SIZE; |
1da177e4 LT |
1082 | |
1083 | if (!options) | |
1084 | return 0; | |
1da177e4 | 1085 | |
e73a75fa RD |
1086 | while ((p = strsep(&options, ",")) != NULL) { |
1087 | int token; | |
b4c07bce LS |
1088 | if (!*p) |
1089 | continue; | |
e73a75fa RD |
1090 | |
1091 | token = match_token(p, tokens, args); | |
1092 | switch (token) { | |
1093 | case Opt_uid: | |
1094 | if (match_int(&args[0], &option)) | |
1095 | goto bad_val; | |
a0eb3a05 EB |
1096 | pconfig->uid = make_kuid(current_user_ns(), option); |
1097 | if (!uid_valid(pconfig->uid)) | |
1098 | goto bad_val; | |
e73a75fa RD |
1099 | break; |
1100 | ||
1101 | case Opt_gid: | |
1102 | if (match_int(&args[0], &option)) | |
1103 | goto bad_val; | |
a0eb3a05 EB |
1104 | pconfig->gid = make_kgid(current_user_ns(), option); |
1105 | if (!gid_valid(pconfig->gid)) | |
1106 | goto bad_val; | |
e73a75fa RD |
1107 | break; |
1108 | ||
1109 | case Opt_mode: | |
1110 | if (match_octal(&args[0], &option)) | |
1111 | goto bad_val; | |
75897d60 | 1112 | pconfig->mode = option & 01777U; |
e73a75fa RD |
1113 | break; |
1114 | ||
1115 | case Opt_size: { | |
e73a75fa RD |
1116 | /* memparse() will accept a K/M/G without a digit */ |
1117 | if (!isdigit(*args[0].from)) | |
1118 | goto bad_val; | |
7ca02d0a MK |
1119 | max_size_opt = memparse(args[0].from, &rest); |
1120 | max_val_type = SIZE_STD; | |
a137e1cc | 1121 | if (*rest == '%') |
7ca02d0a | 1122 | max_val_type = SIZE_PERCENT; |
e73a75fa RD |
1123 | break; |
1124 | } | |
1da177e4 | 1125 | |
e73a75fa RD |
1126 | case Opt_nr_inodes: |
1127 | /* memparse() will accept a K/M/G without a digit */ | |
1128 | if (!isdigit(*args[0].from)) | |
1129 | goto bad_val; | |
1130 | pconfig->nr_inodes = memparse(args[0].from, &rest); | |
1131 | break; | |
1132 | ||
a137e1cc AK |
1133 | case Opt_pagesize: { |
1134 | unsigned long ps; | |
1135 | ps = memparse(args[0].from, &rest); | |
1136 | pconfig->hstate = size_to_hstate(ps); | |
1137 | if (!pconfig->hstate) { | |
9b857d26 | 1138 | pr_err("Unsupported page size %lu MB\n", |
a137e1cc AK |
1139 | ps >> 20); |
1140 | return -EINVAL; | |
1141 | } | |
1142 | break; | |
1143 | } | |
1144 | ||
7ca02d0a MK |
1145 | case Opt_min_size: { |
1146 | /* memparse() will accept a K/M/G without a digit */ | |
1147 | if (!isdigit(*args[0].from)) | |
1148 | goto bad_val; | |
1149 | min_size_opt = memparse(args[0].from, &rest); | |
1150 | min_val_type = SIZE_STD; | |
1151 | if (*rest == '%') | |
1152 | min_val_type = SIZE_PERCENT; | |
1153 | break; | |
1154 | } | |
1155 | ||
e73a75fa | 1156 | default: |
9b857d26 | 1157 | pr_err("Bad mount option: \"%s\"\n", p); |
b4c07bce | 1158 | return -EINVAL; |
e73a75fa RD |
1159 | break; |
1160 | } | |
1da177e4 | 1161 | } |
a137e1cc | 1162 | |
7ca02d0a MK |
1163 | /* |
1164 | * Use huge page pool size (in hstate) to convert the size | |
1165 | * options to number of huge pages. If NO_SIZE, -1 is returned. | |
1166 | */ | |
1167 | pconfig->max_hpages = hugetlbfs_size_to_hpages(pconfig->hstate, | |
1168 | max_size_opt, max_val_type); | |
1169 | pconfig->min_hpages = hugetlbfs_size_to_hpages(pconfig->hstate, | |
1170 | min_size_opt, min_val_type); | |
1171 | ||
1172 | /* | |
1173 | * If max_size was specified, then min_size must be smaller | |
1174 | */ | |
1175 | if (max_val_type > NO_SIZE && | |
1176 | pconfig->min_hpages > pconfig->max_hpages) { | |
1177 | pr_err("minimum size can not be greater than maximum size\n"); | |
1178 | return -EINVAL; | |
a137e1cc AK |
1179 | } |
1180 | ||
1da177e4 | 1181 | return 0; |
e73a75fa RD |
1182 | |
1183 | bad_val: | |
9b857d26 | 1184 | pr_err("Bad value '%s' for mount option '%s'\n", args[0].from, p); |
c12ddba0 | 1185 | return -EINVAL; |
1da177e4 LT |
1186 | } |
1187 | ||
1188 | static int | |
1189 | hugetlbfs_fill_super(struct super_block *sb, void *data, int silent) | |
1190 | { | |
1da177e4 LT |
1191 | int ret; |
1192 | struct hugetlbfs_config config; | |
1193 | struct hugetlbfs_sb_info *sbinfo; | |
1194 | ||
7ca02d0a | 1195 | config.max_hpages = -1; /* No limit on size by default */ |
1da177e4 | 1196 | config.nr_inodes = -1; /* No limit on number of inodes by default */ |
77c70de1 DH |
1197 | config.uid = current_fsuid(); |
1198 | config.gid = current_fsgid(); | |
1da177e4 | 1199 | config.mode = 0755; |
a137e1cc | 1200 | config.hstate = &default_hstate; |
7ca02d0a | 1201 | config.min_hpages = -1; /* No default minimum size */ |
1da177e4 | 1202 | ret = hugetlbfs_parse_options(data, &config); |
1da177e4 LT |
1203 | if (ret) |
1204 | return ret; | |
1205 | ||
1206 | sbinfo = kmalloc(sizeof(struct hugetlbfs_sb_info), GFP_KERNEL); | |
1207 | if (!sbinfo) | |
1208 | return -ENOMEM; | |
1209 | sb->s_fs_info = sbinfo; | |
a137e1cc | 1210 | sbinfo->hstate = config.hstate; |
1da177e4 | 1211 | spin_lock_init(&sbinfo->stat_lock); |
1da177e4 LT |
1212 | sbinfo->max_inodes = config.nr_inodes; |
1213 | sbinfo->free_inodes = config.nr_inodes; | |
90481622 | 1214 | sbinfo->spool = NULL; |
4a25220d DH |
1215 | sbinfo->uid = config.uid; |
1216 | sbinfo->gid = config.gid; | |
1217 | sbinfo->mode = config.mode; | |
1218 | ||
7ca02d0a MK |
1219 | /* |
1220 | * Allocate and initialize subpool if maximum or minimum size is | |
1221 | * specified. Any needed reservations (for minimim size) are taken | |
1222 | * taken when the subpool is created. | |
1223 | */ | |
1224 | if (config.max_hpages != -1 || config.min_hpages != -1) { | |
1225 | sbinfo->spool = hugepage_new_subpool(config.hstate, | |
1226 | config.max_hpages, | |
1227 | config.min_hpages); | |
90481622 DG |
1228 | if (!sbinfo->spool) |
1229 | goto out_free; | |
1230 | } | |
1da177e4 | 1231 | sb->s_maxbytes = MAX_LFS_FILESIZE; |
a137e1cc AK |
1232 | sb->s_blocksize = huge_page_size(config.hstate); |
1233 | sb->s_blocksize_bits = huge_page_shift(config.hstate); | |
1da177e4 LT |
1234 | sb->s_magic = HUGETLBFS_MAGIC; |
1235 | sb->s_op = &hugetlbfs_ops; | |
1236 | sb->s_time_gran = 1; | |
48fde701 AV |
1237 | sb->s_root = d_make_root(hugetlbfs_get_root(sb, &config)); |
1238 | if (!sb->s_root) | |
1da177e4 | 1239 | goto out_free; |
1da177e4 LT |
1240 | return 0; |
1241 | out_free: | |
6e6870d4 | 1242 | kfree(sbinfo->spool); |
1da177e4 LT |
1243 | kfree(sbinfo); |
1244 | return -ENOMEM; | |
1245 | } | |
1246 | ||
3c26ff6e AV |
1247 | static struct dentry *hugetlbfs_mount(struct file_system_type *fs_type, |
1248 | int flags, const char *dev_name, void *data) | |
1da177e4 | 1249 | { |
3c26ff6e | 1250 | return mount_nodev(fs_type, flags, data, hugetlbfs_fill_super); |
1da177e4 LT |
1251 | } |
1252 | ||
1253 | static struct file_system_type hugetlbfs_fs_type = { | |
1254 | .name = "hugetlbfs", | |
3c26ff6e | 1255 | .mount = hugetlbfs_mount, |
1da177e4 LT |
1256 | .kill_sb = kill_litter_super, |
1257 | }; | |
1258 | ||
42d7395f | 1259 | static struct vfsmount *hugetlbfs_vfsmount[HUGE_MAX_HSTATE]; |
1da177e4 | 1260 | |
ef1ff6b8 | 1261 | static int can_do_hugetlb_shm(void) |
1da177e4 | 1262 | { |
a0eb3a05 EB |
1263 | kgid_t shm_group; |
1264 | shm_group = make_kgid(&init_user_ns, sysctl_hugetlb_shm_group); | |
1265 | return capable(CAP_IPC_LOCK) || in_group_p(shm_group); | |
1da177e4 LT |
1266 | } |
1267 | ||
42d7395f AK |
1268 | static int get_hstate_idx(int page_size_log) |
1269 | { | |
af73e4d9 | 1270 | struct hstate *h = hstate_sizelog(page_size_log); |
42d7395f | 1271 | |
42d7395f AK |
1272 | if (!h) |
1273 | return -1; | |
1274 | return h - hstates; | |
1275 | } | |
1276 | ||
be1d2cf5 | 1277 | static const struct dentry_operations anon_ops = { |
118b2302 | 1278 | .d_dname = simple_dname |
0df4d6e5 AV |
1279 | }; |
1280 | ||
af73e4d9 NH |
1281 | /* |
1282 | * Note that size should be aligned to proper hugepage size in caller side, | |
1283 | * otherwise hugetlb_reserve_pages reserves one less hugepages than intended. | |
1284 | */ | |
1285 | struct file *hugetlb_file_setup(const char *name, size_t size, | |
1286 | vm_flags_t acctflag, struct user_struct **user, | |
42d7395f | 1287 | int creat_flags, int page_size_log) |
1da177e4 | 1288 | { |
39b65252 | 1289 | struct file *file = ERR_PTR(-ENOMEM); |
1da177e4 | 1290 | struct inode *inode; |
2c48b9c4 | 1291 | struct path path; |
0df4d6e5 | 1292 | struct super_block *sb; |
1da177e4 | 1293 | struct qstr quick_string; |
42d7395f AK |
1294 | int hstate_idx; |
1295 | ||
1296 | hstate_idx = get_hstate_idx(page_size_log); | |
1297 | if (hstate_idx < 0) | |
1298 | return ERR_PTR(-ENODEV); | |
1da177e4 | 1299 | |
353d5c30 | 1300 | *user = NULL; |
42d7395f | 1301 | if (!hugetlbfs_vfsmount[hstate_idx]) |
5bc98594 AM |
1302 | return ERR_PTR(-ENOENT); |
1303 | ||
ef1ff6b8 | 1304 | if (creat_flags == HUGETLB_SHMFS_INODE && !can_do_hugetlb_shm()) { |
353d5c30 HD |
1305 | *user = current_user(); |
1306 | if (user_shm_lock(size, *user)) { | |
21a3c273 | 1307 | task_lock(current); |
9b857d26 | 1308 | pr_warn_once("%s (%d): Using mlock ulimits for SHM_HUGETLB is deprecated\n", |
21a3c273 DR |
1309 | current->comm, current->pid); |
1310 | task_unlock(current); | |
353d5c30 HD |
1311 | } else { |
1312 | *user = NULL; | |
2584e517 | 1313 | return ERR_PTR(-EPERM); |
353d5c30 | 1314 | } |
2584e517 | 1315 | } |
1da177e4 | 1316 | |
0df4d6e5 | 1317 | sb = hugetlbfs_vfsmount[hstate_idx]->mnt_sb; |
9d66586f | 1318 | quick_string.name = name; |
1da177e4 LT |
1319 | quick_string.len = strlen(quick_string.name); |
1320 | quick_string.hash = 0; | |
0df4d6e5 | 1321 | path.dentry = d_alloc_pseudo(sb, &quick_string); |
2c48b9c4 | 1322 | if (!path.dentry) |
1da177e4 LT |
1323 | goto out_shm_unlock; |
1324 | ||
0df4d6e5 | 1325 | d_set_d_op(path.dentry, &anon_ops); |
42d7395f | 1326 | path.mnt = mntget(hugetlbfs_vfsmount[hstate_idx]); |
39b65252 | 1327 | file = ERR_PTR(-ENOSPC); |
0df4d6e5 | 1328 | inode = hugetlbfs_get_inode(sb, NULL, S_IFREG | S_IRWXUGO, 0); |
1da177e4 | 1329 | if (!inode) |
ce8d2cdf | 1330 | goto out_dentry; |
e1832f29 SS |
1331 | if (creat_flags == HUGETLB_SHMFS_INODE) |
1332 | inode->i_flags |= S_PRIVATE; | |
1da177e4 | 1333 | |
39b65252 | 1334 | file = ERR_PTR(-ENOMEM); |
af73e4d9 NH |
1335 | if (hugetlb_reserve_pages(inode, 0, |
1336 | size >> huge_page_shift(hstate_inode(inode)), NULL, | |
1337 | acctflag)) | |
b45b5bd6 DG |
1338 | goto out_inode; |
1339 | ||
2c48b9c4 | 1340 | d_instantiate(path.dentry, inode); |
1da177e4 | 1341 | inode->i_size = size; |
6d6b77f1 | 1342 | clear_nlink(inode); |
ce8d2cdf | 1343 | |
2c48b9c4 | 1344 | file = alloc_file(&path, FMODE_WRITE | FMODE_READ, |
ce8d2cdf | 1345 | &hugetlbfs_file_operations); |
39b65252 | 1346 | if (IS_ERR(file)) |
b4d232e6 | 1347 | goto out_dentry; /* inode is already attached */ |
ce8d2cdf | 1348 | |
1da177e4 LT |
1349 | return file; |
1350 | ||
b45b5bd6 DG |
1351 | out_inode: |
1352 | iput(inode); | |
1da177e4 | 1353 | out_dentry: |
2c48b9c4 | 1354 | path_put(&path); |
1da177e4 | 1355 | out_shm_unlock: |
353d5c30 HD |
1356 | if (*user) { |
1357 | user_shm_unlock(size, *user); | |
1358 | *user = NULL; | |
1359 | } | |
39b65252 | 1360 | return file; |
1da177e4 LT |
1361 | } |
1362 | ||
1363 | static int __init init_hugetlbfs_fs(void) | |
1364 | { | |
42d7395f | 1365 | struct hstate *h; |
1da177e4 | 1366 | int error; |
42d7395f | 1367 | int i; |
1da177e4 | 1368 | |
457c1b27 | 1369 | if (!hugepages_supported()) { |
9b857d26 | 1370 | pr_info("disabling because there are no supported hugepage sizes\n"); |
457c1b27 NA |
1371 | return -ENOTSUPP; |
1372 | } | |
1373 | ||
d1d5e05f | 1374 | error = -ENOMEM; |
1da177e4 LT |
1375 | hugetlbfs_inode_cachep = kmem_cache_create("hugetlbfs_inode_cache", |
1376 | sizeof(struct hugetlbfs_inode_info), | |
5d097056 | 1377 | 0, SLAB_ACCOUNT, init_once); |
1da177e4 | 1378 | if (hugetlbfs_inode_cachep == NULL) |
e0bf68dd | 1379 | goto out2; |
1da177e4 LT |
1380 | |
1381 | error = register_filesystem(&hugetlbfs_fs_type); | |
1382 | if (error) | |
1383 | goto out; | |
1384 | ||
42d7395f AK |
1385 | i = 0; |
1386 | for_each_hstate(h) { | |
1387 | char buf[50]; | |
1388 | unsigned ps_kb = 1U << (h->order + PAGE_SHIFT - 10); | |
1da177e4 | 1389 | |
42d7395f AK |
1390 | snprintf(buf, sizeof(buf), "pagesize=%uK", ps_kb); |
1391 | hugetlbfs_vfsmount[i] = kern_mount_data(&hugetlbfs_fs_type, | |
1392 | buf); | |
1da177e4 | 1393 | |
42d7395f | 1394 | if (IS_ERR(hugetlbfs_vfsmount[i])) { |
9b857d26 | 1395 | pr_err("Cannot mount internal hugetlbfs for " |
42d7395f AK |
1396 | "page size %uK", ps_kb); |
1397 | error = PTR_ERR(hugetlbfs_vfsmount[i]); | |
1398 | hugetlbfs_vfsmount[i] = NULL; | |
1399 | } | |
1400 | i++; | |
1401 | } | |
1402 | /* Non default hstates are optional */ | |
1403 | if (!IS_ERR_OR_NULL(hugetlbfs_vfsmount[default_hstate_idx])) | |
1404 | return 0; | |
1da177e4 LT |
1405 | |
1406 | out: | |
d1d5e05f | 1407 | kmem_cache_destroy(hugetlbfs_inode_cachep); |
e0bf68dd | 1408 | out2: |
1da177e4 LT |
1409 | return error; |
1410 | } | |
3e89e1c5 | 1411 | fs_initcall(init_hugetlbfs_fs) |